Performance problems in CMake support

[Robert Knight]

Hi,

I notice there are places in the code where QRegExp are constructed with
a constant pattern argument on each function invocation.  

eg.  string.indexOf(QRegExp("somepattern"));

QRegExp initialization is expensive so if possible, moving the QRegExp
to a static class member which is only initialized once should help.
The actual pattern matching against a string seems to be quite fast
(assuming Qt was build with optimizations enabled.  QRegExp is much,
much slower in debug-only builds)  

What I think Aleix is getting at is that a lot of time seems to be spent
in QDir::entryInfoList() to find various libraries and other files.
QDir::entryInfoList() uses QDirIterator which spends a lot of time
constructing QRegExps to do the filter matching.    

For reference, this is the full callgrind output I got from loading the
KDevelop 4 project.

http://www.robertknight.me.uk/files/temp/cmake-parser-profile.tar.gz

Regards,
Robert.


On Tue, 2008-04-01 at 16:35 +0200, Aleix wrote:
> I have removed (and commited) the QRegExp's from the envvar and the
> definitions. I've tested it when loading kdebase and it doesn't make
> any difference.
> 
> I still hope that using the cache it will be a little faster. :P
> 
> Bye,
> Aleix
> 
> On Tue, Apr 1, 2008 at 4:11 PM, Andreas Pakulat <apaku at gmx.de> wrote:
>         On 01.04.08 15:48:39, Aleix wrote:
>         > If you grep QRegExp on the parser directory you will see
>         that it is only on:
>         > - CMakeAst, command parser:
>         >     * To collect the cmake_minimum_required version value.
>         >     * The same for cmake_policy (2.6)
>         > - Condition evaluator:
>         >     * For the MATCHES case (from cmake, can't remove it).
>         > - Project Visitor:
>         >     * To get the information from the envvar list (it could
>         be changed, and
>         > I will, but I don't think it is significant)
>         >     * On the STRING command on regex-related options.
>         >     * ADD_DEFINITION command: to extract info from the
>         definitions. (it
>         > could be changed, and I will, it appears quite often so
>         maybe it could be
>         > significant)
>         >     * REMOVE_DEFINITION command: the same but barely used.
>         >
>         > It is used on the manager 1 time so it is not a problem as
>         well.
>         >
>         > I'll remove now the few I think that could be removed but,
>         as I said, I
>         > think the big load of regex is used by the cmake scripts
>         themselves.
>         
>         
>         Ok, so maybe we should think about not using Qt's regex
>         support, but
>         instead something else, like pcre? Or check what cmake itself
>         uses.
>         
>         Andreas
>         
>         --
>         You are fighting for survival in your own sweet and gentle
>         way.
>         
>         
>         _______________________________________________
>         KDevelop-devel mailing list
>         KDevelop-devel at kdevelop.org
>         https://barney.cs.uni-potsdam.de/mailman/listinfo/kdevelop-devel
>         
> 
> _______________________________________________
> KDevelop-devel mailing list
> KDevelop-devel at kdevelop.org
> https://barney.cs.uni-potsdam.de/mailman/listinfo/kdevelop-devel