Add unified C++/Python coverage via xmsconan 2.15.2

[gagelarsen]

Summary

Adds a two-build instrumented Debug pass that produces both C++ (gcovr) and Python (pytest-cov) coverage reports for every push and pull request to master. Pure additive — no source code changes, only build-config and CI plumbing.

What this PR does

• build.toml: sets [ci].coverage = true and adds a [coverage] table with both thresholds at 0 (report-only mode while we establish the baseline). The existing python_namespaced_dir = "core" is already in place, which is the only other thing the coverage flow needs.
• .github/workflows/Coverage.yaml (new, generated by xmsconan ci): Linux-only job running directly on ubuntu-latest (containerless under xmsconan 2.15.2, since the old container baked xmsconan in and caused silent floor-pin no-ops). Installs gcovr>=7,<9 and xmsconan>=2.15.2, runs xmsconan_coverage, uploads cov-{cpp,py}.xml and coverage-html-{cpp,py}/ as run artifacts.
• .github/workflows/XmsCore-CI.yaml: regenerated, xmsconan pin bumped 2.12.1 → 2.15.2 and all install steps gained --upgrade so the floor is real (not silently overridden by a pre-installed copy).
• .flake8, pytest.ini: new files emitted by xmsconan gen from the same template the rest of the family uses.
• .gitignore: ignores cov-*.xml, cov-*-summary.json, coverage-html-*/ so a developer running coverage locally doesn't accidentally commit a snapshot.
• README.md: short "Coverage" section explaining the workflow + artifacts.

Baseline numbers from the first green Coverage CI run on the two-build flow

From XmsCore-Coverage run 25978408642 against xmsconan==2.15.2:

Layer	Line rate	Lines covered
C++	88.21%	3503 / 3971
Python	72.73%	104 / 143

The C++ jump versus earlier runs of this PR (9.36% on 124/1325 lines) is structural, not a sudden test-writing surge: xmsconan 2.15.2 splits the instrumented build into a CxxTest-driven testing pass and a pytest-driven pybind pass. cov-cpp.xml now reflects coverage from running the CxxTest binaries directly instead of whatever fraction of C++ symbols pytest happened to reach through the pybind ABI. The total line count grew because the testing build compiles the full set of library + testing_sources, whereas the prior pytest-only path only compiled the pybind-linked subset.

Per-file C++ breakdown — current gaps

From the same gcovr report. Files marked "production" are real coverage gaps; testing/TestTools.\* is the test-scaffold library itself, which won't ever be 100% (it has helpers no test currently invokes).

File	Lines	Coverage	Notes
xmscore/misc/DynBitset.cpp	11	0%	production — no test currently exercises it
xmscore/misc/XmLog.cpp	104	60%	production — error-path branches missing
xmscore/misc/Singleton.h	10	70%	template helpers
xmscore/points/functors.h	55	72%	inline math helpers
xmscore/misc/Observer.cpp	129	81%	progress / observer corner cases
xmscore/misc/XmError.cpp	145	81%	ENSURE failure paths
xmscore/misc/StringUtil.cpp	807	89%	format / encoding edge cases
xmscore/points/pt.h	590	91%	inline point ops
xmscore/time/TimeConversion.cpp	155	92%	invalid-input branches
xmscore/dataio/daStreamIo.cpp	1147	96%	stream-failure paths
xmscore/testing/TestTools.cpp	156	14%	test-scaffold — unused helpers
xmscore/testing/TestTools.h	43	39%	test-scaffold — unused helpers

Other files (math.cpp, math.h, Progress.cpp, functors.cpp, pt.cpp, XmError.h) are at 100%.

The thresholds are deliberately 0 so the gate never fails while the baseline is being established. Once test-writing fills the remaining gaps, raise them in a follow-up.

