Document conversion internals and limits

README.md

@@ -111,6 +111,26 @@ id	items[0].qty	items[0].sku	user.name
 逆方向の変換では、`user.name` や `items[0].sku` のような列名から入れ子の
 map/arrayを復元します。
 
+列パスは `.` でmapのキー、`[n]` で0始まりの配列indexを表します。たとえば
+`orders[0].items[1].sku` は、最初の注文に含まれる2番目の商品の `sku` です。
+
+同じ行に `user` と `user.name` のような競合する列がある場合、先に読み込まれた
+値を保持し、後続列で型を上書きしません。曖昧な復元を避けるため、親要素と子要素を
+同時に列として置かないでください。
+
+## セル値の型推定
+
+CSV、TSV、XLSXから構造化形式へ戻す際は、セル文字列を次の順で推定します。
+
+- 空文字は空文字列
+- `true` / `false` はboolean
+- 通常の整数は64-bit integer
+- 小数点または指数表記を含む数値は64-bit floating point
+- それ以外は文字列
+
+郵便番号や商品コードを想定し、`00123` や `-01` のようなゼロ埋め値は文字列のまま
+保持します。日付、時刻、`null` は自動推定しません。
+
 ## CLIオプション
 
 - `-i`: 入力ファイル。省略時はstdin。
@@ -125,6 +145,8 @@ map/arrayを復元します。
 - 表形式では1行目をヘッダーとして扱います。
 - XLSXは指定した1シートのみ読み書きします。
 - セル値の型推定は、空文字、真偽値、整数、小数、文字列の範囲です。
+- 空のmap/arrayは表側で `{}` / `[]` と表示されますが、逆変換時は文字列になります。
+- mapキーに `.`、`[`、`]` を含む場合のescape記法は未対応です。
 - 複雑なExcel書式や数式の保持は目的外です。
 
 ## 開発者向け情報

cmd/dataxl/conversion.go

@@ -0,0 +1,71 @@
+package main
+
+import (
+	"fmt"
+	"path/filepath"
+	"strings"
+)
+
+// convert routes data through either the structured-value or table internal
+// representation. Keeping this matrix in one place makes format adapters
+// independent from CLI and file I/O concerns.
+func convert(input []byte, from, to, sheet string, pretty bool) ([]byte, error) {
+	switch {
+	case isTabular(from) && isTabular(to):
+		t, err := parseTable(input, from, sheet)
+		if err != nil {
+			return nil, err
+		}
+		return encodeTable(t, to, sheet)
+	case isTabular(from):
+		t, err := parseTable(input, from, sheet)
+		if err != nil {
+			return nil, err
+		}
+		return encodeStructured(tableToRecords(t), to, pretty)
+	case isTabular(to):
+		value, err := parseStructured(input, from)
+		if err != nil {
+			return nil, err
+		}
+		return encodeTable(valueToTable(value), to, sheet)
+	default:
+		value, err := parseStructured(input, from)
+		if err != nil {
+			return nil, err
+		}
+		return encodeStructured(value, to, pretty)
+	}
+}
+
+func normalizeFormat(format string) string {
+	format = strings.ToLower(strings.TrimPrefix(strings.TrimSpace(format), "."))
+	switch format {
+	case "yml":
+		return "yaml"
+	case "xlsm", "xls":
+		return "xlsx"
+	default:
+		return format
+	}
+}
+
+func inferFormat(path string) string {
+	if path == "" || path == "-" {
+		return ""
+	}
+	return normalizeFormat(filepath.Ext(path))
+}
+
+func validateFormat(format string) error {
+	switch format {
+	case "json", "yaml", "toml", "csv", "tsv", "xlsx":
+		return nil
+	default:
+		return fmt.Errorf("unsupported format %q", format)
+	}
+}
+
+func isTabular(format string) bool {
+	return format == "csv" || format == "tsv" || format == "xlsx"
+}

cmd/dataxl/doc.go

@@ -0,0 +1,7 @@
+// Package main implements the dataxl command-line converter.
+//
+// Conversion uses two internal representations: structured Go values for
+// JSON/YAML/TOML and table for CSV/TSV/XLSX. Nested structured values cross the
+// boundary through dotted and indexed column paths such as user.name and
+// items[0].sku.
+package main

cmd/dataxl/main.go

@@ -1,25 +1,15 @@
 package main
 
 import (
-	"bytes"
-	"encoding/csv"
-	"encoding/json"
-	"errors"
 	"flag"
 	"fmt"
 	"io"
 	"os"
-	"path/filepath"
-	"regexp"
-	"sort"
-	"strconv"
 	"strings"
-
-	"github.com/BurntSushi/toml"
-	"github.com/xuri/excelize/v2"
-	"gopkg.in/yaml.v3"
 )
 
+// options contains CLI concerns only. Conversion functions receive explicit
+// arguments so they can be reused and tested without constructing a FlagSet.
 type options struct {
 	inFile  string
 	outFile string
@@ -29,11 +19,6 @@ type options struct {
 	pretty  bool
 }
 
