refactor(src): move module aggregators inside their directory as index.sh by Chemaclass · Pull Request #934 · TypedDevs/bashunit

Chemaclass · 2026-07-31T23:18:55Z

🤔 Background

Related #931

ADR-010 put each module's aggregator beside its directory (src/runner.sh + src/runner/*.sh). This amends that decision: the entry point moves inside as index.sh, matching what most ecosystems do (index.ts, __init__.py, mod.rs).

The original reasoning was wrong. It rested on matching the src/assertions.sh → src/assert_*.sh precedent — but assertions.sh aggregates flat files and has no directory. It is not a module aggregator, it is unaffected either way, and the "two conventions in src/" risk it was meant to avoid never existed.

💡 Changes

src/runner.sh → src/runner/index.sh, src/coverage.sh → src/coverage/index.sh; git records both as pure renames (1 line changed each — a reworded header comment).
Fixes a live correctness gap. The "aggregators hold only source lines" rule was enforced against a hand-maintained list, "src/assertions.sh src/runner.sh". refactor(coverage): split src/coverage.sh into a src/coverage/ module #928 added src/coverage.sh and never appended it, so the rule silently stopped covering that file within one module of being introduced. With the entry point at a predictable src/*/index.sh, the test now discovers modules by glob and cannot drift — plus a test asserting the discovery finds every module.
Amends ADR-010 with a dated amendment: chosen option flipped, the withdrawn precedent argument struck through rather than deleted, both options' pros/cons rewritten.
Corrects two references refactor(coverage): split src/coverage.sh into a src/coverage/ module #928 left stale: the branch-helper dynamic-scope note in bash-style.md pointed at src/coverage.sh:818-821, and the unreleased CHANGELOG entry described a shape that no longer ships.

Doing this now costs two renames. #931 adds nine more modules, after which it costs eleven.

✅ Verification

The distributable is unchanged. Diffing the built artifact before and after shows only the two renamed embed markers and the two reworded header comments — same 17627 lines.

The glob test was confirmed to catch what the hardcoded list could not: appending a statement to src/coverage/index.sh was previously green, now red.

Green: sequential · --parallel · --parallel --simple --strict · make sa · make lint · CI-mode ShellCheck (per file, no -x) · bash build.sh bin -v → ✅ Build verified ✅.

…x.sh `src/runner.sh` -> `src/runner/index.sh`, `src/coverage.sh` -> `src/coverage/index.sh`. A module is now one self-contained directory whose entry point is where every other ecosystem puts it (index.ts, __init__.py, mod.rs). Amends ADR-010, which chose "beside" two days ago. The reasoning that first chose "beside" was wrong. It rested on matching the `src/assertions.sh` -> `src/assert_*.sh` precedent, but assertions.sh aggregates *flat* files and has no directory: it is not a module aggregator, it is unaffected by this choice, and the "two conventions in src/" risk it was meant to avoid never existed. What actually decides it is a correctness bug. The rule that aggregators hold only `source` lines is enforced by test_module_aggregators_hold_only_source_lines_and_comments, which named them in a hand-maintained string: "src/assertions.sh src/runner.sh". #928 added src/coverage.sh and never appended it, so the rule silently stopped covering that file within one module of being introduced. With the entry point at a predictable src/*/index.sh the test discovers modules by glob and cannot drift; this commit converts it, and adds a test asserting the discovery finds every module. Verified by appending a statement to src/coverage/index.sh -- previously green, now red. Doing it now costs two renames. #931 adds nine more modules, after which it would cost eleven. The distributable is unchanged: diffing the built artifact before and after shows only the two renamed embed markers and two reworded header comments, same 17627 lines. `bash build.sh bin -v` prints "Build verified". Also corrects two references #928 left stale: the branch-helper dynamic-scope note in bash-style.md pointed at `src/coverage.sh:818-821`, and the unreleased CHANGELOG entry described a shape that no longer ships. Related #931

Chemaclass · 2026-07-31T23:18:55Z

🤔 Background

Related #931

ADR-010 put each module's aggregator beside its directory (src/runner.sh + src/runner/*.sh). This amends that decision: the entry point moves inside as index.sh, matching what most ecosystems do (index.ts, __init__.py, mod.rs).

The original reasoning was wrong. It rested on matching the src/assertions.sh → src/assert_*.sh precedent — but assertions.sh aggregates flat files and has no directory. It is not a module aggregator, it is unaffected either way, and the "two conventions in src/" risk it was meant to avoid never existed.

💡 Changes

src/runner.sh → src/runner/index.sh, src/coverage.sh → src/coverage/index.sh; git records both as pure renames (1 line changed each — a reworded header comment).
Fixes a live correctness gap. The "aggregators hold only source lines" rule was enforced against a hand-maintained list, "src/assertions.sh src/runner.sh". refactor(coverage): split src/coverage.sh into a src/coverage/ module #928 added src/coverage.sh and never appended it, so the rule silently stopped covering that file within one module of being introduced. With the entry point at a predictable src/*/index.sh, the test now discovers modules by glob and cannot drift — plus a test asserting the discovery finds every module.
Amends ADR-010 with a dated amendment: chosen option flipped, the withdrawn precedent argument struck through rather than deleted, both options' pros/cons rewritten.
Corrects two references refactor(coverage): split src/coverage.sh into a src/coverage/ module #928 left stale: the branch-helper dynamic-scope note in bash-style.md pointed at src/coverage.sh:818-821, and the unreleased CHANGELOG entry described a shape that no longer ships.

Doing this now costs two renames. #931 adds nine more modules, after which it costs eleven.

✅ Verification

The distributable is unchanged. Diffing the built artifact before and after shows only the two renamed embed markers and the two reworded header comments — same 17627 lines.

The glob test was confirmed to catch what the hardcoded list could not: appending a statement to src/coverage/index.sh was previously green, now red.

Green: sequential · --parallel · --parallel --simple --strict · make sa · make lint · CI-mode ShellCheck (per file, no -x) · bash build.sh bin -v → ✅ Build verified ✅.

…x.sh `src/runner.sh` -> `src/runner/index.sh`, `src/coverage.sh` -> `src/coverage/index.sh`. A module is now one self-contained directory whose entry point is where every other ecosystem puts it (index.ts, __init__.py, mod.rs). Amends ADR-010, which chose "beside" two days ago. The reasoning that first chose "beside" was wrong. It rested on matching the `src/assertions.sh` -> `src/assert_*.sh` precedent, but assertions.sh aggregates *flat* files and has no directory: it is not a module aggregator, it is unaffected by this choice, and the "two conventions in src/" risk it was meant to avoid never existed. What actually decides it is a correctness bug. The rule that aggregators hold only `source` lines is enforced by test_module_aggregators_hold_only_source_lines_and_comments, which named them in a hand-maintained string: "src/assertions.sh src/runner.sh". #928 added src/coverage.sh and never appended it, so the rule silently stopped covering that file within one module of being introduced. With the entry point at a predictable src/*/index.sh the test discovers modules by glob and cannot drift; this commit converts it, and adds a test asserting the discovery finds every module. Verified by appending a statement to src/coverage/index.sh -- previously green, now red. Doing it now costs two renames. #931 adds nine more modules, after which it would cost eleven. The distributable is unchanged: diffing the built artifact before and after shows only the two renamed embed markers and two reworded header comments, same 17627 lines. `bash build.sh bin -v` prints "Build verified". Also corrects two references #928 left stale: the branch-helper dynamic-scope note in bash-style.md pointed at `src/coverage.sh:818-821`, and the unreleased CHANGELOG entry described a shape that no longer ships. Related #931

Chemaclass · 2026-07-31T23:18:55Z

🤔 Background

Related #931

ADR-010 put each module's aggregator beside its directory (src/runner.sh + src/runner/*.sh). This amends that decision: the entry point moves inside as index.sh, matching what most ecosystems do (index.ts, __init__.py, mod.rs).

The original reasoning was wrong. It rested on matching the src/assertions.sh → src/assert_*.sh precedent — but assertions.sh aggregates flat files and has no directory. It is not a module aggregator, it is unaffected either way, and the "two conventions in src/" risk it was meant to avoid never existed.

💡 Changes

src/runner.sh → src/runner/index.sh, src/coverage.sh → src/coverage/index.sh; git records both as pure renames (1 line changed each — a reworded header comment).
Fixes a live correctness gap. The "aggregators hold only source lines" rule was enforced against a hand-maintained list, "src/assertions.sh src/runner.sh". refactor(coverage): split src/coverage.sh into a src/coverage/ module #928 added src/coverage.sh and never appended it, so the rule silently stopped covering that file within one module of being introduced. With the entry point at a predictable src/*/index.sh, the test now discovers modules by glob and cannot drift — plus a test asserting the discovery finds every module.
Amends ADR-010 with a dated amendment: chosen option flipped, the withdrawn precedent argument struck through rather than deleted, both options' pros/cons rewritten.
Corrects two references refactor(coverage): split src/coverage.sh into a src/coverage/ module #928 left stale: the branch-helper dynamic-scope note in bash-style.md pointed at src/coverage.sh:818-821, and the unreleased CHANGELOG entry described a shape that no longer ships.

Doing this now costs two renames. #931 adds nine more modules, after which it costs eleven.

✅ Verification

The distributable is unchanged. Diffing the built artifact before and after shows only the two renamed embed markers and the two reworded header comments — same 17627 lines.

The glob test was confirmed to catch what the hardcoded list could not: appending a statement to src/coverage/index.sh was previously green, now red.

Green: sequential · --parallel · --parallel --simple --strict · make sa · make lint · CI-mode ShellCheck (per file, no -x) · bash build.sh bin -v → ✅ Build verified ✅.

…x.sh `src/runner.sh` -> `src/runner/index.sh`, `src/coverage.sh` -> `src/coverage/index.sh`. A module is now one self-contained directory whose entry point is where every other ecosystem puts it (index.ts, __init__.py, mod.rs). Amends ADR-010, which chose "beside" two days ago. The reasoning that first chose "beside" was wrong. It rested on matching the `src/assertions.sh` -> `src/assert_*.sh` precedent, but assertions.sh aggregates *flat* files and has no directory: it is not a module aggregator, it is unaffected by this choice, and the "two conventions in src/" risk it was meant to avoid never existed. What actually decides it is a correctness bug. The rule that aggregators hold only `source` lines is enforced by test_module_aggregators_hold_only_source_lines_and_comments, which named them in a hand-maintained string: "src/assertions.sh src/runner.sh". #928 added src/coverage.sh and never appended it, so the rule silently stopped covering that file within one module of being introduced. With the entry point at a predictable src/*/index.sh the test discovers modules by glob and cannot drift; this commit converts it, and adds a test asserting the discovery finds every module. Verified by appending a statement to src/coverage/index.sh -- previously green, now red. Doing it now costs two renames. #931 adds nine more modules, after which it would cost eleven. The distributable is unchanged: diffing the built artifact before and after shows only the two renamed embed markers and two reworded header comments, same 17627 lines. `bash build.sh bin -v` prints "Build verified". Also corrects two references #928 left stale: the branch-helper dynamic-scope note in bash-style.md pointed at `src/coverage.sh:818-821`, and the unreleased CHANGELOG entry described a shape that no longer ships. Related #931

Chemaclass · 2026-07-31T23:18:55Z

🤔 Background

Related #931

ADR-010 put each module's aggregator beside its directory (src/runner.sh + src/runner/*.sh). This amends that decision: the entry point moves inside as index.sh, matching what most ecosystems do (index.ts, __init__.py, mod.rs).

The original reasoning was wrong. It rested on matching the src/assertions.sh → src/assert_*.sh precedent — but assertions.sh aggregates flat files and has no directory. It is not a module aggregator, it is unaffected either way, and the "two conventions in src/" risk it was meant to avoid never existed.

💡 Changes

src/runner.sh → src/runner/index.sh, src/coverage.sh → src/coverage/index.sh; git records both as pure renames (1 line changed each — a reworded header comment).
Fixes a live correctness gap. The "aggregators hold only source lines" rule was enforced against a hand-maintained list, "src/assertions.sh src/runner.sh". refactor(coverage): split src/coverage.sh into a src/coverage/ module #928 added src/coverage.sh and never appended it, so the rule silently stopped covering that file within one module of being introduced. With the entry point at a predictable src/*/index.sh, the test now discovers modules by glob and cannot drift — plus a test asserting the discovery finds every module.
Amends ADR-010 with a dated amendment: chosen option flipped, the withdrawn precedent argument struck through rather than deleted, both options' pros/cons rewritten.
Corrects two references refactor(coverage): split src/coverage.sh into a src/coverage/ module #928 left stale: the branch-helper dynamic-scope note in bash-style.md pointed at src/coverage.sh:818-821, and the unreleased CHANGELOG entry described a shape that no longer ships.

Doing this now costs two renames. #931 adds nine more modules, after which it costs eleven.

✅ Verification

The distributable is unchanged. Diffing the built artifact before and after shows only the two renamed embed markers and the two reworded header comments — same 17627 lines.

The glob test was confirmed to catch what the hardcoded list could not: appending a statement to src/coverage/index.sh was previously green, now red.

Green: sequential · --parallel · --parallel --simple --strict · make sa · make lint · CI-mode ShellCheck (per file, no -x) · bash build.sh bin -v → ✅ Build verified ✅.

…x.sh `src/runner.sh` -> `src/runner/index.sh`, `src/coverage.sh` -> `src/coverage/index.sh`. A module is now one self-contained directory whose entry point is where every other ecosystem puts it (index.ts, __init__.py, mod.rs). Amends ADR-010, which chose "beside" two days ago. The reasoning that first chose "beside" was wrong. It rested on matching the `src/assertions.sh` -> `src/assert_*.sh` precedent, but assertions.sh aggregates *flat* files and has no directory: it is not a module aggregator, it is unaffected by this choice, and the "two conventions in src/" risk it was meant to avoid never existed. What actually decides it is a correctness bug. The rule that aggregators hold only `source` lines is enforced by test_module_aggregators_hold_only_source_lines_and_comments, which named them in a hand-maintained string: "src/assertions.sh src/runner.sh". #928 added src/coverage.sh and never appended it, so the rule silently stopped covering that file within one module of being introduced. With the entry point at a predictable src/*/index.sh the test discovers modules by glob and cannot drift; this commit converts it, and adds a test asserting the discovery finds every module. Verified by appending a statement to src/coverage/index.sh -- previously green, now red. Doing it now costs two renames. #931 adds nine more modules, after which it would cost eleven. The distributable is unchanged: diffing the built artifact before and after shows only the two renamed embed markers and two reworded header comments, same 17627 lines. `bash build.sh bin -v` prints "Build verified". Also corrects two references #928 left stale: the branch-helper dynamic-scope note in bash-style.md pointed at `src/coverage.sh:818-821`, and the unreleased CHANGELOG entry described a shape that no longer ships. Related #931

Chemaclass · 2026-07-31T23:18:55Z

🤔 Background

Related #931

ADR-010 put each module's aggregator beside its directory (src/runner.sh + src/runner/*.sh). This amends that decision: the entry point moves inside as index.sh, matching what most ecosystems do (index.ts, __init__.py, mod.rs).

The original reasoning was wrong. It rested on matching the src/assertions.sh → src/assert_*.sh precedent — but assertions.sh aggregates flat files and has no directory. It is not a module aggregator, it is unaffected either way, and the "two conventions in src/" risk it was meant to avoid never existed.

💡 Changes

src/runner.sh → src/runner/index.sh, src/coverage.sh → src/coverage/index.sh; git records both as pure renames (1 line changed each — a reworded header comment).
Fixes a live correctness gap. The "aggregators hold only source lines" rule was enforced against a hand-maintained list, "src/assertions.sh src/runner.sh". refactor(coverage): split src/coverage.sh into a src/coverage/ module #928 added src/coverage.sh and never appended it, so the rule silently stopped covering that file within one module of being introduced. With the entry point at a predictable src/*/index.sh, the test now discovers modules by glob and cannot drift — plus a test asserting the discovery finds every module.
Amends ADR-010 with a dated amendment: chosen option flipped, the withdrawn precedent argument struck through rather than deleted, both options' pros/cons rewritten.
Corrects two references refactor(coverage): split src/coverage.sh into a src/coverage/ module #928 left stale: the branch-helper dynamic-scope note in bash-style.md pointed at src/coverage.sh:818-821, and the unreleased CHANGELOG entry described a shape that no longer ships.

Doing this now costs two renames. #931 adds nine more modules, after which it costs eleven.

✅ Verification

The distributable is unchanged. Diffing the built artifact before and after shows only the two renamed embed markers and the two reworded header comments — same 17627 lines.

The glob test was confirmed to catch what the hardcoded list could not: appending a statement to src/coverage/index.sh was previously green, now red.

Green: sequential · --parallel · --parallel --simple --strict · make sa · make lint · CI-mode ShellCheck (per file, no -x) · bash build.sh bin -v → ✅ Build verified ✅.

…x.sh `src/runner.sh` -> `src/runner/index.sh`, `src/coverage.sh` -> `src/coverage/index.sh`. A module is now one self-contained directory whose entry point is where every other ecosystem puts it (index.ts, __init__.py, mod.rs). Amends ADR-010, which chose "beside" two days ago. The reasoning that first chose "beside" was wrong. It rested on matching the `src/assertions.sh` -> `src/assert_*.sh` precedent, but assertions.sh aggregates *flat* files and has no directory: it is not a module aggregator, it is unaffected by this choice, and the "two conventions in src/" risk it was meant to avoid never existed. What actually decides it is a correctness bug. The rule that aggregators hold only `source` lines is enforced by test_module_aggregators_hold_only_source_lines_and_comments, which named them in a hand-maintained string: "src/assertions.sh src/runner.sh". #928 added src/coverage.sh and never appended it, so the rule silently stopped covering that file within one module of being introduced. With the entry point at a predictable src/*/index.sh the test discovers modules by glob and cannot drift; this commit converts it, and adds a test asserting the discovery finds every module. Verified by appending a statement to src/coverage/index.sh -- previously green, now red. Doing it now costs two renames. #931 adds nine more modules, after which it would cost eleven. The distributable is unchanged: diffing the built artifact before and after shows only the two renamed embed markers and two reworded header comments, same 17627 lines. `bash build.sh bin -v` prints "Build verified". Also corrects two references #928 left stale: the branch-helper dynamic-scope note in bash-style.md pointed at `src/coverage.sh:818-821`, and the unreleased CHANGELOG entry described a shape that no longer ships. Related #931

Chemaclass · 2026-07-31T23:18:55Z

🤔 Background

Related #931

ADR-010 put each module's aggregator beside its directory (src/runner.sh + src/runner/*.sh). This amends that decision: the entry point moves inside as index.sh, matching what most ecosystems do (index.ts, __init__.py, mod.rs).

The original reasoning was wrong. It rested on matching the src/assertions.sh → src/assert_*.sh precedent — but assertions.sh aggregates flat files and has no directory. It is not a module aggregator, it is unaffected either way, and the "two conventions in src/" risk it was meant to avoid never existed.

💡 Changes

src/runner.sh → src/runner/index.sh, src/coverage.sh → src/coverage/index.sh; git records both as pure renames (1 line changed each — a reworded header comment).
Fixes a live correctness gap. The "aggregators hold only source lines" rule was enforced against a hand-maintained list, "src/assertions.sh src/runner.sh". refactor(coverage): split src/coverage.sh into a src/coverage/ module #928 added src/coverage.sh and never appended it, so the rule silently stopped covering that file within one module of being introduced. With the entry point at a predictable src/*/index.sh, the test now discovers modules by glob and cannot drift — plus a test asserting the discovery finds every module.
Amends ADR-010 with a dated amendment: chosen option flipped, the withdrawn precedent argument struck through rather than deleted, both options' pros/cons rewritten.
Corrects two references refactor(coverage): split src/coverage.sh into a src/coverage/ module #928 left stale: the branch-helper dynamic-scope note in bash-style.md pointed at src/coverage.sh:818-821, and the unreleased CHANGELOG entry described a shape that no longer ships.

Doing this now costs two renames. #931 adds nine more modules, after which it costs eleven.

✅ Verification

The distributable is unchanged. Diffing the built artifact before and after shows only the two renamed embed markers and the two reworded header comments — same 17627 lines.

The glob test was confirmed to catch what the hardcoded list could not: appending a statement to src/coverage/index.sh was previously green, now red.

Green: sequential · --parallel · --parallel --simple --strict · make sa · make lint · CI-mode ShellCheck (per file, no -x) · bash build.sh bin -v → ✅ Build verified ✅.

…x.sh `src/runner.sh` -> `src/runner/index.sh`, `src/coverage.sh` -> `src/coverage/index.sh`. A module is now one self-contained directory whose entry point is where every other ecosystem puts it (index.ts, __init__.py, mod.rs). Amends ADR-010, which chose "beside" two days ago. The reasoning that first chose "beside" was wrong. It rested on matching the `src/assertions.sh` -> `src/assert_*.sh` precedent, but assertions.sh aggregates *flat* files and has no directory: it is not a module aggregator, it is unaffected by this choice, and the "two conventions in src/" risk it was meant to avoid never existed. What actually decides it is a correctness bug. The rule that aggregators hold only `source` lines is enforced by test_module_aggregators_hold_only_source_lines_and_comments, which named them in a hand-maintained string: "src/assertions.sh src/runner.sh". #928 added src/coverage.sh and never appended it, so the rule silently stopped covering that file within one module of being introduced. With the entry point at a predictable src/*/index.sh the test discovers modules by glob and cannot drift; this commit converts it, and adds a test asserting the discovery finds every module. Verified by appending a statement to src/coverage/index.sh -- previously green, now red. Doing it now costs two renames. #931 adds nine more modules, after which it would cost eleven. The distributable is unchanged: diffing the built artifact before and after shows only the two renamed embed markers and two reworded header comments, same 17627 lines. `bash build.sh bin -v` prints "Build verified". Also corrects two references #928 left stale: the branch-helper dynamic-scope note in bash-style.md pointed at `src/coverage.sh:818-821`, and the unreleased CHANGELOG entry described a shape that no longer ships. Related #931

Chemaclass · 2026-07-31T23:18:55Z

🤔 Background

Related #931

ADR-010 put each module's aggregator beside its directory (src/runner.sh + src/runner/*.sh). This amends that decision: the entry point moves inside as index.sh, matching what most ecosystems do (index.ts, __init__.py, mod.rs).

The original reasoning was wrong. It rested on matching the src/assertions.sh → src/assert_*.sh precedent — but assertions.sh aggregates flat files and has no directory. It is not a module aggregator, it is unaffected either way, and the "two conventions in src/" risk it was meant to avoid never existed.

💡 Changes

src/runner.sh → src/runner/index.sh, src/coverage.sh → src/coverage/index.sh; git records both as pure renames (1 line changed each — a reworded header comment).
Fixes a live correctness gap. The "aggregators hold only source lines" rule was enforced against a hand-maintained list, "src/assertions.sh src/runner.sh". refactor(coverage): split src/coverage.sh into a src/coverage/ module #928 added src/coverage.sh and never appended it, so the rule silently stopped covering that file within one module of being introduced. With the entry point at a predictable src/*/index.sh, the test now discovers modules by glob and cannot drift — plus a test asserting the discovery finds every module.
Amends ADR-010 with a dated amendment: chosen option flipped, the withdrawn precedent argument struck through rather than deleted, both options' pros/cons rewritten.
Corrects two references refactor(coverage): split src/coverage.sh into a src/coverage/ module #928 left stale: the branch-helper dynamic-scope note in bash-style.md pointed at src/coverage.sh:818-821, and the unreleased CHANGELOG entry described a shape that no longer ships.

Doing this now costs two renames. #931 adds nine more modules, after which it costs eleven.

✅ Verification

The distributable is unchanged. Diffing the built artifact before and after shows only the two renamed embed markers and the two reworded header comments — same 17627 lines.

The glob test was confirmed to catch what the hardcoded list could not: appending a statement to src/coverage/index.sh was previously green, now red.

Green: sequential · --parallel · --parallel --simple --strict · make sa · make lint · CI-mode ShellCheck (per file, no -x) · bash build.sh bin -v → ✅ Build verified ✅.

…x.sh `src/runner.sh` -> `src/runner/index.sh`, `src/coverage.sh` -> `src/coverage/index.sh`. A module is now one self-contained directory whose entry point is where every other ecosystem puts it (index.ts, __init__.py, mod.rs). Amends ADR-010, which chose "beside" two days ago. The reasoning that first chose "beside" was wrong. It rested on matching the `src/assertions.sh` -> `src/assert_*.sh` precedent, but assertions.sh aggregates *flat* files and has no directory: it is not a module aggregator, it is unaffected by this choice, and the "two conventions in src/" risk it was meant to avoid never existed. What actually decides it is a correctness bug. The rule that aggregators hold only `source` lines is enforced by test_module_aggregators_hold_only_source_lines_and_comments, which named them in a hand-maintained string: "src/assertions.sh src/runner.sh". #928 added src/coverage.sh and never appended it, so the rule silently stopped covering that file within one module of being introduced. With the entry point at a predictable src/*/index.sh the test discovers modules by glob and cannot drift; this commit converts it, and adds a test asserting the discovery finds every module. Verified by appending a statement to src/coverage/index.sh -- previously green, now red. Doing it now costs two renames. #931 adds nine more modules, after which it would cost eleven. The distributable is unchanged: diffing the built artifact before and after shows only the two renamed embed markers and two reworded header comments, same 17627 lines. `bash build.sh bin -v` prints "Build verified". Also corrects two references #928 left stale: the branch-helper dynamic-scope note in bash-style.md pointed at `src/coverage.sh:818-821`, and the unreleased CHANGELOG entry described a shape that no longer ships. Related #931

