diff options
author | Thomas Voss <mail@thomasvoss.com> | 2024-08-10 21:32:27 +0200 |
---|---|---|
committer | Thomas Voss <mail@thomasvoss.com> | 2024-08-10 21:32:27 +0200 |
commit | 2c2f850a29b2e896f640679ed9fcec523ccfdb2d (patch) | |
tree | 412662e9494d00135f0ac3bae5816f7aca53767f /vendor/golang.org/x/text/internal | |
parent | e1c08fce740b54454c4fea09634bf839ef9f76d3 (diff) |
Support sorting countries by name
Diffstat (limited to 'vendor/golang.org/x/text/internal')
-rw-r--r-- | vendor/golang.org/x/text/internal/colltab/collelem.go | 376 | ||||
-rw-r--r-- | vendor/golang.org/x/text/internal/colltab/colltab.go | 105 | ||||
-rw-r--r-- | vendor/golang.org/x/text/internal/colltab/contract.go | 145 | ||||
-rw-r--r-- | vendor/golang.org/x/text/internal/colltab/iter.go | 178 | ||||
-rw-r--r-- | vendor/golang.org/x/text/internal/colltab/numeric.go | 236 | ||||
-rw-r--r-- | vendor/golang.org/x/text/internal/colltab/table.go | 275 | ||||
-rw-r--r-- | vendor/golang.org/x/text/internal/colltab/trie.go | 159 | ||||
-rw-r--r-- | vendor/golang.org/x/text/internal/colltab/weighter.go | 31 |
8 files changed, 1505 insertions, 0 deletions
diff --git a/vendor/golang.org/x/text/internal/colltab/collelem.go b/vendor/golang.org/x/text/internal/colltab/collelem.go new file mode 100644 index 0000000..0c23c8a --- /dev/null +++ b/vendor/golang.org/x/text/internal/colltab/collelem.go @@ -0,0 +1,376 @@ +// Copyright 2012 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 colltab + +import ( + "fmt" + "unicode" +) + +// Level identifies the collation comparison level. +// The primary level corresponds to the basic sorting of text. +// The secondary level corresponds to accents and related linguistic elements. +// The tertiary level corresponds to casing and related concepts. +// The quaternary level is derived from the other levels by the +// various algorithms for handling variable elements. +type Level int + +const ( + Primary Level = iota + Secondary + Tertiary + Quaternary + Identity + + NumLevels +) + +const ( + defaultSecondary = 0x20 + defaultTertiary = 0x2 + maxTertiary = 0x1F + MaxQuaternary = 0x1FFFFF // 21 bits. +) + +// Elem is a representation of a collation element. This API provides ways to encode +// and decode Elems. Implementations of collation tables may use values greater +// or equal to PrivateUse for their own purposes. However, these should never be +// returned by AppendNext. +type Elem uint32 + +const ( + maxCE Elem = 0xAFFFFFFF + PrivateUse = minContract + minContract = 0xC0000000 + maxContract = 0xDFFFFFFF + minExpand = 0xE0000000 + maxExpand = 0xEFFFFFFF + minDecomp = 0xF0000000 +) + +type ceType int + +const ( + ceNormal ceType = iota // ceNormal includes implicits (ce == 0) + ceContractionIndex // rune can be a start of a contraction + ceExpansionIndex // rune expands into a sequence of collation elements + ceDecompose // rune expands using NFKC decomposition +) + +func (ce Elem) ctype() ceType { + if ce <= maxCE { + return ceNormal + } + if ce <= maxContract { + return ceContractionIndex + } else { + if ce <= maxExpand { + return ceExpansionIndex + } + return ceDecompose + } + panic("should not reach here") + return ceType(-1) +} + +// For normal collation elements, we assume that a collation element either has +// a primary or non-default secondary value, not both. +// Collation elements with a primary value are of the form +// +// 01pppppp pppppppp ppppppp0 ssssssss +// - p* is primary collation value +// - s* is the secondary collation value +// 00pppppp pppppppp ppppppps sssttttt, where +// - p* is primary collation value +// - s* offset of secondary from default value. +// - t* is the tertiary collation value +// 100ttttt cccccccc pppppppp pppppppp +// - t* is the tertiar collation value +// - c* is the canonical combining class +// - p* is the primary collation value +// +// Collation elements with a secondary value are of the form +// +// 1010cccc ccccssss ssssssss tttttttt, where +// - c* is the canonical combining class +// - s* is the secondary collation value +// - t* is the tertiary collation value +// 11qqqqqq qqqqqqqq qqqqqqq0 00000000 +// - q* quaternary value +const ( + ceTypeMask = 0xC0000000 + ceTypeMaskExt = 0xE0000000 + ceIgnoreMask = 0xF00FFFFF + ceType1 = 0x40000000 + ceType2 = 0x00000000 + ceType3or4 = 0x80000000 + ceType4 = 0xA0000000 + ceTypeQ = 0xC0000000 + Ignore = ceType4 + firstNonPrimary = 0x80000000 + lastSpecialPrimary = 0xA0000000 + secondaryMask = 0x80000000 + hasTertiaryMask = 0x40000000 + primaryValueMask = 0x3FFFFE00 + maxPrimaryBits = 21 + compactPrimaryBits = 16 + maxSecondaryBits = 12 + maxTertiaryBits = 8 + maxCCCBits = 8 + maxSecondaryCompactBits = 8 + maxSecondaryDiffBits = 4 + maxTertiaryCompactBits = 5 + primaryShift = 9 + compactSecondaryShift = 5 + minCompactSecondary = defaultSecondary - 4 +) + +func makeImplicitCE(primary int) Elem { + return ceType1 | Elem(primary<<primaryShift) | defaultSecondary +} + +// MakeElem returns an Elem for the given values. It will return an error +// if the given combination of values is invalid. +func MakeElem(primary, secondary, tertiary int, ccc uint8) (Elem, error) { + if w := primary; w >= 1<<maxPrimaryBits || w < 0 { + return 0, fmt.Errorf("makeCE: primary weight out of bounds: %x >= %x", w, 1<<maxPrimaryBits) + } + if w := secondary; w >= 1<<maxSecondaryBits || w < 0 { + return 0, fmt.Errorf("makeCE: secondary weight out of bounds: %x >= %x", w, 1<<maxSecondaryBits) + } + if w := tertiary; w >= 1<<maxTertiaryBits || w < 0 { + return 0, fmt.Errorf("makeCE: tertiary weight out of bounds: %x >= %x", w, 1<<maxTertiaryBits) + } + ce := Elem(0) + if primary != 0 { + if ccc != 0 { + if