Matthias Andreas Benkard | 832a54e | 2019-01-29 09:27:38 +0100 | [diff] [blame] | 1 | // Copyright 2015 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 runes |
| 6 | |
| 7 | import ( |
| 8 | "unicode/utf8" |
| 9 | |
| 10 | "golang.org/x/text/transform" |
| 11 | ) |
| 12 | |
| 13 | // Note: below we pass invalid UTF-8 to the tIn and tNotIn transformers as is. |
| 14 | // This is done for various reasons: |
| 15 | // - To retain the semantics of the Nop transformer: if input is passed to a Nop |
| 16 | // one would expect it to be unchanged. |
| 17 | // - It would be very expensive to pass a converted RuneError to a transformer: |
| 18 | // a transformer might need more source bytes after RuneError, meaning that |
| 19 | // the only way to pass it safely is to create a new buffer and manage the |
| 20 | // intermingling of RuneErrors and normal input. |
| 21 | // - Many transformers leave ill-formed UTF-8 as is, so this is not |
| 22 | // inconsistent. Generally ill-formed UTF-8 is only replaced if it is a |
| 23 | // logical consequence of the operation (as for Map) or if it otherwise would |
| 24 | // pose security concerns (as for Remove). |
| 25 | // - An alternative would be to return an error on ill-formed UTF-8, but this |
| 26 | // would be inconsistent with other operations. |
| 27 | |
| 28 | // If returns a transformer that applies tIn to consecutive runes for which |
| 29 | // s.Contains(r) and tNotIn to consecutive runes for which !s.Contains(r). Reset |
| 30 | // is called on tIn and tNotIn at the start of each run. A Nop transformer will |
| 31 | // substitute a nil value passed to tIn or tNotIn. Invalid UTF-8 is translated |
| 32 | // to RuneError to determine which transformer to apply, but is passed as is to |
| 33 | // the respective transformer. |
| 34 | func If(s Set, tIn, tNotIn transform.Transformer) Transformer { |
| 35 | if tIn == nil && tNotIn == nil { |
| 36 | return Transformer{transform.Nop} |
| 37 | } |
| 38 | if tIn == nil { |
| 39 | tIn = transform.Nop |
| 40 | } |
| 41 | if tNotIn == nil { |
| 42 | tNotIn = transform.Nop |
| 43 | } |
| 44 | sIn, ok := tIn.(transform.SpanningTransformer) |
| 45 | if !ok { |
| 46 | sIn = dummySpan{tIn} |
| 47 | } |
| 48 | sNotIn, ok := tNotIn.(transform.SpanningTransformer) |
| 49 | if !ok { |
| 50 | sNotIn = dummySpan{tNotIn} |
| 51 | } |
| 52 | |
| 53 | a := &cond{ |
| 54 | tIn: sIn, |
| 55 | tNotIn: sNotIn, |
| 56 | f: s.Contains, |
| 57 | } |
| 58 | a.Reset() |
| 59 | return Transformer{a} |
| 60 | } |
| 61 | |
| 62 | type dummySpan struct{ transform.Transformer } |
| 63 | |
| 64 | func (d dummySpan) Span(src []byte, atEOF bool) (n int, err error) { |
| 65 | return 0, transform.ErrEndOfSpan |
| 66 | } |
| 67 | |
| 68 | type cond struct { |
| 69 | tIn, tNotIn transform.SpanningTransformer |
| 70 | f func(rune) bool |
| 71 | check func(rune) bool // current check to perform |
| 72 | t transform.SpanningTransformer // current transformer to use |
| 73 | } |
| 74 | |
| 75 | // Reset implements transform.Transformer. |
| 76 | func (t *cond) Reset() { |
| 77 | t.check = t.is |
| 78 | t.t = t.tIn |
| 79 | t.t.Reset() // notIn will be reset on first usage. |
| 80 | } |
| 81 | |
| 82 | func (t *cond) is(r rune) bool { |
| 83 | if t.f(r) { |
| 84 | return true |
| 85 | } |
| 86 | t.check = t.isNot |
| 87 | t.t = t.tNotIn |
