sem: fix processing \0 escaped sequence in string literal

Signed-off-by: Alexander Bezzubov <[email protected]>

Author: bzz

driver/normalizer/strconv.go

@@ -1,6 +1,7 @@
 package normalizer
 
 import (
+	"fmt"
 	"strconv"
 	"unicode/utf8"
 )
@@ -18,7 +19,7 @@ func unquoteSingle(s string) (string, error) {
 	}
 	quote := s[0]
 	if quote != s[n-1] {
-		return "", strconv.ErrSyntax
+		return "", fmt.Errorf("string does not begin and end with a quote")
 	}
 	s = s[1 : len(s)-1]
 
@@ -33,6 +34,7 @@ func unquoteSingle(s string) (string, error) {
 			return s, nil
 		}
 	}
+	s = replaceEscapedMaybe(s, '0', '\x00')
 
 	var runeTmp [utf8.UTFMax]byte
 	buf := make([]byte, 0, 3*len(s)/2) // Try to avoid more allocations.
@@ -62,6 +64,42 @@ func contains(s string, c byte) bool {
 	return false
 }
 
+// replaceEscapedMaybe returns a copy of s with "\\old[^0-9]" replaced by new.
+func replaceEscapedMaybe(s string, old, new rune) string {
+	var runeTmp [utf8.UTFMax]byte
+	n := utf8.EncodeRune(runeTmp[:], new)
+
+	lastCp := 0
+	var buf []byte
+	for i, w := 0, 0; i < len(s); i += w {
+		r1, w1 := utf8.DecodeRuneInString(s[i:])
+		w = w1
+		if r1 == '\\' { // find sequence \\old[^0-9]
+			r2, w2 := utf8.DecodeRuneInString(s[i+w1:])
+			if r2 == old {
+				r3, _ := utf8.DecodeRuneInString(s[i+w1+w2:])
+				if 0 > r3 || r3 > 9 { // not a number after "\\old"
+					w += w2
+					if len(buf) == 0 {
+						buf = make([]byte, 0, 3*len(s)/2)
+					}
+					buf = append(buf, []byte(s[lastCp:i])...)
+					buf = append(buf, runeTmp[:n]...)
+					lastCp = i + w
+				}
+			}
+		}
+	}
+	if lastCp == 0 {
+		return s
+	}
+
+	if 0 < lastCp && lastCp < len(s) {
+		return string(append(buf, []byte(s[lastCp:len(s)])...))
+	}
+	return string(buf)
+}
+
 const lowerhex = "0123456789abcdef"
 
 // quoteSingle is the same as strconv.Quote, but uses ' as a quote.

driver/normalizer/strconv_test.go

@@ -1,6 +1,7 @@
 package normalizer
 
 import (
+	"regexp"
 	"testing"
 
 	"github.com/stretchr/testify/assert"
@@ -40,12 +41,25 @@ func TestUnquoteSingleAndQuoteBack(t *testing.T) {
 	}
 }
 
-func TestUnquoteSingle_StringAndQuoteBack(t *testing.T) {
-	const o = "'rand'"
+func BenchmarkReplacingNullEscape_Iterative(b *testing.B) {
+	b.ReportAllocs()
+	s := testCasesUnquote[3].in
+	for n := 0; n < b.N; n++ {
+		replaceEscapedMaybe(s, '0', '\x00')
+	}
+}
+
+func BenchmarkReplacingNullEscape_Regexp(b *testing.B) {
+	b.ReportAllocs()
+	s := testCasesUnquote[3].in
+	for n := 0; n < b.N; n++ {
+		replaceEscapedMaybeRegexp(s)
+	}
+}
 
-	s, err := unquoteSingle(o)
-	require.NoError(t, err)
-	q := quoteSingle(s)
+var re = regexp.MustCompile(`\\0([^0-9]|$)`)
 
-	require.Equal(t, o, q)
+// replaceEscapedMaybeRegexp is very simple, but slower alternative to normalizer.replaceEscapedMaybe
+func replaceEscapedMaybeRegexp(s string) string {
+	return re.ReplaceAllString(s, "\x00$1")
 }