shiny/text/caret.go - exp - Git at Google

 // Copyright 2016 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 text

 // TODO: do we care about "\n" vs "\r" vs "\r\n"? We only recognize "\n" for
 // now.

 import (
 	"bytes"
 	"errors"
 	"io"
 	"strings"
 	"unicode/utf8"

 	"golang.org/x/image/math/fixed"
 )

 // Caret is a location in a Frame's text, and is the mechanism for adding and
 // removing bytes of text. Conceptually, a Caret and a Frame's text is like an
 // int c and a []byte t such that the text before and after that Caret is t[:c]
 // and t[c:]. That byte-count location remains unchanged even when a Frame is
 // re-sized and laid out into a new tree of Paragraphs, Lines and Boxes.
 //
 // A Frame can have multiple open Carets. For example, the beginning and end of
 // a text selection can be represented by two Carets. Multiple Carets for the
 // one Frame are not safe to use concurrently, but it is valid to interleave
 // such operations sequentially. For example, if two Carets c0 and c1 for the
 // one Frame are positioned at the 10th and 20th byte, and 4 bytes are written
 // to c0, inserting what becomes the equivalent of text[10:14], then c0's
 // position is updated to be 14 but c1's position is also updated to be 24.
 type Caret struct {
 	f *Frame

 	// caretsIndex is the index of this Caret in the f.carets slice.
 	caretsIndex int

 	// seqNum is the Frame f's sequence number for which this Caret's cached p,
 	// l, b and k fields are valid. If f has been modified since then, those
 	// fields will have to be re-calculated based on the pos field (which is
 	// always valid).
 	//
 	// TODO: when re-calculating p, l, b and k, be more efficient than a linear
 	// scan from the start or end?
 	seqNum uint64

 	// p, l and b cache the index of the Caret's Paragraph, Line and Box. None
 	// of these values can be zero.
 	p, l, b int32

 	// k caches the Caret's position in the text, in Frame.text order. It is
 	// valid to index the Frame.text slice with k, analogous to the Box.i and
 	// Box.j fields. For a Caret c, letting bb := c.f.boxes[c.b], an invariant
 	// is that bb.i <= c.k && c.k <= bb.j if the cache is valid (i.e. the
 	// Caret's seqNum equals the Frame's seqNum).
 	k int32

 	// pos is the Caret's position in the text, in layout order. It is the "c"
 	// as in "t[:c]" in the doc comment for type Caret above. It is not valid
 	// to index the Frame.text slice with pos, since the Frame.text slice does
 	// not necessarily hold the textual content in layout order.
 	pos int32

 	tmp [utf8.UTFMax]byte
 }

 // Close closes the Caret.
 func (c *Caret) Close() error {
 	i, j := c.caretsIndex, len(c.f.carets)-1

 	// Swap c with the last element of c.f.carets.
 	if i != j {
 		other := c.f.carets[j]
 		other.caretsIndex = i
 		c.f.carets[i] = other
 	}

 	c.f.carets[j] = nil
 	c.f.carets = c.f.carets[:j]
 	*c = Caret{}
 	return nil
 }

 type leanResult int

 const (
 	// leanOK means that the lean changed the Caret's Box.
 	leanOK leanResult = iota
 	// leanFailedEOF means that the lean did not change the Caret's Box,
 	// because the Caret was already at the end / beginning of the Frame (when
 	// leaning forwards / backwards).
 	leanFailedEOF
 	// leanFailedNotEOB means that the lean did not change the Caret's Box,
 	// because the Caret was not placed at the end / beginning of the Box (when
 	// leaning forwards / backwards).
 	leanFailedNotEOB
 )

 // leanForwards moves the Caret from the right end of one Box to the left end
 // of the next Box, crossing Lines and Paragraphs to find that next Box. It
 // returns whether the Caret moved to a different Box.
 //
 // Diagramatically, suppose we have two adjacent boxes (shown by square
 // brackets below), with the Caret (an integer location called Caret.pos in the
 // Frame's text) in the middle of the "foo2bar3" word:
 //	[foo0 foo1 foo2]^[bar3 bar4 bar5]
 // leanForwards moves Caret.k from fooBox.j to barBox.i, also updating the
 // Caret's p, l and b. Caret.pos remains unchanged.
 func (c *Caret) leanForwards() leanResult {
 	if c.k != c.f.boxes[c.b].j {
 		return leanFailedNotEOB
 	}
 	if nextB := c.f.boxes[c.b].next; nextB != 0 {
 		c.b = nextB
 		c.k = c.f.boxes[c.b].i
 		return leanOK
 	}
 	if nextL := c.f.lines[c.l].next; nextL != 0 {
 		c.l = nextL
 		c.b = c.f.lines[c.l].firstB
 		c.k = c.f.boxes[c.b].i
 		return leanOK
 	}
 	if nextP := c.f.paragraphs[c.p].next; nextP != 0 {
 		c.p = nextP
 		c.l = c.f.paragraphs[c.p].firstL
 		c.b = c.f.lines[c.l].firstB
 		c.k = c.f.boxes[c.b].i
 		return leanOK
 	}
 	return leanFailedEOF
 }

 // leanBackwards is like leanForwards but in the other direction.
 func (c *Caret) leanBackwards() leanResult {
 	if c.k != c.f.boxes[c.b].i {
 		return leanFailedNotEOB
 	}
 	if prevB := c.f.boxes[c.b].prev; prevB != 0 {
 		c.b = prevB
 		c.k = c.f.boxes[c.b].j
 		return leanOK
 	}
 	if prevL := c.f.lines[c.l].prev; prevL != 0 {
 		c.l = prevL
 		c.b = c.f.lines[c.l].lastBox(c.f)
 		c.k = c.f.boxes[c.b].j
 		return leanOK
 	}
 	if prevP := c.f.paragraphs[c.p].prev; prevP != 0 {
 		c.p = prevP
 		c.l = c.f.paragraphs[c.p].lastLine(c.f)
 		c.b = c.f.lines[c.l].lastBox(c.f)
 		c.k = c.f.boxes[c.b].j
 		return leanOK
 	}
 	return leanFailedEOF
 }

 func (c *Caret) seekStart() {
 	c.p = c.f.firstP
 	c.l = c.f.paragraphs[c.p].firstL
 	c.b = c.f.lines[c.l].firstB
 	c.k = c.f.boxes[c.b].i
 	c.pos = 0
 }

 func (c *Caret) seekEnd() {
 	c.p = c.f.lastParagraph()
 	c.l = c.f.paragraphs[c.p].lastLine(c.f)
 	c.b = c.f.lines[c.l].lastBox(c.f)
 	c.k = c.f.boxes[c.b].j
 	c.pos = int32(c.f.len)
 }

