perf(runner): auto-skip per-test timing when the clock forks an interpreter by Chemaclass · Pull Request #769 · TypedDevs/bashunit

Chemaclass · 2026-07-11T13:21:05Z

🤔 Background

Related #765

On shells without a native sub-second clock (Bash 3.2 on macOS), bashunit::clock::now falls back to forking a perl interpreter, so every test pays two perl forks just to measure a duration that a plain run then discards.

💡 Changes

Default BASHUNIT_SHOW_EXECUTION_TIME to auto: per-test times show when the clock is fork-free and hide when measuring would fork an interpreter.
Skip both per-test clock reads unless a consumer needs the number (--profile, --verbose, a report, or an explicit execution-time display).
Keep the total "Time taken" footer visible in auto; only false hides it.
Zero behavior change on cheap-clock platforms (Bash 5 / Linux CI); documented in the CHANGELOG, docs, and ADR-008.

…preter (#765) Default BASHUNIT_SHOW_EXECUTION_TIME to `auto`: measure per-test duration only when a consumer needs it (--profile, --verbose, a report, or an explicit execution-time display). On shells whose clock forks an interpreter (Bash 3.2 macOS -> perl), a plain run now skips both clock reads, removing ~2 perl forks per test. - clock::is_expensive classifies the resolved impl (perl/python/node/ powershell = expensive) - env::is_show_execution_time_enabled resolves auto via the clock cost; env::is_total_execution_time_enabled keeps the footer visible unless explicitly false - runner::needs_test_duration gates the clock reads behind the single set of duration consumers - reports::is_enabled extracted so the predicate can reuse it See adrs/adr-008-auto-skip-per-test-timing.md

Chemaclass · 2026-07-11T13:21:05Z

🤔 Background

Related #765

On shells without a native sub-second clock (Bash 3.2 on macOS), bashunit::clock::now falls back to forking a perl interpreter, so every test pays two perl forks just to measure a duration that a plain run then discards.

💡 Changes

Default BASHUNIT_SHOW_EXECUTION_TIME to auto: per-test times show when the clock is fork-free and hide when measuring would fork an interpreter.
Skip both per-test clock reads unless a consumer needs the number (--profile, --verbose, a report, or an explicit execution-time display).
Keep the total "Time taken" footer visible in auto; only false hides it.
Zero behavior change on cheap-clock platforms (Bash 5 / Linux CI); documented in the CHANGELOG, docs, and ADR-008.

…preter (#765) Default BASHUNIT_SHOW_EXECUTION_TIME to `auto`: measure per-test duration only when a consumer needs it (--profile, --verbose, a report, or an explicit execution-time display). On shells whose clock forks an interpreter (Bash 3.2 macOS -> perl), a plain run now skips both clock reads, removing ~2 perl forks per test. - clock::is_expensive classifies the resolved impl (perl/python/node/ powershell = expensive) - env::is_show_execution_time_enabled resolves auto via the clock cost; env::is_total_execution_time_enabled keeps the footer visible unless explicitly false - runner::needs_test_duration gates the clock reads behind the single set of duration consumers - reports::is_enabled extracted so the predicate can reuse it See adrs/adr-008-auto-skip-per-test-timing.md

Chemaclass · 2026-07-11T13:21:05Z

🤔 Background

Related #765

On shells without a native sub-second clock (Bash 3.2 on macOS), bashunit::clock::now falls back to forking a perl interpreter, so every test pays two perl forks just to measure a duration that a plain run then discards.

💡 Changes

Default BASHUNIT_SHOW_EXECUTION_TIME to auto: per-test times show when the clock is fork-free and hide when measuring would fork an interpreter.
Skip both per-test clock reads unless a consumer needs the number (--profile, --verbose, a report, or an explicit execution-time display).
Keep the total "Time taken" footer visible in auto; only false hides it.
Zero behavior change on cheap-clock platforms (Bash 5 / Linux CI); documented in the CHANGELOG, docs, and ADR-008.

