fail on error (#20)

ci-via-github-actions
This commit is contained in:
Casey Lee authored and GitHub committed 2019-01-23 11:51:22 -08:00
1 parent f98caa2656
commit 320e9b6057
1274 files changed
+398630 -9

No files matched your search

File diff suppressed because it is too large. Load diff
+259
View File
@@ -0,0 +1,259 @@
// Copyright 2011 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 norm
// This file contains Form-specific logic and wrappers for data in tables.go.
// Rune info is stored in a separate trie per composing form. A composing form
// and its corresponding decomposing form share the same trie. Each trie maps
// a rune to a uint16. The values take two forms. For v >= 0x8000:
// bits
// 15: 1 (inverse of NFD_QC bit of qcInfo)
// 13..7: qcInfo (see below). isYesD is always true (no decompostion).
// 6..0: ccc (compressed CCC value).
// For v < 0x8000, the respective rune has a decomposition and v is an index
// into a byte array of UTF-8 decomposition sequences and additional info and
// has the form:
// <header> <decomp_byte>* [<tccc> [<lccc>]]
// The header contains the number of bytes in the decomposition (excluding this
// length byte). The two most significant bits of this length byte correspond
// to bit 5 and 4 of qcInfo (see below). The byte sequence itself starts at v+1.
// The byte sequence is followed by a trailing and leading CCC if the values
// for these are not zero. The value of v determines which ccc are appended
// to the sequences. For v < firstCCC, there are none, for v >= firstCCC,
// the sequence is followed by a trailing ccc, and for v >= firstLeadingCC
// there is an additional leading ccc. The value of tccc itself is the
// trailing CCC shifted left 2 bits. The two least-significant bits of tccc
// are the number of trailing non-starters.
const (
qcInfoMask = 0x3F // to clear all but the relevant bits in a qcInfo
headerLenMask = 0x3F // extract the length value from the header byte
headerFlagsMask = 0xC0 // extract the qcInfo bits from the header byte
)
// Properties provides access to normalization properties of a rune.
type Properties struct {
pos uint8 // start position in reorderBuffer; used in composition.go
size uint8 // length of UTF-8 encoding of this rune
ccc uint8 // leading canonical combining class (ccc if not decomposition)
tccc uint8 // trailing canonical combining class (ccc if not decomposition)
nLead uint8 // number of leading non-starters.
flags qcInfo // quick check flags
index uint16
}
// functions dispatchable per form
type lookupFunc func(b input, i int) Properties
// formInfo holds Form-specific functions and tables.
type formInfo struct {
form Form
composing, compatibility bool // form type
info lookupFunc
nextMain iterFunc
}
var formTable = []*formInfo{{
form: NFC,
composing: true,
compatibility: false,
info: lookupInfoNFC,
nextMain: nextComposed,
}, {
form: NFD,
composing: false,
compatibility: false,
info: lookupInfoNFC,
nextMain: nextDecomposed,
}, {
form: NFKC,
composing: true,
compatibility: true,
info: lookupInfoNFKC,
nextMain: nextComposed,
}, {
form: NFKD,
composing: false,
compatibility: true,
info: lookupInfoNFKC,
nextMain: nextDecomposed,
}}
// We do not distinguish between boundaries for NFC, NFD, etc. to avoid
// unexpected behavior for the user. For example, in NFD, there is a boundary
// after 'a'. However, 'a' might combine with modifiers, so from the application's
// perspective it is not a good boundary. We will therefore always use the
// boundaries for the combining variants.
// BoundaryBefore returns true if this rune starts a new segment and
// cannot combine with any rune on the left.
func (p Properties) BoundaryBefore() bool {
if p.ccc == 0 && !p.combinesBackward() {
return true
}
// We assume that the CCC of the first character in a decomposition
// is always non-zero if different from info.ccc and that we can return
// false at this point. This is verified by maketables.
return false
}
// BoundaryAfter returns true if runes cannot combine with or otherwise
// interact with this or previous runes.
func (p Properties) BoundaryAfter() bool {
// TODO: loosen these conditions.
return p.isInert()
}
// We pack quick check data in 4 bits:
// 5: Combines forward (0 == false, 1 == true)
// 4..3: NFC_QC Yes(00), No (10), or Maybe (11)
// 2: NFD_QC Yes (0) or No (1). No also means there is a decomposition.
// 1..0: Number of trailing non-starters.
