Project loading problems

[Matt Rogers]

On Monday 31 July 2006 18:29, Matt Rogers wrote:
> On Monday 31 July 2006 10:03, mwoehlke wrote:
> > Matt Rogers wrote:
> > > On Friday 28 July 2006 14:50, mwoehlke wrote:
> > >> Matt Rogers wrote:
> > >>> On Thursday 27 July 2006 15:37, mwoehlke wrote:
> > >>>> Before I start filing bug reports, I'd like to check here for any
> > >>>> suggestions in being as helpful as possible.
> > >>>>
> > >>>> I just built KDevelop 3.3.91 (the most recent source tarball I could
> > >>>> find). I have had several crashes while trying to load a rather
> > >>>> large project. Are the backtraces helpful, or is there something I
> > >>>> could do to provide additional information?
> > >>>
> > >>> It would be useful if you could try a checkout of
> > >>> anonsvn.kde.org/home/kde/branches/kdevelop/3.4 and see if the crashes
> > >>> go away. Backtraces would be interesting if you've built kdevelop
> > >>> with debug information.
> > >>
> > >> Ack. Naturally, now that I have rebuilt with --enable-debug=full, it
> > >> doesn't crash. :-)
> > >>
> > >> At least if it does it again, I'll both know what to do, and will have
> > >> a trace from a debug KDevelop.
> > >>
> > >>>> Also, as it takes several minutes to load the project, I decided to
> > >>>> throw callgrind at KDevelop to investigate why this is. Would anyone
> > >>>> be interested in the raw output, or should I just summarize?
> > >>>
> > >>> raw output of the callgrind output posted somewhere would be the most
> > >>> useful, since then we'll be able to open it in kcachegrind, analyze
> > >>> it, and fix the bottlenecks.
> > >>
> > >> Ok, is there anywhere good I can post it? I don't have any of my own
> > >> webspace for that sort of use. (FWIW, I paid attention this last time;
> > >> KDevelop took about 1:40 to fully start when opening my large
> > >> project.) Also, is it expected to have three output files?
> > >>
> > >> I can at least make this observation: there is a 25% self cost to
> > >> "tryit()", called by KMimeMagic::findFileType->process->tryit. About
> > >> 90% of the calls come from methods of CppSupportPart; seems like a
> > >> place to start looking, at least.
> > >>
> > >> Thanks,
> > >
> > > how large is the output? i might be able to find a place for you to put
> > > it
> >
> > All three combined and bzip2'd are 3873340 (just under 3.7 MB). If
> > nothing else, I'd be happy to simply e-mail it to you, but that probably
> > is not the best solution for your end.
>
> bzip2 file requested to be sent to me off list.
> --
> Matt

I've done what I can with the callgrind traces and have eliminated the mime 
type comparisons from the 3.4 branch in subversion. Please let me know if 
this speeds things up for you since it's about the only thing I can do for 
kdevelop 3.4. The other bottlenecks that I saw in my own traces were memory 
allocations and the lexer, which is being replaced for KDevelop 4 anyways.
--
Matt