perf(coverage): compute every file's line stats in one awk invocation by Chemaclass · Pull Request #1089 · TypedDevs/bashunit

Chemaclass · 2026-08-13T05:36:06Z

🤔 Background

Related #1088

precompute_file_stats called compute_file_coverage once per tracked file,
reading each file into a Bash array and classifying it line by line, plus two
more subshells per file to derive the percentage and the class — the last
per-line Bash loop left in the report phase.

💡 Changes

One awk invocation walks a manifest of hits-block/source pairs with getline, so the fork is paid once per run instead of three times per file
Percentage and class fill return slots instead of forking, which the single-file path gets too
The per-file path stays as the fallback whenever the batch pass cannot run or yields nothing for a non-empty tracked list, so a report is never silently empty
2585 ms → 153 ms for 128 files; a --coverage run over src 6.81 s → 3.77 s (9.23 s before perf(coverage): scan function declarations in one awk pass #1085). LCOV and the text report are byte-identical on a fixed corpus

precompute_file_stats called compute_file_coverage once per tracked file, and each call read the whole file into a Bash array and classified it line by line, then took two more subshells to derive the percentage and the class. Three forks and a per-line loop per file: 2585ms for 128 files, the last per-line Bash loop in the report phase. One awk invocation now walks a manifest of hits-block and source pairs with getline, so the fork is paid once for the run: 153ms, and a --coverage run over src goes from 6.81s to 3.77s. Percentage and class fill return slots instead of forking, which the single-file path gets too. The per-file path stays and is used whenever the batch pass cannot run or produces nothing for a non-empty tracked list, so a report is never silently empty. LCOV and the text report are byte-identical on a fixed corpus. Closes#1088

Chemaclass · 2026-08-13T05:36:06Z

🤔 Background

Related #1088

precompute_file_stats called compute_file_coverage once per tracked file,
reading each file into a Bash array and classifying it line by line, plus two
more subshells per file to derive the percentage and the class — the last
per-line Bash loop left in the report phase.

💡 Changes

One awk invocation walks a manifest of hits-block/source pairs with getline, so the fork is paid once per run instead of three times per file
Percentage and class fill return slots instead of forking, which the single-file path gets too
The per-file path stays as the fallback whenever the batch pass cannot run or yields nothing for a non-empty tracked list, so a report is never silently empty
2585 ms → 153 ms for 128 files; a --coverage run over src 6.81 s → 3.77 s (9.23 s before perf(coverage): scan function declarations in one awk pass #1085). LCOV and the text report are byte-identical on a fixed corpus

precompute_file_stats called compute_file_coverage once per tracked file, and each call read the whole file into a Bash array and classified it line by line, then took two more subshells to derive the percentage and the class. Three forks and a per-line loop per file: 2585ms for 128 files, the last per-line Bash loop in the report phase. One awk invocation now walks a manifest of hits-block and source pairs with getline, so the fork is paid once for the run: 153ms, and a --coverage run over src goes from 6.81s to 3.77s. Percentage and class fill return slots instead of forking, which the single-file path gets too. The per-file path stays and is used whenever the batch pass cannot run or produces nothing for a non-empty tracked list, so a report is never silently empty. LCOV and the text report are byte-identical on a fixed corpus. Closes#1088

Chemaclass · 2026-08-13T05:36:06Z

🤔 Background

Related #1088

precompute_file_stats called compute_file_coverage once per tracked file,
reading each file into a Bash array and classifying it line by line, plus two
more subshells per file to derive the percentage and the class — the last
per-line Bash loop left in the report phase.

💡 Changes

One awk invocation walks a manifest of hits-block/source pairs with getline, so the fork is paid once per run instead of three times per file
Percentage and class fill return slots instead of forking, which the single-file path gets too
The per-file path stays as the fallback whenever the batch pass cannot run or yields nothing for a non-empty tracked list, so a report is never silently empty
2585 ms → 153 ms for 128 files; a --coverage run over src 6.81 s → 3.77 s (9.23 s before perf(coverage): scan function declarations in one awk pass #1085). LCOV and the text report are byte-identical on a fixed corpus

precompute_file_stats called compute_file_coverage once per tracked file, and each call read the whole file into a Bash array and classified it line by line, then took two more subshells to derive the percentage and the class. Three forks and a per-line loop per file: 2585ms for 128 files, the last per-line Bash loop in the report phase. One awk invocation now walks a manifest of hits-block and source pairs with getline, so the fork is paid once for the run: 153ms, and a --coverage run over src goes from 6.81s to 3.77s. Percentage and class fill return slots instead of forking, which the single-file path gets too. The per-file path stays and is used whenever the batch pass cannot run or produces nothing for a non-empty tracked list, so a report is never silently empty. LCOV and the text report are byte-identical on a fixed corpus. Closes#1088