//
// When all 4 bits are zero, the character is inert, meaning it is never
// influenced by normalization.
type qcInfo uint8
func (p Properties) isYesC() bool { return p.flags&0x10 == 0 }
func (p Properties) isYesD() bool { return p.flags&0x4 == 0 }
func (p Properties) combinesForward() bool { return p.flags&0x20 != 0 }
func (p Properties) combinesBackward() bool { return p.flags&0x8 != 0 } // == isMaybe
func (p Properties) hasDecomposition() bool { return p.flags&0x4 != 0 } // == isNoD
func (p Properties) isInert() bool {
return p.flags&qcInfoMask == 0 && p.ccc == 0
}
func (p Properties) multiSegment() bool {
return p.index >= firstMulti && p.index < endMulti
}
func (p Properties) nLeadingNonStarters() uint8 {
return p.nLead
}
func (p Properties) nTrailingNonStarters() uint8 {
return uint8(p.flags & 0x03)
}
// Decomposition returns the decomposition for the underlying rune
// or nil if there is none.
func (p Properties) Decomposition() []byte {
// TODO: create the decomposition for Hangul?
if p.index == 0 {
return nil
}
i := p.index
n := decomps[i] & headerLenMask
i++
return decomps[i : i+uint16(n)]
}
// Size returns the length of UTF-8 encoding of the rune.
func (p Properties) Size() int {
return int(p.size)
}
// CCC returns the canonical combining class of the underlying rune.
func (p Properties) CCC() uint8 {
if p.index >= firstCCCZeroExcept {
return 0
}
return ccc[p.ccc]
}
// LeadCCC returns the CCC of the first rune in the decomposition.
// If there is no decomposition, LeadCCC equals CCC.
func (p Properties) LeadCCC() uint8 {
return ccc[p.ccc]
}
// TrailCCC returns the CCC of the last rune in the decomposition.
// If there is no decomposition, TrailCCC equals CCC.
func (p Properties) TrailCCC() uint8 {
return ccc[p.tccc]
}
// Recomposition
// We use 32-bit keys instead of 64-bit for the two codepoint keys.
// This clips off the bits of three entries, but we know this will not
// result in a collision. In the unlikely event that changes to
// UnicodeData.txt introduce collisions, the compiler will catch it.
// Note that the recomposition map for NFC and NFKC are identical.
// combine returns the combined rune or 0 if it doesn't exist.
func combine(a, b rune) rune {
key := uint32(uint16(a))<<16 + uint32(uint16(b))
return recompMap[key]
}
func lookupInfoNFC(b input, i int) Properties {
v, sz := b.charinfoNFC(i)
return compInfo(v, sz)
}
func lookupInfoNFKC(b input, i int) Properties {
v, sz := b.charinfoNFKC(i)
return compInfo(v, sz)
}
// Properties returns properties for the first rune in s.
func (f Form) Properties(s []byte) Properties {
if f == NFC || f == NFD {
return compInfo(nfcData.lookup(s))
}
return compInfo(nfkcData.lookup(s))
}
// PropertiesString returns properties for the first rune in s.
func (f Form) PropertiesString(s string) Properties {
if f == NFC || f == NFD {
return compInfo(nfcData.lookupString(s))
}
return compInfo(nfkcData.lookupString(s))
}
// compInfo converts the information contained in v and sz
// to a Properties. See the comment at the top of the file
// for more information on the format.
func compInfo(v uint16, sz int) Properties {
if v == 0 {
return Properties{size: uint8(sz)}
} else if v >= 0x8000 {
p := Properties{
size: uint8(sz),
ccc: uint8(v),
tccc: uint8(v),
flags: qcInfo(v >> 8),
}
if p.ccc > 0 || p.combinesBackward() {
p.nLead = uint8(p.flags & 0x3)
}
return p
}
// has decomposition
h := decomps[v]
f := (qcInfo(h&headerFlagsMask) >> 2) | 0x4
p := Properties{size: uint8(sz), flags: f, index: v}
if v >= firstCCC {
v += uint16(h&headerLenMask) + 1
c := decomps[v]
p.tccc = c >> 2
p.flags |= qcInfo(c & 0x3)
if v >= firstLeadingCCC {
p.nLead = c & 0x3
if v >= firstStarterWithNLead {
// We were tricked. Remove the decomposition.
