Bug #20284: Regexp::TimeoutError is not triggered by a specific Regexp

• Status: Feedback
• Assignee: -
• Target version: -
• ruby -v: 3.3.0
• Backport: 3.0: UNKNOWN, 3.1: UNKNOWN, 3.2: UNKNOWN, 3.3: UNKNOWN

[jbeschi (jacopo beschi)]

I've noticed that the Regexp::TimeoutError is not triggered for this Regexp:

jacopo-37s-mb 3.3.0 ~ cat fail.rb
require 'benchmark'

p RUBY_DESCRIPTION
Regexp.timeout = 0.1
puts "Regexp.timeout: #{Regexp.timeout}"
re = /((a|aa)+)+b/
10.times do
  p Benchmark.realtime { re.match?("a" * 900000000) }
end

jacopo-37s-mb 3.3.0 ~ ruby fail.rb
"ruby 3.3.0 (2023-12-25 revision 5124f9ac75) [arm64-darwin22]"
Regexp.timeout: 0.1
1.1425850000232458
1.0082139987498522
1.064377998933196
1.0745620001107454
1.0647990014404058
1.6172770000994205
1.396629998460412
1.2450549993664026
1.3292079996317625
1.5961690004915

[mame (Yusuke Endoh)]

• Status changed from Open to Feedback

Thanks for the report.

This takes a long time in the preprocessing part before regexp matching. I think this process is O(n) for the length n of the input string. Since Regexp.timeout= is a countermeasure against super-linear behavior, I feel this is not a big deal. And frankly, I can't think of a good idea to fix it.

It is also known that the timeout do not work for very large regexps that contain no instruction that checks timeouts. Maybe I should update the documentation?

Let me know if anyone has a nice idea to fix it.

[jbeschi (jacopo beschi)]

mame (Yusuke Endoh) wrote in #note-1:

Thanks for the report.

This takes a long time in the preprocessing part before regexp matching. I think this process is O(n) for the length n of the input string. Since Regexp.timeout= is a countermeasure against super-linear behavior, I feel this is not a big deal. And frankly, I can't think of a good idea to fix it.

It is also known that the timeout do not work for very large regexps that contain no instruction that checks timeouts. Maybe I should update the documentation?

Let me know if anyone has a nice idea to fix it.

Got it, makes complete sense. A note in the docs about it would be helpful though.