Re: libc/1427: gprof does not profile threads on linux

[Dan Kegel]

Mark Hahn wrote:
> 
> > Pragmatically, however, there is a simple workaround, documented at
> > http://sam.zoy.org/doc/programming/gprof.html
> > which works by putting a small wrapper around pthread_create.
> > (I haven't tested it myself, but it seems reasonable.)
> 
> [ Which standard defines thread & signal semantics? ]

SUSv2 does; see
http://www.opengroup.org/onlinepubs/007908799/xsh/sigaction.html

> [is the glibc team trying to use the new signal-related support yet? ]

You're referring to CLONE_THREAD and friends, I think.
I don't know, but here's what a bit of google searching finds:

In 2.4-test8-pre1, Linus added a first cut at thread group support.
In March '01, someone summarized progress so far:
http://groups.google.com/groups?hl=en&selm=linux.kernel.6ovg0gq952n.fsf%40calypso.engr.sgi.com
In Oct '01 there was a brief thread indicating further progress:
http://groups.google.com/groups?hl=en&selm=linux.kernel.E15qcwe-00010e-00%40the-village.bc.nu
In Nov '01, IBM released a pthreads implementation for 2.4.9 or later that
uses CLONE_THREAD:
http://www-124.ibm.com/pipermail/pthreads-users/2001-November/000094.html

I don't know what the glibc team has done in this area.  Maybe they
are waiting to see how the IBM approach pans out?

- Dan