| 1 |
/*------------------------------------------------------------------------- |
| 2 |
* |
| 3 |
* parser.c |
| 4 |
* Main entry point/driver for PostgreSQL grammar |
| 5 |
* |
| 6 |
* Note that the grammar is not allowed to perform any table access |
| 7 |
* (since we need to be able to do basic parsing even while inside an |
| 8 |
* aborted transaction). Therefore, the data structures returned by |
| 9 |
* the grammar are "raw" parsetrees that still need to be analyzed by |
| 10 |
* analyze.c and related files. |
| 11 |
* |
| 12 |
* |
| 13 |
* Portions Copyright (c) 1996-2025, PostgreSQL Global Development Group |
| 14 |
* Portions Copyright (c) 1994, Regents of the University of California |
| 15 |
* |
| 16 |
* IDENTIFICATION |
| 17 |
* src/backend/parser/parser.c |
| 18 |
* |
| 19 |
*------------------------------------------------------------------------- |
| 20 |
*/ |
| 21 |
|
| 22 |
#include "postgres.h" |
| 23 |
|
| 24 |
#include "gramparse.h" |
| 25 |
#include "mb/pg_wchar.h" |
| 26 |
#include "parser/parser.h" |
| 27 |
#include "parser/scansup.h" |
| 28 |
|
| 29 |
static bool check_uescapechar(unsigned char escape); |
| 30 |
static char *str_udeescape(const char *str, char escape, |
| 31 |
int position, core_yyscan_t yyscanner); |
| 32 |
|
| 33 |
|
| 34 |
/* |
| 35 |
* raw_parser |
| 36 |
* Given a query in string form, do lexical and grammatical analysis. |
| 37 |
* |
| 38 |
* Returns a list of raw (un-analyzed) parse trees. The contents of the |
| 39 |
* list have the form required by the specified RawParseMode. |
| 40 |
*/ |
| 41 |
List * |
| 42 |
raw_parser(const char *str, RawParseMode mode) |
| 43 |
{ |
| 44 |
core_yyscan_t yyscanner; |
| 45 |
base_yy_extra_type yyextra; |
| 46 |
int yyresult; |
| 47 |
|
| 48 |
/* initialize the flex scanner */ |
| 49 |
yyscanner = scanner_init(str, &yyextra.core_yy_extra, |
| 50 |
&ScanKeywords, ScanKeywordTokens); |
| 51 |
|
| 52 |
/* base_yylex() only needs us to initialize the lookahead token, if any */ |
| 53 |
if (mode == RAW_PARSE_DEFAULT) |
| 54 |
yyextra.have_lookahead = false; |
| 55 |
else |
| 56 |
{ |
| 57 |
/* this array is indexed by RawParseMode enum */ |
| 58 |
static const int mode_token[] = { |
| 59 |
[RAW_PARSE_DEFAULT] = 0, |
| 60 |
[RAW_PARSE_TYPE_NAME] = MODE_TYPE_NAME, |
| 61 |
[RAW_PARSE_PLPGSQL_EXPR] = MODE_PLPGSQL_EXPR, |
| 62 |
[RAW_PARSE_PLPGSQL_ASSIGN1] = MODE_PLPGSQL_ASSIGN1, |
| 63 |
[RAW_PARSE_PLPGSQL_ASSIGN2] = MODE_PLPGSQL_ASSIGN2, |
| 64 |
[RAW_PARSE_PLPGSQL_ASSIGN3] = MODE_PLPGSQL_ASSIGN3, |
| 65 |
}; |
| 66 |
|
| 67 |
yyextra.have_lookahead = true; |
| 68 |
yyextra.lookahead_token = mode_token[mode]; |
| 69 |
yyextra.lookahead_yylloc = 0; |
| 70 |
yyextra.lookahead_end = NULL; |
| 71 |
} |
| 72 |
|
| 73 |
/* initialize the bison parser */ |
| 74 |
parser_init(&yyextra); |
| 75 |
|
| 76 |
/* Parse! */ |
| 77 |
yyresult = base_yyparse(yyscanner); |
| 78 |
|
| 79 |
/* Clean up (release memory) */ |
