Canectic BBS - Changelog
Pending (since v0.9.20b)
Security
- Fixed: a registration on the website could choose its own security level, or end up logged in as someone else. The sign-up was used exactly as it arrived, so one carrying extra fields could give the new account any security level (not the Sysop flag), and one carrying another account's uuid was logged in as that account once it was made. Registration now takes only the fields the form asks for, and logs in the account it has just created. Every earlier release has this, so update if your board takes registrations on the website.
- Fixed: a verification code could be guessed. Nothing limited wrong tries, so a script could work through every six-digit code within the 24 hours one lasts. A code now gets five tries. Each wrong one says how many are left, and after the fifth the user must ask for a new code, which can be sent once a minute. Changing the address being verified no longer gets round the minute.
Files
- Fixed: a file area on the website showed only its newest 50 files. The list said "Showing 50 of 340 files" and offered no way to see the rest. It now has pages, with First, Previous, Next and Last under the list, and you can choose 25, 50, 100 or 200 files to a page. The New view has pages too. The filter box and the column sorting cover the page you are looking at, and the filter says so while there is more than one page.
Admin
- Fixed: Node Spy and Who's Online said "Browsing Web" for everyone on the website. They now say what each web user has open: Reading Messages, Browsing Files, Reading Mail, Admin Panel and so on. Node Spy also shows the address of the page under it. With several tabs open, the one the user is actually using is the one shown. Sysop chat still shows as Chatting. A web user's Duration on Node Spy is now how long their visit has lasted; it was the time since their page last checked in, so it started again with every page.
- Fixed: a user's details said their address was unverified when it was verified, and saving the Edit dialog unverified them. Both dialogs on the Users page read the verified setting wrongly, so Details always said No and Edit always started at No: saving any change to a verified user, even to their location, marked their address unverified. The Verified column in the list was right throughout. Both dialogs now show the user's real setting.
- New: a history page for each user. Click a user on the Users page, then History: everything they have done and everything that has happened to their account, newest first. That covers visits and calls with the address they came from, posts, downloads and uploads, door sessions, wall posts, mail sent, poll votes, moderation, wrong passwords and account changes. Choose how far back to look and which kinds to show. A new setting,
user_history_mail_detailunder Security, decides how much of their mail it shows: the count only, the count and who to (the default), or also the subjects. The text is never shown, and a poll vote never shows the answer. See USER_HISTORY.md. - The board now records what happens to each account. A sysop's change to someone's account (which fields, old and new values, and which sysop), password changes and resets, profile and address changes, renames, verification codes sent, wrong and used, wrong passwords on a real account (with the address they came from), accounts made Sysop, sessions ended, API tokens made or removed, and logouts. Before, most of these were written nowhere a board at the usual log level keeps, and the old values were lost. Kept for a year (
user_events_retention_days). The user's history page reads them. - The user card now shows how a user's address verification went. Under Address verification: when the address was verified, whether a code is waiting (when it was sent, how many of its five tries are used, when it expires), and the last five code mails with what happened to each: sent and when, waiting to send, retrying with the error, or failed with the reason. Mails sent before this update are included.
- Fixed: editing an active bulletin switched it off. The Edit dialog always started with the bulletin set to inactive, so saving it, for any change, took it off the board.
Image of the Day
- New: libraries you fill yourself. An Image of the Day library no longer needs a file area. Leave it at None and add pictures on the library's page: pictures, ZIPs laid out like an echo's (each picture with a .txt of the same name), or a picture chosen with its .txt, up to 100 files at a time. Pictures with no day given queue up from today, one a day in name order (Pictures a day raises that), so a batch comes out a day at a time. A day you choose takes every picture you put on it. Nobody sees a day before it comes, on the website, in the home page block or on the terminal; the library page lists the queue. When 3 days of the queue are left, and again at 1, the sysops get a notification, unless the library is set not to warn (for one a script fills on the day itself).
- New: send a library out into a file echo, a day at a time. Under Send it out, choose a file area to hatch into and a file name prefix. When each day comes, the pictures you added for it go out as
PREFIXplus month and day (NASA1103.ZIP): each picture with its text under the same name, as NASA's were, and a FILE_ID.DIZ. Each day goes out once, from the day you turn it on; pictures from the library's own file area are never sent. Next year's file of the same name replaces this year's, here and on the boards down the line (TIC Replaces). The library page lists what went out, and a day that failed tells the sysops and can be tried again. - New: scripts can add pictures through the API. A token with the new Image of the Day scope can add pictures to a library from a script (
/api/v1/iotd): for pictures from a website or anywhere else that is not an echo. The script gives each picture a key of its own, so running it again adds nothing twice. Sysops can always do this; a library's new Who can add pictures rule lets others, and is Sysop only until you change it. See API.md. - New: an example script for NASA's pictures.
examples/iotd/nasa-iotd.mjsreads NASA's Image of the Day feed and adds each new picture to a library through the API, with NASA's title as its caption and NASA's description as its text. By default each picture joins the end of the library's queue, so the feed's 60 pictures give two months of a picture every day, weekends included; with--same-dayeach goes on the day NASA posted it. It remembers what it has sent, so a picture you delete does not come back, and it skips pictures over the board's 64 MB limit. Run it once a day from a timed event or cron. It is an example and not part of the board, so copy it out of the install before changing it. NASA's Astronomy Picture of the Day, which FSX_IMGE carried, cannot be used this way for now: since it moved to science.nasa.gov its API answers with the wrong pictures. See IMAGE_OF_THE_DAY.md.
v0.9.20b (2026-10-23)
Security
- Fixed: an update could install a release that did not come from us. The release's signature was checked by the release's own updater, so a release built by someone else could leave the check out, and an unsigned release was accepted unless you had turned on
update_require_signature. Anyone able to put a release on the release site, or into the release file echo, could have had their code run by a sysop pressing Update. Your board now checks the signature itself, with the release key it already has, before any of the new release's code runs. A stable release that is not signed by the release key is refused whatever the setting; a development build may still be unsigned on a board on the development channel with the setting off. An archive holding files outside the signed list is refused too, and so is a genuine older release offered as a newer one. This protects every update from this release on: the update to this release is still checked the old way. - New: a revoked release key reaches your board before a release signed with it does. Your board now looks for news of the Canectic release keys on every update check and again just before applying an update: on the release site, on a keyserver (
update_keyserver_url, keys.openpgp.org; HTTP only, blank turns it off), and in the release file echo, where a key file is taken the moment it is tossed. A revocation used to reach a board only inside a later release, after the release it was meant to stop. When your board learns of one it tells you, and refuses releases signed with that key. See Release keys indocs/sysop/UPDATES.md. - Fixed: one release could change which release key a board trusts. A board trusted the key in its installed key file, which every update replaces, so a release signed with a stolen signing key could leave a board trusting a key of the thief's choosing, for good. A board now keeps its own list of the release keys it trusts, in
.canectic-trust/, which no update replaces, and a release can change it only through a signature by a key the board already trusts. Everything the board learns about those keys is kept, so a revoked key stays revoked. - New: a second release key, kept in reserve. If the key that signs releases ever has to be revoked, the reserve signs from then on and your board carries on updating, rather than having nothing left it trusts. It is kept offline and has never been on the machine that builds releases. Your board starts trusting it by itself, because the current release key has vouched for it; its fingerprint is
DE1A 1F1F E233 5146 64DE 7113 8ABB 2D2E C0D1 3127, listed indocs/sysop/INSTALL.mdbeside the current one.
Updates
- New: an update that changes no Node packages needs no internet. It used to run
npm installagainst the npm registry every time. A board with no internet, taking its releases down the file echo, can now update whenever the packages have not changed, and the Update page andupdate.jssay beforehand whether a release changes them (Packages: unchanged or changed). One that does needs the registry, or a mirror set in.npmrc. - Fixed: a release that arrived down the release file echo was never offered. The update check looked for the file in the wrong directory, so a board taking its releases over FTN never saw one.
Installing
- New: install from the release archive alone.
bootstrap.sh --from DIRused to need the index, metadata and checksum list beside the archive, which a file echo does not deliver. Given the archive alone, it now checks theMANIFESTsigned inside it, then every file against thatMANIFEST, refusing anything that does not match or is not listed, before running anything from it. Upper-cased names are fine. See Installing from the archive alone indocs/sysop/INSTALL.md. - Changed: a first install checks the release keys on a keyserver, and needs
gpg. The bootstrap looks the release keys up on keys.openpgp.org first, so a key revoked since it was written is refused, and stops if it cannot ask (--no-keyserveronce you have compared the fingerprint yourself). It now refuses a release whose signature it cannot check at all, where it used to carry on with a warning;gpghas always been in the install instructions.
Newsletters
- New: newsletters (#108). A newsletter is written on your board, or taken from where a network publishes one: an echo (the messages of a named poster, or every message, with numbered parts such as FidoNews's "[00/06]" to "[06/06]" joined back into one issue) or a file echo (the text in each file hatched there, FidoNews's
FNEWTH42.ZIPfor one). Readers find the issues on the website at/newslettersand on the terminal (menu command Newsletters, with a lightbar to pick newsletters and issues), newest first with back issues kept, and a home page block and a login stage show the newest issue when it is new since their last visit. Readers can ask for each new issue by email or netmail, to a verified address, with the notification sending switched on. An issue written here can also be posted into an echo, in parts when it is long, and hatched into a file echo as an 8.3 ZIP with a FILE_ID.DIZ. Set them up under Content > Newsletters; seedocs/sysop/NEWSLETTERS.md.
Polls
- New: polls (#13), the voting booth. A question with up to 20 answers, one to choose or several, voted on from the website at
/polls, from a Polls block on the home page or in the web call flow (voted in without leaving the page), and on the terminal (menu command Polls, and a login stage that asks each poll waiting for the caller's vote). Each account has one vote, which it can change until the poll closes. A poll can let callers add answers of their own, closes when you say, and shows its results after people vote or only once it closes. Only counts are ever shown: who voted for what is shown to nobody, sysops included. Set them up under Content > Polls, where you can also let callers start polls of their own (sysops only until you do); seedocs/sysop/POLLS.md.
Top 10 Lists
- New: a Top 10 page on the website, and a link to it for the navigation bar.
/toptenshows every list with someone on it, with buttons for the last 7 days, the last 30 days and all time, and each list opens on its own page with all 25 it keeps. Add it to the bar under Menus > Web Navigation, A page on this board, Top 10. The lists used to be on the website only as home page or call flow blocks. - Fixed: the terminal's Top 10 lists said only "7 3 A period" at the bottom. They now name each period beside its key,
7 Last 7 days 3 Last 30 days A All time, with the one on screen marked, and Left and Right change the period too.
Menus
- Fixed: a menu hotkey of more than one character never worked, including Ctrl keys written as
^N: the editor took 1 to 4 characters, but the menu matched only one typed key. An entry saved as^Nworks now, as Ctrl+N, without being changed. - New: Ctrl keys and ; commands as hotkeys. Beside each entry's hotkey, choose Key, Ctrl + (a letter pressed with Ctrl, shown as
^N) or ; command (the caller types;, a word of up to 12 letters or digits, and Enter, such as;NODESPY), so sysop commands need not use up a key. A caller with no ; command they may use gets no prompt when they press;. Ctrl keys a terminal uses for something else (H, I, J, M) are refused, and S, Q and C saved with a warning. - New: global commands. Entries that work in every menu, set up once from the Global Commands card on Menus > Menu Editor. Each is shown on every menu or works everywhere without being shown; a menu's own entry with the same hotkey wins in that menu. See Global Commands in
docs/sysop/MENUS.md. - Changed: on a new board, Node Spy is the global command
;SPY, for sysops only, in place of N on the starter Main menu, so it takes no key a caller could want. Boards already set up keep their menus as they are; add;SPYunder Global Commands if you want it.
Notifications
- Fixed: Up and Down did not move the bar in the terminal's notification list, so only the first notification could be opened. They do now.
Upgrade Notes
- Updates need
gpgfrom this release on:sudo apt install gnupg(it has been in the install instructions all along). Without it the next update is refused, saying what to install. System Info shows whether it is there, under External programs.
v0.9.19b (2026-10-10)
Security
- Fixed: some admin pages could be read by anyone, logged in or not. The admin view of a message area and its threads, the message group and file group lists, Node Spy's list of who is online, and the terminal styles page showed their content to any visitor, the messages whatever the area's read security. Every admin page now refuses everyone but a sysop: a visitor is shown the login form, and a user who is not a sysop is told they are not authorized. Their buttons were never usable by anyone else, and Node Spy's live view of a caller's screen always checked.
Notifications
- New: notifications and news by email or netmail. Users can have their notifications (private mail received, their door server registration answered) and your news sent to the email or netmail address on their account, as they happen or once a day in one message, from Notifications on their profile or N in the terminal's User Settings. Only to a verified address, and every message says how to stop. As it happens is limited to
notify_instant_per_hourmessages an hour (6), after which the rest go together. Sysops can have their own notifications sent the same way, chosen by type: problems such as a stopped worker or a file in Bad Files are ticked to start with. Off until you turn onnotify_email_enabledornotify_netmail_enabled(Configuration, System); the daily summary goes atnotify_summary_hour(7). Seedocs/sysop/NOTIFICATIONS.md. - New: send an announcement by email or netmail. Tick Also send it by email or netmail when adding or editing one, and it goes once, when it is published, to everyone who asked for news. The announcement list shows when it went and to how many.
Messages
- New: moderators. Give a user moderation of a message group (every area in it, including areas added later) or of a single area, and they can delete and move anyone's threads and messages there, on the website and in the terminal message reader, as you can. Appoint them on the group's page under Messages > Message Groups: the group's in its card, an area's in the area's info. Users see who moderates an area under its name.
- New: the Moderation page (Messages > Moderation, and on the dashboard). It lists every moderator, and every thread and message deleted or moved by hand, by a moderator or a sysop, on the website, the terminal or the admin pages: when, who, what, where it went, and for a deletion the text that was deleted. Kept for
moderation_log_retention_days(365). Seedocs/sysop/MODERATION.md. - Fixed: deleting a message from the terminal message reader said "Could not delete that message." although it had been deleted, and left it on the screen.
Admin
- Fixed: most admin pages were blank on a phone, the dashboard, Configuration, Users, Networks, Peers, Menus, Doors, Themes, Events, the message and file area pages and others among them: 43 pages showed their content to desktops only. They now show it on every device, and fit a phone screen: columns stack, wide tables scroll sideways inside their card with the search box and paging left in place, and dialogs fit the screen. Desktop layouts are unchanged.
Archives
- New: archives in every common BBS format, not only ZIP. Mail bundles, nodelists, echolists, filebones and infopacks, QWK packets, file descriptions (FILE_ID.DIZ), file previews, Image of the Day and door lists now open ARJ, LHA, RAR, ARC, ZOO, 7z, gzip, bzip2, xz and tar archives as well as ZIP. The format is read from the file itself, never from its name. ZIP needs nothing extra; the others need programs, which nothing installs for you:
sudo apt install 7zip nomarch unar(RAR through 7-Zip also needs7zip-rar). Where a program is missing, Canectic says what to install rather than failing, and the External programs section of System Info has a new archive formats row showing what the board can open. Seedocs/sysop/ARCHIVES.md. - Fixed: a mail bundle packed with LHA, RAR or ARC sat in the inbound directory for good, and an ARJ bundle went to Bad Files unread. They are unpacked now, given the program for their format. A bundle in a format no program on the board opens goes to Bad Files as Bundle needs a program, with the program to install, and Retry unpacks it once you have (an ARJ bundle could not be retried before).
- Fixed: a file named like a mail bundle that was not an archive at all stayed in the inbound directory for good. It now goes to Bad Files once it is a minute old, like a bundle that will not unpack.
- Fixed: unpacking a mail bundle replaced a packet of the same name already in the inbound directory, losing it. The new one is now kept beside it as
NAME_1.PKT. - Fixed: an ARJ, RAR, LHA or 7z upload got no description and no preview, and a picture in one never reached an Image of the Day library.
- Files already in your file areas without a description get one from their FILE_ID.DIZ when you press Regen File Desc HTML on the Events page, once the programs are installed. It now looks in every file that has no description, which includes ZIPs from old versions of PKZIP whose DIZ could not be read before. A short description someone wrote is kept; one that is only the file's name is replaced by the DIZ's first line.
- Fixed: a nodelist packed with ARC, as FidoNet's NODELIST.A## is, could not be imported. NODEDIFF files can now be opened too, but are not yet applied to the nodelist.
- Fixed: the daily echolist and filebone scan picked .zip, .lzh, .rar and .7z files by their names, then opened only the ZIP and 7z ones, logging "Unsupported archive format" for the rest. It now takes the newest archive in the area, whatever its format and name.
Upgrade Notes
- Archives in formats other than ZIP need programs the update does not install:
sudo apt install 7zip nomarch unar, and7zip-rarfor RAR through 7-Zip. Until they are installed ZIP works as before, and a mail bundle in another format goes to Bad Files as Bundle needs a program, where Retry unpacks it once you have; ARJ bundles already in Bad Files can be retried the same way. Once they are installed, press Regen File Desc HTML on the Events page to give files already in your file areas the description in their FILE_ID.DIZ. - A file named like a mail bundle that is not an archive, which used to sit in the inbound directory for good, goes to Bad Files on the first toss after updating.
- Nothing is sent by email or netmail until you turn on
notify_email_enabledornotify_netmail_enabled(Configuration, System). Users can make their choices before then; nothing is saved up to send when you do.
v0.9.18b (2026-10-09)
Security
- Fix: a message from another system could put working HTML on your message pages. Message text that looked like HTML was put on the web page as HTML, on the message reader, the admin message view and private mail, and the echomail, QWK and NNTP imports did not filter it. A message could therefore carry script that ran in the browser of whoever read it, you included, with their login. On the development board a pasted web page in a politics echo was loading an embedded video player from another site. Message text is now always shown as text, whatever it contains, and the update redraws the stored messages this affects (see the upgrade note).
- Fix: an archive's file name could run a command on your board. The daily echolist and filebone scan unpacked archives with a command line that the archive's name could break out of, so an archive arriving in an area the scan reads (a network's echolist, filebone or infopack area, or a list's own scan area), by TIC or by upload, with shell characters in its name ran a command as the user the BBS runs as. Archives are now unpacked without a shell, whatever their names.
src/tools/diagnoseScanners.jshad the same fault and is fixed with it.
Messages
- Fixed: text in angle brackets disappeared from messages on the web, such as the address in
Marc Lewis <user@example.org> wrote:, a<g>, or#include <stdio.h>. It is now shown. - Fixed: ANSI art and coloured messages lost their colours on the web. Bold colours showed in their dark versions; black, dark grey, light grey and white could not be told apart, so shading came out flat; green, grey and white backgrounds showed as coloured text instead; coloured spaces at the end of a line, and lines of nothing but colour, disappeared; and the first characters of every row of art lost their colour. Messages now use the 16 colours a terminal shows, with art on black, the same colours the art previews on the file pages use, so a message looks as it does to a terminal caller. Codes that draw nothing no longer leave text such as
[?7hon the page, and a SAUCE record posted with a piece of art is no longer shown.
Image of the Day
- New: Image of the Day. A library takes the pictures that arrive in one of your file areas, usually a file echo (fsxNet's FSX_IMGE carries NASA's Astronomy Picture of the Day among others), and keeps them by day for as many days as you choose, for the website and for terminal callers. Set them up under Content > Image of the Day: the file area, the days to keep, whether to take videos too, and who sees and who can open each library, by security level and flags, so a library can be for everyone, for users, or for a few. A picture's day is its file's own date, not the day it arrived, so a rescan delivering weeks at once files each on its day. The text that comes with a picture (NASA's write-up, or the file's description) is kept exactly as written, and its title line becomes the picture's caption.
- On the website,
/iotdshows each library a day at a time: the picture first, as large as it fits, with its caption, and the full text beside it on a wide screen or below it on a phone; the days either side, a day list and the recent days. Add a block for a library to the home page or the web call flow, and an Image of the Day link to the navigation bar. - On the terminal, an Image of the Day menu command and call flow stage show the day's picture to callers with sixel graphics (SyncTERM and others), fitted to their screen, with the text behind I and the days behind < and >. Callers without sixel get the picture's web address (set
sitedomainso they are given one), or, if you set the library to, the picture drawn in ANSI blocks: look at a few with As ANSI blocks on the library's page first. Seedocs/sysop/IMAGE_OF_THE_DAY.md.
Terminal
- New: terminal flags. A menu entry, a terminal call flow stage or a door can now depend on what the caller's terminal can do: Sixel, RIP, ANSI, UTF-8, CP437, Browser terminal (the website's terminal) and 80x24 or larger, each set to Any, Required or Prohibited. They belong to the call, found when the caller connects, never to the user, so a user flag with the same name does not pass. A menu entry the caller's terminal fails is hidden; a stage is skipped, and a Login Flow replay says why; a door is left out of the door browser and of what the door server offers, is shown on the website's door pages with the reason in place of Play when the website's terminal cannot run it, and is refused however it is launched, telling a terminal caller why. Set them on the menu entry and door forms (a Terminal row) and with the screen button on a call flow stage. Give the Image of the Day entry or stage Sixel, Required, and callers without sixel no longer see it. See
docs/sysop/TERMINAL_FLAGS.md.
Files
- New: previews on the file pages. The info pop-up and each file's page now show what is in a file: a thumbnail of a picture, the first few pictures of an archive that is mostly pictures (with how many there are), ANSI and ASCII art drawn as a picture (from an .ANS file, or the first piece in an art pack), and an MP3's cover. An archive can also bring its own previews, which are used instead:
FILE_ID.PNG(or .JPG, .GIF, .WEBP) andFILE_ID1toFILE_ID9for more pictures,FILE_ID.ANSfor art, andFILE_ID.MP3or.OGGfor a short sound sample played on the page. That convention is written up for any BBS software to follow indocs/reference/FILE_ID_MEDIA.md, published on the Canectic site. Previews need the same access as a download and do not count as one. They are made the first time a file is looked at and kept infiles/previews. Turn them off withfile_previewsunder Files on the Configuration page. - Fixed: a file's SHA-256 ran past the edge of its page on all but the widest screens.
- Fixed: a coloured FILE_ID.DIZ showed its raw colour codes on the file pages, and a DIZ written for DOS lost its drawing characters. Boxes, blocks and lines were stored as question marks, because the DIZ was read in the wrong character set. Descriptions are now read as DOS text, drawn in colour as a terminal shows them, and set in a fixed-width font on black with the spaces that line them up, on the file page and in the info pop-up; a TIC's long description looks the same. A short description taken from a DIZ skips a decorative top border and leaves out colour codes.
- Fixed: editing a file's long description on the admin page did not change what the website showed. The page kept showing the description as it first arrived.
- Fixed: Regen File Desc HTML on the Events page rewrote every description as plain text. It now draws them as new files are drawn, and a description showing question marks in place of drawing characters is read again from its archive, when the file is still in its area.
Networking
- New: Bad Files. A file that cannot be tossed is still set aside in the bad directory (
ftn/bad), and now you are told why. The notification names the reason (a CRC that does not match, a file echo not carried here, a wrong TIC password, a file that never arrived, and so on), with the detail, who sent it and the area, and links to a new Bad Files page under Networking (and on the dashboard). There each file can be viewed, downloaded, deleted, or retried: Retry puts it back in the inbound directory and tosses it again, for after you have mapped the area or corrected the password, and the page says how that went. Netmail the relay refused is listed too. Files already in the bad directory are listed as "Reason not recorded". New settingbad_file_retention_days(30, 0 = keep) removes old ones with their records. Seedocs/sysop/BAD_FILES.md. - Fixed: a mail packet that could not be read stayed in the inbound directory and failed again on every toss, with nothing to tell you; a mail bundle that would not unpack did the same, and an ARJ bundle was never touched. They are now set aside in the bad directory and listed on the Bad Files page with the reason. A packet or bundle is only set aside once it is a minute old, in case it is still being copied in.
- Fixed: a bad TIC with the same name as one already in the bad directory replaced it, losing the earlier one. Each is now kept under a name of its own.
- Fixed: a TIC that failed with an unexpected error left its file in the inbound directory, and raised no notification. Both now go to the bad directory together, and you are told what the error was.
- Fixed: files from a file echo could arrive twice, with half their TICs in Bad Files as "File did not arrive" and the second copies left in the inbound directory as
NAME_<number>.ZIPfor good. Three faults lined up. The mailer could open two sessions to the same system at once, each sending everything waiting, and accepted a second session from a system already connected; two outbound TIC passes running at the same moment each sent every file, two TICs apiece; and a second copy of a file still waiting in the inbound was saved under a name its TIC could not find. Now a system has one session at a time: a second call to a system already connected is not made, and a second session from one is turned away as busy and comes back later, as other binkp mailers do. Each waiting file is sent by one outbound pass. A file received again unchanged is kept once, a TIC finds a copy kept under another name by its CRC, and a TIC for a file its area already holds goes with the tossed TICs instead of to Bad Files. Seen as a FileFix rescan between two Canectic boards that came through twice. TICs already in Bad Files for this can be retried: they now find their files or are recognised as repeats. - Fixed: Recover and Purge on the Bad Echomail page never said what they had done. The page now shows how many messages were recovered, were dupes or remain, and how many were purged.
Upgrade Notes
- File previews add one library, sharp, which the update downloads (about 20 MB, a ready-built binary for Linux on x64 and ARM). Where none exists for your system, the update still works and pictures simply show no preview; System Info lists the library's version.
- The update redraws the web view of the messages the message fixes affect: those with text that looks like HTML, or with ANSI, pipe or Ctrl-A colour codes, in message areas and private mail. It runs while the board is closed for the update; on the development board it redrew 7,436 of 114,000 messages and the whole database step took under a minute. Message colours are now the terminal's 16, so ANSI blue, for one, is darker than before. If you restyled message colours in
custom.css, message bodies now use theansi-f0toansi-f15(text) andansi-b0toansi-b15(background) classes instead of thefg-andbg-ones. - The update also redraws stored file descriptions in the new style, in a second or two. It cannot repair a description whose drawing characters were already lost as question marks; press Regen File Desc HTML on the Events page afterwards to read those again from their archives.
- Packets and bundles that have been stuck in the inbound directory move to the bad directory on the first toss after updating, each with a notification, and files already in the bad directory are listed on the Bad Files page as "Reason not recorded". After
bad_file_retention_days(30) they are deleted; set it to 0 first to keep them.
v0.9.17b (2026-10-08)
Administration
- New: sysop alerts for a desktop monitor. The board now keeps a short feed of what a sysop wants to hear about as it happens: a terminal caller connecting (only once they are past the terminal screen, so never a bot), a terminal caller logging in, a terminal caller logging off or dropping carrier, a website login or a user coming back to the website, a website Log out, a new user, a file upload, private mail reaching a sysop account, and every sysop notification (these even with
notifications_enabledoff). It is for Canectic Monitor, a desktop app for Linux and Windows that plays a sound for each, downloadable from the releases page with version numbers of its own, and for scripts of your own:/api/v1/monitoras a WebSocket, or/api/v1/monitor/eventsread once. Reading it takes an API token with the new Sysop Monitor scope, which only a sysop account can be given, and which stops working if its user stops being a sysop. Behind a reverse proxy,/api/v1/monitormust be passed through as a WebSocket. See "Sysop monitor" indocs/sysop/API.md. - Getting around the networks pages. The network pages and everything under a network (echolists, filebones, the nodelist and the coordinator) now share one breadcrumb, always "Admin / Networks / FidoNet / ...", and a row of tabs for the network's sections, with a link on the right to the same section across all networks. Several pages had a breadcrumb of a different shape and five had none; the network page linked to neither its nodelist nor its combined views. The network page now opens with a summary of what the network has imported (echolists, filebones, nodelist) and links to each view of it. The nodelist pages link to the flag definitions and back.
- New: orphaned areas across networks. An area mapped to a network but in none of its echolists (or, for file areas, filebones) is counted on the networks list, on All Echolists and All Filebones, and on the network page, and the count links to the network's combined view with the Orphaned tab open. Networks with no echolist or filebone imported say so rather than counting every area they carry as orphaned.
- Fixed: on screens narrower than about 1900 pixels, the admin menu bar covered the top of every admin page, and below about 1550 it pushed Mail, notifications and your account menu (with Logout) off the edge of the screen. The bar now wraps its links onto a second row, and pages start below it however tall it is; links to a section of a page (such as the Test email box) land below it too. The website's menu bar, which carries whatever links you add, behaves the same way.
- New: System Info. A page that shows what the board runs on and checks each piece against what Canectic needs: Node and npm, the operating system, memory and disk, PostgreSQL and
pg_dump, Redis, pm2 and each of the four processes (running or not, uptime, restarts, the Node it runs on), the worker's heartbeat, the boot unit, the installed libraries againstpackage-lock.json, and the external programs with what stops working without each. Problems and warnings say what to do. It catches two things nothing else did: a process still running an old Node after Node was changed, and apg_dumpolder than the PostgreSQL server, which would make an update's database backup fail. Copy report and Download give a plain-text report with no passwords in it, for pasting into an issue or a message. Open it from the dashboard (System Info, first on the System tab);node src/tools/sysinfo.jsprints the same report in a terminal, including when the website is not answering, and exits 1 when anything is a problem. Seedocs/sysop/SYSTEM_INFO.md.
Terminal
- Fixed: the terminal mail reader showed only the newest 50 messages in each folder. Anything older in the Inbox, Sent or Trash could not be reached from the terminal. Moving up past the oldest message shown now brings in the next older ones, Home goes to the oldest of all, the Inbox still opens on the first unread message however far back it is, and the position reads against the whole folder ("31 of 80"). Deleting or restoring a message now keeps your place in the list instead of jumping back to the newest messages.
Web
- Fixed: the mail page showed only the newest 50 messages in each folder. Older messages in the Inbox, Sent or Trash could not be reached on the website. Each folder now says how many it holds ("Showing 50 of 230"), and Load older messages under the list brings in the next 50, which the search box and the column sorting then include. Deleting, restoring or permanently deleting a message now keeps your place: the page used to reload, back to the newest messages of the Inbox. Opening an unread message now lowers the unread count beside Inbox and on the Mail link at once, rather than at the next page load.
- Fixed: a user going through the web call flow was invisible to the rest of the board. They were missing from Who's Online and Node Spy, the time they spent there was not counted towards the Most Time Online lists, a sysop could not start a chat with them, and if the board was closing for maintenance they were not warned until the next page. The call flow page now keeps the same live connection as every other page, so all four work there too.
- Fixed: the Test email box was hard to find. It sat at the bottom of the Email section of the Configuration page, below every setting. It is now at the top of that section, and the admin dashboard has a Test Email card (System tab) that opens the section with the box in view.
Infrastructure
- Redis is now spoken to in RESP3, its current protocol. Both Redis libraries Canectic uses default to RESP3, and Canectic had held them on the older RESP2 so a library upgrade could not change behaviour underneath it. Nothing callers or sysops see changes. On a Redis older than 6.0, which has no RESP3, Canectic uses RESP2 and logs a line in
system.logsaying so. Canectic requires Redis 6.2 or later; System Info and the Update page show the Redis version and warn below it. See "Redis" indocs/sysop/INSTALL.md. - Fixed: the log viewer (
src/tools/logviewer.js) stopped withERR unknown command 'HELLO'on a Redis older than 6.0. It now connects the same way the rest of Canectic does.
Upgrade Notes
- The desktop monitor needs a token, and a WebSocket through your proxy. Nothing changes until you use it. To use Canectic Monitor, make an API token with the Sysop Monitor scope on a sysop account's profile page. Behind a reverse proxy, pass
/api/v1/monitorthrough as a WebSocket, as the board's/ws/paths already are; the nginx block is under "Sysop monitor" indocs/sysop/API.md. - Redis 6.2 or later is now required (it was Redis 6). An older server keeps working, and System Info and the Update page warn about it. To check before updating,
redis-cli INFO servershows it asredis_version.
v0.9.16b (2026-10-06)
Security
- Fixed: two libraries Canectic uses had security advisories. proxy-addr, which works out a caller's address behind a reverse proxy, could be fooled by a trusted subnet written as an IPv4-mapped IPv6 address; this matters only if you set
trust_proxyto a subnet. source-map-js, used by the HTML cleaner, could be made to spin on crafted input. Both are updated to the fixed versions.
Terminal
- Fixed: on a new board, every terminal caller was disconnected straight after logging in. A fresh install had no menus, so the board said "Already at top menu", then "Goodbye", and hung up, the sysop included. A board with no menus now gets three starter menus when it starts: Main (Messages, Files, Doors, Private Mail, Top 10 Lists, User Settings, Verify Address, Node Spy for sysops, Goodbye), Messages (change group or area, read, post, new scan) and Files (change group or area, list, download tagged files, upload). Edit them like any other menu; a board with menus of its own is never touched. A board whose menus exist but include no Main now tells the caller so, and where the sysop makes one, instead of hanging up without a word. See "The Starter Menus" in
docs/sysop/MENUS.md. - New: User Settings on the terminal. A new menu command for callers' own account: username (once per
username_change_cooldowndays), location and gender, by the same rules as the website's profile page, with Change Password and Verify Address run from it. The real name is shown, not editable.
- New: test the outbound email settings. Under the email settings on the Configuration page, a Test email box checks the connection to your relay (connect, TLS and login, sending nothing) or sends one test message to an address you give, and shows the answer at once instead of leaving you to wait for a user's verification email to go missing. When something is wrong it says what to change: a host that does not resolve, a blocked port, a port and TLS setting that disagree, a refused login, a sender address the relay will not send as, and so on. See "Test the settings" in
docs/sysop/MAIL.md.
