Re*Re^2: The High Price of Golf, and A Surprise

[bart]

That's very interesting. I had noticed the same thing. Your conclusion might very well be correct, and I hadn't even thought of that possibility.

Could the next be a special common case, too?

perl -MO=Deparse,x7 -e'sort { $b cmp $a } @x'
sort @x;
-e syntax OK

Well well well, it looks like it. It would explain why there appears to be no overhead for the callback. It can also explain the huge speed difference between "<=>" and "-": only the latter actually uses the callback. Otherwise, it would make no sense: "-" is a very fast operator, and checking the sign of the result should be just as fast. So it must be the overhead of the callback.

Nevertheless, B::Deparse is broken, because the code it produces is obviously not equivalent to the original source code.

[Aristotle]

That would be broken by necessity rather than in implementation. There simply is no way to represent a numerical sort without putting a callback in the source, but the compiled form of that sort doesn't use any callback. The choice is to represent either the original or the compiled form more closely. I prefer the current behaviour of emphasizing the compiled form since that's what the module is for, after all. Maybe that should be, if implementable, a user selectable option.

Good catch about the reason of the - vs <=> slowdown. I didn't make that (so very obvious, in retrospect) connection.

Makeshifts last the longest.

[hv]

Just for the record: firstly, it used to be that even simplistic sort subs such as { $a <=> $b } would call out to perl code, but that was changed to look for certain common cases and substitute a C implementation instead; I think that change first appeared in 5.6.1, judging by comparative timings here.

Secondly, the bug in B::Deparse (which occurred simply because it wasn't immediately updated for the above change) was fixed in perl-5.8.0.

Just to give a feel for the difference it makes spotting and substituting a C-based routine for a perl-based one, I added an extra variant of i_sship tweaked to avoid the optimization:

  i_sshipz => sub { @rin = sort { $a + 0 <=> $b } @nin },

and here is the table I got (running under 5.8.0):

            Rate i_sshipz  f_owtdi  i_owtdi  f_alpha  i_alpha  f_sship
+  i_sship
i_sshipz  16.5/s       --     -27%     -28%     -43%     -52%     -57%
+     -58%
f_owtdi   22.6/s      37%       --      -1%     -22%     -34%     -41%
+     -42%
i_owtdi   22.8/s      38%       1%       --     -21%     -33%     -41%
+     -41%
f_alpha   29.0/s      76%      28%      27%       --     -15%     -25%
+     -26%
i_alpha   34.2/s     107%      51%      50%      18%       --     -11%
+     -12%
f_sship   38.5/s     134%      70%      69%      33%      13%       --
+      -1%
i_sship   38.9/s     136%      72%      71%      34%      14%       1%
+       --

Hugo