Re: bad change to tooltip.el

[Stefan Monnier]

>> I think it is.  Using `fboundp' to see if a function will be available when
>> the code will be run is OK for a quick-and-dirty solution, but it's
>> otherwise just naive.

> I'm not sure I understand that,

Here's an example: if you compile a file which only contains

   (defun foo () (byte-compile-callargs-warn form))

you won't get any warning, no matter how separate and clean an Emacs process
you use for the byte-compilation and even though `form' is not defvar'd
anywhere and byte-compile-callargs-warn will not be defined at runtime.

> but the usual problem with not compiling in a clean Emacs is that you have
> something loaded already (e.g. cl-macs) which masks the fact that the Lisp
> in question needs a require, autoload, or whatever.  That's what I found
> quite a lot in Gnus stuff when it was used without most of Gnus loaded.

I know, but I think the better solution is that in
byte-compile-callargs-warn, instead of using fboundp, we should track the
function name through load-history and things like that to see whether it
will be available at runtime, given the set of requires specified in
the file.


        Stefan