Decode: assigned empty struct to empty defined sections (#879)

Co-authored-by: Thomas Pelletier <thomas@pelletier.codes>

.github/workflows/cifuzz.yml

@@ -0,0 +1,26 @@
+name: CIFuzz
+on: [pull_request]
+jobs:
+  Fuzzing:
+    runs-on: ubuntu-latest
+    steps:
+    - name: Build Fuzzers
+      id: build
+      uses: google/oss-fuzz/infra/cifuzz/actions/build_fuzzers@master
+      with:
+        oss-fuzz-project-name: 'go-toml'
+        dry-run: false
+        language: go
+    - name: Run Fuzzers
+      uses: google/oss-fuzz/infra/cifuzz/actions/run_fuzzers@master
+      with:
+        oss-fuzz-project-name: 'go-toml'
+        fuzz-seconds: 300
+        dry-run: false
+        language: go
+    - name: Upload Crash
+      uses: actions/upload-artifact@v3
+      if: failure() && steps.build.outcome == 'success'
+      with:
+        name: artifacts
+        path: ./out/artifacts

.github/workflows/coverage.yml

@@ -13,8 +13,8 @@ jobs:
         with:
           fetch-depth: 0
       - name: Setup go
-        uses: actions/setup-go@master
+        uses: actions/setup-go@v4
         with:
-          go-version: 1.19
+          go-version: "1.20"
       - name: Run tests with coverage
         run: ./ci.sh coverage -d "${GITHUB_BASE_REF-HEAD}"

.github/workflows/release.yml

@@ -20,9 +20,9 @@ jobs:
         with:
           fetch-depth: 0
       - name: Set up Go
-        uses: actions/setup-go@v2
+        uses: actions/setup-go@v4
         with:
-          go-version: 1.19
+          go-version: "1.20"
       - name: Login to GitHub Container Registry
         uses: docker/login-action@v2
         with:

.github/workflows/workflow.yml

@@ -12,7 +12,7 @@ jobs:
     strategy:
       matrix:
         os: [ 'ubuntu-latest', 'windows-latest', 'macos-latest']
-        go: [ '1.18', '1.19' ]
+        go: [ '1.19', '1.20' ]
     runs-on: ${{ matrix.os }}
     name: ${{ matrix.go }}/${{ matrix.os }}
     steps:
@@ -20,7 +20,7 @@ jobs:
         with:
           fetch-depth: 0
       - name: Setup go ${{ matrix.go }}
-        uses: actions/setup-go@master
+        uses: actions/setup-go@v4
         with:
           go-version: ${{ matrix.go }}
       - name: Run unit tests

README.md

@@ -140,6 +140,17 @@ fmt.Println(string(b))
 
 [marshal]: https://pkg.go.dev/github.com/pelletier/go-toml/v2#Marshal
 
+## Unstable API
+
+This API does not yet follow the backward compatibility guarantees of this
+library. They provide early access to features that may have rough edges or an
+API subject to change.
+
+### Parser
+
+Parser is the unstable API that allows iterative parsing of a TOML document at
+the AST level. See https://pkg.go.dev/github.com/pelletier/go-toml/v2/unstable.
+
 ## Benchmarks
 
 Execution time speedup compared to other Go TOML libraries:
@@ -542,7 +553,7 @@ complete solutions exist out there.
 
 ## Versioning
 
-Go-toml follows [Semantic Versioning](http://semver.org/). The supported version
+Go-toml follows [Semantic Versioning](https://semver.org). The supported version
 of [TOML](https://github.com/toml-lang/toml) is indicated at the beginning of
 this document. The last two major versions of Go are supported
 (see [Go Release Policy](https://golang.org/doc/devel/release.html#policy)).

ci.sh

@@ -77,8 +77,9 @@ cover() {
 
     pushd "$dir"
     go test -covermode=atomic  -coverpkg=./... -coverprofile=coverage.out.tmp ./...
-    cat coverage.out.tmp | grep -v testsuite | grep -v tomltestgen | grep -v gotoml-test-decoder > coverage.out
+    cat coverage.out.tmp | grep -v fuzz | grep -v testsuite | grep -v tomltestgen | grep -v gotoml-test-decoder > coverage.out
     go tool cover -func=coverage.out
+    echo "Coverage profile for ${branch}: ${dir}/coverage.out" >&2
     popd
 
     if [ "${branch}" != "HEAD" ]; then

decode.go

@@ -5,6 +5,8 @@ import (
 	"math"
 	"strconv"
 	"time"
+
+	"github.com/pelletier/go-toml/v2/unstable"
 )
 
 func parseInteger(b []byte) (int64, error) {
@@ -32,7 +34,7 @@ func parseLocalDate(b []byte) (LocalDate, error) {
 	var date LocalDate
 
 	if len(b) != 10 || b[4] != '-' || b[7] != '-' {
-		return date, newDecodeError(b, "dates are expected to have the format YYYY-MM-DD")
+		return date, unstable.NewParserError(b, "dates are expected to have the format YYYY-MM-DD")
 	}
 
 	var err error
@@ -53,7 +55,7 @@ func parseLocalDate(b []byte) (LocalDate, error) {
 	}
 
 	if !isValidDate(date.Year, date.Month, date.Day) {
-		return LocalDate{}, newDecodeError(b, "impossible date")
+		return LocalDate{}, unstable.NewParserError(b, "impossible date")
 	}
 
 	return date, nil
@@ -64,7 +66,7 @@ func parseDecimalDigits(b []byte) (int, error) {
 
 	for i, c := range b {
 		if c < '0' || c > '9' {
-			return 0, newDecodeError(b[i:i+1], "expected digit (0-9)")
+			return 0, unstable.NewParserError(b[i:i+1], "expected digit (0-9)")
 		}
 		v *= 10
 		v += int(c - '0')
@@ -97,7 +99,7 @@ func parseDateTime(b []byte) (time.Time, error) {
 	} else {
 		const dateTimeByteLen = 6
 		if len(b) != dateTimeByteLen {
-			return time.Time{}, newDecodeError(b, "invalid date-time timezone")
+			return time.Time{}, unstable.NewParserError(b, "invalid date-time timezone")
 		}
 		var direction int
 		switch b[0] {
@@ -106,11 +108,11 @@ func parseDateTime(b []byte) (time.Time, error) {
 		case '+':
 			direction = +1
 		default:
-			return time.Time{}, newDecodeError(b[:1], "invalid timezone offset character")
+			return time.Time{}, unstable.NewParserError(b[:1], "invalid timezone offset character")
 		}
 
 		if b[3] != ':' {
-			return time.Time{}, newDecodeError(b[3:4], "expected a : separator")
+			return time.Time{}, unstable.NewParserError(b[3:4], "expected a : separator")
 		}
 
 		hours, err := parseDecimalDigits(b[1:3])
@@ -118,7 +120,7 @@ func parseDateTime(b []byte) (time.Time, error) {
 			return time.Time{}, err
 		}
 		if hours > 23 {
-			return time.Time{}, newDecodeError(b[:1], "invalid timezone offset hours")
+			return time.Time{}, unstable.NewParserError(b[:1], "invalid timezone offset hours")
 		}
 
 		minutes, err := parseDecimalDigits(b[4:6])
@@ -126,7 +128,7 @@ func parseDateTime(b []byte) (time.Time, error) {
 			return time.Time{}, err
 		}
 		if minutes > 59 {
-			return time.Time{}, newDecodeError(b[:1], "invalid timezone offset minutes")
+			return time.Time{}, unstable.NewParserError(b[:1], "invalid timezone offset minutes")
 		}
 
 		seconds := direction * (hours*3600 + minutes*60)
@@ -139,7 +141,7 @@ func parseDateTime(b []byte) (time.Time, error) {
 	}
 
 	if len(b) > 0 {
-		return time.Time{}, newDecodeError(b, "extra bytes at the end of the timezone")
+		return time.Time{}, unstable.NewParserError(b, "extra bytes at the end of the timezone")
 	}
 
 	t := time.Date(
@@ -160,7 +162,7 @@ func parseLocalDateTime(b []byte) (LocalDateTime, []byte, error) {
 
 	const localDateTimeByteMinLen = 11
 	if len(b) < localDateTimeByteMinLen {
-		return dt, nil, newDecodeError(b, "local datetimes are expected to have the format YYYY-MM-DDTHH:MM:SS[.NNNNNNNNN]")
+		return dt, nil, unstable.NewParserError(b, "local datetimes are expected to have the format YYYY-MM-DDTHH:MM:SS[.NNNNNNNNN]")
 	}
 
 	date, err := parseLocalDate(b[:10])
@@ -171,7 +173,7 @@ func parseLocalDateTime(b []byte) (LocalDateTime, []byte, error) {
 
 	sep := b[10]
 	if sep != 'T' && sep != ' ' && sep != 't' {
-		return dt, nil, newDecodeError(b[10:11], "datetime separator is expected to be T or a space")
+		return dt, nil, unstable.NewParserError(b[10:11], "datetime separator is expected to be T or a space")
 	}
 
 	t, rest, err := parseLocalTime(b[11:])
@@ -195,7 +197,7 @@ func parseLocalTime(b []byte) (LocalTime, []byte, error) {
 	// check if b matches to have expected format HH:MM:SS[.NNNNNN]
 	const localTimeByteLen = 8
 	if len(b) < localTimeByteLen {
-		return t, nil, newDecodeError(b, "times are expected to have the format HH:MM:SS[.NNNNNN]")
+		return t, nil, unstable.NewParserError(b, "times are expected to have the format HH:MM:SS[.NNNNNN]")
 	}
 
 	var err error
@@ -206,10 +208,10 @@ func parseLocalTime(b []byte) (LocalTime, []byte, error) {
 	}
 
 	if t.Hour > 23 {
-		return t, nil, newDecodeError(b[0:2], "hour cannot be greater 23")
+		return t, nil, unstable.NewParserError(b[0:2], "hour cannot be greater 23")
 	}
 	if b[2] != ':' {
-		return t, nil, newDecodeError(b[2:3], "expecting colon between hours and minutes")
+		return t, nil, unstable.NewParserError(b[2:3], "expecting colon between hours and minutes")
 	}
 
 	t.Minute, err = parseDecimalDigits(b[3:5])
@@ -217,10 +219,10 @@ func parseLocalTime(b []byte) (LocalTime, []byte, error) {
 		return t, nil, err
 	}
 	if t.Minute > 59 {
-		return t, nil, newDecodeError(b[3:5], "minutes cannot be greater 59")
+		return t, nil, unstable.NewParserError(b[3:5], "minutes cannot be greater 59")
 	}
 	if b[5] != ':' {
-		return t, nil, newDecodeError(b[5:6], "expecting colon between minutes and seconds")
+		return t, nil, unstable.NewParserError(b[5:6], "expecting colon between minutes and seconds")
 	}
 
 	t.Second, err = parseDecimalDigits(b[6:8])
@@ -229,7 +231,7 @@ func parseLocalTime(b []byte) (LocalTime, []byte, error) {
 	}
 
 	if t.Second > 60 {
-		return t, nil, newDecodeError(b[6:8], "seconds cannot be greater 60")
+		return t, nil, unstable.NewParserError(b[6:8], "seconds cannot be greater 60")
 	}
 
 	b = b[8:]
@@ -242,7 +244,7 @@ func parseLocalTime(b []byte) (LocalTime, []byte, error) {
 		for i, c := range b[1:] {
 			if !isDigit(c) {
 				if i == 0 {
-					return t, nil, newDecodeError(b[0:1], "need at least one digit after fraction point")
+					return t, nil, unstable.NewParserError(b[0:1], "need at least one digit after fraction point")
 				}
 				break
 			}
@@ -266,7 +268,7 @@ func parseLocalTime(b []byte) (LocalTime, []byte, error) {
 		}
 
 		if precision == 0 {
-			return t, nil, newDecodeError(b[:1], "nanoseconds need at least one digit")
+			return t, nil, unstable.NewParserError(b[:1], "nanoseconds need at least one digit")
 		}
 
 		t.Nanosecond = frac * nspow[precision]
@@ -289,24 +291,24 @@ func parseFloat(b []byte) (float64, error) {
 	}
 
 	if cleaned[0] == '.' {
-		return 0, newDecodeError(b, "float cannot start with a dot")
+		return 0, unstable.NewParserError(b, "float cannot start with a dot")
 	}
 
 	if cleaned[len(cleaned)-1] == '.' {
-		return 0, newDecodeError(b, "float cannot end with a dot")
+		return 0, unstable.NewParserError(b, "float cannot end with a dot")
 	}
 
 	dotAlreadySeen := false
 	for i, c := range cleaned {
 		if c == '.' {
 			if dotAlreadySeen {
-				return 0, newDecodeError(b[i:i+1], "float can have at most one decimal point")
+				return 0, unstable.NewParserError(b[i:i+1], "float can have at most one decimal point")
 			}
 			if !isDigit(cleaned[i-1]) {
-				return 0, newDecodeError(b[i-1:i+1], "float decimal point must be preceded by a digit")
+				return 0, unstable.NewParserError(b[i-1:i+1], "float decimal point must be preceded by a digit")
 			}
 			if !isDigit(cleaned[i+1]) {
-				return 0, newDecodeError(b[i:i+2], "float decimal point must be followed by a digit")
+				return 0, unstable.NewParserError(b[i:i+2], "float decimal point must be followed by a digit")
 			}
 			dotAlreadySeen = true
 		}
@@ -317,12 +319,12 @@ func parseFloat(b []byte) (float64, error) {
 		start = 1
 	}
 	if cleaned[start] == '0' && isDigit(cleaned[start+1]) {
-		return 0, newDecodeError(b, "float integer part cannot have leading zeroes")
+		return 0, unstable.NewParserError(b, "float integer part cannot have leading zeroes")
 	}
 
 	f, err := strconv.ParseFloat(string(cleaned), 64)
 	if err != nil {
-		return 0, newDecodeError(b, "unable to parse float: %w", err)
+		return 0, unstable.NewParserError(b, "unable to parse float: %w", err)
 	}
 
 	return f, nil
@@ -336,7 +338,7 @@ func parseIntHex(b []byte) (int64, error) {
 
 	i, err := strconv.ParseInt(string(cleaned), 16, 64)
 	if err != nil {
-		return 0, newDecodeError(b, "couldn't parse hexadecimal number: %w", err)
+		return 0, unstable.NewParserError(b, "couldn't parse hexadecimal number: %w", err)
 	}
 
 	return i, nil
@@ -350,7 +352,7 @@ func parseIntOct(b []byte) (int64, error) {
 
 	i, err := strconv.ParseInt(string(cleaned), 8, 64)
 	if err != nil {
-		return 0, newDecodeError(b, "couldn't parse octal number: %w", err)
+		return 0, unstable.NewParserError(b, "couldn't parse octal number: %w", err)
 	}
 
 	return i, nil
@@ -364,7 +366,7 @@ func parseIntBin(b []byte) (int64, error) {
 
 	i, err := strconv.ParseInt(string(cleaned), 2, 64)
 	if err != nil {
-		return 0, newDecodeError(b, "couldn't parse binary number: %w", err)
+		return 0, unstable.NewParserError(b, "couldn't parse binary number: %w", err)
 	}
 
 	return i, nil
@@ -387,12 +389,12 @@ func parseIntDec(b []byte) (int64, error) {
 	}
 
 	if len(cleaned) > startIdx+1 && cleaned[startIdx] == '0' {
-		return 0, newDecodeError(b, "leading zero not allowed on decimal number")
+		return 0, unstable.NewParserError(b, "leading zero not allowed on decimal number")
 	}
 
 	i, err := strconv.ParseInt(string(cleaned), 10, 64)
 	if err != nil {
-		return 0, newDecodeError(b, "couldn't parse decimal number: %w", err)
+		return 0, unstable.NewParserError(b, "couldn't parse decimal number: %w", err)
 	}
 
 	return i, nil
@@ -409,11 +411,11 @@ func checkAndRemoveUnderscoresIntegers(b []byte) ([]byte, error) {
 	}
 
 	if b[start] == '_' {
-		return nil, newDecodeError(b[start:start+1], "number cannot start with underscore")
+		return nil, unstable.NewParserError(b[start:start+1], "number cannot start with underscore")
 	}
 
 	if b[len(b)-1] == '_' {
-		return nil, newDecodeError(b[len(b)-1:], "number cannot end with underscore")
+		return nil, unstable.NewParserError(b[len(b)-1:], "number cannot end with underscore")
 	}
 
 	// fast path
@@ -435,7 +437,7 @@ func checkAndRemoveUnderscoresIntegers(b []byte) ([]byte, error) {
 		c := b[i]
 		if c == '_' {
 			if !before {
-				return nil, newDecodeError(b[i-1:i+1], "number must have at least one digit between underscores")
+				return nil, unstable.NewParserError(b[i-1:i+1], "number must have at least one digit between underscores")
 			}
 			before = false
 		} else {
@@ -449,11 +451,11 @@ func checkAndRemoveUnderscoresIntegers(b []byte) ([]byte, error) {
 
 func checkAndRemoveUnderscoresFloats(b []byte) ([]byte, error) {
 	if b[0] == '_' {
-		return nil, newDecodeError(b[0:1], "number cannot start with underscore")
+		return nil, unstable.NewParserError(b[0:1], "number cannot start with underscore")
 	}
 
 	if b[len(b)-1] == '_' {
-		return nil, newDecodeError(b[len(b)-1:], "number cannot end with underscore")
+		return nil, unstable.NewParserError(b[len(b)-1:], "number cannot end with underscore")
 	}
 
 	// fast path
@@ -476,10 +478,10 @@ func checkAndRemoveUnderscoresFloats(b []byte) ([]byte, error) {
 		switch c {
 		case '_':
 			if !before {
-				return nil, newDecodeError(b[i-1:i+1], "number must have at least one digit between underscores")
+				return nil, unstable.NewParserError(b[i-1:i+1], "number must have at least one digit between underscores")
 			}
 			if i < len(b)-1 && (b[i+1] == 'e' || b[i+1] == 'E') {
-				return nil, newDecodeError(b[i+1:i+2], "cannot have underscore before exponent")
+				return nil, unstable.NewParserError(b[i+1:i+2], "cannot have underscore before exponent")
 			}
 			before = false
 		case '+', '-':
@@ -488,15 +490,15 @@ func checkAndRemoveUnderscoresFloats(b []byte) ([]byte, error) {
 			before = false
 		case 'e', 'E':
 			if i < len(b)-1 && b[i+1] == '_' {
-				return nil, newDecodeError(b[i+1:i+2], "cannot have underscore after exponent")
+				return nil, unstable.NewParserError(b[i+1:i+2], "cannot have underscore after exponent")
 			}
 			cleaned = append(cleaned, c)
 		case '.':
 			if i < len(b)-1 && b[i+1] == '_' {
-				return nil, newDecodeError(b[i+1:i+2], "cannot have underscore after decimal point")
+				return nil, unstable.NewParserError(b[i+1:i+2], "cannot have underscore after decimal point")
 			}
 			if i > 0 && b[i-1] == '_' {
-				return nil, newDecodeError(b[i-1:i], "cannot have underscore before decimal point")
+				return nil, unstable.NewParserError(b[i-1:i], "cannot have underscore before decimal point")
 			}
 			cleaned = append(cleaned, c)
 		default:
@@ -542,3 +544,7 @@ func daysIn(m int, year int) int {
 func isLeap(year int) bool {
 	return year%4 == 0 && (year%100 != 0 || year%400 == 0)
 }
+
+func isDigit(r byte) bool {
+	return r >= '0' && r <= '9'
+}

errors.go

@@ -6,6 +6,7 @@ import (
 	"strings"
 
 	"github.com/pelletier/go-toml/v2/internal/danger"
+	"github.com/pelletier/go-toml/v2/unstable"
 )
 
 // DecodeError represents an error encountered during the parsing or decoding
@@ -55,25 +56,6 @@ func (s *StrictMissingError) String() string {
 
 type Key []string
 
-// internal version of DecodeError that is used as the base to create a
-// DecodeError with full context.
-type decodeError struct {
-	highlight []byte
-	message   string
-	key       Key // optional
-}
-
-func (de *decodeError) Error() string {
-	return de.message
-}
-
-func newDecodeError(highlight []byte, format string, args ...interface{}) error {
-	return &decodeError{
-		highlight: highlight,
-		message:   fmt.Errorf(format, args...).Error(),
-	}
-}
-
 // Error returns the error message contained in the DecodeError.
 func (e *DecodeError) Error() string {
 	return "toml: " + e.message
@@ -105,12 +87,12 @@ func (e *DecodeError) Key() Key {
 // highlight can be freely deallocated.
 //
 //nolint:funlen
-func wrapDecodeError(document []byte, de *decodeError) *DecodeError {
-	offset := danger.SubsliceOffset(document, de.highlight)
+func wrapDecodeError(document []byte, de *unstable.ParserError) *DecodeError {
+	offset := danger.SubsliceOffset(document, de.Highlight)
 
 	errMessage := de.Error()
 	errLine, errColumn := positionAtEnd(document[:offset])
-	before, after := linesOfContext(document, de.highlight, offset, 3)
+	before, after := linesOfContext(document, de.Highlight, offset, 3)
 
 	var buf strings.Builder
 
@@ -140,7 +122,7 @@ func wrapDecodeError(document []byte, de *decodeError) *DecodeError {
 		buf.Write(before[0])
 	}
 
