Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * varlena.c
4 : * Functions for the variable-length built-in types.
5 : *
6 : * Portions Copyright (c) 1996-2024, PostgreSQL Global Development Group
7 : * Portions Copyright (c) 1994, Regents of the University of California
8 : *
9 : *
10 : * IDENTIFICATION
11 : * src/backend/utils/adt/varlena.c
12 : *
13 : *-------------------------------------------------------------------------
14 : */
15 : #include "postgres.h"
16 :
17 : #include <ctype.h>
18 : #include <limits.h>
19 :
20 : #include "access/detoast.h"
21 : #include "access/toast_compression.h"
22 : #include "catalog/pg_collation.h"
23 : #include "catalog/pg_type.h"
24 : #include "common/hashfn.h"
25 : #include "common/int.h"
26 : #include "common/unicode_category.h"
27 : #include "common/unicode_norm.h"
28 : #include "common/unicode_version.h"
29 : #include "funcapi.h"
30 : #include "lib/hyperloglog.h"
31 : #include "libpq/pqformat.h"
32 : #include "miscadmin.h"
33 : #include "nodes/execnodes.h"
34 : #include "parser/scansup.h"
35 : #include "port/pg_bswap.h"
36 : #include "regex/regex.h"
37 : #include "utils/builtins.h"
38 : #include "utils/bytea.h"
39 : #include "utils/guc.h"
40 : #include "utils/lsyscache.h"
41 : #include "utils/memutils.h"
42 : #include "utils/pg_locale.h"
43 : #include "utils/sortsupport.h"
44 : #include "utils/varlena.h"
45 :
46 :
47 : /* GUC variable */
48 : int bytea_output = BYTEA_OUTPUT_HEX;
49 :
50 : typedef struct varlena VarString;
51 :
52 : /*
53 : * State for text_position_* functions.
54 : */
55 : typedef struct
56 : {
57 : bool is_multibyte_char_in_char; /* need to check char boundaries? */
58 :
59 : char *str1; /* haystack string */
60 : char *str2; /* needle string */
61 : int len1; /* string lengths in bytes */
62 : int len2;
63 :
64 : /* Skip table for Boyer-Moore-Horspool search algorithm: */
65 : int skiptablemask; /* mask for ANDing with skiptable subscripts */
66 : int skiptable[256]; /* skip distance for given mismatched char */
67 :
68 : char *last_match; /* pointer to last match in 'str1' */
69 :
70 : /*
71 : * Sometimes we need to convert the byte position of a match to a
72 : * character position. These store the last position that was converted,
73 : * so that on the next call, we can continue from that point, rather than
74 : * count characters from the very beginning.
75 : */
76 : char *refpoint; /* pointer within original haystack string */
77 : int refpos; /* 0-based character offset of the same point */
78 : } TextPositionState;
79 :
80 : typedef struct
81 : {
82 : char *buf1; /* 1st string, or abbreviation original string
83 : * buf */
84 : char *buf2; /* 2nd string, or abbreviation strxfrm() buf */
85 : int buflen1; /* Allocated length of buf1 */
86 : int buflen2; /* Allocated length of buf2 */
87 : int last_len1; /* Length of last buf1 string/strxfrm() input */
88 : int last_len2; /* Length of last buf2 string/strxfrm() blob */
89 : int last_returned; /* Last comparison result (cache) */
90 : bool cache_blob; /* Does buf2 contain strxfrm() blob, etc? */
91 : bool collate_c;
92 : Oid typid; /* Actual datatype (text/bpchar/bytea/name) */
93 : hyperLogLogState abbr_card; /* Abbreviated key cardinality state */
94 : hyperLogLogState full_card; /* Full key cardinality state */
95 : double prop_card; /* Required cardinality proportion */
96 : pg_locale_t locale;
97 : } VarStringSortSupport;
98 :
99 : /*
100 : * Output data for split_text(): we output either to an array or a table.
101 : * tupstore and tupdesc must be set up in advance to output to a table.
102 : */
103 : typedef struct
104 : {
105 : ArrayBuildState *astate;
106 : Tuplestorestate *tupstore;
107 : TupleDesc tupdesc;
108 : } SplitTextOutputData;
109 :
110 : /*
111 : * This should be large enough that most strings will fit, but small enough
112 : * that we feel comfortable putting it on the stack
113 : */
114 : #define TEXTBUFLEN 1024
115 :
116 : #define DatumGetVarStringP(X) ((VarString *) PG_DETOAST_DATUM(X))
117 : #define DatumGetVarStringPP(X) ((VarString *) PG_DETOAST_DATUM_PACKED(X))
118 :
119 : static int varstrfastcmp_c(Datum x, Datum y, SortSupport ssup);
120 : static int bpcharfastcmp_c(Datum x, Datum y, SortSupport ssup);
121 : static int namefastcmp_c(Datum x, Datum y, SortSupport ssup);
122 : static int varlenafastcmp_locale(Datum x, Datum y, SortSupport ssup);
123 : static int namefastcmp_locale(Datum x, Datum y, SortSupport ssup);
124 : static int varstrfastcmp_locale(char *a1p, int len1, char *a2p, int len2, SortSupport ssup);
125 : static Datum varstr_abbrev_convert(Datum original, SortSupport ssup);
126 : static bool varstr_abbrev_abort(int memtupcount, SortSupport ssup);
127 : static int32 text_length(Datum str);
128 : static text *text_catenate(text *t1, text *t2);
129 : static text *text_substring(Datum str,
130 : int32 start,
131 : int32 length,
132 : bool length_not_specified);
133 : static text *text_overlay(text *t1, text *t2, int sp, int sl);
134 : static int text_position(text *t1, text *t2, Oid collid);
135 : static void text_position_setup(text *t1, text *t2, Oid collid, TextPositionState *state);
136 : static bool text_position_next(TextPositionState *state);
137 : static char *text_position_next_internal(char *start_ptr, TextPositionState *state);
138 : static char *text_position_get_match_ptr(TextPositionState *state);
139 : static int text_position_get_match_pos(TextPositionState *state);
140 : static void text_position_cleanup(TextPositionState *state);
141 : static void check_collation_set(Oid collid);
142 : static int text_cmp(text *arg1, text *arg2, Oid collid);
143 : static bytea *bytea_catenate(bytea *t1, bytea *t2);
144 : static bytea *bytea_substring(Datum str,
145 : int S,
146 : int L,
147 : bool length_not_specified);
148 : static bytea *bytea_overlay(bytea *t1, bytea *t2, int sp, int sl);
149 : static void appendStringInfoText(StringInfo str, const text *t);
150 : static bool split_text(FunctionCallInfo fcinfo, SplitTextOutputData *tstate);
151 : static void split_text_accum_result(SplitTextOutputData *tstate,
152 : text *field_value,
153 : text *null_string,
154 : Oid collation);
155 : static text *array_to_text_internal(FunctionCallInfo fcinfo, ArrayType *v,
156 : const char *fldsep, const char *null_string);
157 : static StringInfo makeStringAggState(FunctionCallInfo fcinfo);
158 : static bool text_format_parse_digits(const char **ptr, const char *end_ptr,
159 : int *value);
160 : static const char *text_format_parse_format(const char *start_ptr,
161 : const char *end_ptr,
162 : int *argpos, int *widthpos,
163 : int *flags, int *width);
164 : static void text_format_string_conversion(StringInfo buf, char conversion,
165 : FmgrInfo *typOutputInfo,
166 : Datum value, bool isNull,
167 : int flags, int width);
168 : static void text_format_append_string(StringInfo buf, const char *str,
169 : int flags, int width);
170 :
171 :
172 : /*****************************************************************************
173 : * CONVERSION ROUTINES EXPORTED FOR USE BY C CODE *
174 : *****************************************************************************/
175 :
176 : /*
177 : * cstring_to_text
178 : *
179 : * Create a text value from a null-terminated C string.
180 : *
181 : * The new text value is freshly palloc'd with a full-size VARHDR.
182 : */
183 : text *
184 23859524 : cstring_to_text(const char *s)
185 : {
186 23859524 : return cstring_to_text_with_len(s, strlen(s));
187 : }
188 :
189 : /*
190 : * cstring_to_text_with_len
191 : *
192 : * Same as cstring_to_text except the caller specifies the string length;
193 : * the string need not be null_terminated.
194 : */
195 : text *
196 26422184 : cstring_to_text_with_len(const char *s, int len)
197 : {
198 26422184 : text *result = (text *) palloc(len + VARHDRSZ);
199 :
200 26422184 : SET_VARSIZE(result, len + VARHDRSZ);
201 26422184 : memcpy(VARDATA(result), s, len);
202 :
203 26422184 : return result;
204 : }
205 :
206 : /*
207 : * text_to_cstring
208 : *
209 : * Create a palloc'd, null-terminated C string from a text value.
210 : *
211 : * We support being passed a compressed or toasted text value.
212 : * This is a bit bogus since such values shouldn't really be referred to as
213 : * "text *", but it seems useful for robustness. If we didn't handle that
214 : * case here, we'd need another routine that did, anyway.
215 : */
216 : char *
217 14190852 : text_to_cstring(const text *t)
218 : {
219 : /* must cast away the const, unfortunately */
220 14190852 : text *tunpacked = pg_detoast_datum_packed(unconstify(text *, t));
221 14190852 : int len = VARSIZE_ANY_EXHDR(tunpacked);
222 : char *result;
223 :
224 14190852 : result = (char *) palloc(len + 1);
225 14190852 : memcpy(result, VARDATA_ANY(tunpacked), len);
226 14190852 : result[len] = '\0';
227 :
228 14190852 : if (tunpacked != t)
229 34430 : pfree(tunpacked);
230 :
231 14190852 : return result;
232 : }
233 :
234 : /*
235 : * text_to_cstring_buffer
236 : *
237 : * Copy a text value into a caller-supplied buffer of size dst_len.
238 : *
239 : * The text string is truncated if necessary to fit. The result is
240 : * guaranteed null-terminated (unless dst_len == 0).
241 : *
242 : * We support being passed a compressed or toasted text value.
243 : * This is a bit bogus since such values shouldn't really be referred to as
244 : * "text *", but it seems useful for robustness. If we didn't handle that
245 : * case here, we'd need another routine that did, anyway.
246 : */
247 : void
248 940 : text_to_cstring_buffer(const text *src, char *dst, size_t dst_len)
249 : {
250 : /* must cast away the const, unfortunately */
251 940 : text *srcunpacked = pg_detoast_datum_packed(unconstify(text *, src));
252 940 : size_t src_len = VARSIZE_ANY_EXHDR(srcunpacked);
253 :
254 940 : if (dst_len > 0)
255 : {
256 940 : dst_len--;
257 940 : if (dst_len >= src_len)
258 940 : dst_len = src_len;
259 : else /* ensure truncation is encoding-safe */
260 0 : dst_len = pg_mbcliplen(VARDATA_ANY(srcunpacked), src_len, dst_len);
261 940 : memcpy(dst, VARDATA_ANY(srcunpacked), dst_len);
262 940 : dst[dst_len] = '\0';
263 : }
264 :
265 940 : if (srcunpacked != src)
266 0 : pfree(srcunpacked);
267 940 : }
268 :
269 :
270 : /*****************************************************************************
271 : * USER I/O ROUTINES *
272 : *****************************************************************************/
273 :
274 :
275 : #define VAL(CH) ((CH) - '0')
276 : #define DIG(VAL) ((VAL) + '0')
277 :
278 : /*
279 : * byteain - converts from printable representation of byte array
280 : *
281 : * Non-printable characters must be passed as '\nnn' (octal) and are
282 : * converted to internal form. '\' must be passed as '\\'.
283 : * ereport(ERROR, ...) if bad form.
284 : *
285 : * BUGS:
286 : * The input is scanned twice.
287 : * The error checking of input is minimal.
288 : */
289 : Datum
290 983912 : byteain(PG_FUNCTION_ARGS)
291 : {
292 983912 : char *inputText = PG_GETARG_CSTRING(0);
293 983912 : Node *escontext = fcinfo->context;
294 : char *tp;
295 : char *rp;
296 : int bc;
297 : bytea *result;
298 :
299 : /* Recognize hex input */
300 983912 : if (inputText[0] == '\\' && inputText[1] == 'x')
301 : {
302 111132 : size_t len = strlen(inputText);
303 :
304 111132 : bc = (len - 2) / 2 + VARHDRSZ; /* maximum possible length */
305 111132 : result = palloc(bc);
306 111132 : bc = hex_decode_safe(inputText + 2, len - 2, VARDATA(result),
307 : escontext);
308 111120 : SET_VARSIZE(result, bc + VARHDRSZ); /* actual length */
309 :
310 111120 : PG_RETURN_BYTEA_P(result);
311 : }
312 :
313 : /* Else, it's the traditional escaped style */
314 8069326 : for (bc = 0, tp = inputText; *tp != '\0'; bc++)
315 : {
316 7196558 : if (tp[0] != '\\')
317 7195458 : tp++;
318 1100 : else if ((tp[0] == '\\') &&
319 1100 : (tp[1] >= '0' && tp[1] <= '3') &&
320 1088 : (tp[2] >= '0' && tp[2] <= '7') &&
321 1088 : (tp[3] >= '0' && tp[3] <= '7'))
322 1088 : tp += 4;
323 12 : else if ((tp[0] == '\\') &&
324 12 : (tp[1] == '\\'))
325 0 : tp += 2;
326 : else
327 : {
328 : /*
329 : * one backslash, not followed by another or ### valid octal
330 : */
331 12 : ereturn(escontext, (Datum) 0,
332 : (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
333 : errmsg("invalid input syntax for type %s", "bytea")));
334 : }
335 : }
336 :
337 872768 : bc += VARHDRSZ;
338 :
339 872768 : result = (bytea *) palloc(bc);
340 872768 : SET_VARSIZE(result, bc);
341 :
342 872768 : tp = inputText;
343 872768 : rp = VARDATA(result);
344 8069284 : while (*tp != '\0')
345 : {
346 7196516 : if (tp[0] != '\\')
347 7195428 : *rp++ = *tp++;
348 1088 : else if ((tp[0] == '\\') &&
349 1088 : (tp[1] >= '0' && tp[1] <= '3') &&
350 1088 : (tp[2] >= '0' && tp[2] <= '7') &&
351 1088 : (tp[3] >= '0' && tp[3] <= '7'))
352 : {
353 1088 : bc = VAL(tp[1]);
354 1088 : bc <<= 3;
355 1088 : bc += VAL(tp[2]);
356 1088 : bc <<= 3;
357 1088 : *rp++ = bc + VAL(tp[3]);
358 :
359 1088 : tp += 4;
360 : }
361 0 : else if ((tp[0] == '\\') &&
362 0 : (tp[1] == '\\'))
363 : {
364 0 : *rp++ = '\\';
365 0 : tp += 2;
366 : }
367 : else
368 : {
369 : /*
370 : * We should never get here. The first pass should not allow it.
371 : */
372 0 : ereturn(escontext, (Datum) 0,
373 : (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
374 : errmsg("invalid input syntax for type %s", "bytea")));
375 : }
376 : }
377 :
378 872768 : PG_RETURN_BYTEA_P(result);
379 : }
380 :
381 : /*
382 : * byteaout - converts to printable representation of byte array
383 : *
384 : * In the traditional escaped format, non-printable characters are
385 : * printed as '\nnn' (octal) and '\' as '\\'.
386 : */
387 : Datum
388 158852 : byteaout(PG_FUNCTION_ARGS)
389 : {
390 158852 : bytea *vlena = PG_GETARG_BYTEA_PP(0);
391 : char *result;
392 : char *rp;
393 :
394 158852 : if (bytea_output == BYTEA_OUTPUT_HEX)
395 : {
396 : /* Print hex format */
397 158468 : rp = result = palloc(VARSIZE_ANY_EXHDR(vlena) * 2 + 2 + 1);
398 158468 : *rp++ = '\\';
399 158468 : *rp++ = 'x';
400 158468 : rp += hex_encode(VARDATA_ANY(vlena), VARSIZE_ANY_EXHDR(vlena), rp);
401 : }
402 384 : else if (bytea_output == BYTEA_OUTPUT_ESCAPE)
403 : {
404 : /* Print traditional escaped format */
405 : char *vp;
406 : uint64 len;
407 : int i;
408 :
409 384 : len = 1; /* empty string has 1 char */
410 384 : vp = VARDATA_ANY(vlena);
411 217660 : for (i = VARSIZE_ANY_EXHDR(vlena); i != 0; i--, vp++)
412 : {
413 217276 : if (*vp == '\\')
414 0 : len += 2;
415 217276 : else if ((unsigned char) *vp < 0x20 || (unsigned char) *vp > 0x7e)
416 498 : len += 4;
417 : else
418 216778 : len++;
419 : }
420 :
421 : /*
422 : * In principle len can't overflow uint32 if the input fit in 1GB, but
423 : * for safety let's check rather than relying on palloc's internal
424 : * check.
425 : */
426 384 : if (len > MaxAllocSize)
427 0 : ereport(ERROR,
428 : (errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
429 : errmsg_internal("result of bytea output conversion is too large")));
430 384 : rp = result = (char *) palloc(len);
431 :
432 384 : vp = VARDATA_ANY(vlena);
433 217660 : for (i = VARSIZE_ANY_EXHDR(vlena); i != 0; i--, vp++)
434 : {
435 217276 : if (*vp == '\\')
436 : {
437 0 : *rp++ = '\\';
438 0 : *rp++ = '\\';
439 : }
440 217276 : else if ((unsigned char) *vp < 0x20 || (unsigned char) *vp > 0x7e)
441 498 : {
442 : int val; /* holds unprintable chars */
443 :
444 498 : val = *vp;
445 498 : rp[0] = '\\';
446 498 : rp[3] = DIG(val & 07);
447 498 : val >>= 3;
448 498 : rp[2] = DIG(val & 07);
449 498 : val >>= 3;
450 498 : rp[1] = DIG(val & 03);
451 498 : rp += 4;
452 : }
453 : else
454 216778 : *rp++ = *vp;
455 : }
456 : }
457 : else
458 : {
459 0 : elog(ERROR, "unrecognized bytea_output setting: %d",
460 : bytea_output);
461 : rp = result = NULL; /* keep compiler quiet */
462 : }
463 158852 : *rp = '\0';
464 158852 : PG_RETURN_CSTRING(result);
465 : }
466 :
467 : /*
468 : * bytearecv - converts external binary format to bytea
469 : */
470 : Datum
471 107710 : bytearecv(PG_FUNCTION_ARGS)
472 : {
473 107710 : StringInfo buf = (StringInfo) PG_GETARG_POINTER(0);
474 : bytea *result;
475 : int nbytes;
476 :
477 107710 : nbytes = buf->len - buf->cursor;
478 107710 : result = (bytea *) palloc(nbytes + VARHDRSZ);
479 107710 : SET_VARSIZE(result, nbytes + VARHDRSZ);
480 107710 : pq_copymsgbytes(buf, VARDATA(result), nbytes);
481 107710 : PG_RETURN_BYTEA_P(result);
482 : }
483 :
484 : /*
485 : * byteasend - converts bytea to binary format
486 : *
487 : * This is a special case: just copy the input...
488 : */
489 : Datum
490 68988 : byteasend(PG_FUNCTION_ARGS)
491 : {
492 68988 : bytea *vlena = PG_GETARG_BYTEA_P_COPY(0);
493 :
494 68988 : PG_RETURN_BYTEA_P(vlena);
495 : }
496 :
497 : Datum
498 258774 : bytea_string_agg_transfn(PG_FUNCTION_ARGS)
499 : {
500 : StringInfo state;
501 :
502 258774 : state = PG_ARGISNULL(0) ? NULL : (StringInfo) PG_GETARG_POINTER(0);
503 :
504 : /* Append the value unless null, preceding it with the delimiter. */
505 258774 : if (!PG_ARGISNULL(1))
506 : {
507 243774 : bytea *value = PG_GETARG_BYTEA_PP(1);
508 243774 : bool isfirst = false;
509 :
510 : /*
511 : * You might think we can just throw away the first delimiter, however
512 : * we must keep it as we may be a parallel worker doing partial
513 : * aggregation building a state to send to the main process. We need
514 : * to keep the delimiter of every aggregation so that the combine
515 : * function can properly join up the strings of two separately
516 : * partially aggregated results. The first delimiter is only stripped
517 : * off in the final function. To know how much to strip off the front
518 : * of the string, we store the length of the first delimiter in the
519 : * StringInfo's cursor field, which we don't otherwise need here.
520 : */
521 243774 : if (state == NULL)
522 : {
523 148 : state = makeStringAggState(fcinfo);
524 148 : isfirst = true;
525 : }
526 :
527 243774 : if (!PG_ARGISNULL(2))
528 : {
529 243762 : bytea *delim = PG_GETARG_BYTEA_PP(2);
530 :
531 243762 : appendBinaryStringInfo(state, VARDATA_ANY(delim),
532 243762 : VARSIZE_ANY_EXHDR(delim));
533 243762 : if (isfirst)
534 142 : state->cursor = VARSIZE_ANY_EXHDR(delim);
535 : }
536 :
537 243774 : appendBinaryStringInfo(state, VARDATA_ANY(value),
538 243774 : VARSIZE_ANY_EXHDR(value));
539 : }
540 :
541 : /*
542 : * The transition type for string_agg() is declared to be "internal",
543 : * which is a pass-by-value type the same size as a pointer.
544 : */
545 258774 : if (state)
546 258738 : PG_RETURN_POINTER(state);
547 36 : PG_RETURN_NULL();
548 : }
549 :
550 : Datum
551 154 : bytea_string_agg_finalfn(PG_FUNCTION_ARGS)
552 : {
553 : StringInfo state;
554 :
555 : /* cannot be called directly because of internal-type argument */
556 : Assert(AggCheckCallContext(fcinfo, NULL));
557 :
558 154 : state = PG_ARGISNULL(0) ? NULL : (StringInfo) PG_GETARG_POINTER(0);
559 :
560 154 : if (state != NULL)
561 : {
562 : /* As per comment in transfn, strip data before the cursor position */
563 : bytea *result;
564 148 : int strippedlen = state->len - state->cursor;
565 :
566 148 : result = (bytea *) palloc(strippedlen + VARHDRSZ);
567 148 : SET_VARSIZE(result, strippedlen + VARHDRSZ);
568 148 : memcpy(VARDATA(result), &state->data[state->cursor], strippedlen);
569 148 : PG_RETURN_BYTEA_P(result);
570 : }
571 : else
572 6 : PG_RETURN_NULL();
573 : }
574 :
575 : /*
576 : * textin - converts cstring to internal representation
577 : */
578 : Datum
579 20932254 : textin(PG_FUNCTION_ARGS)
580 : {
581 20932254 : char *inputText = PG_GETARG_CSTRING(0);
582 :
583 20932254 : PG_RETURN_TEXT_P(cstring_to_text(inputText));
584 : }
585 :
586 : /*
587 : * textout - converts internal representation to cstring
588 : */
589 : Datum
590 7102244 : textout(PG_FUNCTION_ARGS)
591 : {
592 7102244 : Datum txt = PG_GETARG_DATUM(0);
593 :
594 7102244 : PG_RETURN_CSTRING(TextDatumGetCString(txt));
595 : }
596 :
597 : /*
598 : * textrecv - converts external binary format to text
599 : */
600 : Datum
601 48 : textrecv(PG_FUNCTION_ARGS)
602 : {
603 48 : StringInfo buf = (StringInfo) PG_GETARG_POINTER(0);
604 : text *result;
605 : char *str;
606 : int nbytes;
607 :
608 48 : str = pq_getmsgtext(buf, buf->len - buf->cursor, &nbytes);
609 :
610 48 : result = cstring_to_text_with_len(str, nbytes);
611 48 : pfree(str);
612 48 : PG_RETURN_TEXT_P(result);
613 : }
614 :
615 : /*
616 : * textsend - converts text to binary format
617 : */
618 : Datum
619 4940 : textsend(PG_FUNCTION_ARGS)
620 : {
621 4940 : text *t = PG_GETARG_TEXT_PP(0);
622 : StringInfoData buf;
623 :
624 4940 : pq_begintypsend(&buf);
625 4940 : pq_sendtext(&buf, VARDATA_ANY(t), VARSIZE_ANY_EXHDR(t));
626 4940 : PG_RETURN_BYTEA_P(pq_endtypsend(&buf));
627 : }
628 :
629 :
630 : /*
631 : * unknownin - converts cstring to internal representation
632 : */
633 : Datum
634 0 : unknownin(PG_FUNCTION_ARGS)
635 : {
636 0 : char *str = PG_GETARG_CSTRING(0);
637 :
638 : /* representation is same as cstring */
639 0 : PG_RETURN_CSTRING(pstrdup(str));
640 : }
641 :
642 : /*
643 : * unknownout - converts internal representation to cstring
644 : */
645 : Datum
646 682 : unknownout(PG_FUNCTION_ARGS)
647 : {
648 : /* representation is same as cstring */
649 682 : char *str = PG_GETARG_CSTRING(0);
650 :
651 682 : PG_RETURN_CSTRING(pstrdup(str));
652 : }
653 :
654 : /*
655 : * unknownrecv - converts external binary format to unknown
656 : */
657 : Datum
658 0 : unknownrecv(PG_FUNCTION_ARGS)
659 : {
660 0 : StringInfo buf = (StringInfo) PG_GETARG_POINTER(0);
661 : char *str;
662 : int nbytes;
663 :
664 0 : str = pq_getmsgtext(buf, buf->len - buf->cursor, &nbytes);
665 : /* representation is same as cstring */
666 0 : PG_RETURN_CSTRING(str);
667 : }
668 :
669 : /*
670 : * unknownsend - converts unknown to binary format
671 : */
672 : Datum
673 0 : unknownsend(PG_FUNCTION_ARGS)
674 : {
675 : /* representation is same as cstring */
676 0 : char *str = PG_GETARG_CSTRING(0);
677 : StringInfoData buf;
678 :
679 0 : pq_begintypsend(&buf);
680 0 : pq_sendtext(&buf, str, strlen(str));
681 0 : PG_RETURN_BYTEA_P(pq_endtypsend(&buf));
682 : }
683 :
684 :
685 : /* ========== PUBLIC ROUTINES ========== */
686 :
687 : /*
688 : * textlen -
689 : * returns the logical length of a text*
690 : * (which is less than the VARSIZE of the text*)
691 : */
692 : Datum
693 430604 : textlen(PG_FUNCTION_ARGS)
694 : {
695 430604 : Datum str = PG_GETARG_DATUM(0);
696 :
697 : /* try to avoid decompressing argument */
698 430604 : PG_RETURN_INT32(text_length(str));
699 : }
700 :
701 : /*
702 : * text_length -
703 : * Does the real work for textlen()
704 : *
705 : * This is broken out so it can be called directly by other string processing
706 : * functions. Note that the argument is passed as a Datum, to indicate that
707 : * it may still be in compressed form. We can avoid decompressing it at all
708 : * in some cases.
709 : */
710 : static int32
711 430616 : text_length(Datum str)
712 : {
713 : /* fastpath when max encoding length is one */
714 430616 : if (pg_database_encoding_max_length() == 1)
715 20 : PG_RETURN_INT32(toast_raw_datum_size(str) - VARHDRSZ);
716 : else
717 : {
718 430596 : text *t = DatumGetTextPP(str);
719 :
720 430596 : PG_RETURN_INT32(pg_mbstrlen_with_len(VARDATA_ANY(t),
721 : VARSIZE_ANY_EXHDR(t)));
722 : }
723 : }
724 :
725 : /*
726 : * textoctetlen -
727 : * returns the physical length of a text*
728 : * (which is less than the VARSIZE of the text*)
729 : */
730 : Datum
731 70 : textoctetlen(PG_FUNCTION_ARGS)
732 : {
733 70 : Datum str = PG_GETARG_DATUM(0);
734 :
735 : /* We need not detoast the input at all */
736 70 : PG_RETURN_INT32(toast_raw_datum_size(str) - VARHDRSZ);
737 : }
738 :
739 : /*
740 : * textcat -
741 : * takes two text* and returns a text* that is the concatenation of
742 : * the two.
743 : *
744 : * Rewritten by Sapa, sapa@hq.icb.chel.su. 8-Jul-96.
745 : * Updated by Thomas, Thomas.Lockhart@jpl.nasa.gov 1997-07-10.
746 : * Allocate space for output in all cases.
747 : * XXX - thomas 1997-07-10
748 : */
749 : Datum
750 1780382 : textcat(PG_FUNCTION_ARGS)
751 : {
752 1780382 : text *t1 = PG_GETARG_TEXT_PP(0);
753 1780382 : text *t2 = PG_GETARG_TEXT_PP(1);
754 :
755 1780382 : PG_RETURN_TEXT_P(text_catenate(t1, t2));
756 : }
757 :
758 : /*
759 : * text_catenate
760 : * Guts of textcat(), broken out so it can be used by other functions
761 : *
762 : * Arguments can be in short-header form, but not compressed or out-of-line
763 : */
764 : static text *
765 1780462 : text_catenate(text *t1, text *t2)
766 : {
767 : text *result;
768 : int len1,
769 : len2,
770 : len;
771 : char *ptr;
772 :
773 1780462 : len1 = VARSIZE_ANY_EXHDR(t1);
774 1780462 : len2 = VARSIZE_ANY_EXHDR(t2);
775 :
776 : /* paranoia ... probably should throw error instead? */
777 1780462 : if (len1 < 0)
778 0 : len1 = 0;
779 1780462 : if (len2 < 0)
780 0 : len2 = 0;
781 :
782 1780462 : len = len1 + len2 + VARHDRSZ;
783 1780462 : result = (text *) palloc(len);
784 :
785 : /* Set size of result string... */
786 1780462 : SET_VARSIZE(result, len);
787 :
788 : /* Fill data field of result string... */
789 1780462 : ptr = VARDATA(result);
790 1780462 : if (len1 > 0)
791 1779644 : memcpy(ptr, VARDATA_ANY(t1), len1);
792 1780462 : if (len2 > 0)
793 1780252 : memcpy(ptr + len1, VARDATA_ANY(t2), len2);
794 :
795 1780462 : return result;
796 : }
797 :
798 : /*
799 : * charlen_to_bytelen()
800 : * Compute the number of bytes occupied by n characters starting at *p
801 : *
802 : * It is caller's responsibility that there actually are n characters;
803 : * the string need not be null-terminated.
804 : */
805 : static int
806 13222 : charlen_to_bytelen(const char *p, int n)
807 : {
808 13222 : if (pg_database_encoding_max_length() == 1)
809 : {
810 : /* Optimization for single-byte encodings */
811 168 : return n;
812 : }
813 : else
814 : {
815 : const char *s;
816 :
817 5947182 : for (s = p; n > 0; n--)
818 5934128 : s += pg_mblen(s);
819 :
820 13054 : return s - p;
821 : }
822 : }
823 :
824 : /*
825 : * text_substr()
826 : * Return a substring starting at the specified position.
827 : * - thomas 1997-12-31
828 : *
829 : * Input:
830 : * - string
831 : * - starting position (is one-based)
832 : * - string length
833 : *
834 : * If the starting position is zero or less, then return from the start of the string
835 : * adjusting the length to be consistent with the "negative start" per SQL.
836 : * If the length is less than zero, return the remaining string.
837 : *
838 : * Added multibyte support.
839 : * - Tatsuo Ishii 1998-4-21
840 : * Changed behavior if starting position is less than one to conform to SQL behavior.
841 : * Formerly returned the entire string; now returns a portion.
842 : * - Thomas Lockhart 1998-12-10
843 : * Now uses faster TOAST-slicing interface
844 : * - John Gray 2002-02-22
845 : * Remove "#ifdef MULTIBYTE" and test for encoding_max_length instead. Change
846 : * behaviors conflicting with SQL to meet SQL (if E = S + L < S throw
847 : * error; if E < 1, return '', not entire string). Fixed MB related bug when
848 : * S > LC and < LC + 4 sometimes garbage characters are returned.
849 : * - Joe Conway 2002-08-10
850 : */
851 : Datum
852 586764 : text_substr(PG_FUNCTION_ARGS)
853 : {
854 586764 : PG_RETURN_TEXT_P(text_substring(PG_GETARG_DATUM(0),
855 : PG_GETARG_INT32(1),
856 : PG_GETARG_INT32(2),
857 : false));
858 : }
859 :
860 : /*
861 : * text_substr_no_len -
862 : * Wrapper to avoid opr_sanity failure due to
863 : * one function accepting a different number of args.
