Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * varlena.c
4 : * Functions for the variable-length built-in types.
5 : *
6 : * Portions Copyright (c) 1996-2019, 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/tuptoaster.h"
21 : #include "catalog/pg_collation.h"
22 : #include "catalog/pg_type.h"
23 : #include "common/int.h"
24 : #include "lib/hyperloglog.h"
25 : #include "libpq/pqformat.h"
26 : #include "miscadmin.h"
27 : #include "parser/scansup.h"
28 : #include "port/pg_bswap.h"
29 : #include "regex/regex.h"
30 : #include "utils/builtins.h"
31 : #include "utils/bytea.h"
32 : #include "utils/hashutils.h"
33 : #include "utils/lsyscache.h"
34 : #include "utils/memutils.h"
35 : #include "utils/pg_locale.h"
36 : #include "utils/sortsupport.h"
37 : #include "utils/varlena.h"
38 :
39 :
40 : /* GUC variable */
41 : int bytea_output = BYTEA_OUTPUT_HEX;
42 :
43 : typedef struct varlena unknown;
44 : typedef struct varlena VarString;
45 :
46 : /*
47 : * State for text_position_* functions.
48 : */
49 : typedef struct
50 : {
51 : bool is_multibyte; /* T if multibyte encoding */
52 : bool is_multibyte_char_in_char;
53 :
54 : char *str1; /* haystack string */
55 : char *str2; /* needle string */
56 : int len1; /* string lengths in bytes */
57 : int len2;
58 :
59 : /* Skip table for Boyer-Moore-Horspool search algorithm: */
60 : int skiptablemask; /* mask for ANDing with skiptable subscripts */
61 : int skiptable[256]; /* skip distance for given mismatched char */
62 :
63 : char *last_match; /* pointer to last match in 'str1' */
64 :
65 : /*
66 : * Sometimes we need to convert the byte position of a match to a
67 : * character position. These store the last position that was converted,
68 : * so that on the next call, we can continue from that point, rather than
69 : * count characters from the very beginning.
70 : */
71 : char *refpoint; /* pointer within original haystack string */
72 : int refpos; /* 0-based character offset of the same point */
73 : } TextPositionState;
74 :
75 : typedef struct
76 : {
77 : char *buf1; /* 1st string, or abbreviation original string
78 : * buf */
79 : char *buf2; /* 2nd string, or abbreviation strxfrm() buf */
80 : int buflen1;
81 : int buflen2;
82 : int last_len1; /* Length of last buf1 string/strxfrm() input */
83 : int last_len2; /* Length of last buf2 string/strxfrm() blob */
84 : int last_returned; /* Last comparison result (cache) */
85 : bool cache_blob; /* Does buf2 contain strxfrm() blob, etc? */
86 : bool collate_c;
87 : Oid typid; /* Actual datatype (text/bpchar/bytea/name) */
88 : hyperLogLogState abbr_card; /* Abbreviated key cardinality state */
89 : hyperLogLogState full_card; /* Full key cardinality state */
90 : double prop_card; /* Required cardinality proportion */
91 : pg_locale_t locale;
92 : } VarStringSortSupport;
93 :
94 : /*
95 : * This should be large enough that most strings will fit, but small enough
96 : * that we feel comfortable putting it on the stack
97 : */
98 : #define TEXTBUFLEN 1024
99 :
100 : #define DatumGetUnknownP(X) ((unknown *) PG_DETOAST_DATUM(X))
101 : #define DatumGetUnknownPCopy(X) ((unknown *) PG_DETOAST_DATUM_COPY(X))
102 : #define PG_GETARG_UNKNOWN_P(n) DatumGetUnknownP(PG_GETARG_DATUM(n))
103 : #define PG_GETARG_UNKNOWN_P_COPY(n) DatumGetUnknownPCopy(PG_GETARG_DATUM(n))
104 : #define PG_RETURN_UNKNOWN_P(x) PG_RETURN_POINTER(x)
105 :
106 : #define DatumGetVarStringP(X) ((VarString *) PG_DETOAST_DATUM(X))
107 : #define DatumGetVarStringPP(X) ((VarString *) PG_DETOAST_DATUM_PACKED(X))
108 :
109 : static int varstrfastcmp_c(Datum x, Datum y, SortSupport ssup);
110 : static int bpcharfastcmp_c(Datum x, Datum y, SortSupport ssup);
111 : static int namefastcmp_c(Datum x, Datum y, SortSupport ssup);
112 : static int varlenafastcmp_locale(Datum x, Datum y, SortSupport ssup);
113 : static int namefastcmp_locale(Datum x, Datum y, SortSupport ssup);
114 : static int varstrfastcmp_locale(char *a1p, int len1, char *a2p, int len2, SortSupport ssup);
115 : static int varstrcmp_abbrev(Datum x, Datum y, SortSupport ssup);
116 : static Datum varstr_abbrev_convert(Datum original, SortSupport ssup);
117 : static bool varstr_abbrev_abort(int memtupcount, SortSupport ssup);
118 : static int32 text_length(Datum str);
119 : static text *text_catenate(text *t1, text *t2);
120 : static text *text_substring(Datum str,
121 : int32 start,
122 : int32 length,
123 : bool length_not_specified);
124 : static text *text_overlay(text *t1, text *t2, int sp, int sl);
125 : static int text_position(text *t1, text *t2, Oid collid);
126 : static void text_position_setup(text *t1, text *t2, Oid collid, TextPositionState *state);
127 : static bool text_position_next(TextPositionState *state);
128 : static char *text_position_next_internal(char *start_ptr, TextPositionState *state);
129 : static char *text_position_get_match_ptr(TextPositionState *state);
130 : static int text_position_get_match_pos(TextPositionState *state);
131 : static void text_position_cleanup(TextPositionState *state);
132 : static void check_collation_set(Oid collid);
133 : static int text_cmp(text *arg1, text *arg2, Oid collid);
134 : static bytea *bytea_catenate(bytea *t1, bytea *t2);
135 : static bytea *bytea_substring(Datum str,
136 : int S,
137 : int L,
138 : bool length_not_specified);
139 : static bytea *bytea_overlay(bytea *t1, bytea *t2, int sp, int sl);
140 : static void appendStringInfoText(StringInfo str, const text *t);
141 : static Datum text_to_array_internal(PG_FUNCTION_ARGS);
142 : static text *array_to_text_internal(FunctionCallInfo fcinfo, ArrayType *v,
143 : const char *fldsep, const char *null_string);
144 : static StringInfo makeStringAggState(FunctionCallInfo fcinfo);
145 : static bool text_format_parse_digits(const char **ptr, const char *end_ptr,
146 : int *value);
147 : static const char *text_format_parse_format(const char *start_ptr,
148 : const char *end_ptr,
149 : int *argpos, int *widthpos,
150 : int *flags, int *width);
151 : static void text_format_string_conversion(StringInfo buf, char conversion,
152 : FmgrInfo *typOutputInfo,
153 : Datum value, bool isNull,
154 : int flags, int width);
155 : static void text_format_append_string(StringInfo buf, const char *str,
156 : int flags, int width);
157 :
158 :
159 : /*****************************************************************************
160 : * CONVERSION ROUTINES EXPORTED FOR USE BY C CODE *
161 : *****************************************************************************/
162 :
163 : /*
164 : * cstring_to_text
165 : *
166 : * Create a text value from a null-terminated C string.
167 : *
168 : * The new text value is freshly palloc'd with a full-size VARHDR.
169 : */
170 : text *
171 4832028 : cstring_to_text(const char *s)
172 : {
173 4832028 : return cstring_to_text_with_len(s, strlen(s));
174 : }
175 :
176 : /*
177 : * cstring_to_text_with_len
178 : *
179 : * Same as cstring_to_text except the caller specifies the string length;
180 : * the string need not be null_terminated.
181 : */
182 : text *
183 8007736 : cstring_to_text_with_len(const char *s, int len)
184 : {
185 8007736 : text *result = (text *) palloc(len + VARHDRSZ);
186 :
187 8007736 : SET_VARSIZE(result, len + VARHDRSZ);
188 8007736 : memcpy(VARDATA(result), s, len);
189 :
190 8007736 : return result;
191 : }
192 :
193 : /*
194 : * text_to_cstring
195 : *
196 : * Create a palloc'd, null-terminated C string from a text value.
197 : *
198 : * We support being passed a compressed or toasted text value.
199 : * This is a bit bogus since such values shouldn't really be referred to as
200 : * "text *", but it seems useful for robustness. If we didn't handle that
201 : * case here, we'd need another routine that did, anyway.
202 : */
203 : char *
204 4100884 : text_to_cstring(const text *t)
205 : {
206 : /* must cast away the const, unfortunately */
207 4100884 : text *tunpacked = pg_detoast_datum_packed(unconstify(text *, t));
208 4100884 : int len = VARSIZE_ANY_EXHDR(tunpacked);
209 : char *result;
210 :
211 4100884 : result = (char *) palloc(len + 1);
212 4100884 : memcpy(result, VARDATA_ANY(tunpacked), len);
213 4100884 : result[len] = '\0';
214 :
215 4100884 : if (tunpacked != t)
216 57160 : pfree(tunpacked);
217 :
218 4100884 : return result;
219 : }
220 :
221 : /*
222 : * text_to_cstring_buffer
223 : *
224 : * Copy a text value into a caller-supplied buffer of size dst_len.
225 : *
226 : * The text string is truncated if necessary to fit. The result is
227 : * guaranteed null-terminated (unless dst_len == 0).
228 : *
229 : * We support being passed a compressed or toasted text value.
230 : * This is a bit bogus since such values shouldn't really be referred to as
231 : * "text *", but it seems useful for robustness. If we didn't handle that
232 : * case here, we'd need another routine that did, anyway.
233 : */
234 : void
235 390 : text_to_cstring_buffer(const text *src, char *dst, size_t dst_len)
236 : {
237 : /* must cast away the const, unfortunately */
238 390 : text *srcunpacked = pg_detoast_datum_packed(unconstify(text *, src));
239 390 : size_t src_len = VARSIZE_ANY_EXHDR(srcunpacked);
240 :
241 390 : if (dst_len > 0)
242 : {
243 390 : dst_len--;
244 390 : if (dst_len >= src_len)
245 390 : dst_len = src_len;
246 : else /* ensure truncation is encoding-safe */
247 0 : dst_len = pg_mbcliplen(VARDATA_ANY(srcunpacked), src_len, dst_len);
248 390 : memcpy(dst, VARDATA_ANY(srcunpacked), dst_len);
249 390 : dst[dst_len] = '\0';
250 : }
251 :
252 390 : if (srcunpacked != src)
253 0 : pfree(srcunpacked);
254 390 : }
255 :
256 :
257 : /*****************************************************************************
258 : * USER I/O ROUTINES *
259 : *****************************************************************************/
260 :
261 :
262 : #define VAL(CH) ((CH) - '0')
263 : #define DIG(VAL) ((VAL) + '0')
264 :
265 : /*
266 : * byteain - converts from printable representation of byte array
267 : *
268 : * Non-printable characters must be passed as '\nnn' (octal) and are
269 : * converted to internal form. '\' must be passed as '\\'.
270 : * ereport(ERROR, ...) if bad form.
271 : *
272 : * BUGS:
273 : * The input is scanned twice.
274 : * The error checking of input is minimal.
275 : */
276 : Datum
277 9994 : byteain(PG_FUNCTION_ARGS)
278 : {
279 9994 : char *inputText = PG_GETARG_CSTRING(0);
280 : char *tp;
281 : char *rp;
282 : int bc;
283 : bytea *result;
284 :
285 : /* Recognize hex input */
286 9994 : if (inputText[0] == '\\' && inputText[1] == 'x')
287 : {
288 44 : size_t len = strlen(inputText);
289 :
290 44 : bc = (len - 2) / 2 + VARHDRSZ; /* maximum possible length */
291 44 : result = palloc(bc);
292 44 : bc = hex_decode(inputText + 2, len - 2, VARDATA(result));
293 36 : SET_VARSIZE(result, bc + VARHDRSZ); /* actual length */
294 :
295 36 : PG_RETURN_BYTEA_P(result);
296 : }
297 :
298 : /* Else, it's the traditional escaped style */
299 198950 : for (bc = 0, tp = inputText; *tp != '\0'; bc++)
300 : {
301 189004 : if (tp[0] != '\\')
302 188362 : tp++;
303 1284 : else if ((tp[0] == '\\') &&
304 1922 : (tp[1] >= '0' && tp[1] <= '3') &&
305 1914 : (tp[2] >= '0' && tp[2] <= '7') &&
306 1276 : (tp[3] >= '0' && tp[3] <= '7'))
307 638 : tp += 4;
308 8 : else if ((tp[0] == '\\') &&
309 4 : (tp[1] == '\\'))
310 0 : tp += 2;
311 : else
312 : {
313 : /*
314 : * one backslash, not followed by another or ### valid octal
315 : */
316 4 : ereport(ERROR,
317 : (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
318 : errmsg("invalid input syntax for type %s", "bytea")));
319 : }
320 : }
321 :
322 9946 : bc += VARHDRSZ;
323 :
324 9946 : result = (bytea *) palloc(bc);
325 9946 : SET_VARSIZE(result, bc);
326 :
327 9946 : tp = inputText;
328 9946 : rp = VARDATA(result);
329 208884 : while (*tp != '\0')
330 : {
331 188992 : if (tp[0] != '\\')
332 188354 : *rp++ = *tp++;
333 1276 : else if ((tp[0] == '\\') &&
334 1914 : (tp[1] >= '0' && tp[1] <= '3') &&
335 1914 : (tp[2] >= '0' && tp[2] <= '7') &&
336 1276 : (tp[3] >= '0' && tp[3] <= '7'))
337 : {
338 638 : bc = VAL(tp[1]);
339 638 : bc <<= 3;
340 638 : bc += VAL(tp[2]);
341 638 : bc <<= 3;
342 638 : *rp++ = bc + VAL(tp[3]);
343 :
344 638 : tp += 4;
345 : }
346 0 : else if ((tp[0] == '\\') &&
347 0 : (tp[1] == '\\'))
348 : {
349 0 : *rp++ = '\\';
350 0 : tp += 2;
351 : }
352 : else
353 : {
354 : /*
355 : * We should never get here. The first pass should not allow it.
356 : */
357 0 : ereport(ERROR,
358 : (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
359 : errmsg("invalid input syntax for type %s", "bytea")));
360 : }
361 : }
362 :
363 9946 : PG_RETURN_BYTEA_P(result);
364 : }
365 :
366 : /*
367 : * byteaout - converts to printable representation of byte array
368 : *
369 : * In the traditional escaped format, non-printable characters are
370 : * printed as '\nnn' (octal) and '\' as '\\'.
371 : */
372 : Datum
373 4864 : byteaout(PG_FUNCTION_ARGS)
374 : {
375 4864 : bytea *vlena = PG_GETARG_BYTEA_PP(0);
376 : char *result;
377 : char *rp;
378 :
379 4864 : if (bytea_output == BYTEA_OUTPUT_HEX)
380 : {
381 : /* Print hex format */
382 4658 : rp = result = palloc(VARSIZE_ANY_EXHDR(vlena) * 2 + 2 + 1);
383 4658 : *rp++ = '\\';
384 4658 : *rp++ = 'x';
385 4658 : rp += hex_encode(VARDATA_ANY(vlena), VARSIZE_ANY_EXHDR(vlena), rp);
386 : }
387 206 : else if (bytea_output == BYTEA_OUTPUT_ESCAPE)
388 : {
389 : /* Print traditional escaped format */
390 : char *vp;
391 : int len;
392 : int i;
393 :
394 206 : len = 1; /* empty string has 1 char */
395 206 : vp = VARDATA_ANY(vlena);
396 2060 : for (i = VARSIZE_ANY_EXHDR(vlena); i != 0; i--, vp++)
397 : {
398 1854 : if (*vp == '\\')
399 0 : len += 2;
400 1854 : else if ((unsigned char) *vp < 0x20 || (unsigned char) *vp > 0x7e)
401 328 : len += 4;
402 : else
403 1526 : len++;
404 : }
405 206 : rp = result = (char *) palloc(len);
406 206 : vp = VARDATA_ANY(vlena);
407 2060 : for (i = VARSIZE_ANY_EXHDR(vlena); i != 0; i--, vp++)
408 : {
409 1854 : if (*vp == '\\')
410 : {
411 0 : *rp++ = '\\';
412 0 : *rp++ = '\\';
413 : }
414 1854 : else if ((unsigned char) *vp < 0x20 || (unsigned char) *vp > 0x7e)
415 328 : {
416 : int val; /* holds unprintable chars */
417 :
418 328 : val = *vp;
419 328 : rp[0] = '\\';
420 328 : rp[3] = DIG(val & 07);
421 328 : val >>= 3;
422 328 : rp[2] = DIG(val & 07);
423 328 : val >>= 3;
424 328 : rp[1] = DIG(val & 03);
425 328 : rp += 4;
426 : }
427 : else
428 1526 : *rp++ = *vp;
429 : }
430 : }
431 : else
432 : {
433 0 : elog(ERROR, "unrecognized bytea_output setting: %d",
434 : bytea_output);
435 : rp = result = NULL; /* keep compiler quiet */
436 : }
437 4864 : *rp = '\0';
438 4864 : PG_RETURN_CSTRING(result);
439 : }
440 :
441 : /*
442 : * bytearecv - converts external binary format to bytea
443 : */
444 : Datum
445 696 : bytearecv(PG_FUNCTION_ARGS)
446 : {
447 696 : StringInfo buf = (StringInfo) PG_GETARG_POINTER(0);
448 : bytea *result;
449 : int nbytes;
450 :
451 696 : nbytes = buf->len - buf->cursor;
452 696 : result = (bytea *) palloc(nbytes + VARHDRSZ);
453 696 : SET_VARSIZE(result, nbytes + VARHDRSZ);
454 696 : pq_copymsgbytes(buf, VARDATA(result), nbytes);
455 696 : PG_RETURN_BYTEA_P(result);
456 : }
457 :
458 : /*
459 : * byteasend - converts bytea to binary format
460 : *
461 : * This is a special case: just copy the input...
462 : */
463 : Datum
464 3760 : byteasend(PG_FUNCTION_ARGS)
465 : {
466 3760 : bytea *vlena = PG_GETARG_BYTEA_P_COPY(0);
467 :
468 3760 : PG_RETURN_BYTEA_P(vlena);
469 : }
470 :
471 : Datum
472 28 : bytea_string_agg_transfn(PG_FUNCTION_ARGS)
473 : {
474 : StringInfo state;
475 :
476 28 : state = PG_ARGISNULL(0) ? NULL : (StringInfo) PG_GETARG_POINTER(0);
477 :
478 : /* Append the value unless null. */
479 28 : if (!PG_ARGISNULL(1))
480 : {
481 28 : bytea *value = PG_GETARG_BYTEA_PP(1);
482 :
483 : /* On the first time through, we ignore the delimiter. */
484 28 : if (state == NULL)
485 16 : state = makeStringAggState(fcinfo);
486 12 : else if (!PG_ARGISNULL(2))
487 : {
488 8 : bytea *delim = PG_GETARG_BYTEA_PP(2);
489 :
490 8 : appendBinaryStringInfo(state, VARDATA_ANY(delim), VARSIZE_ANY_EXHDR(delim));
491 : }
492 :
493 28 : appendBinaryStringInfo(state, VARDATA_ANY(value), VARSIZE_ANY_EXHDR(value));
494 : }
495 :
496 : /*
497 : * The transition type for string_agg() is declared to be "internal",
498 : * which is a pass-by-value type the same size as a pointer.
499 : */
500 28 : PG_RETURN_POINTER(state);
501 : }
502 :
503 : Datum
504 20 : bytea_string_agg_finalfn(PG_FUNCTION_ARGS)
505 : {
506 : StringInfo state;
507 :
508 : /* cannot be called directly because of internal-type argument */
509 : Assert(AggCheckCallContext(fcinfo, NULL));
510 :
511 20 : state = PG_ARGISNULL(0) ? NULL : (StringInfo) PG_GETARG_POINTER(0);
512 :
513 20 : if (state != NULL)
514 : {
515 : bytea *result;
516 :
517 16 : result = (bytea *) palloc(state->len + VARHDRSZ);
518 16 : SET_VARSIZE(result, state->len + VARHDRSZ);
519 16 : memcpy(VARDATA(result), state->data, state->len);
520 16 : PG_RETURN_BYTEA_P(result);
521 : }
522 : else
523 4 : PG_RETURN_NULL();
524 : }
525 :
526 : /*
527 : * textin - converts "..." to internal representation
528 : */
529 : Datum
530 3871780 : textin(PG_FUNCTION_ARGS)
531 : {
532 3871780 : char *inputText = PG_GETARG_CSTRING(0);
533 :
534 3871780 : PG_RETURN_TEXT_P(cstring_to_text(inputText));
535 : }
536 :
537 : /*
538 : * textout - converts internal representation to "..."
539 : */
540 : Datum
541 1725570 : textout(PG_FUNCTION_ARGS)
542 : {
543 1725570 : Datum txt = PG_GETARG_DATUM(0);
544 :
545 1725570 : PG_RETURN_CSTRING(TextDatumGetCString(txt));
546 : }
547 :
548 : /*
549 : * textrecv - converts external binary format to text
550 : */
551 : Datum
552 12 : textrecv(PG_FUNCTION_ARGS)
553 : {
554 12 : StringInfo buf = (StringInfo) PG_GETARG_POINTER(0);
555 : text *result;
556 : char *str;
557 : int nbytes;
558 :
559 12 : str = pq_getmsgtext(buf, buf->len - buf->cursor, &nbytes);
560 :
561 12 : result = cstring_to_text_with_len(str, nbytes);
562 12 : pfree(str);
563 12 : PG_RETURN_TEXT_P(result);
564 : }
565 :
566 : /*
567 : * textsend - converts text to binary format
568 : */
569 : Datum
570 3324 : textsend(PG_FUNCTION_ARGS)
571 : {
572 3324 : text *t = PG_GETARG_TEXT_PP(0);
573 : StringInfoData buf;
574 :
575 3324 : pq_begintypsend(&buf);
576 3324 : pq_sendtext(&buf, VARDATA_ANY(t), VARSIZE_ANY_EXHDR(t));
577 3324 : PG_RETURN_BYTEA_P(pq_endtypsend(&buf));
578 : }
579 :
580 :
581 : /*
582 : * unknownin - converts "..." to internal representation
583 : */
584 : Datum
585 0 : unknownin(PG_FUNCTION_ARGS)
586 : {
587 0 : char *str = PG_GETARG_CSTRING(0);
588 :
589 : /* representation is same as cstring */
590 0 : PG_RETURN_CSTRING(pstrdup(str));
591 : }
592 :
593 : /*
594 : * unknownout - converts internal representation to "..."
