The re module documentation does not do a good job of explaining exactly what \w matches. Quoting https://docs.python.org/3.5/library/re.html :

\w
For Unicode (str) patterns:
Matches Unicode word characters; this includes most characters
that can be part of a word in any language, as well as numbers
and the underscore.

Empirically, this appears to mean "everything in Unicode general categories L* and N*, plus U+005F (underscore)". That is a perfectly sensible definition and the documentation should state it in those terms. "Unicode word characters" could mean any number of different things; note for instance that UTS#18 gives a very different definition.

(Further reading: https://gist.github.com/zackw/3077f387591376c7bf67 plus links therefrom).

I would like to request also a clear explanation be given for the documentation in the 2.7 branch. From https://docs.python.org/2.7/library/re.html :

"\w ... If UNICODE is set, this will match the characters [0-9_] plus whatever is classified as alphanumeric in the Unicode character properties database"

This is ambiguous. Does it mean the "Alphabetic" property from UAX#44? Does it mean something else?

FWIW, the actual behavior of \w matching "everything in Unicode general categories L* and N*, plus U+005F (underscore)" is consistent across all versions I can conveniently test (2.7, 3.4, 3.5).

In 2.7, there are four characters in general category Nl that \w doesn't match, but I believe that is just a bug, not an intentional difference of behavior.

It's too late for the 2.7 docs, but the current docs can still be updated.

Would a change like this be accurate?

Matches Unicode word characters; this includes most alphanumeric characters as well as the underscore. In Unicode, alphanumeric characters are defined to be the general categories L + N (see https://unicode.org/reports/tr44/#General_Category_Values). If the :const:`ASCII` flag is used, only ``[a-zA-Z0-9_]`` is matched.