On Mon, 24 Sep 2012, Takashi Iwai wrote:
At Mon, 24 Sep 2012 16:20:50 +0200 (CEST), Henning Thielemann wrote:
For me valgrind is pretty useless if I have to manually distinguish real problems from false alarms in a long list of valgrind messages. Are you sure that it is generally ok to pass pointers to uninitialized data to ioctl also if in this particular case the pointer is not touched by ioctl?
Only client field is read, and this is the field properly initialized in alsa-lib code. Others are fields to be filled by the kernel, so it doesn't matter at all what's left there.
If yes, then I think valgrind must be somehow teached to ignore this particular case.
I have no idea how to teach valgrind, sorry.
I also asked in the valgrind mailing list and got this answer: http://sourceforge.net/mailarchive/forum.php?thread_name=506089CA.5030309%40...
Of course, I can suppress the message for me, but I expect that more people will stumble on it. Thus I'd prefer a resolution of the problem in ALSA.