Thierry Volpiatto <thierry.volpiatto@gmail.com> writes:

> npostavs@users.sourceforge.net writes:
>
>> I've been attempting an alternate approach which prettyprints the object
>> while scanning it instead of the current way of printing and then
>> reindenting.  Without optimizing, it's about 3 times as fast as the
>> current pp (it's the pp-prin1 in the benchmarks below), though more than
>> 3 times slower than your mapc pp trick.  On the other hand, it also
>> doesn't yet handle function-specific indentation or any compound
>> structure apart from lists, so I'm not sure if it will end up being much
>> faster.
>>
>> (benchmark 1 '(with-temp-buffer (pp-prin1 long-list (current-buffer)) nil)) "Elapsed time: 3.391232s (0.565806s in 11 GCs)"
>> (benchmark 1 '(progn (pp-to-string long-list) nil))                         "Elapsed time: 9.988515s (0.148034s in 3 GCs)"
>> (benchmark 1 '(progn (with-output-to-string (mapc 'pp long-list)) nil))     "Elapsed time: 0.983493s (0.144424s in 3 GCs)"
>> (benchmark 1 '(progn (cl-prin1-to-string long-list) nil))                   "Elapsed time: 0.511617s (0.152483s in 3 GCs)"
>> (benchmark 1 '(progn (prin1-to-string long-list) nil))                      "Elapsed time: 0.029320s"
>
> Interesting, thanks to work on this.

With a couple of minor optimizations it's down to about 1.8s.  The first
is to reuse the tempbuffer instead of letting cl-prin1-to-string make a
new one each time.  Second is to add

    (eval-when-compile
      (cl-proclaim '(optimize (speed 3) (safety 0))))

This also stops these warnings (I guess they're caused by the "safety" code?):

../../emacs-master/lisp/emacs-lisp/pp.el:159:19:Warning: value returned from
    (aref state 6) is unused
../../emacs-master/lisp/emacs-lisp/pp.el:204:24:Warning: value returned from
    (aref state 10) is unused

New times:

(benchmark 1 '(with-temp-buffer (pp-prin1 long-list (current-buffer)) nil)) "Elapsed time: 1.800146s (0.231706s in 6 GCs)"
(benchmark 1 '(progn (pp-to-string long-list) nil))                         "Elapsed time: 9.950225s (0.154100s in 4 GCs)"
(benchmark 1 '(progn (with-output-to-string (mapc 'pp long-list)) nil))     "Elapsed time: 0.980923s (0.149787s in 4 GCs)"

I foolishly neglected to write down what exactly long-list was before,
starting from emacs -Q this seems to approximate it though:

(progn (require 'pp)
       (require 'dabbrev)
       (require 'edebug)
       (require 'cc-mode)
       (require 'vc)
       (setq long-list load-history)
       (length long-list)) ;=> 142