p.flags &= 0x03
p.index = 0
return p
}
p.ccc = decomps[v+1]
}
}
return p
}
+109
View File
@@ -0,0 +1,109 @@
// Copyright 2011 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 norm
import "unicode/utf8"
type input struct {
str string
bytes []byte
}
func inputBytes(str []byte) input {
return input{bytes: str}
}
func inputString(str string) input {
return input{str: str}
}
func (in *input) setBytes(str []byte) {
in.str = ""
in.bytes = str
}
func (in *input) setString(str string) {
in.str = str
in.bytes = nil
}
func (in *input) _byte(p int) byte {
if in.bytes == nil {
return in.str[p]
}
return in.bytes[p]
}
func (in *input) skipASCII(p, max int) int {
if in.bytes == nil {
for ; p < max && in.str[p] < utf8.RuneSelf; p++ {
}
} else {
for ; p < max && in.bytes[p] < utf8.RuneSelf; p++ {
}
}
return p
}
func (in *input) skipContinuationBytes(p int) int {
if in.bytes == nil {
for ; p < len(in.str) && !utf8.RuneStart(in.str[p]); p++ {
}
} else {
for ; p < len(in.bytes) && !utf8.RuneStart(in.bytes[p]); p++ {
}
}
return p
}
func (in *input) appendSlice(buf []byte, b, e int) []byte {
if in.bytes != nil {
return append(buf, in.bytes[b:e]...)
}
for i := b; i < e; i++ {
buf = append(buf, in.str[i])
}
return buf
}
func (in *input) copySlice(buf []byte, b, e int) int {
if in.bytes == nil {
return copy(buf, in.str[b:e])
}
return copy(buf, in.bytes[b:e])
}
func (in *input) charinfoNFC(p int) (uint16, int) {
if in.bytes == nil {
return nfcData.lookupString(in.str[p:])
}
return nfcData.lookup(in.bytes[p:])
}
func (in *input) charinfoNFKC(p int) (uint16, int) {
if in.bytes == nil {
return nfkcData.lookupString(in.str[p:])
}
return nfkcData.lookup(in.bytes[p:])
}
func (in *input) hangul(p int) (r rune) {
var size int
if in.bytes == nil {
if !isHangulString(in.str[p:]) {
return 0
}
r, size = utf8.DecodeRuneInString(in.str[p:])
} else {
if !isHangul(in.bytes[p:]) {
return 0
}
r, size = utf8.DecodeRune(in.bytes[p:])
}
if size != hangulUTF8Size {
return 0
}
return r
}
+457
View File
@@ -0,0 +1,457 @@
// Copyright 2011 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 norm
import (
"fmt"
"unicode/utf8"
)
// MaxSegmentSize is the maximum size of a byte buffer needed to consider any
// sequence of starter and non-starter runes for the purpose of normalization.
const MaxSegmentSize = maxByteBufferSize
// An Iter iterates over a string or byte slice, while normalizing it
// to a given Form.
type Iter struct {
rb reorderBuffer
buf [maxByteBufferSize]byte
info Properties // first character saved from previous iteration
next iterFunc // implementation of next depends on form
asciiF iterFunc
p int // current position in input source
multiSeg []byte // remainder of multi-segment decomposition
}
type iterFunc func(*Iter) []byte
// Init initializes i to iterate over src after normalizing it to Form f.
func (i *Iter) Init(f Form, src []byte) {
i.p = 0
if len(src) == 0 {
i.setDone()
i.rb.nsrc = 0
return
}
i.multiSeg = nil
i.rb.init(f, src)
i.next = i.rb.f.nextMain
i.asciiF = nextASCIIBytes
i.info = i.rb.f.info(i.rb.src, i.p)
i.rb.ss.first(i.info)
}
// InitString initializes i to iterate over src after normalizing it to Form f.
func (i *Iter) InitString(f Form, src string) {
i.p = 0
if len(src) == 0 {
i.setDone()
i.rb.nsrc = 0
return
}
i.multiSeg = nil
i.rb.initString(f, src)
i.next = i.rb.f.nextMain
i.asciiF = nextASCIIString
i.info = i.rb.f.info(i.rb.src, i.p)
i.rb.ss.first(i.info)
}
// Seek sets the segment to be returned by the next call to Next to start
// at position p. It is the responsibility of the caller to set p to the
// start of a segment.
func (i *Iter) Seek(offset int64, whence int) (int64, error) {
var abs int64
switch whence {
case 0:
abs = offset
case 1:
abs = int64(i.p) + offset
case 2:
abs = int64(i.rb.nsrc) + offset
default:
return 0, fmt.Errorf("norm: invalid whence")
}
if abs < 0 {
return 0, fmt.Errorf("norm: negative position")
}
if int(abs) >= i.rb.nsrc {
i.setDone()
return int64(i.p), nil
}
i.p = int(abs)
i.multiSeg = nil
i.next = i.rb.f.nextMain
i.info = i.rb.f.info(i.rb.src, i.p)
i.rb.ss.first(i.info)
return abs, nil
}
// returnSlice returns a slice of the underlying input type as a byte slice.
