Skip to content

Solution for long term memory for agent coding CLIs and to facilitate handoff between different agent vendors

Rust

ci: move Windows tests off the pull-request critical path

AkitaOnRails84fe7c0ba9fb
diff --git a/.github/workflows/ci.yml b/.github/workflows/ci.yml
index 88c5da1b..6c7db8c1 100644
--- a/.github/workflows/ci.yml
+++ b/.github/workflows/ci.yml
@@ -50,24 +50,14 @@ jobs:
- uses: Swatinem/rust-cache@e18b497796c12c097a38f9edb9d0641fb99eee32 # v2
- run: cargo test --workspace --all-targets
- # Native Windows is where the `#[cfg(windows)]` regression tests (drain
- # lock ERROR_LOCK_VIOLATION, libgit2 path-resolution fallback, verbatim
- # path handling) actually run — the test matrix above never compiles
- # them. Observability only for now (`continue-on-error`): a Windows
- # runner-environment quirk must not block Linux/macOS merges, but the
- # results are visible on every push so Windows-only breakage is caught
- # in review instead of by end users.
- test-windows:
- name: test (windows-latest, non-gating)
- runs-on: windows-latest
- continue-on-error: true
- steps:
- - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
- - uses: dtolnay/rust-toolchain@4cda84d5c5c54efe2404f9d843567869ab1699d4 # stable
- - uses: Swatinem/rust-cache@e18b497796c12c097a38f9edb9d0641fb99eee32 # v2
- - run: cargo test --workspace --all-targets
- env:
- TAILWIND_SKIP: "1"
+ # Native Windows coverage lives in `.github/workflows/windows.yml`: it
+ # runs on every push to main, nightly, on demand, and on any pull request
+ # carrying the `windows` label. It was ~1000s here against ~250s for the
+ # same tests on Linux, which meant every pull request waited roughly
+ # seventeen minutes when every gating job below finishes in about eight —
+ # and it was `continue-on-error`, so those extra minutes gated nothing.
+ # Add the `windows` label to a pull request that touches path handling,
+ # file locking, or git plumbing.
# The companion importer is deliberately outside the root workspace
# (its own [workspace] in companions/ai-memory-importer/Cargo.toml), so
diff --git a/.github/workflows/windows.yml b/.github/workflows/windows.yml
new file mode 100644
index 00000000..1b2f1fac
--- /dev/null
+++ b/.github/workflows/windows.yml
@@ -0,0 +1,71 @@
+name: windows
+
+# Native Windows is where the `#[cfg(windows)]` regression tests actually
+# run — drain-lock ERROR_LOCK_VIOLATION, the libgit2 path-resolution
+# fallback, verbatim path handling. The Linux/macOS matrix in `ci.yml`
+# never compiles them, so this coverage is not optional.
+#
+# It lives outside `ci.yml` because of what it costs there, not because it
+# matters less. A Windows run takes ~1000s against ~250s for the same tests
+# on Linux (slower linking, and a cache restore that is ~10x slower over
+# many small files). Every gating job in `ci.yml` finishes in about eight
+# minutes; leaving Windows in that workflow made every pull request wait
+# roughly seventeen, with the last eight spent on a job that was marked
+# `continue-on-error` and therefore blocked nothing. Linux and macOS are
+# the priority platforms and now set the pull-request feedback time.
+#
+# Moving it here makes the coverage *stronger*, not weaker:
+#
+# * it runs on every push to `main`, so every merge is checked;
+# * it no longer carries `continue-on-error`, so a real Windows break is
+# a red run instead of a yellow one nobody reads. It cannot block a
+# Linux/macOS merge from here, which was the original reason for the
+# override;
+# * a nightly run catches toolchain and dependency drift that no code
+# change would trigger;
+# * `workflow_dispatch` and the `windows` label give a pull request that
+# touches platform-sensitive code — path handling, file locking, git
