On Dec 9, 2005, at 10:07 AM, Jean-Marc Lasgouttes wrote:
Bennett> Here's the output from the last linking stage:
It looks reasonable. How was the qt library with profiling named?
Something like libqt-mt_s?
Looking in qt-mac-free-3.3.4/lib, I find libqt-mt.a, which is about 8
times bigger than the version produced without profiling.
Some additional notes:
- it is known that Shark.app provides wrong stacks apparently:
http://lists.apple.com/archives/PerfOptimization-dev/2005/Jun/
msg00266.html
Bennett> ... So does that mean I should do things differently?
No, but version 4.2.3 should have several stack related problems
fixed, so you could try to re-run tests with this version.
OK -- I tried again, applying callstack data mining, charging system
libraries to callers, and charging code without debug info to callers
(which I think is what I was doing before). The result can be found
here: <http://edisk.fandm.edu/bennett.helm/LyX/shark-profile.txt.zip>.
- there is also a tool named QuartzDebug that shows which areas of
a window are updated by brief flashing. It might be interesting to
play around with this.
Bennett> It's a little hard to tell the order of events given that
Bennett> things happen so quickly. But it seems like as you type, the
Bennett> individual character that gets typed gets updated, then the
Bennett> info bar at the bottom of the window is updated, and then the
Bennett> whole text area of the window is updated. It's that last item
Bennett> that seems significant: the entire window is updated with
Bennett> each keystroke.
Even when typing at top-level? We should only update the paragraph at
cursor, I think.
Yes. Perhaps I should add that by "text area" I include the grey area
at the bottom of a document but above the info bar. So even in a 1-
line document, most of the window area flashes when a single
character is typed.
Bennett
