[Bug c/16452] [3.5 regression] Register name lost when compiling with -O2

[Richard Earnshaw]

On Mon, 2004-07-12 at 14:58, matz at suse dot de wrote:
> ------- Additional Comments From matz at suse dot de  2004-07-12 13:58 -------
> Will this catch everything?  I.e. are there other "attributes" of a 
> declaration which are stored in the DECL which also could be removed now 
> but needed later? 

I don't know.  I'm not even 100% certain this is the correct fix.  It
certainly seems wrong that we should clobber the DECL_RTL in this case,
but the front end isn't really my area of expertise.

R.