Checking in vendor folder for ease of using go get.

vendor/golang.org/x/text/unicode/bidi/bidi.go

@@ -0,0 +1,198 @@
+// Copyright 2015 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.
+
+//go:generate go run gen.go gen_trieval.go gen_ranges.go
+
+// Package bidi contains functionality for bidirectional text support.
+//
+// See http://www.unicode.org/reports/tr9.
+//
+// NOTE: UNDER CONSTRUCTION. This API may change in backwards incompatible ways
+// and without notice.
+package bidi // import "golang.org/x/text/unicode/bidi"
+
+// TODO:
+// The following functionality would not be hard to implement, but hinges on
+// the definition of a Segmenter interface. For now this is up to the user.
+// - Iterate over paragraphs
+// - Segmenter to iterate over runs directly from a given text.
+// Also:
+// - Transformer for reordering?
+// - Transformer (validator, really) for Bidi Rule.
+
+// This API tries to avoid dealing with embedding levels for now. Under the hood
+// these will be computed, but the question is to which extent the user should
+// know they exist. We should at some point allow the user to specify an
+// embedding hierarchy, though.
+
+// A Direction indicates the overall flow of text.
+type Direction int
+
+const (
+	// LeftToRight indicates the text contains no right-to-left characters and
+	// that either there are some left-to-right characters or the option
+	// DefaultDirection(LeftToRight) was passed.
+	LeftToRight Direction = iota
+
+	// RightToLeft indicates the text contains no left-to-right characters and
+	// that either there are some right-to-left characters or the option
+	// DefaultDirection(RightToLeft) was passed.
+	RightToLeft
+
+	// Mixed indicates text contains both left-to-right and right-to-left
+	// characters.
+	Mixed
+
+	// Neutral means that text contains no left-to-right and right-to-left
+	// characters and that no default direction has been set.
+	Neutral
+)
+
+type options struct{}
+
+// An Option is an option for Bidi processing.
+type Option func(*options)
+
+// ICU allows the user to define embedding levels. This may be used, for example,
+// to use hierarchical structure of markup languages to define embeddings.
+// The following option may be a way to expose this functionality in this API.
+// // LevelFunc sets a function that associates nesting levels with the given text.
+// // The levels function will be called with monotonically increasing values for p.
+// func LevelFunc(levels func(p int) int) Option {
+// 	panic("unimplemented")
+// }
+
+// DefaultDirection sets the default direction for a Paragraph. The direction is
+// overridden if the text contains directional characters.
+func DefaultDirection(d Direction) Option {
+	panic("unimplemented")
+}
+
+// A Paragraph holds a single Paragraph for Bidi processing.
+type Paragraph struct {
+	// buffers
+}
+
+// SetBytes configures p for the given paragraph text. It replaces text
+// previously set by SetBytes or SetString. If b contains a paragraph separator
+// it will only process the first paragraph and report the number of bytes
+// consumed from b including this separator. Error may be non-nil if options are
+// given.
+func (p *Paragraph) SetBytes(b []byte, opts ...Option) (n int, err error) {
+	panic("unimplemented")
+}
+
+// SetString configures p for the given paragraph text. It replaces text
+// previously set by SetBytes or SetString. If b contains a paragraph separator
+// it will only process the first paragraph and report the number of bytes
+// consumed from b including this separator. Error may be non-nil if options are
+// given.
+func (p *Paragraph) SetString(s string, opts ...Option) (n int, err error) {
+	panic("unimplemented")
+}
+
+// IsLeftToRight reports whether the principle direction of rendering for this
+// paragraphs is left-to-right. If this returns false, the principle direction
+// of rendering is right-to-left.
+func (p *Paragraph) IsLeftToRight() bool {
+	panic("unimplemented")
+}
+
+// Direction returns the direction of the text of this paragraph.
+//
+// The direction may be LeftToRight, RightToLeft, Mixed, or Neutral.
+func (p *Paragraph) Direction() Direction {
+	panic("unimplemented")
+}
+
+// RunAt reports the Run at the given position of the input text.
+//
+// This method can be used for computing line breaks on paragraphs.
+func (p *Paragraph) RunAt(pos int) Run {
+	panic("unimplemented")
+}
+
+// Order computes the visual ordering of all the runs in a Paragraph.
+func (p *Paragraph) Order() (Ordering, error) {
+	panic("unimplemented")
+}
+
+// Line computes the visual ordering of runs for a single line starting and
+// ending at the given positions in the original text.
+func (p *Paragraph) Line(start, end int) (Ordering, error) {
+	panic("unimplemented")
+}
+
+// An Ordering holds the computed visual order of runs of a Paragraph. Calling
+// SetBytes or SetString on the originating Paragraph invalidates an Ordering.
+// The methods of an Ordering should only be called by one goroutine at a time.
+type Ordering struct{}
+
+// Direction reports the directionality of the runs.
+//
+// The direction may be LeftToRight, RightToLeft, Mixed, or Neutral.
+func (o *Ordering) Direction() Direction {
+	panic("unimplemented")
+}
+
+// NumRuns returns the number of runs.
+func (o *Ordering) NumRuns() int {
+	panic("unimplemented")
+}
+
+// Run returns the ith run within the ordering.
+func (o *Ordering) Run(i int) Run {
+	panic("unimplemented")
+}
+
+// TODO: perhaps with options.
+// // Reorder creates a reader that reads the runes in visual order per character.
+// // Modifiers remain after the runes they modify.
+// func (l *Runs) Reorder() io.Reader {
+// 	panic("unimplemented")
+// }
+
+// A Run is a continuous sequence of characters of a single direction.
+type Run struct {
+}
+
+// String returns the text of the run in its original order.
+func (r *Run) String() string {
+	panic("unimplemented")
+}
+
+// Bytes returns the text of the run in its original order.
+func (r *Run) Bytes() []byte {
+	panic("unimplemented")
+}
+
+// TODO: methods for
+// - Display order
+// - headers and footers
+// - bracket replacement.
+
+// Direction reports the direction of the run.
+func (r *Run) Direction() Direction {
+	panic("unimplemented")
+}
+
+// Position of the Run within the text passed to SetBytes or SetString of the
+// originating Paragraph value.
+func (r *Run) Pos() (start, end int) {
+	panic("unimplemented")
+}
+
+// AppendReverse reverses the order of characters of in, appends them to out,
+// and returns the result. Modifiers will still follow the runes they modify.
+// Brackets are replaced with their counterparts.
+func AppendReverse(out, in []byte) []byte {
+	panic("unimplemented")
+}
+
+// ReverseString reverses the order of characters in s and returns a new string.
+// Modifiers will still follow the runes they modify. Brackets are replaced with
+// their counterparts.
+func ReverseString(s string) string {
+	panic("unimplemented")
+}

