fix: float string to number parsing

cast.go

@@ -9,6 +9,7 @@ package cast
 import "time"
 
 const errorMsg = "unable to cast %#v of type %T to %T"
+const errorMsgWith = "unable to cast %#v of type %T to %T: %w"
 
 // Basic is a type parameter constraint for functions accepting basic types.
 //

number.go

@@ -9,6 +9,7 @@ import (
 	"encoding/json"
 	"errors"
 	"fmt"
+	"regexp"
 	"strconv"
 	"strings"
 	"time"
@@ -149,22 +150,22 @@ func toNumberE[T Number](i any, parseFn func(string) (T, error)) (T, error) {
 		}
 
 		v, err := parseFn(s)
-		if err == nil {
-			return v, nil
+		if err != nil {
+			return 0, fmt.Errorf(errorMsgWith, i, i, n, err)
 		}
 
-		return 0, fmt.Errorf(errorMsg, i, i, n)
+		return v, nil
 	case json.Number:
 		if s == "" {
 			return 0, nil
 		}
 
 		v, err := parseFn(string(s))
-		if err == nil {
-			return v, nil
+		if err != nil {
+			return 0, fmt.Errorf(errorMsgWith, i, i, n, err)
 		}
 
-		return 0, fmt.Errorf(errorMsg, i, i, n)
+		return v, nil
 	case float64EProvider:
 		if _, ok := any(n).(float64); !ok {
 			return 0, fmt.Errorf(errorMsg, i, i, n)
@@ -293,22 +294,22 @@ func toUnsignedNumberE[T Number](i any, parseFn func(string) (T, error)) (T, err
 		}
 
 		v, err := parseFn(s)
-		if err == nil {
-			return v, nil
+		if err != nil {
+			return 0, fmt.Errorf(errorMsgWith, i, i, n, err)
 		}
 
-		return 0, fmt.Errorf(errorMsg, i, i, n)
+		return v, nil
 	case json.Number:
 		if s == "" {
 			return 0, nil
 		}
 
 		v, err := parseFn(string(s))
-		if err == nil {
-			return v, nil
+		if err != nil {
+			return 0, fmt.Errorf(errorMsgWith, i, i, n, err)
 		}
 
-		return 0, fmt.Errorf(errorMsg, i, i, n)
+		return v, nil
 	case float64EProvider:
 		if _, ok := any(n).(float64); !ok {
 			return 0, fmt.Errorf(errorMsg, i, i, n)
@@ -413,7 +414,7 @@ func parseInt[T integer](s string) (T, error) {
 }
 
 func parseUint[T unsigned](s string) (T, error) {
-	v, err := strconv.ParseUint(trimDecimal(s), 0, 0)
+	v, err := strconv.ParseUint(strings.TrimLeft(trimDecimal(s), "+"), 0, 0)
 	if err != nil {
 		return 0, err
 	}
@@ -520,13 +521,28 @@ func trimZeroDecimal(s string) string {
 	return s
 }
 
-// trimming decimals seems significantly faster than parsing to float first
-//
-// see BenchmarkDecimal
+var stringNumberRe = regexp.MustCompile(`^([-+]?\d*)(\.\d*)?$`)
+
+// see [BenchmarkDecimal] for details about the implementation
 func trimDecimal(s string) string {
-	// trim the decimal part (if any)
-	if i := strings.Index(s, "."); i >= 0 {
-		s = s[:i]
+	if !strings.Contains(s, ".") {
+		return s
+	}
+
+	matches := stringNumberRe.FindStringSubmatch(s)
+	if matches != nil {
+		// matches[1] is the captured integer part with sign
+		s = matches[1]
+
+		// handle special cases
+		switch s {
+		case "-", "+":
+			s += "0"
+		case "":
+			s = "0"
+		}
+
+		return s
 	}
 
 	return s

number_internal_test.go

@@ -6,7 +6,9 @@
 package cast
 
 import (
+	"regexp"
 	"strconv"
+	"strings"
 	"testing"
 
 	qt "github.com/frankban/quicktest"
@@ -45,42 +47,136 @@ func TestTrimZeroDecimal(t *testing.T) {
 }
 
 func TestTrimDecimal(t *testing.T) {
-	c := qt.New(t)
+	testCases := []struct {
+		input    string
+		expected string
+	}{
+		{"10.0", "10"},
+		{"10.010", "10"},
+		{"00000.00001", "00000"},
+		{"-0001.0", "-0001"},
+		{".5", "0"},
+		{"+12.", "+12"},
+		{"+.25", "+0"},
+		{"-.25", "-0"},
+		{"0.0000000000", "0"},
+		{"0.0000000001", "0"},
+		{"10.0000000000", "10"},
+		{"10.0000000001", "10"},
+		{"10000000000000.0000000000", "10000000000000"},
+
+		{"10...17", "10...17"},
+		{"10.foobar", "10.foobar"},
+		{"10.0i", "10.0i"},
+		{"10.0E9", "10.0E9"},
+	}
 
-	c.Assert(trimDecimal("10.0"), qt.Equals, "10")
-	c.Assert(trimDecimal("10.00"), qt.Equals, "10")
-	c.Assert(trimDecimal("10.010"), qt.Equals, "10")
-	c.Assert(trimDecimal("0.0000000000"), qt.Equals, "0")
-	c.Assert(trimDecimal("0.00000000001"), qt.Equals, "0")
+	for _, testCase := range testCases {
+		// TODO: remove after minimum Go version is >=1.22
+		testCase := testCase
+
+		t.Run(testCase.input, func(t *testing.T) {
+			c := qt.New(t)
+
+			c.Assert(trimDecimal(testCase.input), qt.Equals, testCase.expected)
+		})
+	}
 }
 