595 : */
596 : Datum
597 318 : unknownout(PG_FUNCTION_ARGS)
598 : {
599 : /* representation is same as cstring */
600 318 : char *str = PG_GETARG_CSTRING(0);
601 :
602 318 : PG_RETURN_CSTRING(pstrdup(str));
603 : }
604 :
605 : /*
606 : * unknownrecv - converts external binary format to unknown
607 : */
608 : Datum
609 0 : unknownrecv(PG_FUNCTION_ARGS)
610 : {
611 0 : StringInfo buf = (StringInfo) PG_GETARG_POINTER(0);
612 : char *str;
613 : int nbytes;
614 :
615 0 : str = pq_getmsgtext(buf, buf->len - buf->cursor, &nbytes);
616 : /* representation is same as cstring */
617 0 : PG_RETURN_CSTRING(str);
618 : }
619 :
620 : /*
621 : * unknownsend - converts unknown to binary format
622 : */
623 : Datum
624 0 : unknownsend(PG_FUNCTION_ARGS)
625 : {
626 : /* representation is same as cstring */
627 0 : char *str = PG_GETARG_CSTRING(0);
628 : StringInfoData buf;
629 :
630 0 : pq_begintypsend(&buf);
631 0 : pq_sendtext(&buf, str, strlen(str));
632 0 : PG_RETURN_BYTEA_P(pq_endtypsend(&buf));
633 : }
634 :
635 :
636 : /* ========== PUBLIC ROUTINES ========== */
637 :
638 : /*
639 : * textlen -
640 : * returns the logical length of a text*
641 : * (which is less than the VARSIZE of the text*)
642 : */
643 : Datum
644 202544 : textlen(PG_FUNCTION_ARGS)
645 : {
646 202544 : Datum str = PG_GETARG_DATUM(0);
647 :
648 : /* try to avoid decompressing argument */
649 202544 : PG_RETURN_INT32(text_length(str));
650 : }
651 :
652 : /*
653 : * text_length -
654 : * Does the real work for textlen()
655 : *
656 : * This is broken out so it can be called directly by other string processing
657 : * functions. Note that the argument is passed as a Datum, to indicate that
658 : * it may still be in compressed form. We can avoid decompressing it at all
659 : * in some cases.
660 : */
661 : static int32
662 202552 : text_length(Datum str)
663 : {
664 : /* fastpath when max encoding length is one */
665 202552 : if (pg_database_encoding_max_length() == 1)
666 24 : PG_RETURN_INT32(toast_raw_datum_size(str) - VARHDRSZ);
667 : else
668 : {
669 202528 : text *t = DatumGetTextPP(str);
670 :
671 202528 : PG_RETURN_INT32(pg_mbstrlen_with_len(VARDATA_ANY(t),
672 : VARSIZE_ANY_EXHDR(t)));
673 : }
674 : }
675 :
676 : /*
677 : * textoctetlen -
678 : * returns the physical length of a text*
679 : * (which is less than the VARSIZE of the text*)
680 : */
681 : Datum
682 46 : textoctetlen(PG_FUNCTION_ARGS)
683 : {
684 46 : Datum str = PG_GETARG_DATUM(0);
685 :
686 : /* We need not detoast the input at all */
687 46 : PG_RETURN_INT32(toast_raw_datum_size(str) - VARHDRSZ);
688 : }
689 :
690 : /*
691 : * textcat -
692 : * takes two text* and returns a text* that is the concatenation of
693 : * the two.
694 : *
695 : * Rewritten by Sapa, sapa@hq.icb.chel.su. 8-Jul-96.
696 : * Updated by Thomas, Thomas.Lockhart@jpl.nasa.gov 1997-07-10.
697 : * Allocate space for output in all cases.
698 : * XXX - thomas 1997-07-10
699 : */
700 : Datum
701 1011968 : textcat(PG_FUNCTION_ARGS)
702 : {
703 1011968 : text *t1 = PG_GETARG_TEXT_PP(0);
704 1011968 : text *t2 = PG_GETARG_TEXT_PP(1);
705 :
706 1011968 : PG_RETURN_TEXT_P(text_catenate(t1, t2));
707 : }
708 :
709 : /*
710 : * text_catenate
711 : * Guts of textcat(), broken out so it can be used by other functions
712 : *
713 : * Arguments can be in short-header form, but not compressed or out-of-line
714 : */
715 : static text *
716 1012032 : text_catenate(text *t1, text *t2)
717 : {
718 : text *result;
719 : int len1,
720 : len2,
721 : len;
722 : char *ptr;
723 :
724 1012032 : len1 = VARSIZE_ANY_EXHDR(t1);
725 1012032 : len2 = VARSIZE_ANY_EXHDR(t2);
726 :
727 : /* paranoia ... probably should throw error instead? */
728 1012032 : if (len1 < 0)
729 0 : len1 = 0;
730 1012032 : if (len2 < 0)
731 0 : len2 = 0;
732 :
733 1012032 : len = len1 + len2 + VARHDRSZ;
734 1012032 : result = (text *) palloc(len);
735 :
736 : /* Set size of result string... */
737 1012032 : SET_VARSIZE(result, len);
738 :
739 : /* Fill data field of result string... */
740 1012032 : ptr = VARDATA(result);
741 1012032 : if (len1 > 0)
742 1011520 : memcpy(ptr, VARDATA_ANY(t1), len1);
743 1012032 : if (len2 > 0)
744 1011928 : memcpy(ptr + len1, VARDATA_ANY(t2), len2);
745 :
746 1012032 : return result;
747 : }
748 :
749 : /*
750 : * charlen_to_bytelen()
751 : * Compute the number of bytes occupied by n characters starting at *p
752 : *
753 : * It is caller's responsibility that there actually are n characters;
754 : * the string need not be null-terminated.
755 : */
756 : static int
757 1162 : charlen_to_bytelen(const char *p, int n)
758 : {
759 1162 : if (pg_database_encoding_max_length() == 1)
760 : {
761 : /* Optimization for single-byte encodings */
762 0 : return n;
763 : }
764 : else
765 : {
766 : const char *s;
767 :
768 5669088 : for (s = p; n > 0; n--)
769 5667926 : s += pg_mblen(s);
770 :
771 1162 : return s - p;
772 : }
773 : }
774 :
775 : /*
776 : * text_substr()
777 : * Return a substring starting at the specified position.
778 : * - thomas 1997-12-31
779 : *
780 : * Input:
781 : * - string
782 : * - starting position (is one-based)
783 : * - string length
784 : *
785 : * If the starting position is zero or less, then return from the start of the string
786 : * adjusting the length to be consistent with the "negative start" per SQL.
787 : * If the length is less than zero, return the remaining string.
788 : *
789 : * Added multibyte support.
790 : * - Tatsuo Ishii 1998-4-21
791 : * Changed behavior if starting position is less than one to conform to SQL behavior.
792 : * Formerly returned the entire string; now returns a portion.
793 : * - Thomas Lockhart 1998-12-10
794 : * Now uses faster TOAST-slicing interface
795 : * - John Gray 2002-02-22
796 : * Remove "#ifdef MULTIBYTE" and test for encoding_max_length instead. Change
797 : * behaviors conflicting with SQL to meet SQL (if E = S + L < S throw
798 : * error; if E < 1, return '', not entire string). Fixed MB related bug when
799 : * S > LC and < LC + 4 sometimes garbage characters are returned.
800 : * - Joe Conway 2002-08-10
801 : */
802 : Datum
803 69242 : text_substr(PG_FUNCTION_ARGS)
804 : {
805 69242 : PG_RETURN_TEXT_P(text_substring(PG_GETARG_DATUM(0),
806 : PG_GETARG_INT32(1),
807 : PG_GETARG_INT32(2),
808 : false));
809 : }
810 :
811 : /*
812 : * text_substr_no_len -
813 : * Wrapper to avoid opr_sanity failure due to
814 : * one function accepting a different number of args.
815 : */
816 : Datum
817 26 : text_substr_no_len(PG_FUNCTION_ARGS)
818 : {
819 26 : PG_RETURN_TEXT_P(text_substring(PG_GETARG_DATUM(0),
820 : PG_GETARG_INT32(1),
821 : -1, true));
822 : }
823 :
824 : /*
825 : * text_substring -
826 : * Does the real work for text_substr() and text_substr_no_len()
827 : *
828 : * This is broken out so it can be called directly by other string processing
829 : * functions. Note that the argument is passed as a Datum, to indicate that
830 : * it may still be in compressed/toasted form. We can avoid detoasting all
831 : * of it in some cases.
832 : *
833 : * The result is always a freshly palloc'd datum.
834 : */
835 : static text *
836 94488 : text_substring(Datum str, int32 start, int32 length, bool length_not_specified)
837 : {
838 94488 : int32 eml = pg_database_encoding_max_length();
839 94488 : int32 S = start; /* start position */
840 : int32 S1; /* adjusted start position */
841 : int32 L1; /* adjusted substring length */
842 :
843 : /* life is easy if the encoding max length is 1 */
844 94488 : if (eml == 1)
845 : {
846 12 : S1 = Max(S, 1);
847 :
848 12 : if (length_not_specified) /* special case - get length to end of
849 : * string */
850 0 : L1 = -1;
851 : else
852 : {
853 : /* end position */
854 12 : int E = S + length;
855 :
856 : /*
857 : * A negative value for L is the only way for the end position to
858 : * be before the start. SQL99 says to throw an error.
859 : */
860 12 : if (E < S)
861 0 : ereport(ERROR,
862 : (errcode(ERRCODE_SUBSTRING_ERROR),
863 : errmsg("negative substring length not allowed")));
864 :
865 : /*
866 : * A zero or negative value for the end position can happen if the
867 : * start was negative or one. SQL99 says to return a zero-length
868 : * string.
869 : */
870 12 : if (E < 1)
871 0 : return cstring_to_text("");
872 :
873 12 : L1 = E - S1;
874 : }
875 :
876 : /*
877 : * If the start position is past the end of the string, SQL99 says to
878 : * return a zero-length string -- PG_GETARG_TEXT_P_SLICE() will do
879 : * that for us. Convert to zero-based starting position
880 : */
881 12 : return DatumGetTextPSlice(str, S1 - 1, L1);
882 : }
883 94476 : else if (eml > 1)
884 : {
885 : /*
886 : * When encoding max length is > 1, we can't get LC without
887 : * detoasting, so we'll grab a conservatively large slice now and go
888 : * back later to do the right thing
889 : */
890 : int32 slice_start;
891 : int32 slice_size;
892 : int32 slice_strlen;
893 : text *slice;
894 : int32 E1;
895 : int32 i;
896 : char *p;
897 : char *s;
898 : text *ret;
899 :
900 : /*
901 : * if S is past the end of the string, the tuple toaster will return a
902 : * zero-length string to us
903 : */
904 94476 : S1 = Max(S, 1);
905 :
906 : /*
907 : * We need to start at position zero because there is no way to know
908 : * in advance which byte offset corresponds to the supplied start
909 : * position.
910 : */
911 94476 : slice_start = 0;
912 :
913 94476 : if (length_not_specified) /* special case - get length to end of
914 : * string */
915 58 : slice_size = L1 = -1;
916 : else
917 : {
918 94418 : int E = S + length;
919 :
920 : /*
921 : * A negative value for L is the only way for the end position to
922 : * be before the start. SQL99 says to throw an error.
923 : */
924 94418 : if (E < S)
925 4 : ereport(ERROR,
926 : (errcode(ERRCODE_SUBSTRING_ERROR),
927 : errmsg("negative substring length not allowed")));
928 :
929 : /*
930 : * A zero or negative value for the end position can happen if the
931 : * start was negative or one. SQL99 says to return a zero-length
932 : * string.
933 : */
934 94414 : if (E < 1)
935 0 : return cstring_to_text("");
936 :
937 : /*
938 : * if E is past the end of the string, the tuple toaster will
939 : * truncate the length for us
940 : */
941 94414 : L1 = E - S1;
942 :
943 : /*
944 : * Total slice size in bytes can't be any longer than the start
945 : * position plus substring length times the encoding max length.
946 : */
947 94414 : slice_size = (S1 + L1) * eml;
948 : }
949 :
950 : /*
951 : * If we're working with an untoasted source, no need to do an extra
952 : * copying step.
953 : */
954 188920 : if (VARATT_IS_COMPRESSED(DatumGetPointer(str)) ||
955 94448 : VARATT_IS_EXTERNAL(DatumGetPointer(str)))
956 68 : slice = DatumGetTextPSlice(str, slice_start, slice_size);
957 : else
958 94404 : slice = (text *) DatumGetPointer(str);
959 :
960 : /* see if we got back an empty string */
961 94472 : if (VARSIZE_ANY_EXHDR(slice) == 0)
962 : {
963 0 : if (slice != (text *) DatumGetPointer(str))
964 0 : pfree(slice);
965 0 : return cstring_to_text("");
966 : }
967 :
968 : /* Now we can get the actual length of the slice in MB characters */
969 283416 : slice_strlen = pg_mbstrlen_with_len(VARDATA_ANY(slice),
970 283416 : VARSIZE_ANY_EXHDR(slice));
971 :
972 : /*
973 : * Check that the start position wasn't > slice_strlen. If so, SQL99
974 : * says to return a zero-length string.
975 : */
976 94472 : if (S1 > slice_strlen)
977 : {
978 20 : if (slice != (text *) DatumGetPointer(str))
979 0 : pfree(slice);
980 20 : return cstring_to_text("");
981 : }
982 :
983 : /*
984 : * Adjust L1 and E1 now that we know the slice string length. Again
985 : * remember that S1 is one based, and slice_start is zero based.
986 : */
987 94452 : if (L1 > -1)
988 94414 : E1 = Min(S1 + L1, slice_start + 1 + slice_strlen);
989 : else
990 38 : E1 = slice_start + 1 + slice_strlen;
991 :
992 : /*
993 : * Find the start position in the slice; remember S1 is not zero based
994 : */
995 94452 : p = VARDATA_ANY(slice);
996 3294458 : for (i = 0; i < S1 - 1; i++)
997 3200006 : p += pg_mblen(p);
998 :
999 : /* hang onto a pointer to our start position */
1000 94452 : s = p;
1001 :
1002 : /*
1003 : * Count the actual bytes used by the substring of the requested
1004 : * length.
1005 : */
1006 1641434 : for (i = S1; i < E1; i++)
1007 1546982 : p += pg_mblen(p);
1008 :
1009 94452 : ret = (text *) palloc(VARHDRSZ + (p - s));
1010 94452 : SET_VARSIZE(ret, VARHDRSZ + (p - s));
1011 94452 : memcpy(VARDATA(ret), s, (p - s));
1012 :
1013 94452 : if (slice != (text *) DatumGetPointer(str))
1014 68 : pfree(slice);
1015 :
1016 94452 : return ret;
1017 : }
1018 : else
1019 0 : elog(ERROR, "invalid backend encoding: encoding max length < 1");
1020 :
1021 : /* not reached: suppress compiler warning */
1022 : return NULL;
1023 : }
1024 :
1025 : /*
1026 : * textoverlay
1027 : * Replace specified substring of first string with second
1028 : *
1029 : * The SQL standard defines OVERLAY() in terms of substring and concatenation.
1030 : * This code is a direct implementation of what the standard says.
1031 : */
1032 : Datum
1033 24 : textoverlay(PG_FUNCTION_ARGS)
1034 : {
1035 24 : text *t1 = PG_GETARG_TEXT_PP(0);
1036 24 : text *t2 = PG_GETARG_TEXT_PP(1);
1037 24 : int sp = PG_GETARG_INT32(2); /* substring start position */
1038 24 : int sl = PG_GETARG_INT32(3); /* substring length */
1039 :
1040 24 : PG_RETURN_TEXT_P(text_overlay(t1, t2, sp, sl));
1041 : }
1042 :
1043 : Datum
1044 8 : textoverlay_no_len(PG_FUNCTION_ARGS)
1045 : {
1046 8 : text *t1 = PG_GETARG_TEXT_PP(0);
1047 8 : text *t2 = PG_GETARG_TEXT_PP(1);
1048 8 : int sp = PG_GETARG_INT32(2); /* substring start position */
1049 : int sl;
1050 :
1051 8 : sl = text_length(PointerGetDatum(t2)); /* defaults to length(t2) */
1052 8 : PG_RETURN_TEXT_P(text_overlay(t1, t2, sp, sl));
1053 : }
1054 :
1055 : static text *
1056 32 : text_overlay(text *t1, text *t2, int sp, int sl)
1057 : {
1058 : text *result;
1059 : text *s1;
1060 : text *s2;
1061 : int sp_pl_sl;
1062 :
1063 : /*
1064 : * Check for possible integer-overflow cases. For negative sp, throw a
1065 : * "substring length" error because that's what should be expected
1066 : * according to the spec's definition of OVERLAY().
1067 : */
1068 32 : if (sp <= 0)
1069 0 : ereport(ERROR,
1070 : (errcode(ERRCODE_SUBSTRING_ERROR),
1071 : errmsg("negative substring length not allowed")));
1072 32 : if (pg_add_s32_overflow(sp, sl, &sp_pl_sl))
1073 0 : ereport(ERROR,
1074 : (errcode(ERRCODE_NUMERIC_VALUE_OUT_OF_RANGE),
1075 : errmsg("integer out of range")));
1076 :
1077 32 : s1 = text_substring(PointerGetDatum(t1), 1, sp - 1, false);
1078 32 : s2 = text_substring(PointerGetDatum(t1), sp_pl_sl, -1, true);
1079 32 : result = text_catenate(s1, t2);
1080 32 : result = text_catenate(result, s2);
1081 :
1082 32 : return result;
1083 : }
1084 :
1085 : /*
1086 : * textpos -
1087 : * Return the position of the specified substring.
1088 : * Implements the SQL POSITION() function.
1089 : * Ref: A Guide To The SQL Standard, Date & Darwen, 1997
1090 : * - thomas 1997-07-27
1091 : */
1092 : Datum
1093 80 : textpos(PG_FUNCTION_ARGS)
1094 : {
1095 80 : text *str = PG_GETARG_TEXT_PP(0);
1096 80 : text *search_str = PG_GETARG_TEXT_PP(1);
1097 :
1098 80 : PG_RETURN_INT32((int32) text_position(str, search_str, PG_GET_COLLATION()));
1099 : }
1100 :
1101 : /*
1102 : * text_position -
1103 : * Does the real work for textpos()
1104 : *
1105 : * Inputs:
1106 : * t1 - string to be searched
1107 : * t2 - pattern to match within t1
1108 : * Result:
1109 : * Character index of the first matched char, starting from 1,
1110 : * or 0 if no match.
1111 : *
1112 : * This is broken out so it can be called directly by other string processing
1113 : * functions.
1114 : */
1115 : static int
1116 80 : text_position(text *t1, text *t2, Oid collid)
1117 : {
1118 : TextPositionState state;
1119 : int result;
1120 :
1121 80 : if (VARSIZE_ANY_EXHDR(t1) < 1 || VARSIZE_ANY_EXHDR(t2) < 1)
1122 16 : return 0;
1123 :
1124 64 : text_position_setup(t1, t2, collid, &state);
1125 64 : if (!text_position_next(&state))
1126 22 : result = 0;
1127 : else
1128 42 : result = text_position_get_match_pos(&state);
1129 64 : text_position_cleanup(&state);
1130 64 : return result;
1131 : }
1132 :
1133 :
1134 : /*
1135 : * text_position_setup, text_position_next, text_position_cleanup -
1136 : * Component steps of text_position()
1137 : *
1138 : * These are broken out so that a string can be efficiently searched for
1139 : * multiple occurrences of the same pattern. text_position_next may be
1140 : * called multiple times, and it advances to the next match on each call.
1141 : * text_position_get_match_ptr() and text_position_get_match_pos() return
1142 : * a pointer or 1-based character position of the last match, respectively.
1143 : *
1144 : * The "state" variable is normally just a local variable in the caller.
1145 : *
1146 : * NOTE: text_position_next skips over the matched portion. For example,
1147 : * searching for "xx" in "xxx" returns only one match, not two.
1148 : */
1149 :
1150 : static void
1151 1086 : text_position_setup(text *t1, text *t2, Oid collid, TextPositionState *state)
1152 : {
1153 1086 : int len1 = VARSIZE_ANY_EXHDR(t1);
1154 1086 : int len2 = VARSIZE_ANY_EXHDR(t2);
1155 1086 : pg_locale_t mylocale = 0;
1156 :
1157 1086 : check_collation_set(collid);
1158 :
1159 1086 : if (!lc_collate_is_c(collid) && collid != DEFAULT_COLLATION_OID)
1160 0 : mylocale = pg_newlocale_from_collation(collid);
1161 :
1162 1086 : if (mylocale && !mylocale->deterministic)
1163 0 : ereport(ERROR,
1164 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1165 : errmsg("nondeterministic collations are not supported for substring searches")));
1166 :
1167 : Assert(len1 > 0);
1168 : Assert(len2 > 0);
1169 :
1170 : /*
1171 : * Even with a multi-byte encoding, we perform the search using the raw
1172 : * byte sequence, ignoring multibyte issues. For UTF-8, that works fine,
1173 : * because in UTF-8 the byte sequence of one character cannot contain
1174 : * another character. For other multi-byte encodings, we do the search
1175 : * initially as a simple byte search, ignoring multibyte issues, but
1176 : * verify afterwards that the match we found is at a character boundary,
1177 : * and continue the search if it was a false match.