-	buf.Write(de.highlight)
+	buf.Write(de.Highlight)
 
 	if len(after) > 0 {
 		buf.Write(after[0])
@@ -158,7 +140,7 @@ func wrapDecodeError(document []byte, de *decodeError) *DecodeError {
 		buf.WriteString(strings.Repeat(" ", len(before[0])))
 	}
 
-	buf.WriteString(strings.Repeat("~", len(de.highlight)))
+	buf.WriteString(strings.Repeat("~", len(de.Highlight)))
 
 	if len(errMessage) > 0 {
 		buf.WriteString(" ")
@@ -183,7 +165,7 @@ func wrapDecodeError(document []byte, de *decodeError) *DecodeError {
 		message: errMessage,
 		line:    errLine,
 		column:  errColumn,
-		key:     de.key,
+		key:     de.Key,
 		human:   buf.String(),
 	}
 }

errors_test.go

@@ -7,6 +7,7 @@ import (
 	"strings"
 	"testing"
 
+	"github.com/pelletier/go-toml/v2/unstable"
 	"github.com/stretchr/testify/assert"
 )
 
@@ -170,9 +171,9 @@ line 5`,
 			doc := b.Bytes()
 			hl := doc[start:end]
 
-			err := wrapDecodeError(doc, &decodeError{
-				highlight: hl,
-				message:   e.msg,
+			err := wrapDecodeError(doc, &unstable.ParserError{
+				Highlight: hl,
+				Message:   e.msg,
 			})
 
 			var derr *DecodeError

example_text_marshaling_test.go

@@ -0,0 +1,37 @@
+package toml_test
+
+import (
+	"fmt"
+	"log"
+	"strconv"
+
+	"github.com/pelletier/go-toml/v2"
+)
+
+type customInt int
+
+func (i *customInt) UnmarshalText(b []byte) error {
+	x, err := strconv.ParseInt(string(b), 10, 32)
+	if err != nil {
+		return err
+	}
+	*i = customInt(x * 100)
+	return nil
+}
+
+type doc struct {
+	Value customInt
+}
+
+func ExampleUnmarshal_textUnmarshal() {
+	var x doc
+
+	data := []byte(`value  = "42"`)
+	err := toml.Unmarshal(data, &x)
+	if err != nil {
+		log.Fatal(err)
+	}
+	fmt.Println(x)
+	// Output:
+	// {4200}
+}

fast_test.go

@@ -7,13 +7,20 @@ import (
 	"github.com/stretchr/testify/require"
 )
 
-func TestFastSimple(t *testing.T) {
+func TestFastSimpleInt(t *testing.T) {
 	m := map[string]int64{}
 	err := toml.Unmarshal([]byte(`a = 42`), &m)
 	require.NoError(t, err)
 	require.Equal(t, map[string]int64{"a": 42}, m)
 }
 
+func TestFastSimpleFloat(t *testing.T) {
+	m := map[string]float64{}
+	err := toml.Unmarshal([]byte("a = 42\nb = 1.1\nc = 12341234123412341234123412341234"), &m)
+	require.NoError(t, err)
+	require.Equal(t, map[string]float64{"a": 42, "b": 1.1, "c": 1.2341234123412342e+31}, m)
+}
+
 func TestFastSimpleString(t *testing.T) {
 	m := map[string]string{}
 	err := toml.Unmarshal([]byte(`a = "hello"`), &m)

fuzz_test.go

@@ -1,5 +1,5 @@
-//go:build go1.18 || go1.19
-// +build go1.18 go1.19
+//go:build go1.18 || go1.19 || go1.20
+// +build go1.18 go1.19 go1.20
 
 package toml_test
 

go.mod

@@ -2,4 +2,4 @@ module github.com/pelletier/go-toml/v2
 
 go 1.16
 
-require github.com/stretchr/testify v1.8.0
+require github.com/stretchr/testify v1.8.4

go.sum

@@ -5,9 +5,11 @@ github.com/pmezard/go-difflib v1.0.0 h1:4DBwDE0NGyQoBHbLQYPwSUPoCMWR5BEzIk/f1lZb
 github.com/pmezard/go-difflib v1.0.0/go.mod h1:iKH77koFhYxTK1pcRnkKkqfTogsbg7gZNVY4sRDYZ/4=
 github.com/stretchr/objx v0.1.0/go.mod h1:HFkY916IF+rwdDfMAkV7OtwuqBVzrE8GR6GFx+wExME=
 github.com/stretchr/objx v0.4.0/go.mod h1:YvHI0jy2hoMjB+UWwv71VJQ9isScKT/TqJzVSSt89Yw=
+github.com/stretchr/objx v0.5.0/go.mod h1:Yh+to48EsGEfYuaHDzXPcE3xhTkx73EhmCGUpEOglKo=
 github.com/stretchr/testify v1.7.1/go.mod h1:6Fq8oRcR53rry900zMqJjRRixrwX3KX962/h/Wwjteg=
-github.com/stretchr/testify v1.8.0 h1:pSgiaMZlXftHpm5L7V1+rVB+AZJydKsMxsQBIJw4PKk=
 github.com/stretchr/testify v1.8.0/go.mod h1:yNjHg4UonilssWZ8iaSj1OCr/vHnekPRkoO+kdMU+MU=
+github.com/stretchr/testify v1.8.4 h1:CcVxjf3Q8PM0mHUKJCdn+eZZtm5yQwehR5yeSVQQcUk=
+github.com/stretchr/testify v1.8.4/go.mod h1:sz/lmYIOXD/1dqDmKjjqLyZ2RngseejIcXlSw2iwfAo=
 gopkg.in/check.v1 v0.0.0-20161208181325-20d25e280405 h1:yhCVgyC4o1eVCa2tZl7eS0r+SDo693bJlVdllGtEeKM=
 gopkg.in/check.v1 v0.0.0-20161208181325-20d25e280405/go.mod h1:Co6ibVJAznAaIkqp8huTwlJQCZ016jof/cbN4VW5Yz0=
 gopkg.in/yaml.v3 v3.0.0-20200313102051-9f266ea9e77c/go.mod h1:K4uyk7z7BCEPqu6E+C64Yfv1cQ7kz7rIZviUmN+EgEM=

internal/ast/builder.go

@@ -1,51 +0,0 @@
-package ast
-
-type Reference int
-
-const InvalidReference Reference = -1
-
-func (r Reference) Valid() bool {
-	return r != InvalidReference
-}
-
-type Builder struct {
-	tree    Root
-	lastIdx int
-}
-
-func (b *Builder) Tree() *Root {
-	return &b.tree
-}
-
-func (b *Builder) NodeAt(ref Reference) *Node {
-	return b.tree.at(ref)
-}
-
-func (b *Builder) Reset() {
-	b.tree.nodes = b.tree.nodes[:0]
-	b.lastIdx = 0
-}
-
-func (b *Builder) Push(n Node) Reference {
-	b.lastIdx = len(b.tree.nodes)
-	b.tree.nodes = append(b.tree.nodes, n)
-	return Reference(b.lastIdx)
-}
-
-func (b *Builder) PushAndChain(n Node) Reference {
-	newIdx := len(b.tree.nodes)
-	b.tree.nodes = append(b.tree.nodes, n)
-	if b.lastIdx >= 0 {
-		b.tree.nodes[b.lastIdx].next = newIdx - b.lastIdx
-	}
-	b.lastIdx = newIdx
-	return Reference(b.lastIdx)
-}
-
-func (b *Builder) AttachChild(parent Reference, child Reference) {
-	b.tree.nodes[parent].child = int(child) - int(parent)
-}
-
-func (b *Builder) Chain(from Reference, to Reference) {
-	b.tree.nodes[from].next = int(to) - int(from)
-}

internal/characters/ascii.go

@@ -0,0 +1,42 @@
+package characters
+
+var invalidAsciiTable = [256]bool{
+	0x00: true,
+	0x01: true,
+	0x02: true,
+	0x03: true,
+	0x04: true,
+	0x05: true,
+	0x06: true,
+	0x07: true,
+	0x08: true,
+	// 0x09 TAB
+	// 0x0A LF
+	0x0B: true,
+	0x0C: true,
+	// 0x0D CR
+	0x0E: true,
+	0x0F: true,
+	0x10: true,
+	0x11: true,
+	0x12: true,
+	0x13: true,
+	0x14: true,
+	0x15: true,
+	0x16: true,
+	0x17: true,
+	0x18: true,
+	0x19: true,
+	0x1A: true,
+	0x1B: true,
+	0x1C: true,
+	0x1D: true,
+	0x1E: true,
+	0x1F: true,
+	// 0x20 - 0x7E Printable ASCII characters
+	0x7F: true,
+}
+
+func InvalidAscii(b byte) bool {
+	return invalidAsciiTable[b]
+}

internal/characters/utf8.go

@@ -1,4 +1,4 @@
-package toml
+package characters
 
 import (
 	"unicode/utf8"
@@ -32,7 +32,7 @@ func (u utf8Err) Zero() bool {
 // 0x9 => tab, ok
 // 0xA - 0x1F => invalid
 // 0x7F => invalid
-func utf8TomlValidAlreadyEscaped(p []byte) (err utf8Err) {
+func Utf8TomlValidAlreadyEscaped(p []byte) (err utf8Err) {
 	// Fast path. Check for and skip 8 bytes of ASCII characters per iteration.
 	offset := 0
 	for len(p) >= 8 {
@@ -48,7 +48,7 @@ func utf8TomlValidAlreadyEscaped(p []byte) (err utf8Err) {
 		}
 
 		for i, b := range p[:8] {
-			if invalidAscii(b) {
+			if InvalidAscii(b) {
 				err.Index = offset + i
 				err.Size = 1
 				return
@@ -62,7 +62,7 @@ func utf8TomlValidAlreadyEscaped(p []byte) (err utf8Err) {
 	for i := 0; i < n; {
 		pi := p[i]
 		if pi < utf8.RuneSelf {
-			if invalidAscii(pi) {
+			if InvalidAscii(pi) {
 				err.Index = offset + i
 				err.Size = 1
 				return
@@ -106,11 +106,11 @@ func utf8TomlValidAlreadyEscaped(p []byte) (err utf8Err) {
 }
 
 // Return the size of the next rune if valid, 0 otherwise.
-func utf8ValidNext(p []byte) int {
+func Utf8ValidNext(p []byte) int {
 	c := p[0]
 
 	if c < utf8.RuneSelf {
-		if invalidAscii(c) {
+		if InvalidAscii(c) {
 			return 0
 		}
 		return 1
@@ -140,47 +140,6 @@ func utf8ValidNext(p []byte) int {
 	return size
 }
 
-var invalidAsciiTable = [256]bool{
-	0x00: true,
-	0x01: true,
-	0x02: true,
-	0x03: true,
-	0x04: true,
-	0x05: true,
-	0x06: true,
-	0x07: true,
-	0x08: true,
-	// 0x09 TAB
-	// 0x0A LF
-	0x0B: true,
-	0x0C: true,
-	// 0x0D CR
-	0x0E: true,
-	0x0F: true,
-	0x10: true,
-	0x11: true,
-	0x12: true,
-	0x13: true,
-	0x14: true,
-	0x15: true,
-	0x16: true,
-	0x17: true,
-	0x18: true,
-	0x19: true,
-	0x1A: true,
-	0x1B: true,
-	0x1C: true,
-	0x1D: true,
-	0x1E: true,
-	0x1F: true,
-	// 0x20 - 0x7E Printable ASCII characters
-	0x7F: true,
-}
-
-func invalidAscii(b byte) bool {
-	return invalidAsciiTable[b]
-}
-
 // acceptRange gives the range of valid values for the second byte in a UTF-8
 // sequence.
 type acceptRange struct {

internal/cli/cli.go

@@ -18,7 +18,7 @@ type Program struct {
 	Usage string
 	Fn    ConvertFn
 	// Inplace allows the command to take more than one file as argument and
-	// perform convertion in place on each provided file.
+	// perform conversion in place on each provided file.
 	Inplace bool
 }
 

internal/imported_tests/unmarshal_imported_test.go

@@ -1085,10 +1085,6 @@ func TestUnmarshalCheckConversionFloatInt(t *testing.T) {
 			desc:  "int",
 			input: `I = 1e300`,
 		},
-		{
-			desc:  "float",
-			input: `F = 9223372036854775806`,
-		},
 	}
 
 	for _, test := range testCases {

internal/tracker/key.go

@@ -1,8 +1,6 @@
 package tracker
 
-import (
-	"github.com/pelletier/go-toml/v2/internal/ast"
-)
+import "github.com/pelletier/go-toml/v2/unstable"
 
 // KeyTracker is a tracker that keeps track of the current Key as the AST is
 // walked.
@@ -11,19 +9,19 @@ type KeyTracker struct {
 }
 
 // UpdateTable sets the state of the tracker with the AST table node.
-func (t *KeyTracker) UpdateTable(node *ast.Node) {
+func (t *KeyTracker) UpdateTable(node *unstable.Node) {
 	t.reset()
 	t.Push(node)
 }
 
 // UpdateArrayTable sets the state of the tracker with the AST array table node.
-func (t *KeyTracker) UpdateArrayTable(node *ast.Node) {
+func (t *KeyTracker) UpdateArrayTable(node *unstable.Node) {
 	t.reset()
 	t.Push(node)
 }
 
 // Push the given key on the stack.
-func (t *KeyTracker) Push(node *ast.Node) {
+func (t *KeyTracker) Push(node *unstable.Node) {
 	it := node.Key()
 	for it.Next() {
 		t.k = append(t.k, string(it.Node().Data))
@@ -31,7 +29,7 @@ func (t *KeyTracker) Push(node *ast.Node) {
 }
 
 // Pop key from stack.
-func (t *KeyTracker) Pop(node *ast.Node) {
+func (t *KeyTracker) Pop(node *unstable.Node) {
 	it := node.Key()
 	for it.Next() {
 		t.k = t.k[:len(t.k)-1]

internal/tracker/seen.go

@@ -5,7 +5,7 @@ import (
 	"fmt"
 	"sync"
 
-	"github.com/pelletier/go-toml/v2/internal/ast"
+	"github.com/pelletier/go-toml/v2/unstable"
 )
 
 type keyKind uint8
@@ -150,23 +150,23 @@ func (s *SeenTracker) setExplicitFlag(parentIdx int) {
 // CheckExpression takes a top-level node and checks that it does not contain
 // keys that have been seen in previous calls, and validates that types are
 // consistent.
-func (s *SeenTracker) CheckExpression(node *ast.Node) error {
+func (s *SeenTracker) CheckExpression(node *unstable.Node) error {
 	if s.entries == nil {
 		s.reset()
 	}
 	switch node.Kind {
-	case ast.KeyValue:
+	case unstable.KeyValue:
 		return s.checkKeyValue(node)
-	case ast.Table:
+	case unstable.Table:
 		return s.checkTable(node)
-	case ast.ArrayTable:
+	case unstable.ArrayTable:
 		return s.checkArrayTable(node)
 	default:
 		panic(fmt.Errorf("this should not be a top level node type: %s", node.Kind))
 	}
 }
 
-func (s *SeenTracker) checkTable(node *ast.Node) error {
+func (s *SeenTracker) checkTable(node *unstable.Node) error {
 	if s.currentIdx >= 0 {
 		s.setExplicitFlag(s.currentIdx)
 	}
@@ -219,7 +219,7 @@ func (s *SeenTracker) checkTable(node *ast.Node) error {
 	return nil
 }
 
-func (s *SeenTracker) checkArrayTable(node *ast.Node) error {
+func (s *SeenTracker) checkArrayTable(node *unstable.Node) error {
 	if s.currentIdx >= 0 {
 		s.setExplicitFlag(s.currentIdx)
 	}
@@ -267,7 +267,7 @@ func (s *SeenTracker) checkArrayTable(node *ast.Node) error {
 	return nil
 }
 
-func (s *SeenTracker) checkKeyValue(node *ast.Node) error {
+func (s *SeenTracker) checkKeyValue(node *unstable.Node) error {
 	parentIdx := s.currentIdx
 	it := node.Key()
 
@@ -297,26 +297,26 @@ func (s *SeenTracker) checkKeyValue(node *ast.Node) error {
 	value := node.Value()
 
 	switch value.Kind {
-	case ast.InlineTable:
+	case unstable.InlineTable:
 		return s.checkInlineTable(value)
-	case ast.Array:
+	case unstable.Array:
 		return s.checkArray(value)
 	}
 
 	return nil
 }
 
-func (s *SeenTracker) checkArray(node *ast.Node) error {
+func (s *SeenTracker) checkArray(node *unstable.Node) error {
 	it := node.Children()
 	for it.Next() {
 		n := it.Node()
 		switch n.Kind {
-		case ast.InlineTable:
+		case unstable.InlineTable:
 			err := s.checkInlineTable(n)
 			if err != nil {
 				return err
 			}
-		case ast.Array:
+		case unstable.Array:
 			err := s.checkArray(n)
 			if err != nil {
 				return err
@@ -326,7 +326,7 @@ func (s *SeenTracker) checkArray(node *ast.Node) error {
 	return nil
 }
 
-func (s *SeenTracker) checkInlineTable(node *ast.Node) error {
+func (s *SeenTracker) checkInlineTable(node *unstable.Node) error {
 	if pool.New == nil {
 		pool.New = func() interface{} {
 			return &SeenTracker{}

localtime.go

@@ -4,6 +4,8 @@ import (
 	"fmt"
 	"strings"
 	"time"
+
+	"github.com/pelletier/go-toml/v2/unstable"
 )
 
 // LocalDate represents a calendar day in no specific timezone.
@@ -75,7 +77,7 @@ func (d LocalTime) MarshalText() ([]byte, error) {
 func (d *LocalTime) UnmarshalText(b []byte) error {
 	res, left, err := parseLocalTime(b)
 	if err == nil && len(left) != 0 {
-		err = newDecodeError(left, "extra characters")
+		err = unstable.NewParserError(left, "extra characters")
 	}
 	if err != nil {
 		return err
@@ -109,7 +111,7 @@ func (d LocalDateTime) MarshalText() ([]byte, error) {
 func (d *LocalDateTime) UnmarshalText(data []byte) error {
 	res, left, err := parseLocalDateTime(data)
 	if err == nil && len(left) != 0 {
-		err = newDecodeError(left, "extra characters")
+		err = unstable.NewParserError(left, "extra characters")
 	}
 	if err != nil {
 		return err

marshaler.go

@@ -12,6 +12,8 @@ import (
 	"strings"
 	"time"
 	"unicode"
+
+	"github.com/pelletier/go-toml/v2/internal/characters"
 )
 
 // Marshal serializes a Go value as a TOML document.
@@ -271,7 +273,7 @@ func (enc *Encoder) encode(b []byte, ctx encoderCtx, v reflect.Value) ([]byte, e
 		return enc.encodeMap(b, ctx, v)
 	case reflect.Struct:
 		return enc.encodeStruct(b, ctx, v)
-	case reflect.Slice:
+	case reflect.Slice, reflect.Array:
 		return enc.encodeSlice(b, ctx, v)
 	case reflect.Interface:
 		if v.IsNil() {
@@ -355,9 +357,9 @@ func (enc *Encoder) encodeKv(b []byte, ctx encoderCtx, options valueOptions, v r
 
 	if !ctx.inline {
 		b = enc.encodeComment(ctx.indent, options.comment, b)
+		b = enc.indent(ctx.indent, b)
 	}
 
-	b = enc.indent(ctx.indent, b)
 	b = enc.encodeKey(b, ctx.key)
 	b = append(b, " = "...)
 
