1103824 - Investigate first initialization of IAudioClient used in libcubeb

Status: RESOLVED WONTFIX

(Core :: Audio/Video: cubeb, defect, P3)

Description

[Matthew Gregan [:kinetik]]

While looking at bug 1096717, I noticed that the IAudioClient docs have the following warning:

    Note: In Windows 8, the first use of IAudioClient to access the audio device
          should be on the STA thread. Calls from an MTA thread may result in
          undefined behavior.

It doesn't look like we do anything special to ensure this happens in libcubeb, so we'll need to address this at some stage to avoid surprises.

Comment 1

[Matthew Gregan [:kinetik]]

Attachment: bug1103824_v0.patch

Initialize (and discard, to avoid keeping resources alive via sCubebContext that may never be used) a cubeb context on the main thread.

Change cubeb's auto_com to treat RPC_E_CHANGED_MODE as an error, since that's really telling us that we're asking for (and not getting) MTA on a thread that has already been initialized as STA.

Check that we get an STA initialization in cubeb_init the first time it's called, and use that opportunity to completely initialize (and then discard) an IAudioClient.

Comment 2

[Matthew Gregan [:kinetik]]

Try push looks okay: https://treeherder.mozilla.org/#/jobs?repo=try&revision=52e645df36d1

Comment 3

[Pulsebot]

https://hg.mozilla.org/integration/mozilla-inbound/rev/dd1417716d76

Comment 4

[Pulsebot]

Backout:
https://hg.mozilla.org/integration/mozilla-inbound/rev/e7cd98825197

Comment 5

[Matthew Gregan [:kinetik]]

I looked at the OpenSL failures but not sure what the cause is yet.  I'll reland this with the CubebUtils.cpp changes wrapped in an XP_WIN ifdef (this issue only matters on Windows) and file a followup to address the OpenSL issues.

Comment 6

[Pulsebot]

https://hg.mozilla.org/integration/mozilla-inbound/rev/de34ec0570a3

Comment 7

[Pulsebot]

Backout:
https://hg.mozilla.org/integration/mozilla-inbound/rev/9665b9f50268

Comment 8

[Carsten Book [:Tomcat]]

sorry had to back this out since this seems to have caused somehow https://treeherder.mozilla.org/logviewer.html#?job_id=9992633&repo=mozilla-inbound

Comment 9

[Matthew Gregan [:kinetik]]

Will repush once I've found the cause.

Comment 10

[Matthew Gregan [:kinetik]]

Rather than init and destroy a context in CubebUtils::InitLibrary(), I explicitly make the first CubebUtils::GetCubebContext() from the main thread for the users of libcubeb.  This looks good on try: https://treeherder.mozilla.org/#/jobs?repo=try&revision=47a6687a973f

Comment 11

[Pulsebot]

https://hg.mozilla.org/integration/mozilla-inbound/rev/aeaa70495f1d

Comment 12

[Carsten Book [:Tomcat]]

https://hg.mozilla.org/mozilla-central/rev/aeaa70495f1d

Comment 13

[Pulsebot]

https://hg.mozilla.org/integration/mozilla-inbound/rev/b10c59539ec8

Comment 14

[Matthew Gregan [:kinetik]]

I backed this out due to the regression reported in bug 1176499, along with the fact that this fix was speculative.

Comment 15

[Carsten Book [:Tomcat]]

backed out also from m-c in https://hg.mozilla.org/mozilla-central/rev/b10c59539ec8

Comment 16

[Bulk Bug Changes for mreavy's org]

Mass change P2->P3 to align with new Mozilla triage process.

Comment 17

[Matthew Gregan [:kinetik]]

I don't think we've ever seen evidence that this matters. It's possible it's only relevant in configurations we don't exercise (e.g. exclusive mode).