This is the mail archive of the
gdb-patches@sourceware.org
mailing list for the GDB project.
Re: [RFA 0/7] Improved linker-debugger interface
- From: Gary Benson <gbenson at redhat dot com>
- To: gdb-patches at sourceware dot org
- Date: Thu, 16 May 2013 19:53:11 +0100
- Subject: Re: [RFA 0/7] Improved linker-debugger interface
- References: <20130516144340 dot GA2105 at blade dot nx> <87fvxmzv62 dot fsf at fleche dot redhat dot com>
Tom Tromey wrote:
> >>>>> "Gary" == Gary Benson <gbenson@redhat.com> writes:
> Gary> A quick and dirty speed comparison yielded the following results:
>
> Gary> no of solibs 128 256 512 1024 2048 4096
> Gary> --------------------------------------------------------------
> Gary> old interface 0 1 4 12 47 185
> Gary> new interface 0 0 2 4 10 36
> Gary> (time in seconds)
>
> I know we discussed this before, but I don't recall the outcome.
>
> It still looks like there is a non-linearity in there. Do you know
> why?
Some of it is certainly because GDB stores solibs in linked lists.
There is a little bit of introduced quadratic behaviour in this patch
because target_so_ops->current_sos() is expected to return a freshly
allocated list of libraries to update_solib_list(), which either adds
them to its own list or frees them if it already has them. This makes
sense for the standard interface, which reads everything from the
inferior into a list and returns it. The probes interface already has
a list built by the time target_so_ops->current_sos() is called, but
it has to create a copy to return. I spent some time trying to get
around this in July or August last year but I hit a wall. I may have
another look at this tomorrow, as this latest patch is very different
in that area from what I was working with back then.
> It would be good to have a bug for this.
I will file one (tomorrow!)
Thanks,
Gary
--
http://gbenson.net/