vendor/golang.org/x/text/unicode/bidi/bracket.go

@@ -0,0 +1,335 @@
+// Copyright 2015 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 bidi
+
+import (
+	"container/list"
+	"fmt"
+	"sort"
+)
+
+// This file contains a port of the reference implementation of the
+// Bidi Parentheses Algorithm:
+// http://www.unicode.org/Public/PROGRAMS/BidiReferenceJava/BidiPBAReference.java
+//
+// The implementation in this file covers definitions BD14-BD16 and rule N0
+// of UAX#9.
+//
+// Some preprocessing is done for each rune before data is passed to this
+// algorithm:
+//  - opening and closing brackets are identified
+//  - a bracket pair type, like '(' and ')' is assigned a unique identifier that
+//    is identical for the opening and closing bracket. It is left to do these
+//    mappings.
+//  - The BPA algorithm requires that bracket characters that are canonical
+//    equivalents of each other be able to be substituted for each other.
+//    It is the responsibility of the caller to do this canonicalization.
+//
+// In implementing BD16, this implementation departs slightly from the "logical"
+// algorithm defined in UAX#9. In particular, the stack referenced there
+// supports operations that go beyond a "basic" stack. An equivalent
+// implementation based on a linked list is used here.
+
+// Bidi_Paired_Bracket_Type
+// BD14. An opening paired bracket is a character whose
+// Bidi_Paired_Bracket_Type property value is Open.
+//
+// BD15. A closing paired bracket is a character whose
+// Bidi_Paired_Bracket_Type property value is Close.
+type bracketType byte
+
+const (
+	bpNone bracketType = iota
+	bpOpen
+	bpClose
+)
+
+// bracketPair holds a pair of index values for opening and closing bracket
+// location of a bracket pair.
+type bracketPair struct {
+	opener int
+	closer int
+}
+
+func (b *bracketPair) String() string {
+	return fmt.Sprintf("(%v, %v)", b.opener, b.closer)
+}
+
+// bracketPairs is a slice of bracketPairs with a sort.Interface implementation.
+type bracketPairs []bracketPair
+
+func (b bracketPairs) Len() int           { return len(b) }
+func (b bracketPairs) Swap(i, j int)      { b[i], b[j] = b[j], b[i] }
+func (b bracketPairs) Less(i, j int) bool { return b[i].opener < b[j].opener }
+
+// resolvePairedBrackets runs the paired bracket part of the UBA algorithm.
+//
+// For each rune, it takes the indexes into the original string, the class the
+// bracket type (in pairTypes) and the bracket identifier (pairValues). It also
+// takes the direction type for the start-of-sentence and the embedding level.
+//
+// The identifiers for bracket types are the rune of the canonicalized opening
+// bracket for brackets (open or close) or 0 for runes that are not brackets.
+func resolvePairedBrackets(s *isolatingRunSequence) {
+	p := bracketPairer{
+		sos:              s.sos,
+		openers:          list.New(),
+		codesIsolatedRun: s.types,
+		indexes:          s.indexes,
+	}
+	dirEmbed := L
+	if s.level&1 != 0 {
+		dirEmbed = R
+	}
+	p.locateBrackets(s.p.pairTypes, s.p.pairValues)
+	p.resolveBrackets(dirEmbed, s.p.initialTypes)
+}
+
+type bracketPairer struct {
+	sos Class // direction corresponding to start of sequence
+
+	// The following is a restatement of BD 16 using non-algorithmic language.
+	//
+	// A bracket pair is a pair of characters consisting of an opening
+	// paired bracket and a closing paired bracket such that the
+	// Bidi_Paired_Bracket property value of the former equals the latter,
+	// subject to the following constraints.
+	// - both characters of a pair occur in the same isolating run sequence
+	// - the closing character of a pair follows the opening character
+	// - any bracket character can belong at most to one pair, the earliest possible one
+	// - any bracket character not part of a pair is treated like an ordinary character
+	// - pairs may nest properly, but their spans may not overlap otherwise
+
+	// Bracket characters with canonical decompositions are supposed to be
+	// treated as if they had been normalized, to allow normalized and non-
+	// normalized text to give the same result. In this implementation that step
+	// is pushed out to the caller. The caller has to ensure that the pairValue
+	// slices contain the rune of the opening bracket after normalization for
+	// any opening or closing bracket.
+
+	openers *list.List // list of positions for opening brackets
+
+	// bracket pair positions sorted by location of opening bracket
+	pairPositions bracketPairs
+
+	codesIsolatedRun []Class // directional bidi codes for an isolated run
+	indexes          []int   // array of index values into the original string
+
+}
+
+// matchOpener reports whether characters at given positions form a matching
+// bracket pair.
+func (p *bracketPairer) matchOpener(pairValues []rune, opener, closer int) bool {
+	return pairValues[p.indexes[opener]] == pairValues[p.indexes[closer]]
+}
+
+const maxPairingDepth = 63
+
+// locateBrackets locates matching bracket pairs according to BD16.
+//
+// This implementation uses a linked list instead of a stack, because, while
+// elements are added at the front (like a push) they are not generally removed
+// in atomic 'pop' operations, reducing the benefit of the stack archetype.
+func (p *bracketPairer) locateBrackets(pairTypes []bracketType, pairValues []rune) {
+	// traverse the run
+	// do that explicitly (not in a for-each) so we can record position
+	for i, index := range p.indexes {
+
+		// look at the bracket type for each character
+		if pairTypes[index] == bpNone || p.codesIsolatedRun[i] != ON {
+			// continue scanning
+			continue
+		}
+		switch pairTypes[index] {
+		case bpOpen:
+			// check if maximum pairing depth reached
+			if p.openers.Len() == maxPairingDepth {
+				p.openers.Init()
+				return
+			}
+			// remember opener location, most recent first
+			p.openers.PushFront(i)
+
+		case bpClose:
