On Thu, Aug 1, 2013 at 5:41 PM, Jer Noble <jer.noble@apple.com> wrote:
>
> On Aug 1, 2013, at 7:13 AM, Adam Goode <agoode@google.com> wrote:
>
> That document is a bit misleading/outdated when it comes to data races
> being "harmless". This more recent article illustrates these issues:
>
> http://software.intel.com/en-us/blogs/2013/01/06/benign-data-races-what-could-possibly-go-wrong
>
>
> Perhaps you can clarify something for me. In both the above page and the
> one Ehsan linked to earlier (<http://blog.regehr.org/archives/490>), a
> "data race" occurrs when two threads are writing to the same memory. In
> the WebAudio API "data races" we are discussing, the main thread is
> potentially writing to memory which the audio thread is reading. I.e., the
> audio thread never writes to these buffers.
>
Actually, that's not true. See my example here:
http://people.opera.com/mage/test/webaudio-shared-data.html
(...as discussed in
http://lists.w3.org/Archives/Public/public-audio/2013JulSep/0321.html).
It's definitely a case of the audio thread writing to a buffer that the JS
main thread can observe.
/Marcus
>
> Should this still be considered a "data race"?
>
> -Jer
>
>