864 : */
865 : Datum
866 36 : text_substr_no_len(PG_FUNCTION_ARGS)
867 : {
868 36 : PG_RETURN_TEXT_P(text_substring(PG_GETARG_DATUM(0),
869 : PG_GETARG_INT32(1),
870 : -1, true));
871 : }
872 :
873 : /*
874 : * text_substring -
875 : * Does the real work for text_substr() and text_substr_no_len()
876 : *
877 : * This is broken out so it can be called directly by other string processing
878 : * functions. Note that the argument is passed as a Datum, to indicate that
879 : * it may still be in compressed/toasted form. We can avoid detoasting all
880 : * of it in some cases.
881 : *
882 : * The result is always a freshly palloc'd datum.
883 : */
884 : static text *
885 626672 : text_substring(Datum str, int32 start, int32 length, bool length_not_specified)
886 : {
887 626672 : int32 eml = pg_database_encoding_max_length();
888 626672 : int32 S = start; /* start position */
889 : int32 S1; /* adjusted start position */
890 : int32 L1; /* adjusted substring length */
891 : int32 E; /* end position */
892 :
893 : /*
894 : * SQL99 says S can be zero or negative (which we don't document), but we
895 : * still must fetch from the start of the string.
896 : * https://www.postgresql.org/message-id/170905442373.643.11536838320909376197%40wrigleys.postgresql.org
897 : */
898 626672 : S1 = Max(S, 1);
899 :
900 : /* life is easy if the encoding max length is 1 */
901 626672 : if (eml == 1)
902 : {
903 22 : if (length_not_specified) /* special case - get length to end of
904 : * string */
905 0 : L1 = -1;
906 22 : else if (length < 0)
907 : {
908 : /* SQL99 says to throw an error for E < S, i.e., negative length */
909 0 : ereport(ERROR,
910 : (errcode(ERRCODE_SUBSTRING_ERROR),
911 : errmsg("negative substring length not allowed")));
912 : L1 = -1; /* silence stupider compilers */
913 : }
914 22 : else if (pg_add_s32_overflow(S, length, &E))
915 : {
916 : /*
917 : * L could be large enough for S + L to overflow, in which case
918 : * the substring must run to end of string.
919 : */
920 0 : L1 = -1;
921 : }
922 : else
923 : {
924 : /*
925 : * A zero or negative value for the end position can happen if the
926 : * start was negative or one. SQL99 says to return a zero-length
927 : * string.
928 : */
929 22 : if (E < 1)
930 0 : return cstring_to_text("");
931 :
932 22 : L1 = E - S1;
933 : }
934 :
935 : /*
936 : * If the start position is past the end of the string, SQL99 says to
937 : * return a zero-length string -- DatumGetTextPSlice() will do that
938 : * for us. We need only convert S1 to zero-based starting position.
939 : */
940 22 : return DatumGetTextPSlice(str, S1 - 1, L1);
941 : }
942 626650 : else if (eml > 1)
943 : {
944 : /*
945 : * When encoding max length is > 1, we can't get LC without
946 : * detoasting, so we'll grab a conservatively large slice now and go
947 : * back later to do the right thing
948 : */
949 : int32 slice_start;
950 : int32 slice_size;
951 : int32 slice_strlen;
952 : text *slice;
953 : int32 E1;
954 : int32 i;
955 : char *p;
956 : char *s;
957 : text *ret;
958 :
959 : /*
960 : * We need to start at position zero because there is no way to know
961 : * in advance which byte offset corresponds to the supplied start
962 : * position.
963 : */
964 626650 : slice_start = 0;
965 :
966 626650 : if (length_not_specified) /* special case - get length to end of
967 : * string */
968 76 : slice_size = L1 = -1;
969 626574 : else if (length < 0)
970 : {
971 : /* SQL99 says to throw an error for E < S, i.e., negative length */
972 12 : ereport(ERROR,
973 : (errcode(ERRCODE_SUBSTRING_ERROR),
974 : errmsg("negative substring length not allowed")));
975 : slice_size = L1 = -1; /* silence stupider compilers */
976 : }
977 626562 : else if (pg_add_s32_overflow(S, length, &E))
978 : {
979 : /*
980 : * L could be large enough for S + L to overflow, in which case
981 : * the substring must run to end of string.
982 : */
983 6 : slice_size = L1 = -1;
984 : }
985 : else
986 : {
987 : /*
988 : * A zero or negative value for the end position can happen if the
989 : * start was negative or one. SQL99 says to return a zero-length
990 : * string.
991 : */
992 626556 : if (E < 1)
993 0 : return cstring_to_text("");
994 :
995 : /*
996 : * if E is past the end of the string, the tuple toaster will
997 : * truncate the length for us
998 : */
999 626556 : L1 = E - S1;
1000 :
1001 : /*
1002 : * Total slice size in bytes can't be any longer than the start
1003 : * position plus substring length times the encoding max length.
1004 : * If that overflows, we can just use -1.
1005 : */
1006 626556 : if (pg_mul_s32_overflow(E, eml, &slice_size))
1007 6 : slice_size = -1;
1008 : }
1009 :
1010 : /*
1011 : * If we're working with an untoasted source, no need to do an extra
1012 : * copying step.
1013 : */
1014 626638 : if (VARATT_IS_COMPRESSED(DatumGetPointer(str)) ||
1015 626584 : VARATT_IS_EXTERNAL(DatumGetPointer(str)))
1016 324 : slice = DatumGetTextPSlice(str, slice_start, slice_size);
1017 : else
1018 626314 : slice = (text *) DatumGetPointer(str);
1019 :
1020 : /* see if we got back an empty string */
1021 626638 : if (VARSIZE_ANY_EXHDR(slice) == 0)
1022 : {
1023 0 : if (slice != (text *) DatumGetPointer(str))
1024 0 : pfree(slice);
1025 0 : return cstring_to_text("");
1026 : }
1027 :
1028 : /* Now we can get the actual length of the slice in MB characters */
1029 626638 : slice_strlen = pg_mbstrlen_with_len(VARDATA_ANY(slice),
1030 626638 : VARSIZE_ANY_EXHDR(slice));
1031 :
1032 : /*
1033 : * Check that the start position wasn't > slice_strlen. If so, SQL99
1034 : * says to return a zero-length string.
1035 : */
1036 626638 : if (S1 > slice_strlen)
1037 : {
1038 22 : if (slice != (text *) DatumGetPointer(str))
1039 0 : pfree(slice);
1040 22 : return cstring_to_text("");
1041 : }
1042 :
1043 : /*
1044 : * Adjust L1 and E1 now that we know the slice string length. Again
1045 : * remember that S1 is one based, and slice_start is zero based.
1046 : */
1047 626616 : if (L1 > -1)
1048 626556 : E1 = Min(S1 + L1, slice_start + 1 + slice_strlen);
1049 : else
1050 60 : E1 = slice_start + 1 + slice_strlen;
1051 :
1052 : /*
1053 : * Find the start position in the slice; remember S1 is not zero based
1054 : */
1055 626616 : p = VARDATA_ANY(slice);
1056 5447398 : for (i = 0; i < S1 - 1; i++)
1057 4820782 : p += pg_mblen(p);
1058 :
1059 : /* hang onto a pointer to our start position */
1060 626616 : s = p;
1061 :
1062 : /*
1063 : * Count the actual bytes used by the substring of the requested
1064 : * length.
1065 : */
1066 9777584 : for (i = S1; i < E1; i++)
1067 9150968 : p += pg_mblen(p);
1068 :
1069 626616 : ret = (text *) palloc(VARHDRSZ + (p - s));
1070 626616 : SET_VARSIZE(ret, VARHDRSZ + (p - s));
1071 626616 : memcpy(VARDATA(ret), s, (p - s));
1072 :
1073 626616 : if (slice != (text *) DatumGetPointer(str))
1074 324 : pfree(slice);
1075 :
1076 626616 : return ret;
1077 : }
1078 : else
1079 0 : elog(ERROR, "invalid backend encoding: encoding max length < 1");
1080 :
1081 : /* not reached: suppress compiler warning */
1082 : return NULL;
1083 : }
1084 :
1085 : /*
1086 : * textoverlay
1087 : * Replace specified substring of first string with second
1088 : *
1089 : * The SQL standard defines OVERLAY() in terms of substring and concatenation.
1090 : * This code is a direct implementation of what the standard says.
1091 : */
1092 : Datum
1093 28 : textoverlay(PG_FUNCTION_ARGS)
1094 : {
1095 28 : text *t1 = PG_GETARG_TEXT_PP(0);
1096 28 : text *t2 = PG_GETARG_TEXT_PP(1);
1097 28 : int sp = PG_GETARG_INT32(2); /* substring start position */
1098 28 : int sl = PG_GETARG_INT32(3); /* substring length */
1099 :
1100 28 : PG_RETURN_TEXT_P(text_overlay(t1, t2, sp, sl));
1101 : }
1102 :
1103 : Datum
1104 12 : textoverlay_no_len(PG_FUNCTION_ARGS)
1105 : {
1106 12 : text *t1 = PG_GETARG_TEXT_PP(0);
1107 12 : text *t2 = PG_GETARG_TEXT_PP(1);
1108 12 : int sp = PG_GETARG_INT32(2); /* substring start position */
1109 : int sl;
1110 :
1111 12 : sl = text_length(PointerGetDatum(t2)); /* defaults to length(t2) */
1112 12 : PG_RETURN_TEXT_P(text_overlay(t1, t2, sp, sl));
1113 : }
1114 :
1115 : static text *
1116 40 : text_overlay(text *t1, text *t2, int sp, int sl)
1117 : {
1118 : text *result;
1119 : text *s1;
1120 : text *s2;
1121 : int sp_pl_sl;
1122 :
1123 : /*
1124 : * Check for possible integer-overflow cases. For negative sp, throw a
1125 : * "substring length" error because that's what should be expected
1126 : * according to the spec's definition of OVERLAY().
1127 : */
1128 40 : if (sp <= 0)
1129 0 : ereport(ERROR,
1130 : (errcode(ERRCODE_SUBSTRING_ERROR),
1131 : errmsg("negative substring length not allowed")));
1132 40 : if (pg_add_s32_overflow(sp, sl, &sp_pl_sl))
1133 0 : ereport(ERROR,
1134 : (errcode(ERRCODE_NUMERIC_VALUE_OUT_OF_RANGE),
1135 : errmsg("integer out of range")));
1136 :
1137 40 : s1 = text_substring(PointerGetDatum(t1), 1, sp - 1, false);
1138 40 : s2 = text_substring(PointerGetDatum(t1), sp_pl_sl, -1, true);
1139 40 : result = text_catenate(s1, t2);
1140 40 : result = text_catenate(result, s2);
1141 :
1142 40 : return result;
1143 : }
1144 :
1145 : /*
1146 : * textpos -
1147 : * Return the position of the specified substring.
1148 : * Implements the SQL POSITION() function.
1149 : * Ref: A Guide To The SQL Standard, Date & Darwen, 1997
1150 : * - thomas 1997-07-27
1151 : */
1152 : Datum
1153 106 : textpos(PG_FUNCTION_ARGS)
1154 : {
1155 106 : text *str = PG_GETARG_TEXT_PP(0);
1156 106 : text *search_str = PG_GETARG_TEXT_PP(1);
1157 :
1158 106 : PG_RETURN_INT32((int32) text_position(str, search_str, PG_GET_COLLATION()));
1159 : }
1160 :
1161 : /*
1162 : * text_position -
1163 : * Does the real work for textpos()
1164 : *
1165 : * Inputs:
1166 : * t1 - string to be searched
1167 : * t2 - pattern to match within t1
1168 : * Result:
1169 : * Character index of the first matched char, starting from 1,
1170 : * or 0 if no match.
1171 : *
1172 : * This is broken out so it can be called directly by other string processing
1173 : * functions.
1174 : */
1175 : static int
1176 106 : text_position(text *t1, text *t2, Oid collid)
1177 : {
1178 : TextPositionState state;
1179 : int result;
1180 :
1181 : /* Empty needle always matches at position 1 */
1182 106 : if (VARSIZE_ANY_EXHDR(t2) < 1)
1183 12 : return 1;
1184 :
1185 : /* Otherwise, can't match if haystack is shorter than needle */
1186 94 : if (VARSIZE_ANY_EXHDR(t1) < VARSIZE_ANY_EXHDR(t2))
1187 22 : return 0;
1188 :
1189 72 : text_position_setup(t1, t2, collid, &state);
1190 72 : if (!text_position_next(&state))
1191 24 : result = 0;
1192 : else
1193 48 : result = text_position_get_match_pos(&state);
1194 72 : text_position_cleanup(&state);
1195 72 : return result;
1196 : }
1197 :
1198 :
1199 : /*
1200 : * text_position_setup, text_position_next, text_position_cleanup -
1201 : * Component steps of text_position()
1202 : *
1203 : * These are broken out so that a string can be efficiently searched for
1204 : * multiple occurrences of the same pattern. text_position_next may be
1205 : * called multiple times, and it advances to the next match on each call.
1206 : * text_position_get_match_ptr() and text_position_get_match_pos() return
1207 : * a pointer or 1-based character position of the last match, respectively.
1208 : *
1209 : * The "state" variable is normally just a local variable in the caller.
1210 : *
1211 : * NOTE: text_position_next skips over the matched portion. For example,
1212 : * searching for "xx" in "xxx" returns only one match, not two.
1213 : */
1214 :
1215 : static void
1216 1312 : text_position_setup(text *t1, text *t2, Oid collid, TextPositionState *state)
1217 : {
1218 1312 : int len1 = VARSIZE_ANY_EXHDR(t1);
1219 1312 : int len2 = VARSIZE_ANY_EXHDR(t2);
1220 1312 : pg_locale_t mylocale = 0;
1221 :
1222 1312 : check_collation_set(collid);
1223 :
1224 1312 : if (!lc_collate_is_c(collid))
1225 210 : mylocale = pg_newlocale_from_collation(collid);
1226 :
1227 1312 : if (!pg_locale_deterministic(mylocale))
1228 12 : ereport(ERROR,
1229 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1230 : errmsg("nondeterministic collations are not supported for substring searches")));
1231 :
1232 : Assert(len1 > 0);
1233 : Assert(len2 > 0);
1234 :
1235 : /*
1236 : * Even with a multi-byte encoding, we perform the search using the raw
1237 : * byte sequence, ignoring multibyte issues. For UTF-8, that works fine,
1238 : * because in UTF-8 the byte sequence of one character cannot contain
1239 : * another character. For other multi-byte encodings, we do the search
1240 : * initially as a simple byte search, ignoring multibyte issues, but
1241 : * verify afterwards that the match we found is at a character boundary,
1242 : * and continue the search if it was a false match.
1243 : */
1244 1300 : if (pg_database_encoding_max_length() == 1)
1245 104 : state->is_multibyte_char_in_char = false;
1246 1196 : else if (GetDatabaseEncoding() == PG_UTF8)
1247 1196 : state->is_multibyte_char_in_char = false;
1248 : else
1249 0 : state->is_multibyte_char_in_char = true;
1250 :
1251 1300 : state->str1 = VARDATA_ANY(t1);
1252 1300 : state->str2 = VARDATA_ANY(t2);
1253 1300 : state->len1 = len1;
1254 1300 : state->len2 = len2;
1255 1300 : state->last_match = NULL;
1256 1300 : state->refpoint = state->str1;
1257 1300 : state->refpos = 0;
1258 :
1259 : /*
1260 : * Prepare the skip table for Boyer-Moore-Horspool searching. In these
1261 : * notes we use the terminology that the "haystack" is the string to be
1262 : * searched (t1) and the "needle" is the pattern being sought (t2).
1263 : *
1264 : * If the needle is empty or bigger than the haystack then there is no
1265 : * point in wasting cycles initializing the table. We also choose not to
1266 : * use B-M-H for needles of length 1, since the skip table can't possibly
1267 : * save anything in that case.
1268 : */
1269 1300 : if (len1 >= len2 && len2 > 1)
1270 : {
1271 1092 : int searchlength = len1 - len2;
1272 : int skiptablemask;
1273 : int last;
1274 : int i;
1275 1092 : const char *str2 = state->str2;
1276 :
1277 : /*
1278 : * First we must determine how much of the skip table to use. The
1279 : * declaration of TextPositionState allows up to 256 elements, but for
1280 : * short search problems we don't really want to have to initialize so
1281 : * many elements --- it would take too long in comparison to the
1282 : * actual search time. So we choose a useful skip table size based on
1283 : * the haystack length minus the needle length. The closer the needle
1284 : * length is to the haystack length the less useful skipping becomes.
1285 : *
1286 : * Note: since we use bit-masking to select table elements, the skip
1287 : * table size MUST be a power of 2, and so the mask must be 2^N-1.
1288 : */
1289 1092 : if (searchlength < 16)
1290 54 : skiptablemask = 3;
1291 1038 : else if (searchlength < 64)
1292 16 : skiptablemask = 7;
1293 1022 : else if (searchlength < 128)
1294 14 : skiptablemask = 15;
1295 1008 : else if (searchlength < 512)
1296 202 : skiptablemask = 31;
1297 806 : else if (searchlength < 2048)
1298 662 : skiptablemask = 63;
1299 144 : else if (searchlength < 4096)
1300 74 : skiptablemask = 127;
1301 : else
1302 70 : skiptablemask = 255;
1303 1092 : state->skiptablemask = skiptablemask;
1304 :
1305 : /*
1306 : * Initialize the skip table. We set all elements to the needle
1307 : * length, since this is the correct skip distance for any character
1308 : * not found in the needle.
1309 : */
1310 77884 : for (i = 0; i <= skiptablemask; i++)
1311 76792 : state->skiptable[i] = len2;
1312 :
1313 : /*
1314 : * Now examine the needle. For each character except the last one,
1315 : * set the corresponding table element to the appropriate skip
1316 : * distance. Note that when two characters share the same skip table
1317 : * entry, the one later in the needle must determine the skip
1318 : * distance.
1319 : */
1320 1092 : last = len2 - 1;
1321 :
1322 14876 : for (i = 0; i < last; i++)
1323 13784 : state->skiptable[(unsigned char) str2[i] & skiptablemask] = last - i;
1324 : }
1325 1300 : }
1326 :
1327 : /*
1328 : * Advance to the next match, starting from the end of the previous match
1329 : * (or the beginning of the string, on first call). Returns true if a match
1330 : * is found.
1331 : *
1332 : * Note that this refuses to match an empty-string needle. Most callers
1333 : * will have handled that case specially and we'll never see it here.
1334 : */
1335 : static bool
1336 6184 : text_position_next(TextPositionState *state)
1337 : {
1338 6184 : int needle_len = state->len2;
1339 : char *start_ptr;
1340 : char *matchptr;
1341 :
1342 6184 : if (needle_len <= 0)
1343 0 : return false; /* result for empty pattern */
1344 :
1345 : /* Start from the point right after the previous match. */
1346 6184 : if (state->last_match)
1347 4872 : start_ptr = state->last_match + needle_len;
1348 : else
1349 1312 : start_ptr = state->str1;
1350 :
1351 6184 : retry:
1352 6184 : matchptr = text_position_next_internal(start_ptr, state);
1353 :
1354 6184 : if (!matchptr)
1355 1240 : return false;
1356 :
1357 : /*
1358 : * Found a match for the byte sequence. If this is a multibyte encoding,
1359 : * where one character's byte sequence can appear inside a longer
1360 : * multi-byte character, we need to verify that the match was at a
1361 : * character boundary, not in the middle of a multi-byte character.
1362 : */
1363 4944 : if (state->is_multibyte_char_in_char)
1364 : {
1365 : /* Walk one character at a time, until we reach the match. */
1366 :
1367 : /* the search should never move backwards. */
1368 : Assert(state->refpoint <= matchptr);
1369 :
1370 0 : while (state->refpoint < matchptr)
1371 : {
1372 : /* step to next character. */
1373 0 : state->refpoint += pg_mblen(state->refpoint);
1374 0 : state->refpos++;
1375 :
1376 : /*
1377 : * If we stepped over the match's start position, then it was a
1378 : * false positive, where the byte sequence appeared in the middle
1379 : * of a multi-byte character. Skip it, and continue the search at
1380 : * the next character boundary.
1381 : */
1382 0 : if (state->refpoint > matchptr)
1383 : {
1384 0 : start_ptr = state->refpoint;
1385 0 : goto retry;
1386 : }
1387 : }
1388 : }
1389 :
1390 4944 : state->last_match = matchptr;
1391 4944 : return true;
1392 : }
1393 :
1394 : /*
1395 : * Subroutine of text_position_next(). This searches for the raw byte
1396 : * sequence, ignoring any multi-byte encoding issues. Returns the first
1397 : * match starting at 'start_ptr', or NULL if no match is found.
1398 : */
1399 : static char *
1400 6184 : text_position_next_internal(char *start_ptr, TextPositionState *state)
1401 : {
1402 6184 : int haystack_len = state->len1;
1403 6184 : int needle_len = state->len2;
1404 6184 : int skiptablemask = state->skiptablemask;
1405 6184 : const char *haystack = state->str1;
1406 6184 : const char *needle = state->str2;
1407 6184 : const char *haystack_end = &haystack[haystack_len];
1408 : const char *hptr;
1409 :
1410 : Assert(start_ptr >= haystack && start_ptr <= haystack_end);
1411 :
1412 6184 : if (needle_len == 1)
1413 : {
1414 : /* No point in using B-M-H for a one-character needle */
1415 754 : char nchar = *needle;
1416 :
1417 754 : hptr = start_ptr;
1418 5758 : while (hptr < haystack_end)
1419 : {
1420 5592 : if (*hptr == nchar)
1421 588 : return (char *) hptr;
1422 5004 : hptr++;
1423 : }
1424 : }
1425 : else
1426 : {
1427 5430 : const char *needle_last = &needle[needle_len - 1];
1428 :
1429 : /* Start at startpos plus the length of the needle */
1430 5430 : hptr = start_ptr + needle_len - 1;
1431 142668 : while (hptr < haystack_end)
1432 : {
1433 : /* Match the needle scanning *backward* */
1434 : const char *nptr;
1435 : const char *p;
1436 :
1437 141594 : nptr = needle_last;
1438 141594 : p = hptr;
1439 206866 : while (*nptr == *p)
1440 : {
1441 : /* Matched it all? If so, return 1-based position */
1442 69628 : if (nptr == needle)
1443 4356 : return (char *) p;
1444 65272 : nptr--, p--;
1445 : }
1446 :
1447 : /*
1448 : * No match, so use the haystack char at hptr to decide how far to
1449 : * advance. If the needle had any occurrence of that character
1450 : * (or more precisely, one sharing the same skiptable entry)
1451 : * before its last character, then we advance far enough to align
1452 : * the last such needle character with that haystack position.
1453 : * Otherwise we can advance by the whole needle length.
1454 : */
1455 137238 : hptr += state->skiptable[(unsigned char) *hptr & skiptablemask];
1456 : }
1457 : }
1458 :
1459 1240 : return 0; /* not found */
1460 : }
1461 :
1462 : /*
1463 : * Return a pointer to the current match.
1464 : *
1465 : * The returned pointer points into the original haystack string.
1466 : */
1467 : static char *
1468 4866 : text_position_get_match_ptr(TextPositionState *state)
1469 : {
1470 4866 : return state->last_match;
1471 : }
1472 :
1473 : /*
1474 : * Return the offset of the current match.
1475 : *
1476 : * The offset is in characters, 1-based.
1477 : */
1478 : static int
1479 48 : text_position_get_match_pos(TextPositionState *state)
1480 : {
1481 : /* Convert the byte position to char position. */
1482 96 : state->refpos += pg_mbstrlen_with_len(state->refpoint,
1483 48 : state->last_match - state->refpoint);
1484 48 : state->refpoint = state->last_match;
1485 48 : return state->refpos + 1;
1486 : }
1487 :
1488 : /*
1489 : * Reset search state to the initial state installed by text_position_setup.
1490 : *
1491 : * The next call to text_position_next will search from the beginning
1492 : * of the string.
1493 : */
1494 : static void
1495 12 : text_position_reset(TextPositionState *state)
1496 : {
1497 12 : state->last_match = NULL;
1498 12 : state->refpoint = state->str1;
1499 12 : state->refpos = 0;
1500 12 : }
1501 :
1502 : static void
1503 1300 : text_position_cleanup(TextPositionState *state)
1504 : {
1505 : /* no cleanup needed */
1506 1300 : }
1507 :
1508 :
1509 : static void
1510 12854690 : check_collation_set(Oid collid)
1511 : {
1512 12854690 : if (!OidIsValid(collid))
1513 : {
1514 : /*
1515 : * This typically means that the parser could not resolve a conflict
1516 : * of implicit collations, so report it that way.
1517 : */
1518 30 : ereport(ERROR,
1519 : (errcode(ERRCODE_INDETERMINATE_COLLATION),
1520 : errmsg("could not determine which collation to use for string comparison"),
1521 : errhint("Use the COLLATE clause to set the collation explicitly.")));
1522 : }
1523 12854660 : }
1524 :
1525 : /* varstr_cmp()
1526 : * Comparison function for text strings with given lengths.
1527 : * Includes locale support, but must copy strings to temporary memory
1528 : * to allow null-termination for inputs to strcoll().
1529 : * Returns an integer less than, equal to, or greater than zero, indicating
1530 : * whether arg1 is less than, equal to, or greater than arg2.
1531 : *
1532 : * Note: many functions that depend on this are marked leakproof; therefore,
1533 : * avoid reporting the actual contents of the input when throwing errors.
1534 : * All errors herein should be things that can't happen except on corrupt
1535 : * data, anyway; otherwise we will have trouble with indexing strings that
1536 : * would cause them.
1537 : */
1538 : int
1539 6198838 : varstr_cmp(const char *arg1, int len1, const char *arg2, int len2, Oid collid)
1540 : {
1541 : int result;
1542 :
1543 6198838 : check_collation_set(collid);
1544 :
1545 : /*
1546 : * Unfortunately, there is no strncoll(), so in the non-C locale case we
1547 : * have to do some memory copying. This turns out to be significantly
1548 : * slower, so we optimize the case where LC_COLLATE is C. We also try to
1549 : * optimize relatively-short strings by avoiding palloc/pfree overhead.
1550 : */
1551 6198820 : if (lc_collate_is_c(collid))
1552 : {
1553 2870728 : result = memcmp(arg1, arg2, Min(len1, len2));
1554 2870728 : if ((result == 0) && (len1 != len2))
1555 111648 : result = (len1 < len2) ? -1 : 1;
1556 : }
1557 : else
1558 : {
1559 : pg_locale_t mylocale;
1560 :
1561 3328092 : mylocale = pg_newlocale_from_collation(collid);
1562 :
1563 : /*
1564 : * memcmp() can't tell us which of two unequal strings sorts first,
1565 : * but it's a cheap way to tell if they're equal. Testing shows that
1566 : * memcmp() followed by strcoll() is only trivially slower than
1567 : * strcoll() by itself, so we don't lose much if this doesn't work out
1568 : * very often, and if it does - for example, because there are many
1569 : * equal strings in the input - then we win big by avoiding expensive
1570 : * collation-aware comparisons.
1571 : */
1572 3328092 : if (len1 == len2 && memcmp(arg1, arg2, len1) == 0)
1573 1272954 : return 0;
1574 :
1575 2055138 : result = pg_strncoll(arg1, len1, arg2, len2, mylocale);
1576 :
1577 : /* Break tie if necessary. */
1578 2055138 : if (result == 0 && pg_locale_deterministic(mylocale))
1579 : {
1580 0 : result = memcmp(arg1, arg2, Min(len1, len2));
1581 0 : if ((result == 0) && (len1 != len2))
1582 0 : result = (len1 < len2) ? -1 : 1;
1583 : }
1584 : }
1585 :
1586 4925866 : return result;
1587 : }
1588 :
1589 : /* text_cmp()
1590 : * Internal comparison function for text strings.
1591 : * Returns -1, 0 or 1
1592 : */
1593 : static int
1594 4626428 : text_cmp(text *arg1, text *arg2, Oid collid)
1595 : {
1596 : char *a1p,
1597 : *a2p;
1598 : int len1,
1599 : len2;
1600 :
1601 4626428 : a1p = VARDATA_ANY(arg1);
1602 4626428 : a2p = VARDATA_ANY(arg2);
1603 :
1604 4626428 : len1 = VARSIZE_ANY_EXHDR(arg1);
1605 4626428 : len2 = VARSIZE_ANY_EXHDR(arg2);
1606 :
1607 4626428 : return varstr_cmp(a1p, len1, a2p, len2, collid);
1608 : }
1609 :
1610 : /*
1611 : * Comparison functions for text strings.
1612 : *
1613 : * Note: btree indexes need these routines not to leak memory; therefore,
1614 : * be careful to free working copies of toasted datums. Most places don't
1615 : * need to be so careful.
1616 : */
1617 :
1618 : Datum
1619 6269304 : texteq(PG_FUNCTION_ARGS)
1620 : {
1621 6269304 : Oid collid = PG_GET_COLLATION();
1622 6269304 : bool locale_is_c = false;
1623 6269304 : pg_locale_t mylocale = 0;
1624 : bool result;
1625 :
1626 6269304 : check_collation_set(collid);
1627 :
1628 6269304 : if (lc_collate_is_c(collid))
1629 1273034 : locale_is_c = true;
1630 : else
1631 4996270 : mylocale = pg_newlocale_from_collation(collid);
1632 :
1633 6269304 : if (locale_is_c || pg_locale_deterministic(mylocale))
1634 6268796 : {
1635 6268796 : Datum arg1 = PG_GETARG_DATUM(0);
1636 6268796 : Datum arg2 = PG_GETARG_DATUM(1);
1637 : Size len1,
1638 : len2;
1639 :
1640 : /*
1641 : * Since we only care about equality or not-equality, we can avoid all
1642 : * the expense of strcoll() here, and just do bitwise comparison. In
1643 : * fact, we don't even have to do a bitwise comparison if we can show
1644 : * the lengths of the strings are unequal; which might save us from
1645 : * having to detoast one or both values.