 // calculatePLBK ensures that the Caret's cached p, l, b and k fields are
 // valid.
 func (c *Caret) calculatePLBK() {
 	if c.seqNum != c.f.seqNum {
 		c.seek(c.pos)
 	}
 }

 // Seek satisfies the io.Seeker interface.
 func (c *Caret) Seek(offset int64, whence int) (int64, error) {
 	switch whence {
 	case SeekSet:
 		// No-op.
 	case SeekCur:
 		offset += int64(c.pos)
 	case SeekEnd:
 		offset += int64(c.f.len)
 	default:
 		return 0, errors.New("text: invalid seek whence")
 	}
 	if offset < 0 {
 		return 0, errors.New("text: negative seek position")
 	}
 	if offset > int64(c.f.len) {
 		offset = int64(c.f.len)
 	}
 	c.seek(int32(offset))
 	return offset, nil
 }

 func (c *Caret) seek(off int32) {
 	delta := off - c.pos
 	// If the new offset is closer to the start or the end than to the current
 	// c.pos, or if c's cached {p,l,b,k} values are invalid, move to the start
 	// or end first. In case of a tie, we prefer to seek forwards (i.e. set
 	// delta > 0).
 	if (delta < 0 && -delta >= off) || (c.seqNum != c.f.seqNum) {
 		c.seekStart()
 		delta = off - c.pos
 	}
 	if delta > 0 && delta > int32(c.f.len)-off {
 		c.seekEnd()
 		delta = off - c.pos
 	}

 	if delta != 0 {
 		// Seek forwards.
 		for delta > 0 {
 			if n := c.f.boxes[c.b].j - c.k; n > 0 {
 				if n > delta {
 					n = delta
 				}
 				c.pos += n
 				c.k += n
 				delta -= n
 			} else if c.leanForwards() != leanOK {
 				panic("text: invalid state")
 			}
 		}

 		// Seek backwards.
 		for delta < 0 {
 			if n := c.f.boxes[c.b].i - c.k; n < 0 {
 				if n < delta {
 					n = delta
 				}
 				c.pos += n
 				c.k += n
 				delta -= n
 			} else if c.leanBackwards() != leanOK {
 				panic("text: invalid state")
 			}
 		}

 		// A Caret can't be placed at the end of a Paragraph, unless it is the
 		// final Paragraph. A simple way to enforce this is to lean forwards.
 		c.leanForwards()
 	}

 	c.seqNum = c.f.seqNum
 }

 // Read satisfies the io.Reader interface by copying those bytes after the
 // Caret and incrementing the Caret.
 func (c *Caret) Read(buf []byte) (n int, err error) {
 	c.calculatePLBK()
 	for len(buf) > 0 {
 		if j := c.f.boxes[c.b].j; c.k < j {
 			nn := copy(buf, c.f.text[c.k:j])
 			buf = buf[nn:]
 			n += nn
 			c.pos += int32(nn)
 			c.k += int32(nn)
 		}
 		// A Caret can't be placed at the end of a Paragraph, unless it is the
 		// final Paragraph. A simple way to enforce this is to lean forwards.
 		if c.leanForwards() == leanFailedEOF {
 			break
 		}
 	}
 	if int(c.pos) == c.f.len {
 		err = io.EOF
 	}
 	return n, err
 }

 // ReadByte returns the next byte after the Caret and increments the Caret.
 func (c *Caret) ReadByte() (x byte, err error) {
 	c.calculatePLBK()
 	for {
 		if j := c.f.boxes[c.b].j; c.k < j {
 			x = c.f.text[c.k]
 			c.pos++
 			c.k++
 			// A Caret can't be placed at the end of a Paragraph, unless it is
 			// the final Paragraph. A simple way to enforce this is to lean
 			// forwards.
 			c.leanForwards()
 			return x, nil
 		}
 		if c.leanForwards() == leanFailedEOF {
 			return 0, io.EOF
 		}
 	}
 }

 // ReadRune returns the next rune after the Caret and increments the Caret.
 func (c *Caret) ReadRune() (r rune, size int, err error) {
 	c.calculatePLBK()
 	for {
 		if c.k < c.f.boxes[c.b].j {
 			r, size, c.b, c.k = c.f.readRune(c.b, c.k)
 			c.pos += int32(size)
 			// A Caret can't be placed at the end of a Paragraph, unless it is
 			// the final Paragraph. A simple way to enforce this is to lean
 			// forwards.
 			c.leanForwards()
 			return r, size, nil
 		}
 		if c.leanForwards() == leanFailedEOF {
 			return 0, 0, io.EOF
 		}
 	}
 }

 // WriteByte inserts x into the Frame's text at the Caret and increments the
 // Caret.
 func (c *Caret) WriteByte(x byte) error {
 	c.tmp[0] = x
 	return c.write(c.tmp[:1], "")
 }

 // WriteRune inserts r into the Frame's text at the Caret and increments the
 // Caret.
 func (c *Caret) WriteRune(r rune) (size int, err error) {
 	size = utf8.EncodeRune(c.tmp[:], r)
 	if err = c.write(c.tmp[:size], ""); err != nil {
 		return 0, err
 	}
 	return size, nil
 }

 // WriteString inserts s into the Frame's text at the Caret and increments the
 // Caret.
 func (c *Caret) WriteString(s string) (n int, err error) {
 	for len(s) > 0 {
 		i := 1 + strings.IndexByte(s, '\n')
 		if i == 0 {
 			i = len(s)
 		}
 		if err = c.write(nil, s[:i]); err != nil {
 			break
 		}
 		n += i
 		s = s[i:]
 	}
 	return n, err
 }

 // Write inserts s into the Frame's text at the Caret and increments the Caret.
 func (c *Caret) Write(s []byte) (n int, err error) {
 	for len(s) > 0 {
 		i := 1 + bytes.IndexByte(s, '\n')
 		if i == 0 {
 			i = len(s)
 		}
 		if err = c.write(s[:i], ""); err != nil {
 			break
 		}
 		n += i
 		s = s[i:]
 	}
 	return n, err
 }

 // write inserts a []byte or string into the Frame's text at the Caret.
 //
 // Exactly one of s0 and s1 must be non-empty. That non-empty argument must
 // contain at most one '\n' and if it does contain one, it must be the final
 // byte.
 func (c *Caret) write(s0 []byte, s1 string) error {
 	if m := maxLen - len(c.f.text); len(s0) > m || len(s1) > m {
 		return errors.New("text: insufficient space for writing")
 	}

