Re: make check sometimes hangs

[Stefan Westerfeld <stefan space twc de>]

   Hi!

(sorry my first reply didn't go to the list, resending)

On Thu, Nov 10, 2016 at 11:43:34PM +0100, Tim Janik wrote:
> On 10.11.2016 22:13, Stefan Westerfeld wrote:
> > If I run
> >
> >   cd tests/audio; while :; do make check; done
> >
> > in beast, I'd expect that the audio tests would be done over and over again,
> > forever. However, what happens here is that after some time, make check hangs,
> > nothing happens any more. Looking at pstree there is
> >
> > -bash---python---sh---python---sh---bse2wav.sh---lt-bsetool---2*[{lt-bsetool}]
>
> Hm, I wonder what's python doing there, calling the shell?

Sorry, wrong hanging bsetool process (this was my SpectMorph evaluation
script). The correct pstree output is:

-bash---make---make---lt-bsetool---2*[{lt-bsetool}]

> > so bsetool for some reason hangs. Without CPU usage. Maybe some kind of
> > deadlock, possibly triggered by the use of the null driver?
> >
> > That wasn't always the case. I saw the problem when using a variant of
> > bse2wav.sh for SpectMorph testing - compared to make check in beast, I need to
> > render a lot more wave files here. And that used to work when I made the last
> > SpectMorph release, about six weeks ago.
>
> The main change that went into the DSP logic is parallel processing.
>
> Just to rule that out, you can try disabling it with:
>
> +++ bse/bseenginemaster.cc
> @@ -908,8 +908,8 @@ engine_start_slaves ()
>  {
>    assert_return (slaves_running == false);
>    slaves_running = true;
> -  const uint n_cpus = Rapicorn::ThisThread::online_cpus();
> +  const uint n_cpus = 1;
>    const uint n_slaves = std::max (1u, n_cpus) - 1;
>    for (uint i = 0; i < n_slaves; i++)
>      slave_threads.push_back (new std::thread (engine_run_slave));
>  }

Yes, this seems to be the change to blame; with that applied, make check no
longer hangs.

   Cu... Stefan
-- 
Stefan Westerfeld, http://space.twc.de/~stefan