[Libreoffice] [GSoC] Re: KCachegrind

[Matúš Kukan]

Hi there,

On 7 June 2011 22:57, Michael Meeks <michael.meeks at novell.com> wrote:
>        So - then I wonder if we can catch the problem lower down - inside the
> valgrind 'cachegrind' skin. Now - I know what you're thinking: 'valgrind
> is a complicated nightmare' :-) but you're wrong. Valgrind skins are
> single-threaded, and amazingly simple (in my experience of skin hacking
> - mainly for iogrind admittedly).
>
Did you mean callgrind?

I don't like the code.
It's not a really big problem but I have problems with reading it
because of things like
'if (!CLG_(is_zero_cost)( CLG_(sets).full, currCost->cost ))' and others.

>        I -hope- that (since all the information is in-core at once in the
> valgrind process), that we can get the account right there - last I
> looked ~all of cachegrind was only ~1500 lines of code - hopefully an
> easy read. Presumably adding some --hide-object='libc.so.6,libuno.so'
> etc. is quite do-able there too ?
>
>        What do you think ? [ and sorry this is such a stretching
> experience ;-].
>
I think we should let it be.
It's a shame because it could be helpful but I don't think I am
capable of doing that (in a reasonable amount of time).
I think callgrind is built to produce output of special format and it
could be really hard change the way it's working.
Although I'm not really sure by this. I haven't yet found out how it is working.

At least it is possible to hide self cost of some libraries. But it's
almost the same like you were ignoring them in object view.

regards,
Matus