-type table struct {
-	Header []string
-	Rows   [][]string
-}
-
 func main() {
 	if err := run(os.Args[1:], os.Stdin, os.Stdout, os.Stderr); err != nil {
 		fmt.Fprintln(os.Stderr, "dataxl:", err)
@@ -42,6 +27,26 @@ func main() {
 }
 
 func run(args []string, stdin io.Reader, stdout, stderr io.Writer) error {
+	opt, err := parseOptions(args, stderr)
+	if err != nil {
+		return err
+	}
+	if err := opt.resolveFormats(); err != nil {
+		return err
+	}
+
+	input, err := readInput(opt.inFile, stdin)
+	if err != nil {
+		return err
+	}
+	output, err := convert(input, opt.from, opt.to, opt.sheet, opt.pretty)
+	if err != nil {
+		return err
+	}
+	return writeOutput(opt.outFile, stdout, output)
+}
+
+func parseOptions(args []string, stderr io.Writer) (options, error) {
 	var opt options
 	fs := flag.NewFlagSet("dataxl", flag.ContinueOnError)
 	fs.SetOutput(stderr)
@@ -66,16 +71,15 @@ Notes:
   converting back to json/yaml/toml.`)
 	}
 	if err := fs.Parse(args); err != nil {
-		return err
+		return options{}, err
 	}
 	if fs.NArg() != 0 {
-		return fmt.Errorf("unexpected arguments: %s", strings.Join(fs.Args(), " "))
+		return options{}, fmt.Errorf("unexpected arguments: %s", strings.Join(fs.Args(), " "))
 	}
+	return opt, nil
+}
 
-	input, err := readInput(opt.inFile, stdin)
-	if err != nil {
-		return err
-	}
+func (opt *options) resolveFormats() error {
 	opt.from = normalizeFormat(opt.from)
 	opt.to = normalizeFormat(opt.to)
 	if opt.from == "" {
@@ -85,60 +89,12 @@ Notes:
 		opt.to = inferFormat(opt.outFile)
 	}
 	if opt.from == "" || opt.to == "" {
-		return errors.New("both -from and -to are required when a format cannot be inferred from file names")
+		return fmt.Errorf("both -from and -to are required when a format cannot be inferred from file names")
 	}
 	if err := validateFormat(opt.from); err != nil {
 		return err
 	}
-	if err := validateFormat(opt.to); err != nil {
-		return err
-	}
-
-	var out []byte
-	// Use table as the common representation whenever either side is a
-	// spreadsheet-like format. Structured formats can then share the same
-	// flatten/restore path for CSV, TSV, and XLSX.
-	if isTabular(opt.from) && isTabular(opt.to) {
-		t, err := parseTable(input, opt.from, opt.sheet)
-		if err != nil {
-			return err
-		}
-		out, err = encodeTable(t, opt.to, opt.sheet)
-		if err != nil {
-			return err
-		}
-	} else if isTabular(opt.from) {
-		t, err := parseTable(input, opt.from, opt.sheet)
-		if err != nil {
-			return err
-		}
-		records := tableToRecords(t)
-		out, err = encodeStructured(records, opt.to, opt.pretty)
-		if err != nil {
-			return err
-		}
-	} else if isTabular(opt.to) {
-		value, err := parseStructured(input, opt.from)
-		if err != nil {
-			return err
-		}
-		t := valueToTable(value)
-		out, err = encodeTable(t, opt.to, opt.sheet)
-		if err != nil {
-			return err
-		}
-	} else {
-		value, err := parseStructured(input, opt.from)
-		if err != nil {
-			return err
-		}
-		out, err = encodeStructured(value, opt.to, opt.pretty)
-		if err != nil {
-			return err
-		}
-	}
-
-	return writeOutput(opt.outFile, stdout, out)
+	return validateFormat(opt.to)
 }
 
