Bug 70440 – [Chromium] Export missing symbols from Web*Layer

• Component: New Bugs
• Status: RESOLVED FIXED

[Antoine Labour]

Export missing symbols from Web*Layer

[Antoine Labour]

Created attachment 111673 [details]
Patch

[WebKit Review Bot]

Please wait for approval from [e-mail] before submitting because this patch contains changes to the Chromium public API.

[Darin Fisher (:fishd, Google)]

Comment on attachment 111673 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=111673&action=review

> Source/WebKit/chromium/public/WebContentLayer.h:38
> +class WEBKIT_EXPORT WebContentLayer : public WebLayer {

we don't normally export entire classes.  see other header files.  we just export the non-inline methods (ignore pure virtual methods of course).

[Antoine Labour]

(In reply to comment #3)
> (From update of attachment 111673 [details])
> View in context: https://bugs.webkit.org/attachment.cgi?id=111673&action=review
> 
> > Source/WebKit/chromium/public/WebContentLayer.h:38
> > +class WEBKIT_EXPORT WebContentLayer : public WebLayer {
> 
> we don't normally export entire classes.  see other header files.  we just export the non-inline methods (ignore pure virtual methods of course).

That was necessary, otherwise I was getting undefined symbols for the vtable - it's defined with the first virtual method definition (in this case the destructor, inside of the .so), but is needed in the constructor (inlined in the calling code).
If we're adamant on that, we could try inlining the virtual destructor, or un-inlining the constructors and moving them into the .cpp. I'll see if either one works.

[Darin Fisher (:fishd, Google)]

(In reply to comment #4)
...
> That was necessary, otherwise I was getting undefined symbols for the vtable - it's defined with the first virtual method definition (in this case the destructor, inside of the .so), but is needed in the constructor (inlined in the calling code).
> If we're adamant on that, we could try inlining the virtual destructor, or un-inlining the constructors and moving them into the .cpp. I'll see if either one works.

We haven't had to export classes before.  Something must be unconventional here that can be avoided.

[Darin Fisher (:fishd, Google)]

(In reply to comment #5)
> (In reply to comment #4)
> ...
> > That was necessary, otherwise I was getting undefined symbols for the vtable - it's defined with the first virtual method definition (in this case the destructor, inside of the .so), but is needed in the constructor (inlined in the calling code).
> > If we're adamant on that, we could try inlining the virtual destructor, or un-inlining the constructors and moving them into the .cpp. I'll see if either one works.
> 
> We haven't had to export classes before.  Something must be unconventional here that can be avoided.

perhaps you should inline the WebLayer destructor.  that would probably do the trick.  we normally do not export constructors and destructors.

[Antoine Labour]

Created attachment 111839 [details]
Patch

[Antoine Labour]

New patch does that.

[WebKit Review Bot]

Comment on attachment 111839 [details]
Patch

Clearing flags on attachment: 111839

Committed r98035: <http://trac.webkit.org/changeset/98035>

[WebKit Review Bot]

All reviewed patches have been landed.  Closing bug.