Upgrade Notes
- Your menus are not changed. The starter menus are created only on a board that has no menus at all. User Settings is a new menu command: to offer it, add it to a menu under Menus > Menu Editor.
v0.9.15a (2026-10-04)
Security
- New: scanners on the terminal port are turned away before they get a node. Every new telnet or TCP connection is screened first. A real terminal passes at once by answering the board's cursor query, so callers see nothing different; a client that cannot answer is asked to press ESC twice. A connection whose first words are a router login or command (
root,admin,enable,/bin/busybox, your honeypot names) is closed at once and written tosecurity.logas a honeypot attempt, which the shipped fail2ban honeypot filter already bans; anything else is closed afterterminal_screen_timeoutseconds (15). Screened-out connections no longer fill the caller log, Who's Online or the system log, and Node Spy shows how many were let in and turned away. Switch it off withterminal_screen_enabled. Seedocs/sysop/SECURITY.md, which also covers setting up fail2ban, including banning the port your board really listens on when the router forwards 23 to another.
Web
- New: edit the web call flow. The guided tour of what is new that website users get after logging in is now a flow on the Call Flows page (Flows in the admin menu), marked Web, next to your terminal flows. Drag stages to reorder them, remove the ones you do not want, and add them back from the list. Last Callers, Who's Online and The Wall have settings behind a gear: show them always, or only when someone else has called, is online or has posted since the user's last visit, and how many callers or wall messages to list. Four stages can be added that the web call flow never had: Bulletins (each opens in place, and one new or changed since the user's last visit is marked and opened), System Info, New Mail (their unread private mail, with a button that finishes the tour and opens it) and Verify Address (for users who have not verified, with the days their account has left). Changes take effect at once. It starts out exactly as the call flow was, so nothing changes until you edit it;
callflow_enabledstill turns it off. Seedocs/sysop/CALLFLOWS.md. - New: edit the website's navigation bar. The links across the top of the website (Messages, Files, Doors, Chat, Wall) are now yours to change under Menus > Web Navigation: rename, reorder, hide or remove them, add links to other pages on the board (Bulletins, System Info, QWK and more) or to other sites, and group links under a menu. Each link can be shown to everyone, only to visitors who are not logged in, or only to logged-in users, with the same security and flag rules as terminal menu entries. The right side of the bar (Dashboard, Mail, notifications, your account, Login) stays as it is, so there is always a way into the admin. Hiding a link does not switch the page off. See
docs/sysop/WEBSITE.md. - New: choose what the home page shows, separately for visitors and for logged-in users. Under Content > Home Page, build each from the same blocks as the web call flow (announcements, last callers, who's online, the wall, bulletins, system info, your own Web Pages and more), plus a Log In form for visitors, shown as tabs or one above the other. It starts out as the home page always was, so nothing changes until you edit it. Announcements has a new setting for this, every current announcement or only new ones. See
docs/sysop/WEBSITE.md. - New: write your own pages for the website. Under Content > Web Pages, write pages such as how to connect or house rules in the same editor as bulletins, each at its own address (
/page/how-to-connect), for everyone, visitors only or logged-in users, with security and flags as well. Link one from the navigation bar, or show it in the web call flow with the new Page stage, which can be added once for each page. Links and stages follow a page if its address changes, and are shown only to people who can read it. Seedocs/sysop/WEBSITE.md. - New: time online on the website, and active or idle. Each user's time online is now kept on their account as terminal time and website time. Website time counts only while a page is in front of them and they have used it in the last
web_active_minutes(5): a tab left open in the background counts nothing, and two tabs count once. Who is idle and who is active is kept for each website user online, ready for Who's Online and Node Spy to show. A website visit's entry in the caller log now has a length too, its active minutes. Terminal time starts from your caller log; website time starts at this update. - Fixed: a website visit could be logged as two calls. When a user came back after being idle and their browser made two requests at once, Last Callers showed them twice at the same moment and the caller log counted both. Each visit is now logged once. The Top 10 lists leave out the duplicates already in your caller log. The user's own call count was rarely affected, and is now also counted in a way two logins at once cannot upset.
- Fixed: one failing stage skipped the whole web call flow. If working out any one stage failed (a database error while counting new messages, say), the web call flow quietly marked itself done and sent the user straight on, so they saw no announcements, callers or wall either. Now only that stage is left out, and the failure is written to the system log.
Top 10 Lists
- New: Top 10 lists. Sixteen rankings, each over the last 7 days, the last 30 days and all time: top posters, callers, uploaders and downloaders (by files and by size), time online (terminal, website, and both together), wall posters and door players, and your busiest message areas, hottest threads, most downloaded files, most played doors and doors by play time. They are built once a day. On the website, pick a list from the home page's or the web call flow's Add menu (Top 10: Top Callers) and set its period and how many with the gear. On the terminal, the new Top 10 Lists menu command gives callers a Top 10 menu to choose from, or goes straight to one list you pick, and a terminal call flow can show one list per stage, picked from the Add menu. Content > Top 10 Lists shows every list and, for each list of users, who appears on it: by default everyone except accounts with the new Unranked or the Banned restriction flag, so give your own account Unranked to stay off them all. Message and file areas, their groups, and doors have a new Leave out of Top 10 lists setting for anything you would rather not advertise; what users do there still counts towards their totals. Downloads, door sessions (30 seconds or more is a play) and wall posts are now recorded for the lists, so door and download lists fill up from this update on. Callers from other boards through your door server appear on Top Door Players with their board. No list is shown until you add one. See
docs/sysop/TOPTEN.md.
Doors
- Fixed: BBSLink's Darkness 2.0 was listed as "dkns". The BBSLink door list shipped with 0.9.14a had no name for its newest door, so it showed by its code. A provider that checks its list every day picks up the name by itself; otherwise press Check for changes on the provider's page and apply the list.
Administration
- New: try a call flow without logging out. The new Login Flow menu command runs the stages a caller sees between logging in and the menu, then brings them back to the menu they were on. Its Flow setting is the caller's own flow or any terminal flow that is switched on, so you can try a flow before giving it to a pool. On the website, the navigation bar's new Login Flow link replays the web call flow as the user last saw it, and finishing the replay does not count as finishing the tour. A replay also shows each stage the user would skip, saying why, and so does a terminal Top 10 stage whose list is empty. See
docs/sysop/CALLFLOWS.md. - New: the web block editors say which blocks have nothing to show right now. A stage with nothing to show is skipped, so one just added could look as though it went nowhere, an empty Top 10 list most of all. The web call flow and the home page editors now mark each such block and say why underneath.
- Changes to terminal call flows take effect at once. They used to wait for the next gateway restart. A caller already connected keeps the flow they started with; call pool changes still need a restart.
- Fixed: the Call Flows page threw away a change it refused without saying so. Whatever the page could not save (a pool with no flow chosen, for example) redirected back as though it had worked. It now says what was refused and why. It also now refuses a stage the terminal side cannot run, such as a mistyped name in a new flow's stages. Before, that was saved and every caller on the flow was disconnected on reaching it.
- Fixed: adding a Display Screen stage to a terminal flow never asked which screen. The box for it never appeared, so the stage was added with no screen and had to be set afterwards with the pencil on its badge.
- Fixed: a menu entry the page could not save looked as if it had saved. The Menus page went back to the menu without a word when it refused an entry. It now says why, for example a Top 10 Lists entry naming a list that does not exist.
Upgrade Notes
- Terminal connections are screened. Callers whose terminal program cannot answer a cursor position query are asked to press ESC twice before the board appears.
- The Call Flows page lists a new flow, Web. It is the web call flow, set up as it has always been. It cannot be given to a call pool and has no Active switch;
callflow_enabledturns it off. If your board already had a flow of its own called Web, the new one is called WebLogin. - Saving a terminal call flow changes it straight away. It no longer waits for a gateway restart, so finish an edit before you save it.
- Hand edits to
src/views/includes/user-nav.pugare replaced. The website's navigation bar is now drawn from Menus > Web Navigation, set up as the shipped bar was, so changes made to that file by hand are gone after the update: make them on the new page instead. - Top 10 lists start empty. Nothing shows until you add a Top 10 block or menu entry, and the lists are first built at 5:20 the morning after the update (or straight away with Regenerate Now on Content > Top 10 Lists). Door and download lists fill up from the update on, since those were not recorded before. Every list of users leaves out accounts with the new Unranked restriction flag; nobody has it until you give it, so your own account is listed unless you do.
v0.9.14a (2026-09-29)
Doors
- Fix: Play on the web Doors page ended the call at once for every door not offered to the door server, in 0.9.13a. Your own logged-in users playing in the browser were treated as callers from another BBS, so they could only reach doors offered to the door server, which by default is none. Terminal callers were not affected.
- The doors your door server offers are now chosen by collection. A door collection's new Shown to setting is This board, Door server or Both, and on a collection shown to the door server each door has an Offered switch; doors you add to such a collection are offered at once. A remote BBS's caller who does not name a door now gets a menu of your offered collections and then the doors in the one they pick, through
doorclient.js, RLogin and another Canectic board alike. A collection shown to the door server only is hidden from your own callers. Changing a collection to include the door server first shows you its doors to choose from. The Available to Door Server setting on each door is gone; the doors you had marked are still offered (see Upgrade Notes). A door running on another system can never be offered. - New: doors on other systems as doors of your own. Under Door Server, Doors on other systems now takes door providers of three kinds: another Canectic board, an RLogin door server such as DoorParty or a Synchronet game server (through an SSH tunnel where needed, with the server's key recorded on first connect and a changed key refused), and BBSLink. Buttons fill in DoorParty's, BBSLink's and Synchronet's settings. On a provider's page you add its doors, from its door list (below) or by code, and each becomes a door entry with the new Remote launch type that you put in a collection or a menu like any other door; web play works the same way. Only logged-in callers at the provider's minimum security (the verified level unless you set one) can use them, and a door on another system is never offered through your own door server. RLogin callers each get a password made once and kept, which DoorParty ties their account to; a caller who changes their username starts again there unless you tick Keep sending the name a caller first used. Output is converted for UTF-8 terminals.
- New: door lists for door providers. A provider's page can keep the list of doors it offers, so you tick doors instead of typing codes. The list comes from one shipped with Canectic (DoorParty's 170 doors and BBSLink's 36 are included), from a file in a file area (so a list can arrive by file echo), or, for another Canectic board, from the board itself. A Load button on the provider's page fills in the list that fits it. Tick doors to add them, remove them from the same list, and switch each door off or on. Check for changes shows what a newer list would change before you apply it, and Check every day keeps it current with a notification of each change. A door that leaves the list is marked no longer offered and its door entry made inactive, never deleted; the daily check refuses a list that would remove more than half the doors and tells you instead. Canectic never reads anyone's web pages for a list.
- Fix: Delete on the Door Games page did nothing for a door in any collection, and said nothing either. The door is now removed from its collections and deleted, and a failure is shown.
- Fix: unticking Active on a door collection did not switch it off. The collection stayed active whatever the box said.
Administration
- Every admin page now has a card on the admin dashboard. Door Server (under Doors), Call Flows and Timed Events (under System) and Template Events (under Content) were reachable only from the admin navbar.
Upgrade Notes
- Door server doors are now offered by collection. Each door you had marked Available to Door Server is still offered: it is ticked in every collection it belongs to, a collection holding one now shows to Both, and one that was in no collection is in a new Door Server collection shown to the door server only. Nothing else in those collections is offered. Check Admin > Doors > Door Collections.
v0.9.13a (2026-09-29)
Messages
- Fix: in the terminal message reader, D (delete) and M (move) did nothing for a moderator reading a message full screen, although the bottom bar offered them. They now work there. In the thread list M is Mark Read, as before.
Networking
- Fix: on a board running as a single process (
PROCESS_MODE=embedded), outbound FTN mail waited for the next scheduled scan, and connecting to a system from the nodelist did nothing. The mail process ignored both requests. It now handles them as the standalone mailer does. - Fix: netmail your board relayed carried a Via line with no version, just "Canectic", and the system statistics page and the terminal's system information showed the version as blank. Both now show the version the board is running.
- Fix: mail written in UTF-8 by software that does not say so arrived garbled, with
’stored asΓÇÖ,JörgasJ├╢rg, and whole Cyrillic or Hebrew passages unreadable. A message without a character set declaration was always read as CP437. It is now read as UTF-8 when that is what it contains, which genuine CP437 text, box drawing and ANSI art included, never is. Message bases imported from JAM files are read the same way. Messages already stored garbled can be repaired, see below. - Fix: names and subjects with accented or other non-English letters were mangled on arrival. The To, From and Subject lines of FTN mail lost the top bit of every such character. They are now read in the same character set as the message.
- Fix: netmail to RaidFix was accepted and then silently dropped, with no reply. It is now answered by FileFix, since RAID's RaidFix is its file echo manager.
- Removed: the
tic_auto_forwardsetting, which did nothing (received TIC files are always passed on to the systems subscribed to that area), and the unusednetmail_inandnetmail_outtables from before netmail moved into private mail. The tables are only dropped when empty; if yours hold anything, the update keeps them and says so. - Fix: a file attached to FTN netmail was never sent. The netmail went out marked as carrying a file, but the file itself was not queued for the mailer, so the recipient got a message referring to an attachment that never arrived. The attachment now goes with the next mail session.
- Fix: a peer using EMSI (Wazoo) was sent the same mail on every session. After a Wazoo session sent a file, nothing marked it as sent: packets were not deleted and flow file lines were not marked done, so the next session offered them again. EMSI sessions now clean up exactly as binkp ones do. Mail received over EMSI is also tossed as soon as the session ends, rather than waiting for the next scheduled toss.
- Fix: a file request (FREQ) to a system outside zone 1 was never sent, and the Request Files form on the FREQ page did nothing at all for a host or hub address (node 0, such as 2:203/0). The request was written to the zone 1 outbound whatever the address, where no session for that system looks, and it ignored
MAILER_OUTBOUND_DIR. Requests now go to the outbound the mailer reads for that address. A second request to the same system before it has been sent is now added to the first rather than replacing it. - The FREQ filebox now has its own directory,
ftn/freq, on new installs. It used to default toftn/filebox, which also holds the per-peer outbound fileboxes, so any file left at the top offtn/fileboxcould be requested by any system you link with. Existing boards keep their setting; see the upgrade note to move yours.
Maintenance
- Fix: running
node src/tools/maintenance.js --regen-htmlagain while one was still going ignored the second request, including its--area,--widthand--batch, although the tool said those areas had been queued. A request with different options now queues behind the first; an identical one joins it, and the tool now saysalready runningrather than claiming to have queued it. The same applied to regenerating private mail or file descriptions, purging PM trash or tossed packets with a different--days, and a--dry-runof rethreading or MSGID repair queued alongside a real one. - Fix:
--regen-html --dry-runwas not a dry run. The option never reached the worker, so it regenerated every message for real, and it counted every message as updated either way. A dry run now writes nothing and reports, area by area, how many messages would change, with a few message ids to look at first. A real run now only rewrites messages whose display actually changes. The same goes for--regen-pm-html. - Fix: a maintenance or toss job that ran past its time limit was reported failed but carried on running, and the next job started alongside it. Such a job is now asked to stop: regenerating message HTML, MSGID repair and rethreading stop at the next safe point, and the job is reported as stopped. A job that cannot stop is logged as still running, and the worker health check keeps showing it until it ends. The health check also measured every job against the toss limit (30 minutes), so a maintenance job that was healthy but ran longer than that could get the worker restarted under it; each job is now measured against its own queue's limit.
- New: repair messages stored garbled by the character set problem above.
node src/tools/maintenance.js --fix-charset --dry-runreports, area by area, how many messages are affected and a few message ids to look at; without--dry-runit repairs them, including names and subjects, and redraws them on the web. ANSI art and other genuine CP437 text is left alone. The old text is kept, so--revert-charsetputs everything back. Add--area SHORTCODEto do one area at a time. - Regenerating message HTML now runs one area at a time, from the command line and from the admin Events page. A large board could not finish in a single job before
maintenance_job_timeout(60 minutes) stopped it, so it had to be done by hand in--areabatches. The command waits until every area is done, however long that takes, and reports each one as it finishes.
Administration
- New: a working door server. Other BBSes can now send their callers to the doors you choose. Mark a door "Available to Door Server" in its settings, switch on
doorserver_enabled, and add each BBS under Doors > Door Server; it gets a System ID and a secret, shown once. Any verified user can also register their BBS from the Doors page (doorserver_register_security, which by default is the verified level). Each registration waits for your approval and they are notified when you decide, except for users whose security reachesdoorserver_auto_approve_securityor who hold a flag listed indoorserver_auto_approve_flags, whose registrations are approved at once. You can suspend a BBS, give it a new secret, or limit how many of its callers can be on at once (doorserver_max_concurrent). Remote callers appear in your caller log with their BBS's name. Sysops of other BBS software connect withdoorclient.js, one file needing only Node 22 or later, which they download from the Door Server page and run as a door; it reads the caller's name from DOOR32.SYS or DOOR.SYS and lets them pick a door or go straight to one. Your own callers can visit another Canectic board's doors too: register your board there, add it under Door Server > Remote door servers (the Test button checks the connection and lists its doors), and add a menu entry with the command Remote Door Server. For BBS software that can open RLogin connections but cannot run the client, switch ondoorserver_rlogin_enabled(off by default; portdoorserver_rlogin_port, 5513) and restart: the RLogin user isSYSTEMID:username, the password field is the secret, and the terminal type names the door (oransifor the list). - Fix: an API token issued from Admin > API Tokens was never shown, so it could not be used; tokens are stored only as a hash, so it could not be recovered either. The token is now shown once when it is issued. Messages after saving a file group (such as a file move starting) were not shown either and now are.
- Fix: the First Message Post, Nth Message Post, First File Upload and Nth File Upload template events never fired, so a template attached to one was never sent. They now fire when a user posts a message (on the web, in the terminal or in a QWK reply packet) or uploads a file. Each user's count starts from what your board already held, so nobody with a posting history is greeted as a first-time poster; messages that arrived from a network are never counted.
- Fire Template Event is now in the menu editor's Command list. A menu entry could already fire a message template event, but the editor did not offer the command, so it had to be added by importing menu JSON. Its argument box suggests the event codes your board has.
Files
- Changing where a file area's files are kept now moves them. Changing a file group's base directory, a file area's directory, or the file base directory on the Configuration page used to leave every file behind at the old path, so the areas looked empty. Saving such a change now shows which areas are affected, where their files are and where they would go, and asks whether to move them. They are moved in the background, one at a time, and you get a notification when it is done. A file whose name is already at the new location is left where it is and listed, never overwritten.
- Fix: a file area directory set to a full path (starting with
/or~/) was put inside the group's directory anyway. It is now used as given, like a group's base directory, and moving the group leaves it alone. - Fix: file requests (FREQ) never found files held in file areas. Only the FREQ filebox and magic names worked. Files in the areas a linked system is subscribed to can now be requested.
Security
- Updated morgan, multer and nodemailer for three published advisories: forged lines in the web access log, disk space left behind by uploads that were abandoned partway, and, on a board that sends email through more than one mail server, TLS connections to one server named as another.
- Fix: anyone who could reach your board's web address could start any of your door games. The door server connection (
/ws/doorserver), meant for other BBSes to send their callers to your doors, accepted a built-in test login that was published with the source code, and then ran whatever door it was asked for. It now refuses every connection unless you switch the door server on (doorserver_enabled, off by default), no remote system can log in until you register it, and a remote caller can only run a door you have marked as available to the door server. If you want to close it on a board you have not updated yet, deactivate the Doorserver pool under Admin > Flows and restart the gateway. - Fix: failed logins from mail clients and rejected API tokens could not be banned by fail2ban. A wrong IMAP or SMTP password was logged without the caller's address, and an API request with a bad token was not logged at all. Both are now written to
security.logwith the address, in the form the shipped filter inexamples/fail2ban/already matches, and each line now says which service it came from (via IMAP,via SMTP,via API). A wrong current password when changing your password, on the web or in the terminal, is logged the same way. No change to the filter is needed.
Web
- Dates on the user side of the web read the same way everywhere, in your board's time. Profile, notifications, last callers, announcements, the call flow, the maintenance page and the maintenance banner used to show each viewer's own 12-hour date format, differently on each page. They now use the board's clock and one style: a bare time today (
17:30),Yesterday 17:30,3 days agothrough the week, thenSep 25, andDec 25, 2025once it is more than six months back. Hover over one for the exact time. Tables you can sort by date keep the full date and time. - Opening the notification bell now marks your notifications read. The badge clears as soon as you look, rather than staying lit until you clicked each one or pressed Mark All Read, which is gone from the dropdown (the full Notifications page still has it). New ones stay highlighted until you close the dropdown.
- New private mail no longer shows in the notification bell on the web, since the mail icon already counts it. Callers still see new mail in the terminal's notification list, and reading or deleting the message, on the web, in the terminal, in a mail client or in a QWK packet, now clears it there.
- Fix: finishing the web call flow, or posting to the wall from it, returned an error when the request carried no form data. Browsers always send some, so this was only seen from scripts; the call flow now finishes normally either way.
Upgrade Notes
- The door server is off until you turn on
doorserver_enabled. A Doorserver pool you deactivated to close the hole on 0.9.12a can stay that way; make it active again before switching the door server on. - Repair garbled messages already stored with
node src/tools/maintenance.js --fix-charset --dry-run, then without--dry-run, with the worker running.--revert-charsetputs everything back. - If
freq_filebox_diris stillftn/filebox, move the files you mean to offer intoftn/freqand point the setting there, so the per-peer fileboxes cannot be requested. - The
netmail_inandnetmail_outtables are removed only if empty; the update says so if it keeps yours.
v0.9.12a (2026-09-27)
Updates
- New: switch between stable and development from the Update page. The Channel box at the top of the page now switches, and checks the new channel straight away. Switching to development asks you to confirm, after a warning that development builds get less testing and that there is no going back to the stable you run now. Before, the channel was only on the Configuration page, and a board on development saw stable releases on the Update page with no way to take them. Moving from a development build to stable does not change anything at once, since a build is newer than the stable it was made from; the page now says the board moves onto stable with the next stable release.
- Fix: an update started from 0.9.10a or earlier showed no upgrade notes, before or after. The Update page you update from belongs to the release you are running, and one from before upgrade notes existed passed none on, so the notes never reached the progress page or the Update page afterwards; they were only in the release notes. The new release's updater now fetches and verifies them itself when the page you updated from did not, so they appear on the progress page and under the last update whichever release you came from. If they cannot be verified, the update's warnings say to read the release notes.
- Fix: every update kept its backup for ever, database dump included. Each one is about the size of your database, so a board that had updated a handful of times could be holding gigabytes in
.canectic-update/backup/, plus a staged copy of every release it had updated to. When an update completes it now deletes older backups, keeping the lastupdate_backup_keep(1 by default, the update just finished; 0 keeps them all), and any staged releases left from earlier updates. It never deletes its own backup, nothing is deleted after a failed update, and a failed update's backup is kept until a later update completes.node src/tools/update.js --prunereclaims the space now; add--dry-runto see what it would remove.
Administration
- Fix: changes made on the admin pages recorded nobody as having made them. Creating or editing a message or file group, area, network or peer left "created by" and "updated by" empty; an update started from the Update page recorded no one; and closing or reopening the board for maintenance logged the sysop as
undefined. Each now records the sysop's username. Records written before this release stay empty. A message posted from the admin message pages with its From field left blank is now posted under your username rather than no name at all.
Networking
- Fix: asking for the same job a second time often did nothing. Running
node src/tools/maintenance.js --regen-htmlagain, pressing a connection's pull or push button again, and a second message ID repair, file area scan or user maintenance run were silently ignored for as long as the first run was still listed as finished, which could be days. They now run; a second request is still refused while the first is waiting or running. - You are now told when a connection's polling schedule is removed. When the BBS starts it removes any poll, QWK or NNTP schedule that nothing in the database asks for any more, such as one for a deleted connection or one whose polling is turned off. That was only a line in the log. Connections created before the Polling setting existed have it off, but kept polling until 0.9.6a, when those schedules were removed on the first restart and the connection stopped collecting mail without a word. You now get a notification naming each connection, its network and host, and why its schedule went. If you meant it, nothing to do; if not, turn polling back on for the connection under Peers. Switching polling off yourself does not notify.
- New: a history of every QWK and NNTP transfer, under Networking > QWK and NNTP Transfers, and on each QWK or NNTP connection of a peer's page. Every pull and push is recorded, including ones that never reached the server, with what came of it (mail, nothing waiting, timeout, refused connection, login failure and so on), how far it got, how long it took, how many messages moved and the error. The connection list shows when each one last worked and, separately, when it last actually brought in mail, since a link can pull without error for weeks and bring nothing. History is kept for
network_transfer_retention_days(30 by default). - New: Canectic can be a QWK hub over the internet, and a QWK node of another Canectic board. Add a Hub (Inbound) connection for the node with a node password, and the node connects to your board's web address with its QWK ID and that password, using the new HTTP transport. It sends its replies and collects its packet in one call. A packet is only counted as delivered once the node confirms it has it; if a download fails, the next call gets the same packet, so nothing is lost. A wrong password is logged to
security.logfor fail2ban. Before this a Canectic hub could only swap packets through a shared directory, and never built a packet for its nodes at all. Seedocs/sysop/QWK_NETWORKING.md. - Fix: saving a peer connection's settings made its next NNTP pull start every group again from the first article. The save threw away what the connection had recorded for itself: how far each NNTP newsgroup and QWK conference had been read, and for an FTN link the record of failed connects that holds off repeated calls to a system that is down. They are now kept. The failed-connect record is still cleared when you change the address the link calls (host, port, FTN address, transport or protocol), since those failures were against the old one; the Reset button on the peer page still clears it at any time.
- QWK IDs now follow Synchronet's rules, for each connection's Remote QWK ID and for your own ID on a QWK network: up to 8 characters, none of
space \ / | < > * ? + [ ] : = " ; ,, stored in upper case. Each ID can be used once as your hub and once as your node; before, nothing stopped two connections sharing an ID, and two connections to the same hub fight over the one packet it builds. IDs that start with a digit or contain.or@are accepted with a warning, since netmail cannot be addressed to them. IDs you already have keep working: if any break the rules you get a notification listing them when the board starts. See "QWK IDs" indocs/sysop/QWK_NETWORKING.md. - Fix: a QWK packet download that failed lost its messages. Downloading a personal QWK packet (web, terminal or the API) marked its messages read and its private mail read as soon as the packet was made, so if the download failed or was abandoned, those messages were missing from the next packet. They are now marked only once the download has completed, as Synchronet does; the same goes for files from the download queue.
- Fix: messages in a REP that had to be retried reached the hub twice. When sending a REP to a QWK hub failed and a later push delivered it, the board did not record those messages as sent, so the next push sent them again, private mail included. They are now recorded as sent whichever push delivers them.
- New: personal QWK packets carry files, as Synchronet's do. A user's packet now includes a list of new files (
NEWFILES.DAT) in the areas they scan for new files, the files in their download queue, and attachments to their private mail, each of which their QWK reader can show or save. From the reader, a message toSBBSwith the subjectFILES ON,FILES OFF,ATTACH ON,ATTACH OFForFREQ <file name>switches the list or attachments, or asks for a file, which then comes in the next packet; a reply says what was done. The same two switches are on the QWK page. Both are on by default. The Download button on the QWK page no longer greys out when there are no new messages, since a packet can carry files alone. - New: Canectic can be the FTP hub a Synchronet QWK node calls. Turn on
qwk_ftp_enabled(it listens on port 8021 unless you setqwk_ftp_port) and restart, and a node using Synchronet'sqnet-ftp.js(or other QWK software that calls a hub over FTP) logs in with its QWK ID and the node password on its Hub (Inbound) connection, sends its REP and collects its packet exactly as from a Synchronet hub. A packet counts as delivered only once its download completes. FTP does not encrypt the password, so for nodes on the internet the HTTP hub behind HTTPS remains the safer choice. See "Nodes that call over FTP" indocs/sysop/QWK_NETWORKING.md. - New: QWK netmail reaches systems you do not link with directly, and your board passes it on for others, as a Synchronet QWK hub does. The board learns how to reach every system from the mail that arrives (kept for 90 days), sends by a direct link, a spelled-out path (
HUBID/THEIRID) or a learned route, and passes on netmail other systems send through it, within the network it arrived on. What cannot be passed on goes back to its sender asBADADDR:, and every decision is in the Netmail Relay Log. Each QWK network has a Relay QWK netmail setting, on by default. The send form asks which network to use when a system is known on more than one. Replies to mail that came through other systems go back the way it came. - New: QWK packets carry files, in both directions, as Synchronet has always allowed. Files you put in a link's
out/files/go with its next packet; files that arrive are kept (through your inbound wildcard rules, or in the link'sin/files/) with a notification for each. Before, a file another system sent in a QWK packet was deleted without a word. A REP that failed to upload is now retried at the next push even when there are no new messages; it used to wait for one. - Fix: QWK messages sent from Canectic could arrive at a Synchronet board as one long line. Most messages were marked as UTF-8 in the packet while their text was CP437, and Synchronet, reading them as UTF-8, found no line breaks. The mark now always matches the text. UTF-8 messages from Synchronet boards were also read wrongly, as garbled text on one line, and are now read correctly.
- New: QWK messages that do not fit CP437 go as UTF-8 instead of losing characters, on connections with Send UTF-8 when a message needs it ticked (on by default). Messages that fit CP437 still go as CP437, so nothing changes for them.
- Fix: a Canectic QWK hub did not pass messages between its nodes. Anything that arrived from one system on a QWK network was never sent to any other system on it, so nodes of a Canectic hub only ever saw the hub's own users' messages, and a board that is a node of one system and a hub for others never passed its nodes' messages up. Messages now go to every system on the network except the ones they have already been through, and carry an
@VIAline saying where they came from, as Synchronet does, so netmail replies can find their way back. - Fix: QWK netmail for your own users went undeliverable when it arrived in the forms Synchronet uses (
user@YOURID, orNETMAILwith the address on the first line). It is now delivered. Undeliverable QWK private mail also records its reason, which it did not before. - Each QWK link now has its own packet directory:
qnet/<ID>.hub/for a system that is your hub,qnet/<ID>.node/for one that is your node. Before, both usedqnet/<ID>/, so a system that was both your hub and your node shared one directory, and a half-imported packet from one could be imported as the other's mail. Updating moves what is already there, packets waiting to be sent or imported included, and leavesqnet/<ID>as a link so a node usingqnet-directory.jskeeps working. The peer page shows each connection's packet directory. - New: a QWK network can allow IDs longer than 8 characters (up to 32), for networks where every system accepts them. Off by default, since Synchronet stops at 8.
- Fix: a refused save on a network or peer page looked like it had worked. Adding or editing a connection or an address, or editing a network, returned to the page with no message when the save was refused; the reason was only in the log. It is now shown at the top of the page, with any warnings.
- New: a Canectic QWK hub serves nodes that use a shared directory. Tick Node collects from a shared directory on a node's Hub (Inbound) connection, and every few minutes (
qwk_hub_sweep_minutes, 5 by default) the board imports the REPs the node leaves inqnet/<NODE_ID>.node/in/and builds its packet intoqnet/<NODE_ID>.node/out/for it to collect withqnet-directory.js. The messages in a packet count as delivered once the node has taken it; until then the same packet waits. Before this the board built nothing for such a node, and its REPs sat inin/until something else happened to import them. Seedocs/sysop/QWK_NETWORKING.md. - Fix: QWK private mail went out over whichever QWK connection sent first, so with more than one QWK connection most of it reached the wrong system. Each message now goes only to the system in its address, over the connection whose Remote QWK ID matches, and a reply goes back over the connection the original came in on. A message addressed to a system you have no QWK connection to is returned to its sender as undeliverable instead of being sent somewhere it cannot be delivered.
- Fix: a QWK node with no conferences subscribed never received private mail from a Canectic hub. Its packet is now built for private mail alone.
- Fix: the Force plaintext password box on an FTN connection could not be unticked; saving with it unticked kept it on. It now turns off.
- Fix: saving a QWK or NNTP connection reset how often it polled to 2 hours (QWK) or 1 hour (NNTP), whatever its interval was set to, until the BBS was next restarted. It now keeps the interval you set.
node src/tools/schedules.jsalso compared each QWK and NNTP schedule against those defaults instead of your setting, so it could report a mismatch where there was none and miss a real one. - Fix: a QWK or NNTP pull that failed was logged as though it had worked. The system log said the job finished and that messages were imported whatever happened, and the reason was only in
tosser.log. A failed pull or push is now logged as a failure with its reason, and messages are only reported as imported when some were.
Messages
Message group retention has its own Retention tab, as message areas and file groups do. The Max Age and Max Messages limits for a group were on the Main tab among the other settings, where they were easy to miss when creating or editing a group.
Fix: messages looked double spaced in the web reader, worst at 80 columns on a phone. A message body was laid out as though it were a line of text inside the page, so its lines were spaced for the page's normal text size rather than its own: roughly two and a half lines' worth each at 80 columns on a phone, and about twice what it should be everywhere else. Lines now sit at the message's own spacing, on phones and on desktop, so messages are noticeably more compact than before. Art that is too wide for the screen now scrolls sideways as intended rather than spilling out.
Fix: the message area and file area lists were far wider than a phone screen, so the page scrolled sideways. On a phone they now use the full width, the header's buttons and search wrap onto their own line, and each area's description sits under its name instead of in a column of its own; the message and file counts and the last poster's name are left out on the narrowest screens. Desktop is unchanged.
Fix: the file and group pages were cramped or too wide on a phone: a file area, a single file, new files, file search, upload, and the message and file group pages, plus the front page. They kept a wide margin down each side, and their tables ran past the screen. On a phone they now use the full width, and each file or area shows as one entry with its details underneath and its counts on the right. Tablets and desktops keep the tables.
