Julian Bradfield writes:
>What's wrong with using the approach (and the code) from 21.5
>reads in a charset definition from a Unicode Consortium-format table?
>These are dumpable AFAIK.
Nothing, and that was my first thought. But then I thought that iconv
would have the advantage of making it easy to load the translation tables
on demand - why should I have 20MB (or however much it is) of
translation tables in main memory of every XEmacs instance, when my
typical instance only ever sees Latin-1 and maybe few dozen hanzi?
No reason at all. (Just how dumb do you think Ben Wing is, anyway!?
Don't answer that, I know you know better.<wink>) First of all, here
are the numbers from M-x show-memory-usage (I think you need to
compile --with-debug to get that function in 21.5, and it may not be
available in 21.4 at all):
charset from-unicode to-unicode total
total 528448 358440 886888
Yes, those numbers are *bytes*, not "words", "paragraphs", or TiB.
(You're thinking of the Unihan.txt database, no doubt, which is indeed
almost 30MB.) Second, look at how `load-unicode-tables' is
implemented. You'll see that there is no reason why we can't load a
minimal subset and add more on an as-needed basis. Since this is a
once-per-session thing, who cares if it takes 25ms per table?
There currently is no way to request on-demand loading, but I don't
think that would be hard.
What's the definition of compatibility these days?
"Catering to differences is not too annoying." Differences in basic
Mule functions would be too annoying.
I haven't used fsfmacs for a decade or more, but my impression
that most non-trivial elisp has to cater specially for fsfmacs and
XEmacs. Is that wrong?
No. For one thing, RMS considers following standards or established
practice for the sake of conforming a form of subordination akin to
slavery, and thus ethically unacceptable in the free software movement.
However, and it's very important, the cores of most applications work
fine in both forks. It's stuff like images that is quite different.
Extents are a little bit annoying but there are several wrapper
packages such as overlay.el in fsf-compat, as well as more specialized
but accurate (within the field of interest) emulations in Gnus and VM
(inter alia). Similarly ez-menu or whatever it's called. Heck, you
know how Kyle feels about GNU and its assignment policy, but he's
always maintained compatibility for VM, including use of the menubar
(he'd basically retired by the time GNU got images and toolbars---
scary thought, isn't it).
Several years ago I was going to write a wrapper package for images,
too, implementing the GNUbie interface, but then I realized that
practically speaking the dox would have to be licensed under the
God-Forsaken Documentation License and I lost all appetite for the
project. (I've grown up a little since then, but that stupid piece of
free-advertising-grubbing hypocrisy *still* rubs me the wrong way.)
XEmacs-Beta mailing list