@@ -437,7 +439,7 @@ func (enc *Encoder) encodeString(b []byte, v string, options valueOptions) []byt
 func needsQuoting(v string) bool {
 	// TODO: vectorize
 	for _, b := range []byte(v) {
-		if b == '\'' || b == '\r' || b == '\n' || invalidAscii(b) {
+		if b == '\'' || b == '\r' || b == '\n' || characters.InvalidAscii(b) {
 			return true
 		}
 	}
@@ -575,11 +577,23 @@ func (enc *Encoder) encodeKey(b []byte, k string) []byte {
 	}
 }
 
-func (enc *Encoder) encodeMap(b []byte, ctx encoderCtx, v reflect.Value) ([]byte, error) {
-	if v.Type().Key().Kind() != reflect.String {
-		return nil, fmt.Errorf("toml: type %s is not supported as a map key", v.Type().Key().Kind())
-	}
+func (enc *Encoder) keyToString(k reflect.Value) (string, error) {
+	keyType := k.Type()
+	switch {
+	case keyType.Kind() == reflect.String:
+		return k.String(), nil
 
+	case keyType.Implements(textMarshalerType):
+		keyB, err := k.Interface().(encoding.TextMarshaler).MarshalText()
+		if err != nil {
+			return "", fmt.Errorf("toml: error marshalling key %v from text: %w", k, err)
+		}
+		return string(keyB), nil
+	}
+	return "", fmt.Errorf("toml: type %s is not supported as a map key", keyType.Kind())
+}
+
+func (enc *Encoder) encodeMap(b []byte, ctx encoderCtx, v reflect.Value) ([]byte, error) {
 	var (
 		t                 table
 		emptyValueOptions valueOptions
@@ -587,13 +601,17 @@ func (enc *Encoder) encodeMap(b []byte, ctx encoderCtx, v reflect.Value) ([]byte
 
 	iter := v.MapRange()
 	for iter.Next() {
-		k := iter.Key().String()
 		v := iter.Value()
 
 		if isNil(v) {
 			continue
 		}
 
+		k, err := enc.keyToString(iter.Key())
+		if err != nil {
+			return nil, err
+		}
+
 		if willConvertToTableOrArrayTable(ctx, v) {
 			t.pushTable(k, v, emptyValueOptions)
 		} else {
@@ -912,7 +930,7 @@ func willConvertToTableOrArrayTable(ctx encoderCtx, v reflect.Value) bool {
 		return willConvertToTableOrArrayTable(ctx, v.Elem())
 	}
 
-	if t.Kind() == reflect.Slice {
+	if t.Kind() == reflect.Slice || t.Kind() == reflect.Array {
 		if v.Len() == 0 {
 			// An empty slice should be a kv = [].
 			return false

marshaler_test.go

@@ -15,6 +15,21 @@ import (
 	"github.com/stretchr/testify/require"
 )
 
+type marshalTextKey struct {
+	A string
+	B string
+}
+
+func (k marshalTextKey) MarshalText() ([]byte, error) {
+	return []byte(k.A + "-" + k.B), nil
+}
+
+type marshalBadTextKey struct{}
+
+func (k marshalBadTextKey) MarshalText() ([]byte, error) {
+	return nil, fmt.Errorf("error")
+}
+
 func TestMarshal(t *testing.T) {
 	someInt := 42
 
@@ -97,6 +112,53 @@ also = 'that'
 a = 'test'
 `,
 		},
+		{
+			desc: `map with text key`,
+			v: map[marshalTextKey]string{
+				{A: "a", B: "1"}: "value 1",
+				{A: "a", B: "2"}: "value 2",
+				{A: "b", B: "1"}: "value 3",
+			},
+			expected: `a-1 = 'value 1'
+a-2 = 'value 2'
+b-1 = 'value 3'
+`,
+		},
+		{
+			desc: `table with text key`,
+			v: map[marshalTextKey]map[string]string{
+				{A: "a", B: "1"}: {"value": "foo"},
+			},
+			expected: `[a-1]
+value = 'foo'
+`,
+		},
+		{
+			desc: `map with ptr text key`,
+			v: map[*marshalTextKey]string{
+				{A: "a", B: "1"}: "value 1",
+				{A: "a", B: "2"}: "value 2",
+				{A: "b", B: "1"}: "value 3",
+			},
+			expected: `a-1 = 'value 1'
+a-2 = 'value 2'
+b-1 = 'value 3'
+`,
+		},
+		{
+			desc: `map with bad text key`,
+			v: map[marshalBadTextKey]string{
+				{}: "value 1",
+			},
+			err: true,
+		},
+		{
+			desc: `map with bad ptr text key`,
+			v: map[*marshalBadTextKey]string{
+				{}: "value 1",
+			},
+			err: true,
+		},
 		{
 			desc: "simple string array",
 			v: map[string][]string{
@@ -144,6 +206,45 @@ a = 'test'
 
 [[top]]
 'map2.1' = 'v2.1'
+`,
+		},
+		{
+			desc: "fixed size string array",
+			v: map[string][3]string{
+				"array": {"one", "two", "three"},
+			},
+			expected: `array = ['one', 'two', 'three']
+`,
+		},
+		{
+			desc: "fixed size nested string arrays",
+			v: map[string][2][2]string{
+				"array": {{"one", "two"}, {"three"}},
+			},
+			expected: `array = [['one', 'two'], ['three', '']]
+`,
+		},
+		{
+			desc: "mixed strings and fixed size nested string arrays",
+			v: map[string][]interface{}{
+				"array": {"a string", [2]string{"one", "two"}, "last"},
+			},
+			expected: `array = ['a string', ['one', 'two'], 'last']
+`,
+		},
+		{
+			desc: "fixed size array of maps",
+			v: map[string][2]map[string]string{
+				"ftop": {
+					{"map1.1": "v1.1"},
+					{"map2.1": "v2.1"},
+				},
+			},
+			expected: `[[ftop]]
+'map1.1' = 'v1.1'
+
+[[ftop]]
+'map2.1' = 'v2.1'
 `,
 		},
 		{
@@ -487,9 +588,14 @@ foo = 42
 		},
 		{
 			desc: "invalid map key",
-			v:    map[int]interface{}{},
+			v:    map[int]interface{}{1: "a"},
 			err:  true,
 		},
+		{
+			desc:     "invalid map key but empty",
+			v:        map[int]interface{}{},
+			expected: "",
+		},
 		{
 			desc: "unhandled type",
 			v: struct {
@@ -1190,6 +1296,27 @@ func TestMarshalUint64Overflow(t *testing.T) {
 	require.Error(t, err)
 }
 
+func TestIndentWithInlineTable(t *testing.T) {
+	x := map[string][]map[string]string{
+		"one": []map[string]string{
+			{"0": "0"},
+			{"1": "1"},
+		},
+	}
+	expected := `one = [
+  {0 = '0'},
+  {1 = '1'}
+]
+`
+	var buf bytes.Buffer
+	enc := toml.NewEncoder(&buf)
+	enc.SetIndentTables(true)
+	enc.SetTablesInline(true)
+	enc.SetArraysMultiline(true)
+	require.NoError(t, enc.Encode(x))
+	assert.Equal(t, expected, buf.String())
+}
+
 func ExampleMarshal() {
 	type MyConfig struct {
 		Version int

ossfuzz/fuzz.go

@@ -0,0 +1,45 @@
+//go:build go1.18 || go1.19 || go1.20
+// +build go1.18 go1.19 go1.20
+
+package ossfuzz
+
+import (
+	"fmt"
+	"reflect"
+	"strings"
+
+	"github.com/pelletier/go-toml/v2"
+)
+
+func FuzzToml(data []byte) int {
+	if len(data) >= 2048 {
+		return 0
+	}
+
+	if strings.Contains(string(data), "nan") {
+		return 0
+	}
+
+	var v interface{}
+	err := toml.Unmarshal(data, &v)
+	if err != nil {
+		return 0
+	}
+
+	encoded, err := toml.Marshal(v)
+	if err != nil {
+		panic(fmt.Sprintf("failed to marshal unmarshaled document: %s", err))
+	}
+
+	var v2 interface{}
+	err = toml.Unmarshal(encoded, &v2)
+	if err != nil {
+		panic(fmt.Sprintf("failed round trip: %s", err))
+	}
+
+	if !reflect.DeepEqual(v, v2) {
+		panic(fmt.Sprintf("not equal: %#+v %#+v", v, v2))
+	}
+
+	return 1
+}

parser_test.go

@@ -1,450 +0,0 @@
-package toml
-
-import (
-	"strconv"
-	"strings"
-	"testing"
-
-	"github.com/pelletier/go-toml/v2/internal/ast"
-	"github.com/stretchr/testify/require"
-)
-
-//nolint:funlen
-func TestParser_AST_Numbers(t *testing.T) {
-	examples := []struct {
-		desc  string
-		input string
-		kind  ast.Kind
-		err   bool
-	}{
-		{
-			desc:  "integer just digits",
-			input: `1234`,
-			kind:  ast.Integer,
-		},
-		{
-			desc:  "integer zero",
-			input: `0`,
-			kind:  ast.Integer,
-		},
-		{
-			desc:  "integer sign",
-			input: `+99`,
-			kind:  ast.Integer,
-		},
-		{
-			desc:  "integer hex uppercase",
-			input: `0xDEADBEEF`,
-			kind:  ast.Integer,
-		},
-		{
-			desc:  "integer hex lowercase",
-			input: `0xdead_beef`,
-			kind:  ast.Integer,
-		},
-		{
-			desc:  "integer octal",
-			input: `0o01234567`,
-			kind:  ast.Integer,
-		},
-		{
-			desc:  "integer binary",
-			input: `0b11010110`,
-			kind:  ast.Integer,
-		},
-		{
-			desc:  "float zero",
-			input: `0.0`,
-			kind:  ast.Float,
-		},
-		{
-			desc:  "float positive zero",
-			input: `+0.0`,
-			kind:  ast.Float,
-		},
-		{
-			desc:  "float negative zero",
-			input: `-0.0`,
-			kind:  ast.Float,
-		},
-		{
-			desc:  "float pi",
-			input: `3.1415`,
-			kind:  ast.Float,
-		},
-		{
-			desc:  "float negative",
-			input: `-0.01`,
-			kind:  ast.Float,
-		},
-		{
-			desc:  "float signed exponent",
-			input: `5e+22`,
-			kind:  ast.Float,
-		},
-		{
-			desc:  "float exponent lowercase",
-			input: `1e06`,
-			kind:  ast.Float,
-		},
-		{
-			desc:  "float exponent uppercase",
-			input: `-2E-2`,
-			kind:  ast.Float,
-		},
-		{
-			desc:  "float fractional with exponent",
-			input: `6.626e-34`,
-			kind:  ast.Float,
-		},
-		{
-			desc:  "float underscores",
-			input: `224_617.445_991_228`,
-			kind:  ast.Float,
-		},
-		{
-			desc:  "inf",
-			input: `inf`,
-			kind:  ast.Float,
-		},
-		{
-			desc:  "inf negative",
-			input: `-inf`,
-			kind:  ast.Float,
-		},
-		{
-			desc:  "inf positive",
-			input: `+inf`,
-			kind:  ast.Float,
-		},
-		{
-			desc:  "nan",
-			input: `nan`,
-			kind:  ast.Float,
-		},
-		{
-			desc:  "nan negative",
-			input: `-nan`,
-			kind:  ast.Float,
-		},
-		{
-			desc:  "nan positive",
-			input: `+nan`,
-			kind:  ast.Float,
-		},
-	}
-
-	for _, e := range examples {
-		e := e
-		t.Run(e.desc, func(t *testing.T) {
-			p := parser{}
-			p.Reset([]byte(`A = ` + e.input))
-			p.NextExpression()
-			err := p.Error()
-			if e.err {
-				require.Error(t, err)
-			} else {
-				require.NoError(t, err)
-
-				expected := astNode{
-					Kind: ast.KeyValue,
-					Children: []astNode{
-						{Kind: e.kind, Data: []byte(e.input)},
-						{Kind: ast.Key, Data: []byte(`A`)},
-					},
-				}
-				compareNode(t, expected, p.Expression())
-			}
-		})
-	}
-}
-
-type (
-	astNode struct {
-		Kind     ast.Kind
-		Data     []byte
-		Children []astNode
-	}
-)
-
-func compareNode(t *testing.T, e astNode, n *ast.Node) {
-	t.Helper()
-	require.Equal(t, e.Kind, n.Kind)
-	require.Equal(t, e.Data, n.Data)
-
-	compareIterator(t, e.Children, n.Children())
-}
-
-func compareIterator(t *testing.T, expected []astNode, actual ast.Iterator) {
-	t.Helper()
-	idx := 0
-
-	for actual.Next() {
-		n := actual.Node()
-
-		if idx >= len(expected) {
-			t.Fatal("extra child in actual tree")
-		}
-		e := expected[idx]
-
-		compareNode(t, e, n)
-
-		idx++
-	}
-
-	if idx < len(expected) {
-		t.Fatal("missing children in actual", "idx =", idx, "expected =", len(expected))
-	}
-}
-
-//nolint:funlen
-func TestParser_AST(t *testing.T) {
-	examples := []struct {
-		desc  string
-		input string
-		ast   astNode
-		err   bool
-	}{
-		{
-			desc:  "simple string assignment",
-			input: `A = "hello"`,
-			ast: astNode{
-				Kind: ast.KeyValue,
-				Children: []astNode{
-					{
-						Kind: ast.String,
-						Data: []byte(`hello`),
-					},
-					{
-						Kind: ast.Key,
-						Data: []byte(`A`),
-					},
-				},
-			},
-		},
-		{
-			desc:  "simple bool assignment",
-			input: `A = true`,
-			ast: astNode{
-				Kind: ast.KeyValue,
-				Children: []astNode{
-					{
-						Kind: ast.Bool,
-						Data: []byte(`true`),
-					},
-					{
-						Kind: ast.Key,
-						Data: []byte(`A`),
-					},
-				},
-			},
-		},
-		{
-			desc:  "array of strings",
-			input: `A = ["hello", ["world", "again"]]`,
-			ast: astNode{
-				Kind: ast.KeyValue,
-				Children: []astNode{
-					{
-						Kind: ast.Array,
-						Children: []astNode{
-							{
-								Kind: ast.String,
-								Data: []byte(`hello`),
-							},
-							{
-								Kind: ast.Array,
-								Children: []astNode{
-									{
-										Kind: ast.String,
-										Data: []byte(`world`),
-									},
-									{
-										Kind: ast.String,
-										Data: []byte(`again`),
-									},
-								},
-							},
-						},
-					},
-					{
-						Kind: ast.Key,
-						Data: []byte(`A`),
-					},
-				},
-			},
-		},
-		{
-			desc:  "array of arrays of strings",
-			input: `A = ["hello", "world"]`,
-			ast: astNode{
-				Kind: ast.KeyValue,
-				Children: []astNode{
-					{
-						Kind: ast.Array,
-						Children: []astNode{
-							{
-								Kind: ast.String,
-								Data: []byte(`hello`),
-							},
-							{
-								Kind: ast.String,
-								Data: []byte(`world`),
-							},
-						},
-					},
-					{
-						Kind: ast.Key,
-						Data: []byte(`A`),
-					},
-				},
-			},
-		},
-		{
-			desc:  "inline table",
-			input: `name = { first = "Tom", last = "Preston-Werner" }`,
-			ast: astNode{
-				Kind: ast.KeyValue,
-				Children: []astNode{
-					{
-						Kind: ast.InlineTable,
-						Children: []astNode{
-							{
-								Kind: ast.KeyValue,
-								Children: []astNode{
-									{Kind: ast.String, Data: []byte(`Tom`)},
-									{Kind: ast.Key, Data: []byte(`first`)},
-								},
-							},
-							{
-								Kind: ast.KeyValue,
-								Children: []astNode{
-									{Kind: ast.String, Data: []byte(`Preston-Werner`)},
-									{Kind: ast.Key, Data: []byte(`last`)},
-								},
-							},
-						},
-					},
-					{
-						Kind: ast.Key,
-						Data: []byte(`name`),
-					},
-				},
-			},
-		},
-	}
-
-	for _, e := range examples {
-		e := e
-		t.Run(e.desc, func(t *testing.T) {
-			p := parser{}
-			p.Reset([]byte(e.input))
-			p.NextExpression()
-			err := p.Error()
-			if e.err {
-				require.Error(t, err)
-			} else {
-				require.NoError(t, err)
-				compareNode(t, e.ast, p.Expression())
-			}
-		})
-	}
-}
-
-func BenchmarkParseBasicStringWithUnicode(b *testing.B) {
-	p := &parser{}
-	b.Run("4", func(b *testing.B) {
-		input := []byte(`"\u1234\u5678\u9ABC\u1234\u5678\u9ABC"`)
-		b.ReportAllocs()
-		b.SetBytes(int64(len(input)))
-
-		for i := 0; i < b.N; i++ {
-			p.parseBasicString(input)
-		}
-	})
-	b.Run("8", func(b *testing.B) {
-		input := []byte(`"\u12345678\u9ABCDEF0\u12345678\u9ABCDEF0"`)
-		b.ReportAllocs()
-		b.SetBytes(int64(len(input)))
-
-		for i := 0; i < b.N; i++ {
-			p.parseBasicString(input)
-		}
-	})
-}
-
-func BenchmarkParseBasicStringsEasy(b *testing.B) {
-	p := &parser{}
-
-	for _, size := range []int{1, 4, 8, 16, 21} {
-		b.Run(strconv.Itoa(size), func(b *testing.B) {
-			input := []byte(`"` + strings.Repeat("A", size) + `"`)
-
-			b.ReportAllocs()
-			b.SetBytes(int64(len(input)))
-
-			for i := 0; i < b.N; i++ {
-				p.parseBasicString(input)
-			}
-		})
-	}
-}
-
-func TestParser_AST_DateTimes(t *testing.T) {
-	examples := []struct {
-		desc  string
-		input string
-		kind  ast.Kind
-		err   bool
-	}{
-		{
-			desc:  "offset-date-time with delim 'T' and UTC offset",
-			input: `2021-07-21T12:08:05Z`,
-			kind:  ast.DateTime,
-		},
-		{
-			desc:  "offset-date-time with space delim and +8hours offset",
-			input: `2021-07-21 12:08:05+08:00`,
-			kind:  ast.DateTime,
-		},
-		{
-			desc:  "local-date-time with nano second",
-			input: `2021-07-21T12:08:05.666666666`,
-			kind:  ast.LocalDateTime,
-		},
-		{
-			desc:  "local-date-time",
-			input: `2021-07-21T12:08:05`,
-			kind:  ast.LocalDateTime,
-		},
-		{
-			desc:  "local-date",
-			input: `2021-07-21`,
-			kind:  ast.LocalDate,
-		},
-	}
-
-	for _, e := range examples {
-		e := e
-		t.Run(e.desc, func(t *testing.T) {
-			p := parser{}
-			p.Reset([]byte(`A = ` + e.input))
-			p.NextExpression()
-			err := p.Error()
-			if e.err {
-				require.Error(t, err)
-			} else {
-				require.NoError(t, err)
-
-				expected := astNode{
-					Kind: ast.KeyValue,
-					Children: []astNode{
-						{Kind: e.kind, Data: []byte(e.input)},
-						{Kind: ast.Key, Data: []byte(`A`)},
-					},
-				}
-				compareNode(t, expected, p.Expression())
-			}
-		})
-	}
-}

strict.go

@@ -1,9 +1,9 @@
 package toml
 
 import (
-	"github.com/pelletier/go-toml/v2/internal/ast"
 	"github.com/pelletier/go-toml/v2/internal/danger"
 	"github.com/pelletier/go-toml/v2/internal/tracker"
+	"github.com/pelletier/go-toml/v2/unstable"
 )
 
 type strict struct {
@@ -12,10 +12,10 @@ type strict struct {
 	// Tracks the current key being processed.
 	key tracker.KeyTracker
 
-	missing []decodeError
+	missing []unstable.ParserError
 }
 
-func (s *strict) EnterTable(node *ast.Node) {
+func (s *strict) EnterTable(node *unstable.Node) {
 	if !s.Enabled {
 		return
 	}
@@ -23,7 +23,7 @@ func (s *strict) EnterTable(node *ast.Node) {
 	s.key.UpdateTable(node)
 }
 
-func (s *strict) EnterArrayTable(node *ast.Node) {
+func (s *strict) EnterArrayTable(node *unstable.Node) {
 	if !s.Enabled {
 		return
 	}
@@ -31,7 +31,7 @@ func (s *strict) EnterArrayTable(node *ast.Node) {
 	s.key.UpdateArrayTable(node)
 }
 
-func (s *strict) EnterKeyValue(node *ast.Node) {
+func (s *strict) EnterKeyValue(node *unstable.Node) {
 	if !s.Enabled {
 		return
 	}
@@ -39,7 +39,7 @@ func (s *strict) EnterKeyValue(node *ast.Node) {
 	s.key.Push(node)
 }
 
-func (s *strict) ExitKeyValue(node *ast.Node) {
+func (s *strict) ExitKeyValue(node *unstable.Node) {
 	if !s.Enabled {
 		return
 	}
@@ -47,27 +47,27 @@ func (s *strict) ExitKeyValue(node *ast.Node) {
 	s.key.Pop(node)
 }
 
-func (s *strict) MissingTable(node *ast.Node) {
+func (s *strict) MissingTable(node *unstable.Node) {
 	if !s.Enabled {
 		return
 	}
 
-	s.missing = append(s.missing, decodeError{
-		highlight: keyLocation(node),
-		message:   "missing table",
-		key:       s.key.Key(),
+	s.missing = append(s.missing, unstable.ParserError{
+		Highlight: keyLocation(node),
+		Message:   "missing table",
+		Key:       s.key.Key(),
 	})
 }
 
-func (s *strict) MissingField(node *ast.Node) {
+func (s *strict) MissingField(node *unstable.Node) {
 	if !s.Enabled {
 		return
 	}
 
-	s.missing = append(s.missing, decodeError{
-		highlight: keyLocation(node),
-		message:   "missing field",
-		key:       s.key.Key(),
+	s.missing = append(s.missing, unstable.ParserError{
+		Highlight: keyLocation(node),
+		Message:   "missing field",
+		Key:       s.key.Key(),
 	})
 }
 