Chemaclass · 2026-08-13T05:36:06Z

🤔 Background

Related #1088

precompute_file_stats called compute_file_coverage once per tracked file,
reading each file into a Bash array and classifying it line by line, plus two
more subshells per file to derive the percentage and the class — the last
per-line Bash loop left in the report phase.

💡 Changes

One awk invocation walks a manifest of hits-block/source pairs with getline, so the fork is paid once per run instead of three times per file
Percentage and class fill return slots instead of forking, which the single-file path gets too
The per-file path stays as the fallback whenever the batch pass cannot run or yields nothing for a non-empty tracked list, so a report is never silently empty
2585 ms → 153 ms for 128 files; a --coverage run over src 6.81 s → 3.77 s (9.23 s before perf(coverage): scan function declarations in one awk pass #1085). LCOV and the text report are byte-identical on a fixed corpus

precompute_file_stats called compute_file_coverage once per tracked file, and each call read the whole file into a Bash array and classified it line by line, then took two more subshells to derive the percentage and the class. Three forks and a per-line loop per file: 2585ms for 128 files, the last per-line Bash loop in the report phase. One awk invocation now walks a manifest of hits-block and source pairs with getline, so the fork is paid once for the run: 153ms, and a --coverage run over src goes from 6.81s to 3.77s. Percentage and class fill return slots instead of forking, which the single-file path gets too. The per-file path stays and is used whenever the batch pass cannot run or produces nothing for a non-empty tracked list, so a report is never silently empty. LCOV and the text report are byte-identical on a fixed corpus. Closes#1088

Chemaclass · 2026-08-13T05:36:06Z

🤔 Background

Related #1088

precompute_file_stats called compute_file_coverage once per tracked file,
reading each file into a Bash array and classifying it line by line, plus two
more subshells per file to derive the percentage and the class — the last
per-line Bash loop left in the report phase.

💡 Changes

One awk invocation walks a manifest of hits-block/source pairs with getline, so the fork is paid once per run instead of three times per file
Percentage and class fill return slots instead of forking, which the single-file path gets too
The per-file path stays as the fallback whenever the batch pass cannot run or yields nothing for a non-empty tracked list, so a report is never silently empty
2585 ms → 153 ms for 128 files; a --coverage run over src 6.81 s → 3.77 s (9.23 s before perf(coverage): scan function declarations in one awk pass #1085). LCOV and the text report are byte-identical on a fixed corpus

precompute_file_stats called compute_file_coverage once per tracked file, and each call read the whole file into a Bash array and classified it line by line, then took two more subshells to derive the percentage and the class. Three forks and a per-line loop per file: 2585ms for 128 files, the last per-line Bash loop in the report phase. One awk invocation now walks a manifest of hits-block and source pairs with getline, so the fork is paid once for the run: 153ms, and a --coverage run over src goes from 6.81s to 3.77s. Percentage and class fill return slots instead of forking, which the single-file path gets too. The per-file path stays and is used whenever the batch pass cannot run or produces nothing for a non-empty tracked list, so a report is never silently empty. LCOV and the text report are byte-identical on a fixed corpus. Closes#1088

Chemaclass · 2026-08-13T05:36:06Z

🤔 Background

Related #1088

precompute_file_stats called compute_file_coverage once per tracked file,
reading each file into a Bash array and classifying it line by line, plus two
more subshells per file to derive the percentage and the class — the last
per-line Bash loop left in the report phase.

