Bug 216371 – Some WebAudio tests give different output on different machines

• Component: Web Audio
• Status: RESOLVED FIXED

[Chris Dumez]

Some WebAudio tests give different output on different machines, because they rely on the hardware's preferred sample rate.

[Chris Dumez]

Created attachment 408453 [details]
Patch

[Alex Christensen]

Comment on attachment 408453 [details]
Patch

Could you pass it through the WebPageCreationParameters instead of making new bundle SPI?

[Chris Dumez]

(In reply to Alex Christensen from comment #2)
> Comment on attachment 408453 [details]
> Patch
> 
> Could you pass it through the WebPageCreationParameters instead of making
> new bundle SPI?

I was trying to reduce the amount of plumbing necessary. Also, this hardwareSampleRate() is a global static so I did not really want this to be per page.

[Alex Christensen]

Comment on attachment 408453 [details]
Patch

Why do you pass a WKBundleRef then?
I think we should not be expanding the bundle SPI, even for testing, if we can get away with not expanding it.  In this case we can.

[Chris Dumez]

(In reply to Alex Christensen from comment #4)
> Comment on attachment 408453 [details]
> Patch
> 
> Why do you pass a WKBundleRef then?

> I think we should not be expanding the bundle SPI, even for testing, if we
> can get away with not expanding it.  In this case we can.

Can you clarify your proposal? You want the SPI to be on the WKWebView even though the sample rate is process-wide? Also, how about when the view swaps process? I would need to add logic to forward the sample rate to the new process?

Doing it like you suggest adds more code and complexity.

[Alex Christensen]

Comment on attachment 408453 [details]
Patch

I think we should make a WKWebViewConfigurationPrivateForTesting.h like we have WKWebViewPrivateForTesting.h, add an override there, then copy it to an Optional on WebPageCreationParameters, then in the WebPage constructor if it's not nullopt we set it globally in that process.  When we process swap, a new WebPageCreationParameters is already sent to the new process.  It's not perfect, but it's good enough for test-only SPI.
If you really think that's not a good direction and would prefer increasing our dependence on the bundle, r=me

[Chris Dumez]

Committed r266842: <https://trac.webkit.org/changeset/266842>

[Radar WebKit Bug Importer]

<rdar://problem/68653909>

[Alex Christensen]

I think the best solution would have been to modify the web platform tests to print out a deterministic output.

[Chris Dumez]

Reverted r266842 and r266883 for reason:

Causes some assertions to be hit in debug builds

Committed r266895: <https://trac.webkit.org/changeset/266895>

[Chris Dumez]

Created attachment 408710 [details]
Patch

[Chris Dumez]

(In reply to Chris Dumez from comment #11)
> Created attachment 408710 [details]
> Patch

New approach so requesting review again.

[Alex Christensen]

Comment on attachment 408710 [details]
Patch

For some reason I don't understand, I think this is better than bundle SPI.
I think it would be even better to modify the test to have different output (just PASS/FAIL instead of printing audio sample rates) but this is fine too.

[Chris Dumez]

Comment on attachment 408710 [details]
Patch

Clearing flags on attachment: 408710

Committed r267018: <https://trac.webkit.org/changeset/267018>

[Chris Dumez]

All reviewed patches have been landed.  Closing bug.