 	// Ensure that the Caret is at the end of its Box, and that Box's text is
 	// at the end of the Frame's buffer.
 	c.calculatePLBK()
 	for {
 		bb, n := &c.f.boxes[c.b], int32(len(c.f.text))
 		if c.k == bb.j && c.k == n {
 			break
 		}

 		// If the Box's text is empty, move its empty i:j range to the
 		// equivalent empty range at the end of c.f.text.
 		if bb.i == bb.j {
 			bb.i = n
 			bb.j = n
 			for _, cc := range c.f.carets {
 				if cc.b == c.b {
 					cc.k = n
 				}
 			}
 			continue
 		}

 		// Make the Caret be at the end of its Box.
 		if c.k != bb.j {
 			c.splitBox(true)
 			continue
 		}

 		// Make a new empty Box and move the Caret to it.
 		c.splitBox(true)
 		c.leanForwards()
 	}

 	c.f.invalidateCaches()
 	c.f.paragraphs[c.p].invalidateCaches()
 	c.f.lines[c.l].invalidateCaches()

 	length, nl := len(s0), false
 	if length > 0 {
 		nl = s0[length-1] == '\n'
 		c.f.text = append(c.f.text, s0...)
 	} else {
 		length = len(s1)
 		nl = s1[length-1] == '\n'
 		c.f.text = append(c.f.text, s1...)
 	}
 	c.f.len += length
 	c.f.boxes[c.b].j += int32(length)
 	c.k += int32(length)
 	for _, cc := range c.f.carets {
 		if cc.pos > c.pos {
 			cc.pos += int32(length)
 		}
 	}
 	c.pos += int32(length)
 	oldL := c.l

 	if nl {
 		breakParagraph(c.f, c.p, c.l, c.b)
 		c.p = c.f.paragraphs[c.p].next
 		c.l = c.f.paragraphs[c.p].firstL
 		c.b = c.f.lines[c.l].firstB
 		c.k = c.f.boxes[c.b].i
 	}

 	// TODO: re-layout the new c.p paragraph, if we saw '\n'.
 	layout(c.f, oldL)

 	c.f.seqNum++
 	return nil
 }

 // breakParagraph breaks the Paragraph p into two Paragraphs, just after Box b
 // in Line l in Paragraph p. b's text must end with a '\n'. The new Paragraph
 // is inserted after p.
 func breakParagraph(f *Frame, p, l, b int32) {
 	// Assert that the Box b's text ends with a '\n'.
 	if j := f.boxes[b].j; j == 0 || f.text[j-1] != '\n' {
 		panic("text: invalid state")
 	}

 	// Make a new, empty Paragraph after this Paragraph p.
 	newP, _ := f.newParagraph()
 	nextP := f.paragraphs[p].next
 	if nextP != 0 {
 		f.paragraphs[nextP].prev = newP
 	}
 	f.paragraphs[newP].next = nextP
 	f.paragraphs[newP].prev = p
 	f.paragraphs[p].next = newP

 	// Any Lines in this Paragraph after the break point's Line l move to the
 	// newP Paragraph.
 	if nextL := f.lines[l].next; nextL != 0 {
 		f.lines[l].next = 0
 		f.lines[nextL].prev = 0
 		f.paragraphs[newP].firstL = nextL
 	}

 	// Any Boxes in this Line after the break point's Box b move to a new Line
 	// at the start of the newP Paragraph.
 	if nextB := f.boxes[b].next; nextB != 0 {
 		f.boxes[b].next = 0
 		f.boxes[nextB].prev = 0
 		newL, _ := f.newLine()
 		f.lines[newL].firstB = nextB
 		if newPFirstL := f.paragraphs[newP].firstL; newPFirstL != 0 {
 			f.lines[newL].next = newPFirstL
 			f.lines[newPFirstL].prev = newL
 		}
 		f.paragraphs[newP].firstL = newL
 	}

 	// Make the newP Paragraph's first Line and first Box explicit, since
 	// Carets require an explicit p, l and b.
 	{
 		pp := &f.paragraphs[newP]
 		if pp.firstL == 0 {
 			pp.firstL, _ = f.newLine()
 		}
 		ll := &f.lines[pp.firstL]
 		if ll.firstB == 0 {
 			ll.firstB, _ = f.newBox()
 		}
 	}

 	// TODO: re-layout the newP paragraph.
 }

 // breakLine breaks the Line l at text index k in Box b. The b-and-k index must
 // not be at the start or end of the Line. Text to the right of b-and-k in the
 // Line l will be moved to the start of the next Line in the Paragraph, with
 // that next Line being created if it didn't already exist.
 func breakLine(f *Frame, l, b, k int32) {
 	// Split this Box into two if necessary, so that k equals a Box's j end.
 	bb := &f.boxes[b]
 	if k != bb.j {
 		if k == bb.i {
 			panic("TODO: degenerate split left, possibly adjusting the Line's firstB??")
 		}
 		newB, realloc := f.newBox()
 		if realloc {
 			bb = &f.boxes[b]
 		}
 		nextB := bb.next
 		if nextB != 0 {
 			f.boxes[nextB].prev = newB
 		}
 		f.boxes[newB].next = nextB
 		f.boxes[newB].prev = b
 		f.boxes[newB].i = k
 		f.boxes[newB].j = bb.j
 		bb.next = newB
 		bb.j = k
 	}

 	// Assert that the break point isn't already at the start or end of the Line.
 	if bb.next == 0 || (bb.prev == 0 && k == bb.i) {
 		panic("text: invalid state")
 	}

 	// Insert a line after this one, if one doesn't already exist.
 	ll := &f.lines[l]
 	if ll.next == 0 {
 		newL, realloc := f.newLine()
 		if realloc {
 			ll = &f.lines[l]
 		}
 		f.lines[newL].prev = l
 		ll.next = newL
 	}

 	// Move the remaining boxes to the next line.
 	nextB, nextL := bb.next, ll.next
 	bb.next = 0
 	f.boxes[nextB].prev = 0
 	fb := f.lines[nextL].firstB
 	f.lines[nextL].firstB = nextB

 	// If the next Line already contained Boxes, append them to the end of the
 	// nextB chain, and join the two newly linked Boxes if possible.
 	if fb != 0 {
 		lb := f.lines[nextL].lastBox(f)
 		lbb := &f.boxes[lb]
 		fbb := &f.boxes[fb]
 		lbb.next = fb
 		fbb.prev = lb
 		f.joinBoxes(lb, fb, lbb, fbb)
 	}
 }

