|
| 1 | +#include <stdbool.h> |
| 2 | +#include <stdlib.h> |
| 3 | +#include <string.h> |
| 4 | +#include <wctype.h> |
| 5 | + |
| 6 | +#include "parser.h" |
| 7 | + |
| 8 | +#define MAX_HEREDOCS 10 |
| 9 | +#define DEL_SPACE 512 |
| 10 | + |
| 11 | +typedef struct { |
| 12 | + bool in_heredoc; |
| 13 | + bool stripping_heredoc; |
| 14 | + unsigned heredoc_count; |
| 15 | + char *heredocs[MAX_HEREDOCS]; |
| 16 | +} scanner_state; |
| 17 | + |
| 18 | +enum TokenType { |
| 19 | + HEREDOC_MARKER, |
| 20 | + HEREDOC_LINE, |
| 21 | + HEREDOC_END, |
| 22 | + HEREDOC_NL, |
| 23 | + ERROR_SENTINEL, |
| 24 | +}; |
| 25 | + |
| 26 | +void *tree_sitter_dockerfile_external_scanner_create() { |
| 27 | + scanner_state *state = malloc(sizeof(scanner_state)); |
| 28 | + memset(state, 0, sizeof(scanner_state)); |
| 29 | + return state; |
| 30 | +} |
| 31 | + |
| 32 | +void tree_sitter_dockerfile_external_scanner_destroy(void *payload) { |
| 33 | + if (!payload) |
| 34 | + return; |
| 35 | + |
| 36 | + scanner_state *state = payload; |
| 37 | + for (unsigned i = 0; i < MAX_HEREDOCS; i++) { |
| 38 | + if (state->heredocs[i]) { |
| 39 | + free(state->heredocs[i]); |
| 40 | + } |
| 41 | + } |
| 42 | + |
| 43 | + free(state); |
| 44 | +} |
| 45 | + |
| 46 | +unsigned tree_sitter_dockerfile_external_scanner_serialize(void *payload, |
| 47 | + char *buffer) { |
| 48 | + scanner_state *state = payload; |
| 49 | + |
| 50 | + unsigned pos = 0; |
| 51 | + buffer[pos++] = state->in_heredoc; |
| 52 | + buffer[pos++] = state->stripping_heredoc; |
| 53 | + |
| 54 | + for (unsigned i = 0; i < state->heredoc_count; i++) { |
| 55 | + // Add the ending null byte to the length since we'll have to copy it as |
| 56 | + // well. |
| 57 | + unsigned len = strlen(state->heredocs[i]) + 1; |
| 58 | + |
| 59 | + // If we run out of space, just drop the heredocs that don't fit. |
| 60 | + // We need at least len + 1 bytes space since we'll copy len bytes below |
| 61 | + // and later add a null byte at the end. |
| 62 | + if (pos + len + 1 > TREE_SITTER_SERIALIZATION_BUFFER_SIZE) { |
| 63 | + break; |
| 64 | + } |
| 65 | + |
| 66 | + memcpy(&buffer[pos], state->heredocs[i], len); |
| 67 | + pos += len; |
| 68 | + } |
| 69 | + |
| 70 | + // Add a null byte at the end to make it easy to detect. |
| 71 | + buffer[pos++] = 0; |
| 72 | + return pos; |
| 73 | +} |
| 74 | + |
| 75 | +void tree_sitter_dockerfile_external_scanner_deserialize(void *payload, |
| 76 | + const char *buffer, |
| 77 | + unsigned length) { |
| 78 | + scanner_state *state = payload; |
| 79 | + // Free all current heredocs to avoid leaking memory when we overwrite the |
| 80 | + // array later. |
| 81 | + for (unsigned i = 0; i < state->heredoc_count; i++) { |
| 82 | + free(state->heredocs[i]); |
| 83 | + state->heredocs[i] = NULL; |
| 84 | + } |
| 85 | + |
| 86 | + if (length == 0) { |
| 87 | + state->in_heredoc = false; |
| 88 | + state->stripping_heredoc = false; |
| 89 | + state->heredoc_count = 0; |
| 90 | + } else { |
| 91 | + unsigned pos = 0; |
| 92 | + state->in_heredoc = buffer[pos++]; |
| 93 | + state->stripping_heredoc = buffer[pos++]; |
| 94 | + |
| 95 | + unsigned heredoc_count = 0; |
| 96 | + for (unsigned i = 0; i < MAX_HEREDOCS; i++) { |
| 97 | + unsigned len = strlen(&buffer[pos]); |
| 98 | + |
| 99 | + // We found the ending null byte which means that we're done. |
| 100 | + if (len == 0) |
| 101 | + break; |
| 102 | + |
| 103 | + // Account for the ending null byte in strings (again). |
| 104 | + len++; |
| 105 | + char *heredoc = malloc(len); |
| 106 | + memcpy(heredoc, &buffer[pos], len); |
| 107 | + state->heredocs[i] = heredoc; |
| 108 | + heredoc_count++; |
| 109 | + |
