fix(runner): remove the run-output scratch dir on exit by Chemaclass · Pull Request #811 · TypedDevs/bashunit

Chemaclass · 2026-07-16T03:14:34Z

🤔 Background

Every invocation created $TMPDIR/bashunit/run/<os>/<random> (#801) but nothing removed it — one leaked directory per run (~9.9k had accumulated locally; the acceptance suite alone leaks ~258 per pass).

💡 Changes

Guarded cleanup in env.sh (path-checked like parallel::cleanup), armed via an EXIT trap for early-exit paths (--version, --help, subcommands).
The test-run path replaces that trap, so cmd_test calls the cleanup explicitly after rerun::persist; Ctrl-C and stop-on-failure handlers do the same.
Sourcing errors are now captured in the run dir too — files are sourced sequentially in the parent, so a fixed per-file-truncated path is safe and saves a mktemp + rm fork per test file.
Regression test runs a nested bashunit under a private TMPDIR and asserts nothing is left. SIGKILLed runs still can't clean up (no trap runs); a full suite now leaves ~6 dirs instead of ~258.

Every invocation created $TMPDIR/bashunit/run/<os>/<random> (#801) but nothing removed it, leaking one directory per run — ~9.9k dirs had accumulated locally. - env.sh gains a guarded cleanup (mirrors parallel::cleanup's path check) and arms it via an EXIT trap for early-exit paths (--version, --help, doc, ...). - The test-run path replaces that trap in main.sh, so cmd_test calls the cleanup explicitly after rerun::persist (which reads from the run dir), and the Ctrl-C / stop-on-failure handlers do the same before their exits. - Sourcing errors are captured in the run dir as well: files are sourced sequentially in the parent (workers fork after), so a fixed truncated-per-file path is safe and saves a mktemp and an rm fork per test file. Runs killed with SIGKILL still cannot clean up (no trap runs); a full local suite now leaves ~6 dirs instead of ~258.

Chemaclass · 2026-07-16T03:14:34Z

🤔 Background

Every invocation created $TMPDIR/bashunit/run/<os>/<random> (#801) but nothing removed it — one leaked directory per run (~9.9k had accumulated locally; the acceptance suite alone leaks ~258 per pass).

💡 Changes

Guarded cleanup in env.sh (path-checked like parallel::cleanup), armed via an EXIT trap for early-exit paths (--version, --help, subcommands).
The test-run path replaces that trap, so cmd_test calls the cleanup explicitly after rerun::persist; Ctrl-C and stop-on-failure handlers do the same.
Sourcing errors are now captured in the run dir too — files are sourced sequentially in the parent, so a fixed per-file-truncated path is safe and saves a mktemp + rm fork per test file.
Regression test runs a nested bashunit under a private TMPDIR and asserts nothing is left. SIGKILLed runs still can't clean up (no trap runs); a full suite now leaves ~6 dirs instead of ~258.

Every invocation created $TMPDIR/bashunit/run/<os>/<random> (#801) but nothing removed it, leaking one directory per run — ~9.9k dirs had accumulated locally. - env.sh gains a guarded cleanup (mirrors parallel::cleanup's path check) and arms it via an EXIT trap for early-exit paths (--version, --help, doc, ...). - The test-run path replaces that trap in main.sh, so cmd_test calls the cleanup explicitly after rerun::persist (which reads from the run dir), and the Ctrl-C / stop-on-failure handlers do the same before their exits. - Sourcing errors are captured in the run dir as well: files are sourced sequentially in the parent (workers fork after), so a fixed truncated-per-file path is safe and saves a mktemp and an rm fork per test file. Runs killed with SIGKILL still cannot clean up (no trap runs); a full local suite now leaves ~6 dirs instead of ~258.

Chemaclass · 2026-07-16T03:14:34Z

🤔 Background

Every invocation created $TMPDIR/bashunit/run/<os>/<random> (#801) but nothing removed it — one leaked directory per run (~9.9k had accumulated locally; the acceptance suite alone leaks ~258 per pass).

