URL:
https://github.com/SSSD/sssd/pull/52
Title: #52: CI: Remove dlopen-test from valgrind blacklist
fidencio commented:
"""
On Mon, Oct 17, 2016 at 11:51 PM, lslebodn <notifications(a)github.com> wrote:
On (17/10/16 14:43), fidencio wrote:
>On Mon, Oct 17, 2016 at 11:37 PM, lslebodn <notifications(a)github.com>
wrote:
>
>> On (17/10/16 14:26), fidencio wrote:
>> >On Mon, Oct 17, 2016 at 10:52 PM, lslebodn <notifications(a)github.com>
>> wrote:
>> >
>> >> On (17/10/16 12:34), fidencio wrote:
>> >> >Please, refer to e1a58f3d in the commit message.
>> >> >
>> >> Why? The text is more important. Rest is useless.
>> >>
>> >
>> >Well, you're basically reverting that commit.
>> >But feel free to ignore in any case.
>> >
>> >
>> >>
>> >> >This is a genuine question (even in case it's a dumb one), but
do we
>> >> really need to call dlclose() in our tests? Can't we relax this in
>> order to
>> >> have a meaningful backtrace?
>> >> >
>> >> Let assume:
>> >> * dlclose was not called
>> >> * libraryA is linked with libtalloc and libtevent
>> >> * libraryB is not linked with libtalloc (even though it should be
>> >> * dlopen test test libraries in following order: 1. libraryA; 2.
>> libraryB
>> >>
>> >> Result: missing dependency in libraryB would not be found because
>> >> libraryA and its dependencies are still loaded in memory.
>> >>
>> >
>> >Wouldn't make sense to have two tests then? One as it is nowadays. In
case
>> >the first passes we run the second one, not calling dlclose() and just
>> >checking for leaks?
>> >
>> What would be a purpose of second test?
>> Leaks are not checked in tests itself but by valgrind.
>>
>
>By not calling dlclose() during the second run couldn't you have a "not
>meaningless" (part of the) backtrace?
>
If you can reproduce any valgrind error reported by dlopen-test
then we can consider to do it. Otherwise NACK to the idea.
I do not want to implement something which I cannot verify whether it work
or
no.
Then I'm truly lost here, sorry.
If you think we cannot reproduce any valgrind error reported by dlopen-test
why should we remove dlopen-test from valgrind blacklist in the first place?
"""
See the full comment at
https://github.com/SSSD/sssd/pull/52#issuecomment-254346344