Re: RFC: optimized-out pieces

[Jan Kratochvil <jan.kratochvil@redhat.com>]

Hi Tom,

On Fri, 04 Jun 2010 21:19:28 +0200, Tom Tromey wrote:
> I'd appreciate comments on this patch.  In the absence of comments I
> plan to check it in.

currently it has a lot of regressions, at least tested on Fedora 13 with
-lefence.  Most of them are for x86_64-m32 and i686.  64bit x86_64 is not
regressing much as it has few DW_OP_piece in use.

Good reproducer for x86_64-m32 was for me on current FSF GDB HEAD:
	file gdb.base/store
	b 213
	run
	print u
	print 42

set_value_address sets there value->location.address despite it is an
lval_computed value that time.  Later it crashes on next command:
==4402== Jump to the invalid address stated on the next line
==4402==    at 0x300000: ???
==4402==    by 0x5D6169: free_all_values (value.c:708)
==4402==    by 0x412EED: prepare_execute_command (top.c:339)
==4402==    by 0x412F78: execute_command (top.c:379)
==4402==    by 0x4133B4: command_loop (top.c:524)


x86_64-m32:
	gdb.base/siginfo-obj.exp
	gdb.base/store.exp
x86_64:
	gdb.base/siginfo-obj.exp
	gdb.opt/clobbered-registers-O2.exp - only expected output change
i686:
	gdb.base/siginfo-obj.exp
	gdb.base/store.exp
	gdb.dwarf2/pieces.exp


Regards,
Jan