 // layout inserts a soft return in the Line l if its text measures longer than
 // f.maxWidth and a suitable line break point is found. This may spill text
 // onto the next line, which will also be laid out, and so on recursively.
 func layout(f *Frame, l int32) {
 	if f.maxWidth <= 0 || f.face == nil {
 		return
 	}
 	f.seqNum++

 	for ; l != 0; l = f.lines[l].next {
 		var (
 			firstB     = f.lines[l].firstB
 			reader     = f.lineReader(firstB, f.boxes[firstB].i)
 			breakPoint bAndK
 			prevR      rune
 			prevRValid bool
 			advance    fixed.Int26_6
 		)
 		for {
 			r, _, err := reader.ReadRune()
 			if err != nil || r == '\n' {
 				return
 			}
 			if prevRValid {
 				advance += f.face.Kern(prevR, r)
 			}
 			// TODO: match all whitespace, not just ' '?
 			if r == ' ' {
 				breakPoint = reader.bAndK()
 			}
 			a, ok := f.face.GlyphAdvance(r)
 			if !ok {
 				panic("TODO: is falling back on the U+FFFD glyph the responsibility of the caller or the Face?")
 			}
 			advance += a
 			if r != ' ' && advance > f.maxWidth && breakPoint.b != 0 {
 				breakLine(f, l, breakPoint.b, breakPoint.k)
 				break
 			}
 			prevR, prevRValid = r, true
 		}
 	}
 }

 // Delete deletes nBytes bytes in the specified direction from the Caret's
 // location. It returns the number of bytes deleted, which can be fewer than
 // that requested if it hits the beginning or end of the Frame.
 func (c *Caret) Delete(dir Direction, nBytes int) (dBytes int) {
 	if nBytes <= 0 {
 		return 0
 	}

 	// Convert a backwards delete of n bytes from position p to a forwards
 	// delete of n bytes from position p-n.
 	//
 	// In general, it's easier to delete forwards than backwards. For example,
 	// when crossing paragraph boundaries, it's easier to find the first line
 	// of the next paragraph than the last line of the previous paragraph.
 	if dir == Backwards {
 		newPos := int(c.pos) - nBytes
 		if newPos < 0 {
 			newPos = 0
 			nBytes = int(c.pos)
 			if nBytes == 0 {
 				return 0
 			}
 		}
 		c.seek(int32(newPos))
 	}

 	if int(c.pos) == c.f.len {
 		return 0
 	}

 	c.calculatePLBK()
 	c.leanForwards()
 	if c.f.boxes[c.b].i != c.k && c.splitBox(false) {
 		c.leanForwards()
 	}
 	for nBytes > 0 && int(c.pos) != c.f.len {
 		bb := &c.f.boxes[c.b]
 		n := bb.j - bb.i
 		newLine := n != 0 && c.f.text[bb.j-1] == '\n'
 		if int(n) > nBytes {
 			n = int32(nBytes)
 		}
 		bb.i += n
 		c.k += n
 		dBytes += int(n)
 		nBytes -= int(n)
 		c.f.len -= int(n)

 		if bb.i != bb.j {
 			break
 		}

 		if newLine {
 			c.joinNextParagraph()
 		}
 		c.leanForwards()
 	}

 	// The mergeIntoOneLine will shake out any empty Boxes.
 	l := c.f.mergeIntoOneLine(c.p)
 	layout(c.f, l)
 	c.f.invalidateCaches()

 	// Compact c.f.text if it's large enough and the fraction of deleted text
 	// is above some threshold. The actual threshold value (25%) is arbitrary.
 	// A lower value means more frequent compactions, so less memory on average
 	// but more CPU. A higher value means the opposite.
 	if len(c.f.text) > 4096 && len(c.f.text)/4 < c.f.deletedLen() {
 		c.f.compactText()
 	}

 	c.f.seqNum++
 	for _, cc := range c.f.carets {
 		if cc == c {
 			continue
 		}
 		switch relPos := cc.pos - c.pos; {
 		case relPos <= 0:
 			// No-op.
 		case relPos <= int32(dBytes):
 			cc.pos = c.pos
 		default:
 			cc.pos -= int32(dBytes)
 		}
 	}
 	return dBytes
 }

 // DeleteRunes deletes nRunes runes in the specified direction from the Caret's
 // location. It returns the number of runes and bytes deleted, which can be
 // fewer than that requested if it hits the beginning or end of the Frame.
 func (c *Caret) DeleteRunes(dir Direction, nRunes int) (dRunes, dBytes int) {
 	// Save the current Caret position, move the Caret by nRunes runes to
 	// calculate how many bytes to delete, restore that saved Caret position,
 	// then delete that many bytes.
 	c.calculatePLBK()
 	savedC := *c
 	if dir == Forwards {
 		for dRunes < nRunes {
 			var size int
 			_, size, c.b, c.k = c.f.readRune(c.b, c.k)
 			if size != 0 {
 				dRunes++
 				dBytes += size
 			} else if c.leanForwards() != leanOK {
 				break
 			}
 		}
 	} else {
 		for dRunes < nRunes {
 			var size int
 			_, size, c.b, c.k = c.f.readLastRune(c.b, c.k)
 			if size != 0 {
 				dRunes++
 				dBytes += size
 			} else if c.leanBackwards() != leanOK {
 				break
 			}
 		}
 	}
 	*c = savedC
 	if dBytes != c.Delete(dir, dBytes) {
 		panic("text: invalid state")
 	}
 	return dRunes, dBytes
 }

 // joinNextParagraph joins c's current and next Paragraph. That next Paragraph
 // must exist, and c must be at the last Line of its current Paragraph.
 func (c *Caret) joinNextParagraph() {
 	pp0 := &c.f.paragraphs[c.p]
 	ll0 := &c.f.lines[c.l]
 	if pp0.next == 0 || ll0.next != 0 {
 		panic("text: invalid state")
 	}
 	pp1 := &c.f.paragraphs[pp0.next]
 	l1 := pp1.firstL

 	ll0.next = l1
 	c.f.lines[l1].prev = c.l

 	toFree := pp0.next
 	pp0.next = pp1.next
 	if pp0.next != 0 {
 		c.f.paragraphs[pp0.next].prev = c.p
 	}

 	c.f.freeParagraph(toFree)
 }

 // splitBox splits the Caret's Box into two, at the Caret's location. Unless
 // force is set, it does nothing if the Caret is at either edge of its Box. It
 // returns whether the Box was split. If so, the new Box is created after, not
 // before, the Caret's current Box.
 func (c *Caret) splitBox(force bool) bool {
 	bb := &c.f.boxes[c.b]
 	if !force && (c.k == bb.i || c.k == bb.j) {
 		return false
 	}
 	newB, realloc := c.f.newBox()
 	if realloc {
 		bb = &c.f.boxes[c.b]
 	}
 	nextB := bb.next
 	if nextB != 0 {
 		c.f.boxes[nextB].prev = newB
 	}
 	c.f.boxes[newB] = Box{
 		next: nextB,
 		prev: c.b,
 		i:    c.k,
 		j:    bb.j,
 	}
 	bb.next = newB
 	bb.j = c.k
 	return true
 }
	// Copyright 2016 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 text