+			// see if there is a match
+			count := 0
+			for elem := p.openers.Front(); elem != nil; elem = elem.Next() {
+				count++
+				opener := elem.Value.(int)
+				if p.matchOpener(pairValues, opener, i) {
+					// if the opener matches, add nested pair to the ordered list
+					p.pairPositions = append(p.pairPositions, bracketPair{opener, i})
+					// remove up to and including matched opener
+					for ; count > 0; count-- {
+						p.openers.Remove(p.openers.Front())
+					}
+					break
+				}
+			}
+			sort.Sort(p.pairPositions)
+			// if we get here, the closing bracket matched no openers
+			// and gets ignored
+		}
+	}
+}
+
+// Bracket pairs within an isolating run sequence are processed as units so
+// that both the opening and the closing paired bracket in a pair resolve to
+// the same direction.
+//
+// N0. Process bracket pairs in an isolating run sequence sequentially in
+// the logical order of the text positions of the opening paired brackets
+// using the logic given below. Within this scope, bidirectional types EN
+// and AN are treated as R.
+//
+// Identify the bracket pairs in the current isolating run sequence
+// according to BD16. For each bracket-pair element in the list of pairs of
+// text positions:
+//
+// a Inspect the bidirectional types of the characters enclosed within the
+// bracket pair.
+//
+// b If any strong type (either L or R) matching the embedding direction is
+// found, set the type for both brackets in the pair to match the embedding
+// direction.
+//
+// o [ e ] o -> o e e e o
+//
+// o [ o e ] -> o e o e e
+//
+// o [ NI e ] -> o e NI e e
+//
+// c Otherwise, if a strong type (opposite the embedding direction) is
+// found, test for adjacent strong types as follows: 1 First, check
+// backwards before the opening paired bracket until the first strong type
+// (L, R, or sos) is found. If that first preceding strong type is opposite
+// the embedding direction, then set the type for both brackets in the pair
+// to that type. 2 Otherwise, set the type for both brackets in the pair to
+// the embedding direction.
+//
+// o [ o ] e -> o o o o e
+//
+// o [ o NI ] o -> o o o NI o o
+//
+// e [ o ] o -> e e o e o
+//
+// e [ o ] e -> e e o e e
+//
+// e ( o [ o ] NI ) e -> e e o o o o NI e e
+//
+// d Otherwise, do not set the type for the current bracket pair. Note that
+// if the enclosed text contains no strong types the paired brackets will
+// both resolve to the same level when resolved individually using rules N1
+// and N2.
+//
+// e ( NI ) o -> e ( NI ) o
+
+// getStrongTypeN0 maps character's directional code to strong type as required
+// by rule N0.
+//
+// TODO: have separate type for "strong" directionality.
+func (p *bracketPairer) getStrongTypeN0(index int) Class {
+	switch p.codesIsolatedRun[index] {
+	// in the scope of N0, number types are treated as R
+	case EN, AN, AL, R:
+		return R
+	case L:
+		return L
+	default:
+		return ON
+	}
+}
+
+// classifyPairContent reports the strong types contained inside a Bracket Pair,
+// assuming the given embedding direction.
+//
+// It returns ON if no strong type is found. If a single strong type is found,
+// it returns this this type. Otherwise it returns the embedding direction.
+//
+// TODO: use separate type for "strong" directionality.
+func (p *bracketPairer) classifyPairContent(loc bracketPair, dirEmbed Class) Class {
+	dirOpposite := ON
+	for i := loc.opener + 1; i < loc.closer; i++ {
+		dir := p.getStrongTypeN0(i)
+		if dir == ON {
+			continue
+		}
+		if dir == dirEmbed {
+			return dir // type matching embedding direction found
+		}
+		dirOpposite = dir
+	}
+	// return ON if no strong type found, or class opposite to dirEmbed
+	return dirOpposite
+}
+
+// classBeforePair determines which strong types are present before a Bracket
+// Pair. Return R or L if strong type found, otherwise ON.
+func (p *bracketPairer) classBeforePair(loc bracketPair) Class {
+	for i := loc.opener - 1; i >= 0; i-- {
+		if dir := p.getStrongTypeN0(i); dir != ON {
+			return dir
+		}
+	}
+	// no strong types found, return sos
+	return p.sos
+}
+
+// assignBracketType implements rule N0 for a single bracket pair.
+func (p *bracketPairer) assignBracketType(loc bracketPair, dirEmbed Class, initialTypes []Class) {
+	// rule "N0, a", inspect contents of pair
+	dirPair := p.classifyPairContent(loc, dirEmbed)
+
+	// dirPair is now L, R, or N (no strong type found)
+
+	// the following logical tests are performed out of order compared to
+	// the statement of the rules but yield the same results
+	if dirPair == ON {
+		return // case "d" - nothing to do
+	}
+
+	if dirPair != dirEmbed {
+		// case "c": strong type found, opposite - check before (c.1)
+		dirPair = p.classBeforePair(loc)
+		if dirPair == dirEmbed || dirPair == ON {
+			// no strong opposite type found before - use embedding (c.2)
+			dirPair = dirEmbed
+		}
+	}
+	// else: case "b", strong type found matching embedding,
+	// no explicit action needed, as dirPair is already set to embedding
+	// direction
+
+	// set the bracket types to the type found
+	p.setBracketsToType(loc, dirPair, initialTypes)
+}
+
+func (p *bracketPairer) setBracketsToType(loc bracketPair, dirPair Class, initialTypes []Class) {
+	p.codesIsolatedRun[loc.opener] = dirPair
+	p.codesIsolatedRun[loc.closer] = dirPair
+
+	for i := loc.opener + 1; i < loc.closer; i++ {
+		index := p.indexes[i]
+		if initialTypes[index] != NSM {
+			break
+		}
+		p.codesIsolatedRun[i] = dirPair
+	}
+
+	for i := loc.closer + 1; i < len(p.indexes); i++ {
+		index := p.indexes[i]
+		if initialTypes[index] != NSM {
+			break
+		}
+		p.codesIsolatedRun[i] = dirPair
+	}
+}
+
+// resolveBrackets implements rule N0 for a list of pairs.
+func (p *bracketPairer) resolveBrackets(dirEmbed Class, initialTypes []Class) {
+	for _, loc := range p.pairPositions {
+		p.assignBracketType(loc, dirEmbed, initialTypes)
+	}
+}

