Kept the coverage report from the runs that most need one

[fdesbiens]

A failing test threw away coverage that had already been collected — and the run whose behaviour changed is exactly the run whose coverage is worth reading.

Measured on the failing run of 18 August (32186061073), which uploaded:

ARTIFACTS
test_reports SMP
test_reports ThreadX
test_reports FreeRTOS

No coverage_report-* artifact at all. Compare a passing run, which uploads two.

Two causes, and the workflow one is the smaller

scripts/cmake_bootstrap.sh runs under set -e, so a failing ctest aborted test() before ./coverage.sh was ever reached. The .gcda files exist by that point — nothing was missing except the step that reads them. ctest's status is now captured and returned at the end, and the summary grep is allowed to fail rather than being the thing that stops the coverage behind it.

The serial branch of the test dispatch collected no status either. Under set -e, the first failing configuration stopped the remaining four from being tested at all, and their coverage from being collected. That was cheap while the suites ran in parallel, because the parallel branch already collects exit codes from its background jobs — but #643 moved them to serial, which quietly made one failure cost the other four configurations. The serial branch now collects status the same way the parallel branch does.

Only with both fixed does the report exist, so the workflow steps that publish it can stop skipping on failure. Adding if: guards alone would have changed nothing.

Verified locally

By replacing one test binary with a stub that exits 1:

	Result
before	the 18 Aug run produced no coverage_report artifact
after, one config	run.sh test default_build_coverage exits 8, and produces coverage_report/default_build_coverage.xml — 177 files, 3804/3827 lines
after, all configs	run.sh test all exits 8, and all five configurations run rather than stopping at the first

The failure still fails. Only the reporting around it changed.

Two details worth a reviewer's eye

!cancelled() rather than always() — so a cancelled run still stops promptly. It is the idiom deploy_code_coverage already uses at the job level.

The ${{ }} wrapping is required, not decoration. A bare ! at the start of a YAML scalar opens a tag, and the file does not parse without it:

yaml.scanner.ScannerError: while scanning an anchor
expected alphabetic or numeric character, but found '&'

I hit that on the first attempt; it is worth knowing before anyone adds another negated condition here.

Where this sits

First of the coverage changes, and deliberately first: it has no effect on any coverage number, and the changes that follow — bumping gcovr off the 2018 4.1 pin, fixing the report's absolute paths, and instrumenting all five build configurations instead of one — all move numbers and will produce failing runs while they are iterated on. Doing this one last would mean working blind through exactly the part where the report matters most.