1178 : */
1179 1086 : if (pg_database_encoding_max_length() == 1)
1180 : {
1181 24 : state->is_multibyte = false;
1182 24 : state->is_multibyte_char_in_char = false;
1183 : }
1184 1062 : else if (GetDatabaseEncoding() == PG_UTF8)
1185 : {
1186 1062 : state->is_multibyte = true;
1187 1062 : state->is_multibyte_char_in_char = false;
1188 : }
1189 : else
1190 : {
1191 0 : state->is_multibyte = true;
1192 0 : state->is_multibyte_char_in_char = true;
1193 : }
1194 :
1195 1086 : state->str1 = VARDATA_ANY(t1);
1196 1086 : state->str2 = VARDATA_ANY(t2);
1197 1086 : state->len1 = len1;
1198 1086 : state->len2 = len2;
1199 1086 : state->last_match = NULL;
1200 1086 : state->refpoint = state->str1;
1201 1086 : state->refpos = 0;
1202 :
1203 : /*
1204 : * Prepare the skip table for Boyer-Moore-Horspool searching. In these
1205 : * notes we use the terminology that the "haystack" is the string to be
1206 : * searched (t1) and the "needle" is the pattern being sought (t2).
1207 : *
1208 : * If the needle is empty or bigger than the haystack then there is no
1209 : * point in wasting cycles initializing the table. We also choose not to
1210 : * use B-M-H for needles of length 1, since the skip table can't possibly
1211 : * save anything in that case.
1212 : */
1213 1086 : if (len1 >= len2 && len2 > 1)
1214 : {
1215 984 : int searchlength = len1 - len2;
1216 : int skiptablemask;
1217 : int last;
1218 : int i;
1219 984 : const char *str2 = state->str2;
1220 :
1221 : /*
1222 : * First we must determine how much of the skip table to use. The
1223 : * declaration of TextPositionState allows up to 256 elements, but for
1224 : * short search problems we don't really want to have to initialize so
1225 : * many elements --- it would take too long in comparison to the
1226 : * actual search time. So we choose a useful skip table size based on
1227 : * the haystack length minus the needle length. The closer the needle
1228 : * length is to the haystack length the less useful skipping becomes.
1229 : *
1230 : * Note: since we use bit-masking to select table elements, the skip
1231 : * table size MUST be a power of 2, and so the mask must be 2^N-1.
1232 : */
1233 984 : if (searchlength < 16)
1234 36 : skiptablemask = 3;
1235 948 : else if (searchlength < 64)
1236 4 : skiptablemask = 7;
1237 944 : else if (searchlength < 128)
1238 2 : skiptablemask = 15;
1239 942 : else if (searchlength < 512)
1240 784 : skiptablemask = 31;
1241 158 : else if (searchlength < 2048)
1242 88 : skiptablemask = 63;
1243 70 : else if (searchlength < 4096)
1244 16 : skiptablemask = 127;
1245 : else
1246 54 : skiptablemask = 255;
1247 984 : state->skiptablemask = skiptablemask;
1248 :
1249 : /*
1250 : * Initialize the skip table. We set all elements to the needle
1251 : * length, since this is the correct skip distance for any character
1252 : * not found in the needle.
1253 : */
1254 47784 : for (i = 0; i <= skiptablemask; i++)
1255 46800 : state->skiptable[i] = len2;
1256 :
1257 : /*
1258 : * Now examine the needle. For each character except the last one,
1259 : * set the corresponding table element to the appropriate skip
1260 : * distance. Note that when two characters share the same skip table
1261 : * entry, the one later in the needle must determine the skip
1262 : * distance.
1263 : */
1264 984 : last = len2 - 1;
1265 :
1266 12808 : for (i = 0; i < last; i++)
1267 11824 : state->skiptable[(unsigned char) str2[i] & skiptablemask] = last - i;
1268 : }
1269 1086 : }
1270 :
1271 : /*
1272 : * Advance to the next match, starting from the end of the previous match
1273 : * (or the beginning of the string, on first call). Returns true if a match
1274 : * is found.
1275 : */
1276 : static bool
1277 4106 : text_position_next(TextPositionState *state)
1278 : {
1279 4106 : int needle_len = state->len2;
1280 : char *start_ptr;
1281 : char *matchptr;
1282 :
1283 4106 : if (needle_len <= 0)
1284 0 : return false; /* result for empty pattern */
1285 :
1286 : /* Start from the point right after the previous match. */
1287 4106 : if (state->last_match)
1288 3020 : start_ptr = state->last_match + needle_len;
1289 : else
1290 1086 : start_ptr = state->str1;
1291 :
1292 : retry:
1293 4106 : matchptr = text_position_next_internal(start_ptr, state);
1294 :
1295 4106 : if (!matchptr)
1296 1036 : return false;
1297 :
1298 : /*
1299 : * Found a match for the byte sequence. If this is a multibyte encoding,
1300 : * where one character's byte sequence can appear inside a longer
1301 : * multi-byte character, we need to verify that the match was at a
1302 : * character boundary, not in the middle of a multi-byte character.
1303 : */
1304 3070 : if (state->is_multibyte_char_in_char)
1305 : {
1306 : /* Walk one character at a time, until we reach the match. */
1307 :
1308 : /* the search should never move backwards. */
1309 : Assert(state->refpoint <= matchptr);
1310 :
1311 0 : while (state->refpoint < matchptr)
1312 : {
1313 : /* step to next character. */
1314 0 : state->refpoint += pg_mblen(state->refpoint);
1315 0 : state->refpos++;
1316 :
1317 : /*
1318 : * If we stepped over the match's start position, then it was a
1319 : * false positive, where the byte sequence appeared in the middle
1320 : * of a multi-byte character. Skip it, and continue the search at
1321 : * the next character boundary.
1322 : */
1323 0 : if (state->refpoint > matchptr)
1324 : {
1325 0 : start_ptr = state->refpoint;
1326 0 : goto retry;
1327 : }
1328 : }
1329 : }
1330 :
1331 3070 : state->last_match = matchptr;
1332 3070 : return true;
1333 : }
1334 :
1335 : /*
1336 : * Subroutine of text_position_next(). This searches for the raw byte
1337 : * sequence, ignoring any multi-byte encoding issues. Returns the first
1338 : * match starting at 'start_ptr', or NULL if no match is found.
1339 : */
1340 : static char *
1341 4106 : text_position_next_internal(char *start_ptr, TextPositionState *state)
1342 : {
1343 4106 : int haystack_len = state->len1;
1344 4106 : int needle_len = state->len2;
1345 4106 : int skiptablemask = state->skiptablemask;
1346 4106 : const char *haystack = state->str1;
1347 4106 : const char *needle = state->str2;
1348 4106 : const char *haystack_end = &haystack[haystack_len];
1349 : const char *hptr;
1350 :
1351 : Assert(start_ptr >= haystack && start_ptr <= haystack_end);
1352 :
1353 4106 : if (needle_len == 1)
1354 : {
1355 : /* No point in using B-M-H for a one-character needle */
1356 338 : char nchar = *needle;
1357 :
1358 338 : hptr = start_ptr;
1359 3456 : while (hptr < haystack_end)
1360 : {
1361 3050 : if (*hptr == nchar)
1362 270 : return (char *) hptr;
1363 2780 : hptr++;
1364 : }
1365 : }
1366 : else
1367 : {
1368 3768 : const char *needle_last = &needle[needle_len - 1];
1369 :
1370 : /* Start at startpos plus the length of the needle */
1371 3768 : hptr = start_ptr + needle_len - 1;
1372 103558 : while (hptr < haystack_end)
1373 : {
1374 : /* Match the needle scanning *backward* */
1375 : const char *nptr;
1376 : const char *p;
1377 :
1378 98822 : nptr = needle_last;
1379 98822 : p = hptr;
1380 240416 : while (*nptr == *p)
1381 : {
1382 : /* Matched it all? If so, return 1-based position */
1383 45572 : if (nptr == needle)
1384 2800 : return (char *) p;
1385 42772 : nptr--, p--;
1386 : }
1387 :
1388 : /*
1389 : * No match, so use the haystack char at hptr to decide how far to
1390 : * advance. If the needle had any occurrence of that character
1391 : * (or more precisely, one sharing the same skiptable entry)
1392 : * before its last character, then we advance far enough to align
1393 : * the last such needle character with that haystack position.
1394 : * Otherwise we can advance by the whole needle length.
1395 : */
1396 96022 : hptr += state->skiptable[(unsigned char) *hptr & skiptablemask];
1397 : }
1398 : }
1399 :
1400 1036 : return 0; /* not found */
1401 : }
1402 :
1403 : /*
1404 : * Return a pointer to the current match.
1405 : *
1406 : * The returned pointer points into correct position in the original
1407 : * the haystack string.
1408 : */
1409 : static char *
1410 3028 : text_position_get_match_ptr(TextPositionState *state)
1411 : {
1412 3028 : return state->last_match;
1413 : }
1414 :
1415 : /*
1416 : * Return the offset of the current match.
1417 : *
1418 : * The offset is in characters, 1-based.
1419 : */
1420 : static int
1421 42 : text_position_get_match_pos(TextPositionState *state)
1422 : {
1423 42 : if (!state->is_multibyte)
1424 0 : return state->last_match - state->str1 + 1;
1425 : else
1426 : {
1427 : /* Convert the byte position to char position. */
1428 144 : while (state->refpoint < state->last_match)
1429 : {
1430 60 : state->refpoint += pg_mblen(state->refpoint);
1431 60 : state->refpos++;
1432 : }
1433 : Assert(state->refpoint == state->last_match);
1434 42 : return state->refpos + 1;
1435 : }
1436 : }
1437 :
1438 : static void
1439 1086 : text_position_cleanup(TextPositionState *state)
1440 : {
1441 : /* no cleanup needed */
1442 1086 : }
1443 :
1444 : static void
1445 9035504 : check_collation_set(Oid collid)
1446 : {
1447 9035504 : if (!OidIsValid(collid))
1448 : {
1449 : /*
1450 : * This typically means that the parser could not resolve a conflict
1451 : * of implicit collations, so report it that way.
1452 : */
1453 8 : ereport(ERROR,
1454 : (errcode(ERRCODE_INDETERMINATE_COLLATION),
1455 : errmsg("could not determine which collation to use for string comparison"),
1456 : errhint("Use the COLLATE clause to set the collation explicitly.")));
1457 : }
1458 9035496 : }
1459 :
1460 : /* varstr_cmp()
1461 : * Comparison function for text strings with given lengths.
1462 : * Includes locale support, but must copy strings to temporary memory
1463 : * to allow null-termination for inputs to strcoll().
1464 : * Returns an integer less than, equal to, or greater than zero, indicating
1465 : * whether arg1 is less than, equal to, or greater than arg2.
1466 : */
1467 : int
1468 7125652 : varstr_cmp(const char *arg1, int len1, const char *arg2, int len2, Oid collid)
1469 : {
1470 : int result;
1471 :
1472 7125652 : check_collation_set(collid);
1473 :
1474 : /*
1475 : * Unfortunately, there is no strncoll(), so in the non-C locale case we
1476 : * have to do some memory copying. This turns out to be significantly
1477 : * slower, so we optimize the case where LC_COLLATE is C. We also try to
1478 : * optimize relatively-short strings by avoiding palloc/pfree overhead.
1479 : */
1480 7125648 : if (lc_collate_is_c(collid))
1481 : {
1482 2354330 : result = memcmp(arg1, arg2, Min(len1, len2));
1483 2354330 : if ((result == 0) && (len1 != len2))
1484 51100 : result = (len1 < len2) ? -1 : 1;
1485 : }
1486 : else
1487 : {
1488 : char a1buf[TEXTBUFLEN];
1489 : char a2buf[TEXTBUFLEN];
1490 : char *a1p,
1491 : *a2p;
1492 4771318 : pg_locale_t mylocale = 0;
1493 :
1494 4771318 : if (collid != DEFAULT_COLLATION_OID)
1495 0 : mylocale = pg_newlocale_from_collation(collid);
1496 :
1497 : /*
1498 : * memcmp() can't tell us which of two unequal strings sorts first,
1499 : * but it's a cheap way to tell if they're equal. Testing shows that
1500 : * memcmp() followed by strcoll() is only trivially slower than
1501 : * strcoll() by itself, so we don't lose much if this doesn't work out
1502 : * very often, and if it does - for example, because there are many
1503 : * equal strings in the input - then we win big by avoiding expensive
1504 : * collation-aware comparisons.
1505 : */
1506 4771318 : if (len1 == len2 && memcmp(arg1, arg2, len1) == 0)
1507 2629950 : return 0;
1508 :
1509 : #ifdef WIN32
1510 : /* Win32 does not have UTF-8, so we need to map to UTF-16 */
1511 : if (GetDatabaseEncoding() == PG_UTF8
1512 : && (!mylocale || mylocale->provider == COLLPROVIDER_LIBC))
1513 : {
1514 : int a1len;
1515 : int a2len;
1516 : int r;
1517 :
1518 : if (len1 >= TEXTBUFLEN / 2)
1519 : {
1520 : a1len = len1 * 2 + 2;
1521 : a1p = palloc(a1len);
1522 : }
1523 : else
1524 : {
1525 : a1len = TEXTBUFLEN;
1526 : a1p = a1buf;
1527 : }
1528 : if (len2 >= TEXTBUFLEN / 2)
1529 : {
1530 : a2len = len2 * 2 + 2;
1531 : a2p = palloc(a2len);
1532 : }
1533 : else
1534 : {
1535 : a2len = TEXTBUFLEN;
1536 : a2p = a2buf;
1537 : }
1538 :
1539 : /* stupid Microsloth API does not work for zero-length input */
1540 : if (len1 == 0)
1541 : r = 0;
1542 : else
1543 : {
1544 : r = MultiByteToWideChar(CP_UTF8, 0, arg1, len1,
1545 : (LPWSTR) a1p, a1len / 2);
1546 : if (!r)
1547 : ereport(ERROR,
1548 : (errmsg("could not convert string to UTF-16: error code %lu",
1549 : GetLastError())));
1550 : }
1551 : ((LPWSTR) a1p)[r] = 0;
1552 :
1553 : if (len2 == 0)
1554 : r = 0;
1555 : else
1556 : {
1557 : r = MultiByteToWideChar(CP_UTF8, 0, arg2, len2,
1558 : (LPWSTR) a2p, a2len / 2);
1559 : if (!r)
1560 : ereport(ERROR,
1561 : (errmsg("could not convert string to UTF-16: error code %lu",
1562 : GetLastError())));
1563 : }
1564 : ((LPWSTR) a2p)[r] = 0;
1565 :
1566 : errno = 0;
1567 : #ifdef HAVE_LOCALE_T
1568 : if (mylocale)
1569 : result = wcscoll_l((LPWSTR) a1p, (LPWSTR) a2p, mylocale->info.lt);
1570 : else
1571 : #endif
1572 : result = wcscoll((LPWSTR) a1p, (LPWSTR) a2p);
1573 : if (result == 2147483647) /* _NLSCMPERROR; missing from mingw
1574 : * headers */
1575 : ereport(ERROR,
1576 : (errmsg("could not compare Unicode strings: %m")));
1577 :
1578 : /* Break tie if necessary. */
1579 : if (result == 0 &&
1580 : (!mylocale || mylocale->deterministic))
1581 : {
1582 : result = memcmp(arg1, arg2, Min(len1, len2));
1583 : if ((result == 0) && (len1 != len2))
1584 : result = (len1 < len2) ? -1 : 1;
1585 : }
1586 :
1587 : if (a1p != a1buf)
1588 : pfree(a1p);
1589 : if (a2p != a2buf)
1590 : pfree(a2p);
1591 :
1592 : return result;
1593 : }
1594 : #endif /* WIN32 */
1595 :
1596 2141368 : if (len1 >= TEXTBUFLEN)
1597 200 : a1p = (char *) palloc(len1 + 1);
1598 : else
1599 2141168 : a1p = a1buf;
1600 2141368 : if (len2 >= TEXTBUFLEN)
1601 72 : a2p = (char *) palloc(len2 + 1);
1602 : else
1603 2141296 : a2p = a2buf;
1604 :
1605 2141368 : memcpy(a1p, arg1, len1);
1606 2141368 : a1p[len1] = '\0';
1607 2141368 : memcpy(a2p, arg2, len2);
1608 2141368 : a2p[len2] = '\0';
1609 :
1610 2141368 : if (mylocale)
1611 : {
1612 0 : if (mylocale->provider == COLLPROVIDER_ICU)
1613 : {
1614 : #ifdef USE_ICU
1615 : #ifdef HAVE_UCOL_STRCOLLUTF8
1616 : if (GetDatabaseEncoding() == PG_UTF8)
1617 : {
1618 : UErrorCode status;
1619 :
1620 : status = U_ZERO_ERROR;
1621 : result = ucol_strcollUTF8(mylocale->info.icu.ucol,
1622 : arg1, len1,
1623 : arg2, len2,
1624 : &status);
1625 : if (U_FAILURE(status))
1626 : ereport(ERROR,
1627 : (errmsg("collation failed: %s", u_errorName(status))));
1628 : }
1629 : else
1630 : #endif
1631 : {
1632 : int32_t ulen1,
1633 : ulen2;
1634 : UChar *uchar1,
1635 : *uchar2;
1636 :
1637 : ulen1 = icu_to_uchar(&uchar1, arg1, len1);
1638 : ulen2 = icu_to_uchar(&uchar2, arg2, len2);
1639 :
1640 : result = ucol_strcoll(mylocale->info.icu.ucol,
1641 : uchar1, ulen1,
1642 : uchar2, ulen2);
1643 :
1644 : pfree(uchar1);
1645 : pfree(uchar2);
1646 : }
1647 : #else /* not USE_ICU */
1648 : /* shouldn't happen */
1649 0 : elog(ERROR, "unsupported collprovider: %c", mylocale->provider);
1650 : #endif /* not USE_ICU */
1651 : }
1652 : else
1653 : {
1654 : #ifdef HAVE_LOCALE_T
1655 0 : result = strcoll_l(a1p, a2p, mylocale->info.lt);
1656 : #else
1657 : /* shouldn't happen */
1658 : elog(ERROR, "unsupported collprovider: %c", mylocale->provider);
1659 : #endif
1660 : }
1661 : }
1662 : else
1663 2141368 : result = strcoll(a1p, a2p);
1664 :
1665 : /* Break tie if necessary. */
1666 2141368 : if (result == 0 &&
1667 0 : (!mylocale || mylocale->deterministic))
1668 0 : result = strcmp(a1p, a2p);
1669 :
1670 2141368 : if (a1p != a1buf)
1671 200 : pfree(a1p);
1672 2141368 : if (a2p != a2buf)
1673 72 : pfree(a2p);
1674 : }
1675 :
1676 4495698 : return result;
1677 : }
1678 :
1679 : /* text_cmp()
1680 : * Internal comparison function for text strings.
1681 : * Returns -1, 0 or 1
1682 : */
1683 : static int
1684 6064362 : text_cmp(text *arg1, text *arg2, Oid collid)
1685 : {
1686 : char *a1p,
1687 : *a2p;
1688 : int len1,
1689 : len2;
1690 :
1691 6064362 : a1p = VARDATA_ANY(arg1);
1692 6064362 : a2p = VARDATA_ANY(arg2);
1693 :
1694 6064362 : len1 = VARSIZE_ANY_EXHDR(arg1);
1695 6064362 : len2 = VARSIZE_ANY_EXHDR(arg2);
1696 :
1697 6064362 : return varstr_cmp(a1p, len1, a2p, len2, collid);
1698 : }
1699 :
1700 : /*
1701 : * Comparison functions for text strings.
1702 : *
1703 : * Note: btree indexes need these routines not to leak memory; therefore,
1704 : * be careful to free working copies of toasted datums. Most places don't
1705 : * need to be so careful.
1706 : */
1707 :
1708 : Datum
1709 1581878 : texteq(PG_FUNCTION_ARGS)
1710 : {
1711 1581878 : Oid collid = PG_GET_COLLATION();
1712 : bool result;
1713 :
1714 1581878 : check_collation_set(collid);
1715 :
1716 1581878 : if (lc_collate_is_c(collid) ||
1717 0 : collid == DEFAULT_COLLATION_OID ||
1718 0 : pg_newlocale_from_collation(collid)->deterministic)
1719 1581878 : {
1720 1581878 : Datum arg1 = PG_GETARG_DATUM(0);
1721 1581878 : Datum arg2 = PG_GETARG_DATUM(1);
1722 : Size len1,
1723 : len2;
1724 :
1725 : /*
1726 : * Since we only care about equality or not-equality, we can avoid all
1727 : * the expense of strcoll() here, and just do bitwise comparison. In
1728 : * fact, we don't even have to do a bitwise comparison if we can show
1729 : * the lengths of the strings are unequal; which might save us from
1730 : * having to detoast one or both values.
