|
| 1 | +package app |
| 2 | + |
| 3 | +import ( |
| 4 | + "bufio" |
| 5 | + "bytes" |
| 6 | + "fmt" |
| 7 | + "io" |
| 8 | + "os" |
| 9 | + "sort" |
| 10 | + "strings" |
| 11 | + "sync" |
| 12 | + "unicode" |
| 13 | + "unicode/utf8" |
| 14 | + |
| 15 | + "github.com/tidwall/collate" |
| 16 | +) |
| 17 | + |
| 18 | +const filePerm = 0644 |
| 19 | + |
| 20 | +// splitWordsUnicode splits data into words, using Unicode Letter character class. |
| 21 | +// It works similar to the regular expression "[^\p{L}]+". This is what was used |
| 22 | +// in the original code. Unicode function has slight overhead, but handles UTF-8 |
| 23 | +// correctly. |
| 24 | +// |
| 25 | +// Rust and Python versions split text according to "[\W\d]+" - anything that is |
| 26 | +// not a word or a digit. WTF? |
| 27 | +func splitWordsUnicode(data []byte, atEOF bool) (advance int, token []byte, err error) { |
| 28 | + var start int |
| 29 | + var r rune |
| 30 | + for width := 0; start < len(data); start += width { |
| 31 | + if r, width = utf8.DecodeRune(data[start:]); unicode.IsLetter(r) { |
| 32 | + break |
| 33 | + } |
| 34 | + } |
| 35 | + |
| 36 | + for width, i := 0, start; i < len(data); i += width { |
| 37 | + if r, width = utf8.DecodeRune(data[i:]); !unicode.IsLetter(r) { |
| 38 | + return i + width, data[start:i], nil |
| 39 | + } |
| 40 | + } |
| 41 | + |
| 42 | + if atEOF && len(data) > start { |
| 43 | + return len(data), data[start:], nil |
| 44 | + } |
| 45 | + |
| 46 | + return start, nil, nil |
| 47 | +} |
| 48 | + |
| 49 | +// splitWords splits data into words similar to the "[\W\d]+" regular expression. |
| 50 | +func splitWords(data []byte, atEOF bool) (advance int, token []byte, err error) { |
| 51 | + var start int |
| 52 | + var r rune |
| 53 | + for width := 0; start < len(data); start += width { |
| 54 | + if r, width = utf8.DecodeRune(data[start:]); isLatin(r) { |
| 55 | + break |
| 56 | + } |
| 57 | + } |
| 58 | + |
| 59 | + for width, i := 0, start; i < len(data); i += width { |
| 60 | + if r, width = utf8.DecodeRune(data[i:]); !isLatin(r) { |
| 61 | + return i + width, data[start:i], nil |
| 62 | + } |
| 63 | + } |
| 64 | + |
| 65 | + if atEOF && len(data) > start { |
| 66 | + return len(data), data[start:], nil |
| 67 | + } |
| 68 | + |
| 69 | + return start, nil, nil |
| 70 | +} |
| 71 | + |
| 72 | +func isLatin(r rune) bool { |
| 73 | + if r >= 0x80 || r == 0x00 { |
| 74 | + return false |
| 75 | + } |
| 76 | + |
| 77 | + return ('a' <= r && r <= 'z') || ('A' <= r && r <= 'Z') |
| 78 | +} |
| 79 | + |
| 80 | +func extract(src, dst, lang string, sortResults bool, sem <-chan empty, wg *sync.WaitGroup) { |
| 81 | + defer func() { |
| 82 | + <-sem |
| 83 | + wg.Done() |
| 84 | + }() |
| 85 | + |
| 86 | + fd, err := os.Open(src) |
| 87 | + if err != nil { |
| 88 | + _, _ = fmt.Fprintf(os.Stderr, `extract: opening source file "%s" for reading: %s`, src, err) |
| 89 | + return |
| 90 | + } |
| 91 | + defer fd.Close() |
| 92 | + |
| 93 | + // One of the possible optimisations here is to split file in chunks and process |
| 94 | + // each chunk individually. |
| 95 | + words, err := collectWords(fd) |
| 96 | + if err != nil { |
| 97 | + _, _ = fmt.Fprintf(os.Stderr, `extract: reading input "%s": %s`, src, err) |
