OSDN Git Service

new repo
[bytom/vapor.git] / vendor / golang.org / x / text / unicode / norm / iter.go
1 // Copyright 2011 The Go Authors. All rights reserved.
2 // Use of this source code is governed by a BSD-style
3 // license that can be found in the LICENSE file.
4
5 package norm
6
7 import (
8         "fmt"
9         "unicode/utf8"
10 )
11
12 // MaxSegmentSize is the maximum size of a byte buffer needed to consider any
13 // sequence of starter and non-starter runes for the purpose of normalization.
14 const MaxSegmentSize = maxByteBufferSize
15
16 // An Iter iterates over a string or byte slice, while normalizing it
17 // to a given Form.
18 type Iter struct {
19         rb     reorderBuffer
20         buf    [maxByteBufferSize]byte
21         info   Properties // first character saved from previous iteration
22         next   iterFunc   // implementation of next depends on form
23         asciiF iterFunc
24
25         p        int    // current position in input source
26         multiSeg []byte // remainder of multi-segment decomposition
27 }
28
29 type iterFunc func(*Iter) []byte
30
31 // Init initializes i to iterate over src after normalizing it to Form f.
32 func (i *Iter) Init(f Form, src []byte) {
33         i.p = 0
34         if len(src) == 0 {
35                 i.setDone()
36                 i.rb.nsrc = 0
37                 return
38         }
39         i.multiSeg = nil
40         i.rb.init(f, src)
41         i.next = i.rb.f.nextMain
42         i.asciiF = nextASCIIBytes
43         i.info = i.rb.f.info(i.rb.src, i.p)
44         i.rb.ss.first(i.info)
45 }
46
47 // InitString initializes i to iterate over src after normalizing it to Form f.
48 func (i *Iter) InitString(f Form, src string) {
49         i.p = 0
50         if len(src) == 0 {
51                 i.setDone()
52                 i.rb.nsrc = 0
53                 return
54         }
55         i.multiSeg = nil
56         i.rb.initString(f, src)
57         i.next = i.rb.f.nextMain
58         i.asciiF = nextASCIIString
59         i.info = i.rb.f.info(i.rb.src, i.p)
60         i.rb.ss.first(i.info)
61 }
62
63 // Seek sets the segment to be returned by the next call to Next to start
64 // at position p.  It is the responsibility of the caller to set p to the
65 // start of a segment.
66 func (i *Iter) Seek(offset int64, whence int) (int64, error) {
67         var abs int64
68         switch whence {
69         case 0:
70                 abs = offset
71         case 1:
72                 abs = int64(i.p) + offset
73         case 2:
74                 abs = int64(i.rb.nsrc) + offset
75         default:
76                 return 0, fmt.Errorf("norm: invalid whence")
77         }
78         if abs < 0 {
79                 return 0, fmt.Errorf("norm: negative position")
80         }
81         if int(abs) >= i.rb.nsrc {
82                 i.setDone()
83                 return int64(i.p), nil
84         }
85         i.p = int(abs)
86         i.multiSeg = nil
87         i.next = i.rb.f.nextMain
88         i.info = i.rb.f.info(i.rb.src, i.p)
89         i.rb.ss.first(i.info)
90         return abs, nil
91 }
92
93 // returnSlice returns a slice of the underlying input type as a byte slice.
94 // If the underlying is of type []byte, it will simply return a slice.
95 // If the underlying is of type string, it will copy the slice to the buffer
96 // and return that.
97 func (i *Iter) returnSlice(a, b int) []byte {
98         if i.rb.src.bytes == nil {
99                 return i.buf[:copy(i.buf[:], i.rb.src.str[a:b])]
100         }
101         return i.rb.src.bytes[a:b]
102 }
103
104 // Pos returns the byte position at which the next call to Next will commence processing.
105 func (i *Iter) Pos() int {
106         return i.p
107 }
108
109 func (i *Iter) setDone() {
110         i.next = nextDone
111         i.p = i.rb.nsrc
112 }
113
114 // Done returns true if there is no more input to process.
115 func (i *Iter) Done() bool {
116         return i.p >= i.rb.nsrc
117 }
118
119 // Next returns f(i.input[i.Pos():n]), where n is a boundary of i.input.