vendor/golang.org/x/text/unicode/bidi/core_test.go

@@ -0,0 +1,224 @@
+// Copyright 2015 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 bidi
+
+import (
+	"flag"
+	"fmt"
+	"log"
+	"strconv"
+	"strings"
+	"testing"
+
+	"golang.org/x/text/internal/gen"
+	"golang.org/x/text/internal/testtext"
+	"golang.org/x/text/internal/ucd"
+	"golang.org/x/text/unicode/norm"
+)
+
+var testLevels = flag.Bool("levels", false, "enable testing of levels")
+
+// TestBidiCore performs the tests in BidiTest.txt.
+// See http://www.unicode.org/Public/UCD/latest/ucd/BidiTest.txt.
+func TestBidiCore(t *testing.T) {
+	testtext.SkipIfNotLong(t)
+
+	r := gen.OpenUCDFile("BidiTest.txt")
+	defer r.Close()
+
+	var wantLevels, wantOrder []string
+	p := ucd.New(r, ucd.Part(func(p *ucd.Parser) {
+		s := strings.Split(p.String(0), ":")
+		switch s[0] {
+		case "Levels":
+			wantLevels = strings.Fields(s[1])
+		case "Reorder":
+			wantOrder = strings.Fields(s[1])
+		default:
+			log.Fatalf("Unknown part %q.", s[0])
+		}
+	}))
+
+	for p.Next() {
+		types := []Class{}
+		for _, s := range p.Strings(0) {
+			types = append(types, bidiClass[s])
+		}
+		// We ignore the bracketing part of the algorithm.
+		pairTypes := make([]bracketType, len(types))
+		pairValues := make([]rune, len(types))
+
+		for i := uint(0); i < 3; i++ {
+			if p.Uint(1)&(1<<i) == 0 {
+				continue
+			}
+			lev := level(int(i) - 1)
+			par := newParagraph(types, pairTypes, pairValues, lev)
+
+			if *testLevels {
+				levels := par.getLevels([]int{len(types)})
+				for i, s := range wantLevels {
+					if s == "x" {
+						continue
+					}
+					l, _ := strconv.ParseUint(s, 10, 8)
+					if level(l)&1 != levels[i]&1 {
+						t.Errorf("%s:%d:levels: got %v; want %v", p.String(0), lev, levels, wantLevels)
+						break
+					}
+				}
+			}
+
+			order := par.getReordering([]int{len(types)})
+			gotOrder := filterOrder(types, order)
+			if got, want := fmt.Sprint(gotOrder), fmt.Sprint(wantOrder); got != want {
+				t.Errorf("%s:%d:order: got %v; want %v\noriginal %v", p.String(0), lev, got, want, order)
+			}
+		}
+	}
+	if err := p.Err(); err != nil {
+		log.Fatal(err)
+	}
+}
+
+var removeClasses = map[Class]bool{
+	LRO: true,
+	RLO: true,
+	RLE: true,
+	LRE: true,
+	PDF: true,
+	BN:  true,
+}
+
+// TestBidiCharacters performs the tests in BidiCharacterTest.txt.
+// See http://www.unicode.org/Public/UCD/latest/ucd/BidiCharacterTest.txt
+func TestBidiCharacters(t *testing.T) {
+	testtext.SkipIfNotLong(t)
+
+	ucd.Parse(gen.OpenUCDFile("BidiCharacterTest.txt"), func(p *ucd.Parser) {
+		var (
+			types      []Class
+			pairTypes  []bracketType
+			pairValues []rune
+			parLevel   level
+
+			wantLevel       = level(p.Int(2))
+			wantLevels      = p.Strings(3)
+			wantVisualOrder = p.Strings(4)
+		)
+
+		switch l := p.Int(1); l {
+		case 0, 1:
+			parLevel = level(l)
+		case 2:
+			parLevel = implicitLevel
+		default:
+			// Spec says to ignore unknown parts.
+		}
+
+		runes := p.Runes(0)
+
+		for _, r := range runes {
+			// Assign the bracket type.
+			if d := norm.NFKD.PropertiesString(string(r)).Decomposition(); d != nil {
+				r = []rune(string(d))[0]
+			}
+			p, _ := LookupRune(r)
+
+			// Assign the class for this rune.
+			types = append(types, p.Class())
+
+			switch {
+			case !p.IsBracket():
+				pairTypes = append(pairTypes, bpNone)
+				pairValues = append(pairValues, 0)
+			case p.IsOpeningBracket():
+				pairTypes = append(pairTypes, bpOpen)
+				pairValues = append(pairValues, r)
+			default:
+				pairTypes = append(pairTypes, bpClose)
+				pairValues = append(pairValues, p.reverseBracket(r))
+			}
+		}
+		par := newParagraph(types, pairTypes, pairValues, parLevel)
+
+		// Test results:
+		if got := par.embeddingLevel; got != wantLevel {
+			t.Errorf("%v:level: got %d; want %d", string(runes), got, wantLevel)
+		}
+
+		if *testLevels {
+			gotLevels := getLevelStrings(types, par.getLevels([]int{len(types)}))
+			if got, want := fmt.Sprint(gotLevels), fmt.Sprint(wantLevels); got != want {
+				t.Errorf("%04X %q:%d: got %v; want %v\nval: %x\npair: %v", runes, string(runes), parLevel, got, want, pairValues, pairTypes)
+			}
+		}
+
+		order := par.getReordering([]int{len(types)})
+		order = filterOrder(types, order)
+		if got, want := fmt.Sprint(order), fmt.Sprint(wantVisualOrder); got != want {
+			t.Errorf("%04X %q:%d: got %v; want %v\ngot order: %s", runes, string(runes), parLevel, got, want, reorder(runes, order))
+		}
+	})
+}
+
+func getLevelStrings(cl []Class, levels []level) []string {
+	var results []string
+	for i, l := range levels {
+		if !removeClasses[cl[i]] {
+			results = append(results, fmt.Sprint(l))
+		} else {
+			results = append(results, "x")
+		}
+	}
+	return results
+}
+
+func filterOrder(cl []Class, order []int) []int {
+	no := []int{}
+	for _, o := range order {
+		if !removeClasses[cl[o]] {
+			no = append(no, o)
+		}
+	}
+	return no
+}
+
+func reorder(r []rune, order []int) string {
+	nr := make([]rune, len(order))
+	for i, o := range order {
+		nr[i] = r[o]
+	}
+	return string(nr)
+}
+
+// bidiClass names and codes taken from class "bc" in
+// http://www.unicode.org/Public/8.0.0/ucd/PropertyValueAliases.txt
+var bidiClass = map[string]Class{
+	"AL":  AL,  // classArabicLetter,
+	"AN":  AN,  // classArabicNumber,
+	"B":   B,   // classParagraphSeparator,
+	"BN":  BN,  // classBoundaryNeutral,
+	"CS":  CS,  // classCommonSeparator,
+	"EN":  EN,  // classEuropeanNumber,
+	"ES":  ES,  // classEuropeanSeparator,
+	"ET":  ET,  // classEuropeanTerminator,
+	"L":   L,   // classLeftToRight,
+	"NSM": NSM, // classNonspacingMark,
+	"ON":  ON,  // classOtherNeutral,
+	"R":   R,   // classRightToLeft,
+	"S":   S,   // classSegmentSeparator,
+	"WS":  WS,  // classWhiteSpace,
+
+	"LRO": LRO, // classLeftToRightOverride,
+	"RLO": RLO, // classRightToLeftOverride,
+	"LRE": LRE, // classLeftToRightEmbedding,
+	"RLE": RLE, // classRightToLeftEmbedding,
+	"PDF": PDF, // classPopDirectionalFormat,
+	"LRI": LRI, // classLeftToRightIsolate,
+	"RLI": RLI, // classRightToLeftIsolate,
+	"FSI": FSI, // classFirstStrongIsolate,
+	"PDI": PDI, // classPopDirectionalIsolate,
+}

