Bug 187256 – Allow to remove MediaStreamPrivate observers when iterating over observers

• Component: WebRTC
• Status: RESOLVED FIXED

[youenn fablet]

This will make code clearer to understand

[youenn fablet]

Created attachment 354806 [details]
Patch

[Eric Carlson]

Comment on attachment 354806 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=354806&action=review

> Source/WebCore/ChangeLog:8
> +        Copy the set of observers in a vector before iterating over it.

Nit: s/in a vector/to a vector/

[youenn fablet]

Created attachment 354815 [details]
Patch for landing

[Darin Adler]

Comment on attachment 354806 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=354806&action=review

> Source/WebCore/platform/mediastream/MediaStreamPrivate.cpp:99
> +        if (!m_observers.contains(observer))
> +            continue;

This technique is not guaranteed to work. Code could delete an Observer, then allocate a new Observer and the two might coincidentally use the same memory and have the same address. So checking "contains" on a set of weak pointers doesn’t give a guaranteed accurate answer.

[WebKit Commit Bot]

Comment on attachment 354815 [details]
Patch for landing

Clearing flags on attachment: 354815

Committed r238181: <https://trac.webkit.org/changeset/238181>

[WebKit Commit Bot]

All reviewed patches have been landed.  Closing bug.

[Radar WebKit Bug Importer]

<rdar://problem/46065771>

[youenn fablet]

(In reply to Darin Adler from comment #4)
> Comment on attachment 354806 [details]
> Patch
> 
> View in context:
> https://bugs.webkit.org/attachment.cgi?id=354806&action=review
> 
> > Source/WebCore/platform/mediastream/MediaStreamPrivate.cpp:99
> > +        if (!m_observers.contains(observer))
> > +            continue;
> 
> This technique is not guaranteed to work. Code could delete an Observer,
> then allocate a new Observer and the two might coincidentally use the same
> memory and have the same address. So checking "contains" on a set of weak
> pointers doesn’t give a guaranteed accurate answer.

True, this adds some potential uncertainty when such collision happens.

I guess we could have a HashSet<WeakPtr<Observer>> to fix this issue.
That would add some additional count churning but maybe this is ok since we are creating a Vector already whenever iterating over observers.

I am unsure whether we should fix this uncertainty.