Finish migration of coverage tooling

[LittleHuba]

What

When the coverage tooling was copied from score_communication, it was completed.

This is IMO the worst situation we could end up in. We now have the "same" code in two places, with different states/bugs/maintainership.

This ticket is about finishing the migration that was started. To keep things aligned, the people that started the migration shall also finish it. Only they know why specific pieces were done different/not migrated.

Acceptance Criteria (DoD)

This repository covers all use cases of the original solution in score_communication.
score_communication is migrated to rely on this tooling for coverage.

How

No response

[dcalavrezo-qorix]

Progress update, 2026-09-30

Releases of score_coverage since the last update

• 0.3.0 (2026-09-28): gcov backend for QNX on-target coverage (score_coverage_reporter(backend = "gcov"), counters brought back by score_qnx_unit_tests, same report archive as on Linux; Rust listed as not instrumentable on that backend). Release notes.
• 0.3.1 (2026-09-29): --instrumentation_filter=^//<root>[/:] is now required in every coverage config, and both reporters warn when a report shows the pattern of a library tested from a test/ subpackage without data. Root cause found on baselibs' first QNX report (score/os at 13 % against 80 % on Linux): Bazel guesses the filter from the packages of the test targets and strips only a trailing /tests; libraries tested from a test subpackage were outside the guess, and on the gcov backend Bazel's own collector dropped their counters. With the line, the QNX report went from 82 % to 92 % lines. Tool PR: Require the instrumentation filter; warn when it was left to Bazel #12.

Target version for the rollout is now 0.3.1. Every migration PR sets the instrumentation filter.

Repo	PR	Status	Comparison old → new (record by record)
baselibs	eclipse-score/baselibs#584, #589	merged (0.1.0 → 0.2.0)	identical
baselibs, QNX on-target	eclipse-score/baselibs#626	open, ready for review	QNX report: 649 files, 92.2 % lines with the filter
communication	eclipse-score/communication#1171	open, draft (Linux + QNX, replaces the in-repo pipeline)	347 records identical in status
logging	eclipse-score/logging#338	open	99 → 117 files (headers recovered), 98/99 identical
lifecycle	eclipse-score/lifecycle#715	open	77 → 147 files (headers recovered), 75/77 identical, 2 explained
persistency	eclipse-score/persistency#450	merged	17/17 identical
kyron	eclipse-score/kyron#175	open (rules_rust bumped to 0.68.2-score)	70/74 identical, rest thread-timing variance
time	eclipse-score/time#235	open	84/86 identical, 2 environment effects, same justifications applied
inc_security_crypto	eclipse-score/inc_security_crypto#267	open	239/239 identical
inc_someip_gateway	eclipse-score/inc_someip_gateway#315	open (bump 0.2.0 → 0.3.1, was already on the tool)	48/55 identical, rest IPC timing variance

Every PR carries the before/after table in its description. "Headers recovered" means headers reached through include directories that the old pipeline dropped (coverage_tool#5, fixed in 0.2.0); nothing lost coverage anywhere.

Still to do

• orchestrator: full migration, same recipe.
• reference_integration: one bazelrc line (--run_under=@score_tooling//coverage:llvm_profile_wrapper → @score_coverage//:llvm_profile_wrapper).
• tooling: deprecate and remove coverage/ once the PRs above are merged.

Side findings worth knowing

• Every migrated repo now gets unmapped_files.txt in the archive. Each PR lists the files without any coverage data for the module owners; none of them had data before either, they were simply invisible.
• QNX branch coverage is structurally lower than LLVM's: gcov counts every conditional jump including exception-handling edges (22 % of all branches in baselibs' objects). Gate per platform, do not compare the two figures.
• The QNX toolchain download from qnx.com times out or serves wrong content now and then; several QNX jobs needed a rerun.