Neil Hodgson wrote:
Robert Roessler:
2) is the Pango/GTK on Windows combo doomed?
Do you read Federico Mena Quintero?
http://primates.ximian.com/~federico/news-2005-10.html
Don't know if any of the optimizations have made it to a GTK+
release or onto the Windows port.
A current summary of my (superficial?) performance investigations:
0) this is with the Scintilla GTK widget on Windows XP, attempting to
keep up with a continuous PageDown key stream (repeat rate in BIOS is
30 per sec, I am not sure about XP) to the bottom and then back to the
top; window size is 640x480, file being colored/displayed is ~2300
lines of OCaml
1) there is no funniness with ColouriseCamlDoc - I see a stream of
one-page requests until the bottom of the file - each exactly once;
there are no more when I reverse and go back up to the top
2) the char width cache used for SINGLE-char width computation
requests represents 38% of ALL width calls; those are virtually all
hits, with the only "resets" happening during (at the most) the first
1000 width calls. There are around 28000 width calls, avg is ~5 chars
3) replacing ALL >1 char width calls with "len * 10" gives a somewhat
less than perfect display... and NO perceived speedup; so it looks
like it is the actual rendering and display
4) plugging in the latest [Windows] development builds of GLib and the
Pango/Cairo packages DOES improve perceived snappiness, BUT not enough
to "un-peg" the CPU
So, IS the "Pango/GTK on Windows combo doomed"? Well, there sure are
a *lot* of software layers BEFORE even getting to Windows... I wonder
how this looks in a "native" GTK setting?
Final comment/question (for this posting): if I change my local
SciTE.properties and set
font.base=font:!Bitstream Vera Sans,size:10
font.comment=font:!Bitstream Vera Serif,size:10
as my only "base" fonts (with all other fonts derived from these),
does that make my normal Windows-build SciTE/Scintilla use Pango? If
so, then this should be "like" my testing target platform, WITHOUT
"GTK/GTK on Windows"... in any case, this certainly looks different
(but not the same as my testing platform) - and while "jumping up" the
CPU to maybe 25%, does not really affect the perceived responsiveness.
Robert Roessler
[EMAIL PROTECTED]
http://www.rftp.com
_______________________________________________
Scintilla-interest mailing list
[email protected]
http://mailman.lyra.org/mailman/listinfo/scintilla-interest
