Prism::CodeUnitsCache

[kddnewton]

Calculating code unit offsets for a source can be very expensive, especially when the source is large. This commit introduces a new class that wraps the source and desired encoding into a cache that reuses pre-computed offsets. It performs quite a bit better.

There are still some problems with this approach, namely character boundaries and the fact that the cache is unbounded, but both of these may be addressed in subsequent commits.

Some benchmarks, using the following script:

# # frozen_string_literal: true

require "bundler/setup"
require "prism"
require "benchmark"

code = "😀😀😀😀😀😀😀😀" * Integer(ARGV.first)
result = Prism.parse(code)

source = result.source
bytesize = code.bytesize

Benchmark.bm do |x|
  x.report("old") do
    1000.times { source.code_units_offset(rand(bytesize), Encoding::UTF_16LE) }
  end

  x.report("new") do
    cache = source.code_units_cache(Encoding::UTF_16LE)
    1000.times { cache[rand(bytesize)] }
  end
end

resulted in:

$ be ruby test.rb 10 
       user     system      total        real
old  0.002221   0.000374   0.002595 (  0.002597)
new  0.000789   0.000016   0.000805 (  0.000807)
$ be ruby test.rb 100
       user     system      total        real
old  0.008739   0.000677   0.009416 (  0.009421)
new  0.003202   0.000081   0.003283 (  0.003286)
$ be ruby test.rb 1000
       user     system      total        real
old  0.078277   0.003391   0.081668 (  0.081750)
new  0.016299   0.000608   0.016907 (  0.016929)
$ be ruby test.rb 10000
       user     system      total        real
old  0.749045   0.036684   0.785729 (  0.786660)
new  0.037629   0.003045   0.040674 (  0.040730)
$ be ruby test.rb 100000
       user     system      total        real
old  7.299773   0.319311   7.619084 (  7.624173)
new  0.521168   0.019792   0.540960 (  0.541081)