gcov 12 and 13 / lcov 2.0 version mismatch when collectin coverage with gcc 12 and 13 on ubuntu-24.04

[mknaranja]

Describe the bug

With gcc, gcov, lcov my CI pipeline fails since version mismatches happen.

To Reproduce

Steps to reproduce the behavior:

More or less (see https://github.com/SciCompMod/GMGPolar/tree/main/.github/actions):

cmake -DCMAKE_BUILD_TYPE=Debug -DGMGPOLAR_TEST_COVERAGE=ON .. 
make -j4
cmake --build . --target coverage/fast 

Expected behavior

The code above works locally with lcov-1.14 and, e.g., gcc/gcov 12

Screenshots

With gcc/gcov 12 I get:

Capturing coverage data from .
geninfo cmd: '/usr/bin/geninfo . --output-filename coverage.base --gcov-tool /usr/bin/gcov-12 --base-directory /home/runner/work/GMGPolar/GMGPolar --initial --memory 0'
Found gcov version: 12.3.0
Using intermediate gcov format
Writing temporary data to /tmp/geninfo_datQy7t
Scanning . for .gcno files ...
Found 94 graph files in .
Processing ./tests/CMakeFiles/gmgpolar_tests.dir/comparetest.cpp.gcno
geninfo: ERROR: mismatched end line for _ZN42Test_Parameters_Initialize_Parameters_Test8TestBodyEv at /home/runner/work/GMGPolar/GMGPolar/tests/comparetest.cpp:47: 47 -> 52
	(use "geninfo --ignore-errors mismatch ..." to bypass this error)
gmake[1]: *** [CMakeFiles/coverage.dir/build.make:73: CMakeFiles/coverage] Error 1
gmake: *** [Makefile:174: coverage/fast] Error 2

see: https://github.com/SciCompMod/GMGPolar/actions/runs/12361346502/job/34498552858

With gcc/gcov 13 I get:

stderr:
  /home/runner/work/GMGPolar/GMGPolar/build/tests/CMakeFiles/gmgpolar_tests.dir/comparetest.cpp.gcno:version 'B33*', prefer 'B32*'
geninfo: ERROR: Incompatible GCC/GCOV version found while processing /home/runner/work/GMGPolar/GMGPolar/build/tests/CMakeFiles/gmgpolar_tests.dir/comparetest.cpp.gcno:
	Your test was built with 'B33*'.
	You are trying to capture with gcov tool '/usr/bin/gcov-13' which is version 'B32*'.
	(use "geninfo --ignore-errors version ..." to bypass this error)
gmake[1]: *** [CMakeFiles/coverage.dir/build.make:73: CMakeFiles/coverage] Error 1
gmake: *** [Makefile:174: coverage/fast] Error 2

see: https://github.com/SciCompMod/GMGPolar/actions/runs/12361745308/job/34499718252

Desktop (please complete the following information)

• OS: GitHub ubuntu-latest
• GCC version: gcc-12 and gcc-13 (github apt install gcc-12/13)
• GCOVR version: gcov-12.3 and 13.2 (set with CMAKE and set(GCOV_PATH "/usr/bin/gcov-13"))
• Project directory layout:
  • Roots
  • Object directories

Additional context

Appeared with recent move to ubuntu-24.04.

[henry2cox]

Newer lcov does a fair amount of data consistency checking (...primarily due to sad history of confusing or misleading results).
Your first issue is one such.

• Your function appears in more than one compilation unit (or appears multiple times in this particular gcov output) - and the 'end' line reported is different between the appearances.
• A common cause is an incomplete build or mismatched libraries such that you are trying to merge data from two different versions of source code.
• If you are certain that your build was complete and there is no version mismatch: then another common cause is a bug (or bugs) in your compiler. It turns out that the coverage infrastructure is not the most well tested feature on the planet - and it is extremely difficult to get right. My hat is off to the compiler guys :-)
  Unlike debug data (which simply confuses the debugger and/or the user when it is not right), coverage data leaves artifacts.

Your second issue is almost certainly an incomplete or incorrect build.
gcov is closely tied to gcc - and you must use the same version of both if you want to see correct result. It is possible that mixed versions of gcc/gcov will work - but there is no guarantee.

For both of these: the message also tells you what to do if you want to ignore the errors and continue.
See the --ignore-errors section in the man page for details.

When errors are downgraded: the tool tries to do something/tries its best guess. (There are some controls that you can use to tell lcov what to do in certain situations - see the man pages for details.)
Ignoring or suppressing errors might lead to additional errors and/or to confusing, misleading, or incorrect results. 'end line' errors are likely innocuous (mostly), and really affect only your ability to accurately exclude particular functions and associated coverpoints.

[mknaranja]

Thank you for your response, it actually offers me/us the workaround of ignoring errors but I'd like to avoid such an option/workaround so I was searching for a true solution. We will check again with your comments.

[henry2cox]

If you think about it - the older versions were just silently ignoring all the errors and not telling you about them.

If you were happy with that behaviour: then you might be interested in the --keep-going flag :-)

On the other hand...if results don't have to be correct, then we can get them much faster.