Re: [PATCH v2] [gdb] Support frames inlined into the outer frame

[Luis Machado <luis.machado@linaro.org>]

On 6/4/20 4:23 PM, Simon Marchi wrote:
> On 2020-06-04 12:11 p.m., Simon Marchi wrote:
>> On 2020-03-31 3:18 p.m., Scott Linder wrote:
>>> @@ -763,6 +760,24 @@ frame_id_eq (struct frame_id l, struct frame_id r)
>>>     return eq;
>>>   }
>>>   
>>> +int
>>> +outer_frame_id_p (struct frame_id l)
>>> +{
>>> +  int p;
>>> +
>>> +  /* The artificial_depth can vary so we ignore it when checking if this is
>>> +     an outer_frame_id.  */
>>> +  l.artificial_depth = 0;
>>> +  p = memcmp (&l, &outer_frame_id, sizeof (outer_frame_id));
>>
>> This should be `memcmp (...) == 0`.  Currently, the function returns true when
>> the frame is not an outer frame id, which is the opposite of what it is supposed
>> to do.
>>
>> With this, the test gdb.base/break.exp on AArch64 runs fine.  I will launch a full
>> test run to see if there are any other problems.
>>
>> You can make the new function return "bool" instead of "int", and use true/false instead
>> of zero/non-zero (both in the code and comments).
>>
>> Simon
> 
> With this change, the full test run on AArch64 came out clean.  I'll try a try
> job on the buildbot, but I haven't had much success with it recently.
> 
> Simon
> 

I don't see the same, even with the fixup of memcmp. Though 
gdb.base/break.exp has full passes with the change, the following tests 
internal error with the patch...

gdb.mi/mi-nonstop.exp
gdb.threads/clone-thread_db.exp
gdb.threads/current-lwp-dead.exp
gdb.threads/hand-call-in-threads.exp
gdb.threads/linux-dp.exp
gdb.threads/local-watch-wrong-thread.exp
gdb.threads/queue-signal.exp
gdb.threads/schedlock.exp
gdb.threads/thread_check.exp
gdb.threads/tls.exp

#1  0x0000ffffb7fa1088 in start_thread () from 
/lib/aarch64-linux-gnu/libpthread.so.0
../../../repos/binutils-gdb/gdb/frame.c:551: internal-error: void 
compute_frame_id(frame_info*): Assertion `frame_id_p 
(fi->this_id.value)' failed.

Scott, could you please send a v3 so I can make sure I tested the right 
version? I was initially slightly confused with what version Simon was 
talking about since I had already tested v2.