fix(ci): unbreak GitHub and Forgejo pipelines #1

Merged

hutao merged 3 commits from fix/ci-green into main 2026-09-13 09:40:31 +00:00

Conversation 0 Commits 3 Files changed 7 +74 -39

hutao commented 2026-09-13 08:05:58 +00:00

Owner

Copy link

Every failing job had its own cause; none of them shared a fix.

fmt / clippy — both CIs

• AI_MAX_TOKENS landed unformatted in 4cd0fbf.
• Two lints had gone un-run: a redundant & in a format! argument (src/commands/level_cmds/level.rs:54) and .iter() over a map where only the values are used (src/data/embed_media.rs:147).
• AI_MAX_TOKENS is now gated behind ai. Its only consumer is data::ai::provider, which lives under #[cfg(feature = "ai")] (src/data/mod.rs:1), so the no-feature leg of the matrix failed -D dead-code.

docker — GitHub

Debian bullseye is EOL and its security suite is archived, so apt-get update aborts with:

E: Release file for http://deb.debian.org/debian-security/dists/bullseye-security/InRelease is expired

and exits 100. Both stages move to bookworm.

test / docker — Forgejo

The runner created service containers on Docker's default bridge — the one network with no embedded DNS — so postgres and redis never resolved. The wait loop burned its full timeout and sqlx migrate run then failed with an address-lookup error naming no host.

The runner-side fix is hutao/vps fix/forgejo-runner-network (container.network: ""), which must merge and deploy before this repo's Forgejo test job can pass. The loop here now exits non-zero and names the unreachable host instead of falling through.

The docker job is dropped. Job containers get no Docker socket by design (container.docker_host: "-"), and both ways around that cost more than the job is worth:

• A docker:dind service needs container.privileged: true, which privileges every job container. A workflow's own options: --privileged is silently dropped — the runner's mergeJobOptions (act/container/docker/run.go:430) only honours --health-*, --volume, --tmpfs, --memory and --hostname.
• A host-side dind needs a privileged daemon reachable from the job network, i.e. from every other container on the box.

.github/workflows/full.yml already builds the image on every push to main, which is all this job duplicated.

Verification

All seven matrix legs pass locally (fmt, 3× clippy -D warnings, 3× build). The wait loop was tested both ways — reachable → 0, unreachable → 1 naming the host.

The Docker build is unverified locally (no Docker on the dev machine); this PR's GitHub run is its first real test.

No CHANGELOG.md entry per AGENTS.md:89 — CI/build plumbing and lint fixes, no behavior change.

Co-authored-by: Claude Opus 5 (1M context) noreply@anthropic.com

Every failing job had its own cause; none of them shared a fix. ## fmt / clippy — both CIs - `AI_MAX_TOKENS` landed unformatted in 4cd0fbf. - Two lints had gone un-run: a redundant `&` in a `format!` argument (`src/commands/level_cmds/level.rs:54`) and `.iter()` over a map where only the values are used (`src/data/embed_media.rs:147`). - `AI_MAX_TOKENS` is now gated behind `ai`. Its only consumer is `data::ai::provider`, which lives under `#[cfg(feature = "ai")]` (`src/data/mod.rs:1`), so the no-feature leg of the matrix failed `-D dead-code`. ## docker — GitHub Debian bullseye is EOL and its security suite is archived, so `apt-get update` aborts with: ``` E: Release file for http://deb.debian.org/debian-security/dists/bullseye-security/InRelease is expired ``` and exits 100. Both stages move to bookworm. ## test / docker — Forgejo The runner created service containers on Docker's **default** bridge — the one network with no embedded DNS — so `postgres` and `redis` never resolved. The wait loop burned its full timeout and `sqlx migrate run` then failed with an address-lookup error naming no host. The runner-side fix is hutao/vps `fix/forgejo-runner-network` (`container.network: ""`), **which must merge and deploy before this repo's Forgejo `test` job can pass.** The loop here now exits non-zero and names the unreachable host instead of falling through. The `docker` job is dropped. Job containers get no Docker socket by design (`container.docker_host: "-"`), and both ways around that cost more than the job is worth: - A `docker:dind` **service** needs `container.privileged: true`, which privileges *every* job container. A workflow's own `options: --privileged` is silently dropped — the runner's `mergeJobOptions` (`act/container/docker/run.go:430`) only honours `--health-*`, `--volume`, `--tmpfs`, `--memory` and `--hostname`. - A **host-side dind** needs a privileged daemon reachable from the job network, i.e. from every other container on the box. `.github/workflows/full.yml` already builds the image on every push to main, which is all this job duplicated. ## Verification All seven matrix legs pass locally (`fmt`, 3× `clippy -D warnings`, 3× `build`). The wait loop was tested both ways — reachable → 0, unreachable → 1 naming the host. **The Docker build is unverified locally** (no Docker on the dev machine); this PR's GitHub run is its first real test. No `CHANGELOG.md` entry per `AGENTS.md:89` — CI/build plumbing and lint fixes, no behavior change. Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>

hutao added 1 commit 2026-09-13 08:05:58 +00:00

fix(ci): unbreak GitHub and Forgejo pipelines ... aea631a470

Every job that was failing had its own cause; none of them shared a fix.