@@ -88,7 +88,7 @@ func (s *strict) Error(doc []byte) error {
 	return err
 }
 
-func keyLocation(node *ast.Node) []byte {
+func keyLocation(node *unstable.Node) []byte {
 	k := node.Key()
 
 	hasOne := k.Next()

types.go

@@ -6,9 +6,9 @@ import (
 	"time"
 )
 
-var timeType = reflect.TypeOf(time.Time{})
-var textMarshalerType = reflect.TypeOf(new(encoding.TextMarshaler)).Elem()
-var textUnmarshalerType = reflect.TypeOf(new(encoding.TextUnmarshaler)).Elem()
-var mapStringInterfaceType = reflect.TypeOf(map[string]interface{}{})
-var sliceInterfaceType = reflect.TypeOf([]interface{}{})
+var timeType = reflect.TypeOf((*time.Time)(nil)).Elem()
+var textMarshalerType = reflect.TypeOf((*encoding.TextMarshaler)(nil)).Elem()
+var textUnmarshalerType = reflect.TypeOf((*encoding.TextUnmarshaler)(nil)).Elem()
+var mapStringInterfaceType = reflect.TypeOf(map[string]interface{}(nil))
+var sliceInterfaceType = reflect.TypeOf([]interface{}(nil))
 var stringType = reflect.TypeOf("")

unmarshaler.go

@@ -12,16 +12,16 @@ import (
 	"sync/atomic"
 	"time"
 
-	"github.com/pelletier/go-toml/v2/internal/ast"
 	"github.com/pelletier/go-toml/v2/internal/danger"
 	"github.com/pelletier/go-toml/v2/internal/tracker"
+	"github.com/pelletier/go-toml/v2/unstable"
 )
 
 // Unmarshal deserializes a TOML document into a Go value.
 //
 // It is a shortcut for Decoder.Decode() with the default options.
 func Unmarshal(data []byte, v interface{}) error {
-	p := parser{}
+	p := unstable.Parser{}
 	p.Reset(data)
 	d := decoder{p: &p}
 
@@ -60,7 +60,7 @@ func (d *Decoder) DisallowUnknownFields() *Decoder {
 // are ignored. See Decoder.DisallowUnknownFields() to change this behavior.
 //
 // When a TOML local date, time, or date-time is decoded into a time.Time, its
-// value is represented in time.Local timezone. Otherwise the approriate Local*
+// value is represented in time.Local timezone. Otherwise the appropriate Local*
 // structure is used. For time values, precision up to the nanosecond is
 // supported by truncating extra digits.
 //
@@ -101,7 +101,7 @@ func (d *Decoder) Decode(v interface{}) error {
 		return fmt.Errorf("toml: %w", err)
 	}
 
-	p := parser{}
+	p := unstable.Parser{}
 	p.Reset(b)
 	dec := decoder{
 		p: &p,
@@ -115,7 +115,7 @@ func (d *Decoder) Decode(v interface{}) error {
 
 type decoder struct {
 	// Which parser instance in use for this decoding session.
-	p *parser
+	p *unstable.Parser
 
 	// Flag indicating that the current expression is stashed.
 	// If set to true, calling nextExpr will not actually pull a new expression
@@ -157,7 +157,7 @@ func (d *decoder) typeMismatchError(toml string, target reflect.Type) error {
 	return fmt.Errorf("toml: cannot decode TOML %s into a Go value of type %s", toml, target)
 }
 
-func (d *decoder) expr() *ast.Node {
+func (d *decoder) expr() *unstable.Node {
 	return d.p.Expression()
 }
 
@@ -208,12 +208,12 @@ func (d *decoder) FromParser(v interface{}) error {
 
 	err := d.fromParser(r)
 	if err == nil {
-		return d.strict.Error(d.p.data)
+		return d.strict.Error(d.p.Data())
 	}
 
-	var e *decodeError
+	var e *unstable.ParserError
 	if errors.As(err, &e) {
-		return wrapDecodeError(d.p.data, e)
+		return wrapDecodeError(d.p.Data(), e)
 	}
 
 	return err
@@ -234,16 +234,16 @@ func (d *decoder) fromParser(root reflect.Value) error {
 Rules for the unmarshal code:
 
 - The stack is used to keep track of which values need to be set where.
-- handle* functions <=> switch on a given ast.Kind.
+- handle* functions <=> switch on a given unstable.Kind.
 - unmarshalX* functions need to unmarshal a node of kind X.
 - An "object" is either a struct or a map.
 */
 
-func (d *decoder) handleRootExpression(expr *ast.Node, v reflect.Value) error {
+func (d *decoder) handleRootExpression(expr *unstable.Node, v reflect.Value) error {
 	var x reflect.Value
 	var err error
 
-	if !(d.skipUntilTable && expr.Kind == ast.KeyValue) {
+	if !(d.skipUntilTable && expr.Kind == unstable.KeyValue) {
 		err = d.seen.CheckExpression(expr)
 		if err != nil {
 			return err
@@ -251,16 +251,16 @@ func (d *decoder) handleRootExpression(expr *ast.Node, v reflect.Value) error {
 	}
 
 	switch expr.Kind {
-	case ast.KeyValue:
+	case unstable.KeyValue:
 		if d.skipUntilTable {
 			return nil
 		}
 		x, err = d.handleKeyValue(expr, v)
-	case ast.Table:
+	case unstable.Table:
 		d.skipUntilTable = false
 		d.strict.EnterTable(expr)
 		x, err = d.handleTable(expr.Key(), v)
-	case ast.ArrayTable:
+	case unstable.ArrayTable:
 		d.skipUntilTable = false
 		d.strict.EnterArrayTable(expr)
 		x, err = d.handleArrayTable(expr.Key(), v)
@@ -269,7 +269,7 @@ func (d *decoder) handleRootExpression(expr *ast.Node, v reflect.Value) error {
 	}
 
 	if d.skipUntilTable {
-		if expr.Kind == ast.Table || expr.Kind == ast.ArrayTable {
+		if expr.Kind == unstable.Table || expr.Kind == unstable.ArrayTable {
 			d.strict.MissingTable(expr)
 		}
 	} else if err == nil && x.IsValid() {
@@ -279,14 +279,14 @@ func (d *decoder) handleRootExpression(expr *ast.Node, v reflect.Value) error {
 	return err
 }
 
-func (d *decoder) handleArrayTable(key ast.Iterator, v reflect.Value) (reflect.Value, error) {
+func (d *decoder) handleArrayTable(key unstable.Iterator, v reflect.Value) (reflect.Value, error) {
 	if key.Next() {
 		return d.handleArrayTablePart(key, v)
 	}
 	return d.handleKeyValues(v)
 }
 
-func (d *decoder) handleArrayTableCollectionLast(key ast.Iterator, v reflect.Value) (reflect.Value, error) {
+func (d *decoder) handleArrayTableCollectionLast(key unstable.Iterator, v reflect.Value) (reflect.Value, error) {
 	switch v.Kind() {
 	case reflect.Interface:
 		elem := v.Elem()
@@ -339,13 +339,13 @@ func (d *decoder) handleArrayTableCollectionLast(key ast.Iterator, v reflect.Val
 	case reflect.Array:
 		idx := d.arrayIndex(true, v)
 		if idx >= v.Len() {
-			return v, fmt.Errorf("toml: cannot decode array table into %s at position %d", v.Type(), idx)
+			return v, fmt.Errorf("%s at position %d", d.typeMismatchError("array table", v.Type()), idx)
 		}
 		elem := v.Index(idx)
 		_, err := d.handleArrayTable(key, elem)
 		return v, err
 	default:
-		return reflect.Value{}, fmt.Errorf("toml: cannot decode array table into a %s", v.Type())
+		return reflect.Value{}, d.typeMismatchError("array table", v.Type())
 	}
 }
 
@@ -353,7 +353,7 @@ func (d *decoder) handleArrayTableCollectionLast(key ast.Iterator, v reflect.Val
 // evaluated like a normal key, but if it returns a collection, it also needs to
 // point to the last element of the collection. Unless it is the last part of
 // the key, then it needs to create a new element at the end.
-func (d *decoder) handleArrayTableCollection(key ast.Iterator, v reflect.Value) (reflect.Value, error) {
+func (d *decoder) handleArrayTableCollection(key unstable.Iterator, v reflect.Value) (reflect.Value, error) {
 	if key.IsLast() {
 		return d.handleArrayTableCollectionLast(key, v)
 	}
@@ -390,7 +390,7 @@ func (d *decoder) handleArrayTableCollection(key ast.Iterator, v reflect.Value)
 	case reflect.Array:
 		idx := d.arrayIndex(false, v)
 		if idx >= v.Len() {
-			return v, fmt.Errorf("toml: cannot decode array table into %s at position %d", v.Type(), idx)
+			return v, fmt.Errorf("%s at position %d", d.typeMismatchError("array table", v.Type()), idx)
 		}
 		elem := v.Index(idx)
 		_, err := d.handleArrayTable(key, elem)
@@ -400,7 +400,7 @@ func (d *decoder) handleArrayTableCollection(key ast.Iterator, v reflect.Value)
 	return d.handleArrayTable(key, v)
 }
 
-func (d *decoder) handleKeyPart(key ast.Iterator, v reflect.Value, nextFn handlerFn, makeFn valueMakerFn) (reflect.Value, error) {
+func (d *decoder) handleKeyPart(key unstable.Iterator, v reflect.Value, nextFn handlerFn, makeFn valueMakerFn) (reflect.Value, error) {
 	var rv reflect.Value
 
 	// First, dispatch over v to make sure it is a valid object.
@@ -417,7 +417,10 @@ func (d *decoder) handleKeyPart(key ast.Iterator, v reflect.Value, nextFn handle
 		vt := v.Type()
 
 		// Create the key for the map element. Convert to key type.
-		mk := reflect.ValueOf(string(key.Node().Data)).Convert(vt.Key())
+		mk, err := d.keyFromData(vt.Key(), key.Node().Data)
+		if err != nil {
+			return reflect.Value{}, err
+		}
 
 		// If the map does not exist, create it.
 		if v.IsNil() {
@@ -518,7 +521,7 @@ func (d *decoder) handleKeyPart(key ast.Iterator, v reflect.Value, nextFn handle
 // HandleArrayTablePart navigates the Go structure v using the key v. It is
 // only used for the prefix (non-last) parts of an array-table. When
 // encountering a collection, it should go to the last element.
-func (d *decoder) handleArrayTablePart(key ast.Iterator, v reflect.Value) (reflect.Value, error) {
+func (d *decoder) handleArrayTablePart(key unstable.Iterator, v reflect.Value) (reflect.Value, error) {
 	var makeFn valueMakerFn
 	if key.IsLast() {
 		makeFn = makeSliceInterface
@@ -530,10 +533,10 @@ func (d *decoder) handleArrayTablePart(key ast.Iterator, v reflect.Value) (refle
 
 // HandleTable returns a reference when it has checked the next expression but
 // cannot handle it.
-func (d *decoder) handleTable(key ast.Iterator, v reflect.Value) (reflect.Value, error) {
+func (d *decoder) handleTable(key unstable.Iterator, v reflect.Value) (reflect.Value, error) {
 	if v.Kind() == reflect.Slice {
 		if v.Len() == 0 {
-			return reflect.Value{}, newDecodeError(key.Node().Data, "cannot store a table in a slice")
+			return reflect.Value{}, unstable.NewParserError(key.Node().Data, "cannot store a table in a slice")
 		}
 		elem := v.Index(v.Len() - 1)
 		x, err := d.handleTable(key, elem)
@@ -560,7 +563,7 @@ func (d *decoder) handleKeyValues(v reflect.Value) (reflect.Value, error) {
 	var rv reflect.Value
 	for d.nextExpr() {
 		expr := d.expr()
-		if expr.Kind != ast.KeyValue {
+		if expr.Kind != unstable.KeyValue {
 			// Stash the expression so that fromParser can just loop and use
 			// the right handler.
 			// We could just recurse ourselves here, but at least this gives a
@@ -587,7 +590,7 @@ func (d *decoder) handleKeyValues(v reflect.Value) (reflect.Value, error) {
 }
 
 type (
-	handlerFn    func(key ast.Iterator, v reflect.Value) (reflect.Value, error)
+	handlerFn    func(key unstable.Iterator, v reflect.Value) (reflect.Value, error)
 	valueMakerFn func() reflect.Value
 )
 
@@ -599,11 +602,11 @@ func makeSliceInterface() reflect.Value {
 	return reflect.MakeSlice(sliceInterfaceType, 0, 16)
 }
 
-func (d *decoder) handleTablePart(key ast.Iterator, v reflect.Value) (reflect.Value, error) {
+func (d *decoder) handleTablePart(key unstable.Iterator, v reflect.Value) (reflect.Value, error) {
 	return d.handleKeyPart(key, v, d.handleTable, makeMapStringInterface)
 }
 
-func (d *decoder) tryTextUnmarshaler(node *ast.Node, v reflect.Value) (bool, error) {
+func (d *decoder) tryTextUnmarshaler(node *unstable.Node, v reflect.Value) (bool, error) {
 	// Special case for time, because we allow to unmarshal to it from
 	// different kind of AST nodes.
 	if v.Type() == timeType {
@@ -613,7 +616,7 @@ func (d *decoder) tryTextUnmarshaler(node *ast.Node, v reflect.Value) (bool, err
 	if v.CanAddr() && v.Addr().Type().Implements(textUnmarshalerType) {
 		err := v.Addr().Interface().(encoding.TextUnmarshaler).UnmarshalText(node.Data)
 		if err != nil {
-			return false, newDecodeError(d.p.Raw(node.Raw), "%w", err)
+			return false, unstable.NewParserError(d.p.Raw(node.Raw), "%w", err)
 		}
 
 		return true, nil
@@ -622,7 +625,7 @@ func (d *decoder) tryTextUnmarshaler(node *ast.Node, v reflect.Value) (bool, err
 	return false, nil
 }
 
-func (d *decoder) handleValue(value *ast.Node, v reflect.Value) error {
+func (d *decoder) handleValue(value *unstable.Node, v reflect.Value) error {
 	for v.Kind() == reflect.Ptr {
 		v = initAndDereferencePointer(v)
 	}
@@ -633,32 +636,32 @@ func (d *decoder) handleValue(value *ast.Node, v reflect.Value) error {
 	}
 
 	switch value.Kind {
-	case ast.String:
+	case unstable.String:
 		return d.unmarshalString(value, v)
-	case ast.Integer:
+	case unstable.Integer:
 		return d.unmarshalInteger(value, v)
-	case ast.Float:
+	case unstable.Float:
 		return d.unmarshalFloat(value, v)
-	case ast.Bool:
+	case unstable.Bool:
 		return d.unmarshalBool(value, v)
-	case ast.DateTime:
+	case unstable.DateTime:
 		return d.unmarshalDateTime(value, v)
-	case ast.LocalDate:
+	case unstable.LocalDate:
 		return d.unmarshalLocalDate(value, v)
-	case ast.LocalTime:
+	case unstable.LocalTime:
 		return d.unmarshalLocalTime(value, v)
-	case ast.LocalDateTime:
+	case unstable.LocalDateTime:
 		return d.unmarshalLocalDateTime(value, v)
-	case ast.InlineTable:
+	case unstable.InlineTable:
 		return d.unmarshalInlineTable(value, v)
-	case ast.Array:
+	case unstable.Array:
 		return d.unmarshalArray(value, v)
 	default:
 		panic(fmt.Errorf("handleValue not implemented for %s", value.Kind))
 	}
 }
 
-func (d *decoder) unmarshalArray(array *ast.Node, v reflect.Value) error {
+func (d *decoder) unmarshalArray(array *unstable.Node, v reflect.Value) error {
 	switch v.Kind() {
 	case reflect.Slice:
 		if v.IsNil() {
@@ -729,7 +732,7 @@ func (d *decoder) unmarshalArray(array *ast.Node, v reflect.Value) error {
 	return nil
 }
 
-func (d *decoder) unmarshalInlineTable(itable *ast.Node, v reflect.Value) error {
+func (d *decoder) unmarshalInlineTable(itable *unstable.Node, v reflect.Value) error {
 	// Make sure v is an initialized object.
 	switch v.Kind() {
 	case reflect.Map:
@@ -746,7 +749,7 @@ func (d *decoder) unmarshalInlineTable(itable *ast.Node, v reflect.Value) error
 		}
 		return d.unmarshalInlineTable(itable, elem)
 	default:
-		return newDecodeError(itable.Data, "cannot store inline table in Go type %s", v.Kind())
+		return unstable.NewParserError(d.p.Raw(itable.Raw), "cannot store inline table in Go type %s", v.Kind())
 	}
 
 	it := itable.Children()
@@ -765,7 +768,7 @@ func (d *decoder) unmarshalInlineTable(itable *ast.Node, v reflect.Value) error
 	return nil
 }
 
-func (d *decoder) unmarshalDateTime(value *ast.Node, v reflect.Value) error {
+func (d *decoder) unmarshalDateTime(value *unstable.Node, v reflect.Value) error {
 	dt, err := parseDateTime(value.Data)
 	if err != nil {
 		return err
@@ -775,7 +778,7 @@ func (d *decoder) unmarshalDateTime(value *ast.Node, v reflect.Value) error {
 	return nil
 }
 
-func (d *decoder) unmarshalLocalDate(value *ast.Node, v reflect.Value) error {
+func (d *decoder) unmarshalLocalDate(value *unstable.Node, v reflect.Value) error {
 	ld, err := parseLocalDate(value.Data)
 	if err != nil {
 		return err
@@ -792,28 +795,28 @@ func (d *decoder) unmarshalLocalDate(value *ast.Node, v reflect.Value) error {
 	return nil
 }
 
-func (d *decoder) unmarshalLocalTime(value *ast.Node, v reflect.Value) error {
+func (d *decoder) unmarshalLocalTime(value *unstable.Node, v reflect.Value) error {
 	lt, rest, err := parseLocalTime(value.Data)
 	if err != nil {
 		return err
 	}
 
 	if len(rest) > 0 {
-		return newDecodeError(rest, "extra characters at the end of a local time")
+		return unstable.NewParserError(rest, "extra characters at the end of a local time")
 	}
 
 	v.Set(reflect.ValueOf(lt))
 	return nil
 }
 
-func (d *decoder) unmarshalLocalDateTime(value *ast.Node, v reflect.Value) error {
+func (d *decoder) unmarshalLocalDateTime(value *unstable.Node, v reflect.Value) error {
 	ldt, rest, err := parseLocalDateTime(value.Data)
 	if err != nil {
 		return err
 	}
 
 	if len(rest) > 0 {
-		return newDecodeError(rest, "extra characters at the end of a local date time")
+		return unstable.NewParserError(rest, "extra characters at the end of a local date time")
 	}
 
 	if v.Type() == timeType {
@@ -828,7 +831,7 @@ func (d *decoder) unmarshalLocalDateTime(value *ast.Node, v reflect.Value) error
 	return nil
 }
 
-func (d *decoder) unmarshalBool(value *ast.Node, v reflect.Value) error {
+func (d *decoder) unmarshalBool(value *unstable.Node, v reflect.Value) error {
 	b := value.Data[0] == 't'
 
 	switch v.Kind() {
@@ -837,13 +840,13 @@ func (d *decoder) unmarshalBool(value *ast.Node, v reflect.Value) error {
 	case reflect.Interface:
 		v.Set(reflect.ValueOf(b))
 	default:
-		return newDecodeError(value.Data, "cannot assign boolean to a %t", b)
+		return unstable.NewParserError(value.Data, "cannot assign boolean to a %t", b)
 	}
 
 	return nil
 }
 
-func (d *decoder) unmarshalFloat(value *ast.Node, v reflect.Value) error {
+func (d *decoder) unmarshalFloat(value *unstable.Node, v reflect.Value) error {
 	f, err := parseFloat(value.Data)
 	if err != nil {
 		return err
@@ -854,13 +857,13 @@ func (d *decoder) unmarshalFloat(value *ast.Node, v reflect.Value) error {
 		v.SetFloat(f)
 	case reflect.Float32:
 		if f > math.MaxFloat32 {
-			return newDecodeError(value.Data, "number %f does not fit in a float32", f)
+			return unstable.NewParserError(value.Data, "number %f does not fit in a float32", f)
 		}
 		v.SetFloat(f)
 	case reflect.Interface:
 		v.Set(reflect.ValueOf(f))
 	default:
-		return newDecodeError(value.Data, "float cannot be assigned to %s", v.Kind())
+		return unstable.NewParserError(value.Data, "float cannot be assigned to %s", v.Kind())
 	}
 
 	return nil
@@ -886,7 +889,12 @@ func init() {
 	}
 }
 
-func (d *decoder) unmarshalInteger(value *ast.Node, v reflect.Value) error {
+func (d *decoder) unmarshalInteger(value *unstable.Node, v reflect.Value) error {
+	kind := v.Kind()
+	if kind == reflect.Float32 || kind == reflect.Float64 {
+		return d.unmarshalFloat(value, v)
+	}
+
 	i, err := parseInteger(value.Data)
 	if err != nil {
 		return err
@@ -894,7 +902,7 @@ func (d *decoder) unmarshalInteger(value *ast.Node, v reflect.Value) error {
 
 	var r reflect.Value
 
-	switch v.Kind() {
+	switch kind {
 	case reflect.Int64:
 		v.SetInt(i)
 		return nil
@@ -967,20 +975,20 @@ func (d *decoder) unmarshalInteger(value *ast.Node, v reflect.Value) error {
 	return nil
 }
 