120 // For any input a and b for which f(a) == f(b), subsequent calls
121 // to Next will return the same segments.
122 // Modifying runes are grouped together with the preceding starter, if such a starter exists.
123 // Although not guaranteed, n will typically be the smallest possible n.
124 func (i *Iter) Next() []byte {
125         return i.next(i)
126 }
127
128 func nextASCIIBytes(i *Iter) []byte {
129         p := i.p + 1
130         if p >= i.rb.nsrc {
131                 i.setDone()
132                 return i.rb.src.bytes[i.p:p]
133         }
134         if i.rb.src.bytes[p] < utf8.RuneSelf {
135                 p0 := i.p
136                 i.p = p
137                 return i.rb.src.bytes[p0:p]
138         }
139         i.info = i.rb.f.info(i.rb.src, i.p)
140         i.next = i.rb.f.nextMain
141         return i.next(i)
142 }
143
144 func nextASCIIString(i *Iter) []byte {
145         p := i.p + 1
146         if p >= i.rb.nsrc {
147                 i.buf[0] = i.rb.src.str[i.p]
148                 i.setDone()
149                 return i.buf[:1]
150         }
151         if i.rb.src.str[p] < utf8.RuneSelf {
152                 i.buf[0] = i.rb.src.str[i.p]
153                 i.p = p
154                 return i.buf[:1]
155         }
156         i.info = i.rb.f.info(i.rb.src, i.p)
157         i.next = i.rb.f.nextMain
158         return i.next(i)
159 }
160
161 func nextHangul(i *Iter) []byte {
162         p := i.p
163         next := p + hangulUTF8Size
164         if next >= i.rb.nsrc {
165                 i.setDone()
166         } else if i.rb.src.hangul(next) == 0 {
167                 i.rb.ss.next(i.info)
168                 i.info = i.rb.f.info(i.rb.src, i.p)
169                 i.next = i.rb.f.nextMain
170                 return i.next(i)
171         }
172         i.p = next
173         return i.buf[:decomposeHangul(i.buf[:], i.rb.src.hangul(p))]
174 }
175
176 func nextDone(i *Iter) []byte {
177         return nil
178 }
179
180 // nextMulti is used for iterating over multi-segment decompositions
181 // for decomposing normal forms.
182 func nextMulti(i *Iter) []byte {
183         j := 0
184         d := i.multiSeg
185         // skip first rune
186         for j = 1; j < len(d) && !utf8.RuneStart(d[j]); j++ {
187         }
188         for j < len(d) {
189                 info := i.rb.f.info(input{bytes: d}, j)
190                 if info.BoundaryBefore() {
191                         i.multiSeg = d[j:]
192                         return d[:j]
193                 }
194                 j += int(info.size)
195         }
196         // treat last segment as normal decomposition
197         i.next = i.rb.f.nextMain
198         return i.next(i)
199 }
200
201 // nextMultiNorm is used for iterating over multi-segment decompositions
202 // for composing normal forms.
203 func nextMultiNorm(i *Iter) []byte {
204         j := 0
205         d := i.multiSeg
206         for j < len(d) {
207                 info := i.rb.f.info(input{bytes: d}, j)
208                 if info.BoundaryBefore() {
209                         i.rb.compose()
210                         seg := i.buf[:i.rb.flushCopy(i.buf[:])]
211                         i.rb.insertUnsafe(input{bytes: d}, j, info)
212                         i.multiSeg = d[j+int(info.size):]
213                         return seg
214                 }
215                 i.rb.insertUnsafe(input{bytes: d}, j, info)
216                 j += int(info.size)
217         }
218         i.multiSeg = nil
219         i.next = nextComposed
220         return doNormComposed(i)
221 }
222
223 // nextDecomposed is the implementation of Next for forms NFD and NFKD.
224 func nextDecomposed(i *Iter) (next []byte) {
225         outp := 0
226         inCopyStart, outCopyStart := i.p, 0
227         for {
228                 if sz := int(i.info.size); sz <= 1 {
229                         i.rb.ss = 0
230                         p := i.p
231                         i.p++ // ASCII or illegal byte.  Either way, advance by 1.
