>>>>> "Abdelrazak" == Abdelrazak Younes <[EMAIL PROTECTED]> writes:
Abdelrazak> Maybe my code is wrong but I'd really like to know why.
Abdelrazak> All I am doing is saving the pointer in the table. The
Abdelrazak> second time a font that has the same characteristics is
Abdelrazak> required, instead of creating it again I pass the one
Abdelrazak> created before. So the contruction of a new font is
Abdelrazak> skipped.
I do not think the code is wrong, but it is not better than the old
one (and as Andre' said, it evaluates the array indices twice).
The old code is
/// Get font info (font + metrics) for the given LyX font.
QLFontInfo & fontinfo(LyXFont const & f) {
QLFontInfo * & fi =
fontinfo_[f.family()][f.series()][f.realShape()][f.size()];
if (!fi)
fi = new QLFontInfo(f);
return *fi;
}
The "QLFontInfo * & fi" construct means that fi's address is the table
entry. Therefore, "fi = new QLFontInfo(f);" does set the entry in the
array.
Abdelrazak> If the pointer retrieved in the table is equal to zero
Abdelrazak> that means that this particular font was not used before
Abdelrazak> and we have to create it (via new) and save its pointer to
Abdelrazak> the table.
Abdelrazak> The older code was doing nothing as the table was filled
Abdelrazak> with zero and was never filled in anywhere!
Do you have evidence that it does not? It may be that I read
the code wrong.
Abdelrazak> I am doing just that. This patch is the direct analysis of
Abdelrazak> the profiling done by Bennett, please read that and the
Abdelrazak> analysis I made. Then, please explain me why I am wrong
Abdelrazak> when I think there is a problem with font caching under
Abdelrazak> MacOSX.
I think there is something wrong with the amount of time we compute
width of characters.
- there are remnants of a charwidth caching in the qt3 frontend (for
qt < 3.1, see USE_LYX_FONTCACHE). I remember that at the time I
removed this, Bennett complained that it made LyX feel slower, and I
disregarded his complaints... It may be time to revisit this.
- You asked about computing font width twice. Actually Andre did post
a patch to avoid that,
http://marc.theaimsgroup.com/?l=lyx-devel&m=112825896416687&w=2
but I think the results at the time were not very encouraging.
>> Otherwise, you could maybe find someone who has vtune...
Abdelrazak> I was trying to help MacOS port so I don't see how
Abdelrazak> profiling under wintel will help that as the two
Abdelrazak> implementions of Qt4.1 have very different performances.
The sluggishness of LyX/Qt 1.4 (wrt 1.3) is general, unfortunately.
Only the xforms frontend has reasonable performance.
JMarc
PS: looks like I am making you grumpy. It was not really intended.