vendor/golang.org/x/text/unicode/bidi/gen.go

@@ -0,0 +1,133 @@
+// Copyright 2015 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.
+
+// +build ignore
+
+package main
+
+import (
+	"flag"
+	"log"
+
+	"golang.org/x/text/internal/gen"
+	"golang.org/x/text/internal/triegen"
+	"golang.org/x/text/internal/ucd"
+)
+
+var outputFile = flag.String("out", "tables.go", "output file")
+
+func main() {
+	gen.Init()
+	gen.Repackage("gen_trieval.go", "trieval.go", "bidi")
+	gen.Repackage("gen_ranges.go", "ranges_test.go", "bidi")
+
+	genTables()
+}
+
+// bidiClass names and codes taken from class "bc" in
+// http://www.unicode.org/Public/8.0.0/ucd/PropertyValueAliases.txt
+var bidiClass = map[string]Class{
+	"AL":  AL,  // ArabicLetter
+	"AN":  AN,  // ArabicNumber
+	"B":   B,   // ParagraphSeparator
+	"BN":  BN,  // BoundaryNeutral
+	"CS":  CS,  // CommonSeparator
+	"EN":  EN,  // EuropeanNumber
+	"ES":  ES,  // EuropeanSeparator
+	"ET":  ET,  // EuropeanTerminator
+	"L":   L,   // LeftToRight
+	"NSM": NSM, // NonspacingMark
+	"ON":  ON,  // OtherNeutral
+	"R":   R,   // RightToLeft
+	"S":   S,   // SegmentSeparator
+	"WS":  WS,  // WhiteSpace
+
+	"FSI": Control,
+	"PDF": Control,
+	"PDI": Control,
+	"LRE": Control,
+	"LRI": Control,
+	"LRO": Control,
+	"RLE": Control,
+	"RLI": Control,
+	"RLO": Control,
+}
+
+func genTables() {
+	if numClass > 0x0F {
+		log.Fatalf("Too many Class constants (%#x > 0x0F).", numClass)
+	}
+	w := gen.NewCodeWriter()
+	defer w.WriteVersionedGoFile(*outputFile, "bidi")
+
+	gen.WriteUnicodeVersion(w)
+
+	t := triegen.NewTrie("bidi")
+
+	// Build data about bracket mapping. These bits need to be or-ed with
+	// any other bits.
+	orMask := map[rune]uint64{}
+
+	xorMap := map[rune]int{}
+	xorMasks := []rune{0} // First value is no-op.
+
+	ucd.Parse(gen.OpenUCDFile("BidiBrackets.txt"), func(p *ucd.Parser) {
+		r1 := p.Rune(0)
+		r2 := p.Rune(1)
+		xor := r1 ^ r2
+		if _, ok := xorMap[xor]; !ok {
+			xorMap[xor] = len(xorMasks)
+			xorMasks = append(xorMasks, xor)
+		}
+		entry := uint64(xorMap[xor]) << xorMaskShift
+		switch p.String(2) {
+		case "o":
+			entry |= openMask
+		case "c", "n":
+		default:
+			log.Fatalf("Unknown bracket class %q.", p.String(2))
+		}
+		orMask[r1] = entry
+	})
+
+	w.WriteComment(`
+	xorMasks contains masks to be xor-ed with brackets to get the reverse
+	version.`)
+	w.WriteVar("xorMasks", xorMasks)
+
+	done := map[rune]bool{}
+
+	insert := func(r rune, c Class) {
+		if !done[r] {
+			t.Insert(r, orMask[r]|uint64(c))
+			done[r] = true
+		}
+	}
+
+	// Insert the derived BiDi properties.
+	ucd.Parse(gen.OpenUCDFile("extracted/DerivedBidiClass.txt"), func(p *ucd.Parser) {
+		r := p.Rune(0)
+		class, ok := bidiClass[p.String(1)]
+		if !ok {
+			log.Fatalf("%U: Unknown BiDi class %q", r, p.String(1))
+		}
+		insert(r, class)
+	})
+	visitDefaults(insert)
+
+	// TODO: use sparse blocks. This would reduce table size considerably
+	// from the looks of it.
+
+	sz, err := t.Gen(w)
+	if err != nil {
+		log.Fatal(err)
+	}
+	w.Size += sz
+}
+
+// dummy values to make methods in gen_common compile. The real versions
+// will be generated by this file to tables.go.
+var (
+	xorMasks []rune
+)