| 88 | t.tNotIn.Reset() |
| 89 | return false |
| 90 | } |
| 91 | |
| 92 | func (t *cond) isNot(r rune) bool { |
| 93 | if !t.f(r) { |
| 94 | return true |
| 95 | } |
| 96 | t.check = t.is |
| 97 | t.t = t.tIn |
| 98 | t.tIn.Reset() |
| 99 | return false |
| 100 | } |
| 101 | |
| 102 | // This implementation of Span doesn't help all too much, but it needs to be |
| 103 | // there to satisfy this package's Transformer interface. |
| 104 | // TODO: there are certainly room for improvements, though. For example, if |
| 105 | // t.t == transform.Nop (which will a common occurrence) it will save a bundle |
| 106 | // to special-case that loop. |
| 107 | func (t *cond) Span(src []byte, atEOF bool) (n int, err error) { |
| 108 | p := 0 |
| 109 | for n < len(src) && err == nil { |
| 110 | // Don't process too much at a time as the Spanner that will be |
| 111 | // called on this block may terminate early. |
| 112 | const maxChunk = 4096 |
| 113 | max := len(src) |
| 114 | if v := n + maxChunk; v < max { |
| 115 | max = v |
| 116 | } |
| 117 | atEnd := false |
| 118 | size := 0 |
| 119 | current := t.t |
| 120 | for ; p < max; p += size { |
| 121 | r := rune(src[p]) |
| 122 | if r < utf8.RuneSelf { |
| 123 | size = 1 |
| 124 | } else if r, size = utf8.DecodeRune(src[p:]); size == 1 { |
| 125 | if !atEOF && !utf8.FullRune(src[p:]) { |
| 126 | err = transform.ErrShortSrc |
| 127 | break |
| 128 | } |
| 129 | } |
| 130 | if !t.check(r) { |
| 131 | // The next rune will be the start of a new run. |
| 132 | atEnd = true |
| 133 | break |
| 134 | } |
| 135 | } |
| 136 | n2, err2 := current.Span(src[n:p], atEnd || (atEOF && p == len(src))) |
| 137 | n += n2 |
| 138 | if err2 != nil { |
| 139 | return n, err2 |
| 140 | } |
| 141 | // At this point either err != nil or t.check will pass for the rune at p. |
| 142 | p = n + size |
| 143 | } |
| 144 | return n, err |
| 145 | } |
| 146 | |
| 147 | func (t *cond) Transform(dst, src []byte, atEOF bool) (nDst, nSrc int, err error) { |
| 148 | p := 0 |
| 149 | for nSrc < len(src) && err == nil { |
| 150 | // Don't process too much at a time, as the work might be wasted if the |
| 151 | // destination buffer isn't large enough to hold the result or a |
| 152 | // transform returns an error early. |
| 153 | const maxChunk = 4096 |
| 154 | max := len(src) |
| 155 | if n := nSrc + maxChunk; n < len(src) { |
| 156 | max = n |
| 157 | } |
| 158 | atEnd := false |
| 159 | size := 0 |
| 160 | current := t.t |
| 161 | for ; p < max; p += size { |
| 162 | r := rune(src[p]) |
| 163 | if r < utf8.RuneSelf { |
| 164 | size = 1 |
| 165 | } else if r, size = utf8.DecodeRune(src[p:]); size == 1 { |
| 166 | if !atEOF && !utf8.FullRune(src[p:]) { |
| 167 | err = transform.ErrShortSrc |
| 168 | break |
| 169 | } |
| 170 | } |
| 171 | if !t.check(r) { |
| 172 | // The next rune will be the start of a new run. |
| 173 | atEnd = true |
| 174 | break |
| 175 | } |
| 176 | } |
| 177 | nDst2, nSrc2, err2 := current.Transform(dst[nDst:], src[nSrc:p], atEnd || (atEOF && p == len(src))) |
| 178 | nDst += nDst2 |
| 179 | nSrc += nSrc2 |
| 180 | if err2 != nil { |
| 181 | return nDst, nSrc, err2 |
| 182 | } |
| 183 | // At this point either err != nil or t.check will pass for the rune at p. |
| 184 | p = nSrc + size |
| 185 | } |
| 186 | return nDst, nSrc, err |
| 187 | } |