Fix: the list of threads in an area was unreadable on a phone: five narrow columns, so column headings broke a letter at a time and a subject showed as a few characters. On a phone each thread is now a single entry: the subject in full with who started it and the last reply underneath, and the reply and unread counts in a column of their own on the right. The buttons wrap below the area name. Tablets and desktops keep the table.
Fix: on a phone the last message in an area was often partly off the bottom of the screen, with no way to scroll to it. The message list was sized for a desktop's header and ignored the phone's address bar; it now fits the space actually left on screen, and adjusts when the address bar appears or the phone is turned.
Fix: message headers ran off the side of a phone screen, so the whole page scrolled sideways. A long subject, or a long name next to the Reply and moderation buttons, pushed the header wider than the screen. On a phone the subject now has a line of its own under the header and wraps in full, and a name that does not fit is shortened with an ellipsis. Desktop is unchanged, except that a subject too long for the line is now cut short instead of overflowing.
Fix: a word at the end of a line dropped onto a line of its own at 80 columns. Many systems pad each line with spaces out to the full width, and those spaces were kept, so a line that fitted came out a character or two too wide and its last word wrapped. Trailing spaces are now dropped; nothing else about how a message looks changes. Messages received before updating keep the padding until regenerated; see the upgrade note.
Upgrade Notes
- Regenerate stored messages again, even if you did so for 0.9.11a:
node src/tools/maintenance.js --regen-htmlwith the worker running (--area SHORTCODEfor one area). Messages received before this release keep the trailing padding that wraps their last word at 80 columns until then. - QWK packet directories have moved to
qnet/<ID>.hub/andqnet/<ID>.node/. The update moves what is there and leavesqnet/<ID>as a link to the node directory, so a node collecting withqnet-directory.jskeeps working; point its call-out at the.nodepath shown on its connection when you next change it.
Private Mail
- Fix: on a phone, opening a private message pushed Reply and Delete off the screen. A long subject could not shrink, so the reader's header ran past the edge. On a phone the header now runs in rows (date and buttons, then the subject in full, then From and To); on a desktop a long subject is cut short to fit.
- Dates in the private mail lists are short on a phone, "Sep 25" (or "Sep 2025" once older than six months), so the sender and subject have room; the full time is in the tooltip. Desktop still shows the full date and time.
- Fix: the inbox, sent and trash lists were sorted by the attachment column instead of by date. They looked newest first only because messages arrived in that order; sorting now starts on the date, newest first, and the attachment column can no longer be sorted on.
- The standalone Sent and Trash pages are gone:
/pm/sentand/pm/trashnow open those tabs of the inbox, so a bookmark still works. They were unusable on a phone, and the inbox's own tabs had replaced them. - Fix: the bottom of the inbox was cut off on a phone, for the same reason as the message list: it was sized for a desktop's navigation bar and ignored the phone's address bar. It now fits the space left on screen.
Files
- "New" files now means new since your last visit, everywhere. A file is marked new if it arrived after your previous visit began, on the web and in the terminal alike, and it stays marked for the whole visit; looking at an area or a file no longer clears it. On the web a new visit begins when you log in, or when you come back after
web_idle_thresholdminutes without activity (120 by default). The file lists, the New Files page, the call flow's new-files summary and the terminal's file newscan and area list all agree. Before, each area kept its own marker that viewing moved along, so the counts in one place could disagree with the list in another. - The terminal file browser's M (Mark Read) key is gone: with new meaning since your last visit, there is nothing to mark.
v0.9.11a (2026-09-25)
File Echoes
- Fix: Files sent by an uplink never reached their file areas, and nothing reported an error. Since 0.9.7a, a file whose TIC listed this system under
Seenbywas refused as a loop. But every uplink lists a file's recipients there before sending it, so nearly every inbound file was refused, logged at a low level and counted as a duplicate. OnlyPath, which records the systems a file has actually passed through, is now checked. Files refused in the last week or so can be recovered; see the upgrade note. - Fix: A file that replaced one of the same name was saved as
NAME_1.EXT. Infopacks re-released under their own name, and nodelists a year later, landed beside the file they replaced under a new name, and the next release could not find it to replace, so an area slowly filled with_1,_2,_3copies. A replacement now takes the old file's place, and the old one is deleted once the new one is recorded. Copies already renamed this way are left as they are. - Fix: a file could be sent back to a system that already had it, when that system's address was written with a network domain (
@fidonet) or a.0in the TIC. The same fix means a TIC from such an address is recognised as coming from its peer, so its password is checked. Replaceswith wildcards now works. A TIC sayingReplaces NODELIST.*replaced nothing, since only an exact filename matched, so every version stayed.*and?are now honoured. A wildcard must name at least three characters of the filename before its extension:Replaces *FOO.ZIPworks, but*.*or*.ZIPwould empty the area, so such a line is ignored, logged, and the file imported without replacing anything.- A file whose record could not be saved now goes to
ftn/bad/with its TIC, rather than staying in the area directory unrecorded while the TIC failed again on every toss.
Messages
- Fix: ASCII art and column-aligned posts had their rows run together. Lines were joined into paragraphs whenever they were long enough, which is right for prose and wrong for a drawing made of
| / \ -characters, a MealMaster recipe's section dividers, or a table of figures: two rows became one long line that then wrapped, so the picture broke and extra lines appeared. Lines laid out with columns or wide gaps, or drawn mostly from punctuation, are now left as written. Messages received before updating keep their old layout until regenerated; see the upgrade note. - New: read at 40 or 80 columns on a phone. On a phone the web reader sizes text to 40 columns so it is readable, which breaks anything laid out for 80, ANSI art above all. The reader now has a 40 | 80 switch, remembered per area, and a message recognised as art is shown at 80 on its own, whatever the switch says. Desktop is unchanged at 80.
- Message areas and groups have a "Mostly ANSI art" setting, so an area full of art opens at 80 columns on phones. Set it on the area, or on a group for every area created in it, with Push Group Defaults for areas that already exist. The setting existed in the database without any way to set it.
- Fix: Adding a message area with a
-in its short code did nothing. The form closed and no area appeared, with no reason given. A short code names the area's database tables, so it may only use letters, digits and underscores, up to 52 characters; the form now says so before submitting, and a refused area comes back to the group page with the reason. The same fix means a message group that could not be created now says why, and notices on the group page (such as Push Group Defaults results) appear there instead of on a later page.
Upgrade Notes
- Repair the layout of messages already received with
node src/tools/maintenance.js --regen-html(worker running;--area SHORTCODEto try one area first). ASCII art and column-aligned posts that arrived before this release still show rows run together until then. - Check how your BBS starts at boot, once, after updating. Run
node src/tools/bootUnit.jsas the user the BBS runs as, andnode src/tools/bootUnit.js --installif it reports anything. A board set up withpm2 startupstops starting at boot once the Node version it names is removed, which under nvm happens on an ordinary Node upgrade.--installasks for sudo and restarts nothing; the restart it prints restarts the BBS, so close the board first withnode src/tools/maintenanceMode.js --onif callers are online. - Your existing logs are archived on the first night and deleted 30 days later. The first rotation puts everything your logs hold so far into one archive dated that day, and like any archive it is removed once it is older than
log_retention_days. To keep the history longer, copylogs/production/somewhere else before then, or raiselog_retention_days. - Recover file echo files refused since 0.9.7a. Those received within
ftn_keep_tossed_days(7 by default, after whichftn/tossed/is purged) are still there, each TIC next to its file, and the system log names each one withalready listsfollowed by your address. To import one, move the.ticand its file back intoftn/in/; the next toss imports it. Nodelists and other files that replace their predecessor only need the newest one moving.
Installation
- New:
node src/tools/bootUnit.js --installsets the BBS to start at boot in a way that survives Node upgrades. The unitpm2 startupwrites names one Node version, so under nvm the BBS stops booting once that version is uninstalled. The fix used to be a hand edit described in the install guide; this does it: it installs thecanectic-pm2wrapper, writes the unit that uses it (with sudo, keeping your existingPATHminus its Node version and saving the old unit beside it), and enables it, restarting nothing. Run it once on any board that was set up withpm2 startup.node src/tools/bootUnit.json its own checks a board and says what is wrong; the Update page shows the same warnings. See "Surviving Node Upgrades" indocs/sysop/INSTALL.md. - After a Node upgrade the BBS no longer needs a
pm2 saveto move to the new Node. At boot, the wrapper notices processes that were saved on a Node that has since been replaced, restarts them on the current one and saves. Before, they came back on the old Node, and did not come back at all once it was removed.
Updates
- The Update page now shows a release's upgrade notes before you update. Releases have declared them since 0.9.9a, but only the release notes carried them, so a sysop who updated from the Update page never saw "clear your gateway log" or "retention will be enforced on the first run". A Before you update panel now lists the notes that concern your board, from every release you would skip, most serious first, and the Update button waits for you to tick that you have read them when any is a warning. They are verified against the release's signed checksum list before being shown; when that is not possible you are pointed at the release notes instead. The notes stay on the Update page and the progress page after the update, since several are instructions for afterwards.
node src/tools/update.jsprints them too, and needs--acknowledge-noticesto apply when one is a warning. See "Upgrade notes" indocs/sysop/UPDATES.md. - The "is available" notification now says how many upgrade notes apply to your board.
- Upgrading into this release is the first to keep its notes for afterwards; the panel before the button appears from the update after it, because the page you update from belongs to the release you are running.
Administration
- Fix: turning a log purge off did not stop it. Setting a retention to keep for ever (0, or 999 for tossed packets) left the nightly purge running with the number of days it had before, so it went on deleting what you had asked to keep: tossed packets, notifications, bad echomail, the inbound and outbound file logs, the relay log, mailer session history and update check history. A purge that is switched off is now removed at the next restart, and every purge reads its setting when it runs, so a change of days applies that night rather than after a restart.
- New: log files are rotated daily. Every log in
logs/production/used to grow for ever. Each night at 23:59 each one is compressed intologs/production/archive/YYYY-MM-DD/and started afresh, and archives older thanlog_retention_days(30 by default, 0 keeps them all) are deleted. Settings are under Logging on the configuration page.node src/tools/rotateLogs.js --dry-runshows what a rotation would do, and without--dry-runrotates now. - New: file echo processing has its own log,
tic.log. Every file received or sent by file echo is logged there, with the reason for any it refused. Failures, refusals and each run's totals still appear insystem.logtoo. - Fix: date columns in the admin logs and other tables did not sort by date. Dates showed as
9/22/2026, 5:57:03 PM, or asSeptember 22, 2026andSep 22on some pages, and sorting a column put10/1before9/30and April before March. Every table now shows dates as2026-09-22 17:57:03(or just the date where that is all there is) and sorts on the actual time. Affected: the inbound and outbound file logs, the netmail relay log, API tokens, echolists, filebones, nodelists, message and file areas, users, and on the user side the private mail inbox and file lists. - Times in tables show in your own timezone. They follow your browser's clock, which for a sysop is almost always the board's own; the zone is named when you hover over a time.
v0.9.10a (2026-09-22)
Upgrade Notes
- From 0.9.9a there is nothing to do. A board updating from an earlier release skips 0.9.9a, so the upgrade notes it carried for boards coming from 0.9.8a, 0.9.7a and 0.9.6a or earlier are repeated in this release's notes and shown by the updater.
Networking
- Fix: echomail forwarded to your downlinks lost most of its SEEN-BY. A message usually carries several SEEN-BY lines, and only the last one was kept, so downlinks were not told about most of the systems that already had the message. That is what SEEN-BY is for, and without it they can pass a message to a system that already has it. Every line is now kept, this system is added, and the list goes out as several short lines, as other software writes it. Messages already received keep what was stored; everything tossed from now on is complete.
- Fix: forwarded echomail told downlinks it had come only from you. Its PATH was replaced with your address instead of having it added, and the original PATH was sent as well, at the top of the message where software keeping the last PATH line ignored it.
- Fix: every message you forwarded carried your timezone instead of its author's, and posts made on the board carried a fixed
-0600whatever your timezone and whatever the time of year. The author's timezone now passes through untouched, and posts made here carry this system's real offset for the date. - Fix: forwarded echomail gained a second tear line after its Origin line, which showed as a stray
---at the end of the message on every system downstream. - Your tosser now names itself with its real version in the
TIDline, where it saidCanectic Tosser v0.1.
Updates
- The Update page now says when the release you are running has been withdrawn, with the reason and, once there is one, the release that replaces it. The update check has known since withdrawals began and nothing showed it, so a board on a pulled release looked exactly like one that was up to date.
- The update check now notifies you when the release you are running has been withdrawn. Before, only the Update page showed it, so a board whose sysop never opened that page was never told. The notification carries the reason and what to move to, and is sent once per release rather than every morning.
- Fix: a withdrawn release could name a replacement that had been withdrawn too. 0.9.7a names 0.9.8a, which was withdrawn two days later, so a board on 0.9.7a was told to move to a release it would be refused. The Update page, the notification, the update tool and the bootstrap now follow the chain of replacements and name the newest release still listed, 0.9.9a in that case.
- An update now tells you about a one-off repair it left for you. A few database repairs are too slow to run without asking, so the update reports them and leaves them to you, and until now that report was only in the update's log: an update could say COMPLETE with one still waiting. It is now a warning on the run, shown on the Update page and by
node src/tools/update.js --status, with the commands to run. The one boards set up before this spring are most likely to see is the MSGID split:node src/tools/maintenance.js --fix-msgids --dry-runto see what it would change, then without--dry-runto do it. - Fix: an update on a board installed before releases carried a MANIFEST warned that "every changed data file was kept as yours" when none had been. It now says so only when a data file really was kept, with how many; each one is also named, as before.
v0.9.9a (2026-09-22)
Upgrade Notes
- From 0.9.8a, this is the first update that gives notice, so it takes longer. Callers and logged-in web users are warned and given
maintenance_grace_seconds(5 minutes unless you have changed it) before the BBS stops, and the update then waits for binkp sessions and jobs in progress to finish. Set a shorter grace on the Config page first if you would rather. - If an update to 0.9.8a stopped partway, finish it before updating again. The BBS is stopped, the board is closed for maintenance, and the update's log ends in
Unhandled 'error' eventandECONNRESET, usually during the dependencies step. From the install directory, run the steps it had not reached, in order:npm install --omit=dev,node src/tools/dbmigrations.js,pm2 start ecosystem.config.cjs, and once the processes are up,node src/tools/maintenanceMode.js --off. Then update to this release as normal.
Updates
- Fix: an update could stop partway through if anyone had the board open in a web browser. While the BBS is stopped, the update answers on the board's web port itself, and a browser tab trying to reconnect to the board could make the update itself stop, in the middle, leaving the BBS stopped and the Update page still saying an update was running. Updates into 0.9.8a were affected. An update always runs the updater of the release it installs, so the update into this release is not.
- An update that fails unexpectedly now says so. A visitor's browser dropping a connection is ignored and the update carries on; any other unexpected error is recorded against the step the update was on, with whether the BBS was left stopped, before the updater exits. The Update page and
node src/tools/update.js --statusthen show it failed, where before a run that had died went on showing as running. Two quieter ways for an update to vanish without a record, both in how it starts itself, are closed as well. - Fix: the Update page offered the version you had just updated to. It showed the result of the last check, which ran before the update, until you pressed Check now. It now compares what that check found with what is running, and says when the check is older than the board. The update itself also refuses a release that is not newer than the one running, so a page left open cannot reinstall it, and
node src/tools/update.json a board that is up to date says so instead of reporting that it cannot move yet.
Installation
- The bootstrap will not install a withdrawn release. Asked for one by name, it stops and says why the release was withdrawn and what replaces it. Left to choose, it takes the newest release that has not been withdrawn.
Release Engineering
- A release can be withdrawn the moment a fault is found, before its fix exists.
node scripts/release.mjs --withdraw <version> --reason "..."takes it out of the channel index's release list into a separate withdrawn list, so no board and no bootstrap is offered it again, including ones written before withdrawal existed: those take the first release listed, and marking an entry in place (the format used once, for 0.9.7a) did nothing for them.--supersede <version> --by <replacement>names the fix once it is cut. An index still in the old format is converted the next time it is written. See "Withdrawing a release" indocs/reference/RELEASE_PROCESS.md. - Fix:
--supersede --signsigned nothing and said it had. It returned before the step that chooses the signing key, so it rewroteSHA256SUMS, removed the old signature and printed "rewritten and signed". A withdrawal now proves it can sign before it changes anything.
v0.9.8a (2026-09-22)
Upgrade Notes
- Clear your gateway's log after updating. Until this release, every login on the web wrote the user's name and password, in plain text, to the gateway's pm2 output log (
logs/production/gateway-out.logon a standard install). The logging is gone, but the log still holds whatever was written before. Empty it or delete it once you have updated, and treat any copies or backups of it the same way. - This update cannot give your callers notice; later ones will. Updates now close the board for maintenance, warn the people on it and wait for the network work in progress to finish before stopping the BBS. The processes this update replaces predate that, so they cannot be asked: after about twenty seconds the update carries on and stops the BBS as updates always have. Update at a quiet hour. Every update after this one drains properly. For the same reason, if you start this update from the web, the progress page cannot recognise you while the BBS is stopped and you will see the maintenance page like everybody else; follow it from the Update page once the BBS is back, or with
node src/tools/update.js --status. - If an update to 0.9.7a failed at its schema step, finish it before updating again. That update stopped with
spawnSync is not definedafter it had replaced the files and installed the dependencies, then started the BBS again on 0.9.7a's code over a database that had not been brought up to date. Runnode src/tools/dbmigrations.jsin the install directory and restart the BBS, which completes what the update was doing, then update to this release as normal.
Maintenance Mode
- New: close the board for maintenance, and know when it is safe to start. Admin > Maintenance, or
node src/tools/maintenanceMode.js, closes the board to everyone but you for an import, work on areas, database work by hand, or anything else that needs it to itself. Callers and logged-in web users get a countdown and are then disconnected or shown a maintenance page; visitors see your reason and when you expect to be back; mail clients and binkp links are told to try later, so no mail is lost; the worker stops taking new jobs, which queue up for when you reopen. A drain table shows what each part of the board still has in flight, and says so when nothing is.--waiton the tool holds until then, for scripts. Seedocs/sysop/MAINTENANCE.md. - Follow an update from the web while the BBS is stopped. The Update button now takes you to a progress page that follows the update from start to finish, including the stretch where the BBS itself is stopped: the update answers on the board's web port in its place, showing the progress to the sysops who were logged in when it started and the maintenance page to everybody else. It uses the same port as the BBS, so it works behind a reverse proxy unchanged. See "Updating from the web" in
docs/sysop/UPDATING.md. - Updates close the board themselves. An update warns callers and web users when it starts, waits until the board has drained before stopping it, and reopens it only once the update has checked every file it wrote. An update that fails after stopping the board leaves it closed, so nobody calls in while you find out what happened; reopen it from the Maintenance page when you are satisfied.
Updates
- Fix: an update into 0.9.7a stopped at its schema step with
spawnSync is not defined. By then the files had been replaced and the dependencies installed, which is past the point where an update puts the old files back, so the board was left on the new code over a database that had not been brought up to date. An update always runs the updater of the release it is installing, so the update into this release is not affected. See the upgrade note if your board is in that state. - A release can now be withdrawn after it is published, and the update check says so. A withdrawn release is no longer offered, and a board running one is told, with the reason, even before a replacement exists. This works from this release onwards: a board on 0.9.7a checks with 0.9.7a's code, which offers this release as an ordinary update without mentioning the withdrawal.
- An installed board now keeps the signature of the release it runs (
MANIFEST.asc) beside itsMANIFEST. Installs never received it, so a board kept whatever it was first installed with, or nothing. Updating to an unsigned development build removes the old one rather than leaving a signature over the previous release in place. - Fix: every install and update reported failure at its very last step, after everything had worked. The last step checks that every file landed as released, and
npm install, which runs just before it, rewritespackage-lock.jsonas a matter of course. A fresh install was told it had not finished and that removing it and starting again was safe; an update was left closed for maintenance. The check now compares the lockfile by what it actually pins, which npm does not change, and names any file that genuinely did not land. If you installed 0.9.6a or 0.9.7a and saw "the install did not finish" at the verify step, your board was installed correctly. update_require_signaturenow does what it says. It was read by nothing. An update now checks the release's signature against the release key your board already has before applying anything: a signature that fails is always refused, and with the setting on, so is an unsigned release (which includes every development build). See "Signatures" indocs/sysop/UPDATES.md.- Fix: an update started from the Admin UI stopped partway, leaving half the BBS stopped. The gateway and mail processes went down, the mailer and worker kept running, and the update itself disappeared at its "stop" step with nothing on the Update page to say why. The update was being stopped along with the BBS it was updating. It now takes itself out of the way first. Updates started from a shell were not affected. This fix travels with the new release's updater, so the update INTO this release already benefits from it, including from the Admin UI.
Security
- The bootstrap now checks release signatures against the Canectic release key itself. It accepted a signature made by any key in your keyring. It now fetches the release key from the release directory, requires the signature to have been made by the key whose fingerprint is built into it and printed in the install guide, and stops before unpacking anything when a signature fails or comes from another key. Importing the key yourself is no longer needed.
- Fix: web logins wrote passwords to the log. See the upgrade note above.
Networking
- Fix: a binkp link connecting over WebSocket while your mailer was at its session limit received an empty busy message. The reason text was never copied into the frame. It now reads "System busy, try later", as it always did over TCP.
- Fix: binkp sessions failed straight after logging in when a link presented more than one address and you had a connection for another of them with the same session password. Grouping those connections into one session, which is what the matching password is for, threw an error instead.
- Fix: automatic echolist and filebone scanning failed for a network whose area held no archive to unpack. It logged that it was still looking for the lists, then threw instead of reporting what it had found.
Doors
- Fix: a door using the
canectic.sysdrop file never started. Writing the drop file failed on the board name, so the door was never launched.
Administration
Fix: with no mail domain configured, the mail process restarted every few seconds until pm2 gave up on it, which read as a fault on every new board. It now waits quietly, and saving a mail domain in /admin/config starts the mail servers without a restart.
The Update page now counts terminal callers. It counted people on the web only, so a board with three callers in a door and nobody on the web told you nobody was online.
Errors in the web interface are now written to the system log. Before, a page that failed showed you an error page and recorded nothing, so if a save on an admin page did not take, the logs could not tell you whether it had failed or had never been sent. Each server error is now logged with the page it happened on and where in the code it was raised. Pages that simply do not exist (the steady stream of 404s from scanners probing for WordPress and the like) are deliberately not logged, so they cannot bury the errors that matter.
Release Archives
- A release's
FILE_ID.DIZnow describes the software and nothing else. It carried the first sentence of each upgrade warning, which is not what a BBS listing is for and not where a sysop reads them. Upgrade notes are in the release notes and the release's metadata.
v0.9.7a (2026-09-20)
Upgrade Notes
Everything in this block is something that happens to your board on the first run after updating. Read it before updating a live system.
- Embedded mode only (
PROCESS_MODE=embedded): maintenance that has never run starts running. The embedded worker recognised 7 of 33 maintenance jobs and silently completed the rest without doing them, so log purges, message and file retention, nodelist and echolist scanning, timed events and the update check have all been idle. They begin on the first daily run after you update, and clear the whole backlog at once. The part to check before updating is retention: if you ever setmax_age_daysormax_messageson a message or file area, it has been ignored until now and will be enforced on that first run. The Retention Preview on an area's admin page reports what a limit would remove without removing anything. Boards on the normal four-process setup are unaffected — their worker has always run these. - This update rewrites stored message bodies. Three parts of the system had been storing text three different ways in the same database columns, which is what allowed relayed messages to be corrupted. The update normalises them, automatically, asking nothing of you. On the development board that was 4,707 messages across 9 areas, all received over QWK; your figures will differ. The previous copy of every rewritten message is retained in the old column, unread, so the change is reversible. Messages already displaying incorrectly before this update are not repaired by it.
- If you carry file echoes, expect a burst of outbound files on the first toss after updating. A file arriving from an uplink was correctly queued for your downlinks and then never sent, because nothing ran the outbound pass for file echoes. That queue has been growing for as long as you have been receiving files, so a downlink that has had nothing from you for months may receive a great deal at once. Nothing was lost and nothing needs doing; it is the volume that may surprise you and your links. To meter it, update at a quiet hour and watch the first pass on the peer page, which now shows pending, sent and failed counts per connection.
Messages
Message bodies are now stored one way regardless of where they came from. Three importers had been storing text three different ways in the same two database columns, which is what allowed messages relayed between networks to be corrupted. An automatic migration rewrites the affected rows when you update: on the development board it touched 4,707 messages across 9 areas, all of them QWK-sourced. Nothing is asked of you and nothing needs checking afterwards. The older column is deliberately left in place for this release, unread, so the rewrite can be reversed if it ever needs to be.
Fix: messages received over NNTP displayed as mojibake on a UTF-8 terminal. The reader converted them a second time, from a character set they were never in.
Fix: messages relayed between networks were corrupted on the way out. The QWK packer decided how to read a stored message body from which database column was populated, and that test is only correct for messages that arrived over QWK. A message that came in over FTN and went out in a QWK packet had every extended character replaced with rubbish. NNTP had the same fault in a different form: outgoing articles declared no character set at all and were written in a way that discarded anything outside plain Latin-1, and incoming UTF-8 articles were stored mangled because they were never properly decoded. Outgoing NNTP articles now declare UTF-8, encode their headers correctly, and carry their text intact.
Fix: every message this system sent carried the wrong character encoding. FTN assumes CP437 unless a message says otherwise, and Canectic was writing UTF-8 bytes while saying nothing. Receivers followed the convention, decoded UTF-8 through the CP437 table, and got mojibake: a block of ANSI art becomes a wall of
Γûê, an accented name comes out wrong, a subject line looks corrupted. It affected names, subjects and message bodies alike; art simply makes it obvious, being entirely extended characters. Outbound mail is now encoded properly and declares what it carries with a CHRS kludge.Message areas have an Outbound Character Set setting, and message groups carry a default for it. Automatic (the default) sends CP437 whenever the message fits it, which is nearly all traffic, and switches to UTF-8 only for content CP437 cannot represent — where the alternative was replacing those characters with
?. An area can also be pinned to Always CP437 for an echo whose members run software that ignores charset declarations, or Always UTF-8 for a modern network that has agreed on it. Set it once on a group and every area created in it inherits it, with Push Group Defaults available for areas that already exist (unticked by default, since it replaces a per-area choice); an explicit choice on an area always wins. The setting is per area rather than per link because the encoding of an echo is something everyone carrying it agrees on, not a property of any one link.Inbound messages are now decoded by what they declare rather than always as CP437, so mail from a system that correctly announces UTF-8 no longer arrives mangled. An undeclared message is still treated as CP437, which is the network default.
Fix: kludge lines were destroyed on every message this system relayed. Stored control lines were being read back in the wrong form, so each one was written out as
undefined:with its name and contents gone. A single captured packet carried 772 of them across 157 messages. The visible effects were charset declarations vanishing (which is what made relayed art unreadable) and reply linkage being lost, so downstream systems could not thread the messages. Already-damaged lines are now discarded rather than passed on.Fix: ANSI art drawn with block characters was flattened into a single line. Paragraph reflow joins consecutive lines that are close to the full width, which is how text from a wider editor gets reassembled — but every row of 78-to-80 column art looks exactly like that. Two faults combined: art detection only recognised art that moves the cursor, so art drawn as colour plus block characters read as ordinary text, and on the web the reflow ran before anything asked whether the body was art at all. Three rows of art became one line, and since the web copy of a message is rendered once when it arrives, nothing later could put the breaks back. Art detection now also recognises block-drawn art, and the web asks before it reflows rather than after, which is what the terminal already did.
Fixing this does not repair messages already stored. Run
node src/tools/maintenance.js --regen-html --area <SHORTCODE>on an affected area, and note that it can only recover art whose plain-text copy survived import intact.One deliberate trade: a message that is ordinary prose but carries two or more full-width rules now reads as art and will not be reflowed. Nothing about its shape distinguishes it from a drawing, and reflow was previously merging such a rule onto the end of the sentence above it, so leaving it alone is not obviously the worse outcome.
Thread mode can now be set on a message group. It was per area only, so a sysop wanting a whole network to read flat had to set every area by hand and set each new one again afterwards. The group carries a default that new areas inherit, an explicit choice on the area form still wins, and Push Group Defaults gained a Thread Mode tickbox for applying it to areas that already exist. It is unticked by default, like Retention, because pushing it replaces whatever each area was set to. Existing groups default to Standard, so nothing changes until you set one.
File Echoes
Fix: files received from an uplink were never passed on to your downlinks. A TIC file arriving from a peer was filed into the local area correctly and its outbound rows were written correctly, and then nothing ever staged them. The periodic outbound sweep, which is the safety net that catches whatever was not sent on demand, covered echomail, QWK, NNTP and email but had never included file echoes; and the inbound toss did not signal an outbound one the way the design had always described. The only things that actually staged a file echo were a local upload, the admin Trigger Event button and a manual toss, so a downlink received files only when somebody on your board happened to upload one. Nothing failed and nothing was logged. Inbound tossing now signals outbound immediately, and file echoes are part of the outbound sweep, so a board with no local uploads at all still feeds its downlinks.
Fix: a downlink's
%RESCANfor a file area replied cheerfully and sent nothing. A file rescan writes pending rows rather than rewinding a pointer, so unlike a message rescan there is nothing that picks them up on the next pass. The admin Rescan button carried the same promise. Both now signal an outbound toss once the rows are written.Fix: file echoes did not work in either direction in embedded mode (
PROCESS_MODE=embedded). The embedded worker had no case for the TIC jobs, so BullMQ handed each one over, the switch fell through to its default, and the job was marked complete without the work being done. Outbound email and the outbound sweep were missing for the same reason. A test now holds the two workers to the same set of job names, since a job nothing handles is indistinguishable from one that succeeded.Fix: nothing stopped a file echo circulating. A file that reads as new is imported and then hatched onward to every subscribed downlink, so in a mesh a file that is not recognised does not merely get stored twice, it goes round again and each system that fails to recognise it passes it on. Two brakes now exist where there was effectively one.
The standard FTN guard was missing entirely: a TIC carries Seenby and Path, this system stamps itself into both on the way out, and nothing checked for its own address on the way back in. That check is now the first thing done, before the file is imported or forwarded, and it matches on the bare address so a chain that writes the address with a 5D domain on one hop and without it on the next is still recognised.
The dupe check, which was the only brake, had a hole that defeated it exactly when it mattered. It compared filenames case-sensitively and required an exact match even when the checksum matched, so a file returning from a chain that had uppercased or renamed it read as a completely new file with an identical CRC. Filenames are now compared case-insensitively, and a file whose checksum and size both match one already held is recognised whatever it has been renamed to. Size is part of that comparison deliberately: a CRC-32 on its own is not enough to reject a file over, since across a large area the odds of some pair of unrelated files colliding are around one in a hundred.
The two guards answer different questions and both are needed. The dupe check asks whether this area holds the file now, which retention can make false; the Seenby stamp asks whether the file has been through this system before, and travels with the file rather than living in a table that gets pruned.
Fix: a peer subscribed to a file area that is not carried on its network collected a dead delivery row for every file, for ever. The outbound pass needs an areatag to build a TIC descriptor and looks one up by the peer's network; with no mapping it marked the transfer failed, and a failed transfer was never retried by anything. Nothing filtered the subscriber list against that, so the rows piled up invisibly and one downlink simply never received anything from that area. Such a subscriber is now left out at the point the rows are created, on all three paths that create them (inbound tossing, web upload and terminal upload), and the mismatch is logged naming the peer, the area and the network so it can be fixed rather than silently skipped.
A peer's file echo delivery is now visible, and failures can be retried. The peer page shows pending, sent, failed and skipped counts per connection, lists the most recent failures with the file, area, reason and attempt count, and offers a Retry Failed button that re-queues them and runs an outbound toss. Previously a failed transfer was terminal and invisible: there was no page that showed one and nothing in the system that would try it again.
Fix: the command line tosser never moved file echoes.
node src/tools/tosser.js --ftnenqueued the echomail and netmail jobs only, though the documentation listed it as one of the triggers for TIC processing. A sysop tossing by hand moved messages, moved no files, and was told nothing had gone wrong.--ftnnow covers file echoes in both directions, as it always claimed to.
Security
- adm-zip updated to 0.6.1, clearing two advisories against 0.6.0: extraction followed symlinks in the destination, allowing a crafted archive to overwrite files outside the target directory, and a declared uncompressed size could be used to exhaust memory. This matters here because both archive paths read files that arrive from outside the board: FILE_ID.DIZ extraction from uploaded and tossed files, and infopack scanning for echolists and filebones. The fix is a patch release inside the range already declared, so
npm installpicks it up with no configuration change.
Embedded Mode
- Fix: the clean shutdown message never printed, and shutdown ended on an unhandled error instead. Both the gateway and the door server tested a bare
logtoconsolewhen the last server stopped, but that name is a private constant inside the logger and was never in scope, so the check threwReferenceErrorfrom inside a timer callback. The logger now exposes the setting properly. - Fix: an embedded board ran almost no maintenance at all.
PROCESS_MODE=embeddedstarts the worker inside the gateway, and that in-process worker recognised 7 of the 33 maintenance jobs. Schedules are registered in both modes, so the other 26 fired on time, were handed over, matched nothing, and were marked complete. Nothing was logged at a normal log level and the queue diagnostics reported every scheduler healthy. In practice that meant no log or retention purges of any kind (bad echomail, inbound and outbound file logs, mailer sessions, notifications, relay log, tossed packets, update checks), no message or file retention limits, no periodic inbound scan, no timed events, no update check, no nodelist or echolist scanning, and admin buttons such as Rescan Area, Rethread and Batch Subscribe reporting success while doing nothing. Boards running the default four-process setup were unaffected. The maintenance jobs now live in one module both workers call, so the two cannot drift apart again, and a test holds them to the list of jobs anything can actually enqueue.
