Why do we have two ways of finding sniffers?

Why do we have two ways of finding sniffers?

[Jim Blandy]

Looking at frame-unwind.c and frame-unwind.h, it seems like there are
two mostly equivalent ways to register frame unwinding methods.  Why
do we have two ways, instead of just one?  The comments in
frame-unwind.h don't really explain why it's helpful.

The only function that scans the list of unwinders is
frame_unwind_by_frame: it takes a NEXT frame, and returns a 'struct
frame_unwind *' if it can find one.

It works by walking the NEXT frame's gdbarch's list of frame
unwinders.  Each entry in the list is one of two possible kinds:

- If the entry points to a frame_unwind_sniffer_ftype function, that
  takes the next frame's frame_info; if it likes it, then it returns a
  'struct frame_unwind *' to use for this frame.

- If the entry points to a 'struct frame_unwind' object, that contains
  its own sniffer, which takes a pointer to the 'struct frame_unwind',
  the next frame's frame_info, and a prologue cache, and just returns
  a boolean indicating whether this is an appropriate frame_unwind for
  it.

I don't understand why we need both alternatives.  Shouldn't it be
sufficient to simply have each entry in the list point to a function
that expects the next frame's frame_info and a prologue cache, and
returns a 'struct frame_unwind *' or zero?

Re: Why do we have two ways of finding sniffers?

[Jim Blandy]

Jim Blandy <jimb@redhat.com> writes:
> I don't understand why we need both alternatives.  Shouldn't it be
> sufficient to simply have each entry in the list point to a function
> that expects the next frame's frame_info and a prologue cache, and
> returns a 'struct frame_unwind *' or zero?

I guess we'd also need a void * passed through to the function, to
support things like tramp_frame_prepend_unwinder.

Re: Why do we have two ways of finding sniffers?

[Mark Kettenis]

> From: Jim Blandy <jimb@redhat.com>
> Date: Wed, 19 Oct 2005 17:16:46 -0700
> 
> I don't understand why we need both alternatives.  Shouldn't it be
> sufficient to simply have each entry in the list point to a function
> that expects the next frame's frame_info and a prologue cache, and
> returns a 'struct frame_unwind *' or zero?

I don't think there is a *technical* reason why we need both
alternatives.  It's more a matter that we had (and still have to some
extent) a pretty long list of basically unmainted targets.  So
converting frame_unwind_append_sniffer() into
frame_unwind_append_unwinder() is difficult to accomplish.  But you
should really answer Andrew about this since he wrote that code.

Mark

Re: Why do we have two ways of finding sniffers?

[Jim Blandy]

Mark Kettenis <mark.kettenis@xs4all.nl> writes:
>> From: Jim Blandy <jimb@redhat.com>
>> Date: Wed, 19 Oct 2005 17:16:46 -0700
>> 
>> I don't understand why we need both alternatives.  Shouldn't it be
>> sufficient to simply have each entry in the list point to a function
>> that expects the next frame's frame_info and a prologue cache, and
>> returns a 'struct frame_unwind *' or zero?
>
> I don't think there is a *technical* reason why we need both
> alternatives.  It's more a matter that we had (and still have to some
> extent) a pretty long list of basically unmainted targets.  So
> converting frame_unwind_append_sniffer() into
> frame_unwind_append_unwinder() is difficult to accomplish.  But you
> should really answer Andrew about this since he wrote that code.

Okay.  I figured all the frame-unwind.c code was relatively new, so
diversity was probably deliberate.