-func (d *decoder) unmarshalString(value *ast.Node, v reflect.Value) error {
+func (d *decoder) unmarshalString(value *unstable.Node, v reflect.Value) error {
 	switch v.Kind() {
 	case reflect.String:
 		v.SetString(string(value.Data))
 	case reflect.Interface:
 		v.Set(reflect.ValueOf(string(value.Data)))
 	default:
-		return newDecodeError(d.p.Raw(value.Raw), "cannot store TOML string into a Go %s", v.Kind())
+		return unstable.NewParserError(d.p.Raw(value.Raw), "cannot store TOML string into a Go %s", v.Kind())
 	}
 
 	return nil
 }
 
-func (d *decoder) handleKeyValue(expr *ast.Node, v reflect.Value) (reflect.Value, error) {
+func (d *decoder) handleKeyValue(expr *unstable.Node, v reflect.Value) (reflect.Value, error) {
 	d.strict.EnterKeyValue(expr)
 
 	v, err := d.handleKeyValueInner(expr.Key(), expr.Value(), v)
@@ -994,7 +1002,7 @@ func (d *decoder) handleKeyValue(expr *ast.Node, v reflect.Value) (reflect.Value
 	return v, err
 }
 
-func (d *decoder) handleKeyValueInner(key ast.Iterator, value *ast.Node, v reflect.Value) (reflect.Value, error) {
+func (d *decoder) handleKeyValueInner(key unstable.Iterator, value *unstable.Node, v reflect.Value) (reflect.Value, error) {
 	if key.Next() {
 		// Still scoping the key
 		return d.handleKeyValuePart(key, value, v)
@@ -1004,7 +1012,32 @@ func (d *decoder) handleKeyValueInner(key ast.Iterator, value *ast.Node, v refle
 	return reflect.Value{}, d.handleValue(value, v)
 }
 
-func (d *decoder) handleKeyValuePart(key ast.Iterator, value *ast.Node, v reflect.Value) (reflect.Value, error) {
+func (d *decoder) keyFromData(keyType reflect.Type, data []byte) (reflect.Value, error) {
+	switch {
+	case stringType.AssignableTo(keyType):
+		return reflect.ValueOf(string(data)), nil
+
+	case stringType.ConvertibleTo(keyType):
+		return reflect.ValueOf(string(data)).Convert(keyType), nil
+
+	case keyType.Implements(textUnmarshalerType):
+		mk := reflect.New(keyType.Elem())
+		if err := mk.Interface().(encoding.TextUnmarshaler).UnmarshalText(data); err != nil {
+			return reflect.Value{}, fmt.Errorf("toml: error unmarshalling key type %s from text: %w", stringType, err)
+		}
+		return mk, nil
+
+	case reflect.PtrTo(keyType).Implements(textUnmarshalerType):
+		mk := reflect.New(keyType)
+		if err := mk.Interface().(encoding.TextUnmarshaler).UnmarshalText(data); err != nil {
+			return reflect.Value{}, fmt.Errorf("toml: error unmarshalling key type %s from text: %w", stringType, err)
+		}
+		return mk.Elem(), nil
+	}
+	return reflect.Value{}, fmt.Errorf("toml: cannot convert map key of type %s to expected type %s", stringType, keyType)
+}
+
+func (d *decoder) handleKeyValuePart(key unstable.Iterator, value *unstable.Node, v reflect.Value) (reflect.Value, error) {
 	// contains the replacement for v
 	var rv reflect.Value
 
@@ -1014,16 +1047,9 @@ func (d *decoder) handleKeyValuePart(key ast.Iterator, value *ast.Node, v reflec
 	case reflect.Map:
 		vt := v.Type()
 
-		mk := reflect.ValueOf(string(key.Node().Data))
-		mkt := stringType
-
-		keyType := vt.Key()
-		if !mkt.AssignableTo(keyType) {
-			if !mkt.ConvertibleTo(keyType) {
-				return reflect.Value{}, fmt.Errorf("toml: cannot convert map key of type %s to expected type %s", mkt, keyType)
-			}
-
-			mk = mk.Convert(keyType)
+		mk, err := d.keyFromData(vt.Key(), key.Node().Data)
+		if err != nil {
+			return reflect.Value{}, err
 		}
 
 		// If the map does not exist, create it.
@@ -1034,15 +1060,9 @@ func (d *decoder) handleKeyValuePart(key ast.Iterator, value *ast.Node, v reflec
 
 		mv := v.MapIndex(mk)
 		set := false
-		if !mv.IsValid() {
+		if !mv.IsValid() || key.IsLast() {
 			set = true
 			mv = reflect.New(v.Type().Elem()).Elem()
-		} else {
-			if key.IsLast() {
-				var x interface{}
-				mv = reflect.ValueOf(&x).Elem()
-				set = true
-			}
 		}
 
 		nv, err := d.handleKeyValueInner(key, value, mv)
@@ -1072,6 +1092,19 @@ func (d *decoder) handleKeyValuePart(key ast.Iterator, value *ast.Node, v reflec
 		d.errorContext.Field = path
 
 		f := fieldByIndex(v, path)
+
+		if !f.CanSet() {
+			// If the field is not settable, need to take a slower path and make a copy of
+			// the struct itself to a new location.
+			nvp := reflect.New(v.Type())
+			nvp.Elem().Set(v)
+			v = nvp.Elem()
+			_, err := d.handleKeyValuePart(key, value, v)
+			if err != nil {
+				return reflect.Value{}, err
+			}
+			return nvp.Elem(), nil
+		}
 		x, err := d.handleKeyValueInner(key, value, f)
 		if err != nil {
 			return reflect.Value{}, err
@@ -1137,10 +1170,10 @@ func initAndDereferencePointer(v reflect.Value) reflect.Value {
 
 // Same as reflect.Value.FieldByIndex, but creates pointers if needed.
 func fieldByIndex(v reflect.Value, path []int) reflect.Value {
-	for i, x := range path {
+	for _, x := range path {
 		v = v.Field(x)
 
-		if i < len(path)-1 && v.Kind() == reflect.Pointer {
+		if v.Kind() == reflect.Ptr {
 			if v.IsNil() {
 				v.Set(reflect.New(v.Type().Elem()))
 			}
@@ -1208,7 +1241,7 @@ func forEachField(t reflect.Type, path []int, do func(name string, path []int))
 
 		if f.Anonymous && name == "" {
 			t2 := f.Type
-			if t2.Kind() == reflect.Pointer {
+			if t2.Kind() == reflect.Ptr {
 				t2 = t2.Elem()
 			}
 

unmarshaler_test.go

@@ -16,6 +16,27 @@ import (
 	"github.com/stretchr/testify/require"
 )
 
+type unmarshalTextKey struct {
+	A string
+	B string
+}
+
+func (k *unmarshalTextKey) UnmarshalText(text []byte) error {
+	parts := strings.Split(string(text), "-")
+	if len(parts) != 2 {
+		return fmt.Errorf("invalid text key: %s", text)
+	}
+	k.A = parts[0]
+	k.B = parts[1]
+	return nil
+}
+
+type unmarshalBadTextKey struct{}
+
+func (k *unmarshalBadTextKey) UnmarshalText(text []byte) error {
+	return fmt.Errorf("error")
+}
+
 func ExampleDecoder_DisallowUnknownFields() {
 	type S struct {
 		Key1 string
@@ -69,7 +90,6 @@ func ExampleUnmarshal() {
 	fmt.Println("version:", cfg.Version)
 	fmt.Println("name:", cfg.Name)
 	fmt.Println("tags:", cfg.Tags)
-
 	// Output:
 	// version: 2
 	// name: go-toml
@@ -315,6 +335,7 @@ func TestUnmarshal(t *testing.T) {
 		target   interface{}
 		expected interface{}
 		err      bool
+		assert   func(t *testing.T, test test)
 	}
 	examples := []struct {
 		skip  bool
@@ -350,6 +371,96 @@ func TestUnmarshal(t *testing.T) {
 				}
 			},
 		},
+		{
+			desc:  "kv text key",
+			input: `a-1 = "foo"`,
+			gen: func() test {
+				type doc = map[unmarshalTextKey]string
+
+				return test{
+					target:   &doc{},
+					expected: &doc{{A: "a", B: "1"}: "foo"},
+				}
+			},
+		},
+		{
+			desc: "table text key",
+			input: `["a-1"]
+foo = "bar"`,
+			gen: func() test {
+				type doc = map[unmarshalTextKey]map[string]string
+
+				return test{
+					target:   &doc{},
+					expected: &doc{{A: "a", B: "1"}: map[string]string{"foo": "bar"}},
+				}
+			},
+		},
+		{
+			desc:  "kv ptr text key",
+			input: `a-1 = "foo"`,
+			gen: func() test {
+				type doc = map[*unmarshalTextKey]string
+
+				return test{
+					target:   &doc{},
+					expected: &doc{{A: "a", B: "1"}: "foo"},
+					assert: func(t *testing.T, test test) {
+						// Despite the documentation:
+						//     Pointer variable equality is determined based on the equality of the
+						// 		 referenced values (as opposed to the memory addresses).
+						// assert.Equal does not work properly with maps with pointer keys
+						// https://github.com/stretchr/testify/issues/1143
+						expected := make(map[unmarshalTextKey]string)
+						for k, v := range *(test.expected.(*doc)) {
+							expected[*k] = v
+						}
+						got := make(map[unmarshalTextKey]string)
+						for k, v := range *(test.target.(*doc)) {
+							got[*k] = v
+						}
+						assert.Equal(t, expected, got)
+					},
+				}
+			},
+		},
+		{
+			desc:  "kv bad text key",
+			input: `a-1 = "foo"`,
+			gen: func() test {
+				type doc = map[unmarshalBadTextKey]string
+
+				return test{
+					target: &doc{},
+					err:    true,
+				}
+			},
+		},
+		{
+			desc:  "kv bad ptr text key",
+			input: `a-1 = "foo"`,
+			gen: func() test {
+				type doc = map[*unmarshalBadTextKey]string
+
+				return test{
+					target: &doc{},
+					err:    true,
+				}
+			},
+		},
+		{
+			desc: "table bad text key",
+			input: `["a-1"]
+foo = "bar"`,
+			gen: func() test {
+				type doc = map[unmarshalBadTextKey]map[string]string
+
+				return test{
+					target: &doc{},
+					err:    true,
+				}
+			},
+		},
 		{
 			desc:  "time.time with negative zone",
 			input: `a = 1979-05-27T00:32:00-07:00 `, // space intentional
@@ -1166,6 +1277,64 @@ B = "data"`,
 				}
 			},
 		},
+		{
+			desc: "array table into maps with pointer on last key",
+			input: `[[foo]]
+			bar = "hello"`,
+			gen: func() test {
+				type doc struct {
+					Foo **[]interface{}
+				}
+				x := &[]interface{}{
+					map[string]interface{}{
+						"bar": "hello",
+					},
+				}
+				return test{
+					target: &doc{},
+					expected: &doc{
+						Foo: &x,
+					},
+				}
+			},
+		},
+		{
+			desc: "array table into maps with pointer on intermediate key",
+			input: `[[foo.foo2]]
+			bar = "hello"`,
+			gen: func() test {
+				type doc struct {
+					Foo **map[string]interface{}
+				}
+				x := &map[string]interface{}{
+					"foo2": []interface{}{
+						map[string]interface{}{
+							"bar": "hello",
+						},
+					},
+				}
+				return test{
+					target: &doc{},
+					expected: &doc{
+						Foo: &x,
+					},
+				}
+			},
+		},
+		{
+			desc: "array table into maps with pointer on last key with invalid leaf type",
+			input: `[[foo]]
+			bar = "hello"`,
+			gen: func() test {
+				type doc struct {
+					Foo **[]map[string]int
+				}
+				return test{
+					target: &doc{},
+					err:    true,
+				}
+			},
+		},
 		{
 			desc:  "unexported struct fields are ignored",
 			input: `foo = "bar"`,
@@ -1521,6 +1690,16 @@ B = "data"`,
 				}
 			},
 		},
+		{
+			desc:  "empty map into map with invalid key type",
+			input: ``,
+			gen: func() test {
+				return test{
+					target:   &map[int]string{},
+					expected: &map[int]string{},
+				}
+			},
+		},
 		{
 			desc:  "into map with convertible key type",
 			input: `A = "hello"`,
@@ -1777,7 +1956,11 @@ B = "data"`,
 				require.Error(t, err)
 			} else {
 				require.NoError(t, err)
-				assert.Equal(t, test.expected, test.target)
+				if test.assert != nil {
+					test.assert(t, test)
+				} else {
+					assert.Equal(t, test.expected, test.target)
+				}
 			}
 		})
 	}
@@ -2466,6 +2649,91 @@ func TestIssue807(t *testing.T) {
 	require.Equal(t, "foo", m.Name)
 }
 
+func TestIssue850(t *testing.T) {
+	data := make(map[string]string)
+	err := toml.Unmarshal([]byte("foo = {}"), &data)
+	require.Error(t, err)
+}
+
+func TestIssue851(t *testing.T) {
+	type Target struct {
+		Params map[string]string `toml:"params"`
+	}
+
+	content := "params = {a=\"1\",b=\"2\"}"
+	var target Target
+	err := toml.Unmarshal([]byte(content), &target)
+	require.NoError(t, err)
+	require.Equal(t, map[string]string{"a": "1", "b": "2"}, target.Params)
+	err = toml.Unmarshal([]byte(content), &target)
+	require.NoError(t, err)
+	require.Equal(t, map[string]string{"a": "1", "b": "2"}, target.Params)
+}
+
+func TestIssue866(t *testing.T) {
+	type Pipeline struct {
+		Mapping map[string]struct {
+			Req [][]string `toml:"req"`
+			Res [][]string `toml:"res"`
+		} `toml:"mapping"`
+	}
+
+	type Pipelines struct {
+		PipelineMapping map[string]*Pipeline `toml:"pipelines"`
+	}
+
+	var badToml = `
+[pipelines.register]
+mapping.inst.req = [
+    ["param1", "value1"],
+]
+mapping.inst.res = [
+    ["param2", "value2"],
+]
+`
+
+	pipelines := new(Pipelines)
+	if err := toml.NewDecoder(bytes.NewBufferString(badToml)).DisallowUnknownFields().Decode(pipelines); err != nil {
+		t.Fatal(err)
+	}
+	if pipelines.PipelineMapping["register"].Mapping["inst"].Req[0][0] != "param1" {
+		t.Fatal("unmarshal failed with mismatch value")
+	}
+
+	var goodTooToml = `
+[pipelines.register]
+mapping.inst.req = [
+    ["param1", "value1"],
+]
+`
+
+	pipelines = new(Pipelines)
+	if err := toml.NewDecoder(bytes.NewBufferString(goodTooToml)).DisallowUnknownFields().Decode(pipelines); err != nil {
+		t.Fatal(err)
+	}
+	if pipelines.PipelineMapping["register"].Mapping["inst"].Req[0][0] != "param1" {
+		t.Fatal("unmarshal failed with mismatch value")
+	}
+
+	var goodToml = `
+[pipelines.register.mapping.inst]
+req = [
+    ["param1", "value1"],
+]
+res = [
+    ["param2", "value2"],
+]
+`
+
+	pipelines = new(Pipelines)
+	if err := toml.NewDecoder(bytes.NewBufferString(goodToml)).DisallowUnknownFields().Decode(pipelines); err != nil {
+		t.Fatal(err)
+	}
+	if pipelines.PipelineMapping["register"].Mapping["inst"].Req[0][0] != "param1" {
+		t.Fatal("unmarshal failed with mismatch value")
+	}
+}
+
 func TestUnmarshalDecodeErrors(t *testing.T) {
 	examples := []struct {
 		desc string
@@ -2742,7 +3010,7 @@ world'`,
 			data: "a = \"aaaa\xE2\x80\x00\"",
 		},
 		{
-			desc: "invalid 4rd byte of 4-byte utf8 character in string with no escape sequence",
+			desc: "invalid 4th byte of 4-byte utf8 character in string with no escape sequence",
 			data: "a = \"aaaa\xF2\x81\x81\x00\"",
 		},
 		{
@@ -2758,7 +3026,7 @@ world'`,
 			data: "a = 'aaaa\xE2\x80\x00'",
 		},
 		{
-			desc: "invalid 4rd byte of 4-byte utf8 character in literal string",
+			desc: "invalid 4th byte of 4-byte utf8 character in literal string",
 			data: "a = 'aaaa\xF2\x81\x81\x00'",
 		},
 		{
@@ -3301,3 +3569,47 @@ func TestUnmarshalEmbedNonString(t *testing.T) {
 	require.NoError(t, err)
 	require.Nil(t, d.Foo)
 }
+
+func TestUnmarshal_Nil(t *testing.T) {
+	type Foo struct {
+		Foo *Foo `toml:"foo,omitempty"`
+		Bar *Foo `toml:"bar,omitempty"`
+	}
+
+	examples := []struct {
+		desc     string
+		input    string
+		expected string
+		err      bool
+	}{
+		{
+			desc:     "empty",
+			input:    ``,
+			expected: ``,
+		},
+		{
+			desc: "simplest",
+			input: `
+            [foo]
+            [foo.foo]
+            `,
+			expected: "[foo]\n[foo.foo]\n",
+		},
+	}
+
+	for _, ex := range examples {
+		e := ex
+		t.Run(e.desc, func(t *testing.T) {
+			foo := Foo{}
+			err := toml.Unmarshal([]byte(e.input), &foo)
+			if e.err {
+				require.Error(t, err)
+			} else {
+				require.NoError(t, err)
+				j, err := toml.Marshal(foo)
+				require.NoError(t, err)
+				assert.Equal(t, e.expected, string(j))
+			}
+		})
+	}
+}

unstable/ast.go

@@ -1,4 +1,4 @@
-package ast
+package unstable
 
 import (
 	"fmt"
@@ -7,13 +7,16 @@ import (
 	"github.com/pelletier/go-toml/v2/internal/danger"
 )
 
-// Iterator starts uninitialized, you need to call Next() first.
+// Iterator over a sequence of nodes.
+//
+// Starts uninitialized, you need to call Next() first.
 //
 // For example:
 //
 //	it := n.Children()
 //	for it.Next() {
-//			it.Node()
+//		n := it.Node()
+//		// do something with n
 //	}
 type Iterator struct {
 	started bool
@@ -32,42 +35,31 @@ func (c *Iterator) Next() bool {
 }
 
 // IsLast returns true if the current node of the iterator is the last
-// one.  Subsequent call to Next() will return false.
+// one.  Subsequent calls to Next() will return false.
 func (c *Iterator) IsLast() bool {
 	return c.node.next == 0
 }
 
-// Node returns a copy of the node pointed at by the iterator.
+// Node returns a pointer to the node pointed at by the iterator.
 func (c *Iterator) Node() *Node {
 	return c.node
 }
 
-// Root contains a full AST.
+// Node in a TOML expression AST.
 //
-// It is immutable once constructed with Builder.
-type Root struct {
-	nodes []Node
-}
-
-// Iterator over the top level nodes.
-func (r *Root) Iterator() Iterator {
-	it := Iterator{}
-	if len(r.nodes) > 0 {
-		it.node = &r.nodes[0]
-	}
-	return it
-}
-
-func (r *Root) at(idx Reference) *Node {
-	return &r.nodes[idx]
-}
-
-// Arrays have one child per element in the array.  InlineTables have
-// one child per key-value pair in the table.  KeyValues have at least
-// two children. The first one is the value. The rest make a
-// potentially dotted key.  Table and Array table have one child per
-// element of the key they represent (same as KeyValue, but without
-// the last node being the value).
+// Depending on Kind, its sequence of children should be interpreted
+// differently.
+//
+//   - Array have one child per element in the array.
+//   - InlineTable have one child per key-value in the table (each of kind
+//     InlineTable).
+//   - KeyValue have at least two children. The first one is the value. The rest
+//     make a potentially dotted key.
+//   - Table and ArrayTable's children represent a dotted key (same as
+//     KeyValue, but without the first node being the value).
+//
+// When relevant, Raw describes the range of bytes this node is referring to in
+// the input document. Use Parser.Raw() to retrieve the actual bytes.
 type Node struct {
 	Kind Kind
 	Raw  Range  // Raw bytes from the input.
@@ -80,13 +72,13 @@ type Node struct {
 	child int // 0 if no child
 }
 
+// Range of bytes in the document.
 type Range struct {
 	Offset uint32
 	Length uint32
 }
 
