Andre Poenitz wrote:
On Tue, Nov 07, 2006 at 04:43:07PM +0100, Abdelrazak Younes wrote:
As my painting investigation continue, it seems that this problem has
_nothing_ to do with painting on screen at all. Modifying a text line
between two huge formula exhibits the slow behaviour even if only the
text line is repainted on screen (you can use "-dbg painting" to make
sure of that.
As I said. Mathed is essentially uchanged since 1.2.0. It was completely
different before, but speed was acceptable in the 1.2.x and 1.3.x
series. So whatever triggers that behaviour is some recent interaction.
Indeed, the problem only appears with 1.4.
Note that I am not saying the data model is perfect, especially since
I changed my stance about reference counting and deep copies vs COW
lately.
I guess this could be related to the data memory models used in mathed.
Maybe some big tables created/deleted on the fly?
Hm... the macro tables?
I've tried to disable buildMacro() but that didn't help much.
Also, this comment from Michael Wojcik* makes me wonder if maybe mathed
does something special with lyxerr... any clue someone?
We should have a macro encapsulating lyxerr that guarantees minimal
overhead...
Agreed. And I agree also with your macro proposal.
"So, for example, if LyX were running with debug enabled, the debug
lines being written to the LyX console window might cause a noticeable
jump in csrss load. But I haven't tested that."
[Console output is terribly slow on Windows btw.]
And maybe that is the main problem. IIRC correctly LyX is compiled as a
console application; maybe, _maybe_, the problem with csrss.exe will
disappear if we change that.
Abdel.
