[Python-Dev] is anyone using the ucnhash C API?

M.-A. Lemburg mal@lemburg.com
Sat, 15 Jul 2000 19:03:30 +0200

Fredrik Lundh wrote:
> (except for the unicode object module, that is).

I don't think so... it's very new and the C API isn't
> as I've mentioned earlier, I'm working on an alternative unicode
> database, which includes a ucnhash replacement.
> my first thought was to emulate the existing api, but it has two
> small problems:
> -- it returns pointers into the database.  this means that you
>    cannot compress the names at all, if you want to be thread-
>    safe (which you have to be, in this case).
> -- it exposes low-level implementation details, such as the hash
>    table size, etc.  I'm not sure if I can emulate that in any sane
>    way under the new database design.
> so if nobody uses this, I'd rather change the unicodeobject.c
> file to use the new API directly...

Please don't drop the idea of an on-demand loadable module
though. The current design used by Bill Tutt is to load the
ucnhash module (which contains the mappings) only when it
is needed -- most code won't need it.

BTW, does you database use the on-demand loading feature ?
I don't think that having the Unicode database linked statically
in the Python interpreter is a good idea -- at least not until
Unicode takes over ;-)

Marc-Andre Lemburg
Business:                                      http://www.lemburg.com/
Python Pages:                           http://www.lemburg.com/python/