LWN: Comments on "Poke-a-hole and friends"

Poke-a-hole and friends

acme — Mon, 22 Jun 2009 19:33:41 +0000

Most of the time it was used in C codebases, but it support most C++ constructs.

The most recent activity is in fact related to C++: holes are being miscalculated after ancestor classes due to the way this info is being encoded by gcc: base classes sometimes are just forward declared in the debug section for an object file, so one can't find its size.

See the current discussion in the dwarves mailing list at:

http://article.gmane.org/gmane.comp.debugging.dwarves/109

Please feel free to submit problems, I seldom write C++ code so C++ support in the dwarves comes from interacting with people trying to use it, like in past interactions with developers from KDE, CERN, Mozilla, etc.

Poke-a-hole and friends

johill — Sun, 21 Jun 2009 11:27:20 +0000

> pfunct --cc_inlined shows functions which are declared inline but have been uninlined by the compiler. Such functions are good candidates for a second look, or for removing the inline declaration altogether. Fortunately, pfunct --cc_uninlined on vmlinux (only) did not list any functions.

I think that should be 'pfunct --cc_unlinined' in both instances, rather than just the second.

Poke-a-hole and friends

tdz — Sun, 21 Jun 2009 09:22:28 +0000

I use pahole for some of my programs and really like it. Thanks to Arnaldo Carvalho de Melo for making this little gem available.

Regards, Thomas

Poke-a-hole and friends

sigra — Sat, 20 Jun 2009 11:08:11 +0000

Is this mature enough yet to be let loose on a whole application, to find
things that could easily be optimized? Something that could be run along
with cppcheck, krazy2 and other tools that a project may have integrated.
Does it have 0 false positives? Does it work for C++?