Bug 141641 – [GTK] WebKitFrame objects are never released

• Status: RESOLVED
• Resolution: FIXED
• Component: WebKitGTK

[Carlos Garcia Campos]

We relay on willDestroyFrame callback of WKBundlePageLoaderClient, but this callback is never emitted and can't be emitted. It was my fault, I thought I had tested it when I added it, but the thing is that willDestroyFrame is called from WebFrameLoaderClient::frameLoaderDestroyed(), but at that point the frame has already been detached from the page, so WebFrame::page() always returns nullptr at that point. But even if we had a WebPage pointer at that time, when the page is closed, the loader client is reset before FrameLoader::detachFromParent() si called, so when WebFrameLoaderClient::frameLoaderDestroyed() is called then loader client doesn't have callbacks attached.

[Carlos Garcia Campos]

Created attachment 246647 [details]
Patch

[Martin Robinson]

Comment on attachment 246647 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=246647&action=review

> Source/WebKit2/WebProcess/InjectedBundle/API/gtk/WebKitWebPage.cpp:81
> +class WebKitFrameWrapper final: public FrameDestructionObserver {

I wonder if it makes sense to simply add the FrameDestructionObserver as a member of WebKitFrame? I worry about the possibility of introducing reference cycles, by wrapping the frame itself.

> Source/WebKit2/WebProcess/InjectedBundle/API/gtk/WebKitWebPage.cpp:123
> +static void webkitFrameDestroy(WebFrame* webFrame)
> +{
> +    webFrameMap().remove(webFrame);

This should probably be called webkitFrameDestroyed or webkitFrameDestroyedCallback, I think. It doesn't destroy the frame, it just responds to the destruction.

[Carlos Garcia Campos]

(In reply to comment #2)
> Comment on attachment 246647 [details]
> Patch
> 
> View in context:
> https://bugs.webkit.org/attachment.cgi?id=246647&action=review
> 
> > Source/WebKit2/WebProcess/InjectedBundle/API/gtk/WebKitWebPage.cpp:81
> > +class WebKitFrameWrapper final: public FrameDestructionObserver {
> 
> I wonder if it makes sense to simply add the FrameDestructionObserver as a
> member of WebKitFrame? I worry about the possibility of introducing
> reference cycles, by wrapping the frame itself.

There's no possibility of ref cycles, because the only reference of WebKitFrame is owned by the wrapper that is deleted when the WebCore::Frame is destroyed. That hasn't actually changed, since current the only references is owned by the WebFrameMap. The only difference is that now the WebFrameMap contains a WebKitFrameWrapper instead of a WebKitFrame directly. 
The destructor is in WebKitWebPage because frames are created in WebKitWebPage, which is where the WebFrameMap is.

> > Source/WebKit2/WebProcess/InjectedBundle/API/gtk/WebKitWebPage.cpp:123
> > +static void webkitFrameDestroy(WebFrame* webFrame)
> > +{
> > +    webFrameMap().remove(webFrame);
> 
> This should probably be called webkitFrameDestroyed or
> webkitFrameDestroyedCallback, I think. It doesn't destroy the frame, it just
> responds to the destruction.

It's confusing, it responds to a WebCore::Frame destruction to destroy the WebKitFrame object. But I agree it's confusing, webFrameDestroyed() could work better.

[Martin Robinson]

Comment on attachment 246647 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=246647&action=review

>>> Source/WebKit2/WebProcess/InjectedBundle/API/gtk/WebKitWebPage.cpp:123
>>> +    webFrameMap().remove(webFrame);
>> 
>> This should probably be called webkitFrameDestroyed or webkitFrameDestroyedCallback, I think. It doesn't destroy the frame, it just responds to the destruction.
> 
> It's confusing, it responds to a WebCore::Frame destruction to destroy the WebKitFrame object. But I agree it's confusing, webFrameDestroyed() could work better.

webFrameDestroyed sounds like a good name as well.

[Carlos Garcia Campos]

(In reply to comment #4)
> Comment on attachment 246647 [details]
> Patch
> 
> View in context:
> https://bugs.webkit.org/attachment.cgi?id=246647&action=review
> 
> >>> Source/WebKit2/WebProcess/InjectedBundle/API/gtk/WebKitWebPage.cpp:123
> >>> +    webFrameMap().remove(webFrame);
> >> 
> >> This should probably be called webkitFrameDestroyed or webkitFrameDestroyedCallback, I think. It doesn't destroy the frame, it just responds to the destruction.
> > 
> > It's confusing, it responds to a WebCore::Frame destruction to destroy the WebKitFrame object. But I agree it's confusing, webFrameDestroyed() could work better.
> 
> webFrameDestroyed sounds like a good name as well.

Ok, that's what I proposed, so any other concern?

[Martin Robinson]

Comment on attachment 246647 [details]
Patch

No, this looks okay.

[Carlos Garcia Campos]

(In reply to comment #6)
> Comment on attachment 246647 [details]
> Patch
> 
> No, this looks okay.

Thank you!

[Carlos Garcia Campos]

Committed r180211: <http://trac.webkit.org/changeset/180211>