[mkarlesky]

Supersedes #13. Adds native code coverage to the Testing view. Run with Coverage runs tests through Ceedling's gcov plugin, then shows each source file's line, branch, and function coverage in VS Code's coverage views. It reads the Cobertura report from gcovr or ReportGenerator.

Credit

The design and the first implementation are by @jake-kim-sndk (Jake Jooncheol Kim), in #13. His pull request closed so it could be retargeted, and his fork has since been deleted. His commit survived at refs/pull/13/head, and it is preserved two ways:

• In this branch's history, unchanged as aa07df0, with his author, date, and message. A merge commit brings it in, so GitHub lists it under his name.
• On archive/pr-13-gcov-coverage, pushed to this repo as a durable copy.

The commit that ports his engine and controller wiring onto the reworked code carries a Co-authored-by trailer for him. CHANGELOG credits him in its existing "contributed by" style. The new docs/Acknowledgments.md credits him after the README's account of the extension's origin.

Commits

Commit	What
aa07df0	Jake's original commit, exactly as submitted
🔀 Merge	Brings in aa07df0. next_version reworked the engine and controller it changed, so conflicts resolve to next_version's side. His new module, tests, and fixtures arrive exactly as written.
🧪 Fixtures	capture-gcov.sh plus captures from Ceedling 1.2.0 with its own gcovr 7.2 and ReportGenerator 5.5.11. Each tool's Cobertura report, and dumpconfig with the gcov plugin's defaults.
✨ Port (co-authored by Jake)	Coverage through the reworked engine. Coverage state comes from the resolved configuration, so no report path is hard-coded. Coverage builds run under the project directory's lock. The profile appears only while some project enables the gcov plugin.
✨ ReportGenerator	The module is renamed coberturaReport.ts, so its history follows. Several <source> roots resolve to the candidate path that exists. Real-capture tests.
✅ Tests	Engine edge cases. Coverage in VS Code against the fake Ceedling: the profile's lifecycle, file coverage, and its details.
📝 Docs	The README "Code coverage" how-to, the CHANGELOG, BreakingChanges, Acknowledgments, Development.md, and a manual checklist with tests/manual/mixins/.
🐛 Review fixes	The final review's findings (below).

Behavior

• Which report: gcovr's, when both tools write one. ReportGenerator's is ReportGenerator/Cobertura.xml, or ReportGenerator/Cobertura/Cobertura.xml with several report types.
• One run, one measurement: each request clears the coverage counters earlier runs left (.gcda, plus ReportGenerator's .gcov inputs), once per build directory. The new ceedlingExplorer.coverageResetsCounters setting turns that off to accumulate coverage across runs.
• :report_task: TRUE: the run adds report:gcov to its own command line.
• No coverage: a project without the plugin, without a Cobertura report, with a report covering no sources, or whose report tool is missing gets a notification saying why, once a session. Its tests still run.
• gcov: task names: Ceedling 1.2.0's gcov plugin still takes only a bare file name, unlike its test: tasks. That is reported as gcov: tasks don't accept project-relative paths, unlike test: tasks in 1.2.0 Ceedling#1362. Coverage runs pass the bare name before 1.2.1, warning when same-named test files make that ambiguous, and the path from 1.2.1 (pathQualifiedCoverageTasks).

Minimum VS Code is now 1.88.0, up from 1.74.0. 1.88 finalized the test coverage API. Typechecking runs against @types/vscode 1.88.0, so no newer API is used.

Verification

• Lint: 0 findings.
• Tests: 437 unit and 18 integration pass in the sidecar.
• Real-Ceedling gate, a one-off script: the compiled engine against Ceedling 1.2.0 with real gcovr and ReportGenerator.
  • Exact checks, one run at a time:
    • gcovr preferred;
    • per-run hits with the counter reset on, accumulated hits with it off;
    • function and project runs;
    • ReportGenerator resolving both a source inside the project and one above it.
  • Then 90 to 120 seconds of overlapping coverage runs, plain runs, discovery, reparses, and Clean, with random cancellation, at two seeds and up to 8 in flight. Result: 0 violations, 0 unhandled rejections, no leftover processes or temp files.

Final review

Four parallel reviews covered the engine and coverage state, the controller, the parser and merge, and the tests and docs. Each finding was checked against the code, and each behavior change was made test-first.

Fixed:

• A function run's whole-file fallback counted its first attempt's hits twice.
• Coverage was lost when a run threw partway.
• A failed build gave no explanation for a configuration that can't produce coverage.
• An empty report said nothing.
• Line 0, negative values, and implausible branch counts could crash VS Code's positions.
• Untrimmed sources.
• Repeated notifications.
• Duplicated XML helpers, and path keys taken from the host rather than the engine.
• Doc accuracy: ceedling clean also clears counters, ReportGenerator lists earlier builds' sources at 0%, and same-named test files are ambiguous. Plus style.

Deferred, with reason: two coverage runs at once in one project share its counters. Keeping them apart would need the engine to track when each request ends. That is rare, and the README documents it.

Manual checks

docs/Development.md §5 gains coverage checks for gcovr, ReportGenerator, per-run coverage, and a project without a Cobertura report.

🤖 Generated with Claude Code