Files on the Web
- Fix: viewing a file area did not clear its new-file badge. The area page advances your scan pointer to the newest file it listed, but it picked that file by comparing ids straight out of the database, and PostgreSQL returns them as text. After the first comparison the rest were being sorted as words rather than numbers, where "9" comes after "37", so the pointer was set to the lowest-numbered file on the page instead of the highest. Pointers only ever move forward, so the wrong value was discarded and the pointer simply stopped, leaving a count that nothing could clear. The same mistake made the terminal file browser stop marking files as new once you had opened one.
- New files are now visible on the web, not just counted.
/fileshas a New Files button showing the total across every area you are subscribed to, and/files/newlists those areas with a preview of what has arrived in each. Each file area has an All / New toggle, and new files are marked in the listing. The new-only listing shows the oldest unseen file first and walks forward on each visit, so a backlog larger than one page is worked through rather than skipped.
Release Engineering
- A cut refuses when the changelog carries more than one
## Pendingheading. Only the first is read and its body stops at the next##, so entries written under a second one are dropped from the release they were written for without a word. This changelog grew a duplicate from a hand edit and nothing noticed; it was found by reading the file. - Fix: the GitLab release step could create an unsigned tag at the wrong commit, and nothing about the result looked wrong.
glab release createdoes not require the tag to exist: given a name it does not recognise, it creates one, as a lightweight tag at the remote's current head. Run at the end of a cut, before anything has been pushed, that head is the commit before the release commit. So the signed annotated tag stayed on the maintainer's machine, the remote got an unsigned lightweight one pointing somewhere else, and the release page hung off that. The page was there and the assets were right; the signature that is the entire point of signing was not anywhere a reader could check it. The step now readsgit ls-remotefirst and refuses on a tag that is absent or lightweight, naming the command to fix it. - A cut still does not push, deliberately, so the GitLab step normally refuses at the end of one.
--publish <version>does that half on its own once the tag is pushed, and the cut says so rather than leaving it to be remembered. - Publishing a second time over the same release now says what happened: GitLab refuses an asset link whose name is already on the release, so the files are not replaced, though the name and notes are updated. It names the command to remove the release if replacing them is what was meant.
- Releases are published with
node scripts/publish.mjs, which sends the release files, this changelog and each release's notes to canectic.tranqsoft.com, then downloads every file back to confirm it is served byte for byte. The changelog appears at/docs/changelogand each release's notes at/releases/<version>/notes, and the changelog is refreshed between releases as well. Nothing is ever deleted from the release host, because a development build cannot be rebuilt without the stable release it is based on.
v0.9.6a (2026-09-16)
Upgrade Notes
Everything in this block is an action, not a description. Read it before updating a live board.
Node 22 or later is now required, and Node 26 is the tested target.
better-sqlite3moved to v13, the first release built on N-API, so its native binary is stable across Node major versions instead of being tied to each version's V8 internals. Itsenginesrange is what sets the floor. nvm is the recommended way to manage this;INSTALL.mdleads with it and keeps a native Node install documented as the supported alternative.npm 12, which ships with Node 26, will not run dependency install scripts unless the package is allowlisted. A plain
npm installon a new box blocks six packages, three of them the native modules whose install script is the thing that resolves the binary.package.jsonnow approvesbcrypt,node-pty,@serialport/bindings-cppandbetter-sqlite3, pinned to version, and deniesmsgpackr-extract,cpu-featuresandssh2deliberately (all three have working pure-JS paths). Pinning means upgrading one of the four approved packages needs a deliberate re-approval, which is exactly when a native module's install script deserves another look. Without the allowlist, password hashing, door games and DOS doors over socat fail at runtime rather than at install time.allow_netmail_relayis deprecated and ignored. It is replaced bynetmail_relay_mode, takingoff,nodelistoron, defaulting tonodelist. Every upgrade lands onnodelistregardless of what the old boolean said, because it cannot be honoured:configStoreseeds every default into thesettingstable, so a sysop who deliberately set it false is indistinguishable from one who never touched it. Since relay never worked correctly before this release anyway (the message and packet headers were swapped on every transit message) there is no working configuration to preserve. A system that wants full transit relay sets the mode deliberately on/admin/config. Readdocs/sysop/NETMAIL_RELAY.mdbefore choosingon, which is considerably broader than it reads.The one-time MSGID tuple split no longer runs during migrations. Its predicate is
msg_id LIKE '% %', which no index helps, so it full-scanned every per-area message table on every update to establish there was nothing to do. On a hub with hundreds of areas that is minutes per pull.dbmigrations.jsnow checks a marker insettings(a single lookup) and prints instructions instead. If your message base predates themsg_id/msg_orig_addrsplit, run this once with the worker running:node src/tools/maintenance.js --fix-msgids --dry-run (report only) node src/tools/maintenance.js --fix-msgids (repair)A fresh install with no per-area message tables is marked complete immediately and never sees the notice. Only a full, non-dry run sets the marker: an
--arearun repairs what it was pointed at and says nothing about the rest, so it may not silence the notice on a system that is still half repaired. Legacy rows do not match new-format ones for dupe checks or threading, so leaving this undone is not cosmetic.Run
node src/tools/dbmigrations.jsafter updating. New columns this release:importedon every per-area messages table (this is what stops imported history being tossed back out to peers),max_age_daysandmax_messagesonmessage_areasandmessage_groups,destination_addressonechomail_bad, and thesort_order/post_ascolumns that were previously created by runtimeALTER TABLEstatements at gateway startup and are now declared indb/schema. New tables:mailer_sessions,outbound_file_log,netmail_relay_log,update_checks.Changing the pm2 child interpreter needs a full re-register, not a restart. The four apps now pin their interpreter at config-eval time instead of resolving a bare
nodefrom whatever environment the daemon inherited.pm2 resurrectreplays the olddump.pm2and will not pick it up:pm2 delete all && pm2 start ecosystem.config.cjs && pm2 saveIf you boot via systemd, route the unit through the wrapper. The unit
pm2 startupgenerates bakes an absolute Node path into bothPATHand everyExec*line, which pins the service to whichever Node was current the day it was generated. Afternvm alias default 27the BBS still boots on the old version, and once that version is uninstalled it stops booting at all. Install the new wrapper outside the repo, becauseupdate.shtreatssetup/as a code directory and wipes it:install -m 755 setup/canectic-pm2.sh ~/.local/bin/canectic-pm2Then point the unit's
Exec*lines at~/.local/bin/canectic-pm2. It resolves Node at start time from nvm'sdefaultalias when nvm is present and from plain PATH when it is not, so one unit shape serves both install types.A database created before this release may be SQL_ASCII.
create_pg_db_user.shissued a bareCREATE DATABASEwith noENCODING, so it inheritedtemplate1, which on a cluster whose initdb ran under a C or POSIX locale is SQL_ASCII. That encoding performs no validation at all: invalid byte sequences are stored silently and handed back unchanged, so any component writing non-UTF-8 corrupts data without raising an error. Message bodies carry CP437 art and accented names, so this matters in practice. New installs now create withTEMPLATE template0 ENCODING 'UTF8' LC_COLLATE 'C' LC_CTYPE 'C'. Existing installs convert withsetup/convert_db_encoding.sh, which scans every text column for invalid bytes and refuses up front rather than failing halfway through a restore, verifies row counts table by table, and then swaps the database names so.envneeds no edit.--preflight-onlyanswers whether you need it at all.Upgrading the host OS: read
docs/sysop/OS_UPGRADE.mdfirst. An Ubuntu release upgrade installs a new PostgreSQL major but does not migrate the data. The new cluster comes up healthy and empty, nothing errors, and every message base, user record and per-area table stays in the old cluster until someone drops it. The runbook covers thepg_dropcluster/pg_upgradeclustersequence with port continuity and row-count verification, Redis 7 to 8 (RDB is forward-only, so the dataset does not survive a rollback), native addons, third-party PPAs, and rollback. One thing it records that is easy to miss: a release upgrade can silently removebuild-essential, which only surfaces on the next cleannpm install.
Prerequisites
socatwas not named in any install instruction anywhere, despite being what the DosemuPty launch type needs to give a DOS door a serial port to talk to the caller over, and DosemuPty is the mode recommended for interactive DOS doors. A board set up by following the documentation exactly would have had DOS doors fail the first time a caller ran one.dosemu2was documented assudo apt install dosemu2, which fails on a stock Ubuntu because it exists only in a PPA. A command that does not work is worse than no command, because the reader concludes the package is gone rather than that they need the repository.INSTALL.mdand the README now give theadd-apt-repositoryline, and say plainly thatsocatis needed alongside it.postgresql-clientis now named too. A local PostgreSQL brings it along, so this only bites a board whose database is remote or managed, and what it costs is specific:pg_dumpis what an update uses to back the database up immediately before migrating, which is the thing that makes a failed migration recoverable.- The installer now names what is missing. Every external program Canectic uses is declared in one place with what it enables and what is lost without it, and the installer reports the absent ones with the command to install each. None of them block an install, because none of them are needed to run a board: that is precisely why each was previously discovered at the moment a caller first tried the feature it belonged to, where a DOS door that exits instantly looks like a Canectic bug rather than a missing package.
/admin/updatereports them as well, which is not a duplicate of the install check. An OS release upgrade disables third-party repositories, which is wheredosemu2comes from, and has been observed removingbuild-essentialoutright, so a board can arrive at an update having quietly lost a tool it had when it was installed.- The declaration is what the documentation is tested against, so a program added to the list and not written down now fails the suite. Prose drifts from the code that needs it, and that drift is the whole reason
socatwent unmentioned for as long as it did.
Node 26 and Dependency Upgrade
- Every dependency is now at its current release.
npm outdatedis empty andnpm auditreports 0 vulnerabilities, down from 17 advisories at the start of the pass. Majors taken: bullmq 5 to 6, ioredis 5 to 6, redis 5 to 6, nodemailer 8 to 10, uuid 13 to 14, chalk 5 to 6, file-type 21 to 22, basic-ftp 5 to 6, mocha 11 to 12, better-sqlite3 12 to 13, adm-zip 0.5 to 0.6. Removedmongodb, which had zero references, andpath, an inert shim for the built-in module. - The advisories cleared were genuinely reachable in a deployed system, not theoretical:
ws(uninitialised memory disclosure and a fragment denial of service against internet-facing WebSocket servers),multer(denial of service via deeply nested field names on the upload endpoints),qsandbody-parser(every request passes through them via Express), andadm-zipCVE-2026-39244, a crafted archive forcing a 4GB allocation, which is reachable because we extract archives arriving over FTN. The four remaining advisories reach us only through mocha, a dev dependency, or through a nodemailer option we never pass. - Both Redis clients are now explicitly pinned to RESP2 (
RESP: 2insrc/middleware/db/redis.js,protocol: 2insrc/queues/connection.js). node-redis 6 and ioredis 6 both default to RESP3, which changes reply shapes:hGetAllresults,existstruthiness, and everythingsessionstore.jsreads. Pinning means the dependency upgrade could not silently change behaviour. Adopting RESP3 remains available as its own deliberate change. - BullMQ 6 needed no code change. It moved ioredis to an optional peer dependency, and we already depend on ioredis directly and build the connection ourselves.
Configuration
Tier 1 is now the
.envfile read from disk, notprocess.env. Editing.envand restarting had no effect, with no error and no warning. Two behaviours combine:pm2 savesnapshots the environment intodump.pm2per app and replays it into each child on restart or resurrect, anddotenv.config()then declines to correct it because it does not overwrite a key already present inprocess.env. The contract was written about a file and implemented against an ambient environment. Precedence is now: in the file and ambient, the file wins and the disagreement is reported; file only, the file; ambient only, ambient, so one-off command-line overrides still work; neither, the database and then the default. Bootstrap keys keep ambient precedence, since they are needed before the file can be located andPROCESS_MODE=embedded node bin/gateway.jsis supported. Divergence surfaces three ways: a per-key warning at load and reload naming both values and giving the pm2 remedy,appConfig.getEnvDivergence(), and a banner plus badge tooltip on/admin/config. Values for keys matching pass, secret, token or credential are redacted in all three.A
.envedit can now be applied without restarting. A Reload button beside Exit on/admin/config(POST/admin/config/reload) re-reads both the file and the database, then publishesCONFIG_CHANGEDto the other three processes. Keys markedrestart: truestill need a real restart.Settings with a fixed set of legal values now render as a select with a hint for the value in effect, and the save route refuses anything outside the list rather than storing it, naming the key it left alone. Every consumer of a setting like this reads it as text and falls back to something safe on an unrecognised value, so a typo did not fail loudly: relay would quietly resolve to
offwhile the settings page displayed "nodelists" as though it were in effect. Currentlynetmail_relay_modeandzmodem_default_crc.New sysop guide
docs/sysop/NETMAIL_RELAY.md, since the relay modes had no sysop-facing documentation at all. It covers which mode to pick, whyonis broader than it reads given that route resolution walks the nodelist and the coordinator tree, what nodelist mode actually checks, the fact that no nodelist imported for a network means no relay for that network, what it does not protect against, and theallow_netmail_relayupgrade.The settings table now records whether a value was chosen or merely seeded.
configStoreseeds every default intosettingson first run, so a row existed for every key whether or not anyone had touched it, and a deliberate choice was indistinguishable from an untouched key. That is whyallow_netmail_relayhad to be deprecated and ignored rather than migrated: honouring a storedfalsewould have pinned every upgrade to the old behaviour, and honouringtruewould have switched relay on for someone who had turned it off. A newsourcecolumn recordsdefaultorsysop, andconfigStore.wasSetBySysop(key)answers the question. This does not reopen the relay decision, which was settled on separate grounds (relay was broken before, so there was no working configuration to preserve); it is there for the next deprecation.Fix: the source badge on
/admin/configclaimed the database for settings nobody had ever touched. Any row present insettingswas reported as coming from the database, and since every default is seeded there on first run, that was very nearly every key: on a real install, 122 of 158 keys were sitting at exactly what the release ships while the page said otherwise. A stored value now reads as coming from the database when a sysop set it, or when what is stored differs from what this release ships. The second half matters because a seeded value goes stale when a default changes between releases, and that stored value is still the one taking effect, so reporting it as the default would be wrong in the other direction.Existing installs are backfilled by migration
0004-settings-provenanceusing two signals that are exact rather than guessed: a row written after it was inserted (updated_at > created_at, which the seed cannot produce), and a key that is not inconfigDefaultsat all and therefore was never seeded. On the development box that identified 36 settings as chosen and 122 as shipped defaults. One case is genuinely unrecoverable and is not guessed at: a sysop who deliberately set a key to exactly its default value leaves no evidence that they did.Fix: an upgraded install reported the version it was first installed at, not the one it was running.
currentversionis derived frompackage.json, butconfigStoreseeds defaults withON CONFLICT DO NOTHING, so the row is written once on first run and never refreshed. On the development box it was seeded on 2026-03-20 saying0.1aand still said0.1asix months later while the code was0.9.5a. Everything that read the config path reported the stale value: the@VR@art code, the terminal system information display, and the@PIDkludge on every FTN message posted. Any install that has been upgraded rather than freshly installed has the same problem.systeminfo.softwareVersion()now readspackage.jsonfirst and treats the config value as a fallback for the case where that file cannot be read. A version is not a preference: setting it does not change what is running, it only makes the system misreport itself to its users and its peers. Migration0005-refresh-stale-versioncorrects the stored row as well, since/admin/configdisplays it, and touches it only when the sysop never set it deliberately, which is the first use of the provenance the previous change made available.Fix: the binkp and EMSI mailers announced
Canectic BBS Mailer/0.1to every peer.PRODUCT_VERSIONwas hardcoded and had never been updated, so every FTN system this software has ever called was told it was running version 0.1. Both now take the version fromsysteminfoat the point the banner is built rather than at import, since config is not loaded when those modules are first imported.The version had six readers going three different ways with three different fallbacks (
0.0.0,0.1a, and the hardcoded0.1).artCodesandmessages.jseach calledappConfigdirectly rather than going throughsysteminfo, anddbmigrationsreadpackage.jsonitself, twice. Everything now goes throughsysteminfo, which also gainedversionInfo()for callers that compare rather than print, andcanonicalVersion()so a strayvprefix or a retired spelling never reaches another system. A test asserts that nothing outsideconfigDefaultsandsysteminforeads the version for itself.
Message Bases: Import, Retention and Moderation
- JAM message base import alongside the existing Synchronet SMB support, for migrating a Mystic hub. Struct offsets follow the JAM(mbp) spec as published in JAM.H;
MSG_DELETEDis0x80000000, the sign bit, so attribute tests go through a helper rather than bare truthiness. Bulk import scans a directory, matches each base to an area by areatag then shortcode, and runs on the worker queue with a mapping form that carries decisions rather than filesystem paths. - Both parsers now stream.
iterateJAManditerateSMBare generators yielding one message at a time, reading each body from a file descriptor rather than holding the whole text file plus every decoded message in memory. This was not theoretical: the first bulk run exhausted the worker heap twice at around 5,000 messages. A 393 MB base with 100,000 messages now peaks at 4 MB of heap, and a real 106,000 message import ran at 157 MB RSS. - Fix: SMB import stored the full
"address serial"MSGID string inmsg_idand leftmsg_orig_addrempty, so a message imported from SMB would not dupe-match the same message arriving later over FTN. It also never setsource_network_id, leaving imported history eligible to be tossed back out to peers. Both now match whatftnToDbdoes. - Fix:
statSMBandstatJAMread into a pre-zeroed fixed buffer and handed the whole allocation to the parser, so a half-copied file looked like a valid empty base. - Optional retention limits on message areas.
max_age_daysandmax_messagesonmessage_areas, with the same pair onmessage_groupsas defaults seeded into new areas. Both NULL by default, and NULL means keep everything, so every existing area and every fresh install prune nothing until a sysop opts in. Enforced by a newmessage-maintenancejob at 04:00 daily. Age measures the message's own date rather than when it arrived, which is what a sysop means by "keep 90 days" and matches Synchronet and Mystic. That carries a sharp edge: a JAM or SMB import keeps its original dates, and real imports show up to 1,999 days between message date and import date, so setting a limit on a freshly imported base prunes nearly all of it on the next run.GET /admin/msgArea/:shortcode/retention/previewaccepts proposed limits as query parameters, reports what would go, and deletes nothing. Preview and the real run sharepruneArea()so they cannot disagree. - Pruning is per message, not per thread. In FTN a reply whose parent has rolled off is the normal state of the network. Survivors get
message_countandlast_messagerecomputed, emptied threads are dropped along with theiruser_thread_pointers, and area counters are refreshed, because the reader lists threads straight from the thread table and orders bylast_message. - Retention now outranks peer high water marks. Pruning previously stopped at the lowest enabled
msg_idxso nothing was deleted before every peer had collected it. Right instinct, wrong rule: one downlink that stopped polling a year ago pins the floor at its old position and retention silently does nothing, for ever. Export runsWHERE id > msg_idx, so a pruned row is simply absent and the peer receives what survives; any subscription left below the pruned range is advanced to the highest pruned id. Both the run and the preview reportpeersAdvancedand apeersBehindlist carryingpreviousIdxandmessagesLostper peer, and each advance is logged naming the peer, so the cost is visible before and after. - Group defaults can now push retention to existing areas, via a Retention tickbox in the Push Group Defaults modal. It is unticked by default, unlike the others: both retention columns are nullable and null means no pruning, so pushing retention from a group with no limits set turns pruning off on every area in the group. The modal says so explicitly when the group is in that state.
- Fix: both push paths, message and file, did
pushDefaultsToAreas(shortcode, fields || ['all']). The form posts one entry per ticked checkbox, so unticking everything sent no fields, fell through to['all'], and pushed every setting. With retention in the list that could silently disable pruning across a whole group in one click. The routes now stop with a warning and both models treat an explicitly empty array as push-nothing. - Moderation is now a property of the message. Every view that shows a message was being handed a page-level
canModerateflag and left to work out what it meant, so the rule lived in the routes and in the browser and the terminal reader never got any of it.messages.jsanswers once and the answer travels with the item:moderationForAreagives the area answer including every destination,moderationFornarrows it per message or thread, andattachModerationstamps.moderationacross a list in one pass including a refusal, so a view never has to tell "not computed" apart from "no". No consumer reads permission flags orsource_network_idany more. - Delete controls sit where the content is visible: a delete in the thread header for the whole thread and one on every message in the thread view, and one per message in the flat view. Not on the area thread list, which shows subject lines only, so deleting from it would be blind. The confirm dialog says plainly that deleting the first message takes the whole thread with it, and that a deletion is local: anything already sent to peers stays where it is and a rescan can bring the thread back. Gated on
areaAccess.canModerateArea(user, area), which is the Sysop flag today and is its own function because per-group and per-area moderators are wanted next. The check is in the routes as well as the templates: hiding a button is presentation, not access control. - The terminal reader gets delete and move on D and M, confirmed, with a scope question when a thread is on screen and a lightbar list of the destinations that message's own answer allows. The dumb-terminal path gets the same operations with a numbered list.
- Fix: the lazy-load endpoints built their JSON message shape by hand and left out
uuid, so messages appended by "load more" carried a delete control addressing nothing while the first thirty on the page worked. Both endpoints now go throughmessages.messageForClient, one tested definition of the wire shape. - Fix:
buildMoveTargetsresolved the networked flag with a query per area, so a board carrying two hundred areas ran two hundred queries on every thread page. - Fix: message dates that cannot be true are rejected at both ingest points. Chasing 2060 dates in an imported COOKING base: all ten were from one author, all carried tz -0400, their MSGIDs decoded to July 2024, and every one was congruent to 16 seconds mod 60, so systematic corruption at the source rather than a wrong clock. Faithfully importing bad data is still a problem, because the damage is downstream and silent: the area's
last_messagebecame 2060-10-14, which pins it to the top of the recently-active sort permanently, keeps it inside every retention window so it is never pruned, and renders as a plausible "Oct 14" in the list column.sanitiseMessageDatenow falls back to the receive time for a date more than two days ahead, before 1980, or unparseable, and logs why. Applied inechomail.ftnToDbandsmbImport. The two-day tolerance is deliberate: clock skew of minutes or hours between FTN systems is routine, years ahead is not. This guards new arrivals only; rows already stored keep their bad dates.
Message Networking: Netmail Relay
- Netmail relay is now
netmail_relay_mode:off,nodelistoron, defaulting tonodelist. The old boolean's "on" was far more open than it read.resolveRoutefinds a way to anything dialable in the nodelist and walks up to a hub, host, RC or ZC for anything else, soallow_netmail_relay=truemeant transit for the whole of every network this system carries, in any direction, including between two of them. A node on one othernet could reach any node on another through this system, which is an open relay however well trusted each individual link is. Nodelists here are scoped per network, so requiring both ends to be listed in the same one is the whole fix: cross-network relay is refused because the ends resolve to different networks, invented addresses are refused because they are listed nowhere, and a zone we hold no address in is refused because we do not carry that network. Ordinary routed netmail is untouched. The rule is network-scoped rather than net-scoped on purpose, since a coordinator routes outbound mail for their nodes and FidoNet spans zones 1 to 6. Points are stripped before the lookup, as points are never listed. An unrecognised mode value fails closed tooffrather than to the default. Refusals now say which rule refused them and name the address they could not find, instead of always reporting "relay is disabled". - Fix: relayed netmail was claiming the wrong origin in both directions. A netmail carries two "from" addresses answering different questions: the message header says who wrote it, the packet header says who is handing the packet over. Those are the same address for mail we originated and different for transit, and they were backwards.
TossOutboundNetmailstamped our address over the message header regardless oftransport_data.relayed, whileWritePktsderived the packet header from the INTL kludge, which on transit mail names the original endpoints. So a relayed netmail went out claiming to be written by us and sent by the originator, with replies going back to the wrong system.resolveOutboundSourcenow answers both questions in one tested place, andWritePktsgroups on an explicitPacketSourcerather than guessing from INTL. - Fix: no Via line was ever written, so an enabled relay had no loop protection at all. Two systems routing to each other would pass the same message back and forth until a sysop noticed.
relayNetmailnow refuses a message that has already been through any of our addresses and stamps its own hop before queueing. Matching is on the bare address and case insensitive, because a system may write itself into a chain with a domain on one hop and without it on the next, and a loop check that misses is worse than none since it reads as protection. - Fix: the inbound Via parser kept two tokens and dropped the rest, losing the timestamp or the software from every line it read depending on whether the writer used the optional
@. A truncated chain still relays and still looks like a path, so nothing complained. The raw line is now kept so another system's hop survives a relay verbatim. - Netmail addressed to Ping now gets a Pong back showing the route it took. A convention rather than a numbered standard, but Husky, Mystic and Synchronet all answer it, so other sysops expect it to work; until now a ping aimed here fell through to the undeliverable queue and generated a sysop notification. The reply comes from the address the ping was aimed at rather than the default AKA, so a node pinging one of several addresses can tell which answered, and it carries no Via lines of its own or our own loop check would refuse it. New key
netmail_ping_enabled, default true. - Netmail relay decision log. Relay left traces before this, but no two of them in the same place: a success wrote an alog line and a
pm_outboxrow, a refusal wrote a different alog line, a loop wrote an undeliverable row, and an unroutable address wrote a text dump intoftn/bad. Two of those four roll over, so "did we relay what we should have, and did it leave" was not answerable after the fact.netmail_relay_lognow takes one row per decision, written at the decision point, whatever the decision was. Refusals are recorded on purpose: a gate nobody can see refusing anything is a gate nobody can validate. Each row keeps who wrote the message apart from who handed us the packet, which are the same on originated mail and different on transit, and are the only way to see which link is actually feeding you transit traffic. Also recorded: the Via chain as it arrived, the AKA we stamped, the next hop and how the route resolved, the mode in effect, and thepm_outboxuuid the message became. That uuid makes the useful query possible:pendingDelivery()finds relays that were queued and never sent, which nothing else could show, because queueing looked like success at every point the old logging covered and a message then sitting unsent produces no further event of any kind. Retentionnetmail_relay_log_retention_days, default 180, longer than the other logs because relay disputes surface weeks later. - Admin page at
/admin/relay-log, under Networking alongside the file logs it resembles. It leads with the thing worth acting on, which is messages accepted for relay and queued more than N hours ago whose outbound row still has not gone. Below that, counts by decision over a selectable window, every decision with a per-row detail view showing the Via chain as it arrived and how the route resolved, filter buttons per decision, and transit mail grouped by who handed us the packet, which is where a misconfigured or probing link shows up. CLI equivalent atnode src/tools/relayLog.js. - Sender authentication is deliberately not attempted. FTN cannot verify a claimed origin, and per-link trust rules buy little against the real exposure, which is one of your own authenticated links injecting forged mail. That waits for a network whose nodelist can carry public keys.
Message Networking: AreaFix, FileFix and Echolists
%RESCANon both AreaFix and FileFix. Neither fixer had a rescan, so a downlink that subscribed to an area got nothing until the next new message arrived and had no way to ask for backfill. Grammar, identical on both:%RESCAN(every subscribed area, default window),%RESCAN ALLor*,%RESCAN AREATAG,%RESCAN AREATAG 30for a day count, and%RESCAN ALL /FULLfor everything still held. Bare%RESCANmeaning "all" is what SBBSecho, Mystic and Squish implement, so it must not return an unknown-command reply./FULLis the documented flag because a leading dash already means unsubscribe in the AreaFix grammar. Defaulting to a window rather than everything is deliberate, since a full rescan of a busy area pushes thousands of messages down a link:areafix_rescan_default_days(60),areafix_rescan_max_days(0, no ceiling, does not restrict/FULL) andareafix_rescan_enabledmake that a sysop decision. Two limits are stated in the response text rather than left to surprise the downlink: retention caps what can be replayed, and imported history is never replayed.- Only the parser is shared between the two sides, because the mechanisms genuinely differ:
peer_area_subscriptionshas amsg_idxhigh water mark so a message rescan rewinds a pointer, while file subscriptions have no pointer at all and distribution is per file per peer, so a file rescan re-queuesfile_distributionrows to pending and creates them where a file was never queued for that peer. The pointer anchors on the oldest in-window id minus one rather than the newest row older than the cutoff, becausemsg_idxis an id and the window is a date and the two do not move together: an imported message carries a high id and a very old date, so the obvious anchor lands above in-window messages with lower ids and silently skips exactly what was asked for. A rescan only ever moves a pointer backwards. - Rescan button on every peer connection row in the admin peer page, opening a modal taking an areatag (blank or ALL for everything), a window in days, and a Full checkbox. It mirrors the netmail grammar and calls the same model functions, so the UI and
%RESCANcannot drift apart. FTN connections previously had no reset control at all: the existing Reset Idx buttons appear only for NNTP and QWK, so the main case had no way to resend held mail from the UI. - Batch subscribe for echolists. Subscribing from an echolist was one modal per area, and BACKBONE.NA carries 241 areatags. New page at
/admin/echolist/:networkId/batch(with?echolistId=Nto scope to one list) lists only uncarried areatags with a checkbox each, a filter box, and a select-all that respects the active filter so a prefix can be taken without touching the rest. One shared settings block covers the whole batch: target group, read and post security, post_as, newscan default, the four flag lists, and the peer allow/enable toggles. The work runs asecholist:batch-subscribeon the maintenance queue rather than in the request, because each area costs a uniqueness check, a QWK conference allocation, an INSERT, two backing tables, a scan pointer per active user, the network mapping and a peer sync. A sysop notification reports created, failed, skipped and renamed counts when the run lands. Create-only: mapping an existing local area onto an areatag is a per-area pairing decision that does not generalise to a batch. planBatch()is a read-only pre-flight exported separately, so a dry run cannot disagree with what the job does. It resolves shortcode collisions against both the database and codes claimed earlier in the same run, since sanitising is lossy and FSX.GEN and FSX-GEN both want FSX_GEN, and it dedupes repeated areatags, because an areatag is unique per network so a repeat means a doubled checkbox rather than a second area. Failures are isolated per area so one bad tag does not abandon the rest.- Batch subscribe for filebones, at
/admin/filebone/:networkId/batch, jobfilebone:batch-subscribe. Kept as a parallel module rather than folded into a shared abstraction, matching the convention thatfileEcholist.jsmirrorsnetworkEcholist.js: file areas have no backing tables, creating and mapping are separate calls, peer sync takes both flags in one pass, there is no QWK conference number, and it adds browse security, browse flags and themax_age_days/max_filesretention limits.sub_diris deliberately not a batch setting, sincenewArea()defaults it to the lowercased shortcode and one shared value would put every area in one directory. - Fix: deleting an area left echolist entries pointing at it.
removeAreacleaned up the area record, its two backing tables, its peer subscriptions and its network_areas mapping, and leftlocal_area_idonnetwork_echolist_areasalone. That field is what makes an areatag read as locally carried throughout the admin UI. The daily reconciliation would eventually clear it, but only for active entries and only by matching areatag, so an inactive entry kept the pointer for ever and every entry kept it until the next daily run. The window is the part that bites: area ids come from a sequence, so a stale pointer does not stay visibly broken, it eventually names whichever area took the id, and nothing in the UI can tell that from the truth. BothremoveAreafunctions now clear by id at the point of deletion, which needs no areatag match and covers inactive rows, and both reconciliations now clear any pointer whose area no longer exists at all, which repairs databases already carrying dangles. - Fix:
removeAreareported success when the final delete failed.postgres.writereports a failure in its return value rather than throwing, so thetry/catcharound it could never fire: the area stayed and the caller was told it had gone. Both the message and file versions now check the returned error, which is also what makes the echolist clearing above meaningful, since a cleanup that cannot be seen to fail is not a cleanup.