// If the underlying is of type []byte, it will simply return a slice.
// If the underlying is of type string, it will copy the slice to the buffer
// and return that.
func (i *Iter) returnSlice(a, b int) []byte {
if i.rb.src.bytes == nil {
return i.buf[:copy(i.buf[:], i.rb.src.str[a:b])]
}
return i.rb.src.bytes[a:b]
}
// Pos returns the byte position at which the next call to Next will commence processing.
func (i *Iter) Pos() int {
return i.p
}
func (i *Iter) setDone() {
i.next = nextDone
i.p = i.rb.nsrc
}
// Done returns true if there is no more input to process.
func (i *Iter) Done() bool {
return i.p >= i.rb.nsrc
}
// Next returns f(i.input[i.Pos():n]), where n is a boundary of i.input.
// For any input a and b for which f(a) == f(b), subsequent calls
// to Next will return the same segments.
// Modifying runes are grouped together with the preceding starter, if such a starter exists.
// Although not guaranteed, n will typically be the smallest possible n.
func (i *Iter) Next() []byte {
return i.next(i)
}
func nextASCIIBytes(i *Iter) []byte {
p := i.p + 1
if p >= i.rb.nsrc {
i.setDone()
return i.rb.src.bytes[i.p:p]
}
if i.rb.src.bytes[p] < utf8.RuneSelf {
p0 := i.p
i.p = p
return i.rb.src.bytes[p0:p]
}
i.info = i.rb.f.info(i.rb.src, i.p)
i.next = i.rb.f.nextMain
return i.next(i)
}
func nextASCIIString(i *Iter) []byte {
p := i.p + 1
if p >= i.rb.nsrc {
i.buf[0] = i.rb.src.str[i.p]
i.setDone()
return i.buf[:1]
}
if i.rb.src.str[p] < utf8.RuneSelf {
i.buf[0] = i.rb.src.str[i.p]
i.p = p
return i.buf[:1]
}
i.info = i.rb.f.info(i.rb.src, i.p)
i.next = i.rb.f.nextMain
return i.next(i)
}
func nextHangul(i *Iter) []byte {
p := i.p
next := p + hangulUTF8Size
if next >= i.rb.nsrc {
i.setDone()
} else if i.rb.src.hangul(next) == 0 {
i.rb.ss.next(i.info)
i.info = i.rb.f.info(i.rb.src, i.p)
i.next = i.rb.f.nextMain
return i.next(i)
}
i.p = next
return i.buf[:decomposeHangul(i.buf[:], i.rb.src.hangul(p))]
}
func nextDone(i *Iter) []byte {
return nil
}
// nextMulti is used for iterating over multi-segment decompositions
// for decomposing normal forms.
func nextMulti(i *Iter) []byte {
j := 0
d := i.multiSeg
// skip first rune
for j = 1; j < len(d) && !utf8.RuneStart(d[j]); j++ {
}
for j < len(d) {
info := i.rb.f.info(input{bytes: d}, j)
if info.BoundaryBefore() {
i.multiSeg = d[j:]
return d[:j]
}
j += int(info.size)
}
// treat last segment as normal decomposition
i.next = i.rb.f.nextMain
return i.next(i)
}
// nextMultiNorm is used for iterating over multi-segment decompositions
// for composing normal forms.
func nextMultiNorm(i *Iter) []byte {
j := 0
d := i.multiSeg
for j < len(d) {
info := i.rb.f.info(input{bytes: d}, j)
if info.BoundaryBefore() {
i.rb.compose()
seg := i.buf[:i.rb.flushCopy(i.buf[:])]
i.rb.insertUnsafe(input{bytes: d}, j, info)
i.multiSeg = d[j+int(info.size):]
return seg
}
i.rb.insertUnsafe(input{bytes: d}, j, info)
j += int(info.size)
}
i.multiSeg = nil
i.next = nextComposed
return doNormComposed(i)
}
// nextDecomposed is the implementation of Next for forms NFD and NFKD.
