http://gcc.gnu.org/bugzilla/show_bug.cgi?id=47471
--- Comment #3 from Jan Kratochvil <jan.kratochvil at redhat dot com>
2011-03-29 17:35:43 UTC ---
(In reply to comment #2)
The patch has regressions, even against: GNU C 4.6.1 20110329 (prerelease)
One of many:
-PASS: gdb.base/break.exp: run until quoted breakpoint
+FAIL: gdb.base/break.exp: run until quoted breakpoint
0000000000400783 <marker2>:
int marker2 (a) int a; { return (1); } /* set breakpoint 9 here */
400783: 55 push %rbp
400784: 48 89 e5 mov %rsp,%rbp
400787: 89 7d fc mov %edi,-0x4(%rbp)
40078a: b8 01 00 00 00 mov $0x1,%eax
40078f: 5d pop %rbp
400790: c3 retq
FSF GCC correctly sets the prologue line boundary:
- Special opcode 104: advance Address by 7 to 0x400783 and Line by 1 to 47
- Special opcode 103: advance Address by 7 to 0x40078a and Line by 0 to 47
- Special opcode 104: advance Address by 7 to 0x400791 and Line by 1 to 48
but the patched FSF GCC does not:
+ Special opcode 160: advance Address by 11 to 0x400783 and Line by 1 to 47
+ Special opcode 202: advance Address by 14 to 0x400791 and Line by 1 to 48
GCC could also start emitting DW_LNS_set_prologue_end (GDB cannot parse it yet,
though) instead of these line number magics. Unfortunately the prologue end is
still needed for GDB as during -O0 -g the VTA is not in effect and
<50a> DW_AT_location : 2 byte block: 91 6c (DW_OP_fbreg: -20)
is in such case invalid at the entry PC (0x400783).
