On Fri, 13 May 2016 14:26:59 +0200, Ricard Wanderlof wrote:
On Thu, 12 May 2016, Takashi Iwai wrote:
Yeah, that's why I stated the primary question. The atomic operation there is mostly cosmetic, not really helpful. In the whole other operation, we don't guarantee the thread safety at all. So, implementing the atomic op only in a very small part of the whole code appears like a superfluous optimization to me.
The question is, if we decide to remove this atomic stuff, how do we verify that it actually causes no problems, on all architectures, etc?
It's not what one can show in 100%, as you know. And, it's the reason why the code is still left over years...
IMO, we just need to make a clear statement about the thread safety of alsa-lib. For a single threaded op, it can't give any regression, of course.
Takashi