The database stores no NUL on either engine, but the application still held
the value a client sent. A key or a check made from it was then a different
value from the one the database compared: `victim@x.test` followed by NUL
was a fresh key for the per-email password-reset throttle while the lookup
found the victim's row, so every spelling mailed a new link and voided the
last one.
Every input now loses its NUL where it enters TREK: the nulFreeRequest
middleware after the body parsers (path, query string, body, cookies), the
first global pipe for what a route parses itself (a multipart body), the
websocket adapter for every frame, and the MCP registry through a new
host-supplied input mapper, since /mcp keeps its raw body. Headers need
nothing: Node refuses a NUL in one with a 400. A string under a key naming a
password stays as sent: it is only ever hashed, and stripping it would lock
out an account whose password was set with one. The stripping in the
database layer stays as the backstop for what does not come from a request.
createFromBaseline created the migration history's table through MikroORM's
storage before, and outside, the transaction that runs the baseline. Without a
master transaction MikroORM commits that table on its own, so a first boot that
failed in the baseline (a server without ICU, a timeout, a dropped connection)
left an empty mikro_orm_migrations behind. The next start with
RESTORE_FROM_BACKUP took that for a database in use, skipped the restore with a
warning and brought up an empty instance, after which the restore never ran.
The table is now created on the baseline's own transaction, so a failed first
boot leaves the database empty. The boot restore also treats a lone migration
history that records nothing as no database, which a restore clears anyway; any
other answer keeps it, so the check fails closed. Database-Postgres says both.
Plugin statements on Postgres went over the extended query protocol through a
marker parameter and a hook that replaced query() on every client of TREK's
pool. That kept a plugin to one statement per round trip, but it changed
TREK's own hot path for a concern only plugin data has.
The plugin data store now checks a pg client out of the ORM's pool for each
transaction, through MikroORM's getNativeClient, and runs the transaction on
it directly. Every plugin write goes as a query with queryMode 'extended' and
every read through a pg-cursor on that client, so Postgres parses exactly one
statement either way. The settings the transaction makes are local, and
before the client goes back to the pool the store drops temporary tables,
cursors, listeners, session advisory locks and sequence state, or closes a
client it cannot bring back to that state. The row cap, quota, time budget,
untimed migrate, per-plugin queue, pool share and script commits are as
before, and plugin args are still sent without NUL.
The marker, the hook and its install in the Postgres driver are gone.
lint:dialect missed several classes of SQL that SQLite reads and Postgres does
not, or reads differently:
- group_concat, ifnull, instr, iif, printf, typeof, total(), changes() and
json_object/json_array/json_valid/json_type/json_quote, as calls in SQL text
and as the bare name handed to eb.fn.
- LIKE in SQL text, Kysely's 'like' operator and MikroORM's $like, which ignore
case on SQLite only. A LIKE that is exact on purpose now says so in a
like-exact comment; the six existing ones (setting keys, storage paths, the
lowercased reference search, sqlite_master) carry it.
- The sqlite member of engineDdl was not read at all, so a trigger or a data
backfill put there never reached Postgres. It may now hold only the schema
change and a rebuild's copy and sqlite_sequence bookkeeping, and SQLite
statements with an empty postgres member need a sqlite-only reason.
- execBestEffort, bestEffort and execUnlessTableMissing in a post-cut
migration: on Postgres a swallowed failure aborts the migration transaction.
- The post-cut rules follow POSTGRES_FIRST_CUT as well as the manifest, so
regenerating the baseline over a new migration no longer releases it.
src/db/engine/sqlite/ is SQLite's own code and is no longer scanned.
Postgres text cannot hold U+0000, so any write or filter carrying one (a
request body, a query string, an import, a feed, a forwarded mail, a
plugin's own SQL) answered a 500 on Postgres while SQLite stored it. Both
engines now drop the character where every value leaves for the database
(src/db/nul-free.ts): the MikroORM platforms when they inline a string or
serialize a JSON value, the Kysely dialects for every parameter, a
plugin's own SQLite file for its arguments, and the move from SQLite,
which used to refuse such text. The same request now answers the same
status and leaves the same row on both engines, so the places NUL suite
runs on Postgres as well and leaves SQLITE_ONLY.
The release queue narrows each owner's rows with LIKE '%run%' before it
parses them, and relied on SQLite's LIKE ignoring ASCII case. On Postgres
LIKE is exact, so a row naming an object in uppercase was not found and
the release deleted an object that row still names. likeAny now matches
LOWER(column) against the lowercase run, which reads alike on both
engines. RELI-010 fails on Postgres without it.
