José Matos wrote:
On Monday 27 October 2008 15:29:05 Jean-Marc Lasgouttes wrote:
José Matos <[EMAIL PROTECTED]> writes:
So yes it should be possible to speed the conversion but there are
reasons for the slow response. Up until this moment we have focused on
making it correct, and just after we should make it fast. :-)
It seems that this time has come :)

OK. I think that I have found the culprit.

Attached follows a patch to that cut the conversion time from 63 to 10 seconds in one of my recent slowest examples.

FWIW the problem was that the unicode symbols were being read on each call of the functions that need them. In the case above that happened 918 times (!).

I would appreciate the testing of this patch. The patch is straightforward and it should be correct.

Mea culpa. I think I must have had it the way you have it at some point. I do remember thinking about this. (Maybe what I did, actually, was move it out of a loop.) Anyway, I can't imagine that this patch isn't correct.


Reply via email to