page.go

// Copyright 2014 The Go Authors.  All rights reserved.
// Use of this source code is governed by a BSD-style
// license that can be found in the LICENSE file.

package pdf

import (
	"bytes"
	"errors"
	"fmt"
	"io"
	"sort"
	"strings"
)

// A Page represent a single page in a PDF file.
// The methods interpret a Page dictionary stored in V.
type Page struct {
	V Value
	F map[string]*Font
}

// Page returns the page for the given page number.
// Page numbers are indexed starting at 1, not 0.
// If the page is not found, Page returns a Page with p.V.IsNull().
func (r *Reader) Page(num int) Page {
	num-- // now 0-indexed
	page := r.Trailer().Key("Root").Key("Pages")
Search:
	for page.Key("Type").Name() == "Pages" {
		count := int(page.Key("Count").Int64())
		if count < num {
			return Page{}
		}
		kids := page.Key("Kids")
		for i := 0; i < kids.Len(); i++ {
			kid := kids.Index(i)
			if kid.Key("Type").Name() == "Pages" {
				c := int(kid.Key("Count").Int64())
				if num < c {
					page = kid
					continue Search
				}
				num -= c
				continue
			}
			if kid.Key("Type").Name() == "Page" {
				if num == 0 {
					return Page{kid, getFonts(kid)}
				}
				num--
			}
		}
		break
	}
	return Page{}
}

func getFonts(v Value) map[string]*Font {
	fonts := make(map[string]*Font)

	// get inherited resources associated with the page
	for parent := v; !parent.IsNull(); parent = parent.Key("Parent") {
		// get resources dictionary associated with the page
		if resources := parent.Key("Resources"); !resources.IsNull() {
			// get font dictionary associated with the page
			font := resources.Key("Font")

			for _, name := range font.Keys() {
				fontDict := font.Key(name)
				f := Font{V: fontDict, enc: nil}

				if fontDict.Key("Subtype").Name() == "Type0" {
					f.IsCIDFont = true
					desc := fontDict.Key("DescendantFonts").Index(0)
					f.DW = desc.Key("DW").Float64() // Default width
					f.W = extractWidths(desc.Key("W"))
				}
				fonts[name] = &f
			}
		}
	}

	return fonts
}

// NumPage returns the number of pages in the PDF file.
func (r *Reader) NumPage() int {
	return int(r.Trailer().Key("Root").Key("Pages").Key("Count").Int64())
}

// GetPlainText returns all the text in the PDF file
func (r *Reader) GetPlainText() (reader io.Reader, err error) {
	pages := r.NumPage()
	var buf bytes.Buffer
	for i := 1; i <= pages; i++ {
		p := r.Page(i)
		text, err := p.GetPlainText()
		if err != nil {
			return &bytes.Buffer{}, err
		}
		buf.WriteString(text)
	}
	return &buf, nil
}

// Font returns the font with the given name associated with the page.
func (p Page) Font(name string) *Font {
	return p.F[name]
}

func extractWidths(w Value) map[int]float64 {
	widths := make(map[int]float64)

	for i := 0; i < w.Len(); i++ {
		firstCID := int(w.Index(i).Int64())
		i++
		next := w.Index(i)

		if next.Kind() == Array {
			// Case where an array follows the first CID, specifying individual widths
			for j := 0; j < next.Len(); j++ {
				widths[firstCID+j] = next.Index(j).Float64()
			}
		} else {
			// Case where a single width is specified for a range of CIDs
			lastCID := int(next.Int64())
			i++
			width := w.Index(i).Float64()

			for cid := firstCID; cid <= lastCID; cid++ {
				widths[cid] = width
			}
		}
	}

	return widths
}

// A Font represent a font in a PDF file.
// The methods interpret a Font dictionary stored in V.
type Font struct {
	V   Value
	enc TextEncoding

	IsCIDFont bool
	DW        float64
	W         map[int]float64
}

// BaseFont returns the font's name (BaseFont property).
func (f Font) BaseFont() string {
	return f.V.Key("BaseFont").Name()
}