1731 : */
1732 1581878 : len1 = toast_raw_datum_size(arg1);
1733 1581878 : len2 = toast_raw_datum_size(arg2);
1734 1581878 : if (len1 != len2)
1735 269124 : result = false;
1736 : else
1737 : {
1738 1312754 : text *targ1 = DatumGetTextPP(arg1);
1739 1312754 : text *targ2 = DatumGetTextPP(arg2);
1740 :
1741 1312754 : result = (memcmp(VARDATA_ANY(targ1), VARDATA_ANY(targ2),
1742 : len1 - VARHDRSZ) == 0);
1743 :
1744 1312754 : PG_FREE_IF_COPY(targ1, 0);
1745 1312754 : PG_FREE_IF_COPY(targ2, 1);
1746 : }
1747 : }
1748 : else
1749 : {
1750 0 : text *arg1 = PG_GETARG_TEXT_PP(0);
1751 0 : text *arg2 = PG_GETARG_TEXT_PP(1);
1752 :
1753 0 : result = (text_cmp(arg1, arg2, collid) == 0);
1754 :
1755 0 : PG_FREE_IF_COPY(arg1, 0);
1756 0 : PG_FREE_IF_COPY(arg2, 1);
1757 : }
1758 :
1759 1581878 : PG_RETURN_BOOL(result);
1760 : }
1761 :
1762 : Datum
1763 9832 : textne(PG_FUNCTION_ARGS)
1764 : {
1765 9832 : Oid collid = PG_GET_COLLATION();
1766 : bool result;
1767 :
1768 9832 : check_collation_set(collid);
1769 :
1770 9832 : if (lc_collate_is_c(collid) ||
1771 0 : collid == DEFAULT_COLLATION_OID ||
1772 0 : pg_newlocale_from_collation(collid)->deterministic)
1773 9832 : {
1774 9832 : Datum arg1 = PG_GETARG_DATUM(0);
1775 9832 : Datum arg2 = PG_GETARG_DATUM(1);
1776 : Size len1,
1777 : len2;
1778 :
1779 : /* See comment in texteq() */
1780 9832 : len1 = toast_raw_datum_size(arg1);
1781 9832 : len2 = toast_raw_datum_size(arg2);
1782 9832 : if (len1 != len2)
1783 506 : result = true;
1784 : else
1785 : {
1786 9326 : text *targ1 = DatumGetTextPP(arg1);
1787 9326 : text *targ2 = DatumGetTextPP(arg2);
1788 :
1789 9326 : result = (memcmp(VARDATA_ANY(targ1), VARDATA_ANY(targ2),
1790 : len1 - VARHDRSZ) != 0);
1791 :
1792 9326 : PG_FREE_IF_COPY(targ1, 0);
1793 9326 : PG_FREE_IF_COPY(targ2, 1);
1794 : }
1795 : }
1796 : else
1797 : {
1798 0 : text *arg1 = PG_GETARG_TEXT_PP(0);
1799 0 : text *arg2 = PG_GETARG_TEXT_PP(1);
1800 :
1801 0 : result = (text_cmp(arg1, arg2, collid) != 0);
1802 :
1803 0 : PG_FREE_IF_COPY(arg1, 0);
1804 0 : PG_FREE_IF_COPY(arg2, 1);
1805 : }
1806 :
1807 9832 : PG_RETURN_BOOL(result);
1808 : }
1809 :
1810 : Datum
1811 73590 : text_lt(PG_FUNCTION_ARGS)
1812 : {
1813 73590 : text *arg1 = PG_GETARG_TEXT_PP(0);
1814 73590 : text *arg2 = PG_GETARG_TEXT_PP(1);
1815 : bool result;
1816 :
1817 73590 : result = (text_cmp(arg1, arg2, PG_GET_COLLATION()) < 0);
1818 :
1819 73586 : PG_FREE_IF_COPY(arg1, 0);
1820 73586 : PG_FREE_IF_COPY(arg2, 1);
1821 :
1822 73586 : PG_RETURN_BOOL(result);
1823 : }
1824 :
1825 : Datum
1826 55538 : text_le(PG_FUNCTION_ARGS)
1827 : {
1828 55538 : text *arg1 = PG_GETARG_TEXT_PP(0);
1829 55538 : text *arg2 = PG_GETARG_TEXT_PP(1);
1830 : bool result;
1831 :
1832 55538 : result = (text_cmp(arg1, arg2, PG_GET_COLLATION()) <= 0);
1833 :
1834 55538 : PG_FREE_IF_COPY(arg1, 0);
1835 55538 : PG_FREE_IF_COPY(arg2, 1);
1836 :
1837 55538 : PG_RETURN_BOOL(result);
1838 : }
1839 :
1840 : Datum
1841 34634 : text_gt(PG_FUNCTION_ARGS)
1842 : {
1843 34634 : text *arg1 = PG_GETARG_TEXT_PP(0);
1844 34634 : text *arg2 = PG_GETARG_TEXT_PP(1);
1845 : bool result;
1846 :
1847 34634 : result = (text_cmp(arg1, arg2, PG_GET_COLLATION()) > 0);
1848 :
1849 34634 : PG_FREE_IF_COPY(arg1, 0);
1850 34634 : PG_FREE_IF_COPY(arg2, 1);
1851 :
1852 34634 : PG_RETURN_BOOL(result);
1853 : }
1854 :
1855 : Datum
1856 40204 : text_ge(PG_FUNCTION_ARGS)
1857 : {
1858 40204 : text *arg1 = PG_GETARG_TEXT_PP(0);
1859 40204 : text *arg2 = PG_GETARG_TEXT_PP(1);
1860 : bool result;
1861 :
1862 40204 : result = (text_cmp(arg1, arg2, PG_GET_COLLATION()) >= 0);
1863 :
1864 40204 : PG_FREE_IF_COPY(arg1, 0);
1865 40204 : PG_FREE_IF_COPY(arg2, 1);
1866 :
1867 40204 : PG_RETURN_BOOL(result);
1868 : }
1869 :
1870 : Datum
1871 25132 : text_starts_with(PG_FUNCTION_ARGS)
1872 : {
1873 25132 : Datum arg1 = PG_GETARG_DATUM(0);
1874 25132 : Datum arg2 = PG_GETARG_DATUM(1);
1875 25132 : Oid collid = PG_GET_COLLATION();
1876 25132 : pg_locale_t mylocale = 0;
1877 : bool result;
1878 : Size len1,
1879 : len2;
1880 :
1881 25132 : check_collation_set(collid);
1882 :
1883 25132 : if (!lc_collate_is_c(collid) && collid != DEFAULT_COLLATION_OID)
1884 0 : mylocale = pg_newlocale_from_collation(collid);
1885 :
1886 25132 : if (mylocale && !mylocale->deterministic)
1887 0 : ereport(ERROR,
1888 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1889 : errmsg("nondeterministic collations are not supported for substring searches")));
1890 :
1891 25132 : len1 = toast_raw_datum_size(arg1);
1892 25132 : len2 = toast_raw_datum_size(arg2);
1893 25132 : if (len2 > len1)
1894 0 : result = false;
1895 : else
1896 : {
1897 25132 : text *targ1 = text_substring(arg1, 1, len2, false);
1898 25132 : text *targ2 = DatumGetTextPP(arg2);
1899 :
1900 50264 : result = (memcmp(VARDATA_ANY(targ1), VARDATA_ANY(targ2),
1901 50264 : VARSIZE_ANY_EXHDR(targ2)) == 0);
1902 :
1903 25132 : PG_FREE_IF_COPY(targ1, 0);
1904 25132 : PG_FREE_IF_COPY(targ2, 1);
1905 : }
1906 :
1907 25132 : PG_RETURN_BOOL(result);
1908 : }
1909 :
1910 : Datum
1911 5776242 : bttextcmp(PG_FUNCTION_ARGS)
1912 : {
1913 5776242 : text *arg1 = PG_GETARG_TEXT_PP(0);
1914 5776242 : text *arg2 = PG_GETARG_TEXT_PP(1);
1915 : int32 result;
1916 :
1917 5776242 : result = text_cmp(arg1, arg2, PG_GET_COLLATION());
1918 :
1919 5776242 : PG_FREE_IF_COPY(arg1, 0);
1920 5776242 : PG_FREE_IF_COPY(arg2, 1);
1921 :
1922 5776242 : PG_RETURN_INT32(result);
1923 : }
1924 :
1925 : Datum
1926 51284 : bttextsortsupport(PG_FUNCTION_ARGS)
1927 : {
1928 51284 : SortSupport ssup = (SortSupport) PG_GETARG_POINTER(0);
1929 51284 : Oid collid = ssup->ssup_collation;
1930 : MemoryContext oldcontext;
1931 :
1932 51284 : oldcontext = MemoryContextSwitchTo(ssup->ssup_cxt);
1933 :
1934 : /* Use generic string SortSupport */
1935 51284 : varstr_sortsupport(ssup, TEXTOID, collid);
1936 :
1937 51280 : MemoryContextSwitchTo(oldcontext);
1938 :
1939 51280 : PG_RETURN_VOID();
1940 : }
1941 :
1942 : /*
1943 : * Generic sortsupport interface for character type's operator classes.
1944 : * Includes locale support, and support for BpChar semantics (i.e. removing
1945 : * trailing spaces before comparison).
1946 : *
1947 : * Relies on the assumption that text, VarChar, BpChar, and bytea all have the
1948 : * same representation. Callers that always use the C collation (e.g.
1949 : * non-collatable type callers like bytea) may have NUL bytes in their strings;
1950 : * this will not work with any other collation, though.
1951 : */
1952 : void
1953 96282 : varstr_sortsupport(SortSupport ssup, Oid typid, Oid collid)
1954 : {
1955 96282 : bool abbreviate = ssup->abbreviate;
1956 96282 : bool collate_c = false;
1957 : VarStringSortSupport *sss;
1958 96282 : pg_locale_t locale = 0;
1959 :
1960 96282 : check_collation_set(collid);
1961 :
1962 : /*
1963 : * If possible, set ssup->comparator to a function which can be used to
1964 : * directly compare two datums. If we can do this, we'll avoid the
1965 : * overhead of a trip through the fmgr layer for every comparison, which
1966 : * can be substantial.
1967 : *
1968 : * Most typically, we'll set the comparator to varlenafastcmp_locale,
1969 : * which uses strcoll() to perform comparisons. We use that for the
1970 : * BpChar case too, but type NAME uses namefastcmp_locale. However, if
1971 : * LC_COLLATE = C, we can make things quite a bit faster with
1972 : * varstrfastcmp_c, bpcharfastcmp_c, or namefastcmp_c, all of which use
1973 : * memcmp() rather than strcoll().
1974 : */
1975 96278 : if (lc_collate_is_c(collid))
1976 : {
1977 68220 : if (typid == BPCHAROID)
1978 12 : ssup->comparator = bpcharfastcmp_c;
1979 68208 : else if (typid == NAMEOID)
1980 : {
1981 44066 : ssup->comparator = namefastcmp_c;
1982 : /* Not supporting abbreviation with type NAME, for now */
1983 44066 : abbreviate = false;
1984 : }
1985 : else
1986 24142 : ssup->comparator = varstrfastcmp_c;
1987 :
1988 68220 : collate_c = true;
1989 : }
1990 : else
1991 : {
1992 : /*
1993 : * We need a collation-sensitive comparison. To make things faster,
1994 : * we'll figure out the collation based on the locale id and cache the
1995 : * result.
1996 : */
1997 28058 : if (collid != DEFAULT_COLLATION_OID)
1998 0 : locale = pg_newlocale_from_collation(collid);
1999 :
2000 : /*
2001 : * There is a further exception on Windows. When the database
2002 : * encoding is UTF-8 and we are not using the C collation, complex
2003 : * hacks are required. We don't currently have a comparator that
2004 : * handles that case, so we fall back on the slow method of having the
2005 : * sort code invoke bttextcmp() (in the case of text) via the fmgr
2006 : * trampoline. ICU locales work just the same on Windows, however.
2007 : */
2008 : #ifdef WIN32
2009 : if (GetDatabaseEncoding() == PG_UTF8 &&
2010 : !(locale && locale->provider == COLLPROVIDER_ICU))
2011 : return;
2012 : #endif
2013 :
2014 : /*
2015 : * We use varlenafastcmp_locale except for type NAME.
2016 : */
2017 28058 : if (typid == NAMEOID)
2018 : {
2019 0 : ssup->comparator = namefastcmp_locale;
2020 : /* Not supporting abbreviation with type NAME, for now */
2021 0 : abbreviate = false;
2022 : }
2023 : else
2024 28058 : ssup->comparator = varlenafastcmp_locale;
2025 : }
2026 :
2027 : /*
2028 : * Unfortunately, it seems that abbreviation for non-C collations is
2029 : * broken on many common platforms; testing of multiple versions of glibc
2030 : * reveals that, for many locales, strcoll() and strxfrm() do not return
2031 : * consistent results, which is fatal to this optimization. While no
2032 : * other libc other than Cygwin has so far been shown to have a problem,
2033 : * we take the conservative course of action for right now and disable
2034 : * this categorically. (Users who are certain this isn't a problem on
2035 : * their system can define TRUST_STRXFRM.)
2036 : *
2037 : * Even apart from the risk of broken locales, it's possible that there
2038 : * are platforms where the use of abbreviated keys should be disabled at
2039 : * compile time. Having only 4 byte datums could make worst-case
2040 : * performance drastically more likely, for example. Moreover, macOS's
2041 : * strxfrm() implementation is known to not effectively concentrate a
2042 : * significant amount of entropy from the original string in earlier
2043 : * transformed blobs. It's possible that other supported platforms are
2044 : * similarly encumbered. So, if we ever get past disabling this
2045 : * categorically, we may still want or need to disable it for particular
2046 : * platforms.
2047 : */
2048 : #ifndef TRUST_STRXFRM
2049 96278 : if (!collate_c && !(locale && locale->provider == COLLPROVIDER_ICU))
2050 28058 : abbreviate = false;
2051 : #endif
2052 :
2053 : /*
2054 : * If we're using abbreviated keys, or if we're using a locale-aware
2055 : * comparison, we need to initialize a StringSortSupport object. Both
2056 : * cases will make use of the temporary buffers we initialize here for
2057 : * scratch space (and to detect requirement for BpChar semantics from
2058 : * caller), and the abbreviation case requires additional state.
2059 : */
2060 96278 : if (abbreviate || !collate_c)
2061 : {
2062 29272 : sss = palloc(sizeof(VarStringSortSupport));
2063 29272 : sss->buf1 = palloc(TEXTBUFLEN);
2064 29272 : sss->buflen1 = TEXTBUFLEN;
2065 29272 : sss->buf2 = palloc(TEXTBUFLEN);
2066 29272 : sss->buflen2 = TEXTBUFLEN;
2067 : /* Start with invalid values */
2068 29272 : sss->last_len1 = -1;
2069 29272 : sss->last_len2 = -1;
2070 : /* Initialize */
2071 29272 : sss->last_returned = 0;
2072 29272 : sss->locale = locale;
2073 :
2074 : /*
2075 : * To avoid somehow confusing a strxfrm() blob and an original string,
2076 : * constantly keep track of the variety of data that buf1 and buf2
2077 : * currently contain.
2078 : *
2079 : * Comparisons may be interleaved with conversion calls. Frequently,
2080 : * conversions and comparisons are batched into two distinct phases,
2081 : * but the correctness of caching cannot hinge upon this. For
2082 : * comparison caching, buffer state is only trusted if cache_blob is
2083 : * found set to false, whereas strxfrm() caching only trusts the state
2084 : * when cache_blob is found set to true.
2085 : *
2086 : * Arbitrarily initialize cache_blob to true.
2087 : */
2088 29272 : sss->cache_blob = true;
2089 29272 : sss->collate_c = collate_c;
2090 29272 : sss->typid = typid;
2091 29272 : ssup->ssup_extra = sss;
2092 :
2093 : /*
2094 : * If possible, plan to use the abbreviated keys optimization. The
2095 : * core code may switch back to authoritative comparator should
2096 : * abbreviation be aborted.
2097 : */
2098 29272 : if (abbreviate)
2099 : {
2100 1214 : sss->prop_card = 0.20;
2101 1214 : initHyperLogLog(&sss->abbr_card, 10);
2102 1214 : initHyperLogLog(&sss->full_card, 10);
2103 1214 : ssup->abbrev_full_comparator = ssup->comparator;
2104 1214 : ssup->comparator = varstrcmp_abbrev;
2105 1214 : ssup->abbrev_converter = varstr_abbrev_convert;
2106 1214 : ssup->abbrev_abort = varstr_abbrev_abort;
2107 : }
2108 : }
2109 96278 : }
2110 :
2111 : /*
2112 : * sortsupport comparison func (for C locale case)
2113 : */
2114 : static int
2115 51630868 : varstrfastcmp_c(Datum x, Datum y, SortSupport ssup)
2116 : {
2117 51630868 : VarString *arg1 = DatumGetVarStringPP(x);
2118 51630868 : VarString *arg2 = DatumGetVarStringPP(y);
2119 : char *a1p,
2120 : *a2p;
2121 : int len1,
2122 : len2,
2123 : result;
2124 :
2125 51630868 : a1p = VARDATA_ANY(arg1);
2126 51630868 : a2p = VARDATA_ANY(arg2);
2127 :
2128 51630868 : len1 = VARSIZE_ANY_EXHDR(arg1);
2129 51630868 : len2 = VARSIZE_ANY_EXHDR(arg2);
2130 :
2131 51630868 : result = memcmp(a1p, a2p, Min(len1, len2));
2132 51630868 : if ((result == 0) && (len1 != len2))
2133 1046308 : result = (len1 < len2) ? -1 : 1;
2134 :
2135 : /* We can't afford to leak memory here. */
2136 51630868 : if (PointerGetDatum(arg1) != x)
2137 0 : pfree(arg1);
2138 51630868 : if (PointerGetDatum(arg2) != y)
2139 0 : pfree(arg2);
2140 :
2141 51630868 : return result;
2142 : }
2143 :
2144 : /*
2145 : * sortsupport comparison func (for BpChar C locale case)
2146 : *
2147 : * BpChar outsources its sortsupport to this module. Specialization for the
2148 : * varstr_sortsupport BpChar case, modeled on
2149 : * internal_bpchar_pattern_compare().
2150 : */
2151 : static int
2152 16 : bpcharfastcmp_c(Datum x, Datum y, SortSupport ssup)
2153 : {
2154 16 : BpChar *arg1 = DatumGetBpCharPP(x);
2155 16 : BpChar *arg2 = DatumGetBpCharPP(y);
2156 : char *a1p,
2157 : *a2p;
2158 : int len1,
2159 : len2,
2160 : result;
2161 :
2162 16 : a1p = VARDATA_ANY(arg1);
2163 16 : a2p = VARDATA_ANY(arg2);
2164 :
2165 16 : len1 = bpchartruelen(a1p, VARSIZE_ANY_EXHDR(arg1));
2166 16 : len2 = bpchartruelen(a2p, VARSIZE_ANY_EXHDR(arg2));
2167 :
2168 16 : result = memcmp(a1p, a2p, Min(len1, len2));
2169 16 : if ((result == 0) && (len1 != len2))
2170 0 : result = (len1 < len2) ? -1 : 1;
2171 :
2172 : /* We can't afford to leak memory here. */
2173 16 : if (PointerGetDatum(arg1) != x)
2174 0 : pfree(arg1);
2175 16 : if (PointerGetDatum(arg2) != y)
2176 0 : pfree(arg2);
2177 :
2178 16 : return result;
2179 : }
2180 :
2181 : /*
2182 : * sortsupport comparison func (for NAME C locale case)
2183 : */
2184 : static int
2185 65224724 : namefastcmp_c(Datum x, Datum y, SortSupport ssup)
2186 : {
2187 65224724 : Name arg1 = DatumGetName(x);
2188 65224724 : Name arg2 = DatumGetName(y);
2189 :
2190 65224724 : return strncmp(NameStr(*arg1), NameStr(*arg2), NAMEDATALEN);
2191 : }
2192 :
2193 : /*
2194 : * sortsupport comparison func (for locale case with all varlena types)
2195 : */
2196 : static int
2197 28869422 : varlenafastcmp_locale(Datum x, Datum y, SortSupport ssup)
2198 : {
2199 28869422 : VarString *arg1 = DatumGetVarStringPP(x);
2200 28869422 : VarString *arg2 = DatumGetVarStringPP(y);
2201 : char *a1p,
2202 : *a2p;
2203 : int len1,
2204 : len2,
2205 : result;
2206 :
2207 28869422 : a1p = VARDATA_ANY(arg1);
2208 28869422 : a2p = VARDATA_ANY(arg2);
2209 :
2210 28869422 : len1 = VARSIZE_ANY_EXHDR(arg1);
2211 28869422 : len2 = VARSIZE_ANY_EXHDR(arg2);
2212 :
2213 28869422 : result = varstrfastcmp_locale(a1p, len1, a2p, len2, ssup);
2214 :
2215 : /* We can't afford to leak memory here. */
2216 28869422 : if (PointerGetDatum(arg1) != x)
2217 0 : pfree(arg1);
2218 28869422 : if (PointerGetDatum(arg2) != y)
2219 0 : pfree(arg2);
2220 :
2221 28869422 : return result;
2222 : }
2223 :
2224 : /*
2225 : * sortsupport comparison func (for locale case with NAME type)
2226 : */
2227 : static int
2228 0 : namefastcmp_locale(Datum x, Datum y, SortSupport ssup)
2229 : {
2230 0 : Name arg1 = DatumGetName(x);
2231 0 : Name arg2 = DatumGetName(y);
2232 :
2233 0 : return varstrfastcmp_locale(NameStr(*arg1), strlen(NameStr(*arg1)),
2234 0 : NameStr(*arg2), strlen(NameStr(*arg2)),
2235 : ssup);
2236 : }
2237 :
2238 : /*
2239 : * sortsupport comparison func for locale cases
2240 : */
2241 : static int
2242 28869422 : varstrfastcmp_locale(char *a1p, int len1, char *a2p, int len2, SortSupport ssup)
2243 : {
2244 28869422 : VarStringSortSupport *sss = (VarStringSortSupport *) ssup->ssup_extra;
2245 : int result;
2246 : bool arg1_match;
2247 :
2248 : /* Fast pre-check for equality, as discussed in varstr_cmp() */
2249 28869422 : if (len1 == len2 && memcmp(a1p, a2p, len1) == 0)
2250 : {
2251 : /*
2252 : * No change in buf1 or buf2 contents, so avoid changing last_len1 or
2253 : * last_len2. Existing contents of buffers might still be used by
2254 : * next call.
2255 : *
2256 : * It's fine to allow the comparison of BpChar padding bytes here,
2257 : * even though that implies that the memcmp() will usually be
2258 : * performed for BpChar callers (though multibyte characters could
2259 : * still prevent that from occurring). The memcmp() is still very
2260 : * cheap, and BpChar's funny semantics have us remove trailing spaces
2261 : * (not limited to padding), so we need make no distinction between
2262 : * padding space characters and "real" space characters.
2263 : */
2264 12537158 : return 0;
2265 : }
2266 :
2267 16332264 : if (sss->typid == BPCHAROID)
2268 : {
2269 : /* Get true number of bytes, ignoring trailing spaces */
2270 58852 : len1 = bpchartruelen(a1p, len1);
2271 58852 : len2 = bpchartruelen(a2p, len2);
2272 : }
2273 :
2274 16332264 : if (len1 >= sss->buflen1)
2275 : {
2276 0 : pfree(sss->buf1);
2277 0 : sss->buflen1 = Max(len1 + 1, Min(sss->buflen1 * 2, MaxAllocSize));
2278 0 : sss->buf1 = MemoryContextAlloc(ssup->ssup_cxt, sss->buflen1);
2279 : }
2280 16332264 : if (len2 >= sss->buflen2)
2281 : {
2282 0 : pfree(sss->buf2);
2283 0 : sss->buflen2 = Max(len2 + 1, Min(sss->buflen2 * 2, MaxAllocSize));
2284 0 : sss->buf2 = MemoryContextAlloc(ssup->ssup_cxt, sss->buflen2);
2285 : }
2286 :
2287 : /*
2288 : * We're likely to be asked to compare the same strings repeatedly, and
2289 : * memcmp() is so much cheaper than strcoll() that it pays to try to cache
2290 : * comparisons, even though in general there is no reason to think that
2291 : * that will work out (every string datum may be unique). Caching does
2292 : * not slow things down measurably when it doesn't work out, and can speed
2293 : * things up by rather a lot when it does. In part, this is because the
2294 : * memcmp() compares data from cachelines that are needed in L1 cache even
2295 : * when the last comparison's result cannot be reused.