	// TODO: do we care about "\n" vs "\r" vs "\r\n"? We only recognize "\n" for
	// now.

	import (
	"bytes"
	"errors"
	"io"
	"strings"
	"unicode/utf8"

	"golang.org/x/image/math/fixed"
	)

	// Caret is a location in a Frame's text, and is the mechanism for adding and
	// removing bytes of text. Conceptually, a Caret and a Frame's text is like an
	// int c and a []byte t such that the text before and after that Caret is t[:c]
	// and t[c:]. That byte-count location remains unchanged even when a Frame is
	// re-sized and laid out into a new tree of Paragraphs, Lines and Boxes.
	//
	// A Frame can have multiple open Carets. For example, the beginning and end of
	// a text selection can be represented by two Carets. Multiple Carets for the
	// one Frame are not safe to use concurrently, but it is valid to interleave
	// such operations sequentially. For example, if two Carets c0 and c1 for the
	// one Frame are positioned at the 10th and 20th byte, and 4 bytes are written
	// to c0, inserting what becomes the equivalent of text[10:14], then c0's
	// position is updated to be 14 but c1's position is also updated to be 24.
	type Caret struct {
	f *Frame

	// caretsIndex is the index of this Caret in the f.carets slice.
	caretsIndex int

	// seqNum is the Frame f's sequence number for which this Caret's cached p,
	// l, b and k fields are valid. If f has been modified since then, those
	// fields will have to be re-calculated based on the pos field (which is
	// always valid).
	//
	// TODO: when re-calculating p, l, b and k, be more efficient than a linear
	// scan from the start or end?
	seqNum uint64

	// p, l and b cache the index of the Caret's Paragraph, Line and Box. None
	// of these values can be zero.
	p, l, b int32

	// k caches the Caret's position in the text, in Frame.text order. It is
	// valid to index the Frame.text slice with k, analogous to the Box.i and
	// Box.j fields. For a Caret c, letting bb := c.f.boxes[c.b], an invariant
	// is that bb.i <= c.k && c.k <= bb.j if the cache is valid (i.e. the
	// Caret's seqNum equals the Frame's seqNum).
	k int32

	// pos is the Caret's position in the text, in layout order. It is the "c"
	// as in "t[:c]" in the doc comment for type Caret above. It is not valid
	// to index the Frame.text slice with pos, since the Frame.text slice does
	// not necessarily hold the textual content in layout order.
	pos int32

	tmp [utf8.UTFMax]byte
	}

	// Close closes the Caret.
	func (c *Caret) Close() error {
	i, j := c.caretsIndex, len(c.f.carets)-1

	// Swap c with the last element of c.f.carets.
	if i != j {
	other := c.f.carets[j]
	other.caretsIndex = i
	c.f.carets[i] = other
	}

	c.f.carets[j] = nil
	c.f.carets = c.f.carets[:j]
	*c = Caret{}
	return nil
	}

	type leanResult int

	const (
	// leanOK means that the lean changed the Caret's Box.
	leanOK leanResult = iota
	// leanFailedEOF means that the lean did not change the Caret's Box,
	// because the Caret was already at the end / beginning of the Frame (when
	// leaning forwards / backwards).
	leanFailedEOF
	// leanFailedNotEOB means that the lean did not change the Caret's Box,
	// because the Caret was not placed at the end / beginning of the Box (when
	// leaning forwards / backwards).
	leanFailedNotEOB
	)

	// leanForwards moves the Caret from the right end of one Box to the left end
	// of the next Box, crossing Lines and Paragraphs to find that next Box. It
	// returns whether the Caret moved to a different Box.
	//
	// Diagramatically, suppose we have two adjacent boxes (shown by square
	// brackets below), with the Caret (an integer location called Caret.pos in the
	// Frame's text) in the middle of the "foo2bar3" word:
	// [foo0 foo1 foo2]^[bar3 bar4 bar5]
	// leanForwards moves Caret.k from fooBox.j to barBox.i, also updating the
	// Caret's p, l and b. Caret.pos remains unchanged.
	func (c *Caret) leanForwards() leanResult {
	if c.k != c.f.boxes[c.b].j {
	return leanFailedNotEOB
	}
	if nextB := c.f.boxes[c.b].next; nextB != 0 {
	c.b = nextB
	c.k = c.f.boxes[c.b].i
	return leanOK
	}
	if nextL := c.f.lines[c.l].next; nextL != 0 {
	c.l = nextL
	c.b = c.f.lines[c.l].firstB
	c.k = c.f.boxes[c.b].i
	return leanOK
	}
	if nextP := c.f.paragraphs[c.p].next; nextP != 0 {
	c.p = nextP
	c.l = c.f.paragraphs[c.p].firstL
	c.b = c.f.lines[c.l].firstB
	c.k = c.f.boxes[c.b].i
	return leanOK
	}
	return leanFailedEOF
	}

	// leanBackwards is like leanForwards but in the other direction.
	func (c *Caret) leanBackwards() leanResult {
	if c.k != c.f.boxes[c.b].i {
	return leanFailedNotEOB
	}
	if prevB := c.f.boxes[c.b].prev; prevB != 0 {
	c.b = prevB
	c.k = c.f.boxes[c.b].j
	return leanOK
	}
	if prevL := c.f.lines[c.l].prev; prevL != 0 {
	c.l = prevL
	c.b = c.f.lines[c.l].lastBox(c.f)
	c.k = c.f.boxes[c.b].j
	return leanOK
	}
	if prevP := c.f.paragraphs[c.p].prev; prevP != 0 {
	c.p = prevP
	c.l = c.f.paragraphs[c.p].lastLine(c.f)
	c.b = c.f.lines[c.l].lastBox(c.f)
	c.k = c.f.boxes[c.b].j
	return leanOK
	}
	return leanFailedEOF
	}

	func (c *Caret) seekStart() {
	c.p = c.f.firstP
	c.l = c.f.paragraphs[c.p].firstL
	c.b = c.f.lines[c.l].firstB
	c.k = c.f.boxes[c.b].i
	c.pos = 0
	}

	func (c *Caret) seekEnd() {
	c.p = c.f.lastParagraph()
	c.l = c.f.paragraphs[c.p].lastLine(c.f)
	c.b = c.f.lines[c.l].lastBox(c.f)
	c.k = c.f.boxes[c.b].j
	c.pos = int32(c.f.len)
	}

