Module coverage 

bynkc test --coverage — remap V8 line coverage onto .bynk source.

The test runner already owns the two artefacts a coverage tool needs and a user cannot reconstruct: it launches the node process that executes the suite, and it holds the source maps from .bynk → emitted .ts. This module is the “one genuinely new piece” the coverage proposal (issue #854, ADR recorded at merge) calls out: it reads the raw V8 coverage the runtime writes to NODE_V8_COVERAGE, and attributes each executed / unexecuted line back to .bynk source through two line-level source-map hops:

1. out-js/**/*.js.map — tsc’s map, .js line → emitted .ts line. tsc does not chain input maps, so this hop only reaches the .ts.
2. out/**/*.ts.map — the emitter’s map (ADR 0103), .ts line → .bynk line. Statement-anchored and line-level (generated column always 0).

Composed, a covered .js line lands on a .bynk line. Emitted glue with no .bynk origin (codec wrappers, capability injection, the module header) is unmapped in hop 2, so it contributes nothing — it is counted as out-of-scope, never as uncovered user code (the proposal’s map-fidelity mitigation, for free).

Decisions realised here (recorded in the ADR): line/statement coverage only, no branch coverage (DECISION B) — a .bynk line is covered if any generated line mapping to it executed; and the measured set excludes the tests/ tree and the workers scaffold (DECISION D), filtered once on the out-js-relative path of the executed .js — the emitted tree’s own top-level tests//workers/ dirs — before the maps are even consulted. The .bynk side is deliberately not re-filtered, so a user source that merely lives under a dir named tests/workers is still measured.

Structs

CoverageReport

A whole-run coverage report: one FileCoverage per measured .bynk file, sorted by path, plus the derived totals.

FileCoverage

Per-.bynk-file line coverage, keyed by a project-relative display path.

Functions

collect_coverage

Collect coverage from a finished run and attribute it to .bynk source.

format_ranges

Compress a sorted line list into comma-separated ranges: [6,7,8,51] → "6-8, 51".

percent

Coverage percentage of covered/total, rounded; total == 0 → 100. Only a genuinely complete run reads 100%: round-half-up would report 995/1000 as 100% while lines are still uncovered — self-contradicting in the table and a false green for a CI gate keyed on the JSON percent — so a run with any uncovered line is clamped to at most 99.

render_rich

Render the human coverage summary — a per-file bar, percentage, covered/total line count, and the uncovered-line ranges — plus a total row. Trailing newline. An empty report renders a single “no coverage attributed” note.