| 110 | + pos += len; |
| 111 | + } |
| 112 | + |
| 113 | + state->heredoc_count = heredoc_count; |
| 114 | + } |
| 115 | +} |
| 116 | + |
| 117 | +static void skip_whitespace(TSLexer *lexer) { |
| 118 | + while (lexer->lookahead != '\0' && lexer->lookahead != '\n' && |
| 119 | + iswspace(lexer->lookahead)) |
| 120 | + lexer->advance(lexer, true); |
| 121 | +} |
| 122 | + |
| 123 | +static bool scan_marker(scanner_state *state, TSLexer *lexer) { |
| 124 | + skip_whitespace(lexer); |
| 125 | + |
| 126 | + if (lexer->lookahead != '<') |
| 127 | + return false; |
| 128 | + lexer->advance(lexer, false); |
| 129 | + |
| 130 | + if (lexer->lookahead != '<') |
| 131 | + return false; |
| 132 | + lexer->advance(lexer, false); |
| 133 | + |
| 134 | + bool stripping = false; |
| 135 | + if (lexer->lookahead == '-') { |
| 136 | + stripping = true; |
| 137 | + lexer->advance(lexer, false); |
| 138 | + } |
| 139 | + |
| 140 | + int32_t quote = 0; |
| 141 | + if (lexer->lookahead == '"' || lexer->lookahead == '\'') { |
| 142 | + quote = lexer->lookahead; |
| 143 | + lexer->advance(lexer, false); |
| 144 | + } |
| 145 | + |
| 146 | + // Reserve a reasonable amount of space for the heredoc delimiter string. |
| 147 | + // Most heredocs (like EOF, EOT, EOS, FILE, etc.) are pretty short so we'll |
| 148 | + // usually only need a few bytes. We're also limited to less than 1024 bytes |
| 149 | + // by tree-sitter since our state has to fit in |
| 150 | + // TREE_SITTER_SERIALIZATION_BUFFER_SIZE. |
| 151 | + char delimiter[DEL_SPACE]; |
| 152 | + |
| 153 | + // We start recording the actual string at position 1 since we store whether |
| 154 | + // it's a stripping heredoc in the first position (with either a dash or a |
| 155 | + // space). |
| 156 | + unsigned del_idx = 1; |
| 157 | + |
| 158 | + while (lexer->lookahead != '\0' && |
| 159 | + (quote ? lexer->lookahead != quote : !iswspace(lexer->lookahead))) { |
| 160 | + if (lexer->lookahead == '\\') { |
| 161 | + lexer->advance(lexer, false); |
| 162 | + |
| 163 | + if (lexer->lookahead == '\0') { |
| 164 | + return false; |
| 165 | + } |
| 166 | + } |
| 167 | + |
| 168 | + if (del_idx > 0) { |
| 169 | + delimiter[del_idx++] = lexer->lookahead; |
| 170 | + } |
| 171 | + lexer->advance(lexer, false); |
| 172 | + |
| 173 | + // If we run out of space, stop recording the delimiter but keep |
| 174 | + // advancing the lexer to ensure that we at least parse the marker |
| 175 | + // correctly. Reserve two bytes: one for the strip indicator and |
| 176 | + // one for the terminating null byte. |
| 177 | + if (del_idx >= DEL_SPACE - 2) { |
| 178 | + del_idx = 0; |
| 179 | + } |
| 180 | + } |
| 181 | + |
| 182 | + if (quote) { |
| 183 | + if (lexer->lookahead != quote) { |
| 184 | + return false; |
| 185 | + } |
| 186 | + lexer->advance(lexer, false); |
| 187 | + } |
| 188 | + |
| 189 | + if (del_idx == 0) { |
| 190 | + lexer->result_symbol = HEREDOC_MARKER; |
| 191 | + return true; |
| 192 | + } |
| 193 | + |
| 194 | + delimiter[0] = stripping ? '-' : ' '; |
| 195 | + delimiter[del_idx] = '\0'; |
| 196 | + |
| 197 | + // We copy the delimiter string to the heap here since we can't store our |
| 198 | + // stack-allocated string in our state (which is stored on the heap). |
| 199 | + char *del_copy = malloc(del_idx + 1); |
| 200 | + memcpy(del_copy, delimiter, del_idx + 1); |
| 201 | + |
| 202 | + if (state->heredoc_count == 0) { |
| 203 | + state->heredoc_count = 1; |
| 204 | + state->heredocs[0] = del_copy; |
| 205 | + state->stripping_heredoc = stripping; |
| 206 | + } else if (state->heredoc_count >= MAX_HEREDOCS) { |
| 207 | + free(del_copy); |
| 208 | + } else { |