2296 : */
2297 16332264 : arg1_match = true;
2298 16332264 : if (len1 != sss->last_len1 || memcmp(sss->buf1, a1p, len1) != 0)
2299 : {
2300 12116966 : arg1_match = false;
2301 12116966 : memcpy(sss->buf1, a1p, len1);
2302 12116966 : sss->buf1[len1] = '\0';
2303 12116966 : sss->last_len1 = len1;
2304 : }
2305 :
2306 : /*
2307 : * If we're comparing the same two strings as last time, we can return the
2308 : * same answer without calling strcoll() again. This is more likely than
2309 : * it seems (at least with moderate to low cardinality sets), because
2310 : * quicksort compares the same pivot against many values.
2311 : */
2312 16332264 : if (len2 != sss->last_len2 || memcmp(sss->buf2, a2p, len2) != 0)
2313 : {
2314 2547784 : memcpy(sss->buf2, a2p, len2);
2315 2547784 : sss->buf2[len2] = '\0';
2316 2547784 : sss->last_len2 = len2;
2317 : }
2318 13784480 : else if (arg1_match && !sss->cache_blob)
2319 : {
2320 : /* Use result cached following last actual strcoll() call */
2321 3327978 : return sss->last_returned;
2322 : }
2323 :
2324 13004286 : if (sss->locale)
2325 : {
2326 0 : if (sss->locale->provider == COLLPROVIDER_ICU)
2327 : {
2328 : #ifdef USE_ICU
2329 : #ifdef HAVE_UCOL_STRCOLLUTF8
2330 : if (GetDatabaseEncoding() == PG_UTF8)
2331 : {
2332 : UErrorCode status;
2333 :
2334 : status = U_ZERO_ERROR;
2335 : result = ucol_strcollUTF8(sss->locale->info.icu.ucol,
2336 : a1p, len1,
2337 : a2p, len2,
2338 : &status);
2339 : if (U_FAILURE(status))
2340 : ereport(ERROR,
2341 : (errmsg("collation failed: %s", u_errorName(status))));
2342 : }
2343 : else
2344 : #endif
2345 : {
2346 : int32_t ulen1,
2347 : ulen2;
2348 : UChar *uchar1,
2349 : *uchar2;
2350 :
2351 : ulen1 = icu_to_uchar(&uchar1, a1p, len1);
2352 : ulen2 = icu_to_uchar(&uchar2, a2p, len2);
2353 :
2354 : result = ucol_strcoll(sss->locale->info.icu.ucol,
2355 : uchar1, ulen1,
2356 : uchar2, ulen2);
2357 :
2358 : pfree(uchar1);
2359 : pfree(uchar2);
2360 : }
2361 : #else /* not USE_ICU */
2362 : /* shouldn't happen */
2363 0 : elog(ERROR, "unsupported collprovider: %c", sss->locale->provider);
2364 : #endif /* not USE_ICU */
2365 : }
2366 : else
2367 : {
2368 : #ifdef HAVE_LOCALE_T
2369 0 : result = strcoll_l(sss->buf1, sss->buf2, sss->locale->info.lt);
2370 : #else
2371 : /* shouldn't happen */
2372 : elog(ERROR, "unsupported collprovider: %c", sss->locale->provider);
2373 : #endif
2374 : }
2375 : }
2376 : else
2377 13004286 : result = strcoll(sss->buf1, sss->buf2);
2378 :
2379 : /* Break tie if necessary. */
2380 13004286 : if (result == 0 &&
2381 0 : (!sss->locale || sss->locale->deterministic))
2382 0 : result = strcmp(sss->buf1, sss->buf2);
2383 :
2384 : /* Cache result, perhaps saving an expensive strcoll() call next time */
2385 13004286 : sss->cache_blob = false;
2386 13004286 : sss->last_returned = result;
2387 13004286 : return result;
2388 : }
2389 :
2390 : /*
2391 : * Abbreviated key comparison func
2392 : */
2393 : static int
2394 3183246 : varstrcmp_abbrev(Datum x, Datum y, SortSupport ssup)
2395 : {
2396 : /*
2397 : * When 0 is returned, the core system will call varstrfastcmp_c()
2398 : * (bpcharfastcmp_c() in BpChar case) or varlenafastcmp_locale(). Even a
2399 : * strcmp() on two non-truncated strxfrm() blobs cannot indicate *equality*
2400 : * authoritatively, for the same reason that there is a strcoll()
2401 : * tie-breaker call to strcmp() in varstr_cmp().
2402 : */
2403 3183246 : if (x > y)
2404 1443392 : return 1;
2405 1739854 : else if (x == y)
2406 421540 : return 0;
2407 : else
2408 1318314 : return -1;
2409 : }
2410 :
2411 : /*
2412 : * Conversion routine for sortsupport. Converts original to abbreviated key
2413 : * representation. Our encoding strategy is simple -- pack the first 8 bytes
2414 : * of a strxfrm() blob into a Datum (on little-endian machines, the 8 bytes are
2415 : * stored in reverse order), and treat it as an unsigned integer. When the "C"
2416 : * locale is used, or in case of bytea, just memcpy() from original instead.
2417 : */
2418 : static Datum
2419 330020 : varstr_abbrev_convert(Datum original, SortSupport ssup)
2420 : {
2421 330020 : VarStringSortSupport *sss = (VarStringSortSupport *) ssup->ssup_extra;
2422 330020 : VarString *authoritative = DatumGetVarStringPP(original);
2423 330020 : char *authoritative_data = VARDATA_ANY(authoritative);
2424 :
2425 : /* working state */
2426 : Datum res;
2427 : char *pres;
2428 : int len;
2429 : uint32 hash;
2430 :
2431 330020 : pres = (char *) &res;
2432 : /* memset(), so any non-overwritten bytes are NUL */
2433 330020 : memset(pres, 0, sizeof(Datum));
2434 330020 : len = VARSIZE_ANY_EXHDR(authoritative);
2435 :
2436 : /* Get number of bytes, ignoring trailing spaces */
2437 330020 : if (sss->typid == BPCHAROID)
2438 0 : len = bpchartruelen(authoritative_data, len);
2439 :
2440 : /*
2441 : * If we're using the C collation, use memcpy(), rather than strxfrm(), to
2442 : * abbreviate keys. The full comparator for the C locale is always
2443 : * memcmp(). It would be incorrect to allow bytea callers (callers that
2444 : * always force the C collation -- bytea isn't a collatable type, but this
2445 : * approach is convenient) to use strxfrm(). This is because bytea
2446 : * strings may contain NUL bytes. Besides, this should be faster, too.
2447 : *
2448 : * More generally, it's okay that bytea callers can have NUL bytes in
2449 : * strings because varstrcmp_abbrev() need not make a distinction between
2450 : * terminating NUL bytes, and NUL bytes representing actual NULs in the
2451 : * authoritative representation. Hopefully a comparison at or past one
2452 : * abbreviated key's terminating NUL byte will resolve the comparison
2453 : * without consulting the authoritative representation; specifically, some
2454 : * later non-NUL byte in the longer string can resolve the comparison
2455 : * against a subsequent terminating NUL in the shorter string. There will
2456 : * usually be what is effectively a "length-wise" resolution there and
2457 : * then.
2458 : *
2459 : * If that doesn't work out -- if all bytes in the longer string
2460 : * positioned at or past the offset of the smaller string's (first)
2461 : * terminating NUL are actually representative of NUL bytes in the
2462 : * authoritative binary string (perhaps with some *terminating* NUL bytes
2463 : * towards the end of the longer string iff it happens to still be small)
2464 : * -- then an authoritative tie-breaker will happen, and do the right
2465 : * thing: explicitly consider string length.
2466 : */
2467 330020 : if (sss->collate_c)
2468 330020 : memcpy(pres, authoritative_data, Min(len, sizeof(Datum)));
2469 : else
2470 : {
2471 : Size bsize;
2472 : #ifdef USE_ICU
2473 : int32_t ulen = -1;
2474 : UChar *uchar = NULL;
2475 : #endif
2476 :
2477 : /*
2478 : * We're not using the C collation, so fall back on strxfrm or ICU
2479 : * analogs.
2480 : */
2481 :
2482 : /* By convention, we use buffer 1 to store and NUL-terminate */
2483 0 : if (len >= sss->buflen1)
2484 : {
2485 0 : pfree(sss->buf1);
2486 0 : sss->buflen1 = Max(len + 1, Min(sss->buflen1 * 2, MaxAllocSize));
2487 0 : sss->buf1 = palloc(sss->buflen1);
2488 : }
2489 :
2490 : /* Might be able to reuse strxfrm() blob from last call */
2491 0 : if (sss->last_len1 == len && sss->cache_blob &&
2492 0 : memcmp(sss->buf1, authoritative_data, len) == 0)
2493 : {
2494 0 : memcpy(pres, sss->buf2, Min(sizeof(Datum), sss->last_len2));
2495 : /* No change affecting cardinality, so no hashing required */
2496 0 : goto done;
2497 : }
2498 :
2499 0 : memcpy(sss->buf1, authoritative_data, len);
2500 :
2501 : /*
2502 : * Just like strcoll(), strxfrm() expects a NUL-terminated string. Not
2503 : * necessary for ICU, but doesn't hurt.
2504 : */
2505 0 : sss->buf1[len] = '\0';
2506 0 : sss->last_len1 = len;
2507 :
2508 : #ifdef USE_ICU
2509 : /* When using ICU and not UTF8, convert string to UChar. */
2510 : if (sss->locale && sss->locale->provider == COLLPROVIDER_ICU &&
2511 : GetDatabaseEncoding() != PG_UTF8)
2512 : ulen = icu_to_uchar(&uchar, sss->buf1, len);
2513 : #endif
2514 :
2515 : /*
2516 : * Loop: Call strxfrm() or ucol_getSortKey(), possibly enlarge buffer,
2517 : * and try again. Both of these functions have the result buffer
2518 : * content undefined if the result did not fit, so we need to retry
2519 : * until everything fits, even though we only need the first few bytes
2520 : * in the end. When using ucol_nextSortKeyPart(), however, we only
2521 : * ask for as many bytes as we actually need.
2522 : */
2523 : for (;;)
2524 : {
2525 : #ifdef USE_ICU
2526 : if (sss->locale && sss->locale->provider == COLLPROVIDER_ICU)
2527 : {
2528 : /*
2529 : * When using UTF8, use the iteration interface so we only
2530 : * need to produce as many bytes as we actually need.
2531 : */
2532 : if (GetDatabaseEncoding() == PG_UTF8)
2533 : {
2534 : UCharIterator iter;
2535 : uint32_t state[2];
2536 : UErrorCode status;
2537 :
2538 : uiter_setUTF8(&iter, sss->buf1, len);
2539 : state[0] = state[1] = 0; /* won't need that again */
2540 : status = U_ZERO_ERROR;
2541 : bsize = ucol_nextSortKeyPart(sss->locale->info.icu.ucol,
2542 : &iter,
2543 : state,
2544 : (uint8_t *) sss->buf2,
2545 : Min(sizeof(Datum), sss->buflen2),
2546 : &status);
2547 : if (U_FAILURE(status))
2548 : ereport(ERROR,
2549 : (errmsg("sort key generation failed: %s",
2550 : u_errorName(status))));
2551 : }
2552 : else
2553 : bsize = ucol_getSortKey(sss->locale->info.icu.ucol,
2554 : uchar, ulen,
2555 : (uint8_t *) sss->buf2, sss->buflen2);
2556 : }
2557 : else
2558 : #endif
2559 : #ifdef HAVE_LOCALE_T
2560 0 : if (sss->locale && sss->locale->provider == COLLPROVIDER_LIBC)
2561 0 : bsize = strxfrm_l(sss->buf2, sss->buf1,
2562 0 : sss->buflen2, sss->locale->info.lt);
2563 : else
2564 : #endif
2565 0 : bsize = strxfrm(sss->buf2, sss->buf1, sss->buflen2);
2566 :
2567 0 : sss->last_len2 = bsize;
2568 0 : if (bsize < sss->buflen2)
2569 0 : break;
2570 :
2571 : /*
2572 : * Grow buffer and retry.
2573 : */
2574 0 : pfree(sss->buf2);
2575 0 : sss->buflen2 = Max(bsize + 1,
2576 : Min(sss->buflen2 * 2, MaxAllocSize));
2577 0 : sss->buf2 = palloc(sss->buflen2);
2578 : }
2579 :
2580 : /*
2581 : * Every Datum byte is always compared. This is safe because the
2582 : * strxfrm() blob is itself NUL terminated, leaving no danger of
2583 : * misinterpreting any NUL bytes not intended to be interpreted as
2584 : * logically representing termination.
2585 : *
2586 : * (Actually, even if there were NUL bytes in the blob it would be
2587 : * okay. See remarks on bytea case above.)
2588 : */
2589 0 : memcpy(pres, sss->buf2, Min(sizeof(Datum), bsize));
2590 :
2591 : #ifdef USE_ICU
2592 : if (uchar)
2593 : pfree(uchar);
2594 : #endif
2595 : }
2596 :
2597 : /*
2598 : * Maintain approximate cardinality of both abbreviated keys and original,
2599 : * authoritative keys using HyperLogLog. Used as cheap insurance against
2600 : * the worst case, where we do many string transformations for no saving
2601 : * in full strcoll()-based comparisons. These statistics are used by
2602 : * varstr_abbrev_abort().
2603 : *
2604 : * First, Hash key proper, or a significant fraction of it. Mix in length
2605 : * in order to compensate for cases where differences are past
2606 : * PG_CACHE_LINE_SIZE bytes, so as to limit the overhead of hashing.
2607 : */
2608 330020 : hash = DatumGetUInt32(hash_any((unsigned char *) authoritative_data,
2609 : Min(len, PG_CACHE_LINE_SIZE)));
2610 :
2611 330020 : if (len > PG_CACHE_LINE_SIZE)
2612 6 : hash ^= DatumGetUInt32(hash_uint32((uint32) len));
2613 :
2614 330020 : addHyperLogLog(&sss->full_card, hash);
2615 :
2616 : /* Hash abbreviated key */
2617 : #if SIZEOF_DATUM == 8
2618 : {
2619 : uint32 lohalf,
2620 : hihalf;
2621 :
2622 330020 : lohalf = (uint32) res;
2623 330020 : hihalf = (uint32) (res >> 32);
2624 330020 : hash = DatumGetUInt32(hash_uint32(lohalf ^ hihalf));
2625 : }
2626 : #else /* SIZEOF_DATUM != 8 */
2627 : hash = DatumGetUInt32(hash_uint32((uint32) res));
2628 : #endif
2629 :
2630 330020 : addHyperLogLog(&sss->abbr_card, hash);
2631 :
2632 : /* Cache result, perhaps saving an expensive strxfrm() call next time */
2633 330020 : sss->cache_blob = true;
2634 : done:
2635 :
2636 : /*
2637 : * Byteswap on little-endian machines.
2638 : *
2639 : * This is needed so that varstrcmp_abbrev() (an unsigned integer 3-way
2640 : * comparator) works correctly on all platforms. If we didn't do this,
2641 : * the comparator would have to call memcmp() with a pair of pointers to
2642 : * the first byte of each abbreviated key, which is slower.
2643 : */
2644 330020 : res = DatumBigEndianToNative(res);
2645 :
2646 : /* Don't leak memory here */
2647 330020 : if (PointerGetDatum(authoritative) != original)
2648 0 : pfree(authoritative);
2649 :
2650 330020 : return res;
2651 : }
2652 :
2653 : /*
2654 : * Callback for estimating effectiveness of abbreviated key optimization, using
2655 : * heuristic rules. Returns value indicating if the abbreviation optimization
2656 : * should be aborted, based on its projected effectiveness.
2657 : */
2658 : static bool
2659 892 : varstr_abbrev_abort(int memtupcount, SortSupport ssup)
2660 : {
2661 892 : VarStringSortSupport *sss = (VarStringSortSupport *) ssup->ssup_extra;
2662 : double abbrev_distinct,
2663 : key_distinct;
2664 :
2665 : Assert(ssup->abbreviate);
2666 :
2667 : /* Have a little patience */
2668 892 : if (memtupcount < 100)
2669 402 : return false;
2670 :
2671 490 : abbrev_distinct = estimateHyperLogLog(&sss->abbr_card);
2672 490 : key_distinct = estimateHyperLogLog(&sss->full_card);
2673 :
2674 : /*
2675 : * Clamp cardinality estimates to at least one distinct value. While
2676 : * NULLs are generally disregarded, if only NULL values were seen so far,
2677 : * that might misrepresent costs if we failed to clamp.
2678 : */
2679 490 : if (abbrev_distinct <= 1.0)
2680 0 : abbrev_distinct = 1.0;
2681 :
2682 490 : if (key_distinct <= 1.0)
2683 0 : key_distinct = 1.0;
2684 :
2685 : /*
2686 : * In the worst case all abbreviated keys are identical, while at the same
2687 : * time there are differences within full key strings not captured in
2688 : * abbreviations.
2689 : */
2690 : #ifdef TRACE_SORT
2691 490 : if (trace_sort)
2692 : {
2693 0 : double norm_abbrev_card = abbrev_distinct / (double) memtupcount;
2694 :
2695 0 : elog(LOG, "varstr_abbrev: abbrev_distinct after %d: %f "
2696 : "(key_distinct: %f, norm_abbrev_card: %f, prop_card: %f)",
2697 : memtupcount, abbrev_distinct, key_distinct, norm_abbrev_card,
2698 : sss->prop_card);
2699 : }
2700 : #endif
2701 :
2702 : /*
2703 : * If the number of distinct abbreviated keys approximately matches the
2704 : * number of distinct authoritative original keys, that's reason enough to
2705 : * proceed. We can win even with a very low cardinality set if most
2706 : * tie-breakers only memcmp(). This is by far the most important
2707 : * consideration.
2708 : *
2709 : * While comparisons that are resolved at the abbreviated key level are
2710 : * considerably cheaper than tie-breakers resolved with memcmp(), both of
2711 : * those two outcomes are so much cheaper than a full strcoll() once
2712 : * sorting is underway that it doesn't seem worth it to weigh abbreviated
2713 : * cardinality against the overall size of the set in order to more
2714 : * accurately model costs. Assume that an abbreviated comparison, and an
2715 : * abbreviated comparison with a cheap memcmp()-based authoritative
2716 : * resolution are equivalent.
2717 : */
2718 490 : if (abbrev_distinct > key_distinct * sss->prop_card)
2719 : {
2720 : /*
2721 : * When we have exceeded 10,000 tuples, decay required cardinality
2722 : * aggressively for next call.
2723 : *
2724 : * This is useful because the number of comparisons required on
2725 : * average increases at a linearithmic rate, and at roughly 10,000
2726 : * tuples that factor will start to dominate over the linear costs of
2727 : * string transformation (this is a conservative estimate). The decay
2728 : * rate is chosen to be a little less aggressive than halving -- which
2729 : * (since we're called at points at which memtupcount has doubled)
2730 : * would never see the cost model actually abort past the first call
2731 : * following a decay. This decay rate is mostly a precaution against
2732 : * a sudden, violent swing in how well abbreviated cardinality tracks
2733 : * full key cardinality. The decay also serves to prevent a marginal
2734 : * case from being aborted too late, when too much has already been
2735 : * invested in string transformation.
2736 : *
2737 : * It's possible for sets of several million distinct strings with
2738 : * mere tens of thousands of distinct abbreviated keys to still
2739 : * benefit very significantly. This will generally occur provided
2740 : * each abbreviated key is a proxy for a roughly uniform number of the
2741 : * set's full keys. If it isn't so, we hope to catch that early and
2742 : * abort. If it isn't caught early, by the time the problem is
2743 : * apparent it's probably not worth aborting.
2744 : */
2745 490 : if (memtupcount > 10000)
2746 0 : sss->prop_card *= 0.65;
2747 :
2748 490 : return false;
2749 : }
2750 :
2751 : /*
2752 : * Abort abbreviation strategy.
2753 : *
2754 : * The worst case, where all abbreviated keys are identical while all
2755 : * original strings differ will typically only see a regression of about
2756 : * 10% in execution time for small to medium sized lists of strings.
2757 : * Whereas on modern CPUs where cache stalls are the dominant cost, we can
2758 : * often expect very large improvements, particularly with sets of strings
2759 : * of moderately high to high abbreviated cardinality. There is little to
2760 : * lose but much to gain, which our strategy reflects.
2761 : */
2762 : #ifdef TRACE_SORT
2763 0 : if (trace_sort)
2764 0 : elog(LOG, "varstr_abbrev: aborted abbreviation at %d "
2765 : "(abbrev_distinct: %f, key_distinct: %f, prop_card: %f)",
2766 : memtupcount, abbrev_distinct, key_distinct, sss->prop_card);
2767 : #endif
2768 :
2769 0 : return true;
2770 : }
2771 :
2772 : Datum
2773 43082 : text_larger(PG_FUNCTION_ARGS)
2774 : {
2775 43082 : text *arg1 = PG_GETARG_TEXT_PP(0);
2776 43082 : text *arg2 = PG_GETARG_TEXT_PP(1);
2777 : text *result;
2778 :
2779 43082 : result = ((text_cmp(arg1, arg2, PG_GET_COLLATION()) > 0) ? arg1 : arg2);
2780 :
2781 43082 : PG_RETURN_TEXT_P(result);
2782 : }
2783 :
2784 : Datum
2785 41072 : text_smaller(PG_FUNCTION_ARGS)
2786 : {
2787 41072 : text *arg1 = PG_GETARG_TEXT_PP(0);
2788 41072 : text *arg2 = PG_GETARG_TEXT_PP(1);
2789 : text *result;
2790 :
2791 41072 : result = ((text_cmp(arg1, arg2, PG_GET_COLLATION()) < 0) ? arg1 : arg2);
2792 :
2793 41072 : PG_RETURN_TEXT_P(result);
2794 : }
2795 :
2796 :
2797 : /*
2798 : * Cross-type comparison functions for types text and name.