Upstream prerequisites

This works because xmsconan 2.15.2 finally has the coverage flow wired up correctly. The rollout went 2.14.0 → 2.15.2 over several releases, each one uncovering the next downstream gap once the previous one was fixed:

• 2.14.0 — initial unified coverage release.
• 2.14.1 (#61 / #62 / #64 / #65 via PR #63): generate_configurations Debug+pybind gate, filter shape, testing/pybind matrix exclusivity, multi-Python pinning.
• 2.14.2 (#66 via PR #67): conan cache path --folder=source reference-shape fix.
• 2.14.3 (#69 and #71 via PR #70 + PR #72): XMS_COVERAGE into CMake buildenv; rglob for pytest-cov artifacts; loud-fail guard for empty gcovr summaries.
• 2.14.4 (direct on master, commit 2e978fe): pass XMS_COVERAGE to CMake as a -D cache variable since conan's [buildenv] activation isn't reaching the CMake subprocess.
• 2.14.5 (direct on master, commit 0131f6d): anchor gcovr --root and the doubled filter against the build folder, not the conan source folder — sources are copied into the build folder before compilation, so .gcno paths live there.
• 2.15.2 (PR #76): the combined pybind=True + testing=True + Debug config was structurally fragile — the pybind .so could pick up undefined CxxTest symbols (CxxTest::charToString) at dlopen. Replaced with two independent builds (CxxTest-only testing build for cov-cpp.xml; pybind-only build for cov-py.xml). Also drops the Coverage container so the canary always exercises whatever xmsconan is on devpi, and adds --upgrade to all CI install steps so the in-repo floor is real.

Test plan

• XmsCore-CI workflow on feature/coverage passes on all three platforms (Mac, Linux, Windows × py3.10/3.13).
• XmsCore-Coverage workflow produces real artifacts and exits 0.
• cov-cpp.xml is non-empty Cobertura XML with line-rate="0.8821" and lines-valid="3971".
• cov-py.xml is non-empty Cobertura XML with line-rate="0.7273" and lines-valid="143".
• CxxTest binary actually runs during the coverage flow (Total Test time (real) = 1.48 sec in the Coverage job log), and pytest actually runs (25 passed in 0.47s) — both contribute to their respective cov-{cpp,py}.xml outputs.

Follow-ups (not in this PR)

• Backfill tests for the real C++ gaps: misc/DynBitset.cpp (11 lines, 0%) first, then the error-path branches in XmLog.cpp (60%), Observer.cpp (81%), and XmError.cpp (81%). testing/TestTools.{cpp,h} is the test-scaffold and is expected to stay partially covered.
• Raise [coverage].cpp_threshold and [coverage].python_threshold once the remaining gaps land, so regressions break the build. cpp_threshold could reasonably go to 80 already.

🤖 Generated with Claude Code

[gagelarsen]

Code review of PR #126

Scope: changes in this PR only — pre-existing code is out of scope per the user's review policy.

Pure additive build-config/CI plumbing — no production C++ or Python source touched, so coverage/testing policy has nothing to enforce on this diff. The PR body is unusually well-grounded (numbers, per-file table, and upstream issue chain all verified against the latest XmsCore-Coverage run 25978408642). All three CI legs are green at 634155b9. No blockers. One scope-correct observation worth flagging: the new .flake8 is effectively inert under the (pre-existing) CI lint invocation.

Blockers

None.

Recommendations

• .flake8 (new file) is not consulted by the existing CI lint step — .github/workflows/XmsCore-CI.yaml already invokes flake8 --isolated … with a hand-rolled argument list (that line is pre-existing, not changed by this PR — git diff master..feature/coverage -- .github/workflows/XmsCore-CI.yaml only touches the three xmsconan>=2.15.2 --upgrade install lines). Because --isolated ignores every config file, the rules the new .flake8 introduces beyond the CLI args — the B028, W503 additions to ignore, the banned-modules = osgeo.*=… rule, and the tests/files/* / tests/fixtures/* excludes — are not actually enforced on PRs. Scope-respecting recommendation: either drop .flake8 from this PR (since nothing exercises it), or accept that it's local-dev-only and document that in a comment at the top of the file. A CI-side change to honour .flake8 (removing --isolated and the inline args) is the cleaner long-term fix but is out of scope for this PR — file it as a follow-up. Not a release blocker.

