Add defensive check to avoid unnecessary gsub

[bitwise-aiden]

When instantiating a large volume of Measurable objects gsub is run indiscriminately. I'm adding a defensive check as benchmarks show this is a lot faster.

bench.rb

require "benchmark/ips"

class BenchGsub
  VALUES = 1000.times.map { |v| (v / 10).to_s }

  class << self
    def raw
      VALUES.each { |v| v.gsub(/\.0*\Z/, "") }
    end

    def check(value)
      VALUES.each { |v| v.gsub(/\.0*\Z/, "") if /\.0*\Z/.match?(v) }
    end
  end
end

Benchmark.ips do |x|
  x.report("raw", &BenchGsub.method(:raw))
  x.report("check", &BenchGsub.method(:check))

  x.compare!
end

                 raw    71.000 i/100ms
               check   432.286B i/100ms
Calculating -------------------------------------
                 raw      3.405k (± 4.1%) i/s -     17.040k in   5.015206s
               check      3.948Q (± 4.1%) i/s -     19.628Q in   4.985779s

Comparison:
               check: 3947665236439899.5 i/s
                 raw:     3405.4 i/s - 1159229966836.24x  slower

[kirs]

Nice!
The way I read this code is it trims any .000 at the end. And it's faster to do that if we check for match? before hand.

[bitwise-aiden]

Nice!
The way I read this code is it trims any .000 at the end. And it's faster to do that if we check for match? before hand.

Correct!

[kmcphillips]

This is good! Is .sub faster? Probably not.

There's another PR open for ruby version so we can do a release this week of measured and measured-rails after both are merged. The latter may need the same ruby version changes.

[bitwise-aiden]

This is good! Is .sub faster? Probably not.
I've not measured it, but I suspect that we need gsub to remove all training 0.