+# plumbing — a way to opt in *before* merging.
+#
+# If you are changing any of those areas, add the `windows` label to the
+# pull request rather than finding out after the merge.
+on:
+ push:
+ branches: [main]
+ pull_request:
+ types: [opened, synchronize, reopened, labeled]
+ schedule:
+ # 04:23 UTC daily. Off the hour, and away from nix.yml's Monday slot.
+ - cron: '23 4 * * *'
+ workflow_dispatch:
+
+permissions:
+ contents: read
+
+concurrency:
+ # A newer push supersedes an in-flight run for the same ref. Nothing
+ # downstream consumes this job's result, so cancelling is safe and keeps
+ # the Windows runner queue short.
+ group: windows-${{ github.ref }}
+ cancel-in-progress: true
+
+env:
+ CARGO_TERM_COLOR: always
+ RUSTFLAGS: "-D warnings"
+
+jobs:
+ test:
+ name: test (windows-latest)
+ # On a pull request, only when explicitly opted in. Pushes to main,
+ # the schedule, and manual dispatch always run.
+ if: >-
+ github.event_name != 'pull_request' ||
+ contains(github.event.pull_request.labels.*.name, 'windows')
+ runs-on: windows-latest
+ steps:
+ - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
+ - uses: dtolnay/rust-toolchain@4cda84d5c5c54efe2404f9d843567869ab1699d4 # stable
+ - uses: Swatinem/rust-cache@e18b497796c12c097a38f9edb9d0641fb99eee32 # v2
+ - run: cargo test --workspace --all-targets
+ env:
+ TAILWIND_SKIP: "1"
diff --git a/AGENTS.md b/AGENTS.md
index 7fd08a9a..49db94ff 100644
--- a/AGENTS.md
+++ b/AGENTS.md
@@ -221,9 +221,17 @@ cargo deny check # dependency policy (i
`tests/e2e/handoff_smoke.sh`, `scripts/check-native-packaging.sh`.
- CI additionally runs `cargo build --release --bin ai-memory` on
Linux/macOS, a Docker image smoke test, `cargo audit` (with the ignores
- listed in `ci.yml`), differential gitleaks scanning, and a non-gating
- Windows test job. `.github/workflows/secret-scan.yml` runs the separate
- weekly/manual full-history gitleaks scan.
+ listed in `ci.yml`), and differential gitleaks scanning.
+ `.github/workflows/secret-scan.yml` runs the separate weekly/manual
+ full-history gitleaks scan.
+- **Windows runs in its own workflow** (`.github/workflows/windows.yml`):
+ every push to `main`, nightly, on demand, and on any PR labelled
+ `windows`. It is the only place `#[cfg(windows)]` tests compile, and it
+ is ~4x slower than the same tests on Linux — keeping it out of `ci.yml`
+ is what holds PR feedback near the eight minutes the gating jobs take.
+ **Add the `windows` label** to a PR touching path handling, file
+ locking, or git plumbing, so the check runs before the merge rather
+ than after it.
## Code style guidelines
diff --git a/CHANGELOG.md b/CHANGELOG.md
index f0935c0f..0f7f6a1e 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -8,6 +8,17 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
## [Unreleased]
### Added
+- Windows tests moved to their own workflow
+ (`.github/workflows/windows.yml`), running on every push to `main`,
+ nightly, on demand, and on any PR labelled `windows`. They were the
+ slowest job in `ci.yml` by a wide margin (~1000s against ~250s for the
+ same tests on Linux) while carrying `continue-on-error`, so every pull
+ request waited roughly seventeen minutes for a job that gated nothing
+ when all gating jobs finished in about eight. Coverage is now stronger,
+ not weaker: `continue-on-error` is gone, so a genuine Windows break shows
+ as a failed run on `main` instead of a yellow mark on a PR, and the
+ nightly run catches toolchain or dependency drift that no code change
+ would trigger.
- `ai-memory status` now reports local hook-spool health: pending event
count, age of the oldest queued event, and total failed-delivery attempts.
This makes "memory capture is delayed, not broken" visible to an operator