| 80 |
scanner_finish(yyscanner); |
| 81 |
|
| 82 |
if (yyresult) /* error */ |
| 83 |
return NIL; |
| 84 |
|
| 85 |
return yyextra.parsetree; |
| 86 |
} |
| 87 |
|
| 88 |
|
| 89 |
/* |
| 90 |
* Intermediate filter between parser and core lexer (core_yylex in scan.l). |
| 91 |
* |
| 92 |
* This filter is needed because in some cases the standard SQL grammar |
| 93 |
* requires more than one token lookahead. We reduce these cases to one-token |
| 94 |
* lookahead by replacing tokens here, in order to keep the grammar LALR(1). |
| 95 |
* |
| 96 |
* Using a filter is simpler than trying to recognize multiword tokens |
| 97 |
* directly in scan.l, because we'd have to allow for comments between the |
| 98 |
* words. Furthermore it's not clear how to do that without re-introducing |
| 99 |
* scanner backtrack, which would cost more performance than this filter |
| 100 |
* layer does. |
| 101 |
* |
| 102 |
* We also use this filter to convert UIDENT and USCONST sequences into |
| 103 |
* plain IDENT and SCONST tokens. While that could be handled by additional |
| 104 |
* productions in the main grammar, it's more efficient to do it like this. |
| 105 |
* |
| 106 |
* The filter also provides a convenient place to translate between |
| 107 |
* the core_YYSTYPE and YYSTYPE representations (which are really the |
| 108 |
* same thing anyway, but notationally they're different). |
| 109 |
*/ |
| 110 |
int |
| 111 |
base_yylex(YYSTYPE *lvalp, YYLTYPE *llocp, core_yyscan_t yyscanner) |
| 112 |
{ |
| 113 |
base_yy_extra_type *yyextra = pg_yyget_extra(yyscanner); |
| 114 |
int cur_token; |
| 115 |
int next_token; |
| 116 |
int cur_token_length; |
| 117 |
YYLTYPE cur_yylloc; |
| 118 |
|
| 119 |
/* Get next token --- we might already have it */ |
| 120 |
if (yyextra->have_lookahead) |
| 121 |
{ |
| 122 |
cur_token = yyextra->lookahead_token; |
| 123 |
lvalp->core_yystype = yyextra->lookahead_yylval; |
| 124 |
*llocp = yyextra->lookahead_yylloc; |
| 125 |
if (yyextra->lookahead_end) |
| 126 |
*(yyextra->lookahead_end) = yyextra->lookahead_hold_char; |
| 127 |
yyextra->have_lookahead = false; |
| 128 |
} |
| 129 |
else |
| 130 |
cur_token = core_yylex(&(lvalp->core_yystype), llocp, yyscanner); |
| 131 |
|
| 132 |
/* |
| 133 |
* If this token isn't one that requires lookahead, just return it. If it |
| 134 |
* does, determine the token length. (We could get that via strlen(), but |
| 135 |
* since we have such a small set of possibilities, hardwiring seems |
| 136 |
* feasible and more efficient --- at least for the fixed-length cases.) |
| 137 |
*/ |
| 138 |
switch (cur_token) |
| 139 |
{ |
| 140 |
case FORMAT: |
| 141 |
cur_token_length = 6; |
| 142 |
break; |
| 143 |
case NOT: |
| 144 |
cur_token_length = 3; |
| 145 |
break; |
| 146 |
case NULLS_P: |
| 147 |
cur_token_length = 5; |
| 148 |
break; |
| 149 |
case WITH: |
| 150 |
cur_token_length = 4; |
| 151 |
break; |
| 152 |
case UIDENT: |
| 153 |
