Re: duplicated user destructor for lookahead

[Joel E. Denny]

On Thu, 22 Sep 2005, Paul Eggert wrote:

> "Joel E. Denny" <address@hidden> writes:
>
> > In any case, I believe the following patch is cleaner than my last one.
> > Its approach is simply to synchronize the shift for all stacks.  Embedded
> > comments should explain the rest.
>
> That looks good to me.

Thanks.  If no one objects, then it should be ready for install.  It's a 
separate bug and should be a separate patch from the one discussed below.

> > that
> > `If stack is well-formed' comment has me scared to do this.  Can anyone
> > explain this comment?
>
> It's meant to cover the case where the stack is only partially allocated
> (it takes more than one malloc call to create it, and perhaps a later
> malloc failed).  But this can happen only during initialization; the
> parser proper isn't invoked if the stack never got built right.

I see it now.  Thanks.

The patch looks straight-forward, but I hope to devise a test case that 
can reliably prove the bug and the patch.

Joel