	// calculatePLBK ensures that the Caret's cached p, l, b and k fields are
	// valid.
	func (c *Caret) calculatePLBK() {
	if c.seqNum != c.f.seqNum {
	c.seek(c.pos)
	}
	}

	// Seek satisfies the io.Seeker interface.
	func (c *Caret) Seek(offset int64, whence int) (int64, error) {
	switch whence {
	case SeekSet:
	// No-op.
	case SeekCur:
	offset += int64(c.pos)
	case SeekEnd:
	offset += int64(c.f.len)
	default:
	return 0, errors.New("text: invalid seek whence")
	}
	if offset < 0 {
	return 0, errors.New("text: negative seek position")
	}
	if offset > int64(c.f.len) {
	offset = int64(c.f.len)
	}
	c.seek(int32(offset))
	return offset, nil
	}

	func (c *Caret) seek(off int32) {
	delta := off - c.pos
	// If the new offset is closer to the start or the end than to the current
	// c.pos, or if c's cached {p,l,b,k} values are invalid, move to the start
	// or end first. In case of a tie, we prefer to seek forwards (i.e. set
	// delta > 0).
	if (delta < 0 && -delta >= off) \|\| (c.seqNum != c.f.seqNum) {
	c.seekStart()
	delta = off - c.pos
	}
	if delta > 0 && delta > int32(c.f.len)-off {
	c.seekEnd()
	delta = off - c.pos
	}

	if delta != 0 {
	// Seek forwards.
	for delta > 0 {
	if n := c.f.boxes[c.b].j - c.k; n > 0 {
	if n > delta {
	n = delta
	}
	c.pos += n
	c.k += n
	delta -= n
	} else if c.leanForwards() != leanOK {
	panic("text: invalid state")
	}
	}

	// Seek backwards.
	for delta < 0 {
	if n := c.f.boxes[c.b].i - c.k; n < 0 {
	if n < delta {
	n = delta
	}
	c.pos += n
	c.k += n
	delta -= n
	} else if c.leanBackwards() != leanOK {
	panic("text: invalid state")
	}
	}

	// A Caret can't be placed at the end of a Paragraph, unless it is the
	// final Paragraph. A simple way to enforce this is to lean forwards.
	c.leanForwards()
	}

	c.seqNum = c.f.seqNum
	}

	// Read satisfies the io.Reader interface by copying those bytes after the
	// Caret and incrementing the Caret.
	func (c *Caret) Read(buf []byte) (n int, err error) {
	c.calculatePLBK()
	for len(buf) > 0 {
	if j := c.f.boxes[c.b].j; c.k < j {
	nn := copy(buf, c.f.text[c.k:j])
	buf = buf[nn:]
	n += nn
	c.pos += int32(nn)
	c.k += int32(nn)
	}
	// A Caret can't be placed at the end of a Paragraph, unless it is the
	// final Paragraph. A simple way to enforce this is to lean forwards.
	if c.leanForwards() == leanFailedEOF {
	break
	}
	}
	if int(c.pos) == c.f.len {
	err = io.EOF
	}
	return n, err
	}

	// ReadByte returns the next byte after the Caret and increments the Caret.
	func (c *Caret) ReadByte() (x byte, err error) {
	c.calculatePLBK()
	for {
	if j := c.f.boxes[c.b].j; c.k < j {
	x = c.f.text[c.k]
	c.pos++
	c.k++
	// A Caret can't be placed at the end of a Paragraph, unless it is
	// the final Paragraph. A simple way to enforce this is to lean
	// forwards.
	c.leanForwards()
	return x, nil
	}
	if c.leanForwards() == leanFailedEOF {
	return 0, io.EOF
	}
	}
	}

	// ReadRune returns the next rune after the Caret and increments the Caret.
	func (c *Caret) ReadRune() (r rune, size int, err error) {
	c.calculatePLBK()
	for {
	if c.k < c.f.boxes[c.b].j {
	r, size, c.b, c.k = c.f.readRune(c.b, c.k)
	c.pos += int32(size)
	// A Caret can't be placed at the end of a Paragraph, unless it is
	// the final Paragraph. A simple way to enforce this is to lean
	// forwards.
	c.leanForwards()
	return r, size, nil
	}
	if c.leanForwards() == leanFailedEOF {
	return 0, 0, io.EOF
	}
	}
	}

	// WriteByte inserts x into the Frame's text at the Caret and increments the
	// Caret.
	func (c *Caret) WriteByte(x byte) error {
	c.tmp[0] = x
	return c.write(c.tmp[:1], "")
	}

	// WriteRune inserts r into the Frame's text at the Caret and increments the
	// Caret.
	func (c *Caret) WriteRune(r rune) (size int, err error) {
	size = utf8.EncodeRune(c.tmp[:], r)
	if err = c.write(c.tmp[:size], ""); err != nil {
	return 0, err
	}
	return size, nil
	}

	// WriteString inserts s into the Frame's text at the Caret and increments the
	// Caret.
	func (c *Caret) WriteString(s string) (n int, err error) {
	for len(s) > 0 {
	i := 1 + strings.IndexByte(s, '\n')
	if i == 0 {
	i = len(s)
	}
	if err = c.write(nil, s[:i]); err != nil {
	break
	}
	n += i
	s = s[i:]
	}
	return n, err
	}

	// Write inserts s into the Frame's text at the Caret and increments the Caret.
	func (c *Caret) Write(s []byte) (n int, err error) {
	for len(s) > 0 {
	i := 1 + bytes.IndexByte(s, '\n')
	if i == 0 {
	i = len(s)
	}
	if err = c.write(s[:i], ""); err != nil {
	break
	}
	n += i
	s = s[i:]
	}
	return n, err
	}

	// write inserts a []byte or string into the Frame's text at the Caret.
	//
	// Exactly one of s0 and s1 must be non-empty. That non-empty argument must
	// contain at most one '\n' and if it does contain one, it must be the final
	// byte.
	func (c *Caret) write(s0 []byte, s1 string) error {
	if m := maxLen - len(c.f.text); len(s0) > m \|\| len(s1) > m {
	return errors.New("text: insufficient space for writing")
	}

