Adds a link to CustomComponent/LocalAi in the packaged schema's Settings ›
Spam Filter, above LLM Classifier, and updates the schema hash so admins
fetch the new layout rather than a cached one.
INBUXA Admin draws the page (feature/local-ai-setup); this makes it
reachable. An admin from before that page would show "Unknown component"
here, so this lands after the admin release that carries it.
The server fetched upstream's latest published rules from GitHub at run
time: a version nobody here tested, code-like expressions from an account
we don't control, and the upstream name as a default in the admin form.
The published rules of spam-filter v3.0.2 are now embedded
(resources/spam-filter/, MIT, in THIRD-PARTY.md) and used whenever no other
source is configured. An empty setting and upstream's old default both mean
the bundled rules, so existing installs switch without a settings change;
the URL stays an operator override (https:// or file://). The schema default
is dropped and its description says what empty means, and the strip's
rename pass does the same to each import.
Rules load on first boot as before, and again whenever the bundled version
differs from the last one loaded, which only adds missing rules and tags.
That brings the AI classifier's LLM_* scores to installs that predate them:
production has none today.
upstream-watch now also opens an issue when spam-filter publishes a newer
release; resources/spam-filter/README.md says how to take it.
The antispam test now runs on the bundled rules, the path production
takes; SPAM_RULES_URL tests another set. Unit tests cover the URL handling
and that the bundled rules parse and score the AI tags as the AI spec says.
The rename pass vendored a patched sieve-rs and pointed Cargo.toml's
[patch.crates-io] at vendor/sieve-rs. .dockerignore ignores everything and
re-includes a short list that did not have vendor on it, so the image build
had no such directory and stopped at
failed to load source for dependency `sieve-rs`
failed to read /build/vendor/sieve-rs/Cargo.toml
CI could not have caught that: it builds from a checkout, where the
directory is simply there, and only the image build has a context to prune.
The first that was known about it was a tag that had already been pushed.
So: vendor is re-included, and tools/fork/context-check.py now asserts the
thing that was quietly assumed -- every path a [patch] section names exists
and survives .dockerignore. It runs beside the other fork checks and takes
no toolchain.
Also, the comments in .dockerignore started with // , which Docker does not
read as a comment: they were patterns that happened to match nothing. They
are # now.
v2026.9.24 was tagged on a commit CI had passed, and its release build could
not compile crates/scim at all:
error: queries overflow the depth limit!
= note: query depth increased by 130 when computing layout of
{async fn body of context::<impl ...>::writable_domain()}
The crate is ours, and the failure is profile-dependent: the release profile
computes those async fn layouts in one go and goes past rustc's default query
depth, while the dev profile never gets that far. CI builds dev, so CI was
green on a commit that could not be released. The tag produced no image and
no release, which is the one merciful part.
Two changes:
- #![recursion_limit = "256"] on the crate, which is what rustc itself
suggests, with a note saying why it only shows up in release. Proved by
building -p scim in release locally: it now finishes.
- CI builds the release profile too, on pushes to main. Pull requests stay
on dev, where the wait is worth less. A few minutes per merge is cheaper
than learning this from a tag, which throws away a multi-architecture
build and leaves a version half-cut.
The server sets a deferred recipient's next retry from its clock when the
attempt defers, in whole seconds. The test subtracted its own clock taken
when the loop next saw the message, after saving and reporting, so
whenever that lag crossed a second boundary the 2 s retry measured 1 s
and the test failed. Under load, after the other SMTP tests, that was
most runs.
It now measures from when the test started the attempt, which the
server's deferral can only follow, by under a second: each retry is its
interval or one more. Each position is still checked against its own
interval, so a wrong schedule still fails.
It listens on HTTP 19048, as the dkim2 DSN and report tests do. Those are
marked serial, this wasn't, so when the SMTP tests ran together (as
upstream's CI runs them) it could start beside them and requests reached
whichever server had the port: missing JMAP creates here, and 'You must
authenticate first' in dkim2_dsn_is_signed.
It failed everywhere but upstream's machines, for two reasons:
- The spam rules, which carry every score, came from a path on an
upstream developer's own disk. Without SPAM_RULES_URL none loaded, every
score was 0.00 and the combined case came out ham instead of spam at
13.70. The published rules of spam-filter v3.0.2 are now pinned beside
the test cases (Apache-2.0 or MIT, taken as MIT; in THIRD-PARTY.md).
SPAM_RULES_URL still overrides.
- The first combined case expects a Pyzor hit, and its digest (that of an
empty body) wasn't among the three the test mode answers, so it went to
a public Pyzor server: it failed offline and would drift with that
server's counts. Test mode now answers every digest from a fixed table,
with the empty body's added, and never reaches the network.
The test passes online and offline, alone and with the rest of the SMTP
tests. queue_retry, unrelated, still fails when it runs after the others
in one process, though it passes alone every time.