// FirstChar returns the code point of the first character in the font.
func (f Font) FirstChar() int {
	return int(f.V.Key("FirstChar").Int64())
}

// LastChar returns the code point of the last character in the font.
func (f Font) LastChar() int {
	return int(f.V.Key("LastChar").Int64())
}

// Widths returns the widths of the glyphs in the font.
// In a well-formed PDF, len(f.Widths()) == f.LastChar()+1 - f.FirstChar().
func (f Font) Widths() []float64 {
	x := f.V.Key("Widths")
	var out []float64
	for i := 0; i < x.Len(); i++ {
		out = append(out, x.Index(i).Float64())
	}
	return out
}

// Width returns the width of the given code point.
func (f Font) Width(code int) float64 {
	if f.IsCIDFont {
		if width, ok := f.W[code]; ok {
			return width
		}
		return f.DW
	}
	first := f.FirstChar()
	last := f.LastChar()
	if code < first || last < code {
		return 0
	}
	return f.V.Key("Widths").Index(code - first).Float64()
}

// Encoder returns the encoding between font code point sequences and UTF-8.
func (f Font) Encoder() TextEncoding {
	if f.enc == nil { // caching the Encoder so we don't have to continually parse charmap
		f.enc = f.getEncoder()
	}
	return f.enc
}

func (f Font) getEncoder() TextEncoding {
	enc := f.V.Key("Encoding")
	switch enc.Kind() {
	case Name:
		switch enc.Name() {
		case "WinAnsiEncoding":
			return &byteEncoder{&winAnsiEncoding}
		case "MacRomanEncoding":
			return &byteEncoder{&macRomanEncoding}
		case "Identity-H":
			return f.charmapEncoding()
		default:
			if DebugOn {
				println("unknown encoding", enc.Name())
			}
			return &nopEncoder{}
		}
	case Dict:
		return &dictEncoder{enc.Key("Differences")}
	case Null:
		return f.charmapEncoding()
	default:
		if DebugOn {
			println("unexpected encoding", enc.String())
		}
		return &nopEncoder{}
	}
}

func (f *Font) charmapEncoding() TextEncoding {
	toUnicode := f.V.Key("ToUnicode")
	if toUnicode.Kind() == Stream {
		m := readCmap(toUnicode)
		if m == nil {
			return &nopEncoder{}
		}
		return m
	}

	return &byteEncoder{&pdfDocEncoding}
}

type dictEncoder struct {
	v Value
}

func (e *dictEncoder) Decode(raw string) (text string) {
	r := make([]rune, 0, len(raw))
	for i := 0; i < len(raw); i++ {
		ch := rune(raw[i])
		n := -1
		for j := 0; j < e.v.Len(); j++ {
			x := e.v.Index(j)
			if x.Kind() == Integer {
				n = int(x.Int64())
				continue
			}
			if x.Kind() == Name {
				if int(raw[i]) == n {
					r := nameToRune[x.Name()]
					if r != 0 {
						ch = r
						break
					}
				}
				n++
			}
		}
		r = append(r, ch)
	}
	return string(r)
}

// A TextEncoding represents a mapping between
// font code points and UTF-8 text.
type TextEncoding interface {
	// Decode returns the UTF-8 text corresponding to
	// the sequence of code points in raw.
	Decode(raw string) (text string)
}

type nopEncoder struct {
}

func (e *nopEncoder) Decode(raw string) (text string) {
	return raw
}

type byteEncoder struct {
	table *[256]rune
}

func (e *byteEncoder) Decode(raw string) (text string) {
	r := make([]rune, 0, len(raw))
	for i := 0; i < len(raw); i++ {
		r = append(r, e.table[raw[i]])
	}
	return string(r)
}

type byteRange struct {
	low  string
	high string
}

type bfchar struct {
	orig string
	repl string
}

type bfrange struct {
	lo  string
	hi  string
	dst Value
}

type cmap struct {
	space   [4][]byteRange // codespace range
	bfrange []bfrange
	bfchar  []bfchar
}

