Bug#670586: iceweasel:[regression 3.5.16-13 > 14] JavaScript SIGSEGV

[Mike Hommey]

On Thu, May 10, 2012 at 09:21:19PM +0200, Mike Hommey wrote:
> On Wed, May 09, 2012 at 01:58:27PM +0100, Steven Chamberlain wrote:
> > On 09/05/12 08:43, Mike Hommey wrote:
> > > On Tue, May 08, 2012 at 11:51:45PM +0100, Steven Chamberlain wrote:
> > >> I've narrowed the regression down to this (somewhat secretive) patch:
> > >> squeeze-patches/Bug-732951.-r-bsmedberg-a-akeybl.patch
> > 
> > > Could you check with the attached patch?
> > 
> > Sorry, hasn't worked (had to apply it on top of the problem patch, but
> > it did not make this regression go away).
> 
> Actually, it did change the crash signature, which allowed me to go
> further and find the right fixes.
> 
> Now, as it changes public ABI, I need to be extra cautious. Fortunately,
> it is "only" changing 2 functions that returned "void" to return a
> PRBool (so, essentially, int). Now the question is, when there is call
> to a function returning void, is the compiler allowed to assume that
> the register normally used for return values is not going to be
> modified? If it is, then it's not so safe of a change.

Erf, forget it, even before considering this, there's the obvious change
in symbol mangling that prevents the change from being safe.

Mike