 func readInput(path string, stdin io.Reader) ([]byte, error) {
@@ -155,459 +111,3 @@ func writeOutput(path string, stdout io.Writer, data []byte) error {
 	}
 	return os.WriteFile(path, data, 0o644)
 }
-
-func normalizeFormat(format string) string {
-	format = strings.ToLower(strings.TrimPrefix(strings.TrimSpace(format), "."))
-	switch format {
-	case "yml":
-		return "yaml"
-	case "xlsm", "xls":
-		return "xlsx"
-	default:
-		return format
-	}
-}
-
-func inferFormat(path string) string {
-	if path == "" || path == "-" {
-		return ""
-	}
-	return normalizeFormat(strings.TrimPrefix(filepath.Ext(path), "."))
-}
-
-func validateFormat(format string) error {
-	switch format {
-	case "json", "yaml", "toml", "csv", "tsv", "xlsx":
-		return nil
-	default:
-		return fmt.Errorf("unsupported format %q", format)
-	}
-}
-
-func isTabular(format string) bool {
-	return format == "csv" || format == "tsv" || format == "xlsx"
-}
-
-func parseStructured(input []byte, format string) (any, error) {
-	var value any
-	switch format {
-	case "json":
-		dec := json.NewDecoder(bytes.NewReader(input))
-		dec.UseNumber()
-		if err := dec.Decode(&value); err != nil {
-			return nil, err
-		}
-	case "yaml":
-		if err := yaml.Unmarshal(input, &value); err != nil {
-			return nil, err
-		}
-		value = normalizeYAML(value)
-	case "toml":
-		var m map[string]any
-		if err := toml.Unmarshal(input, &m); err != nil {
-			return nil, err
-		}
-		value = m
-	default:
-		return nil, fmt.Errorf("format %q is not structured", format)
-	}
-	return value, nil
-}
-
-func encodeStructured(value any, format string, pretty bool) ([]byte, error) {
-	switch format {
-	case "json":
-		if pretty {
-			return json.MarshalIndent(value, "", "  ")
-		}
-		return json.Marshal(value)
-	case "yaml":
-		return yaml.Marshal(value)
-	case "toml":
-		m, ok := value.(map[string]any)
-		if !ok {
-			m = map[string]any{"rows": value}
-		}
-		var b bytes.Buffer
-		err := toml.NewEncoder(&b).Encode(m)
-		return b.Bytes(), err
-	default:
-		return nil, fmt.Errorf("format %q is not structured", format)
-	}
-}
-
-func normalizeYAML(value any) any {
-	switch v := value.(type) {
-	case map[string]any:
-		out := make(map[string]any, len(v))
-		for key, child := range v {
-			out[key] = normalizeYAML(child)
-		}
-		return out
-	case map[any]any:
-		out := make(map[string]any, len(v))
-		for key, child := range v {
-			out[fmt.Sprint(key)] = normalizeYAML(child)
-		}
-		return out
-	case []any:
-		for i := range v {
-			v[i] = normalizeYAML(v[i])
-		}
-	}
-	return value
-}
-
-func parseTable(input []byte, format, sheet string) (table, error) {
-	switch format {
-	case "csv":
-		return readDelimited(input, ',')
-	case "tsv":
-		return readDelimited(input, '\t')
-	case "xlsx":
-		f, err := excelize.OpenReader(bytes.NewReader(input))
-		if err != nil {
-			return table{}, err
-		}
-		defer f.Close()
-		rows, err := f.GetRows(sheet)
-		if err != nil {
-			return table{}, err
-		}
-		return rowsToTable(rows), nil
-	default:
-		return table{}, fmt.Errorf("format %q is not tabular", format)
-	}
-}
-
-func readDelimited(input []byte, comma rune) (table, error) {
-	r := csv.NewReader(bytes.NewReader(input))
-	r.Comma = comma
-	r.FieldsPerRecord = -1
-	r.TrimLeadingSpace = comma == ','
-	rows, err := r.ReadAll()
-	if err != nil {
-		return table{}, err
-	}
-	return rowsToTable(rows), nil
-}
-
-func rowsToTable(rows [][]string) table {
-	if len(rows) == 0 {
-		return table{}
-	}
-	width := 0
-	for _, row := range rows {
-		if len(row) > width {
-			width = len(row)
-		}
-	}
-	header := padRow(rows[0], width)
-	body := make([][]string, 0, len(rows)-1)
-	for _, row := range rows[1:] {
-		body = append(body, padRow(row, width))
-	}
-	return table{Header: header, Rows: body}
-}
-
-func padRow(row []string, width int) []string {
-	out := make([]string, width)
-	copy(out, row)
-	return out
-}
-
-func encodeTable(t table, format, sheet string) ([]byte, error) {
-	switch format {
-	case "csv":
-		return writeDelimited(t, ',')
-	case "tsv":
-		return writeDelimited(t, '\t')
-	case "xlsx":
-		return writeXLSX(t, sheet)
-	default:
-		return nil, fmt.Errorf("format %q is not tabular", format)
-	}
-}
-
-func writeDelimited(t table, comma rune) ([]byte, error) {
-	var b bytes.Buffer
-	w := csv.NewWriter(&b)
-	w.Comma = comma
-	if err := w.Write(t.Header); err != nil {
-		return nil, err
-	}
-	for _, row := range t.Rows {
-		if err := w.Write(row); err != nil {
-			return nil, err
-		}
-	}
-	w.Flush()
-	return b.Bytes(), w.Error()
-}
-
-func writeXLSX(t table, sheet string) ([]byte, error) {
-	f := excelize.NewFile()
-	defaultSheet := f.GetSheetName(0)
-	if sheet == "" {
-		sheet = "Sheet1"
-	}
-	if defaultSheet != sheet {
-		if err := f.SetSheetName(defaultSheet, sheet); err != nil {
-			return nil, err
-		}
-	}
-	rows := append([][]string{t.Header}, t.Rows...)
-	for r, row := range rows {
-		for c, value := range row {
-			cell, err := excelize.CoordinatesToCellName(c+1, r+1)
-			if err != nil {
-				return nil, err
-			}
-			if err := f.SetCellValue(sheet, cell, value); err != nil {
-				return nil, err
-			}
-		}
-	}
-	if len(t.Header) > 0 {
-		// The generated workbook is meant for editing, so keep the header row
-		// visible and visually distinct.
-		end, _ := excelize.CoordinatesToCellName(len(t.Header), 1)
-		style, _ := f.NewStyle(&excelize.Style{Font: &excelize.Font{Bold: true}})
-		_ = f.SetCellStyle(sheet, "A1", end, style)
-		_ = f.SetPanes(sheet, &excelize.Panes{
-			Freeze:      true,
-			Split:       false,
-			XSplit:      0,
