RE: [PATCH v11 10/10] gdb: Filter trampoline frames in backtrace when using Python frame-filters.

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

Thanks Tom, for the feedback.

Please see the inline reply below:

>> +  if (skip_trampoline_functions)
>> +    {
>> +      for (int i = 0; (SAFE_TRAMPOLINE_CHAIN (i, frame)
>> +		       && in_trampoline_frame (frame)); ++i)
>> +	{
>> +	  frame = get_prev_frame (frame);
>> +	  if (frame ==  nullptr)
>> +	    return nullptr;
>> +	}
>> +    }

Tom> This seems suspicious to me.
Tom> For one thing, suppose you have A -> Trampoline -> B.  Then in Python you fetch the frame B.  What does B.newer() return?
Tom> I think with this code it will just return B again.

Abdul> You are right that newer () returns the same frame again, and it is a real problem, though for the caller: with A calling the trampoline and the trampoline calling B, it is A.newer () that returns A again, while B.newer () is not affected.  Since every gdb.Frame of a trampoline is turned into the next outer frame, the frame object that A.newer () returns for the trampoline behaves as A, so walking newer () never gets back to B.

For example, with ifx and gdb.fortran/func-trampoline.f90, the call stack is:
  func_trampoline -> first_.t95p.t96p -> first -> second_.t74p.t75p -> second

where first_.t95p.t96p and second_.t74p.t75p are the trampolines. Stopped in "second":

  (gdb) python f2 = gdb.selected_frame().older()
  (gdb) python print(f2.level(), f2.name())
  2 first
  (gdb) python print(f2.newer().level(), f2.newer().name())
  2 first

Here f2.newer () should return "second" (level 0), or the trampoline second_.t74p.t75p at level 1, but not "first" again.

Tom> For DAP, and frame filters, I tend to think it would be better for Python to simply see the frames as they are, but then add a new attribute -- on the frame, but also maybe on the symbol -- indicating if it's been marked as a trampoline.
Tom> Then maybe _FrameDecoratorBase.elided could return True if the frame is a non-"live" trampoline frame.

Abdul> Agreed.  For v12 I will drop the change to frame_object_to_frame_info, so that Python sees the frames as they are, and add new attributes:

- gdb.Symbol.is_trampoline, an attribute like Symbol.is_function;
- gdb.Frame.is_trampoline (), a method like the other gdb.Frame  accessors, true if the frame's function is marked as a trampoline;

For the frame filters, elided () returns the frames elided under a frame, so I would do it the other way round: the frame that a trampoline leads to returns the trampoline frames as its elided frames, except for the live frame, which is never elided.  Frame filters and DAP then handle them like other elided frames.  This depends on what we decide for the backtrace command in 04/10, and I will keep the Python side consistent with that.

Thanks & Best Regards,
Abdul Basit

-----Original Message-----
From: Tom Tromey <tom@tromey.com> 
Sent: Wednesday, October 21, 2026 11:03 PM
To: Ijaz, Abdul B <abdul.b.ijaz@intel.com>
Cc: gdb-patches@sourceware.org; JiniSusan.George@amd.com; tom@tromey.com; eliz@gnu.org
Subject: Re: [PATCH v11 10/10] gdb: Filter trampoline frames in backtrace when using Python frame-filters.

>>>>> Abdul Basit Ijaz <abdul.b.ijaz@intel.com> writes:

> In case of the Python frame-filters, this change filters trampoline 
> functions when the option 'skip-trampoline-functions' is set to 'on'.  
> GDB recently added the option to hide frames that are marked by the 
> compiler with "DW_AT_trampoline", see commit 'Skip trampoline frames 
> for the backtrace command'.  This implements the same functionality 
> when python frame filters are in use.

> diff --git a/gdb/python/py-frame.c b/gdb/python/py-frame.c index 
> afa89f0d112..098b37230e1 100644
> --- a/gdb/python/py-frame.c
> +++ b/gdb/python/py-frame.c
> @@ -59,6 +59,17 @@ frame_object_to_frame_info (PyObject *obj)
>    if (frame == NULL)
>      return NULL;
 
> +  if (skip_trampoline_functions)
> +    {
> +      for (int i = 0; (SAFE_TRAMPOLINE_CHAIN (i, frame)
> +		       && in_trampoline_frame (frame)); ++i)
> +	{
> +	  frame = get_prev_frame (frame);
> +	  if (frame ==  nullptr)
> +	    return nullptr;
> +	}
> +    }

This seems suspicious to me.

For one thing, suppose you have A -> Trampoline -> B.  Then in Python you fetch the frame B.  What does B.newer() return?

I think with this code it will just return B again.

For DAP, and frame filters, I tend to think it would be better for Python to simply see the frames as they are, but then add a new attribute -- on the frame, but also maybe on the symbol -- indicating if it's been marked as a trampoline.

Then maybe _FrameDecoratorBase.elided could return True if the frame is a non-"live" trampoline frame.

Tom
________________________________________
Intel Deutschland GmbH 

Registered Address: Dornacher Strasse 1, 85622 Feldkirchen, Germany 

Tel: +49 (89) 99143-0 

www.intel.de 

Managing Directors: Candice Moore, Jeffrey Schneiderman, Ramachandran Sitaraman

Chairperson of the Supervisory Board: Sonja Pierer

Registered Seat: Munich Commercial Register B: Amtsgericht Munich HRB 186928

This e-mail and any attachments may contain confidential material for
the sole use of the intended recipient(s). Any review or distribution
by others is strictly prohibited. If you are not the intended
recipient, please contact the sender and delete all copies.