case USCONST: |
| 154 |
cur_token_length = strlen(yyextra->core_yy_extra.scanbuf + *llocp); |
| 155 |
break; |
| 156 |
case WITHOUT: |
| 157 |
cur_token_length = 7; |
| 158 |
break; |
| 159 |
default: |
| 160 |
return cur_token; |
| 161 |
} |
| 162 |
|
| 163 |
/* |
| 164 |
* Identify end+1 of current token. core_yylex() has temporarily stored a |
| 165 |
* '\0' here, and will undo that when we call it again. We need to redo |
| 166 |
* it to fully revert the lookahead call for error reporting purposes. |
| 167 |
*/ |
| 168 |
yyextra->lookahead_end = yyextra->core_yy_extra.scanbuf + |
| 169 |
*llocp + cur_token_length; |
| 170 |
Assert(*(yyextra->lookahead_end) == '\0'); |
| 171 |
|
| 172 |
/* |
| 173 |
* Save and restore *llocp around the call. It might look like we could |
| 174 |
* avoid this by just passing &lookahead_yylloc to core_yylex(), but that |
| 175 |
* does not work because flex actually holds onto the last-passed pointer |
| 176 |
* internally, and will use that for error reporting. We need any error |
| 177 |
* reports to point to the current token, not the next one. |
| 178 |
*/ |
| 179 |
cur_yylloc = *llocp; |
| 180 |
|
| 181 |
/* Get next token, saving outputs into lookahead variables */ |
| 182 |
next_token = core_yylex(&(yyextra->lookahead_yylval), llocp, yyscanner); |
| 183 |
yyextra->lookahead_token = next_token; |
| 184 |
yyextra->lookahead_yylloc = *llocp; |
| 185 |
|
| 186 |
*llocp = cur_yylloc; |
| 187 |
|
| 188 |
/* Now revert the un-truncation of the current token */ |
| 189 |
yyextra->lookahead_hold_char = *(yyextra->lookahead_end); |
| 190 |
*(yyextra->lookahead_end) = '\0'; |
| 191 |
|
| 192 |
yyextra->have_lookahead = true; |
| 193 |
|
| 194 |
/* Replace cur_token if needed, based on lookahead */ |
| 195 |
switch (cur_token) |
| 196 |
{ |
| 197 |
case FORMAT: |
| 198 |
/* Replace FORMAT by FORMAT_LA if it's followed by JSON */ |
| 199 |
switch (next_token) |
| 200 |
{ |
| 201 |
case JSON: |
| 202 |
cur_token = FORMAT_LA; |
| 203 |
break; |
| 204 |
} |
| 205 |
break; |
| 206 |
|
| 207 |
case NOT: |
| 208 |
/* Replace NOT by NOT_LA if it's followed by BETWEEN, IN, etc */ |
| 209 |
switch (next_token) |
| 210 |
{ |
| 211 |
case BETWEEN: |
| 212 |
case IN_P: |
| 213 |
case LIKE: |
| 214 |
case ILIKE: |
| 215 |
case SIMILAR: |
| 216 |
cur_token = NOT_LA; |
| 217 |
break; |
| 218 |
} |
| 219 |
break; |
| 220 |
|
| 221 |
case NULLS_P: |
| 222 |
/* Replace NULLS_P by NULLS_LA if it's followed by FIRST or LAST */ |
| 223 |
switch (next_token) |
| 224 |
{ |
| 225 |
case FIRST_P: |
| 226 |
case LAST_P: |
| 227 |
cur_token = NULLS_LA; |
| 228 |
break; |
| 229 |
} |
| 230 |
break; |
| 231 |
|
| 232 |
case WITH: |
| 233 |
/* Replace WITH by WITH_LA if it's followed by TIME or ORDINALITY */ |
| 234 |
switch (next_token) |
| 235 |
{ |
| 236 |
case TIME: |
| 237 |
case ORDINALITY: |
| 238 |
cur_token = WITH_LA; |
| 239 |
break; |
| 240 |
} |
| 241 |
break; |
| 242 |
|