-			YSplit:      1,
-			TopLeftCell: "A2",
-			ActivePane:  "bottomLeft",
-		})
-	}
-	var b bytes.Buffer
-	if err := f.Write(&b); err != nil {
-		return nil, err
-	}
-	return b.Bytes(), nil
-}
-
-func valueToTable(value any) table {
-	records := recordsFromValue(value)
-	flatRows := make([]map[string]string, 0, len(records))
-	seen := map[string]bool{}
-	var header []string
-	for _, record := range records {
-		flat := map[string]string{}
-		flatten("", record, flat)
-		for key := range flat {
-			if !seen[key] {
-				seen[key] = true
-				header = append(header, key)
-			}
-		}
-		flatRows = append(flatRows, flat)
-	}
-	// Stable column ordering keeps generated CSV/TSV/XLSX diffs predictable.
-	sort.Strings(header)
-	rows := make([][]string, 0, len(flatRows))
-	for _, flat := range flatRows {
-		row := make([]string, len(header))
-		for i, key := range header {
-			row[i] = flat[key]
-		}
-		rows = append(rows, row)
-	}
-	return table{Header: header, Rows: rows}
-}
-
-func recordsFromValue(value any) []any {
-	switch v := value.(type) {
-	case []any:
-		return v
-	case map[string]any:
-		// Common wrapper keys let TOML and object-shaped inputs represent a
-		// table without adding a format-specific flag.
-		for _, key := range []string{"rows", "records", "items"} {
-			if rows, ok := v[key].([]any); ok {
-				return rows
-			}
-		}
-		return []any{v}
-	default:
-		return []any{v}
-	}
-}
-
-// flatten converts nested values into spreadsheet-safe column paths such as
-// user.name and items[0].sku.
-func flatten(prefix string, value any, out map[string]string) {
-	switch v := value.(type) {
-	case map[string]any:
-		if len(v) == 0 && prefix != "" {
-			out[prefix] = "{}"
-			return
-		}
-		keys := make([]string, 0, len(v))
-		for key := range v {
-			keys = append(keys, key)
-		}
-		sort.Strings(keys)
-		for _, key := range keys {
-			childPrefix := key
-			if prefix != "" {
-				childPrefix = prefix + "." + key
-			}
-			flatten(childPrefix, v[key], out)
-		}
-	case []any:
-		if len(v) == 0 && prefix != "" {
-			out[prefix] = "[]"
-			return
-		}
-		for i, child := range v {
-			flatten(fmt.Sprintf("%s[%d]", prefix, i), child, out)
-		}
-	default:
-		if prefix == "" {
-			prefix = "value"
-		}
-		out[prefix] = scalarString(v)
-	}
-}
-
-func scalarString(value any) string {
-	switch v := value.(type) {
-	case nil:
-		return ""
-	case json.Number:
-		return v.String()
-	case string:
-		return v
-	case bool:
-		return strconv.FormatBool(v)
-	case int, int8, int16, int32, int64, uint, uint8, uint16, uint32, uint64, float32, float64:
-		return fmt.Sprint(v)
-	default:
-		data, err := json.Marshal(v)
-		if err != nil {
-			return fmt.Sprint(v)
-		}
-		return string(data)
-	}
-}
-
-// tableToRecords restores each row by interpreting header cells as path
-// expressions. Empty headers are ignored so spare spreadsheet columns are safe.
-func tableToRecords(t table) []map[string]any {
-	var records []map[string]any
-	for _, row := range t.Rows {
-		record := map[string]any{}
-		for i, header := range t.Header {
-			header = strings.TrimSpace(header)
-			if header == "" || i >= len(row) {
-				continue
-			}
-			setPath(record, header, parseCell(row[i]))
-		}
-		records = append(records, record)
-	}
-	return records
-}
-
-// parseCell keeps spreadsheet round trips useful while avoiding broad type
-// inference that could surprise users editing IDs or codes.
-func parseCell(s string) any {
-	s = strings.TrimSpace(s)
-	if s == "" {
-		return ""
-	}
-	if s == "true" {
-		return true
-	}
-	if s == "false" {
-		return false
-	}
-	if i, err := strconv.ParseInt(s, 10, 64); err == nil {
-		return i
-	}
-	if f, err := strconv.ParseFloat(s, 64); err == nil && strings.ContainsAny(s, ".eE") {
-		return f
-	}
-	return s
-}
-
-var pathTokenRE = regexp.MustCompile(`([^\.\[\]]+)|\[(\d+)\]`)
-
-// setPath creates maps and slices as needed for dotted and indexed header
-// paths. Invalid intermediate shapes are left unchanged instead of guessing.
-func setPath(root map[string]any, path string, value any) {
-	tokens := parsePath(path)
-	if len(tokens) == 0 {
-		return
-	}
-	var cur any = root
-	for i, token := range tokens {
-		last := i == len(tokens)-1
-		nextIsIndex := !last && tokens[i+1].isIndex
-		if token.isIndex {
-			continue
-		}
-		m, ok := cur.(map[string]any)
-		if !ok {
-			return
-		}
-		if last {
-			m[token.key] = value
-			return
-		}
-		if _, ok := m[token.key]; !ok {
-			if nextIsIndex {
-				m[token.key] = []any{}
-			} else {
-				m[token.key] = map[string]any{}
-			}
-		}
-		if nextIsIndex {
-			slice, _ := m[token.key].([]any)
-			index := tokens[i+1].index
-			for len(slice) <= index {
-				if i+2 < len(tokens) && !tokens[i+2].isIndex {
-					slice = append(slice, map[string]any{})
-				} else {
-					slice = append(slice, nil)
-				}
-			}
-			m[token.key] = slice
-			if i+2 == len(tokens) {
-				slice[index] = value
-				return
-			}
-			if slice[index] == nil {
-				slice[index] = map[string]any{}
-			}
-			cur = slice[index]
-			i++
-			continue
-		}
-		cur = m[token.key]
-	}
-}
-
-type pathToken struct {
-	key     string
-	index   int
-	isIndex bool
-}
-
-func parsePath(path string) []pathToken {
-	matches := pathTokenRE.FindAllStringSubmatch(path, -1)
-	tokens := make([]pathToken, 0, len(matches))
-	for _, m := range matches {
-		if m[1] != "" {
-			tokens = append(tokens, pathToken{key: m[1]})
-			continue
-		}
-		index, _ := strconv.Atoi(m[2])
-		tokens = append(tokens, pathToken{index: index, isIndex: true})
-	}
-	return tokens
-}

