On Sat, Apr 4, 2020 at 5:56 PM Petr Jelinek <petr@2ndquadrant.com> wrote:
> On 04/04/2020 07:25, Tom Lane wrote:
> > Petr Jelinek <petr@2ndquadrant.com> writes:
> >> On 03/04/2020 17:51, Tom Lane wrote:
> >>> But the forked-off children have to write the gcov files independently,
> >>> don't they?
> >
> >> Hmm that's very good point. I did see these missing coverage issue when
> >> running tests that explicitly start more instances of postgres before
> >> though. And with some quick googling, parallel testing seems to be issue
> >> with gcov for more people.
> >
> > I poked around and found this:
> >
> > https://gcc.gnu.org/legacy-ml/gcc-help/2005-11/msg00074.html
> >
> > which says
> >
> > gcov instrumentation is multi-process safe, but not multi-thread
> > safe. The multi-processing safety relies on OS level file locking,
> > which is not available on some systems.
> >
> > That would explain why it works for me, but then there's a question
> > of why it doesn't work for you ...
>
> Hmm, I wonder if it has something to do with docker then (I rarely run
> any tests directly on the main system nowadays). But that does not
> explain why it does not work for Amit either.
One thing to I must clarify: coverage for most of pgoutput.c looks
okay on each run. I am concerned that the coverage for the code added
by the patch is shown to be close to zero, which is a mystery to me,
because I can confirm by other means such as debugging elogs() to next
to the new code that the newly added tests do cover them.
--
Thank you,
Amit Langote
EnterpriseDB: http://www.enterprisedb.com