232                         if i.p >= i.rb.nsrc {
233                                 i.setDone()
234                                 return i.returnSlice(p, i.p)
235                         } else if i.rb.src._byte(i.p) < utf8.RuneSelf {
236                                 i.next = i.asciiF
237                                 return i.returnSlice(p, i.p)
238                         }
239                         outp++
240                 } else if d := i.info.Decomposition(); d != nil {
241                         // Note: If leading CCC != 0, then len(d) == 2 and last is also non-zero.
242                         // Case 1: there is a leftover to copy.  In this case the decomposition
243                         // must begin with a modifier and should always be appended.
244                         // Case 2: no leftover. Simply return d if followed by a ccc == 0 value.
245                         p := outp + len(d)
246                         if outp > 0 {
247                                 i.rb.src.copySlice(i.buf[outCopyStart:], inCopyStart, i.p)
248                                 // TODO: this condition should not be possible, but we leave it
249                                 // in for defensive purposes.
250                                 if p > len(i.buf) {
251                                         return i.buf[:outp]
252                                 }
253                         } else if i.info.multiSegment() {
254                                 // outp must be 0 as multi-segment decompositions always
255                                 // start a new segment.
256                                 if i.multiSeg == nil {
257                                         i.multiSeg = d
258                                         i.next = nextMulti
259                                         return nextMulti(i)
260                                 }
261                                 // We are in the last segment.  Treat as normal decomposition.
262                                 d = i.multiSeg
263                                 i.multiSeg = nil
264                                 p = len(d)
265                         }
266                         prevCC := i.info.tccc
267                         if i.p += sz; i.p >= i.rb.nsrc {
268                                 i.setDone()
269                                 i.info = Properties{} // Force BoundaryBefore to succeed.
270                         } else {
271                                 i.info = i.rb.f.info(i.rb.src, i.p)
272                         }
273                         switch i.rb.ss.next(i.info) {
274                         case ssOverflow:
275                                 i.next = nextCGJDecompose
276                                 fallthrough
277                         case ssStarter:
278                                 if outp > 0 {
279                                         copy(i.buf[outp:], d)
280                                         return i.buf[:p]
281                                 }
282                                 return d
283                         }
284                         copy(i.buf[outp:], d)
285                         outp = p
286                         inCopyStart, outCopyStart = i.p, outp
287                         if i.info.ccc < prevCC {
288                                 goto doNorm
289                         }
290                         continue
291                 } else if r := i.rb.src.hangul(i.p); r != 0 {
292                         outp = decomposeHangul(i.buf[:], r)
293                         i.p += hangulUTF8Size
294                         inCopyStart, outCopyStart = i.p, outp
295                         if i.p >= i.rb.nsrc {
296                                 i.setDone()
297                                 break
298                         } else if i.rb.src.hangul(i.p) != 0 {
299                                 i.next = nextHangul
300                                 return i.buf[:outp]
301                         }
302                 } else {
303                         p := outp + sz
304                         if p > len(i.buf) {
305                                 break
306                         }
307                         outp = p
308                         i.p += sz
309                 }
310                 if i.p >= i.rb.nsrc {
311                         i.setDone()
312                         break
313                 }
314                 prevCC := i.info.tccc
315                 i.info = i.rb.f.info(i.rb.src, i.p)
316                 if v := i.rb.ss.next(i.info); v == ssStarter {
317                         break
318                 } else if v == ssOverflow {
319                         i.next = nextCGJDecompose
320                         break
321                 }
322                 if i.info.ccc < prevCC {
323                         goto doNorm
324                 }
325         }
326         if outCopyStart == 0 {
327                 return i.returnSlice(inCopyStart, i.p)
328         } else if inCopyStart < i.p {
329                 i.rb.src.copySlice(i.buf[outCopyStart:], inCopyStart, i.p)
330         }
331         return i.buf[:outp]
332 doNorm:
333         // Insert what we have decomposed so far in the reorderBuffer.
334         // As we will only reorder, there will always be enough room.