Message Networking: FTN Tosser and Packet Fixes
- Fix: a hub was suppressing nearly all forwarded echomail.
getOutboundForConnectionskipped any message whosesource_network_idmatched the connection's network, on the reasoning that a message should not go back to the network it came from. That conflates the network with the node: a hub receives from its uplink on network N and must forward to its downlinks, which are also on network N. That is what echomail distribution is. SincehandleInboundEchoMessagestampssource_network_idon every inbound message, the filter matched nearly everything: on the reporting system, 127 of 129 candidate messages for a downlink were suppressed, and the two survivors were locally posted. Only mail the sysop typed themselves was leaving the system. Not re-sending to the node a message came from was already covered twice, by the explicit source-address check and by full SEEN-BY parsing per FTS-0004. With the filter gone, the same 129 candidates break down as 126 stopped because that peer sent them, 2 by SEEN-BY, 1 sendable. The one useful job the filter was doing, keeping imported history from being re-exported, is now an explicitimportedcolumn on the area tables. - Outbound eligibility is extracted into
evaluateOutbound(msg, dest, self)so it can be asked "would this be sent?" without sending it. It previously lived inline in a function that also writes packets and advances watermarks, which is precisely why a filter suppressing 127 of 129 messages went unnoticed for as long as it did. There is a test assertingsource_network_idis not consulted, so the outage cannot be reintroduced by someone reading the field name and assuming it belongs. - Fix: message area creation was broken outright. Appending the
importedcolumn tosrc/templates/msgTable.confleft the file with a trailing newline;getColumnsFromFilesplit on newline with no filtering, so the trailing newline became an empty column entry andCREATE TABLEemitted", )". Any new message area created would fail after writing its row, andnewArearolls the row back, so the symptom was a create reporting failure with a half-made area.getColumnsFromFilenow drops blank lines so the format is not one stray keystroke from breaking area creation again. - Fix: the import seal could jump over mail that arrived mid-import, which was the originally reported symptom.
advanceOutboundIdxForAreasealedmsg_idxtoMAX(id)as of the seal rather than to the last row the import wrote. Inbound tossing runs on the toss worker while a bulk import runs on the maintenance worker, so ordinary mail lands interleaved with imported rows, and sealing toMAX(id)movedmsg_idxpast that mail so it was never sent downstream, silently. During a long bulk import that is the expected condition, not a rare race. The seal now stops one short of the oldest non-imported row above the floor. - Fix: the netmail writer stamped no packet password. Two writers produce packets: echomail goes through
WritePktToNode, which reads the peer's ownpacket_passwordfromconnection_data, while netmail went throughWritePkts, which read the globalPacketPwdkey and nothing else. A sysop who set a packet password on a link therefore got authenticated echomail and netmail with eight zero bytes where the password belongs. Found from two AreaFix requests an uplink returned: the password was correctly in the subject, which is where AreaFix wants it, and the packet header field was empty, which is what the uplink was actually rejecting. The peer lookup is keyed on the packet destination rather than the message destination, since that is the system we hand the packet to and therefore whose password applies, which also means relayed mail gets the hop's password. Address matching drops a 5D domain suffix and a trailing.0. The global key stays as the fallback for links with none of their own. The AreaFix/FileFix/Ping response path was also sending unauthenticated replies and had stopped reporting queued responses because it was not awaiting the now-async writer. A packet going out with no password at all is now logged, which is worth knowing before the far end tells you. - Fix: the inbound backstop only ever noticed literal
.pktfiles.TossInboundalso processes arcmail bundles, matched on the FTS-5000 filename pattern, so a bundle sitting in the inbound triggered nothing. In normal operation that was invisible, because the mailer enqueues a toss itself the moment a binkp session receives files. It only shows when that enqueue did not happen: a worker that was down, a restart between receipt and enqueue, a bundle put there by hand or recovered from a filebox. Those are exactly the cases a backstop exists for, and mail could sit unclaimed indefinitely with nothing reporting it. Both paths now shareftnTosser.isTossableInboundFileso they cannot drift apart again, and the scan names the file that triggered it. - Fix: the inbound scanner threw a stack trace on every pass if the inbound held a subdirectory, which sysops routinely keep (
ftn/in/Oldis the usual one). The type sniffer read every directory entry withreadFile, so each pass threw EISDIR and dumped a trace throughconsole.error, outside the log bus and outside the configured level. At the inbound scan interval that is a trace every few minutes for ever, which on a live hub buried the lines that mattered. It now stats first and treats anything that is not a regular file as "not ours". It also sniffs withfileTypeFromFilerather than reading the whole file: magic numbers live in the first few bytes, but an inbound holding arcmail bundles or TIC attachments was being loaded into memory in full, every file, on every scan. - Fix: a colon-form INTL kludge crashed the whole packet read.
processMessageTextthrew "Assignment to constant variable" on anyINTL: <dest> <orig>control line, because the dedicatedINTLbranch reassigned consts declared above the switch and ES modules are strict mode, so it failed the whole packet rather than one message. The branch was wrong in three further ways even without the throw. Removing it is the fix: the default branch already parses both forms correctly.
Mailer
The
[binkp DIAG]tracing is gone. Fifteen call sites across the session controller, the binkp protocol and the TCP transport, left in deliberately while the multi-batch file session work was being proved against a live BinkIt peer. They wrote magenta lines into every mailer log on every session. Removing them also took_t0insendGOT, which existed only to time them, and the'end'listener in the TCP transport, which existed only to log peer FIN: Node closes a socket on FIN by itself under the defaultallowHalfOpen, so an empty listener was dead code rather than load-bearing. Closes #32.Mailer session history. The mailer knew all of this while a session was running and kept none of it. The only durable trace was the backoff state on
connection_data, which is overwritten on every attempt, so it answers "is this peer in trouble now" and cannot answer "what happened the last five times we called".mailer_sessionsrecords one row per attempt in both directions, with outcome and reason, files and bytes each way, duration, the peer's M_NUL VER/SYS/ZYZ banner, every AKA it presented, and the host. Attempts that never opened a socket are recorded too, which matters more than the rest: the peer failing hardest is the one that never gets far enough to have a session, and recording only completed handshakes would leave it looking like the quietest node on the board. Writes are diagnostic, so a failed insert is swallowed rather than taking a session down with it. Retentionmailer_session_retention_days, default 30. Admin views at/admin/mailer.outbound_file_logmirrorsinbound_file_log, written on M_GOT so a row means the peer acknowledged the file. Kept as its own table rather than a direction column, because the FTN tosser looks packets up ininbound_file_logby filename to trace senders. Retentionoutbound_log_retention_days, default 90.Fix: a peer configured to poll every 240 minutes was being called every minute. The poll schedule was correct and had nothing to do with it. The BSO scanner decides a node has mail from the filename, never the contents, which is fine for packets and wrong for flow files, which are a list of things to send and can end up listing nothing. A spent flow file therefore read as pending work for ever: the scanner triggered a session, the session skipped every line, nothing was sent, nothing deleted the file, and it happened again on the next scan. At the default 60 second scan interval that is one connection a minute indefinitely, and
poll_intervalis not involved because this path calls_connectToPeerdirectly. Two ways a flow file gets into that state: every line already marked~, because the delete was gated on a line changing during that call, or lines naming files that no longer exist, which were skipped silently and left unmarked for ever.src/utils/floFile.jsnow owns this:pruneFlowFile()marks vanished lines and deletes an exhausted file, called from the scanner before it counts an address as having mail and from both outbound list builders so the tidy-up does not depend on any one caller running.pendingCount()is the honest answer to whether a node has mail, as against whether a flow file exists.Fix: connect backoff was doing nothing whenever mail was waiting. The guard was in
pollConnection()only, and the BSO scanner calls_connectToPeerdirectly, so a peer refusing connections was retried every scan interval. The guard now lives in_connectToPeeritself, which every outbound path comes through;triggerPollpassesforce: true, because a sysop pressing the poll button has said what they want. The scanner keeps its own check so it can report backoff skips in the scan summary, but enforcement no longer depends on any caller remembering.Fix: nodelist-routed addresses had no backoff of any kind, which is the worse half of that. Backoff state lives in a peer connection's
connection_dataand these have no peer row, so nothing counted their failures and nothing limited the retries: a dead route cost a connection a minute indefinitely. The manager now keeps an in-memory map for them with the same doubling and cap, fed by both the connect error and the session end and cleared when a session completes. In memory rather than persisted because these are transient routes rather than configured links, andmailer_sessionsis the durable record.Fix: a poll interval changed in the admin UI sat in the database doing nothing until the next gateway restart.
updateConnectionScheduleexisted, its doc comment said to call it from the admin routes, and nothing did. Schedules were only ever written byregisterSchedules()at startup, while the peer page showed the new value the whole time. The same gap left orphans behind: registration skips connections whose polling is disabled rather than clearing them, and nothing removed a schedule when a connection was deleted, so both kept firing indefinitely. The create, update and delete routes now maintain schedules directly, andupdateConnectionScheduletakes the connection row rather than loose fields so a caller cannot pass one with polling disabled and get a schedule anyway.
Worker Health and Queue Diagnostics
- A hub went four days without tossing while pm2 reported the worker online for the entire period. The process was alive, its event loop was turning, and every concurrency slot was held by a job that would never return. The worker owns both the toss and maintenance queues and nothing else drains them, so losing it stops all message tossing, all TIC processing and every maintenance sweep while the web and the terminal carry on normally, which is why it goes unnoticed. There are three failure modes and only the first is the one a supervisor sees: the process exits (pm2 restarts it), the process is alive with its consumer closed (nothing), and the process is alive with a job that never returns (nothing). BullMQ's stalled-job detection cannot cover the third, because a job parked on an await renews its lock exactly like a healthy one; at the default toss concurrency of three, three of those and the queue is stopped with nothing logged.
- Both queue processors are now wrapped in a timeout, so a job past its limit rejects, BullMQ retries it, and the slot comes back.
toss_job_timeout30 minutes,maintenance_job_timeout60 minutes (higher because imports and batch subscribes legitimately run long), 0 disables. - The worker publishes a heartbeat carrying pid, uptime, per-queue consumer status, when it last finished a job, and the age of the longest job in flight. A clean shutdown writes a
stoppedmarker, without which the watchdog cannot tell a sysop runningpm2 stop workerfrom a worker falling over. The worker also self-checks: a closed consumer that has not recovered across several beats exits non-zero so the supervisor brings back a clean process. - The watchdog runs in the gateway, because a dead gateway is obvious in minutes and a dead worker is silent for days (it is skipped in embedded mode). It escalates: log, notify the sysop once per incident, then restart. The restart is gated by grace window, cooldown and attempt cap, because a worker that cannot stay up is not fixed by restarting it a hundred times and a restart loop is worse than the outage. A bare
pm2inworker_restart_commandis resolved next to the running Node binary before falling back to PATH, since under nvm the daemon's PATH may never have reached the gateway. New keys:worker_health_enabled,worker_heartbeat_interval,worker_health_check_interval,worker_health_grace,worker_selfcheck_max_failures,worker_restart_enabled,worker_restart_command,worker_restart_cooldown,worker_restart_max_attempts,worker_schedule_lag_grace. - pm2 config: the worker had
max_restarts15 with a flat 2s delay and a 3smin_uptime, so a worker exiting at startup because PostgreSQL or Redis was briefly away burned all fifteen restarts inside a minute and pm2 then left it stopped for good. Exponential backoff and a 30smin_uptimeinstead. - New queue diagnostic:
node src/tools/schedules.js. It reports the worker's own heartbeat, per-queue depth with the age of every active job, and every registered scheduler with its interval, next run and iteration count, then compares that against what the database says it should be and flags mismatches and orphans.--syncmakes them match without a gateway restart,--pruneremoves orphans only. Next run is the headline and matters more than it reads: a BullMQ scheduler is not a timer, the next iteration is created when a worker picks up the current delayed job, so a stopped or wedged worker freezes every schedule where it stands while the registered intervals still look perfect.--syncdeliberately skips overdue entries, because overdue is a consumer problem and re-registering would look like a fix and change nothing. The watchdog warns on an overdue schedule alongside a healthy worker, which usually means the processes disagree aboutREDIS_BULLMQ_DB. - Orphan job schedules are now removed automatically at gateway startup. Registration rebuilds the whole managed set from the database on every start, and anything in a managed family it did not just write has no row behind it: a connection that was deleted, or one whose polling was turned off, which registration skips rather than clears. Those kept firing their job for ever, once per interval, and the only way to find them was to run the diagnostic by hand. Five were live on the development box, two of them for connections that had been deleted outright. The sweep covers mailer polls, QWK and NNTP pull/push, timed events and coordinator stub generation, and is confined to those families by an explicit allowlist, so the fixed daily maintenance schedules can never be candidates however the expected set is built. It runs only after a registration pass that completed, since pruning against a half-built list would delete live schedules.
src/tools/schedules.js --prunestill exists for clearing them without a restart. src/tools/schedules.jsnow knows about timed event and coordinator segment schedules, which it did not previously list as expected at all. It shares the orphan rule with registration so the two cannot disagree about what is managed, and it compares cron patterns as well as intervals: a timed event registered with the wrong pattern used to read as matching, because both sides were comparing an interval that neither had.--syncdeliberately skips those two families, since their job payloads differ and gateway startup rewrites them anyway.
Admin Panel
- Tabbed layout for the message group and file group create/edit forms, and for all three file area forms, bringing them in line with the message area modals. The three file area forms were near-identical copies differing only by the group selector and the immutable short code, and now share a
fileAreaFormFields(prefix, mode, opts)mixin. - Fix: file areas could not be created or edited through the web at all. The modal opened, every field worked, and there was nothing to save with. The mixin conversion above left the submit button behind in each caller, so all three forms rendered without one. That conversion was verified by diffing the set of element ids rendered before and after, and a submit input has no id, so the check passed while the button was gone. The button now lives in the mixin, keyed on mode, so a caller cannot forget it, and it sits outside the tab content deliberately: inside, it would appear only while one tab was open, which looks the same as the bug from the other two.
tests/adminFormSubmit.test.jsrenders the real templates and asserts that every admin form has something to submit it with. - Fix: every refused admin request returned 500 instead of 403.
requireAdminrenderedadmin/unauthorized, and there was no such template. The render threw, Express caught it, and the caller got a server error, while the alog line the guard had already written said the request was refused. The logs and the response disagreed on every admin route using the middleware, and a refusal looked like a crash. The page now exists, with the same guard chain every other admin view uses: the unauthorized panel for a logged-in user without access, and the login form for a caller with no session, since a refusal to someone not logged in should offer them the way in. A refusal to a caller that asked for JSON (a.jsonpath,X-Requested-With, or an Accept header naming JSON and not HTML) now answers JSON, because a JSON caller handed an HTML error page reports a parse error rather than the 403 that actually happened.tests/adminRenderTargets.test.jschecks every literalres.render()target insrc/routesagainst the views directory, which is the only way a missing template on an error path is caught before a user finds it. - Post As is now on the new-group modal for message groups. The backend already handled it; the form just never sent the field, so a group created there silently took the
usernamedefault and could only be changed after opening the group.
Web and Terminal
- The message areas list gets a sort toggle beside the search box: the sysop's ordering, or most recently active first. Sorting happens within each group rather than across the page, so the grouping still means something. It is client-side because the rows are already in the browser, and a round trip would drop the current search text and the Subscribed/All toggle. Areas that have never carried a message sort last rather than jumbling in among dated ones. The choice is remembered in
localStorageunder a namespaced key, with reads and writes wrapped, sincelocalStoragethrows outright in a private window and a sort control is not worth breaking a page over. - The last-message column shows the year once a date stops being recent: "Mar 15" while recent, "Mar 2021" once not, which is what
ls -lhas done with the same problem for decades. Previously an area whose last message arrived in 2021 read exactly as freshly as one from last week. The threshold is six months rather than a calendar-year test, and that distinction is the point: "is this a previous year" would drop the day from a six-week-old December message the moment the year rolled over. The exact value is in a tooltip, so the abbreviated form never hides anything. - Fix: an older browser window could be sent back to the call flow on every page, endlessly, while a newer window on the same account was fine.
last_callflow_atis cached on the session so the trigger check does not query the database per request, and that cache has no expiry. Three things had to hold for it to be endless and all three did. The username index dropped the session:session_index:<username>was written once at login with an expiry of twicesession_ttland never refreshed, while the session hash was renewed on every page view, so a Remember Me session outlived its own index entry by days. That also silently broke every other broadcast, including security level changes and config pushes, just less visibly. Completion broadcast to the index and fell back to the current session only when the broadcast had updated nothing at all, so a user with any newer session got a non-zero count and no fallback. And nothing re-read the authoritative value, because the lazy database lookup is gated on a flag set once and never cleared. All three are fixed: the index is re-added on session renew, completion writes the current session first and unconditionally, andGET /callflowreconciles its cached timestamp against the database on load. - Fix: art rendering could stall the terminal outright.
@TC@,@TU@and@ON@reach PostgreSQL or Redis andprocessArtCodesruns on every menu redraw. The resolvers had try/catch and were described as graceful fallback, but a catch only handles rejection: both clients retry internally, so an outage produces a hang, which nothing caught. It was not only an outage problem either, since every redraw ranCOUNT(*)over calls and users plus ansMembersand one Redis get per online user, paid continuously while healthy.resolveStatnow caches per code with a TTL and races a refresh againstart_stat_timeout_ms(new key, default 1500), returning the last known value on timeout while leaving the refresh running to fill the cache for the next render. Once a render gives up on a particular refresh, later renders do not wait on it again: measured with Redis down, 1502ms once and then 0ms. Failures no longer overwrite good data, where previously a failed COUNT cached '0' as though it were true.@ON@splits into a cached web presence count and a terminal node count computed locally, so a Redis outage still reports connected terminal users instead of zero.
Release Engineering
- Cutting a release is now one command.
node scripts/release.mjs --stable 0.9.6arefuses a dirty tree, a failing suite, an empty changelogPendingsection or a version that is not newer than what the tree already claims; closesPendinginto the version and opens a fresh one; writes a release note skeleton; bumpspackage.json; commits and tags; and then builds the artifact. v0.9.5a shipped with no release notes at all and they had to be written retrospectively five months later from the changelog, which is the specific gap this closes: after a cut there is always a file carrying the upgrade actions and the change list, waiting for the prose. - The bookkeeping deliberately happens before the artifact is staged, so a release archive contains its own release notes and its own changelog entry. Building first would ship an archive whose changelog says the release is still pending.
- Upgrade steps and notices are now declared rather than only written as prose. This release's Upgrade Notes carry nine actions and the update script runs two; the rest are the sysop's job to read, match against their install and perform, and several are silent when missed.
docs/releases/pending.jsonaccumulates the same actions in a form a machine reads: a step has something to run and a condition saying whether this install needs it, a notice has nothing to run and is a behaviour that changed underneath the sysop. Nothing consumes them yet, which is the update check and the updater, but a release published from here on carries them. - The two halves are cross checked at cut time and never derived from one another. Parsing a changelog to decide what to execute on somebody's live board works right up until a bullet is reworded, and the failure is silent. What the script reports instead is a whole half missing: upgrade notes with nothing declared, or declarations with no upgrade notes.
- A release now publishes a content-hash manifest of every shipped file.
update.shcompares mtimes to decide whether a sysop edited a file, which says when it was written and nothing about whether the contents differ from what shipped. The manifest also makes deletions expressible: it describes the complete target tree, so a file present locally and absent from the manifest is one to remove. A delta expressed only as changed files cannot say "this file is gone", which is exactly whyupdate.shreaches forrm -rfon code directories. - Development builds ship as deltas against the last stable.
--devderives its version frompackage.jsonplus today's date, packs only the files that differ from the base, and carries the full manifest so the rest is reconstructible. A build touching five files ships five files rather than the whole tree, which matters down a file echo. Always against stable and never against the previous development build, so the base cannot be ambiguous and one pruned artifact cannot break a chain. - The build stamps the computed version into the
package.jsoninside the artifact. Without it every development build would install claiming to be the stable it was based on, be told by an update check that it is current, and never see a newer build at all. A consequence worth knowing: that one file always differs from the base, so "nothing has changed since the last build" needs the staged file re-hashed with the base's version put back, which is how the script refuses to publish a build with nothing in it. - What ships, and how each file may be treated on an install, now has one definition.
src/release/shippedTree.jsreplaces the directory lists that were written twice ininstall.shandupdate.shand drifted, which is what letinstall.shdelete every sysop-added asset underpublic/. The categories are behavioural claims (overwrite and delete strays, overlay, seed once, merge, skip art) and travel per file inside the manifest rather than being re-derived, so a release that adds a directory carries the rule for it. A path with no rule does not ship at all rather than defaulting: defaulting a new top-level directory to code would mean the first release after somebody adds one deletes whatever a sysop keeps in it. - A stable cut refuses to publish if it drops a migration its own development builds already applied. Because a development build names its base in its own version, the set of builds that put migrations into people's databases this cycle is enumerable straight out of the release index. If a migration went out on development and got deferred out of the stable cut, everyone running one of those builds is about to be refused the upgrade by the runtime move rule; the script names the builds and the migrations and stops. The design called for a warning here; a warning in script output is one nobody reads, and
--allow-migration-gapcovers the case where retiring the migration is deliberate. - Also published per release: a per-channel index listing every release newest first with its migration ids, so a client answers "is there anything newer, and can I move to it" in one fetch, and
SHA256SUMSover everything, optionally GPG signed. Getting the files onto the host is still by hand.docs/reference/RELEASE_PROCESS.mdis the maintainer guide. - A release is a ZIP rather than a tarball, for one reason that matters to this audience:
FILE_ID.DIZ. Every BBS file area system knows how to pull a description out of a ZIP and almost none can do it from a tarball, so a release travelling down a file echo as a tarball sits there with no description, which is a file nobody downloads. The usual objection to ZIP is Unix permissions and it does not apply here, becauseMANIFESTrecords each file's mode and the updater chmods from it. bootstrap.shand its checksum are now published with the release.INSTALL.mdtells a new sysop to fetch both and runsha256sum -c bootstrap.sh.sha256before running anything, and neither file was produced by anything, so the documented path could not be followed. The release script now copies the bootstrap out ofsetup/on every cut, writes the checksum beside it in the format that command reads, and publishesINSTALL.mdalongside so the install can be read about without downloading a release first. The bootstrap is the one file that cannot ship inside a release, because resolving a release is its job, and a stale copy on the host is invisible: it resolves the newest release perfectly well and then hands off to an installer it was never checked against.- Fix:
SHA256SUMSlisted the metadata sidecars and not the software. It was built from a pattern naming.tar.gz, which went on matching nothing after the artifact became a ZIP, and which never matched the channel index either. Signing that list would have signed everything except the thing worth signing. The list is now built by exclusion, meaning everything in the publish directory except the list itself and a signature over it, so nothing new can be left out of it by omission. Artifact integrity was never affected: the bootstrap and the updater both verify the archive against the checksum in its own metadata, which was correct throughout. - Fix: a release cut in the evening west of Greenwich was dated the following day, in the changelog heading and the release metadata both, because the date came from UTC while the development build stamp beside it came from the local calendar. Both now use the local date, which is the one the person cutting the release means.
- Releases are signed, starting with this one. The public key is published beside
bootstrap.shand committed in the tree, so it also travels inside every release. Import it once and the bootstrap checks the signature on everything you ever fetch; without it you get the same checksum verification as before and a note saying the release was not signature checked. Nothing is refused for being unsigned yet. The fingerprint isF3F1 6D36 59AF 25FB F947 4C6D B5D8 AB14 3016 761Fand it is printed inINSTALL.md, the README and these notes so it can be compared against more than one place. - Two signatures, because they reach different people.
SHA256SUMS.ascsits at the release root on the host and says the publish directory is ours before anything is unpacked.MANIFEST.ascrides inside the archive and says the tree is ours after, which is the one that reaches a board fed by a file echo: a checksum list at a web address never travels down an echo, so a release signed only at the host could never be verified by the boards the FTN transport exists for. The second is also the stronger, since it vouches for the content rather than the container and survives a chain renaming or re-archiving the file. - What a board will pin is the primary key's fingerprint rather than a copy of the key. A stored copy is a snapshot: a signature from an expired signing subkey still verifies against it, but one from a rotated subkey does not, so every routine key rotation would strand exactly the boards that cannot reach out to refresh. Pinning the fingerprint and taking the key material from each release makes rotation a non-event and reserves refusal for a change of primary key, which is the case that should be refused. The verifying half of this lands next release; what matters now is that a release published today carries what it needs.
- Fix: a signed cut created a correct, verifiable tag and then died on the line after it. The tag command was given inherited output so the passphrase prompt would reach the terminal, and Node returns nothing rather than a string when it is not capturing output, so the helper that trims every git result turned a command that succeeded into a crash. The artifact was never built and the repository was left holding a release commit with no release. The helper tolerates it now, and the tag no longer asks for inherited output at all, because the passphrase prompt is drawn on the terminal named by
GPG_TTYrather than on the command's own output. - The release script's orchestration is now importable and tested rather than only runnable. It used to carry a comment saying the orchestration was the part a test could not usefully assert; two bugs in one evening landed in exactly that part, both of them in the handful of lines between the suite passing and the artifact existing, which is the worst place in it to be wrong because the repository has already been written to by then.
- A signed cut proves it can sign before it writes anything. The first attempt at one got as far as closing the changelog, clearing the pending declarations, bumping
package.jsonand committing, and then failed ongit tag -sbecause gpg could not reach a terminal to ask for the passphrase, leaving a repository holding a release commit for a release that does not exist. Having a key is not the same as being able to use it, and the difference only shows up at the moment of use, so the cut now uses it first on a throwaway file. The failure it hit most often names an ioctl rather than the thing to fix, so the message names the thing to fix:export GPG_TTY=$(tty). - Fix: the bootstrap could be talked into unpacking a substituted release. It checked the archive against the
sha256inside the archive's own metadata file, then verified the signature overSHA256SUMSand never opened it. The document the archive was actually trusted against was therefore vouched for by nothing, while the sysop was told "signature verified". Anyone able to write to the release host swapped the archive and its metadata together, left the genuine signed checksum list alone, and both checks passed. The order is now signature, then checksum list, then metadata, then archive, with every link checked and a refusal naming the file that disagreed. A release delivered the way a file echo delivers one, with no checksum list beside it, still proceeds on its own checksum and says that is what it did. - Fix: the release script would build a delta against whatever manifest it was pointed at, without checking it described the tree the metadata names. The version and the
basefield are both derived from the version string, so they agree with each other whichever manifest was loaded, and a--base-manifestaimed at the wrong place produced a published delta that reconstructs into a tree no release corresponds to. The manifest states its own version, so the check is one comparison. - Cutting a release now also puts it on the GitLab releases page, with the release notes as its description and the real artifacts attached: the archive, its metadata and manifest,
SHA256SUMS, the signature over it, and the public key. The page has carried an entry for every tag since v0.9.0-alpha.1, all of them made by hand, so the first release cut by script would have been the first one missing from the page people actually look at. - Missing would not have been the worst of it. GitLab generates a "Source code (zip)" for any tag by itself, so a release page looks like it offers a download when what it offers is a git tree: no
MANIFEST, noCANECTIC.ID, noFILE_ID.DIZ, and installing it takes the from-source path rather than the bootstrap one. Every release before this one offered exactly that and nothing else. - The release host stays canonical.
update_check_urlpoints there, the bootstrap fetches from there, and a file echo carries the same archive; GitLab is where a person finds a release and reads what changed. A second fetch location is safe precisely because the release is signed: the signature is over the content rather than over a location, so a copy elsewhere cannot say anything the original does not. Mirroring unsigned artifacts would be adding a second thing that can drift. bootstrap.shis deliberately not attached there. It defaults to fetching from the release host, so a copy downloaded from a release page would be a trap for as long as that host is not serving, and the archive on its own installs perfectly well.- Creating the page entry is the last thing a cut does and it is never fatal. By then the archive exists and has been verified, the commit is made and the tag is signed, so a
glabthat is missing or unauthenticated is reported with the command to run by hand rather than failing a release that is already complete.--no-gitlabskips it.
Update Checking
- Canectic can now tell you when there is a newer release.
/admin/updatereports what is available on each channel and whether this system can take it, and an opt-in daily job raises a sysop notification when something newer appears. It is off by default, because an outbound request from a sysop's box should be something they turned on rather than something they have to discover and turn off. It never downloads an artifact and never applies anything: installing an update is stillsetup/update.sh, and the in-BBS updater is the next phase. - No telemetry. The check is a plain GET for a static JSON file, with no query string, no identifier, no version and nothing about the board or its users. A server logging it learns that an IP address asked for a file, which is what any web server learns about any fetch. With the check off and nobody pressing the button, this system makes no outbound request at all.
- A release is reported as blocked, with reasons, rather than hidden. Three things block a move: this database has applied a one-time migration the target release does not ship, the install is below the target's declared upgrade floor, or Node is older than the release needs. The first is the one that matters, and it is the case a version comparison cannot see: work that went out on a development build and then got deferred out of a stable cut makes the move forward by number and impossible in practice. "You cannot move to stable" is not actionable, so the page and the notification both name the migration.
- The migration rule reads only migrations the ledger recorded as
applied.skippedmeans the migration decided this system did not need it andpreinstalledmeans the database was born current, so in neither case is there a transformation for a target to understand. Reading every settled id instead would have refused every upgrade on a brand new board, since a fresh database records the lot as preinstalled. - A system on the development channel is also told when stable is available, because converging back onto stable is a move worth being offered and would otherwise never be seen. A system on stable is never told about development builds: that is a channel they did not choose, and offering it would be the software talking somebody into running prerelease code.
- Notifications deduplicate on version and verdict. A daily job that says the same thing every morning is a job a sysop turns off, after which they hear about nothing at all. Both halves are in the key because a release that was blocked and is now reachable, because they ran the migration or upgraded Node, is news even though the version has not moved, and that is the notification they have been waiting for.
- A channel with no index published reads as "not resolved" rather than as an error. A project that has cut no development builds has no development index, and a 404 there is the normal state of affairs; putting a red error on every stable-only install's page daily would train sysops to ignore the page. A timeout, a server error, or a document that is not an index is a real error and is recorded as one. One channel failing does not discard the other.
- Every check is recorded, whether it found anything or not, and shown as history on the page. A page that shows only the last result cannot tell a check that found nothing from a check that stopped running, and the second is the failure worth seeing: an opt-in daily job that silently stopped looks exactly like a project that stopped releasing. Retention is
update_check_retention_days, default 90. - The daily schedule is registered only when the check is enabled, rather than registered and skipped.
src/tools/schedules.jscompares every registered scheduler against what the database says should exist, and a permanently inert entry would be a line that means nothing on the one page whose value is that everything on it means something. Turning the check on therefore needs a gateway restart orschedules.js --sync, like every other schedule-affecting setting. - Check now on
/admin/updateruns regardless of the setting. The opt-out is about this system reaching out unprompted; a person pressing a button is not that, and refusing them would mean the only way to ask a question is to change a setting and restart. - Only the settings that do something are configurable:
update_check_enabled,update_channel,update_check_urlandupdate_check_retention_days. The FTN transport keys in the design are deliberately absent until that transport exists, since a setting whose other value silently does nothing is exactly what the options mechanism was added to prevent. New sysop guidedocs/sysop/UPDATES.md.
Updating: Staging and Applying
- A release can now be fetched, proved and landed on an installation by one tool.
node src/tools/update.jsresolves a release, downloads it, verifies its checksum before unpacking anything, reconstructs a development delta against its base stable, checks the result against the manifest, works out exactly what applying it would do, and with--applydoes it. Without--applyit writes nothing to the installation. - This is half of the updater and says so. It does not stop the BBS, install dependencies, migrate the database, run the release's declared upgrade steps, or restart anything, and it prints those as remaining work when it finishes rather than implying they are done.
setup/update.shis still the supported path for a normal update. New sysop guidedocs/sysop/UPDATING.md. - What an update does to a file is now a value a sysop can read, not control flow in a shell script.
buildApplyPlanis pure and returns a decision and a reason per file, so a dry run is the same code path with the writes turned off. This is the module that can destroy somebody's work:install.shexpressed the same rules as a sequence ofrm -rfandcpcalls, which is why nobody noticed that its update mode deleted every sysop-added asset underpublic/. - Each category is a claim about whose file it is. Code is ours: overwritten, and strays in those directories are deleted, because a module the new release does not know about is still importable. Overlay is ours to add to and theirs to extend.
public/css/custom.cssis theirs the moment it exists. Config is theirs once touched. Art is theirs. Removal happens only in code directories, since anywhere else a file we did not ship and do not ship now is the sysop's, and the general form of that rule is exactly the bug that deleted their door icons and branding. - Whether a sysop edited a config file is now answerable rather than guessed.
update.shcompares mtimes, which say when a file was written and nothing about whether it differs from what shipped, so it has to stop and ask about every data file. Each release ships a manifest of content hashes and leaves a copy in the installation, so "still exactly what the last release shipped" means nobody touched it and the update is safe to take. An install predating this machinery has no manifest and cannot answer the question, so it fails to conflict: their file is kept and they are told. Guessing the other way would overwrite edits nobody can recover. - A category the code has no rule for also falls to conflict rather than being written over. The updater runs from the new release's tree onto an older install, so the categoriser in play may well predate a directory the release added, and "I do not know what this is" has to read as "leave it alone".
- A development build is reconstructed from its base, never patched into place. The delta is laid over a freshly extracted base in staging, pruned to the manifest so deletions travel, and verified before anything is applied. The case that forces this: a file changed in an earlier development build and put back in a later one appears in neither delta at its correct content, so patching leaves it silently wrong while every individual step looked right.
- Everything the apply overwrites or deletes is copied into a backup directory at its own relative path before anything is touched, and files are written to a temporary name and renamed into place so a process reading one during an update sees the old file or the new file and never half of one. A failed staging run cannot leave an installation half updated, because nothing outside the staging directory is written until the tree has been verified.
- Two bugs found while testing this against real artifacts, both of which would have compounded silently. The installation's own manifest was not being rewritten after an apply, so the next update would have compared config files against the previous release's hashes and reported every legitimately-changed file as a sysop edit, for ever. And the manifest was not recorded in the backup, so a rollback left the install claiming the release it had just backed out of, which is the one lie that makes the following update reason from the wrong hashes.
- The move rule is checked before anything is downloaded, not after. Refusing a release after fetching twenty megabytes would be rude and refusing after extraction would be worse. Naming a version explicitly still goes through the release index, because the one time somebody names a version by hand is when they are trying to get past something.
Updating: The Handoff
- An update can now be handed to the release being installed.
--handoffstages a release, then launches the new version's own updater as a detached process and returns. Two reasons. A process cannot replace its own files while running, so an update started from the running BBS has the thing doing the work also being the thing replaced; and applying an upgrade is version specific and is where the bugs are, so doing it with the new release's code means a release can ship a fixed updater and it takes effect for that upgrade rather than the one after.--applystill does it synchronously for a sysop at the console. - Detachment is the load-bearing part, because
pm2 stop allsignals the whole process group and an updater launched without it dies with the thing it was updating, halfway through. Verified by SIGKILLing the launcher's entire process group immediately after the spawn and confirming the update still completed and the installation came out at the new version. - The contract between the launching release and the applying release is a JSON run file, not a command line. One argument,
--run <path>. Those two are different versions of this software and both will keep changing: a newer updater reading an older release's file defaults what is missing, an older one ignores what it does not recognise, and a positional argument list can do neither. - Staging now lives inside the installation, under
.canectic-update/. Not tidiness:node_modulesresolves by walking up from a module's own directory and a release artifact ships none, so an updater staged in/tmpcannot import anything with a dependency, which includesalogger. The dot directory is not a shipped category, so nothing scans, replaces or backs it up. The updater's working directory is the installation, where.envandnode_moduleslive; only its code comes from staging. - A release cut before this existed ships no updater to hand off to. The launcher checks, refuses, and names
--applyas the alternative rather than failing obscurely on spawn. That is not hypothetical, it is every release to date. - Progress is a durable file, not a database row. The updater is replacing
src/, which includes the database layer, and may be running while the BBS is stopped, so making the one record of what happened depend on Postgres would put it out of reach exactly when it is wanted. It lives underlogs/update/, writes are atomic, andnode src/tools/update.js --statusreads it long after the run finished and the process that started it was replaced. - A run refuses to report success over a failed or still-running phase. Reporting a half-applied tree as a successful update is the worst outcome available, because the files are on disk, the processes are down, and the only record says it went fine. "Never reached" and "deliberately not done" are also kept distinct: this updater performs four of the ten phases and records the other six as skipped with the reason and the command to run by hand, so the run reads as incomplete on purpose rather than as stuck.