	// Ensure that the Caret is at the end of its Box, and that Box's text is
	// at the end of the Frame's buffer.
	c.calculatePLBK()
	for {
	bb, n := &c.f.boxes[c.b], int32(len(c.f.text))
	if c.k == bb.j && c.k == n {
	break
	}

	// If the Box's text is empty, move its empty i:j range to the
	// equivalent empty range at the end of c.f.text.
	if bb.i == bb.j {
	bb.i = n
	bb.j = n
	for _, cc := range c.f.carets {
	if cc.b == c.b {
	cc.k = n
	}
	}
	continue
	}

	// Make the Caret be at the end of its Box.
	if c.k != bb.j {
	c.splitBox(true)
	continue
	}

	// Make a new empty Box and move the Caret to it.
	c.splitBox(true)
	c.leanForwards()
	}

	c.f.invalidateCaches()
	c.f.paragraphs[c.p].invalidateCaches()
	c.f.lines[c.l].invalidateCaches()

	length, nl := len(s0), false
	if length > 0 {
	nl = s0[length-1] == '\n'
	c.f.text = append(c.f.text, s0...)
	} else {
	length = len(s1)
	nl = s1[length-1] == '\n'
	c.f.text = append(c.f.text, s1...)
	}
	c.f.len += length
	c.f.boxes[c.b].j += int32(length)
	c.k += int32(length)
	for _, cc := range c.f.carets {
	if cc.pos > c.pos {
	cc.pos += int32(length)
	}
	}
	c.pos += int32(length)
	oldL := c.l

	if nl {
	breakParagraph(c.f, c.p, c.l, c.b)
	c.p = c.f.paragraphs[c.p].next
	c.l = c.f.paragraphs[c.p].firstL
	c.b = c.f.lines[c.l].firstB
	c.k = c.f.boxes[c.b].i
	}

	// TODO: re-layout the new c.p paragraph, if we saw '\n'.
	layout(c.f, oldL)

	c.f.seqNum++
	return nil
	}

	// breakParagraph breaks the Paragraph p into two Paragraphs, just after Box b
	// in Line l in Paragraph p. b's text must end with a '\n'. The new Paragraph
	// is inserted after p.
	func breakParagraph(f *Frame, p, l, b int32) {
	// Assert that the Box b's text ends with a '\n'.
	if j := f.boxes[b].j; j == 0 \|\| f.text[j-1] != '\n' {
	panic("text: invalid state")
	}

	// Make a new, empty Paragraph after this Paragraph p.
	newP, _ := f.newParagraph()
	nextP := f.paragraphs[p].next
	if nextP != 0 {
	f.paragraphs[nextP].prev = newP
	}
	f.paragraphs[newP].next = nextP
	f.paragraphs[newP].prev = p
	f.paragraphs[p].next = newP

	// Any Lines in this Paragraph after the break point's Line l move to the
	// newP Paragraph.
	if nextL := f.lines[l].next; nextL != 0 {
	f.lines[l].next = 0
	f.lines[nextL].prev = 0
	f.paragraphs[newP].firstL = nextL
	}

	// Any Boxes in this Line after the break point's Box b move to a new Line
	// at the start of the newP Paragraph.
	if nextB := f.boxes[b].next; nextB != 0 {
	f.boxes[b].next = 0
	f.boxes[nextB].prev = 0
	newL, _ := f.newLine()
	f.lines[newL].firstB = nextB
	if newPFirstL := f.paragraphs[newP].firstL; newPFirstL != 0 {
	f.lines[newL].next = newPFirstL
	f.lines[newPFirstL].prev = newL
	}
	f.paragraphs[newP].firstL = newL
	}

	// Make the newP Paragraph's first Line and first Box explicit, since
	// Carets require an explicit p, l and b.
	{
	pp := &f.paragraphs[newP]
	if pp.firstL == 0 {
	pp.firstL, _ = f.newLine()
	}
	ll := &f.lines[pp.firstL]
	if ll.firstB == 0 {
	ll.firstB, _ = f.newBox()
	}
	}

	// TODO: re-layout the newP paragraph.
	}

	// breakLine breaks the Line l at text index k in Box b. The b-and-k index must
	// not be at the start or end of the Line. Text to the right of b-and-k in the
	// Line l will be moved to the start of the next Line in the Paragraph, with
	// that next Line being created if it didn't already exist.
	func breakLine(f *Frame, l, b, k int32) {
	// Split this Box into two if necessary, so that k equals a Box's j end.
	bb := &f.boxes[b]
	if k != bb.j {
	if k == bb.i {
	panic("TODO: degenerate split left, possibly adjusting the Line's firstB??")
	}
	newB, realloc := f.newBox()
	if realloc {
	bb = &f.boxes[b]
	}
	nextB := bb.next
	if nextB != 0 {
	f.boxes[nextB].prev = newB
	}
	f.boxes[newB].next = nextB
	f.boxes[newB].prev = b
	f.boxes[newB].i = k
	f.boxes[newB].j = bb.j
	bb.next = newB
	bb.j = k
	}

	// Assert that the break point isn't already at the start or end of the Line.
	if bb.next == 0 \|\| (bb.prev == 0 && k == bb.i) {
	panic("text: invalid state")
	}

	// Insert a line after this one, if one doesn't already exist.
	ll := &f.lines[l]
	if ll.next == 0 {
	newL, realloc := f.newLine()
	if realloc {
	ll = &f.lines[l]
	}
	f.lines[newL].prev = l
	ll.next = newL
	}

	// Move the remaining boxes to the next line.
	nextB, nextL := bb.next, ll.next
	bb.next = 0
	f.boxes[nextB].prev = 0
	fb := f.lines[nextL].firstB
	f.lines[nextL].firstB = nextB

	// If the next Line already contained Boxes, append them to the end of the
	// nextB chain, and join the two newly linked Boxes if possible.
	if fb != 0 {
	lb := f.lines[nextL].lastBox(f)
	lbb := &f.boxes[lb]
	fbb := &f.boxes[fb]
	lbb.next = fb
	fbb.prev = lb
	f.joinBoxes(lb, fb, lbb, fbb)
	}
	}