primary >= 1<<compactPrimaryBits { + return 0, fmt.Errorf("makeCE: primary weight with non-zero CCC out of bounds: %x >= %x", primary, 1<<compactPrimaryBits) + } + if secondary != defaultSecondary { + return 0, fmt.Errorf("makeCE: cannot combine non-default secondary value (%x) with non-zero CCC (%x)", secondary, ccc) + } + ce = Elem(tertiary << (compactPrimaryBits + maxCCCBits)) + ce |= Elem(ccc) << compactPrimaryBits + ce |= Elem(primary) + ce |= ceType3or4 + } else if tertiary == defaultTertiary { + if secondary >= 1<<maxSecondaryCompactBits { + return 0, fmt.Errorf("makeCE: secondary weight with non-zero primary out of bounds: %x >= %x", secondary, 1<<maxSecondaryCompactBits) + } + ce = Elem(primary<<(maxSecondaryCompactBits+1) + secondary) + ce |= ceType1 + } else { + d := secondary - defaultSecondary + maxSecondaryDiffBits + if d >= 1<<maxSecondaryDiffBits || d < 0 { + return 0, fmt.Errorf("makeCE: secondary weight diff out of bounds: %x < 0 || %x > %x", d, d, 1<<maxSecondaryDiffBits) + } + if tertiary >= 1<<maxTertiaryCompactBits { + return 0, fmt.Errorf("makeCE: tertiary weight with non-zero primary out of bounds: %x > %x", tertiary, 1<<maxTertiaryCompactBits) + } + ce = Elem(primary<<maxSecondaryDiffBits + d) + ce = ce<<maxTertiaryCompactBits + Elem(tertiary) + } + } else { + ce = Elem(secondary<<maxTertiaryBits + tertiary) + ce += Elem(ccc) << (maxSecondaryBits + maxTertiaryBits) + ce |= ceType4 + } + return ce, nil +} + +// MakeQuaternary returns an Elem with the given quaternary value. +func MakeQuaternary(v int) Elem { + return ceTypeQ | Elem(v<<primaryShift) +} + +// Mask sets weights for any level smaller than l to 0. +// The resulting Elem can be used to test for equality with +// other Elems to which the same mask has been applied. +func (ce Elem) Mask(l Level) uint32 { + return 0 +} + +// CCC returns the canonical combining class associated with the underlying character, +// if applicable, or 0 otherwise. +func (ce Elem) CCC() uint8 { + if ce&ceType3or4 != 0 { + if ce&ceType4 == ceType3or4 { + return uint8(ce >> 16) + } + return uint8(ce >> 20) + } + return 0 +} + +// Primary returns the primary collation weight for ce. +func (ce Elem) Primary() int { + if ce >= firstNonPrimary { + if ce > lastSpecialPrimary { + return 0 + } + return int(uint16(ce)) + } + return int(ce&primaryValueMask) >> primaryShift +} + +// Secondary returns the secondary collation weight for ce. +func (ce Elem) Secondary() int { + switch ce & ceTypeMask { + case ceType1: + return int(uint8(ce)) + case ceType2: + return minCompactSecondary + int((ce>>compactSecondaryShift)&0xF) + case ceType3or4: + if ce < ceType4 { + return defaultSecondary + } + return int(ce>>8) & 0xFFF + case ceTypeQ: + return 0 + } + panic("should not reach here") +} + +// Tertiary returns the tertiary collation weight for ce. +func (ce Elem) Tertiary() uint8 { + if ce&hasTertiaryMask == 0 { + if ce&ceType3or4 == 0 { + return uint8(ce & 0x1F) + } + if ce&ceType4 == ceType4 { + return uint8(ce) + } + return uint8(ce>>24) & 0x1F // type 2 + } else if ce&ceTypeMask == ceType1 { + return defaultTertiary + } + // ce is a quaternary value. + return 0 +} + +func (ce Elem) updateTertiary(t uint8) Elem { + if ce&ceTypeMask == ceType1 { + // convert to type 4 + nce := ce & primaryValueMask + nce |= Elem(uint8(ce)-minCompactSecondary) << compactSecondaryShift + ce = nce + } else if ce&ceTypeMaskExt == ceType3or4 { + ce &= ^Elem(maxTertiary << 24) + return ce | (Elem(t) << 24) + } else { + // type 2 or 4 + ce &= ^Elem(maxTertiary) + } + return ce | Elem(t) +} + +// Quaternary returns the quaternary value if explicitly specified, +// 0 if ce == Ignore, or MaxQuaternary otherwise. +// Quaternary values are used only for shifted variants. +func (ce Elem) Quaternary() int { + if ce&ceTypeMask == ceTypeQ { + return int(ce&primaryValueMask) >> primaryShift + } else if ce&ceIgnoreMask == Ignore { + return 0 + } + return MaxQuaternary +} + +// Weight returns the collation weight for the given level. +func (ce Elem) Weight(l Level) int { + switch l { + case Primary: + return ce.Primary() + case Secondary: + return ce.Secondary() + case Tertiary: + return int(ce.Tertiary()) + case Quaternary: + return ce.Quaternary() + } + return 0 // return 0 (ignore) for undefined levels. +} + +// For contractions, collation elements are of the form +// 110bbbbb bbbbbbbb iiiiiiii iiiinnnn, where +// - n* is the size of the first node in the contraction trie. +// - i* is the index of the first node in the contraction trie. +// - b* is the offset into the contraction collation element table. +// +// See contract.go for details on the contraction trie. +const ( + maxNBits = 4 + maxTrieIndexBits = 12 + maxContractOffsetBits = 13 +) + +func splitContractIndex(ce Elem) (index, n, offset int) { + n = int(ce & (1<<maxNBits - 1)) + ce >>= maxNBits + index = int(ce & (1<<maxTrieIndexBits - 1)) + ce >>= maxTrieIndexBits + offset = int(ce & (1<<maxContractOffsetBits - 1)) + return +} + +// For expansions, Elems are of the form 11100000 00000000 bbbbbbbb bbbbbbbb, +// where b* is the index into the expansion sequence table. +const maxExpandIndexBits = 16 + +func splitExpandIndex(ce Elem) (index int) { + return int(uint16(ce)) +} + +// Some runes can be expanded using NFKD decomposition. Instead of storing the full +// sequence of collation elements, we decompose the rune and lookup the collation +// elements for each rune in the decomposition and modify the tertiary weights. +// The Elem, in this case, is of the form 11110000 00000000 wwwwwwww vvvvvvvv, where +// - v* is the replacement tertiary weight for the first rune, +// - w* is the replacement tertiary weight