-// Next returns a copy of the next node, or an invalid Node if there
-// is no next node.
+// Next returns a pointer to the next node, or nil if there is no next node.
 func (n *Node) Next() *Node {
 	if n.next == 0 {
 		return nil
@@ -96,9 +88,9 @@ func (n *Node) Next() *Node {
 	return (*Node)(danger.Stride(ptr, size, n.next))
 }
 
-// Child returns a copy of the first child node of this node. Other
-// children can be accessed calling Next on the first child.  Returns
-// an invalid Node if there is none.
+// Child returns a pointer to the first child node of this node. Other children
+// can be accessed calling Next on the first child.  Returns an nil if this Node
+// has no child.
 func (n *Node) Child() *Node {
 	if n.child == 0 {
 		return nil
@@ -113,9 +105,9 @@ func (n *Node) Valid() bool {
 	return n != nil
 }
 
-// Key returns the child nodes making the Key on a supported
-// node. Panics otherwise.  They are guaranteed to be all be of the
-// Kind Key. A simple key would return just one element.
+// Key returns the children nodes making the Key on a supported node. Panics
+// otherwise.  They are guaranteed to be all be of the Kind Key. A simple key
+// would return just one element.
 func (n *Node) Key() Iterator {
 	switch n.Kind {
 	case KeyValue:

unstable/benchmark_test.go

@@ -1,4 +1,4 @@
-package toml
+package unstable
 
 import (
 	"bytes"
@@ -55,7 +55,7 @@ func BenchmarkParseLiteralStringValid(b *testing.B) {
 
 	for name, input := range inputs {
 		b.Run(name, func(b *testing.B) {
-			p := parser{}
+			p := Parser{}
 			b.SetBytes(int64(len(input)))
 			b.ReportAllocs()
 			b.ResetTimer()

unstable/builder.go

@@ -0,0 +1,71 @@
+package unstable
+
+// root contains a full AST.
+//
+// It is immutable once constructed with Builder.
+type root struct {
+	nodes []Node
+}
+
+// Iterator over the top level nodes.
+func (r *root) Iterator() Iterator {
+	it := Iterator{}
+	if len(r.nodes) > 0 {
+		it.node = &r.nodes[0]
+	}
+	return it
+}
+
+func (r *root) at(idx reference) *Node {
+	return &r.nodes[idx]
+}
+
+type reference int
+
+const invalidReference reference = -1
+
+func (r reference) Valid() bool {
+	return r != invalidReference
+}
+
+type builder struct {
+	tree    root
+	lastIdx int
+}
+
+func (b *builder) Tree() *root {
+	return &b.tree
+}
+
+func (b *builder) NodeAt(ref reference) *Node {
+	return b.tree.at(ref)
+}
+
+func (b *builder) Reset() {
+	b.tree.nodes = b.tree.nodes[:0]
+	b.lastIdx = 0
+}
+
+func (b *builder) Push(n Node) reference {
+	b.lastIdx = len(b.tree.nodes)
+	b.tree.nodes = append(b.tree.nodes, n)
+	return reference(b.lastIdx)
+}
+
+func (b *builder) PushAndChain(n Node) reference {
+	newIdx := len(b.tree.nodes)
+	b.tree.nodes = append(b.tree.nodes, n)
+	if b.lastIdx >= 0 {
+		b.tree.nodes[b.lastIdx].next = newIdx - b.lastIdx
+	}
+	b.lastIdx = newIdx
+	return reference(b.lastIdx)
+}
+
+func (b *builder) AttachChild(parent reference, child reference) {
+	b.tree.nodes[parent].child = int(child) - int(parent)
+}
+
+func (b *builder) Chain(from reference, to reference) {
+	b.tree.nodes[from].next = int(to) - int(from)
+}

unstable/doc.go

@@ -0,0 +1,3 @@
+// Package unstable provides APIs that do not meet the backward compatibility
+// guarantees yet.
+package unstable

unstable/kind.go

@@ -1,25 +1,26 @@
-package ast
+package unstable
 
 import "fmt"
 
+// Kind represents the type of TOML structure contained in a given Node.
 type Kind int
 
 const (
-	// meta
+	// Meta
 	Invalid Kind = iota
 	Comment
 	Key
 
-	// top level structures
+	// Top level structures
 	Table
 	ArrayTable
 	KeyValue
 
-	// containers values
+	// Containers values
 	Array
 	InlineTable
 
-	// values
+	// Values
 	String
 	Bool
 	Float
@@ -30,6 +31,7 @@ const (
 	DateTime
 )
 
+// String implementation of fmt.Stringer.
 func (k Kind) String() string {
 	switch k {
 	case Invalid:

unstable/parser.go

@@ -1,50 +1,108 @@
-package toml
+package unstable
 
 import (
 	"bytes"
+	"fmt"
 	"unicode"
 
-	"github.com/pelletier/go-toml/v2/internal/ast"
+	"github.com/pelletier/go-toml/v2/internal/characters"
 	"github.com/pelletier/go-toml/v2/internal/danger"
 )
 
-type parser struct {
-	builder ast.Builder
-	ref     ast.Reference
+// ParserError describes an error relative to the content of the document.
+//
+// It cannot outlive the instance of Parser it refers to, and may cause panics
+// if the parser is reset.
+type ParserError struct {
+	Highlight []byte
+	Message   string
+	Key       []string // optional
+}
+
+// Error is the implementation of the error interface.
+func (e *ParserError) Error() string {
+	return e.Message
+}
+
+// NewParserError is a convenience function to create a ParserError
+//
+// Warning: Highlight needs to be a subslice of Parser.data, so only slices
+// returned by Parser.Raw are valid candidates.
+func NewParserError(highlight []byte, format string, args ...interface{}) error {
+	return &ParserError{
+		Highlight: highlight,
+		Message:   fmt.Errorf(format, args...).Error(),
+	}
+}
+
+// Parser scans over a TOML-encoded document and generates an iterative AST.
+//
+// To prime the Parser, first reset it with the contents of a TOML document.
+// Then, process all top-level expressions sequentially. See Example.
+//
+// Don't forget to check Error() after you're done parsing.
+//
+// Each top-level expression needs to be fully processed before calling
+// NextExpression() again. Otherwise, calls to various Node methods may panic if
+// the parser has moved on the next expression.
+//
+// For performance reasons, go-toml doesn't make a copy of the input bytes to
+// the parser. Make sure to copy all the bytes you need to outlive the slice
+// given to the parser.
+type Parser struct {
 	data    []byte
+	builder builder
+	ref     reference
 	left    []byte
 	err     error
 	first   bool
+
+	KeepComments bool
 }
 
-func (p *parser) Range(b []byte) ast.Range {
-	return ast.Range{
+// Data returns the slice provided to the last call to Reset.
+func (p *Parser) Data() []byte {
+	return p.data
+}
+
+// Range returns a range description that corresponds to a given slice of the
+// input. If the argument is not a subslice of the parser input, this function
+// panics.
+func (p *Parser) Range(b []byte) Range {
+	return Range{
 		Offset: uint32(danger.SubsliceOffset(p.data, b)),
 		Length: uint32(len(b)),
 	}
 }
 
-func (p *parser) Raw(raw ast.Range) []byte {
+// Raw returns the slice corresponding to the bytes in the given range.
+func (p *Parser) Raw(raw Range) []byte {
 	return p.data[raw.Offset : raw.Offset+raw.Length]
 }
 
-func (p *parser) Reset(b []byte) {
+// Reset brings the parser to its initial state for a given input. It wipes an
+// reuses internal storage to reduce allocation.
+func (p *Parser) Reset(b []byte) {
 	p.builder.Reset()
-	p.ref = ast.InvalidReference
+	p.ref = invalidReference
 	p.data = b
 	p.left = b
 	p.err = nil
 	p.first = true
 }
 
-//nolint:cyclop
-func (p *parser) NextExpression() bool {
+// NextExpression parses the next top-level expression. If an expression was
+// successfully parsed, it returns true. If the parser is at the end of the
+// document or an error occurred, it returns false.
+//
+// Retrieve the parsed expression with Expression().
+func (p *Parser) NextExpression() bool {
 	if len(p.left) == 0 || p.err != nil {
 		return false
 	}
 
 	p.builder.Reset()
-	p.ref = ast.InvalidReference
+	p.ref = invalidReference
 
 	for {
 		if len(p.left) == 0 || p.err != nil {
@@ -73,15 +131,56 @@ func (p *parser) NextExpression() bool {
 	}
 }
 
-func (p *parser) Expression() *ast.Node {
+// Expression returns a pointer to the node representing the last successfully
+// parsed expression.
+func (p *Parser) Expression() *Node {
 	return p.builder.NodeAt(p.ref)
 }
 
-func (p *parser) Error() error {
+// Error returns any error that has occurred during parsing.
+func (p *Parser) Error() error {
 	return p.err
 }
 
-func (p *parser) parseNewline(b []byte) ([]byte, error) {
+// Position describes a position in the input.
+type Position struct {
+	// Number of bytes from the beginning of the input.
+	Offset int
+	// Line number, starting at 1.
+	Line int
+	// Column number, starting at 1.
+	Column int
+}
+
+// Shape describes the position of a range in the input.
+type Shape struct {
+	Start Position
+	End   Position
+}
+
+func (p *Parser) position(b []byte) Position {
+	offset := danger.SubsliceOffset(p.data, b)
+
+	lead := p.data[:offset]
+
+	return Position{
+		Offset: offset,
+		Line:   bytes.Count(lead, []byte{'\n'}) + 1,
+		Column: len(lead) - bytes.LastIndex(lead, []byte{'\n'}),
+	}
+}
+
+// Shape returns the shape of the given range in the input.  Will
+// panic if the range is not a subslice of the input.
+func (p *Parser) Shape(r Range) Shape {
+	raw := p.Raw(r)
+	return Shape{
+		Start: p.position(raw),
+		End:   p.position(raw[r.Length:]),
+	}
+}
+
+func (p *Parser) parseNewline(b []byte) ([]byte, error) {
 	if b[0] == '\n' {
 		return b[1:], nil
 	}
@@ -91,14 +190,27 @@ func (p *parser) parseNewline(b []byte) ([]byte, error) {
 		return rest, err
 	}
 
-	return nil, newDecodeError(b[0:1], "expected newline but got %#U", b[0])
+	return nil, NewParserError(b[0:1], "expected newline but got %#U", b[0])
 }
 
-func (p *parser) parseExpression(b []byte) (ast.Reference, []byte, error) {
+func (p *Parser) parseComment(b []byte) (reference, []byte, error) {
+	ref := invalidReference
+	data, rest, err := scanComment(b)
+	if p.KeepComments && err == nil {
+		ref = p.builder.Push(Node{
+			Kind: Comment,
+			Raw:  p.Range(data),
+			Data: data,
+		})
+	}
+	return ref, rest, err
+}
+
+func (p *Parser) parseExpression(b []byte) (reference, []byte, error) {
 	// expression =  ws [ comment ]
 	// expression =/ ws keyval ws [ comment ]
 	// expression =/ ws table ws [ comment ]
-	ref := ast.InvalidReference
+	ref := invalidReference
 
 	b = p.parseWhitespace(b)
 
@@ -107,7 +219,7 @@ func (p *parser) parseExpression(b []byte) (ast.Reference, []byte, error) {
 	}
 
 	if b[0] == '#' {
-		_, rest, err := scanComment(b)
+		ref, rest, err := p.parseComment(b)
 		return ref, rest, err
 	}
 
@@ -129,14 +241,17 @@ func (p *parser) parseExpression(b []byte) (ast.Reference, []byte, error) {
 	b = p.parseWhitespace(b)
 
 	if len(b) > 0 && b[0] == '#' {
-		_, rest, err := scanComment(b)
+		cref, rest, err := p.parseComment(b)
+		if cref != invalidReference {
+			p.builder.Chain(ref, cref)
+		}
 		return ref, rest, err
 	}
 
 	return ref, b, nil
 }
 
-func (p *parser) parseTable(b []byte) (ast.Reference, []byte, error) {
+func (p *Parser) parseTable(b []byte) (reference, []byte, error) {
 	// table = std-table / array-table
 	if len(b) > 1 && b[1] == '[' {
 		return p.parseArrayTable(b)
@@ -145,12 +260,12 @@ func (p *parser) parseTable(b []byte) (ast.Reference, []byte, error) {
 	return p.parseStdTable(b)
 }
 
-func (p *parser) parseArrayTable(b []byte) (ast.Reference, []byte, error) {
+func (p *Parser) parseArrayTable(b []byte) (reference, []byte, error) {
 	// array-table = array-table-open key array-table-close
 	// array-table-open  = %x5B.5B ws  ; [[ Double left square bracket
 	// array-table-close = ws %x5D.5D  ; ]] Double right square bracket
-	ref := p.builder.Push(ast.Node{
-		Kind: ast.ArrayTable,
+	ref := p.builder.Push(Node{
+		Kind: ArrayTable,
 	})
 
 	b = b[2:]
@@ -174,12 +289,12 @@ func (p *parser) parseArrayTable(b []byte) (ast.Reference, []byte, error) {
 	return ref, b, err
 }
 
-func (p *parser) parseStdTable(b []byte) (ast.Reference, []byte, error) {
+func (p *Parser) parseStdTable(b []byte) (reference, []byte, error) {
 	// std-table = std-table-open key std-table-close
 	// std-table-open  = %x5B ws     ; [ Left square bracket
 	// std-table-close = ws %x5D     ; ] Right square bracket
-	ref := p.builder.Push(ast.Node{
-		Kind: ast.Table,
+	ref := p.builder.Push(Node{
+		Kind: Table,
 	})
 
 	b = b[1:]
@@ -199,15 +314,15 @@ func (p *parser) parseStdTable(b []byte) (ast.Reference, []byte, error) {
 	return ref, b, err
 }
 
-func (p *parser) parseKeyval(b []byte) (ast.Reference, []byte, error) {
+func (p *Parser) parseKeyval(b []byte) (reference, []byte, error) {
 	// keyval = key keyval-sep val
-	ref := p.builder.Push(ast.Node{
-		Kind: ast.KeyValue,
+	ref := p.builder.Push(Node{
+		Kind: KeyValue,
 	})
 
 	key, b, err := p.parseKey(b)
 	if err != nil {
-		return ast.InvalidReference, nil, err
+		return invalidReference, nil, err
 	}
 
 	// keyval-sep = ws %x3D ws ; =
@@ -215,12 +330,12 @@ func (p *parser) parseKeyval(b []byte) (ast.Reference, []byte, error) {
 	b = p.parseWhitespace(b)
 
 	if len(b) == 0 {
-		return ast.InvalidReference, nil, newDecodeError(b, "expected = after a key, but the document ends there")
+		return invalidReference, nil, NewParserError(b, "expected = after a key, but the document ends there")
 	}
 
 	b, err = expect('=', b)
 	if err != nil {
-		return ast.InvalidReference, nil, err
+		return invalidReference, nil, err
 	}
 
 	b = p.parseWhitespace(b)
@@ -237,12 +352,12 @@ func (p *parser) parseKeyval(b []byte) (ast.Reference, []byte, error) {
 }
 
 //nolint:cyclop,funlen
-func (p *parser) parseVal(b []byte) (ast.Reference, []byte, error) {
+func (p *Parser) parseVal(b []byte) (reference, []byte, error) {
 	// val = string / boolean / array / inline-table / date-time / float / integer
-	ref := ast.InvalidReference
+	ref := invalidReference
 
 	if len(b) == 0 {
-		return ref, nil, newDecodeError(b, "expected value, not eof")
+		return ref, nil, NewParserError(b, "expected value, not eof")
 	}
 
 	var err error
@@ -259,8 +374,8 @@ func (p *parser) parseVal(b []byte) (ast.Reference, []byte, error) {
 		}
 
 		if err == nil {
-			ref = p.builder.Push(ast.Node{
-				Kind: ast.String,
+			ref = p.builder.Push(Node{
+				Kind: String,
 				Raw:  p.Range(raw),
 				Data: v,
 			})
@@ -277,8 +392,8 @@ func (p *parser) parseVal(b []byte) (ast.Reference, []byte, error) {
 		}
 
 		if err == nil {
-			ref = p.builder.Push(ast.Node{
-				Kind: ast.String,
+			ref = p.builder.Push(Node{
+				Kind: String,
 				Raw:  p.Range(raw),
 				Data: v,
 			})
@@ -287,22 +402,22 @@ func (p *parser) parseVal(b []byte) (ast.Reference, []byte, error) {
 		return ref, b, err
 	case 't':
 		if !scanFollowsTrue(b) {
-			return ref, nil, newDecodeError(atmost(b, 4), "expected 'true'")
+			return ref, nil, NewParserError(atmost(b, 4), "expected 'true'")
 		}
 
-		ref = p.builder.Push(ast.Node{
-			Kind: ast.Bool,
+		ref = p.builder.Push(Node{
+			Kind: Bool,
 			Data: b[:4],
 		})
 
 		return ref, b[4:], nil
 	case 'f':
 		if !scanFollowsFalse(b) {
-			return ref, nil, newDecodeError(atmost(b, 5), "expected 'false'")
+			return ref, nil, NewParserError(atmost(b, 5), "expected 'false'")
 		}
 
-		ref = p.builder.Push(ast.Node{
-			Kind: ast.Bool,
+		ref = p.builder.Push(Node{
+			Kind: Bool,
 			Data: b[:5],
 		})
 
@@ -324,7 +439,7 @@ func atmost(b []byte, n int) []byte {
 	return b[:n]
 }
 
-func (p *parser) parseLiteralString(b []byte) ([]byte, []byte, []byte, error) {
+func (p *Parser) parseLiteralString(b []byte) ([]byte, []byte, []byte, error) {
 	v, rest, err := scanLiteralString(b)
 	if err != nil {
 		return nil, nil, nil, err
@@ -333,19 +448,20 @@ func (p *parser) parseLiteralString(b []byte) ([]byte, []byte, []byte, error) {
 	return v, v[1 : len(v)-1], rest, nil
 }
 
-func (p *parser) parseInlineTable(b []byte) (ast.Reference, []byte, error) {
+func (p *Parser) parseInlineTable(b []byte) (reference, []byte, error) {
 	// inline-table = inline-table-open [ inline-table-keyvals ] inline-table-close
 	// inline-table-open  = %x7B ws     ; {
 	// inline-table-close = ws %x7D     ; }
 	// inline-table-sep   = ws %x2C ws  ; , Comma
 	// inline-table-keyvals = keyval [ inline-table-sep inline-table-keyvals ]
-	parent := p.builder.Push(ast.Node{
-		Kind: ast.InlineTable,
+	parent := p.builder.Push(Node{
+		Kind: InlineTable,
+		Raw:  p.Range(b[:1]),
 	})
 
 	first := true
 
-	var child ast.Reference
+	var child reference
 
 	b = b[1:]
 
@@ -356,7 +472,7 @@ func (p *parser) parseInlineTable(b []byte) (ast.Reference, []byte, error) {
 		b = p.parseWhitespace(b)
 
 		if len(b) == 0 {
-			return parent, nil, newDecodeError(previousB[:1], "inline table is incomplete")
+			return parent, nil, NewParserError(previousB[:1], "inline table is incomplete")
 		}
 
 		if b[0] == '}' {
@@ -371,7 +487,7 @@ func (p *parser) parseInlineTable(b []byte) (ast.Reference, []byte, error) {
 			b = p.parseWhitespace(b)
 		}
 
-		var kv ast.Reference
+		var kv reference
 
 		kv, b, err = p.parseKeyval(b)
 		if err != nil {
@@ -394,7 +510,7 @@ func (p *parser) parseInlineTable(b []byte) (ast.Reference, []byte, error) {
 }
 
 //nolint:funlen,cyclop
-func (p *parser) parseValArray(b []byte) (ast.Reference, []byte, error) {
+func (p *Parser) parseValArray(b []byte) (reference, []byte, error) {
 	// array = array-open [ array-values ] ws-comment-newline array-close
 	// array-open =  %x5B ; [
 	// array-close = %x5D ; ]
@@ -405,23 +521,39 @@ func (p *parser) parseValArray(b []byte) (ast.Reference, []byte, error) {
 	arrayStart := b
 	b = b[1:]
 
-	parent := p.builder.Push(ast.Node{
-		Kind: ast.Array,
+	parent := p.builder.Push(Node{
+		Kind: Array,
 	})
 
+	// First indicates whether the parser is looking for the first element
+	// (non-comment) of the array.
 	first := true
 
-	var lastChild ast.Reference
+	lastChild := invalidReference
+
+	addChild := func(valueRef reference) {
+		if lastChild == invalidReference {
+			p.builder.AttachChild(parent, valueRef)
+		} else {
+			p.builder.Chain(lastChild, valueRef)
+		}
+		lastChild = valueRef
+	}
 
 	var err error
 	for len(b) > 0 {
-		b, err = p.parseOptionalWhitespaceCommentNewline(b)
+		cref := invalidReference
+		cref, b, err = p.parseOptionalWhitespaceCommentNewline(b)
 		if err != nil {
 			return parent, nil, err
 		}
 