…preter (#765) Default BASHUNIT_SHOW_EXECUTION_TIME to `auto`: measure per-test duration only when a consumer needs it (--profile, --verbose, a report, or an explicit execution-time display). On shells whose clock forks an interpreter (Bash 3.2 macOS -> perl), a plain run now skips both clock reads, removing ~2 perl forks per test. - clock::is_expensive classifies the resolved impl (perl/python/node/ powershell = expensive) - env::is_show_execution_time_enabled resolves auto via the clock cost; env::is_total_execution_time_enabled keeps the footer visible unless explicitly false - runner::needs_test_duration gates the clock reads behind the single set of duration consumers - reports::is_enabled extracted so the predicate can reuse it See adrs/adr-008-auto-skip-per-test-timing.md

Chemaclass · 2026-07-11T13:21:05Z

🤔 Background

Related #765

On shells without a native sub-second clock (Bash 3.2 on macOS), bashunit::clock::now falls back to forking a perl interpreter, so every test pays two perl forks just to measure a duration that a plain run then discards.

💡 Changes

Default BASHUNIT_SHOW_EXECUTION_TIME to auto: per-test times show when the clock is fork-free and hide when measuring would fork an interpreter.
Skip both per-test clock reads unless a consumer needs the number (--profile, --verbose, a report, or an explicit execution-time display).
Keep the total "Time taken" footer visible in auto; only false hides it.
Zero behavior change on cheap-clock platforms (Bash 5 / Linux CI); documented in the CHANGELOG, docs, and ADR-008.

…preter (#765) Default BASHUNIT_SHOW_EXECUTION_TIME to `auto`: measure per-test duration only when a consumer needs it (--profile, --verbose, a report, or an explicit execution-time display). On shells whose clock forks an interpreter (Bash 3.2 macOS -> perl), a plain run now skips both clock reads, removing ~2 perl forks per test. - clock::is_expensive classifies the resolved impl (perl/python/node/ powershell = expensive) - env::is_show_execution_time_enabled resolves auto via the clock cost; env::is_total_execution_time_enabled keeps the footer visible unless explicitly false - runner::needs_test_duration gates the clock reads behind the single set of duration consumers - reports::is_enabled extracted so the predicate can reuse it See adrs/adr-008-auto-skip-per-test-timing.md

Chemaclass · 2026-07-11T13:21:05Z

🤔 Background

Related #765

On shells without a native sub-second clock (Bash 3.2 on macOS), bashunit::clock::now falls back to forking a perl interpreter, so every test pays two perl forks just to measure a duration that a plain run then discards.

💡 Changes

Default BASHUNIT_SHOW_EXECUTION_TIME to auto: per-test times show when the clock is fork-free and hide when measuring would fork an interpreter.
Skip both per-test clock reads unless a consumer needs the number (--profile, --verbose, a report, or an explicit execution-time display).
Keep the total "Time taken" footer visible in auto; only false hides it.
Zero behavior change on cheap-clock platforms (Bash 5 / Linux CI); documented in the CHANGELOG, docs, and ADR-008.

…preter (#765) Default BASHUNIT_SHOW_EXECUTION_TIME to `auto`: measure per-test duration only when a consumer needs it (--profile, --verbose, a report, or an explicit execution-time display). On shells whose clock forks an interpreter (Bash 3.2 macOS -> perl), a plain run now skips both clock reads, removing ~2 perl forks per test. - clock::is_expensive classifies the resolved impl (perl/python/node/ powershell = expensive) - env::is_show_execution_time_enabled resolves auto via the clock cost; env::is_total_execution_time_enabled keeps the footer visible unless explicitly false - runner::needs_test_duration gates the clock reads behind the single set of duration consumers - reports::is_enabled extracted so the predicate can reuse it See adrs/adr-008-auto-skip-per-test-timing.md

Chemaclass · 2026-07-11T13:21:05Z

🤔 Background

Related #765

On shells without a native sub-second clock (Bash 3.2 on macOS), bashunit::clock::now falls back to forking a perl interpreter, so every test pays two perl forks just to measure a duration that a plain run then discards.

