Bug #8444: Regexp vars $~ and friends are not thread local

• Status: Assigned
• Assignee: ko1 (Koichi Sasada)
• Target version: -
• ruby -v: trunk
• Tags: core

[jamespharaoh (James Pharaoh)]

In the docs for the Regexp special variables, $~ and friends, it says "These global variables are thread-local and method-local variables". However the following gives an unexpected result:

def get_proc
  proc do |str|
    str =~ /(.+)/
    sleep 0.1
    puts "got #{$1} from #{str}\n"
  end
end
proc = get_proc
t1 = Thread.new { proc.call "abc" }
t2 = Thread.new { proc.call "def" }
t1.join
t2.join

This outputs the following:

got abc from abc
got abc from def

The expected result is of course:

got abc from abc
got def from def

Clearly the variables are being scoped to the get_proc method and are being shared by both threads. This runs contrary to the documentation and also to expectations.

This behaviour should either be changed, or the documentation updated to reflect the actual behaviour.

Interestingly, the following does work as expected:

proc = proc do |str|
  str =~ /(.+)/
  sleep 0.1
  puts "got #{$1} from #{str}\n"
end
t1 = Thread.new { proc.call "abc" }
t2 = Thread.new { proc.call "def" }
t1.join
t2.join

[jeremyevans0 (Jeremy Evans)]

• Backport deleted (1.9.3: UNKNOWN, 2.0.0: UNKNOWN)

This behavior is still present in the master branch and it does seem like a bug to me. It does seems specific to returning a closure from a method and having two separate threads call that closure.

It doesn't occur if you pass the proc directly to Thread.new:

def get_proc
lambda do |str|
str =~ /(.+)/
sleep 0.1
puts "got #{$1} from #{str}\n"
end
end
proc = get_proc
t1 = Thread.new("abc", &proc)
t2 = Thread.new("def", &proc)
t1.join
t2.join

or if you switch to calling a method directly:

def get_proc str
str =~ /(.+)/
sleep 0.1
puts "got #{$1} from #{str}\n"
end
t1 = Thread.new { get_proc "abc" }
t2 = Thread.new { get_proc "def" }
t1.join
t2.join

From some debugging, one difference between the two cases in the original post is that different branches are taken in lep_svar, with the failing case taking the if and the passing case taking the else.

[jeremyevans0 (Jeremy Evans)]

• Related to Bug #12689: Thread isolation of $~ and $_ added

[ko1 (Koichi Sasada)]

• Assignee set to ko1 (Koichi Sasada)

[jeremyevans0 (Jeremy Evans)]

• Has duplicate Bug #14364: Regexp last match variable in procs added

[jeremyevans0 (Jeremy Evans)]

• Related to Bug #16448: regex global variables are working differently depending on scope added

RubyConf Hack Day: suggest to close

[jemmai (Jemma Issroff)]

• Status changed from Open to Closed

[jemmai (Jemma Issroff)]

• Status changed from Closed to Open

[nobu (Nobuyoshi Nakada)]

• Description updated (diff)

"Thread-local" in that sentence was wrong, and removed at 932dd9f10e684fa99b059054fbc934607d85b45a, but "method-local" is removed too.

[Dan0042 (Daniel DeLorme)]

nobu (Nobuyoshi Nakada) wrote in #note-9:

"Thread-local" in that sentence was wrong, and removed at 932dd9f10e684fa99b059054fbc934607d85b45a, but "method-local" is removed too.

Indeed, the documentation used to say "Regexp Global Variables" or "Special Global Variables" and this has been reduced to just "Global Variables" which I think is quite misleading. Sure they have the $ sigil, but they're not global at all. If anything, "Special Variables" would be more accurate.

[hsbt (Hiroshi SHIBATA)]

• Status changed from Open to Assigned

[jeremyevans0 (Jeremy Evans)]

• Related to Bug #20807: String#gsub fails when called from string subclass with a block passed added