1646 : */
1647 6268796 : len1 = toast_raw_datum_size(arg1);
1648 6268796 : len2 = toast_raw_datum_size(arg2);
1649 6268796 : if (len1 != len2)
1650 2999664 : result = false;
1651 : else
1652 : {
1653 3269132 : text *targ1 = DatumGetTextPP(arg1);
1654 3269132 : text *targ2 = DatumGetTextPP(arg2);
1655 :
1656 3269132 : result = (memcmp(VARDATA_ANY(targ1), VARDATA_ANY(targ2),
1657 : len1 - VARHDRSZ) == 0);
1658 :
1659 3269132 : PG_FREE_IF_COPY(targ1, 0);
1660 3269132 : PG_FREE_IF_COPY(targ2, 1);
1661 : }
1662 : }
1663 : else
1664 : {
1665 508 : text *arg1 = PG_GETARG_TEXT_PP(0);
1666 508 : text *arg2 = PG_GETARG_TEXT_PP(1);
1667 :
1668 508 : result = (text_cmp(arg1, arg2, collid) == 0);
1669 :
1670 508 : PG_FREE_IF_COPY(arg1, 0);
1671 508 : PG_FREE_IF_COPY(arg2, 1);
1672 : }
1673 :
1674 6269304 : PG_RETURN_BOOL(result);
1675 : }
1676 :
1677 : Datum
1678 20404 : textne(PG_FUNCTION_ARGS)
1679 : {
1680 20404 : Oid collid = PG_GET_COLLATION();
1681 20404 : bool locale_is_c = false;
1682 20404 : pg_locale_t mylocale = 0;
1683 : bool result;
1684 :
1685 20404 : check_collation_set(collid);
1686 :
1687 20404 : if (lc_collate_is_c(collid))
1688 5710 : locale_is_c = true;
1689 : else
1690 14694 : mylocale = pg_newlocale_from_collation(collid);
1691 :
1692 20404 : if (locale_is_c || pg_locale_deterministic(mylocale))
1693 20380 : {
1694 20380 : Datum arg1 = PG_GETARG_DATUM(0);
1695 20380 : Datum arg2 = PG_GETARG_DATUM(1);
1696 : Size len1,
1697 : len2;
1698 :
1699 : /* See comment in texteq() */
1700 20380 : len1 = toast_raw_datum_size(arg1);
1701 20380 : len2 = toast_raw_datum_size(arg2);
1702 20380 : if (len1 != len2)
1703 2722 : result = true;
1704 : else
1705 : {
1706 17658 : text *targ1 = DatumGetTextPP(arg1);
1707 17658 : text *targ2 = DatumGetTextPP(arg2);
1708 :
1709 17658 : result = (memcmp(VARDATA_ANY(targ1), VARDATA_ANY(targ2),
1710 : len1 - VARHDRSZ) != 0);
1711 :
1712 17658 : PG_FREE_IF_COPY(targ1, 0);
1713 17658 : PG_FREE_IF_COPY(targ2, 1);
1714 : }
1715 : }
1716 : else
1717 : {
1718 24 : text *arg1 = PG_GETARG_TEXT_PP(0);
1719 24 : text *arg2 = PG_GETARG_TEXT_PP(1);
1720 :
1721 24 : result = (text_cmp(arg1, arg2, collid) != 0);
1722 :
1723 24 : PG_FREE_IF_COPY(arg1, 0);
1724 24 : PG_FREE_IF_COPY(arg2, 1);
1725 : }
1726 :
1727 20404 : PG_RETURN_BOOL(result);
1728 : }
1729 :
1730 : Datum
1731 206300 : text_lt(PG_FUNCTION_ARGS)
1732 : {
1733 206300 : text *arg1 = PG_GETARG_TEXT_PP(0);
1734 206300 : text *arg2 = PG_GETARG_TEXT_PP(1);
1735 : bool result;
1736 :
1737 206300 : result = (text_cmp(arg1, arg2, PG_GET_COLLATION()) < 0);
1738 :
1739 206282 : PG_FREE_IF_COPY(arg1, 0);
1740 206282 : PG_FREE_IF_COPY(arg2, 1);
1741 :
1742 206282 : PG_RETURN_BOOL(result);
1743 : }
1744 :
1745 : Datum
1746 315980 : text_le(PG_FUNCTION_ARGS)
1747 : {
1748 315980 : text *arg1 = PG_GETARG_TEXT_PP(0);
1749 315980 : text *arg2 = PG_GETARG_TEXT_PP(1);
1750 : bool result;
1751 :
1752 315980 : result = (text_cmp(arg1, arg2, PG_GET_COLLATION()) <= 0);
1753 :
1754 315980 : PG_FREE_IF_COPY(arg1, 0);
1755 315980 : PG_FREE_IF_COPY(arg2, 1);
1756 :
1757 315980 : PG_RETURN_BOOL(result);
1758 : }
1759 :
1760 : Datum
1761 195854 : text_gt(PG_FUNCTION_ARGS)
1762 : {
1763 195854 : text *arg1 = PG_GETARG_TEXT_PP(0);
1764 195854 : text *arg2 = PG_GETARG_TEXT_PP(1);
1765 : bool result;
1766 :
1767 195854 : result = (text_cmp(arg1, arg2, PG_GET_COLLATION()) > 0);
1768 :
1769 195854 : PG_FREE_IF_COPY(arg1, 0);
1770 195854 : PG_FREE_IF_COPY(arg2, 1);
1771 :
1772 195854 : PG_RETURN_BOOL(result);
1773 : }
1774 :
1775 : Datum
1776 177902 : text_ge(PG_FUNCTION_ARGS)
1777 : {
1778 177902 : text *arg1 = PG_GETARG_TEXT_PP(0);
1779 177902 : text *arg2 = PG_GETARG_TEXT_PP(1);
1780 : bool result;
1781 :
1782 177902 : result = (text_cmp(arg1, arg2, PG_GET_COLLATION()) >= 0);
1783 :
1784 177902 : PG_FREE_IF_COPY(arg1, 0);
1785 177902 : PG_FREE_IF_COPY(arg2, 1);
1786 :
1787 177902 : PG_RETURN_BOOL(result);
1788 : }
1789 :
1790 : Datum
1791 37914 : text_starts_with(PG_FUNCTION_ARGS)
1792 : {
1793 37914 : Datum arg1 = PG_GETARG_DATUM(0);
1794 37914 : Datum arg2 = PG_GETARG_DATUM(1);
1795 37914 : Oid collid = PG_GET_COLLATION();
1796 37914 : pg_locale_t mylocale = 0;
1797 : bool result;
1798 : Size len1,
1799 : len2;
1800 :
1801 37914 : check_collation_set(collid);
1802 :
1803 37914 : if (!lc_collate_is_c(collid))
1804 25276 : mylocale = pg_newlocale_from_collation(collid);
1805 :
1806 37914 : if (!pg_locale_deterministic(mylocale))
1807 0 : ereport(ERROR,
1808 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1809 : errmsg("nondeterministic collations are not supported for substring searches")));
1810 :
1811 37914 : len1 = toast_raw_datum_size(arg1);
1812 37914 : len2 = toast_raw_datum_size(arg2);
1813 37914 : if (len2 > len1)
1814 0 : result = false;
1815 : else
1816 : {
1817 37914 : text *targ1 = text_substring(arg1, 1, len2, false);
1818 37914 : text *targ2 = DatumGetTextPP(arg2);
1819 :
1820 37914 : result = (memcmp(VARDATA_ANY(targ1), VARDATA_ANY(targ2),
1821 37914 : VARSIZE_ANY_EXHDR(targ2)) == 0);
1822 :
1823 37914 : PG_FREE_IF_COPY(targ1, 0);
1824 37914 : PG_FREE_IF_COPY(targ2, 1);
1825 : }
1826 :
1827 37914 : PG_RETURN_BOOL(result);
1828 : }
1829 :
1830 : Datum
1831 3414224 : bttextcmp(PG_FUNCTION_ARGS)
1832 : {
1833 3414224 : text *arg1 = PG_GETARG_TEXT_PP(0);
1834 3414224 : text *arg2 = PG_GETARG_TEXT_PP(1);
1835 : int32 result;
1836 :
1837 3414224 : result = text_cmp(arg1, arg2, PG_GET_COLLATION());
1838 :
1839 3414224 : PG_FREE_IF_COPY(arg1, 0);
1840 3414224 : PG_FREE_IF_COPY(arg2, 1);
1841 :
1842 3414224 : PG_RETURN_INT32(result);
1843 : }
1844 :
1845 : Datum
1846 71268 : bttextsortsupport(PG_FUNCTION_ARGS)
1847 : {
1848 71268 : SortSupport ssup = (SortSupport) PG_GETARG_POINTER(0);
1849 71268 : Oid collid = ssup->ssup_collation;
1850 : MemoryContext oldcontext;
1851 :
1852 71268 : oldcontext = MemoryContextSwitchTo(ssup->ssup_cxt);
1853 :
1854 : /* Use generic string SortSupport */
1855 71268 : varstr_sortsupport(ssup, TEXTOID, collid);
1856 :
1857 71256 : MemoryContextSwitchTo(oldcontext);
1858 :
1859 71256 : PG_RETURN_VOID();
1860 : }
1861 :
1862 : /*
1863 : * Generic sortsupport interface for character type's operator classes.
1864 : * Includes locale support, and support for BpChar semantics (i.e. removing
1865 : * trailing spaces before comparison).
1866 : *
1867 : * Relies on the assumption that text, VarChar, BpChar, and bytea all have the
1868 : * same representation. Callers that always use the C collation (e.g.
1869 : * non-collatable type callers like bytea) may have NUL bytes in their strings;
1870 : * this will not work with any other collation, though.
1871 : */
1872 : void
1873 107554 : varstr_sortsupport(SortSupport ssup, Oid typid, Oid collid)
1874 : {
1875 107554 : bool abbreviate = ssup->abbreviate;
1876 107554 : bool collate_c = false;
1877 : VarStringSortSupport *sss;
1878 107554 : pg_locale_t locale = 0;
1879 :
1880 107554 : check_collation_set(collid);
1881 :
1882 : /*
1883 : * If possible, set ssup->comparator to a function which can be used to
1884 : * directly compare two datums. If we can do this, we'll avoid the
1885 : * overhead of a trip through the fmgr layer for every comparison, which
1886 : * can be substantial.
1887 : *
1888 : * Most typically, we'll set the comparator to varlenafastcmp_locale,
1889 : * which uses strcoll() to perform comparisons. We use that for the
1890 : * BpChar case too, but type NAME uses namefastcmp_locale. However, if
1891 : * LC_COLLATE = C, we can make things quite a bit faster with
1892 : * varstrfastcmp_c, bpcharfastcmp_c, or namefastcmp_c, all of which use
1893 : * memcmp() rather than strcoll().
1894 : */
1895 107542 : if (lc_collate_is_c(collid))
1896 : {
1897 72204 : if (typid == BPCHAROID)
1898 278 : ssup->comparator = bpcharfastcmp_c;
1899 71926 : else if (typid == NAMEOID)
1900 : {
1901 35324 : ssup->comparator = namefastcmp_c;
1902 : /* Not supporting abbreviation with type NAME, for now */
1903 35324 : abbreviate = false;
1904 : }
1905 : else
1906 36602 : ssup->comparator = varstrfastcmp_c;
1907 :
1908 72204 : collate_c = true;
1909 : }
1910 : else
1911 : {
1912 : /*
1913 : * We need a collation-sensitive comparison. To make things faster,
1914 : * we'll figure out the collation based on the locale id and cache the
1915 : * result.
1916 : */
1917 35338 : locale = pg_newlocale_from_collation(collid);
1918 :
1919 : /*
1920 : * We use varlenafastcmp_locale except for type NAME.
1921 : */
1922 35338 : if (typid == NAMEOID)
1923 : {
1924 0 : ssup->comparator = namefastcmp_locale;
1925 : /* Not supporting abbreviation with type NAME, for now */
1926 0 : abbreviate = false;
1927 : }
1928 : else
1929 35338 : ssup->comparator = varlenafastcmp_locale;
1930 : }
1931 :
1932 : /*
1933 : * Unfortunately, it seems that abbreviation for non-C collations is
1934 : * broken on many common platforms; see pg_strxfrm_enabled().
1935 : *
1936 : * Even apart from the risk of broken locales, it's possible that there
1937 : * are platforms where the use of abbreviated keys should be disabled at
1938 : * compile time. Having only 4 byte datums could make worst-case
1939 : * performance drastically more likely, for example. Moreover, macOS's
1940 : * strxfrm() implementation is known to not effectively concentrate a
1941 : * significant amount of entropy from the original string in earlier
1942 : * transformed blobs. It's possible that other supported platforms are
1943 : * similarly encumbered. So, if we ever get past disabling this
1944 : * categorically, we may still want or need to disable it for particular
1945 : * platforms.
1946 : */
1947 107542 : if (!collate_c && !pg_strxfrm_enabled(locale))
1948 34776 : abbreviate = false;
1949 :
1950 : /*
1951 : * If we're using abbreviated keys, or if we're using a locale-aware
1952 : * comparison, we need to initialize a VarStringSortSupport object. Both
1953 : * cases will make use of the temporary buffers we initialize here for
1954 : * scratch space (and to detect requirement for BpChar semantics from
1955 : * caller), and the abbreviation case requires additional state.
1956 : */
1957 107542 : if (abbreviate || !collate_c)
1958 : {
1959 57164 : sss = palloc(sizeof(VarStringSortSupport));
1960 57164 : sss->buf1 = palloc(TEXTBUFLEN);
1961 57164 : sss->buflen1 = TEXTBUFLEN;
1962 57164 : sss->buf2 = palloc(TEXTBUFLEN);
1963 57164 : sss->buflen2 = TEXTBUFLEN;
1964 : /* Start with invalid values */
1965 57164 : sss->last_len1 = -1;
1966 57164 : sss->last_len2 = -1;
1967 : /* Initialize */
1968 57164 : sss->last_returned = 0;
1969 57164 : sss->locale = locale;
1970 :
1971 : /*
1972 : * To avoid somehow confusing a strxfrm() blob and an original string,
1973 : * constantly keep track of the variety of data that buf1 and buf2
1974 : * currently contain.
1975 : *
1976 : * Comparisons may be interleaved with conversion calls. Frequently,
1977 : * conversions and comparisons are batched into two distinct phases,
1978 : * but the correctness of caching cannot hinge upon this. For
1979 : * comparison caching, buffer state is only trusted if cache_blob is
1980 : * found set to false, whereas strxfrm() caching only trusts the state
1981 : * when cache_blob is found set to true.
1982 : *
1983 : * Arbitrarily initialize cache_blob to true.
1984 : */
1985 57164 : sss->cache_blob = true;
1986 57164 : sss->collate_c = collate_c;
1987 57164 : sss->typid = typid;
1988 57164 : ssup->ssup_extra = sss;
1989 :
1990 : /*
1991 : * If possible, plan to use the abbreviated keys optimization. The
1992 : * core code may switch back to authoritative comparator should
1993 : * abbreviation be aborted.
1994 : */
1995 57164 : if (abbreviate)
1996 : {
1997 22340 : sss->prop_card = 0.20;
1998 22340 : initHyperLogLog(&sss->abbr_card, 10);
1999 22340 : initHyperLogLog(&sss->full_card, 10);
2000 22340 : ssup->abbrev_full_comparator = ssup->comparator;
2001 22340 : ssup->comparator = ssup_datum_unsigned_cmp;
2002 22340 : ssup->abbrev_converter = varstr_abbrev_convert;
2003 22340 : ssup->abbrev_abort = varstr_abbrev_abort;
2004 : }
2005 : }
2006 107542 : }
2007 :
2008 : /*
2009 : * sortsupport comparison func (for C locale case)
2010 : */
2011 : static int
2012 33441760 : varstrfastcmp_c(Datum x, Datum y, SortSupport ssup)
2013 : {
2014 33441760 : VarString *arg1 = DatumGetVarStringPP(x);
2015 33441760 : VarString *arg2 = DatumGetVarStringPP(y);
2016 : char *a1p,
2017 : *a2p;
2018 : int len1,
2019 : len2,
2020 : result;
2021 :
2022 33441760 : a1p = VARDATA_ANY(arg1);
2023 33441760 : a2p = VARDATA_ANY(arg2);
2024 :
2025 33441760 : len1 = VARSIZE_ANY_EXHDR(arg1);
2026 33441760 : len2 = VARSIZE_ANY_EXHDR(arg2);
2027 :
2028 33441760 : result = memcmp(a1p, a2p, Min(len1, len2));
2029 33441760 : if ((result == 0) && (len1 != len2))
2030 912940 : result = (len1 < len2) ? -1 : 1;
2031 :
2032 : /* We can't afford to leak memory here. */
2033 33441760 : if (PointerGetDatum(arg1) != x)
2034 2 : pfree(arg1);
2035 33441760 : if (PointerGetDatum(arg2) != y)
2036 2 : pfree(arg2);
2037 :
2038 33441760 : return result;
2039 : }
2040 :
2041 : /*
2042 : * sortsupport comparison func (for BpChar C locale case)
2043 : *
2044 : * BpChar outsources its sortsupport to this module. Specialization for the
2045 : * varstr_sortsupport BpChar case, modeled on
2046 : * internal_bpchar_pattern_compare().
2047 : */
2048 : static int
2049 61994 : bpcharfastcmp_c(Datum x, Datum y, SortSupport ssup)
2050 : {
2051 61994 : BpChar *arg1 = DatumGetBpCharPP(x);
2052 61994 : BpChar *arg2 = DatumGetBpCharPP(y);
2053 : char *a1p,
2054 : *a2p;
2055 : int len1,
2056 : len2,
2057 : result;
2058 :
2059 61994 : a1p = VARDATA_ANY(arg1);
2060 61994 : a2p = VARDATA_ANY(arg2);
2061 :
2062 61994 : len1 = bpchartruelen(a1p, VARSIZE_ANY_EXHDR(arg1));
2063 61994 : len2 = bpchartruelen(a2p, VARSIZE_ANY_EXHDR(arg2));
2064 :
2065 61994 : result = memcmp(a1p, a2p, Min(len1, len2));
2066 61994 : if ((result == 0) && (len1 != len2))
2067 4 : result = (len1 < len2) ? -1 : 1;
2068 :
2069 : /* We can't afford to leak memory here. */
2070 61994 : if (PointerGetDatum(arg1) != x)
2071 0 : pfree(arg1);
2072 61994 : if (PointerGetDatum(arg2) != y)
2073 0 : pfree(arg2);
2074 :
2075 61994 : return result;
2076 : }
2077 :
2078 : /*
2079 : * sortsupport comparison func (for NAME C locale case)
2080 : */
2081 : static int
2082 30351614 : namefastcmp_c(Datum x, Datum y, SortSupport ssup)
2083 : {
2084 30351614 : Name arg1 = DatumGetName(x);
2085 30351614 : Name arg2 = DatumGetName(y);
2086 :
2087 30351614 : return strncmp(NameStr(*arg1), NameStr(*arg2), NAMEDATALEN);
2088 : }
2089 :
2090 : /*
2091 : * sortsupport comparison func (for locale case with all varlena types)
2092 : */
2093 : static int
2094 25616992 : varlenafastcmp_locale(Datum x, Datum y, SortSupport ssup)
2095 : {
2096 25616992 : VarString *arg1 = DatumGetVarStringPP(x);
2097 25616992 : VarString *arg2 = DatumGetVarStringPP(y);
2098 : char *a1p,
2099 : *a2p;
2100 : int len1,
2101 : len2,
2102 : result;
2103 :
2104 25616992 : a1p = VARDATA_ANY(arg1);
2105 25616992 : a2p = VARDATA_ANY(arg2);
2106 :
2107 25616992 : len1 = VARSIZE_ANY_EXHDR(arg1);
2108 25616992 : len2 = VARSIZE_ANY_EXHDR(arg2);
2109 :
2110 25616992 : result = varstrfastcmp_locale(a1p, len1, a2p, len2, ssup);
2111 :
2112 : /* We can't afford to leak memory here. */
2113 25616992 : if (PointerGetDatum(arg1) != x)
2114 4 : pfree(arg1);
2115 25616992 : if (PointerGetDatum(arg2) != y)
2116 4 : pfree(arg2);
2117 :
2118 25616992 : return result;
2119 : }
2120 :
2121 : /*
2122 : * sortsupport comparison func (for locale case with NAME type)
2123 : */
2124 : static int
2125 0 : namefastcmp_locale(Datum x, Datum y, SortSupport ssup)
2126 : {
2127 0 : Name arg1 = DatumGetName(x);
2128 0 : Name arg2 = DatumGetName(y);
2129 :
2130 0 : return varstrfastcmp_locale(NameStr(*arg1), strlen(NameStr(*arg1)),
2131 0 : NameStr(*arg2), strlen(NameStr(*arg2)),
2132 : ssup);
2133 : }
2134 :
2135 : /*
2136 : * sortsupport comparison func for locale cases
2137 : */
2138 : static int
2139 25616992 : varstrfastcmp_locale(char *a1p, int len1, char *a2p, int len2, SortSupport ssup)
2140 : {
2141 25616992 : VarStringSortSupport *sss = (VarStringSortSupport *) ssup->ssup_extra;
2142 : int result;
2143 : bool arg1_match;
2144 :
2145 : /* Fast pre-check for equality, as discussed in varstr_cmp() */
2146 25616992 : if (len1 == len2 && memcmp(a1p, a2p, len1) == 0)
2147 : {
2148 : /*
2149 : * No change in buf1 or buf2 contents, so avoid changing last_len1 or
2150 : * last_len2. Existing contents of buffers might still be used by
2151 : * next call.
2152 : *
2153 : * It's fine to allow the comparison of BpChar padding bytes here,
2154 : * even though that implies that the memcmp() will usually be
2155 : * performed for BpChar callers (though multibyte characters could
2156 : * still prevent that from occurring). The memcmp() is still very
2157 : * cheap, and BpChar's funny semantics have us remove trailing spaces
2158 : * (not limited to padding), so we need make no distinction between
2159 : * padding space characters and "real" space characters.
2160 : */
2161 9141390 : return 0;
2162 : }
2163 :
2164 16475602 : if (sss->typid == BPCHAROID)
2165 : {
2166 : /* Get true number of bytes, ignoring trailing spaces */
2167 33012 : len1 = bpchartruelen(a1p, len1);
2168 33012 : len2 = bpchartruelen(a2p, len2);
2169 : }
2170 :
2171 16475602 : if (len1 >= sss->buflen1)
2172 : {
2173 14 : sss->buflen1 = Max(len1 + 1, Min(sss->buflen1 * 2, MaxAllocSize));
2174 14 : sss->buf1 = repalloc(sss->buf1, sss->buflen1);
2175 : }
2176 16475602 : if (len2 >= sss->buflen2)
2177 : {
2178 10 : sss->buflen2 = Max(len2 + 1, Min(sss->buflen2 * 2, MaxAllocSize));
2179 10 : sss->buf2 = repalloc(sss->buf2, sss->buflen2);
2180 : }
2181 :
2182 : /*
2183 : * We're likely to be asked to compare the same strings repeatedly, and
2184 : * memcmp() is so much cheaper than strcoll() that it pays to try to cache
2185 : * comparisons, even though in general there is no reason to think that
2186 : * that will work out (every string datum may be unique). Caching does
2187 : * not slow things down measurably when it doesn't work out, and can speed
2188 : * things up by rather a lot when it does. In part, this is because the
2189 : * memcmp() compares data from cachelines that are needed in L1 cache even
2190 : * when the last comparison's result cannot be reused.
2191 : */
2192 16475602 : arg1_match = true;
2193 16475602 : if (len1 != sss->last_len1 || memcmp(sss->buf1, a1p, len1) != 0)
2194 : {
2195 14760550 : arg1_match = false;
2196 14760550 : memcpy(sss->buf1, a1p, len1);
2197 14760550 : sss->buf1[len1] = '\0';
2198 14760550 : sss->last_len1 = len1;
2199 : }
2200 :
2201 : /*
2202 : * If we're comparing the same two strings as last time, we can return the
2203 : * same answer without calling strcoll() again. This is more likely than
2204 : * it seems (at least with moderate to low cardinality sets), because
2205 : * quicksort compares the same pivot against many values.
2206 : */
2207 16475602 : if (len2 != sss->last_len2 || memcmp(sss->buf2, a2p, len2) != 0)
2208 : {
2209 2427396 : memcpy(sss->buf2, a2p, len2);
2210 2427396 : sss->buf2[len2] = '\0';
2211 2427396 : sss->last_len2 = len2;
2212 : }
2213 14048206 : else if (arg1_match && !sss->cache_blob)
2214 : {
2215 : /* Use result cached following last actual strcoll() call */
2216 1444674 : return sss->last_returned;
2217 : }
2218 :
2219 15030928 : result = pg_strcoll(sss->buf1, sss->buf2, sss->locale);
2220 :
2221 : /* Break tie if necessary. */
2222 15030928 : if (result == 0 && pg_locale_deterministic(sss->locale))
2223 0 : result = strcmp(sss->buf1, sss->buf2);
2224 :
2225 : /* Cache result, perhaps saving an expensive strcoll() call next time */
2226 15030928 : sss->cache_blob = false;
2227 15030928 : sss->last_returned = result;
2228 15030928 : return result;
2229 : }
2230 :
2231 : /*
2232 : * Conversion routine for sortsupport. Converts original to abbreviated key
2233 : * representation. Our encoding strategy is simple -- pack the first 8 bytes
2234 : * of a strxfrm() blob into a Datum (on little-endian machines, the 8 bytes are
2235 : * stored in reverse order), and treat it as an unsigned integer. When the "C"
2236 : * locale is used, or in case of bytea, just memcpy() from original instead.
2237 : */
2238 : static Datum
2239 818546 : varstr_abbrev_convert(Datum original, SortSupport ssup)
2240 : {
2241 818546 : const size_t max_prefix_bytes = sizeof(Datum);
2242 818546 : VarStringSortSupport *sss = (VarStringSortSupport *) ssup->ssup_extra;
2243 818546 : VarString *authoritative = DatumGetVarStringPP(original);
2244 818546 : char *authoritative_data = VARDATA_ANY(authoritative);
2245 :
2246 : /* working state */
2247 : Datum res;
2248 : char *pres;
2249 : int len;
2250 : uint32 hash;
2251 :
2252 818546 : pres = (char *) &res;
2253 : /* memset(), so any non-overwritten bytes are NUL */
2254 818546 : memset(pres, 0, max_prefix_bytes);
2255 818546 : len = VARSIZE_ANY_EXHDR(authoritative);
2256 :
2257 : /* Get number of bytes, ignoring trailing spaces */
2258 818546 : if (sss->typid == BPCHAROID)
2259 1010 : len = bpchartruelen(authoritative_data, len);
2260 :
2261 : /*
2262 : * If we're using the C collation, use memcpy(), rather than strxfrm(), to
2263 : * abbreviate keys. The full comparator for the C locale is always
2264 : * memcmp(). It would be incorrect to allow bytea callers (callers that
2265 : * always force the C collation -- bytea isn't a collatable type, but this
2266 : * approach is convenient) to use strxfrm(). This is because bytea
2267 : * strings may contain NUL bytes. Besides, this should be faster, too.
2268 : *
2269 : * More generally, it's okay that bytea callers can have NUL bytes in
2270 : * strings because abbreviated cmp need not make a distinction between
2271 : * terminating NUL bytes, and NUL bytes representing actual NULs in the
2272 : * authoritative representation. Hopefully a comparison at or past one
2273 : * abbreviated key's terminating NUL byte will resolve the comparison
2274 : * without consulting the authoritative representation; specifically, some
2275 : * later non-NUL byte in the longer string can resolve the comparison
2276 : * against a subsequent terminating NUL in the shorter string. There will
2277 : * usually be what is effectively a "length-wise" resolution there and
2278 : * then.
2279 : *
2280 : * If that doesn't work out -- if all bytes in the longer string
2281 : * positioned at or past the offset of the smaller string's (first)
2282 : * terminating NUL are actually representative of NUL bytes in the
2283 : * authoritative binary string (perhaps with some *terminating* NUL bytes
2284 : * towards the end of the longer string iff it happens to still be small)
2285 : * -- then an authoritative tie-breaker will happen, and do the right
2286 : * thing: explicitly consider string length.
2287 : */
2288 818546 : if (sss->collate_c)
2289 816752 : memcpy(pres, authoritative_data, Min(len, max_prefix_bytes));
2290 : else
2291 : {
2292 : Size bsize;
2293 :
2294 : /*
2295 : * We're not using the C collation, so fall back on strxfrm or ICU
2296 : * analogs.
2297 : */
2298 :
2299 : /* By convention, we use buffer 1 to store and NUL-terminate */
2300 1794 : if (len >= sss->buflen1)
2301 : {
2302 0 : sss->buflen1 = Max(len + 1, Min(sss->buflen1 * 2, MaxAllocSize));
2303 0 : sss->buf1 = repalloc(sss->buf1, sss->buflen1);
2304 : }
2305 :
2306 : /* Might be able to reuse strxfrm() blob from last call */
2307 1794 : if (sss->last_len1 == len && sss->cache_blob &&
2308 924 : memcmp(sss->buf1, authoritative_data, len) == 0)
2309 : {
2310 168 : memcpy(pres, sss->buf2, Min(max_prefix_bytes, sss->last_len2));
2311 : /* No change affecting cardinality, so no hashing required */
2312 168 : goto done;
2313 : }
2314 :
2315 1626 : memcpy(sss->buf1, authoritative_data, len);
2316 :
2317 : /*
2318 : * pg_strxfrm() and pg_strxfrm_prefix expect NUL-terminated strings.
2319 : */
2320 1626 : sss->buf1[len] = '\0';
2321 1626 : sss->last_len1 = len;
2322 :
2323 1626 : if (pg_strxfrm_prefix_enabled(sss->locale))
2324 : {
2325 1626 : if (sss->buflen2 < max_prefix_bytes)
2326 : {
2327 0 : sss->buflen2 = Max(max_prefix_bytes,
2328 : Min(sss->buflen2 * 2, MaxAllocSize));
2329 0 : sss->buf2 = repalloc(sss->buf2, sss->buflen2);
2330 : }
2331 :
2332 1626 : bsize = pg_strxfrm_prefix(sss->buf2, sss->buf1,
2333 : max_prefix_bytes, sss->locale);
2334 1626 : sss->last_len2 = bsize;
2335 : }
2336 : else
2337 : {
2338 : /*
2339 : * Loop: Call pg_strxfrm(), possibly enlarge buffer, and try
2340 : * again. The pg_strxfrm() function leaves the result buffer
2341 : * content undefined if the result did not fit, so we need to
2342 : * retry until everything fits, even though we only need the first
2343 : * few bytes in the end.
2344 : */
2345 : for (;;)
2346 : {
2347 0 : bsize = pg_strxfrm(sss->buf2, sss->buf1, sss->buflen2,
2348 : sss->locale);
2349 :
2350 0 : sss->last_len2 = bsize;
2351 0 : if (bsize < sss->buflen2)
2352 0 : break;
2353 :
2354 : /*
2355 : * Grow buffer and retry.
2356 : */
2357 0 : sss->buflen2 = Max(bsize + 1,
2358 : Min(sss->buflen2 * 2, MaxAllocSize));
2359 0 : sss->buf2 = repalloc(sss->buf2, sss->buflen2);
2360 : }
2361 : }
2362 :
2363 : /*
2364 : * Every Datum byte is always compared. This is safe because the
2365 : * strxfrm() blob is itself NUL terminated, leaving no danger of
2366 : * misinterpreting any NUL bytes not intended to be interpreted as
2367 : * logically representing termination.
2368 : *
2369 : * (Actually, even if there were NUL bytes in the blob it would be
2370 : * okay. See remarks on bytea case above.)
2371 : */
2372 1626 : memcpy(pres, sss->buf2, Min(max_prefix_bytes, bsize));
2373 : }
2374 :
2375 : /*
2376 : * Maintain approximate cardinality of both abbreviated keys and original,
2377 : * authoritative keys using HyperLogLog. Used as cheap insurance against
2378 : * the worst case, where we do many string transformations for no saving
2379 : * in full strcoll()-based comparisons. These statistics are used by
2380 : * varstr_abbrev_abort().
2381 : *
2382 : * First, Hash key proper, or a significant fraction of it. Mix in length
2383 : * in order to compensate for cases where differences are past
2384 : * PG_CACHE_LINE_SIZE bytes, so as to limit the overhead of hashing.
2385 : */
2386 818378 : hash = DatumGetUInt32(hash_any((unsigned char *) authoritative_data,
2387 : Min(len, PG_CACHE_LINE_SIZE)));
2388 :
2389 818378 : if (len > PG_CACHE_LINE_SIZE)
2390 160 : hash ^= DatumGetUInt32(hash_uint32((uint32) len));
2391 :
2392 818378 : addHyperLogLog(&sss->full_card, hash);
2393 :
2394 : /* Hash abbreviated key */
2395 : #if SIZEOF_DATUM == 8
2396 : {
2397 : uint32 lohalf,
2398 : hihalf;
2399 :
2400 818378 : lohalf = (uint32) res;
2401 818378 : hihalf = (uint32) (res >> 32);
2402 818378 : hash = DatumGetUInt32(hash_uint32(lohalf ^ hihalf));
2403 : }
2404 : #else /* SIZEOF_DATUM != 8 */
2405 : hash = DatumGetUInt32(hash_uint32((uint32) res));
2406 : #endif
2407 :
2408 818378 : addHyperLogLog(&sss->abbr_card, hash);
2409 :
2410 : /* Cache result, perhaps saving an expensive strxfrm() call next time */
2411 818378 : sss->cache_blob = true;
2412 818546 : done:
2413 :
2414 : /*
2415 : * Byteswap on little-endian machines.
2416 : *
2417 : * This is needed so that ssup_datum_unsigned_cmp() (an unsigned integer
2418 : * 3-way comparator) works correctly on all platforms. If we didn't do
2419 : * this, the comparator would have to call memcmp() with a pair of
2420 : * pointers to the first byte of each abbreviated key, which is slower.
2421 : */
2422 818546 : res = DatumBigEndianToNative(res);
2423 :
2424 : /* Don't leak memory here */
2425 818546 : if (PointerGetDatum(authoritative) != original)
2426 4 : pfree(authoritative);
2427 :
2428 818546 : return res;
2429 : }
2430 :
2431 : /*
2432 : * Callback for estimating effectiveness of abbreviated key optimization, using
2433 : * heuristic rules. Returns value indicating if the abbreviation optimization
2434 : * should be aborted, based on its projected effectiveness.
2435 : */
2436 : static bool
2437 2146 : varstr_abbrev_abort(int memtupcount, SortSupport ssup)
2438 : {
2439 2146 : VarStringSortSupport *sss = (VarStringSortSupport *) ssup->ssup_extra;
2440 : double abbrev_distinct,
2441 : key_distinct;
2442 :
2443 : Assert(ssup->abbreviate);
2444 :
2445 : /* Have a little patience */
2446 2146 : if (memtupcount < 100)
2447 1172 : return false;
2448 :
2449 974 : abbrev_distinct = estimateHyperLogLog(&sss->abbr_card);
2450 974 : key_distinct = estimateHyperLogLog(&sss->full_card);
2451 :
2452 : /*
2453 : * Clamp cardinality estimates to at least one distinct value. While
2454 : * NULLs are generally disregarded, if only NULL values were seen so far,
2455 : * that might misrepresent costs if we failed to clamp.