💡 Changes

One awk invocation walks a manifest of hits-block/source pairs with getline, so the fork is paid once per run instead of three times per file
Percentage and class fill return slots instead of forking, which the single-file path gets too
The per-file path stays as the fallback whenever the batch pass cannot run or yields nothing for a non-empty tracked list, so a report is never silently empty
2585 ms → 153 ms for 128 files; a --coverage run over src 6.81 s → 3.77 s (9.23 s before perf(coverage): scan function declarations in one awk pass #1085). LCOV and the text report are byte-identical on a fixed corpus

precompute_file_stats called compute_file_coverage once per tracked file, and each call read the whole file into a Bash array and classified it line by line, then took two more subshells to derive the percentage and the class. Three forks and a per-line loop per file: 2585ms for 128 files, the last per-line Bash loop in the report phase. One awk invocation now walks a manifest of hits-block and source pairs with getline, so the fork is paid once for the run: 153ms, and a --coverage run over src goes from 6.81s to 3.77s. Percentage and class fill return slots instead of forking, which the single-file path gets too. The per-file path stays and is used whenever the batch pass cannot run or produces nothing for a non-empty tracked list, so a report is never silently empty. LCOV and the text report are byte-identical on a fixed corpus. Closes#1088

Chemaclass · 2026-08-13T05:36:06Z

🤔 Background

Related #1088

precompute_file_stats called compute_file_coverage once per tracked file,
reading each file into a Bash array and classifying it line by line, plus two
more subshells per file to derive the percentage and the class — the last
per-line Bash loop left in the report phase.

💡 Changes

One awk invocation walks a manifest of hits-block/source pairs with getline, so the fork is paid once per run instead of three times per file
Percentage and class fill return slots instead of forking, which the single-file path gets too
The per-file path stays as the fallback whenever the batch pass cannot run or yields nothing for a non-empty tracked list, so a report is never silently empty
2585 ms → 153 ms for 128 files; a --coverage run over src 6.81 s → 3.77 s (9.23 s before perf(coverage): scan function declarations in one awk pass #1085). LCOV and the text report are byte-identical on a fixed corpus

precompute_file_stats called compute_file_coverage once per tracked file, and each call read the whole file into a Bash array and classified it line by line, then took two more subshells to derive the percentage and the class. Three forks and a per-line loop per file: 2585ms for 128 files, the last per-line Bash loop in the report phase. One awk invocation now walks a manifest of hits-block and source pairs with getline, so the fork is paid once for the run: 153ms, and a --coverage run over src goes from 6.81s to 3.77s. Percentage and class fill return slots instead of forking, which the single-file path gets too. The per-file path stays and is used whenever the batch pass cannot run or produces nothing for a non-empty tracked list, so a report is never silently empty. LCOV and the text report are byte-identical on a fixed corpus. Closes#1088

Chemaclass · 2026-08-13T05:36:06Z

🤔 Background

Related #1088

precompute_file_stats called compute_file_coverage once per tracked file,
reading each file into a Bash array and classifying it line by line, plus two
more subshells per file to derive the percentage and the class — the last
per-line Bash loop left in the report phase.

💡 Changes

One awk invocation walks a manifest of hits-block/source pairs with getline, so the fork is paid once per run instead of three times per file
Percentage and class fill return slots instead of forking, which the single-file path gets too
The per-file path stays as the fallback whenever the batch pass cannot run or yields nothing for a non-empty tracked list, so a report is never silently empty
2585 ms → 153 ms for 128 files; a --coverage run over src 6.81 s → 3.77 s (9.23 s before perf(coverage): scan function declarations in one awk pass #1085). LCOV and the text report are byte-identical on a fixed corpus

precompute_file_stats called compute_file_coverage once per tracked file, and each call read the whole file into a Bash array and classified it line by line, then took two more subshells to derive the percentage and the class. Three forks and a per-line loop per file: 2585ms for 128 files, the last per-line Bash loop in the report phase. One awk invocation now walks a manifest of hits-block and source pairs with getline, so the fork is paid once for the run: 153ms, and a --coverage run over src goes from 6.81s to 3.77s. Percentage and class fill return slots instead of forking, which the single-file path gets too. The per-file path stays and is used whenever the batch pass cannot run or produces nothing for a non-empty tracked list, so a report is never silently empty. LCOV and the text report are byte-identical on a fixed corpus. Closes#1088

Chemaclass added the enhancement New feature or request label Aug 13, 2026

Chemaclass self-assigned this Aug 13, 2026

Chemaclass merged commit eac53cc into mainAug 13, 2026
37 checks passed

Chemaclass deleted the perf/1088-one-awk-precompute branch August 13, 2026 05:43