💡 Changes

Guarded cleanup in env.sh (path-checked like parallel::cleanup), armed via an EXIT trap for early-exit paths (--version, --help, subcommands).
The test-run path replaces that trap, so cmd_test calls the cleanup explicitly after rerun::persist; Ctrl-C and stop-on-failure handlers do the same.
Sourcing errors are now captured in the run dir too — files are sourced sequentially in the parent, so a fixed per-file-truncated path is safe and saves a mktemp + rm fork per test file.
Regression test runs a nested bashunit under a private TMPDIR and asserts nothing is left. SIGKILLed runs still can't clean up (no trap runs); a full suite now leaves ~6 dirs instead of ~258.

Every invocation created $TMPDIR/bashunit/run/<os>/<random> (#801) but nothing removed it, leaking one directory per run — ~9.9k dirs had accumulated locally. - env.sh gains a guarded cleanup (mirrors parallel::cleanup's path check) and arms it via an EXIT trap for early-exit paths (--version, --help, doc, ...). - The test-run path replaces that trap in main.sh, so cmd_test calls the cleanup explicitly after rerun::persist (which reads from the run dir), and the Ctrl-C / stop-on-failure handlers do the same before their exits. - Sourcing errors are captured in the run dir as well: files are sourced sequentially in the parent (workers fork after), so a fixed truncated-per-file path is safe and saves a mktemp and an rm fork per test file. Runs killed with SIGKILL still cannot clean up (no trap runs); a full local suite now leaves ~6 dirs instead of ~258.

Chemaclass · 2026-07-16T03:14:34Z

🤔 Background

Every invocation created $TMPDIR/bashunit/run/<os>/<random> (#801) but nothing removed it — one leaked directory per run (~9.9k had accumulated locally; the acceptance suite alone leaks ~258 per pass).

💡 Changes

Guarded cleanup in env.sh (path-checked like parallel::cleanup), armed via an EXIT trap for early-exit paths (--version, --help, subcommands).
The test-run path replaces that trap, so cmd_test calls the cleanup explicitly after rerun::persist; Ctrl-C and stop-on-failure handlers do the same.
Sourcing errors are now captured in the run dir too — files are sourced sequentially in the parent, so a fixed per-file-truncated path is safe and saves a mktemp + rm fork per test file.
Regression test runs a nested bashunit under a private TMPDIR and asserts nothing is left. SIGKILLed runs still can't clean up (no trap runs); a full suite now leaves ~6 dirs instead of ~258.

Every invocation created $TMPDIR/bashunit/run/<os>/<random> (#801) but nothing removed it, leaking one directory per run — ~9.9k dirs had accumulated locally. - env.sh gains a guarded cleanup (mirrors parallel::cleanup's path check) and arms it via an EXIT trap for early-exit paths (--version, --help, doc, ...). - The test-run path replaces that trap in main.sh, so cmd_test calls the cleanup explicitly after rerun::persist (which reads from the run dir), and the Ctrl-C / stop-on-failure handlers do the same before their exits. - Sourcing errors are captured in the run dir as well: files are sourced sequentially in the parent (workers fork after), so a fixed truncated-per-file path is safe and saves a mktemp and an rm fork per test file. Runs killed with SIGKILL still cannot clean up (no trap runs); a full local suite now leaves ~6 dirs instead of ~258.

Chemaclass · 2026-07-16T03:14:34Z

🤔 Background

Every invocation created $TMPDIR/bashunit/run/<os>/<random> (#801) but nothing removed it — one leaked directory per run (~9.9k had accumulated locally; the acceptance suite alone leaks ~258 per pass).

💡 Changes

Guarded cleanup in env.sh (path-checked like parallel::cleanup), armed via an EXIT trap for early-exit paths (--version, --help, subcommands).
The test-run path replaces that trap, so cmd_test calls the cleanup explicitly after rerun::persist; Ctrl-C and stop-on-failure handlers do the same.
Sourcing errors are now captured in the run dir too — files are sourced sequentially in the parent, so a fixed per-file-truncated path is safe and saves a mktemp + rm fork per test file.
Regression test runs a nested bashunit under a private TMPDIR and asserts nothing is left. SIGKILLed runs still can't clean up (no trap runs); a full suite now leaves ~6 dirs instead of ~258.

