<regex>: regex_traits<_Elem> uses an inadmissible value of type char_class_type to represent character class "w"

[muellerj2]

regex_traits<_Elem> uses static_cast<ctype_base::mask>(-1) to represent the character class "w":

STL/stl/inc/regex

Line 354 in 8124540

_REGEX_CHAR_CLASS_NAME("w", static_cast<ctype_base::mask>(-1)),

This is an inadmissible choice, because it violates [re.grammar]/9:

The results from multiple calls to traits_inst.lookup_classname can be bitwise or'ed together and subsequently passed to traits_inst.isctype.

Specifically, or'ing the char_class_type for "w" with the char_class_type for any other character class always produces the value for "w" again, even if the combination should match more characters.

Additional remarks

I think resolving this issue will break ABI. However, it should be possible to mitigate the problems caused by this issue.