Case insensitive FastRegexMatcher doesn't check all folds

[colega]

What did you do?

I ran the mimir-prometheus FuzzFastRegexMatcher_WithStaticallyDefinedRegularExpressions for a while, and found a bug.

The input string "ſſs" is not matched by the case-insensitive regular expression "(?i:(foo1|foo2|aaa|bbb|ccc|ddd|eee|fff|ggg|hhh|iii|lll|mmm|nnn|ooo|ppp|qqq|rrr|sss|ttt|uuu|vvv|www|xxx|yyy|zzz))",.

What did you expect to see?

I expected that string to be matched, as the golang's regexp package matches it.

What did you see instead? Under which circumstances?

It wasn't matched.

[colega]

A very quick diff that fixes the problem in the least performant way:

diff --git model/labels/regexp.go model/labels/regexp.go
index 79e340984..7681483e7 100644
--- model/labels/regexp.go
+++ model/labels/regexp.go
@@ -791,7 +791,15 @@ func (m *equalMultiStringMapMatcher) Matches(s string) bool {
        }

        _, ok := m.values[s]
-       return ok
+       if ok || m.caseSensitive {
+               return ok
+       }
+       for k := range m.values {
+               if strings.EqualFold(s, k) {
+                       return true
+               }
+       }
+       return false
 }

[colega]

I'd also consider just going through the standard regular expression engine if the matcher is case insensitive and the string contains unicode runes.

[colega]

The case-insensitive regular expression "(?i:(sss))" should match the string "ſſs" but it does not.

(Edit: this comment was a response to a deleted comment).

[kushalShukla-web]

you mean this matches function is incomplete right and it should have a check for a string to be matched .

[kushalShukla-web]

i think here we should use "regexp" package instead of "string" package .
here on this line it will always return false as its not checking for a certain pattern strings.EqualFold(s, k) .

[kushalShukla-web]

will create pr for this today .

[bboreham]

I observe it fails because ToLower("ſſs") is "ſſs".
However if we flip all the ToLowers to ToUpper, this case does work.

Maybe we can use unicode.SimpleFold? (which I found used inside Go's regexp package)

[Ranveer777]

@colega @bboreham
I'm interested in picking up this bug.