Repository navigation
fix(ci): run the Windows native lane with its process boundary fixture - #2336
Conversation
`verify-windows-native.mjs` takes the boundary executable as its second argument and skips every observation that needs it when the argument is absent. CI passed no argument, so the breakaway denial, caller and child token comparison, Windows environment semantics and in-place ancestor reparse attempt have never run there; they only ran for an operator who built the fixture by hand, which `native/windows/README.md` described as optional. The fixture is an ordinary Windows executable independent of the addon, so the lane that already installs MinGW to cross-build the artifact cross-builds it too and uploads it beside the existing inspection fixture. The Windows job downloads it and passes its path, matching how that inspection fixture already crosses the same job boundary. No observation, assertion or threshold changes; this only stops four of them from being silently skipped. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
|
Codex usage limits have been reached for code reviews. Please check with the admins of this repo to increase the limits by adding credits. |
|
@greptileai review |
|
This PR is set for a Plus review. TREX isn't supported with Plus or Apex reviews yet, so Greptile reviews this PR at Base while TREX is on. |
|
… lane `real-ghidra-windows.yml` also runs the native lane without the boundary executable, so its proof run skipped the same breakaway, caller and child token, environment semantics and ancestor reparse observations. Its artifact job already installs MinGW, so it cross-builds and uploads the fixture the same way, and the self-hosted Windows job downloads it and passes its path. This makes the README's claim about both Windows workflows true rather than narrowing it to one lane. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
|
All four red checks here are inherited from Static checks — Test shard (4/6) — Test (coverage + timing) and CI required are downstream of those two: the first merges the shard reports, the second is the aggregate gate. What this PR's own run does show, which is the part I could not verify locally:
Happy to rebase once #2691 lands if a green run is wanted before review. |
Source-reviewed changes. Tests, builds and check reruns were skipped at maintainer request.
Node timers can fire while performance.now() is still a fraction of a millisecond short of the monotonic deadline. ProviderStartupDeadline then reported interruption "timeout" and an aborted signal while remainingMs() returned ceil(0.x) = 1, so polling loops guarded by remainingMs() > 0 (Hopper and Ghidra startup) could run past an elapsed deadline, and the real-timer test "aborts a bounded wait when its real monotonic deadline expires" failed intermittently in CI (expected 1 to be 0). Once the deadline's own timer has classified a timeout, report zero remaining time. A deterministic regression test fires the timer with the monotonic clock 0.4 ms short of the deadline. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> (cherry picked from commit dfdaa0c)
…dows-boundary-fixture # Conflicts: # .github/workflows/ci.yml
…Rinkia/rea into fix/ci-windows-boundary-fixture
Problem
scripts/verify-windows-native.mjstakes the process boundary executable asprocess.argv[3]and gates a block of observations on it:CI runs the lane as
npm run verify:windows-native, which passes no arguments, so that block — and the reparse scenario gated on the same argument earlier in the file — has never executed in CI. These are the observations that never ran:processes.breakawayDeniedCREATE_BREAKAWAY_FROM_JOBis refused, so an owned child cannot leave the Job Objectprocesses.childOwnerNormalizedWithoutChangingCaller+callerTokendeepEqualbefore and after)processes.windowsEnvironmentSemanticscontrols.inPlaceAncestorReparseDeniedThey ran only for an operator who built the fixture by hand, and
native/windows/README.mddescribed it as "optional", so there was no signal that CI coverage depended on it.Change
The fixture is an ordinary Windows executable that reports its own token, environment and breakaway results; it does not link the addon. So the job that already installs
g++-mingw-w64-x86-64to cross-build the artifact cross-builds the fixture too and uploads it, and the Windows job downloads it and passes its path — the same waywindows-native-inspection-fixturealready crosses that job boundary.No observation, assertion, threshold or fixture source changes. The only behavioural difference is that four observations stop being silently skipped.
native/windows/README.mdno longer calls the fixture optional, says plainly that the lane skips those observations without it, and keeps the manual invocation for local runs.Validation
Windows 11 Pro 26100 x64, Node v24.16.0, on this branch:
actionlint 1.7.7with-shellcheck= -pyflakes=, the pinned version and flags from theworkflowsjob: clean.Artifact and fixture cross-built from this branch with the documented lane (Linux + MinGW-w64 g++ 12 in a container, since this host has no C++ toolchain).
npm run verify:windows-native:artifact→ok: true, SHA-2561eecd8e6…01219e.The lane run exactly as the workflow now invokes it:
→
ok: true, 11 process observations and 20 filesystem controls all true, including the four above, which reporttrueinstead of being absent. Confirms the--argument passthrough reachesprocess.argv[2]/[3]through npm on Windows.npm run build:cachedandnode scripts/check-doc-facts.mjs: pass.Not verified here: the GitHub-hosted runners themselves, so the artifact upload/download hop between the Linux and Windows jobs is exercised for the first time by this PR's own CI run. The uploaded executable loses its POSIX execute bit in transit, which does not matter on Windows, where
.exeis launched by extension.Independent of #2202 and based on current
main; if that lands, its caller-affinity observation is picked up by this same wiring with no further change.🤖 Generated with Claude Code