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

 >> Given that switch-to-buffer returns its argument, /and/ given that
 >> mapc returns the sequence it’s given, I suggest that the (mapcar
 >> 'switch-to-buffer LIST) forms in lisp/files.el be replaced with
 >> (mapc 'switch-to-buffer LIST), – if only to avoid the unnecessary
 >> consing when the list is effectively copied in the mapcar case.

 > Thanks, I think it's indeed a valid/correct optimization, but I
 > really dislike relying on mapc's return value (it really should not
 > return any value at all).

	I tend to disagree with that last part, – it seems like a common
	idiom for a function (or, generally, – a /form/; setq does that,
	for one thing) that’s used “solely” for its side-effects to
	return its “primary” argument, thus allowing for easy
	“chaining”, like:

   (foo (bar 42 (baz data)))

	Instead of:

   (progn
     (foo    data)
     (bar 42 data)
     (baz    data))

	Or something even more crude, like:

   (mapc (lambda (fn)
           (if (consp fn)
               (apply (car fn) `(,@(cdr fn) ,data))
             (funcall fn data)))
         '(foo (bar 42) baz))

	Surely, ‘mapc’ fits that pattern pretty well; cf.:

   (foo (bar  42   (baz lst)))
   (foo (mapc 'qux (baz lst)))

 > In this case, the optimization doesn't seem worth the inconvenient of
 > having a very unusual code (relying on mapc's return value), since
 > those few cons cells we save are drowned in the noise of all the code
 > run by switch-to-buffer.

	Yes.  However, I believe that the last two hunks of the one
	another variant of the diff (MIMEd) actually make the intent to
	return the reverse of the list returned by find-file-noselect
	/clearer,/ – although at the expense of adding one extra LoC in
	each case.

-- 
FSF associate member #7257  http://boycottsystemd.org/  … 3013 B6A0 230E 334A