2456 : */
2457 974 : if (abbrev_distinct <= 1.0)
2458 0 : abbrev_distinct = 1.0;
2459 :
2460 974 : if (key_distinct <= 1.0)
2461 0 : key_distinct = 1.0;
2462 :
2463 : /*
2464 : * In the worst case all abbreviated keys are identical, while at the same
2465 : * time there are differences within full key strings not captured in
2466 : * abbreviations.
2467 : */
2468 : #ifdef TRACE_SORT
2469 974 : if (trace_sort)
2470 : {
2471 0 : double norm_abbrev_card = abbrev_distinct / (double) memtupcount;
2472 :
2473 0 : elog(LOG, "varstr_abbrev: abbrev_distinct after %d: %f "
2474 : "(key_distinct: %f, norm_abbrev_card: %f, prop_card: %f)",
2475 : memtupcount, abbrev_distinct, key_distinct, norm_abbrev_card,
2476 : sss->prop_card);
2477 : }
2478 : #endif
2479 :
2480 : /*
2481 : * If the number of distinct abbreviated keys approximately matches the
2482 : * number of distinct authoritative original keys, that's reason enough to
2483 : * proceed. We can win even with a very low cardinality set if most
2484 : * tie-breakers only memcmp(). This is by far the most important
2485 : * consideration.
2486 : *
2487 : * While comparisons that are resolved at the abbreviated key level are
2488 : * considerably cheaper than tie-breakers resolved with memcmp(), both of
2489 : * those two outcomes are so much cheaper than a full strcoll() once
2490 : * sorting is underway that it doesn't seem worth it to weigh abbreviated
2491 : * cardinality against the overall size of the set in order to more
2492 : * accurately model costs. Assume that an abbreviated comparison, and an
2493 : * abbreviated comparison with a cheap memcmp()-based authoritative
2494 : * resolution are equivalent.
2495 : */
2496 974 : if (abbrev_distinct > key_distinct * sss->prop_card)
2497 : {
2498 : /*
2499 : * When we have exceeded 10,000 tuples, decay required cardinality
2500 : * aggressively for next call.
2501 : *
2502 : * This is useful because the number of comparisons required on
2503 : * average increases at a linearithmic rate, and at roughly 10,000
2504 : * tuples that factor will start to dominate over the linear costs of
2505 : * string transformation (this is a conservative estimate). The decay
2506 : * rate is chosen to be a little less aggressive than halving -- which
2507 : * (since we're called at points at which memtupcount has doubled)
2508 : * would never see the cost model actually abort past the first call
2509 : * following a decay. This decay rate is mostly a precaution against
2510 : * a sudden, violent swing in how well abbreviated cardinality tracks
2511 : * full key cardinality. The decay also serves to prevent a marginal
2512 : * case from being aborted too late, when too much has already been
2513 : * invested in string transformation.
2514 : *
2515 : * It's possible for sets of several million distinct strings with
2516 : * mere tens of thousands of distinct abbreviated keys to still
2517 : * benefit very significantly. This will generally occur provided
2518 : * each abbreviated key is a proxy for a roughly uniform number of the
2519 : * set's full keys. If it isn't so, we hope to catch that early and
2520 : * abort. If it isn't caught early, by the time the problem is
2521 : * apparent it's probably not worth aborting.
2522 : */
2523 974 : if (memtupcount > 10000)
2524 4 : sss->prop_card *= 0.65;
2525 :
2526 974 : return false;
2527 : }
2528 :
2529 : /*
2530 : * Abort abbreviation strategy.
2531 : *
2532 : * The worst case, where all abbreviated keys are identical while all
2533 : * original strings differ will typically only see a regression of about
2534 : * 10% in execution time for small to medium sized lists of strings.
2535 : * Whereas on modern CPUs where cache stalls are the dominant cost, we can
2536 : * often expect very large improvements, particularly with sets of strings
2537 : * of moderately high to high abbreviated cardinality. There is little to
2538 : * lose but much to gain, which our strategy reflects.
2539 : */
2540 : #ifdef TRACE_SORT
2541 0 : if (trace_sort)
2542 0 : elog(LOG, "varstr_abbrev: aborted abbreviation at %d "
2543 : "(abbrev_distinct: %f, key_distinct: %f, prop_card: %f)",
2544 : memtupcount, abbrev_distinct, key_distinct, sss->prop_card);
2545 : #endif
2546 :
2547 0 : return true;
2548 : }
2549 :
2550 : /*
2551 : * Generic equalimage support function for character type's operator classes.
2552 : * Disables the use of deduplication with nondeterministic collations.
2553 : */
2554 : Datum
2555 8076 : btvarstrequalimage(PG_FUNCTION_ARGS)
2556 : {
2557 : /* Oid opcintype = PG_GETARG_OID(0); */
2558 8076 : Oid collid = PG_GET_COLLATION();
2559 :
2560 8076 : check_collation_set(collid);
2561 :
2562 8076 : if (lc_collate_is_c(collid) ||
2563 44 : collid == DEFAULT_COLLATION_OID ||
2564 44 : get_collation_isdeterministic(collid))
2565 8056 : PG_RETURN_BOOL(true);
2566 : else
2567 20 : PG_RETURN_BOOL(false);
2568 : }
2569 :
2570 : Datum
2571 229560 : text_larger(PG_FUNCTION_ARGS)
2572 : {
2573 229560 : text *arg1 = PG_GETARG_TEXT_PP(0);
2574 229560 : text *arg2 = PG_GETARG_TEXT_PP(1);
2575 : text *result;
2576 :
2577 229560 : result = ((text_cmp(arg1, arg2, PG_GET_COLLATION()) > 0) ? arg1 : arg2);
2578 :
2579 229560 : PG_RETURN_TEXT_P(result);
2580 : }
2581 :
2582 : Datum
2583 86076 : text_smaller(PG_FUNCTION_ARGS)
2584 : {
2585 86076 : text *arg1 = PG_GETARG_TEXT_PP(0);
2586 86076 : text *arg2 = PG_GETARG_TEXT_PP(1);
2587 : text *result;
2588 :
2589 86076 : result = ((text_cmp(arg1, arg2, PG_GET_COLLATION()) < 0) ? arg1 : arg2);
2590 :
2591 86076 : PG_RETURN_TEXT_P(result);
2592 : }
2593 :
2594 :
2595 : /*
2596 : * Cross-type comparison functions for types text and name.
2597 : */
2598 :
2599 : Datum
2600 203896 : nameeqtext(PG_FUNCTION_ARGS)
2601 : {
2602 203896 : Name arg1 = PG_GETARG_NAME(0);
2603 203896 : text *arg2 = PG_GETARG_TEXT_PP(1);
2604 203896 : size_t len1 = strlen(NameStr(*arg1));
2605 203896 : size_t len2 = VARSIZE_ANY_EXHDR(arg2);
2606 203896 : Oid collid = PG_GET_COLLATION();
2607 : bool result;
2608 :
2609 203896 : check_collation_set(collid);
2610 :
2611 203896 : if (collid == C_COLLATION_OID)
2612 302032 : result = (len1 == len2 &&
2613 131680 : memcmp(NameStr(*arg1), VARDATA_ANY(arg2), len1) == 0);
2614 : else
2615 33544 : result = (varstr_cmp(NameStr(*arg1), len1,
2616 33544 : VARDATA_ANY(arg2), len2,
2617 : collid) == 0);
2618 :
2619 203896 : PG_FREE_IF_COPY(arg2, 1);
2620 :
2621 203896 : PG_RETURN_BOOL(result);
2622 : }
2623 :
2624 : Datum
2625 7338 : texteqname(PG_FUNCTION_ARGS)
2626 : {
2627 7338 : text *arg1 = PG_GETARG_TEXT_PP(0);
2628 7338 : Name arg2 = PG_GETARG_NAME(1);
2629 7338 : size_t len1 = VARSIZE_ANY_EXHDR(arg1);
2630 7338 : size_t len2 = strlen(NameStr(*arg2));
2631 7338 : Oid collid = PG_GET_COLLATION();
2632 : bool result;
2633 :
2634 7338 : check_collation_set(collid);
2635 :
2636 7338 : if (collid == C_COLLATION_OID)
2637 564 : result = (len1 == len2 &&
2638 180 : memcmp(VARDATA_ANY(arg1), NameStr(*arg2), len1) == 0);
2639 : else
2640 6954 : result = (varstr_cmp(VARDATA_ANY(arg1), len1,
2641 6954 : NameStr(*arg2), len2,
2642 : collid) == 0);
2643 :
2644 7338 : PG_FREE_IF_COPY(arg1, 0);
2645 :
2646 7338 : PG_RETURN_BOOL(result);
2647 : }
2648 :
2649 : Datum
2650 36 : namenetext(PG_FUNCTION_ARGS)
2651 : {
2652 36 : Name arg1 = PG_GETARG_NAME(0);
2653 36 : text *arg2 = PG_GETARG_TEXT_PP(1);
2654 36 : size_t len1 = strlen(NameStr(*arg1));
2655 36 : size_t len2 = VARSIZE_ANY_EXHDR(arg2);
2656 36 : Oid collid = PG_GET_COLLATION();
2657 : bool result;
2658 :
2659 36 : check_collation_set(collid);
2660 :
2661 36 : if (collid == C_COLLATION_OID)
2662 18 : result = !(len1 == len2 &&
2663 0 : memcmp(NameStr(*arg1), VARDATA_ANY(arg2), len1) == 0);
2664 : else
2665 18 : result = !(varstr_cmp(NameStr(*arg1), len1,
2666 18 : VARDATA_ANY(arg2), len2,
2667 : collid) == 0);
2668 :
2669 36 : PG_FREE_IF_COPY(arg2, 1);
2670 :
2671 36 : PG_RETURN_BOOL(result);
2672 : }
2673 :
2674 : Datum
2675 18 : textnename(PG_FUNCTION_ARGS)
2676 : {
2677 18 : text *arg1 = PG_GETARG_TEXT_PP(0);
2678 18 : Name arg2 = PG_GETARG_NAME(1);
2679 18 : size_t len1 = VARSIZE_ANY_EXHDR(arg1);
2680 18 : size_t len2 = strlen(NameStr(*arg2));
2681 18 : Oid collid = PG_GET_COLLATION();
2682 : bool result;
2683 :
2684 18 : check_collation_set(collid);
2685 :
2686 18 : if (collid == C_COLLATION_OID)
2687 0 : result = !(len1 == len2 &&
2688 0 : memcmp(VARDATA_ANY(arg1), NameStr(*arg2), len1) == 0);
2689 : else
2690 18 : result = !(varstr_cmp(VARDATA_ANY(arg1), len1,
2691 18 : NameStr(*arg2), len2,
2692 : collid) == 0);
2693 :
2694 18 : PG_FREE_IF_COPY(arg1, 0);
2695 :
2696 18 : PG_RETURN_BOOL(result);
2697 : }
2698 :
2699 : Datum
2700 145896 : btnametextcmp(PG_FUNCTION_ARGS)
2701 : {
2702 145896 : Name arg1 = PG_GETARG_NAME(0);
2703 145896 : text *arg2 = PG_GETARG_TEXT_PP(1);
2704 : int32 result;
2705 :
2706 291792 : result = varstr_cmp(NameStr(*arg1), strlen(NameStr(*arg1)),
2707 291792 : VARDATA_ANY(arg2), VARSIZE_ANY_EXHDR(arg2),
2708 : PG_GET_COLLATION());
2709 :
2710 145896 : PG_FREE_IF_COPY(arg2, 1);
2711 :
2712 145896 : PG_RETURN_INT32(result);
2713 : }
2714 :
2715 : Datum
2716 0 : bttextnamecmp(PG_FUNCTION_ARGS)
2717 : {
2718 0 : text *arg1 = PG_GETARG_TEXT_PP(0);
2719 0 : Name arg2 = PG_GETARG_NAME(1);
2720 : int32 result;
2721 :
2722 0 : result = varstr_cmp(VARDATA_ANY(arg1), VARSIZE_ANY_EXHDR(arg1),
2723 0 : NameStr(*arg2), strlen(NameStr(*arg2)),
2724 : PG_GET_COLLATION());
2725 :
2726 0 : PG_FREE_IF_COPY(arg1, 0);
2727 :
2728 0 : PG_RETURN_INT32(result);
2729 : }
2730 :
2731 : #define CmpCall(cmpfunc) \
2732 : DatumGetInt32(DirectFunctionCall2Coll(cmpfunc, \
2733 : PG_GET_COLLATION(), \
2734 : PG_GETARG_DATUM(0), \
2735 : PG_GETARG_DATUM(1)))
2736 :
2737 : Datum
2738 49818 : namelttext(PG_FUNCTION_ARGS)
2739 : {
2740 49818 : PG_RETURN_BOOL(CmpCall(btnametextcmp) < 0);
2741 : }
2742 :
2743 : Datum
2744 0 : nameletext(PG_FUNCTION_ARGS)
2745 : {
2746 0 : PG_RETURN_BOOL(CmpCall(btnametextcmp) <= 0);
2747 : }
2748 :
2749 : Datum
2750 0 : namegttext(PG_FUNCTION_ARGS)
2751 : {
2752 0 : PG_RETURN_BOOL(CmpCall(btnametextcmp) > 0);
2753 : }
2754 :
2755 : Datum
2756 31350 : namegetext(PG_FUNCTION_ARGS)
2757 : {
2758 31350 : PG_RETURN_BOOL(CmpCall(btnametextcmp) >= 0);
2759 : }
2760 :
2761 : Datum
2762 0 : textltname(PG_FUNCTION_ARGS)
2763 : {
2764 0 : PG_RETURN_BOOL(CmpCall(bttextnamecmp) < 0);
2765 : }
2766 :
2767 : Datum
2768 0 : textlename(PG_FUNCTION_ARGS)
2769 : {
2770 0 : PG_RETURN_BOOL(CmpCall(bttextnamecmp) <= 0);
2771 : }
2772 :
2773 : Datum
2774 0 : textgtname(PG_FUNCTION_ARGS)
2775 : {
2776 0 : PG_RETURN_BOOL(CmpCall(bttextnamecmp) > 0);
2777 : }
2778 :
2779 : Datum
2780 0 : textgename(PG_FUNCTION_ARGS)
2781 : {
2782 0 : PG_RETURN_BOOL(CmpCall(bttextnamecmp) >= 0);
2783 : }
2784 :
2785 : #undef CmpCall
2786 :
2787 :
2788 : /*
2789 : * The following operators support character-by-character comparison
2790 : * of text datums, to allow building indexes suitable for LIKE clauses.
2791 : * Note that the regular texteq/textne comparison operators, and regular
2792 : * support functions 1 and 2 with "C" collation are assumed to be
2793 : * compatible with these!
2794 : */
2795 :
2796 : static int
2797 152158 : internal_text_pattern_compare(text *arg1, text *arg2)
2798 : {
2799 : int result;
2800 : int len1,
2801 : len2;
2802 :
2803 152158 : len1 = VARSIZE_ANY_EXHDR(arg1);
2804 152158 : len2 = VARSIZE_ANY_EXHDR(arg2);
2805 :
2806 152158 : result = memcmp(VARDATA_ANY(arg1), VARDATA_ANY(arg2), Min(len1, len2));
2807 152158 : if (result != 0)
2808 152092 : return result;
2809 66 : else if (len1 < len2)
2810 0 : return -1;
2811 66 : else if (len1 > len2)
2812 18 : return 1;
2813 : else
2814 48 : return 0;
2815 : }
2816 :
2817 :
2818 : Datum
2819 39580 : text_pattern_lt(PG_FUNCTION_ARGS)
2820 : {
2821 39580 : text *arg1 = PG_GETARG_TEXT_PP(0);
2822 39580 : text *arg2 = PG_GETARG_TEXT_PP(1);
2823 : int result;
2824 :
2825 39580 : result = internal_text_pattern_compare(arg1, arg2);
2826 :
2827 39580 : PG_FREE_IF_COPY(arg1, 0);
2828 39580 : PG_FREE_IF_COPY(arg2, 1);
2829 :
2830 39580 : PG_RETURN_BOOL(result < 0);
2831 : }
2832 :
2833 :
2834 : Datum
2835 37510 : text_pattern_le(PG_FUNCTION_ARGS)
2836 : {
2837 37510 : text *arg1 = PG_GETARG_TEXT_PP(0);
2838 37510 : text *arg2 = PG_GETARG_TEXT_PP(1);
2839 : int result;
2840 :
2841 37510 : result = internal_text_pattern_compare(arg1, arg2);
2842 :
2843 37510 : PG_FREE_IF_COPY(arg1, 0);
2844 37510 : PG_FREE_IF_COPY(arg2, 1);
2845 :
2846 37510 : PG_RETURN_BOOL(result <= 0);
2847 : }
2848 :
2849 :
2850 : Datum
2851 37534 : text_pattern_ge(PG_FUNCTION_ARGS)
2852 : {
2853 37534 : text *arg1 = PG_GETARG_TEXT_PP(0);
2854 37534 : text *arg2 = PG_GETARG_TEXT_PP(1);
2855 : int result;
2856 :
2857 37534 : result = internal_text_pattern_compare(arg1, arg2);
2858 :
2859 37534 : PG_FREE_IF_COPY(arg1, 0);
2860 37534 : PG_FREE_IF_COPY(arg2, 1);
2861 :
2862 37534 : PG_RETURN_BOOL(result >= 0);
2863 : }
2864 :
2865 :
2866 : Datum
2867 37510 : text_pattern_gt(PG_FUNCTION_ARGS)
2868 : {
2869 37510 : text *arg1 = PG_GETARG_TEXT_PP(0);
2870 37510 : text *arg2 = PG_GETARG_TEXT_PP(1);
2871 : int result;
2872 :
2873 37510 : result = internal_text_pattern_compare(arg1, arg2);
2874 :
2875 37510 : PG_FREE_IF_COPY(arg1, 0);
2876 37510 : PG_FREE_IF_COPY(arg2, 1);
2877 :
2878 37510 : PG_RETURN_BOOL(result > 0);
2879 : }
2880 :
2881 :
2882 : Datum
2883 24 : bttext_pattern_cmp(PG_FUNCTION_ARGS)
2884 : {
2885 24 : text *arg1 = PG_GETARG_TEXT_PP(0);
2886 24 : text *arg2 = PG_GETARG_TEXT_PP(1);
2887 : int result;
2888 :
2889 24 : result = internal_text_pattern_compare(arg1, arg2);
2890 :
2891 24 : PG_FREE_IF_COPY(arg1, 0);
2892 24 : PG_FREE_IF_COPY(arg2, 1);
2893 :
2894 24 : PG_RETURN_INT32(result);
2895 : }
2896 :
2897 :
2898 : Datum
2899 116 : bttext_pattern_sortsupport(PG_FUNCTION_ARGS)
2900 : {
2901 116 : SortSupport ssup = (SortSupport) PG_GETARG_POINTER(0);
2902 : MemoryContext oldcontext;
2903 :
2904 116 : oldcontext = MemoryContextSwitchTo(ssup->ssup_cxt);
2905 :
2906 : /* Use generic string SortSupport, forcing "C" collation */
2907 116 : varstr_sortsupport(ssup, TEXTOID, C_COLLATION_OID);
2908 :
2909 116 : MemoryContextSwitchTo(oldcontext);
2910 :
2911 116 : PG_RETURN_VOID();
2912 : }
2913 :
2914 :
2915 : /*-------------------------------------------------------------
2916 : * byteaoctetlen
2917 : *
2918 : * get the number of bytes contained in an instance of type 'bytea'
2919 : *-------------------------------------------------------------
2920 : */
2921 : Datum
2922 458 : byteaoctetlen(PG_FUNCTION_ARGS)
2923 : {
2924 458 : Datum str = PG_GETARG_DATUM(0);
2925 :
2926 : /* We need not detoast the input at all */
2927 458 : PG_RETURN_INT32(toast_raw_datum_size(str) - VARHDRSZ);
2928 : }
2929 :
2930 : /*
2931 : * byteacat -
2932 : * takes two bytea* and returns a bytea* that is the concatenation of
2933 : * the two.
2934 : *
2935 : * Cloned from textcat and modified as required.
2936 : */
2937 : Datum
2938 1520 : byteacat(PG_FUNCTION_ARGS)
2939 : {
2940 1520 : bytea *t1 = PG_GETARG_BYTEA_PP(0);
2941 1520 : bytea *t2 = PG_GETARG_BYTEA_PP(1);
2942 :
2943 1520 : PG_RETURN_BYTEA_P(bytea_catenate(t1, t2));
2944 : }
2945 :
2946 : /*
2947 : * bytea_catenate
2948 : * Guts of byteacat(), broken out so it can be used by other functions
2949 : *
2950 : * Arguments can be in short-header form, but not compressed or out-of-line
2951 : */
2952 : static bytea *
2953 1556 : bytea_catenate(bytea *t1, bytea *t2)
2954 : {
2955 : bytea *result;
2956 : int len1,
2957 : len2,
2958 : len;
2959 : char *ptr;
2960 :
2961 1556 : len1 = VARSIZE_ANY_EXHDR(t1);
2962 1556 : len2 = VARSIZE_ANY_EXHDR(t2);
2963 :
2964 : /* paranoia ... probably should throw error instead? */
2965 1556 : if (len1 < 0)
2966 0 : len1 = 0;
2967 1556 : if (len2 < 0)
2968 0 : len2 = 0;
2969 :
2970 1556 : len = len1 + len2 + VARHDRSZ;
2971 1556 : result = (bytea *) palloc(len);
2972 :
2973 : /* Set size of result string... */
2974 1556 : SET_VARSIZE(result, len);
2975 :
2976 : /* Fill data field of result string... */
2977 1556 : ptr = VARDATA(result);
2978 1556 : if (len1 > 0)
2979 1556 : memcpy(ptr, VARDATA_ANY(t1), len1);
2980 1556 : if (len2 > 0)
2981 1538 : memcpy(ptr + len1, VARDATA_ANY(t2), len2);
2982 :
2983 1556 : return result;
2984 : }
2985 :
2986 : #define PG_STR_GET_BYTEA(str_) \
2987 : DatumGetByteaPP(DirectFunctionCall1(byteain, CStringGetDatum(str_)))
2988 :
2989 : /*
2990 : * bytea_substr()
2991 : * Return a substring starting at the specified position.
2992 : * Cloned from text_substr and modified as required.
2993 : *
2994 : * Input:
2995 : * - string
2996 : * - starting position (is one-based)
2997 : * - string length (optional)
2998 : *
2999 : * If the starting position is zero or less, then return from the start of the string
3000 : * adjusting the length to be consistent with the "negative start" per SQL.
3001 : * If the length is less than zero, an ERROR is thrown. If no third argument
3002 : * (length) is provided, the length to the end of the string is assumed.
3003 : */
3004 : Datum
3005 86 : bytea_substr(PG_FUNCTION_ARGS)
3006 : {
3007 86 : PG_RETURN_BYTEA_P(bytea_substring(PG_GETARG_DATUM(0),
3008 : PG_GETARG_INT32(1),
3009 : PG_GETARG_INT32(2),
3010 : false));
3011 : }
3012 :
3013 : /*
3014 : * bytea_substr_no_len -
3015 : * Wrapper to avoid opr_sanity failure due to
3016 : * one function accepting a different number of args.
3017 : */
3018 : Datum
3019 3900 : bytea_substr_no_len(PG_FUNCTION_ARGS)
3020 : {
3021 3900 : PG_RETURN_BYTEA_P(bytea_substring(PG_GETARG_DATUM(0),
3022 : PG_GETARG_INT32(1),
3023 : -1,
3024 : true));
3025 : }
3026 :
3027 : static bytea *
3028 4022 : bytea_substring(Datum str,
3029 : int S,
3030 : int L,
3031 : bool length_not_specified)
3032 : {
3033 : int32 S1; /* adjusted start position */
3034 : int32 L1; /* adjusted substring length */
3035 : int32 E; /* end position */
3036 :
3037 : /*
3038 : * The logic here should generally match text_substring().
3039 : */
3040 4022 : S1 = Max(S, 1);
3041 :
3042 4022 : if (length_not_specified)
3043 : {
3044 : /*
3045 : * Not passed a length - DatumGetByteaPSlice() grabs everything to the
3046 : * end of the string if we pass it a negative value for length.
3047 : */
3048 3918 : L1 = -1;
3049 : }
3050 104 : else if (L < 0)
3051 : {
3052 : /* SQL99 says to throw an error for E < S, i.e., negative length */
3053 12 : ereport(ERROR,
3054 : (errcode(ERRCODE_SUBSTRING_ERROR),
3055 : errmsg("negative substring length not allowed")));
3056 : L1 = -1; /* silence stupider compilers */
3057 : }
3058 92 : else if (pg_add_s32_overflow(S, L, &E))
3059 : {
3060 : /*
3061 : * L could be large enough for S + L to overflow, in which case the
3062 : * substring must run to end of string.
3063 : */
3064 6 : L1 = -1;
3065 : }
3066 : else
3067 : {
3068 : /*
3069 : * A zero or negative value for the end position can happen if the
3070 : * start was negative or one. SQL99 says to return a zero-length
3071 : * string.
3072 : */
3073 86 : if (E < 1)
3074 0 : return PG_STR_GET_BYTEA("");
3075 :
3076 86 : L1 = E - S1;
3077 : }
3078 :
3079 : /*
3080 : * If the start position is past the end of the string, SQL99 says to
3081 : * return a zero-length string -- DatumGetByteaPSlice() will do that for
3082 : * us. We need only convert S1 to zero-based starting position.
3083 : */
3084 4010 : return DatumGetByteaPSlice(str, S1 - 1, L1);
3085 : }
3086 :
3087 : /*
3088 : * byteaoverlay
3089 : * Replace specified substring of first string with second
3090 : *
3091 : * The SQL standard defines OVERLAY() in terms of substring and concatenation.
3092 : * This code is a direct implementation of what the standard says.
3093 : */
3094 : Datum
3095 6 : byteaoverlay(PG_FUNCTION_ARGS)
3096 : {
3097 6 : bytea *t1 = PG_GETARG_BYTEA_PP(0);
3098 6 : bytea *t2 = PG_GETARG_BYTEA_PP(1);
3099 6 : int sp = PG_GETARG_INT32(2); /* substring start position */
3100 6 : int sl = PG_GETARG_INT32(3); /* substring length */
3101 :
3102 6 : PG_RETURN_BYTEA_P(bytea_overlay(t1, t2, sp, sl));
3103 : }
3104 :
3105 : Datum
3106 12 : byteaoverlay_no_len(PG_FUNCTION_ARGS)
3107 : {
3108 12 : bytea *t1 = PG_GETARG_BYTEA_PP(0);
3109 12 : bytea *t2 = PG_GETARG_BYTEA_PP(1);
3110 12 : int sp = PG_GETARG_INT32(2); /* substring start position */
3111 : int sl;
3112 :
3113 12 : sl = VARSIZE_ANY_EXHDR(t2); /* defaults to length(t2) */
3114 12 : PG_RETURN_BYTEA_P(bytea_overlay(t1, t2, sp, sl));
3115 : }
3116 :
3117 : static bytea *
3118 18 : bytea_overlay(bytea *t1, bytea *t2, int sp, int sl)
3119 : {
3120 : bytea *result;
3121 : bytea *s1;
3122 : bytea *s2;
3123 : int sp_pl_sl;
3124 :
3125 : /*
3126 : * Check for possible integer-overflow cases. For negative sp, throw a
3127 : * "substring length" error because that's what should be expected
3128 : * according to the spec's definition of OVERLAY().
3129 : */
3130 18 : if (sp <= 0)
3131 0 : ereport(ERROR,
3132 : (errcode(ERRCODE_SUBSTRING_ERROR),
3133 : errmsg("negative substring length not allowed")));
3134 18 : if (pg_add_s32_overflow(sp, sl, &sp_pl_sl))
3135 0 : ereport(ERROR,
3136 : (errcode(ERRCODE_NUMERIC_VALUE_OUT_OF_RANGE),
3137 : errmsg("integer out of range")));
3138 :
3139 18 : s1 = bytea_substring(PointerGetDatum(t1), 1, sp - 1, false);
3140 18 : s2 = bytea_substring(PointerGetDatum(t1), sp_pl_sl, -1, true);
3141 18 : result = bytea_catenate(s1, t2);
3142 18 : result = bytea_catenate(result, s2);
3143 :
3144 18 : return result;
3145 : }
3146 :
3147 : /*
3148 : * bit_count
3149 : */
3150 : Datum
3151 6 : bytea_bit_count(PG_FUNCTION_ARGS)
3152 : {
3153 6 : bytea *t1 = PG_GETARG_BYTEA_PP(0);
3154 :
3155 6 : PG_RETURN_INT64(pg_popcount(VARDATA_ANY(t1), VARSIZE_ANY_EXHDR(t1)));
3156 : }
3157 :
3158 : /*
3159 : * byteapos -
3160 : * Return the position of the specified substring.
3161 : * Implements the SQL POSITION() function.
3162 : * Cloned from textpos and modified as required.
3163 : */
3164 : Datum
3165 0 : byteapos(PG_FUNCTION_ARGS)
3166 : {
3167 0 : bytea *t1 = PG_GETARG_BYTEA_PP(0);
3168 0 : bytea *t2 = PG_GETARG_BYTEA_PP(1);
3169 : int pos;
3170 : int px,
3171 : p;
3172 : int len1,
3173 : len2;
3174 : char *p1,
3175 : *p2;
3176 :
3177 0 : len1 = VARSIZE_ANY_EXHDR(t1);
3178 0 : len2 = VARSIZE_ANY_EXHDR(t2);
3179 :
3180 0 : if (len2 <= 0)
3181 0 : PG_RETURN_INT32(1); /* result for empty pattern */
3182 :
3183 0 : p1 = VARDATA_ANY(t1);
3184 0 : p2 = VARDATA_ANY(t2);
3185 :
3186 0 : pos = 0;
3187 0 : px = (len1 - len2);
3188 0 : for (p = 0; p <= px; p++)
3189 : {
3190 0 : if ((*p2 == *p1) && (memcmp(p1, p2, len2) == 0))
3191 : {
3192 0 : pos = p + 1;
3193 0 : break;
3194 : };
3195 0 : p1++;
3196 : };
3197 :
3198 0 : PG_RETURN_INT32(pos);
3199 : }
3200 :
3201 : /*-------------------------------------------------------------
3202 : * byteaGetByte
3203 : *
3204 : * this routine treats "bytea" as an array of bytes.
3205 : * It returns the Nth byte (a number between 0 and 255).
3206 : *-------------------------------------------------------------
3207 : */
3208 : Datum
3209 60 : byteaGetByte(PG_FUNCTION_ARGS)
3210 : {
3211 60 : bytea *v = PG_GETARG_BYTEA_PP(0);
3212 60 : int32 n = PG_GETARG_INT32(1);
3213 : int len;
3214 : int byte;
3215 :
3216 60 : len = VARSIZE_ANY_EXHDR(v);
3217 :
3218 60 : if (n < 0 || n >= len)
3219 6 : ereport(ERROR,
3220 : (errcode(ERRCODE_ARRAY_SUBSCRIPT_ERROR),
3221 : errmsg("index %d out of valid range, 0..%d",
3222 : n, len - 1)));
3223 :
3224 54 : byte = ((unsigned char *) VARDATA_ANY(v))[n];
3225 :
3226 54 : PG_RETURN_INT32(byte);
3227 : }
3228 :
3229 : /*-------------------------------------------------------------
3230 : * byteaGetBit
3231 : *
3232 : * This routine treats a "bytea" type like an array of bits.
3233 : * It returns the value of the Nth bit (0 or 1).
3234 : *
3235 : *-------------------------------------------------------------
3236 : */
3237 : Datum
3238 12 : byteaGetBit(PG_FUNCTION_ARGS)
3239 : {
3240 12 : bytea *v = PG_GETARG_BYTEA_PP(0);
3241 12 : int64 n = PG_GETARG_INT64(1);
3242 : int byteNo,
3243 : bitNo;
3244 : int len;
3245 : int byte;
3246 :
3247 12 : len = VARSIZE_ANY_EXHDR(v);
3248 :
3249 12 : if (n < 0 || n >= (int64) len * 8)
3250 6 : ereport(ERROR,
3251 : (errcode(ERRCODE_ARRAY_SUBSCRIPT_ERROR),
3252 : errmsg("index %lld out of valid range, 0..%lld",
3253 : (long long) n, (long long) len * 8 - 1)));
3254 :
3255 : /* n/8 is now known < len, so safe to cast to int */
3256 6 : byteNo = (int) (n / 8);
3257 6 : bitNo = (int) (n % 8);
3258 :
3259 6 : byte = ((unsigned char *) VARDATA_ANY(v))[byteNo];
3260 :
3261 6 : if (byte & (1 << bitNo))
3262 6 : PG_RETURN_INT32(1);
3263 : else
3264 0 : PG_RETURN_INT32(0);
3265 : }
3266 :
3267 : /*-------------------------------------------------------------
3268 : * byteaSetByte
3269 : *
3270 : * Given an instance of type 'bytea' creates a new one with
3271 : * the Nth byte set to the given value.
