Le 29/04/2010 14:27, Stephan Hennig a écrit :
I'm not really closely following the current restructuring efforts, but
could you please point out what makes first-class support for patterns
from dehyph-exptl (german-x et al.) desirable?
Nothing but the desire (at least from my side) to avoid any (even the slightest)
regression in the changes we're currently making unless necessary.
Currently, the cost of supporting "special" (see below) languages by handling
them the old way is:
- essentially zero in terms of code complexity,
- winning only 2,2 second when we could win 2.4 in terms of format loading time
(that is, getting "only" 90% of the maximal win).
How do the timestamped
patterns break your scheme for all the other languages?
Short version: our choice is to use only plain text files for patterns loaded at
runtime, similar to those found at [1], kindly provided by Mojca. If you want to
provide such files in dehyph-exptl, then we'll happily support dynamic loading
of these patterns.
[1] http://www.tug.org/svn/texhyphen/trunk/plain/
IMO, in the long run, it is the most desirable solution. It is probably the
format that will be required (or at least recommended) when proper dynamic
loading will be supported by a babel extension or polyglossia (IMO, AFAICS, etc).
Best,
Manuel.
