gettext needs a rewrite

[Rich Felker]

On Mon, Mar 26, 2007 at 02:39:34PM +0800, Rogelio Serrano wrote:
> im just wondering if anybody else besides me is thinking of a gettext rewrite.
> 
> or is catgets a better thing to do?

catgets is a much better design, and the obvious implementation is
absolutely tiny (like 1k of machine code, at most). i'm planning on
writing one for my libc implementation since i want to support most or
all of susv3. however, all the applications i know use gettext
instead, so catgets has little practical purpose except for trying to
convince application authors to use it.

gettext cannot be made efficient; you always need at least one strcmp
per lookup to make sure you got the right hit. this is inherent in
using strings as keys. it also means that the catalogue file for each
language must duplicate all the english strings, which wastes a huge
amount of space.

this trouble _can_ be avoided if you're willing to sacrifice
compatibility of message catalogues between different versions of your
application. you could choose a hash function based on the set of
english strings in your application, such that there are no
collisions. this is always possible for a fixed set of strings and
very practical if the set is reasonably small. then, you can simply
use the hashes as indices into the message catalogue files as if you
were using catgets instead of gettext. some precaution should be taken
to ensure that the hash function for the application binary and the
catalogue file match.

rich