Every invocation created $TMPDIR/bashunit/run/<os>/<random> (#801) but nothing removed it, leaking one directory per run — ~9.9k dirs had accumulated locally. - env.sh gains a guarded cleanup (mirrors parallel::cleanup's path check) and arms it via an EXIT trap for early-exit paths (--version, --help, doc, ...). - The test-run path replaces that trap in main.sh, so cmd_test calls the cleanup explicitly after rerun::persist (which reads from the run dir), and the Ctrl-C / stop-on-failure handlers do the same before their exits. - Sourcing errors are captured in the run dir as well: files are sourced sequentially in the parent (workers fork after), so a fixed truncated-per-file path is safe and saves a mktemp and an rm fork per test file. Runs killed with SIGKILL still cannot clean up (no trap runs); a full local suite now leaves ~6 dirs instead of ~258.

Chemaclass · 2026-07-16T03:14:34Z

🤔 Background

Every invocation created $TMPDIR/bashunit/run/<os>/<random> (#801) but nothing removed it — one leaked directory per run (~9.9k had accumulated locally; the acceptance suite alone leaks ~258 per pass).

💡 Changes

Guarded cleanup in env.sh (path-checked like parallel::cleanup), armed via an EXIT trap for early-exit paths (--version, --help, subcommands).
The test-run path replaces that trap, so cmd_test calls the cleanup explicitly after rerun::persist; Ctrl-C and stop-on-failure handlers do the same.
Sourcing errors are now captured in the run dir too — files are sourced sequentially in the parent, so a fixed per-file-truncated path is safe and saves a mktemp + rm fork per test file.
Regression test runs a nested bashunit under a private TMPDIR and asserts nothing is left. SIGKILLed runs still can't clean up (no trap runs); a full suite now leaves ~6 dirs instead of ~258.

Every invocation created $TMPDIR/bashunit/run/<os>/<random> (#801) but nothing removed it, leaking one directory per run — ~9.9k dirs had accumulated locally. - env.sh gains a guarded cleanup (mirrors parallel::cleanup's path check) and arms it via an EXIT trap for early-exit paths (--version, --help, doc, ...). - The test-run path replaces that trap in main.sh, so cmd_test calls the cleanup explicitly after rerun::persist (which reads from the run dir), and the Ctrl-C / stop-on-failure handlers do the same before their exits. - Sourcing errors are captured in the run dir as well: files are sourced sequentially in the parent (workers fork after), so a fixed truncated-per-file path is safe and saves a mktemp and an rm fork per test file. Runs killed with SIGKILL still cannot clean up (no trap runs); a full local suite now leaves ~6 dirs instead of ~258.

Chemaclass · 2026-07-16T03:14:34Z

🤔 Background

Every invocation created $TMPDIR/bashunit/run/<os>/<random> (#801) but nothing removed it — one leaked directory per run (~9.9k had accumulated locally; the acceptance suite alone leaks ~258 per pass).

💡 Changes

Guarded cleanup in env.sh (path-checked like parallel::cleanup), armed via an EXIT trap for early-exit paths (--version, --help, subcommands).
The test-run path replaces that trap, so cmd_test calls the cleanup explicitly after rerun::persist; Ctrl-C and stop-on-failure handlers do the same.
Sourcing errors are now captured in the run dir too — files are sourced sequentially in the parent, so a fixed per-file-truncated path is safe and saves a mktemp + rm fork per test file.
Regression test runs a nested bashunit under a private TMPDIR and asserts nothing is left. SIGKILLed runs still can't clean up (no trap runs); a full suite now leaves ~6 dirs instead of ~258.

Every invocation created $TMPDIR/bashunit/run/<os>/<random> (#801) but nothing removed it, leaking one directory per run — ~9.9k dirs had accumulated locally. - env.sh gains a guarded cleanup (mirrors parallel::cleanup's path check) and arms it via an EXIT trap for early-exit paths (--version, --help, doc, ...). - The test-run path replaces that trap in main.sh, so cmd_test calls the cleanup explicitly after rerun::persist (which reads from the run dir), and the Ctrl-C / stop-on-failure handlers do the same before their exits. - Sourcing errors are captured in the run dir as well: files are sourced sequentially in the parent (workers fork after), so a fixed truncated-per-file path is safe and saves a mktemp and an rm fork per test file. Runs killed with SIGKILL still cannot clean up (no trap runs); a full local suite now leaves ~6 dirs instead of ~258.

Chemaclass added the bug Something isn't working label Jul 16, 2026

Chemaclass self-assigned this Jul 16, 2026

Chemaclass mentioned this pull request Jul 16, 2026
docs: add fork-budget rule with census method and measured limits #812
Merged