| 243 |
case WITHOUT: |
| 244 |
/* Replace WITHOUT by WITHOUT_LA if it's followed by TIME */ |
| 245 |
switch (next_token) |
| 246 |
{ |
| 247 |
case TIME: |
| 248 |
cur_token = WITHOUT_LA; |
| 249 |
break; |
| 250 |
} |
| 251 |
break; |
| 252 |
|
| 253 |
case UIDENT: |
| 254 |
case USCONST: |
| 255 |
/* Look ahead for UESCAPE */ |
| 256 |
if (next_token == UESCAPE) |
| 257 |
{ |
| 258 |
/* Yup, so get third token, which had better be SCONST */ |
| 259 |
const char *escstr; |
| 260 |
|
| 261 |
/* Again save and restore *llocp */ |
| 262 |
cur_yylloc = *llocp; |
| 263 |
|
| 264 |
/* Un-truncate current token so errors point to third token */ |
| 265 |
*(yyextra->lookahead_end) = yyextra->lookahead_hold_char; |
| 266 |
|
| 267 |
/* Get third token */ |
| 268 |
next_token = core_yylex(&(yyextra->lookahead_yylval), |
| 269 |
llocp, yyscanner); |
| 270 |
|
| 271 |
/* If we throw error here, it will point to third token */ |
| 272 |
if (next_token != SCONST) |
| 273 |
scanner_yyerror("UESCAPE must be followed by a simple string literal", |
| 274 |
yyscanner); |
| 275 |
|
| 276 |
escstr = yyextra->lookahead_yylval.str; |
| 277 |
if (strlen(escstr) != 1 || !check_uescapechar(escstr[0])) |
| 278 |
scanner_yyerror("invalid Unicode escape character", |
| 279 |
yyscanner); |
| 280 |
|
| 281 |
/* Now restore *llocp; errors will point to first token */ |
| 282 |
*llocp = cur_yylloc; |
| 283 |
|
| 284 |
/* Apply Unicode conversion */ |
| 285 |
lvalp->core_yystype.str = |
| 286 |
str_udeescape(lvalp->core_yystype.str, |
| 287 |
escstr[0], |
| 288 |
*llocp, |
| 289 |
yyscanner); |
| 290 |
|
| 291 |
/* |
| 292 |
* We don't need to revert the un-truncation of UESCAPE. What |
| 293 |
* we do want to do is clear have_lookahead, thereby consuming |
| 294 |
* all three tokens. |
| 295 |
*/ |
| 296 |
yyextra->have_lookahead = false; |
| 297 |
} |
| 298 |
else |
| 299 |
{ |
| 300 |
/* No UESCAPE, so convert using default escape character */ |
| 301 |
lvalp->core_yystype.str = |
| 302 |
str_udeescape(lvalp->core_yystype.str, |
| 303 |
'\\', |
| 304 |
*llocp, |
| 305 |
yyscanner); |
| 306 |
} |
| 307 |
|
| 308 |
if (cur_token == UIDENT) |
| 309 |
{ |
| 310 |
/* It's an identifier, so truncate as appropriate */ |
| 311 |
truncate_identifier(lvalp->core_yystype.str, |
| 312 |
strlen(lvalp->core_yystype.str), |
| 313 |
true); |
| 314 |
cur_token = IDENT; |
| 315 |
} |
| 316 |
else if (cur_token == USCONST) |
| 317 |
{ |
| 318 |
cur_token = SCONST; |
| 319 |
} |
| 320 |
break; |
| 321 |
} |
| 322 |
|
| 323 |
return cur_token; |
| 324 |
} |
| 325 |
|
| 326 |
/* convert hex digit (caller should have verified that) to value */ |
| 327 |
static unsigned int |
| 328 |
hexval(unsigned char c) |
| 329 |
{ |
| 330 |
if (c >= '0' && c <= '9') |
| 331 |
return c - '0'; |
| 332 |
if (c >= 'a' && c <= 'f') |
| 333 |
return c - 'a' + 0xA; |
| 334 |
if (c >= 'A' && c <= 'F') |
| 335 |
return c - 'A' + 0xA; |
| 336 |
