bug#6997: Loading w32-fns under X11 signals an error

[Stefan Monnier <monnier@iro.umontreal.ca>]

>> They can define new functions and variables, as well as load other
>> files, and even set a few things up but only to *add* behavior, not to
>> *change* behavior.
> Not sure how instrumental that definition is.  Doesn't addition
> constitute a change?

I already mentioned this loop-hole:

   they do change the semantics of the language but in a way that's
   usually accepted as "harmless": it only gives meaning to things which
   earlier were errors.

I'm not sure exactly what you're trying to argue here.  It's been policy
for a long time now that loading files should be harmless, so many files
were changed so that (load "foo") doesn't enable foo-mode any more, but
instead just defines foo-mode (and then foo-mode is marked auto-loaded
so the user can just call (foo-mode 1) rather than (load "foo") in his
.emacs).
Clearly, load foo.el will make changes in that it will define new
functions and variables, may add things to debug-ignored-errors and
whatnot.  I can't believe you really don't know what I mean by
"harmless".
   
> Let's take a practical example: w32-convert-standard-filename.  Do you
> want it to become effective or not?

What means "effective"?  Do I want it to be defined just by loading
w32-fns.el?  Very much so, yes.
Do I want convert-standard-filename to use it just because I've loaded
w32-fns.el?  Well, no since it would then change the behavior of Emacs.

And indeed the current code behaves correctly in this respect, AFAICT.

> If your answer is NO, then what about someone (not called Stefan) who
> wants to test some issue for which w32-convert-standard-filename must
> be in effect?

He'll have to weak the code/variables that decide whether to call
w32-convert-standard-filename or not.

>> > IOW, can you come up with a general enough definition of the reason(s)
>> > for loading w32-fns on non-w32 systems?
>> The same reason why I might want to load diff-mode.el without ever
>> wanting to use that mode.
>> E.g. to complete function/variable names, or to browse customize data, ...
> That's rather a narrow need.  Sounds like lots of labor to gain very
> little (why does it matter to have symbols that begin with a `w32-'?).

Yes, it's a lot of work for fairly little gain.  Note that I bumped into
it because my code couldn't care less that "w32-" is special, not
because I specifically wanted to have symbols starting with "w32-".


        Stefan