`str.count("\n")` is 1.3-170 times faster than `str.lines.count` or `str.each_line.count` depending on the string size

[ilyazub]

str.count("\n") is 1.3-170 times faster than str.lines.count or str.each_line.count (ref: https://serpapi.com/blog/lines-count-failed-deployments/). The speed difference grows with the lines count.

$ ruby tmp/string_count_benchmark.rb
Warming up --------------------------------------
 String#count('\n') 86.000 i/100ms
 String#lines.size 1.000 i/100ms
 String#lines.count 1.000 i/100ms
String#each_line.count
 1.000 i/100ms
Calculating -------------------------------------
 String#count('\n') 771.031 (± 6.6%) i/s - 3.870k in 5.041849s
 String#lines.size 4.785 (± 0.0%) i/s - 24.000 in 5.037242s
 String#lines.count 4.513 (± 0.0%) i/s - 23.000 in 5.112095s
String#each_line.count
 4.763 (± 0.0%) i/s - 24.000 in 5.075882s

Comparison:
 String#count('\n'): 771.0 i/s
 String#lines.size: 4.8 i/s - 161.12x (± 0.00) slower
String#each_line.count: 4.8 i/s - 161.87x (± 0.00) slower
 String#lines.count: 4.5 i/s - 170.86x (± 0.00) slower

Benchmark code:

require "benchmark/ips"

HTML = "\nruby\n" * 1024 * 1024

def fastest
 HTML.count("\n")
end

def faster
 HTML.each_line.count
end

def fast
 HTML.lines.length
end

def slow
 HTML.lines.size
end

Benchmark.ips do |x|
 x.report("String#count('\\n')") { fastest }
 x.report("String#lines.size") { faster }
 x.report("String#lines.count") { fast }
 x.report("String#each_line.count") { slow }
 x.compare!
end

I'd like to add this benchmark to fast-ruby. Wdyt?

---

Based on our updates to the @guilhermesimoes' very helpful gist: https://gist.github.com/guilhermesimoes/d69e547884e556c3dc95?permalink_comment_id=4687645#gistcomment-4687645

[ilyazub]

@JuanVqz @etagwerker what do you think about a benchmark for String#count vs String#lines.count vs String#each_line.count?

[JuanVqz]

This seems a Rails related benchmark, I wonder if we are adding framework-related benchmarks

[ilyazub]

It doesn't depend on Rails (ruby/ruby#4001 (comment)). I updated the benchmark code to work with the plain Ruby.

[ilyazub]

@JuanVqz I updated the benchmark code above to work with the plain Ruby.

What do you think?

[ixti]

IMHO using $/ is bad - it's less obvious than "\n". More than that, $/ can be redefined, so the code may become broken.

[ilyazub]

@ixti Sounds good. The performance difference is in the String#count vs other methods. $/ vs \n did't impact performance here.