💡 Changes

Default BASHUNIT_SHOW_EXECUTION_TIME to auto: per-test times show when the clock is fork-free and hide when measuring would fork an interpreter.
Skip both per-test clock reads unless a consumer needs the number (--profile, --verbose, a report, or an explicit execution-time display).
Keep the total "Time taken" footer visible in auto; only false hides it.
Zero behavior change on cheap-clock platforms (Bash 5 / Linux CI); documented in the CHANGELOG, docs, and ADR-008.

…preter (#765) Default BASHUNIT_SHOW_EXECUTION_TIME to `auto`: measure per-test duration only when a consumer needs it (--profile, --verbose, a report, or an explicit execution-time display). On shells whose clock forks an interpreter (Bash 3.2 macOS -> perl), a plain run now skips both clock reads, removing ~2 perl forks per test. - clock::is_expensive classifies the resolved impl (perl/python/node/ powershell = expensive) - env::is_show_execution_time_enabled resolves auto via the clock cost; env::is_total_execution_time_enabled keeps the footer visible unless explicitly false - runner::needs_test_duration gates the clock reads behind the single set of duration consumers - reports::is_enabled extracted so the predicate can reuse it See adrs/adr-008-auto-skip-per-test-timing.md

Chemaclass · 2026-07-11T13:21:05Z

🤔 Background

Related #765

On shells without a native sub-second clock (Bash 3.2 on macOS), bashunit::clock::now falls back to forking a perl interpreter, so every test pays two perl forks just to measure a duration that a plain run then discards.

💡 Changes

Default BASHUNIT_SHOW_EXECUTION_TIME to auto: per-test times show when the clock is fork-free and hide when measuring would fork an interpreter.
Skip both per-test clock reads unless a consumer needs the number (--profile, --verbose, a report, or an explicit execution-time display).
Keep the total "Time taken" footer visible in auto; only false hides it.
Zero behavior change on cheap-clock platforms (Bash 5 / Linux CI); documented in the CHANGELOG, docs, and ADR-008.

…preter (#765) Default BASHUNIT_SHOW_EXECUTION_TIME to `auto`: measure per-test duration only when a consumer needs it (--profile, --verbose, a report, or an explicit execution-time display). On shells whose clock forks an interpreter (Bash 3.2 macOS -> perl), a plain run now skips both clock reads, removing ~2 perl forks per test. - clock::is_expensive classifies the resolved impl (perl/python/node/ powershell = expensive) - env::is_show_execution_time_enabled resolves auto via the clock cost; env::is_total_execution_time_enabled keeps the footer visible unless explicitly false - runner::needs_test_duration gates the clock reads behind the single set of duration consumers - reports::is_enabled extracted so the predicate can reuse it See adrs/adr-008-auto-skip-per-test-timing.md

Chemaclass · 2026-07-11T13:21:05Z

🤔 Background

Related #765

On shells without a native sub-second clock (Bash 3.2 on macOS), bashunit::clock::now falls back to forking a perl interpreter, so every test pays two perl forks just to measure a duration that a plain run then discards.

💡 Changes

Default BASHUNIT_SHOW_EXECUTION_TIME to auto: per-test times show when the clock is fork-free and hide when measuring would fork an interpreter.
Skip both per-test clock reads unless a consumer needs the number (--profile, --verbose, a report, or an explicit execution-time display).
Keep the total "Time taken" footer visible in auto; only false hides it.
Zero behavior change on cheap-clock platforms (Bash 5 / Linux CI); documented in the CHANGELOG, docs, and ADR-008.

…preter (#765) Default BASHUNIT_SHOW_EXECUTION_TIME to `auto`: measure per-test duration only when a consumer needs it (--profile, --verbose, a report, or an explicit execution-time display). On shells whose clock forks an interpreter (Bash 3.2 macOS -> perl), a plain run now skips both clock reads, removing ~2 perl forks per test. - clock::is_expensive classifies the resolved impl (perl/python/node/ powershell = expensive) - env::is_show_execution_time_enabled resolves auto via the clock cost; env::is_total_execution_time_enabled keeps the footer visible unless explicitly false - runner::needs_test_duration gates the clock reads behind the single set of duration consumers - reports::is_enabled extracted so the predicate can reuse it See adrs/adr-008-auto-skip-per-test-timing.md