3272 : *
3273 : *-------------------------------------------------------------
3274 : */
3275 : Datum
3276 12 : byteaSetByte(PG_FUNCTION_ARGS)
3277 : {
3278 12 : bytea *res = PG_GETARG_BYTEA_P_COPY(0);
3279 12 : int32 n = PG_GETARG_INT32(1);
3280 12 : int32 newByte = PG_GETARG_INT32(2);
3281 : int len;
3282 :
3283 12 : len = VARSIZE(res) - VARHDRSZ;
3284 :
3285 12 : if (n < 0 || n >= len)
3286 6 : ereport(ERROR,
3287 : (errcode(ERRCODE_ARRAY_SUBSCRIPT_ERROR),
3288 : errmsg("index %d out of valid range, 0..%d",
3289 : n, len - 1)));
3290 :
3291 : /*
3292 : * Now set the byte.
3293 : */
3294 6 : ((unsigned char *) VARDATA(res))[n] = newByte;
3295 :
3296 6 : PG_RETURN_BYTEA_P(res);
3297 : }
3298 :
3299 : /*-------------------------------------------------------------
3300 : * byteaSetBit
3301 : *
3302 : * Given an instance of type 'bytea' creates a new one with
3303 : * the Nth bit set to the given value.
3304 : *
3305 : *-------------------------------------------------------------
3306 : */
3307 : Datum
3308 12 : byteaSetBit(PG_FUNCTION_ARGS)
3309 : {
3310 12 : bytea *res = PG_GETARG_BYTEA_P_COPY(0);
3311 12 : int64 n = PG_GETARG_INT64(1);
3312 12 : int32 newBit = PG_GETARG_INT32(2);
3313 : int len;
3314 : int oldByte,
3315 : newByte;
3316 : int byteNo,
3317 : bitNo;
3318 :
3319 12 : len = VARSIZE(res) - VARHDRSZ;
3320 :
3321 12 : if (n < 0 || n >= (int64) len * 8)
3322 6 : ereport(ERROR,
3323 : (errcode(ERRCODE_ARRAY_SUBSCRIPT_ERROR),
3324 : errmsg("index %lld out of valid range, 0..%lld",
3325 : (long long) n, (long long) len * 8 - 1)));
3326 :
3327 : /* n/8 is now known < len, so safe to cast to int */
3328 6 : byteNo = (int) (n / 8);
3329 6 : bitNo = (int) (n % 8);
3330 :
3331 : /*
3332 : * sanity check!
3333 : */
3334 6 : if (newBit != 0 && newBit != 1)
3335 0 : ereport(ERROR,
3336 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
3337 : errmsg("new bit must be 0 or 1")));
3338 :
3339 : /*
3340 : * Update the byte.
3341 : */
3342 6 : oldByte = ((unsigned char *) VARDATA(res))[byteNo];
3343 :
3344 6 : if (newBit == 0)
3345 6 : newByte = oldByte & (~(1 << bitNo));
3346 : else
3347 0 : newByte = oldByte | (1 << bitNo);
3348 :
3349 6 : ((unsigned char *) VARDATA(res))[byteNo] = newByte;
3350 :
3351 6 : PG_RETURN_BYTEA_P(res);
3352 : }
3353 :
3354 :
3355 : /* text_name()
3356 : * Converts a text type to a Name type.
3357 : */
3358 : Datum
3359 30558 : text_name(PG_FUNCTION_ARGS)
3360 : {
3361 30558 : text *s = PG_GETARG_TEXT_PP(0);
3362 : Name result;
3363 : int len;
3364 :
3365 30558 : len = VARSIZE_ANY_EXHDR(s);
3366 :
3367 : /* Truncate oversize input */
3368 30558 : if (len >= NAMEDATALEN)
3369 6 : len = pg_mbcliplen(VARDATA_ANY(s), len, NAMEDATALEN - 1);
3370 :
3371 : /* We use palloc0 here to ensure result is zero-padded */
3372 30558 : result = (Name) palloc0(NAMEDATALEN);
3373 30558 : memcpy(NameStr(*result), VARDATA_ANY(s), len);
3374 :
3375 30558 : PG_RETURN_NAME(result);
3376 : }
3377 :
3378 : /* name_text()
3379 : * Converts a Name type to a text type.
3380 : */
3381 : Datum
3382 637170 : name_text(PG_FUNCTION_ARGS)
3383 : {
3384 637170 : Name s = PG_GETARG_NAME(0);
3385 :
3386 637170 : PG_RETURN_TEXT_P(cstring_to_text(NameStr(*s)));
3387 : }
3388 :
3389 :
3390 : /*
3391 : * textToQualifiedNameList - convert a text object to list of names
3392 : *
3393 : * This implements the input parsing needed by nextval() and other
3394 : * functions that take a text parameter representing a qualified name.
3395 : * We split the name at dots, downcase if not double-quoted, and
3396 : * truncate names if they're too long.
3397 : */
3398 : List *
3399 1418 : textToQualifiedNameList(text *textval)
3400 : {
3401 : char *rawname;
3402 1418 : List *result = NIL;
3403 : List *namelist;
3404 : ListCell *l;
3405 :
3406 : /* Convert to C string (handles possible detoasting). */
3407 : /* Note we rely on being able to modify rawname below. */
3408 1418 : rawname = text_to_cstring(textval);
3409 :
3410 1418 : if (!SplitIdentifierString(rawname, '.', &namelist))
3411 0 : ereport(ERROR,
3412 : (errcode(ERRCODE_INVALID_NAME),
3413 : errmsg("invalid name syntax")));
3414 :
3415 1418 : if (namelist == NIL)
3416 0 : ereport(ERROR,
3417 : (errcode(ERRCODE_INVALID_NAME),
3418 : errmsg("invalid name syntax")));
3419 :
3420 2946 : foreach(l, namelist)
3421 : {
3422 1528 : char *curname = (char *) lfirst(l);
3423 :
3424 1528 : result = lappend(result, makeString(pstrdup(curname)));
3425 : }
3426 :
3427 1418 : pfree(rawname);
3428 1418 : list_free(namelist);
3429 :
3430 1418 : return result;
3431 : }
3432 :
3433 : /*
3434 : * SplitIdentifierString --- parse a string containing identifiers
3435 : *
3436 : * This is the guts of textToQualifiedNameList, and is exported for use in
3437 : * other situations such as parsing GUC variables. In the GUC case, it's
3438 : * important to avoid memory leaks, so the API is designed to minimize the
3439 : * amount of stuff that needs to be allocated and freed.
3440 : *
3441 : * Inputs:
3442 : * rawstring: the input string; must be overwritable! On return, it's
3443 : * been modified to contain the separated identifiers.
3444 : * separator: the separator punctuation expected between identifiers
3445 : * (typically '.' or ','). Whitespace may also appear around
3446 : * identifiers.
3447 : * Outputs:
3448 : * namelist: filled with a palloc'd list of pointers to identifiers within
3449 : * rawstring. Caller should list_free() this even on error return.
3450 : *
3451 : * Returns true if okay, false if there is a syntax error in the string.
3452 : *
3453 : * Note that an empty string is considered okay here, though not in
3454 : * textToQualifiedNameList.
3455 : */
3456 : bool
3457 124254 : SplitIdentifierString(char *rawstring, char separator,
3458 : List **namelist)
3459 : {
3460 124254 : char *nextp = rawstring;
3461 124254 : bool done = false;
3462 :
3463 124254 : *namelist = NIL;
3464 :
3465 124260 : while (scanner_isspace(*nextp))
3466 6 : nextp++; /* skip leading whitespace */
3467 :
3468 124254 : if (*nextp == '\0')
3469 24286 : return true; /* allow empty string */
3470 :
3471 : /* At the top of the loop, we are at start of a new identifier. */
3472 : do
3473 : {
3474 : char *curname;
3475 : char *endp;
3476 :
3477 167146 : if (*nextp == '"')
3478 : {
3479 : /* Quoted name --- collapse quote-quote pairs, no downcasing */
3480 32682 : curname = nextp + 1;
3481 : for (;;)
3482 : {
3483 32686 : endp = strchr(nextp + 1, '"');
3484 32684 : if (endp == NULL)
3485 0 : return false; /* mismatched quotes */
3486 32684 : if (endp[1] != '"')
3487 32682 : break; /* found end of quoted name */
3488 : /* Collapse adjacent quotes into one quote, and look again */
3489 2 : memmove(endp, endp + 1, strlen(endp));
3490 2 : nextp = endp;
3491 : }
3492 : /* endp now points at the terminating quote */
3493 32682 : nextp = endp + 1;
3494 : }
3495 : else
3496 : {
3497 : /* Unquoted name --- extends to separator or whitespace */
3498 : char *downname;
3499 : int len;
3500 :
3501 134464 : curname = nextp;
3502 1102292 : while (*nextp && *nextp != separator &&
3503 967830 : !scanner_isspace(*nextp))
3504 967828 : nextp++;
3505 134464 : endp = nextp;
3506 134464 : if (curname == nextp)
3507 0 : return false; /* empty unquoted name not allowed */
3508 :
3509 : /*
3510 : * Downcase the identifier, using same code as main lexer does.
3511 : *
3512 : * XXX because we want to overwrite the input in-place, we cannot
3513 : * support a downcasing transformation that increases the string
3514 : * length. This is not a problem given the current implementation
3515 : * of downcase_truncate_identifier, but we'll probably have to do
3516 : * something about this someday.
3517 : */
3518 134464 : len = endp - curname;
3519 134464 : downname = downcase_truncate_identifier(curname, len, false);
3520 : Assert(strlen(downname) <= len);
3521 134464 : strncpy(curname, downname, len); /* strncpy is required here */
3522 134464 : pfree(downname);
3523 : }
3524 :
3525 167148 : while (scanner_isspace(*nextp))
3526 2 : nextp++; /* skip trailing whitespace */
3527 :
3528 167146 : if (*nextp == separator)
3529 : {
3530 67178 : nextp++;
3531 116288 : while (scanner_isspace(*nextp))
3532 49110 : nextp++; /* skip leading whitespace for next */
3533 : /* we expect another name, so done remains false */
3534 : }
3535 99968 : else if (*nextp == '\0')
3536 99966 : done = true;
3537 : else
3538 2 : return false; /* invalid syntax */
3539 :
3540 : /* Now safe to overwrite separator with a null */
3541 167144 : *endp = '\0';
3542 :
3543 : /* Truncate name if it's overlength */
3544 167144 : truncate_identifier(curname, strlen(curname), false);
3545 :
3546 : /*
3547 : * Finished isolating current name --- add it to list
3548 : */
3549 167144 : *namelist = lappend(*namelist, curname);
3550 :
3551 : /* Loop back if we didn't reach end of string */
3552 167144 : } while (!done);
3553 :
3554 99966 : return true;
3555 : }
3556 :
3557 :
3558 : /*
3559 : * SplitDirectoriesString --- parse a string containing file/directory names
3560 : *
3561 : * This works fine on file names too; the function name is historical.
3562 : *
3563 : * This is similar to SplitIdentifierString, except that the parsing
3564 : * rules are meant to handle pathnames instead of identifiers: there is
3565 : * no downcasing, embedded spaces are allowed, the max length is MAXPGPATH-1,
3566 : * and we apply canonicalize_path() to each extracted string. Because of the
3567 : * last, the returned strings are separately palloc'd rather than being
3568 : * pointers into rawstring --- but we still scribble on rawstring.
3569 : *
3570 : * Inputs:
3571 : * rawstring: the input string; must be modifiable!
3572 : * separator: the separator punctuation expected between directories
3573 : * (typically ',' or ';'). Whitespace may also appear around
3574 : * directories.
3575 : * Outputs:
3576 : * namelist: filled with a palloc'd list of directory names.
3577 : * Caller should list_free_deep() this even on error return.
3578 : *
3579 : * Returns true if okay, false if there is a syntax error in the string.
3580 : *
3581 : * Note that an empty string is considered okay here.
3582 : */
3583 : bool
3584 1484 : SplitDirectoriesString(char *rawstring, char separator,
3585 : List **namelist)
3586 : {
3587 1484 : char *nextp = rawstring;
3588 1484 : bool done = false;
3589 :
3590 1484 : *namelist = NIL;
3591 :
3592 1484 : while (scanner_isspace(*nextp))
3593 0 : nextp++; /* skip leading whitespace */
3594 :
3595 1484 : if (*nextp == '\0')
3596 2 : return true; /* allow empty string */
3597 :
3598 : /* At the top of the loop, we are at start of a new directory. */
3599 : do
3600 : {
3601 : char *curname;
3602 : char *endp;
3603 :
3604 1482 : if (*nextp == '"')
3605 : {
3606 : /* Quoted name --- collapse quote-quote pairs */
3607 0 : curname = nextp + 1;
3608 : for (;;)
3609 : {
3610 0 : endp = strchr(nextp + 1, '"');
3611 0 : if (endp == NULL)
3612 0 : return false; /* mismatched quotes */
3613 0 : if (endp[1] != '"')
3614 0 : break; /* found end of quoted name */
3615 : /* Collapse adjacent quotes into one quote, and look again */
3616 0 : memmove(endp, endp + 1, strlen(endp));
3617 0 : nextp = endp;
3618 : }
3619 : /* endp now points at the terminating quote */
3620 0 : nextp = endp + 1;
3621 : }
3622 : else
3623 : {
3624 : /* Unquoted name --- extends to separator or end of string */
3625 1482 : curname = endp = nextp;
3626 24866 : while (*nextp && *nextp != separator)
3627 : {
3628 : /* trailing whitespace should not be included in name */
3629 23384 : if (!scanner_isspace(*nextp))
3630 23384 : endp = nextp + 1;
3631 23384 : nextp++;
3632 : }
3633 1482 : if (curname == endp)
3634 0 : return false; /* empty unquoted name not allowed */
3635 : }
3636 :
3637 1482 : while (scanner_isspace(*nextp))
3638 0 : nextp++; /* skip trailing whitespace */
3639 :
3640 1482 : if (*nextp == separator)
3641 : {
3642 0 : nextp++;
3643 0 : while (scanner_isspace(*nextp))
3644 0 : nextp++; /* skip leading whitespace for next */
3645 : /* we expect another name, so done remains false */
3646 : }
3647 1482 : else if (*nextp == '\0')
3648 1482 : done = true;
3649 : else
3650 0 : return false; /* invalid syntax */
3651 :
3652 : /* Now safe to overwrite separator with a null */
3653 1482 : *endp = '\0';
3654 :
3655 : /* Truncate path if it's overlength */
3656 1482 : if (strlen(curname) >= MAXPGPATH)
3657 0 : curname[MAXPGPATH - 1] = '\0';
3658 :
3659 : /*
3660 : * Finished isolating current name --- add it to list
3661 : */
3662 1482 : curname = pstrdup(curname);
3663 1482 : canonicalize_path(curname);
3664 1482 : *namelist = lappend(*namelist, curname);
3665 :
3666 : /* Loop back if we didn't reach end of string */
3667 1482 : } while (!done);
3668 :
3669 1482 : return true;
3670 : }
3671 :
3672 :
3673 : /*
3674 : * SplitGUCList --- parse a string containing identifiers or file names
3675 : *
3676 : * This is used to split the value of a GUC_LIST_QUOTE GUC variable, without
3677 : * presuming whether the elements will be taken as identifiers or file names.
3678 : * We assume the input has already been through flatten_set_variable_args(),
3679 : * so that we need never downcase (if appropriate, that was done already).
3680 : * Nor do we ever truncate, since we don't know the correct max length.
3681 : * We disallow embedded whitespace for simplicity (it shouldn't matter,
3682 : * because any embedded whitespace should have led to double-quoting).
3683 : * Otherwise the API is identical to SplitIdentifierString.
3684 : *
3685 : * XXX it's annoying to have so many copies of this string-splitting logic.
3686 : * However, it's not clear that having one function with a bunch of option
3687 : * flags would be much better.
3688 : *
3689 : * XXX there is a version of this function in src/bin/pg_dump/dumputils.c.
3690 : * Be sure to update that if you have to change this.
3691 : *
3692 : * Inputs:
3693 : * rawstring: the input string; must be overwritable! On return, it's
3694 : * been modified to contain the separated identifiers.
3695 : * separator: the separator punctuation expected between identifiers
3696 : * (typically '.' or ','). Whitespace may also appear around
3697 : * identifiers.
3698 : * Outputs:
3699 : * namelist: filled with a palloc'd list of pointers to identifiers within
3700 : * rawstring. Caller should list_free() this even on error return.
3701 : *
3702 : * Returns true if okay, false if there is a syntax error in the string.
3703 : */
3704 : bool
3705 3268 : SplitGUCList(char *rawstring, char separator,
3706 : List **namelist)
3707 : {
3708 3268 : char *nextp = rawstring;
3709 3268 : bool done = false;
3710 :
3711 3268 : *namelist = NIL;
3712 :
3713 3268 : while (scanner_isspace(*nextp))
3714 0 : nextp++; /* skip leading whitespace */
3715 :
3716 3268 : if (*nextp == '\0')
3717 3202 : return true; /* allow empty string */
3718 :
3719 : /* At the top of the loop, we are at start of a new identifier. */
3720 : do
3721 : {
3722 : char *curname;
3723 : char *endp;
3724 :
3725 92 : if (*nextp == '"')
3726 : {
3727 : /* Quoted name --- collapse quote-quote pairs */
3728 24 : curname = nextp + 1;
3729 : for (;;)
3730 : {
3731 36 : endp = strchr(nextp + 1, '"');
3732 30 : if (endp == NULL)
3733 0 : return false; /* mismatched quotes */
3734 30 : if (endp[1] != '"')
3735 24 : break; /* found end of quoted name */
3736 : /* Collapse adjacent quotes into one quote, and look again */
3737 6 : memmove(endp, endp + 1, strlen(endp));
3738 6 : nextp = endp;
3739 : }
3740 : /* endp now points at the terminating quote */
3741 24 : nextp = endp + 1;
3742 : }
3743 : else
3744 : {
3745 : /* Unquoted name --- extends to separator or whitespace */
3746 68 : curname = nextp;
3747 638 : while (*nextp && *nextp != separator &&
3748 570 : !scanner_isspace(*nextp))
3749 570 : nextp++;
3750 68 : endp = nextp;
3751 68 : if (curname == nextp)
3752 0 : return false; /* empty unquoted name not allowed */
3753 : }
3754 :
3755 92 : while (scanner_isspace(*nextp))
3756 0 : nextp++; /* skip trailing whitespace */
3757 :
3758 92 : if (*nextp == separator)
3759 : {
3760 26 : nextp++;
3761 44 : while (scanner_isspace(*nextp))
3762 18 : nextp++; /* skip leading whitespace for next */
3763 : /* we expect another name, so done remains false */
3764 : }
3765 66 : else if (*nextp == '\0')
3766 66 : done = true;
3767 : else
3768 0 : return false; /* invalid syntax */
3769 :
3770 : /* Now safe to overwrite separator with a null */
3771 92 : *endp = '\0';
3772 :
3773 : /*
3774 : * Finished isolating current name --- add it to list
3775 : */
3776 92 : *namelist = lappend(*namelist, curname);
3777 :
3778 : /* Loop back if we didn't reach end of string */
3779 92 : } while (!done);
3780 :
3781 66 : return true;
3782 : }
3783 :
3784 :
3785 : /*****************************************************************************
3786 : * Comparison Functions used for bytea
3787 : *
3788 : * Note: btree indexes need these routines not to leak memory; therefore,
3789 : * be careful to free working copies of toasted datums. Most places don't
3790 : * need to be so careful.
3791 : *****************************************************************************/
3792 :
3793 : Datum
3794 10390 : byteaeq(PG_FUNCTION_ARGS)
3795 : {
3796 10390 : Datum arg1 = PG_GETARG_DATUM(0);
3797 10390 : Datum arg2 = PG_GETARG_DATUM(1);
3798 : bool result;
3799 : Size len1,
3800 : len2;
3801 :
3802 : /*
3803 : * We can use a fast path for unequal lengths, which might save us from
3804 : * having to detoast one or both values.
3805 : */
3806 10390 : len1 = toast_raw_datum_size(arg1);
3807 10390 : len2 = toast_raw_datum_size(arg2);
3808 10390 : if (len1 != len2)
3809 4316 : result = false;
3810 : else
3811 : {
3812 6074 : bytea *barg1 = DatumGetByteaPP(arg1);
3813 6074 : bytea *barg2 = DatumGetByteaPP(arg2);
3814 :
3815 6074 : result = (memcmp(VARDATA_ANY(barg1), VARDATA_ANY(barg2),
3816 : len1 - VARHDRSZ) == 0);
3817 :
3818 6074 : PG_FREE_IF_COPY(barg1, 0);
3819 6074 : PG_FREE_IF_COPY(barg2, 1);
3820 : }
3821 :
3822 10390 : PG_RETURN_BOOL(result);
3823 : }
3824 :
3825 : Datum
3826 768 : byteane(PG_FUNCTION_ARGS)
3827 : {
3828 768 : Datum arg1 = PG_GETARG_DATUM(0);
3829 768 : Datum arg2 = PG_GETARG_DATUM(1);
3830 : bool result;
3831 : Size len1,
3832 : len2;
3833 :
3834 : /*
3835 : * We can use a fast path for unequal lengths, which might save us from
3836 : * having to detoast one or both values.
3837 : */
3838 768 : len1 = toast_raw_datum_size(arg1);
3839 768 : len2 = toast_raw_datum_size(arg2);
3840 768 : if (len1 != len2)
3841 0 : result = true;
3842 : else
3843 : {
3844 768 : bytea *barg1 = DatumGetByteaPP(arg1);
3845 768 : bytea *barg2 = DatumGetByteaPP(arg2);
3846 :
3847 768 : result = (memcmp(VARDATA_ANY(barg1), VARDATA_ANY(barg2),
3848 : len1 - VARHDRSZ) != 0);
3849 :
3850 768 : PG_FREE_IF_COPY(barg1, 0);
3851 768 : PG_FREE_IF_COPY(barg2, 1);
3852 : }
3853 :
3854 768 : PG_RETURN_BOOL(result);
3855 : }
3856 :
3857 : Datum
3858 8318 : bytealt(PG_FUNCTION_ARGS)
3859 : {
3860 8318 : bytea *arg1 = PG_GETARG_BYTEA_PP(0);
3861 8318 : bytea *arg2 = PG_GETARG_BYTEA_PP(1);
3862 : int len1,
3863 : len2;
3864 : int cmp;
3865 :
3866 8318 : len1 = VARSIZE_ANY_EXHDR(arg1);
3867 8318 : len2 = VARSIZE_ANY_EXHDR(arg2);
3868 :
3869 8318 : cmp = memcmp(VARDATA_ANY(arg1), VARDATA_ANY(arg2), Min(len1, len2));
3870 :
3871 8318 : PG_FREE_IF_COPY(arg1, 0);
3872 8318 : PG_FREE_IF_COPY(arg2, 1);
3873 :
3874 8318 : PG_RETURN_BOOL((cmp < 0) || ((cmp == 0) && (len1 < len2)));
3875 : }
3876 :
3877 : Datum
3878 6356 : byteale(PG_FUNCTION_ARGS)
3879 : {
3880 6356 : bytea *arg1 = PG_GETARG_BYTEA_PP(0);
3881 6356 : bytea *arg2 = PG_GETARG_BYTEA_PP(1);
3882 : int len1,
3883 : len2;
3884 : int cmp;
3885 :
3886 6356 : len1 = VARSIZE_ANY_EXHDR(arg1);
3887 6356 : len2 = VARSIZE_ANY_EXHDR(arg2);
3888 :
3889 6356 : cmp = memcmp(VARDATA_ANY(arg1), VARDATA_ANY(arg2), Min(len1, len2));
3890 :
3891 6356 : PG_FREE_IF_COPY(arg1, 0);
3892 6356 : PG_FREE_IF_COPY(arg2, 1);
3893 :
3894 6356 : PG_RETURN_BOOL((cmp < 0) || ((cmp == 0) && (len1 <= len2)));
3895 : }
3896 :
3897 : Datum
3898 6228 : byteagt(PG_FUNCTION_ARGS)
3899 : {
3900 6228 : bytea *arg1 = PG_GETARG_BYTEA_PP(0);
3901 6228 : bytea *arg2 = PG_GETARG_BYTEA_PP(1);
3902 : int len1,
3903 : len2;
3904 : int cmp;
3905 :
3906 6228 : len1 = VARSIZE_ANY_EXHDR(arg1);
3907 6228 : len2 = VARSIZE_ANY_EXHDR(arg2);
3908 :
3909 6228 : cmp = memcmp(VARDATA_ANY(arg1), VARDATA_ANY(arg2), Min(len1, len2));
3910 :
3911 6228 : PG_FREE_IF_COPY(arg1, 0);
3912 6228 : PG_FREE_IF_COPY(arg2, 1);
3913 :
3914 6228 : PG_RETURN_BOOL((cmp > 0) || ((cmp == 0) && (len1 > len2)));
3915 : }
3916 :
3917 : Datum
3918 5010 : byteage(PG_FUNCTION_ARGS)
3919 : {
3920 5010 : bytea *arg1 = PG_GETARG_BYTEA_PP(0);
3921 5010 : bytea *arg2 = PG_GETARG_BYTEA_PP(1);
3922 : int len1,
3923 : len2;
3924 : int cmp;
3925 :
3926 5010 : len1 = VARSIZE_ANY_EXHDR(arg1);
3927 5010 : len2 = VARSIZE_ANY_EXHDR(arg2);
3928 :
3929 5010 : cmp = memcmp(VARDATA_ANY(arg1), VARDATA_ANY(arg2), Min(len1, len2));
3930 :
3931 5010 : PG_FREE_IF_COPY(arg1, 0);
3932 5010 : PG_FREE_IF_COPY(arg2, 1);
3933 :
3934 5010 : PG_RETURN_BOOL((cmp > 0) || ((cmp == 0) && (len1 >= len2)));
3935 : }
3936 :
3937 : Datum
3938 87440 : byteacmp(PG_FUNCTION_ARGS)
3939 : {
3940 87440 : bytea *arg1 = PG_GETARG_BYTEA_PP(0);
3941 87440 : bytea *arg2 = PG_GETARG_BYTEA_PP(1);
3942 : int len1,
3943 : len2;
3944 : int cmp;
3945 :
3946 87440 : len1 = VARSIZE_ANY_EXHDR(arg1);
3947 87440 : len2 = VARSIZE_ANY_EXHDR(arg2);
3948 :
3949 87440 : cmp = memcmp(VARDATA_ANY(arg1), VARDATA_ANY(arg2), Min(len1, len2));
3950 87440 : if ((cmp == 0) && (len1 != len2))
3951 14644 : cmp = (len1 < len2) ? -1 : 1;
3952 :
3953 87440 : PG_FREE_IF_COPY(arg1, 0);
3954 87440 : PG_FREE_IF_COPY(arg2, 1);
3955 :
3956 87440 : PG_RETURN_INT32(cmp);
3957 : }
3958 :
3959 : Datum
3960 20 : bytea_sortsupport(PG_FUNCTION_ARGS)
3961 : {
3962 20 : SortSupport ssup = (SortSupport) PG_GETARG_POINTER(0);
3963 : MemoryContext oldcontext;
3964 :
3965 20 : oldcontext = MemoryContextSwitchTo(ssup->ssup_cxt);
3966 :
3967 : /* Use generic string SortSupport, forcing "C" collation */
3968 20 : varstr_sortsupport(ssup, BYTEAOID, C_COLLATION_OID);
3969 :
3970 20 : MemoryContextSwitchTo(oldcontext);
3971 :
3972 20 : PG_RETURN_VOID();
3973 : }
3974 :
3975 : /*
3976 : * appendStringInfoText
3977 : *
3978 : * Append a text to str.
3979 : * Like appendStringInfoString(str, text_to_cstring(t)) but faster.
3980 : */
3981 : static void
3982 1684440 : appendStringInfoText(StringInfo str, const text *t)
3983 : {
3984 1684440 : appendBinaryStringInfo(str, VARDATA_ANY(t), VARSIZE_ANY_EXHDR(t));
3985 1684440 : }
3986 :
3987 : /*
3988 : * replace_text
3989 : * replace all occurrences of 'old_sub_str' in 'orig_str'
3990 : * with 'new_sub_str' to form 'new_str'
3991 : *
3992 : * returns 'orig_str' if 'old_sub_str' == '' or 'orig_str' == ''
3993 : * otherwise returns 'new_str'
3994 : */
3995 : Datum
3996 1052 : replace_text(PG_FUNCTION_ARGS)
3997 : {
3998 1052 : text *src_text = PG_GETARG_TEXT_PP(0);
3999 1052 : text *from_sub_text = PG_GETARG_TEXT_PP(1);
4000 1052 : text *to_sub_text = PG_GETARG_TEXT_PP(2);
4001 : int src_text_len;
4002 : int from_sub_text_len;
4003 : TextPositionState state;
4004 : text *ret_text;
4005 : int chunk_len;
4006 : char *curr_ptr;
4007 : char *start_ptr;
4008 : StringInfoData str;
4009 : bool found;
4010 :
4011 1052 : src_text_len = VARSIZE_ANY_EXHDR(src_text);
4012 1052 : from_sub_text_len = VARSIZE_ANY_EXHDR(from_sub_text);
4013 :
4014 : /* Return unmodified source string if empty source or pattern */
4015 1052 : if (src_text_len < 1 || from_sub_text_len < 1)
4016 : {
4017 0 : PG_RETURN_TEXT_P(src_text);
4018 : }
4019 :
4020 1052 : text_position_setup(src_text, from_sub_text, PG_GET_COLLATION(), &state);
4021 :
4022 1052 : found = text_position_next(&state);
4023 :
4024 : /* When the from_sub_text is not found, there is nothing to do. */
4025 1052 : if (!found)
4026 : {
4027 250 : text_position_cleanup(&state);
4028 250 : PG_RETURN_TEXT_P(src_text);
4029 : }
4030 802 : curr_ptr = text_position_get_match_ptr(&state);
4031 802 : start_ptr = VARDATA_ANY(src_text);
4032 :
4033 802 : initStringInfo(&str);
4034 :
4035 : do
4036 : {
4037 4342 : CHECK_FOR_INTERRUPTS();
4038 :
4039 : /* copy the data skipped over by last text_position_next() */
4040 4342 : chunk_len = curr_ptr - start_ptr;
4041 4342 : appendBinaryStringInfo(&str, start_ptr, chunk_len);
4042 :
4043 4342 : appendStringInfoText(&str, to_sub_text);
4044 :
4045 4342 : start_ptr = curr_ptr + from_sub_text_len;
4046 :
4047 4342 : found = text_position_next(&state);
4048 4342 : if (found)
4049 3540 : curr_ptr = text_position_get_match_ptr(&state);
4050 : }
4051 4342 : while (found);
4052 :
4053 : /* copy trailing data */
4054 802 : chunk_len = ((char *) src_text + VARSIZE_ANY(src_text)) - start_ptr;
4055 802 : appendBinaryStringInfo(&str, start_ptr, chunk_len);
4056 :
4057 802 : text_position_cleanup(&state);
4058 :
4059 802 : ret_text = cstring_to_text_with_len(str.data, str.len);
4060 802 : pfree(str.data);
4061 :
4062 802 : PG_RETURN_TEXT_P(ret_text);
4063 : }
4064 :
4065 : /*
4066 : * check_replace_text_has_escape
4067 : *
4068 : * Returns 0 if text contains no backslashes that need processing.
4069 : * Returns 1 if text contains backslashes, but not regexp submatch specifiers.
4070 : * Returns 2 if text contains regexp submatch specifiers (\1 .. \9).
4071 : */
4072 : static int
4073 11188 : check_replace_text_has_escape(const text *replace_text)
4074 : {
4075 11188 : int result = 0;
4076 11188 : const char *p = VARDATA_ANY(replace_text);
4077 11188 : const char *p_end = p + VARSIZE_ANY_EXHDR(replace_text);
4078 :
4079 11232 : while (p < p_end)
4080 : {
4081 : /* Find next escape char, if any. */
4082 10372 : p = memchr(p, '\\', p_end - p);
4083 10372 : if (p == NULL)
4084 9550 : break;
4085 822 : p++;
4086 : /* Note: a backslash at the end doesn't require extra processing. */
4087 822 : if (p < p_end)
4088 : {
4089 822 : if (*p >= '1' && *p <= '9')
4090 778 : return 2; /* Found a submatch specifier, so done */
4091 44 : result = 1; /* Found some other sequence, keep looking */
4092 44 : p++;
4093 : }
4094 : }
4095 10410 : return result;
4096 : }
4097 :
4098 : /*
4099 : * appendStringInfoRegexpSubstr
4100 : *
4101 : * Append replace_text to str, substituting regexp back references for
4102 : * \n escapes. start_ptr is the start of the match in the source string,
4103 : * at logical character position data_pos.
