Re: AbiWord 0.5.2

Shaw Terwilliger (sterwill@postman.abisource.com)
Mon, 22 Mar 1999 14:55:48 -0600


Darren Benham wrote:
> > perhaps you have a different version of ispell than we have.
> > we only use the /usr/lib/ispell directory for the .hash files.
> > grab a copy of our american.hash* on the download page and
> > see if it works (put it in some other direcorty so you don't
> > mess up ispell).
>
> Yes, it worked. Here's the stats on the american.hash we use.. what can I
> do so that this works on a general basis? I *could* keep a seperate copy
> of the american.hash around for abiword but that's less then ideal.

There are lots of problems with the way ispell makes dictionaries,
and the way different Unix platforms build/use them. For example,
ispell's dictionary hashes depend on the architecture's endianness
which means one would have to provide both big endian and little
endian hashes for SPARC/PowerPC/M68K vs. Intel/Alpha. I guess VAX
and its endianness it its own category. Anyway, the problems don't
stop there... I wasn't successful getting my Intel-built, 32-bit
hash working on my 64-bit Alpha... AbiWord is built in native 64-bit
code, but the ispell stuff won't read that dictionary from the
32-bit machine; both Intel and Alpha are little endian.

There is also the matter of the languages the dictionaries were
build to handle. English has 26 "flags" in the dictionary, whereas
other languages might have double the number (52 "flags").
Now, one would think that American.hash would always have 26 flags,
but the dictionaries Debian (and maybe Red Hat) installs have 52.
Hence the "hash options don't agree" problem.

The interim solutions I can think of is just to build custom
ispell dictionaries on all the architectures having problems with
each other. I guess we defer the problem of building custom user
hashes during execution. :)

-- 
Shaw Terwilliger


This archive was generated by hypermail 1.03b2.