Further performance optimization

search.go

@@ -12,6 +12,7 @@ import (
 	"sync"
 	"time"
 	"unicode"
+	"unicode/utf8"
 )
 
 // searchSectionCap bounds how many rows each results section renders. The
@@ -46,7 +47,7 @@ type searchPageData struct {
 	PageTotal    int
 	FileTotal    int
 	IndexBuiltAt time.Time
-	RenderMS     int64
+	renderTimer
 }
 
 // indexEntry is the shared scoreable core of both folder and file index
@@ -132,7 +133,7 @@ func (h *handler) handleSearch(w http.ResponseWriter, r *http.Request) {
 		IndexBuiltAt: builtAt,
 	}
 	w.Header().Set("Content-Type", "text/html; charset=utf-8")
-	data.RenderMS = elapsedMS(r)
+	data.renderTimer = renderTimer{requestStart(r)}
 	if err := searchTmpl.ExecuteTemplate(w, "layout", data); err != nil {
 		log.Printf("search template error: %v", err)
 	}
@@ -278,7 +279,7 @@ func scoreName(nameLower string, nameTokens []string, qLower string, qTokens []s
 				if best < 20 {
 					best = 20
 				}
-			case fuzzy && levenshtein(w, qt) <= 2:
+			case fuzzy && withinTwoEdits(w, qt):
 				if best < 5 {
 					best = 5
 				}
@@ -575,6 +576,17 @@ func tokenize(s string) []string {
 	return tokens
 }
 
+// withinTwoEdits reports whether a and b are at most 2 edits apart. Lengths
+// differing by more than 2 rule that out without running levenshtein, which
+// skips nearly every word pair in the per-query scoring loop.
+func withinTwoEdits(a, b string) bool {
+	d := utf8.RuneCountInString(a) - utf8.RuneCountInString(b)
+	if d < -2 || d > 2 {
+		return false
+	}
+	return levenshtein(a, b) <= 2
+}
+
 // levenshtein returns the edit distance between a and b. Operates on runes so
 // multi-byte characters count as one edit.
 func levenshtein(a, b string) int {