Neil Hodgson wrote:
Robert Roessler:
As always, it would be nice if someone that runs SciTE in GTK-land
would try this version of Scintilla's PlatGTK.cxx to see how much it
does for "real" (not on Windows) GTK - I would do it myself if I had
GTK development and runtime environments...
Compiled both and visually there is no obvious winner: on an AMD64
3500 running Fedora Core 4 in a maximized window on a 1680x1050
monitor opening scintilla/src/Editor.cxx and paging down. I
uncommented the measurement code in Editor::Paint and the numbers
aren't really consistent but the cached painting is around 10%
*slower* than the original 1.68 code. Layout is 25% slower and I'm not
sure why but it could be a result of memory pressure, locking, or
interactions between caching layers.
Thanks for looking at this in some detail, Neil... while things sound
sort of grim, it isn't really clear your results make any sense - you
can see how simple the code is, and memory-to-memory pixmap copying
will presumably not cost *more* than full text rendering of the same
string (even *if* the text rendering engine is caching some of the
glyphs).
So. In the case of "GTK on Windows", I see a clear improvement - but
you find that in "real" or "native" GTK, my current approach to deal
with sucky - no, make that totally abysmal - text rendering
performance is not a winner... that presumably means I need to look
elsewhere (or possibly gain a better understanding of your results).
BTW, I notice in a different post that you mention using VMware for
*nix testing - was that the case here, or was this a Fedora Core 4
running "on the metal"?
Robert Roessler
[EMAIL PROTECTED]
http://www.rftp.com
_______________________________________________
Scintilla-interest mailing list
[email protected]
http://mailman.lyra.org/mailman/listinfo/scintilla-interest