func nextDecomposed(i *Iter) (next []byte) {
outp := 0
inCopyStart, outCopyStart := i.p, 0
for {
if sz := int(i.info.size); sz <= 1 {
i.rb.ss = 0
p := i.p
i.p++ // ASCII or illegal byte. Either way, advance by 1.
if i.p >= i.rb.nsrc {
i.setDone()
return i.returnSlice(p, i.p)
} else if i.rb.src._byte(i.p) < utf8.RuneSelf {
i.next = i.asciiF
return i.returnSlice(p, i.p)
}
outp++
} else if d := i.info.Decomposition(); d != nil {
// Note: If leading CCC != 0, then len(d) == 2 and last is also non-zero.
// Case 1: there is a leftover to copy. In this case the decomposition
// must begin with a modifier and should always be appended.
// Case 2: no leftover. Simply return d if followed by a ccc == 0 value.
p := outp + len(d)
if outp > 0 {
i.rb.src.copySlice(i.buf[outCopyStart:], inCopyStart, i.p)
// TODO: this condition should not be possible, but we leave it
// in for defensive purposes.
if p > len(i.buf) {
return i.buf[:outp]
}
} else if i.info.multiSegment() {
// outp must be 0 as multi-segment decompositions always
// start a new segment.
if i.multiSeg == nil {
i.multiSeg = d
i.next = nextMulti
return nextMulti(i)
}
// We are in the last segment. Treat as normal decomposition.
d = i.multiSeg
i.multiSeg = nil
p = len(d)
}
prevCC := i.info.tccc
if i.p += sz; i.p >= i.rb.nsrc {
i.setDone()
i.info = Properties{} // Force BoundaryBefore to succeed.
} else {
i.info = i.rb.f.info(i.rb.src, i.p)
}
switch i.rb.ss.next(i.info) {
case ssOverflow:
i.next = nextCGJDecompose
fallthrough
case ssStarter:
if outp > 0 {
copy(i.buf[outp:], d)
return i.buf[:p]
}
return d
}
copy(i.buf[outp:], d)
outp = p
inCopyStart, outCopyStart = i.p, outp
if i.info.ccc < prevCC {
goto doNorm
}
continue
} else if r := i.rb.src.hangul(i.p); r != 0 {
outp = decomposeHangul(i.buf[:], r)
i.p += hangulUTF8Size
inCopyStart, outCopyStart = i.p, outp
if i.p >= i.rb.nsrc {
i.setDone()
break
} else if i.rb.src.hangul(i.p) != 0 {
i.next = nextHangul
return i.buf[:outp]
}
} else {
p := outp + sz
if p > len(i.buf) {
break
}
outp = p
i.p += sz
}
if i.p >= i.rb.nsrc {
i.setDone()
break
}
prevCC := i.info.tccc
i.info = i.rb.f.info(i.rb.src, i.p)
if v := i.rb.ss.next(i.info); v == ssStarter {
break
} else if v == ssOverflow {
i.next = nextCGJDecompose
break
}
if i.info.ccc < prevCC {
goto doNorm
}
}
if outCopyStart == 0 {
return i.returnSlice(inCopyStart, i.p)
} else if inCopyStart < i.p {
i.rb.src.copySlice(i.buf[outCopyStart:], inCopyStart, i.p)
}
return i.buf[:outp]
doNorm:
// Insert what we have decomposed so far in the reorderBuffer.
// As we will only reorder, there will always be enough room.
i.rb.src.copySlice(i.buf[outCopyStart:], inCopyStart, i.p)
i.rb.insertDecomposed(i.buf[0:outp])
return doNormDecomposed(i)
}
func doNormDecomposed(i *Iter) []byte {
for {
i.rb.insertUnsafe(i.rb.src, i.p, i.info)
if i.p += int(i.info.size); i.p >= i.rb.nsrc {
i.setDone()
break
}
i.info = i.rb.f.info(i.rb.src, i.p)
if i.info.ccc == 0 {
break
}
if s := i.rb.ss.next(i.info); s == ssOverflow {
i.next = nextCGJDecompose
break
}
}
// new segment or too many combining characters: exit normalization
return i.buf[:i.rb.flushCopy(i.buf[:])]
}
func nextCGJDecompose(i *Iter) []byte {
i.rb.ss = 0
i.rb.insertCGJ()
i.next = nextDecomposed
i.rb.ss.first(i.info)
buf := doNormDecomposed(i)
return buf
}
// nextComposed is the implementation of Next for forms NFC and NFKC.