2799 : */
2800 :
2801 : Datum
2802 94664 : nameeqtext(PG_FUNCTION_ARGS)
2803 : {
2804 94664 : Name arg1 = PG_GETARG_NAME(0);
2805 94664 : text *arg2 = PG_GETARG_TEXT_PP(1);
2806 94664 : size_t len1 = strlen(NameStr(*arg1));
2807 94664 : size_t len2 = VARSIZE_ANY_EXHDR(arg2);
2808 94664 : Oid collid = PG_GET_COLLATION();
2809 : bool result;
2810 :
2811 94664 : check_collation_set(collid);
2812 :
2813 94664 : if (collid == C_COLLATION_OID)
2814 179146 : result = (len1 == len2 &&
2815 84482 : memcmp(NameStr(*arg1), VARDATA_ANY(arg2), len1) == 0);
2816 : else
2817 0 : result = (varstr_cmp(NameStr(*arg1), len1,
2818 0 : VARDATA_ANY(arg2), len2,
2819 : collid) == 0);
2820 :
2821 94664 : PG_FREE_IF_COPY(arg2, 1);
2822 :
2823 94664 : PG_RETURN_BOOL(result);
2824 : }
2825 :
2826 : Datum
2827 260 : texteqname(PG_FUNCTION_ARGS)
2828 : {
2829 260 : text *arg1 = PG_GETARG_TEXT_PP(0);
2830 260 : Name arg2 = PG_GETARG_NAME(1);
2831 260 : size_t len1 = VARSIZE_ANY_EXHDR(arg1);
2832 260 : size_t len2 = strlen(NameStr(*arg2));
2833 260 : Oid collid = PG_GET_COLLATION();
2834 : bool result;
2835 :
2836 260 : check_collation_set(collid);
2837 :
2838 260 : if (collid == C_COLLATION_OID)
2839 380 : result = (len1 == len2 &&
2840 120 : memcmp(VARDATA_ANY(arg1), NameStr(*arg2), len1) == 0);
2841 : else
2842 0 : result = (varstr_cmp(VARDATA_ANY(arg1), len1,
2843 0 : NameStr(*arg2), len2,
2844 : collid) == 0);
2845 :
2846 260 : PG_FREE_IF_COPY(arg1, 0);
2847 :
2848 260 : PG_RETURN_BOOL(result);
2849 : }
2850 :
2851 : Datum
2852 0 : namenetext(PG_FUNCTION_ARGS)
2853 : {
2854 0 : Name arg1 = PG_GETARG_NAME(0);
2855 0 : text *arg2 = PG_GETARG_TEXT_PP(1);
2856 0 : size_t len1 = strlen(NameStr(*arg1));
2857 0 : size_t len2 = VARSIZE_ANY_EXHDR(arg2);
2858 0 : Oid collid = PG_GET_COLLATION();
2859 : bool result;
2860 :
2861 0 : check_collation_set(collid);
2862 :
2863 0 : if (collid == C_COLLATION_OID)
2864 0 : result = !(len1 == len2 &&
2865 0 : memcmp(NameStr(*arg1), VARDATA_ANY(arg2), len1) == 0);
2866 : else
2867 0 : result = !(varstr_cmp(NameStr(*arg1), len1,
2868 0 : VARDATA_ANY(arg2), len2,
2869 : collid) == 0);
2870 :
2871 0 : PG_FREE_IF_COPY(arg2, 1);
2872 :
2873 0 : PG_RETURN_BOOL(result);
2874 : }
2875 :
2876 : Datum
2877 0 : textnename(PG_FUNCTION_ARGS)
2878 : {
2879 0 : text *arg1 = PG_GETARG_TEXT_PP(0);
2880 0 : Name arg2 = PG_GETARG_NAME(1);
2881 0 : size_t len1 = VARSIZE_ANY_EXHDR(arg1);
2882 0 : size_t len2 = strlen(NameStr(*arg2));
2883 0 : Oid collid = PG_GET_COLLATION();
2884 : bool result;
2885 :
2886 0 : check_collation_set(collid);
2887 :
2888 0 : if (collid == C_COLLATION_OID)
2889 0 : result = !(len1 == len2 &&
2890 0 : memcmp(VARDATA_ANY(arg1), NameStr(*arg2), len1) == 0);
2891 : else
2892 0 : result = !(varstr_cmp(VARDATA_ANY(arg1), len1,
2893 0 : NameStr(*arg2), len2,
2894 : collid) == 0);
2895 :
2896 0 : PG_FREE_IF_COPY(arg1, 0);
2897 :
2898 0 : PG_RETURN_BOOL(result);
2899 : }
2900 :
2901 : Datum
2902 67420 : btnametextcmp(PG_FUNCTION_ARGS)
2903 : {
2904 67420 : Name arg1 = PG_GETARG_NAME(0);
2905 67420 : text *arg2 = PG_GETARG_TEXT_PP(1);
2906 : int32 result;
2907 :
2908 337100 : result = varstr_cmp(NameStr(*arg1), strlen(NameStr(*arg1)),
2909 269680 : VARDATA_ANY(arg2), VARSIZE_ANY_EXHDR(arg2),
2910 : PG_GET_COLLATION());
2911 :
2912 67420 : PG_FREE_IF_COPY(arg2, 1);
2913 :
2914 67420 : PG_RETURN_INT32(result);
2915 : }
2916 :
2917 : Datum
2918 0 : bttextnamecmp(PG_FUNCTION_ARGS)
2919 : {
2920 0 : text *arg1 = PG_GETARG_TEXT_PP(0);
2921 0 : Name arg2 = PG_GETARG_NAME(1);
2922 : int32 result;
2923 :
2924 0 : result = varstr_cmp(VARDATA_ANY(arg1), VARSIZE_ANY_EXHDR(arg1),
2925 0 : NameStr(*arg2), strlen(NameStr(*arg2)),
2926 : PG_GET_COLLATION());
2927 :
2928 0 : PG_FREE_IF_COPY(arg1, 0);
2929 :
2930 0 : PG_RETURN_INT32(result);
2931 : }
2932 :
2933 : #define CmpCall(cmpfunc) \
2934 : DatumGetInt32(DirectFunctionCall2Coll(cmpfunc, \
2935 : PG_GET_COLLATION(), \
2936 : PG_GETARG_DATUM(0), \
2937 : PG_GETARG_DATUM(1)))
2938 :
2939 : Datum
2940 21342 : namelttext(PG_FUNCTION_ARGS)
2941 : {
2942 21342 : PG_RETURN_BOOL(CmpCall(btnametextcmp) < 0);
2943 : }
2944 :
2945 : Datum
2946 0 : nameletext(PG_FUNCTION_ARGS)
2947 : {
2948 0 : PG_RETURN_BOOL(CmpCall(btnametextcmp) <= 0);
2949 : }
2950 :
2951 : Datum
2952 0 : namegttext(PG_FUNCTION_ARGS)
2953 : {
2954 0 : PG_RETURN_BOOL(CmpCall(btnametextcmp) > 0);
2955 : }
2956 :
2957 : Datum
2958 20492 : namegetext(PG_FUNCTION_ARGS)
2959 : {
2960 20492 : PG_RETURN_BOOL(CmpCall(btnametextcmp) >= 0);
2961 : }
2962 :
2963 : Datum
2964 0 : textltname(PG_FUNCTION_ARGS)
2965 : {
2966 0 : PG_RETURN_BOOL(CmpCall(bttextnamecmp) < 0);
2967 : }
2968 :
2969 : Datum
2970 0 : textlename(PG_FUNCTION_ARGS)
2971 : {
2972 0 : PG_RETURN_BOOL(CmpCall(bttextnamecmp) <= 0);
2973 : }
2974 :
2975 : Datum
2976 0 : textgtname(PG_FUNCTION_ARGS)
2977 : {
2978 0 : PG_RETURN_BOOL(CmpCall(bttextnamecmp) > 0);
2979 : }
2980 :
2981 : Datum
2982 0 : textgename(PG_FUNCTION_ARGS)
2983 : {
2984 0 : PG_RETURN_BOOL(CmpCall(bttextnamecmp) >= 0);
2985 : }
2986 :
2987 : #undef CmpCall
2988 :
2989 :
2990 : /*
2991 : * The following operators support character-by-character comparison
2992 : * of text datums, to allow building indexes suitable for LIKE clauses.
2993 : * Note that the regular texteq/textne comparison operators, and regular
2994 : * support functions 1 and 2 with "C" collation are assumed to be
2995 : * compatible with these!
2996 : */
2997 :
2998 : static int
2999 100640 : internal_text_pattern_compare(text *arg1, text *arg2, Oid collid)
3000 : {
3001 : int result;
3002 : int len1,
3003 : len2;
3004 :
3005 100640 : check_collation_set(collid);
3006 :
3007 : /*
3008 : * XXX We cannot use a text_pattern_ops index for nondeterministic
3009 : * collations, because these operators intentionally ignore the collation.
3010 : * However, the planner has no way to know that, so it might choose such
3011 : * an index for an "=" clause, which would lead to wrong results. This
3012 : * check here doesn't prevent choosing the index, but it will at least
3013 : * error out if the index is chosen. A text_pattern_ops index on a column
3014 : * with nondeterministic collation is pretty useless anyway, since LIKE
3015 : * etc. won't work there either. A future possibility would be to
3016 : * annotate the operator class or its members in the catalog to avoid the
3017 : * index. Another alternative is to stay away from the *_pattern_ops
3018 : * operator classes and prefer creating LIKE-supporting indexes with
3019 : * COLLATE "C".
3020 : */
3021 100640 : if (!get_collation_isdeterministic(collid))
3022 0 : ereport(ERROR,
3023 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3024 : errmsg("nondeterministic collations are not supported for operator class \"%s\"",
3025 : "text_pattern_ops")));
3026 :
3027 100640 : len1 = VARSIZE_ANY_EXHDR(arg1);
3028 100640 : len2 = VARSIZE_ANY_EXHDR(arg2);
3029 :
3030 100640 : result = memcmp(VARDATA_ANY(arg1), VARDATA_ANY(arg2), Min(len1, len2));
3031 100640 : if (result != 0)
3032 100604 : return result;
3033 36 : else if (len1 < len2)
3034 0 : return -1;
3035 36 : else if (len1 > len2)
3036 12 : return 1;
3037 : else
3038 24 : return 0;
3039 : }
3040 :
3041 :
3042 : Datum
3043 25608 : text_pattern_lt(PG_FUNCTION_ARGS)
3044 : {
3045 25608 : text *arg1 = PG_GETARG_TEXT_PP(0);
3046 25608 : text *arg2 = PG_GETARG_TEXT_PP(1);
3047 : int result;
3048 :
3049 25608 : result = internal_text_pattern_compare(arg1, arg2, PG_GET_COLLATION());
3050 :
3051 25608 : PG_FREE_IF_COPY(arg1, 0);
3052 25608 : PG_FREE_IF_COPY(arg2, 1);
3053 :
3054 25608 : PG_RETURN_BOOL(result < 0);
3055 : }
3056 :
3057 :
3058 : Datum
3059 25008 : text_pattern_le(PG_FUNCTION_ARGS)
3060 : {
3061 25008 : text *arg1 = PG_GETARG_TEXT_PP(0);
3062 25008 : text *arg2 = PG_GETARG_TEXT_PP(1);
3063 : int result;
3064 :
3065 25008 : result = internal_text_pattern_compare(arg1, arg2, PG_GET_COLLATION());
3066 :
3067 25008 : PG_FREE_IF_COPY(arg1, 0);
3068 25008 : PG_FREE_IF_COPY(arg2, 1);
3069 :
3070 25008 : PG_RETURN_BOOL(result <= 0);
3071 : }
3072 :
3073 :
3074 : Datum
3075 25008 : text_pattern_ge(PG_FUNCTION_ARGS)
3076 : {
3077 25008 : text *arg1 = PG_GETARG_TEXT_PP(0);
3078 25008 : text *arg2 = PG_GETARG_TEXT_PP(1);
3079 : int result;
3080 :
3081 25008 : result = internal_text_pattern_compare(arg1, arg2, PG_GET_COLLATION());
3082 :
3083 25008 : PG_FREE_IF_COPY(arg1, 0);
3084 25008 : PG_FREE_IF_COPY(arg2, 1);
3085 :
3086 25008 : PG_RETURN_BOOL(result >= 0);
3087 : }
3088 :
3089 :
3090 : Datum
3091 25008 : text_pattern_gt(PG_FUNCTION_ARGS)
3092 : {
3093 25008 : text *arg1 = PG_GETARG_TEXT_PP(0);
3094 25008 : text *arg2 = PG_GETARG_TEXT_PP(1);
3095 : int result;
3096 :
3097 25008 : result = internal_text_pattern_compare(arg1, arg2, PG_GET_COLLATION());
3098 :
3099 25008 : PG_FREE_IF_COPY(arg1, 0);
3100 25008 : PG_FREE_IF_COPY(arg2, 1);
3101 :
3102 25008 : PG_RETURN_BOOL(result > 0);
3103 : }
3104 :
3105 :
3106 : Datum
3107 8 : bttext_pattern_cmp(PG_FUNCTION_ARGS)
3108 : {
3109 8 : text *arg1 = PG_GETARG_TEXT_PP(0);
3110 8 : text *arg2 = PG_GETARG_TEXT_PP(1);
3111 : int result;
3112 :
3113 8 : result = internal_text_pattern_compare(arg1, arg2, PG_GET_COLLATION());
3114 :
3115 8 : PG_FREE_IF_COPY(arg1, 0);
3116 8 : PG_FREE_IF_COPY(arg2, 1);
3117 :
3118 8 : PG_RETURN_INT32(result);
3119 : }
3120 :
3121 :
3122 : Datum
3123 78 : bttext_pattern_sortsupport(PG_FUNCTION_ARGS)
3124 : {
3125 78 : SortSupport ssup = (SortSupport) PG_GETARG_POINTER(0);
3126 78 : Oid collid = ssup->ssup_collation;
3127 : MemoryContext oldcontext;
3128 :
3129 78 : check_collation_set(collid);
3130 :
3131 78 : if (!get_collation_isdeterministic(collid))
3132 0 : ereport(ERROR,
3133 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3134 : errmsg("nondeterministic collations are not supported for operator class \"%s\"",
3135 : "text_pattern_ops")));
3136 :
3137 78 : oldcontext = MemoryContextSwitchTo(ssup->ssup_cxt);
3138 :
3139 : /* Use generic string SortSupport, forcing "C" collation */
3140 78 : varstr_sortsupport(ssup, TEXTOID, C_COLLATION_OID);
3141 :
3142 78 : MemoryContextSwitchTo(oldcontext);
3143 :
3144 78 : PG_RETURN_VOID();
3145 : }
3146 :
3147 :
3148 : /*-------------------------------------------------------------
3149 : * byteaoctetlen
3150 : *
3151 : * get the number of bytes contained in an instance of type 'bytea'
3152 : *-------------------------------------------------------------
3153 : */
3154 : Datum
3155 22 : byteaoctetlen(PG_FUNCTION_ARGS)
3156 : {
3157 22 : Datum str = PG_GETARG_DATUM(0);
3158 :
3159 : /* We need not detoast the input at all */
3160 22 : PG_RETURN_INT32(toast_raw_datum_size(str) - VARHDRSZ);
3161 : }
3162 :
3163 : /*
3164 : * byteacat -
3165 : * takes two bytea* and returns a bytea* that is the concatenation of
3166 : * the two.
3167 : *
3168 : * Cloned from textcat and modified as required.
3169 : */
3170 : Datum
3171 0 : byteacat(PG_FUNCTION_ARGS)
3172 : {
3173 0 : bytea *t1 = PG_GETARG_BYTEA_PP(0);
3174 0 : bytea *t2 = PG_GETARG_BYTEA_PP(1);
3175 :
3176 0 : PG_RETURN_BYTEA_P(bytea_catenate(t1, t2));
3177 : }
3178 :
3179 : /*
3180 : * bytea_catenate
3181 : * Guts of byteacat(), broken out so it can be used by other functions
3182 : *
3183 : * Arguments can be in short-header form, but not compressed or out-of-line
3184 : */
3185 : static bytea *
3186 24 : bytea_catenate(bytea *t1, bytea *t2)
3187 : {
3188 : bytea *result;
3189 : int len1,
3190 : len2,
3191 : len;
3192 : char *ptr;
3193 :
3194 24 : len1 = VARSIZE_ANY_EXHDR(t1);
3195 24 : len2 = VARSIZE_ANY_EXHDR(t2);
3196 :
3197 : /* paranoia ... probably should throw error instead? */
3198 24 : if (len1 < 0)
3199 0 : len1 = 0;
3200 24 : if (len2 < 0)
3201 0 : len2 = 0;
3202 :
3203 24 : len = len1 + len2 + VARHDRSZ;
3204 24 : result = (bytea *) palloc(len);
3205 :
3206 : /* Set size of result string... */
3207 24 : SET_VARSIZE(result, len);
3208 :
3209 : /* Fill data field of result string... */
3210 24 : ptr = VARDATA(result);
3211 24 : if (len1 > 0)
3212 24 : memcpy(ptr, VARDATA_ANY(t1), len1);
3213 24 : if (len2 > 0)
3214 12 : memcpy(ptr + len1, VARDATA_ANY(t2), len2);
3215 :
3216 24 : return result;
3217 : }
3218 :
3219 : #define PG_STR_GET_BYTEA(str_) \
3220 : DatumGetByteaPP(DirectFunctionCall1(byteain, CStringGetDatum(str_)))
3221 :
3222 : /*
3223 : * bytea_substr()
3224 : * Return a substring starting at the specified position.
3225 : * Cloned from text_substr and modified as required.
3226 : *
3227 : * Input:
3228 : * - string
3229 : * - starting position (is one-based)
3230 : * - string length (optional)
3231 : *
3232 : * If the starting position is zero or less, then return from the start of the string
3233 : * adjusting the length to be consistent with the "negative start" per SQL.
3234 : * If the length is less than zero, an ERROR is thrown. If no third argument
3235 : * (length) is provided, the length to the end of the string is assumed.
3236 : */
3237 : Datum
3238 36 : bytea_substr(PG_FUNCTION_ARGS)
3239 : {
3240 36 : PG_RETURN_BYTEA_P(bytea_substring(PG_GETARG_DATUM(0),
3241 : PG_GETARG_INT32(1),
3242 : PG_GETARG_INT32(2),
3243 : false));
3244 : }
3245 :
3246 : /*
3247 : * bytea_substr_no_len -
3248 : * Wrapper to avoid opr_sanity failure due to
3249 : * one function accepting a different number of args.
3250 : */
3251 : Datum
3252 16 : bytea_substr_no_len(PG_FUNCTION_ARGS)
3253 : {
3254 16 : PG_RETURN_BYTEA_P(bytea_substring(PG_GETARG_DATUM(0),
3255 : PG_GETARG_INT32(1),
3256 : -1,
3257 : true));
3258 : }
3259 :
3260 : static bytea *
3261 76 : bytea_substring(Datum str,
3262 : int S,
3263 : int L,
3264 : bool length_not_specified)
3265 : {
3266 : int S1; /* adjusted start position */
3267 : int L1; /* adjusted substring length */
3268 :
3269 76 : S1 = Max(S, 1);
3270 :
3271 76 : if (length_not_specified)
3272 : {
3273 : /*
3274 : * Not passed a length - DatumGetByteaPSlice() grabs everything to the
3275 : * end of the string if we pass it a negative value for length.
3276 : */
3277 28 : L1 = -1;
3278 : }
3279 : else
3280 : {
3281 : /* end position */
3282 48 : int E = S + L;
3283 :
3284 : /*
3285 : * A negative value for L is the only way for the end position to be
3286 : * before the start. SQL99 says to throw an error.
3287 : */
3288 48 : if (E < S)
3289 4 : ereport(ERROR,
3290 : (errcode(ERRCODE_SUBSTRING_ERROR),
3291 : errmsg("negative substring length not allowed")));
3292 :
3293 : /*
3294 : * A zero or negative value for the end position can happen if the
3295 : * start was negative or one. SQL99 says to return a zero-length
3296 : * string.
3297 : */
3298 44 : if (E < 1)
3299 0 : return PG_STR_GET_BYTEA("");
3300 :
3301 44 : L1 = E - S1;
3302 : }
3303 :
3304 : /*
3305 : * If the start position is past the end of the string, SQL99 says to
3306 : * return a zero-length string -- DatumGetByteaPSlice() will do that for
3307 : * us. Convert to zero-based starting position
3308 : */
3309 72 : return DatumGetByteaPSlice(str, S1 - 1, L1);
3310 : }
3311 :
3312 : /*
3313 : * byteaoverlay
3314 : * Replace specified substring of first string with second
3315 : *
3316 : * The SQL standard defines OVERLAY() in terms of substring and concatenation.
3317 : * This code is a direct implementation of what the standard says.
3318 : */
3319 : Datum
3320 4 : byteaoverlay(PG_FUNCTION_ARGS)
3321 : {
3322 4 : bytea *t1 = PG_GETARG_BYTEA_PP(0);
3323 4 : bytea *t2 = PG_GETARG_BYTEA_PP(1);
3324 4 : int sp = PG_GETARG_INT32(2); /* substring start position */
3325 4 : int sl = PG_GETARG_INT32(3); /* substring length */
3326 :
3327 4 : PG_RETURN_BYTEA_P(bytea_overlay(t1, t2, sp, sl));
3328 : }
3329 :
3330 : Datum
3331 8 : byteaoverlay_no_len(PG_FUNCTION_ARGS)
3332 : {
3333 8 : bytea *t1 = PG_GETARG_BYTEA_PP(0);
3334 8 : bytea *t2 = PG_GETARG_BYTEA_PP(1);
3335 8 : int sp = PG_GETARG_INT32(2); /* substring start position */
3336 : int sl;
3337 :
3338 8 : sl = VARSIZE_ANY_EXHDR(t2); /* defaults to length(t2) */
3339 8 : PG_RETURN_BYTEA_P(bytea_overlay(t1, t2, sp, sl));
3340 : }
3341 :
3342 : static bytea *
3343 12 : bytea_overlay(bytea *t1, bytea *t2, int sp, int sl)
3344 : {
3345 : bytea *result;
3346 : bytea *s1;
3347 : bytea *s2;
3348 : int sp_pl_sl;
3349 :
3350 : /*
3351 : * Check for possible integer-overflow cases. For negative sp, throw a
3352 : * "substring length" error because that's what should be expected
3353 : * according to the spec's definition of OVERLAY().
