Re: [patch] for PR 31676

[Zdenek Dvorak <rakdver at atrey dot karlin dot mff dot cuni dot cz>]

Hello,

> > 2) We forget to create edges in CFG from non-local gotos to the
> >    non-local goto receivers inside the same function.
> 
> I think that this is not supposed to happen, see e.g. expand_builtin_longjmp
>   /* Search backwards and mark the jump insn as a non-local goto.
>      Note that this precludes the use of __builtin_longjmp to a
>      __builtin_setjmp target in the same function.  However, we've
>      already cautioned the user that these functions are for
>      internal exception handling use only.  */
> 
> Could you clarify why you need this for execute/built-in-setjmp.c?

OK; during debugging, I was working with simplified testcase where I
manually inlined the sub2 function; I did not realize this was not
allowed.  This piece of code is not needed, then.

Zdenek