> we should have a release soon. To do :
> 0.0.5 task list
> o get opf_filter and friends working OK
it work actually?
> o fix showing all counters for oprofpp -s
I prefer to not fix this and than nobody try to fix it until my
oprofpp/opf_filter change are commit.
> o clean up starting GUI (currently non-intuitive, I'm on it)
> o check documentation
> o re-do performance tests
really needed for a cvs version?
> o autoconf for sstream - Philippe, what do we need to check for here ?
> What fails/ what doesn't with which gcc versions ?
sstream is the only problem (for gcc < =2.95.2). But I prefer to make the check
generic, like (not tested):
add sys_include/sstream in the distrib
try in autconf
if fail add a -I sys_include to CFLAGS. This is not really clean because
I made assumption than preprocessor -I directory are tried after
all predefined system include dir. Strictly speaking this not portable
> Anything else ? Philippe, I'd prefer oprofpp changes to be minimal
> unless you think the best way to fix opf_filter problems is via a re-jig of
> the code.
I get too many problem by trying minimal change but we retrieve the
same code inside class member. They remains a dozen of un-readable
lines like :
"vma += xxxx + zzz - vvvv; then call a function that make vma -= zzzz + .... "
there is two class: one to encapsulate bfd stuff and one for samples files.
I have finish this and it work, now I need to use this inside opf_filter.
The two main effects are: remove the issue "output of oprofpp and reading
code inside opf_filter must match" and that remove a great part of pipeline
problem in op_to_source
My change fix also a few buglet inside filename handling of oprofpp.
> And for anyone who wants to add to these :
> 0.0.6 task list
I have planned to doxygen'ize c++ file, this imply review/recomment opf_filter.
> o make hashmap use string pool method
+ fix zero dname map hash entry bug + trick to avoid rebuilding pathname
> o implement proc->old_exec
> o post-prof GUI ? (big job)
+ a gui to handle a notion of profiling session (not relied to the
session_number variable used actually that I have miss-named)
allowing easy maintenance of session and comparing results of
session. I think at something like: opening a new sesssion create
a sub dir in samples dir and all samples files go in samples/session_name.
Then add tools to allows comparing two session. I prefer to have
limitation on what can be compared: something like: allow binary to be
different but source file of the binary must be very near from one session
to another, this is related to the problem to make a powerfull op_diff tools.
These tools can be very usefull for common case: try different compiler,
different compiler options, try to tune system/application through configuration
I am go away 5/6 days from friday 24. I am unsure if we need my change
to oprofpp.c / opf_filter.cpp before or after 0.05.