func nextComposed(i *Iter) []byte {
outp, startp := 0, i.p
var prevCC uint8
for {
if !i.info.isYesC() {
goto doNorm
}
prevCC = i.info.tccc
sz := int(i.info.size)
if sz == 0 {
sz = 1 // illegal rune: copy byte-by-byte
}
p := outp + sz
if p > len(i.buf) {
break
}
outp = p
i.p += sz
if i.p >= i.rb.nsrc {
i.setDone()
break
} else if i.rb.src._byte(i.p) < utf8.RuneSelf {
i.rb.ss = 0
i.next = i.asciiF
break
}
i.info = i.rb.f.info(i.rb.src, i.p)
if v := i.rb.ss.next(i.info); v == ssStarter {
break
} else if v == ssOverflow {
i.next = nextCGJCompose
break
}
if i.info.ccc < prevCC {
goto doNorm
}
}
return i.returnSlice(startp, i.p)
doNorm:
// reset to start position
i.p = startp
i.info = i.rb.f.info(i.rb.src, i.p)
i.rb.ss.first(i.info)
if i.info.multiSegment() {
d := i.info.Decomposition()
info := i.rb.f.info(input{bytes: d}, 0)
i.rb.insertUnsafe(input{bytes: d}, 0, info)
i.multiSeg = d[int(info.size):]
i.next = nextMultiNorm
return nextMultiNorm(i)
}
i.rb.ss.first(i.info)
i.rb.insertUnsafe(i.rb.src, i.p, i.info)
return doNormComposed(i)
}
func doNormComposed(i *Iter) []byte {
// First rune should already be inserted.
for {
if i.p += int(i.info.size); i.p >= i.rb.nsrc {
i.setDone()
break
}
i.info = i.rb.f.info(i.rb.src, i.p)
if s := i.rb.ss.next(i.info); s == ssStarter {
break
} else if s == ssOverflow {
i.next = nextCGJCompose
break
}
i.rb.insertUnsafe(i.rb.src, i.p, i.info)
}
i.rb.compose()
seg := i.buf[:i.rb.flushCopy(i.buf[:])]
return seg
}
func nextCGJCompose(i *Iter) []byte {
i.rb.ss = 0 // instead of first
i.rb.insertCGJ()
i.next = nextComposed
// Note that we treat any rune with nLeadingNonStarters > 0 as a non-starter,
// even if they are not. This is particularly dubious for U+FF9E and UFF9A.
// If we ever change that, insert a check here.
i.rb.ss.first(i.info)
i.rb.insertUnsafe(i.rb.src, i.p, i.info)
return doNormComposed(i)
}
File diff suppressed because it is too large. Load diff
File diff suppressed because it is too large. Load diff
+125
View File
@@ -0,0 +1,125 @@
// Copyright 2011 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 norm
import "io"
type normWriter struct {
rb reorderBuffer
w io.Writer
buf []byte
}
// Write implements the standard write interface. If the last characters are
// not at a normalization boundary, the bytes will be buffered for the next
// write. The remaining bytes will be written on close.
func (w *normWriter) Write(data []byte) (n int, err error) {
// Process data in pieces to keep w.buf size bounded.
const chunk = 4000
for len(data) > 0 {
// Normalize into w.buf.
m := len(data)
if m > chunk {
m = chunk
}
w.rb.src = inputBytes(data[:m])
w.rb.nsrc = m
w.buf = doAppend(&w.rb, w.buf, 0)
data = data[m:]
n += m
// Write out complete prefix, save remainder.
// Note that lastBoundary looks back at most 31 runes.
i := lastBoundary(&w.rb.f, w.buf)
if i == -1 {
i = 0
}
if i > 0 {
if _, err = w.w.Write(w.buf[:i]); err != nil {
break
}
bn := copy(w.buf, w.buf[i:])
w.buf = w.buf[:bn]
}
}
return n, err
}
// Close forces data that remains in the buffer to be written.
func (w *normWriter) Close() error {
if len(w.buf) > 0 {
_, err := w.w.Write(w.buf)
if err != nil {
return err
}
}
return nil
}
// Writer returns a new writer that implements Write(b)
// by writing f(b) to w. The returned writer may use an
// an internal buffer to maintain state across Write calls.