	// layout inserts a soft return in the Line l if its text measures longer than
	// f.maxWidth and a suitable line break point is found. This may spill text
	// onto the next line, which will also be laid out, and so on recursively.
	func layout(f *Frame, l int32) {
	if f.maxWidth <= 0 \|\| f.face == nil {
	return
	}
	f.seqNum++

	for ; l != 0; l = f.lines[l].next {
	var (
	firstB = f.lines[l].firstB
	reader = f.lineReader(firstB, f.boxes[firstB].i)
	breakPoint bAndK
	prevR rune
	prevRValid bool
	advance fixed.Int26_6
	)
	for {
	r, _, err := reader.ReadRune()
	if err != nil \|\| r == '\n' {
	return
	}
	if prevRValid {
	advance += f.face.Kern(prevR, r)
	}
	// TODO: match all whitespace, not just ' '?
	if r == ' ' {
	breakPoint = reader.bAndK()
	}
	a, ok := f.face.GlyphAdvance(r)
	if !ok {
	panic("TODO: is falling back on the U+FFFD glyph the responsibility of the caller or the Face?")
	}
	advance += a
	if r != ' ' && advance > f.maxWidth && breakPoint.b != 0 {
	breakLine(f, l, breakPoint.b, breakPoint.k)
	break
	}
	prevR, prevRValid = r, true
	}
	}
	}

	// Delete deletes nBytes bytes in the specified direction from the Caret's
	// location. It returns the number of bytes deleted, which can be fewer than
	// that requested if it hits the beginning or end of the Frame.
	func (c *Caret) Delete(dir Direction, nBytes int) (dBytes int) {
	if nBytes <= 0 {
	return 0
	}

	// Convert a backwards delete of n bytes from position p to a forwards
	// delete of n bytes from position p-n.
	//
	// In general, it's easier to delete forwards than backwards. For example,
	// when crossing paragraph boundaries, it's easier to find the first line
	// of the next paragraph than the last line of the previous paragraph.
	if dir == Backwards {
	newPos := int(c.pos) - nBytes
	if newPos < 0 {
	newPos = 0
	nBytes = int(c.pos)
	if nBytes == 0 {
	return 0
	}
	}
	c.seek(int32(newPos))
	}

	if int(c.pos) == c.f.len {
	return 0
	}

	c.calculatePLBK()
	c.leanForwards()
	if c.f.boxes[c.b].i != c.k && c.splitBox(false) {
	c.leanForwards()
	}
	for nBytes > 0 && int(c.pos) != c.f.len {
	bb := &c.f.boxes[c.b]
	n := bb.j - bb.i
	newLine := n != 0 && c.f.text[bb.j-1] == '\n'
	if int(n) > nBytes {
	n = int32(nBytes)
	}
	bb.i += n
	c.k += n
	dBytes += int(n)
	nBytes -= int(n)
	c.f.len -= int(n)

	if bb.i != bb.j {
	break
	}

	if newLine {
	c.joinNextParagraph()
	}
	c.leanForwards()
	}

	// The mergeIntoOneLine will shake out any empty Boxes.
	l := c.f.mergeIntoOneLine(c.p)
	layout(c.f, l)
	c.f.invalidateCaches()

	// Compact c.f.text if it's large enough and the fraction of deleted text
	// is above some threshold. The actual threshold value (25%) is arbitrary.
	// A lower value means more frequent compactions, so less memory on average
	// but more CPU. A higher value means the opposite.
	if len(c.f.text) > 4096 && len(c.f.text)/4 < c.f.deletedLen() {
	c.f.compactText()
	}

	c.f.seqNum++
	for _, cc := range c.f.carets {
	if cc == c {
	continue
	}
	switch relPos := cc.pos - c.pos; {
	case relPos <= 0:
	// No-op.
	case relPos <= int32(dBytes):
	cc.pos = c.pos
	default:
	cc.pos -= int32(dBytes)
	}
	}
	return dBytes
	}

	// DeleteRunes deletes nRunes runes in the specified direction from the Caret's
	// location. It returns the number of runes and bytes deleted, which can be
	// fewer than that requested if it hits the beginning or end of the Frame.
	func (c *Caret) DeleteRunes(dir Direction, nRunes int) (dRunes, dBytes int) {
	// Save the current Caret position, move the Caret by nRunes runes to
	// calculate how many bytes to delete, restore that saved Caret position,
	// then delete that many bytes.
	c.calculatePLBK()
	savedC := *c
	if dir == Forwards {
	for dRunes < nRunes {
	var size int
	_, size, c.b, c.k = c.f.readRune(c.b, c.k)
	if size != 0 {
	dRunes++
	dBytes += size
	} else if c.leanForwards() != leanOK {
	break
	}
	}
	} else {
	for dRunes < nRunes {
	var size int
	_, size, c.b, c.k = c.f.readLastRune(c.b, c.k)
	if size != 0 {
	dRunes++
	dBytes += size
	} else if c.leanBackwards() != leanOK {
	break
	}
	}
	}
	*c = savedC
	if dBytes != c.Delete(dir, dBytes) {
	panic("text: invalid state")
	}
	return dRunes, dBytes
	}

	// joinNextParagraph joins c's current and next Paragraph. That next Paragraph
	// must exist, and c must be at the last Line of its current Paragraph.
	func (c *Caret) joinNextParagraph() {
	pp0 := &c.f.paragraphs[c.p]
	ll0 := &c.f.lines[c.l]
	if pp0.next == 0 \|\| ll0.next != 0 {
	panic("text: invalid state")
	}
	pp1 := &c.f.paragraphs[pp0.next]
	l1 := pp1.firstL

	ll0.next = l1
	c.f.lines[l1].prev = c.l

	toFree := pp0.next
	pp0.next = pp1.next
	if pp0.next != 0 {
	c.f.paragraphs[pp0.next].prev = c.p
	}

	c.f.freeParagraph(toFree)
	}

	// splitBox splits the Caret's Box into two, at the Caret's location. Unless
	// force is set, it does nothing if the Caret is at either edge of its Box. It
	// returns whether the Box was split. If so, the new Box is created after, not
	// before, the Caret's current Box.
	func (c *Caret) splitBox(force bool) bool {
	bb := &c.f.boxes[c.b]
	if !force && (c.k == bb.i \|\| c.k == bb.j) {
	return false
	}
	newB, realloc := c.f.newBox()
	if realloc {
	bb = &c.f.boxes[c.b]
	}
	nextB := bb.next
	if nextB != 0 {
	c.f.boxes[nextB].prev = newB
	}
	c.f.boxes[newB] = Box{
	next: nextB,
	prev: c.b,
	i: c.k,
	j: bb.j,
	}
	bb.next = newB
	bb.j = c.k
	return true
	}