func (m *cmap) Decode(raw string) (text string) {
	var r []rune
Parse:
	for len(raw) > 0 {
		for n := 1; n <= 4 && n <= len(raw); n++ { // number of digits in character replacement (1-4 possible)
			for _, space := range m.space[n-1] { // find matching codespace Ranges for number of digits
				if space.low <= raw[:n] && raw[:n] <= space.high { // see if value is in range
					text := raw[:n]
					raw = raw[n:]
					for _, bfchar := range m.bfchar { // check for matching bfchar
						if len(bfchar.orig) == n && bfchar.orig == text {
							r = append(r, []rune(utf16Decode(bfchar.repl))...)
							continue Parse
						}
					}
					for _, bfrange := range m.bfrange { // check for matching bfrange
						if len(bfrange.lo) == n && bfrange.lo <= text && text <= bfrange.hi {
							if bfrange.dst.Kind() == String {
								s := bfrange.dst.RawString()
								if bfrange.lo != text { // value isn't at the beginning of the range so scale result
									b := []byte(s)
									b[len(b)-1] += text[len(text)-1] - bfrange.lo[len(bfrange.lo)-1] // increment last byte by difference
									s = string(b)
								}
								r = append(r, []rune(utf16Decode(s))...)
								continue Parse
							}
							if bfrange.dst.Kind() == Array {
								n := text[len(text)-1] - bfrange.lo[len(bfrange.lo)-1]
								v := bfrange.dst.Index(int(n))
								if v.Kind() == String {
									s := v.RawString()
									r = append(r, []rune(utf16Decode(s))...)
									continue Parse
								}
								if DebugOn {
									fmt.Printf("array %v\n", bfrange.dst)
								}
							} else {
								if DebugOn {
									fmt.Printf("unknown dst %v\n", bfrange.dst)
								}
							}
							r = append(r, noRune)
							continue Parse
						}
					}
					r = append(r, noRune)
					continue Parse
				}
			}
		}
		if DebugOn {
			println("no code space found")
		}
		r = append(r, noRune)
		raw = raw[1:]
	}
	return string(r)
}

func readCmap(toUnicode Value) *cmap {
	n := -1
	var m cmap
	ok := true
	Interpret(toUnicode, func(stk *Stack, op string) {
		if !ok {
			return
		}
		switch op {
		case "findresource":
			stk.Pop() // category
			stk.Pop() // key
			stk.Push(newDict())
		case "begincmap":
			stk.Push(newDict())
		case "endcmap":
			stk.Pop()
		case "begincodespacerange":
			n = int(stk.Pop().Int64())
		case "endcodespacerange":
			if n < 0 {
				if DebugOn {
					println("missing begincodespacerange")
				}
				ok = false
				return
			}
			for i := 0; i < n; i++ {
				hi, lo := stk.Pop().RawString(), stk.Pop().RawString()
				if len(lo) == 0 || len(lo) != len(hi) {
					if DebugOn {
						println("bad codespace range")
					}
					ok = false
					return
				}
				m.space[len(lo)-1] = append(m.space[len(lo)-1], byteRange{lo, hi})
			}
			n = -1
		case "beginbfchar":
			n = int(stk.Pop().Int64())
		case "endbfchar":
			if n < 0 {
				panic("missing beginbfchar")
			}
			for i := 0; i < n; i++ {
				repl, orig := stk.Pop().RawString(), stk.Pop().RawString()
				m.bfchar = append(m.bfchar, bfchar{orig, repl})
			}
		case "beginbfrange":
			n = int(stk.Pop().Int64())
		case "endbfrange":
			if n < 0 {
				panic("missing beginbfrange")
			}
			for i := 0; i < n; i++ {
				dst, srcHi, srcLo := stk.Pop(), stk.Pop().RawString(), stk.Pop().RawString()
				m.bfrange = append(m.bfrange, bfrange{srcLo, srcHi, dst})
			}
		case "defineresource":
			stk.Pop().Name() // category
			value := stk.Pop()
			stk.Pop().Name() // key
			stk.Push(value)
		default:
			if DebugOn {
				println("interp\t", op)
			}
		}
	})
	if !ok {
		return nil
	}
	return &m
}

type matrix [3][3]float64

var ident = matrix{{1, 0, 0}, {0, 1, 0}, {0, 0, 1}}

func (x matrix) mul(y matrix) matrix {
	var z matrix
	for i := 0; i < 3; i++ {
		for j := 0; j < 3; j++ {
			for k := 0; k < 3; k++ {
				z[i][j] += x[i][k] * y[k][j]
			}
		}
	}
	return z
}