335         i.rb.src.copySlice(i.buf[outCopyStart:], inCopyStart, i.p)
336         i.rb.insertDecomposed(i.buf[0:outp])
337         return doNormDecomposed(i)
338 }
339
340 func doNormDecomposed(i *Iter) []byte {
341         for {
342                 i.rb.insertUnsafe(i.rb.src, i.p, i.info)
343                 if i.p += int(i.info.size); i.p >= i.rb.nsrc {
344                         i.setDone()
345                         break
346                 }
347                 i.info = i.rb.f.info(i.rb.src, i.p)
348                 if i.info.ccc == 0 {
349                         break
350                 }
351                 if s := i.rb.ss.next(i.info); s == ssOverflow {
352                         i.next = nextCGJDecompose
353                         break
354                 }
355         }
356         // new segment or too many combining characters: exit normalization
357         return i.buf[:i.rb.flushCopy(i.buf[:])]
358 }
359
360 func nextCGJDecompose(i *Iter) []byte {
361         i.rb.ss = 0
362         i.rb.insertCGJ()
363         i.next = nextDecomposed
364         i.rb.ss.first(i.info)
365         buf := doNormDecomposed(i)
366         return buf
367 }
368
369 // nextComposed is the implementation of Next for forms NFC and NFKC.
370 func nextComposed(i *Iter) []byte {
371         outp, startp := 0, i.p
372         var prevCC uint8
373         for {
374                 if !i.info.isYesC() {
375                         goto doNorm
376                 }
377                 prevCC = i.info.tccc
378                 sz := int(i.info.size)
379                 if sz == 0 {
380                         sz = 1 // illegal rune: copy byte-by-byte
381                 }
382                 p := outp + sz
383                 if p > len(i.buf) {
384                         break
385                 }
386                 outp = p
387                 i.p += sz
388                 if i.p >= i.rb.nsrc {
389                         i.setDone()
390                         break
391                 } else if i.rb.src._byte(i.p) < utf8.RuneSelf {
392                         i.rb.ss = 0
393                         i.next = i.asciiF
394                         break
395                 }
396                 i.info = i.rb.f.info(i.rb.src, i.p)
397                 if v := i.rb.ss.next(i.info); v == ssStarter {
398                         break
399                 } else if v == ssOverflow {
400                         i.next = nextCGJCompose
401                         break
402                 }
403                 if i.info.ccc < prevCC {
404                         goto doNorm
405                 }
406         }
407         return i.returnSlice(startp, i.p)
408 doNorm:
409         // reset to start position
410         i.p = startp
411         i.info = i.rb.f.info(i.rb.src, i.p)
412         i.rb.ss.first(i.info)
413         if i.info.multiSegment() {
414                 d := i.info.Decomposition()
415                 info := i.rb.f.info(input{bytes: d}, 0)
416                 i.rb.insertUnsafe(input{bytes: d}, 0, info)
417                 i.multiSeg = d[int(info.size):]
418                 i.next = nextMultiNorm
419                 return nextMultiNorm(i)
420         }
421         i.rb.ss.first(i.info)
422         i.rb.insertUnsafe(i.rb.src, i.p, i.info)
423         return doNormComposed(i)
424 }
425
426 func doNormComposed(i *Iter) []byte {
427         // First rune should already be inserted.
428         for {
429                 if i.p += int(i.info.size); i.p >= i.rb.nsrc {
430                         i.setDone()
431                         break
432                 }
433                 i.info = i.rb.f.info(i.rb.src, i.p)
434                 if s := i.rb.ss.next(i.info); s == ssStarter {
435                         break
436                 } else if s == ssOverflow {
437                         i.next = nextCGJCompose
438                         break
439                 }
440                 i.rb.insertUnsafe(i.rb.src, i.p, i.info)
441         }
442         i.rb.compose()
443         seg := i.buf[:i.rb.flushCopy(i.buf[:])]
444         return seg
445 }
446
447 func nextCGJCompose(i *Iter) []byte {
448         i.rb.ss = 0 // instead of first
449         i.rb.insertCGJ()
450         i.next = nextComposed
451         // Note that we treat any rune with nLeadingNonStarters > 0 as a non-starter,
452         // even if they are not. This is particularly dubious for U+FF9E and UFF9A.
453         // If we ever change that, insert a check here.
454         i.rb.ss.first(i.info)
455         i.rb.insertUnsafe(i.rb.src, i.p, i.info)
456         return doNormComposed(i)
457 }