Every integer column is bigint on Postgres. SQLite stored a fraction sent
to one (a packing quantity of 1.5, a budget item for 2.5 days, an id of
3.5) as a REAL, while Postgres rounded it in a statement MikroORM renders
and refused it with a 500 in a Kysely statement, so the engines held
different data for the same request.
The request schemas of those fields now take whole numbers only, and
INTCOL (tests/unit/contracts/integer-columns.test.ts) holds the class: it
reads the integer columns from the entities and fails on any DTO field or
inline MCP tool field of that name that takes any number.
Most writes start their notification without waiting for it. NotificationsService now tracks each send until it settles, and settled() resolves once every send started before it has finished. The channel check moves to channel-gate.ts, which keeps the service under the size limit.
On Postgres a check read before a transaction, or outside one, no longer
holds at the write that follows: the pool runs other requests' writes in
between. Moved into the unit of work that writes:
- the last-admin guard of an admin's role edit and of an account deletion
(deleteUserCompletely takes a precondition run in its transaction)
- the login lockout guards of the auth settings save and the SSO settings
save
- the first-account check of an SSO sign-up, which starts over as an
ordinary sign-up when another account won
- the OAuth client caps, the token cap and the mail intake limits
- the year check before a vacay year is opened, and the label name check of
a rename
A password sign-up that loses the last use of an invite is now refused, as
the SSO sign-up already was, and a reset link is burned by one conditional
update, so two resets with one link cannot both change the password.
UnitOfWork took the engine's write lock, but four repositories opened
transactions of their own: the demo addon set, the plugin user erasure's
savepoints, and the plugin metadata and scheduler cap checks. On Postgres
those ran beside the others on the pool.
writeTransaction (src/db/engine/write-transaction.ts) is now the one
transaction that writes: UnitOfWork wraps it, and the demo seed and reset,
which run outside the container, call it with the engine the environment
names. The capped plugin writes run in their caller's unit of work instead
of a transaction of their own, and the plugin erasure runs in the erasure's
transaction without savepoints, since its three tables come with the
migrations on both engines. lint:tx counts writeTransaction as a
transaction.
PLUGIN_DATA binds the plugin data backend to the engine (the same
{ sqlite, postgres }[engine.id] binding the backup port uses): the SQLite
file per plugin, unchanged, or the Postgres schema per plugin. The host
opens a plugin's store through it and an uninstall that deletes data erases
through it, which drops the schema on Postgres. Nothing outside db/engine
asks which engine runs.
A plugin's SQL is passed through as written, so a plugin needs to know the
dialect: db.engine (db:own) answers 'sqlite' or 'postgres', and the SDK
exposes it as ctx.db.engine(), in the mock host (dbEngine option) and in
dev. The plugin facts are regenerated and the SDK goes to 1.10.0.
The Postgres lane's PGPLUG suite installs, uses and erases plugin data on
Postgres and shows that a plugin reaches neither TREK's users table nor
another plugin's schema.
The baseline was travel-baseline.db on both engines and went straight into
the swap, so a SQLite baseline left in the data directory made every hourly
reset of a Postgres demo fail inside pg_restore, and never got replaced. It
is now named after the engine's archive entry (travel-baseline.db on
SQLite, travel-baseline.pgdump on Postgres), and the backup port's new
checkCopy refuses one the engine cannot restore before the swap; the reset
then saves a fresh baseline and logs the reason instead of failing hourly.
meta.set and scheduler.set checked their caps in a transaction of the
repository's own, which keeps check and write together on SQLite's single
writer but not on Postgres, where two pool connections both read the same
count. Both RPCs now run the capped write inside UnitOfWork.transactional,
which takes the engine's write lock; the repository's transaction becomes a
savepoint of it.
- The client tools are looked for newest first, from 18 down: a client
handles every server up to its own major and refuses a newer one. A dump
or a restore now refuses a client older than the server up front, naming
both versions, and the boot refuses a server outside PostgreSQL 14 to 18.
- sslrootcert=system is used only by a client from libpq 16 on; an older
client is refused with a message that names sslrootcert and the bin dir.
- The restore's CVE-2025-8714 table covers 18.
- A restore refuses a role that is a superuser or may run programs or write
files on the database host, since the archive's SQL runs as that role;
the boot warns about such a role.
- pg_restore's script is streamed to its work file and stopped once it
passes BACKUP_MAX_DECOMPRESSED_MB, so a small dump can no longer expand
without limit on the data volume.
- A failed tool's error keeps only the summarized stderr; its cause names
the exit code and signal instead of carrying the process error, whose
message and stderr quoted row data into the log.
- The migration history is looked for in the current schema, where the
tables and the backups are, so a role whose search_path is not public
boots a second time.