for the second rune, +// +// Tertiary weights of subsequent runes should be replaced with maxTertiary. +// See https://www.unicode.org/reports/tr10/#Compatibility_Decompositions for more details. +func splitDecompose(ce Elem) (t1, t2 uint8) { + return uint8(ce), uint8(ce >> 8) +} + +const ( + // These constants were taken from https://www.unicode.org/versions/Unicode6.0.0/ch12.pdf. + minUnified rune = 0x4E00 + maxUnified = 0x9FFF + minCompatibility = 0xF900 + maxCompatibility = 0xFAFF + minRare = 0x3400 + maxRare = 0x4DBF +) +const ( + commonUnifiedOffset = 0x10000 + rareUnifiedOffset = 0x20000 // largest rune in common is U+FAFF + otherOffset = 0x50000 // largest rune in rare is U+2FA1D + illegalOffset = otherOffset + int(unicode.MaxRune) + maxPrimary = illegalOffset + 1 +) + +// implicitPrimary returns the primary weight for the a rune +// for which there is no entry for the rune in the collation table. +// We take a different approach from the one specified in +// https://unicode.org/reports/tr10/#Implicit_Weights, +// but preserve the resulting relative ordering of the runes. +func implicitPrimary(r rune) int { + if unicode.Is(unicode.Ideographic, r) { + if r >= minUnified && r <= maxUnified { + // The most common case for CJK. + return int(r) + commonUnifiedOffset + } + if r >= minCompatibility && r <= maxCompatibility { + // This will typically not hit. The DUCET explicitly specifies mappings + // for all characters that do not decompose. + return int(r) + commonUnifiedOffset + } + return int(r) + rareUnifiedOffset + } + return int(r) + otherOffset +} diff --git a/vendor/golang.org/x/text/internal/colltab/colltab.go b/vendor/golang.org/x/text/internal/colltab/colltab.go new file mode 100644 index 0000000..02f2247 --- /dev/null +++ b/vendor/golang.org/x/text/internal/colltab/colltab.go @@ -0,0 +1,105 @@ +// 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 colltab contains functionality related to collation tables. +// It is only to be used by the collate and search packages. +package colltab // import "golang.org/x/text/internal/colltab" + +import ( + "sort" + + "golang.org/x/text/language" +) + +// MatchLang finds the index of t in tags, using a matching algorithm used for +// collation and search. tags[0] must be language.Und, the remaining tags should +// be sorted alphabetically. +// +// Language matching for collation and search is different from the matching +// defined by language.Matcher: the (inferred) base language must be an exact +// match for the relevant fields. For example, "gsw" should not match "de". +// Also the parent relation is different, as a parent may have a different +// script. So usually the parent of zh-Hant is und, whereas for MatchLang it is +// zh. +func MatchLang(t language.Tag, tags []language.Tag) int { + // Canonicalize the values, including collapsing macro languages. + t, _ = language.All.Canonicalize(t) + + base, conf := t.Base() + // Estimate the base language, but only use high-confidence values. + if conf < language.High { + // The root locale supports "search" and "standard". We assume that any + // implementation will only use one of both. + return 0 + } + + // Maximize base and script and normalize the tag. + if _, s, r := t.Raw(); (r != language.Region{}) { + p, _ := language.Raw.Compose(base, s, r) + // Taking the parent forces the script to be maximized. + p = p.Parent() + // Add back region and extensions. + t, _ = language.Raw.Compose(p, r, t.Extensions()) + } else { + // Set the maximized base language. + t, _ = language.Raw.Compose(base, s, t.Extensions()) + } + + // Find start index of the language tag. + start := 1 + sort.Search(len(tags)-1, func(i int) bool { + b, _, _ := tags[i+1].Raw() + return base.String() <= b.String() + }) + if start < len(tags) { + if b, _, _ := tags[start].Raw(); b != base { + return 0 + } + } + + // Besides the base language, script and region, only the collation type and + // the custom variant defined in the 'u' extension are used to distinguish a + // locale. + // Strip all variants and extensions and add back the custom variant. + tdef, _ := language.Raw.Compose(t.Raw()) + tdef, _ = tdef.SetTypeForKey("va", t.TypeForKey("va")) + + // First search for a specialized collation type, if present. + try := []language.Tag{tdef} + if co := t.TypeForKey("co"); co != "" { + tco, _ := tdef.SetTypeForKey("co", co) + try = []language.Tag{tco, tdef} + } + + for _, tx := range try { + for ; tx != language.Und; tx = parent(tx) { + for i, t := range tags[start:] { + if b, _, _ := t.Raw(); b != base { + break + } + if tx == t { + return start + i + } + } + } + } + return 0 +} + +// parent computes the structural parent. This means inheritance may change +// script. So, unlike the CLDR parent, parent(zh-Hant) == zh. +func parent(t language.Tag) language.Tag { + if t.TypeForKey("va") != "" { + t, _ = t.SetTypeForKey("va", "") + return t + } + result := language.Und + if b, s, r := t.Raw(); (r != language.Region{}) { + result, _ = language.Raw.Compose(b, s, t.Extensions()) + } else if (s != language.Script{}) { + result, _ = language.Raw.Compose(b, t.Extensions()) + } else if (b != language.Base{}) { + result, _ = language.Raw.Compose(t.Extensions()) + } + return result +} diff --git a/vendor/golang.org/x/text/internal/colltab/contract.go b/vendor/golang.org/x/text/internal/colltab/contract.go new file mode 100644 index 0000000..25649d4 --- /dev/null +++ b/vendor/golang.org/x/text/internal/colltab/contract.go @@ -0,0 +1,145 @@ +// Copyright 2012 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 colltab + +import "unicode/utf8" + +// For a description of ContractTrieSet, see text/collate/build/contract.go. + +type