elog(ERROR, "invalid hexadecimal digit"); |
| 337 |
return 0; /* not reached */ |
| 338 |
} |
| 339 |
|
| 340 |
/* is Unicode code point acceptable? */ |
| 341 |
static void |
| 342 |
check_unicode_value(pg_wchar c) |
| 343 |
{ |
| 344 |
if (!is_valid_unicode_codepoint(c)) |
| 345 |
ereport(ERROR, |
| 346 |
(errcode(ERRCODE_SYNTAX_ERROR), |
| 347 |
errmsg("invalid Unicode escape value"))); |
| 348 |
} |
| 349 |
|
| 350 |
/* is 'escape' acceptable as Unicode escape character (UESCAPE syntax) ? */ |
| 351 |
static bool |
| 352 |
check_uescapechar(unsigned char escape) |
| 353 |
{ |
| 354 |
if (isxdigit(escape) |
| 355 |
|| escape == '+' |
| 356 |
|| escape == '\'' |
| 357 |
|| escape == '"' |
| 358 |
|| scanner_isspace(escape)) |
| 359 |
return false; |
| 360 |
else |
| 361 |
return true; |
| 362 |
} |
| 363 |
|
| 364 |
/* |
| 365 |
* Process Unicode escapes in "str", producing a palloc'd plain string |
| 366 |
* |
| 367 |
* escape: the escape character to use |
| 368 |
* position: start position of U&'' or U&"" string token |
| 369 |
* yyscanner: context information needed for error reports |
| 370 |
*/ |
| 371 |
static char * |
| 372 |
str_udeescape(const char *str, char escape, |
| 373 |
int position, core_yyscan_t yyscanner) |
| 374 |
{ |
| 375 |
const char *in; |
| 376 |
char *new, |
| 377 |
*out; |
| 378 |
size_t new_len; |
| 379 |
pg_wchar pair_first = 0; |
| 380 |
ScannerCallbackState scbstate; |
| 381 |
|
| 382 |
/* |
| 383 |
* Guesstimate that result will be no longer than input, but allow enough |
| 384 |
* padding for Unicode conversion. |
| 385 |
*/ |
| 386 |
new_len = strlen(str) + MAX_UNICODE_EQUIVALENT_STRING + 1; |
| 387 |
new = palloc(new_len); |
| 388 |
|
| 389 |
in = str; |
| 390 |
out = new; |
| 391 |
while (*in) |
| 392 |
{ |
| 393 |
/* Enlarge string if needed */ |
| 394 |
size_t out_dist = out - new; |
| 395 |
|
| 396 |
if (out_dist > new_len - (MAX_UNICODE_EQUIVALENT_STRING + 1)) |
| 397 |
{ |
| 398 |
new_len *= 2; |
| 399 |
new = repalloc(new, new_len); |
| 400 |
out = new + out_dist; |
| 401 |
} |
| 402 |
|
| 403 |
if (in[0] == escape) |
| 404 |
{ |
| 405 |
/* |
| 406 |
* Any errors reported while processing this escape sequence will |
| 407 |
* have an error cursor pointing at the escape. |
| 408 |
*/ |
| 409 |
setup_scanner_errposition_callback(&scbstate, yyscanner, |
| 410 |
in - str + position + 3); /* 3 for U&" */ |
| 411 |
if (in[1] == escape) |
| 412 |
{ |
| 413 |
if (pair_first) |
| 414 |
goto invalid_pair; |
| 415 |
*out++ = escape; |
| 416 |
in += 2; |
| 417 |
} |
| 418 |
else if (isxdigit((unsigned char) in[1]) && |
| 419 |
isxdigit((unsigned char) in[2]) && |
| 420 |
isxdigit((unsigned char) in[3]) && |
| 421 |
isxdigit((unsigned char) in[4])) |
| 422 |
{ |
| 423 |
pg_wchar unicode; |
| 424 |
|
| 425 |
unicode = (hexval(in[1]) << 12) + |
| 426 |
(hexval(in[2]) << 8) + |
| 427 |
(hexval(in[3]) << 4) + |
| 428 |
hexval(in[4]); |
| 429 |
check_unicode_value(unicode); |