4104 : */
4105 : static void
4106 236 : appendStringInfoRegexpSubstr(StringInfo str, text *replace_text,
4107 : regmatch_t *pmatch,
4108 : char *start_ptr, int data_pos)
4109 : {
4110 236 : const char *p = VARDATA_ANY(replace_text);
4111 236 : const char *p_end = p + VARSIZE_ANY_EXHDR(replace_text);
4112 :
4113 574 : while (p < p_end)
4114 : {
4115 518 : const char *chunk_start = p;
4116 : int so;
4117 : int eo;
4118 :
4119 : /* Find next escape char, if any. */
4120 518 : p = memchr(p, '\\', p_end - p);
4121 518 : if (p == NULL)
4122 174 : p = p_end;
4123 :
4124 : /* Copy the text we just scanned over, if any. */
4125 518 : if (p > chunk_start)
4126 318 : appendBinaryStringInfo(str, chunk_start, p - chunk_start);
4127 :
4128 : /* Done if at end of string, else advance over escape char. */
4129 518 : if (p >= p_end)
4130 174 : break;
4131 344 : p++;
4132 :
4133 344 : if (p >= p_end)
4134 : {
4135 : /* Escape at very end of input. Treat same as unexpected char */
4136 6 : appendStringInfoChar(str, '\\');
4137 6 : break;
4138 : }
4139 :
4140 338 : if (*p >= '1' && *p <= '9')
4141 278 : {
4142 : /* Use the back reference of regexp. */
4143 278 : int idx = *p - '0';
4144 :
4145 278 : so = pmatch[idx].rm_so;
4146 278 : eo = pmatch[idx].rm_eo;
4147 278 : p++;
4148 : }
4149 60 : else if (*p == '&')
4150 : {
4151 : /* Use the entire matched string. */
4152 18 : so = pmatch[0].rm_so;
4153 18 : eo = pmatch[0].rm_eo;
4154 18 : p++;
4155 : }
4156 42 : else if (*p == '\\')
4157 : {
4158 : /* \\ means transfer one \ to output. */
4159 36 : appendStringInfoChar(str, '\\');
4160 36 : p++;
4161 36 : continue;
4162 : }
4163 : else
4164 : {
4165 : /*
4166 : * If escape char is not followed by any expected char, just treat
4167 : * it as ordinary data to copy. (XXX would it be better to throw
4168 : * an error?)
4169 : */
4170 6 : appendStringInfoChar(str, '\\');
4171 6 : continue;
4172 : }
4173 :
4174 296 : if (so >= 0 && eo >= 0)
4175 : {
4176 : /*
4177 : * Copy the text that is back reference of regexp. Note so and eo
4178 : * are counted in characters not bytes.
4179 : */
4180 : char *chunk_start;
4181 : int chunk_len;
4182 :
4183 : Assert(so >= data_pos);
4184 296 : chunk_start = start_ptr;
4185 296 : chunk_start += charlen_to_bytelen(chunk_start, so - data_pos);
4186 296 : chunk_len = charlen_to_bytelen(chunk_start, eo - so);
4187 296 : appendBinaryStringInfo(str, chunk_start, chunk_len);
4188 : }
4189 : }
4190 236 : }
4191 :
4192 : /*
4193 : * replace_text_regexp
4194 : *
4195 : * replace substring(s) in src_text that match pattern with replace_text.
4196 : * The replace_text can contain backslash markers to substitute
4197 : * (parts of) the matched text.
4198 : *
4199 : * cflags: regexp compile flags.
4200 : * collation: collation to use.
4201 : * search_start: the character (not byte) offset in src_text at which to
4202 : * begin searching.
4203 : * n: if 0, replace all matches; if > 0, replace only the N'th match.
4204 : */
4205 : text *
4206 11188 : replace_text_regexp(text *src_text, text *pattern_text,
4207 : text *replace_text,
4208 : int cflags, Oid collation,
4209 : int search_start, int n)
4210 : {
4211 : text *ret_text;
4212 : regex_t *re;
4213 11188 : int src_text_len = VARSIZE_ANY_EXHDR(src_text);
4214 11188 : int nmatches = 0;
4215 : StringInfoData buf;
4216 : regmatch_t pmatch[10]; /* main match, plus \1 to \9 */
4217 11188 : int nmatch = lengthof(pmatch);
4218 : pg_wchar *data;
4219 : size_t data_len;
4220 : int data_pos;
4221 : char *start_ptr;
4222 : int escape_status;
4223 :
4224 11188 : initStringInfo(&buf);
4225 :
4226 : /* Convert data string to wide characters. */
4227 11188 : data = (pg_wchar *) palloc((src_text_len + 1) * sizeof(pg_wchar));
4228 11188 : data_len = pg_mb2wchar_with_len(VARDATA_ANY(src_text), data, src_text_len);
4229 :
4230 : /* Check whether replace_text has escapes, especially regexp submatches. */
4231 11188 : escape_status = check_replace_text_has_escape(replace_text);
4232 :
4233 : /* If no regexp submatches, we can use REG_NOSUB. */
4234 11188 : if (escape_status < 2)
4235 : {
4236 10410 : cflags |= REG_NOSUB;
4237 : /* Also tell pg_regexec we only want the whole-match location. */
4238 10410 : nmatch = 1;
4239 : }
4240 :
4241 : /* Prepare the regexp. */
4242 11188 : re = RE_compile_and_cache(pattern_text, cflags, collation);
4243 :
4244 : /* start_ptr points to the data_pos'th character of src_text */
4245 11188 : start_ptr = (char *) VARDATA_ANY(src_text);
4246 11188 : data_pos = 0;
4247 :
4248 16224 : while (search_start <= data_len)
4249 : {
4250 : int regexec_result;
4251 :
4252 16218 : CHECK_FOR_INTERRUPTS();
4253 :
4254 16218 : regexec_result = pg_regexec(re,
4255 : data,
4256 : data_len,
4257 : search_start,
4258 : NULL, /* no details */
4259 : nmatch,
4260 : pmatch,
4261 : 0);
4262 :
4263 16218 : if (regexec_result == REG_NOMATCH)
4264 9764 : break;
4265 :
4266 6454 : if (regexec_result != REG_OKAY)
4267 : {
4268 : char errMsg[100];
4269 :
4270 0 : pg_regerror(regexec_result, re, errMsg, sizeof(errMsg));
4271 0 : ereport(ERROR,
4272 : (errcode(ERRCODE_INVALID_REGULAR_EXPRESSION),
4273 : errmsg("regular expression failed: %s", errMsg)));
4274 : }
4275 :
4276 : /*
4277 : * Count matches, and decide whether to replace this match.
4278 : */
4279 6454 : nmatches++;
4280 6454 : if (n > 0 && nmatches != n)
4281 : {
4282 : /*
4283 : * No, so advance search_start, but not start_ptr/data_pos. (Thus,
4284 : * we treat the matched text as if it weren't matched, and copy it
4285 : * to the output later.)
4286 : */
4287 60 : search_start = pmatch[0].rm_eo;
4288 60 : if (pmatch[0].rm_so == pmatch[0].rm_eo)
4289 0 : search_start++;
4290 60 : continue;
4291 : }
4292 :
4293 : /*
4294 : * Copy the text to the left of the match position. Note we are given
4295 : * character not byte indexes.
4296 : */
4297 6394 : if (pmatch[0].rm_so - data_pos > 0)
4298 : {
4299 : int chunk_len;
4300 :
4301 6236 : chunk_len = charlen_to_bytelen(start_ptr,
4302 6236 : pmatch[0].rm_so - data_pos);
4303 6236 : appendBinaryStringInfo(&buf, start_ptr, chunk_len);
4304 :
4305 : /*
4306 : * Advance start_ptr over that text, to avoid multiple rescans of
4307 : * it if the replace_text contains multiple back-references.
4308 : */
4309 6236 : start_ptr += chunk_len;
4310 6236 : data_pos = pmatch[0].rm_so;
4311 : }
4312 :
4313 : /*
4314 : * Copy the replace_text, processing escapes if any are present.
4315 : */
4316 6394 : if (escape_status > 0)
4317 236 : appendStringInfoRegexpSubstr(&buf, replace_text, pmatch,
4318 : start_ptr, data_pos);
4319 : else
4320 6158 : appendStringInfoText(&buf, replace_text);
4321 :
4322 : /* Advance start_ptr and data_pos over the matched text. */
4323 12788 : start_ptr += charlen_to_bytelen(start_ptr,
4324 6394 : pmatch[0].rm_eo - data_pos);
4325 6394 : data_pos = pmatch[0].rm_eo;
4326 :
4327 : /*
4328 : * If we only want to replace one occurrence, we're done.
4329 : */
4330 6394 : if (n > 0)
4331 1418 : break;
4332 :
4333 : /*
4334 : * Advance search position. Normally we start the next search at the
4335 : * end of the previous match; but if the match was of zero length, we
4336 : * have to advance by one character, or we'd just find the same match
4337 : * again.
4338 : */
4339 4976 : search_start = data_pos;
4340 4976 : if (pmatch[0].rm_so == pmatch[0].rm_eo)
4341 12 : search_start++;
4342 : }
4343 :
4344 : /*
4345 : * Copy the text to the right of the last match.
4346 : */
4347 11188 : if (data_pos < data_len)
4348 : {
4349 : int chunk_len;
4350 :
4351 10682 : chunk_len = ((char *) src_text + VARSIZE_ANY(src_text)) - start_ptr;
4352 10682 : appendBinaryStringInfo(&buf, start_ptr, chunk_len);
4353 : }
4354 :
4355 11188 : ret_text = cstring_to_text_with_len(buf.data, buf.len);
4356 11188 : pfree(buf.data);
4357 11188 : pfree(data);
4358 :
4359 11188 : return ret_text;
4360 : }
4361 :
4362 : /*
4363 : * split_part
4364 : * parse input string based on provided field separator
4365 : * return N'th item (1 based, negative counts from end)
4366 : */
4367 : Datum
4368 102 : split_part(PG_FUNCTION_ARGS)
4369 : {
4370 102 : text *inputstring = PG_GETARG_TEXT_PP(0);
4371 102 : text *fldsep = PG_GETARG_TEXT_PP(1);
4372 102 : int fldnum = PG_GETARG_INT32(2);
4373 : int inputstring_len;
4374 : int fldsep_len;
4375 : TextPositionState state;
4376 : char *start_ptr;
4377 : char *end_ptr;
4378 : text *result_text;
4379 : bool found;
4380 :
4381 : /* field number is 1 based */
4382 102 : if (fldnum == 0)
4383 6 : ereport(ERROR,
4384 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
4385 : errmsg("field position must not be zero")));
4386 :
4387 96 : inputstring_len = VARSIZE_ANY_EXHDR(inputstring);
4388 96 : fldsep_len = VARSIZE_ANY_EXHDR(fldsep);
4389 :
4390 : /* return empty string for empty input string */
4391 96 : if (inputstring_len < 1)
4392 12 : PG_RETURN_TEXT_P(cstring_to_text(""));
4393 :
4394 : /* handle empty field separator */
4395 84 : if (fldsep_len < 1)
4396 : {
4397 : /* if first or last field, return input string, else empty string */
4398 24 : if (fldnum == 1 || fldnum == -1)
4399 12 : PG_RETURN_TEXT_P(inputstring);
4400 : else
4401 12 : PG_RETURN_TEXT_P(cstring_to_text(""));
4402 : }
4403 :
4404 : /* find the first field separator */
4405 60 : text_position_setup(inputstring, fldsep, PG_GET_COLLATION(), &state);
4406 :
4407 60 : found = text_position_next(&state);
4408 :
4409 : /* special case if fldsep not found at all */
4410 60 : if (!found)
4411 : {
4412 12 : text_position_cleanup(&state);
4413 : /* if first or last field, return input string, else empty string */
4414 12 : if (fldnum == 1 || fldnum == -1)
4415 6 : PG_RETURN_TEXT_P(inputstring);
4416 : else
4417 6 : PG_RETURN_TEXT_P(cstring_to_text(""));
4418 : }
4419 :
4420 : /*
4421 : * take care of a negative field number (i.e. count from the right) by
4422 : * converting to a positive field number; we need total number of fields
4423 : */
4424 48 : if (fldnum < 0)
4425 : {
4426 : /* we found a fldsep, so there are at least two fields */
4427 24 : int numfields = 2;
4428 :
4429 36 : while (text_position_next(&state))
4430 12 : numfields++;
4431 :
4432 : /* special case of last field does not require an extra pass */
4433 24 : if (fldnum == -1)
4434 : {
4435 6 : start_ptr = text_position_get_match_ptr(&state) + fldsep_len;
4436 6 : end_ptr = VARDATA_ANY(inputstring) + inputstring_len;
4437 6 : text_position_cleanup(&state);
4438 6 : PG_RETURN_TEXT_P(cstring_to_text_with_len(start_ptr,
4439 : end_ptr - start_ptr));
4440 : }
4441 :
4442 : /* else, convert fldnum to positive notation */
4443 18 : fldnum += numfields + 1;
4444 :
4445 : /* if nonexistent field, return empty string */
4446 18 : if (fldnum <= 0)
4447 : {
4448 6 : text_position_cleanup(&state);
4449 6 : PG_RETURN_TEXT_P(cstring_to_text(""));
4450 : }
4451 :
4452 : /* reset to pointing at first match, but now with positive fldnum */
4453 12 : text_position_reset(&state);
4454 12 : found = text_position_next(&state);
4455 : Assert(found);
4456 : }
4457 :
4458 : /* identify bounds of first field */
4459 36 : start_ptr = VARDATA_ANY(inputstring);
4460 36 : end_ptr = text_position_get_match_ptr(&state);
4461 :
4462 66 : while (found && --fldnum > 0)
4463 : {
4464 : /* identify bounds of next field */
4465 30 : start_ptr = end_ptr + fldsep_len;
4466 30 : found = text_position_next(&state);
4467 30 : if (found)
4468 18 : end_ptr = text_position_get_match_ptr(&state);
4469 : }
4470 :
4471 36 : text_position_cleanup(&state);
4472 :
4473 36 : if (fldnum > 0)
4474 : {
4475 : /* N'th field separator not found */
4476 : /* if last field requested, return it, else empty string */
4477 12 : if (fldnum == 1)
4478 : {
4479 6 : int last_len = start_ptr - VARDATA_ANY(inputstring);
4480 :
4481 6 : result_text = cstring_to_text_with_len(start_ptr,
4482 : inputstring_len - last_len);
4483 : }
4484 : else
4485 6 : result_text = cstring_to_text("");
4486 : }
4487 : else
4488 : {
4489 : /* non-last field requested */
4490 24 : result_text = cstring_to_text_with_len(start_ptr, end_ptr - start_ptr);
4491 : }
4492 :
4493 36 : PG_RETURN_TEXT_P(result_text);
4494 : }
4495 :
4496 : /*
4497 : * Convenience function to return true when two text params are equal.
4498 : */
4499 : static bool
4500 348 : text_isequal(text *txt1, text *txt2, Oid collid)
4501 : {
4502 348 : return DatumGetBool(DirectFunctionCall2Coll(texteq,
4503 : collid,
4504 : PointerGetDatum(txt1),
4505 : PointerGetDatum(txt2)));
4506 : }
4507 :
4508 : /*
4509 : * text_to_array
4510 : * parse input string and return text array of elements,
4511 : * based on provided field separator
4512 : */
4513 : Datum
4514 146 : text_to_array(PG_FUNCTION_ARGS)
4515 : {
4516 : SplitTextOutputData tstate;
4517 :
4518 : /* For array output, tstate should start as all zeroes */
4519 146 : memset(&tstate, 0, sizeof(tstate));
4520 :
4521 146 : if (!split_text(fcinfo, &tstate))
4522 6 : PG_RETURN_NULL();
4523 :
4524 128 : if (tstate.astate == NULL)
4525 6 : PG_RETURN_ARRAYTYPE_P(construct_empty_array(TEXTOID));
4526 :
4527 122 : PG_RETURN_DATUM(makeArrayResult(tstate.astate,
4528 : CurrentMemoryContext));
4529 : }
4530 :
4531 : /*
4532 : * text_to_array_null
4533 : * parse input string and return text array of elements,
4534 : * based on provided field separator and null string
4535 : *
4536 : * This is a separate entry point only to prevent the regression tests from
4537 : * complaining about different argument sets for the same internal function.
4538 : */
4539 : Datum
4540 60 : text_to_array_null(PG_FUNCTION_ARGS)
4541 : {
4542 60 : return text_to_array(fcinfo);
4543 : }
4544 :
4545 : /*
4546 : * text_to_table
4547 : * parse input string and return table of elements,
4548 : * based on provided field separator
4549 : */
4550 : Datum
4551 84 : text_to_table(PG_FUNCTION_ARGS)
4552 : {
4553 84 : ReturnSetInfo *rsi = (ReturnSetInfo *) fcinfo->resultinfo;
4554 : SplitTextOutputData tstate;
4555 :
4556 84 : tstate.astate = NULL;
4557 84 : InitMaterializedSRF(fcinfo, MAT_SRF_USE_EXPECTED_DESC);
4558 84 : tstate.tupstore = rsi->setResult;
4559 84 : tstate.tupdesc = rsi->setDesc;
4560 :
4561 84 : (void) split_text(fcinfo, &tstate);
4562 :
4563 84 : return (Datum) 0;
4564 : }
4565 :
4566 : /*
4567 : * text_to_table_null
4568 : * parse input string and return table of elements,
4569 : * based on provided field separator and null string
4570 : *
4571 : * This is a separate entry point only to prevent the regression tests from
4572 : * complaining about different argument sets for the same internal function.
4573 : */
4574 : Datum
4575 24 : text_to_table_null(PG_FUNCTION_ARGS)
4576 : {
4577 24 : return text_to_table(fcinfo);
4578 : }
4579 :
4580 : /*
4581 : * Common code for text_to_array, text_to_array_null, text_to_table
4582 : * and text_to_table_null functions.
4583 : *
4584 : * These are not strict so we have to test for null inputs explicitly.
4585 : * Returns false if result is to be null, else returns true.
4586 : *
4587 : * Note that if the result is valid but empty (zero elements), we return
4588 : * without changing *tstate --- caller must handle that case, too.
4589 : */
4590 : static bool
4591 230 : split_text(FunctionCallInfo fcinfo, SplitTextOutputData *tstate)
4592 : {
4593 : text *inputstring;
4594 : text *fldsep;
4595 : text *null_string;
4596 230 : Oid collation = PG_GET_COLLATION();
4597 : int inputstring_len;
4598 : int fldsep_len;
4599 : char *start_ptr;
4600 : text *result_text;
4601 :
4602 : /* when input string is NULL, then result is NULL too */
4603 230 : if (PG_ARGISNULL(0))
4604 12 : return false;
4605 :
4606 218 : inputstring = PG_GETARG_TEXT_PP(0);
4607 :
4608 : /* fldsep can be NULL */
4609 218 : if (!PG_ARGISNULL(1))
4610 188 : fldsep = PG_GETARG_TEXT_PP(1);
4611 : else
4612 30 : fldsep = NULL;
4613 :
4614 : /* null_string can be NULL or omitted */
4615 218 : if (PG_NARGS() > 2 && !PG_ARGISNULL(2))
4616 84 : null_string = PG_GETARG_TEXT_PP(2);
4617 : else
4618 134 : null_string = NULL;
4619 :
4620 218 : if (fldsep != NULL)
4621 : {
4622 : /*
4623 : * Normal case with non-null fldsep. Use the text_position machinery
4624 : * to search for occurrences of fldsep.
4625 : */
4626 : TextPositionState state;
4627 :
4628 188 : inputstring_len = VARSIZE_ANY_EXHDR(inputstring);
4629 188 : fldsep_len = VARSIZE_ANY_EXHDR(fldsep);
4630 :
4631 : /* return empty set for empty input string */
4632 188 : if (inputstring_len < 1)
4633 60 : return true;
4634 :
4635 : /* empty field separator: return input string as a one-element set */
4636 176 : if (fldsep_len < 1)
4637 : {
4638 48 : split_text_accum_result(tstate, inputstring,
4639 : null_string, collation);
4640 48 : return true;
4641 : }
4642 :
4643 128 : text_position_setup(inputstring, fldsep, collation, &state);
4644 :
4645 116 : start_ptr = VARDATA_ANY(inputstring);
4646 :
4647 : for (;;)
4648 464 : {
4649 : bool found;
4650 : char *end_ptr;
4651 : int chunk_len;
4652 :
4653 580 : CHECK_FOR_INTERRUPTS();
4654 :
4655 580 : found = text_position_next(&state);
4656 580 : if (!found)
4657 : {
4658 : /* fetch last field */
4659 116 : chunk_len = ((char *) inputstring + VARSIZE_ANY(inputstring)) - start_ptr;
4660 116 : end_ptr = NULL; /* not used, but some compilers complain */
4661 : }
4662 : else
4663 : {
4664 : /* fetch non-last field */
4665 464 : end_ptr = text_position_get_match_ptr(&state);
4666 464 : chunk_len = end_ptr - start_ptr;
4667 : }
4668 :
4669 : /* build a temp text datum to pass to split_text_accum_result */
4670 580 : result_text = cstring_to_text_with_len(start_ptr, chunk_len);
4671 :
4672 : /* stash away this field */
4673 580 : split_text_accum_result(tstate, result_text,
4674 : null_string, collation);
4675 :
4676 580 : pfree(result_text);
4677 :
4678 580 : if (!found)
4679 116 : break;
4680 :
4681 464 : start_ptr = end_ptr + fldsep_len;
4682 : }
4683 :
4684 116 : text_position_cleanup(&state);
4685 : }
4686 : else
4687 : {
4688 : /*
4689 : * When fldsep is NULL, each character in the input string becomes a
4690 : * separate element in the result set. The separator is effectively
4691 : * the space between characters.
4692 : */
4693 30 : inputstring_len = VARSIZE_ANY_EXHDR(inputstring);
4694 :
4695 30 : start_ptr = VARDATA_ANY(inputstring);
4696 :
4697 252 : while (inputstring_len > 0)
4698 : {
4699 222 : int chunk_len = pg_mblen(start_ptr);
4700 :
4701 222 : CHECK_FOR_INTERRUPTS();
4702 :
4703 : /* build a temp text datum to pass to split_text_accum_result */
4704 222 : result_text = cstring_to_text_with_len(start_ptr, chunk_len);
4705 :
4706 : /* stash away this field */
4707 222 : split_text_accum_result(tstate, result_text,
4708 : null_string, collation);
4709 :
4710 222 : pfree(result_text);
4711 :
4712 222 : start_ptr += chunk_len;
4713 222 : inputstring_len -= chunk_len;
4714 : }
4715 : }
4716 :
4717 146 : return true;
4718 : }
4719 :
4720 : /*
4721 : * Add text item to result set (table or array).
4722 : *
4723 : * This is also responsible for checking to see if the item matches
4724 : * the null_string, in which case we should emit NULL instead.
4725 : */
4726 : static void
4727 850 : split_text_accum_result(SplitTextOutputData *tstate,
4728 : text *field_value,
4729 : text *null_string,
4730 : Oid collation)
4731 : {
4732 850 : bool is_null = false;
4733 :
4734 850 : if (null_string && text_isequal(field_value, null_string, collation))
4735 60 : is_null = true;
4736 :
4737 850 : if (tstate->tupstore)
4738 : {
4739 : Datum values[1];
4740 : bool nulls[1];
4741 :
4742 228 : values[0] = PointerGetDatum(field_value);
4743 228 : nulls[0] = is_null;
4744 :
4745 228 : tuplestore_putvalues(tstate->tupstore,
4746 : tstate->tupdesc,
4747 : values,
4748 : nulls);
4749 : }
4750 : else
4751 : {
4752 622 : tstate->astate = accumArrayResult(tstate->astate,
4753 : PointerGetDatum(field_value),
4754 : is_null,
4755 : TEXTOID,
4756 : CurrentMemoryContext);
4757 : }
4758 850 : }
4759 :
4760 : /*
4761 : * array_to_text
4762 : * concatenate Cstring representation of input array elements
4763 : * using provided field separator
4764 : */
4765 : Datum
4766 65394 : array_to_text(PG_FUNCTION_ARGS)
4767 : {
4768 65394 : ArrayType *v = PG_GETARG_ARRAYTYPE_P(0);
4769 65394 : char *fldsep = text_to_cstring(PG_GETARG_TEXT_PP(1));
4770 :
4771 65394 : PG_RETURN_TEXT_P(array_to_text_internal(fcinfo, v, fldsep, NULL));
4772 : }
4773 :
4774 : /*
4775 : * array_to_text_null
4776 : * concatenate Cstring representation of input array elements
4777 : * using provided field separator and null string
4778 : *
4779 : * This version is not strict so we have to test for null inputs explicitly.
4780 : */
4781 : Datum
4782 12 : array_to_text_null(PG_FUNCTION_ARGS)
4783 : {
4784 : ArrayType *v;
4785 : char *fldsep;
4786 : char *null_string;
4787 :
4788 : /* returns NULL when first or second parameter is NULL */
4789 12 : if (PG_ARGISNULL(0) || PG_ARGISNULL(1))
4790 0 : PG_RETURN_NULL();
4791 :
4792 12 : v = PG_GETARG_ARRAYTYPE_P(0);
4793 12 : fldsep = text_to_cstring(PG_GETARG_TEXT_PP(1));
4794 :
4795 : /* NULL null string is passed through as a null pointer */
4796 12 : if (!PG_ARGISNULL(2))
4797 6 : null_string = text_to_cstring(PG_GETARG_TEXT_PP(2));
4798 : else
4799 6 : null_string = NULL;
4800 :
4801 12 : PG_RETURN_TEXT_P(array_to_text_internal(fcinfo, v, fldsep, null_string));
4802 : }
4803 :
4804 : /*
4805 : * common code for array_to_text and array_to_text_null functions
4806 : */
4807 : static text *
4808 65424 : array_to_text_internal(FunctionCallInfo fcinfo, ArrayType *v,
4809 : const char *fldsep, const char *null_string)
4810 : {
4811 : text *result;
4812 : int nitems,
4813 : *dims,
4814 : ndims;
4815 : Oid element_type;
4816 : int typlen;
4817 : bool typbyval;
4818 : char typalign;
4819 : StringInfoData buf;
4820 65424 : bool printed = false;
4821 : char *p;
4822 : bits8 *bitmap;
4823 : int bitmask;
4824 : int i;
4825 : ArrayMetaState *my_extra;
4826 :
4827 65424 : ndims = ARR_NDIM(v);
4828 65424 : dims = ARR_DIMS(v);
4829 65424 : nitems = ArrayGetNItems(ndims, dims);
4830 :
4831 : /* if there are no elements, return an empty string */
4832 65424 : if (nitems == 0)
4833 41284 : return cstring_to_text_with_len("", 0);
4834 :
4835 24140 : element_type = ARR_ELEMTYPE(v);
4836 24140 : initStringInfo(&buf);
4837 :
4838 : /*
4839 : * We arrange to look up info about element type, including its output
4840 : * conversion proc, only once per series of calls, assuming the element
4841 : * type doesn't change underneath us.
4842 : */
4843 24140 : my_extra = (ArrayMetaState *) fcinfo->flinfo->fn_extra;
4844 24140 : if (my_extra == NULL)
4845 : {
4846 1376 : fcinfo->flinfo->fn_extra = MemoryContextAlloc(fcinfo->flinfo->fn_mcxt,
4847 : sizeof(ArrayMetaState));
4848 1376 : my_extra = (ArrayMetaState *) fcinfo->flinfo->fn_extra;
4849 1376 : my_extra->element_type = ~element_type;
4850 : }
4851 :
4852 24140 : if (my_extra->element_type != element_type)
4853 : {
4854 : /*
4855 : * Get info about element type, including its output conversion proc
4856 : */
4857 1376 : get_type_io_data(element_type, IOFunc_output,
4858 : &my_extra->typlen, &my_extra->typbyval,
4859 : &my_extra->typalign, &my_extra->typdelim,
4860 : &my_extra->typioparam, &my_extra->typiofunc);
4861 1376 : fmgr_info_cxt(my_extra->typiofunc, &my_extra->proc,
4862 1376 : fcinfo->flinfo->fn_mcxt);
4863 1376 : my_extra->element_type = element_type;
4864 : }
4865 24140 : typlen = my_extra->typlen;
4866 24140 : typbyval = my_extra->typbyval;
4867 24140 : typalign = my_extra->typalign;
4868 :
4869 24140 : p = ARR_DATA_PTR(v);
4870 24140 : bitmap = ARR_NULLBITMAP(v);
4871 24140 : bitmask = 1;
4872 :
4873 82222 : for (i = 0; i < nitems; i++)
4874 : {
4875 : Datum itemvalue;
4876 : char *value;
4877 :
4878 : /* Get source element, checking for NULL */
4879 58082 : if (bitmap && (*bitmap & bitmask) == 0)
4880 : {
4881 : /* if null_string is NULL, we just ignore null elements */
4882 18 : if (null_string != NULL)
4883 : {
4884 6 : if (printed)
4885 6 : appendStringInfo(&buf, "%s%s", fldsep, null_string);
4886 : else
4887 0 : appendStringInfoString(&buf, null_string);
4888 6 : printed = true;
4889 : }
4890 : }
4891 : else
4892 : {
4893 58064 : itemvalue = fetch_att(p, typbyval, typlen);
4894 :
4895 58064 : value = OutputFunctionCall(&my_extra->proc, itemvalue);
4896 :
4897 58064 : if (printed)
4898 33924 : appendStringInfo(&buf, "%s%s", fldsep, value);
4899 : else
4900 24140 : appendStringInfoString(&buf, value);
4901 58064 : printed = true;
4902 :
4903 58064 : p = att_addlength_pointer(p, typlen, p);
4904 58064 : p = (char *) att_align_nominal(p, typalign);
4905 : }
4906 :
4907 : /* advance bitmap pointer if any */
4908 58082 : if (bitmap)
4909 : {
4910 108 : bitmask <<= 1;
4911 108 : if (bitmask == 0x100)
4912 : {
4913 0 : bitmap++;
4914 0 : bitmask = 1;
4915 : }
4916 : }
4917 : }
4918 :
4919 24140 : result = cstring_to_text_with_len(buf.data, buf.len);
4920 24140 : pfree(buf.data);
4921 :
4922 24140 : return result;
4923 : }
4924 :
4925 : /*
4926 : * Workhorse for to_bin, to_oct, and to_hex. Note that base must be > 1 and <=
4927 : * 16.
4928 : */
4929 : static inline text *
4930 38750 : convert_to_base(uint64 value, int base)
4931 : {
4932 38750 : const char *digits = "0123456789abcdef";
4933 :
4934 : /* We size the buffer for to_bin's longest possible return value. */
4935 : char buf[sizeof(uint64) * BITS_PER_BYTE];
4936 38750 : char *const end = buf + sizeof(buf);
4937 38750 : char *ptr = end;
4938 :
4939 : Assert(base > 1);
4940 : Assert(base <= 16);
4941 :
4942 : do
4943 : {
4944 75974 : *--ptr = digits[value % base];
4945 75974 : value /= base;
4946 75974 : } while (ptr > buf && value);
4947 :
4948 38750 : return cstring_to_text_with_len(ptr, end - ptr);
4949 : }
4950 :
4951 : /*
4952 : * Convert an integer to a string containing a base-2 (binary) representation
4953 : * of the number.
4954 : */
4955 : Datum
4956 12 : to_bin32(PG_FUNCTION_ARGS)
4957 : {
4958 12 : uint64 value = (uint32) PG_GETARG_INT32(0);
4959 :
4960 12 : PG_RETURN_TEXT_P(convert_to_base(value, 2));
4961 : }
4962 : Datum
4963 12 : to_bin64(PG_FUNCTION_ARGS)
4964 : {
4965 12 : uint64 value = (uint64) PG_GETARG_INT64(0);
4966 :
4967 12 : PG_RETURN_TEXT_P(convert_to_base(value, 2));
4968 : }
4969 :
4970 : /*
4971 : * Convert an integer to a string containing a base-8 (oct) representation of
4972 : * the number.
