RE: [PATCH v4 4/4] gdb: Skip trampoline frames in the stack for printing or finish command.

["Ijaz, Abdul B" <abdul.b.ijaz@intel.com>]

Hi Tom,

Thanks for the feedback. Added replies below:

Tom> If I am stopped in a frame and go "up", will I be in the trampoline frame anyway?  And will "bt" then not print the selected frame?  But "frame" will?

Yes in current implementation bt will not print those trampoline frames and user may see from the frame number in bt since in output  frame numbers will be missing/elided for those trampoline calls. So "up" command will still take it to trampoline call. Do we need to handle "up" call ? . One sample run from program where first function call the second function and DW_at_trampoline exist for both of them.

22        second = x * y ! second-breakpt
(gdb) bt
#0  second (x=20, y=9) at db.fortran/func-trampoline.f90:22
#2  0x0000000000405203 in first (num1=16, num2=3) at gdb.fortran/func-trampoline.f90:29
#4  0x0000000000405254 in func_trampoline () at gdb.fortran/func-trampoline.f90:35
(gdb) up
#1  0x0000000000405229 in second_.t74p.t75p () at gdb.fortran/func-trampoline.f90:30
(gdb) frame 1
#1  0x0000000000405229 in second_.t74p.t75p () at gdb.fortran/func-trampoline.f90:30
30      end function

Tom > Also this approach ignores frame filters -- if one is in use, then I think trampoline frames will show up again.  You can try this by writing a dummy frame filter that just returns all frames as-is.

Yes in case of frame filter these trampolines frames are shown up.  Do we need to cover it in this change since I think we need to handle in such case in frame filter script or something need to be handled here for this ?

Thanks & Best Regards
Abdul Basit

-----Original Message-----
From: Tom Tromey <tom@tromey.com> 
Sent: Wednesday, August 2, 2023 10:42 PM
To: Abdul Basit Ijaz via Gdb-patches <gdb-patches@sourceware.org>
Cc: Ijaz, Abdul B <abdul.b.ijaz@intel.com>; JiniSusan.George@amd.com; tom@tromey.com; eliz@gnu.org; blarsen@redhat.com
Subject: Re: [PATCH v4 4/4] gdb: Skip trampoline frames in the stack for printing or finish command.

>>>>> "Abdul" == Abdul Basit Ijaz via Gdb-patches <gdb-patches@sourceware.org> writes:

Abdul> Before the change, GDB prints the frames indicated by the 
Abdul> compiler with DIE "DW_AT_trampoline" in the backtrace and finish 
Abdul> command, but for better user experience, all such frames can be 
Abdul> hidden from the user.  So, after this change, now such frames are 
Abdul> not printed any more in the backtrace command and also the 'finish' command skips the trampoline calls.

Abdul> So far, this DIE is added to DWARF only by the IFX compiler, so 
Abdul> gdb.fortran/mixed-lang-stack test used to fail for this compiler 
Abdul> because of these extra trampoline frames in the backtrace.  After 
Abdul> the commit, those trampoline frames are filtered so test is 
Abdul> updated accordingly to handle the frame level of the filtered frames.

Thanks for the patch.

I'm not totally sure how I feel about the stack trace part of this one.

On the one hand, normally I think it makes sense to elide these frames.
They aren't very interesting.

On the other hand, we have frame filters for frame elision, but then this works in a totally different way.

If I am stopped in a frame and go "up", will I be in the trampoline frame anyway?  And will "bt" then not print the selected frame?  But "frame" will?

Abdul> +  if (skip_trampoline_functions)
Abdul> +    {
Abdul> +      for (int i = 0; i < MAX_TRAMPOLINE_CHAIN_SIZE
Abdul> +		      && (frame != nullptr)
Abdul> +		      && in_trampoline_frame (frame); ++i)
Abdul> +	frame = get_prev_frame (frame);

Formatting looks weird here.

Abdul> +++ b/gdb/stack.c
Abdul> @@ -2054,6 +2054,14 @@ backtrace_command_1 (const frame_print_options &fp_opts,
Abdul>  	{
Abdul>  	  QUIT;
 
Abdul> +	  if (in_trampoline_frame (fi))
Abdul> +	    {
Abdul> +	      /* Trampoline frames are not printed so they are not counted in
Abdul> +		 the backtrace limit.  */
Abdul> +	      count++;
Abdul> +	      continue;

I guess the frame numbers will skip as well, because the frame level is a property of the frame, not of the loop displaying frames.

Also this approach ignores frame filters -- if one is in use, then I think trampoline frames will show up again.  You can try this by writing a dummy frame filter that just returns all frames as-is.

Tom
Intel Deutschland GmbH
Registered Address: Am Campeon 10, 85579 Neubiberg, Germany
Tel: +49 89 99 8853-0, www.intel.de <http://www.intel.de>
Managing Directors: Christin Eisenschmid, Sharon Heck, Tiffany Doon Silva  
Chairperson of the Supervisory Board: Nicole Lau
Registered Office: Munich
Commercial Register: Amtsgericht Muenchen HRB 186928