Hi, list,
This is something that I wanted to add to the incubator-nuttx-testing,
but it appears that no issues are supported in that repository. That is
a problem; issues should be enabled for the testing repository as well.
Anyway...
I was looking at some successful logs from recent checks of one of my
PRs and saw undetected errors. These were not declared as "errors" by
GCC, but only as "warnings" so they did not cause the build the fail.
This leads me think that we should fail the check if there are new
warnings introduced by build.
Detecting new warnings might be tricky, however. I did this manually
when I ran the build tests myself: I just kept the last good test log
and diff'ed the current test log with that known good log. Any new
warnings would show up in the diffs. The tricky part is that there is a
lot of other kruft that shows up in the diffs too like data differences
and line number differences.
So I don't know if this would be worth the effort. The benefits would be:
1) We could find undetected errors that currently only manifest as a
warnings.
2) We have too many warnings now and we should not permit new warnings
to creep into the build. I would like to be able to someday trick all
warnings as errors. This would lead to higher quality code. The number
of unique warnings is really rather small now (and most generated by
explicit #warning commands). If we keep the number of new warnings
down, then getting to zero warnings could be achievable.
Greg
