patches for memprof on current glibc

View: New views
3 Messages — Rating Filter:   Alert me  

patches for memprof on current glibc

by John Berthels :: Rate this Message:

Reply to Author | View Threaded | Show Only this Message

Hi folks,

I've been trying to get GNOME cvs memprof building and running on ubuntu 5.10.

There was a compilation issue with newer binutils (already fixed in
the ubuntu patchset) and also a problem caused by the fact that
backtrace() in glibc 2.3.5 calls dlopen(), which calls malloc and
hence recurses.

By sticking in a simple-minded malloc for use during startup (and
forcing initialisation of backtrace() at init time), I can get memprof
running (with the colour display working) and the 'leaks' button
working.

The 'profile' button currently causes a segv (looks like it might be
gtk_list_store related at the moment) and I haven't tested on anything
apart from simple single-threaded apps.

Would patches along these lines be welcome? Sent to this list or the
current maintainer (I'm afraid I'm not sure who that is). I don't have
"production quality" patches yet, but I can clean up + post what I
have if there is any interest.

regards,

jb
_______________________________________________
memprof-list mailing list
memprof-list@...
http://mail.gnome.org/mailman/listinfo/memprof-list

Re: patches for memprof on current glibc

by Jochen Baier :: Rate this Message:

Reply to Author | View Threaded | Show Only this Message

John Berthels wrote:

>Hi folks,
>
>I've been trying to get GNOME cvs memprof building and running on ubuntu 5.10.
>
>There was a compilation issue with newer binutils (already fixed in
>the ubuntu patchset) and also a problem caused by the fact that
>backtrace() in glibc 2.3.5 calls dlopen(), which calls malloc and
>hence recurses.
>
>By sticking in a simple-minded malloc for use during startup (and
>forcing initialisation of backtrace() at init time), I can get memprof
>running (with the colour display working) and the 'leaks' button
>working.
>
>The 'profile' button currently causes a segv (looks like it might be
>gtk_list_store related at the moment) and I haven't tested on anything
>apart from simple single-threaded apps.
>
>Would patches along these lines be welcome? Sent to this list or the
>current maintainer (I'm afraid I'm not sure who that is). I don't have
>"production quality" patches yet, but I can clean up + post what I
>have if there is any interest.
>
>regards,
>
>jb
>_______________________________________________
>memprof-list mailing list
>memprof-list@...
>http://mail.gnome.org/mailman/listinfo/memprof-list
>
>  
>
hi,

iam really interested in a working memprof. this app is really great.
can you upload the patches/motified version somewhere ? or sent it to me
or to the list.


thanks for your work, jochen


_______________________________________________
memprof-list mailing list
memprof-list@...
http://mail.gnome.org/mailman/listinfo/memprof-list

Re: patches for memprof on current glibc

by John Berthels :: Rate this Message:

Reply to Author | View Threaded | Show Only this Message

> hi,
>
> iam really interested in a working memprof. this app is really great.
> can you upload the patches/motified version somewhere ? or sent it to me
> or to the list.

OK. This is a "works for me" patch against current GNOME CVS. Also
profiling seems to work if I build against my installed gnome libs
rather than gnome cvs libs. I've not tested multi-threaded or
multi-process scenarios.

>From the changelog:

* Pull in bfdutils.c compilation fix for new binutils from the
ubuntu 5.10/debian patchset (?from Matej Vela)

* Pull in mi_atomic_increment() changes in intercept.c from
Michael Meeks

* Use a dumb 'bozo allocator' during initialisation instead of
returning NULL and force a backtrace during initialisation so we
successfully init backtrace() in new glibc (2.3.5).

regards,

jb


_______________________________________________
memprof-list mailing list
memprof-list@...
http://mail.gnome.org/mailman/listinfo/memprof-list

memprof.patch.gz (2K) Download Attachment