vendor/golang.org/x/text/unicode/bidi/gen_ranges.go

@@ -0,0 +1,57 @@
+// Copyright 2015 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.
+
+// +build ignore
+
+package main
+
+import (
+	"unicode"
+
+	"golang.org/x/text/internal/gen"
+	"golang.org/x/text/internal/ucd"
+	"golang.org/x/text/unicode/rangetable"
+)
+
+// These tables are hand-extracted from:
+// http://www.unicode.org/Public/8.0.0/ucd/extracted/DerivedBidiClass.txt
+func visitDefaults(fn func(r rune, c Class)) {
+	// first write default values for ranges listed above.
+	visitRunes(fn, AL, []rune{
+		0x0600, 0x07BF, // Arabic
+		0x08A0, 0x08FF, // Arabic Extended-A
+		0xFB50, 0xFDCF, // Arabic Presentation Forms
+		0xFDF0, 0xFDFF,
+		0xFE70, 0xFEFF,
+		0x0001EE00, 0x0001EEFF, // Arabic Mathematical Alpha Symbols
+	})
+	visitRunes(fn, R, []rune{
+		0x0590, 0x05FF, // Hebrew
+		0x07C0, 0x089F, // Nko et al.
+		0xFB1D, 0xFB4F,
+		0x00010800, 0x00010FFF, // Cypriot Syllabary et. al.
+		0x0001E800, 0x0001EDFF,
+		0x0001EF00, 0x0001EFFF,
+	})
+	visitRunes(fn, ET, []rune{ // European Terminator
+		0x20A0, 0x20Cf, // Currency symbols
+	})
+	rangetable.Visit(unicode.Noncharacter_Code_Point, func(r rune) {
+		fn(r, BN) // Boundary Neutral
+	})
+	ucd.Parse(gen.OpenUCDFile("DerivedCoreProperties.txt"), func(p *ucd.Parser) {
+		if p.String(1) == "Default_Ignorable_Code_Point" {
+			fn(p.Rune(0), BN) // Boundary Neutral
+		}
+	})
+}
+
+func visitRunes(fn func(r rune, c Class), c Class, runes []rune) {
+	for i := 0; i < len(runes); i += 2 {
+		lo, hi := runes[i], runes[i+1]
+		for j := lo; j <= hi; j++ {
+			fn(j, c)
+		}
+	}
+}

vendor/golang.org/x/text/unicode/bidi/gen_trieval.go

@@ -0,0 +1,64 @@
+// Copyright 2015 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.
+
+// +build ignore
+
+package main
+
+// Class is the Unicode BiDi class. Each rune has a single class.
+type Class uint
+
+const (
+	L       Class = iota // LeftToRight
+	R                    // RightToLeft
+	EN                   // EuropeanNumber
+	ES                   // EuropeanSeparator
+	ET                   // EuropeanTerminator
+	AN                   // ArabicNumber
+	CS                   // CommonSeparator
+	B                    // ParagraphSeparator
+	S                    // SegmentSeparator
+	WS                   // WhiteSpace
+	ON                   // OtherNeutral
+	BN                   // BoundaryNeutral
+	NSM                  // NonspacingMark
+	AL                   // ArabicLetter
+	Control              // Control LRO - PDI
+
+	numClass
+
+	LRO // LeftToRightOverride
+	RLO // RightToLeftOverride
+	LRE // LeftToRightEmbedding
+	RLE // RightToLeftEmbedding
+	PDF // PopDirectionalFormat
+	LRI // LeftToRightIsolate
+	RLI // RightToLeftIsolate
+	FSI // FirstStrongIsolate
+	PDI // PopDirectionalIsolate
+
+	unknownClass = ^Class(0)
+)
+
+var controlToClass = map[rune]Class{
+	0x202D: LRO, // LeftToRightOverride,
+	0x202E: RLO, // RightToLeftOverride,
+	0x202A: LRE, // LeftToRightEmbedding,
+	0x202B: RLE, // RightToLeftEmbedding,
+	0x202C: PDF, // PopDirectionalFormat,
+	0x2066: LRI, // LeftToRightIsolate,
+	0x2067: RLI, // RightToLeftIsolate,
+	0x2068: FSI, // FirstStrongIsolate,
+	0x2069: PDI, // PopDirectionalIsolate,
+}
+
+// A trie entry has the following bits:
+// 7..5  XOR mask for brackets
+// 4     1: Bracket open, 0: Bracket close
+// 3..0  Class type
+
+const (
+	openMask     = 0x10
+	xorMaskShift = 5
+)

