Re: [tree-ssa] Bootstrap failure on powerpc-apple-darwin

[Jan Hubicka <hubicka at ucw dot cz>]

> The tree-ssa branch has a bootstrap failure on 
> powerpc-apple-darwin7.0.0.
> 
> The compiler goes into an infinite loop in decl_function_context with 
> the following reduced source code:
> extern void gen_movstrsi_8reg (void);
> int expand_block_move (int i)
> {
>     int bytes;
>     {
>         union {
>             void (*movstrsi) (void);
>         } gen_func;
>         gen_func.movstrsi = gen_movstrsi_8reg;
>         (*gen_func.movstrsi) ();
>     }
> 
>     return 1;
> }

Hi,
I got suck here.  It seems to be related to Richard's double linking
work, but I am not quite sure.
The problem is that we get block whose supercontext is pointing to
itself.  This happens because we re-gimplify bind_expr.  I am attaching
simple patch to make re-gimplification happen properly, but the reason
why we re-gimplify is still behind me.
It happens wen the function itself does:
  lower_stmt_body (&BIND_EXPR_BODY (stmt), data);
the first stamement of BIND_EXPR_BODY seems to be the BIND_EXPR itself
that looks very wrong, but I can't figure out how it happens.
I am also no longer able to dump the insn chains.  How this is done?

Honza
> 
> Thanks,
> Andrew Pinski