fmt/clippy (both CIs): `AI_MAX_TOKENS` landed unformatted in 4cd0fbf, and
two lints had gone un-run — a redundant `&` in a `format!` argument and an
`.iter()` over a map where only the values are used. `AI_MAX_TOKENS` is also
gated behind `ai` now: its only consumer is `data::ai::provider`, which lives
under `#[cfg(feature = "ai")]`, so the no-feature leg of the matrix failed
`-D dead-code`.

docker (GitHub): Debian bullseye is EOL and its security suite is archived,
so `apt-get update` now aborts with "Release file ... is expired" and exit
100. Both stages move to bookworm.

test/docker (Forgejo): the runner creates service containers on docker's
DEFAULT bridge, the one network with no embedded DNS, so `postgres` and
`redis` never resolved — the wait loop burned its full timeout and
`sqlx migrate run` then failed with an address-lookup error naming no host.
Fixed on the runner side (vps: container.network ""); the loop here now
exits non-zero and names the unreachable host instead of falling through.
The `docker` job is dropped: job containers get no docker socket by design
(container.docker_host "-"), and both ways around that cost more than the
job is worth. A `docker:dind` service needs container.privileged, which
privileges EVERY job container — a workflow's own `options: --privileged` is
silently dropped, since the runner's mergeJobOptions only honours
--health-*, --volume, --tmpfs, --memory and --hostname. A host-side dind
needs a privileged daemon reachable from the job network, i.e. from every
other container on the box. GitHub already builds the image on every push to
main, which is all this job duplicated.

No CHANGELOG entry: CI/build plumbing and lint fixes, no behavior change.

Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>

hutao added 2 commits 2026-09-13 09:14:14 +00:00

fix(ci): drop the /ai-review coverage gate, which aborted the script ... a99c4ffd61

`check_coverage "src/data/ai/review/guilds" 90` outlived the code it
gated — /ai-review was removed in #42 and `src/data/ai/review/` with it.
The pattern now matches no SF: line, so `lf` stays 0 and check_coverage
returns 1 on "no lines found".

Under `set -e` that aborts the whole script at that line, so the global
`check_coverage "src/" 70` below it has not run since #42. Both CIs mark
the step `continue-on-error: true`, and GitHub reports a continue-on-error
step as `success`, so nothing surfaced it — Forgejo printing
`exitcode '1': failure` is what made it visible.

Verified against the lcov.info artifact from the last green run: with the
stale gate gone all three checks run and pass, the global one at 78%.

The stale `src/data/ai/review/(agent|client|github)\.rs` entry in
DISCORD_BOUND goes too — same removed module.

Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>

perf(ci): drop dev debuginfo from the clippy and build matrices ... 240ba821d8

Measured on --all-features, clean: 375s -> 357s wall, and the target
directory 2.3G -> 1.4G. The time is noise; the 39% smaller directory is
the point, because that is what the Actions cache has to upload and
download on every job once caching is enabled (vps#9).

Not set on `test`. That job runs cargo llvm-cov, whose line mapping is
worth more than a smaller cache round-trip.

Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>

hutao merged commit 6f5d32ce76 into main 2026-09-13 09:40:31 +00:00

hutao referenced this pull request from a commit 2026-09-13 09:40:32 +00:00

Merge pull request 'fix(ci): unbreak GitHub and Forgejo pipelines' (#1) from fix/ci-green into main

hutao referenced this pull request from a commit 2026-10-27 12:58:58 +00:00

Merge pull request #1 from 1Git2Clone/feat/llm-crate

hutao referenced this pull request from a commit 2026-10-27 12:58:58 +00:00

feat(custom-reactions): per-guild numbers, list command, URL validation

hutao referenced this pull request from a commit 2026-10-27 12:58:58 +00:00

Merge pull request 'fix(ci): unbreak GitHub and Forgejo pipelines' (#1) from fix/ci-green into main

Sign in to join this conversation.

Reviewers

No reviewers

Labels

Clear labels

No items

No labels

Milestone

Clear milestone

No items

No milestone

Projects

Clear projects

No items

No project

Assignees

Clear assignees

No assignees

1 participant

Notifications

Subscribe

Due date

The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set

Reference

hutao/serenity-discord-bot!1

No description provided.