vendor/golang.org/x/text/unicode/bidi/prop.go

@@ -0,0 +1,206 @@
+// 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 bidi
+
+import "unicode/utf8"
+
+// Properties provides access to BiDi properties of runes.
+type Properties struct {
+	entry uint8
+	last  uint8
+}
+
+var trie = newBidiTrie(0)
+
+// TODO: using this for bidirule reduces the running time by about 5%. Consider
+// if this is worth exposing or if we can find a way to speed up the Class
+// method.
+//
+// // CompactClass is like Class, but maps all of the BiDi control classes
+// // (LRO, RLO, LRE, RLE, PDF, LRI, RLI, FSI, PDI) to the class Control.
+// func (p Properties) CompactClass() Class {
+// 	return Class(p.entry & 0x0F)
+// }
+
+// Class returns the Bidi class for p.
+func (p Properties) Class() Class {
+	c := Class(p.entry & 0x0F)
+	if c == Control {
+		c = controlByteToClass[p.last&0xF]
+	}
+	return c
+}
+
+// IsBracket reports whether the rune is a bracket.
+func (p Properties) IsBracket() bool { return p.entry&0xF0 != 0 }
+
+// IsOpeningBracket reports whether the rune is an opening bracket.
+// IsBracket must return true.
+func (p Properties) IsOpeningBracket() bool { return p.entry&openMask != 0 }
+
+// TODO: find a better API and expose.
+func (p Properties) reverseBracket(r rune) rune {
+	return xorMasks[p.entry>>xorMaskShift] ^ r
+}
+
+var controlByteToClass = [16]Class{
+	0xD: LRO, // U+202D LeftToRightOverride,
+	0xE: RLO, // U+202E RightToLeftOverride,
+	0xA: LRE, // U+202A LeftToRightEmbedding,
+	0xB: RLE, // U+202B RightToLeftEmbedding,
+	0xC: PDF, // U+202C PopDirectionalFormat,
+	0x6: LRI, // U+2066 LeftToRightIsolate,
+	0x7: RLI, // U+2067 RightToLeftIsolate,
+	0x8: FSI, // U+2068 FirstStrongIsolate,
+	0x9: PDI, // U+2069 PopDirectionalIsolate,
+}
+
+// LookupRune returns properties for r.
+func LookupRune(r rune) (p Properties, size int) {
+	var buf [4]byte
+	n := utf8.EncodeRune(buf[:], r)
+	return Lookup(buf[:n])
+}
+
+// TODO: these lookup methods are based on the generated trie code. The returned
+// sizes have slightly different semantics from the generated code, in that it
+// always returns size==1 for an illegal UTF-8 byte (instead of the length
+// of the maximum invalid subsequence). Most Transformers, like unicode/norm,
+// leave invalid UTF-8 untouched, in which case it has performance benefits to
+// do so (without changing the semantics). Bidi requires the semantics used here
+// for the bidirule implementation to be compatible with the Go semantics.
+//  They ultimately should perhaps be adopted by all trie implementations, for
+// convenience sake.
+// This unrolled code also boosts performance of the secure/bidirule package by
+// about 30%.
+// So, to remove this code:
+//   - add option to trie generator to define return type.
+//   - always return 1 byte size for ill-formed UTF-8 runes.
+
+// Lookup returns properties for the first rune in s and the width in bytes of
+// its encoding. The size will be 0 if s does not hold enough bytes to complete
+// the encoding.
+func Lookup(s []byte) (p Properties, sz int) {
+	c0 := s[0]
+	switch {
+	case c0 < 0x80: // is ASCII
+		return Properties{entry: bidiValues[c0]}, 1
+	case c0 < 0xC2:
+		return Properties{}, 1
+	case c0 < 0xE0: // 2-byte UTF-8
+		if len(s) < 2 {
+			return Properties{}, 0
+		}
+		i := bidiIndex[c0]
+		c1 := s[1]
+		if c1 < 0x80 || 0xC0 <= c1 {
+			return Properties{}, 1
+		}
+		return Properties{entry: trie.lookupValue(uint32(i), c1)}, 2
+	case c0 < 0xF0: // 3-byte UTF-8
+		if len(s) < 3 {
+			return Properties{}, 0
+		}
+		i := bidiIndex[c0]
+		c1 := s[1]
+		if c1 < 0x80 || 0xC0 <= c1 {
+			return Properties{}, 1
+		}
+		o := uint32(i)<<6 + uint32(c1)
+		i = bidiIndex[o]
+		c2 := s[2]
+		if c2 < 0x80 || 0xC0 <= c2 {
+			return Properties{}, 1
+		}
+		return Properties{entry: trie.lookupValue(uint32(i), c2), last: c2}, 3
+	case c0 < 0xF8: // 4-byte UTF-8
+		if len(s) < 4 {
+			return Properties{}, 0
+		}
+		i := bidiIndex[c0]
+		c1 := s[1]
+		if c1 < 0x80 || 0xC0 <= c1 {
+			return Properties{}, 1
+		}
+		o := uint32(i)<<6 + uint32(c1)
+		i = bidiIndex[o]
+		c2 := s[2]
+		if c2 < 0x80 || 0xC0 <= c2 {
+			return Properties{}, 1
+		}
+		o = uint32(i)<<6 + uint32(c2)
+		i = bidiIndex[o]
+		c3 := s[3]
+		if c3 < 0x80 || 0xC0 <= c3 {
+			return Properties{}, 1
+		}
+		return Properties{entry: trie.lookupValue(uint32(i), c3)}, 4
+	}
+	// Illegal rune
+	return Properties{}, 1
+}
+
+// LookupString returns properties for the first rune in s and the width in
+// bytes of its encoding. The size will be 0 if s does not hold enough bytes to
+// complete the encoding.
+func LookupString(s string) (p Properties, sz int) {
+	c0 := s[0]
+	switch {
+	case c0 < 0x80: // is ASCII
+		return Properties{entry: bidiValues[c0]}, 1
+	case c0 < 0xC2:
+		return Properties{}, 1
+	case c0 < 0xE0: // 2-byte UTF-8
+		if len(s) < 2 {
+			return Properties{}, 0
+		}
+		i := bidiIndex[c0]
+		c1 := s[1]
+		if c1 < 0x80 || 0xC0 <= c1 {
+			return Properties{}, 1
+		}
+		return Properties{entry: trie.lookupValue(uint32(i), c1)}, 2
+	case c0 < 0xF0: // 3-byte UTF-8
+		if len(s) < 3 {
+			return Properties{}, 0
+		}
+		i := bidiIndex[c0]
+		c1 := s[1]
+		if c1 < 0x80 || 0xC0 <= c1 {
+			return Properties{}, 1
+		}
+		o := uint32(i)<<6 + uint32(c1)
+		i = bidiIndex[o]
+		c2 := s[2]
+		if c2 < 0x80 || 0xC0 <= c2 {
+			return Properties{}, 1
+		}
+		return Properties{entry: trie.lookupValue(uint32(i), c2), last: c2}, 3
+	case c0 < 0xF8: // 4-byte UTF-8
+		if len(s) < 4 {
+			return Properties{}, 0
+		}
+		i := bidiIndex[c0]
+		c1 := s[1]
+		if c1 < 0x80 || 0xC0 <= c1 {
+			return Properties{}, 1
+		}
+		o := uint32(i)<<6 + uint32(c1)
+		i = bidiIndex[o]
+		c2 := s[2]
+		if c2 < 0x80 || 0xC0 <= c2 {
+			return Properties{}, 1
+		}
+		o = uint32(i)<<6 + uint32(c2)
+		i = bidiIndex[o]
+		c3 := s[3]
+		if c3 < 0x80 || 0xC0 <= c3 {
+			return Properties{}, 1
+		}
+		return Properties{entry: trie.lookupValue(uint32(i), c3)}, 4
+	}
+	// Illegal rune
+	return Properties{}, 1
+}

