Merge different languages for language stats (#24900) (#24921)

Backport #24900

Fix #24896

If users set different languages by `linguist-language`, the `stats` map
could be: `java: 100, Java: 200`.

Language stats are stored as case-insensitive in database and there is a
unique key.

So, the different language names should be merged to one unique name:
`Java: 300`

(cherry picked from commit a83d597989)

modules/git/repo_language_stats.go

@@ -3,7 +3,46 @@
 
 package git
 
+import (
+	"strings"
+	"unicode"
+)
+
 const (
 	fileSizeLimit int64 = 16 * 1024   // 16 KiB
 	bigFileSize   int64 = 1024 * 1024 // 1 MiB
 )
+
+// mergeLanguageStats mergers language names with different cases. The name with most upper case letters is used.
+func mergeLanguageStats(stats map[string]int64) map[string]int64 {
+	names := map[string]struct {
+		uniqueName string
+		upperCount int
+	}{}
+
+	countUpper := func(s string) (count int) {
+		for _, r := range s {
+			if unicode.IsUpper(r) {
+				count++
+			}
+		}
+		return count
+	}
+
+	for name := range stats {
+		cnt := countUpper(name)
+		lower := strings.ToLower(name)
+		if cnt >= names[lower].upperCount {
+			names[lower] = struct {
+				uniqueName string
+				upperCount int
+			}{uniqueName: name, upperCount: cnt}
+		}
+	}
+
+	res := make(map[string]int64, len(names))
+	for name, num := range stats {
+		res[names[strings.ToLower(name)].uniqueName] += num
+	}
+	return res
+}