ContractTrieSet []struct{ L, H, N, I uint8 } + +// ctScanner is used to match a trie to an input sequence. +// A contraction may match a non-contiguous sequence of bytes in an input string. +// For example, if there is a contraction for <a, combining_ring>, it should match +// the sequence <a, combining_cedilla, combining_ring>, as combining_cedilla does +// not block combining_ring. +// ctScanner does not automatically skip over non-blocking non-starters, but rather +// retains the state of the last match and leaves it up to the user to continue +// the match at the appropriate points. +type ctScanner struct { + states ContractTrieSet + s []byte + n int + index int + pindex int + done bool +} + +type ctScannerString struct { + states ContractTrieSet + s string + n int + index int + pindex int + done bool +} + +func (t ContractTrieSet) scanner(index, n int, b []byte) ctScanner { + return ctScanner{s: b, states: t[index:], n: n} +} + +func (t ContractTrieSet) scannerString(index, n int, str string) ctScannerString { + return ctScannerString{s: str, states: t[index:], n: n} +} + +// result returns the offset i and bytes consumed p so far. If no suffix +// matched, i and p will be 0. +func (s *ctScanner) result() (i, p int) { + return s.index, s.pindex +} + +func (s *ctScannerString) result() (i, p int) { + return s.index, s.pindex +} + +const ( + final = 0 + noIndex = 0xFF +) + +// scan matches the longest suffix at the current location in the input +// and returns the number of bytes consumed. +func (s *ctScanner) scan(p int) int { + pr := p // the p at the rune start + str := s.s + states, n := s.states, s.n + for i := 0; i < n && p < len(str); { + e := states[i] + c := str[p] + // TODO: a significant number of contractions are of a form that + // cannot match discontiguous UTF-8 in a normalized string. We could let + // a negative value of e.n mean that we can set s.done = true and avoid + // the need for additional matches. + if c >= e.L { + if e.L == c { + p++ + if e.I != noIndex { + s.index = int(e.I) + s.pindex = p + } + if e.N != final { + i, states, n = 0, states[int(e.H)+n:], int(e.N) + if p >= len(str) || utf8.RuneStart(str[p]) { + s.states, s.n, pr = states, n, p + } + } else { + s.done = true + return p + } + continue + } else if e.N == final && c <= e.H { + p++ + s.done = true + s.index = int(c-e.L) + int(e.I) + s.pindex = p + return p + } + } + i++ + } + return pr +} + +// scan is a verbatim copy of ctScanner.scan. +func (s *ctScannerString) scan(p int) int { + pr := p // the p at the rune start + str := s.s + states, n := s.states, s.n + for i := 0; i < n && p < len(str); { + e := states[i] + c := str[p] + // TODO: a significant number of contractions are of a form that + // cannot match discontiguous UTF-8 in a normalized string. We could let + // a negative value of e.n mean that we can set s.done = true and avoid + // the need for additional matches. + if c >= e.L { + if e.L == c { + p++ + if e.I != noIndex { + s.index = int(e.I) + s.pindex = p + } + if e.N != final { + i, states, n = 0, states[int(e.H)+n:], int(e.N) + if p >= len(str) || utf8.RuneStart(str[p]) { + s.states, s.n, pr = states, n, p + } + } else { + s.done = true + return p + } + continue + } else if e.N == final && c <= e.H { + p++ + s.done = true + s.index = int(c-e.L) + int(e.I) + s.pindex = p + return p + } + } + i++ + } + return pr +} diff --git a/vendor/golang.org/x/text/internal/colltab/iter.go b/vendor/golang.org/x/text/internal/colltab/iter.go new file mode 100644 index 0000000..c1b1ba8 --- /dev/null +++ b/vendor/golang.org/x/text/internal/colltab/iter.go @@ -0,0 +1,178 @@ +// 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 colltab + +// An Iter incrementally converts chunks of the input text to collation +// elements, while ensuring that the collation elements are in normalized order +// (that is, they are in the order as if the input text were normalized first). +type Iter struct { + Weighter Weighter + Elems []Elem + // N is the number of elements in Elems that will not be reordered on + // subsequent iterations, N <= len(Elems). + N int + + bytes []byte + str string + // Because the Elems buffer may contain collation elements that are needed + // for look-ahead, we need two positions in the text (bytes or str): one for + // the end position in the text for the current iteration and one for the + // start of the next call to appendNext. + pEnd int // end position in text corresponding to N. + pNext int // pEnd <= pNext. +} + +// Reset sets the position in the current input text to p and discards any +// results obtained so far. +func (i *Iter) Reset(p int) { + i.Elems = i.Elems[:0] + i.N = 0 + i.pEnd = p + i.pNext = p +} + +// Len returns the length of the input text. +func (i *Iter) Len() int { + if i.bytes != nil { + return len(i.bytes) + } + return len(i.str) +} + +// Discard removes the collation elements up to N. +func (i *Iter) Discard() { + // TODO: change this such that only modifiers following starters will have + // to be copied. + i.Elems = i.Elems[:copy(i.Elems, i.Elems[i.N:])] + i.N = 0 +} + +// End returns the end position of the input text for which Next has returned +// results. +func (i *Iter) End() int { + return i.pEnd +} + +// SetInput resets i to input s. +func (i *Iter) SetInput(s []byte) { + i.bytes = s + i.str = "" + i.Reset(0) +} + +// SetInputString resets i to input s. +func (i *Iter) SetInputString(s string) { + i.str = s + i.bytes = nil + i.Reset(0) +} + +func (i *Iter) done() bool { + return i.pNext >= len(i.str) && i.pNext >= len(i.bytes) +} + +func (i *Iter) appendNext() bool { + if i.done() { + return false + } + var sz int + if