cmd/dataxl/main_test.go

@@ -5,6 +5,7 @@ import (
 	"encoding/json"
 	"os"
 	"path/filepath"
+	"reflect"
 	"strings"
 	"testing"
 )
@@ -88,6 +89,76 @@ func TestJSONToXLSXAndBack(t *testing.T) {
 	}
 }
 
+func TestResolveFormatsFromFileExtensions(t *testing.T) {
+	opt := options{inFile: "input.YML", outFile: "output.XLSM"}
+	if err := opt.resolveFormats(); err != nil {
+		t.Fatal(err)
+	}
+	if opt.from != "yaml" || opt.to != "xlsx" {
+		t.Fatalf("resolved formats = %q -> %q, want yaml -> xlsx", opt.from, opt.to)
+	}
+}
+
+func TestConvertStructuredToStructured(t *testing.T) {
+	got, err := convert([]byte("name: Alice\nactive: true\n"), "yaml", "json", "Sheet1", false)
+	if err != nil {
+		t.Fatal(err)
+	}
+	if string(got) != `{"active":true,"name":"Alice"}` {
+		t.Fatalf("JSON = %s", got)
+	}
+}
+
+func TestRowsToTablePadsRaggedRows(t *testing.T) {
+	got := rowsToTable([][]string{{"a", "b"}, {"1"}, {"2", "3", "4"}})
+	want := table{
+		Header: []string{"a", "b", ""},
+		Rows:   [][]string{{"1", "", ""}, {"2", "3", "4"}},
+	}
+	if !reflect.DeepEqual(got, want) {
+		t.Fatalf("table = %#v, want %#v", got, want)
+	}
+}
+
+func TestParseCellConservativeInference(t *testing.T) {
+	tests := map[string]any{
+		"":      "",
+		"true":  true,
+		"42":    int64(42),
+		"3.14":  3.14,
+		"00123": "00123",
+		"1e3":   1000.0,
+		"Alice": "Alice",
+	}
+	for input, want := range tests {
+		if got := parseCell(input); !reflect.DeepEqual(got, want) {
+			t.Errorf("parseCell(%q) = %#v, want %#v", input, got, want)
+		}
+	}
+}
+
+func TestSetPathDoesNotOverwriteConflictingShape(t *testing.T) {
+	record := map[string]any{}
+	setPath(record, "user", "Alice")
+	setPath(record, "user.name", "Bob")
+	if got := record["user"]; got != "Alice" {
+		t.Fatalf("user = %#v, want original scalar", got)
+	}
+}
+
+func TestSetPathRestoresNestedArrays(t *testing.T) {
+	record := map[string]any{}
+	setPath(record, "orders[0].items[1].sku", "B-002")
+
+	orders := record["orders"].([]any)
+	order := orders[0].(map[string]any)
+	items := order["items"].([]any)
+	item := items[1].(map[string]any)
+	if item["sku"] != "B-002" {
+		t.Fatalf("nested sku = %#v, want B-002", item["sku"])
+	}
+}
+
 type ioDiscard struct{}
 
