You may have noticed make check still fails on a buildbot (mx1). This is because mx1 host has no sound card, so SDL_OpenAudio fails, and the SDL callback *never* arrives.
Now, our problem is that the core lib relies on the assumption that sound will be fetched at regular intervals, while SDL doesn't guarantee that. We have two options here: 1) force SDL to always fetch sound at regular interval (starting our own thread if the SDL one doesn't start). 2) have sound fetched in main thread on core ::advance. Option 2 has the advantage to drop another forced thread in the core. The main thread might be filling a queue for the system mixer to consume, in whatever way the hosting application decide to use (threading or event...). To allow threading, the only thing to mutex-protect would be the samples queue. Downside would be that if libcore is too slow in pushing samples we'd have a sample buffer underrun and if it is too fast we'd have an overrun (or we'd grow the queue in unexpected ways). Both occurrences (underrun and overrun) would need to be signalled somehow back to core for triggering some synchronization mechanism. Option 1 will require us to have NullSoundHandler also spawn a thread for fetching samples. And we'd need to implement a TestingSoundHandler for manual fetching. (the TestingSoundHandler would not be needed in option 2 as the NullSoundHandler would just consume all samples on ::advance, from a libcore callback). Comments ? Ideas ? --strk; () ASCII Ribbon Campaign /\ Keep it simple! _______________________________________________ Gnash-dev mailing list [email protected] http://lists.gnu.org/mailman/listinfo/gnash-dev
