From mboxrd@z Thu Jan  1 00:00:00 1970
From: Jan Hubicka <jh@suse.cz>
To: Jason Merrill <jason_merrill@redhat.com>
Cc: Jan Hubicka <jh@suse.cz>, Richard Kenner <kenner@vlsi1.ultra.nyu.edu>, gcc-patches@gcc.gnu.org
Subject: Re: spec2000 regression
Date: Wed, 15 Aug 2001 06:52:00 -0000
Message-id: <20010815155130.J19872@atrey.karlin.mff.cuni.cz>
References: <10108151302.AA17923@vlsi1.ultra.nyu.edu> <20010815150251.Y19872@atrey.karlin.mff.cuni.cz> <m3hev9qwc1.fsf@prospero.cambridge.redhat.com>
X-SW-Source: 2001-08/msg00911.html

bad style or not, I believe it would be somewhat dificult to avoid getting it into
gcc. Such bugs can be dificult to find, as they waste only the performance.
Thanks Andreas for his automated tester :)
> 
> On the contrary, it is needed for the case where a variable gets a class
> different from that of its type, so recomputing it would give the wrong
> answer.  How does this work for you?
> 
> 2001-08-15  Jason Merrill  <jason_merrill@redhat.com>
> 
> 	* varasm.c (make_decl_rtl): Don't SET_DECL_RTL until we're done
> 	modifying the rtl.
> 
Yes, I've made the same patch before the one I sent to list and it solves
this particular problem.
Note that function.c contains other similar code - just take a look for
set_memory_attributes calls contains DECL_RTL in the argument.

Honza