3354 : */
3355 12 : if (sp <= 0)
3356 0 : ereport(ERROR,
3357 : (errcode(ERRCODE_SUBSTRING_ERROR),
3358 : errmsg("negative substring length not allowed")));
3359 12 : if (pg_add_s32_overflow(sp, sl, &sp_pl_sl))
3360 0 : ereport(ERROR,
3361 : (errcode(ERRCODE_NUMERIC_VALUE_OUT_OF_RANGE),
3362 : errmsg("integer out of range")));
3363 :
3364 12 : s1 = bytea_substring(PointerGetDatum(t1), 1, sp - 1, false);
3365 12 : s2 = bytea_substring(PointerGetDatum(t1), sp_pl_sl, -1, true);
3366 12 : result = bytea_catenate(s1, t2);
3367 12 : result = bytea_catenate(result, s2);
3368 :
3369 12 : return result;
3370 : }
3371 :
3372 : /*
3373 : * byteapos -
3374 : * Return the position of the specified substring.
3375 : * Implements the SQL POSITION() function.
3376 : * Cloned from textpos and modified as required.
3377 : */
3378 : Datum
3379 0 : byteapos(PG_FUNCTION_ARGS)
3380 : {
3381 0 : bytea *t1 = PG_GETARG_BYTEA_PP(0);
3382 0 : bytea *t2 = PG_GETARG_BYTEA_PP(1);
3383 : int pos;
3384 : int px,
3385 : p;
3386 : int len1,
3387 : len2;
3388 : char *p1,
3389 : *p2;
3390 :
3391 0 : len1 = VARSIZE_ANY_EXHDR(t1);
3392 0 : len2 = VARSIZE_ANY_EXHDR(t2);
3393 :
3394 0 : if (len2 <= 0)
3395 0 : PG_RETURN_INT32(1); /* result for empty pattern */
3396 :
3397 0 : p1 = VARDATA_ANY(t1);
3398 0 : p2 = VARDATA_ANY(t2);
3399 :
3400 0 : pos = 0;
3401 0 : px = (len1 - len2);
3402 0 : for (p = 0; p <= px; p++)
3403 : {
3404 0 : if ((*p2 == *p1) && (memcmp(p1, p2, len2) == 0))
3405 : {
3406 0 : pos = p + 1;
3407 0 : break;
3408 : };
3409 0 : p1++;
3410 : };
3411 :
3412 0 : PG_RETURN_INT32(pos);
3413 : }
3414 :
3415 : /*-------------------------------------------------------------
3416 : * byteaGetByte
3417 : *
3418 : * this routine treats "bytea" as an array of bytes.
3419 : * It returns the Nth byte (a number between 0 and 255).
3420 : *-------------------------------------------------------------
3421 : */
3422 : Datum
3423 0 : byteaGetByte(PG_FUNCTION_ARGS)
3424 : {
3425 0 : bytea *v = PG_GETARG_BYTEA_PP(0);
3426 0 : int32 n = PG_GETARG_INT32(1);
3427 : int len;
3428 : int byte;
3429 :
3430 0 : len = VARSIZE_ANY_EXHDR(v);
3431 :
3432 0 : if (n < 0 || n >= len)
3433 0 : ereport(ERROR,
3434 : (errcode(ERRCODE_ARRAY_SUBSCRIPT_ERROR),
3435 : errmsg("index %d out of valid range, 0..%d",
3436 : n, len - 1)));
3437 :
3438 0 : byte = ((unsigned char *) VARDATA_ANY(v))[n];
3439 :
3440 0 : PG_RETURN_INT32(byte);
3441 : }
3442 :
3443 : /*-------------------------------------------------------------
3444 : * byteaGetBit
3445 : *
3446 : * This routine treats a "bytea" type like an array of bits.
3447 : * It returns the value of the Nth bit (0 or 1).
3448 : *
3449 : *-------------------------------------------------------------
3450 : */
3451 : Datum
3452 0 : byteaGetBit(PG_FUNCTION_ARGS)
3453 : {
3454 0 : bytea *v = PG_GETARG_BYTEA_PP(0);
3455 0 : int32 n = PG_GETARG_INT32(1);
3456 : int byteNo,
3457 : bitNo;
3458 : int len;
3459 : int byte;
3460 :
3461 0 : len = VARSIZE_ANY_EXHDR(v);
3462 :
3463 0 : if (n < 0 || n >= len * 8)
3464 0 : ereport(ERROR,
3465 : (errcode(ERRCODE_ARRAY_SUBSCRIPT_ERROR),
3466 : errmsg("index %d out of valid range, 0..%d",
3467 : n, len * 8 - 1)));
3468 :
3469 0 : byteNo = n / 8;
3470 0 : bitNo = n % 8;
3471 :
3472 0 : byte = ((unsigned char *) VARDATA_ANY(v))[byteNo];
3473 :
3474 0 : if (byte & (1 << bitNo))
3475 0 : PG_RETURN_INT32(1);
3476 : else
3477 0 : PG_RETURN_INT32(0);
3478 : }
3479 :
3480 : /*-------------------------------------------------------------
3481 : * byteaSetByte
3482 : *
3483 : * Given an instance of type 'bytea' creates a new one with
3484 : * the Nth byte set to the given value.
3485 : *
3486 : *-------------------------------------------------------------
3487 : */
3488 : Datum
3489 0 : byteaSetByte(PG_FUNCTION_ARGS)
3490 : {
3491 0 : bytea *res = PG_GETARG_BYTEA_P_COPY(0);
3492 0 : int32 n = PG_GETARG_INT32(1);
3493 0 : int32 newByte = PG_GETARG_INT32(2);
3494 : int len;
3495 :
3496 0 : len = VARSIZE(res) - VARHDRSZ;
3497 :
3498 0 : if (n < 0 || n >= len)
3499 0 : ereport(ERROR,
3500 : (errcode(ERRCODE_ARRAY_SUBSCRIPT_ERROR),
3501 : errmsg("index %d out of valid range, 0..%d",
3502 : n, len - 1)));
3503 :
3504 : /*
3505 : * Now set the byte.
3506 : */
3507 0 : ((unsigned char *) VARDATA(res))[n] = newByte;
3508 :
3509 0 : PG_RETURN_BYTEA_P(res);
3510 : }
3511 :
3512 : /*-------------------------------------------------------------
3513 : * byteaSetBit
3514 : *
3515 : * Given an instance of type 'bytea' creates a new one with
3516 : * the Nth bit set to the given value.
3517 : *
3518 : *-------------------------------------------------------------
3519 : */
3520 : Datum
3521 0 : byteaSetBit(PG_FUNCTION_ARGS)
3522 : {
3523 0 : bytea *res = PG_GETARG_BYTEA_P_COPY(0);
3524 0 : int32 n = PG_GETARG_INT32(1);
3525 0 : int32 newBit = PG_GETARG_INT32(2);
3526 : int len;
3527 : int oldByte,
3528 : newByte;
3529 : int byteNo,
3530 : bitNo;
3531 :
3532 0 : len = VARSIZE(res) - VARHDRSZ;
3533 :
3534 0 : if (n < 0 || n >= len * 8)
3535 0 : ereport(ERROR,
3536 : (errcode(ERRCODE_ARRAY_SUBSCRIPT_ERROR),
3537 : errmsg("index %d out of valid range, 0..%d",
3538 : n, len * 8 - 1)));
3539 :
3540 0 : byteNo = n / 8;
3541 0 : bitNo = n % 8;
3542 :
3543 : /*
3544 : * sanity check!
3545 : */
3546 0 : if (newBit != 0 && newBit != 1)
3547 0 : ereport(ERROR,
3548 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
3549 : errmsg("new bit must be 0 or 1")));
3550 :
3551 : /*
3552 : * Update the byte.
3553 : */
3554 0 : oldByte = ((unsigned char *) VARDATA(res))[byteNo];
3555 :
3556 0 : if (newBit == 0)
3557 0 : newByte = oldByte & (~(1 << bitNo));
3558 : else
3559 0 : newByte = oldByte | (1 << bitNo);
3560 :
3561 0 : ((unsigned char *) VARDATA(res))[byteNo] = newByte;
3562 :
3563 0 : PG_RETURN_BYTEA_P(res);
3564 : }
3565 :
3566 :
3567 : /* text_name()
3568 : * Converts a text type to a Name type.
3569 : */
3570 : Datum
3571 826 : text_name(PG_FUNCTION_ARGS)
3572 : {
3573 826 : text *s = PG_GETARG_TEXT_PP(0);
3574 : Name result;
3575 : int len;
3576 :
3577 826 : len = VARSIZE_ANY_EXHDR(s);
3578 :
3579 : /* Truncate oversize input */
3580 826 : if (len >= NAMEDATALEN)
3581 4 : len = pg_mbcliplen(VARDATA_ANY(s), len, NAMEDATALEN - 1);
3582 :
3583 : /* We use palloc0 here to ensure result is zero-padded */
3584 826 : result = (Name) palloc0(NAMEDATALEN);
3585 826 : memcpy(NameStr(*result), VARDATA_ANY(s), len);
3586 :
3587 826 : PG_RETURN_NAME(result);
3588 : }
3589 :
3590 : /* name_text()
3591 : * Converts a Name type to a text type.
3592 : */
3593 : Datum
3594 488336 : name_text(PG_FUNCTION_ARGS)
3595 : {
3596 488336 : Name s = PG_GETARG_NAME(0);
3597 :
3598 488336 : PG_RETURN_TEXT_P(cstring_to_text(NameStr(*s)));
3599 : }
3600 :
3601 :
3602 : /*
3603 : * textToQualifiedNameList - convert a text object to list of names
3604 : *
3605 : * This implements the input parsing needed by nextval() and other
3606 : * functions that take a text parameter representing a qualified name.
3607 : * We split the name at dots, downcase if not double-quoted, and
3608 : * truncate names if they're too long.
3609 : */
3610 : List *
3611 788 : textToQualifiedNameList(text *textval)
3612 : {
3613 : char *rawname;
3614 788 : List *result = NIL;
3615 : List *namelist;
3616 : ListCell *l;
3617 :
3618 : /* Convert to C string (handles possible detoasting). */
3619 : /* Note we rely on being able to modify rawname below. */
3620 788 : rawname = text_to_cstring(textval);
3621 :
3622 788 : if (!SplitIdentifierString(rawname, '.', &namelist))
3623 0 : ereport(ERROR,
3624 : (errcode(ERRCODE_INVALID_NAME),
3625 : errmsg("invalid name syntax")));
3626 :
3627 788 : if (namelist == NIL)
3628 0 : ereport(ERROR,
3629 : (errcode(ERRCODE_INVALID_NAME),
3630 : errmsg("invalid name syntax")));
3631 :
3632 1652 : foreach(l, namelist)
3633 : {
3634 864 : char *curname = (char *) lfirst(l);
3635 :
3636 864 : result = lappend(result, makeString(pstrdup(curname)));
3637 : }
3638 :
3639 788 : pfree(rawname);
3640 788 : list_free(namelist);
3641 :
3642 788 : return result;
3643 : }
3644 :
3645 : /*
3646 : * SplitIdentifierString --- parse a string containing identifiers
3647 : *
3648 : * This is the guts of textToQualifiedNameList, and is exported for use in
3649 : * other situations such as parsing GUC variables. In the GUC case, it's
3650 : * important to avoid memory leaks, so the API is designed to minimize the
3651 : * amount of stuff that needs to be allocated and freed.
3652 : *
3653 : * Inputs:
3654 : * rawstring: the input string; must be overwritable! On return, it's
3655 : * been modified to contain the separated identifiers.
3656 : * separator: the separator punctuation expected between identifiers
3657 : * (typically '.' or ','). Whitespace may also appear around
3658 : * identifiers.
3659 : * Outputs:
3660 : * namelist: filled with a palloc'd list of pointers to identifiers within
3661 : * rawstring. Caller should list_free() this even on error return.
3662 : *
3663 : * Returns true if okay, false if there is a syntax error in the string.
3664 : *
3665 : * Note that an empty string is considered okay here, though not in
3666 : * textToQualifiedNameList.
3667 : */
3668 : bool
3669 70198 : SplitIdentifierString(char *rawstring, char separator,
3670 : List **namelist)
3671 : {
3672 70198 : char *nextp = rawstring;
3673 70198 : bool done = false;
3674 :
3675 70198 : *namelist = NIL;
3676 :
3677 140396 : while (scanner_isspace(*nextp))
3678 0 : nextp++; /* skip leading whitespace */
3679 :
3680 70198 : if (*nextp == '\0')
3681 8640 : return true; /* allow empty string */
3682 :
3683 : /* At the top of the loop, we are at start of a new identifier. */
3684 : do
3685 : {
3686 : char *curname;
3687 : char *endp;
3688 :
3689 97472 : if (*nextp == '"')
3690 : {
3691 : /* Quoted name --- collapse quote-quote pairs, no downcasing */
3692 13406 : curname = nextp + 1;
3693 : for (;;)
3694 : {
3695 13410 : endp = strchr(nextp + 1, '"');
3696 13408 : if (endp == NULL)
3697 0 : return false; /* mismatched quotes */
3698 13408 : if (endp[1] != '"')
3699 13406 : break; /* found end of quoted name */
3700 : /* Collapse adjacent quotes into one quote, and look again */
3701 2 : memmove(endp, endp + 1, strlen(endp));
3702 2 : nextp = endp;
3703 : }
3704 : /* endp now points at the terminating quote */
3705 13406 : nextp = endp + 1;
3706 : }
3707 : else
3708 : {
3709 : /* Unquoted name --- extends to separator or whitespace */
3710 : char *downname;
3711 : int len;
3712 :
3713 84066 : curname = nextp;
3714 1442310 : while (*nextp && *nextp != separator &&
3715 637090 : !scanner_isspace(*nextp))
3716 637088 : nextp++;
3717 84066 : endp = nextp;
3718 84066 : if (curname == nextp)
3719 0 : return false; /* empty unquoted name not allowed */
3720 :
3721 : /*
3722 : * Downcase the identifier, using same code as main lexer does.
3723 : *
3724 : * XXX because we want to overwrite the input in-place, we cannot
3725 : * support a downcasing transformation that increases the string
3726 : * length. This is not a problem given the current implementation
3727 : * of downcase_truncate_identifier, but we'll probably have to do
3728 : * something about this someday.
3729 : */
3730 84066 : len = endp - curname;
3731 84066 : downname = downcase_truncate_identifier(curname, len, false);
3732 : Assert(strlen(downname) <= len);
3733 84066 : strncpy(curname, downname, len); /* strncpy is required here */
3734 84066 : pfree(downname);
3735 : }
3736 :
3737 194946 : while (scanner_isspace(*nextp))
3738 2 : nextp++; /* skip trailing whitespace */
3739 :
3740 97472 : if (*nextp == separator)
3741 : {
3742 35914 : nextp++;
3743 91630 : while (scanner_isspace(*nextp))
3744 19802 : nextp++; /* skip leading whitespace for next */
3745 : /* we expect another name, so done remains false */
3746 : }
3747 61558 : else if (*nextp == '\0')
3748 61556 : done = true;
3749 : else
3750 2 : return false; /* invalid syntax */
3751 :
3752 : /* Now safe to overwrite separator with a null */
3753 97470 : *endp = '\0';
3754 :
3755 : /* Truncate name if it's overlength */
3756 97470 : truncate_identifier(curname, strlen(curname), false);
3757 :
3758 : /*
3759 : * Finished isolating current name --- add it to list
3760 : */
3761 97470 : *namelist = lappend(*namelist, curname);
3762 :
3763 : /* Loop back if we didn't reach end of string */
3764 97470 : } while (!done);
3765 :
3766 61556 : return true;
3767 : }
3768 :
3769 :
3770 : /*
3771 : * SplitDirectoriesString --- parse a string containing file/directory names
3772 : *
3773 : * This works fine on file names too; the function name is historical.
3774 : *
3775 : * This is similar to SplitIdentifierString, except that the parsing
3776 : * rules are meant to handle pathnames instead of identifiers: there is
3777 : * no downcasing, embedded spaces are allowed, the max length is MAXPGPATH-1,
3778 : * and we apply canonicalize_path() to each extracted string. Because of the
3779 : * last, the returned strings are separately palloc'd rather than being
3780 : * pointers into rawstring --- but we still scribble on rawstring.
3781 : *
3782 : * Inputs:
3783 : * rawstring: the input string; must be modifiable!
3784 : * separator: the separator punctuation expected between directories
3785 : * (typically ',' or ';'). Whitespace may also appear around
3786 : * directories.
3787 : * Outputs:
3788 : * namelist: filled with a palloc'd list of directory names.
3789 : * Caller should list_free_deep() this even on error return.
3790 : *
3791 : * Returns true if okay, false if there is a syntax error in the string.
3792 : *
3793 : * Note that an empty string is considered okay here.
3794 : */
3795 : bool
3796 580 : SplitDirectoriesString(char *rawstring, char separator,
3797 : List **namelist)
3798 : {
3799 580 : char *nextp = rawstring;
3800 580 : bool done = false;
3801 :
3802 580 : *namelist = NIL;
3803 :
3804 1160 : while (scanner_isspace(*nextp))
3805 0 : nextp++; /* skip leading whitespace */
3806 :
3807 580 : if (*nextp == '\0')
3808 2 : return true; /* allow empty string */
3809 :
3810 : /* At the top of the loop, we are at start of a new directory. */
3811 : do
3812 : {
3813 : char *curname;
3814 : char *endp;
3815 :
3816 578 : if (*nextp == '"')
3817 : {
3818 : /* Quoted name --- collapse quote-quote pairs */
3819 0 : curname = nextp + 1;
3820 : for (;;)
3821 : {
3822 0 : endp = strchr(nextp + 1, '"');
3823 0 : if (endp == NULL)
3824 0 : return false; /* mismatched quotes */
3825 0 : if (endp[1] != '"')
3826 0 : break; /* found end of quoted name */
3827 : /* Collapse adjacent quotes into one quote, and look again */
3828 0 : memmove(endp, endp + 1, strlen(endp));
3829 0 : nextp = endp;
3830 : }
3831 : /* endp now points at the terminating quote */
3832 0 : nextp = endp + 1;
3833 : }
3834 : else
3835 : {
3836 : /* Unquoted name --- extends to separator or end of string */
3837 578 : curname = endp = nextp;
3838 11096 : while (*nextp && *nextp != separator)
3839 : {
3840 : /* trailing whitespace should not be included in name */
3841 9940 : if (!scanner_isspace(*nextp))
3842 9940 : endp = nextp + 1;
3843 9940 : nextp++;
3844 : }
3845 578 : if (curname == endp)
3846 0 : return false; /* empty unquoted name not allowed */
3847 : }
3848 :
3849 1156 : while (scanner_isspace(*nextp))
3850 0 : nextp++; /* skip trailing whitespace */
3851 :
3852 578 : if (*nextp == separator)
3853 : {
3854 0 : nextp++;
3855 0 : while (scanner_isspace(*nextp))
3856 0 : nextp++; /* skip leading whitespace for next */
3857 : /* we expect another name, so done remains false */
3858 : }
3859 578 : else if (*nextp == '\0')
3860 578 : done = true;
3861 : else
3862 0 : return false; /* invalid syntax */
3863 :
3864 : /* Now safe to overwrite separator with a null */
3865 578 : *endp = '\0';
3866 :
3867 : /* Truncate path if it's overlength */
3868 578 : if (strlen(curname) >= MAXPGPATH)
3869 0 : curname[MAXPGPATH - 1] = '\0';
3870 :
3871 : /*
3872 : * Finished isolating current name --- add it to list
3873 : */
3874 578 : curname = pstrdup(curname);
3875 578 : canonicalize_path(curname);
3876 578 : *namelist = lappend(*namelist, curname);
3877 :
3878 : /* Loop back if we didn't reach end of string */
3879 578 : } while (!done);
3880 :
3881 578 : return true;
3882 : }
3883 :
3884 :
3885 : /*
3886 : * SplitGUCList --- parse a string containing identifiers or file names
3887 : *
3888 : * This is used to split the value of a GUC_LIST_QUOTE GUC variable, without
3889 : * presuming whether the elements will be taken as identifiers or file names.
3890 : * We assume the input has already been through flatten_set_variable_args(),
3891 : * so that we need never downcase (if appropriate, that was done already).
3892 : * Nor do we ever truncate, since we don't know the correct max length.
3893 : * We disallow embedded whitespace for simplicity (it shouldn't matter,
3894 : * because any embedded whitespace should have led to double-quoting).
3895 : * Otherwise the API is identical to SplitIdentifierString.
3896 : *
3897 : * XXX it's annoying to have so many copies of this string-splitting logic.
3898 : * However, it's not clear that having one function with a bunch of option
3899 : * flags would be much better.
3900 : *
3901 : * XXX there is a version of this function in src/bin/pg_dump/dumputils.c.
3902 : * Be sure to update that if you have to change this.
3903 : *
3904 : * Inputs:
3905 : * rawstring: the input string; must be overwritable! On return, it's
3906 : * been modified to contain the separated identifiers.
3907 : * separator: the separator punctuation expected between identifiers
3908 : * (typically '.' or ','). Whitespace may also appear around
3909 : * identifiers.
3910 : * Outputs:
3911 : * namelist: filled with a palloc'd list of pointers to identifiers within
3912 : * rawstring. Caller should list_free() this even on error return.
3913 : *
3914 : * Returns true if okay, false if there is a syntax error in the string.