The name is inbuxa, lowercase, like the wordmark; INBUXA reads as an
acronym. The admin and webmail already changed. Here that's everything
the server shows people: the brand macro behind the protocol greetings,
the HTTP and SCIM realms, the startup banner and the calendar and contact
PRODID; the first-party OAuth client descriptions; the legacy-protocol
refusals; the default calendar and address book names and the SMTP
greeting default, in the code and the schema served to the admin
(checksum regenerated); startup and shutdown events; the User-Agent;
the sign-in and RSVP pages; the service units; the OpenAPI realm; the
crate descriptions and the README, where it's set in bold.
Identifiers that are uppercase for their own reasons stay: INBUXA_*
settings, SUBSPACE_INBUXA. So do code comments and the AGPL 5(a) notice
lines.
Tests follow: the IMAP ID name, the default collection names, the PRODID
in the iTIP fixtures and the CalDAV free-busy expectations, and the e2e
legacy-protocol refusals. The webdav, imap and jmap suites pass, so do
the unit tests of every crate touched, and 73 of 75 SMTP tests; of the
other two, antispam fails on main too, and queue_retry is a timing flake
that passes on its own.
Carries upstream 0.16.23 -- the DSN, POP3, Sieve, DMARC-report, ACME and
DNSSEC-resolver fixes in its own change log -- with the files it changed
marked under AGPL section 5(a), and one upstream test dropped that the fork's
routing makes meaningless.
It is also the first release whose tag attaches binaries: a host install can
now fetch inbuxa-linux-amd64.tar.gz or inbuxa-linux-arm64.tar.gz instead of
pulling the image and copying the file out of it.
Everything clients, users and operators meet now carries the fork's name,
with no aliases (SPEC.md §2.4, changed here from "protocol identifiers
stay"):
- JMAP: upstream's registry capability is urn:inbuxa:jmap:registry, beside
the fork's own urn:inbuxa:jmap.
- WebDAV lock and sync tokens are urn:inbuxa:dav*; clients resync once.
- Sieve: vnd.inbuxa.while and vnd.inbuxa.expressions. sieve-rs spells these
into its compiler, so it's vendored (vendor/sieve-rs, 0.7.3) and patched in;
a unit test fails if Cargo.lock ever moves past the vendored copy. The
trusted runtime now names itself too, rather than answering sieve-rs's
default.
- The web interface's OAuth client is inbuxa-webui. On every start the old
stalwart-webui client is removed and any application naming it is moved
over.
- The spam filter's blobs are INBUXA_SPAM_*; every start moves any left
under the old keys, so a trained model survives.
- SQL stores and log files default to inbuxa, in the code and in the
schema served to the admin (checksum regenerated).
- Settings are INBUXA_* only. A STALWART_* variable that's set where its
INBUXA_* one isn't stops the server at startup, naming it.
- The version-upgrade messages link docs.inbuxa.org's migration page, and
the OpenAPI description, smtp crate metadata and web-push test fixtures
lose the name.
Kept on purpose, allowlisted with reasons: the OAuth key-derivation
contexts (renaming them would end every session and invalidate every
sealed client id) and the hashed application prefix.
Also fixes a latent start-up failure: ensure_client updated an existing
first-party client with a revision of 0, which the registry's assertion
never matches, so adding a redirect URI or changing the webmail secret
failed start-up. And the principal session test now expects
legacyProtocols (C-1, added 2026-09-21), which it had missed.
Tested: the server builds without warnings; common's 106 unit tests,
including the vendoring check; a new integration test for the two
start-up migrations; and the webdav, jmap, imap and SMTP Sieve suites.
A release published an image and nothing else, so there was nothing for a
host install to download -- the only way to get the binary was to pull the
image and copy it out, which makes "install without Docker" depend on
Docker.
Each release now carries inbuxa-linux-amd64.tar.gz, inbuxa-linux-arm64.tar.gz
and SHA256SUMS, named as stalwart-migrator's are.
They are taken out of the image this pipeline just pushed rather than
compiled again. A second Rust build per architecture is the slowest thing
here, and it would leave two artifacts that are meant to be the same build
and only probably are. Extracting makes that identity a fact: the binary in
the tarball is the file the image runs. `docker create` starts nothing, so
copying a file out of an arm64 image on an amd64 runner needs no emulation.
One thing the extraction cannot carry: the image grants the binary
cap_net_bind_service, and a tar archive does not keep that xattr. The
release body says so, and says what to do instead -- setcap, or
AmbientCapabilities in the unit -- because a server that cannot bind 25 and
does not say why is a bad first hour.
Checked by hand against v2026.9.23 before this landed: both architectures
extract to the right ELF, and the amd64 binary runs on a bare Debian 13 with
every library resolved and reports its own version.
strip.py compiles the stripped tree, so a dual-licensed file that only
serves an Enterprise feature fails the import instead of the merge, as
v0.16.23's tests/src/directory/issuer.rs does. Upstream's tests of the
features the fork rebuilt are expected not to compile there and are listed
in build-check-known.txt; an error anywhere else fails the run. Checked
against both imports: v0.16.22 passes with its 16 expected errors, v0.16.23
fails on issuer.rs alone. Imports the strip leaves unused are reported.
