From: Benoit Daloze <eregontp@...>
Date: 2012-06-19T00:54:20+09:00
Subject: [ruby-core:45701] Re: A Proposal to Improve the Method String.concat

Hello,

On 18 June 2012 15:16, Martin Vahi <martin.vahi@softf1.com> wrote:
>
> I'm sorry, if I'm double-posting or even triple-posting.
> It seems that one has to perform mail based registration
> before the posts from the http://www.ruby-forum.com/ reach the list.
>
> What regards to performance optimizations, then
> I have Ruby and PHP demo scripts at
>
> https://github.com/martinvahi/mmmv_notes/tree/master/mmmv_notes/phenomenon_scrutinization/string_concatenation
>
> that demonstrates at least 20% performance difference
> between plain string concatenation algorithm and
> the versions that can be downloaded from there.

I believe this is not worth it.
It seems you are not aware of String#<<, and that Ruby has mutable strings.
Which means there is no need for optimized repeated String#+, one can
just #dup the original (if needed) and use #<<(aliased to #concat)
afterwards.

In numbers:

n = 100000
String#concat: 0.11s user 0.01s system 97% cpu 0.115 total
watershed concatenation: 0.18s user 0.03s system 97% cpu 0.215 total
pseudo-watershed concatenation: 4.00s user 2.98s system 98% cpu 7.083 total
plain loop concatenation: 7.33s user 7.54s system 98% cpu 15.081 total

n = 1000000,
String#concat: 0.88s user 0.01s system 99% cpu 0.896 total
watershed concatenation: 2.04s user 0.43s system 99% cpu 2.471 total

But maybe I missed something.