Re: 3.4 PATCH: Fix IRIX 6 bootstrap failure

["Kaveh R. Ghazi" <ghazi at caip dot rutgers dot edu>]

 > From: Rainer Orth <ro@TechFak.Uni-Bielefeld.DE>
 > 
 > > (Still dying in cp/decl2.c though.)
 > 
 > Indeed: this could be worked around by backing out revision 1.267 of
 > fold-const.c (which caused bootstrap failures on all 64-bit platforms).
 > The patch has been reverted in the meantime.
 > 	Rainer

Ok I'm past that, my irix6.5 box is back to bootstrapville.  However
I'm getting a bootstrap failure in stage2 building libgcc2 on irix6.2:

 > libgcc2.c: In function `__floatdidf':
 > libgcc2.c:1251: internal compiler error: Bus error

It's dying in fold:

 > GNU C version 3.4 20030630 (experimental) (mips-sgi-irix6.2)
 >         compiled by GNU C version 3.4 20030630 (experimental).
 > GGC heuristics: --param ggc-min-expand=63 --param
 >         ggc-min-heapsize=62878
 > 
 > Program received signal SIGSEGV, Segmentation fault.
 > 0x10320044 in fold (expr=0x1041e9b0) at
 > ../../egcc-CVS20030629/gcc/fold-const.c:5218
 > 5218          if (TREE_CODE (TREE_OPERAND (t, 0)) == NOP_EXPR
 > (gdb) bt
 > #0  0x10320044 in fold (expr=0x1041e9b0) at ../../egcc-CVS20030629/gcc/fold-const.c:5218
 > #1  0x1006eccc in build_binary_op (code=PLUS_EXPR, orig_op0=0x4172380, orig_op1=0x41755c0, convert_p=76)
 >     at ../../egcc-CVS20030629/gcc/c-typeck.c:2566
 > #2  0x10074b90 in build_modify_expr (lhs=0x4172380, modifycode=PLUS_EXPR, rhs=0x41755c0)
 >     at ../../egcc-CVS20030629/gcc/c-typeck.c:3767
 > #3  0x100406d0 in yyparse () at c-parse.y:601
 > #4  0x10099e04 in c_common_parse_file (set_yydebug=2147428224) at ../../egcc-CVS20030629/gcc/c-opts.c:1179
 > #5  0x100f1550 in compile_file () at ../../egcc-CVS20030629/gcc/toplev.c:2070
 > #6  0x100f8690 in do_compile () at ../../egcc-CVS20030629/gcc/toplev.c:4960
 > #7  0x100f8760 in toplev_main (argc=18, argv=0x7fff2ee4) at ../../egcc-CVS20030629/gcc/toplev.c:4989
 > #8  0x100c6134 in main (argc=2147428224, argv=0xc8) at ../../egcc-CVS20030629/gcc/main.c:37


Since stage1 cc1 compiles libgcc2 fine, I thought it might be a
miscompilation somewhere in stage2 cc1.  So I substituted
stage1/fold-const.o and relinked cc1.  By doing that the compilation
succeeds.  It's now into stage3, (but will probably die again in the
same place.)

I don't know when this happened, my last successful irix6.2 bootstrap
on mainline was a few weeks ago:
http://gcc.gnu.org/ml/gcc-testresults/2003-06/msg00484.html

I was wondering if you or anyone else with irix6.2 has seen this also
and/or has any suggestions.

		Thanks,
		--Kaveh
--
Kaveh R. Ghazi			ghazi@caip.rutgers.edu