Verified

• All three install steps in XmsCore-CI.yaml that this PR touches pin xmsconan>=2.15.2 with --upgrade and consistent quoting (mac line 108, linux line 242, windows line 387). These are the only lines this PR changes in that file.
• Coverage.yaml (new) is containerless on ubuntu-latest, uses actions/setup-python@v5, installs gcovr>=7,<9 and xmsconan>=2.15.2 --upgrade, and uploads exactly the two artifact bundles the README documents.
• build.toml diff adds [ci].coverage = true and a [coverage] table with thresholds of 0, matching the PR body's "report-only while baseline is being established" framing.
• .gitignore diff covers all six local artifact patterns: cov-{cpp,py}.xml, cov-{cpp,py}-summary.json, coverage-html-{cpp,py}/.
• README.md diff adds only a new Coverage section that accurately describes the new flow (Linux-only, two HTML reports + two XML files, thresholds at 0). No other README sections are touched by this PR.
• pytest.ini (new) defines testpaths = _package/tests and python_files = test_*.py *_test.py *_pyt.py, matching what Coverage CI consumes.
• PR body numbers cross-check exactly against the Coverage run log: gcovr TOTAL 3971 3503 88%, pytest 25 passed in 0.47s, CxxTest Total Test time (real) = 1.48 sec. Per-file table matches the gcovr report line-for-line (DynBitset 11/0/0%, XmLog 104/63/60%, Singleton 10/7/70%, functors.h 55/40/72%, Observer 129/105/81%, XmError 145/118/81%, StringUtil 807/720/89%, pt.h 590/540/91%, TimeConversion 155/144/92%, daStreamIo 1147/1108/96%, TestTools.cpp 156/22/14%, TestTools.h 43/17/39%). "Other files at 100%" claim also verified: math.cpp 58, math.h 26, Progress.cpp 86, functors.cpp 105, pt.cpp 342, XmError.h 2 — all 100% in the log.
• Two-build flow visible in the run log: first build with pybind=False testing=True (testing pass → cov-cpp.xml from CxxTest runner), second build with pybind enabled (pytest pass → cov-py.xml).
• Branch history is clean — only commits relevant to the coverage feature: 22d4df7 (gitignore), 90199f7 (initial 2.14.0 wire-up), e013bce (2.15.1 bump), e3c8003 (--upgrade fix), 634155b (regenerate at 2.15.2).
• git diff master..feature/coverage --stat is 147 lines added across 7 files. No accidentally committed binaries, secrets, or local config.
• Code conventions: nothing to check — no C++/Python production code in the diff.

Notes

• .flake8 line 20 sets isolated = true as a config-file key. That key has no effect inside a config file (it's a CLI flag); harmless but slightly misleading if someone later edits .flake8 expecting that knob to mean something. In-scope because .flake8 is new in this PR.
• Coverage.yaml (new in this PR) uses actions/checkout@v2 while it uses @v5 for setup-python and @v4 for upload-artifact. The newly-added file is internally inconsistent on action versions; worth aligning to @v4 at minimum within this PR's file.
• Per the PR body's own follow-up list, raising cpp_threshold to 80 once the gaps land is the obvious next step — and DynBitset.cpp at 0% / 11 lines is the cheapest gap to close first. Worth filing as an issue tied to this PR's merge so the follow-up doesn't get lost.

---

Scope correction note: A prior version of this review flagged (a) the --isolated flag in XmsCore-CI.yaml's lint step and (b) the README Testing section recommending python -m unittest discover. Both of those targets are pre-existing code that this PR does not modify, so per the user's "changes-only" review policy they have been dropped from the recommendations. The .flake8-is-inert observation is retained but reframed to avoid prescribing changes to pre-existing CI lines.