- Fix: the update tool reported the version of the tree it was running from rather than the tree it was about to modify. Identical whenever
--targetis the current directory, which is exactly what anyone testing by hand would use, and wrong for every other case: the move rule was comparing the wrong system's migrations and upgrade floor. Reading an installation is now its own module with a target argument, and it also answers the two-manifest question the data-file rule depends on. - Fix: naming
--versionexplicitly skipped the move rule. The release was taken at the caller's word instead of being looked up, so the one time somebody names a version by hand, which is when they are trying to get past something, was the one time nothing checked. It now always resolves through the release index. - Fix: the tool would hang indefinitely on a stopped database. The pg pool retries internally, so asking the migration ledger a question on an unreachable database never returns. It is now bounded, and an unreadable ledger comes back as "unknown" rather than as an empty list, because treating "we do not know" as "nothing applied" would clear every migration blocker and wave through precisely the move the rule exists to stop. The upgrade floor and the Node check still run since neither needs the ledger, and only
--applyand--handoffare refused.
Updating: The Whole Sequence
- An update now does the whole job. Stops the BBS, replaces the files, installs dependencies, dumps the database, migrates it, runs the release's automatic upgrade steps, and starts the BBS again, with each phase recorded as it happens. Anything it skips it names, along with the command the sysop now owes.
node src/tools/update.js --handoff, or the Update button on/admin/update. - One implementation of the phases, used by both ways in.
--applyruns them in process and the detached updater runs them after a handoff, through the same module. Two programs that both claim to apply an update is the exact shape that letinstall.shandupdate.shdrift until one of them deleted every sysop-added asset underpublic/, and there was no reason to build it a second time here. - The file backup happens before the processes are stopped, since shipped files do not change while the BBS runs and there is no reason to pay downtime for it. The database dump happens after the stop and immediately before the migration, because it exists to protect the migration: taken any earlier it would not contain the writes in between, and taken before the stop that window would be as long as the dump itself.
- A database dump is taken by default and can be declined. Free space is checked against the actual database size first, so it fails before starting rather than halfway through leaving a truncated file that looks like a backup. Requiring one was rejected: a hub whose message base outgrows its free disk would then be unable to update at all. Warning only was rejected too, because the one time it matters there would be nothing to restore from.
- A failure partway through rolls the files back, but only while that is still the right thing to do. Up to and including the dependency install the only thing that changed is files, so the backup goes back and the BBS is started again. From the schema phase onwards it does not: old code over a migrated database is a worse state than the failure, so the update stops, says where the backup and the dump are, and leaves the decision to a person. The list of phases on the safe side is written out rather than derived from the order, because a phase drifting across that line is silent and only shows up during a failure.
- A failure that stopped the BBS starts it again, and says so. A board left down by a failure partway through is its own problem, separate from whatever failed.
dbmigrations.jsdoes the declarative pass and the migration ledger in one invocation, so themigrationsphase is recorded as skipped with that as the reason rather than marked done. Marking it done would imply a second thing happened. It also meansdbmigrationsis no longer a declared upgrade step: the schema phase is structural and runs on every update, while steps are for release-specific one-offs.- Process control is
update_process_control,update_stop_commandandupdate_start_command. A barepm2is resolved next to the running Node before PATH, reusing the worker watchdog's helper, because under nvm the supervisor's PATH may never have held the version the BBS is running on. Starting from the ecosystem file starts every process in it, including one deliberately stopped, which is a known consequence of the simple option; the pm2 state is recorded so--statusshows what was running beforehand. - Stable release artifacts are now cached, so a development update does not re-fetch its base every time. A delta is always one hop against the stable it names, and a sysop tracking development needs that same tarball for every build cut on it. The cache keeps two by default and never evicts the base the installed version deltas against, nor lets protecting it consume the budget. A cached base is re-verified against its checksum on every single use and discarded if it fails: a local file is no more trustworthy than a downloaded one, it can rot on disk, and it is the base everything else is rebuilt from, so trusting it would silently produce a tree that passes its own manifest check while not being the release.
- Everything the updater keeps now has one home and three lifetimes.
.canectic-update/staging/goes with each update,base/survives them,backup/lasts until reclaimed. Backups moved here fromlogs/update/, which now holds only the record: small, readable and worth keeping for ever, against bytes that are large and reclaimable./admin/updateshows the size of each and what deleting it costs, so "I need that space back" has an answer that is not guesswork. - Preflight runs before an update is offered, not after it has started. Several of its checks are only answerable while the board is up: once the processes are stopped, nobody is online and nothing is in flight, and asking then is asking a question the asking has already changed. So it runs in the gateway with the board running, and the updater re-checks only the cheap invariants it can still see for itself. Blocking is reserved for things that make the update fail; connected callers only warn, because whether three people mid-door is a reason to wait is the sysop's call, and a preflight that refused on judgement calls is one they learn to bypass. Free space that cannot be measured warns rather than blocking, since blocking would make every filesystem without
statfsun-updatable. /admin/updatehas an Update button, gated on preflight passing and on no run already live, with a confirmation that says plainly what it will do and how many users it will disconnect. The ordering is the trick: check, respond, and only then stage and hand off. The gateway is about to spawn something that stops the gateway, so the response has to be gone first, and the page reports afterwards from the run file rather than from the connection it was serving on. The run record exists from the moment the button is pressed, so a failure during staging shows up on the page instead of looking like nothing happened.
Installing: One Command
- A new sysop now runs one script and nothing else.
setup/bootstrap.shresolves a release, downloads it, verifies its checksum before unpacking anything, extracts it, and hands off to that release's own installer. Nogit clone, and no source directory to look after afterwards. The old shape had a second copy of the software whose only purpose was to install the first, which could be stale, on the wrong branch or at an untagged commit with nothing checking. - Install is update from nothing, and that is now literal. The file placement is the updater's own planner against an empty directory, so every file comes out as new and the rules that protect a sysop's work on an update simply have nothing to protect. There is no second copier. The first one was
install.sh, and it drifted fromupdate.shuntil it deleted every sysop-added asset underpublic/. The phase vocabulary is shared too: an install skips backup because there is nothing to preserve and stop because nothing is running, while an update skips the .env phase because that file is the sysop's. - Fix, found by the first end-to-end run: the installer could not run on a machine with no dependencies. It failed on
Cannot find package 'luxon'before doing anything. Obvious in hindsight and easy to miss in development, wherenode_modulesis always there: a fresh machine has an extracted release and nothing else, sonpm installis something the installer RUNS and therefore cannot be something it needs.updateApplylost its logging import and the command helpers moved out of the update runner, which reachesappConfigand sodotenv. A test now walks the installer's import graph and fails on anything that is not a Node builtin or one of our own files, because this is exactly the kind of rule that holds until somebody adds one convenient import. - Runtime directories are now created on updates as well as installs. A release ships no empty directories, because an empty directory is not a file, so the FTN spool, the logs and the art tree have to be made by something. That something was a block of
mkdircalls ininstall.shwhich an update never ran, so a release that starts writing somewhere new worked on a fresh install and failed on every upgraded one, at the moment the feature was first used rather than at update time. .envis written once from the shipped template with the values the installer was given, at mode 600 because it holds the database password, and is never rewritten: a directory that already has one belongs to somebody who set it up. A password is never defaulted or generated, since a credential nobody was told about in a file nobody looked at is worse than being asked for one.- Installing over an existing installation is refused, naming the version found and printing the updater command instead, because it would treat every one of the sysop's files as new.
--forceoverrides it. - Automation stops where it stops being honest. The PostgreSQL role and database need
sudo -u postgres, which is restricted on plenty of hosts and absent on managed Postgres, so--create-databaseruns the script that already handles that properly and without it the command is printed. Everything after the credentials exist is automated. A failed install is deliberately not rolled back: there was nothing underneath, so deleting the directory and starting again is simpler and safer than unpicking it, and the run says so. - A development build cannot be installed from nothing, since it is a delta against a stable release. The bootstrap refuses one and names the stable to install first, rather than reimplementing reconstruction that the updater already does and has tests for.
bootstrap.sh --extract-only DIRunpacks a verified release and stops. The documentation and the setup scripts live inside the release, and the bootstrap unpacked into a temporary directory and deleted it on the way out, so there was no moment at which a sysop could readINSTALL.mdor runcreate_pg_db_user.shbefore committing to an install. It now prints the three things to do next.--download-onlystill stops one step earlier and leaves the archive itself.- The installer checks PostgreSQL is accepting connections before it writes anything. Neither the bootstrap nor the installer's preflight asked about the database, so on a machine without one the install laid down the whole tree, created the runtime directories, wrote
.env, rannpm installand only then failed at the schema phase, which is the worst place to stop: everything is on disk and nothing works.pg_isreadyanswers it, and its absence is reported rather than treated as a failure, since a board on a managed or remote database has no local client and is perfectly fine. - The documented path is verify, then run.
INSTALL.mdleads with downloading the bootstrap and checking its published checksum; piping it into a shell is shown as the alternative rather than the headline. This audience cares about provenance and it costs one line. Signature checking is wired for when a key is published: aSHA256SUMS.ascthat is present but unverifiable warns loudly, and its absence is not yet a refusal.
Updating: Over a File Echo
- Releases can now arrive down a file echo, with no outbound connection of any kind. Set
update_check_transporttoftnand pointupdate_stable_file_areaat the area carrying the release echo. This is how BBS software was distributed for twenty years, it works for a board with an FTN link and no outbound HTTP, and it does not depend on a web host being reachable. The echoes areCANECTICfor announcements,CANECTIC_FILESfor stable andCANECTIC_DEVfor development builds. - Almost nothing new happens on this path, which is why it was built last. By the time a release is interesting the TIC tosser has received it, checked it against the TIC, hashed it and filed it, so detection is a database query and downloading is a copy. The directory source the updater already had was written to take a directory as well as a URL for exactly this.
- A file echo carries files, not indexes, so the channel index is derived from what the area holds. Each archive carries a
CANECTIC.IDand its own metadata at its root, which is what those sidecars are for. An archive that cannot identify itself is dropped rather than offered: its migration list would be empty, an empty migration list clears every blocker in the move rule, and an unidentifiable file would then be presented as safe to install against a database it says nothing about. The filename is treated as a hint only, since it survives a distribution chain that may rename, truncate or uppercase it; uppercased names are normalised, because 8.3 habits are still alive in parts of the network. - What FTN integrity guarantees is now written down rather than assumed. The TIC's CRC-32, checked on arrival, and the SHA-256 Canectic computes and stores both prove a file has not changed since it got here. Neither proves it is the file that was built and published. So the locally computed hash is never put in the field that means "published checksum", and staging a release without one requires the caller to say it accepts that: refusing outright would make the transport useless, and skipping quietly would make the HTTP path a lie. The manifest end-state check runs either way, so a corrupted or truncated archive is still caught before anything is applied.
- Metadata that is absent is normal on a file echo, because two TIC'd files travel independently and one can be lost; the copy inside the archive covers it. Metadata that is present and corrupt stops the update instead, because that is somebody telling you something is wrong and quietly using a different copy would hide it. The two are told apart by a flag on the error rather than by catching everything, which is the difference between a fallback and a swallow.
- Publishing stays by hand, consistent with the release script: the artifact goes into a file area and the existing TIC machinery distributes it.
Schema and Setup
message_groups.post_aswas created by a runtimeALTER TABLE ... IF NOT EXISTSat gateway startup rather than declared indb/schema, sodbmigrationsneither knew about it nor reported it. It was not alone: twoensureSortOrderfunctions called frombin/gateway.json every boot issued eleven such ALTERs between them, six redundant and five genuinely missing. All five are now declared with the same types and defaults, and both functions are removed. Schema lives in one place.Dropped the
netmail_badplaceholder, which was listed intablenames.confwith a zero-byte column definition file, sodbmigrationsreported "no column definitions" on every run and never created the table. Bad inbound netmail is already handled bypm_undeliverable, which carries a full review workflow. The schema test that let this sit unnoticed skipped empty.conffiles as valid placeholders; that skip is now an assertion, so a table listed with no column definitions fails the suite instead of producing a migration error on every run.setup/create_pg_db_user.shcreates the database as UTF8 explicitly (see upgrade notes), and warns when it adopts a pre-existing database that is not.setup/convert_db_encoding.shis new. README gains a manual database creation section, because the script needssudo -u postgres psqland fails where that is restricted; it covers the three things that actually bite, which are the missing ENCODING clause, PostgreSQL 15+ revoking default public schema grants (the cause of permission errors fromdbmigrations.js), and peer authentication.Migrations and startup schema maintenance are roughly ten times faster. Both the static tables and the per-area message tables asked
information_schemawhether each column existed, one round trip per column. A message area has two backing tables of about 28 columns each, so maintaining one area cost roughly 114 queries, andtablenames.confcarries about 127 more tables on top. Measured on a 31 area development box: 6.9 seconds for the area tables and 3.2 for the static ones, paid on everydbmigrations.jsrun and every process that calls it, almost always to establish that nothing needed doing. On a hub with several hundred areas that is minutes per update. Both paths now read a table's columns in one query and compare in memory, and share the reader so they cannot drift. Same 31 areas afterwards: 0.47 seconds and 0.95. This was the other half of the slow update that the MSGID split fix addressed; that half is now done too.The two
ALTER TABLE ... ALTER COLUMN seen_by TYPE TEXTstatements that ran for every message area on every maintenance pass are now conditional on the column still beingvarchar. The template has said TEXT for a long time, so this was twoACCESS EXCLUSIVElocks per area per run to change nothing. The decision costs no extra query, since the column read above already carries the type. A failure to widen is now reported rather than discarded:postgres.writereturns the error instead of throwing, so the old code could not tell a column it could not widen from one that did not need it.Fix: a column definition separated by a tab rather than a space could never be matched or added. The name was taken with
split(" ")[0]and then had whitespace stripped, soseen_by\tTEXTproduced the nameseen_byTEXT. A definition with leading whitespace produced an empty name and was skipped in silence. Neither happens in a shipped definition file today, which is the only reason it went unnoticed; both are now parsed correctly and tested.Fix: a column definition with a capital letter would be re-added on every run, for ever. Postgres folds an unquoted identifier, so
MixedCase TEXTcreates a column namedmixedcase, and the exact-match existence check never found it again: every migration emittedADD COLUMN MixedCase, which failed, and the failure was one more error line in a long run. Column matching is now case-insensitive on both the add and the remove side.One-time database migrations are now recorded, so they run once and can be reasoned about. The declarative schema in
db/schemaconverges on a shape no matter what state it finds, which is exactly why it cannot express an event: "renumber every QWK conference once" is not a shape. Asking a converging system to perform an event had produced four different answers to the same question, one per case (schema sniffing for the menu layout, a marker row insettingsfor the MSGID split, nothing at all forqwkConfRenumber, and an unanswerable question forallow_netmail_relay). There is now aschema_migrationstable, adb/migrations/directory of orderedNNNN-slug.jsmodules, and a runner insrc/maintenance/migrationLedger.js. Once a migration is recorded it is never considered again.Migrations declare a phase,
preorpostrelative to the declarative schema pass. Theprephase is not a nicety: the menu layout migration has to readentries_x_offsetbefore that column is dropped, and running it afterwards would convert data that is already gone.A migration can be manual (
automatic: false), meaning it is reported with instructions and never run. That is how the MSGID tuple split is carried now: its predicate full scans every per-area message table, so a sysop chooses when to pay for it, andmaintenance.js --fix-msgidsrecords the completion in the ledger when it finishes. A system that ran the repair before the ledger existed is recognised by its oldsettingsmarker and recorded as not needing it, so nobody is told to redo expensive work they have already done.One row per attempt rather than one per migration, with no unique constraint on the id, following
mailer_sessions: a failed attempt is the history worth reading later.failedis deliberately not a terminal outcome, so fixing the cause and running again picks up exactly where it stopped, and everything that already succeeded is left alone.A fresh database records every migration as never needed instead of running it, since a database created by this release is born current and has no legacy data to repair. That is decided from whether the
userstable existed before the run, checked before anything creates tables. Getting it wrong in the other direction is the dangerous one: an existing install treated as fresh would skip every repair it actually needed, in silence.Migrations run from
dbmigrations.js, which is a CLI invoked during an update, and not at gateway startup. A migration that fails fails an update and cannot stop the BBS from booting. The run stops at the first failure rather than continuing, because a later migration may depend on an earlier one and carrying on is how a half-migrated database happens.The QWK conference renumber is now a migration rather than an upgrade step to remember. It was documented as "after pulling this release and before restarting, run
node src/tools/qwkConfRenumber.jsonce". Leaving it undone is not cosmetic: a group with noqwk_conf_startcannot have a new area created in it at all, becausenextConfNumberInGroupthrows and the sysop sees a failed area creation pointing at a tool they were supposed to have run. It now runs automatically as0003-qwk-conference-blocks, in default mode only.Default mode is what made that safe to automate: it preserves every conference block already assigned and fills only the gaps, and inside a block it keeps each area at the offset it already holds. An area that is already correct is never renumbered, because every renumber invalidates a conference list somebody is holding in a mail reader.
--force, which discards and repacks everything, stays a deliberate choice made at the command line and is never run by the migration.A group holding more areas than
qwk_group_strideallows now stops the update with the group named, rather than exiting 1 from a tool a sysop may not have been watching. That is a real decision for them (raise the stride or split the group) and half a renumber is worse than none.The planning logic moved out of the CLI into
src/maintenance/qwkConfRenumber.js, whereplanRenumber()is pure and has tests for the first time. The tool was a 404-line script that planned, printed and wrote in one pass and calledprocess.exitat every turn, so nothing else could use it and none of its rules were covered. Running the tool by hand still works and now records itself in the migration ledger, so a manual run and an automatic one leave the same trace.Fix:
setup/install.shupdate mode deleted every sysop-added asset underpublic/. It detected an existing installation and offered to update it, and its update path did a fullrm -rfreplace of each code directory withpublicstill in that list.setup/update.shhad movedpublicinto an overlay so door icons, collection images and custom branding survive an update, and added a skiplist sopublic/css/custom.cssis seeded once and never overwritten.install.shnever received either change, so running it against a live board destroyed all of it, and the script invited that by detecting the mode itself and offering. It now refuses and points atupdate.sh, saying what that script preserves rather than only that it is different. Issue #44.The cause was two programs claiming to update, with duplicated directory categorisation that drifted when only one of them was changed.
do_updateandmerge_data_dirare removed frominstall.shrather than brought back into line, so there is no second copy left to drift. Installing is all that script does now.
Testing
- 3,831 tests passing, several hundred of them added over this block. New coverage where it was load-bearing rather than decorative: the JAM and SMB parsers and adapters (including streaming equivalence and file descriptor leak checks on early consumer break), message retention, the import seal with an interleaved fixture, table template parsing, outbound eligibility, AreaFix end to end from netmail arrival to response body, the rescan grammar and pointer arithmetic, batch subscribe planning, group default pushes, moderation and the wire shape, the Via chain, relay decisions, worker health detection, flow file pruning, schedule registration against a real BullMQ queue, and every admin form having a submit control.
- Redis and BullMQ now have integration coverage. Nothing in the suite exercised either, so the ioredis, node-redis and bullmq major upgrades passed the whole suite while proving nothing about the code paths they changed. The load-bearing assertion is that hashes come back as plain objects, which is what a silent revert to RESP3 would break without failing a single unit test.
- The release machinery is tested by pulling each rule the other way and checking that the suite notices: reversing the manifest diff, letting an empty manifest parse, dropping the seed-once rule on
public/css/custom.css, sorting the release index oldest first, removing the delta-base agreement check, and letting a manual upgrade step ship with no command. Ten such inversions, ten failures. A rule about what may be deleted from somebody else's installation is worth more than a test that only passes. - The same treatment on the update check, where every rule is directional and every reversal returns a plausible answer: the upgrade floor compared the wrong way, the migration rule with its arguments swapped (which refuses every upgrade there is, because every release ships migrations the database has not run), the newest release picked by string order, a stable install told about development builds, and the notification dedupe keyed on version alone. One of those reversals passed the first time, which is how a coverage hole gets found: string ordering and version ordering only disagree from the tenth release onward, so
0.9.10aagainst0.9.9ais now asserted directly. - The updater's destructive rules get the same treatment, and here the reversals are the actual history of this project rather than hypotheticals: removal not limited to code directories (which is what deleted sysop assets under
public/),custom.csstreated as an ordinary overlay, art overwritten, a missing manifest assumed to mean untouched, the installation's manifest left stale after an apply, a delta not pruned so deletions never travel, and the checksum not verified before unpacking. Seven reversals, seven failures. - The handoff's rules get the same treatment: a run completing over a failed phase, a skip with no reason, a phase run twice, a staged tree inside a directory the update replaces, no check that the release ships an updater at all, the run state written only after the spawn, and reading the wrong tree's version. Nine reversals, nine failures.
- The FTN derivation rules were pulled the other way as well: an unidentifiable archive offered anyway, the locally computed hash presented as the published one, a wrong-channel artifact offered, the index ordered lexically, a missing checksum accepted without permission, and a corrupt sibling silently fallen back from. One of those passed first time and for the wrong reason, which is the more useful outcome: the unidentifiable-archive rule was only being caught by the channel filter that happened to run after it, so the check moved to the front where it belongs and is now asserted with no channel filter at all.
- The installer's constraint that it must work with no
node_modulesis checked by walking its import graph rather than by remembering it, and the six destructive install rules were pulled the other way to confirm the suite notices: the .env overwritten when one already exists, the .env left world readable, an existing installation clobbered, the runtime directories not created, a dependency added to the install path, and the art tree enumerated on only one axis. - The phase sequence is tested by injecting the commands, because without a way to make npm fail on demand there is no way to assert that a failure at the dependency install rolls the files back while one at the steps phase does not, and that rule is the whole reason the sequence has a middle. Nine more reversals checked: the rollback window extended past the migration, no rollback at all, the dump taken before the files rather than before the migration, manual steps run automatically, the board left down after a failure, a cached base trusted without verifying, the protected base evicted, unmeasurable free space treated as none, and connected callers blocking rather than warning.
- The update tool now has tests at the command line, not only under it. Two of the fixes above had no coverage because they lived in a CLI
main(): the version it reports and whether a named version is checked are both only observable by running the thing. Those tests publish miniature releases into a temp directory and spawn the tool against a temp installation, and they declare whatever migrations the local ledger happens to hold so that a developer's own database cannot decide whether they pass. - Two real bugs came out of writing these rather than out of review.
NaN || 0in the Node version comparison turned an unreadable version into zero, so a system whose Node version could not be parsed would have been told every release needed a newer one. And the update check importedappConfigas a namespace when it is a default-export singleton, which no unit test would have caught and which failed on the first end-to-end run. npm testnow skips cleanly when PostgreSQL or Redis is unavailable rather than failing 22 tests, so a clean checkout on a machine lacking either can still tell a real regression from a missing service. Both probes race against a bound and swallow the hang as well as the rejection, because each client retries internally and awaiting one in abeforehook turns a missing service into a hook timeout, which mocha reports as a failure rather than a skip.
Features
Message areas can now be moved between groups from the admin UI. The area info modal (popup when you click a row in
/admin/msgAreasor/admin/msgGroup/<shortcode>) gains a new "Move" button alongside View/Edit/Delete. Clicking it opens a confirmation modal showing the area, its current group, a target-group dropdown (filtered to exclude the current group), and an optional "Apply new group's default settings to this area" checkbox. Unchecked moves only change the group assignment. Checked moves also overwrite the area's security levels, age limits, permission/restriction flags, post-as, and newscan default with the target group's values. Message content, user read pointers, network mappings, peer subscriptions, and all statistics are preserved in either case. The QWK conference number is reassigned from the target group's block (lowest unused offset) so the area continues to cluster with its new groupmates in mail reader conference lists; if the target block is full the move aborts cleanly. Single-area moves only for now; bulk and drag-and-drop reorganization remain out of scope.BREAKING -- UPGRADE STEP REQUIRED: user QWK packets now reserve conference numbers 1-100 for Canectic system content, and message areas are numbered in per-group blocks so areas in the same group cluster together in QWK reader conference lists. Each
message_groupsrow owns a contiguous block ofqwk_group_stridenumbers (config, default 1000) starting at the newqwk_conf_startcolumn. Areas in the group take consecutive offsets within[qwk_conf_start, qwk_conf_start + stride - 1]. The first group's block starts at the first stride boundary above the reserved range (e.g. 1000 for stride=1000); subsequent groups stack atMAX(existing) + stride. Three system conferences ship in this release: announcements (default conf 1, read-only), wall (default conf 2, accepts one post per REP upload), and bulletins (default conf 3, read-only). Group create/edit modals expose a new "QWK Conference Start" field, pre-filled with the next available block on create. Changing the start on an existing group renumbers every area in the group, preserving each area's offset within the new block, with full collision checking against areas in other groups. The admin area editor still exposes a per-areaqwk_conf_numberoverride field as an escape hatch -- the override is allowed to live outside the group's block as long as it stays globally unique and above the reserved range. New areas auto-pick the lowest unused offset in their parent group's block; if the block is full, creation fails with a "block full" error and the sysop must increaseqwk_group_strideor move areas out of the group.moveAreareassigns the conference number from the target group's block before any other writes; if that block is full, the move aborts cleanly. Upgrade action: after pulling this release and before restarting the gateway/worker, runnode src/tools/qwkConfRenumber.jsonce against the live database. Default mode preserves any existingqwk_conf_startvalues and only fills in groups missing one, then walks each group's areas to assign offsets. Use--forceto discard everything and rebuild from scratch (recommended for installs that have not yet shipped any QWK packets to users). The tool ensures the schema column exists before reading, supports--dry-runfor preview, and uses a two-phase write (park at -id, then write final values) to dodge UNIQUE-constraint collisions when the new range overlaps the old. Any downloaded QWK packets held by testers will have stale conference numbers and should be re-downloaded after the tool runs. New config key:qwk_group_stride(default 1000). Larger stride leaves more room for future areas; smaller stride packs groups tighter. Changing this value does NOT renumber existing groups -- run the tool with--forceto re-pack. System conferences are emitted first in CONTROL.DAT so they appear at the top of mail-client conference lists. Content is filtered by the newusers.last_qwk_download_attimestamp (updated after each successful build) so each download only carries material new since the previous sync. REP imports targeting announcements/bulletins conferences are rejected; the wall accepts the first post and rejects extras withWall posts are limited to one per REP upload. Any rejections from a REP import generate a LOCAL system PM back to the user summarising the failures -- most mail clients don't surface per-message rejection details, so this is the only realistic feedback loop for offline-only users. Three new config keys (qwk_announcements_conf,qwk_wall_conf,qwk_bulletins_conf, defaults 1/2/3,0disables each). Wall post handling: first line only (multi-line bodies take the first line), whitespace stripped, silently truncated towall_max_length.wall.canPost()access control (Unverified restriction etc) applies. Seedocs/reference/QWK_USER_DESIGN.mdfor the full lifecycle.PM trash now uses a two-stage hard delete with a background cleanup worker so user actions stay non-blocking even on bulk purge. The lifecycle gains a third stage:
deleted=false/purged=false(live),deleted=true/purged=false(in trash, recoverable),purged=true(invisible, queued for cleanup, not recoverable). Deleting from Trash via the web button OR via IMAPSTORE \Deleted+EXPUNGEagainst the Trash mailbox tags the row and returns instantly -- a Thunderbird "Empty Trash" against hundreds of messages now responds in milliseconds instead of blocking on per-row file unlinks. The actual disk + DB work runs in a newpm-purged-cleanupBullMQ job (hourly schedule + on-demand enqueue after each user purge for sub-minute reclaim). The existing daily 03:00 retention sweep is repurposed as tag-only -- it flipspurged=trueon rows whosedeleted_atis pastpm_trash_retention_daysinstead of doing the cleanup itself, centralising all hard-delete work inexecutePurgedCleanup. Per-row attachment cleanup failures now retry on the next cycle instead of becoming permanent orphans, and the dynamic-import ofpmAttachments.jslogs at level 1 on failure instead of the silent catch the old purge had. Schema gainspurged BOOLEAN DEFAULT falseandpurged_at TIMESTAMPTZonpm_inboxandpm_outbox(auto-add migration handles rollout). Read-path filters added acrossgetInbox,getTrash,getMessage,getUnreadCount,getSentMessages,getSentMessage,restoreMessage,restoreSentMessage,imapAdapter.getMailboxMessages/getFullMessage/getMailboxStatus,qwkUserunread-PM scan, and the/pm/attachment/:idownership check, so tagged rows are invisible everywhere a user (web, IMAP, terminal QWK, attachment download) might encounter them. The web Inbox split-pane reader now shows both Restore and Delete buttons in trash mode, with the Delete button routing to the new/pm/trash/:uuid/deleteendpoint. Soft-delete from inbox/sent is unchanged.New sysop guide
docs/sysop/MAIL.mdwalks through the full mail subsystem end-to-end: outbound SMTP relay setup (with known-good configs for AWS SES, Gmail, Mailgun, local Postfix), system vs user sender identity, TLS port/secure gotchas and auto-reconciliation, inbound gateway with IPv4 CIDR allowlisting, upstream MTA integration (Postfix relay_domains, Proxmox MG), recipient verification probes (VRFY + MAIL FROM/RCPT TO), mail client access (IMAP + SMTP submission for Thunderbird/Apple Mail/K-9), user email verification flow, a troubleshooting section covering the common failure modes (wrong version number, cached recipient rejections, SPF/DKIM, auth failures), and a complete configuration key reference.INSTALL.mdandTEMPLATES.mdnow cross-link to it.Recipient resolver (
privateMessages.resolveRecipient) now matchesfirst.laststyle local-parts against the fullname column via a dot-to-space fallback. The existing exact-username and exact-fullname matches still fire first, and a dotted-username that also matches literally (e.g.dr.smith) isn't shadowed. This makes external mail addressed to<first.last@mail_domain>round-trip correctly through VRFY, RCPT TO, and every other path that callsresolveRecipient— matching theimapAdapter.networkEmailAddressconvention used on outbound so there's one consistent form across the whole mail network.Trusted inbound SMTP server now supports recipient address verification for upstream mail gateways: VRFY command is advertised in EHLO and accepts bare username,
user@domain,first.last,first.last@domain, or<user@domain>formats, returning250 <canonical>for known local users and550 5.1.1 User unknownfor unknown. MAIL FROM / RCPT TO verification probes (the "MAIL FROM: <> / RCPT TO:/ RSET" pattern used by Postfix reject_unverified_recipient) are now explicitly supported -- empty sender no longer triggers "503 Need MAIL FROM first", and every probe is logged at level 3 on the tosser log channel so sysops can watch verification activity in real time. Letting upstream relays verify addresses against Canectic means one source of truth for "does this recipient exist?" across the whole mail network.Message area create/edit modals redesigned with a tabbed layout (Main/Access/Network) via a shared Pug mixin (
_areaFormFields.pug) to reduce duplication and make field additions safer. Both the global areas index and the per-group view share the same partial.New "Poll a Node" control on the Jobs & Maintenance page (
/admin/events, Mailer tab) triggers an outbound mailer session on demand, so a poll no longer requires shell access tosrc/tools/trigger-poll.js. The field accepts either an FTN address (4D or 5D, e.g.1:134/101or21:1/100@fsxnet) or a peer connection UUID, with a dropdown pre-filled from your active FTN peers showing peer name, network, and a[HOLD]marker where applicable. A UUID polls that connection directly. An address is resolved throughftnRouter.resolveRoute(), the same cascade the BSO scanner uses, so it will call a peer connection when one matches and fall through to a nodelist-dialable node when there is no peer record at all -- handy for reaching a node you have not linked yet. Both paths bypass the poll schedule and the connect backoff, since a manual poll is explicit sysop intent. Resolution happens server-side and reports the route method and next hop (or hostname and port for a nodelist call) so an unexpected route is visible immediately; the poll itself is handed to the mailer process via the mailer queue, so the button returns as soon as the job is queued and session results appear in the mailer log. Requires the mailer process to be running, otherwise the job waits in the queue.
