featurep instead of bound tests (was: Changes to emacs/lisp/gnus/ChangeLog, v)

featurep instead of bound tests (was: Changes to emacs/lisp/gnus/ChangeLog, v)

[Reiner Steib]

On Thu, Jan 24 2008, Dan Nicolaescu on emacs-diffs:

> --- gnus/ChangeLog	24 Jan 2008 07:47:37 -0000	1.633
> +++ gnus/ChangeLog	24 Jan 2008 15:14:50 -0000	1.634
> @@ -1,3 +1,9 @@
> +2008-01-24  Dan Nicolaescu  <dann@ics.uci.edu>
> +
> +	* sieve.el (sieve-make-overlay, sieve-overlay-put, sieve-overlays-at):
> +	* message.el (message-beginning-of-line): Use featurep instead of bound
> +	tests in order to resolve conditionals at compile time.

In the past, I've been told that it's preferable to use fbound tests
instead of Emacs flavor or version tests.  So I'm surprised by changes
like these...

> --- gnus/sieve.el	8 Jan 2008 20:45:19 -0000	1.12
> +++ gnus/sieve.el	24 Jan 2008 15:14:49 -0000	1.13
> @@ -290,15 +290,15 @@
>    (get-char-property (or pos (point)) 'script-name))
>
>  (eval-and-compile
> -  (defalias 'sieve-make-overlay (if (fboundp 'make-overlay)
> -				    'make-overlay
> -				  'make-extent))
> -  (defalias 'sieve-overlay-put (if (fboundp 'overlay-put)
> -				   'overlay-put
> -				 'set-extent-property))
> -  (defalias 'sieve-overlays-at (if (fboundp 'overlays-at)
> -				   'overlays-at
> -				 'extents-at)))
> +  (defalias 'sieve-make-overlay (if (featurep 'xemacs)
> +				    'make-extent
> +				  'make-overlay))
> +  (defalias 'sieve-overlay-put (if (featurep 'xemacs)
> +				   'set-extent-property
> +				 'overlay-put))
> +  (defalias 'sieve-overlays-at (if  (featurep 'xemacs)
> +				   'extents-at
> +				 'overlays-at)))

Bye, Reiner.
-- 
       ,,,
      (o o)
---ooO-(_)-Ooo---  |  PGP key available  |  http://rsteib.home.pages.de/

Re: featurep instead of bound tests (was: Changes to emacs/lisp/gnus/ChangeLog, v)

[Dan Nicolaescu]

Reiner Steib <reinersteib+gmane@imap.cc> writes:

  > On Thu, Jan 24 2008, Dan Nicolaescu on emacs-diffs:
  > 
  > > --- gnus/ChangeLog	24 Jan 2008 07:47:37 -0000	1.633
  > > +++ gnus/ChangeLog	24 Jan 2008 15:14:50 -0000	1.634
  > > @@ -1,3 +1,9 @@
  > > +2008-01-24  Dan Nicolaescu  <dann@ics.uci.edu>
  > > +
  > > +	* sieve.el (sieve-make-overlay, sieve-overlay-put, sieve-overlays-at):
  > > +	* message.el (message-beginning-of-line): Use featurep instead of bound
  > > +	tests in order to resolve conditionals at compile time.
  > 
  > In the past, I've been told that it's preferable to use fbound tests
  > instead of Emacs flavor or version tests.  So I'm surprised by changes
  > like these...

In general probably yes, but these particular cases are about functions
that emacs will probably never get because it has similar facilities:
overlays vs extents.


