String#to_i slower than MRI when trailing garbage

[headius]

From http://jira.codehaus.org/browse/JRUBY-5620

The rough perf numbers reported toward the end of that bug show that only the trailing-garbage case of String#to_i still runs slower than other impls, and it does so due to a wasted BigInteger along the way. I file it here because it is still an issue and probably can be improved pretty easily.

The benchmark from that bug is still valid, but may have updates in the Rubinius repository.

[headius]

Current numbers:

~/projects/jruby $ jruby -Xcompile.invokedynamic=true blah.rb
Calculating -------------------------------------
#to_i with an integer in a string
                        83.570k i/100ms
#to_i with a float in a string
                       139.337k i/100ms
#to_i with an empty string
                       157.295k i/100ms
#to_i with an integer and extra text
                       113.454k i/100ms
-------------------------------------------------
#to_i with an integer in a string
                         16.610M (±13.9%) i/s -     80.561M
#to_i with a float in a string
                         15.420M (±14.8%) i/s -     74.685M
#to_i with an empty string
                         37.093M (±17.9%) i/s -    176.957M
#to_i with an integer and extra text
                          7.003M (±10.6%) i/s -     34.490M

[Who828]

Looking into it.

[headius]

Fixed by #2464.