Repository navigation
fix(ci): repair generated checks and obsolete recovery tests - #6529
Conversation
Generated-by: OpenAI Codex
Astro-Han
left a comment
There was a problem hiding this comment.
Automated review notice: This comment was posted by an automated review agent. It is not an independent human review and does not replace one.
I reviewed the complete 11-file diff at 9dfc8e9caa3cd92fc53423e8b943a9853758a273 and checked all changed file contents against its merged result 37ca6f16; they match. I found no confirmed P0–P2 defect in this source review.
The wire tests retain capture/admission/charge-release and cancellation assertions while moving initialization/teardown outside the 10-second progress watchdog; the catalog sharing/schema-projection assertions remain in their existing test with a smaller description fixture. The launcher test now applies a 60-second watchdog to each actual update/cleanup case, rather than both full fixture lifecycles together. These are deliberate changes to what is timed, not evidence that runtime initialization became faster. The continuation test saves a draft before mounting the UI, observes recovery of the original receipt, requires that draft to survive and retains the exactly-two-model-requests assertion. Removed desktop controls and the reply-discard helper are no longer asserted. I also checked the assertion lint change, generated SDK comment and archive-selection log line.
For notices, an independent structured comparison shows 11 CUA component identities move from 7dfc429b to 9e385e66, no common component digest changes, and notice-document differences consist exactly of that source-revision substitution. I did not independently regenerate dependency licenses or accept the external CUA implementation.
I executed the 13 existing automatic-recovery JavaScript tests against current Rust 50687a74's production TypeScript modules under Node 22.23.2: all pass. Those tests use controlled NSH/Host boundaries and mock timers, not a native renderer or real cross-process journal. I did not execute the changed Rust fixtures, generated-contract toolchain or actual old-store/crash recovery locally.
The previous failures are concrete: old #6520/#6523 runs each ended at 1,980 passed, one failed and 26 skipped, with the launcher test's 60-second timeout; another d513a72e run failed one wire progress timeout. None establishes PR-specific causality, and this source adjustment does not itself prove those failures resolved. Current merged-Rust CI remains in progress. This review is not a full audit of #6527's 81-file recovery change.
Automated review notice: This comment was posted by an automated review agent operated by Astro-Han. It is not an independent human review and does not replace one.
Summary
Restore rust-preview checks after the automatic-recovery changes, and remove obsolete test behavior. The real-Host continuation test now recovers the original lost receipt automatically while preserving a draft saved before the UI mounts. The removed desktop test exercised manual check/dismiss/feedback controls that no longer exist; its unused reply-discard helper is removed as well.
Wire admission tests keep their charge-release, counting-work and cancellation assertions with smaller fixtures. Their progress watchdog excludes Root initialization and teardown. The launcher update test similarly measures each actual update and physical cleanup rather than two complete Host fixture lifecycles together. Existing timeout values are unchanged.
Also fix the two reported Clippy warnings, regenerate SDK declarations and pinned Cua notices, and log the actual root selection count for archive coverage auditing.
Verification
AI use
Tool(s) and scope: OpenAI Codex implemented, validated and reviewed the changes.
Checklist
Does this PR entail a change in behavior?