| 209 | + state->heredocs[state->heredoc_count++] = del_copy; |
| 210 | + } |
| 211 | + |
| 212 | + lexer->result_symbol = HEREDOC_MARKER; |
| 213 | + return true; |
| 214 | +} |
| 215 | + |
| 216 | +static bool scan_content(scanner_state *state, TSLexer *lexer, |
| 217 | + const bool *valid_symbols) { |
| 218 | + if (state->heredoc_count == 0) { |
| 219 | + state->in_heredoc = false; |
| 220 | + return false; |
| 221 | + } |
| 222 | + |
| 223 | + state->in_heredoc = true; |
| 224 | + |
| 225 | + if (state->stripping_heredoc) { |
| 226 | + skip_whitespace(lexer); |
| 227 | + } |
| 228 | + |
| 229 | + if (valid_symbols[HEREDOC_END]) { |
| 230 | + unsigned delim_idx = 1; |
| 231 | + // Look for the current heredoc delimiter. |
| 232 | + while (state->heredocs[0][delim_idx] != '\0' && |
| 233 | + lexer->lookahead != '\0' && |
| 234 | + lexer->lookahead == state->heredocs[0][delim_idx]) { |
| 235 | + lexer->advance(lexer, false); |
| 236 | + delim_idx++; |
| 237 | + } |
| 238 | + |
| 239 | + // Check if the entire string matched. |
| 240 | + if (state->heredocs[0][delim_idx] == '\0') { |
| 241 | + lexer->result_symbol = HEREDOC_END; |
| 242 | + |
| 243 | + // Shift the first heredoc off the list. |
| 244 | + free(state->heredocs[0]); |
| 245 | + |
| 246 | + for (unsigned i = 1; i < state->heredoc_count; i++) { |
| 247 | + state->heredocs[i - 1] = state->heredocs[i]; |
| 248 | + } |
| 249 | + state->heredocs[state->heredoc_count - 1] = NULL; |
| 250 | + state->heredoc_count--; |
| 251 | + |
| 252 | + if (state->heredoc_count > 0) { |
| 253 | + state->stripping_heredoc = state->heredocs[0][0] == '-'; |
| 254 | + } else { |
| 255 | + state->in_heredoc = false; |
| 256 | + } |
| 257 | + |
| 258 | + return true; |
| 259 | + } |
| 260 | + } |
| 261 | + |
| 262 | + if (!valid_symbols[HEREDOC_LINE]) |
| 263 | + return false; |
| 264 | + |
| 265 | + lexer->result_symbol = HEREDOC_LINE; |
| 266 | + |
| 267 | + for (;;) { |
| 268 | + switch (lexer->lookahead) { |
| 269 | + case '\0': |
| 270 | + if (lexer->eof(lexer)) { |
| 271 | + state->in_heredoc = false; |
| 272 | + return true; |
| 273 | + } |
| 274 | + lexer->advance(lexer, false); |
| 275 | + break; |
| 276 | + |
| 277 | + case '\n': |
| 278 | + return true; |
| 279 | + |
| 280 | + default: |
| 281 | + lexer->advance(lexer, false); |
| 282 | + } |
| 283 | + } |
| 284 | +} |
| 285 | + |
| 286 | +bool tree_sitter_dockerfile_external_scanner_scan(void *payload, TSLexer *lexer, |
| 287 | + const bool *valid_symbols) { |
| 288 | + scanner_state *state = payload; |
| 289 | + |
| 290 | + if (valid_symbols[ERROR_SENTINEL]) { |
| 291 | + if (state->in_heredoc) { |
| 292 | + return scan_content(state, lexer, valid_symbols); |
| 293 | + } else { |
| 294 | + return scan_marker(state, lexer); |
| 295 | + } |
| 296 | + } |
| 297 | + |
| 298 | + // HEREDOC_NL only matches a linebreak if there are open heredocs. This is |
| 299 | + // necessary to avoid a conflict in the grammar since a normal line break |
| 300 | + // could either be the start of a heredoc or the end of an instruction. |
| 301 | + if (valid_symbols[HEREDOC_NL]) { |
| 302 | + if (state->heredoc_count > 0 && lexer->lookahead == '\n') { |
| 303 | + lexer->result_symbol = HEREDOC_NL; |
| 304 | + lexer->advance(lexer, false); |
| 305 | + return true; |
| 306 | + } |
| 307 | + } |
| 308 | + |
| 309 | + if (valid_symbols[HEREDOC_MARKER]) { |
| 310 | + return scan_marker(state, lexer); |
| 311 | + } |
| 312 | + |
| 313 | + if (valid_symbols[HEREDOC_LINE] || valid_symbols[HEREDOC_END]) { |
| 314 | + return scan_content(state, lexer, valid_symbols); |
| 315 | + } |
| 316 | + |
| 317 | + return false; |
| 318 | +} |
0 commit comments