conversion: optimize Scan() scientific scanning (#127)

elee1766 · holiman · web-flow · commit c385c612a93a · 2023-03-16T14:44:54.000-04:00
This change optimizes the Scan method, by making utilizing seek instead of string-split, adds some test and makes the function alloc-free. 

---------

Co-authored-by: Martin Holst Swende &lt;martin@swende.se&gt;
diff --git a/.gitignore b/.gitignore
@@ -1 +1,7 @@
 /.idea
+
+
+# go output files
+*.test
+*.out
+*.pb.gz
diff --git a/conversion.go b/conversion.go
@@ -615,33 +615,40 @@ func (dst *Int) Scan(src interface{}) error {
 	}
 	switch src := src.(type) {
 	case string:
-		splt := strings.SplitN(src, "e", 2)
-		if len(splt) == 1 {
-			return dst.SetFromDecimal(src)
-		}
-		if err := dst.SetFromDecimal(splt[0]); err != nil {
-			return err
-		}
-		if splt[1] == "0" {
-			return nil
-		}
-		exp := new(Int)
-		if err := exp.SetFromDecimal(splt[1]); err != nil {
-			return err
-		}
-		if !exp.IsUint64() || exp.Uint64() > uint64(len(twoPow256Sub1)) {
-			return ErrBig256Range
-		}
-		exp.Exp(NewInt(10), exp)
-		_, overflow := dst.MulOverflow(dst, exp)
-		if overflow {
-			return ErrBig256Range
-		}
-		return nil
+		return dst.scanScientificFromString(src)
 	case []byte:
-		return dst.SetFromDecimal(string(src))
+		return dst.scanScientificFromString(string(src))
+	}
+	return errors.New("unsupported type")
+}
+
+func (dst *Int) scanScientificFromString(src string) error {
+	if len(src) == 0 {
+		dst.Clear()
+		return nil
 	}
-	return fmt.Errorf("cannot scan %T", src)
+	idx := strings.IndexByte(src, 'e')
+	if idx == -1 {
+		return dst.SetFromDecimal(src)
+	}
+	if err := dst.SetFromDecimal(src[:idx]); err != nil {
+		return err
+	}
+	if src[(idx+1):] == "0" {
+		return nil
+	}
+	exp := new(Int)
+	if err := exp.SetFromDecimal(src[(idx + 1):]); err != nil {
+		return err
+	}
+	if exp.GtUint64(77) { // 10**78 is larger than 2**256
+		return ErrBig256Range
+	}
+	exp.Exp(NewInt(10), exp)
+	if _, overflow := dst.MulOverflow(dst, exp); overflow {
+		return ErrBig256Range
+	}
+	return nil
 }
 
 // Value implements the database/sql/driver Valuer interface.
diff --git a/conversion_test.go b/conversion_test.go
@@ -84,10 +84,34 @@ func TestScanScientific(t *testing.T) {
 		exp *Int
 		err string
 	}{
+		{
+			in:  "",
+			exp: new(Int),
+		},
+		{
+			in:  "e30",
+			err: "EOF",
+		},
+		{
+			in:  "30e",
+			err: "EOF",
+		},
+		{
+			in:  twoPow256Sub1 + "e",
+			err: "EOF",
+		},
 		{
 			in:  "14e30",
 			exp: new(Int).Mul(NewInt(14), new(Int).Exp(NewInt(10), NewInt(30))),
 		},
+		{ // 0xdd15fe86affad800000000000000000000000000000000000000000000000000
+			in:  "1e77",
+			exp: new(Int).Mul(NewInt(1), new(Int).Exp(NewInt(10), NewInt(77))),
+		},
+		{ // 0x8a2dbf142dfcc8000000000000000000000000000000000000000000000000000
+			in:  "1e78",
+			err: ErrBig256Range.Error(),
+		},
 		{
 			in:  "1455522523e31",
 			exp: new(Int).Mul(NewInt(1455522523), new(Int).Exp(NewInt(10), NewInt(31))),
@@ -169,6 +193,40 @@ func TestToBig(t *testing.T) {
 	}
 }
 
+func BenchmarkScanScientific(b *testing.B) {
+	intsub1 := new(Int)
+	_ = intsub1.fromDecimal(twoPow256Sub1)
+	cases := []struct {
+		in  string
+		exp *Int
+		err string
+	}{
+		{
+			in:  "14e30",
+			exp: new(Int).Mul(NewInt(14), new(Int).Exp(NewInt(10), NewInt(30))),
+		},
+		{
+			in:  "1455522523e31",
+			exp: new(Int).Mul(NewInt(1455522523), new(Int).Exp(NewInt(10), NewInt(31))),
+		},
+		{
+			in:  twoPow256Sub1 + "e0",
+			exp: intsub1,
+		},
+		{
+			in:  "1e00000000000000000",
+			exp: NewInt(1),
+		},
+	}
+	i := new(Int)
+	b.ResetTimer()
+	for idx := 0; idx < b.N; idx++ {
+		for _, v := range cases {
+			_ = i.Scan(v.in)
+		}
+	}
+}
+
 func benchSetFromBig(bench *testing.B, b *big.Int) Int {
 	var f Int
 	for i := 0; i < bench.N; i++ {

-Original file line number
+Diff line change
@@ @@ -1 +1,7 @@ @@
 /.idea
++
++
 +# go output files
 +*.test
 +*.out
 +*.pb.gz