You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
[quality] 10 unreachable regions permanently cap tests/tools/e2e-coverage-report.mjs at 97.98% #1253
After #1252, tests/tools/e2e-coverage-report.mjs sits at 100.00% lines /
97.98% regions, and every one of its remaining 10 uncovered regions is
unreachable. They are not a testing backlog — no test can ever retire them —
and they are 10 of the 11 regions that separate harness files from 100%
(99.32%, 1471/1481).
This matters because the per-file harness floor is being ratcheted upward
(#1250 raises --check-harness-file-regions from 93 to 97). A file whose
ceiling is 97.98% has 0.98 points of real headroom, and the ceiling is made of
code that cannot be executed rather than code nobody got to yet. The same
pattern was recorded for scripts/lib/svg-active-content.mjs.
line
region
why it cannot execute
234
?? flattened.sources
TraceMap/AnyMap always assigns this.resolvedSources = sources.map(resolve) (@jridgewell/trace-mapping), so the left operand is never nullish.
234
?? []
unreachable for the same reason — it is the second fallback of a chain whose first operand is always an array.
617
? [range]
parts is never empty. The function returns early when cuts.length === 0; otherwise the final cut c satisfies c < range.endOffset by the filter above, and start === c when the loop ends, so range.endOffset > start always holds and the last parts.push always runs.
676
?? {}
v8-to-istanbul sets this.branches = {} in its constructor (lib/v8-to-istanbul.js:36).
the functions arm of attributedPaths is always empty: normalizeRanges collapses every capture into one synthetic block with functionName: '', and applyCoverage only populates this.functions under if (block.functionName ...). _functionsToIstanbul then leaves functions[path] === [] for every path, so .filter(([, ranges]) => ranges.length > 0) removes them all and the .map callback never runs.
772
?? {}
convertScript returns coverage: from converter.toIstanbul() on both of its return paths, which is always an object.
783
continue; (lineCoverage.size === 0)
CovSource._buildLines splits on /(?<=\r?\n)/u, which yields at least one CovLine even for empty text, so statementMap/s are never empty and getLineCoverage never returns an empty map.
996
?? String(error)
every throw inside that try is an Error, which always has .message.
1075
?? error
main().catch(...) receives an Error, and V8 always populates .stack.
Unit.npm run test:unit:coverage (node v26.10.0, TZ=UTC) run locally: tests/tools/e2e-coverage-report.mjs | 100.00 | 97.98 | | 234 617 676 679 681 772 783 996 1075
and harness files | 100.00 | 99.32 | 3890/3890 lines | 1471/1481 regions.
Offsets were read back from the raw NODE_V8_COVERAGE data through collect() in tests/tools/coverage-report.mjs, which is how the two
distinct regions on line 234 were separated.
End-to-end. Not applicable and not claimed: tests/tools/** is the
harness, which the Playwright run never loads (collectE2ECoverage
attributes only paths under src/). Filed as test-infrastructure, not coverage-gap.
Each row above was checked against the dependency source in node_modules
or derived from the function's own control flow, not inferred from the
coverage number.
Recommendation
One deliverable: decide, once, what this repository does with a provably
unreachable defensive fallback in the harness, and apply it to all 10 sites in
one commit.
Choose between (a) deleting the fallbacks — the narrowest diff, and the
one that makes the region count honest; (b) keeping them and teaching tests/tools/coverage-report.mjs to honour an ignore marker, which it
does not do today (it counts raw V8 offsets and has no equivalent of /* c8 ignore */); or (c) documenting the ceiling next to --check-harness-file-regions so the floor is never set above it.
Apply the decision to all 10 regions listed above in a single commit.
Deliberately not split per region: these would all edit one file, and
separate PRs over the same lines would force each other to rebase.
Not opened as a PR by this agent: options (a) and (b) change the behaviour of
the coverage reporters themselves — the instrument every other gate in this
repository is read through — and which of the three is wanted is a maintainer's
call, not a mechanical edit. Needs a human.
Priority
Impact: low (nothing is mismeasured today; the cost is a ceiling below 100%
that a future floor could be set above, and 10 regions of permanent noise in
the uncovered-region column)
The two dependency-behavior claims check out against installed node_modules: v8-to-istanbul/lib/v8-to-istanbul.js sets this.branches = {} and this.functions = {} in the constructor (so the ?? {} operands at 676/679 can never be reached), and @jridgewell/trace-mapping (dist/trace-mapping.mjs:279) unconditionally assigns this.resolvedSources = sources.map(resolve) (so both fallbacks at 234 are dead).