// A Text represents a single piece of text drawn on a page.
type Text struct {
	Font     string  // the font used
	FontSize float64 // the font size, in points (1/72 of an inch)
	X        float64 // the X coordinate, in points, increasing left to right
	Y        float64 // the Y coordinate, in points, increasing bottom to top
	W        float64 // the width of the text, in points
	S        string  // the actual UTF-8 text
}

// A Rect represents a rectangle.
type Rect struct {
	Min, Max Point
}

// A Point represents an X, Y pair.
type Point struct {
	X float64
	Y float64
}

// Content describes the basic content on a page: the text and any drawn rectangles.
type Content struct {
	Text []Text
	Rect []Rect
}

type gstate struct {
	Tc    float64
	Tw    float64
	Th    float64
	Tl    float64
	Tf    *Font
	Tfs   float64
	Tmode int
	Trise float64
	Tm    matrix
	Tlm   matrix
	Trm   matrix
	CTM   matrix
}

// GetPlainText returns the page's all text without format.
// fonts can be passed in (to improve parsing performance) or left nil
func (p Page) GetPlainText() (result string, err error) {
	defer func() {
		if r := recover(); r != nil {
			result = ""
			err = errors.New(fmt.Sprint(r))
		}
	}()

	strm := p.V.Key("Contents")
	var enc TextEncoding = &nopEncoder{}

	var textBuilder bytes.Buffer
	showText := func(s string) {
		for _, ch := range enc.Decode(s) {
			_, err := textBuilder.WriteRune(ch)
			if err != nil {
				panic(err)
			}
		}
	}

	Interpret(strm, func(stk *Stack, op string) {
		n := stk.Len()
		args := make([]Value, n)
		for i := n - 1; i >= 0; i-- {
			args[i] = stk.Pop()
		}

		switch op {
		default:
			return
		case "T*": // move to start of next line
			showText("\n")
		case "Tf": // set text font and size
			if len(args) != 2 {
				panic("bad TL")
			}
			if font, ok := p.F[args[0].Name()]; ok {
				enc = font.Encoder()
			} else {
				enc = &nopEncoder{}
			}
		case "\"": // set spacing, move to next line, and show text
			if len(args) != 3 {
				panic("bad \" operator")
			}
			fallthrough
		case "'": // move to next line and show text
			if len(args) != 1 {
				panic("bad ' operator")
			}
			fallthrough
		case "Tj": // show text
			if len(args) != 1 {
				panic("bad Tj operator")
			}
			showText(args[0].RawString())
		case "TJ": // show text, allowing individual glyph positioning
			v := args[0]
			for i := 0; i < v.Len(); i++ {
				x := v.Index(i)
				if x.Kind() == String {
					showText(x.RawString())
				}
			}
		}
	})
	return textBuilder.String(), nil
}

// Column represents the contents of a column
type Column struct {
	Position int64
	Content  TextVertical
}

// Columns is a list of column
type Columns []*Column

// GetTextByColumn returns the page's all text grouped by column
func (p Page) GetTextByColumn() (Columns, error) {
	result := Columns{}
	var err error

	defer func() {
		if r := recover(); r != nil {
			result = Columns{}
			err = errors.New(fmt.Sprint(r))
		}
	}()

	showText := func(enc TextEncoding, currentX, currentY float64, s string) {
		var textBuilder bytes.Buffer

		for _, ch := range enc.Decode(s) {
			_, err := textBuilder.WriteRune(ch)
			if err != nil {
				panic(err)
			}
		}
		text := Text{
			S: textBuilder.String(),
			X: currentX,
			Y: currentY,
		}

