>>>> In <87d6h1gm50.fsf(a)tleepslib.sk.tsukuba.ac.jp>
>>>> "Stephen J. Turnbull" <stephen(a)xemacs.org> wrote:
KY> In principle, APEL does never redefine existing functions having
KY> no problems,
In practice in XEmacs it regularly does.
Yep, my recognition was possibly lacking.
We should build APEL under 21.1 (in practice I think we're
building under 21.4), or some packages depending on APEL may not run
on 21.1, because APEL decides what to provide at compile time, not
at load time. This means that anything added or fixed in 21.4 or
21.5 gets redefined by APEL.
I was really often troubled with the problem (since I'm now using
only 21.4, I tend to forget it, though). It is mainly caused by
the def*-maybe macros which switch function definitions at the
compile time. Although it was probably a rational technique for
early days, it is not suitable to the XEmacs package. In
addition, def*-maybe began to be abused by other programs.
I and Shuhei KOBAYASHI used to try to overcome the problem. The
plan was to make def*-maybe macros switch function definitions
at the load time by default. The code already exists in the
branch of cvs.m17n.org
, but it is not fully verified.
We could of course just remove APEL and everything it depends on
the packages and tell the users to build their own. This is better
for the users, but rather annoying for the developers. Or we could
rebuild APEL for each version of XEmacs, but aside from being extra
work for Norbert, this would be hard to make work in installations
providing multiple versions of core XEmacs.
Is there any XEmacs package using APEL actively? It seems to be
very few to me. I think separating APEL from XEmacs is good
idea even though I am one of the APEL developers.
KY> but it fixes bugs in old function definitions, emulates functions
KY> which aren't available in old Emacsen, ...
Right. So that means when I do C-x C-e on the sexp provided by the
(non-APEL-using) Original Poster, if APEL has been installed in my
XEmacs I will see up-to-date interfaces and behavior, and typically if
I then do C-x f on the function, it looks like the version distributed
in core XEmacs because it doesn't get defun'ed by APEL, instead the
function cell is set directly. ISTR that APEL even replaces subrs in
the function cell of some symbols without saying so in the docstring.
This is very confusing when debugging.
I sometimes want to keep away from APEL for the same reason.
All this could be fixed either in APEL or in XEmacs, but I think it
would require a fair amount of work either way.
It will be easier to abolish APEL from XEmacs. I am writing so
in the hope that other APEL developers bring counterargument. :p
Katsumi Yamaoka <yamaoka(a)jpl.org>