why are re group names so restrictive?

[Skip Montanaro]

Why does re restrict the characters in group names to be Python identifiers?
They seem to only be used where strings are allowed, thus the character set
should only exclude ">" and ")" (and possibly "<" and ")" for symmetry).  If
I try to create a group name which doesn't look like a Python identifier, it
complains (as documented):

    >>> regex = re.compile("(?P<p+>.*)")
    Traceback (most recent call last):
      File "<stdin>", line 1, in ?
      File "/Users/skip/local/lib/python2.3/sre.py", line 179, in compile
        return _compile(pattern, flags)
      File "/Users/skip/local/lib/python2.3/sre.py", line 229, in _compile
        raise error, v # invalid expression
    sre_constants.error: bad character in group name
    >>> regex = re.compile("(?P<9p>.*)")
    Traceback (most recent call last):
      File "<stdin>", line 1, in ?
      File "/Users/skip/local/lib/python2.3/sre.py", line 179, in compile
        return _compile(pattern, flags)
      File "/Users/skip/local/lib/python2.3/sre.py", line 229, in _compile
        raise error, v # invalid expression
    sre_constants.error: bad character in group name
    >>> regex = re.compile("(?P<p9>.*)")

Are there some contexts where group names are used like Python identifiers
which force this restriction?  I could understand the restriction if groups
could be accessed as attributes of a match object, e.g.:

    >>> regex = re.compile("(?P<p9>.*)")
    >>> mat = regex.match("abc 123")
    >>> mat.group(1)
    'abc 123'
    >>> mat.group("p9")
    'abc 123'
    >>> mat.p9
    Traceback (most recent call last):
      File "<stdin>", line 1, in ?
    AttributeError: p9

but that isn't possible.

Thx,

Skip