TREK runs on PostgreSQL when TREK_DB_URL is set, and on SQLite as before
otherwise: an engine port under src/db/engine, the dialect helpers under
src/db/dialect, a Postgres baseline for the migrations up to the cut,
pg_dump/pg_restore backups, a probe of every repository method against a
real server and a Postgres test lane in CI.
A place keeps its one category, which decides its marker, and can now carry
any number of tags as extra labels to filter by.
- Tags belong to the trip (tags.trip_id): every member sees the same list and
whoever may edit the trip's places adds, renames and deletes. Personal tags
stay what they were.
- The place form picks them on desktop and mobile through one shared hook,
usePlaceTags, and can add a new tag to the trip on the spot.
- The existing category filter of the places list lists the tags as well.
- Tags show on the place (inspector pills, phone place sheet); the marker is
untouched.
- MCP: create_place, create_and_assign_place and update_place take `tags`,
and list/create/update/delete_trip_tag manage a trip's tags.
- /api/trips/:tripId/tags and the tripTags:changed event keep every client
of the trip in step.
Since 1.22 Dawarich sends a visit's display name, which is "Unknown Location"
for a visit without a place. TREK showed that as the name, although the points
recorded during the stay are reverse geocoded. A stay without a real name now
takes its place's name or address, and otherwise the city most of its points
fall in. A visit without a place also takes the middle of its points as its
coordinate.
Closes#3105
A move keeps the entry's resolution and changes only its trip and created
ids, so an undo that read them before a move landed still claimed the entry,
removed nothing and left the moved booking behind. The claim is now bound to
the trip and ids it was read with.
With two booked sectors under one flight number, the second is in its watch
window for the whole first one, and the aircraft flying the first sector was
read as the second having departed, with its departure notice hours early.
An aircraft that fits a sibling sector better is now left to it.
isBareHeaderBlock rescanned the header run below every sender line, so a
plain-text mail of nothing but "From:" lines without a subject took
quadratic time and held the event loop for seconds before the address was
even checked. The block starts are now worked out for all lines in one pass
from the bottom up, with the same rule.
Clearing a common item's owner and dropping the bringer row now happen inside
one unit of work, so a failure cannot leave one without the other. The lint
and RTL baselines follow the rows that lost their last physical sides.
A common item only let you pledge yourself. Nobody else could be named,
and a bringer could never be removed, so an item that went personal and back
kept its owner's avatar for good. The item menu now lists every member to
tick on or off, a click on the avatar opens it there, and the add route takes
an optional user_id limited to the trip's members. Taking the owner off
clears the owner. The remove route now checks visibility for the caller, not
for the user named in the path.
A nullable day_assignments.stay_minutes holds the stay of one visit; a visit without one keeps reading the place's duration, so existing trips drive as before. REST PUT /trips/:tripId/assignments/:id/stay and the MCP tool set_assignment_stay write it with day_edit.
Follow-up messages failed with HTTP 400: a whitespace-only text block, two model messages in a row where the turn had pushed for an answer, or a tool call renamed after its provider state was recorded. The rebuilt history now matches what the provider saw during the first answer.
- The mail import's status tiles (which, labels, tones) were written out
in the desktop and the mobile card alike; statTiles in the shared model
holds them, each card keeps its own look.
- The journey share link set 'copied' back two seconds after a copy even
when the dialog had closed, which also broke a client test run (window
gone at teardown). useCopied, the trip share's hook, moves to hooks/ and
serves both; it clears its timer on unmount.
- PlanLlmConfig is not exported: only the union is used outside.
Only an upload started a processing run; a forwarded mail waited for the
next two-minute tick, and one that came in right after a run waited for two.
The IMAP and relay runs start one run once they took anything in, the
webhook once per mail it accepted.
Stalwart writes Delivered-To as the mailbox itself, its sub-address dropped,
so a forwarded mail to trek+<token>@ was refused as unknown_address: the
first delivery header won even though it carried no token. The headers are
now read in their order of trust, and the first that names a live address
decides.
A ChatGPT sign-in only ever ran the assistant: parsing's clients call an
endpoint with a key, and a plan has none. AI connections now lends the plan
as a reader. Each document is one Responses call with the schema as the one
tool the model has to call (tool_choice required). The token is read and
refreshed per call by ChatgptAuthService, and a sign-in that has to be
renewed lends nothing.
- ResolvedLlmConfig is a union: ApiLlmConfig (what an admin configures) or
PlanLlmConfig (provider chatgpt, a read function, never stored).
- ConnectionReadClient sends the text plus photo or scanned pages, and reads
the filled shape, or the shape written as text; prose is unreadable.