Bugfixes
- Fix: binkp sessions containing a file transfer were being logged as
failedby binkd peers, which then re-queued every file they had just delivered — a redelivery loop, despite the data arriving and tossing correctly on our side. Root cause was a missing binkp/1.1 feature, not a teardown race. A binkp/1.1 session is a sequence of batches: each ends with an M_EOB from both sides, and the session ends only on a batch that carried nothing but those two M_EOBs. binkd counts every command frame in either direction asmsgs_in_batch(data frames excluded, handshake excluded) and, when a batch closes with more than two, clears both EOB flags and opens another batch with a fresh M_EOB. Canectic treated the first mutual EOB as end-of-session and closed, leaving binkd withremote_EOB == 0at its post-loop check, so it loggedfailedand calledcreate_poll()to restore the poll. Empty sessions land on exactlymsgs_in_batch == 2and terminate after one exchange, which is why only sessions carrying a file ever failed and why the bug looked like a close-timing race._checkSessionEnd()insrc/mailer/protocols/binkp.jsnow mirrors binkd's accounting exactly: per-batch command-frame counter, both EOB flags cleared at a batch boundary, and the continuation M_EOB sent by the protocol itself. Verified against a binkd state-machine emulator transcribed fromprotocol.c— the same file session that producedfailedbefore the change now producesOK. Guarded for interop with mailers that don't implement 1.1 batching: the continuation is gated on the peer advertisingbinkp/1.1in itsM_NUL VERline (the identicalmajor*100+minor > 100test binkd applies), so binkp/1.0 and version-less peers keep byte-identical single-batch behaviour. A peer that closes after at least one completed EOB exchange with nothing pending now ends the sessioncompleterather thanerror— binkd does exactly that whenever a batch held only the two M_EOBs. A 10s idle guard ends the session cleanly if a 1.1-advertising peer neither opens a second batch nor closes, and an 8-batch ceiling backstops runaway loops. - Fix: Synchronet BinkIt peers were being sent
M_ERR "Forced Shutdown"at the end of every file session with us. BinkIt callsreset_eob()on each non-M_EOB frame it receives, so an M_GOT arriving after our M_EOB zeroed itsgoteobcounter and itsclose()took theif (!this.goteob)branch. Two changes fix it: our M_EOB is now held until every M_GOT owed for the batch has been written, and the binkp/1.1 continuation batch above ends on a bare M_EOB with nothing following it. BinkIt sessions now close gracefully with no M_ERR. - Fix: the mailer would deadlock against any peer that also waited for the remote M_EOB before sending its own (including another Canectic instance with nothing to exchange), holding the connection until the session timer fired. Our M_EOB is now sent as soon as the send queue drains, matching binkd, rather than waiting on the peer's.
- Fix: Bad echomail recovery ("Recover from bad echomail") was failing with
invalid input syntax for type integer: ""for every held message.recoverMessagesrebuilt the FTN message object without settingSource/Destinationaddress strings;echomail.ftnToDbthen evaluatedftnMessage.Source ? ftnMessage.Source.split(":")[0] : ""and inserted the empty string into thesource_zone/destination_zoneINTEGER columns. Fixed: recovery now restoresSourceandDestinationfrom stored addresses and parses Zone/Net/Node from them.ftnToDbhardened to coerce all address integer columns viatoIntOrNulland fall back to the integerSourceZone/DestinationZonefields when strings are absent. Schema gainsdestination_addressonechomail_badso future bad messages store the full destination for recovery fidelity. Rundbmigrations.jsafter update to add the column. - Fix: Outbound FTN echomail
@REPLYkludge was emitting only the parent's serial with no origin address, because the reply path from both the terminal msgEditor and the web flat/thread/admin reply modals was passing only the bare serial (msg_id) toreplyToThread.splitFullMsgidon a bare serial correctly produced{msg_orig_addr: '', msg_id: <serial>}, which storedreply_orig_addras empty; the outbounddbToFtnthen joined empty-addr + serial back to just the serial for the kludge. Reworked the contract: callers now passformObj.reply_to_uuid(the parent message's DBuuid) andreplyToThreadlooks upmsg_id/msg_orig_addrfrom the authoritative row in the area's msgs table, eliminating the client-assembled tuple string and the trust on whatever the form posted. Added a fallback to the thread root'sthreadid(full "ADDR SERIAL") when no parent uuid is provided (e.g. QWK REP imports that only match by subject). Web views (messages/areas/flat.pug,thread.pug,admin/msgAreas/thread.pug) updated: Reply buttons carrydata-bs-msguuidand the hidden input isreply_to_uuid. The terminal msgEditor passesthis._replyToMsg.uuid.getMessages/getMessagesForThread/getMessagesForAreaSELECTs includemsg_orig_addr+reply_orig_addrfor display consistency.@REPLYkludges on new echomail/netmail replies are now FTS-0009 compliant. - Fix: outbound netmail addressed through the 5D form of the SMTP/IMAP pseudo-TLD leaked the FTN domain into the
INTLkludge. Sending touser@21.1.100.fsxnet.ftnstoresto_addressas21:1/100@fsxnet, which flowed unmodified intomsg.Destinationand then intoINTL 21:1/100@fsxnet <source>, where FTS-0001 specifies barezone:net/node. The damage was not limited to the kludge text:WritePktsparses theINTLvalue back apart to build its packet grouping keys and overwritesmessage.Source/message.Destinationwith what it finds, so a domain-bearing address also split packet groups and wrote the 5D string back into the message's own address fields. NewbareFtnAddress()helper insrc/middleware/tossers/ftnTosser.jsstrips the@domainsuffix (preserving the point) and is applied to both halves of theINTLvalue at construction. 4D addressing was never affected. Note that network selection on outbound is still driven by the destination zone viasystemAddresses.resolveSourceForDestinationZone(), not by the domain label, so the label remains informational for routing purposes. - Fix: AreaFix and FileFix
+TAG/-TAG/%COMMANDbody parsing was splitting on/\r?\n/, which never matches in FTN messages. FTN message bodies use bare\ras the line delimiter (set inftnTosser.processMessageTextasbodyLines.join('\r')), so the entire body collapsed into a single line. The tear-linestartsWith('---')guard never fired, the+branch swallowed everything after the tag (kludges, tear line, origin) into the areatag string, the SQLUPPER(na.areatag) = $2lookup missed, and the requester sawTAG\r--- TearLine: not found on this networkrendered as a visually-broken two-line response while no row was ever written topeer_area_subscriptions. Split now uses/\r\n|\r|\n/in bothsrc/models/areafix.jsandsrc/models/filefix.js. - Fix: Nodelist auto-scan was hardcoded to only match files literally starting with
NODELIST., silently skipping every network that uses a different naming convention (AgoraNet shipsagoranet.z00, FsxNetFSXNET.Z00, MicronetMICRONET.Z00, etc.). Additionally, the day-number gating (num > currentNumber) permanently rejected files using the.Z00placeholder suffix (extracted day=0, gate0 > 0= false). Rewritten: Pass 1 prefersNODELIST.*filenames; Pass 2 falls back to the newest non-junk file in the configured nodelist area. Both passes gate onupload_date > MAX(last_seen)from theftn_nodelisttable, handling year wraparound and placeholder extensions. Junk filter (FILE_ID.DIZ, readmes, .nfo, .txt, etc.) prevents misc files from being fed to the nodelist parser.extractNodelistFromZipapplies the same junk filter when selecting the fallback file inside an infopack ZIP. - Fix: Echolist/filebone auto-scan reported "0 list(s) updated" and silently re-parsed unchanged archives on every scheduled run because
_importFilereturned{unchanged: true}before callingexecuteImport, leavinglast_imported_atuntouched. AddedtouchLastImported()helper onnetworkEcholistandfileEcholistmodels that bumps the timestamp on no-change scans so the next run correctly skips the file via_isNewer. Scanner summary now reports "X updated, Y verified (no changes)" so sysops see that work happened even when nothing changed. - Fix: Echolist scanner second-pass (unzip archive + match configured names) had silent fallthroughs for several failure modes. Added L2 diagnostic logging when: an archive is in
file_entriesbut missing on disk (with resolved path), no archive is found at all, and most importantly when the unzip succeeds but no extracted file matches any configured echolist/filebone name (logs extracted contents and configured names side by side so the sysop can see exactly what's inside vs. what's expected). - Feat: MSGID counter now lives in a dedicated
msgid_day_watermarkPostgres table, one row per day-bucket, replacing the earlier Redis-backed counter. Every serial allocation is a single atomicINSERT ... ON CONFLICT ... DO UPDATE SET max_counter = max_counter + 1 RETURNINGupsert, with Postgres row-level locking handling cross-process coordination natively. No more Redis TTL management, no more startup re-scan via outbound table enumeration, no more dual source of truth. On first deployment the table is auto-backfilled from existing current-format msg_ids (grouped by day prefix, max counter per day) so the watermark reflects real history immediately. Legacy 11+ char serials from pre-fix installs are ignored during backfill. NewallocateHistoricalSerial(date, sourceTag)helper lets import tools allocate MSGIDs at the correct chronological day for historical message bases without conflicting with live generation.getWatermarkForDay(day)returns the current max counter, message count, and source tag for diagnostics.sourcecolumn distinguishes'live'/'backfill'/'import_<label>'provenance. The watermark table appears indbmigrations.jsas an auto-created table viatablenames.conf. - Feat: FTN MSGID system rebuilt around a Canectic epoch (December 8, 1980 UTC, deliberately pre-FidoNet for historical import support) and a tuple storage model. The 32-bit FTS-0009 serial now encodes
days_since_epochin the high 16 bits and a shared Redis-backed per-day counter in the low 16 bits, giving a 179-year horizon (wraps in 2160) and 65,536 MSGIDs per day aggregated across all four Canectic processes. On process startup,initializeMsgidCounter()seeds the Redis counter from a DB scan of today's highest serial (+100 safety margin) so cold starts, Redis flushes, and fresh installs can't regenerate into used values.msg_idstorage is split into(msg_id, msg_orig_addr)tuple columns so cross-network dupe detection, import-vs-generated collision safety, and cross-post discovery can all be reasoned about via the separate fields.reply_togets the same split for threading consistency.threadidstays as a full-string opaque key. The tosser/compose/import paths all update their dupe checks and reply_to lookups to compare tuples. Legacy"address serial"rows are backfilled via an idempotent migration indbmigrations.js. NewsplitFullMsgid/joinMsgidTuple/parseSerial/dayBucketForDatehelpers expose the format for reuse. Full design is documented indocs/reference/NETWORKING_ARCHITECTURE.mdanddocs/reference/CTN_DESIGN.md. - Fix: FTN MSGID kludge serial number was violating FTS-0009 by emitting 11-character hex strings instead of the required 8.
genMsgId()andgetMilliHex()insrc/utils/ftnUtils.jswere hex-encodingDate.now()directly, and thepadStart(8, '0')only padded up -- it never truncated, so every MSGID ever emitted by Canectic after 2004 (when epoch milliseconds exceeded 32 bits) has been non-compliant. Fixed to pack 16 bits of wall-clock seconds with a 16-bit per-process monotonic counter into a single 32-bit value, hex-encoded to exactly 8 characters. Collision requires both halves to wrap simultaneously (~18 hours × 65536 calls per second). Tests rewritten from "at least 8 chars" to "EXACTLY 8 chars" so the bug can't come back silently. Impact: outbound FTN echomail and netmail now produce^AMSGID: 1:2/3 abcdef01instead of^AMSGID: 1:2/3 19d79e4bf38. Existing stored MSGIDs on inbound messages (still 11 chars) are unaffected because they're treated as opaque strings for dupe-checking; only what we emit changes. - Fix: PM2 child processes and the alogger timestamp formatter now honour a single authoritative timezone.
ecosystem.config.cjsloads.envat config-eval time, resolves an IANA zone fromprocess.env.TZ(falling back toIntl.DateTimeFormat().resolvedOptions().timeZone), and spreads it into every app's env block so alogger's prettytime output uses local time even on minimal server images where/etc/localtimeisn't configured. Also fixes a latent bug inalogger.writeClogwhere it was calling.toLocaleString(DateTime.DATETIME_MED)on a native JSDate— the Luxon preset was silently ignored, falling back to default locale format. Switched toLuxon.DateTime.now()so the preset actually applies. NOTE: pm2's ownlog_date_formatprefix on captured stdout is stamped by the pm2 daemon, not the child — if your pm2 logs still show UTC aftergit pull && pm2 restart all, restart the daemon itself withpm2 kill && TZ=<zone> pm2 start ecosystem.config.cjsto propagate the zone to the daemon.examples/env.exampledocuments this inline. - Fix: Email tosser was querying a non-existent
realnamecolumn when looking up the sender's display name for outbound PMs (the actual column isusers.fullname). The try/catch swallowed the error so mail still delivered with the fallbackpm.from_userdisplay name, but every outbound cycle loggedDB query error: column "realname" does not existand user-generated emails ended up with the username instead of the user's full display name. Query fixed, display name now reflects the realusers.fullname. - Fix: Email tosser was deriving the outbound email local-part from
pm.from_user, which is NOT guaranteed to be a username — the per-grouppost_assetting allows users to post under either their username OR their fullname, andvalidateFromUserin the compose routes accepts either. For a user posting under their real name the local-part was being derived from the (potentially non-unique) fullname. Address now always comes fromusers.username(looked up bypm.user_id, the guaranteed-unique field) while the display name continues to preferusers.fullname. The identity-picking logic is factored into a pure helper_pickSenderIdentityso the authority rules are pinned by unit tests and one source of truth. - Fix: Inbound SMTP server was rejecting verification probes from upstream mail gateways that use the "MAIL FROM: <> / RCPT TO:
/ RSET" pattern. The _cmdRcptTohandler did a truthy check (if (!this.mailFrom)) which treated an empty-string sender (legitimate for bounces and probes) the same as a missing sender. Changed to a strict null check so empty senders are allowed to proceed. - Fix: EHLO capability advertisement was leaking inappropriate features between modes. Submission mode (authenticated) previously exposed VRFY; trusted mode (IP-allowlisted) previously exposed AUTH. Now submission mode advertises
AUTH PLAIN LOGIN, trusted mode advertisesVRFY, and both advertisePIPELINING,8BITMIME, andSIZE. Trusted-mode EHLO no longer looks like a submission server to confused probes. - Fix: RCPT TO responses now include RFC 3463 enhanced status codes (
250 2.1.5for accepted,550 5.1.1 <addr> Recipient address unknownfor unknown) so downstream relays cache the correct verdict and log useful reasons. - Fix:
email_trusted_hostsnow accepts IPv4 CIDR blocks (e.g.10.0.0.0/8,192.168.1.0/24) in addition to bare IP addresses. Mixed lists work:127.0.0.1, 10.0.0.0/8, 192.168.1.5. Entries with non-zero host bits (e.g.10.0.0.1/8) are normalised automatically. Invalid CIDR entries (bad prefix, non-IP base) log a warning and are skipped so a typo in one entry doesn't kill the whole allowlist. IPv6 still requires exact addresses. Inbound-SMTP rejection log is now level 2 (visible in default logs) with the full allowlist and a pointer to the config field, and the 554 response includes the offending IP in the 5.7.1 reply so remote postfix logs show "access denied" instead of an opaque "connection refused". - Fix: Email tosser was using the configured
email_from_address/email_from_namefor every outbound message including user-to-user PMs. Those should only be used for system-generated email (verification, password reset, template notifications, sysop broadcasts). User-generated PMs now use<username>@<mail_domain>as the from address with the sender's realname (falling back to username) for the display name. Reply-To is set to the same value so external recipients can reply directly to the BBS user. System detection checkstransport_data.system_notification === true,from_user === 'SYSTEM', oruser_idnull/0 (the existing pattern used bymailVerification.jsand friends). Ifmail_domainisn't configured we fall back to the system from-address with a warning so the tosser still makes forward progress. - Fix: Email tosser now auto-reconciles
email_relay_portagainstemail_relay_secureto prevent the classic OpenSSL "wrong version number" handshake error. Port 465 expects implicit TLS (secure=true); ports 587 and 25 expect STARTTLS (secure=false). Mismatched combinations are logged as a warning and overridden for the current cycle so a single misconfigured field doesn't spin the retry queue forever. Non-standard ports are left alone, trusting the sysop's explicit intent. Also setsrequireTLS: trueon port 587 so nodemailer refuses to fall back to plaintext if the relay stops advertising STARTTLS. TLS mismatch errors now log a diagnostic hint pointing at the config fields, and the cached transporter is force-rebuilt on error so post-fix config changes take effect on the next cycle without a worker restart. - Fix: FTN tosser kludge lines were missing the space after the colon (e.g.
^AMSGID:1:2/3 abcdef01instead of the standard^AMSGID: 1:2/3 abcdef01). Standard FTS-0004 colon-space separator is now inserted for all kludges except INTL/TOPT/FMPT/Via, which correctly use the bare-space separator per FTS-0001. NewformatKludge()helper centralises the rule and has regression tests. - Fix: FTN tosser no longer unzips non-mail-bundle ZIP files in the inbound directory. Previously it detected ZIPs by magic bytes and extracted anything it found, which destroyed TIC-attached ZIPs (and nodelist/infopack bundles) before the TIC tosser could process them. The extractor now gates on the FTN BSO arcmail bundle filename pattern (
.mo0,.tu5,.su9, etc. per FTS-5000) and leaves all other archives untouched for their respective processors. - Fix: FTN tosser bundle extraction switched from
pexec unzipto adm-zip in-process. No shell-injection surface from odd filenames, deterministic and testable behaviour, faithful extraction of every entry inside a legitimate mail bundle (no content filtering -- stray files inside a bundle are still extracted so nothing gets silently lost). Archive decisions and per-bundle extraction counts are now logged via alogger on thetosserchannel so sysops can verify the filename gate is active at runtime. - Fix: Notification dropdown populated with recent unread notifications on connect (#1)
- Fix: Sysop notifications create one row per sysop (not shared broadcast rows) for correct per-user read state (#1)
- Fix: Menu JSON import NaN error from null colour values (#2)
- Fix: FTN NC addresses (node 0, e.g. 1:134/0) no longer rejected as invalid (#3)
- Fix: Message group creation security field handling hardened (#4)
- Fix: Message group create modal had single "Required Security" field instead of separate Read/Post Security + flags (#4)
- Fix: Duplicate admin view template files (flat file shadowing index.pug) consolidated
- Fix: Removing a network address crashed with "Cannot read properties of null" due to
postgres.write()return value mismatch (#7) - Fix: Missing
aloggerimport in menus.js caused "alogger is not defined" error on menu import/duplicate - Fix: Call flow completion loop when redirected from /callflow (return URL now sanitized to avoid redirect back to callflow)
- Fix: User-to-user telegram chat close button not appearing (CSS class specificity overriding inline style)
- Fix: Nodelist import crash on duplicate addresses in same batch (e.g. ZC and RC both listed as zone:net/0)
- Fix: SMB import crash on missing CP437 export name (convertCP437toUTF8 vs cp437ToUtf8)
- Fix: SMB import empty string for integer columns (source/destination zone/net/node now null when no FTN address)
- Fix: SMB import null bytes in text fields causing PostgreSQL rejection (0x00 stripped from all string fields)
- Fix: SMB parser header layout corrected (dataOffset at 64, totalDfields at 68, dfields-before-hfields order)
- Fix: SMB parser encoding changed from binary (Latin-1) to utf8
- Fix: Message table seen_by/path columns widened from VARCHAR(255) to TEXT (FTN paths can be thousands of chars)
- Fix: Callflow completion not updating session (session_index missing for pre-index sessions, added direct session ID fallback)
- Fix: Callflow force redirect restricted to GET only and excludes admin/pm/notification routes
- Fix: Callflow pending flag cleared on page render to break redirect loops
- Fix: SMB index record size corrected (20 bytes, not 28) -- was only finding 18 of 125 messages
- Fix: SMB import threading: flush thread stats every 10 messages during import for accurate subject matching
- Fix: Threading fallback: when reply parent is missing, falls back to subject matching instead of orphaning
- Fix: Threading window widened to 365 days for messages with "Re:" subject (was 30 days, missed long-running threads)
- Fix: Outbound FTN packets missing REPLY kludge -- replies sent from Canectic had no @REPLY, breaking threading on downstream systems
- Fix: Maintenance jobs blocked re-runs after completion (BullMQ jobId deduplication, added removeOnComplete/removeOnFail)
- Fix: Echolist/filebone import not back-filling local_area_id for pre-existing network area mappings
- Fix: Echolist reconciliation SQL using invalid table alias in UPDATE FROM clause
- Fix: Combined echolist/filebone view page scroll caused by modals inside main-wrapper (moved to page level)
- Fix: Filebone upload form field name mismatch with multer (filebone_file vs echolist_file)
- Fix: Filebone view/index routes matching :networkId before view/:id (Express route ordering)
- Fix: Filebone templates using renamed variable names (filebone/filebones) but routes passing echolist/echolists
- Fix: Area list parser not handling % comment lines or SBBSecho "Area AREATAG flags description" filebone format
- Fix: Cron schedule patterns had minute/hour fields swapped (hour 30 invalid)
- Fix: File group/area ~ path expansion (Node.js fs treats ~ literally, not as home directory)
- Fix: Echolist/filebone duplicate creation (same name on same network now re-uses existing entry)
- Fix: Thread view was advancing the area-level scan pointer, marking all older threads as read (#30).
advanceThreadPointernow only advances the per-thread pointer. Floor consolidation after each thread view detects when the area pointer can safely advance (all messages below a boundary are covered by thread pointers) and prunes redundantuser_thread_pointersrows. - Fix: New user registration logged two calls -- the registration handler logged a call but didn't set the
web_activityRedis key, so the webActivity middleware logged a second call on the redirect - Fix: Mail verification didn't update the active session, so the new security level and cleared restriction flags only took effect after re-login. Session is now synced immediately after verification (security, permission flags, restriction flags)
- Fix: Hard-delete from PM trash now requires confirmation. The web inbox reader's Delete button in trash mode went straight to a permanent, irreversible delete with no prompt. Added a Bootstrap confirmation modal ("Permanently Delete?") that must be acknowledged before the delete form is submitted. Soft-deletes from inbox/sent (which move to trash and are recoverable) still submit without a prompt.
- Fix: File wildcard rules weren't being applied to inbound files after admin UI changes.
fileWildcardscached the ruleset per process, but admin CRUD runs in the gateway while live matching runs in the mailer's TransferEngine -- the mailer's cache never saw the new rules, so everymatchFilenamereturned null and files fell through to the default inbound dir. Wired cross-process invalidation via a newWILDCARDS_CHANGEDRedis pub/sub event: CRUD publishes, mailer/worker/gateway subscribers call localinvalidateCache()on receipt. Same fix applied to the new netmail-routes model (NETMAIL_ROUTES_CHANGEDevent) so the admin UI's/admin/netmail-routeschanges propagate to the worker's tosser. - Fix: Timed events of type
Dosemuweren't actually running the batch -- they exited in under 2 seconds with code 0 and no side effects. The runner built the command stringdosemu -dumb -E c:\bat\something.batand passed it toexec()withshell:true, which runs throughbash -c. Bash strips backslashes in unquoted arguments (\b,\aetc. are shell escapes), soc:\bat\something.batarrived at DOSemu asc:batsomethingbat. DOSemu couldn't find the file and exited silently. Fix is to double-quote the batch path so backslashes survive:dosemu -dumb -E "c:\bat\something.bat". Embedded quotes in the command are escaped defensively. - Fix: NativeShell door launches failed with
execvp(3) failed.: No such file or directorywhenever the configured Command included arguments (e.g../USURPER /N1).pty.spawn()takes a literal executable path as its first arg -- it doesn't parse the string as a shell command, so the entire string./USURPER /N1was tried as a filename. NativeShell now runs the command through/bin/sh -cso argument parsing, quoting, and env var refs work the same way they do for NativeCommand. Admin-configureddoorArgs(separate list field) are POSIX-quoted and appended to the shell command line so args containing spaces survive too. - Fix: Native PTY doors (NativeShell launch type) could fail at spawn with a cryptic
chdir(2) failed: No such file or directorywhen the door's Door Directory was configured with a leading~/. Thesubstitute()helper inconfigvars.jshad a fast-path early return for templates with no%Xtokens, which bypassed the tilde-expansion at the end. Paths using%D(which forced the slow path) expanded correctly, so dropfiles wrote to/home/user/doors/mygame/while the pty spawn tried tochdirinto~/doors/mygameverbatim and failed. Fast path now handles~/expansion too. Dosemu / Dosemu-PTY / NativeCommand paths get the same fix for free since they all route through the same helper. - Fix:
setup/update.shwas wiping sysop-added content underpublic/. The script treatedpublicas a code directory (rm -rfthencp -afrom source) on every update, so any custom subdirectories likepublic/images/doors/,public/images/collections/, or custom branding files were deleted. Split the update model into three categories: CODE_DIRS (full overwrite, as before, forbin/src/db/docs/tests/examples/setup), OVERLAY_DIRS (shipped files overwrite target, target-only files preserved, used forpublic), and DATA_DIRS (merge with per-file prompt, as before).publicnow walks source files and copies them in place without touching anything the sysop added. If you lost custom images in a previous update, restore them from backups and they'll survive future updates. - Fix: Saving a door via admin edit reset its position in every collection it belonged to. The update route did remove-all-collection-memberships followed by re-add with default sort_order=0, which wiped any drag-to-reorder positions the sysop had set on the collection view. The update flow now diffs selected collections against current memberships: unchanged entries keep their existing
sort_order, newly-selected collections get appended to the end (max(sort_order) + 1) instead of the top, and only deselected collections are removed.doors/newanddoors/:id/duplicatealso now append-on-add so new or duplicated doors don't leapfrog existing ordering. - Fix: New door creation via admin UI silently skipped collection assignment. The route checked
result.insertIdbut the model returnsresult.doorId, so the collection-sync block never ran -- a sysop adding a door with collections selected at creation time would see the door appear with no collections, and have to edit and re-save to attach them. Fixed to use the correct field. Editing existing doors was unaffected. - Fix: System notification PMs (undeliverable netmail alerts, AreaFix/FileFix activity, etc.) didn't appear in IMAP Trash.
notifySysops/template/bounce paths calleddeliverMessagedirectly without supplying amsg_id, leavingpm_inbox.msg_idNULL. Some IMAP clients filter or hide messages without a meaningfulMessage-IDheader (or with collisions resulting from synthetic IDs), so the rows showed in web/terminal but not via Thunderbird et al.deliverMessagenow generates a UUID fallback formsg_idwhen the caller doesn't provide one, matching the existing conventionqueueMessageuses for user-to-user PMs. Everypm_inboxrow now has a non-nullmsg_idand renders consistently across all clients. - Fix: binkp handshake was silently rejecting peers whose mailer NUL-terminates command bodies (Argus observed, likely others). The M_PWD handler was
.trim()ing the received password, but JS.trim()strips whitespace only -- not NUL bytes. A peer sending"115898\0"was compared against a stored"115898"and failed the equality check. Per FTS-1026, command bodies MAY be NUL-terminated; trailing NULs are now stripped at payload extraction so the fix covers M_PWD, M_ADR, M_NUL and every other command consistently. Also added a level-5 auth-detail log line on mismatch showing matched peer, mode (plain/CRAM-MD5), and byte lengths so the next peer-auth bug is easier to diagnose.
Features
- Sysop-managed custom CSS: new
public/css/custom.cssfile, loaded last in every layout (user, admin, public, callflow) so it overrides Canectic's shipped styles, Bootstrap, Darkly, and DataTables. Ships with a self-documenting header on fresh installs.setup/update.shhas a newOVERLAY_SKIPlist that seeds the file on first install but never overwrites it on subsequent updates -- sysop branding/customisations survive indefinitely. Put any installation-specific tweaks (nav colours, font overrides, per-site palette) here instead of editingstyles.cssdirectly. - Door collections admin: drag-and-drop reordering in the collections list. Same pattern the doors-within-a-collection view and the file/message group lists already use -- drag handle icon in the leftmost column, drop indicator border while dragging,
Save Orderbutton appears once you've moved something, saves via a JSON POST to/admin/doorCollections/reorder. Affects the order collections appear in both the admin list and the user-facing/doorspage. - Door collection cards on
/doorsnow use a hero-image layout matching the door cards on/doors/collection/:slug: big icon up top (up to 140px tall) on a dark panel, title/description/game-count below. Replaces the old cramped 64x64 icon-beside-text layout so sysop-supplied collection icons actually get visual space. Also cleaned up a nested-anchor (invalid HTML) where the inner "Browse Collection" link sat inside the already-anchored card. - Door collections admin: description field gains TinyMCE editor with the same ANSI colour palette as doors, so collection listings can be styled with colours/formatting to match the rest of the door UI. Both the New and Edit modals use the rich editor (lazy-init on modal show so hidden textareas don't break TinyMCE rendering). Existing plain-text descriptions continue to work unchanged. Collection description is now rendered unescaped (
!=) on both admin views and the user-facing/doors+/doors/collection/:slugpages. - Doors admin: each row gets a Duplicate button (between Edit and Delete) that creates an inactive copy with a unique slug (
<slug>-copy,<slug>-copy-2, ...) and "(Copy)" / "(Copy N)" name. All settings carry over (cmd, dirs, args, launch type, dropfile config, real flags, max nodes, descriptions, icon, sort order). Collection memberships are copied too so the duplicate appears in the same browse categories. Copy lands inactive so the sysop can review/adjust before exposing it. Slug normalization strips an existing-copy[-N]suffix so duplicating a copy doesn't accumulate-copy-copy-copy. - Doors admin: live filter input above the table. Search matches name / slug / cmd / script / launch type as you type, with a row count showing "X of Y shown". Pure client-side -- no server round trip per keystroke.
- Mailer connect backoff: repeated outbound connect failures to a peer no longer hammer the remote every cycle. After each failure (TCP error, auth failure, unexpected disconnect), Canectic doubles the retry window from the peer's base
poll_interval, with ±10% jitter, capped atmailer_backoff_max_hours(default 24h). A successful session (state=complete) clears the counter immediately. Scheduled polls skip peers currently in backoff; admin manual "trigger poll" bypasses the guard so a sysop can force a retry. Per-peer state lives inconnection_dataJSONB (consecutive_failures,last_failure_at,last_failure_reason,next_allowed_poll_at,last_success_at) so restarts don't forget where we were. Admin peer detail page shows a Backoff row with failure count, next-allowed timestamp, last failure reason, and a Reset button. Config keys:mailer_backoff_enabled(default true),mailer_backoff_max_hours(default 24). Prevents a misconfigured or offline peer from generating unbounded failed-poll log noise and wasted cycles. - Per-peer "force plaintext password" toggle in the FTN peer connection admin form. When enabled, Canectic sends plaintext
M_PWDto the remote even if the answerer offered CRAM-MD5. Escape hatch for peers whose CRAM-MD5 verifier rejects valid digests (observed with some Argus configurations) or whose passwords are asymmetric in ways Canectic can't replicate. Plaintext is less secure than CRAM-MD5 so only enable per-peer when needed. A level-5 diagnostic log line now also prints the full CRAM-MD5 challenge, password length, and computed digest on outbound sessions so sysops can independently verify HMAC output viaopenssl dgst -md5 -mac HMACwhen chasing mismatches. - Custom netmail routes: sysop-configurable rules that match inbound FTN netmail by
to_user(required, exact or glob, case-insensitive) with optionalfrom_user/from_addrnarrowing filters, and apply one of three actions: alias (rewriteto_userand fall through to normal dispatch, useful for aliasingareamanagertoareafixor custom handles to real users), dump (write the netmail as JSON to a target directory for an external handler, with%Xsubstitution and relative paths anchored underfilebox_base_dir-- one file per message, attached files left alone for wildcard/TIC handling), ignore (drop the netmail silently, no PM delivery, no undeliverable queue, no response). Intended for IBBS game noise likeBRE Systemwhere the netmail pointer is useless but an attached.REP/.ZIPshould still be picked up by file wildcards. Rules evaluated insort_order, first-match-wins, cached in-process and invalidated on CRUD. Per-rulelog_leveldefaults to 0 (silent) so ignore rules don't re-introduce the noise they're meant to cut. Admin UI at/admin/netmail-routeswith drag-to-reorder, edit modal, and a dry-run test form. Sysop guide:docs/sysop/NETMAIL_ROUTES.md. - User profile: first/last name fields are now read-only (sysop change only). Users are directed to contact the sysop to change their name.
- Group and area ordering: drag-to-reorder on message groups, message areas, file groups, file areas with Save Order button (#6)
- Mocha upgraded to 11.3.0 (fixes diff DoS vulnerability) (#5)
- sbbsecho.ini import: two-step migration tool to import linked FTN nodes from Synchronet, with network/address resolution, drag-and-drop peer grouping, and per-connection network selection
- Migrations hub: admin page consolidating all import and export tools at /admin/migrations
- Network echolist registry: per-network echolist management with import, diff preview, historical tracking, local-carried status, sysop notifications on area removals, multiple echolists per network with default flag
- Filebone registry: file echo list management (mirrors echolist system) with import, diff, historical tracking, sysop notifications on removals
- Admin dashboard redesign: tabbed layout with icon cards and instant search, System tab as default, all tools accessible
- Admin navbar reorganised: dedicated Networking dropdown (Networks, Peers, Nodelists, Echolists, Filebones, FREQ, Migrations), Config removed from nav
- Legacy admin netmail routes and views removed (netmail flows through PM system)
- User navbar: added Chat (links to who's online/telegrams) and Wall links
- PostgreSQL setup: install guide now includes manual DB/user creation walkthrough with PG 15+ schema grant instructions
- Admin message/thread deletion: per-message delete buttons in thread view, per-thread delete buttons in area view, Delete Thread in sidebar with confirmation modals
- Thread rebuild maintenance tool:
--rethreadCLI flag, admin Rebuild Threads button, re-processes all messages in chronological order - Queue status panel on Jobs & Maintenance page: view active/waiting/delayed/failed jobs, remove stuck jobs, retry/remove failed, drain queues
- Combined echolist view: tabbed (All/Carried/Available/Orphaned) with counts on tabs, subscribe and unsubscribe/unmap actions, DataTables search on All tab
- Combined filebone view: same tabbed layout with browse/download/upload security fields and flag selectors
- Echolist/filebone subscribe modal: config-based security defaults, permission/restriction flag selectors, group auto-populate, separate Allow/Enable peer checkboxes
- Echolist/filebone unmap modal: cascading delete options (remove area, optionally delete tables/files)
- Echolist/filebone create from file upload: name derived from filename, import in one step
- Echolist/filebone local_area_id reconciliation: daily maintenance job + admin button, bidirectional sync
- Select All / Deselect All buttons on peer connection subscription modals (message and file, allowed and enabled)
- Message group post_as setting (username/realname) with push defaults to areas
- Configurable default security levels: config keys for message read/post and file browse/download/upload defaults
- Door collection membership: Collections tab on New Door and Edit Door modals with checkbox selection
- File group directory check/create: Check button on base_dir field shows whether directory exists, offers inline Create button
- Filebone subscribe modal: sub directory field for file area disk location
- Delete confirmation modals on all-echolists and all-filebones listing pages (replaces alert confirms)
- Standalone update.sh script: copies code, merges data, runs npm install and db migrations automatically
- 57 new unit tests for sbbsecho parser (2,016 total)
- Per-peer outbound fileboxes: optional
outbox_diron FTN peer connections, BSO scanner picks up files and sends with delete-after-send, admin UI field on connection forms - Inbound file wildcards: sysop-configurable glob pattern rules route inbound files to specific directories, skip toss by default with still_toss override, admin CRUD at /admin/wildcards with drag-to-reorder and test tool
- %X variable substitution in filebox paths, wildcard target directories, file group base_dir, and file area sub_dir (e.g. %M for DOSemu drive_c)
filebox_base_dirconfig key for per-peer filebox parent directory- Inbound Wildcards link added to admin Networking dropdown
- Automated echolist/filebone scanning: daily maintenance job with four-tier priority: per-list scan_area, network echolist_area, network filebone_area, network infopack_area. Each tier checks for direct file match then newest ZIP. Auto-imports via diff pipeline with sysop notifications on locally-carried removals.