4973 : */
4974 : Datum
4975 12 : to_oct32(PG_FUNCTION_ARGS)
4976 : {
4977 12 : uint64 value = (uint32) PG_GETARG_INT32(0);
4978 :
4979 12 : PG_RETURN_TEXT_P(convert_to_base(value, 8));
4980 : }
4981 : Datum
4982 12 : to_oct64(PG_FUNCTION_ARGS)
4983 : {
4984 12 : uint64 value = (uint64) PG_GETARG_INT64(0);
4985 :
4986 12 : PG_RETURN_TEXT_P(convert_to_base(value, 8));
4987 : }
4988 :
4989 : /*
4990 : * Convert an integer to a string containing a base-16 (hex) representation of
4991 : * the number.
4992 : */
4993 : Datum
4994 38690 : to_hex32(PG_FUNCTION_ARGS)
4995 : {
4996 38690 : uint64 value = (uint32) PG_GETARG_INT32(0);
4997 :
4998 38690 : PG_RETURN_TEXT_P(convert_to_base(value, 16));
4999 : }
5000 : Datum
5001 12 : to_hex64(PG_FUNCTION_ARGS)
5002 : {
5003 12 : uint64 value = (uint64) PG_GETARG_INT64(0);
5004 :
5005 12 : PG_RETURN_TEXT_P(convert_to_base(value, 16));
5006 : }
5007 :
5008 : /*
5009 : * Return the size of a datum, possibly compressed
5010 : *
5011 : * Works on any data type
5012 : */
5013 : Datum
5014 122 : pg_column_size(PG_FUNCTION_ARGS)
5015 : {
5016 122 : Datum value = PG_GETARG_DATUM(0);
5017 : int32 result;
5018 : int typlen;
5019 :
5020 : /* On first call, get the input type's typlen, and save at *fn_extra */
5021 122 : if (fcinfo->flinfo->fn_extra == NULL)
5022 : {
5023 : /* Lookup the datatype of the supplied argument */
5024 122 : Oid argtypeid = get_fn_expr_argtype(fcinfo->flinfo, 0);
5025 :
5026 122 : typlen = get_typlen(argtypeid);
5027 122 : if (typlen == 0) /* should not happen */
5028 0 : elog(ERROR, "cache lookup failed for type %u", argtypeid);
5029 :
5030 122 : fcinfo->flinfo->fn_extra = MemoryContextAlloc(fcinfo->flinfo->fn_mcxt,
5031 : sizeof(int));
5032 122 : *((int *) fcinfo->flinfo->fn_extra) = typlen;
5033 : }
5034 : else
5035 0 : typlen = *((int *) fcinfo->flinfo->fn_extra);
5036 :
5037 122 : if (typlen == -1)
5038 : {
5039 : /* varlena type, possibly toasted */
5040 122 : result = toast_datum_size(value);
5041 : }
5042 0 : else if (typlen == -2)
5043 : {
5044 : /* cstring */
5045 0 : result = strlen(DatumGetCString(value)) + 1;
5046 : }
5047 : else
5048 : {
5049 : /* ordinary fixed-width type */
5050 0 : result = typlen;
5051 : }
5052 :
5053 122 : PG_RETURN_INT32(result);
5054 : }
5055 :
5056 : /*
5057 : * Return the compression method stored in the compressed attribute. Return
5058 : * NULL for non varlena type or uncompressed data.
5059 : */
5060 : Datum
5061 162 : pg_column_compression(PG_FUNCTION_ARGS)
5062 : {
5063 : int typlen;
5064 : char *result;
5065 : ToastCompressionId cmid;
5066 :
5067 : /* On first call, get the input type's typlen, and save at *fn_extra */
5068 162 : if (fcinfo->flinfo->fn_extra == NULL)
5069 : {
5070 : /* Lookup the datatype of the supplied argument */
5071 108 : Oid argtypeid = get_fn_expr_argtype(fcinfo->flinfo, 0);
5072 :
5073 108 : typlen = get_typlen(argtypeid);
5074 108 : if (typlen == 0) /* should not happen */
5075 0 : elog(ERROR, "cache lookup failed for type %u", argtypeid);
5076 :
5077 108 : fcinfo->flinfo->fn_extra = MemoryContextAlloc(fcinfo->flinfo->fn_mcxt,
5078 : sizeof(int));
5079 108 : *((int *) fcinfo->flinfo->fn_extra) = typlen;
5080 : }
5081 : else
5082 54 : typlen = *((int *) fcinfo->flinfo->fn_extra);
5083 :
5084 162 : if (typlen != -1)
5085 0 : PG_RETURN_NULL();
5086 :
5087 : /* get the compression method id stored in the compressed varlena */
5088 162 : cmid = toast_get_compression_id((struct varlena *)
5089 162 : DatumGetPointer(PG_GETARG_DATUM(0)));
5090 162 : if (cmid == TOAST_INVALID_COMPRESSION_ID)
5091 6 : PG_RETURN_NULL();
5092 :
5093 : /* convert compression method id to compression method name */
5094 156 : switch (cmid)
5095 : {
5096 66 : case TOAST_PGLZ_COMPRESSION_ID:
5097 66 : result = "pglz";
5098 66 : break;
5099 90 : case TOAST_LZ4_COMPRESSION_ID:
5100 90 : result = "lz4";
5101 90 : break;
5102 0 : default:
5103 0 : elog(ERROR, "invalid compression method id %d", cmid);
5104 : }
5105 :
5106 156 : PG_RETURN_TEXT_P(cstring_to_text(result));
5107 : }
5108 :
5109 : /*
5110 : * Return the chunk_id of the on-disk TOASTed value. Return NULL if the value
5111 : * is un-TOASTed or not on-disk.
5112 : */
5113 : Datum
5114 12 : pg_column_toast_chunk_id(PG_FUNCTION_ARGS)
5115 : {
5116 : int typlen;
5117 : struct varlena *attr;
5118 : struct varatt_external toast_pointer;
5119 :
5120 : /* On first call, get the input type's typlen, and save at *fn_extra */
5121 12 : if (fcinfo->flinfo->fn_extra == NULL)
5122 : {
5123 : /* Lookup the datatype of the supplied argument */
5124 12 : Oid argtypeid = get_fn_expr_argtype(fcinfo->flinfo, 0);
5125 :
5126 12 : typlen = get_typlen(argtypeid);
5127 12 : if (typlen == 0) /* should not happen */
5128 0 : elog(ERROR, "cache lookup failed for type %u", argtypeid);
5129 :
5130 12 : fcinfo->flinfo->fn_extra = MemoryContextAlloc(fcinfo->flinfo->fn_mcxt,
5131 : sizeof(int));
5132 12 : *((int *) fcinfo->flinfo->fn_extra) = typlen;
5133 : }
5134 : else
5135 0 : typlen = *((int *) fcinfo->flinfo->fn_extra);
5136 :
5137 12 : if (typlen != -1)
5138 0 : PG_RETURN_NULL();
5139 :
5140 12 : attr = (struct varlena *) DatumGetPointer(PG_GETARG_DATUM(0));
5141 :
5142 12 : if (!VARATT_IS_EXTERNAL_ONDISK(attr))
5143 6 : PG_RETURN_NULL();
5144 :
5145 6 : VARATT_EXTERNAL_GET_POINTER(toast_pointer, attr);
5146 :
5147 6 : PG_RETURN_OID(toast_pointer.va_valueid);
5148 : }
5149 :
5150 : /*
5151 : * string_agg - Concatenates values and returns string.
5152 : *
5153 : * Syntax: string_agg(value text, delimiter text) RETURNS text
5154 : *
5155 : * Note: Any NULL values are ignored. The first-call delimiter isn't
5156 : * actually used at all, and on subsequent calls the delimiter precedes
5157 : * the associated value.
5158 : */
5159 :
5160 : /* subroutine to initialize state */
5161 : static StringInfo
5162 2300 : makeStringAggState(FunctionCallInfo fcinfo)
5163 : {
5164 : StringInfo state;
5165 : MemoryContext aggcontext;
5166 : MemoryContext oldcontext;
5167 :
5168 2300 : if (!AggCheckCallContext(fcinfo, &aggcontext))
5169 : {
5170 : /* cannot be called directly because of internal-type argument */
5171 0 : elog(ERROR, "string_agg_transfn called in non-aggregate context");
5172 : }
5173 :
5174 : /*
5175 : * Create state in aggregate context. It'll stay there across subsequent
5176 : * calls.
5177 : */
5178 2300 : oldcontext = MemoryContextSwitchTo(aggcontext);
5179 2300 : state = makeStringInfo();
5180 2300 : MemoryContextSwitchTo(oldcontext);
5181 :
5182 2300 : return state;
5183 : }
5184 :
5185 : Datum
5186 852018 : string_agg_transfn(PG_FUNCTION_ARGS)
5187 : {
5188 : StringInfo state;
5189 :
5190 852018 : state = PG_ARGISNULL(0) ? NULL : (StringInfo) PG_GETARG_POINTER(0);
5191 :
5192 : /* Append the value unless null, preceding it with the delimiter. */
5193 852018 : if (!PG_ARGISNULL(1))
5194 : {
5195 836970 : text *value = PG_GETARG_TEXT_PP(1);
5196 836970 : bool isfirst = false;
5197 :
5198 : /*
5199 : * You might think we can just throw away the first delimiter, however
5200 : * we must keep it as we may be a parallel worker doing partial
5201 : * aggregation building a state to send to the main process. We need
5202 : * to keep the delimiter of every aggregation so that the combine
5203 : * function can properly join up the strings of two separately
5204 : * partially aggregated results. The first delimiter is only stripped
5205 : * off in the final function. To know how much to strip off the front
5206 : * of the string, we store the length of the first delimiter in the
5207 : * StringInfo's cursor field, which we don't otherwise need here.
5208 : */
5209 836970 : if (state == NULL)
5210 : {
5211 1912 : state = makeStringAggState(fcinfo);
5212 1912 : isfirst = true;
5213 : }
5214 :
5215 836970 : if (!PG_ARGISNULL(2))
5216 : {
5217 836970 : text *delim = PG_GETARG_TEXT_PP(2);
5218 :
5219 836970 : appendStringInfoText(state, delim);
5220 836970 : if (isfirst)
5221 1912 : state->cursor = VARSIZE_ANY_EXHDR(delim);
5222 : }
5223 :
5224 836970 : appendStringInfoText(state, value);
5225 : }
5226 :
5227 : /*
5228 : * The transition type for string_agg() is declared to be "internal",
5229 : * which is a pass-by-value type the same size as a pointer.
5230 : */
5231 852018 : if (state)
5232 851940 : PG_RETURN_POINTER(state);
5233 78 : PG_RETURN_NULL();
5234 : }
5235 :
5236 : /*
5237 : * string_agg_combine
5238 : * Aggregate combine function for string_agg(text) and string_agg(bytea)
5239 : */
5240 : Datum
5241 120 : string_agg_combine(PG_FUNCTION_ARGS)
5242 : {
5243 : StringInfo state1;
5244 : StringInfo state2;
5245 : MemoryContext agg_context;
5246 :
5247 120 : if (!AggCheckCallContext(fcinfo, &agg_context))
5248 0 : elog(ERROR, "aggregate function called in non-aggregate context");
5249 :
5250 120 : state1 = PG_ARGISNULL(0) ? NULL : (StringInfo) PG_GETARG_POINTER(0);
5251 120 : state2 = PG_ARGISNULL(1) ? NULL : (StringInfo) PG_GETARG_POINTER(1);
5252 :
5253 120 : if (state2 == NULL)
5254 : {
5255 : /*
5256 : * NULL state2 is easy, just return state1, which we know is already
5257 : * in the agg_context
5258 : */
5259 0 : if (state1 == NULL)
5260 0 : PG_RETURN_NULL();
5261 0 : PG_RETURN_POINTER(state1);
5262 : }
5263 :
5264 120 : if (state1 == NULL)
5265 : {
5266 : /* We must copy state2's data into the agg_context */
5267 : MemoryContext old_context;
5268 :
5269 120 : old_context = MemoryContextSwitchTo(agg_context);
5270 120 : state1 = makeStringAggState(fcinfo);
5271 120 : appendBinaryStringInfo(state1, state2->data, state2->len);
5272 120 : state1->cursor = state2->cursor;
5273 120 : MemoryContextSwitchTo(old_context);
5274 : }
5275 0 : else if (state2->len > 0)
5276 : {
5277 : /* Combine ... state1->cursor does not change in this case */
5278 0 : appendBinaryStringInfo(state1, state2->data, state2->len);
5279 : }
5280 :
5281 120 : PG_RETURN_POINTER(state1);
5282 : }
5283 :
5284 : /*
5285 : * string_agg_serialize
5286 : * Aggregate serialize function for string_agg(text) and string_agg(bytea)
5287 : *
5288 : * This is strict, so we need not handle NULL input
5289 : */
5290 : Datum
5291 120 : string_agg_serialize(PG_FUNCTION_ARGS)
5292 : {
5293 : StringInfo state;
5294 : StringInfoData buf;
5295 : bytea *result;
5296 :
5297 : /* cannot be called directly because of internal-type argument */
5298 : Assert(AggCheckCallContext(fcinfo, NULL));
5299 :
5300 120 : state = (StringInfo) PG_GETARG_POINTER(0);
5301 :
5302 120 : pq_begintypsend(&buf);
5303 :
5304 : /* cursor */
5305 120 : pq_sendint(&buf, state->cursor, 4);
5306 :
5307 : /* data */
5308 120 : pq_sendbytes(&buf, state->data, state->len);
5309 :
5310 120 : result = pq_endtypsend(&buf);
5311 :
5312 120 : PG_RETURN_BYTEA_P(result);
5313 : }
5314 :
5315 : /*
5316 : * string_agg_deserialize
5317 : * Aggregate deserial function for string_agg(text) and string_agg(bytea)
5318 : *
5319 : * This is strict, so we need not handle NULL input
5320 : */
5321 : Datum
5322 120 : string_agg_deserialize(PG_FUNCTION_ARGS)
5323 : {
5324 : bytea *sstate;
5325 : StringInfo result;
5326 : StringInfoData buf;
5327 : char *data;
5328 : int datalen;
5329 :
5330 : /* cannot be called directly because of internal-type argument */
5331 : Assert(AggCheckCallContext(fcinfo, NULL));
5332 :
5333 120 : sstate = PG_GETARG_BYTEA_PP(0);
5334 :
5335 : /*
5336 : * Initialize a StringInfo so that we can "receive" it using the standard
5337 : * recv-function infrastructure.
5338 : */
5339 120 : initReadOnlyStringInfo(&buf, VARDATA_ANY(sstate),
5340 120 : VARSIZE_ANY_EXHDR(sstate));
5341 :
5342 120 : result = makeStringAggState(fcinfo);
5343 :
5344 : /* cursor */
5345 120 : result->cursor = pq_getmsgint(&buf, 4);
5346 :
5347 : /* data */
5348 120 : datalen = VARSIZE_ANY_EXHDR(sstate) - 4;
5349 120 : data = (char *) pq_getmsgbytes(&buf, datalen);
5350 120 : appendBinaryStringInfo(result, data, datalen);
5351 :
5352 120 : pq_getmsgend(&buf);
5353 :
5354 120 : PG_RETURN_POINTER(result);
5355 : }
5356 :
5357 : Datum
5358 1984 : string_agg_finalfn(PG_FUNCTION_ARGS)
5359 : {
5360 : StringInfo state;
5361 :
5362 : /* cannot be called directly because of internal-type argument */
5363 : Assert(AggCheckCallContext(fcinfo, NULL));
5364 :
5365 1984 : state = PG_ARGISNULL(0) ? NULL : (StringInfo) PG_GETARG_POINTER(0);
5366 :
5367 1984 : if (state != NULL)
5368 : {
5369 : /* As per comment in transfn, strip data before the cursor position */
5370 1912 : PG_RETURN_TEXT_P(cstring_to_text_with_len(&state->data[state->cursor],
5371 : state->len - state->cursor));
5372 : }
5373 : else
5374 72 : PG_RETURN_NULL();
5375 : }
5376 :
5377 : /*
5378 : * Prepare cache with fmgr info for the output functions of the datatypes of
5379 : * the arguments of a concat-like function, beginning with argument "argidx".
5380 : * (Arguments before that will have corresponding slots in the resulting
5381 : * FmgrInfo array, but we don't fill those slots.)
5382 : */
5383 : static FmgrInfo *
5384 46 : build_concat_foutcache(FunctionCallInfo fcinfo, int argidx)
5385 : {
5386 : FmgrInfo *foutcache;
5387 : int i;
5388 :
5389 : /* We keep the info in fn_mcxt so it survives across calls */
5390 46 : foutcache = (FmgrInfo *) MemoryContextAlloc(fcinfo->flinfo->fn_mcxt,
5391 46 : PG_NARGS() * sizeof(FmgrInfo));
5392 :
5393 220 : for (i = argidx; i < PG_NARGS(); i++)
5394 : {
5395 : Oid valtype;
5396 : Oid typOutput;
5397 : bool typIsVarlena;
5398 :
5399 174 : valtype = get_fn_expr_argtype(fcinfo->flinfo, i);
5400 174 : if (!OidIsValid(valtype))
5401 0 : elog(ERROR, "could not determine data type of concat() input");
5402 :
5403 174 : getTypeOutputInfo(valtype, &typOutput, &typIsVarlena);
5404 174 : fmgr_info_cxt(typOutput, &foutcache[i], fcinfo->flinfo->fn_mcxt);
5405 : }
5406 :
5407 46 : fcinfo->flinfo->fn_extra = foutcache;
5408 :
5409 46 : return foutcache;
5410 : }
5411 :
5412 : /*
5413 : * Implementation of both concat() and concat_ws().
5414 : *
5415 : * sepstr is the separator string to place between values.
5416 : * argidx identifies the first argument to concatenate (counting from zero);
5417 : * note that this must be constant across any one series of calls.
5418 : *
5419 : * Returns NULL if result should be NULL, else text value.
5420 : */
5421 : static text *
5422 114 : concat_internal(const char *sepstr, int argidx,
5423 : FunctionCallInfo fcinfo)
5424 : {
5425 : text *result;
5426 : StringInfoData str;
5427 : FmgrInfo *foutcache;
5428 114 : bool first_arg = true;
5429 : int i;
5430 :
5431 : /*
5432 : * concat(VARIADIC some-array) is essentially equivalent to
5433 : * array_to_text(), ie concat the array elements with the given separator.
5434 : * So we just pass the case off to that code.
5435 : */
5436 114 : if (get_fn_expr_variadic(fcinfo->flinfo))
5437 : {
5438 : ArrayType *arr;
5439 :
5440 : /* Should have just the one argument */
5441 : Assert(argidx == PG_NARGS() - 1);
5442 :
5443 : /* concat(VARIADIC NULL) is defined as NULL */
5444 30 : if (PG_ARGISNULL(argidx))
5445 12 : return NULL;
5446 :
5447 : /*
5448 : * Non-null argument had better be an array. We assume that any call
5449 : * context that could let get_fn_expr_variadic return true will have
5450 : * checked that a VARIADIC-labeled parameter actually is an array. So
5451 : * it should be okay to just Assert that it's an array rather than
5452 : * doing a full-fledged error check.
5453 : */
5454 : Assert(OidIsValid(get_base_element_type(get_fn_expr_argtype(fcinfo->flinfo, argidx))));
5455 :
5456 : /* OK, safe to fetch the array value */
5457 18 : arr = PG_GETARG_ARRAYTYPE_P(argidx);
5458 :
5459 : /*
5460 : * And serialize the array. We tell array_to_text to ignore null
5461 : * elements, which matches the behavior of the loop below.
5462 : */
5463 18 : return array_to_text_internal(fcinfo, arr, sepstr, NULL);
5464 : }
5465 :
5466 : /* Normal case without explicit VARIADIC marker */
5467 84 : initStringInfo(&str);
5468 :
5469 : /* Get output function info, building it if first time through */
5470 84 : foutcache = (FmgrInfo *) fcinfo->flinfo->fn_extra;
5471 84 : if (foutcache == NULL)
5472 46 : foutcache = build_concat_foutcache(fcinfo, argidx);
5473 :
5474 372 : for (i = argidx; i < PG_NARGS(); i++)
5475 : {
5476 288 : if (!PG_ARGISNULL(i))
5477 : {
5478 210 : Datum value = PG_GETARG_DATUM(i);
5479 :
5480 : /* add separator if appropriate */
5481 210 : if (first_arg)
5482 78 : first_arg = false;
5483 : else
5484 132 : appendStringInfoString(&str, sepstr);
5485 :
5486 : /* call the appropriate type output function, append the result */
5487 210 : appendStringInfoString(&str,
5488 210 : OutputFunctionCall(&foutcache[i], value));
5489 : }
5490 : }
5491 :
5492 84 : result = cstring_to_text_with_len(str.data, str.len);
5493 84 : pfree(str.data);
5494 :
5495 84 : return result;
5496 : }
5497 :
5498 : /*
5499 : * Concatenate all arguments. NULL arguments are ignored.
5500 : */
5501 : Datum
5502 36 : text_concat(PG_FUNCTION_ARGS)
5503 : {
5504 : text *result;
5505 :
5506 36 : result = concat_internal("", 0, fcinfo);
5507 36 : if (result == NULL)
5508 6 : PG_RETURN_NULL();
5509 30 : PG_RETURN_TEXT_P(result);
5510 : }
5511 :
5512 : /*
5513 : * Concatenate all but first argument value with separators. The first
5514 : * parameter is used as the separator. NULL arguments are ignored.
5515 : */
5516 : Datum
5517 84 : text_concat_ws(PG_FUNCTION_ARGS)
5518 : {
5519 : char *sep;
5520 : text *result;
5521 :
5522 : /* return NULL when separator is NULL */
5523 84 : if (PG_ARGISNULL(0))
5524 6 : PG_RETURN_NULL();
5525 78 : sep = text_to_cstring(PG_GETARG_TEXT_PP(0));
5526 :
5527 78 : result = concat_internal(sep, 1, fcinfo);
5528 78 : if (result == NULL)
5529 6 : PG_RETURN_NULL();
5530 72 : PG_RETURN_TEXT_P(result);
5531 : }
5532 :
5533 : /*
5534 : * Return first n characters in the string. When n is negative,
5535 : * return all but last |n| characters.
5536 : */
5537 : Datum
5538 1908 : text_left(PG_FUNCTION_ARGS)
5539 : {
5540 1908 : int n = PG_GETARG_INT32(1);
5541 :
5542 1908 : if (n < 0)
5543 : {
5544 30 : text *str = PG_GETARG_TEXT_PP(0);
5545 30 : const char *p = VARDATA_ANY(str);
5546 30 : int len = VARSIZE_ANY_EXHDR(str);
5547 : int rlen;
5548 :
5549 30 : n = pg_mbstrlen_with_len(p, len) + n;
5550 30 : rlen = pg_mbcharcliplen(p, len, n);
5551 30 : PG_RETURN_TEXT_P(cstring_to_text_with_len(p, rlen));
5552 : }
5553 : else
5554 1878 : PG_RETURN_TEXT_P(text_substring(PG_GETARG_DATUM(0), 1, n, false));
5555 : }
5556 :
5557 : /*
5558 : * Return last n characters in the string. When n is negative,
5559 : * return all but first |n| characters.
5560 : */
5561 : Datum
5562 66 : text_right(PG_FUNCTION_ARGS)
5563 : {
5564 66 : text *str = PG_GETARG_TEXT_PP(0);
5565 66 : const char *p = VARDATA_ANY(str);
5566 66 : int len = VARSIZE_ANY_EXHDR(str);
5567 66 : int n = PG_GETARG_INT32(1);
5568 : int off;
5569 :
5570 66 : if (n < 0)
5571 30 : n = -n;
5572 : else
5573 36 : n = pg_mbstrlen_with_len(p, len) - n;
5574 66 : off = pg_mbcharcliplen(p, len, n);
5575 :
5576 66 : PG_RETURN_TEXT_P(cstring_to_text_with_len(p + off, len - off));
5577 : }
5578 :
5579 : /*
5580 : * Return reversed string
5581 : */
5582 : Datum
5583 6 : text_reverse(PG_FUNCTION_ARGS)
5584 : {
5585 6 : text *str = PG_GETARG_TEXT_PP(0);
5586 6 : const char *p = VARDATA_ANY(str);
5587 6 : int len = VARSIZE_ANY_EXHDR(str);
5588 6 : const char *endp = p + len;
5589 : text *result;
5590 : char *dst;
5591 :
5592 6 : result = palloc(len + VARHDRSZ);
5593 6 : dst = (char *) VARDATA(result) + len;
5594 6 : SET_VARSIZE(result, len + VARHDRSZ);
5595 :
5596 6 : if (pg_database_encoding_max_length() > 1)
5597 : {
5598 : /* multibyte version */
5599 36 : while (p < endp)
5600 : {
5601 : int sz;
5602 :
5603 30 : sz = pg_mblen(p);
5604 30 : dst -= sz;
5605 30 : memcpy(dst, p, sz);
5606 30 : p += sz;
5607 : }
5608 : }
5609 : else
5610 : {
5611 : /* single byte version */
5612 0 : while (p < endp)
5613 0 : *(--dst) = *p++;
5614 : }
5615 :
5616 6 : PG_RETURN_TEXT_P(result);
5617 : }
5618 :
5619 :
5620 : /*
5621 : * Support macros for text_format()
5622 : */
5623 : #define TEXT_FORMAT_FLAG_MINUS 0x0001 /* is minus flag present? */
5624 :
5625 : #define ADVANCE_PARSE_POINTER(ptr,end_ptr) \
5626 : do { \
5627 : if (++(ptr) >= (end_ptr)) \
5628 : ereport(ERROR, \
5629 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE), \
5630 : errmsg("unterminated format() type specifier"), \
5631 : errhint("For a single \"%%\" use \"%%%%\"."))); \
5632 : } while (0)
5633 :
5634 : /*
5635 : * Returns a formatted string
5636 : */
5637 : Datum
5638 26132 : text_format(PG_FUNCTION_ARGS)
5639 : {
5640 : text *fmt;
5641 : StringInfoData str;
5642 : const char *cp;
5643 : const char *start_ptr;
5644 : const char *end_ptr;
5645 : text *result;
5646 : int arg;
5647 : bool funcvariadic;
5648 : int nargs;
5649 26132 : Datum *elements = NULL;
5650 26132 : bool *nulls = NULL;
5651 26132 : Oid element_type = InvalidOid;
5652 26132 : Oid prev_type = InvalidOid;
5653 26132 : Oid prev_width_type = InvalidOid;
5654 : FmgrInfo typoutputfinfo;
5655 : FmgrInfo typoutputinfo_width;
5656 :
5657 : /* When format string is null, immediately return null */
5658 26132 : if (PG_ARGISNULL(0))
5659 6 : PG_RETURN_NULL();
5660 :
5661 : /* If argument is marked VARIADIC, expand array into elements */
5662 26126 : if (get_fn_expr_variadic(fcinfo->flinfo))
5663 : {
5664 : ArrayType *arr;
5665 : int16 elmlen;
5666 : bool elmbyval;
5667 : char elmalign;
5668 : int nitems;
5669 :
5670 : /* Should have just the one argument */
5671 : Assert(PG_NARGS() == 2);
5672 :
5673 : /* If argument is NULL, we treat it as zero-length array */
5674 48 : if (PG_ARGISNULL(1))
5675 6 : nitems = 0;
5676 : else
5677 : {
5678 : /*
5679 : * Non-null argument had better be an array. We assume that any
5680 : * call context that could let get_fn_expr_variadic return true
5681 : * will have checked that a VARIADIC-labeled parameter actually is
5682 : * an array. So it should be okay to just Assert that it's an
5683 : * array rather than doing a full-fledged error check.
5684 : */
5685 : Assert(OidIsValid(get_base_element_type(get_fn_expr_argtype(fcinfo->flinfo, 1))));
5686 :
5687 : /* OK, safe to fetch the array value */
5688 42 : arr = PG_GETARG_ARRAYTYPE_P(1);
5689 :
5690 : /* Get info about array element type */
5691 42 : element_type = ARR_ELEMTYPE(arr);
5692 42 : get_typlenbyvalalign(element_type,
5693 : &elmlen, &elmbyval, &elmalign);
5694 :
5695 : /* Extract all array elements */
5696 42 : deconstruct_array(arr, element_type, elmlen, elmbyval, elmalign,
5697 : &elements, &nulls, &nitems);
5698 : }
5699 :
5700 48 : nargs = nitems + 1;
5701 48 : funcvariadic = true;
5702 : }
5703 : else
5704 : {
5705 : /* Non-variadic case, we'll process the arguments individually */
5706 26078 : nargs = PG_NARGS();
5707 26078 : funcvariadic = false;
5708 : }
5709 :
5710 : /* Setup for main loop. */
5711 26126 : fmt = PG_GETARG_TEXT_PP(0);
5712 26126 : start_ptr = VARDATA_ANY(fmt);
5713 26126 : end_ptr = start_ptr + VARSIZE_ANY_EXHDR(fmt);
5714 26126 : initStringInfo(&str);
5715 26126 : arg = 1; /* next argument position to print */
5716 :
5717 : /* Scan format string, looking for conversion specifiers. */
5718 746632 : for (cp = start_ptr; cp < end_ptr; cp++)
5719 : {
5720 : int argpos;
5721 : int widthpos;
5722 : int flags;
5723 : int width;
5724 : Datum value;
5725 : bool isNull;
5726 : Oid typid;
5727 :
5728 : /*
5729 : * If it's not the start of a conversion specifier, just copy it to
5730 : * the output buffer.
5731 : */
5732 720566 : if (*cp != '%')
5733 : {
5734 659820 : appendStringInfoCharMacro(&str, *cp);
5735 659838 : continue;
5736 : }
5737 :
5738 60746 : ADVANCE_PARSE_POINTER(cp, end_ptr);
5739 :
5740 : /* Easy case: %% outputs a single % */
5741 60746 : if (*cp == '%')
5742 : {
5743 18 : appendStringInfoCharMacro(&str, *cp);
5744 18 : continue;
5745 : }
5746 :
5747 : /* Parse the optional portions of the format specifier */
5748 60728 : cp = text_format_parse_format(cp, end_ptr,
5749 : &argpos, &widthpos,
5750 : &flags, &width);
5751 :
5752 : /*
5753 : * Next we should see the main conversion specifier. Whether or not
5754 : * an argument position was present, it's known that at least one
5755 : * character remains in the string at this point. Experience suggests
5756 : * that it's worth checking that that character is one of the expected
5757 : * ones before we try to fetch arguments, so as to produce the least
5758 : * confusing response to a mis-formatted specifier.
5759 : */
5760 60704 : if (strchr("sIL", *cp) == NULL)
5761 6 : ereport(ERROR,
5762 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
5763 : errmsg("unrecognized format() type specifier \"%.*s\"",
5764 : pg_mblen(cp), cp),
5765 : errhint("For a single \"%%\" use \"%%%%\".")));
5766 :
5767 : /* If indirect width was specified, get its value */
5768 60698 : if (widthpos >= 0)
5769 : {
5770 : /* Collect the specified or next argument position */
5771 42 : if (widthpos > 0)
5772 36 : arg = widthpos;
5773 42 : if (arg >= nargs)
5774 0 : ereport(ERROR,
5775 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
5776 : errmsg("too few arguments for format()")));
5777 :
5778 : /* Get the value and type of the selected argument */
5779 42 : if (!funcvariadic)
5780 : {
5781 42 : value = PG_GETARG_DATUM(arg);
5782 42 : isNull = PG_ARGISNULL(arg);
5783 42 : typid = get_fn_expr_argtype(fcinfo->flinfo, arg);
5784 : }
5785 : else
5786 : {
5787 0 : value = elements[arg - 1];
5788 0 : isNull = nulls[arg - 1];
5789 0 : typid = element_type;
5790 : }
5791 42 : if (!OidIsValid(typid))
5792 0 : elog(ERROR, "could not determine data type of format() input");
5793 :
5794 42 : arg++;
5795 :
5796 : /* We can treat NULL width the same as zero */
5797 42 : if (isNull)
5798 6 : width = 0;
5799 36 : else if (typid == INT4OID)
5800 36 : width = DatumGetInt32(value);
5801 0 : else if (typid == INT2OID)
5802 0 : width = DatumGetInt16(value);
5803 : else
5804 : {
5805 : /* For less-usual datatypes, convert to text then to int */
5806 : char *str;
5807 :
5808 0 : if (typid != prev_width_type)
5809 : {
5810 : Oid typoutputfunc;
5811 : bool typIsVarlena;
5812 :
5813 0 : getTypeOutputInfo(typid, &typoutputfunc, &typIsVarlena);
5814 0 : fmgr_info(typoutputfunc, &typoutputinfo_width);
5815 0 : prev_width_type = typid;
5816 : }
5817 :
5818 0 : str = OutputFunctionCall(&typoutputinfo_width, value);
5819 :
5820 : /* pg_strtoint32 will complain about bad data or overflow */
5821 0 : width = pg_strtoint32(str);
5822 :
5823 0 : pfree(str);
5824 : }
5825 : }
5826 :
5827 : /* Collect the specified or next argument position */
5828 60698 : if (argpos > 0)
5829 132 : arg = argpos;
5830 60698 : if (arg >= nargs)
5831 24 : ereport(ERROR,
5832 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
5833 : errmsg("too few arguments for format()")));
5834 :
5835 : /* Get the value and type of the selected argument */
5836 60674 : if (!funcvariadic)
5837 : {
5838 59402 : value = PG_GETARG_DATUM(arg);
5839 59402 : isNull = PG_ARGISNULL(arg);
5840 59402 : typid = get_fn_expr_argtype(fcinfo->flinfo, arg);
5841 : }
5842 : else
5843 : {
5844 1272 : value = elements[arg - 1];
5845 1272 : isNull = nulls[arg - 1];
5846 1272 : typid = element_type;
5847 : }
5848 60674 : if (!OidIsValid(typid))
5849 0 : elog(ERROR, "could not determine data type of format() input");
5850 :
5851 60674 : arg++;
5852 :
5853 : /*
5854 : * Get the appropriate typOutput function, reusing previous one if
5855 : * same type as previous argument. That's particularly useful in the
5856 : * variadic-array case, but often saves work even for ordinary calls.
