more consistent line splitting

change the scanner to split lines according to expected
behavior, e.g: split lines on LF, CR or CRLF

remove need for extra function to guess that after the facts

Co-authored-by: Luis Davim <luis.davim@gmail.com>

gotenv.go

@@ -3,6 +3,7 @@ package gotenv
 
 import (
 	"bufio"
+	"bytes"
 	"fmt"
 	"io"
 	"os"
@@ -174,9 +175,38 @@ func Write(env Env, filename string) error {
 	return file.Sync()
 }
 
-func strictParse(r io.Reader, override bool) (Env, error) {
-	env := make(Env)
+// splitLines is a valid SplitFunc for a bufio.Scanner. It will split lines on CR ('\r'), LF ('\n') or CRLF (any of the three sequences).
+// If a CR is immediately followed by a LF, it is treated as a CRLF (one single line break).
+func splitLines(data []byte, atEOF bool) (advance int, token []byte, err error) {
+	if atEOF && len(data) == 0 {
+		return 0, nil, bufio.ErrFinalToken
+	}
+
+	idx := bytes.IndexAny(data, "\r\n")
+	switch {
+	case atEOF && idx < 0:
+		return len(data), data, bufio.ErrFinalToken
+
+	case idx < 0:
+		return 0, nil, nil
+	}
+
+	// consume CR or LF
+	eol := idx + 1
+	// detect CRLF
+	if len(data) > eol && data[eol-1] == '\r' && data[eol] == '\n' {
+		eol++
+	}
+
+	return eol, data[:idx], nil
+}
+
+func strictParse(r io.Reader, env Env, override bool) (Env, error) {
+	if env == nil {
+		env = make(Env)
+	}
 	scanner := bufio.NewScanner(r)
+	scanner.Split(splitLines)
 
 	firstLine := true
 
@@ -283,7 +313,6 @@ func parseLine(s string, env Env, override bool) error {
 			return varReplacement(s, hsq, env, override)
 		}
 		val = varRgx.ReplaceAllStringFunc(val, fv)
-		val = parseVal(val, env, hdq, override)
 	}
 
 	env[key] = val
@@ -352,18 +381,3 @@ func checkFormat(s string, env Env) error {
 
 	return fmt.Errorf("line `%s` doesn't match format", s)
 }
-
-func parseVal(val string, env Env, ignoreNewlines bool, override bool) string {
-	if strings.Contains(val, "=") && !ignoreNewlines {
-		kv := strings.Split(val, "\r")
-
-		if len(kv) > 1 {
-			val = kv[0]
-			for _, l := range kv[1:] {
-				_ = parseLine(l, env, override)
-			}
-		}
-	}
-
-	return val
-}

scanner_test.go

@@ -0,0 +1,78 @@
+package gotenv
+
+import (
+	"bufio"
+	"strings"
+	"testing"
+
+	"github.com/stretchr/testify/assert"
+)
+
+func TestScanner(t *testing.T) {
+
+	type testCase struct {
+		name string
+		in   string
+		exp  []string
+	}
+
+	testCases := []testCase{
+		{
+			"regular LF split with trailing LF",
+			"aa\nbb\ncc\n",
+			[]string{"aa", "bb", "cc", ""},
+		},
+		{
+			"regular LF split with no trailing LF",
+			"aa\nbb\ncc",
+			[]string{"aa", "bb", "cc"},
+		},
+
+		{
+			"regular CR split with trailing CR",
+			"aa\rbb\rcc\r",
+			[]string{"aa", "bb", "cc", ""},
+		},
+		{
+			"regular CR split with no trailing CR",
+			"aa\rbb\rcc",
+			[]string{"aa", "bb", "cc"},
+		},
+
+		{
+			"regular CRLF split with trailing CRLF",
+			"aa\r\nbb\r\ncc\r\n",
+			[]string{"aa", "bb", "cc", ""},
+		},
+		{
+			"regular CRLF split with no trailing CRLF",
+			"aa\r\nbb\r\ncc",
+			[]string{"aa", "bb", "cc"},
+		},
+
+		{
+			"mix of possible line endings",
+			"aa\r\nbb\ncc\rdd",
+			[]string{"aa", "bb", "cc", "dd"},
+		},
+	}
+
+	for _, tc := range testCases {
+		s := bufio.NewScanner(strings.NewReader(tc.in))
+		s.Split(splitLines)
+
+		i := 0
+		for s.Scan() {
+			if i >= len(tc.exp) {
+				assert.Fail(t, "unexpected line", "testCase: %s - got extra line: %q", tc.name, s.Text())
+			} else {
+				got := s.Text()
+				assert.Equal(t, tc.exp[i], got, "testCase: %s - line %d", tc.name, i)
+			}
+			i++
+		}
+
+		assert.NoError(t, s.Err(), "testCase: %s", tc.name)
+		assert.Equal(t, len(tc.exp), i, "testCase: %s - expected to have the correct line count", tc.name)
+	}
+}