Bug 157433 – [COCOA] Disable HAVE_DTRACE at build time

• Status: RESOLVED
• Resolution: FIXED
• Component: WebCore Misc.

[Chris Dumez]

Disable HAVE_DTRACE at build time by default for performance reasons. Also remove the JSC-related code which seems unused.

[Chris Dumez]

<rdar://problem/26148841>

[Chris Dumez]

Created attachment 278278 [details]
Patch

[Chris Dumez]

Created attachment 278385 [details]
Patch

[Mark Lam]

The EWS bots are sad.  Please fix.

[Chris Dumez]

Created attachment 278386 [details]
Patch

[Mark Lam]

Comment on attachment 278386 [details]
Patch

r=me if the bots are happy.

[WebKit Commit Bot]

Comment on attachment 278386 [details]
Patch

Clearing flags on attachment: 278386

Committed r200568: <http://trac.webkit.org/changeset/200568>

[WebKit Commit Bot]

All reviewed patches have been landed.  Closing bug.

(In reply to comment #7)
> Comment on attachment 278386 [details]
> Patch
> 
> Clearing flags on attachment: 278386
> 
> Committed r200568: <http://trac.webkit.org/changeset/200568>

Changes to WebKit2 need to be reviewed and or authored by one of the people listed in <http://trac.webkit.org/browser/trunk/Source/WebKit2/Owners?rev=198224>.

Comment on attachment 278386 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=278386&action=review

> Source/WebKit2/Platform/IPC/MessageRecorder.cpp:30
> +#if HAVE(DTRACE)
> +

It seems like the complaint here was that UUID generation in the MessageRecorder constructor had a measurable impact on performance. The right way to address this would be to either use a cheaper unique identifier (a sequence number in the low bits and the process ID in the upper bits, for example) or the generate the unique identifiers only when the DTrace probe is enabled.

Pretending that OS X and iOS don’t have DTrace is an unnecessarily heavy-handed approach to take.

[Chris Dumez]

(In reply to comment #10)
> Comment on attachment 278386 [details]
> Patch
> 
> View in context:
> https://bugs.webkit.org/attachment.cgi?id=278386&action=review
> 
> > Source/WebKit2/Platform/IPC/MessageRecorder.cpp:30
> > +#if HAVE(DTRACE)
> > +
> 
> It seems like the complaint here was that UUID generation in the
> MessageRecorder constructor had a measurable impact on performance. The
> right way to address this would be to either use a cheaper unique identifier
> (a sequence number in the low bits and the process ID in the upper bits, for
> example) or the generate the unique identifiers only when the DTrace probe
> is enabled.
> 
> Pretending that OS X and iOS don’t have DTrace is an unnecessarily
> heavy-handed approach

Sure, we should make this less expensive if we need to re enable DTRACE. I did discuss this with Sam and Tim (who added this code) and they agreed to disable it for now. It does not seem people (besides Tim) were using this code, it has also never really worked on iOS as far as I know.