Bumped gcovr off the 4.1 pin it had been held on since 2018

[fdesbiens]

scripts/install.sh pinned gcovr==4.1, released in 2018. That version has neither --json nor --add-tracefile, which is exactly how the five build configurations would be merged into a single coverage report — so the pin is a prerequisite for the rest of the coverage work, not just staleness.

This moves it to 8.6, the current release. The pin stays exact, so a coverage percentage remains comparable between runs, and it stays hand-moved: it lives in a shell script that no Dependabot ecosystem can parse. The new comment in install.sh says so, since #662 might otherwise imply this is now automated.

Nothing else is in this pull request on purpose — the number movement it might cause should not be confused with one caused by a later change.

It moves no number

Measured on the default_build_coverage tree of test/tx, over the same .gcda with the same gcov, varying only the gcovr version:

gcovr	lines-valid	branches-valid	files
4.1	3827	1994	177
7.0	3827	1994	177
8.3	3827	1994	177
8.6	3827	1994	177

The denominator does not move with the tool. The plan behind this work expected 3822 → 3827 across the bump; that figure does not reproduce — not under gcc-13, not under gcc-14, and not with --object-directory dropped. The only variant that changes the count is dropping the -f ../../../common/src filter, which collapses the report to zero files.

Two findings worth more than the bump

The coverage numerator is not deterministic. On an identical tree with an identical compiler, three consecutive runs of the full suite — all 96 tests passing every time — reported:

run	lines	branches	uncovered
1	3826/3827	1992/1994	tx_thread_system_resume.c:529
2	3827/3827	1993/1994	none
3	3827/3827	1993/1994	none

That line is _tx_thread_system_return(); — the "preemption is needed, return to the system" path of _tx_thread_system_resume — and it takes its guarding branch with it. It has been characterised as never executed by the suite. It is executed on some runs and not others. Two consequences: a coverage floor set at the observed maximum would flake, and the real fix is a test that takes that path deliberately rather than by timing.

--xml-pretty and --object-directory still work on 8.6, but only as deprecated aliases for --cobertura-pretty and --gcov-object-directory. Nothing to change here — worth knowing for whoever removes --object-directory, which measurably does nothing anyway.

Verification

Reading gcc-13 output — the compiler the runners actually use, since install.sh installs plain gcc and nothing sets CC — gcovr 8.6 runs the existing test/tx/cmake/coverage.sh unchanged end to end: Cobertura XML plus 181 HTML files, same 177 classes, exit 0. 96/96 tests passed.

Assisted-by: Claude Opus 5 noreply@anthropic.com