\d does not work anymore since 0.2.7

[tobias47n9e]

Up until 0.2.6 (?P<user_agent>[a-zA-Z\d-_]+/[0-9\.]+) was a valid regex. Since 0.2.7 \d does not work and only (?P<user_agent>[a-zA-Z0-9-_]+/[0-9\.]+) works. Is it possible that that is a regression because as far as I understand RE2 it should be valid.

The issue was raised here:
https://gitlab.com/tobias47n9e/wikibase_rs/issues/1

[BurntSushi]

Ug. That is definitely not an (intentionally, IIRC) valid regex, but it looks like the old parser accepted it anyway. The problem is that your syntax is attempting to write a character range, where the end of the range is _ but the beginning of the range is \d. Starting a range with a class itself isn't meaningful, so the parser today returns an error. In general, the - character is always interpreted as indicating a range unless it is escaped or unless it appears at the beginning or end of a character class. e.g., While [\d-_] is invalid, [\d_-] is valid and does what you want.

At the very least, the error message could be better here and should call out the fact that it's trying to interpret a range.

Note that it is also possible that the current behavior is actually undesirable and we might want to be a bit more flexible in allowing - to be treated as a literal - if either end of its range isn't actually a valid start or end of a range.

[BurntSushi]

@tobias47n9e Also, note that \d isn't actually [0-9]. It is much bigger than that, and accepts any codepoint in the \p{Digit} Unicode property.

[tobias47n9e]

Oh. That makes a lot of sense 😄 Not sure what most people except from the single dash (literal or range). Stricter interpreting with a more helpful error message does sound good to me.

About the 2nd comment: Not sure if I even need \d at the moment because I am just trying to restrict the types of user agents people use. I can always expand it when people complain 😄