		var currentColumn *Column
		columnFound := false
		for _, column := range result {
			if int64(currentX) == column.Position {
				currentColumn = column
				columnFound = true
				break
			}
		}

		if !columnFound {
			currentColumn = &Column{
				Position: int64(currentX),
				Content:  TextVertical{},
			}
			result = append(result, currentColumn)
		}

		currentColumn.Content = append(currentColumn.Content, text)
	}

	p.walkTextBlocks(showText)

	for _, column := range result {
		sort.Sort(column.Content)
	}

	sort.Slice(result, func(i, j int) bool {
		return result[i].Position < result[j].Position
	})

	return result, err
}

// Row represents the contents of a row
type Row struct {
	Position int64
	Content  TextHorizontal
}

// Rows is a list of rows
type Rows []*Row

// GetTextByRow returns the page's all text grouped by rows
func (p Page) GetTextByRow() (Rows, error) {
	result := Rows{}
	var err error

	defer func() {
		if r := recover(); r != nil {
			result = Rows{}
			err = errors.New(fmt.Sprint(r))
		}
	}()

	showText := func(enc TextEncoding, currentX, currentY float64, s string) {
		var textBuilder bytes.Buffer
		for _, ch := range enc.Decode(s) {
			_, err := textBuilder.WriteRune(ch)
			if err != nil {
				panic(err)
			}
		}

		// if DebugOn {
		// 	fmt.Println(textBuilder.String())
		// }

		text := Text{
			S: textBuilder.String(),
			X: currentX,
			Y: currentY,
		}

		var currentRow *Row
		rowFound := false
		for _, row := range result {
			if int64(currentY) == row.Position {
				currentRow = row
				rowFound = true
				break
			}
		}

		if !rowFound {
			currentRow = &Row{
				Position: int64(currentY),
				Content:  TextHorizontal{},
			}
			result = append(result, currentRow)
		}

		currentRow.Content = append(currentRow.Content, text)
	}

	p.walkTextBlocks(showText)

	for _, row := range result {
		sort.Sort(row.Content)
	}

	sort.Slice(result, func(i, j int) bool {
		return result[i].Position > result[j].Position
	})

	return result, err
}

func (p Page) walkTextBlocks(walker func(enc TextEncoding, x, y float64, s string)) {
	strm := p.V.Key("Contents")

	var enc TextEncoding = &nopEncoder{}
	var currentX, currentY float64
	Interpret(strm, func(stk *Stack, op string) {
		n := stk.Len()
		args := make([]Value, n)
		for i := n - 1; i >= 0; i-- {
			args[i] = stk.Pop()
		}

		// if DebugOn {
		// 	fmt.Println(op, "->", args)
		// }

		switch op {
		default:
			return
		case "T*": // move to start of next line
		case "Tf": // set text font and size
			if len(args) != 2 {
				panic("bad TL")
			}

			if font, ok := p.F[args[0].Name()]; ok {
				enc = font.Encoder()
			} else {
				enc = &nopEncoder{}
			}
		case "\"": // set spacing, move to next line, and show text
			if len(args) != 3 {
				panic("bad \" operator")
			}
			fallthrough
		case "'": // move to next line and show text
			if len(args) != 1 {
				panic("bad ' operator")
			}
			fallthrough
		case "Tj": // show text
			if len(args) != 1 {
				panic("bad Tj operator")
			}

			walker(enc, currentX, currentY, args[0].RawString())
		case "TJ": // show text, allowing individual glyph positioning
			v := args[0]
			for i := 0; i < v.Len(); i++ {
				x := v.Index(i)
				if x.Kind() == String {
					walker(enc, currentX, currentY, x.RawString())
				}
			}
		case "Td":
			walker(enc, currentX, currentY, "")
		case "Tm":
			currentX = args[4].Float64()
			currentY = args[5].Float64()
		}
	})
}

