Re: byte compiler warnings for different Emacs versions

[Stefan Monnier]

> If I use a compile-time test and I compile on Emacs 20, then the only code
> that shows up in the .elc is the Emacs 20 version of the definition.  Running
> the .elc on Emacs 21 then tries to run the Emacs 20 version, which is
> inappropriate and can result in errors.

Generally the Emacs-20 code will work fine.  The problem you've shown until
now had to do with Emacs-21-specific code (i.e. the define-minor-mode) being
compiled by Emacs-20.

> I offered this file as an example to look at wrt your statement that "If a
> file byte-compiled on Emacs-20 doesn't run on Emacs-21 it's generally
> considered as a bug (but not the other way around)."  Is this perhaps not
> part of your "generally" case (I guess so).

Maybe.  I can't know since you don't say which specific problem you're
thinking of.

> Anyway, this is the case that I was discussing from the beginning: I would
> like a _runtime_ (top-level) test in compiled code that will make the right
> function definition based on the runtime Emacs major version.

No problem there, except that a macro (or a special form) can't be properly
byte-compiled with an Emacs that doesn't define this macro, so you can't
properly compile with Emacs-20 a piece of code that uses
`define-minor-mode'.
There's nothing Emacs-21 can do about it.
I'd recommend either you don't use define-minor-mode, or if you do, use
it unconditionally and provide your own definition of it if the Emacs
you're running on doesn't have it.

> (when (>= emacs-major-version 21)
>   (define-key menu-bar-dired-operate-menu [touch]
>     '(menu-item "Change Timestamp..." dired-do-touch
>                 :help "Change timestamp of marked files")))

As usual, (>= emacs-major-version 21) is the wrong test:
you want to use something like (fboundp 'dired-do-touch).

> I had the impression that this also worked for defun in place of define-key.

It does.

> I now think I was perhaps mistaken about that (please confirm this). And
> apparently the following does not work, in any case: (if (...)
> (define-minor-mode...) (defun...)). You say that both defun and
> define-minor-mode suffer similarly, and I guess you're right.

What I meant by `defun' suffering similarly, is that if you byte-compile
a file with an Emacs that doesn't have `defun', then the piece of
byte-compiled code that uses `defun' will never run right, even on an Emacs
that does provide `defun'.  I know it's a bad example since all Emacsen
define `defun'.

> BTW, defun is referred to in Emacs 20 (via C-h f defun) as a "built-in
> function". That is why I referred to it that way.  And define-key is
> referred to in both versions as a "built-in function".

I know.  In Emacs-CVS (and Emacs-21.1 as well IIRC), this is fixed and
C-h f defun RET says "defun is a special form ...".


        Stefan