 func (ioDiscard) Write(p []byte) (int, error) { return len(p), nil }

cmd/dataxl/path.go

@@ -0,0 +1,134 @@
+package main
+
+import (
+	"regexp"
+	"strconv"
+	"strings"
+)
+
+// tableToRecords interprets non-empty headers as path expressions. Blank
+// headers are intentionally ignored so spare spreadsheet columns are harmless.
+func tableToRecords(t table) []map[string]any {
+	records := make([]map[string]any, 0, len(t.Rows))
+	for _, row := range t.Rows {
+		record := make(map[string]any)
+		for i, header := range t.Header {
+			header = strings.TrimSpace(header)
+			if header == "" || i >= len(row) {
+				continue
+			}
+			setPath(record, header, parseCell(row[i]))
+		}
+		records = append(records, record)
+	}
+	return records
+}
+
+// parseCell deliberately performs narrow inference. In particular, strings
+// such as 00123 remain strings because converting identifiers is surprising.
+func parseCell(s string) any {
+	s = strings.TrimSpace(s)
+	if s == "" {
+		return ""
+	}
+	if s == "true" {
+		return true
+	}
+	if s == "false" {
+		return false
+	}
+	// Preserve zero-padded identifiers such as postal codes and product codes.
+	leadingZero := len(s) > 1 && s[0] == '0'
+	negativeLeadingZero := len(s) > 2 && s[0] == '-' && s[1] == '0'
+	if i, err := strconv.ParseInt(s, 10, 64); err == nil && !leadingZero && !negativeLeadingZero {
+		return i
+	}
+	if f, err := strconv.ParseFloat(s, 64); err == nil && strings.ContainsAny(s, ".eE") {
+		return f
+	}
+	return s
+}
+
+var pathTokenRE = regexp.MustCompile(`([^\.\[\]]+)|\[(\d+)\]`)
+
+type pathToken struct {
+	key     string
+	index   int
+	isIndex bool
+}
+
+func parsePath(path string) []pathToken {
+	matches := pathTokenRE.FindAllStringSubmatch(path, -1)
+	tokens := make([]pathToken, 0, len(matches))
+	for _, match := range matches {
+		if match[1] != "" {
+			tokens = append(tokens, pathToken{key: match[1]})
+			continue
+		}
+		index, _ := strconv.Atoi(match[2]) // regexp guarantees decimal digits.
+		tokens = append(tokens, pathToken{index: index, isIndex: true})
+	}
+	return tokens
+}
+
+// setPath creates maps and slices while walking a dotted/indexed header. The
+// supported grammar is a sequence of map keys with optional array indices,
+// e.g. "orders[0].items[1].sku". Conflicting intermediate shapes are left
+// unchanged rather than silently overwriting data from an earlier column.
+func setPath(root map[string]any, path string, value any) {
+	tokens := parsePath(path)
+	if len(tokens) == 0 {
+		return
+	}
+	var current any = root
+	for i := 0; i < len(tokens); i++ {
+		token := tokens[i]
+		last := i == len(tokens)-1
+		nextIsIndex := !last && tokens[i+1].isIndex
+		if token.isIndex {
+			continue
+		}
+		m, ok := current.(map[string]any)
+		if !ok {
+			return
+		}
+		if last {
+			m[token.key] = value
+			return
+		}
+		if _, exists := m[token.key]; !exists {
+			if nextIsIndex {
+				m[token.key] = []any{}
+			} else {
+				m[token.key] = map[string]any{}
+			}
+		}
+		if !nextIsIndex {
+			current = m[token.key]
+			continue
+		}
+
+		slice, ok := m[token.key].([]any)
+		if !ok {
+			return
+		}
+		index := tokens[i+1].index
+		for len(slice) <= index {
+			if i+2 < len(tokens) && !tokens[i+2].isIndex {
+				slice = append(slice, map[string]any{})
+			} else {
+				slice = append(slice, nil)
+			}
+		}
+		m[token.key] = slice
+		if i+2 == len(tokens) {
+			slice[index] = value
+			return
+		}
+		if slice[index] == nil {
+			slice[index] = map[string]any{}
+		}
+		current = slice[index]
+		i++ // The array index was consumed together with its key.
+	}
+}

cmd/dataxl/structured.go

@@ -0,0 +1,85 @@
+package main
+
+import (
+	"bytes"
+	"encoding/json"
+	"fmt"
+
+	"github.com/BurntSushi/toml"
+	"gopkg.in/yaml.v3"
+)
+
+// parseStructured decodes a format into maps, slices and scalar Go values.
+// json.Number is retained so large JSON integers do not pass through float64.
+func parseStructured(input []byte, format string) (any, error) {
+	var value any
+	switch format {
+	case "json":
+		dec := json.NewDecoder(bytes.NewReader(input))
+		dec.UseNumber()
+		if err := dec.Decode(&value); err != nil {
+			return nil, err
+		}
+	case "yaml":
+		if err := yaml.Unmarshal(input, &value); err != nil {
+			return nil, err
+		}
+		value = normalizeYAML(value)
+	case "toml":
+		var m map[string]any
+		if err := toml.Unmarshal(input, &m); err != nil {
+			return nil, err
+		}
+		value = m
+	default:
+		return nil, fmt.Errorf("format %q is not structured", format)
+	}
+	return value, nil
+}
+
+func encodeStructured(value any, format string, pretty bool) ([]byte, error) {
+	switch format {
+	case "json":
+		if pretty {
+			return json.MarshalIndent(value, "", "  ")
+		}
+		return json.Marshal(value)
+	case "yaml":
+		return yaml.Marshal(value)
+	case "toml":
+		m, ok := value.(map[string]any)
+		if !ok {
+			// TOML has no top-level array, so preserve it under a documented key.
+			m = map[string]any{"rows": value}
+		}
+		var b bytes.Buffer
+		err := toml.NewEncoder(&b).Encode(m)
+		return b.Bytes(), err
+	default:
+		return nil, fmt.Errorf("format %q is not structured", format)
+	}
+}
+
+// normalizeYAML converts yaml.v3's possible map[any]any values into the
+// string-keyed maps used by the rest of the conversion pipeline.
+func normalizeYAML(value any) any {
+	switch v := value.(type) {
+	case map[string]any:
+		out := make(map[string]any, len(v))
+		for key, child := range v {
+			out[key] = normalizeYAML(child)
+		}
+		return out
+	case map[any]any:
+		out := make(map[string]any, len(v))
+		for key, child := range v {
+			out[fmt.Sprint(key)] = normalizeYAML(child)
+		}
+		return out
+	case []any:
+		for i := range v {
+			v[i] = normalizeYAML(v[i])
+		}
+	}
+	return value
+}