| 430 |
if (pair_first) |
| 431 |
{ |
| 432 |
if (is_utf16_surrogate_second(unicode)) |
| 433 |
{ |
| 434 |
unicode = surrogate_pair_to_codepoint(pair_first, unicode); |
| 435 |
pair_first = 0; |
| 436 |
} |
| 437 |
else |
| 438 |
goto invalid_pair; |
| 439 |
} |
| 440 |
else if (is_utf16_surrogate_second(unicode)) |
| 441 |
goto invalid_pair; |
| 442 |
|
| 443 |
if (is_utf16_surrogate_first(unicode)) |
| 444 |
pair_first = unicode; |
| 445 |
else |
| 446 |
{ |
| 447 |
pg_unicode_to_server(unicode, (unsigned char *) out); |
| 448 |
out += strlen(out); |
| 449 |
} |
| 450 |
in += 5; |
| 451 |
} |
| 452 |
else if (in[1] == '+' && |
| 453 |
isxdigit((unsigned char) in[2]) && |
| 454 |
isxdigit((unsigned char) in[3]) && |
| 455 |
isxdigit((unsigned char) in[4]) && |
| 456 |
isxdigit((unsigned char) in[5]) && |
| 457 |
isxdigit((unsigned char) in[6]) && |
| 458 |
isxdigit((unsigned char) in[7])) |
| 459 |
{ |
| 460 |
pg_wchar unicode; |
| 461 |
|
| 462 |
unicode = (hexval(in[2]) << 20) + |
| 463 |
(hexval(in[3]) << 16) + |
| 464 |
(hexval(in[4]) << 12) + |
| 465 |
(hexval(in[5]) << 8) + |
| 466 |
(hexval(in[6]) << 4) + |
| 467 |
hexval(in[7]); |
| 468 |
check_unicode_value(unicode); |
| 469 |
if (pair_first) |
| 470 |
{ |
| 471 |
if (is_utf16_surrogate_second(unicode)) |
| 472 |
{ |
| 473 |
unicode = surrogate_pair_to_codepoint(pair_first, unicode); |
| 474 |
pair_first = 0; |
| 475 |
} |
| 476 |
else |
| 477 |
goto invalid_pair; |
| 478 |
} |
| 479 |
else if (is_utf16_surrogate_second(unicode)) |
| 480 |
goto invalid_pair; |
| 481 |
|
| 482 |
if (is_utf16_surrogate_first(unicode)) |
| 483 |
pair_first = unicode; |
| 484 |
else |
| 485 |
{ |
| 486 |
pg_unicode_to_server(unicode, (unsigned char *) out); |
| 487 |
out += strlen(out); |
| 488 |
} |
| 489 |
in += 8; |
| 490 |
} |
| 491 |
else |
| 492 |
ereport(ERROR, |
| 493 |
(errcode(ERRCODE_SYNTAX_ERROR), |
| 494 |
errmsg("invalid Unicode escape"), |
| 495 |
errhint("Unicode escapes must be \\XXXX or \\+XXXXXX."))); |
| 496 |
|
| 497 |
cancel_scanner_errposition_callback(&scbstate); |
| 498 |
} |
| 499 |
else |
| 500 |
{ |
| 501 |
if (pair_first) |
| 502 |
goto invalid_pair; |
| 503 |
|
| 504 |
*out++ = *in++; |
| 505 |
} |
| 506 |
} |
| 507 |
|
| 508 |
/* unfinished surrogate pair? */ |
| 509 |
if (pair_first) |
| 510 |
goto invalid_pair; |
| 511 |
|
| 512 |
*out = '\0'; |
| 513 |
return new; |
| 514 |
|
| 515 |
/* |
| 516 |
* We might get here with the error callback active, or not. Call |
| 517 |
* scanner_errposition to make sure an error cursor appears; if the |
| 518 |
* callback is active, this is duplicative but harmless. |
| 519 |
*/ |
| 520 |
invalid_pair: |
| 521 |
ereport(ERROR, |
| 522 |
(errcode(ERRCODE_SYNTAX_ERROR), |
| 523 |
errmsg("invalid Unicode surrogate pair"), |
| 524 |
scanner_errposition(in - str + position + 3, /* 3 for U&" */ |
| 525 |
yyscanner))); |
| 526 |
return NULL; /* keep compiler quiet */ |
| 527 |
} |