Reduce regexps computed for parsing kramdown lists

[ashmaroli]

Rationale

Interpolated regular expressions within a frequently called method allocates memory for each generated expression. When possible, interpolation can be limited to just once by using the /o flag. But that is not possible in this situation.

While profiling the build for Jekyll's documentation site (via memory_profiler), I came across this hot path where the regex /^( {0,1}[+*-])([\t| ].*?\n)/ was allocated a significant amount of times closely followed by /^( {0,2}[+*-])([\t| ].*?\n)/ among a few others.

If it is of any relevance, the summary from the profiling is:

Before

Total allocated: 158.09 MB (1168556 objects)
Total retained:  10.3 MB (92547 objects)

After

Total allocated: 157.29 MB (1163537 objects)
Total retained:  10.3 MB (92556 objects)

[gettalong]

@ashmaroli Could you profile/benchmakr this change again? The change is quite large for a ~0,4% win on allocated objects.

[ashmaroli]

Using the following script to profile the changes:

# frozen_string_literal: true

require 'memory_profiler'
require 'bundler/setup'
require 'kramdown'

CONTENT = (<<~TEXT * 500)
  Once upon a time in a far away land

  * hello
  * world

  And this other time it was

    * hello again
    * worldwide

TEXT

class Profiler
  def self.profile(label, &block)
    report = MemoryProfiler.report(&block)
    total_allocated_output = report.scale_bytes(report.total_allocated_memsize)
    total_retained_output  = report.scale_bytes(report.total_retained_memsize)
    puts "Total allocated: #{total_allocated_output} (#{report.total_allocated} objects)"
    puts "Total retained:  #{total_retained_output} (#{report.total_retained} objects)"
    report.pretty_print(to_file: "#{label}-memprof.tmp", scale_bytes: true )
  end
end

Profiler.profile(ARGV[0]) { Kramdown::Document.new(CONTENT).to_html }

ruby profile.rb master

Total allocated: 29.11 MB (220240 objects)
Total retained:  840.45 kB (2971 objects)

and on this branch (with updates on master pulled)

ruby profile.rb pull-request

Total allocated: 26.07 MB (202381 objects)
Total retained:  843.01 kB (2988 objects)

[gettalong]

@ashmaroli Thanks for the profiling information!

So it seems that for very heavy usage of lists (nearly only lists), there can be a win of about 10% on allocated objects but there is no difference in execution as far as I could measure.

I also profiled some "standard" kramdown documents and there the win is also only 0,4% , like with your original profiling stats.

Nonetheless, I have merged your changes! Thanks again!

[ashmaroli]

Thank you for merging :)