Skip to content

verifier: JOBS_SIGNAL_OP signals with SIGUSR1/SIGUSR2, not SIGINT - #673

Closed
jaredLunde wants to merge 1 commit into
mainfrom
verifier-signals-not-int
Closed

jaredLunde wants to merge 1 commit into
mainfrom
verifier-signals-not-int

Conversation

@jaredLunde

Copy link
Copy Markdown
Contributor

Found by the Beyond pin bump's W12.1 on E2B: through the front, the signal op reaches a job's group failed (job printed ready, never the SIGINT handler's output) while 38 other checks passed. sandbox-exec there is started by a shell's cmd &, which starts it with SIGINT ignored; an ignored signal cannot be trapped, so no correct endpoint could pass. The check (and the jobs.signal-op.delivers-the-named-signal scenario) now deliver SIGUSR2 then SIGUSR1, which no launcher pre-ignores. execd conformance and the rust-driver scenario pass locally. Needs a v0.1.23 tag for the bump.

🤖 Generated with Claude Code

https://claude.ai/code/session_01LBGuEqmuH3r35k31s2EJ8X

…her pre-ignores

A sandbox-exec started by a shell's `cmd &` (the E2B provider) has SIGINT ignored, and an ignored
signal cannot be trapped, so the check could never see a SIGINT handler run on a correct endpoint
(it failed through the Beyond front on E2B: the job printed "ready" and never "I"). The signal op
is the same; only the handler signals change, in the verifier and in the matrix scenario.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01LBGuEqmuH3r35k31s2EJ8X
@jaredLunde

Copy link
Copy Markdown
Contributor Author

Superseded: the verifier was right. A job's SIGINT does nothing on an endpoint started with SIGINT ignored, which is the product bug; fixing execd instead (new PR).

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant