feat(coverage): report loop constructs as branch points by Chemaclass · Pull Request #858 · TypedDevs/bashunit

Chemaclass · 2026-07-24T07:05:33Z

🤔 Background

Related #855

Branch coverage only understood if/elif/else and case (ADR-007 MVP). Loops were invisible, so a never-entered while/for body counted as fully covered.

💡 Changes

Extend static branch extraction to while/until/for/select: a loop is a single-arm branch (its body), taken iff the loop ran at least once — a zero-iteration loop now surfaces as an uncovered branch.
Nested loops pair done with the correct opener; the existing line-hit inference and report path consume the new loop records unchanged (no runtime cost).
Same-line short-circuit guards stay out of scope (line-granularity limit); documented in ADR-007.
Covered by extraction tests (while/for/until/nested) and end-to-end compute_branch_hits tests (body taken + zero-iteration not-taken).

Extend static branch extraction to while/until/for/select loops. A loop is a single-arm branch — its body range — taken iff an executable body line was hit (the loop ran at least once), so a never-entered zero-iteration loop now surfaces as an uncovered branch. Every done-closed opener pushes a frame so done pairs with the right opener when nested; the existing line-hit inference and report path consume the new 'loop' records unchanged. Same-line short-circuit guards ([ cond ] && action) remain out of scope: the DEBUG trap records line hits, not sub-line outcomes, so both outcomes share one line. Documented in adr-007. Closes#855

Chemaclass · 2026-07-24T07:05:33Z

🤔 Background

Related #855

Branch coverage only understood if/elif/else and case (ADR-007 MVP). Loops were invisible, so a never-entered while/for body counted as fully covered.

💡 Changes

Extend static branch extraction to while/until/for/select: a loop is a single-arm branch (its body), taken iff the loop ran at least once — a zero-iteration loop now surfaces as an uncovered branch.
Nested loops pair done with the correct opener; the existing line-hit inference and report path consume the new loop records unchanged (no runtime cost).
Same-line short-circuit guards stay out of scope (line-granularity limit); documented in ADR-007.
Covered by extraction tests (while/for/until/nested) and end-to-end compute_branch_hits tests (body taken + zero-iteration not-taken).

Extend static branch extraction to while/until/for/select loops. A loop is a single-arm branch — its body range — taken iff an executable body line was hit (the loop ran at least once), so a never-entered zero-iteration loop now surfaces as an uncovered branch. Every done-closed opener pushes a frame so done pairs with the right opener when nested; the existing line-hit inference and report path consume the new 'loop' records unchanged. Same-line short-circuit guards ([ cond ] && action) remain out of scope: the DEBUG trap records line hits, not sub-line outcomes, so both outcomes share one line. Documented in adr-007. Closes#855

Chemaclass · 2026-07-24T07:05:33Z

🤔 Background

Related #855

Branch coverage only understood if/elif/else and case (ADR-007 MVP). Loops were invisible, so a never-entered while/for body counted as fully covered.

💡 Changes

Extend static branch extraction to while/until/for/select: a loop is a single-arm branch (its body), taken iff the loop ran at least once — a zero-iteration loop now surfaces as an uncovered branch.
Nested loops pair done with the correct opener; the existing line-hit inference and report path consume the new loop records unchanged (no runtime cost).
Same-line short-circuit guards stay out of scope (line-granularity limit); documented in ADR-007.
Covered by extraction tests (while/for/until/nested) and end-to-end compute_branch_hits tests (body taken + zero-iteration not-taken).

Extend static branch extraction to while/until/for/select loops. A loop is a single-arm branch — its body range — taken iff an executable body line was hit (the loop ran at least once), so a never-entered zero-iteration loop now surfaces as an uncovered branch. Every done-closed opener pushes a frame so done pairs with the right opener when nested; the existing line-hit inference and report path consume the new 'loop' records unchanged. Same-line short-circuit guards ([ cond ] && action) remain out of scope: the DEBUG trap records line hits, not sub-line outcomes, so both outcomes share one line. Documented in adr-007. Closes#855

Chemaclass · 2026-07-24T07:05:33Z

🤔 Background

Related #855

Branch coverage only understood if/elif/else and case (ADR-007 MVP). Loops were invisible, so a never-entered while/for body counted as fully covered.

💡 Changes

Extend static branch extraction to while/until/for/select: a loop is a single-arm branch (its body), taken iff the loop ran at least once — a zero-iteration loop now surfaces as an uncovered branch.
Nested loops pair done with the correct opener; the existing line-hit inference and report path consume the new loop records unchanged (no runtime cost).
Same-line short-circuit guards stay out of scope (line-granularity limit); documented in ADR-007.
Covered by extraction tests (while/for/until/nested) and end-to-end compute_branch_hits tests (body taken + zero-iteration not-taken).

Extend static branch extraction to while/until/for/select loops. A loop is a single-arm branch — its body range — taken iff an executable body line was hit (the loop ran at least once), so a never-entered zero-iteration loop now surfaces as an uncovered branch. Every done-closed opener pushes a frame so done pairs with the right opener when nested; the existing line-hit inference and report path consume the new 'loop' records unchanged. Same-line short-circuit guards ([ cond ] && action) remain out of scope: the DEBUG trap records line hits, not sub-line outcomes, so both outcomes share one line. Documented in adr-007. Closes#855

Chemaclass · 2026-07-24T07:05:33Z

🤔 Background

Related #855

Branch coverage only understood if/elif/else and case (ADR-007 MVP). Loops were invisible, so a never-entered while/for body counted as fully covered.

💡 Changes

Extend static branch extraction to while/until/for/select: a loop is a single-arm branch (its body), taken iff the loop ran at least once — a zero-iteration loop now surfaces as an uncovered branch.
Nested loops pair done with the correct opener; the existing line-hit inference and report path consume the new loop records unchanged (no runtime cost).
Same-line short-circuit guards stay out of scope (line-granularity limit); documented in ADR-007.
Covered by extraction tests (while/for/until/nested) and end-to-end compute_branch_hits tests (body taken + zero-iteration not-taken).

Extend static branch extraction to while/until/for/select loops. A loop is a single-arm branch — its body range — taken iff an executable body line was hit (the loop ran at least once), so a never-entered zero-iteration loop now surfaces as an uncovered branch. Every done-closed opener pushes a frame so done pairs with the right opener when nested; the existing line-hit inference and report path consume the new 'loop' records unchanged. Same-line short-circuit guards ([ cond ] && action) remain out of scope: the DEBUG trap records line hits, not sub-line outcomes, so both outcomes share one line. Documented in adr-007. Closes#855

Chemaclass · 2026-07-24T07:05:33Z

🤔 Background

Related #855

Branch coverage only understood if/elif/else and case (ADR-007 MVP). Loops were invisible, so a never-entered while/for body counted as fully covered.

💡 Changes

Extend static branch extraction to while/until/for/select: a loop is a single-arm branch (its body), taken iff the loop ran at least once — a zero-iteration loop now surfaces as an uncovered branch.
Nested loops pair done with the correct opener; the existing line-hit inference and report path consume the new loop records unchanged (no runtime cost).
Same-line short-circuit guards stay out of scope (line-granularity limit); documented in ADR-007.
Covered by extraction tests (while/for/until/nested) and end-to-end compute_branch_hits tests (body taken + zero-iteration not-taken).

Extend static branch extraction to while/until/for/select loops. A loop is a single-arm branch — its body range — taken iff an executable body line was hit (the loop ran at least once), so a never-entered zero-iteration loop now surfaces as an uncovered branch. Every done-closed opener pushes a frame so done pairs with the right opener when nested; the existing line-hit inference and report path consume the new 'loop' records unchanged. Same-line short-circuit guards ([ cond ] && action) remain out of scope: the DEBUG trap records line hits, not sub-line outcomes, so both outcomes share one line. Documented in adr-007. Closes#855

Chemaclass · 2026-07-24T07:05:33Z

🤔 Background

Related #855

Branch coverage only understood if/elif/else and case (ADR-007 MVP). Loops were invisible, so a never-entered while/for body counted as fully covered.

💡 Changes

Extend static branch extraction to while/until/for/select: a loop is a single-arm branch (its body), taken iff the loop ran at least once — a zero-iteration loop now surfaces as an uncovered branch.
Nested loops pair done with the correct opener; the existing line-hit inference and report path consume the new loop records unchanged (no runtime cost).
Same-line short-circuit guards stay out of scope (line-granularity limit); documented in ADR-007.
Covered by extraction tests (while/for/until/nested) and end-to-end compute_branch_hits tests (body taken + zero-iteration not-taken).

Extend static branch extraction to while/until/for/select loops. A loop is a single-arm branch — its body range — taken iff an executable body line was hit (the loop ran at least once), so a never-entered zero-iteration loop now surfaces as an uncovered branch. Every done-closed opener pushes a frame so done pairs with the right opener when nested; the existing line-hit inference and report path consume the new 'loop' records unchanged. Same-line short-circuit guards ([ cond ] && action) remain out of scope: the DEBUG trap records line hits, not sub-line outcomes, so both outcomes share one line. Documented in adr-007. Closes#855

Chemaclass · 2026-07-24T07:05:33Z

🤔 Background

Related #855

Branch coverage only understood if/elif/else and case (ADR-007 MVP). Loops were invisible, so a never-entered while/for body counted as fully covered.

💡 Changes

Extend static branch extraction to while/until/for/select: a loop is a single-arm branch (its body), taken iff the loop ran at least once — a zero-iteration loop now surfaces as an uncovered branch.
Nested loops pair done with the correct opener; the existing line-hit inference and report path consume the new loop records unchanged (no runtime cost).
Same-line short-circuit guards stay out of scope (line-granularity limit); documented in ADR-007.
Covered by extraction tests (while/for/until/nested) and end-to-end compute_branch_hits tests (body taken + zero-iteration not-taken).

Extend static branch extraction to while/until/for/select loops. A loop is a single-arm branch — its body range — taken iff an executable body line was hit (the loop ran at least once), so a never-entered zero-iteration loop now surfaces as an uncovered branch. Every done-closed opener pushes a frame so done pairs with the right opener when nested; the existing line-hit inference and report path consume the new 'loop' records unchanged. Same-line short-circuit guards ([ cond ] && action) remain out of scope: the DEBUG trap records line hits, not sub-line outcomes, so both outcomes share one line. Documented in adr-007. Closes#855

Chemaclass added the enhancement New feature or request label Jul 24, 2026

Chemaclass self-assigned this Jul 24, 2026

Chemaclass merged commit 61d605f into mainJul 24, 2026
37 checks passed