cmd/dataxl/table.go

@@ -0,0 +1,265 @@
+package main
+
+import (
+	"bytes"
+	"encoding/csv"
+	"encoding/json"
+	"fmt"
+	"sort"
+	"strconv"
+
+	"github.com/xuri/excelize/v2"
+)
+
+// table is the common representation for CSV, TSV and XLSX. Rows are padded
+// to Header width when read, which keeps subsequent conversions rectangular.
+type table struct {
+	Header []string
+	Rows   [][]string
+}
+
+func parseTable(input []byte, format, sheet string) (table, error) {
+	switch format {
+	case "csv":
+		return readDelimited(input, ',')
+	case "tsv":
+		return readDelimited(input, '\t')
+	case "xlsx":
+		f, err := excelize.OpenReader(bytes.NewReader(input))
+		if err != nil {
+			return table{}, err
+		}
+		defer f.Close()
+		rows, err := f.GetRows(sheet)
+		if err != nil {
+			return table{}, err
+		}
+		return rowsToTable(rows), nil
+	default:
+		return table{}, fmt.Errorf("format %q is not tabular", format)
+	}
+}
+
+func readDelimited(input []byte, comma rune) (table, error) {
+	r := csv.NewReader(bytes.NewReader(input))
+	r.Comma = comma
+	r.FieldsPerRecord = -1
+	r.TrimLeadingSpace = comma == ','
+	rows, err := r.ReadAll()
+	if err != nil {
+		return table{}, err
+	}
+	return rowsToTable(rows), nil
+}
+
+func rowsToTable(rows [][]string) table {
+	if len(rows) == 0 {
+		return table{}
+	}
+	width := 0
+	for _, row := range rows {
+		width = max(width, len(row))
+	}
+	header := padRow(rows[0], width)
+	body := make([][]string, 0, len(rows)-1)
+	for _, row := range rows[1:] {
+		body = append(body, padRow(row, width))
+	}
+	return table{Header: header, Rows: body}
+}
+
+func padRow(row []string, width int) []string {
+	out := make([]string, width)
+	copy(out, row)
+	return out
+}
+
+func encodeTable(t table, format, sheet string) ([]byte, error) {
+	switch format {
+	case "csv":
+		return writeDelimited(t, ',')
+	case "tsv":
+		return writeDelimited(t, '\t')
+	case "xlsx":
+		return writeXLSX(t, sheet)
+	default:
+		return nil, fmt.Errorf("format %q is not tabular", format)
+	}
+}
+
+func writeDelimited(t table, comma rune) ([]byte, error) {
+	var b bytes.Buffer
+	w := csv.NewWriter(&b)
+	w.Comma = comma
+	if err := w.Write(t.Header); err != nil {
+		return nil, err
+	}
+	for _, row := range t.Rows {
+		if err := w.Write(row); err != nil {
+			return nil, err
+		}
+	}
+	w.Flush()
+	return b.Bytes(), w.Error()
+}
+
+func writeXLSX(t table, sheet string) ([]byte, error) {
+	f := excelize.NewFile()
+	defer f.Close()
+	defaultSheet := f.GetSheetName(0)
+	if sheet == "" {
+		sheet = "Sheet1"
+	}
+	if defaultSheet != sheet {
+		if err := f.SetSheetName(defaultSheet, sheet); err != nil {
+			return nil, err
+		}
+	}
+	rows := append([][]string{t.Header}, t.Rows...)
+	for rowIndex, row := range rows {
+		for columnIndex, value := range row {
+			cell, err := excelize.CoordinatesToCellName(columnIndex+1, rowIndex+1)
+			if err != nil {
+				return nil, err
+			}
+			if err := f.SetCellValue(sheet, cell, value); err != nil {
+				return nil, err
+			}
+		}
+	}
+	if err := styleHeader(f, sheet, len(t.Header)); err != nil {
+		return nil, err
+	}
+	var b bytes.Buffer
+	if err := f.Write(&b); err != nil {
+		return nil, err
+	}
+	return b.Bytes(), nil
+}
+
+// styleHeader applies editing conveniences without trying to preserve or
+// emulate arbitrary workbook formatting.
+func styleHeader(f *excelize.File, sheet string, width int) error {
+	if width == 0 {
+		return nil
+	}
+	end, err := excelize.CoordinatesToCellName(width, 1)
+	if err != nil {
+		return err
+	}
+	style, err := f.NewStyle(&excelize.Style{Font: &excelize.Font{Bold: true}})
+	if err != nil {
+		return err
+	}
+	if err := f.SetCellStyle(sheet, "A1", end, style); err != nil {
+		return err
+	}
+	return f.SetPanes(sheet, &excelize.Panes{
+		Freeze: true, YSplit: 1, TopLeftCell: "A2", ActivePane: "bottomLeft",
+	})
+}
+
+func valueToTable(value any) table {
+	records := recordsFromValue(value)
+	flatRows := make([]map[string]string, 0, len(records))
+	seen := make(map[string]bool)
+	var header []string
+	for _, record := range records {
+		flat := make(map[string]string)
+		flatten("", record, flat)
+		for key := range flat {
+			if !seen[key] {
+				seen[key] = true
+				header = append(header, key)
+			}
+		}
+		flatRows = append(flatRows, flat)
+	}
+	// Stable column ordering makes generated files and their diffs predictable.
+	sort.Strings(header)
+	rows := make([][]string, 0, len(flatRows))
+	for _, flat := range flatRows {
+		row := make([]string, len(header))
+		for i, key := range header {
+			row[i] = flat[key]
+		}
+		rows = append(rows, row)
+	}
+	return table{Header: header, Rows: rows}
+}
+
+func recordsFromValue(value any) []any {
+	switch v := value.(type) {
+	case []any:
+		return v
+	case map[string]any:
+		// Wrapper keys let object-shaped formats such as TOML carry row arrays.
+		for _, key := range []string{"rows", "records", "items"} {
+			if rows, ok := v[key].([]any); ok {
+				return rows
+			}
+		}
+		return []any{v}
+	default:
+		return []any{v}
+	}
+}
+
+// flatten converts nested values to column paths such as user.name and
+// items[0].sku. Empty maps and arrays use explicit textual markers; these are
+// visible to editors but currently return as strings when converted back.
+func flatten(prefix string, value any, out map[string]string) {
+	switch v := value.(type) {
+	case map[string]any:
+		if len(v) == 0 && prefix != "" {
+			out[prefix] = "{}"
+			return
+		}
+		keys := make([]string, 0, len(v))
+		for key := range v {
+			keys = append(keys, key)
+		}
+		sort.Strings(keys)
+		for _, key := range keys {
+			childPrefix := key
+			if prefix != "" {
+				childPrefix = prefix + "." + key
+			}
+			flatten(childPrefix, v[key], out)
+		}
+	case []any:
+		if len(v) == 0 && prefix != "" {
+			out[prefix] = "[]"
+			return
+		}
+		for i, child := range v {
+			flatten(fmt.Sprintf("%s[%d]", prefix, i), child, out)
+		}
+	default:
+		if prefix == "" {
+			prefix = "value"
+		}
+		out[prefix] = scalarString(v)
+	}
+}
+
+func scalarString(value any) string {
+	switch v := value.(type) {
+	case nil:
+		return ""
+	case json.Number:
+		return v.String()
+	case string:
+		return v
+	case bool:
+		return strconv.FormatBool(v)
+	case int, int8, int16, int32, int64, uint, uint8, uint16, uint32, uint64, float32, float64:
+		return fmt.Sprint(v)
+	default:
+		data, err := json.Marshal(v)
+		if err != nil {
+			return fmt.Sprint(v)
+		}
+		return string(data)
+	}
+}