- AiChatClientService.chatgptPlan is the one place that builds the plan's
client, for the assistant and for parsing.
- Wiki: a ChatGPT sign-in reads bookings on its plan.
- The upload posted its FormData through the client's JSON default, so the
form went as JSON and the server saw no file (400). It goes through
postMultipart now, like every other upload.
- reservation-origin.ts becomes reservation-origin.contract.ts: the booking
import sets an origin, and a contract is what another domain may import.
- CI: the client's strict type check holds the whole program with strict
checks and outgrew the default ~4 GB heap; it gets 8 GB like the server's.
The own address was made only when asked for, and the card had no way to ask
before one existed, so it showed none. Once a template is set and the
person's import is on, the first look provisions it (idempotent; read after).
- auth.service: how an admin form value is stored (validated, encrypted, a
masked echo kept) moves to the app settings registry as storedAdminValue
and isMaskEcho.
- reservations.service: a booking's origin (withoutOrigin, originColumns)
moves to reservation-origin.ts, its input shapes to reservation-inputs.ts,
exported on from the service so importers stay as they were.
- Reservations.repository: the joined projection, the origin lookup and the
origin columns move to _shared/, like the Wallet pass context before.
- test-sql: MailIngestApplyService.prepare is marked as no SQL statement.
- The pending bookings dialog opens centred, and says 'nothing waiting' once.
- mime-walker: trim each line of a stripped HTML body, so the space a tag
leaves is not read as a folded header line that swallows the booking
under an Outlook forward's header block.
- settings: an unset daily AI limit is the default of ten, not zero
(Number(null) is 0), which kept the AI from ever reading a mail.
- env: MAIL_INGEST_IMAP_PASS and MAIL_INGEST_WEBHOOK_SECRET are secrets
and never printed in the boot report.
- pipeline test: the subject is kept without its forward prefixes, the
demo move adds its booking before the second trip exists, the typed
notification mock, and AI reading stays offered while the person
switched it off.
- e2e: the pipeline tick runs in a request context of its own, as the
cron registrar runs it, and the 401 checks run without awaiting in a loop.
ai_available folded the person's own switch in, so switching AI reading off
greyed the switch out and it never came back. It now says only whether AI
reading works on the instance (the AI Parsing addon and a connection). When
it does not, the settings card says it needs the AI Parsing addon, by its
localized name, in all 28 locales; the wiki says the same.
- lint:tx: the mail transports and the intake document why their writes stand
alone (@txIndependent, network I/O between them, one copy per address); the
JSDoc of NotificationsService.respond is back above it after the response
actions were registered in between.
- lint:boundaries: booking-import asks McpToolGuardsService whether the person
may edit the trip's bookings instead of reading the Users table; its unit
test builds the service through one named factory.
- lint:query-api: findIdByExternal reads through Kysely.
- lint:test-mocks: the mail e2e hands its own snapshot db to the ORM instead
of mocking the database module.
- Entities regenerated (index order only).
- Client: the review row reads the live booking, the shells share
useMailInboxShell, and the admin card's buttons are one size down.
The cron callbacks hand back nothing; the cursor's lookup is findCursor so it
no longer shadows the base repository's find(); the draft's suggested trip
uses the shared contract's type; endpoints spell out their optional columns;
the relay page's running count is typed. Prettier over every new file outside
the generated entities.
A plan without the import switches it off from outside the server, and an
admin can switch it off too; the plan promises a "mail import paused" note.
An hourly job sends mail_ingest_paused to each holder of a live address with
their own import on, notes the pause first, and forgets it once it is back.
The settings event was registered with no sender; renewing or revoking an
address and changing the switches now send it. The client lists both events
as not yet shown, which keeps the registry-parity test green until the mail
screens move them to their listener.
The webhook refuses headers that cannot be signed right before reading a
byte, limits each address to 120 requests a minute and reads four bodies at
once. The admin save refuses a webhook secret under 32 characters, and the
MCP settings tool registers its input strict like every other tool.
Anyone could add an Authentication-Results header with dmarc=pass. The hint
now needs a DKIM pass signed by the From domain or a DMARC pass naming it,
and reads it only for IMAP, where the receiving server's header is topmost;
the Cloud relay, the webhook and uploads never show a sender as verified.
A co-member could move the forwarder's booking into a private trip, carrying
their mail along; undo stays open to editors. Move and filing refuse demo
accounts, and filing claims the entry before writing documents, putting it
back into the inbox when nothing could be filed.
An upload made an address and stored a mail the pipeline then put aside;
it now answers 409 instead. The run an upload asks for at once starts only
where the crons run, so it can no longer outlive a test's database.