Re: [Gnash-commit] [SCM] Gnash branch, master, updated. release_0_8_9_start-91-gc03ea7e

[Bastiaan Jacques]

On Sat, 19 Feb 2011, Sandro Santilli wrote:

> On Fri, Feb 18, 2011 at 11:33:05PM +0100, Bastiaan Jacques wrote:
> Worth triple checking all accesses to that function, with pencil and paper.
> Valgrind might have a say too (what's it called ? hellgrind ?)

Actually, Valgrind led me to the cause of the bug. It doesn't complain
anymore after the patch.

Helgrind shows lots of potential threading errors. Most are lock order
violations. Some are potential data races due to methods also not being
mutex protected. (For example, getFinalVolume() and setFinalVolume().) I
see no errors relating to unplugInputStream.

> In particular, I'm concerned about deadlocks (locking overridden
> functions of sound_handler_sdl triggering a call to unplugInputStream
> again)

So far as I can tell, unplugInputStream is called only by Sound_as and
Netstream_as, who do not lock sound_handler_sdl's mutex.

Bastiaan