Gilles Chanteperdrix wrote:
>> I think my workflow was misleading regarding this. Not necessarily all
>> stuff I push into my 'for-upstream' branch is done with all tests at the
>> time when I push. Official pull request mails are supposed to mark this.
>> I'm pushing to save my work and, of course, to expose it. So I
>> appreciate comments, but I surely don't expect them within minutes.
>
> No. You still refuse to acknowledge the truth: I receive pull requests
> for untested stuff. Read John's mail again. He cherry picked changes,
> for which I received a pull request, he ran the mutex-torture tests,
> which the patchset had modified, and the test was not passing. So, the
> conclusion is that you did not run mutex-torture, which the commit had
> modified before sending the pull request. I did not have to dig long in
> the git logs to find this example. It was just yesterday. Think about it.
>
> That is precisely the situation I want to avoid from now on.
Ok. It turns out that yes, the timed mutex tests had a bug, but probably
only visible on uniprocessor machines. And I guess you test on SMP
machines. So, my apologies, the patch was tested, not enough, but tested.
>> PS: Note that this u_mode fix exposed an issue of the sigtest:
>> cancel_with_signals fails if there is some TSD destructor registered
>> that issues a Xenomai syscall. Not sure yet if it is a fundamental issue
>> or just related to the test itself.
>
> The sigtest test probably assumes that no parasite syscall occurs. Yet
> another reason for not using this u_mode implementation.
Remove that test case. I do not know what it was testing anyway.
--
Gilles.
_______________________________________________
Xenomai-core mailing list
[email protected]
https://mail.gna.org/listinfo/xenomai-core