- Network file area config: echolist_area, filebone_area, and infopack_area fields on network edit form for configuring where network list files and infopacks are received
- Per-echolist/filebone scan_area: optional file area override for individual list scanning (highest priority)
- Scan Echolists/Filebones button on Jobs & Maintenance page for manual trigger
- Echolist reconciliation schedule moved to 4:30 AM (after echolist scan at 4:15 AM)
- File area retention limits: optional max_age_days and max_files on file areas and file groups (group defaults inherited by new areas), enforced by daily file-maintenance job (removes oldest files from DB and disk). Useful for nodelist/nodediff/infopack areas. 0 or blank clears the limit.
- Fix: fileAreas.newArea not returning insertId -- file areas created from filebone combined subscribe were not getting their network mapping created
- Fix: Binkp session race -- TransferEngine was created after async multi-address discovery, so M_FILE offers from fast remotes (e.g. binkd) during the await were lost, causing 60s timeout
- Fix: Peer polling with no remote_host now falls back to nodelist lookup for connection info
- Fix: FTN address matching normalizes .0 point suffix (169:1/107.0 matches 169:1/107) for mailers like Argus that advertise point 0 explicitly
- Fix: New message area with network mapping now sets allowed=true for peers (was leaving both allowed and enabled as false)
- Fix: Combined echolist/filebone subscribe allow/enable peer checkboxes -- both tabs had same-named checkboxes in one form, Express received array instead of string, sync was silently skipped
- Fix: FTN outbound packet buffer overflow on messages with long SEEN-BY/PATH lines (fixed 10KB buffer replaced with dynamic sizing)
- Fix: FTN outbound zone awareness -- packets and .flo files were always written to ftn/out/ regardless of destination zone, now uses BSO zone directories (ftn/out.015/ for zone 21, etc.)
- Fix: TIC outbound delivery broken -- staged TIC descriptors and attached files had no .flo flow file entries, so the mailer never discovered or sent them. Now creates .flo entries with zone-aware directories.
- Timed events system: sysop-configurable scheduled commands for door maintenance, backups, etc. Launch types: Native Command (shell) and Dosemu (headless). Cron scheduling via BullMQ, %X variable substitution, timeout enforcement, output capture. Admin CRUD at /admin/timed-events with run-now button.
- Update script: txtfiles/ now completely skipped during updates (was copying new files which could still interfere with custom art). Use --include-art to merge art files explicitly.
- Call flow editor: DB-backed call flows and call pools, replacing hardcoded arrays. Admin UI at /admin/callflows with drag-to-reorder stage editor, add/remove stages, pool-to-flow assignment, WS path config. System flows (Standard, Doorserver, WebDoor) seeded on first run, protected from deletion.
- displayScreen menu command: display art files from menus and call flow stages with type:Welcome/Goodbye random selection or specific subdir/filename targeting
- Missing menu commands added to admin dropdown: QWK Transfer, News Viewer, Notifications
- Honeypot login detection with fail2ban integration: login attempts with known-bad usernames (root, admin, etc.) logged to security.log with [HONEYPOT] tag. Normal auth failures logged with [AUTHFAIL] tag. Terminal connections using honeypot names are immediately disconnected. Honeypot names blocked from registration (show as "taken"). Config keys: honeypot_enabled, honeypot_usernames, trust_proxy. Example fail2ban filter + jail configs in examples/fail2ban/.
- Reverse proxy support: trust_proxy config key for correct client IP resolution behind nginx/Caddy/HAProxy (X-Forwarded-For headers)
- Fix: Terminal login and registration now accept special characters in passwords (was restricted to alphanumeric only)
- Bad echomail handling: messages received for unknown areatags are stored in a holding table instead of being silently lost. Tracked per-areatag with first/last seen, counts, and source addresses. Admin UI at /admin/bad-echomail with per-tag message browser, recover (import to area when created), and purge. Daily auto-purge via bad_echo_retention_days config (default 30). Fixes packets retrying endlessly when they contain messages for areas we don't carry.
- Inbound file log: tracks every file received via binkp/EMSI sessions with peer connection info. Tosser uses it to populate received_from on bad echomail entries. Config: inbound_log_retention_days (default 90 days).
- Newscan default on message groups and areas: new column (default false) on groups and areas, admin UI on all group/area create/edit forms and echolist subscribe modal. Groups inherit to new areas.
- Combined area browser for messages and files: single page grouped by group with Subscribed/All toggle, instant search, AJAX subscribe/unsubscribe buttons, and help modal. Replaces two-page group list + area list navigation.
- Fix: Message body_html with unclosed tags (del/s) no longer cascades strikethrough to subsequent messages (CSS containment on .msg-card)
- Network Coordinator: segment definitions with role (Hub/NC/RC/ZC), managed nodes, and stub generation. Segments define what you coordinate with upstream address, distribution method (filebox/BSO/manual), and cron schedule. Admin at /admin/coordinator/:networkId with segment overview and per-segment node management. Stub generation in standard FTN nodelist format. Distribution via peer filebox, BSO flow file delivery, or manual download. Scheduled and on-demand generation via "Send" button.
- Network Coordinator redesign: nested segment hierarchy with parent/child relationships, internal/external source tracking, per-network role configuration (coordinator_roles in network config JSONB). Tree view admin UI with indented cards, source badges (Internal/External), context-aware "Add Child" with role defaults. External segments display received stub content read-only. Recursive tree-walk nodelist compilation.
- Network Coordinator node editor: three-tab layout (Identity, Connectivity, Flags). Role-aware keyword filtering, auto-populated zone/net from segment, hub routing dropdown. Progressive disclosure with separate tabs for address, network, and flags.
- Managed nodelist flag definitions: DB-backed flag definitions table replacing hardcoded COMMON_FLAGS. Each flag has code, description, category (internet/mail/modem/custom), has_argument, argument_hint, auto_managed, sort_order, active. Seeded from config/default/nodelist-flags.json with full FTS-5001 flag set (49 flags). Admin CRUD at /admin/nodelist-flags. Node editor Flags tab groups by category with inline argument inputs for flags that accept values (e.g. port numbers). Auto-managed flags (INA, IBN, ITN) shown as informational display.
- Push defaults for file groups: new Push Defaults button and modal with security, flags, newscan_default, and retention options. Mirrors existing message group push defaults.
- Newscan default added to message group push defaults modal.
- Seed sync tool:
node src/tools/seedSync.jsupserts seed data from JSON files into database tables. Compares by match key, inserts new records, updates changed columns. Supports --dry-run preview and --seedtargeting. - Fix: Unified message threading across all three tossers (FTN, QWK, NNTP). All now call shared messages.threadMessage() instead of maintaining separate implementations. Algorithm guarantees every message gets a thread (reply_to lookup -> subject match -> create new thread). No message can be inserted with NULL threadid.
- Fix: Post-batch orphan reconciliation: after each toss job, re-checks messages whose parents arrived in the same batch and re-links them to the correct thread. Fixes out-of-order FTN delivery where replies arrived before their parent.
- Fix: applyThreadingForArea() implemented (was a TODO stub). Scans orphaned messages and re-runs threading. Used by rethread maintenance and available for manual repair.
- Fix: Removed duplicate threading code from QWK tosser (
105 lines) and NNTP tosser (75 lines). - Refactor: Consolidated echomail.js + messages.js. All message reads, posting, threading helpers, thread maintenance, and inbound handling moved into messages.js as the single canonical model. echomail.js reduced to FTN protocol helpers only (ftnToDb, dbToFtn, getOutboundForConnection). Eliminates 8 duplicate function definitions that had caused recurring bugs where fixes got applied to one path and missed on the other. Net reduction of ~385 lines while guaranteeing a single source of truth.
- Fix: Origin line format now uses parentheses everywhere. Previously the messages.js posting path (web/admin) still used the old "BoardName 1:2/3" format while only the echomail.js path had been updated. Both paths now produce "BoardName (1:2/3)" and append the address in parentheses to custom origin lines if not already present.
- Fix: formatBodyHtml divergence resolved. messages.js no longer has its own simpler copy -- it now uses the canonical version from terminalText.js which does paragraph reflow, ANSI art detection, and noWrap mode. Web message rendering now matches import-time rendering.
- Fix: FTN inbound messages were getting duplicate line breaks. processMessageText() split on \r but joined with \r\n, leaving orphaned \n chars from \r\n pairs. Normalised to \r before splitting.
- Fix: FTN toss crash from syntax error when passing network areatag (e.g. FSX_GEN) to reconcileOrphans instead of local shortcode. Moved pushImportedMsgId into handleInboundEchoMessage where the correct shortcode is already resolved.
- Fix: %D (door directory) and %F (dropfile directory) variables now correctly resolve nested %X tokens. Previously, setting dropfile_dir to %D in a door config returned the raw, unsubstituted doorDir string (e.g. "%!/lord") because %D resolved to the pre-substitution value. Door dropfiles were landing in invalid paths instead of the configured door directory. substitute() now recursively re-expands resolved values that still contain %X tokens, capped at 8 levels to prevent infinite loops.
- Message templates system: sysops can write reusable message templates that fire on user lifecycle events (registered, verified, password reset, purge warning), activity milestones (first login, Nth login, login anniversary), and scheduled events (birthday, account anniversary, dormant users). Templates support {var} substitution with standard vars (boardname, sysopname, date, user.*) and per-event vars. Each template can deliver via PM, email, netmail, or any combination -- per-transport skip when address unavailable. Templates can filter with JSON conditions (security_min, verified, nth_count, days_inactive, permission/restriction_flags). Admin CRUD at /admin/templates and /admin/template-events. Custom events can be fired from the admin UI Fire page, menu command (fireEvent), CLI tool (node src/tools/fireEvent.js), or timed events (new FireEvent launch type with optional user filter query).
- REST API v1 with Bearer token auth: new
/api/v1namespace exposes programmatic access to Canectic for scripted clients (QWK readers, automation, personal tooling). Users issue tokens on their profile page; sysops can audit and revoke all tokens at/admin/api-tokens. Tokens arecanectic_-prefixed, bcrypt-hashed at rest, shown exactly once at creation. Scopes:qwk(download packets),qwk_rw(download + upload REP). First endpoints:POST /api/v1/qwk/downloadreturns an octet-stream QWK packet for the caller's subscribed areas + unread PMs,POST /api/v1/qwk/uploadimports a REP packet via multipart or raw body. Each successful API call is recorded in the caller log (source QWK-API) with logoncount throttled bump (default once every 4 hours,api_call_count_interval_hoursconfig) so active users still appear as regulars without inflating metrics from polling clients. Fires newuser_qwk_fetchedtemplate event on every call. CSRF exempt; auth failures return JSON. Sysop guide:docs/sysop/API.md. Architecture reference:docs/reference/API_DESIGN.md.
v0.9.5a (2026-04-04)
Notification System
- Centralised notification model (
notificationstable) with user-targeted and sysop-broadcast types - Bell icon in both admin and user nav bars with unread count badge and dropdown (10 recent items)
/notificationsfull history page with pagination- Real-time push via PresenceSocket: unread count on connect, new notifications pushed instantly
- Terminal
notificationReadermenu command: lightbar list, detail view, mark all read - Notification producers wired: sysop receives
tic_error,pm_undeliverable,areafix_request,filefix_request,user_registered; users receivepm_received - Daily purge job for old read notifications
- Config keys:
notifications_enabled,notifications_retention_days
Call Flow Improvements
- Login always routes through callflow regardless of redirect target; original destination preserved and restored after completion
- Time-based triggers: 2+ hours shows prominent banner and redirects home page; 24+ hours force-redirects all pages
- Prominent "Welcome back" banner: blue bar with stars icon, large "See What's New" button, dismiss button
last_callflow_atcached on session (loaded from DB once per session, not every request)- Callflow completion updates all sessions for user via
sessionUpdateUser()(multi-tab/device safe)
WebSocket Background Tab Fixes
- Background tabs no longer attempt WebSocket reconnect (deferred to visibilitychange event)
- Server prune tolerance increased to 5 missed cycles for better background tab survival
- Session expired redirect deferred for background tabs until they become visible
PM File Attachments
- File attachments on private messages across all transport types
pm_attachmentstable linked bymessage_uuid,has_attachmentsboolean on pm_inbox/pm_outbox- Attachment storage:
{pm_attachment_dir}/{YYYY-MM}/{uuid}/{filename} - Config keys:
pm_attachment_dir,pm_max_attachment_size(10MB default),pm_max_attachments(5 default) - LOCAL: shared file on disk, refs in both inbox and outbox
- EMAIL inbound: SMTP MIME multipart extraction with recursive nested multipart, base64/QP decode
- EMAIL outbound: nodemailer attachment passthrough
- FTN inbound: file-attach flag (0x0010) parsed from type-2 packet, files linked from ftn/in
- FTN outbound: flag set on attribute word, .flo flow file entries written for binkp delivery
- Web: multer upload on compose/reply forms, download route with ownership check
- IMAP: dynamic BODYSTRUCTURE (multipart/mixed), full RFC 2822 rendering with base64 attachment parts
- Terminal: text indicator "[Attachments available -- use web or IMAP to download]"
Session Management
- Remember Me checkbox on login (configurable via
remember_me_days, default 14) - Session TTL default 12 hours (
session_ttlconfig key), stored per-session for correct renewal - Log Out All Other Sessions on user profile (password protected)
- Admin Kill Sessions button on user management page
- Session expiry detection: PresenceSocket heartbeat validates session, expired sessions redirect to login
- WebSocket reconnect with exponential backoff (5s to 60s max) on both user and admin layouts
- Node >= 18.0.0 engine requirement added to package.json
Web Art and Seasonal Art
- Web art system: images served from
txtfiles/webart/{arttype}/(PNG, JPG, SVG, GIF, WebP) - Random image selection per page load for call flow welcome splash
- Seasonal/holiday art overrides: admin-configurable date ranges at Content > Seasonal Art
- When today's date matches an active season, art from
{arttype}/{season}/is used before default pool - Works for both terminal (ANSI/ASCII) and web (image) art
- Seeded defaults: Christmas (Dec 15 - Jan 2), Halloween (Oct 25-31), New Year (Jan 1-3)
- Auto-creates directories across all art formats when a new season is added
- Prompts whether to delete directories when a season is removed
- "Welcome" art type added to terminal (drawrandomscreen, drawscreen, menu display file types)
- Call flow welcome stage shows random web art + BBS name + greeting before data stages
- Install script creates all standard art directories on fresh install
Menu Management
- Duplicate menu: copies all settings, layout, colours, and entries to a new menu identifier
- Export to JSON: download a menu and all entries as a portable
.jsonfile (DB IDs stripped) - Import from JSON: upload a previously exported file to create a menu, with optional name override
- Buttons on menu list page (Import) and individual menu page (Duplicate, Export JSON)
News Files
- Admin-configurable news/score files at Content > News Files
- Displays ANSI/ASCII art files from door games, scores, or BBS news in the call flow
- Terminal:
newsViewermenu command displays files in sequence with any-key advance, ESC to skip - Web: rendered as HTML via
formatBodyHtmlin monospace cards with black background - Optional centering for oversized terminals (configurable source width per file, default 80)
- Drag-to-reorder, active/inactive toggle, add/edit/delete modals
- Integrates into both web and terminal call flows as a "News" stage
Home Page
- Tabbed layout with Announcements and Wall tabs
- Font size bump to 1.2rem site-wide for improved readability
Web Call Flow
- Guided login experience at /callflow mirroring the terminal call flow
- Stages: new announcements, last callers, who's online, wall (with inline post), new message summary, new file summary
- Auto-triggered on web login, prompted on idle session resume with "Welcome back" bar
- AJAX-driven single page with smooth stage transitions, no full page reloads
- Stages with no data auto-skip (e.g. no new announcements = stage skipped)
- Keyboard navigable: Enter/Space to advance, S/Escape to skip all
- New message and file summaries link directly to the relevant areas
users.last_callflow_atcolumn tracks "new since last visit" for announcement filtering- Config keys:
callflow_enabled,callflow_on_resume
Theme System
- PostgreSQL-backed terminal colour themes with admin CRUD at /admin/themes
- Colour category accordion editor (General, Menu, Content, Messages, Files, Terminal)
- Cascade: per-entry colour > per-menu override > theme default > handler fallback
- Duplicate theme, set default, reset to theme defaults per menu
- Seeded from
config/themes/canectic.jsonon first run - TinyMCE "Theme Default" colour bridge:
#010101sentinel maps to theme content colour via CSS - Web colour bridge:
resolveWebColour(theme, context)maps page contexts to theme properties for web content containers - Themes button added to admin dashboard
Menu System Overhaul
- Margin-based entry regions with configurable top/bottom/left/right margins
- Auto-column sizing: system calculates how many columns fit based on region width and entry text
- Max columns setting, column gap, fill order toggle (columns vs rows)
- Vertical centering within the entry region
- Lightbar mode: arrow-key highlight bar navigation with Home/End/PgUp/PgDn
- Featured entries rendered in a separate colour
- Art-defined entry positioning mode via
@Enn@/@Fnn@codes in ANSI art
Art Code System
@XX@codes embedded in ANSI/ASCII art files, replaced inline at render time- Universal codes:
@BN@board name,@UN@username,@DT@date/time,@ND@node, etc. - Width specifier:
@BN20@for wider fields (pad/truncate to width) - Menu codes:
@Enn@/@Fnn@entry slot positioning,@MT@/@HK@/@PR@/@ER@UI positions - Async pipeline for DB-backed codes (
@TC@total calls,@TU@total users,@ON@online count) - Fast path skips processing when no
@byte found drawfileWithCodes()/drawscreenWithCodes()async rendering entry points
Size-Aware Art Discovery
- Art file walk order: exact size match (
filename-WxH.ext) > column match (within 5 cols, best row fit) > default (filename.ext) - Size fit beats format capability (sixel > utf8 > ansi > ascii)
- Applies to all art rendering system-wide (menus, headers, screens)
- Sixel companion overlay:
.six+.anspairing for art-defined menus on sixel terminals
Door Collections
- Many-to-many category grouping for door games
- Admin CRUD at /admin/doorCollections with drag-to-reorder
- Terminal door browser with collection navigation and description panel
- Web door listing with collection cards and play/info buttons
- Door browser auto-resume: returns to collection after door exit
Door Runner Consolidation
- Five launch types: NativeShell (PTY), NativeCommand (piped stdio), Dosemu (console), DosemuPty (full), SyncJS
- Legacy ExecNative/SpawnNative/SyncJS types mapped to the consolidated set
- All launch types support output parser (CP437 conversion + DECSTBM centering)
NodeSpy System
- Web dashboard at /admin/nodespy with auto-updating card grid (4-second AJAX poll)
- Live terminal snoop via xterm.js with OutputRingBuffer catch-up history
- Sysop-to-terminal chat: split-screen DECSTBM on target, web chat bar for sysop
- Chat queue: if user not at menu, chat fires automatically when they return
- One-time Redis token authentication for WebSocket connections
- Terminal
nodeSpymenu command: node list, live watch, chat initiate
Sysop Chat
- Break-in chat from NodeSpy (both web dashboard and terminal command)
- SplitScreenChat engine: DECSTBM scroll region, conversation history, input line, peek line
- Target input hijacked via
call.rawmodeduring active chat - 5 beeps on chat start, ESC to exit (sysop only)
- Chat queue indicator on both web (pulsing badge) and terminal (bold status badge)
Web Sysop Chat
- Sysop can initiate real-time chat with web-browsing users from NodeSpy dashboard
- NodeSpy dashboard now shows web presence users alongside terminal nodes
- Chat routed through PresenceSocket (
/ws/presence) with Redis-backed session persistence - Floating chat overlay widget on all pages (both user and admin layouts)
- Chat survives page navigation via
chat_restorewith full message history from Redis - Minimise to badge with unread count, typing indicators, notification sounds
- Config keys:
web_chat_enabled,web_chat_sound
User-to-User Telegrams
- Any authenticated web user can send a real-time message to another online web user
- "Message" button on Who's Online page next to web users
- Same chat overlay and PresenceSocket infrastructure as sysop chat
- Two chat modes:
sysop(admin-only, target can only minimise) andtelegram(any user, both can close)
Wall / Oneliners
- Terminal wall handler: display recent messages, post prompt, call flow stage + menu command
- Web wall at /wall with post form, also embedded on home page
- Access control:
wall_post_security,wall_post_required_flags,wall_post_restricted_flags - Auto-purge: oldest messages removed when count exceeds
wall_max_messages(default 60) - Admin management at /admin/wall with delete capability
- One-post-per-session soft limit (web form hidden after posting, terminal skips prompt)
- Terminal colour highlight for own posts
- Config keys:
wall_display_count,wall_max_length,wall_max_messages
Bulletin System
- DB-backed bulletins with admin CRUD at /admin/bulletins (Content menu)
- TinyMCE editor for bulletin content, drag-to-reorder, active/inactive toggle
- Terminal
bulletinViewermenu command: numbered selection, art file display with scrolling - Web bulletin listing at /bulletins with detail view
- Art file association: bulletins can reference ANSI art files for terminal display
System Statistics
- Terminal
sysInfomenu command + call flow stage (SubModuleAdapter) - Web /stats page with system info card
- Aggregated statistics: total users, total calls, total messages, total files, online count
formatFileSize()utility for human-readable file sizes
File Subscriptions and Newscan
- User file area subscriptions: T key toggle in file area selector (mirrors message area pattern)
user_file_subscriptionstable with user override > areanewscan_defaultresolutionfileNewscanterminal menu command: scan subscribed areas for new files since last visit- File area selector shows new file counts per area
Custom Permission and Restriction Flags
- Sysops can define custom flags beyond the built-in set (Sysop, Mod, Adult, etc.)
- Config keys:
custom_permission_flags,custom_restriction_flags(comma-separated) - Custom flags appear in all flag selectors: area access, menu entries, wall posting
- Built-in flags always present, custom flags appended
User QWK/REP Packet Handling
- Personal QWK packet download: subscribed message areas + unread PMs, uses existing scan pointers
- REP packet upload: replies posted to message areas via standard posting path, PM replies delivered
- Web interface at /qwk with download button and upload form
- Terminal menu command
qwkTransferwith D/U/Q hotkey menu and ZMODEM/XMODEM protocol selection - Conference numbers via
message_areas.qwk_conf_numbercolumn (defaults to area ID, sysop-editable) - Config keys:
qwk_enabled,qwk_bbsid,qwk_max_msgs_per_area
areas.bbs Import
- Admin tool at Messages > Import areas.bbs for batch-creating message areas from FTN hub area lists
- Upload areas.bbs file, select target message group and network
- Preview with per-area controls: include/exclude, action (Create New or map existing unmapped area), allow peers, enable peers
- Bulk controls: Select All, Deselect All, Toggle All Allow, Toggle All Enable
- New areas inherit group security level and permission/restriction flags
Who's Online and Activity Tracking
- Web activity middleware: session renewal, Redis presence with
setActivity()for page-specific labels - PresenceSocket (
/ws/presence): lightweight WebSocket for heartbeats, presence tracking, chat - Activity map: central display label resolution for both terminal and web activities
- Who's Online page: merged terminal nodes + web presence, deduplication, activity labels
- Who's Online partial reusable for home page and call flow embedding
- Last Callers page with source badge (Web/Terminal), reusable partial
Goodbye Handler
- Confirm logoff prompt with batch download check
- If user has tagged files in batch, prompts to download before disconnecting
- Protocol selection (XMODEM/ZMODEM) for batch download on exit
Web Activity Tracking
- Centralized session renewal in
webActivity.jsmiddleware - Redis presence with
setActivity()for page-specific activity labels getOnlineUsers()API merging terminal and web sources- Configurable idle threshold for new call record logging
File Area Maintenance
- Duplicate file cleanup (hash-based detection)
- Missing file removal (DB entries for files no longer on disk)
Testing
- 471 new unit tests (1,649 total, up from 1,178)
- New test coverage: areaAccess, activityMap, wall, bulletins, announcements, systemStats, artCodes, qwkUser
- Bulletin
updateByIdargument order bugfix found and fixed via testing
AreaFix / FileFix
- Automated echomail area management via FTN netmail (AreaFix)
- Automated TIC file area management via FTN netmail (FileFix)
- Commands: +AREATAG (subscribe), -AREATAG (unsubscribe), %LIST, %QUERY, %AREAS, %HELP, %PAUSE, %RESUME
- Authentication via
areafix_passwordin peer connection_data (password in Subject line) - Auto-generates response netmail queued as outbound packet
- Processor routing in
ftnNetmailAdapter.js: 'areafix'/'areamgr' and 'filefix'/'filemgr'/'tic'
FREQ (File Request)
- Three-layer file resolver: magic aliases > filebox directory > file entries database
- Inbound: binkp session detects
.reqfiles, resolves and queues files for sending back - Outbound: admin UI at Messages > File Requests to create
.reqfor any FTN peer - Magic alias CRUD with file/filelist/nodelist types (seeded with FILES=filelist)
- Auto-generated BBS-wide file listing (magic type
filelist, written to filebox as_FILES.LST) - Filebox browser showing current files in
ftn/filebox - Config keys:
freq_enabled,freq_filebox_dir,freq_max_files(default 10)
areas.bbs / files.bbs Export
generateAreasBbs()utility for generating areas.bbs/files.bbs from network mappings- Admin download routes:
/admin/export/areasbbs/:networkId,/admin/export/filesbbs/:networkId - Download buttons on network detail pages in admin
%AREAScommand in AreaFix and FileFix returns the listing in areas.bbs/files.bbs format
Tossed Packet Retention
ftn_keep_tossed_daysconfig key controls retention of tossed FTN packets (default 7 days)- 0 = delete immediately after tossing, 999 = keep forever
- Daily maintenance job
purge-tossedat 2 AM removes expired packets - Tosser reads config at runtime instead of using hardcoded constants
Admin Panel
- Scan Nodelists and Purge Tossed Packets buttons added to Jobs & Maintenance page
Multi-Address Binkp Sessions
- Binkp sessions now handle multiple peer connections when the remote advertises multiple AKAs in M_ADR
- Password-filtered grouping: additional connections only included if their session_password matches the authenticated password and is non-empty
- Outbound file list built independently for each grouped connection (no cross-contamination)
- All active FTN addresses advertised in M_ADR with defaults first
- Session deduplication checks all grouped connections to prevent duplicate outbound polls
- Backward compatible: single-address sessions work identically
Bugfixes
- Fixed DataTables column mismatch on nodelist view (missing Connects Via column in AJAX rebuild)
- Fixed bulletin
updateByIdargument order (arguments were reversed) - Fixed duplicate web call logging (activity timestamp now set on login)
- Fixed stale web sessions causing reconnect loops (session expiry detection added)
Documentation
- Door server design doc captured (design for multi-BBS door game sharing, not yet built)
- Process separation architecture note in
PROCESS_ARCHITECTURE.md(future web/terminal split guidance)
Infrastructure
- Version read from
package.jsonvia config system (not per-call file reads) - Shared access check utility (
src/utils/areaAccess.js) extracted from route code areasBbsParser.jsutility for areas.bbs file parsing and generation- Web presence
userSocketsreverse index for targeted WebSocket message delivery - Redis-backed chat sessions with TTL safety net and automatic reconnect restore
v0.9.1-alpha.1 (2026-03-29)
Web Door Launcher
- Browser-based door game access via xterm.js embedded terminal
- Token-based WebSocket authentication (server-generated, one-time use, 60s TTL)
- Fixed 80x25 terminal with auto-scaling font in centered card layout
- Door listing page at /doors with play links
- Auto-redirect to door list on door exit
Terminal Scaling and Centering
- DECSTBM scroll region for vertical + horizontal centering of door output on oversized terminals
- Output parser extended to all door launch types (NativeShell, ExecNative, SpawnNative, Dosemu, DosemuPty, SyncJS)
- Per-door
fixed_sizeflag in admin door editor for doors that need forced 80x25 centering - CP437-to-UTF8 conversion in output parser without byte mangling on raw CP437 terminals
- Form feed (0x0C) intercepted and converted to proper ESC[2J clear for xterm.js compatibility
- Scroll region preserved across door screen clears, reset on parser destroy
Encoding Detection
- Fixed UTF-8 test character sending 4 bytes instead of 2 (JS string encoding issue)
- TTYPE-based encoding override for modern terminals (xterm, gnome, konsole, etc.)
- TTYPE response now stored on call object from telnet subnegotiation
- Populated UTF-8 charset glyphs (30 characters) and box drawing characters (40 characters)
- Fixed getboxes() returning wrong array for UTF8 encoding
User Verification
- 6-digit verification code sent via email (SMTP relay) or FTN netmail on registration
- Auto-send on registration and address change
- Terminal verification handler (call flow + menu command) with send/enter/change address
- Web verification page with send, enter code, change address, days remaining countdown
- Unverified restriction flag added to new accounts on creation
- Security level bump on verification (configurable via verified_security)
- Sysop auto-verified on promotion
- Unverified account purge (configurable retention period, daily maintenance job)
- Mail type auto-detection for email vs netmail (handles 5D FTN addresses)
Password Reset
- Pre-login: code-based reset from logon screen (username, send code, enter 8-char code, set password)
- Web: forgot password page with token link, password change form
- Post-login: change password menu command (current password required)
- 4-hour token expiry, 2-minute rate limit between sends
- "Forgot Password?" link on login page
Profile Editing
- Editable fields: username (with configurable cooldown, default 90 days), name, location, gender
- Inline password change section on profile page
- Verification status badge with link to verify/change address
- Username change updates session immediately
Admin Panel
- Jobs & Maintenance page with tabbed manual triggers for all worker jobs
- Announcement drag-to-reorder with save button
- Menu entry drag-to-reorder with save button
- User maintenance button on admin users page
- SMB message base import (upload .shd/.sdt/.sid, preview, import to target area)
- Door editor: fixed_size toggle for per-door centering control
- TinyMCE GPL license key configured
Infrastructure
- Empty-string DB config values now fall back to defaults (was overriding with blank)
- Menu entry permission/restriction flag parsing handles JSON strings
- User maintenance worker (mailtype reconciliation + unverified account purge)
- Daily user maintenance schedule (5:00 AM)
- Legacy TCP/Telnet servers start when their ports are explicitly set
- Default menu entries (redrawMenu on ?) added to new menus automatically
- Hotkey prefix on menu entries using menu/theme hotkey color
Synchronet JS Doors
- Output parser wired for SyncJS doors (CP437 conversion + optional centering)
- Fixed noyes() return value (true=No, false=Yes) matching Synchronet convention
- Fixed deny() to pass through noyes() directly
- Documentation warning about yesno()/noyes() return value asymmetry
v0.9.0-alpha.1 "Handshake" (2026-03-24)
First public alpha release.
Core Systems
- Terminal server with auto-detect TCP/Telnet and WebSocket support
- ANSI/CP437 and UTF-8 terminal with auto-detection and wide terminal support
- Hotkey menu system with configurable colors and layout
- Threaded message reader with scrolling viewport and quoting
- Full-screen message editor with word wrap and quote system
- File browser with native XMODEM and ZMODEM transfer engines
- Batch file downloads with tag-and-download
- Door game support (NativeShell, ExecNative, SpawnNative, Dosemu, DosemuPty, SyncJS)
- Private messaging across 5 transport types (LOCAL, FTN, QWK, EMAIL)
Web Interface
- Responsive web forum with Bootstrap 5 Darkly theme
- File browsing, download, and multipart upload with hash verification
- Webmail-style private messaging with split-pane inbox
- 6-step registration wizard with AJAX validation and gender field
- Landing page with announcement/bulletin system
- Admin dashboard with full CRUD for all system entities
Networking
- FTN type-2 packet tosser with binkp mailer (TCP + WebSocket)
- BSO outbound scanner with nodelist routing
- QWK client pull/push and hub mode
- NNTP article pull/push
- TIC file echo distribution with auto-forwarding
- Internet email transport (outbound SMTP relay, inbound trusted gateway)
Mail Client Access
- IMAP4rev1 server for PM access from mail clients
- SMTP submission server with multi-transport addressing
- Cross-process IMAP IDLE push notifications
Infrastructure
- Four-process model (gateway, worker, mail, mailer) managed by PM2
- BullMQ job queues with scheduled toss cycles and maintenance
- Three-tier configuration (.env > database > defaults) with admin UI
- Declarative database migrations
- CSRF protection, startup DB/Redis validation, health check endpoint
- Install/update script, .env.example template
- 1,178 tests passing