"Desmond, Colin" wrote:
I am working on the KProf front-end for gprof, as well as some other
profilers, and I was wondering what the future plans are for gprof, in terms
of additional functionality.
For starters:
* support for multiple threads
* support for large programs
(see http://www.kegel.com/gprof.html )
I can also imagine
* ability to start and stop profiling during execution
The problem is, gprof is partly in binutils, and partly in glibc.
The glibc maintainer is very, very picky about what patches
he'll accept. In particular, he has not yet accepted the perfectly
reasonable patch
http://www.kegel.com/gmon.patch
which fixes support for large programs. I haven't tried hard
enough to get it accepted; perhaps some readers of this list
could advise me on how to proceed.
I don't know that the above patch is `perfectly reasonable' :-) It
modifies a gprof data structure in a way that makes it incompatible with
every other gprof implementation (all the ones I've seen came from the
same source). I think it would be better to define a new data
structures that could handle this.