docs/architecture.md

@@ -12,17 +12,22 @@
 - table -> structured
 - table -> table
 
-## Format Adapters
+## Source Layout
 
-各形式の処理は `cmd/dataxl/main.go` の以下の関数に集約しています。
+実装は `cmd/dataxl` package内で責務別に分割しています。
 
-- `parseStructured`
-- `encodeStructured`
-- `parseTable`
-- `encodeTable`
+- `main.go`: CLI option、stdin/stdout、ファイル入出力
+- `conversion.go`: 変換経路の選択、形式名の正規化・推定
+- `structured.go`: JSON/YAML/TOML adapter
+- `table.go`: CSV/TSV/XLSX adapter、table model、flatten
+- `path.go`: セル値の型推定、列パスのparse、unflatten
 
-現在はCLIが小さいため単一ファイルに置いています。形式やオプションが増えたら、
-`internal/format` や `internal/table` へ分割する余地があります。
+`convert` はファイル入出力から独立しているため、CLIを経由せず変換matrixをテストできます。
+形式固有処理は `parseStructured` / `encodeStructured` または
+`parseTable` / `encodeTable` に閉じ込めます。
+
+現状は単一commandだけが利用するため同じpackageに置いています。別commandやlibrary APIから
+再利用する段階になったら、安定させたい境界を見極めたうえで `internal` packageへ移します。
 
 ## Table Model
 
@@ -79,6 +84,21 @@ items[0].sku
 - 小数または指数表記: float64
 - その他: string
 
+ゼロ埋め整数は、IDやコードを壊さないためstringとして保持します。複数列が同じパスで
+異なる中間型を要求する場合は、先に構築された値を後続列で上書きしません。
+
+### Path grammar
+
+現在の列パスは次の要素を扱います。
+
+```text
+path  = key, { ".", key | "[", index, "]" };
+index = digit, { digit };
+```
+
+実例は `user.name`、`items[0].sku`、`orders[0].items[1].qty` です。
+区切り文字を含むmap keyのescapeは未対応です。
+
 ## TOML Output
 
 TOMLはトップレベル配列を直接表せないため、表からTOMLへ出力する場合など、
@@ -91,3 +111,9 @@ TOMLはトップレベル配列を直接表せないため、表からTOMLへ出
 - `github.com/BurntSushi/toml`: TOML読み書き
 
 Go 1.24以上を前提にしています。
+
+## Error handling
+
+- 未対応形式、decode失敗、workbook/sheet操作失敗は呼び出し元へerrorを返します。
+- path復元中の型競合は既存値を保護するため、その列の適用を中止します。
+- XLSXのstyle・pane設定も通常の変換errorとして扱い、不完全なworkbookを成功扱いしません。

docs/development.md

@@ -51,6 +51,11 @@ The current tests cover:
 - YAML -> TSV flattening
 - TSV -> JSON path restoration
 - JSON -> XLSX -> JSON round trip
+- structured -> structured conversion without CLI/file I/O
+- extension normalization and format inference
+- ragged table row padding
+- conservative cell type inference, including zero-padded identifiers
+- conflicting unflatten paths
 
 When adding a new format or path rule, add tests around both directions where
 possible.
@@ -96,3 +101,6 @@ go install git.rumginger.org/agent/dataxl/cmd/dataxl@v0.1.0
 - Preserve headers as the contract between spreadsheet data and structured data.
 - Prefer explicit errors over silent best-effort conversion when a format is unsupported.
 - Keep dependencies small unless a format needs a mature parser/writer.
+- Keep CLI/file I/O in `main.go`; conversion behavior should remain testable through `convert`.
+- Keep format-specific behavior in its structured or table adapter.
+- Document round-trip limitations when a representation cannot preserve a value exactly.