It also renames the upstream name where clients, users or operators meet
it as an identifier, from tools/fork/renames.py: wire-protocol names, the
web interface's client id, store keys, configuration defaults and the
served schema. main is renamed with the same module, so a re-import
arrives purged and those lines don't conflict.
notice-check.py fails CI when an upstream file the fork changed, measured
against the upstream branch, lacks its AGPL 5(a) notice; --fix adds it.
It runs beside the name check in a renamed fork-checks job.
Also commits v0.16.23's strip report under docs/fork/strip-reports/, which
the import in #18 left out.
tests/src/directory/issuer.rs, new in v0.16.23, tests routing a bearer token
to a directory by its issuer. That routing is Enterprise-only upstream (the
body of get_directory_for_issuer), and the fork doesn't build it: a token
naming no address gets the server default (DIR-2). The test also calls a
helper from upstream's Enterprise-only OIDC test, so it can't compile here.
mta.rs imported types::id::Id for code inside an Enterprise snippet; the
stripped tree leaves it unused, upstream's as well as ours.
These upstream files were changed after the fork marked the files it had
modified, and never got the notice: six by the listener and schema-cache
work on 2026-09-20, two by the name check. Found by diffing against the
upstream snapshot branch, as before.
Five conflicts, resolved:
- crates/common/src/auth/authentication.rs: upstream's get_directory_for_token
and JwtClaims replace extract_jwt_domain; the per-domain directory code
(DIR-1, DIR-5 to DIR-7) is kept, and the token lookup routes through it.
The release's one new Enterprise snippet was the body of
get_directory_for_issuer, which stays returning None: a token naming no
address gets the server default, as DIR-2 specifies and as v0.16.22 did.
- crates/common/src/manager/application.rs: upstream's rewrite of the tests,
with the temp directory names renamed again, and the 5(a) notice the
name-purge change should have added.
- crates/common/src/network/mta.rs: both sides' imports.
- crates/main/Cargo.toml: the AGPL-only license kept, version 0.16.23.
- Cargo.lock: upstream's, with the fork's crates added by Cargo.
tools/fork/name-check.py reads every string literal in crates/ (comments
and test directories skipped) and fails on any that carries the upstream
name without an entry in name-allowlist.txt. An upstream merge can bring
such strings in without a conflict, so it runs on every push and PR.
The first run found three the earlier sweeps missed, fixed here: the SMTP
HELP reply pointed at upstream's website (now brand_url!), the event
collector thread was named after upstream, and the FreeBSD default data
path still said /var/db/stalwart/ where Linux already had /var/lib/inbuxa/.
Two operator-visible defaults are allowlisted as open, pending a decision:
the log file prefix and the SQL stores' default database and user.
Reads metadata only: upstream's releases list from GitHub's API and the
head of the upstream branch from Gitea's. Nothing of upstream's is
fetched, so its history can't land here. Daily at 06:17 UTC.
The first-party application descriptions and the telemetry service name and
instrumentation scope are shown to operators, and the unpacked-application
temp directory carried the name too.
Left alone deliberately: the OAuth key-derivation contexts (renaming them
would invalidate every sealed token and client id), the migration defaults
that read an upstream installation, links to upstream's upgrade guide, the
wire-protocol identifiers, and upstream's own license and templates.
brand_version_full! is user-visible -- --version, the startup banner, the
console, telemetry and the JMAP session's implementation field -- and the
name belongs only in copyright notices and the lineage line.
publish.yml replaces .github/workflows/publish.yml: on a v* tag it checks the
tag equals v<brand_version!> and is on main, builds the linux/amd64+arm64
image in one buildx run (the Dockerfile already cross-compiles, so only its
final stage goes through QEMU), pushes :<version> and :latest to the
registry, links the package, and creates the tag's release if it has none.
weekly-release.yml ports .github/workflows/release.yml: bump brand_version!
through the contents API, then create the release and so the tag, which
starts publish.yml. It only dry-runs until RELEASE_LIVE=1 and a
RELEASE_TOKEN secret exist.
GitHub took the organization's repos and GHCR offline on 2026-09-20. Repo,
release, raw-file and clone links now go to Gitea at git.coffeylabs.org,
container images to registry.coffeylabs.org, and GitLab-style /-/blob paths
to Gitea's /src/branch form. Go module paths are identifiers and stay as
they are; links to GitHub issues and pull requests are left as history.
The build now mounts the named volume inbuxa-server-cargo at /cache and keeps
CARGO_HOME and CARGO_TARGET_DIR there, so a push reuses the compiled
dependency tree (RocksDB included) instead of rebuilding it from scratch.
Both runners allow that one volume; each host keeps its own copy.
With the cache in place the job moves to runs-on: light, so it can run on
host2 as well. Cargo's parallelism now follows the job's CPU cap rather than
the host's core count, and the target dir is dropped past 25 GB.
jcoffey-dev