[magicdrive]

Problem

git-parallel always exits with 0, even when the command fails in some or all repositories. This is documented as historical behavior (test_error_propagation), so scripts cannot rely on the exit status to detect a failed pull/fetch.

Root Cause

Each batch is waited for with a bare wait, which returns 0 whatever the jobs returned, and each job (git::parallel::execute) ends with echo, so the command's own status is lost as well.

Solution

Add an opt-in --strict; the default is unchanged.

• Every job is waited for with wait "$pid" (in both modes) and, with --strict, its status is collected. execute now returns the command's status; a frozen repository returns 0.
• Decisions:
  • Default: unchanged, exit 0 (historical behavior).
  • Exit status: 0 when every repository succeeded (frozen ones and "no repository" count as success); 1 when at least one failed. Usage errors also exit 1 but happen before anything runs.
  • Partial failure / multiple repositories: all repositories still run (no fail-fast); stdout is the same as without --strict; stderr lists the failed repositories in order with their own status, also under --silent:

    git-parallel: the command failed in 2 of 5 repositories:
      /src/api (exit 1)
      /src/web (exit 128)
    

  • Cancellation: a failure does not cancel other repositories; only signals cancel.
  • SIGINT/SIGTERM (and HUP/QUIT): running commands are killed, temp files removed, git-parallel: interrupted by SIG<NAME> on stderr, exit 128 + signal (130/143/129/131). Without --strict the old handler (SIGKILL of the whole tree, git-parallel included) is kept.
  • JSON: git-parallel has no JSON output; none added.
• Help, man page and README document the exit status.

Security Impact

None. No new input is evaluated; command handling is unchanged.

Compatibility

Default behavior, output and exit status are unchanged; the existing 20 tests are untouched and pass. --strict is a new long option only.

Tests

8 new tests in tests/run.sh: success, partial failure with --jobs 3/2/1 (failure in every batch position), all fail (128) and command not found (127), distinct statuses per repository across batches (3, 42) and the last job of the last batch, frozen repositories, --silent, no repositories, SIGTERM (143) and SIGINT (130) with commands killed and temp files removed, --help.

• GREEN: make check (bash 5.3) and TEST_BASH=/bin/bash tests/run.sh (bash 3.2): 28 passed, 0 failed.
• RED: the new tests fail against main's script (43 assertion failures), and against this branch with the per-PID wait replaced by a bare wait (partial failure, statuses and frozen/silent tests fail).

CI

The existing matrix (Linux/macOS, bash 3.2/4/5) runs the new tests.

Migration

None. Use git parallel --strict -- ... where the exit status matters.

Risks

Low. The default path now waits per PID instead of a bare wait, which waits for the same set of jobs.

https://claude.ai/code/session_01RbGbwRHvCEhsKLZfbvbbRg