// Calling its Close method writes any buffered data to w.
func (f Form) Writer(w io.Writer) io.WriteCloser {
wr := &normWriter{rb: reorderBuffer{}, w: w}
wr.rb.init(f, nil)
return wr
}
type normReader struct {
rb reorderBuffer
r io.Reader
inbuf []byte
outbuf []byte
bufStart int
lastBoundary int
err error
}
// Read implements the standard read interface.
func (r *normReader) Read(p []byte) (int, error) {
for {
if r.lastBoundary-r.bufStart > 0 {
n := copy(p, r.outbuf[r.bufStart:r.lastBoundary])
r.bufStart += n
if r.lastBoundary-r.bufStart > 0 {
return n, nil
}
return n, r.err
}
if r.err != nil {
return 0, r.err
}
outn := copy(r.outbuf, r.outbuf[r.lastBoundary:])
r.outbuf = r.outbuf[0:outn]
r.bufStart = 0
n, err := r.r.Read(r.inbuf)
r.rb.src = inputBytes(r.inbuf[0:n])
r.rb.nsrc, r.err = n, err
if n > 0 {
r.outbuf = doAppend(&r.rb, r.outbuf, 0)
}
if err == io.EOF {
r.lastBoundary = len(r.outbuf)
} else {
r.lastBoundary = lastBoundary(&r.rb.f, r.outbuf)
if r.lastBoundary == -1 {
r.lastBoundary = 0
}
}
}
}
// Reader returns a new reader that implements Read
// by reading data from r and returning f(data).
func (f Form) Reader(r io.Reader) io.Reader {
const chunk = 4000
buf := make([]byte, chunk)
rr := &normReader{rb: reorderBuffer{}, r: r, inbuf: buf}
rr.rb.init(f, buf)
return rr
}
File diff suppressed because it is too large. Load diff
File diff suppressed because it is too large. Load diff
+88
View File
@@ -0,0 +1,88 @@
// 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 norm
import (
"unicode/utf8"
"golang.org/x/text/transform"
)
// Reset implements the Reset method of the transform.Transformer interface.
func (Form) Reset() {}
// Transform implements the Transform method of the transform.Transformer
// interface. It may need to write segments of up to MaxSegmentSize at once.
// Users should either catch ErrShortDst and allow dst to grow or have dst be at
// least of size MaxTransformChunkSize to be guaranteed of progress.
func (f Form) Transform(dst, src []byte, atEOF bool) (nDst, nSrc int, err error) {
n := 0
// Cap the maximum number of src bytes to check.
b := src
eof := atEOF
if ns := len(dst); ns < len(b) {
err = transform.ErrShortDst
eof = false
b = b[:ns]
}
i, ok := formTable[f].quickSpan(inputBytes(b), n, len(b), eof)
n += copy(dst[n:], b[n:i])
if !ok {
nDst, nSrc, err = f.transform(dst[n:], src[n:], atEOF)
return nDst + n, nSrc + n, err
}
if n < len(src) && !atEOF {
err = transform.ErrShortSrc
}
return n, n, err
}
func flushTransform(rb *reorderBuffer) bool {
// Write out (must fully fit in dst, or else it is an ErrShortDst).
if len(rb.out) < rb.nrune*utf8.UTFMax {
return false
}
rb.out = rb.out[rb.flushCopy(rb.out):]
return true
}
var errs = []error{nil, transform.ErrShortDst, transform.ErrShortSrc}
// transform implements the transform.Transformer interface. It is only called
// when quickSpan does not pass for a given string.
func (f Form) transform(dst, src []byte, atEOF bool) (nDst, nSrc int, err error) {
// TODO: get rid of reorderBuffer. See CL 23460044.
rb := reorderBuffer{}
rb.init(f, src)
for {
// Load segment into reorder buffer.
rb.setFlusher(dst[nDst:], flushTransform)
end := decomposeSegment(&rb, nSrc, atEOF)
if end < 0 {
return nDst, nSrc, errs[-end]
}
nDst = len(dst) - len(rb.out)
nSrc = end
// Next quickSpan.
