[Edbrowse-dev] AutoCompartment other window

[Edbrowse-dev] AutoCompartment other window

[Karl Dahlke]

Well, it only took a couple hours after my cleanup push to run into
a reproducible segfault.
I tracked it down to an AutoCompartment call.
Remember I said they were all gone; there is one that we actually need in html.cpp,
just before we destroy the object associated with the tag.
It's part of freeWindow().
Naturally I used the same old line,
but that sets the compartment of the current window;
we are not always freeing the current window.
q5 will quit session 5, for example.
So this is the only AutoCompartment that is not based on cw.
I fixed this and the segfault went away.

A question for you in jsdom.cpp function domLink().
This line

        JS::RootedObject owner_root(cw->jss->jcx, owner);

Then you use owner_root instead of owner thereafter.
Is this necessary?
owner is safely rooted on the heap and isn't going to go anywhere.
We could just use owner throughout and it would be much clearer.

Karl Dahlke

Re: [Edbrowse-dev] AutoCompartment other window

[Adam Thompson]

On Tue, Feb 11, 2014 at 06:55:22AM -0500, Karl Dahlke wrote:
> A question for you in jsdom.cpp function domLink().
> This line
> 
>         JS::RootedObject owner_root(cw->jss->jcx, owner);
> 
> Then you use owner_root instead of owner thereafter.
> Is this necessary?
> owner is safely rooted on the heap and isn't going to go anywhere.
> We could just use owner throughout and it would be much clearer.

That's from when the html parser wasn't js-aware,
I've just pushed a change rectifying this (removing owner_root).

Cheers,
Adam.