// Content returns the page's content.
func (p Page) Content() Content {
	strm := p.V.Key("Contents")
	var enc TextEncoding = &nopEncoder{}

	var g = gstate{
		Th:  1,
		CTM: ident,
	}

	var text []Text
	showText := func(s string) {
		n := 1
		decoded := enc.Decode(s)
		for _, ch := range decoded {
			var w0 float64
			if n < len(s) {
				w0 = g.Tf.Width(int(s[n]))
			}
			n += 2

			f := g.Tf.BaseFont()
			if i := strings.Index(f, "+"); i >= 0 {
				f = f[i+1:]
			}

			Trm := matrix{{g.Tfs * g.Th, 0, 0}, {0, g.Tfs, 0}, {0, g.Trise, 1}}.mul(g.Tm).mul(g.CTM)
			text = append(text, Text{f, Trm[0][0], Trm[2][0], Trm[2][1], w0 / 1000 * Trm[0][0], string(ch)})

			tx := w0/1000*g.Tfs + g.Tc
			tx *= g.Th
			g.Tm = matrix{{1, 0, 0}, {0, 1, 0}, {tx, 0, 1}}.mul(g.Tm)
		}
	}

	var rect []Rect
	var gstack []gstate
	Interpret(strm, func(stk *Stack, op string) {
		n := stk.Len()
		args := make([]Value, n)
		for i := n - 1; i >= 0; i-- {
			args[i] = stk.Pop()
		}
		switch op {
		default:
			// if DebugOn {
			// 	fmt.Println(op, args)
			// }
			return

		case "cm": // update g.CTM
			if len(args) != 6 {
				panic("bad g.Tm")
			}
			var m matrix
			for i := 0; i < 6; i++ {
				m[i/2][i%2] = args[i].Float64()
			}
			m[2][2] = 1
			g.CTM = m.mul(g.CTM)

		case "gs": // set parameters from graphics state resource
			//gs := p.Resources().Key("ExtGState").Key(args[0].Name())
			//font := gs.Key("Font")
			//if font.Kind() == Array && font.Len() == 2 {
			// if DebugOn {
			// 	fmt.Println("FONT", font)
			// }
			//}

		case "f": // fill
		case "g": // setgray
		case "l": // lineto
		case "m": // moveto

		case "cs": // set colorspace non-stroking
		case "scn": // set color non-stroking

		case "re": // append rectangle to path
			if len(args) != 4 {
				panic("bad re")
			}
			x, y, w, h := args[0].Float64(), args[1].Float64(), args[2].Float64(), args[3].Float64()
			rect = append(rect, Rect{Point{x, y}, Point{x + w, y + h}})

		case "q": // save graphics state
			gstack = append(gstack, g)

		case "Q": // restore graphics state
			n := len(gstack) - 1
			g = gstack[n]
			gstack = gstack[:n]

		case "BT": // begin text (reset text matrix and line matrix)
			g.Tm = ident
			g.Tlm = g.Tm

		case "ET": // end text

		case "T*": // move to start of next line
			x := matrix{{1, 0, 0}, {0, 1, 0}, {0, -g.Tl, 1}}
			g.Tlm = x.mul(g.Tlm)
			g.Tm = g.Tlm

		case "Tc": // set character spacing
			if len(args) != 1 {
				panic("bad g.Tc")
			}
			g.Tc = args[0].Float64()

		case "TD": // move text position and set leading
			if len(args) != 2 {
				panic("bad Td")
			}
			g.Tl = -args[1].Float64()
			fallthrough
		case "Td": // move text position
			if len(args) != 2 {
				panic("bad Td")
			}
			tx := args[0].Float64()
			ty := args[1].Float64()
			x := matrix{{1, 0, 0}, {0, 1, 0}, {tx, ty, 1}}
			g.Tlm = x.mul(g.Tlm)
			g.Tm = g.Tlm

		case "Tf": // set text font and size
			if len(args) != 2 {
				panic("bad TL")
			}
			f := args[0].Name()
			g.Tf = p.Font(f)
			if g.Tf == nil {
				if DebugOn {
					println("no font data for", f)
				}
				g.Tf = &Font{}
			}
			enc = g.Tf.Encoder()
			if enc == nil {
				if DebugOn {
					println("no cmap for", f)
				}
				enc = &nopEncoder{}
			}
			g.Tfs = args[1].Float64()

