perf(runner): order test functions in pure bash, drop 5 forks per file by Chemaclass · Pull Request #809 · TypedDevs/bashunit

Chemaclass · 2026-07-16T02:51:34Z

🤔 Background

Continuing the fork-reduction series (#801-#808). Ground-truth profiling with a PATH shim (bash -x trace counts overcount by re-echoed test output) showed 7 awk + 2 sort forks per test file, dominated by discovery: functions_for_script ran an awk | sort | awk pipeline twice per file, and the duplicate check piped awk through sort.

💡 Changes

functions_for_script: filter + insertion-sort the file's functions by definition line in pure bash (a file's function list is tens of names); extdebug now enabled only inside the capture subshell, leaving the caller's setting untouched.
check_duplicate_functions: sort the (usually empty) duplicate list inside the awk END block instead of piping through sort.
9 → 5 forks per test file; shim-based regression test + unit tests for ordering, cross-file filtering, and extdebug preservation. No behaviour change.

Two per-file fork clusters in discovery became pure bash / awk-internal: - functions_for_script piped 'declare -F' through awk | sort | awk (three forks) and the pipeline ran twice per file. A file's function list is small (tens of names), so filter + insertion-sort by definition line in bash — and enable extdebug only inside the capture subshell so the caller's setting is untouched. - check_duplicate_functions piped its awk pass through 'sort' just to keep the (usually empty) duplicate list deterministic; the awk END block sorts it now. 9 -> 5 forks per test file (awk 7 -> 5, sort 2 -> 0), on every file of every run including the acceptance suite's ~258 nested ones. Measured with a PATH shim: bash -x trace counts also pick up re-echoed test output and overcount. No behaviour change.

Chemaclass · 2026-07-16T02:51:34Z

🤔 Background

Continuing the fork-reduction series (#801-#808). Ground-truth profiling with a PATH shim (bash -x trace counts overcount by re-echoed test output) showed 7 awk + 2 sort forks per test file, dominated by discovery: functions_for_script ran an awk | sort | awk pipeline twice per file, and the duplicate check piped awk through sort.

💡 Changes

functions_for_script: filter + insertion-sort the file's functions by definition line in pure bash (a file's function list is tens of names); extdebug now enabled only inside the capture subshell, leaving the caller's setting untouched.
check_duplicate_functions: sort the (usually empty) duplicate list inside the awk END block instead of piping through sort.
9 → 5 forks per test file; shim-based regression test + unit tests for ordering, cross-file filtering, and extdebug preservation. No behaviour change.

Two per-file fork clusters in discovery became pure bash / awk-internal: - functions_for_script piped 'declare -F' through awk | sort | awk (three forks) and the pipeline ran twice per file. A file's function list is small (tens of names), so filter + insertion-sort by definition line in bash — and enable extdebug only inside the capture subshell so the caller's setting is untouched. - check_duplicate_functions piped its awk pass through 'sort' just to keep the (usually empty) duplicate list deterministic; the awk END block sorts it now. 9 -> 5 forks per test file (awk 7 -> 5, sort 2 -> 0), on every file of every run including the acceptance suite's ~258 nested ones. Measured with a PATH shim: bash -x trace counts also pick up re-echoed test output and overcount. No behaviour change.

Chemaclass · 2026-07-16T02:51:34Z

🤔 Background

Continuing the fork-reduction series (#801-#808). Ground-truth profiling with a PATH shim (bash -x trace counts overcount by re-echoed test output) showed 7 awk + 2 sort forks per test file, dominated by discovery: functions_for_script ran an awk | sort | awk pipeline twice per file, and the duplicate check piped awk through sort.

💡 Changes

functions_for_script: filter + insertion-sort the file's functions by definition line in pure bash (a file's function list is tens of names); extdebug now enabled only inside the capture subshell, leaving the caller's setting untouched.
check_duplicate_functions: sort the (usually empty) duplicate list inside the awk END block instead of piping through sort.
9 → 5 forks per test file; shim-based regression test + unit tests for ordering, cross-file filtering, and extdebug preservation. No behaviour change.

Two per-file fork clusters in discovery became pure bash / awk-internal: - functions_for_script piped 'declare -F' through awk | sort | awk (three forks) and the pipeline ran twice per file. A file's function list is small (tens of names), so filter + insertion-sort by definition line in bash — and enable extdebug only inside the capture subshell so the caller's setting is untouched. - check_duplicate_functions piped its awk pass through 'sort' just to keep the (usually empty) duplicate list deterministic; the awk END block sorts it now. 9 -> 5 forks per test file (awk 7 -> 5, sort 2 -> 0), on every file of every run including the acceptance suite's ~258 nested ones. Measured with a PATH shim: bash -x trace counts also pick up re-echoed test output and overcount. No behaviour change.

Chemaclass · 2026-07-16T02:51:34Z

🤔 Background

Continuing the fork-reduction series (#801-#808). Ground-truth profiling with a PATH shim (bash -x trace counts overcount by re-echoed test output) showed 7 awk + 2 sort forks per test file, dominated by discovery: functions_for_script ran an awk | sort | awk pipeline twice per file, and the duplicate check piped awk through sort.

💡 Changes

functions_for_script: filter + insertion-sort the file's functions by definition line in pure bash (a file's function list is tens of names); extdebug now enabled only inside the capture subshell, leaving the caller's setting untouched.
check_duplicate_functions: sort the (usually empty) duplicate list inside the awk END block instead of piping through sort.
9 → 5 forks per test file; shim-based regression test + unit tests for ordering, cross-file filtering, and extdebug preservation. No behaviour change.

