[ruby-core:105613] [Ruby master Bug#17507] Regexp capture groups ignored sometimes in some multithreaded environments (possible race condition)

Issue #17507 has been updated by nagachika (Tomoyuki Chikanaga).

Backport changed from 2.6: UNKNOWN, 2.7: REQUIRED, 3.0: REQUIRED to 2.6: UNKNOWN, 2.7: REQUIRED, 3.0: DONE

ruby_3_0 cfad0583eb18bb4505f28ee5cfab703a6b9987be merged revision(s) abc0304cb28cb9dcc3476993bc487884c139fd11.

----------------------------------------
Bug #17507: Regexp capture groups ignored sometimes in some multithreaded environments (possible race condition)
https://bugs.ruby-lang.org/issues/17507#change-94102

* Author: byteit101 (Patrick Plenefisch)
* Status: Closed
* Priority: Normal
* ruby -v: ruby 3.0.0p0 (2020-12-25 revision 95aff21468) [x86_64-linux]
* Backport: 2.6: UNKNOWN, 2.7: REQUIRED, 3.0: DONE
----------------------------------------
Same behavior from 2.6-3.0
OS: Debian 10

Ruby Script:

``` ruby
TargetStr = "a-x-foo-bar-baz-z-b"

worker = lambda do
	# For more hits, use File.read here instead of TargetStr
	m = TargetStr.match(/foo-([A-Za-z0-9_\.]+)-baz/) # more cases in the []+ means more hits
	raise "Error, #{m.inspect}" if m[1].nil?
	File.exist? "/tmp"
	TargetStr.gsub(/foo-bar-baz/, "foo-abc-baz") # must match the same as the first match
end

def threaded_test(worker)
	6.times.map {Thread.new {10_000.times {worker.call}}}.each(&:join)
end
threaded_test(worker) # must be a function calling a block or proc or lambda. Change any of that and it doesn't hit this

puts "No Error Hits"
```

Result of script (number of exceptions varies per-run):

```
#<Thread:0x00005616663dba08 bug.rb:12 run> terminated with exception (report_on_exception is true):
bug.rb:6:in `block in <main>': Error, #<MatchData "foo-bar-baz"> (RuntimeError)
	from bug.rb:12:in `block (3 levels) in threaded_test'
	from bug.rb:12:in `times'
	from bug.rb:12:in `block (2 levels) in threaded_test'
#<Thread:0x00005616663db5f8 bug.rb:12 run> terminated with exception (report_on_exception is true):
bug.rb:6:in `block in <main>': Error, #<MatchData "foo-bar-baz"> (RuntimeError)
	from bug.rb:12:in `block (3 levels) in threaded_test'
	from bug.rb:12:in `times'
	from bug.rb:12:in `block (2 levels) in threaded_test'
#<Thread:0x00005616663dbb98 bug.rb:12 run> terminated with exception (report_on_exception is true):
bug.rb:6:in `block in <main>': Error, #<MatchData "foo-bar-baz"> (RuntimeError)
	from bug.rb:12:in `block (3 levels) in threaded_test'
	from bug.rb:12:in `times'
	from bug.rb:12:in `block (2 levels) in threaded_test'
bug.rb:6:in `block in <main>': Error, #<MatchData "foo-bar-baz"> (RuntimeError)
	from bug.rb:12:in `block (3 levels) in threaded_test'
	from bug.rb:12:in `times'
	from bug.rb:12:in `block (2 levels) in threaded_test'
```

Expected Result:

```
No Error Hits
```

Details:

As is, it usually errors (most of the time it fails about 300-500 iterations in)
A success is what is expected, m == #&lt;MatchData "foo-bar-baz" 1:"bar">
An error is when m == #&lt;MatchData "foo-bar-baz"> (no capture saved)

* If threaded_test is only one thread (1.times) it always works. For more than one thread, it usually errors.
* If File.exist? is removed, it always works.
* If the gsub is removed, it always works.
* If the gsub is changed to not match the match() call, it always works.
* If the match expression is simpler (/foo-(bar)-baz/, or even /foo-([a-z]+)-baz/), it usually works.
* If the threaded_test is not a method, but executed inline (ie, comment out the def, end, call, on lines 11 & 13-14) it always works.
* If worker is a block, or a lambda (as provided), it usually errors.
* If worker is put inline into the 10_000.times call, it always works.
* If TargetStr matches on index 0, it always works.





-- 
https://bugs.ruby-lang.org/

Unsubscribe: <mailto:[e-mail]?subject=unsubscribe>
<http://lists.ruby-lang.org/cgi-bin/mailman/options/ruby-core>