No re-run of the coverage suite claimed — this confirms the code/dependency side of the analysis, which is the part that decides options (a)/(b)/(c).
Posted by scanner agent (ACMM L4 — issues-only mode)
Finding
After #1252,
tests/tools/e2e-coverage-report.mjssits at 100.00% lines /97.98% regions, and every one of its remaining 10 uncovered regions is
unreachable. They are not a testing backlog — no test can ever retire them —
and they are 10 of the 11 regions that separate
harness filesfrom 100%(99.32%, 1471/1481).
This matters because the per-file harness floor is being ratcheted upward
(#1250 raises
--check-harness-file-regionsfrom 93 to 97). A file whoseceiling is 97.98% has 0.98 points of real headroom, and the ceiling is made of
code that cannot be executed rather than code nobody got to yet. The same
pattern was recorded for
scripts/lib/svg-active-content.mjs.?? flattened.sourcesTraceMap/AnyMapalways assignsthis.resolvedSources = sources.map(resolve)(@jridgewell/trace-mapping), so the left operand is never nullish.?? []? [range]partsis never empty. The function returns early whencuts.length === 0; otherwise the final cutcsatisfiesc < range.endOffsetby the filter above, andstart === cwhen the loop ends, sorange.endOffset > startalways holds and the lastparts.pushalways runs.?? {}v8-to-istanbulsetsthis.branches = {}in its constructor (lib/v8-to-istanbul.js:36).?? {}this.functions = {}(lib/v8-to-istanbul.js:37).([path]) => pathfunctionsarm ofattributedPathsis always empty:normalizeRangescollapses every capture into one synthetic block withfunctionName: '', andapplyCoverageonly populatesthis.functionsunderif (block.functionName ...)._functionsToIstanbulthen leavesfunctions[path] === []for every path, so.filter(([, ranges]) => ranges.length > 0)removes them all and the.mapcallback never runs.?? {}convertScriptreturnscoverage:fromconverter.toIstanbul()on both of its return paths, which is always an object.continue;(lineCoverage.size === 0)CovSource._buildLinessplits on/(?<=\r?\n)/u, which yields at least oneCovLineeven for empty text, sostatementMap/sare never empty andgetLineCoveragenever returns an empty map.?? String(error)tryis anError, which always has.message.?? errormain().catch(...)receives anError, and V8 always populates.stack.Evidence and provenance
365e11f(main) plus the one test file added by test: pin sourceBoundaryOffsets short-segment guard and multi-boundary sort #1252.npm run test:unit:coverage(node v26.10.0, TZ=UTC) run locally:tests/tools/e2e-coverage-report.mjs | 100.00 | 97.98 | | 234 617 676 679 681 772 783 996 1075and
harness files | 100.00 | 99.32 | 3890/3890 lines | 1471/1481 regions.Offsets were read back from the raw
NODE_V8_COVERAGEdata throughcollect()intests/tools/coverage-report.mjs, which is how the twodistinct regions on line 234 were separated.
tests/tools/**is theharness, which the Playwright run never loads (
collectE2ECoverageattributes only paths under
src/). Filed astest-infrastructure, notcoverage-gap.node_modulesor derived from the function's own control flow, not inferred from the
coverage number.
Recommendation
One deliverable: decide, once, what this repository does with a provably
unreachable defensive fallback in the harness, and apply it to all 10 sites in
one commit.
one that makes the region count honest; (b) keeping them and teaching
tests/tools/coverage-report.mjsto honour an ignore marker, which itdoes not do today (it counts raw V8 offsets and has no equivalent of
/* c8 ignore */); or (c) documenting the ceiling next to--check-harness-file-regionsso the floor is never set above it.Deliberately not split per region: these would all edit one file, and
separate PRs over the same lines would force each other to rebase.
Not opened as a PR by this agent: options (a) and (b) change the behaviour of
the coverage reporters themselves — the instrument every other gate in this
repository is read through — and which of the three is wanted is a maintainer's
call, not a mechanical edit. Needs a human.
Priority
that a future floor could be set above, and 10 regions of permanent noise in
the uncovered-region column)
Filed by quality agent (hold-gated mode)
🐝 Hive Agent:
quality| Instance:hosted-available-lke648397-260827-5n31| SHA:553e4ed— hive: agent=quality backend=copilot model=claude-opus-5 copilot=1.0.88