Ian Lance Taylor - Re: PATCH: fix ICE while producing libgcc/_muldi3.o (original) (raw)

This is the mail archive of the gcc-patches@gcc.gnu.orgmailing list for the GCC project.

• From: Ian Lance Taylor
• To: Sandra Loosemore
• Cc: GCC Patches
• Date: 08 Feb 2007 16:07:41 -0800
• Subject: Re: PATCH: fix ICE while producing libgcc/_muldi3.o
• References: 45BFEAF1.7010300@codesourcery.com <m3hctwk6ab.fsf@localhost.localdomain> <45CBBAD2.2070500@codesourcery.com>

Sandra Loosemore sandra@codesourcery.com writes:

Ian Lance Taylor wrote:

Sandra Loosemore sandra@codesourcery.com writes:

2007-01-30 Sandra Loosemore sandra@codesourcery.com

* fold-const.c (fold_checksum_tree): Avoid referencing nonexistent
TREE_TYPE field of GIMPLE_MODIFY_STMT objects.

It doesn't seem right to just return. I think we should checksum the fields of the statement just as we do for any other tree. We just have to avoid accessing TREE_TYPE.

It's not just accessing TREE_TYPE; the next thing it does is look at TREE_CHAIN, which doesn't make sense for tcc_gimple_stmt, either. And there is no case for tcc_gimple_stmt in the switch that follows that, either. It could be made to do something useful, but the patch I posted is sufficient to get rid of the ICE when building the gcc libraries with checking enabled. :-)

I agree that your patch gets rid of the ICE. But GIMPLE_MODIFY_STMT does have operands, so it seems to me that fold_checksum_tree should get their checksum. At least, I don't see why it shouldn't.

Ian

• References:
  • Re: PATCH: fix ICE while producing libgcc/_muldi3.o
    * From: Ian Lance Taylor
  • Re: PATCH: fix ICE while producing libgcc/_muldi3.o
    * From: Sandra Loosemore