Jonas H. reported about slow horizontal scrolling on talk.
http://rubyforge.org/pipermail/sup-talk/2010-December/004400.html
I looked into it and found out that scrolling is pretty much fully
dependant on Buffer#write and main cpu hogs within it are
String#display_length (uses String#scan when on 1.8) and
Ncurses::WINDOW#method_missing (wide/normal dispatching).
Caching String#display_length cuts down String#scan calls by ~30%.
That and hardwiring Ncurses::WINDOW#mvaddstr and #attrset cut average
Buffer#write call to half of what it was.
Also having configurable COL_JUMP would help people who need to scroll
horizontally a lot. Included in the sup patch.
Perceivable difference of these modifications is very small, at least
to me. Please have a look if these patches (one of them against sup
and another against ncursesw.rb in ncursesw gem) make any sense.
====
--- /usr/lib/ruby/gems/1.8/gems/ncursesw-1.2.4.2/lib/ncursesw.rb 2010-12-
31 00:37:31.000000000 +0200
+++ lib/ncurses.rb 2010-12-30 23:50:26.000000000 +0200
@@ -59,6 +58,29 @@
module Destroy_checker; def destroyed?; @destroyed; end; end
class WINDOW
include Destroy_checker
+
+ @@mvwaddstr = Ncurses.respond_to? "mvwaddstr"
+ # This would be handled by #method_missing below, but for
+ # performance reasons it is extracted here. #mvaddstr is the beef
+ # of Redwood::Buffer#write which does all the drawing.
+ def mvaddstr(*args)
+ if @@mvwaddstr
+ Ncurses.mvwaddstr(*args)
+ else
+ Ncurses.mvaddstr(*args)
+ end
+ end
+
+ @@wattrset = Ncurses.respond_to? "wattrset"
+ # See #mvaddstr abowe.
+ def attrset(*args)
+ if @@wattrset
+ Ncurses.wattrset(*args)
+ else
+ Ncurses.attrset(*args)
+ end
+ end
+
def method_missing(name, *args)
name = name.to_s
if (name[0,2] == "mv")
====
--
Tero Tilus ## 050 3635 235 ## http://tero.tilus.net/
Attachment:
0002-Performance-and-configurability-of-horizontal-scroll.patch
Description: Binary data
_______________________________________________ Sup-devel mailing list Sup-devel@rubyforge.org http://rubyforge.org/mailman/listinfo/sup-devel