i.bytes == nil { + i.Elems, sz = i.Weighter.AppendNextString(i.Elems, i.str[i.pNext:]) + } else { + i.Elems, sz = i.Weighter.AppendNext(i.Elems, i.bytes[i.pNext:]) + } + if sz == 0 { + sz = 1 + } + i.pNext += sz + return true +} + +// Next appends Elems to the internal array. On each iteration, it will either +// add starters or modifiers. In the majority of cases, an Elem with a primary +// value > 0 will have a CCC of 0. The CCC values of collation elements are also +// used to detect if the input string was not normalized and to adjust the +// result accordingly. +func (i *Iter) Next() bool { + if i.N == len(i.Elems) && !i.appendNext() { + return false + } + + // Check if the current segment starts with a starter. + prevCCC := i.Elems[len(i.Elems)-1].CCC() + if prevCCC == 0 { + i.N = len(i.Elems) + i.pEnd = i.pNext + return true + } else if i.Elems[i.N].CCC() == 0 { + // set i.N to only cover part of i.Elems for which prevCCC == 0 and + // use rest for the next call to next. + for i.N++; i.N < len(i.Elems) && i.Elems[i.N].CCC() == 0; i.N++ { + } + i.pEnd = i.pNext + return true + } + + // The current (partial) segment starts with modifiers. We need to collect + // all successive modifiers to ensure that they are normalized. + for { + p := len(i.Elems) + i.pEnd = i.pNext + if !i.appendNext() { + break + } + + if ccc := i.Elems[p].CCC(); ccc == 0 || len(i.Elems)-i.N > maxCombiningCharacters { + // Leave the starter for the next iteration. This ensures that we + // do not return sequences of collation elements that cross two + // segments. + // + // TODO: handle large number of combining characters by fully + // normalizing the input segment before iteration. This ensures + // results are consistent across the text repo. + i.N = p + return true + } else if ccc < prevCCC { + i.doNorm(p, ccc) // should be rare, never occurs for NFD and FCC. + } else { + prevCCC = ccc + } + } + + done := len(i.Elems) != i.N + i.N = len(i.Elems) + return done +} + +// nextNoNorm is the same as next, but does not "normalize" the collation +// elements. +func (i *Iter) nextNoNorm() bool { + // TODO: remove this function. Using this instead of next does not seem + // to improve performance in any significant way. We retain this until + // later for evaluation purposes. + if i.done() { + return false + } + i.appendNext() + i.N = len(i.Elems) + return true +} + +const maxCombiningCharacters = 30 + +// doNorm reorders the collation elements in i.Elems. +// It assumes that blocks of collation elements added with appendNext +// either start and end with the same CCC or start with CCC == 0. +// This allows for a single insertion point for the entire block. +// The correctness of this assumption is verified in builder.go. +func (i *Iter) doNorm(p int, ccc uint8) { + n := len(i.Elems) + k := p + for p--; p > i.N && ccc < i.Elems[p-1].CCC(); p-- { + } + i.Elems = append(i.Elems, i.Elems[p:k]...) + copy(i.Elems[p:], i.Elems[k:]) + i.Elems = i.Elems[:n] +} diff --git a/vendor/golang.org/x/text/internal/colltab/numeric.go b/vendor/golang.org/x/text/internal/colltab/numeric.go new file mode 100644 index 0000000..53b819c --- /dev/null +++ b/vendor/golang.org/x/text/internal/colltab/numeric.go @@ -0,0 +1,236 @@ +// 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 colltab + +import ( + "unicode" + "unicode/utf8" +) + +// NewNumericWeighter wraps w to replace individual digits to sort based on their +// numeric value. +// +// Weighter w must have a free primary weight after the primary weight for 9. +// If this is not the case, numeric value will sort at the same primary level +// as the first primary sorting after 9. +func NewNumericWeighter(w Weighter) Weighter { + getElem := func(s string) Elem { + elems, _ := w.AppendNextString(nil, s) + return elems[0] + } + nine := getElem("9") + + // Numbers should order before zero, but the DUCET has no room for this. + // TODO: move before zero once we use fractional collation elements. + ns, _ := MakeElem(nine.Primary()+1, nine.Secondary(), int(nine.Tertiary()), 0) + + return &numericWeighter{ + Weighter: w, + + // We assume that w sorts digits of different kinds in order of numeric + // value and that the tertiary weight order is preserved. + // + // TODO: evaluate whether it is worth basing the ranges on the Elem + // encoding itself once the move to fractional weights is complete. + zero: getElem("0"), + zeroSpecialLo: getElem("0"), // U+FF10 FULLWIDTH DIGIT ZERO + zeroSpecialHi: getElem("₀"), // U+2080 SUBSCRIPT ZERO + nine: nine, + nineSpecialHi: getElem("₉"), // U+2089 SUBSCRIPT NINE + numberStart: ns, + } +} + +// A numericWeighter translates a stream of digits into a stream of weights +// representing the numeric value. +type numericWeighter struct { + Weighter + + // The Elems below all demarcate boundaries of specific ranges. With the + // current element encoding digits are in two ranges: normal (default + // tertiary value) and special. For most languages, digits have collation + // elements in the normal range. + // + // Note: the range tests are very specific for the element encoding used by + // this implementation. The tests in collate_test.go are designed to fail + // if this code is not updated when an encoding has changed. + + zero Elem // normal digit zero + zeroSpecialLo Elem // special digit zero, low tertiary value + zeroSpecialHi Elem // special digit zero, high tertiary value + nine Elem // normal digit nine + nineSpecialHi Elem // special digit nine + numberStart Elem +} + +// AppendNext calls the namesake of the underlying weigher, but replaces single +// digits with weights representing their value. +func (nw *numericWeighter) AppendNext(buf []Elem, s []byte) (ce []Elem, n int) { + ce, n = nw.Weighter.AppendNext(buf, s) + nc := numberConverter{ + elems: buf, + w: nw, + b: s, + } + isZero, ok := nc.checkNextDigit(ce) + if !ok { + return ce, n + } + // ce might have been grown already, so take it instead of buf. + nc.init(ce, len(buf), isZero) + for n < len(s) { + ce, sz := nw.Weighter.AppendNext(nc.elems, s[n:]) + nc.b = s + n += sz + if !nc.update(ce) { + break + } + } + return nc.result(), n +} + +// AppendNextString calls the namesake of the underlying weigher, but replaces +// single digits with weights representing their value. +func (nw *numericWeighter) AppendNextString(buf []Elem, s string) (ce []Elem, n int) { + ce, n = nw.Weighter.AppendNextString(buf, s) + nc := numberConverter{ + elems: buf, + w: nw, + s: s, + } + isZero, ok := nc.checkNextDigit(ce) + if !ok { + return ce, n + } + nc.init(ce, len(buf), isZero) + for n < len(s) { + ce, sz := nw.Weighter.AppendNextString(nc.elems, s[n:]) + nc.s = s + n += sz + if !nc.update(ce) { + break + } + } + return nc.result(), n +} + +type numberConverter struct { + w *numericWeighter + + elems []Elem + nDigits int + lenIndex int + + s string // set if the input was of type string + b []byte // set if the input was of type []byte +} + +// init completes initialization of a numberConverter and prepares it for adding +// more digits. elems is assumed to have a digit starting at oldLen. +func (nc *numberConverter) init(elems []Elem, oldLen int, isZero bool) { + // Insert a marker indicating the start of a number and a placeholder + // for the number of digits. + if isZero { + elems = append(elems[:oldLen], nc.w.numberStart, 0) + } else { + elems = append(elems, 0, 0) + copy(elems[oldLen+2:], elems[oldLen:]) + elems[oldLen] = nc.w.numberStart + elems[oldLen+1] = 0 + + nc.nDigits = 1 + } + nc.elems = elems + nc.lenIndex = oldLen + 1 +} + +// checkNextDigit reports whether bufNew adds a single digit relative to the old +// buffer. If it does, it also reports whether this digit is zero. +func (nc *numberConverter) checkNextDigit(bufNew []Elem) (isZero, ok bool) { + if len(nc.elems) >= len(bufNew) { + return false, false + } + e := bufNew[len(nc.elems)] + if e < nc.w.zeroSpecialLo || nc.w.nine < e { + // Not a number. + return false, false + } + if e < nc.w.zero { + if e > nc.w.nineSpecialHi { + // Not a number. + return false, false + } + if !nc.isDigit() { + return false, false + } + isZero = e <= nc.w.zeroSpecialHi + } else { + // This is the common case if we encounter a digit. + isZero = e == nc.w.zero + } + // Test the remaining added collation elements have a zero primary value. + if n := len(bufNew) - len(nc.elems); n > 1 { + for i := len(nc.elems) + 1; i < len(bufNew); i++ { + if bufNew[i].Primary() != 0 { + return false, false + } + } + // In some rare cases, collation elements will encode runes in + // unicode.No as a digit. For example Ethiopic digits (U+1369 - U+1371) + // are not in Nd. Also some digits that clearly belong in unicode.No, + // like U+0C78 TELUGU FRACTION DIGIT ZERO FOR ODD POWERS OF FOUR, have + // collation elements indistinguishable from normal digits. + // Unfortunately, this means we need to make this check for nearly all + // non-Latin digits. + // + // TODO: check the performance impact and find something better if it is + // an issue. + if !nc.isDigit() { + return false, false + } + } + return isZero, true +} + +func (nc *numberConverter) isDigit() bool { + if nc.b != nil { + r, _ := utf8.DecodeRune(nc.b) + return unicode.In(r, unicode.Nd) + } + r, _ := utf8.DecodeRuneInString(nc.s) + return unicode.In(r, unicode.Nd) +} + +// We currently support a maximum of about 2M digits (the number of primary +// values). Such numbers will compare correctly against small numbers, but their +// comparison against other large numbers is undefined. +// +// TODO: define a proper fallback, such as comparing large numbers textually or +// actually allowing numbers of unlimited length. +// +// TODO: cap this to a lower number (like 100) and maybe allow a larger number +// in an option? +const maxDigits = 1<<maxPrimaryBits - 1 + +func (nc *numberConverter) update(elems []Elem) bool { + isZero, ok := nc.checkNextDigit(elems) + if nc.nDigits == 0 && isZero { + return true + } + nc.elems = elems + if !ok { + return false + } + nc.nDigits++ + return nc.nDigits < maxDigits +} + +// result fills in the length element for the digit sequence and returns the +// completed collation elements. +func (nc *numberConverter) result() []Elem { + e, _ := MakeElem(nc.nDigits, defaultSecondary, defaultTertiary, 0) + nc.elems[nc.lenIndex] = e + return nc.elems +} diff --git a/vendor/golang.org/x/text/internal/colltab/table.go b/vendor/golang.org/x/text/internal/colltab/table.go new file mode 100644 index 0000000..e26e36d --- /dev/null +++ b/vendor/golang.org/x/text/internal/colltab/table.go @@ -0,0 +1,275 @@ +// Copyright 2012 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 colltab + +import ( + "unicode/utf8" + + "golang.org/x/text/unicode/norm" +) + +// Table holds all collation data for a given collation ordering. +type Table struct { + Index Trie // main trie + + // expansion info + ExpandElem []uint32 + + // contraction info + ContractTries ContractTrieSet + ContractElem []uint32 + MaxContractLen int + VariableTop uint32 +} + +func (t *Table) AppendNext(w []Elem, b []byte) (res []Elem, n int) { + return t.appendNext(w, source{bytes: b}) +} + +func (t *Table) AppendNextString(w []Elem, s string) (res []Elem, n int) { + return t.appendNext(w, source{str: s}) +} + +func (t *Table) Start(p int, b []byte) int { + // TODO: implement + panic("not implemented") +} + +func (t *Table) StartString(p int, s string) int { + // TODO: implement + panic("not implemented") +} + +func (t *Table) Domain() []string { + // TODO: implement + panic("not implemented") +} + +func (t *Table) Top() uint32 { + return t.VariableTop +} + +type source struct { + str string + bytes []byte +} + +func (src *source) lookup(t *Table) (ce Elem, sz int) { + if src.bytes == nil { + return t.Index.lookupString(src.str) + } + return t.Index.lookup(src.bytes) +} + +func (src *source) tail(sz int) { + if src.bytes == nil { + src.str = src.str[sz:] + } else { + src.bytes = src.bytes[sz:] + } +} + +func (src *source) nfd(buf []byte, end int) []byte { + if src.bytes == nil { + return norm.NFD.AppendString(buf[:0], src.str[:end]) + } + return norm.NFD.Append(buf[:0], src.bytes[:end]...) +} + +func (src *source) rune() (r rune, sz int) { + if src.bytes == nil { + return utf8.DecodeRuneInString(src.str) + } + return utf8.DecodeRune(src.bytes) +} + +func (src *source) properties(f norm.Form) norm.Properties { + if src.bytes == nil { + return f.PropertiesString(src.str) + } + return f.Properties(src.bytes) +} + +// appendNext appends the weights corresponding to the next rune or +// contraction in s. If a contraction is matched to a discontinuous +// sequence of runes, the weights for the interstitial runes are +// appended as well. It returns a new slice that includes the appended +// weights and the number of bytes consumed from s. +func (t *Table) appendNext(w []Elem, src source) (res []Elem, n int) { + ce, sz := src.lookup(t) + tp := ce.ctype() + if tp == ceNormal { + if ce == 0 { + r, _ := src.rune() + const ( + hangulSize = 3 + firstHangul = 0xAC00 + lastHangul = 0xD7A3 + ) + if r >= firstHangul && r <= lastHangul { + // TODO: performance can be considerably improved here. + n = sz + var buf [16]byte // Used for decomposing Hangul. + for b := src.nfd(buf[:0], hangulSize); len(b) > 0; b = b[sz:] { + ce, sz = t.Index.lookup(b) + w = append(w, ce) + } + return w, n + } + ce = makeImplicitCE(implicitPrimary(r)) + } + w = append(w, ce) + } else if tp == ceExpansionIndex { + w = t.appendExpansion(w, ce) + } else if tp == ceContractionIndex { + n := 0 + src.tail(sz) + if src.bytes == nil { + w, n = t.matchContractionString(w, ce, src.str) + } else { + w, n = t.matchContraction(w, ce, src.bytes) + } + sz += n + } else if tp == ceDecompose { + // Decompose using NFKD and replace tertiary weights. + t1, t2 := splitDecompose(ce) + i := len(w) + nfkd := src.properties(norm.NFKD).Decomposition() + for p := 0; len(nfkd) > 0; nfkd = nfkd[p:] { + w, p = t.appendNext(w, source{bytes: nfkd}) + } + w[i] = w[i].updateTertiary(t1) + if i++; i < len(w) { + w[i] = w[i].updateTertiary(t2) + for i++; i < len(w); i++ { + w[i] = w[i].updateTertiary(maxTertiary) + } + } + } + return w, sz +} + +func (t *Table) appendExpansion(w []Elem, ce Elem) []Elem { + i := splitExpandIndex(ce) + n := int(t.ExpandElem[i]) + i++ + for _, ce := range t.ExpandElem[i : i+n] { + w = append(w, Elem(ce)) + } + return w +} + +func (t *Table) matchContraction(w []Elem, ce Elem, suffix []byte) ([]Elem, int) { + index, n, offset := splitContractIndex(ce) + + scan := t.ContractTries.scanner(index, n, suffix) + buf := [norm.MaxSegmentSize]byte{} + bufp := 0 + p := scan.scan(0) + + if !scan.done && p < len(suffix) && suffix[p] >= utf8.RuneSelf { + // By now we should have filtered most cases. + p0 := p + bufn := 0 + rune := norm.NFD.Properties(suffix[p:]) + p += rune.Size() + if rune.LeadCCC() != 0 { + prevCC := rune.TrailCCC() + // A gap may only occur in the last normalization segment. + // This also ensures that len(scan.s) < norm.MaxSegmentSize. + if end := norm.NFD.FirstBoundary(suffix[p:]); end != -1 { + scan.s = suffix[:p+end] + } + for p < len(suffix) && !scan.done && suffix[p] >= utf8.RuneSelf { + rune = norm.NFD.Properties(suffix[p:]) + if ccc := rune.LeadCCC(); ccc == 0 || prevCC >= ccc { + break + } + prevCC = rune.TrailCCC() + if pp := scan.scan(p); pp != p { + // Copy the interstitial runes for later processing. + bufn += copy(buf[bufn:], suffix[p0:p]) + if scan.pindex == pp { + bufp = bufn + } + p, p0 = pp, pp + } else { + p += rune.Size() + } + } + } + } + // Append weights for the matched contraction, which may be an expansion. + i, n := scan.result() + ce = Elem(t.ContractElem[i+offset]) + if ce.ctype() == ceNormal { + w = append(w, ce) + } else { + w = t.appendExpansion(w, ce) + } + // Append weights for the runes in the segment not part of the contraction. + for b, p := buf[:bufp], 0; len(b) > 0; b = b[p:] { + w, p = t.appendNext(w, source{bytes: b}) + } + return w, n +} + +// TODO: unify the two implementations. This is best done after first simplifying +// the algorithm taking into account the inclusion of both NFC and NFD forms +// in the table. +func (t *Table) matchContractionString(w []Elem, ce Elem, suffix string) ([]Elem, int) { + index, n, offset := splitContractIndex(ce) + + scan := t.ContractTries.scannerString(index, n, suffix) + buf := [norm.MaxSegmentSize]byte{} + bufp := 0 + p := scan.scan(0) + + if !scan.done && p < len(suffix) && suffix[p] >= utf8.RuneSelf { + // By now we should have filtered most cases. + p0 := p + bufn := 0 + rune := norm.NFD.PropertiesString(suffix[p:]) + p += rune.Size() + if rune.LeadCCC() != 0 { + prevCC := rune.TrailCCC() + // A gap may only occur in the last normalization segment. + // This also ensures that len(scan.s) < norm.MaxSegmentSize. + if end := norm.NFD.FirstBoundaryInString(suffix[p:]); end != -1 { + scan.s = suffix[:p+end] + } + for p < len(suffix) && !scan.done && suffix[p] >= utf8.RuneSelf { + rune = norm.NFD.PropertiesString(suffix[p:]) + if ccc := rune.LeadCCC(); ccc == 0 || prevCC >= ccc { + break + } + prevCC = rune.TrailCCC() + if pp := scan.scan(p); pp != p { + // Copy the interstitial runes for later processing. + bufn += copy(buf[bufn:], suffix[p0:p]) + if scan.pindex == pp { + bufp = bufn + } + p, p0 = pp, pp + } else { + p += rune.Size() + } + } + } + } + // Append weights for the matched contraction, which may be an expansion. + i, n := scan.result() + ce = Elem(t.ContractElem[i+offset]) + if ce.ctype() == ceNormal { + w = append(w, ce) + } else { + w = t.appendExpansion(w, ce) + } + // Append weights for the runes in the segment not part of the contraction. + for b, p := buf[:bufp], 0; len(b) > 0; b = b[p:] { + w, p = t.appendNext(w, source{bytes: b}) + } + return w, n +} diff --git a/vendor/golang.org/x/text/internal/colltab/trie.go b/vendor/golang.org/x/text/internal/colltab/trie.go new file mode 100644 index 0000000..a0eaa0d --- /dev/null +++ b/vendor/golang.org/x/text/internal/colltab/trie.go @@ -0,0 +1,159 @@ +// Copyright 2012 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. + +// The trie in this file is used to associate the first full character in an +// UTF-8 string to a collation element. All but the last byte in a UTF-8 byte +// sequence are used to lookup offsets in the index table to be used for the +// next byte. The last byte is used to index into a table of collation elements. +// For a full description, see go.text/collate/build/trie.go. + +package colltab + +const blockSize = 64 + +type Trie struct { + Index0 []uint16 // index for first byte (0xC0-0xFF) + Values0 []uint32 // index for first byte (0x00-0x7F) + Index []uint16 + Values []uint32 +} + +const ( + t1 = 0x00 // 0000 0000 + tx = 0x80 // 1000 0000 + t2 = 0xC0 // 1100 0000 + t3 = 0xE0 // 1110 0000 + t4 = 0xF0 // 1111 0000 + t5 = 0xF8 // 1111 1000 + t6 = 0xFC // 1111 1100 + te = 0xFE // 1111 1110 +) + +func (t *Trie) lookupValue(n uint16, b byte) Elem { + return Elem(t.Values[int(n)<<6+int(b)]) +} + +// lookup returns the trie value for the first UTF-8 encoding in s and +// the width in bytes of this encoding. The size will be 0 if s does not +// hold enough bytes to complete the encoding. len(s) must be greater than 0. +func (t *Trie) lookup(s []byte) (v Elem, sz int) { + c0 := s[0] + switch { + case c0 < tx: + return Elem(t.Values0[c0]), 1 + case c0 < t2: + return 0, 1 + case c0 < t3: + if len(s) < 2 { + return 0, 0 + } + i := t.Index0[c0] + c1 := s[1] + if c1 < tx || t2 <= c1 { + return 0, 1 + } + return t.lookupValue(i, c1), 2 + case c0 < t4: + if len(s) < 3 { + return 0, 0 + } + i := t.Index0[c0] + c1 := s[1] + if c1 < tx || t2 <= c1 { + return 0, 1 + } + o := int(i)<<6 + int(c1) + i = t.Index[o] + c2 := s[2] + if c2 < tx || t2 <= c2 { + return 0, 2 + } + return t.lookupValue(i, c2), 3 + case c0 < t5: + if len(s) < 4 { + return 0, 0 + } + i := t.Index0[c0] + c1 := s[1] + if c1 < tx || t2 <= c1 { + return 0, 1 + } + o := int(i)<<6 + int(c1) + i = t.Index[o] + c2 := s[2] + if c2 < tx || t2 <= c2 { + return 0, 2 + } + o = int(i)<<6 + int(c2) + i = t.Index[o] + c3 := s[3] + if c3 < tx || t2 <= c3 { + return 0, 3 + } + return t.lookupValue(i, c3), 4 + } + // Illegal rune + return 0, 1 +} + +// The body of lookupString is a verbatim copy of that of lookup. +func (t *Trie) lookupString(s string) (v Elem, sz int) { + c0 := s[0] + switch { + case c0 < tx: + return Elem(t.Values0[c0]), 1 + case c0 < t2: + return 0, 1 + case c0 < t3: + if len(s) < 2 { + return 0, 0 + } + i := t.Index0[c0] + c1 := s[1] + if c1 < tx || t2 <= c1 { + return 0, 1 + } + return t.lookupValue(i, c1), 2 + case c0 < t4: + if len(s) < 3 { + return 0, 0 + } + i := t.Index0[c0] + c1 := s[1] + if c1 < tx || t2 <= c1 { + return 0, 1 + } + o := int(i)<<6 + int(c1) + i = t.Index[o] + c2 := s[2] + if c2 < tx || t2 <= c2 { + return 0, 2 + } + return t.lookupValue(i, c2), 3 + case c0 < t5: + if len(s) < 4 { + return 0, 0 + } + i := t.Index0[c0] + c1 := s[1] + if c1 < tx || t2 <= c1 { + return 0, 1 + } + o := int(i)<<6 + int(c1) + i = t.Index[o] + c2 := s[2] + if c2 < tx || t2 <= c2 { + return 0, 2 + } + o = int(i)<<6 + int(c2) + i = t.Index[o] + c3 := s[3] + if c3 < tx || t2 <= c3 { + return 0, 3 + } + return t.lookupValue(i, c3), 4 + } + // Illegal rune + return 0, 1 +} diff --git a/vendor/golang.org/x/text/internal/colltab/weighter.go b/vendor/golang.org/x/text/internal/colltab/weighter.go new file mode 100644 index 0000000..f1ec45f --- /dev/null +++ b/vendor/golang.org/x/text/internal/colltab/weighter.go @@ -0,0 +1,31 @@ +// Copyright 2013 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 colltab // import "golang.org/x/text/internal/colltab" + +// A Weighter can be used as a source for Collator and Searcher. +type Weighter interface { + // Start finds the start of the segment that includes position p. + Start(p int, b []byte) int + + // StartString finds the start of the segment that includes position p. + StartString(p int, s string) int + + // AppendNext appends Elems to buf corresponding to the longest match + // of a single character or contraction from the start of s. + // It returns the new buf and the number of bytes consumed. + AppendNext(buf []Elem, s []byte) (ce []Elem, n int) + + // AppendNextString appends Elems to buf corresponding to the longest match + // of a single character or contraction from the start of s. + // It returns the new buf and the number of bytes consumed. + AppendNextString(buf []Elem, s string) (ce []Elem, n int) + + // Domain returns a slice of all single characters and contractions for which + // collation elements are defined in this table. + Domain() []string + + // Top returns the highest variable primary value. + Top() uint32 +} |