This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: Advise needed on varasm.c constant recording bug
- From: David Edelsohn <dje at watson dot ibm dot com>
- To: kenner at vlsi1 dot ultra dot nyu dot edu (Richard Kenner)
- Cc: gcc at gcc dot gnu dot org
- Date: Mon, 31 Dec 2001 18:08:23 -0500
- Subject: Re: Advise needed on varasm.c constant recording bug
>>>>> Richard Kenner writes:
Richard> When we compare tree constants for equality in varasm.c, we hash the
Richard> address of the type and record that address.
Richard> We *could* fix it by teaching GC to look inside the constant descriptors,
Richard> but I think that's a mess. Indeed, we don't *need* those constant
Richard> descriptors anymore. We can simply keep a hash table of the tree nodes for
Richard> the constants themselves and let GC know about it. Indeed, this was on
Richard> my list of things to do for 3.1, but it didn't get done by December 15.
Shouldn't we move away from hashing internal GCC addresses or is
that what you are suggesting? Hashing internal addresses repeatedly
causes problems when GCC's memory layout differs on different hosts, or
even between bootstrap stages causing comparison failures.
David