		case "\"": // set spacing, move to next line, and show text
			if len(args) != 3 {
				panic("bad \" operator")
			}
			g.Tw = args[0].Float64()
			g.Tc = args[1].Float64()
			args = args[2:]
			fallthrough
		case "'": // move to next line and show text
			if len(args) != 1 {
				panic("bad ' operator")
			}
			x := matrix{{1, 0, 0}, {0, 1, 0}, {0, -g.Tl, 1}}
			g.Tlm = x.mul(g.Tlm)
			g.Tm = g.Tlm
			fallthrough
		case "Tj": // show text
			if len(args) != 1 {
				panic("bad Tj operator")
			}
			showText(args[0].RawString())

		case "TJ": // show text, allowing individual glyph positioning
			v := args[0]
			for i := 0; i < v.Len(); i++ {
				x := v.Index(i)
				if x.Kind() == String {
					showText(x.RawString())
				} else {
					tx := -x.Float64() / 1000 * g.Tfs * g.Th
					g.Tm = matrix{{1, 0, 0}, {0, 1, 0}, {tx, 0, 1}}.mul(g.Tm)
				}
			}
			showText("\n")

		case "TL": // set text leading
			if len(args) != 1 {
				panic("bad TL")
			}
			g.Tl = args[0].Float64()

		case "Tm": // set text matrix and line matrix
			if len(args) != 6 {
				panic("bad g.Tm")
			}
			var m matrix
			for i := 0; i < 6; i++ {
				m[i/2][i%2] = args[i].Float64()
			}
			m[2][2] = 1
			g.Tm = m
			g.Tlm = m

		case "Tr": // set text rendering mode
			if len(args) != 1 {
				panic("bad Tr")
			}
			g.Tmode = int(args[0].Int64())

		case "Ts": // set text rise
			if len(args) != 1 {
				panic("bad Ts")
			}
			g.Trise = args[0].Float64()

		case "Tw": // set word spacing
			if len(args) != 1 {
				panic("bad g.Tw")
			}
			g.Tw = args[0].Float64()

		case "Tz": // set horizontal text scaling
			if len(args) != 1 {
				panic("bad Tz")
			}
			g.Th = args[0].Float64() / 100
		}
	})
	return Content{text, rect}
}

// TextVertical implements sort.Interface for sorting
// a slice of Text values in vertical order, top to bottom,
// and then left to right within a line.
type TextVertical []Text

func (x TextVertical) Len() int      { return len(x) }
func (x TextVertical) Swap(i, j int) { x[i], x[j] = x[j], x[i] }
func (x TextVertical) Less(i, j int) bool {
	if x[i].Y != x[j].Y {
		return x[i].Y > x[j].Y
	}
	return x[i].X < x[j].X
}

// TextHorizontal implements sort.Interface for sorting
// a slice of Text values in horizontal order, left to right,
// and then top to bottom within a column.
type TextHorizontal []Text

func (x TextHorizontal) Len() int      { return len(x) }
func (x TextHorizontal) Swap(i, j int) { x[i], x[j] = x[j], x[i] }
func (x TextHorizontal) Less(i, j int) bool {
	if x[i].X != x[j].X {
		return x[i].X < x[j].X
	}
	return x[i].Y > x[j].Y
}

// An Outline is a tree describing the outline (also known as the table of contents)
// of a document.
type Outline struct {
	Title string    // title for this element
	Child []Outline // child elements
}

// Outline returns the document outline.
// The Outline returned is the root of the outline tree and typically has no Title itself.
// That is, the children of the returned root are the top-level entries in the outline.
func (r *Reader) Outline() Outline {
	return buildOutline(r.Trailer().Key("Root").Key("Outlines"))
}

func buildOutline(entry Value) Outline {
	var x Outline
	x.Title = entry.Key("Title").Text()
	for child := entry.Key("First"); child.Kind() == Dict; child = child.Key("Next") {
		x.Child = append(x.Child, buildOutline(child))
	}
	return x
}