+		if cref != invalidReference {
+			addChild(cref)
+		}
+
 		if len(b) == 0 {
-			return parent, nil, newDecodeError(arrayStart[:1], "array is incomplete")
+			return parent, nil, NewParserError(arrayStart[:1], "array is incomplete")
 		}
 
 		if b[0] == ']' {
@@ -430,16 +562,19 @@ func (p *parser) parseValArray(b []byte) (ast.Reference, []byte, error) {
 
 		if b[0] == ',' {
 			if first {
-				return parent, nil, newDecodeError(b[0:1], "array cannot start with comma")
+				return parent, nil, NewParserError(b[0:1], "array cannot start with comma")
 			}
 			b = b[1:]
 
-			b, err = p.parseOptionalWhitespaceCommentNewline(b)
+			cref, b, err = p.parseOptionalWhitespaceCommentNewline(b)
 			if err != nil {
 				return parent, nil, err
 			}
+			if cref != invalidReference {
+				addChild(cref)
+			}
 		} else if !first {
-			return parent, nil, newDecodeError(b[0:1], "array elements must be separated by commas")
+			return parent, nil, NewParserError(b[0:1], "array elements must be separated by commas")
 		}
 
 		// TOML allows trailing commas in arrays.
@@ -447,23 +582,22 @@ func (p *parser) parseValArray(b []byte) (ast.Reference, []byte, error) {
 			break
 		}
 
-		var valueRef ast.Reference
+		var valueRef reference
 		valueRef, b, err = p.parseVal(b)
 		if err != nil {
 			return parent, nil, err
 		}
 
-		if first {
-			p.builder.AttachChild(parent, valueRef)
-		} else {
-			p.builder.Chain(lastChild, valueRef)
-		}
-		lastChild = valueRef
+		addChild(valueRef)
 
-		b, err = p.parseOptionalWhitespaceCommentNewline(b)
+		cref, b, err = p.parseOptionalWhitespaceCommentNewline(b)
 		if err != nil {
 			return parent, nil, err
 		}
+		if cref != invalidReference {
+			addChild(cref)
+		}
+
 		first = false
 	}
 
@@ -472,15 +606,34 @@ func (p *parser) parseValArray(b []byte) (ast.Reference, []byte, error) {
 	return parent, rest, err
 }
 
-func (p *parser) parseOptionalWhitespaceCommentNewline(b []byte) ([]byte, error) {
+func (p *Parser) parseOptionalWhitespaceCommentNewline(b []byte) (reference, []byte, error) {
+	rootCommentRef := invalidReference
+	latestCommentRef := invalidReference
+
+	addComment := func(ref reference) {
+		if rootCommentRef == invalidReference {
+			rootCommentRef = ref
+		} else if latestCommentRef == invalidReference {
+			p.builder.AttachChild(rootCommentRef, ref)
+			latestCommentRef = ref
+		} else {
+			p.builder.Chain(latestCommentRef, ref)
+			latestCommentRef = ref
+		}
+	}
+
 	for len(b) > 0 {
 		var err error
 		b = p.parseWhitespace(b)
 
 		if len(b) > 0 && b[0] == '#' {
-			_, b, err = scanComment(b)
+			var ref reference
+			ref, b, err = p.parseComment(b)
 			if err != nil {
-				return nil, err
+				return invalidReference, nil, err
+			}
+			if ref != invalidReference {
+				addComment(ref)
 			}
 		}
 
@@ -491,17 +644,17 @@ func (p *parser) parseOptionalWhitespaceCommentNewline(b []byte) ([]byte, error)
 		if b[0] == '\n' || b[0] == '\r' {
 			b, err = p.parseNewline(b)
 			if err != nil {
-				return nil, err
+				return invalidReference, nil, err
 			}
 		} else {
 			break
 		}
 	}
 
-	return b, nil
+	return rootCommentRef, b, nil
 }
 
-func (p *parser) parseMultilineLiteralString(b []byte) ([]byte, []byte, []byte, error) {
+func (p *Parser) parseMultilineLiteralString(b []byte) ([]byte, []byte, []byte, error) {
 	token, rest, err := scanMultilineLiteralString(b)
 	if err != nil {
 		return nil, nil, nil, err
@@ -520,7 +673,7 @@ func (p *parser) parseMultilineLiteralString(b []byte) ([]byte, []byte, []byte,
 }
 
 //nolint:funlen,gocognit,cyclop
-func (p *parser) parseMultilineBasicString(b []byte) ([]byte, []byte, []byte, error) {
+func (p *Parser) parseMultilineBasicString(b []byte) ([]byte, []byte, []byte, error) {
 	// ml-basic-string = ml-basic-string-delim [ newline ] ml-basic-body
 	// ml-basic-string-delim
 	// ml-basic-string-delim = 3quotation-mark
@@ -551,11 +704,11 @@ func (p *parser) parseMultilineBasicString(b []byte) ([]byte, []byte, []byte, er
 
 	if !escaped {
 		str := token[startIdx:endIdx]
-		verr := utf8TomlValidAlreadyEscaped(str)
+		verr := characters.Utf8TomlValidAlreadyEscaped(str)
 		if verr.Zero() {
 			return token, str, rest, nil
 		}
-		return nil, nil, nil, newDecodeError(str[verr.Index:verr.Index+verr.Size], "invalid UTF-8")
+		return nil, nil, nil, NewParserError(str[verr.Index:verr.Index+verr.Size], "invalid UTF-8")
 	}
 
 	var builder bytes.Buffer
@@ -635,13 +788,13 @@ func (p *parser) parseMultilineBasicString(b []byte) ([]byte, []byte, []byte, er
 				builder.WriteRune(x)
 				i += 8
 			default:
-				return nil, nil, nil, newDecodeError(token[i:i+1], "invalid escaped character %#U", c)
+				return nil, nil, nil, NewParserError(token[i:i+1], "invalid escaped character %#U", c)
 			}
 			i++
 		} else {
-			size := utf8ValidNext(token[i:])
+			size := characters.Utf8ValidNext(token[i:])
 			if size == 0 {
-				return nil, nil, nil, newDecodeError(token[i:i+1], "invalid character %#U", c)
+				return nil, nil, nil, NewParserError(token[i:i+1], "invalid character %#U", c)
 			}
 			builder.Write(token[i : i+size])
 			i += size
@@ -651,7 +804,7 @@ func (p *parser) parseMultilineBasicString(b []byte) ([]byte, []byte, []byte, er
 	return token, builder.Bytes(), rest, nil
 }
 
-func (p *parser) parseKey(b []byte) (ast.Reference, []byte, error) {
+func (p *Parser) parseKey(b []byte) (reference, []byte, error) {
 	// key = simple-key / dotted-key
 	// simple-key = quoted-key / unquoted-key
 	//
@@ -662,11 +815,11 @@ func (p *parser) parseKey(b []byte) (ast.Reference, []byte, error) {
 	// dot-sep   = ws %x2E ws  ; . Period
 	raw, key, b, err := p.parseSimpleKey(b)
 	if err != nil {
-		return ast.InvalidReference, nil, err
+		return invalidReference, nil, err
 	}
 
-	ref := p.builder.Push(ast.Node{
-		Kind: ast.Key,
+	ref := p.builder.Push(Node{
+		Kind: Key,
 		Raw:  p.Range(raw),
 		Data: key,
 	})
@@ -681,8 +834,8 @@ func (p *parser) parseKey(b []byte) (ast.Reference, []byte, error) {
 				return ref, nil, err
 			}
 
-			p.builder.PushAndChain(ast.Node{
-				Kind: ast.Key,
+			p.builder.PushAndChain(Node{
+				Kind: Key,
 				Raw:  p.Range(raw),
 				Data: key,
 			})
@@ -694,9 +847,9 @@ func (p *parser) parseKey(b []byte) (ast.Reference, []byte, error) {
 	return ref, b, nil
 }
 
-func (p *parser) parseSimpleKey(b []byte) (raw, key, rest []byte, err error) {
+func (p *Parser) parseSimpleKey(b []byte) (raw, key, rest []byte, err error) {
 	if len(b) == 0 {
-		return nil, nil, nil, newDecodeError(b, "expected key but found none")
+		return nil, nil, nil, NewParserError(b, "expected key but found none")
 	}
 
 	// simple-key = quoted-key / unquoted-key
@@ -711,12 +864,12 @@ func (p *parser) parseSimpleKey(b []byte) (raw, key, rest []byte, err error) {
 		key, rest = scanUnquotedKey(b)
 		return key, key, rest, nil
 	default:
-		return nil, nil, nil, newDecodeError(b[0:1], "invalid character at start of key: %c", b[0])
+		return nil, nil, nil, NewParserError(b[0:1], "invalid character at start of key: %c", b[0])
 	}
 }
 
 //nolint:funlen,cyclop
-func (p *parser) parseBasicString(b []byte) ([]byte, []byte, []byte, error) {
+func (p *Parser) parseBasicString(b []byte) ([]byte, []byte, []byte, error) {
 	// basic-string = quotation-mark *basic-char quotation-mark
 	// quotation-mark = %x22            ; "
 	// basic-char = basic-unescaped / escaped
@@ -744,11 +897,11 @@ func (p *parser) parseBasicString(b []byte) ([]byte, []byte, []byte, error) {
 	// validate the string and return a direct reference to the buffer.
 	if !escaped {
 		str := token[startIdx:endIdx]
-		verr := utf8TomlValidAlreadyEscaped(str)
+		verr := characters.Utf8TomlValidAlreadyEscaped(str)
 		if verr.Zero() {
 			return token, str, rest, nil
 		}
-		return nil, nil, nil, newDecodeError(str[verr.Index:verr.Index+verr.Size], "invalid UTF-8")
+		return nil, nil, nil, NewParserError(str[verr.Index:verr.Index+verr.Size], "invalid UTF-8")
 	}
 
 	i := startIdx
@@ -795,13 +948,13 @@ func (p *parser) parseBasicString(b []byte) ([]byte, []byte, []byte, error) {
 				builder.WriteRune(x)
 				i += 8
 			default:
-				return nil, nil, nil, newDecodeError(token[i:i+1], "invalid escaped character %#U", c)
+				return nil, nil, nil, NewParserError(token[i:i+1], "invalid escaped character %#U", c)
 			}
 			i++
 		} else {
-			size := utf8ValidNext(token[i:])
+			size := characters.Utf8ValidNext(token[i:])
 			if size == 0 {
-				return nil, nil, nil, newDecodeError(token[i:i+1], "invalid character %#U", c)
+				return nil, nil, nil, NewParserError(token[i:i+1], "invalid character %#U", c)
 			}
 			builder.Write(token[i : i+size])
 			i += size
@@ -813,7 +966,7 @@ func (p *parser) parseBasicString(b []byte) ([]byte, []byte, []byte, error) {
 
 func hexToRune(b []byte, length int) (rune, error) {
 	if len(b) < length {
-		return -1, newDecodeError(b, "unicode point needs %d character, not %d", length, len(b))
+		return -1, NewParserError(b, "unicode point needs %d character, not %d", length, len(b))
 	}
 	b = b[:length]
 
@@ -828,19 +981,19 @@ func hexToRune(b []byte, length int) (rune, error) {
 		case 'A' <= c && c <= 'F':
 			d = uint32(c - 'A' + 10)
 		default:
-			return -1, newDecodeError(b[i:i+1], "non-hex character")
+			return -1, NewParserError(b[i:i+1], "non-hex character")
 		}
 		r = r*16 + d
 	}
 
 	if r > unicode.MaxRune || 0xD800 <= r && r < 0xE000 {
-		return -1, newDecodeError(b, "escape sequence is invalid Unicode code point")
+		return -1, NewParserError(b, "escape sequence is invalid Unicode code point")
 	}
 
 	return rune(r), nil
 }
 
-func (p *parser) parseWhitespace(b []byte) []byte {
+func (p *Parser) parseWhitespace(b []byte) []byte {
 	// ws = *wschar
 	// wschar =  %x20  ; Space
 	// wschar =/ %x09  ; Horizontal tab
@@ -850,24 +1003,24 @@ func (p *parser) parseWhitespace(b []byte) []byte {
 }
 
 //nolint:cyclop
-func (p *parser) parseIntOrFloatOrDateTime(b []byte) (ast.Reference, []byte, error) {
+func (p *Parser) parseIntOrFloatOrDateTime(b []byte) (reference, []byte, error) {
 	switch b[0] {
 	case 'i':
 		if !scanFollowsInf(b) {
-			return ast.InvalidReference, nil, newDecodeError(atmost(b, 3), "expected 'inf'")
+			return invalidReference, nil, NewParserError(atmost(b, 3), "expected 'inf'")
 		}
 
-		return p.builder.Push(ast.Node{
-			Kind: ast.Float,
+		return p.builder.Push(Node{
+			Kind: Float,
 			Data: b[:3],
 		}), b[3:], nil
 	case 'n':
 		if !scanFollowsNan(b) {
-			return ast.InvalidReference, nil, newDecodeError(atmost(b, 3), "expected 'nan'")
+			return invalidReference, nil, NewParserError(atmost(b, 3), "expected 'nan'")
 		}
 
-		return p.builder.Push(ast.Node{
-			Kind: ast.Float,
+		return p.builder.Push(Node{
+			Kind: Float,
 			Data: b[:3],
 		}), b[3:], nil
 	case '+', '-':
@@ -898,7 +1051,7 @@ func (p *parser) parseIntOrFloatOrDateTime(b []byte) (ast.Reference, []byte, err
 	return p.scanIntOrFloat(b)
 }
 
-func (p *parser) scanDateTime(b []byte) (ast.Reference, []byte, error) {
+func (p *Parser) scanDateTime(b []byte) (reference, []byte, error) {
 	// scans for contiguous characters in [0-9T:Z.+-], and up to one space if
 	// followed by a digit.
 	hasDate := false
@@ -941,30 +1094,30 @@ byteLoop:
 		}
 	}
 
-	var kind ast.Kind
+	var kind Kind
 
 	if hasTime {
 		if hasDate {
 			if hasTz {
-				kind = ast.DateTime
+				kind = DateTime
 			} else {
-				kind = ast.LocalDateTime
+				kind = LocalDateTime
 			}
 		} else {
-			kind = ast.LocalTime
+			kind = LocalTime
 		}
 	} else {
-		kind = ast.LocalDate
+		kind = LocalDate
 	}
 
-	return p.builder.Push(ast.Node{
+	return p.builder.Push(Node{
 		Kind: kind,
 		Data: b[:i],
 	}), b[i:], nil
 }
 
 //nolint:funlen,gocognit,cyclop
-func (p *parser) scanIntOrFloat(b []byte) (ast.Reference, []byte, error) {
+func (p *Parser) scanIntOrFloat(b []byte) (reference, []byte, error) {
 	i := 0
 
 	if len(b) > 2 && b[0] == '0' && b[1] != '.' && b[1] != 'e' && b[1] != 'E' {
@@ -990,8 +1143,8 @@ func (p *parser) scanIntOrFloat(b []byte) (ast.Reference, []byte, error) {
 			}
 		}
 
-		return p.builder.Push(ast.Node{
-			Kind: ast.Integer,
+		return p.builder.Push(Node{
+			Kind: Integer,
 			Data: b[:i],
 		}), b[i:], nil
 	}
@@ -1013,40 +1166,40 @@ func (p *parser) scanIntOrFloat(b []byte) (ast.Reference, []byte, error) {
 
 		if c == 'i' {
 			if scanFollowsInf(b[i:]) {
-				return p.builder.Push(ast.Node{
-					Kind: ast.Float,
+				return p.builder.Push(Node{
+					Kind: Float,
 					Data: b[:i+3],
 				}), b[i+3:], nil
 			}
 
-			return ast.InvalidReference, nil, newDecodeError(b[i:i+1], "unexpected character 'i' while scanning for a number")
+			return invalidReference, nil, NewParserError(b[i:i+1], "unexpected character 'i' while scanning for a number")
 		}
 
 		if c == 'n' {
 			if scanFollowsNan(b[i:]) {
-				return p.builder.Push(ast.Node{
-					Kind: ast.Float,
+				return p.builder.Push(Node{
+					Kind: Float,
 					Data: b[:i+3],
 				}), b[i+3:], nil
 			}
 
-			return ast.InvalidReference, nil, newDecodeError(b[i:i+1], "unexpected character 'n' while scanning for a number")
+			return invalidReference, nil, NewParserError(b[i:i+1], "unexpected character 'n' while scanning for a number")
 		}
 
 		break
 	}
 
 	if i == 0 {
-		return ast.InvalidReference, b, newDecodeError(b, "incomplete number")
+		return invalidReference, b, NewParserError(b, "incomplete number")
 	}
 
-	kind := ast.Integer
+	kind := Integer
 
 	if isFloat {
-		kind = ast.Float
+		kind = Float
 	}
 
-	return p.builder.Push(ast.Node{
+	return p.builder.Push(Node{
 		Kind: kind,
 		Data: b[:i],
 	}), b[i:], nil
@@ -1075,11 +1228,11 @@ func isValidBinaryRune(r byte) bool {
 
 func expect(x byte, b []byte) ([]byte, error) {
 	if len(b) == 0 {
-		return nil, newDecodeError(b, "expected character %c but the document ended here", x)
+		return nil, NewParserError(b, "expected character %c but the document ended here", x)
 	}
 
 	if b[0] != x {
-		return nil, newDecodeError(b[0:1], "expected character %c", x)
+		return nil, NewParserError(b[0:1], "expected character %c", x)
 	}
 
