Bug 46220 – Fix some Objective-C GC problems and use RetainPtr instead of HardRetain/Release

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

[Darin Adler]

Fix some Objective-C GC problems and use RetainPtr instead of HardRetain/Release

[Darin Adler]

Created attachment 68299 [details]
Patch

[Early Warning System Bot]

Attachment 68299 [details] did not build on qt:
Build output: http://queues.webkit.org/results/3996095

[WebKit Review Bot]

Attachment 68299 [details] did not build on gtk:
Build output: http://queues.webkit.org/results/3996096

[WebKit Review Bot]

Attachment 68299 [details] did not build on chromium:
Build output: http://queues.webkit.org/results/4081046

[Darin Adler]

Created attachment 68312 [details]
Patch

[Anders Carlsson]

Comment on attachment 68312 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=68312&action=review

> WebCore/bindings/objc/ObjCEventListener.h:59
> +        typedef id <DOMEventListener> ObjCListener;

There shouldn't be a space after id here.

> WebCore/bridge/objc/objc_instance.mm:88
> +    s_exception = nil;

How will not retaining here work in a non-GC environment?

[Darin Adler]

(In reply to comment #6)
> > WebCore/bridge/objc/objc_instance.mm:88
> > +    s_exception = nil;
> 
> How will not retaining here work in a non-GC environment?

This is a global in an Objective-C++ source file; the compiler and garbage collection runtime already handles this correctly without explicit retain/release. We have many other globals like this. An earlier version of my patch attempted to “fix” all those, but then I talked to Tim Hatcher, read some documentation, and did a little testing, and saw this worked fine.

In fact, the problem in ThemeMac.mm quite possibly is Leopard-only because it seems that the SnowLeopard version of GC can handle modifying an Objective-C global through a pointer even though the Leopard version cannot. But the fix is OK for all platforms.

[Darin Adler]

(In reply to comment #6)
> > WebCore/bindings/objc/ObjCEventListener.h:59
> > +        typedef id <DOMEventListener> ObjCListener;
> 
> There shouldn't be a space after id here.

I did a quick grep and I see that we have id without a space in 6 places in WebCore and id with a space in about 50 places, including public headers.

I think we should standardize on one or the other, but for this patch I will leave this site alone -- it matches what used to be there.

[Darin Adler]

(In reply to comment #7)
> (In reply to comment #6)
> > > WebCore/bridge/objc/objc_instance.mm:88
> > > +    s_exception = nil;
> > 
> > How will not retaining here work in a non-GC environment?
> 
> This is a global in an Objective-C++ source file; the compiler and garbage collection runtime already handles this correctly without explicit retain/release. We have many other globals like this. An earlier version of my patch attempted to “fix” all those, but then I talked to Tim Hatcher, read some documentation, and did a little testing, and saw this worked fine.
> 
> In fact, the problem in ThemeMac.mm quite possibly is Leopard-only because it seems that the SnowLeopard version of GC can handle modifying an Objective-C global through a pointer even though the Leopard version cannot. But the fix is OK for all platforms.

Oops! You said “non-GC”!

That change was bad. I have to retain/release. Fixed now.

[Darin Adler]

Committed r68054: <http://trac.webkit.org/changeset/68054>