[bug:1721353] Run 'line-coverage' regression runs on a latest fedora machine (say fedora30).

[gluster-ant]

URL: https://bugzilla.redhat.com/1721353
Creator: atumball at redhat
Time: 20190618T04:49:37

Description of problem:

I suspect the code coverage tool with centos7 is not covering all the details.

Check https://build.gluster.org/job/line-coverage/lastCompletedBuild/Line_20Coverage_20Report/libglusterfs/src/glusterfs/stack.h.gcov.html for example, and you can see that it says, 17/41 functions are covered. But if you notice there are only 17 inline functions, and all of them are actually covered.

If it reported it properly, we should have had 100% coverage there.

With that detail, I hope having newer version would get this sorted.

Also note, we recently fixed all the issues with python3 in regression runs too, so moving to fedora should help us identify issues sooner with python3 (if any).

Version-Release number of selected component (if applicable):
master

How reproducible:
100%

Expected results:
Nightly line-coverage runs to run on fedora systems.

[gluster-ant]

Time: 20190618T06:23:14
atumball at redhat commented:
Ok, when I used the lcov tool with the same commands as that of lcov.sh from build-jobs repo, I got below numbers for stack.h (which I used as an example above).

on current builder : stack.h - lines(262/497 - 52.7%), functions(17/41 - 41.5%)
on fedora29 (local): stack.h - lines(94/111 - 84.7%), functions(6/7 - 85.7%)

I hope just by running the regression on fedora, we would get more up-to-date information, and more coverage details. Just note that I suspect this to be more of an header file specific details, and even then, up-to-date information is better than stale info.

[gluster-ant]

Time: 20190618T07:31:30
khiremat at redhat commented:
I have also witnessed the small difference when I was working to improve lcov of glusterd-georep.c. On fedora 30, I used to see 70.1 % but on centos 69.9 %.
I didn't spend time debugging that though. Didn't expect it was platform dependent.