 	return b[1:], nil

unstable/parser_test.go

@@ -0,0 +1,629 @@
+package unstable
+
+import (
+	"fmt"
+	"strconv"
+	"strings"
+	"testing"
+
+	"github.com/stretchr/testify/require"
+)
+
+func TestParser_AST_Numbers(t *testing.T) {
+	examples := []struct {
+		desc  string
+		input string
+		kind  Kind
+		err   bool
+	}{
+		{
+			desc:  "integer just digits",
+			input: `1234`,
+			kind:  Integer,
+		},
+		{
+			desc:  "integer zero",
+			input: `0`,
+			kind:  Integer,
+		},
+		{
+			desc:  "integer sign",
+			input: `+99`,
+			kind:  Integer,
+		},
+		{
+			desc:  "integer hex uppercase",
+			input: `0xDEADBEEF`,
+			kind:  Integer,
+		},
+		{
+			desc:  "integer hex lowercase",
+			input: `0xdead_beef`,
+			kind:  Integer,
+		},
+		{
+			desc:  "integer octal",
+			input: `0o01234567`,
+			kind:  Integer,
+		},
+		{
+			desc:  "integer binary",
+			input: `0b11010110`,
+			kind:  Integer,
+		},
+		{
+			desc:  "float zero",
+			input: `0.0`,
+			kind:  Float,
+		},
+		{
+			desc:  "float positive zero",
+			input: `+0.0`,
+			kind:  Float,
+		},
+		{
+			desc:  "float negative zero",
+			input: `-0.0`,
+			kind:  Float,
+		},
+		{
+			desc:  "float pi",
+			input: `3.1415`,
+			kind:  Float,
+		},
+		{
+			desc:  "float negative",
+			input: `-0.01`,
+			kind:  Float,
+		},
+		{
+			desc:  "float signed exponent",
+			input: `5e+22`,
+			kind:  Float,
+		},
+		{
+			desc:  "float exponent lowercase",
+			input: `1e06`,
+			kind:  Float,
+		},
+		{
+			desc:  "float exponent uppercase",
+			input: `-2E-2`,
+			kind:  Float,
+		},
+		{
+			desc:  "float fractional with exponent",
+			input: `6.626e-34`,
+			kind:  Float,
+		},
+		{
+			desc:  "float underscores",
+			input: `224_617.445_991_228`,
+			kind:  Float,
+		},
+		{
+			desc:  "inf",
+			input: `inf`,
+			kind:  Float,
+		},
+		{
+			desc:  "inf negative",
+			input: `-inf`,
+			kind:  Float,
+		},
+		{
+			desc:  "inf positive",
+			input: `+inf`,
+			kind:  Float,
+		},
+		{
+			desc:  "nan",
+			input: `nan`,
+			kind:  Float,
+		},
+		{
+			desc:  "nan negative",
+			input: `-nan`,
+			kind:  Float,
+		},
+		{
+			desc:  "nan positive",
+			input: `+nan`,
+			kind:  Float,
+		},
+	}
+
+	for _, e := range examples {
+		e := e
+		t.Run(e.desc, func(t *testing.T) {
+			p := Parser{}
+			p.Reset([]byte(`A = ` + e.input))
+			p.NextExpression()
+			err := p.Error()
+			if e.err {
+				require.Error(t, err)
+			} else {
+				require.NoError(t, err)
+
+				expected := astNode{
+					Kind: KeyValue,
+					Children: []astNode{
+						{Kind: e.kind, Data: []byte(e.input)},
+						{Kind: Key, Data: []byte(`A`)},
+					},
+				}
+				compareNode(t, expected, p.Expression())
+			}
+		})
+	}
+}
+
+type (
+	astNode struct {
+		Kind     Kind
+		Data     []byte
+		Children []astNode
+	}
+)
+
+func compareNode(t *testing.T, e astNode, n *Node) {
+	t.Helper()
+	require.Equal(t, e.Kind, n.Kind)
+	require.Equal(t, e.Data, n.Data)
+
+	compareIterator(t, e.Children, n.Children())
+}
+
+func compareIterator(t *testing.T, expected []astNode, actual Iterator) {
+	t.Helper()
+	idx := 0
+
+	for actual.Next() {
+		n := actual.Node()
+
+		if idx >= len(expected) {
+			t.Fatal("extra child in actual tree")
+		}
+		e := expected[idx]
+
+		compareNode(t, e, n)
+
+		idx++
+	}
+
+	if idx < len(expected) {
+		t.Fatal("missing children in actual", "idx =", idx, "expected =", len(expected))
+	}
+}
+
+//nolint:funlen
+func TestParser_AST(t *testing.T) {
+	examples := []struct {
+		desc  string
+		input string
+		ast   astNode
+		err   bool
+	}{
+		{
+			desc:  "simple string assignment",
+			input: `A = "hello"`,
+			ast: astNode{
+				Kind: KeyValue,
+				Children: []astNode{
+					{
+						Kind: String,
+						Data: []byte(`hello`),
+					},
+					{
+						Kind: Key,
+						Data: []byte(`A`),
+					},
+				},
+			},
+		},
+		{
+			desc:  "simple bool assignment",
+			input: `A = true`,
+			ast: astNode{
+				Kind: KeyValue,
+				Children: []astNode{
+					{
+						Kind: Bool,
+						Data: []byte(`true`),
+					},
+					{
+						Kind: Key,
+						Data: []byte(`A`),
+					},
+				},
+			},
+		},
+		{
+			desc:  "array of strings",
+			input: `A = ["hello", ["world", "again"]]`,
+			ast: astNode{
+				Kind: KeyValue,
+				Children: []astNode{
+					{
+						Kind: Array,
+						Children: []astNode{
+							{
+								Kind: String,
+								Data: []byte(`hello`),
+							},
+							{
+								Kind: Array,
+								Children: []astNode{
+									{
+										Kind: String,
+										Data: []byte(`world`),
+									},
+									{
+										Kind: String,
+										Data: []byte(`again`),
+									},
+								},
+							},
+						},
+					},
+					{
+						Kind: Key,
+						Data: []byte(`A`),
+					},
+				},
+			},
+		},
+		{
+			desc:  "array of arrays of strings",
+			input: `A = ["hello", "world"]`,
+			ast: astNode{
+				Kind: KeyValue,
+				Children: []astNode{
+					{
+						Kind: Array,
+						Children: []astNode{
+							{
+								Kind: String,
+								Data: []byte(`hello`),
+							},
+							{
+								Kind: String,
+								Data: []byte(`world`),
+							},
+						},
+					},
+					{
+						Kind: Key,
+						Data: []byte(`A`),
+					},
+				},
+			},
+		},
+		{
+			desc:  "inline table",
+			input: `name = { first = "Tom", last = "Preston-Werner" }`,
+			ast: astNode{
+				Kind: KeyValue,
+				Children: []astNode{
+					{
+						Kind: InlineTable,
+						Children: []astNode{
+							{
+								Kind: KeyValue,
+								Children: []astNode{
+									{Kind: String, Data: []byte(`Tom`)},
+									{Kind: Key, Data: []byte(`first`)},
+								},
+							},
+							{
+								Kind: KeyValue,
+								Children: []astNode{
+									{Kind: String, Data: []byte(`Preston-Werner`)},
+									{Kind: Key, Data: []byte(`last`)},
+								},
+							},
+						},
+					},
+					{
+						Kind: Key,
+						Data: []byte(`name`),
+					},
+				},
+			},
+		},
+	}
+
+	for _, e := range examples {
+		e := e
+		t.Run(e.desc, func(t *testing.T) {
+			p := Parser{}
+			p.Reset([]byte(e.input))
+			p.NextExpression()
+			err := p.Error()
+			if e.err {
+				require.Error(t, err)
+			} else {
+				require.NoError(t, err)
+				compareNode(t, e.ast, p.Expression())
+			}
+		})
+	}
+}
+
+func BenchmarkParseBasicStringWithUnicode(b *testing.B) {
+	p := &Parser{}
+	b.Run("4", func(b *testing.B) {
+		input := []byte(`"\u1234\u5678\u9ABC\u1234\u5678\u9ABC"`)
+		b.ReportAllocs()
+		b.SetBytes(int64(len(input)))
+
+		for i := 0; i < b.N; i++ {
+			p.parseBasicString(input)
+		}
+	})
+	b.Run("8", func(b *testing.B) {
+		input := []byte(`"\u12345678\u9ABCDEF0\u12345678\u9ABCDEF0"`)
+		b.ReportAllocs()
+		b.SetBytes(int64(len(input)))
+
+		for i := 0; i < b.N; i++ {
+			p.parseBasicString(input)
+		}
+	})
+}
+
+func BenchmarkParseBasicStringsEasy(b *testing.B) {
+	p := &Parser{}
+
+	for _, size := range []int{1, 4, 8, 16, 21} {
+		b.Run(strconv.Itoa(size), func(b *testing.B) {
+			input := []byte(`"` + strings.Repeat("A", size) + `"`)
+
+			b.ReportAllocs()
+			b.SetBytes(int64(len(input)))
+
+			for i := 0; i < b.N; i++ {
+				p.parseBasicString(input)
+			}
+		})
+	}
+}
+
+func TestParser_AST_DateTimes(t *testing.T) {
+	examples := []struct {
+		desc  string
+		input string
+		kind  Kind
+		err   bool
+	}{
+		{
+			desc:  "offset-date-time with delim 'T' and UTC offset",
+			input: `2021-07-21T12:08:05Z`,
+			kind:  DateTime,
+		},
+		{
+			desc:  "offset-date-time with space delim and +8hours offset",
+			input: `2021-07-21 12:08:05+08:00`,
+			kind:  DateTime,
+		},
+		{
+			desc:  "local-date-time with nano second",
+			input: `2021-07-21T12:08:05.666666666`,
+			kind:  LocalDateTime,
+		},
+		{
+			desc:  "local-date-time",
+			input: `2021-07-21T12:08:05`,
+			kind:  LocalDateTime,
+		},
+		{
+			desc:  "local-date",
+			input: `2021-07-21`,
+			kind:  LocalDate,
+		},
+	}
+
+	for _, e := range examples {
+		e := e
+		t.Run(e.desc, func(t *testing.T) {
+			p := Parser{}
+			p.Reset([]byte(`A = ` + e.input))
+			p.NextExpression()
+			err := p.Error()
+			if e.err {
+				require.Error(t, err)
+			} else {
+				require.NoError(t, err)
+
+				expected := astNode{
+					Kind: KeyValue,
+					Children: []astNode{
+						{Kind: e.kind, Data: []byte(e.input)},
+						{Kind: Key, Data: []byte(`A`)},
+					},
+				}
+				compareNode(t, expected, p.Expression())
+			}
+		})
+	}
+}
+
+// This example demonstrates how to parse a TOML document and preserving
+// comments.  Comments are stored in the AST as Comment nodes. This example
+// displays the structure of the full AST generated by the parser using the
+// following structure:
+//
+//  1. Each root-level expression is separated by three dashes.
+//  2. Bytes associated to a node are displayed in square brackets.
+//  3. Siblings have the same indentation.
+//  4. Children of a node are indented one level.
+func ExampleParser_comments() {
+	doc := `# Top of the document comment.
+# Optional, any amount of lines.
+
+# Above table.
+[table] # Next to table.
+# Above simple value.
+key = "value" # Next to simple value.
+# Below simple value.
+
+# Some comment alone.
+
+# Multiple comments, on multiple lines.
+
+# Above inline table.
+name = { first = "Tom", last = "Preston-Werner" } # Next to inline table.
+# Below inline table.
+
+# Above array.
+array = [ 1, 2, 3 ] # Next to one-line array.
+# Below array.
+
+# Above multi-line array.
+key5 = [ # Next to start of inline array.
+  # Second line before array content.
+  1, # Next to first element.
+  # After first element.
+  # Before second element.
+  2,
+  3, # Next to last element
+  # After last element.
+] # Next to end of array.
+# Below multi-line array.
+
+# Before array table.
+[[products]] # Next to array table.
+# After array table.
+`
+
+	var printGeneric func(*Parser, int, *Node)
+	printGeneric = func(p *Parser, indent int, e *Node) {
+		if e == nil {
+			return
+		}
+		s := p.Shape(e.Raw)
+		x := fmt.Sprintf("%d:%d->%d:%d (%d->%d)", s.Start.Line, s.Start.Column, s.End.Line, s.End.Column, s.Start.Offset, s.End.Offset)
+		fmt.Printf("%-25s | %s%s [%s]\n", x, strings.Repeat("  ", indent), e.Kind, e.Data)
+		printGeneric(p, indent+1, e.Child())
+		printGeneric(p, indent, e.Next())
+	}
+
+	printTree := func(p *Parser) {
+		for p.NextExpression() {
+			e := p.Expression()
+			fmt.Println("---")
+			printGeneric(p, 0, e)
+		}
+		if err := p.Error(); err != nil {
+			panic(err)
+		}
+	}
+
+	p := &Parser{
+		KeepComments: true,
+	}
+	p.Reset([]byte(doc))
+	printTree(p)
+
+	// Output:
+	// ---
+	// 1:1->1:31 (0->30)         | Comment [# Top of the document comment.]
+	// ---
+	// 2:1->2:33 (31->63)        | Comment [# Optional, any amount of lines.]
+	// ---
+	// 4:1->4:15 (65->79)        | Comment [# Above table.]
+	// ---
+	// 1:1->1:1 (0->0)           | Table []
+	// 5:2->5:7 (81->86)         |   Key [table]
+	// 5:9->5:25 (88->104)       | Comment [# Next to table.]
+	// ---
+	// 6:1->6:22 (105->126)      | Comment [# Above simple value.]
+	// ---
+	// 1:1->1:1 (0->0)           | KeyValue []
+	// 7:7->7:14 (133->140)      |   String [value]
+	// 7:1->7:4 (127->130)       |   Key [key]
+	// 7:15->7:38 (141->164)     | Comment [# Next to simple value.]
+	// ---
+	// 8:1->8:22 (165->186)      | Comment [# Below simple value.]
+	// ---
+	// 10:1->10:22 (188->209)    | Comment [# Some comment alone.]
+	// ---
+	// 12:1->12:40 (211->250)    | Comment [# Multiple comments, on multiple lines.]
+	// ---
+	// 14:1->14:22 (252->273)    | Comment [# Above inline table.]
+	// ---
+	// 1:1->1:1 (0->0)           | KeyValue []
+	// 15:8->15:9 (281->282)     |   InlineTable []
+	// 1:1->1:1 (0->0)           |     KeyValue []
+	// 15:18->15:23 (291->296)   |       String [Tom]
+	// 15:10->15:15 (283->288)   |       Key [first]
+	// 1:1->1:1 (0->0)           |     KeyValue []
+	// 15:32->15:48 (305->321)   |       String [Preston-Werner]
+	// 15:25->15:29 (298->302)   |       Key [last]
+	// 15:1->15:5 (274->278)     |   Key [name]
+	// 15:51->15:74 (324->347)   | Comment [# Next to inline table.]
+	// ---
+	// 16:1->16:22 (348->369)    | Comment [# Below inline table.]
+	// ---
+	// 18:1->18:15 (371->385)    | Comment [# Above array.]
+	// ---
+	// 1:1->1:1 (0->0)           | KeyValue []
+	// 1:1->1:1 (0->0)           |   Array []
+	// 1:1->1:1 (0->0)           |     Integer [1]
+	// 1:1->1:1 (0->0)           |     Integer [2]
+	// 1:1->1:1 (0->0)           |     Integer [3]
+	// 19:1->19:6 (386->391)     |   Key [array]
+	// 19:21->19:46 (406->431)   | Comment [# Next to one-line array.]
+	// ---
+	// 20:1->20:15 (432->446)    | Comment [# Below array.]
+	// ---
+	// 22:1->22:26 (448->473)    | Comment [# Above multi-line array.]
+	// ---
+	// 1:1->1:1 (0->0)           | KeyValue []
+	// 1:1->1:1 (0->0)           |   Array []
+	// 23:10->23:42 (483->515)   |     Comment [# Next to start of inline array.]
+	// 24:3->24:38 (518->553)    |       Comment [# Second line before array content.]
+	// 1:1->1:1 (0->0)           |     Integer [1]
+	// 25:6->25:30 (559->583)    |     Comment [# Next to first element.]
+	// 26:3->26:25 (586->608)    |       Comment [# After first element.]
+	// 27:3->27:27 (611->635)    |       Comment [# Before second element.]
+	// 1:1->1:1 (0->0)           |     Integer [2]
+	// 1:1->1:1 (0->0)           |     Integer [3]
+	// 29:6->29:28 (646->668)    |     Comment [# Next to last element]
+	// 30:3->30:24 (671->692)    |       Comment [# After last element.]
+	// 23:1->23:5 (474->478)     |   Key [key5]
+	// 31:3->31:26 (695->718)    | Comment [# Next to end of array.]
+	// ---
+	// 32:1->32:26 (719->744)    | Comment [# Below multi-line array.]
+	// ---
+	// 34:1->34:22 (746->767)    | Comment [# Before array table.]
+	// ---
+	// 1:1->1:1 (0->0)           | ArrayTable []
+	// 35:3->35:11 (770->778)    |   Key [products]
+	// 35:14->35:36 (781->803)   | Comment [# Next to array table.]
+	// ---
+	// 36:1->36:21 (804->824)    | Comment [# After array table.]
+}
+
+func ExampleParser() {
+	doc := `
+	hello = "world"
+	value = 42
+	`
+	p := Parser{}
+	p.Reset([]byte(doc))
+	for p.NextExpression() {
+		e := p.Expression()
+		fmt.Printf("Expression: %s\n", e.Kind)
+		value := e.Value()
+		it := e.Key()
+		k := it.Node() // shortcut: we know there is no dotted key in the example
+		fmt.Printf("%s -> (%s) %s\n", k.Data, value.Kind, value.Data)
+	}
+
+	// Output:
+	// Expression: KeyValue
+	// hello -> (String) world
+	// Expression: KeyValue
+	// value -> (Integer) 42
+}

unstable/scanner.go

@@ -1,4 +1,6 @@
-package toml
+package unstable
+
+import "github.com/pelletier/go-toml/v2/internal/characters"
 
 func scanFollows(b []byte, pattern string) bool {
 	n := len(pattern)
@@ -54,16 +56,16 @@ func scanLiteralString(b []byte) ([]byte, []byte, error) {
 		case '\'':
 			return b[:i+1], b[i+1:], nil
 		case '\n', '\r':
-			return nil, nil, newDecodeError(b[i:i+1], "literal strings cannot have new lines")
+			return nil, nil, NewParserError(b[i:i+1], "literal strings cannot have new lines")
 		}
-		size := utf8ValidNext(b[i:])
+		size := characters.Utf8ValidNext(b[i:])
 		if size == 0 {
-			return nil, nil, newDecodeError(b[i:i+1], "invalid character")
+			return nil, nil, NewParserError(b[i:i+1], "invalid character")
 		}
 		i += size
 	}
 
-	return nil, nil, newDecodeError(b[len(b):], "unterminated literal string")
+	return nil, nil, NewParserError(b[len(b):], "unterminated literal string")
 }
 
 func scanMultilineLiteralString(b []byte) ([]byte, []byte, error) {
@@ -98,39 +100,39 @@ func scanMultilineLiteralString(b []byte) ([]byte, []byte, error) {
 				i++
 
 				if i < len(b) && b[i] == '\'' {
-					return nil, nil, newDecodeError(b[i-3:i+1], "''' not allowed in multiline literal string")
+					return nil, nil, NewParserError(b[i-3:i+1], "''' not allowed in multiline literal string")
 				}
 
 				return b[:i], b[i:], nil
 			}
 		case '\r':
 			if len(b) < i+2 {
-				return nil, nil, newDecodeError(b[len(b):], `need a \n after \r`)
+				return nil, nil, NewParserError(b[len(b):], `need a \n after \r`)
 			}
 			if b[i+1] != '\n' {
-				return nil, nil, newDecodeError(b[i:i+2], `need a \n after \r`)
+				return nil, nil, NewParserError(b[i:i+2], `need a \n after \r`)
 			}
 			i += 2 // skip the \n
 			continue
 		}
-		size := utf8ValidNext(b[i:])
+		size := characters.Utf8ValidNext(b[i:])
 		if size == 0 {
-			return nil, nil, newDecodeError(b[i:i+1], "invalid character")
+			return nil, nil, NewParserError(b[i:i+1], "invalid character")
 		}
 		i += size
 	}
 
-	return nil, nil, newDecodeError(b[len(b):], `multiline literal string not terminated by '''`)
+	return nil, nil, NewParserError(b[len(b):], `multiline literal string not terminated by '''`)
 }
 
 func scanWindowsNewline(b []byte) ([]byte, []byte, error) {
 	const lenCRLF = 2
 	if len(b) < lenCRLF {
-		return nil, nil, newDecodeError(b, "windows new line expected")
+		return nil, nil, NewParserError(b, "windows new line expected")
 	}
 
 	if b[1] != '\n' {
-		return nil, nil, newDecodeError(b, `windows new line should be \r\n`)
+		return nil, nil, NewParserError(b, `windows new line should be \r\n`)
 	}
 
 	return b[:lenCRLF], b[lenCRLF:], nil
@@ -149,7 +151,6 @@ func scanWhitespace(b []byte) ([]byte, []byte) {
 	return b, b[len(b):]
 }
 
-//nolint:unparam
 func scanComment(b []byte) ([]byte, []byte, error) {
 	// comment-start-symbol = %x23 ; #
 	// non-ascii = %x80-D7FF / %xE000-10FFFF
@@ -165,11 +166,11 @@ func scanComment(b []byte) ([]byte, []byte, error) {
 			if i+1 < len(b) && b[i+1] == '\n' {
 				return b[:i+1], b[i+1:], nil
 			}
-			return nil, nil, newDecodeError(b[i:i+1], "invalid character in comment")
+			return nil, nil, NewParserError(b[i:i+1], "invalid character in comment")
 		}
-		size := utf8ValidNext(b[i:])
+		size := characters.Utf8ValidNext(b[i:])
 		if size == 0 {
-			return nil, nil, newDecodeError(b[i:i+1], "invalid character in comment")
+			return nil, nil, NewParserError(b[i:i+1], "invalid character in comment")
 		}
 
 		i += size
@@ -192,17 +193,17 @@ func scanBasicString(b []byte) ([]byte, bool, []byte, error) {
 		case '"':
 			return b[:i+1], escaped, b[i+1:], nil
 		case '\n', '\r':
-			return nil, escaped, nil, newDecodeError(b[i:i+1], "basic strings cannot have new lines")
+			return nil, escaped, nil, NewParserError(b[i:i+1], "basic strings cannot have new lines")
 		case '\\':
 			if len(b) < i+2 {
-				return nil, escaped, nil, newDecodeError(b[i:i+1], "need a character after \\")
+				return nil, escaped, nil, NewParserError(b[i:i+1], "need a character after \\")
 			}
 			escaped = true
 			i++ // skip the next character
 		}
 	}
 
-	return nil, escaped, nil, newDecodeError(b[len(b):], `basic string not terminated by "`)
+	return nil, escaped, nil, NewParserError(b[len(b):], `basic string not terminated by "`)
 }
 
 func scanMultilineBasicString(b []byte) ([]byte, bool, []byte, error) {
@@ -243,27 +244,27 @@ func scanMultilineBasicString(b []byte) ([]byte, bool, []byte, error) {
 				i++
 
 				if i < len(b) && b[i] == '"' {
-					return nil, escaped, nil, newDecodeError(b[i-3:i+1], `""" not allowed in multiline basic string`)
+					return nil, escaped, nil, NewParserError(b[i-3:i+1], `""" not allowed in multiline basic string`)
 				}
 
 				return b[:i], escaped, b[i:], nil
 			}
 		case '\\':
 			if len(b) < i+2 {
-				return nil, escaped, nil, newDecodeError(b[len(b):], "need a character after \\")
+				return nil, escaped, nil, NewParserError(b[len(b):], "need a character after \\")
 			}
 			escaped = true
 			i++ // skip the next character
 		case '\r':
 			if len(b) < i+2 {
-				return nil, escaped, nil, newDecodeError(b[len(b):], `need a \n after \r`)
+				return nil, escaped, nil, NewParserError(b[len(b):], `need a \n after \r`)
 			}
 			if b[i+1] != '\n' {
-				return nil, escaped, nil, newDecodeError(b[i:i+2], `need a \n after \r`)
+				return nil, escaped, nil, NewParserError(b[i:i+2], `need a \n after \r`)
 			}
 			i++ // skip the \n
 		}
 	}
 
-	return nil, escaped, nil, newDecodeError(b[len(b):], `multiline basic string not terminated by """`)
+	return nil, escaped, nil, NewParserError(b[len(b):], `multiline basic string not terminated by """`)
 }