Re: nXML

[Stephen J. Turnbull]

Stefan Monnier writes:

 > As for `char-names/unicode', that would fit better in `etc/unicode' as well,

My experience suggests you want to avoid the identifier "unicode" in
packages, either as a full name or as a component.  It's just going to
give annoying collisions when emacs-unicode gets merged, and the worst
part is that unicode is such basic and useful functionality that the
packaged stuff will have all kinds of clients, both in emacs proper
and in third party packages.

 > all those char-name/unicode/*.el files should be auto-generated from
 > emacs/admin/unidata/Unicode.txt (maybe even on-the-fly ;-).

XEmacs 21.5 does something similar, reading the mapping files for
charsets, at startup.  There's a measurable but not really painful
performance hit (750 ms on a 750MHz machine; I suspect it's I/O bound
so you won't get a proportional benefit from higher CPU speeds).
N.B. We have C support for this.

As long as you put it in C, I expect "measurable but not really
painful" perfomance hit for this too.