ci: cache compilation and cancel superseded validation #400
No reviewers
Labels
No labels
code-review
correctness
dos
performance
security
severity/high
severity/low
severity/medium
tech-debt
Kind/Breaking
Kind/Bug
Kind/Documentation
Kind/Enhancement
Kind/Feature
Kind/Security
Kind/Testing
Priority
Critical
Priority
High
Priority
Low
Priority
Medium
Reviewed
Confirmed
Reviewed
Duplicate
Reviewed
Invalid
Reviewed
Won't Fix
Status
Abandoned
Status
Blocked
Status
Need More Info
No milestone
No project
No assignees
1 participant
Notifications
Due date
No due date set.
Dependencies
No dependencies set
Reference
h-dv/code-index!400
Loading…
Reference in a new issue
No description provided.
Delete branch "codex/ci-compiler-cache"
Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?
Linux validation downloaded Rust 1.98.0 over a 1.94.0 image, rebuilt without the runner's shared compiler cache, and left superseded core/package/pilot runs occupying the queue. This update pins the matching toolchain image, enables compiler reuse in four Linux core lanes, cancels obsolete validation by event and ref, and starts package fan-out after shared lint to give core checks runner capacity.
All 39 Rust-image references use ci-rust:1.98.0-v7 at registry digest
sha256:8de376754f48f8cbe01c52e9a7163bb00be353226f55ff9016192114f7c145f4. It carries Rust 1.98.0, rustfmt and Clippy. The image was built from the existing ci-images recipe at ed2bd54f, withBASE_IMAGE=git.h-dv.de/h-dv/ci-min:1.94.0-v7andRUST_VERSION=1.98.0. Only its new versioned tag was published. The tool baseline was re-probed; a manifest read after registry logout and repository formatting with networking disabled both passed.Clippy, workspace tests, daemon tests and corpus opt into the runner-supplied Redis cache after disk pre-flight. Each job owns a Unix socket. Missing tools/backend or failed startup leaves ordinary compilation enabled; successful startup configures server-I/O fallback. Statistics omit the backend URL. Both package matrices wait for their shared lint job, allowing independent core checks and pilots to start first. Their conditions still produce scope reports after failed/skipped lint and stop cancelled runs. This Forgejo instance did not enforce max-parallel during live validation; scheduling uses an explicit dependency. All eleven packages remain in both matrices. Core/package/pilot cancellation groups include both event and ref, preserving scheduled measurements. Release/channel writers retain their shared, non-cancelling publishing group.
The first Windows run failed before initialize returned in
concise_search_text_keeps_the_snippet_origin_disclosure: 4,760 of 4,761 ordinary tests passed, with no product error on inherited stderr. The harness now includes the child's OS exit status on stdout EOF. A real-child test verifies exit code 2 from a CLI refusal is retained. #401 tracks the original startup failure, whose cause remains unknown; the diagnostic is not a root-cause fix. The obsolete runs were cancelled with their logs retained, and the replacement head requires all normal CI gates.Local validation of the replacement tree:
ci; its nine timing cases remain assigned to the serial timing lane. All 15 affected-target tests passed in snapshot and daemon modes.The final-head live Clippy job passed with 345 compiler-cache hits (294 Rust, 50 C/C++, one assembler), zero misses and zero cache errors; compilation took 5m17s. The earlier cold run populated those 345 entries with zero cache errors. These are compiler-reuse measurements; full-run speed depends on runner load and remaining compilation. Native Windows completed successfully in 39m29s, including all 4,762 ordinary tests, 27 serial timing tests, doctests and plugin qualification. Both pilots passed in 4m25s. All eleven package lanes and eleven real-corpus lanes passed in 55m10s. Linux core has passed all 4,980 ordinary tests, 37 serial timing tests, doctests, candidate C#/Timeline qualification and the broad corpus lane. Daemon transport qualification also passed; the complete Linux workflow finished in 1h23m46s. Its cache recorded 203 hits / 189 misses with zero errors. Workspace caching recorded 131 hits / 460 misses; corpus recorded 19 hits / 130 misses; both reported zero cache errors. The package duration does not establish an end-to-end CI speedup; repeated compilation remains a bottleneck. Test commands, required rows, thresholds, retry/incremental policies and runtime/plugin-host build selections retain their existing behavior.
Fixes #399
ci: use pinned Rust image and shared compiler cacheto ci: cache compilation and cancel superseded validation