5857 : */
5858 60674 : if (typid != prev_type)
5859 : {
5860 : Oid typoutputfunc;
5861 : bool typIsVarlena;
5862 :
5863 29240 : getTypeOutputInfo(typid, &typoutputfunc, &typIsVarlena);
5864 29240 : fmgr_info(typoutputfunc, &typoutputfinfo);
5865 29240 : prev_type = typid;
5866 : }
5867 :
5868 : /*
5869 : * And now we can format the value.
5870 : */
5871 60674 : switch (*cp)
5872 : {
5873 60674 : case 's':
5874 : case 'I':
5875 : case 'L':
5876 60674 : text_format_string_conversion(&str, *cp, &typoutputfinfo,
5877 : value, isNull,
5878 : flags, width);
5879 60668 : break;
5880 0 : default:
5881 : /* should not get here, because of previous check */
5882 0 : ereport(ERROR,
5883 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
5884 : errmsg("unrecognized format() type specifier \"%.*s\"",
5885 : pg_mblen(cp), cp),
5886 : errhint("For a single \"%%\" use \"%%%%\".")));
5887 : break;
5888 : }
5889 : }
5890 :
5891 : /* Don't need deconstruct_array results anymore. */
5892 26066 : if (elements != NULL)
5893 42 : pfree(elements);
5894 26066 : if (nulls != NULL)
5895 42 : pfree(nulls);
5896 :
5897 : /* Generate results. */
5898 26066 : result = cstring_to_text_with_len(str.data, str.len);
5899 26066 : pfree(str.data);
5900 :
5901 26066 : PG_RETURN_TEXT_P(result);
5902 : }
5903 :
5904 : /*
5905 : * Parse contiguous digits as a decimal number.
5906 : *
5907 : * Returns true if some digits could be parsed.
5908 : * The value is returned into *value, and *ptr is advanced to the next
5909 : * character to be parsed.
5910 : *
5911 : * Note parsing invariant: at least one character is known available before
5912 : * string end (end_ptr) at entry, and this is still true at exit.
5913 : */
5914 : static bool
5915 121420 : text_format_parse_digits(const char **ptr, const char *end_ptr, int *value)
5916 : {
5917 121420 : bool found = false;
5918 121420 : const char *cp = *ptr;
5919 121420 : int val = 0;
5920 :
5921 121732 : while (*cp >= '0' && *cp <= '9')
5922 : {
5923 318 : int8 digit = (*cp - '0');
5924 :
5925 318 : if (unlikely(pg_mul_s32_overflow(val, 10, &val)) ||
5926 318 : unlikely(pg_add_s32_overflow(val, digit, &val)))
5927 0 : ereport(ERROR,
5928 : (errcode(ERRCODE_NUMERIC_VALUE_OUT_OF_RANGE),
5929 : errmsg("number is out of range")));
5930 318 : ADVANCE_PARSE_POINTER(cp, end_ptr);
5931 312 : found = true;
5932 : }
5933 :
5934 121414 : *ptr = cp;
5935 121414 : *value = val;
5936 :
5937 121414 : return found;
5938 : }
5939 :
5940 : /*
5941 : * Parse a format specifier (generally following the SUS printf spec).
5942 : *
5943 : * We have already advanced over the initial '%', and we are looking for
5944 : * [argpos][flags][width]type (but the type character is not consumed here).
5945 : *
5946 : * Inputs are start_ptr (the position after '%') and end_ptr (string end + 1).
5947 : * Output parameters:
5948 : * argpos: argument position for value to be printed. -1 means unspecified.
5949 : * widthpos: argument position for width. Zero means the argument position
5950 : * was unspecified (ie, take the next arg) and -1 means no width
5951 : * argument (width was omitted or specified as a constant).
5952 : * flags: bitmask of flags.
5953 : * width: directly-specified width value. Zero means the width was omitted
5954 : * (note it's not necessary to distinguish this case from an explicit
5955 : * zero width value).
5956 : *
5957 : * The function result is the next character position to be parsed, ie, the
5958 : * location where the type character is/should be.
5959 : *
5960 : * Note parsing invariant: at least one character is known available before
5961 : * string end (end_ptr) at entry, and this is still true at exit.
5962 : */
5963 : static const char *
5964 60728 : text_format_parse_format(const char *start_ptr, const char *end_ptr,
5965 : int *argpos, int *widthpos,
5966 : int *flags, int *width)
5967 : {
5968 60728 : const char *cp = start_ptr;
5969 : int n;
5970 :
5971 : /* set defaults for output parameters */
5972 60728 : *argpos = -1;
5973 60728 : *widthpos = -1;
5974 60728 : *flags = 0;
5975 60728 : *width = 0;
5976 :
5977 : /* try to identify first number */
5978 60728 : if (text_format_parse_digits(&cp, end_ptr, &n))
5979 : {
5980 174 : if (*cp != '$')
5981 : {
5982 : /* Must be just a width and a type, so we're done */
5983 24 : *width = n;
5984 24 : return cp;
5985 : }
5986 : /* The number was argument position */
5987 150 : *argpos = n;
5988 : /* Explicit 0 for argument index is immediately refused */
5989 150 : if (n == 0)
5990 6 : ereport(ERROR,
5991 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
5992 : errmsg("format specifies argument 0, but arguments are numbered from 1")));
5993 144 : ADVANCE_PARSE_POINTER(cp, end_ptr);
5994 : }
5995 :
5996 : /* Handle flags (only minus is supported now) */
5997 60722 : while (*cp == '-')
5998 : {
5999 30 : *flags |= TEXT_FORMAT_FLAG_MINUS;
6000 30 : ADVANCE_PARSE_POINTER(cp, end_ptr);
6001 : }
6002 :
6003 60692 : if (*cp == '*')
6004 : {
6005 : /* Handle indirect width */
6006 48 : ADVANCE_PARSE_POINTER(cp, end_ptr);
6007 48 : if (text_format_parse_digits(&cp, end_ptr, &n))
6008 : {
6009 : /* number in this position must be closed by $ */
6010 42 : if (*cp != '$')
6011 0 : ereport(ERROR,
6012 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
6013 : errmsg("width argument position must be ended by \"$\"")));
6014 : /* The number was width argument position */
6015 42 : *widthpos = n;
6016 : /* Explicit 0 for argument index is immediately refused */
6017 42 : if (n == 0)
6018 6 : ereport(ERROR,
6019 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
6020 : errmsg("format specifies argument 0, but arguments are numbered from 1")));
6021 36 : ADVANCE_PARSE_POINTER(cp, end_ptr);
6022 : }
6023 : else
6024 6 : *widthpos = 0; /* width's argument position is unspecified */
6025 : }
6026 : else
6027 : {
6028 : /* Check for direct width specification */
6029 60644 : if (text_format_parse_digits(&cp, end_ptr, &n))
6030 30 : *width = n;
6031 : }
6032 :
6033 : /* cp should now be pointing at type character */
6034 60680 : return cp;
6035 : }
6036 :
6037 : /*
6038 : * Format a %s, %I, or %L conversion
6039 : */
6040 : static void
6041 60674 : text_format_string_conversion(StringInfo buf, char conversion,
6042 : FmgrInfo *typOutputInfo,
6043 : Datum value, bool isNull,
6044 : int flags, int width)
6045 : {
6046 : char *str;
6047 :
6048 : /* Handle NULL arguments before trying to stringify the value. */
6049 60674 : if (isNull)
6050 : {
6051 342 : if (conversion == 's')
6052 270 : text_format_append_string(buf, "", flags, width);
6053 72 : else if (conversion == 'L')
6054 66 : text_format_append_string(buf, "NULL", flags, width);
6055 6 : else if (conversion == 'I')
6056 6 : ereport(ERROR,
6057 : (errcode(ERRCODE_NULL_VALUE_NOT_ALLOWED),
6058 : errmsg("null values cannot be formatted as an SQL identifier")));
6059 336 : return;
6060 : }
6061 :
6062 : /* Stringify. */
6063 60332 : str = OutputFunctionCall(typOutputInfo, value);
6064 :
6065 : /* Escape. */
6066 60332 : if (conversion == 'I')
6067 : {
6068 : /* quote_identifier may or may not allocate a new string. */
6069 3106 : text_format_append_string(buf, quote_identifier(str), flags, width);
6070 : }
6071 57226 : else if (conversion == 'L')
6072 : {
6073 3232 : char *qstr = quote_literal_cstr(str);
6074 :
6075 3232 : text_format_append_string(buf, qstr, flags, width);
6076 : /* quote_literal_cstr() always allocates a new string */
6077 3232 : pfree(qstr);
6078 : }
6079 : else
6080 53994 : text_format_append_string(buf, str, flags, width);
6081 :
6082 : /* Cleanup. */
6083 60332 : pfree(str);
6084 : }
6085 :
6086 : /*
6087 : * Append str to buf, padding as directed by flags/width
6088 : */
6089 : static void
6090 60668 : text_format_append_string(StringInfo buf, const char *str,
6091 : int flags, int width)
6092 : {
6093 60668 : bool align_to_left = false;
6094 : int len;
6095 :
6096 : /* fast path for typical easy case */
6097 60668 : if (width == 0)
6098 : {
6099 60584 : appendStringInfoString(buf, str);
6100 60584 : return;
6101 : }
6102 :
6103 84 : if (width < 0)
6104 : {
6105 : /* Negative width: implicit '-' flag, then take absolute value */
6106 6 : align_to_left = true;
6107 : /* -INT_MIN is undefined */
6108 6 : if (width <= INT_MIN)
6109 0 : ereport(ERROR,
6110 : (errcode(ERRCODE_NUMERIC_VALUE_OUT_OF_RANGE),
6111 : errmsg("number is out of range")));
6112 6 : width = -width;
6113 : }
6114 78 : else if (flags & TEXT_FORMAT_FLAG_MINUS)
6115 24 : align_to_left = true;
6116 :
6117 84 : len = pg_mbstrlen(str);
6118 84 : if (align_to_left)
6119 : {
6120 : /* left justify */
6121 30 : appendStringInfoString(buf, str);
6122 30 : if (len < width)
6123 30 : appendStringInfoSpaces(buf, width - len);
6124 : }
6125 : else
6126 : {
6127 : /* right justify */
6128 54 : if (len < width)
6129 54 : appendStringInfoSpaces(buf, width - len);
6130 54 : appendStringInfoString(buf, str);
6131 : }
6132 : }
6133 :
6134 : /*
6135 : * text_format_nv - nonvariadic wrapper for text_format function.
6136 : *
6137 : * note: this wrapper is necessary to pass the sanity check in opr_sanity,
6138 : * which checks that all built-in functions that share the implementing C
6139 : * function take the same number of arguments.
6140 : */
6141 : Datum
6142 30 : text_format_nv(PG_FUNCTION_ARGS)
6143 : {
6144 30 : return text_format(fcinfo);
6145 : }
6146 :
6147 : /*
6148 : * Helper function for Levenshtein distance functions. Faster than memcmp(),
6149 : * for this use case.
6150 : */
6151 : static inline bool
6152 0 : rest_of_char_same(const char *s1, const char *s2, int len)
6153 : {
6154 0 : while (len > 0)
6155 : {
6156 0 : len--;
6157 0 : if (s1[len] != s2[len])
6158 0 : return false;
6159 : }
6160 0 : return true;
6161 : }
6162 :
6163 : /* Expand each Levenshtein distance variant */
6164 : #include "levenshtein.c"
6165 : #define LEVENSHTEIN_LESS_EQUAL
6166 : #include "levenshtein.c"
6167 :
6168 :
6169 : /*
6170 : * The following *ClosestMatch() functions can be used to determine whether a
6171 : * user-provided string resembles any known valid values, which is useful for
6172 : * providing hints in log messages, among other things. Use these functions
6173 : * like so:
6174 : *
6175 : * initClosestMatch(&state, source_string, max_distance);
6176 : *
6177 : * for (int i = 0; i < num_valid_strings; i++)
6178 : * updateClosestMatch(&state, valid_strings[i]);
6179 : *
6180 : * closestMatch = getClosestMatch(&state);
6181 : */
6182 :
6183 : /*
6184 : * Initialize the given state with the source string and maximum Levenshtein
6185 : * distance to consider.
6186 : */
6187 : void
6188 56 : initClosestMatch(ClosestMatchState *state, const char *source, int max_d)
6189 : {
6190 : Assert(state);
6191 : Assert(max_d >= 0);
6192 :
6193 56 : state->source = source;
6194 56 : state->min_d = -1;
6195 56 : state->max_d = max_d;
6196 56 : state->match = NULL;
6197 56 : }
6198 :
6199 : /*
6200 : * If the candidate string is a closer match than the current one saved (or
6201 : * there is no match saved), save it as the closest match.
6202 : *
6203 : * If the source or candidate string is NULL, empty, or too long, this function
6204 : * takes no action. Likewise, if the Levenshtein distance exceeds the maximum
6205 : * allowed or more than half the characters are different, no action is taken.
6206 : */
6207 : void
6208 336 : updateClosestMatch(ClosestMatchState *state, const char *candidate)
6209 : {
6210 : int dist;
6211 :
6212 : Assert(state);
6213 :
6214 336 : if (state->source == NULL || state->source[0] == '\0' ||
6215 336 : candidate == NULL || candidate[0] == '\0')
6216 0 : return;
6217 :
6218 : /*
6219 : * To avoid ERROR-ing, we check the lengths here instead of setting
6220 : * 'trusted' to false in the call to varstr_levenshtein_less_equal().
6221 : */
6222 336 : if (strlen(state->source) > MAX_LEVENSHTEIN_STRLEN ||
6223 336 : strlen(candidate) > MAX_LEVENSHTEIN_STRLEN)
6224 0 : return;
6225 :
6226 336 : dist = varstr_levenshtein_less_equal(state->source, strlen(state->source),
6227 336 : candidate, strlen(candidate), 1, 1, 1,
6228 : state->max_d, true);
6229 336 : if (dist <= state->max_d &&
6230 56 : dist <= strlen(state->source) / 2 &&
6231 14 : (state->min_d == -1 || dist < state->min_d))
6232 : {
6233 14 : state->min_d = dist;
6234 14 : state->match = candidate;
6235 : }
6236 : }
6237 :
6238 : /*
6239 : * Return the closest match. If no suitable candidates were provided via
6240 : * updateClosestMatch(), return NULL.
6241 : */
6242 : const char *
6243 56 : getClosestMatch(ClosestMatchState *state)
6244 : {
6245 : Assert(state);
6246 :
6247 56 : return state->match;
6248 : }
6249 :
6250 :
6251 : /*
6252 : * Unicode support
6253 : */
6254 :
6255 : static UnicodeNormalizationForm
6256 186 : unicode_norm_form_from_string(const char *formstr)
6257 : {
6258 186 : UnicodeNormalizationForm form = -1;
6259 :
6260 : /*
6261 : * Might as well check this while we're here.
6262 : */
6263 186 : if (GetDatabaseEncoding() != PG_UTF8)
6264 0 : ereport(ERROR,
6265 : (errcode(ERRCODE_SYNTAX_ERROR),
6266 : errmsg("Unicode normalization can only be performed if server encoding is UTF8")));
6267 :
6268 186 : if (pg_strcasecmp(formstr, "NFC") == 0)
6269 66 : form = UNICODE_NFC;
6270 120 : else if (pg_strcasecmp(formstr, "NFD") == 0)
6271 36 : form = UNICODE_NFD;
6272 84 : else if (pg_strcasecmp(formstr, "NFKC") == 0)
6273 36 : form = UNICODE_NFKC;
6274 48 : else if (pg_strcasecmp(formstr, "NFKD") == 0)
6275 36 : form = UNICODE_NFKD;
6276 : else
6277 12 : ereport(ERROR,
6278 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
6279 : errmsg("invalid normalization form: %s", formstr)));
6280 :
6281 174 : return form;
6282 : }
6283 :
6284 : /*
6285 : * Returns version of Unicode used by Postgres in "major.minor" format (the
6286 : * same format as the Unicode version reported by ICU). The third component
6287 : * ("update version") never involves additions to the character repertiore and
6288 : * is unimportant for most purposes.
6289 : *
6290 : * See: https://unicode.org/versions/
6291 : */
6292 : Datum
6293 6 : unicode_version(PG_FUNCTION_ARGS)
6294 : {
6295 6 : PG_RETURN_TEXT_P(cstring_to_text(PG_UNICODE_VERSION));
6296 : }
6297 :
6298 : /*
6299 : * Returns version of Unicode used by ICU, if enabled; otherwise NULL.
6300 : */
6301 : Datum
6302 2 : icu_unicode_version(PG_FUNCTION_ARGS)
6303 : {
6304 : #ifdef USE_ICU
6305 2 : PG_RETURN_TEXT_P(cstring_to_text(U_UNICODE_VERSION));
6306 : #else
6307 : PG_RETURN_NULL();
6308 : #endif
6309 : }
6310 :
6311 : /*
6312 : * Check whether the string contains only assigned Unicode code
6313 : * points. Requires that the database encoding is UTF-8.
6314 : */
6315 : Datum
6316 12 : unicode_assigned(PG_FUNCTION_ARGS)
6317 : {
6318 12 : text *input = PG_GETARG_TEXT_PP(0);
6319 : unsigned char *p;
6320 : int size;
6321 :
6322 12 : if (GetDatabaseEncoding() != PG_UTF8)
6323 0 : ereport(ERROR,
6324 : (errmsg("Unicode categorization can only be performed if server encoding is UTF8")));
6325 :
6326 : /* convert to pg_wchar */
6327 12 : size = pg_mbstrlen_with_len(VARDATA_ANY(input), VARSIZE_ANY_EXHDR(input));
6328 12 : p = (unsigned char *) VARDATA_ANY(input);
6329 48 : for (int i = 0; i < size; i++)
6330 : {
6331 42 : pg_wchar uchar = utf8_to_unicode(p);
6332 42 : int category = unicode_category(uchar);
6333 :
6334 42 : if (category == PG_U_UNASSIGNED)
6335 6 : PG_RETURN_BOOL(false);
6336 :
6337 36 : p += pg_utf_mblen(p);
6338 : }
6339 :
6340 6 : PG_RETURN_BOOL(true);
6341 : }
6342 :
6343 : Datum
6344 48 : unicode_normalize_func(PG_FUNCTION_ARGS)
6345 : {
6346 48 : text *input = PG_GETARG_TEXT_PP(0);
6347 48 : char *formstr = text_to_cstring(PG_GETARG_TEXT_PP(1));
6348 : UnicodeNormalizationForm form;
6349 : int size;
6350 : pg_wchar *input_chars;
6351 : pg_wchar *output_chars;
6352 : unsigned char *p;
6353 : text *result;
6354 : int i;
6355 :
6356 48 : form = unicode_norm_form_from_string(formstr);
6357 :
6358 : /* convert to pg_wchar */
6359 42 : size = pg_mbstrlen_with_len(VARDATA_ANY(input), VARSIZE_ANY_EXHDR(input));
6360 42 : input_chars = palloc((size + 1) * sizeof(pg_wchar));
6361 42 : p = (unsigned char *) VARDATA_ANY(input);
6362 168 : for (i = 0; i < size; i++)
6363 : {
6364 126 : input_chars[i] = utf8_to_unicode(p);
6365 126 : p += pg_utf_mblen(p);
6366 : }
6367 42 : input_chars[i] = (pg_wchar) '\0';
6368 : Assert((char *) p == VARDATA_ANY(input) + VARSIZE_ANY_EXHDR(input));
6369 :
6370 : /* action */
6371 42 : output_chars = unicode_normalize(form, input_chars);
6372 :
6373 : /* convert back to UTF-8 string */
6374 42 : size = 0;
6375 162 : for (pg_wchar *wp = output_chars; *wp; wp++)
6376 : {
6377 : unsigned char buf[4];
6378 :
6379 120 : unicode_to_utf8(*wp, buf);
6380 120 : size += pg_utf_mblen(buf);
6381 : }
6382 :
6383 42 : result = palloc(size + VARHDRSZ);
6384 42 : SET_VARSIZE(result, size + VARHDRSZ);
6385 :
6386 42 : p = (unsigned char *) VARDATA_ANY(result);
6387 162 : for (pg_wchar *wp = output_chars; *wp; wp++)
6388 : {
6389 120 : unicode_to_utf8(*wp, p);
6390 120 : p += pg_utf_mblen(p);
6391 : }
6392 : Assert((char *) p == (char *) result + size + VARHDRSZ);
6393 :
6394 42 : PG_RETURN_TEXT_P(result);
6395 : }
6396 :
6397 : /*
6398 : * Check whether the string is in the specified Unicode normalization form.
6399 : *
6400 : * This is done by converting the string to the specified normal form and then
6401 : * comparing that to the original string. To speed that up, we also apply the
6402 : * "quick check" algorithm specified in UAX #15, which can give a yes or no
6403 : * answer for many strings by just scanning the string once.
6404 : *
6405 : * This function should generally be optimized for the case where the string
6406 : * is in fact normalized. In that case, we'll end up looking at the entire
6407 : * string, so it's probably not worth doing any incremental conversion etc.
6408 : */
6409 : Datum
6410 138 : unicode_is_normalized(PG_FUNCTION_ARGS)
6411 : {
6412 138 : text *input = PG_GETARG_TEXT_PP(0);
6413 138 : char *formstr = text_to_cstring(PG_GETARG_TEXT_PP(1));
6414 : UnicodeNormalizationForm form;
6415 : int size;
6416 : pg_wchar *input_chars;
6417 : pg_wchar *output_chars;
6418 : unsigned char *p;
6419 : int i;
6420 : UnicodeNormalizationQC quickcheck;
6421 : int output_size;
6422 : bool result;
6423 :
6424 138 : form = unicode_norm_form_from_string(formstr);
6425 :
6426 : /* convert to pg_wchar */
6427 132 : size = pg_mbstrlen_with_len(VARDATA_ANY(input), VARSIZE_ANY_EXHDR(input));
6428 132 : input_chars = palloc((size + 1) * sizeof(pg_wchar));
6429 132 : p = (unsigned char *) VARDATA_ANY(input);
6430 504 : for (i = 0; i < size; i++)
6431 : {
6432 372 : input_chars[i] = utf8_to_unicode(p);
6433 372 : p += pg_utf_mblen(p);
6434 : }
6435 132 : input_chars[i] = (pg_wchar) '\0';
6436 : Assert((char *) p == VARDATA_ANY(input) + VARSIZE_ANY_EXHDR(input));
6437 :
6438 : /* quick check (see UAX #15) */
6439 132 : quickcheck = unicode_is_normalized_quickcheck(form, input_chars);
6440 132 : if (quickcheck == UNICODE_NORM_QC_YES)
6441 42 : PG_RETURN_BOOL(true);
6442 90 : else if (quickcheck == UNICODE_NORM_QC_NO)
6443 12 : PG_RETURN_BOOL(false);
6444 :
6445 : /* normalize and compare with original */
6446 78 : output_chars = unicode_normalize(form, input_chars);
6447 :
6448 78 : output_size = 0;
6449 324 : for (pg_wchar *wp = output_chars; *wp; wp++)
6450 246 : output_size++;
6451 :
6452 114 : result = (size == output_size) &&
6453 36 : (memcmp(input_chars, output_chars, size * sizeof(pg_wchar)) == 0);
6454 :
6455 78 : PG_RETURN_BOOL(result);
6456 : }
6457 :
6458 : /*
6459 : * Check if first n chars are hexadecimal digits
6460 : */
6461 : static bool
6462 156 : isxdigits_n(const char *instr, size_t n)
6463 : {
6464 660 : for (size_t i = 0; i < n; i++)
6465 570 : if (!isxdigit((unsigned char) instr[i]))
6466 66 : return false;
6467 :
6468 90 : return true;
6469 : }
6470 :
6471 : static unsigned int
6472 504 : hexval(unsigned char c)
6473 : {
6474 504 : if (c >= '0' && c <= '9')
6475 384 : return c - '0';
6476 120 : if (c >= 'a' && c <= 'f')
6477 60 : return c - 'a' + 0xA;
6478 60 : if (c >= 'A' && c <= 'F')
6479 60 : return c - 'A' + 0xA;
6480 0 : elog(ERROR, "invalid hexadecimal digit");
6481 : return 0; /* not reached */
6482 : }
6483 :
6484 : /*
6485 : * Translate string with hexadecimal digits to number
6486 : */
6487 : static unsigned int
6488 90 : hexval_n(const char *instr, size_t n)
6489 : {
6490 90 : unsigned int result = 0;
6491 :
6492 594 : for (size_t i = 0; i < n; i++)
6493 504 : result += hexval(instr[i]) << (4 * (n - i - 1));
6494 :
6495 90 : return result;
6496 : }
6497 :
6498 : /*
6499 : * Replaces Unicode escape sequences by Unicode characters
6500 : */
6501 : Datum
6502 66 : unistr(PG_FUNCTION_ARGS)
6503 : {
6504 66 : text *input_text = PG_GETARG_TEXT_PP(0);
6505 : char *instr;
6506 : int len;
6507 : StringInfoData str;
6508 : text *result;
6509 66 : pg_wchar pair_first = 0;
6510 : char cbuf[MAX_UNICODE_EQUIVALENT_STRING + 1];
6511 :
6512 66 : instr = VARDATA_ANY(input_text);
6513 66 : len = VARSIZE_ANY_EXHDR(input_text);
6514 :
6515 66 : initStringInfo(&str);
6516 :
6517 510 : while (len > 0)
6518 : {
6519 486 : if (instr[0] == '\\')
6520 : {
6521 102 : if (len >= 2 &&
6522 102 : instr[1] == '\\')
6523 : {
6524 6 : if (pair_first)
6525 0 : goto invalid_pair;
6526 6 : appendStringInfoChar(&str, '\\');
6527 6 : instr += 2;
6528 6 : len -= 2;
6529 : }
6530 96 : else if ((len >= 5 && isxdigits_n(instr + 1, 4)) ||
6531 66 : (len >= 6 && instr[1] == 'u' && isxdigits_n(instr + 2, 4)))
6532 30 : {
6533 : pg_wchar unicode;
6534 42 : int offset = instr[1] == 'u' ? 2 : 1;
6535 :
6536 42 : unicode = hexval_n(instr + offset, 4);
6537 :
6538 42 : if (!is_valid_unicode_codepoint(unicode))
6539 0 : ereport(ERROR,
6540 : errcode(ERRCODE_INVALID_PARAMETER_VALUE),
6541 : errmsg("invalid Unicode code point: %04X", unicode));
6542 :
6543 42 : if (pair_first)
6544 : {
6545 12 : if (is_utf16_surrogate_second(unicode))
6546 : {
6547 0 : unicode = surrogate_pair_to_codepoint(pair_first, unicode);
6548 0 : pair_first = 0;
6549 : }
6550 : else
6551 12 : goto invalid_pair;
6552 : }
6553 30 : else if (is_utf16_surrogate_second(unicode))
6554 0 : goto invalid_pair;
6555 :
6556 30 : if (is_utf16_surrogate_first(unicode))
6557 18 : pair_first = unicode;
6558 : else
6559 : {
6560 12 : pg_unicode_to_server(unicode, (unsigned char *) cbuf);
6561 12 : appendStringInfoString(&str, cbuf);
6562 : }
6563 :
6564 30 : instr += 4 + offset;
6565 30 : len -= 4 + offset;
6566 : }
6567 54 : else if (len >= 8 && instr[1] == '+' && isxdigits_n(instr + 2, 6))
6568 12 : {
6569 : pg_wchar unicode;
6570 :
6571 24 : unicode = hexval_n(instr + 2, 6);
6572 :
6573 24 : if (!is_valid_unicode_codepoint(unicode))
6574 6 : ereport(ERROR,
6575 : errcode(ERRCODE_INVALID_PARAMETER_VALUE),
6576 : errmsg("invalid Unicode code point: %04X", unicode));
6577 :
6578 18 : if (pair_first)
6579 : {
6580 6 : if (is_utf16_surrogate_second(unicode))
6581 : {
6582 0 : unicode = surrogate_pair_to_codepoint(pair_first, unicode);
6583 0 : pair_first = 0;
6584 : }
6585 : else
6586 6 : goto invalid_pair;
6587 : }
6588 12 : else if (is_utf16_surrogate_second(unicode))
6589 0 : goto invalid_pair;
6590 :
6591 12 : if (is_utf16_surrogate_first(unicode))
6592 6 : pair_first = unicode;
6593 : else
6594 : {
6595 6 : pg_unicode_to_server(unicode, (unsigned char *) cbuf);
6596 6 : appendStringInfoString(&str, cbuf);
6597 : }
6598 :
6599 12 : instr += 8;
6600 12 : len -= 8;
6601 : }
6602 30 : else if (len >= 10 && instr[1] == 'U' && isxdigits_n(instr + 2, 8))
6603 12 : {
6604 : pg_wchar unicode;
6605 :
6606 24 : unicode = hexval_n(instr + 2, 8);
6607 :
6608 24 : if (!is_valid_unicode_codepoint(unicode))
6609 6 : ereport(ERROR,
6610 : errcode(ERRCODE_INVALID_PARAMETER_VALUE),
6611 : errmsg("invalid Unicode code point: %04X", unicode));
6612 :
6613 18 : if (pair_first)
6614 : {
6615 6 : if (is_utf16_surrogate_second(unicode))
6616 : {
6617 0 : unicode = surrogate_pair_to_codepoint(pair_first, unicode);
6618 0 : pair_first = 0;
6619 : }
6620 : else
6621 6 : goto invalid_pair;
6622 : }
6623 12 : else if (is_utf16_surrogate_second(unicode))
6624 0 : goto invalid_pair;
6625 :
6626 12 : if (is_utf16_surrogate_first(unicode))
6627 6 : pair_first = unicode;
6628 : else
6629 : {
6630 6 : pg_unicode_to_server(unicode, (unsigned char *) cbuf);
6631 6 : appendStringInfoString(&str, cbuf);
6632 : }
6633 :
6634 12 : instr += 10;
6635 12 : len -= 10;
6636 : }
6637 : else
6638 6 : ereport(ERROR,
6639 : (errcode(ERRCODE_SYNTAX_ERROR),
6640 : errmsg("invalid Unicode escape"),
6641 : errhint("Unicode escapes must be \\XXXX, \\+XXXXXX, \\uXXXX, or \\UXXXXXXXX.")));
6642 : }
6643 : else
6644 : {
6645 384 : if (pair_first)
6646 0 : goto invalid_pair;
6647 :
6648 384 : appendStringInfoChar(&str, *instr++);
6649 384 : len--;
6650 : }
6651 : }
6652 :
6653 : /* unfinished surrogate pair? */
6654 24 : if (pair_first)
6655 6 : goto invalid_pair;
6656 :
6657 18 : result = cstring_to_text_with_len(str.data, str.len);
6658 18 : pfree(str.data);
6659 :
6660 18 : PG_RETURN_TEXT_P(result);
6661 :
6662 30 : invalid_pair:
6663 30 : ereport(ERROR,
6664 : (errcode(ERRCODE_SYNTAX_ERROR),
6665 : errmsg("invalid Unicode surrogate pair")));
6666 : PG_RETURN_NULL(); /* keep compiler quiet */
6667 : }
|