gdb.base/pending.exp failures

[Ulrich Weigand]

FAIL: gdb.base/pending.exp: run to resolved breakpoint 1 (without symbols)
FAIL: gdb.base/pending.exp: continue to resolved breakpoint 1
FAIL: gdb.base/pending.exp: continue to resolved breakpoint 3 (the program exited)

Further analysis required.

[Yao Qi (yao-codesourcery)]

The first failure is caused by breakpoint setting by mistake. In test case, breakpoint is set on function 'pendfunc1', and expected to set/hit on line 22, however, breakpoint is hit on line 23.

Reproduce it in another way,
$ ./gdb /home/yao/Work/LP615989/pendshr.c.o
(gdb) b pendfunc1
During symbol reading, DW_AT_name missing from DW_TAG_base_type.
Breakpoint 1 at 0x6: file /home/yao/maverick/home/yao/cvs/src/gdb/testsuite/gdb.base/pendshr.c, line 23.

You can see that breakpoint is set at address 0x6, which is associated with line 23. We can prove this by dumping .debug_line, as follows,

  Special opcode 49: advance Address by 6 to 0x6 and Line by 2 to 23 // <---- here,
  Special opcode 45: advance Address by 6 to 0xc and Line by -2 to 21
  Special opcode 20: advance Address by 2 to 0xe and Line by 1 to 22
   Special opcode 62: advance Address by 8 to 0x16 and Line by 1 to 23

Disasm pendfunc1 shows that, ldr insn (on 0x6) loads a const.

00000000 <pendfunc1>:
   0: b580 push {r7, lr}
   2: b084 sub sp, #16
   4: af00 add r7, sp, #0
   6: 4b09 ldr r3, [pc, #36] ; (2c <pendfunc1+0x2c>)
   8: a200 add r2, pc, #0 ; (adr r2, c <pendfunc1+0xc>)
   a: 4413 add r3, r2

[Yao Qi (yao-codesourcery)]

pendfunc1:
        @ args = 0, pretend = 0, frame = 16
        @ frame_needed = 1, uses_anonymous_args = 0
        push {r7, lr} @ 27 *push_multi [length = 4]
        sub sp, sp, #16 @ 28 *arm_addsi3/5 [length = 4]
        add r7, sp, #0 @ 30 *arm_addsi3/1 [length = 4]
        ldr r3, .L3 @ 9 pic_load_addr_thumb2 [length = 4]

Insn ldr on address 0x6 is generated from rtl 'pic_load_addr_thumb2', so it should *not* be associated with line 23,

So this entry
      'Special opcode 49: advance Address by 6 to 0x6 and Line by 2 to 23'
in .debug_line is wrong. AFAICS, the first failure is caused by a compiler bug.

[Yao Qi (yao-codesourcery)]

Run this test with gcc-linaro-4.5-2010.08-1, failures go away!!

$ make check RUNTESTFLAGS="pending.exp CC_FOR_TARGET='/home/yao/gcc-linaro-4.5-2010.08-1-armv7l-maverick-cbuild2-pavo2/bin/armv7l-unknown-linux-gnueabi-gcc'"
Running /home/yao/maverick/home/yao/cvs/src/gdb/testsuite/gdb.base/pending.exp ...

  === gdb Summary ===

# of expected passes 29

This case still fails with gcc-4.5 (Ubuntu 4.5.1-1ubuntu1) 4.5.1, gcc-4.4 (Ubuntu/Linaro 4.4.4-8ubuntu1) 4.4.5 20100728, and gcc-linaro-4.4-2010.08-0.

[Yao Qi (yao-codesourcery)]

Open bug LP:617384 for linaro gcc 4.4.

[Ulrich Weigand (uweigand)]

The underlying compiler bug is now fixed; closing the GDB bug as Invalid.