bizarre problem with minor mode defined using define-minor-mode

[ken manheimer <ken.manheimer@gmail.com>]

[-- Attachment #1: Type: text/plain, Size: 3840 bytes --]

i'm encountering a serious problem with a function created using
define-minor-mode - and the problem is only occurring when operating with
the byte-compiled version of the function.  i've tried to debug the problem,
but am stumped.  i'm able to reproduce the problem in a few versions of
emacs, though, and have created a bzr branch on launchpad with a copy of the
code.

one of the final steps in my updates to allout.el is migrating allout-mode
minor mode to `define-minor-mode'.  however, some key substitutions way deep
in my code do only happen properly when i use a explicitly evaluated (via
eval-defun or eval-when-compile) version of the define-minor-mode to define
the allout-mode function.  if the minor mode was established by a
byte-compiled version of the function the same key substitutions do not show
in the mode.  (there are some other keybinding problems that occur, as
well.)

i wonder whether i'm somehow misusing define-minor-mode in a dumb way, but
am entirely missing it if that's the case.

the problem happens identically on the x toolkit and chiba-u versions of
emacs, both running on macox X snow leopard:

 GNU Emacs 23.2.91.1 (x86_64-apple-darwin10.5.0, Carbon Version 1.6.0 AppKit
1038.35) of 2010-12-20 on crisp

 GNU Emacs 23.1.90.2 (x86_64-apple-darwin10.4.0, X toolkit) of 2010-11-05 on
crisp

i can reproduce the problem by wrapping the define-minor-mode in either
`eval-and-compile' or `eval-when-compile' and doing an
`emacs-lisp-byte-compile-and-load' to show the different behaviors.  i've
created a bzr branch on launchpad.net with a prepared version (details,
below) of allout.el for others to try:


http://bazaar.launchpad.net/~ken-manheimer/emacs/alloutdev/annotate/head:/lisp/allout.el?start_revid=102483

in this version i've wrapped the define-minor-mode with eval-and-compile.
 that can be changed to test the different behaviors.

when byte-compiling and loading with the eval-when-compile, invoking the
mode yields the key substitutions, as it ought.  (of course, the step of
loading the byte-compiled code is irrelevant at that point, because the
function isn't in the byte-compiled code.  it's only defined because it was
evaluated during the byte-compile.)   with the eval-and-compile, loading the
byte-compiled allout.elc wipes out the key substitutions - they're no longer
in effect.  of course, the lack of the function in the byte-compiled code
when using eval-when-compile makes the byte-compiled code useless, so i have
no workaround for the problem, just a weird demonstration of it.

i've tried tracking down the failed key substitutions, by
wrapping define-key with some advice that noted when the key defines
happened, and they do every time the mode is activated, whether or not it's
via the byte-compiled version of the mode function (when defined
by define-minor-mode).  my function which does the key
substitutions (allout-setup-mode-map) uses fset to ensure that the mode map
is properly globally established, so it's already kind of complicated.  i'm
also wondering whether i'm just misunderstanding something about the way
define-minor-mode is supposed to work.

to see the problem, visit this version (see link, above) of allout.el and
start allout-mode.  then emacs-lisp-byte-compile-and-load the code, and
you'll find that \C-a is bound to move-beginning-of-line, as it normally is
- not replaced by allout-beginning-of-line, as it should be.

then change the `eval-and-compile' that's wrapped around
the define-minor-mode allout-mode to `eval-when-compile',
do emacs-lisp-byte-compile-and-load, and you'll find that \C-a is properly
bound to allout-beginning-of-line.  (you can also just manually evaluate the
define-minor-mode - that will also the keysubsitutions in the keymap
immediately.)

help would be much appreciated!
--
ken manheimer
http://myriadicity.net

[-- Attachment #2: Type: text/html, Size: 4263 bytes --]