vendor/golang.org/x/text/unicode/bidi/ranges_test.go

@@ -0,0 +1,53 @@
+// Code generated by running "go generate" in golang.org/x/text. DO NOT EDIT.
+
+package bidi
+
+import (
+	"unicode"
+
+	"golang.org/x/text/internal/gen"
+	"golang.org/x/text/internal/ucd"
+	"golang.org/x/text/unicode/rangetable"
+)
+
+// These tables are hand-extracted from:
+// http://www.unicode.org/Public/8.0.0/ucd/extracted/DerivedBidiClass.txt
+func visitDefaults(fn func(r rune, c Class)) {
+	// first write default values for ranges listed above.
+	visitRunes(fn, AL, []rune{
+		0x0600, 0x07BF, // Arabic
+		0x08A0, 0x08FF, // Arabic Extended-A
+		0xFB50, 0xFDCF, // Arabic Presentation Forms
+		0xFDF0, 0xFDFF,
+		0xFE70, 0xFEFF,
+		0x0001EE00, 0x0001EEFF, // Arabic Mathematical Alpha Symbols
+	})
+	visitRunes(fn, R, []rune{
+		0x0590, 0x05FF, // Hebrew
+		0x07C0, 0x089F, // Nko et al.
+		0xFB1D, 0xFB4F,
+		0x00010800, 0x00010FFF, // Cypriot Syllabary et. al.
+		0x0001E800, 0x0001EDFF,
+		0x0001EF00, 0x0001EFFF,
+	})
+	visitRunes(fn, ET, []rune{ // European Terminator
+		0x20A0, 0x20Cf, // Currency symbols
+	})
+	rangetable.Visit(unicode.Noncharacter_Code_Point, func(r rune) {
+		fn(r, BN) // Boundary Neutral
+	})
+	ucd.Parse(gen.OpenUCDFile("DerivedCoreProperties.txt"), func(p *ucd.Parser) {
+		if p.String(1) == "Default_Ignorable_Code_Point" {
+			fn(p.Rune(0), BN) // Boundary Neutral
+		}
+	})
+}
+
+func visitRunes(fn func(r rune, c Class), c Class, runes []rune) {
+	for i := 0; i < len(runes); i += 2 {
+		lo, hi := runes[i], runes[i+1]
+		for j := lo; j <= hi; j++ {
+			fn(j, c)
+		}
+	}
+}

vendor/golang.org/x/text/unicode/bidi/tables_test.go

@@ -0,0 +1,82 @@
+// Copyright 2015 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 bidi
+
+import (
+	"testing"
+
+	"golang.org/x/text/internal/gen"
+	"golang.org/x/text/internal/testtext"
+	"golang.org/x/text/internal/ucd"
+)
+
+var labels = []string{
+	AL:  "AL",
+	AN:  "AN",
+	B:   "B",
+	BN:  "BN",
+	CS:  "CS",
+	EN:  "EN",
+	ES:  "ES",
+	ET:  "ET",
+	L:   "L",
+	NSM: "NSM",
+	ON:  "ON",
+	R:   "R",
+	S:   "S",
+	WS:  "WS",
+
+	LRO: "LRO",
+	RLO: "RLO",
+	LRE: "LRE",
+	RLE: "RLE",
+	PDF: "PDF",
+	LRI: "LRI",
+	RLI: "RLI",
+	FSI: "FSI",
+	PDI: "PDI",
+}
+
+func TestTables(t *testing.T) {
+	testtext.SkipIfNotLong(t)
+
+	ucd.Parse(gen.OpenUCDFile("BidiBrackets.txt"), func(p *ucd.Parser) {
+		r1 := p.Rune(0)
+		want := p.Rune(1)
+
+		e, _ := LookupRune(r1)
+		if got := e.reverseBracket(r1); got != want {
+			t.Errorf("Reverse(%U) = %U; want %U", r1, got, want)
+		}
+	})
+
+	done := map[rune]bool{}
+	test := func(name string, r rune, want string) {
+		str := string(r)
+		e, _ := LookupString(str)
+		if got := labels[e.Class()]; got != want {
+			t.Errorf("%s:%U: got %s; want %s", name, r, got, want)
+		}
+		if e2, sz := LookupRune(r); e != e2 || sz != len(str) {
+			t.Errorf("LookupRune(%U) = %v, %d; want %v, %d", r, e2, e, sz, len(str))
+		}
+		if e2, sz := Lookup([]byte(str)); e != e2 || sz != len(str) {
+			t.Errorf("Lookup(%U) = %v, %d; want %v, %d", r, e2, e, sz, len(str))
+		}
+		done[r] = true
+	}
+
+	// Insert the derived BiDi properties.
+	ucd.Parse(gen.OpenUCDFile("extracted/DerivedBidiClass.txt"), func(p *ucd.Parser) {
+		r := p.Rune(0)
+		test("derived", r, p.String(1))
+	})
+	visitDefaults(func(r rune, c Class) {
+		if !done[r] {
+			test("default", r, labels[c])
+		}
+	})
+
+}

vendor/golang.org/x/text/unicode/bidi/trieval.go

@@ -0,0 +1,60 @@
+// Code generated by running "go generate" in golang.org/x/text. DO NOT EDIT.
+
+package bidi
+
+// Class is the Unicode BiDi class. Each rune has a single class.
+type Class uint
+
+const (
+	L       Class = iota // LeftToRight
+	R                    // RightToLeft
+	EN                   // EuropeanNumber
+	ES                   // EuropeanSeparator
+	ET                   // EuropeanTerminator
+	AN                   // ArabicNumber
+	CS                   // CommonSeparator
+	B                    // ParagraphSeparator
+	S                    // SegmentSeparator
+	WS                   // WhiteSpace
+	ON                   // OtherNeutral
+	BN                   // BoundaryNeutral
+	NSM                  // NonspacingMark
+	AL                   // ArabicLetter
+	Control              // Control LRO - PDI
+
+	numClass
+
+	LRO // LeftToRightOverride
+	RLO // RightToLeftOverride
+	LRE // LeftToRightEmbedding
+	RLE // RightToLeftEmbedding
+	PDF // PopDirectionalFormat
+	LRI // LeftToRightIsolate
+	RLI // RightToLeftIsolate
+	FSI // FirstStrongIsolate
+	PDI // PopDirectionalIsolate
+
+	unknownClass = ^Class(0)
+)
+
+var controlToClass = map[rune]Class{
+	0x202D: LRO, // LeftToRightOverride,
+	0x202E: RLO, // RightToLeftOverride,
+	0x202A: LRE, // LeftToRightEmbedding,
+	0x202B: RLE, // RightToLeftEmbedding,
+	0x202C: PDF, // PopDirectionalFormat,
+	0x2066: LRI, // LeftToRightIsolate,
+	0x2067: RLI, // RightToLeftIsolate,
+	0x2068: FSI, // FirstStrongIsolate,
+	0x2069: PDI, // PopDirectionalIsolate,
+}
+
+// A trie entry has the following bits:
+// 7..5  XOR mask for brackets
+// 4     1: Bracket open, 0: Bracket close
+// 3..0  Class type
+
+const (
+	openMask     = 0x10
+	xorMaskShift = 5
+)