lib/fs: Add ASCII fastpath for normalization

lib/fs/folding.go

@@ -9,41 +9,45 @@ package fs
 import (
  "strings"
  "unicode"
- "unicode/utf8"
 
  "golang.org/x/text/unicode/norm"
 )
 
+type caseType int
+
+const (
+ asciiLower caseType = iota
+ asciiMixed
+ nonAscii
+)
+
 // UnicodeLowercaseNormalized returns the Unicode lower case variant of s,
 // having also normalized it to normalization form C.
 func UnicodeLowercaseNormalized(s string) string {
- i := firstCaseChange(s)
- if i == -1 {
- return norm.NFC.String(s)
+ switch checkCase(s) {
+ case asciiLower:
+ return s
+ case asciiMixed:
+ return strings.ToLower(s)
+ default:
+ return norm.NFC.String(strings.Map(toLower, s))
  }
+}
 
- var rs strings.Builder
- // WriteRune always reserves utf8.UTFMax bytes for non-ASCII runes,
- // even if it doesn't need all that space. Overallocate now to prevent
- // it from ever triggering a reallocation.
- rs.Grow(utf8.UTFMax - 1 + len(s))
- rs.WriteString(s[:i])
-
- for _, r := range s[i:] {
- rs.WriteRune(unicode.ToLower(unicode.ToUpper(r)))
- }
- return norm.NFC.String(rs.String())
+func toLower(r rune) rune {
+ return unicode.ToLower(unicode.ToUpper(r))
 }
 
-// Byte index of the first rune r s.t. lower(upper(r)) != r.
-func firstCaseChange(s string) int {
- for i, r := range s {
- if r <= unicode.MaxASCII && (r < 'A' || r > 'Z') {
- continue
+func checkCase(s string) caseType {
+ c := asciiLower
+ for i := 0; i < len(s); i++ {
+ b := s[i]
+ if b > unicode.MaxASCII {
+ return nonAscii
  }
- if unicode.ToLower(unicode.ToUpper(r)) != r {
- return i
+ if 'A' <= b && b <= 'Z' {
+ c = asciiMixed
  }
  }
- return -1
+ return c
 }