Two per-file fork clusters in discovery became pure bash / awk-internal: - functions_for_script piped 'declare -F' through awk | sort | awk (three forks) and the pipeline ran twice per file. A file's function list is small (tens of names), so filter + insertion-sort by definition line in bash — and enable extdebug only inside the capture subshell so the caller's setting is untouched. - check_duplicate_functions piped its awk pass through 'sort' just to keep the (usually empty) duplicate list deterministic; the awk END block sorts it now. 9 -> 5 forks per test file (awk 7 -> 5, sort 2 -> 0), on every file of every run including the acceptance suite's ~258 nested ones. Measured with a PATH shim: bash -x trace counts also pick up re-echoed test output and overcount. No behaviour change.

Chemaclass · 2026-07-16T02:51:34Z

🤔 Background

Continuing the fork-reduction series (#801-#808). Ground-truth profiling with a PATH shim (bash -x trace counts overcount by re-echoed test output) showed 7 awk + 2 sort forks per test file, dominated by discovery: functions_for_script ran an awk | sort | awk pipeline twice per file, and the duplicate check piped awk through sort.

💡 Changes

functions_for_script: filter + insertion-sort the file's functions by definition line in pure bash (a file's function list is tens of names); extdebug now enabled only inside the capture subshell, leaving the caller's setting untouched.
check_duplicate_functions: sort the (usually empty) duplicate list inside the awk END block instead of piping through sort.
9 → 5 forks per test file; shim-based regression test + unit tests for ordering, cross-file filtering, and extdebug preservation. No behaviour change.

Two per-file fork clusters in discovery became pure bash / awk-internal: - functions_for_script piped 'declare -F' through awk | sort | awk (three forks) and the pipeline ran twice per file. A file's function list is small (tens of names), so filter + insertion-sort by definition line in bash — and enable extdebug only inside the capture subshell so the caller's setting is untouched. - check_duplicate_functions piped its awk pass through 'sort' just to keep the (usually empty) duplicate list deterministic; the awk END block sorts it now. 9 -> 5 forks per test file (awk 7 -> 5, sort 2 -> 0), on every file of every run including the acceptance suite's ~258 nested ones. Measured with a PATH shim: bash -x trace counts also pick up re-echoed test output and overcount. No behaviour change.

Chemaclass · 2026-07-16T02:51:34Z

🤔 Background

Continuing the fork-reduction series (#801-#808). Ground-truth profiling with a PATH shim (bash -x trace counts overcount by re-echoed test output) showed 7 awk + 2 sort forks per test file, dominated by discovery: functions_for_script ran an awk | sort | awk pipeline twice per file, and the duplicate check piped awk through sort.

💡 Changes

functions_for_script: filter + insertion-sort the file's functions by definition line in pure bash (a file's function list is tens of names); extdebug now enabled only inside the capture subshell, leaving the caller's setting untouched.
check_duplicate_functions: sort the (usually empty) duplicate list inside the awk END block instead of piping through sort.
9 → 5 forks per test file; shim-based regression test + unit tests for ordering, cross-file filtering, and extdebug preservation. No behaviour change.

Two per-file fork clusters in discovery became pure bash / awk-internal: - functions_for_script piped 'declare -F' through awk | sort | awk (three forks) and the pipeline ran twice per file. A file's function list is small (tens of names), so filter + insertion-sort by definition line in bash — and enable extdebug only inside the capture subshell so the caller's setting is untouched. - check_duplicate_functions piped its awk pass through 'sort' just to keep the (usually empty) duplicate list deterministic; the awk END block sorts it now. 9 -> 5 forks per test file (awk 7 -> 5, sort 2 -> 0), on every file of every run including the acceptance suite's ~258 nested ones. Measured with a PATH shim: bash -x trace counts also pick up re-echoed test output and overcount. No behaviour change.

Chemaclass · 2026-07-16T02:51:34Z

🤔 Background

Continuing the fork-reduction series (#801-#808). Ground-truth profiling with a PATH shim (bash -x trace counts overcount by re-echoed test output) showed 7 awk + 2 sort forks per test file, dominated by discovery: functions_for_script ran an awk | sort | awk pipeline twice per file, and the duplicate check piped awk through sort.

💡 Changes

functions_for_script: filter + insertion-sort the file's functions by definition line in pure bash (a file's function list is tens of names); extdebug now enabled only inside the capture subshell, leaving the caller's setting untouched.
check_duplicate_functions: sort the (usually empty) duplicate list inside the awk END block instead of piping through sort.
9 → 5 forks per test file; shim-based regression test + unit tests for ordering, cross-file filtering, and extdebug preservation. No behaviour change.

Two per-file fork clusters in discovery became pure bash / awk-internal: - functions_for_script piped 'declare -F' through awk | sort | awk (three forks) and the pipeline ran twice per file. A file's function list is small (tens of names), so filter + insertion-sort by definition line in bash — and enable extdebug only inside the capture subshell so the caller's setting is untouched. - check_duplicate_functions piped its awk pass through 'sort' just to keep the (usually empty) duplicate list deterministic; the awk END block sorts it now. 9 -> 5 forks per test file (awk 7 -> 5, sort 2 -> 0), on every file of every run including the acceptance suite's ~258 nested ones. Measured with a PATH shim: bash -x trace counts also pick up re-echoed test output and overcount. No behaviour change.

Chemaclass added the enhancement New feature or request label Jul 16, 2026

Chemaclass self-assigned this Jul 16, 2026

Chemaclass merged commit 0ee8c62 into mainJul 16, 2026
37 checks passed

Chemaclass deleted the perf/functions-for-script-pure-bash branch July 16, 2026 02:57