Re: featurep instead of bound tests

[Stephen J. Turnbull]

Dan Nicolaescu writes:

 > Stefan Monnier writes:

 >   > Agreed.  I see 2 better solutions:
 >   > 1 - use (require 'outline)

I have no opinion on that, but

 >   > 2 - teach the byte-optimizer that (fboundp 'make-overlay) will
 >   >     always be t.

is not useful.

(fboundp 'make-overlay) is always the wrong thing to do (unless the
code gracefully exits, doing nothing, without it).  XEmacs *does* have
`make-overlay' (it's deprecated and not autoloaded or dumped, but it
exists in a compatibility library that can be require'd).  I imagine
there are some people who have emulated `make-extent' with
`make-overlay' for Emacs, too.

However, emulations are rarely perfect (XEmacs's `make-overlay' is
explicitly not intended to be), so many developers are going to wish
to use the native facilities.  That *requires* a (featurep 'xemacs) or
equivalent to know which is native when both are fboundp.

 > [Optimizing (fboundp 'make-overlay)] has the problem that XEmacs
 > would have to teach its compiler that it is always false.

No, that's not a problem.  The compiler doesn't need to know, it's an
optimization.  In any case, it's not always false in XEmacs.