end = rb.nsrc
eof := atEOF
if n := nSrc + len(dst) - nDst; n < end {
err = transform.ErrShortDst
end = n
eof = false
}
end, ok := rb.f.quickSpan(rb.src, nSrc, end, eof)
n := copy(dst[nDst:], rb.src.bytes[nSrc:end])
nSrc += n
nDst += n
if ok {
if n < rb.nsrc && !atEOF {
err = transform.ErrShortSrc
}
return nDst, nSrc, err
}
}
}
+54
View File
@@ -0,0 +1,54 @@
// Copyright 2011 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 norm
type valueRange struct {
value uint16 // header: value:stride
lo, hi byte // header: lo:n
}
type sparseBlocks struct {
values []valueRange
offset []uint16
}
var nfcSparse = sparseBlocks{
values: nfcSparseValues[:],
offset: nfcSparseOffset[:],
}
var nfkcSparse = sparseBlocks{
values: nfkcSparseValues[:],
offset: nfkcSparseOffset[:],
}
var (
nfcData = newNfcTrie(0)
nfkcData = newNfkcTrie(0)
)
// lookupValue determines the type of block n and looks up the value for b.
// For n < t.cutoff, the block is a simple lookup table. Otherwise, the block
// is a list of ranges with an accompanying value. Given a matching range r,
// the value for b is by r.value + (b - r.lo) * stride.
func (t *sparseBlocks) lookup(n uint32, b byte) uint16 {
offset := t.offset[n]
header := t.values[offset]
lo := offset + 1
hi := lo + uint16(header.lo)
for lo < hi {
m := lo + (hi-lo)/2
r := t.values[m]
if r.lo <= b && b <= r.hi {
return r.value + uint16(b-r.lo)*header.value
}
if b < r.lo {
hi = m
} else {
lo = m + 1
}
}
return 0
}
+117
View File
@@ -0,0 +1,117 @@
// Copyright 2011 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
// Trie table generator.
// Used by make*tables tools to generate a go file with trie data structures
// for mapping UTF-8 to a 16-bit value. 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 with 16-bit values.
package main
import (
"fmt"
"io"
)
const maxSparseEntries = 16
type normCompacter struct {
sparseBlocks [][]uint64
sparseOffset []uint16
sparseCount int
name string
}
func mostFrequentStride(a []uint64) int {
counts := make(map[int]int)
var v int
for _, x := range a {
if stride := int(x) - v; v != 0 && stride >= 0 {
counts[stride]++
}
v = int(x)
}
var maxs, maxc int
for stride, cnt := range counts {
if cnt > maxc || (cnt == maxc && stride < maxs) {
maxs, maxc = stride, cnt
}
}
return maxs
}
func countSparseEntries(a []uint64) int {
stride := mostFrequentStride(a)
var v, count int
for _, tv := range a {
if int(tv)-v != stride {
if tv != 0 {
count++
}
}
v = int(tv)
}
return count
}
func (c *normCompacter) Size(v []uint64) (sz int, ok bool) {
if n := countSparseEntries(v); n <= maxSparseEntries {
return (n+1)*4 + 2, true
}
return 0, false
}
func (c *normCompacter) Store(v []uint64) uint32 {
h := uint32(len(c.sparseOffset))
c.sparseBlocks = append(c.sparseBlocks, v)
c.sparseOffset = append(c.sparseOffset, uint16(c.sparseCount))
c.sparseCount += countSparseEntries(v) + 1
return h
}
func (c *normCompacter) Handler() string {
return c.name + "Sparse.lookup"
}
func (c *normCompacter) Print(w io.Writer) (retErr error) {
p := func(f string, x ...interface{}) {
if _, err := fmt.Fprintf(w, f, x...); retErr == nil && err != nil {
retErr = err
}
}
ls := len(c.sparseBlocks)
p("// %sSparseOffset: %d entries, %d bytes\n", c.name, ls, ls*2)
p("var %sSparseOffset = %#v\n\n", c.name, c.sparseOffset)
ns := c.sparseCount
p("// %sSparseValues: %d entries, %d bytes\n", c.name, ns, ns*4)
p("var %sSparseValues = [%d]valueRange {", c.name, ns)
for i, b := range c.sparseBlocks {
p("\n// Block %#x, offset %#x", i, c.sparseOffset[i])
var v int
stride := mostFrequentStride(b)
n := countSparseEntries(b)
p("\n{value:%#04x,lo:%#02x},", stride, uint8(n))
for i, nv := range b {
if int(nv)-v != stride {
if v != 0 {
p(",hi:%#02x},", 0x80+i-1)
}
if nv != 0 {
p("\n{value:%#04x,lo:%#02x", nv, 0x80+i)
}
}
v = int(nv)
}
if v != 0 {
p(",hi:%#02x},", 0x80+len(b)-1)
}
}
p("\n}\n\n")
return
}