Bug 226001 - Use CheckedLock more in WebKit2 code
Summary: Use CheckedLock more in WebKit2 code
Status: RESOLVED FIXED
Alias: None
Product: WebKit
Classification: Unclassified
Component: WebKit2 (show other bugs)
Version: WebKit Nightly Build
Hardware: Unspecified Unspecified
: P2 Normal
Assignee: Chris Dumez
URL:
Keywords: InRadar
Depends on:
Blocks:
 
Reported: 2021-05-19 19:29 PDT by Chris Dumez
Modified: 2021-05-20 09:12 PDT (History)
10 users (show)

See Also:

Attachments
Patch (48.00 KB, patch)
2021-05-19 19:30 PDT, Chris Dumez 		no flags Details | Formatted Diff | Diff
Patch (50.13 KB, patch)
2021-05-19 23:09 PDT, Chris Dumez 		no flags Details | Formatted Diff | Diff
Patch (49.05 KB, patch)
2021-05-20 07:37 PDT, Chris Dumez 		no flags Details | Formatted Diff | Diff
Show Obsolete (2) View All Add an attachment (proposed patch, testcase, etc.)

Note You need to log in before you can comment on or make changes to this bug.
Description Chris Dumez 2021-05-19 19:29:07 PDT 
Use CheckedLock more in WebKit2 code to benefit for Clang Thread Safety Analysis.
Comment 1 Chris Dumez 2021-05-19 19:30:59 PDT 
Created attachment 429127 [details]
Patch
Comment 2 Alex Christensen 2021-05-19 20:52:55 PDT 
Comment on attachment 429127 [details]
Patch

I like WTF_GUARDED_BY_LOCK
Comment 3 Chris Dumez 2021-05-19 23:09:39 PDT 
Created attachment 429140 [details]
Patch
Comment 4 Kimmo Kinnunen 2021-05-20 01:21:52 PDT 
Comment on attachment 429140 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=429140&action=review

great work.
maybe the lockable() could be reverted / not added.

> Source/WTF/ChangeLog:8
> +        Add lockable() getter to Locker<CheckedLock> to match the regular Locker<T>.

I didn't see the use case to add that.
That enables the anti-pattern that is being invoked also in this patch.

> Source/WebKit/WebProcess/GPU/webrtc/LibWebRTCCodecs.cpp:177
> +void LibWebRTCCodecs::ensureGPUProcessConnectionOnMainThread(Locker<CheckedLock>& locker)

For CheckedLock, this sort of unused parameter pattern is now an anti-pattern.

The annotation in the header file would be :
   void LibWebRTCCodecs::ensureGPUProcessConnectionOnMainThread() WTF_REQUIRES_LOCK(m_connectionLock);
Comment 5 Chris Dumez 2021-05-20 07:37:17 PDT 
Created attachment 429171 [details]
Patch
Comment 6 Chris Dumez 2021-05-20 07:38:48 PDT 
(In reply to Kimmo Kinnunen from comment #4)
> Comment on attachment 429140 [details]
> Patch
> 
> View in context:
> https://bugs.webkit.org/attachment.cgi?id=429140&action=review
> 
> great work.
> maybe the lockable() could be reverted / not added.
> 
> > Source/WTF/ChangeLog:8
> > +        Add lockable() getter to Locker<CheckedLock> to match the regular Locker<T>.
> 
> I didn't see the use case to add that.
> That enables the anti-pattern that is being invoked also in this patch.

Yes, definitely not needed if we get rid of the "Passing Locker as parameter" pattern. I dropped this.

> > Source/WebKit/WebProcess/GPU/webrtc/LibWebRTCCodecs.cpp:177
> > +void LibWebRTCCodecs::ensureGPUProcessConnectionOnMainThread(Locker<CheckedLock>& locker)
> 
> For CheckedLock, this sort of unused parameter pattern is now an
> anti-pattern.
> 
> The annotation in the header file would be :
>    void LibWebRTCCodecs::ensureGPUProcessConnectionOnMainThread()
> WTF_REQUIRES_LOCK(m_connectionLock);

Fixed.
Comment 7 EWS 2021-05-20 09:11:32 PDT 
Committed r277787 (237948@main): <https://commits.webkit.org/237948@main>

All reviewed patches have been landed. Closing bug and clearing flags on attachment 429171 [details].
Comment 8 Radar WebKit Bug Importer 2021-05-20 09:12:17 PDT 
<rdar://problem/78261485>