Chemaclass · 2026-07-31T23:18:55Z

🤔 Background

Related #931

ADR-010 put each module's aggregator beside its directory (src/runner.sh + src/runner/*.sh). This amends that decision: the entry point moves inside as index.sh, matching what most ecosystems do (index.ts, __init__.py, mod.rs).

The original reasoning was wrong. It rested on matching the src/assertions.sh → src/assert_*.sh precedent — but assertions.sh aggregates flat files and has no directory. It is not a module aggregator, it is unaffected either way, and the "two conventions in src/" risk it was meant to avoid never existed.

💡 Changes

src/runner.sh → src/runner/index.sh, src/coverage.sh → src/coverage/index.sh; git records both as pure renames (1 line changed each — a reworded header comment).
Fixes a live correctness gap. The "aggregators hold only source lines" rule was enforced against a hand-maintained list, "src/assertions.sh src/runner.sh". refactor(coverage): split src/coverage.sh into a src/coverage/ module #928 added src/coverage.sh and never appended it, so the rule silently stopped covering that file within one module of being introduced. With the entry point at a predictable src/*/index.sh, the test now discovers modules by glob and cannot drift — plus a test asserting the discovery finds every module.
Amends ADR-010 with a dated amendment: chosen option flipped, the withdrawn precedent argument struck through rather than deleted, both options' pros/cons rewritten.
Corrects two references refactor(coverage): split src/coverage.sh into a src/coverage/ module #928 left stale: the branch-helper dynamic-scope note in bash-style.md pointed at src/coverage.sh:818-821, and the unreleased CHANGELOG entry described a shape that no longer ships.

Doing this now costs two renames. #931 adds nine more modules, after which it costs eleven.

✅ Verification

The distributable is unchanged. Diffing the built artifact before and after shows only the two renamed embed markers and the two reworded header comments — same 17627 lines.

The glob test was confirmed to catch what the hardcoded list could not: appending a statement to src/coverage/index.sh was previously green, now red.

Green: sequential · --parallel · --parallel --simple --strict · make sa · make lint · CI-mode ShellCheck (per file, no -x) · bash build.sh bin -v → ✅ Build verified ✅.

…x.sh `src/runner.sh` -> `src/runner/index.sh`, `src/coverage.sh` -> `src/coverage/index.sh`. A module is now one self-contained directory whose entry point is where every other ecosystem puts it (index.ts, __init__.py, mod.rs). Amends ADR-010, which chose "beside" two days ago. The reasoning that first chose "beside" was wrong. It rested on matching the `src/assertions.sh` -> `src/assert_*.sh` precedent, but assertions.sh aggregates *flat* files and has no directory: it is not a module aggregator, it is unaffected by this choice, and the "two conventions in src/" risk it was meant to avoid never existed. What actually decides it is a correctness bug. The rule that aggregators hold only `source` lines is enforced by test_module_aggregators_hold_only_source_lines_and_comments, which named them in a hand-maintained string: "src/assertions.sh src/runner.sh". #928 added src/coverage.sh and never appended it, so the rule silently stopped covering that file within one module of being introduced. With the entry point at a predictable src/*/index.sh the test discovers modules by glob and cannot drift; this commit converts it, and adds a test asserting the discovery finds every module. Verified by appending a statement to src/coverage/index.sh -- previously green, now red. Doing it now costs two renames. #931 adds nine more modules, after which it would cost eleven. The distributable is unchanged: diffing the built artifact before and after shows only the two renamed embed markers and two reworded header comments, same 17627 lines. `bash build.sh bin -v` prints "Build verified". Also corrects two references #928 left stale: the branch-helper dynamic-scope note in bash-style.md pointed at `src/coverage.sh:818-821`, and the unreleased CHANGELOG entry described a shape that no longer ships. Related #931

Chemaclass added the refactoring Refactoring or cleaning related label Jul 31, 2026

Chemaclass self-assigned this Jul 31, 2026

Chemaclass merged commit afd070a into mainJul 31, 2026
37 checks passed