Lars Ingebrigtsen <larsi@gnus.org> writes:

> Hm...  I'm not quite sure I understand these benchmarks.  There are no
> completion predicates for any eww-* symbols, I think?  So there
> shouldn't be any differences here at all, if I read the patch correctly.

Oops, sorry, I conflated completion-predicate with command-modes.

The eww completion is only one half of the benchmark; the rest operates
over all commands in obarray (but less than 1% of those have a
completion-predicate, hence the (failed) attempt to prefilter obarray).

> What I was wondering was -- if (at some point in the future) we have,
> say, 2K symbols with a completion predicate -- would putting that
> condition-case inside the loop make things slower or not?  My gut
> feeling says that it's probably not an issue (so we should just go ahead
> and do it), but it'd be nice to know...

Here's the attached benchmark on 16K commands (warning: takes a while):

  == TAB RET ==

  bench-vanilla-no-pred    3.533359 100 2.058253
  bench-vanilla-no-pred    3.531937 100 2.054076
  bench-vanilla-no-pred    3.529376 100 2.052615
  bench-vanilla-with-pred  3.627937 105 2.127567
  bench-vanilla-with-pred  3.616596 105 2.117001
  bench-vanilla-with-pred  3.612720 105 2.114463
  bench-vanilla-safe-pred  3.639222 105 2.132278
  bench-vanilla-safe-pred  3.647615 105 2.134284
  bench-vanilla-safe-pred  3.636989 105 2.127953
  bench-safe-no-pred       3.573388 100 2.087055
  bench-safe-no-pred       3.555347 100 2.069622
  bench-safe-no-pred       3.545747 100 2.063041
  bench-safe-with-pred     3.653066 105 2.145138
  bench-safe-with-pred     3.643867 105 2.140400
  bench-safe-with-pred     3.659689 105 2.150498

  == b e n c h - TAB C-a C-k RET ==

  bench-vanilla-no-pred    0.649064  0 0.000000
  bench-vanilla-no-pred    0.652256  0 0.000000
  bench-vanilla-no-pred    0.677194  0 0.000000
  bench-vanilla-with-pred  1.652515 40 0.750009
  bench-vanilla-with-pred  1.653861 40 0.753575
  bench-vanilla-with-pred  1.672873 40 0.747151
  bench-vanilla-safe-pred  1.669653 40 0.742116
  bench-vanilla-safe-pred  1.673809 40 0.743255
  bench-vanilla-safe-pred  1.672750 40 0.741095
  bench-safe-no-pred       0.646114  0 0.000000
  bench-safe-no-pred       0.644843  0 0.000000
  bench-safe-no-pred       0.646764  0 0.000000
  bench-safe-with-pred     1.672326 40 0.743054
  bench-safe-with-pred     1.677331 40 0.742163
  bench-safe-with-pred     1.711009 40 0.759025

So condition-case still looks negligible relative to the rest of the
work, and my vote's for putting it around the call to
read-extended-command-predicate.

-- 
Basil