+// Analysis (in the order of performance):
+//
+// - Trimming decimals based on decimal point yields a lot of incorrectly parsed values.
+// - Parsing to float might be better, but we still need to cast the number, it might overflow, problematic.
+// - Regex parsing is an order of magnitude slower, but it yields correct results.
 func BenchmarkDecimal(b *testing.B) {
-	testCases := []string{"10.0", "10.00", "10.010", "0.0000000000", "0.0000000001", "10.0000000000", "10.0000000001", "10000000000000.0000000000"}
+	testCases := []struct {
+		input       string
+		expectError bool
+	}{
+		{"10.0", false},
+		{"10.00", false},
+		{"10.010", false},
+		{"0.0000000000", false},
+		{"0.0000000001", false},
+		{"10.0000000000", false},
+		{"10.0000000001", false},
+		{"10000000000000.0000000000", false},
+
+		// {"10...17", true},
+		// {"10.foobar", true},
+		// {"10.0i", true},
+		// {"10.0E9", true},
+	}
+
+	trimDecimalString := func(s string) string {
+		// trim the decimal part (if any)
+		if i := strings.Index(s, "."); i >= 0 {
+			s = s[:i]
+		}
+
+		return s
+	}
+
+	re := regexp.MustCompile(`^([-+]?\d*)(\.\d*)?$`)
+
+	trimDecimalRegex := func(s string) string {
+		matches := re.FindStringSubmatch(s)
+		if matches != nil {
+			// matches[1] is the captured integer part with sign
+			return matches[1]
+		}
+
+		return s
+	}
 
 	for _, testCase := range testCases {
 		// TODO: remove after minimum Go version is >=1.22
 		testCase := testCase
 
-		b.Run(testCase, func(b *testing.B) {
+		b.Run(testCase.input, func(b *testing.B) {
 			b.Run("ParseFloat", func(b *testing.B) {
 				// TODO: use b.Loop() once updated to Go 1.24
 				for i := 0; i < b.N; i++ {
-					v, err := strconv.ParseFloat(testCase, 64)
-					if err != nil {
-						b.Fatal(err)
+					v, err := strconv.ParseFloat(testCase.input, 64)
+					if (err != nil) != testCase.expectError {
+						if err != nil {
+							b.Fatal(err)
+						}
+
+						b.Fatal("expected error, but got none")
 					}
 
 					n := int64(v)
 					_ = n
 				}
 			})
 
-			b.Run("TrimDecimal", func(b *testing.B) {
+			b.Run("TrimDecimalString", func(b *testing.B) {
+				// TODO: use b.Loop() once updated to Go 1.24
+				for i := 0; i < b.N; i++ {
+					v, err := strconv.ParseInt(trimDecimalString(testCase.input), 0, 0)
+					if (err != nil) != testCase.expectError {
+						if err != nil {
+							b.Fatal(err)
+						}
+
+						b.Fatal("expected error, but got none")
+					}
+
+					_ = v
+				}
+			})
+
+			b.Run("TrimDecimalRegex", func(b *testing.B) {
 				// TODO: use b.Loop() once updated to Go 1.24
 				for i := 0; i < b.N; i++ {
-					v, err := strconv.ParseInt(trimDecimal(testCase), 0, 0)
-					if err != nil {
-						b.Fatal(err)
+					v, err := strconv.ParseInt(trimDecimalRegex(testCase.input), 0, 0)
+					if (err != nil) != testCase.expectError {
+						if err != nil {
+							b.Fatal(err)
+						}
+
+						b.Fatal("expected error, but got none")
 					}
 
 					_ = v

number_test.go

@@ -150,6 +150,8 @@ var numberContexts = map[string]numberContext{
 	},
 }
 
+// TODO: separate test and failure cases?
+// Kinda hard to track cases right now.
 func generateNumberTestCases(samples []any) []testCase {
 	zero := samples[0]
 	one := samples[1]
@@ -169,7 +171,9 @@ func generateNumberTestCases(samples []any) []testCase {
 	_ = overflowString
 
 	kind := reflect.TypeOf(zero).Kind()
+	isSint := kind == reflect.Int || kind == reflect.Int8 || kind == reflect.Int16 || kind == reflect.Int32 || kind == reflect.Int64
 	isUint := kind == reflect.Uint || kind == reflect.Uint8 || kind == reflect.Uint16 || kind == reflect.Uint32 || kind == reflect.Uint64
+	isInt := isSint || isUint
 
 	// Some precision is lost when converting from float64 to float32.
 	eightPoint31_32 := eightPoint31
@@ -231,6 +235,27 @@ func generateNumberTestCases(samples []any) []testCase {
 		// Failure cases
 		{"test", zero, true},
 		{testing.T{}, zero, true},
+
+		{"10...17", zero, true},
+		{"10.foobar", zero, true},
+		{"10.0i", zero, true},
+	}
+
+	if isInt {
+		testCases = append(
+			testCases,
+
+			testCase{".5", zero, false},
+			testCase{"+8.", eight, false},
+			testCase{"+.25", zero, false},
+			testCase{"-.25", zero, isUint},
+
+			testCase{"10.0E9", zero, true},
+		)
+	} else if kind == reflect.Float32 {
+		testCases = append(testCases, testCase{"10.0E9", float32(10000000000.000000), false})
+	} else if kind == reflect.Float64 {
+		testCases = append(testCases, testCase{"10.0E9", float64(10000000000.000000), false})
 	}
 
 	if isUint && underflowString != nil {