[tree-ssa] dom1 bug.

[Daniel Berlin]

>
>> Which looks right to me.
>
> It is, its just not what I was getting. huh.
>
> Andrew
>

There is something going on with dump files ATM,   You can get it to 
crash on some files with the right dump flags. But it's not the dumpers 
themselves, it seems to be writing to closed dump files.

valgrind'ing cc1 -O2 -fdump-tree-all-all on any preprocessed source i 
have around gives:

==29412== Invalid read of size 1
==29412==    at 0x402D5D5F: _IO_vfprintf_internal (in 
/lib/i686/libc-2.3.3.so)
==29412==    by 0x402DEC1E: __GI_fprintf (in /lib/i686/libc-2.3.3.so)
==29412==    by 0x814764B: dump_var_map (tree-ssa-live.c:1645)
==29412==    by 0x81048C5: remove_ssa_form (tree-ssa.c:2542)
==29412==    Address 0x4218B4EE is 70 bytes inside a block of size 364 
free'd
==29412==    at 0x40013A21: free (vg_replace_malloc.c:189)
==29412==    by 0x402EF13D: _IO_fclose@@GLIBC_2.1 (in 
/lib/i686/libc-2.3.3.so)
==29412==    by 0x8435EED: dump_end (tree-dump.c:761)
==29412==    by 0x8100150: rewrite_into_ssa (tree-ssa.c:466)
==29412==
==29412== Invalid read of size 4
==29412==    at 0x402D5D75: _IO_vfprintf_internal (in 
/lib/i686/libc-2.3.3.so)
==29412==    by 0x402DEC1E: __GI_fprintf (in /lib/i686/libc-2.3.3.so)
==29412==    by 0x814764B: dump_var_map (tree-ssa-live.c:1645)
==29412==    by 0x81048C5: remove_ssa_form (tree-ssa.c:2542)
==29412==    Address 0x4218B504 is 92 bytes inside a block of size 364 
free'd
==29412==    at 0x40013A21: free (vg_replace_malloc.c:189)
==29412==    by 0x402EF13D: _IO_fclose@@GLIBC_2.1 (in 
/lib/i686/libc-2.3.3.so)
==29412==    by 0x8435EED: dump_end (tree-dump.c:761)
==29412==    by 0x8100150: rewrite_into_ssa (tree-ssa.c:466)

etc

Maybe this problem is why you aren't seeing the right info in the dump 
file (IE it wasn't open when it was writing to it)