Re: [PATCH] gdb: New frame_cache_cleared observer.

[Doug Evans <dje@google.com>]

On Thu, Mar 26, 2015 at 5:50 AM, Andrew Burgess
<andrew.burgess@embecosm.com> wrote:
> Doug, Pedro,
>
> * Pedro Alves <palves@redhat.com> [2015-03-26 09:24:27 +0000]:
>
>> On 03/25/2015 11:18 PM, Doug Evans wrote:
>> > On Wed, Mar 25, 2015 at 10:11 AM, Andrew Burgess
>> > <andrew.burgess@embecosm.com> wrote:
>> >> This adds a new observer for the frame cache cleared event.
>> >>
>> >> While working on a new gdb port I found that I wanted to cache machine
>> >> state that was gathered as part of the register read process.  The
>> >> most appropriate time to discard this cached information is when the
>> >> frame cache is flushed.
>> >>
>> >> However, as I don't have an actual use for this observer that I can
>> >> post upstream (yet) I don't know if this will be acceptable, but given
>> >> it's a fairly small change I thought I'd try.
>
>> Right.  We delete dead code all the time.  So it's better to wait until
>> is has a use, because otherwise someone could well end up stumbling on it,
>> noticing it has no uses and decides to send a patch that garbage
>> collects it.
>
> Thanks for looking at my patch, and I understand why you've rejected
> it for now.

It's easy enough to prevent people errantly spending cycles submitting
a patch to delete such code. I think the larger worry is how long will the
code stay in the yet-to-be-used state, and how does one manage
things as the quantity of such code grows. IOW, can we manage
opening the gates without it turning into a flood? It's just easier to
keep the gate shut.

> I do have one followup: as far as I can tell the observers
> register_changed, inferior_call_pre, and inferior_call_post are only
> used by the python bindings to make the events available in python.
> As far as I can tell[1] these event bindings are only used within the
> test suite.
>
> ... and a question: If I made frame_cache_cleared a python accessible
> event, and added a test would this be sufficient to keep the code
> alive?
>
> Thanks for your time,
> Andrew
>
> [1] I could easily be wrong!

Yeah, they're events exported to python and were added
because someone had a need for them from python.

Original posting is here:
https://sourceware.org/ml/gdb-patches/2013-06/msg00889.html
Last review cycle begins here:
https://sourceware.org/ml/gdb-patches/2014-10/msg00573.html

That they're only used by the testsuite is typical of most of
python support: It's there for users writing python scripts,
gdb itself doesn't use it.

If a good use-case can be made for python wanting
to know when the frame-cache is cleared then I'd
support such a patch.

OTOH, I'm wondering if a frame-cache-cleared event
is the right one for your use-case.
I'm guessing this isn't for frame unwinding,
otherwise you could just use the existing mechanism
(e.g., frame_unwind.dealloc_cache).