[tree-ssa] kill handling of non-gimple address

[Jan Hubicka]

> In message <xypr7zuic4t.fsf@dhcp64-198.boston.redhat.com>, Jason Merrill writes
> :
>  >On Tue, 25 Nov 2003 23:56:45 +0100, Jan Hubicka <hubicka@ucw.cz> wrote:
>  >
>  >> Hi,
>  >> after Richard's change, this should be no longer needed.
>  >
>  >> 	* fold-const.c (nondestructive_fold_binary_to_constant): Avoid unnecesa
>  >ry
>  >> 	handling of non-gimple address expressions.
>  >
>  >Is this function only called for gimple code?
> I don't think it's safe to make that assumption.
OK, ignore the patch then.

In general, I think such uncertainess is that is the problem of whole
fold code.  It seems to me that we should make clean split in between
functions that are accpeting gimple and producing gimple, accpting
generic and producing generic and possibly accepting same kind of IL as
producing to allow some sharing in between implementations of folders.

If we had some longer term plan how to deal with this, I would happily
help working on it, as fold bugs are stopping me in at least two other
plans I am working on and fixing them is not at all easy in current
code.  So please let me know if there is something decided about this.

Honza
> 
> jeff