  > > --- gnus/sieve.el	8 Jan 2008 20:45:19 -0000	1.12
  > > +++ gnus/sieve.el	24 Jan 2008 15:14:49 -0000	1.13
  > > @@ -290,15 +290,15 @@
  > >    (get-char-property (or pos (point)) 'script-name))
  > >
  > >  (eval-and-compile
  > > -  (defalias 'sieve-make-overlay (if (fboundp 'make-overlay)
  > > -				    'make-overlay
  > > -				  'make-extent))
  > > -  (defalias 'sieve-overlay-put (if (fboundp 'overlay-put)
  > > -				   'overlay-put
  > > -				 'set-extent-property))
  > > -  (defalias 'sieve-overlays-at (if (fboundp 'overlays-at)
  > > -				   'overlays-at
  > > -				 'extents-at)))
  > > +  (defalias 'sieve-make-overlay (if (featurep 'xemacs)
  > > +				    'make-extent
  > > +				  'make-overlay))
  > > +  (defalias 'sieve-overlay-put (if (featurep 'xemacs)
  > > +				   'set-extent-property
  > > +				 'overlay-put))
  > > +  (defalias 'sieve-overlays-at (if  (featurep 'xemacs)
  > > +				   'extents-at
  > > +				 'overlays-at)))
  > 
  > Bye, Reiner.
  > -- 
  >        ,,,
  >       (o o)
  > ---ooO-(_)-Ooo---  |  PGP key available  |  http://rsteib.home.pages.de/

Re: featurep instead of bound tests

[Stefan Monnier]

>> --- gnus/ChangeLog	24 Jan 2008 07:47:37 -0000	1.633
>> +++ gnus/ChangeLog	24 Jan 2008 15:14:50 -0000	1.634
>> @@ -1,3 +1,9 @@
>> +2008-01-24  Dan Nicolaescu  <dann@ics.uci.edu>
>> +
>> +	* sieve.el (sieve-make-overlay, sieve-overlay-put, sieve-overlays-at):
>> +	* message.el (message-beginning-of-line): Use featurep instead of bound
>> +	tests in order to resolve conditionals at compile time.

> In the past, I've been told that it's preferable to use fbound tests
> instead of Emacs flavor or version tests.  So I'm surprised by changes
> like these...

Agreed.  I see 2 better solutions:
1 - use (require 'outline)
2 - teach the byte-optimizer that (fboundp 'make-overlay) will always be t.


        Stefan

Re: featurep instead of bound tests

[Dan Nicolaescu]

Stefan Monnier <monnier@iro.umontreal.ca> writes:

  > >> --- gnus/ChangeLog	24 Jan 2008 07:47:37 -0000	1.633
  > >> +++ gnus/ChangeLog	24 Jan 2008 15:14:50 -0000	1.634
  > >> @@ -1,3 +1,9 @@
  > >> +2008-01-24  Dan Nicolaescu  <dann@ics.uci.edu>
  > >> +
  > >> +	* sieve.el (sieve-make-overlay, sieve-overlay-put, sieve-overlays-at):
  > >> +	* message.el (message-beginning-of-line): Use featurep instead of bound
  > >> +	tests in order to resolve conditionals at compile time.
  > 
  > > In the past, I've been told that it's preferable to use fbound tests
  > > instead of Emacs flavor or version tests.  So I'm surprised by changes
  > > like these...
  > 
  > Agreed.  I see 2 better solutions:
  > 1 - use (require 'outline)
  > 2 - teach the byte-optimizer that (fboundp 'make-overlay) will always be t.

That has the problem that XEmacs would have to teach its compiler that
it is always false. Or if the test was (fboundp 'make-extent) emacs will
have to know it's always false.
IMO for extents vs overlays and zmacs-* the featurep test is the best
option.

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.

Re: featurep instead of bound tests

[Richard Stallman]

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

That is a good idea.

Re: featurep instead of bound tests (was: Changes to emacs/lisp/gnus/ChangeLog, v)

[Richard Stallman]

    In the past, I've been told that it's preferable to use fbound tests
    instead of Emacs flavor or version tests.

Yes, often the fboundp tests are prefereable.

    In general probably yes, but these particular cases are about functions
    that emacs will probably never get because it has similar facilities:
    overlays vs extents.

Even so, the fboundp tests are sometimes better.