Below, the second regexp seems just as guilty as the first to me.

Python 2.5.1 (r251:54869, Apr 18 2007, 22:08:04) 
[GCC 4.0.1 (Apple Computer, Inc. build 5367)] on darwin
Type "help", "copyright", "credits" or "license" for more information.
>>> import re
>>> re.compile(r'((x|y)*)*')
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File
"/Library/Frameworks/Python.framework/Versions/2.5/lib/python2.5/re.py",
line 180, in compile
    return _compile(pattern, flags)
  File
"/Library/Frameworks/Python.framework/Versions/2.5/lib/python2.5/re.py",
line 233, in _compile
    raise error, v # invalid expression
sre_constants.error: nothing to repeat
>>> re.compile(r'((x|y+)*)*')
<_sre.SRE_Pattern object at 0x18548>

I don't know if that error is to protect the sre engine from bad
patterns or just a courtesy to users.  If the former, it could be a
serious bug.

I'm almost tempted to call the first of these a bug:  isn't '((x|y)*)*'
a perfectly valid (albeit somewhat redundant) regular expression?  What
am I missing here?

Even if there are issues with capturing, shouldn't the version without
capturing subexpressions still work?

I get:

>>> re.compile(r'(?:(?:x|y)*)*')
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/usr/lib/python2.5/re.py", line 180, in compile
    return _compile(pattern, flags)
  File "/usr/lib/python2.5/re.py", line 233, in _compile
    raise error, v # invalid expression
sre_constants.error: nothing to repeat

Huh.  Maybe you're right.  JavaScript, Ruby, and Perl all accept both
regexes, although no two agree on what should be captured:

js> "xyyzy".replace(/((x|y)*)*/, "($1, $2)") 
(xyy, y)zy
js> "xyyzy".replace(/((x|y+)*)*/, "($1, $2)")
(xyy, yy)zy

>> "xyyzy".sub(/((x|y)*)*/, "(\\1, \\2)")
=> "(, y)zy"
>> "xyyzy".sub(/((x|y+)*)*/, "(\\1, \\2)")
=> "(, yy)zy"

  DB<1> $_ = 'xyyzy'; s/((x|y)*)*/(\1 \2)/; print
( )zy
  DB<2> $_ = 'xyyzy'; s/((x|y+)*)*/(\1 \2)/; print
( yy)zy

Ruby's behavior seems best to me.

> Ruby's behavior seems best to me.

We can obtain the Ruby behavior easily.  There is one check in sre_compile.py in the '_simple' function that needs to be removed (see attached patch).  Whether or not the Ruby behavior is the "correct" behavior I am still not sure.  In any case, I think throwing an exception is to aggressive for this case.

The re module is addressed in issue #2636.

BTW, my regex module behaves like Ruby:

>>> regex.sub(r"((x|y)*)*", "(\\1, \\2)", "xyyzy", count=1)
'(, y)zy'
>>> regex.sub(r"((x|y+)*)*", "(\\1, \\2)", "xyyzy", count=1)
'(, yy)zy'

> The re module is addressed in issue #2636.

Wow, that issue thread is massive...  What about the 're' module is addressed?  Is 'regex' replacing 're'?  Is 'regex' being rolled into 're'?  Are they both going to exist?

The issue started about updating the re module and adding features that other languages already possess in their regex implementations (the last time any significant work was done on it was in 2003).

The hope is that the new regex implementation will eventually replace the existing one, and putting it initially in a module called 'regex' allows it to be tested more easily.

You can do:

import regex as re

and existing code should still work.