tree a0d0b905e4145dc725b0f7f4fd41fac091609091
parent 487de66b54298c1949a47e6a8a242dda9da92703
author Filip Niksic <fniksic@google.com> 1790171123 -0700
committer Copybara-Service <copybara-worker@google.com> 1790171166 -0700

Skip redundant second replay for single-input non-persistent crash batches.

During crash replay (`ReplayCrash`), Centipede executes a batch containing a single crashing input (`input_vec.size() == 1`). Previously, when that input crashed, `Centipede::ReportCrash` re-executed the binary a second time to isolate the reproducer from the batch, and skipped writing crash metadata (`<hash>.desc` and `<hash>.sig`) if the second execution timed out (for example, due to slow stack trace printing in the signal handler), failed to reproduce a flaky crash, or produced a different signature. When crash metadata files were not written, `GetCrashesFromWorkdir` returned zero crashes (`Total crashes: 0`) and `CrashSummary` omitted the reproduced crash.

Disable `persistent_mode` in `ReplayCrash` so the single crashing input runs in an isolated process, and update `Centipede::ExecuteAndReportCrash` and `Centipede::ReportCrash` to save the reproducer and crash metadata directly from `batch_result` without re-executing the binary when `input_vec.size() == 1` and `persistent_mode` is disabled.

PiperOrigin-RevId: 986722767