| 98 | + return |
| 99 | + } |
| 100 | + |
| 101 | + if sortResults { |
| 102 | + less := collate.IndexString(lang) |
| 103 | + sort.Slice(words, func(i, j int) bool { |
| 104 | + return less(words[i], words[j]) |
| 105 | + }) |
| 106 | + } |
| 107 | + |
| 108 | + wd, err := os.OpenFile(dst, os.O_WRONLY|os.O_CREATE|os.O_TRUNC, filePerm) |
| 109 | + if err != nil { |
| 110 | + _, _ = fmt.Fprintf(os.Stderr, `extract: opening destination file "%s" for writing: %s`, src, err) |
| 111 | + return |
| 112 | + } |
| 113 | + defer fd.Close() |
| 114 | + |
| 115 | + // Writing word by word can result in too many writes, hence, it is slow. |
| 116 | + // Let's add some steroids ... |
| 117 | + wr := bufio.NewWriter(wd) |
| 118 | + |
| 119 | + if err = writeResults(wr, words); err != nil { |
| 120 | + _, _ = fmt.Fprintf(os.Stderr, `extract: writing results "%s": %s`, dst, err) |
| 121 | + return |
| 122 | + } |
| 123 | + if err = wr.Flush(); err != nil { |
| 124 | + _, _ = fmt.Fprintf(os.Stderr, `extract: writing results "%s": %s`, dst, err) |
| 125 | + return |
| 126 | + } |
| 127 | + |
| 128 | + // _, _ = fmt.Fprintf(os.Stdout, "Saved %s\n", dst) |
| 129 | +} |
| 130 | +func collectWords(r io.Reader) ([]string, error) { |
| 131 | + scanner := bufio.NewScanner(r) |
| 132 | + |
| 133 | + scanner.Split(splitWordsUnicode) |
| 134 | + |
| 135 | + dict := make(map[string]empty) |
| 136 | + for scanner.Scan() { |
| 137 | + word := strings.ToLower(scanner.Text()) |
| 138 | + if _, ok := dict[word]; ok { |
| 139 | + continue // duplicate detected |
| 140 | + } |
| 141 | + |
| 142 | + dict[word] = empty{} |
| 143 | + |
| 144 | + // Theoretically, if sorting is not needed, we can write right here and |
| 145 | + // skip words slice preparation below. |
| 146 | + } |
| 147 | + if err := scanner.Err(); err != nil { |
| 148 | + return nil, err |
| 149 | + } |
| 150 | + |
| 151 | + // This is expensive ... |
| 152 | + words := make([]string, len(dict)) |
| 153 | + |
| 154 | + var i int |
| 155 | + for w := range dict { |
| 156 | + words[i] = w |
| 157 | + i++ |
| 158 | + |
| 159 | + delete(dict, w) |
| 160 | + } |
| 161 | + |
| 162 | + return words, nil |
| 163 | +} |
| 164 | + |
| 165 | +func writeResults(w io.Writer, words []string) error { |
| 166 | + // This is to preallocate memory once for "string => []byte + \n" conversion |
| 167 | + // and reuse it on every iteration. |
| 168 | + var buf bytes.Buffer |
| 169 | + for _, word := range words { |
| 170 | + buf.WriteString(word) |
| 171 | + buf.WriteRune('\n') |
| 172 | + |
| 173 | + if _, err := buf.WriteTo(w); err != nil { |
| 174 | + return err |
| 175 | + } |
| 176 | + |
| 177 | + buf.Reset() |
| 178 | + } |
| 179 | + |
| 180 | + return nil |
| 181 | +} |
| 182 | + |
| 183 | +func ExtractUniqueWords(content string, lang string) ([]string, error) { |
| 184 | + r := strings.NewReader(content) |
| 185 | + words, err := collectWords(r) |
| 186 | + if err != nil { |
| 187 | + _, _ = fmt.Fprintf(os.Stderr, `collectWords error: %s`, err) |
| 188 | + return nil, err |
| 189 | + } |
| 190 | + less := collate.IndexString(lang) |
| 191 | + sort.Slice(words, func(i, j int) bool { |
| 192 | + return less(words[i], words[j]) |
| 193 | + }) |
| 194 | + return words, nil |
| 195 | +} |
0 commit comments