Another performance improvement on the Formatter by using Kernel#sprintf over String#%

[amatsuda]

String#% takes the arguments as an Array object, which requires us to create an extra object (like we're doing here).

Also, while both String#% and Kernel#sprintf internally falls back rb_str_format in the current CRuby implementation, the former seems to be performing a little bit of argument checking, which possibly causes a very subtle performance difference (?).

Anyway, here's a benchmark of String#% vs Kernel#sprintf.

require 'benchmark/ips'

Benchmark.ips do |x|
  Format = "%s, [%s #%d] %5s -- %s: %s\n"
  DatetimeFormat = "%Y-%m-%dT%H:%M:%S.%6N"
  time = Time.now.strftime(DatetimeFormat)
  pid = Process.pid

  x.report('String#%') { Format % ['D', time, pid, 'DEBUG', 'hello', 'world'] }
  x.report('sprintf') { sprintf(Format, 'D', time, pid, 'DEBUG', 'hello', 'world') }

  x.compare!
end

Warming up --------------------------------------
            String#%   205.492k i/100ms
             sprintf   219.627k i/100ms
Calculating -------------------------------------
            String#%      2.057M (± 0.9%) i/s -     10.480M in   5.094731s
             sprintf      2.193M (± 0.8%) i/s -     10.981M in   5.006767s

Comparison:
             sprintf:  2193434.0 i/s
            String#%:  2057237.4 i/s - 1.07x  (± 0.00) slower

P.S. I have no strong preference on whether to use Kernel#sprintf or Kernel#format. I'm happy to rewrite the patch to call format(...) (or Kernel.sprintf(...) or Kernel.format(...)) if the maintainer has any preference.