Bug 111375 – [v8] Binding integrity non-functional on windows

• Component: WebCore Misc.
• Status: RESOLVED WONTFIX

[Thomas Sepez]

Tracking bug to get V8 Binding integrity changes landed for windows.

[Thomas Sepez]

Created attachment 191348 [details]
WIP.

[Thomas Sepez]

Created attachment 191360 [details]
Patch.

[Adam Barth]

Comment on attachment 191360 [details]
Patch.

View in context: https://bugs.webkit.org/attachment.cgi?id=191360&action=review

> Source/WebCore/bindings/scripts/CodeGeneratorV8.pm:3722
> +    return "ActiveDOMObject\@1\@" if ($typename eq "AbstractWorker");
> +    return "ActiveDOMObject\@1\@" if ($typename eq "AudioContext");

We don't want to special-case all these interfaces in the code generator.

[Kentaro Hara]

Comment on attachment 191360 [details]
Patch.

View in context: https://bugs.webkit.org/attachment.cgi?id=191360&action=review

> Source/WebCore/bindings/scripts/CodeGeneratorV8.pm:3717
> +sub GetWinMangledExtensionForType

This looks really fragile. Wouldn't there be any alternative idea to avoid hard-coding?

[Kentaro Hara]

Comment on attachment 191360 [details]
Patch.

oops, my comment has overwritten abarth's r-.

[Thomas Sepez]

(In reply to comment #3)
> (From update of attachment 191360 [details])
> View in context: https://bugs.webkit.org/attachment.cgi?id=191360&action=review
> 
> > Source/WebCore/bindings/scripts/CodeGeneratorV8.pm:3722
> > +    return "ActiveDOMObject\@1\@" if ($typename eq "AbstractWorker");
> > +    return "ActiveDOMObject\@1\@" if ($typename eq "AudioContext");
> 
> We don't want to special-case all these interfaces in the code generator.
Ok, so would an .in file be acceptable?

[Thomas Sepez]

(In reply to comment #4)
> (From update of attachment 191360 [details])
> View in context: https://bugs.webkit.org/attachment.cgi?id=191360&action=review
> 
> > Source/WebCore/bindings/scripts/CodeGeneratorV8.pm:3717
> > +sub GetWinMangledExtensionForType
> 
> This looks really fragile. Wouldn't there be any alternative idea to avoid hard-coding?

I've toyed with the idea of simply creating an instance of the class the first time one is wrapped and extracting the pointer from it directly via its first member rather than playing linker tricks.

The downside with that is bloat and performance, but not a big deal when we are doing it at wrap time, rather than toNative() time.

The obstacle is that each of these would need a no-arg constructor to make such a object from which to extract the pointer, which currently isn't the case.

[Adam Barth]

Comment on attachment 191360 [details]
Patch.

View in context: https://bugs.webkit.org/attachment.cgi?id=191360&action=review

>>> Source/WebCore/bindings/scripts/CodeGeneratorV8.pm:3722
>>> +    return "ActiveDOMObject\@1\@" if ($typename eq "AudioContext");
>> 
>> We don't want to special-case all these interfaces in the code generator.
> 
> Ok, so would an .in file be acceptable?

What is the relationship between the two strings?

What I'm worried about here is that folks who add or change these interfaces or objects will need to be able to keep this information up-to-date or else they'll get some kind of strange compile error on Windows.  We need to set them up for success by making this as easy and automatic as possible.

[Thomas Sepez]

> > Ok, so would an .in file be acceptable?
> 
> What is the relationship between the two strings?
> 
Its truly icky and a burden which may put this out of reach.  It the name of the first superclass that declares a new virtual function, where first is the leftmost class in transitive expansion of the class C : public A, public B ... inheritance list.  If that makes any sense at all.

[Adam Barth]

I see.  It's the class name for the first vtable in the object.

[Anders Carlsson]

V8 is gone from WebKit.