3915 : */
3916 : bool
3917 8 : SplitGUCList(char *rawstring, char separator,
3918 : List **namelist)
3919 : {
3920 8 : char *nextp = rawstring;
3921 8 : bool done = false;
3922 :
3923 8 : *namelist = NIL;
3924 :
3925 16 : while (scanner_isspace(*nextp))
3926 0 : nextp++; /* skip leading whitespace */
3927 :
3928 8 : if (*nextp == '\0')
3929 0 : return true; /* allow empty string */
3930 :
3931 : /* At the top of the loop, we are at start of a new identifier. */
3932 : do
3933 : {
3934 : char *curname;
3935 : char *endp;
3936 :
3937 20 : if (*nextp == '"')
3938 : {
3939 : /* Quoted name --- collapse quote-quote pairs */
3940 16 : curname = nextp + 1;
3941 : for (;;)
3942 : {
3943 24 : endp = strchr(nextp + 1, '"');
3944 20 : if (endp == NULL)
3945 0 : return false; /* mismatched quotes */
3946 20 : if (endp[1] != '"')
3947 16 : break; /* found end of quoted name */
3948 : /* Collapse adjacent quotes into one quote, and look again */
3949 4 : memmove(endp, endp + 1, strlen(endp));
3950 4 : nextp = endp;
3951 : }
3952 : /* endp now points at the terminating quote */
3953 16 : nextp = endp + 1;
3954 : }
3955 : else
3956 : {
3957 : /* Unquoted name --- extends to separator or whitespace */
3958 4 : curname = nextp;
3959 88 : while (*nextp && *nextp != separator &&
3960 40 : !scanner_isspace(*nextp))
3961 40 : nextp++;
3962 4 : endp = nextp;
3963 4 : if (curname == nextp)
3964 0 : return false; /* empty unquoted name not allowed */
3965 : }
3966 :
3967 40 : while (scanner_isspace(*nextp))
3968 0 : nextp++; /* skip trailing whitespace */
3969 :
3970 20 : if (*nextp == separator)
3971 : {
3972 12 : nextp++;
3973 36 : while (scanner_isspace(*nextp))
3974 12 : nextp++; /* skip leading whitespace for next */
3975 : /* we expect another name, so done remains false */
3976 : }
3977 8 : else if (*nextp == '\0')
3978 8 : done = true;
3979 : else
3980 0 : return false; /* invalid syntax */
3981 :
3982 : /* Now safe to overwrite separator with a null */
3983 20 : *endp = '\0';
3984 :
3985 : /*
3986 : * Finished isolating current name --- add it to list
3987 : */
3988 20 : *namelist = lappend(*namelist, curname);
3989 :
3990 : /* Loop back if we didn't reach end of string */
3991 20 : } while (!done);
3992 :
3993 8 : return true;
3994 : }
3995 :
3996 :
3997 : /*****************************************************************************
3998 : * Comparison Functions used for bytea
3999 : *
4000 : * Note: btree indexes need these routines not to leak memory; therefore,
4001 : * be careful to free working copies of toasted datums. Most places don't
4002 : * need to be so careful.
4003 : *****************************************************************************/
4004 :
4005 : Datum
4006 7954 : byteaeq(PG_FUNCTION_ARGS)
4007 : {
4008 7954 : Datum arg1 = PG_GETARG_DATUM(0);
4009 7954 : Datum arg2 = PG_GETARG_DATUM(1);
4010 : bool result;
4011 : Size len1,
4012 : len2;
4013 :
4014 : /*
4015 : * We can use a fast path for unequal lengths, which might save us from
4016 : * having to detoast one or both values.
4017 : */
4018 7954 : len1 = toast_raw_datum_size(arg1);
4019 7954 : len2 = toast_raw_datum_size(arg2);
4020 7954 : if (len1 != len2)
4021 4304 : result = false;
4022 : else
4023 : {
4024 3650 : bytea *barg1 = DatumGetByteaPP(arg1);
4025 3650 : bytea *barg2 = DatumGetByteaPP(arg2);
4026 :
4027 3650 : result = (memcmp(VARDATA_ANY(barg1), VARDATA_ANY(barg2),
4028 : len1 - VARHDRSZ) == 0);
4029 :
4030 3650 : PG_FREE_IF_COPY(barg1, 0);
4031 3650 : PG_FREE_IF_COPY(barg2, 1);
4032 : }
4033 :
4034 7954 : PG_RETURN_BOOL(result);
4035 : }
4036 :
4037 : Datum
4038 528 : byteane(PG_FUNCTION_ARGS)
4039 : {
4040 528 : Datum arg1 = PG_GETARG_DATUM(0);
4041 528 : Datum arg2 = PG_GETARG_DATUM(1);
4042 : bool result;
4043 : Size len1,
4044 : len2;
4045 :
4046 : /*
4047 : * We can use a fast path for unequal lengths, which might save us from
4048 : * having to detoast one or both values.
4049 : */
4050 528 : len1 = toast_raw_datum_size(arg1);
4051 528 : len2 = toast_raw_datum_size(arg2);
4052 528 : if (len1 != len2)
4053 0 : result = true;
4054 : else
4055 : {
4056 528 : bytea *barg1 = DatumGetByteaPP(arg1);
4057 528 : bytea *barg2 = DatumGetByteaPP(arg2);
4058 :
4059 528 : result = (memcmp(VARDATA_ANY(barg1), VARDATA_ANY(barg2),
4060 : len1 - VARHDRSZ) != 0);
4061 :
4062 528 : PG_FREE_IF_COPY(barg1, 0);
4063 528 : PG_FREE_IF_COPY(barg2, 1);
4064 : }
4065 :
4066 528 : PG_RETURN_BOOL(result);
4067 : }
4068 :
4069 : Datum
4070 7300 : bytealt(PG_FUNCTION_ARGS)
4071 : {
4072 7300 : bytea *arg1 = PG_GETARG_BYTEA_PP(0);
4073 7300 : bytea *arg2 = PG_GETARG_BYTEA_PP(1);
4074 : int len1,
4075 : len2;
4076 : int cmp;
4077 :
4078 7300 : len1 = VARSIZE_ANY_EXHDR(arg1);
4079 7300 : len2 = VARSIZE_ANY_EXHDR(arg2);
4080 :
4081 7300 : cmp = memcmp(VARDATA_ANY(arg1), VARDATA_ANY(arg2), Min(len1, len2));
4082 :
4083 7300 : PG_FREE_IF_COPY(arg1, 0);
4084 7300 : PG_FREE_IF_COPY(arg2, 1);
4085 :
4086 7300 : PG_RETURN_BOOL((cmp < 0) || ((cmp == 0) && (len1 < len2)));
4087 : }
4088 :
4089 : Datum
4090 5556 : byteale(PG_FUNCTION_ARGS)
4091 : {
4092 5556 : bytea *arg1 = PG_GETARG_BYTEA_PP(0);
4093 5556 : bytea *arg2 = PG_GETARG_BYTEA_PP(1);
4094 : int len1,
4095 : len2;
4096 : int cmp;
4097 :
4098 5556 : len1 = VARSIZE_ANY_EXHDR(arg1);
4099 5556 : len2 = VARSIZE_ANY_EXHDR(arg2);
4100 :
4101 5556 : cmp = memcmp(VARDATA_ANY(arg1), VARDATA_ANY(arg2), Min(len1, len2));
4102 :
4103 5556 : PG_FREE_IF_COPY(arg1, 0);
4104 5556 : PG_FREE_IF_COPY(arg2, 1);
4105 :
4106 5556 : PG_RETURN_BOOL((cmp < 0) || ((cmp == 0) && (len1 <= len2)));
4107 : }
4108 :
4109 : Datum
4110 5214 : byteagt(PG_FUNCTION_ARGS)
4111 : {
4112 5214 : bytea *arg1 = PG_GETARG_BYTEA_PP(0);
4113 5214 : bytea *arg2 = PG_GETARG_BYTEA_PP(1);
4114 : int len1,
4115 : len2;
4116 : int cmp;
4117 :
4118 5214 : len1 = VARSIZE_ANY_EXHDR(arg1);
4119 5214 : len2 = VARSIZE_ANY_EXHDR(arg2);
4120 :
4121 5214 : cmp = memcmp(VARDATA_ANY(arg1), VARDATA_ANY(arg2), Min(len1, len2));
4122 :
4123 5214 : PG_FREE_IF_COPY(arg1, 0);
4124 5214 : PG_FREE_IF_COPY(arg2, 1);
4125 :
4126 5214 : PG_RETURN_BOOL((cmp > 0) || ((cmp == 0) && (len1 > len2)));
4127 : }
4128 :
4129 : Datum
4130 4394 : byteage(PG_FUNCTION_ARGS)
4131 : {
4132 4394 : bytea *arg1 = PG_GETARG_BYTEA_PP(0);
4133 4394 : bytea *arg2 = PG_GETARG_BYTEA_PP(1);
4134 : int len1,
4135 : len2;
4136 : int cmp;
4137 :
4138 4394 : len1 = VARSIZE_ANY_EXHDR(arg1);
4139 4394 : len2 = VARSIZE_ANY_EXHDR(arg2);
4140 :
4141 4394 : cmp = memcmp(VARDATA_ANY(arg1), VARDATA_ANY(arg2), Min(len1, len2));
4142 :
4143 4394 : PG_FREE_IF_COPY(arg1, 0);
4144 4394 : PG_FREE_IF_COPY(arg2, 1);
4145 :
4146 4394 : PG_RETURN_BOOL((cmp > 0) || ((cmp == 0) && (len1 >= len2)));
4147 : }
4148 :
4149 : Datum
4150 87402 : byteacmp(PG_FUNCTION_ARGS)
4151 : {
4152 87402 : bytea *arg1 = PG_GETARG_BYTEA_PP(0);
4153 87402 : bytea *arg2 = PG_GETARG_BYTEA_PP(1);
4154 : int len1,
4155 : len2;
4156 : int cmp;
4157 :
4158 87402 : len1 = VARSIZE_ANY_EXHDR(arg1);
4159 87402 : len2 = VARSIZE_ANY_EXHDR(arg2);
4160 :
4161 87402 : cmp = memcmp(VARDATA_ANY(arg1), VARDATA_ANY(arg2), Min(len1, len2));
4162 87402 : if ((cmp == 0) && (len1 != len2))
4163 14500 : cmp = (len1 < len2) ? -1 : 1;
4164 :
4165 87402 : PG_FREE_IF_COPY(arg1, 0);
4166 87402 : PG_FREE_IF_COPY(arg2, 1);
4167 :
4168 87402 : PG_RETURN_INT32(cmp);
4169 : }
4170 :
4171 : Datum
4172 86 : bytea_sortsupport(PG_FUNCTION_ARGS)
4173 : {
4174 86 : SortSupport ssup = (SortSupport) PG_GETARG_POINTER(0);
4175 : MemoryContext oldcontext;
4176 :
4177 86 : oldcontext = MemoryContextSwitchTo(ssup->ssup_cxt);
4178 :
4179 : /* Use generic string SortSupport, forcing "C" collation */
4180 86 : varstr_sortsupport(ssup, BYTEAOID, C_COLLATION_OID);
4181 :
4182 86 : MemoryContextSwitchTo(oldcontext);
4183 :
4184 86 : PG_RETURN_VOID();
4185 : }
4186 :
4187 : /*
4188 : * appendStringInfoText
4189 : *
4190 : * Append a text to str.
4191 : * Like appendStringInfoString(str, text_to_cstring(t)) but faster.
4192 : */
4193 : static void
4194 1230954 : appendStringInfoText(StringInfo str, const text *t)
4195 : {
4196 1230954 : appendBinaryStringInfo(str, VARDATA_ANY(t), VARSIZE_ANY_EXHDR(t));
4197 1230954 : }
4198 :
4199 : /*
4200 : * replace_text
4201 : * replace all occurrences of 'old_sub_str' in 'orig_str'
4202 : * with 'new_sub_str' to form 'new_str'
4203 : *
4204 : * returns 'orig_str' if 'old_sub_str' == '' or 'orig_str' == ''
4205 : * otherwise returns 'new_str'
4206 : */
4207 : Datum
4208 966 : replace_text(PG_FUNCTION_ARGS)
4209 : {
4210 966 : text *src_text = PG_GETARG_TEXT_PP(0);
4211 966 : text *from_sub_text = PG_GETARG_TEXT_PP(1);
4212 966 : text *to_sub_text = PG_GETARG_TEXT_PP(2);
4213 : int src_text_len;
4214 : int from_sub_text_len;
4215 : TextPositionState state;
4216 : text *ret_text;
4217 : int chunk_len;
4218 : char *curr_ptr;
4219 : char *start_ptr;
4220 : StringInfoData str;
4221 : bool found;
4222 :
4223 966 : src_text_len = VARSIZE_ANY_EXHDR(src_text);
4224 966 : from_sub_text_len = VARSIZE_ANY_EXHDR(from_sub_text);
4225 :
4226 : /* Return unmodified source string if empty source or pattern */
4227 966 : if (src_text_len < 1 || from_sub_text_len < 1)
4228 : {
4229 0 : PG_RETURN_TEXT_P(src_text);
4230 : }
4231 :
4232 966 : text_position_setup(src_text, from_sub_text, PG_GET_COLLATION(), &state);
4233 :
4234 966 : found = text_position_next(&state);
4235 :
4236 : /* When the from_sub_text is not found, there is nothing to do. */
4237 966 : if (!found)
4238 : {
4239 762 : text_position_cleanup(&state);
4240 762 : PG_RETURN_TEXT_P(src_text);
4241 : }
4242 204 : curr_ptr = text_position_get_match_ptr(&state);
4243 204 : start_ptr = VARDATA_ANY(src_text);
4244 :
4245 204 : initStringInfo(&str);
4246 :
4247 : do
4248 : {
4249 2800 : CHECK_FOR_INTERRUPTS();
4250 :
4251 : /* copy the data skipped over by last text_position_next() */
4252 2800 : chunk_len = curr_ptr - start_ptr;
4253 2800 : appendBinaryStringInfo(&str, start_ptr, chunk_len);
4254 :
4255 2800 : appendStringInfoText(&str, to_sub_text);
4256 :
4257 2800 : start_ptr = curr_ptr + from_sub_text_len;
4258 :
4259 2800 : found = text_position_next(&state);
4260 2800 : if (found)
4261 2596 : curr_ptr = text_position_get_match_ptr(&state);
4262 : }
4263 2800 : while (found);
4264 :
4265 : /* copy trailing data */
4266 204 : chunk_len = ((char *) src_text + VARSIZE_ANY(src_text)) - start_ptr;
4267 204 : appendBinaryStringInfo(&str, start_ptr, chunk_len);
4268 :
4269 204 : text_position_cleanup(&state);
4270 :
4271 204 : ret_text = cstring_to_text_with_len(str.data, str.len);
4272 204 : pfree(str.data);
4273 :
4274 204 : PG_RETURN_TEXT_P(ret_text);
4275 : }
4276 :
4277 : /*
4278 : * check_replace_text_has_escape_char
4279 : *
4280 : * check whether replace_text contains escape char.
4281 : */
4282 : static bool
4283 988 : check_replace_text_has_escape_char(const text *replace_text)
4284 : {
4285 988 : const char *p = VARDATA_ANY(replace_text);
4286 988 : const char *p_end = p + VARSIZE_ANY_EXHDR(replace_text);
4287 :
4288 988 : if (pg_database_encoding_max_length() == 1)
4289 : {
4290 12 : for (; p < p_end; p++)
4291 : {
4292 0 : if (*p == '\\')
4293 0 : return true;
4294 : }
4295 : }
4296 : else
4297 : {
4298 3050 : for (; p < p_end; p += pg_mblen(p))
4299 : {
4300 2182 : if (*p == '\\')
4301 108 : return true;
4302 : }
4303 : }
4304 :
4305 880 : return false;
4306 : }
4307 :
4308 : /*
4309 : * appendStringInfoRegexpSubstr
4310 : *
4311 : * Append replace_text to str, substituting regexp back references for
4312 : * \n escapes. start_ptr is the start of the match in the source string,
4313 : * at logical character position data_pos.
4314 : */
4315 : static void
4316 60 : appendStringInfoRegexpSubstr(StringInfo str, text *replace_text,
4317 : regmatch_t *pmatch,
4318 : char *start_ptr, int data_pos)
4319 : {
4320 60 : const char *p = VARDATA_ANY(replace_text);
4321 60 : const char *p_end = p + VARSIZE_ANY_EXHDR(replace_text);
4322 60 : int eml = pg_database_encoding_max_length();
4323 :
4324 : for (;;)
4325 122 : {
4326 182 : const char *chunk_start = p;
4327 : int so;
4328 : int eo;
4329 :
4330 : /* Find next escape char. */
4331 182 : if (eml == 1)
4332 : {
4333 0 : for (; p < p_end && *p != '\\'; p++)
4334 : /* nothing */ ;
4335 : }
4336 : else
4337 : {
4338 182 : for (; p < p_end && *p != '\\'; p += pg_mblen(p))
4339 : /* nothing */ ;
4340 : }
4341 :
4342 : /* Copy the text we just scanned over, if any. */
4343 182 : if (p > chunk_start)
4344 72 : appendBinaryStringInfo(str, chunk_start, p - chunk_start);
4345 :
4346 : /* Done if at end of string, else advance over escape char. */
4347 182 : if (p >= p_end)
4348 60 : break;
4349 122 : p++;
4350 :
4351 122 : if (p >= p_end)
4352 : {
4353 : /* Escape at very end of input. Treat same as unexpected char */
4354 0 : appendStringInfoChar(str, '\\');
4355 0 : break;
4356 : }
4357 :
4358 122 : if (*p >= '1' && *p <= '9')
4359 104 : {
4360 : /* Use the back reference of regexp. */
4361 104 : int idx = *p - '0';
4362 :
4363 104 : so = pmatch[idx].rm_so;
4364 104 : eo = pmatch[idx].rm_eo;
4365 104 : p++;
4366 : }
4367 18 : else if (*p == '&')
4368 : {
4369 : /* Use the entire matched string. */
4370 0 : so = pmatch[0].rm_so;
4371 0 : eo = pmatch[0].rm_eo;
4372 0 : p++;
4373 : }
4374 18 : else if (*p == '\\')
4375 : {
4376 : /* \\ means transfer one \ to output. */
4377 18 : appendStringInfoChar(str, '\\');
4378 18 : p++;
4379 18 : continue;
4380 : }
4381 : else
4382 : {
4383 : /*
4384 : * If escape char is not followed by any expected char, just treat
4385 : * it as ordinary data to copy. (XXX would it be better to throw
4386 : * an error?)
4387 : */
4388 0 : appendStringInfoChar(str, '\\');
4389 0 : continue;
4390 : }
4391 :
4392 104 : if (so != -1 && eo != -1)
4393 : {
4394 : /*
4395 : * Copy the text that is back reference of regexp. Note so and eo
4396 : * are counted in characters not bytes.
4397 : */
4398 : char *chunk_start;
4399 : int chunk_len;
4400 :
4401 : Assert(so >= data_pos);
4402 104 : chunk_start = start_ptr;
4403 104 : chunk_start += charlen_to_bytelen(chunk_start, so - data_pos);
4404 104 : chunk_len = charlen_to_bytelen(chunk_start, eo - so);
4405 104 : appendBinaryStringInfo(str, chunk_start, chunk_len);
4406 : }
4407 : }
4408 60 : }
4409 :
4410 : #define REGEXP_REPLACE_BACKREF_CNT 10
4411 :
4412 : /*
4413 : * replace_text_regexp
4414 : *
4415 : * replace text that matches to regexp in src_text to replace_text.
4416 : *
4417 : * Note: to avoid having to include regex.h in builtins.h, we declare
4418 : * the regexp argument as void *, but really it's regex_t *.
4419 : */
4420 : text *
4421 988 : replace_text_regexp(text *src_text, void *regexp,
4422 : text *replace_text, bool glob)
4423 : {
4424 : text *ret_text;
4425 988 : regex_t *re = (regex_t *) regexp;
4426 988 : int src_text_len = VARSIZE_ANY_EXHDR(src_text);
4427 : StringInfoData buf;
4428 : regmatch_t pmatch[REGEXP_REPLACE_BACKREF_CNT];
4429 : pg_wchar *data;
4430 : size_t data_len;
4431 : int search_start;
4432 : int data_pos;
4433 : char *start_ptr;
4434 : bool have_escape;
4435 :
4436 988 : initStringInfo(&buf);
4437 :
4438 : /* Convert data string to wide characters. */
4439 988 : data = (pg_wchar *) palloc((src_text_len + 1) * sizeof(pg_wchar));
4440 988 : data_len = pg_mb2wchar_with_len(VARDATA_ANY(src_text), data, src_text_len);
4441 :
4442 : /* Check whether replace_text has escape char. */
4443 988 : have_escape = check_replace_text_has_escape_char(replace_text);
4444 :
4445 : /* start_ptr points to the data_pos'th character of src_text */
4446 988 : start_ptr = (char *) VARDATA_ANY(src_text);
4447 988 : data_pos = 0;
4448 :
4449 988 : search_start = 0;
4450 2314 : while (search_start <= data_len)
4451 : {
4452 : int regexec_result;
4453 :
4454 1322 : CHECK_FOR_INTERRUPTS();
4455 :
4456 1322 : regexec_result = pg_regexec(re,
4457 : data,
4458 : data_len,
4459 : search_start,
4460 : NULL, /* no details */
4461 : REGEXP_REPLACE_BACKREF_CNT,
4462 : pmatch,
4463 : 0);
4464 :
4465 1322 : if (regexec_result == REG_NOMATCH)
4466 814 : break;
4467 :
4468 508 : if (regexec_result != REG_OKAY)
4469 : {
4470 : char errMsg[100];
4471 :
4472 0 : CHECK_FOR_INTERRUPTS();
4473 0 : pg_regerror(regexec_result, re, errMsg, sizeof(errMsg));
4474 0 : ereport(ERROR,
4475 : (errcode(ERRCODE_INVALID_REGULAR_EXPRESSION),
4476 : errmsg("regular expression failed: %s", errMsg)));
4477 : }
4478 :
4479 : /*
4480 : * Copy the text to the left of the match position. Note we are given
4481 : * character not byte indexes.
4482 : */
4483 508 : if (pmatch[0].rm_so - data_pos > 0)
4484 : {
4485 : int chunk_len;
4486 :
4487 446 : chunk_len = charlen_to_bytelen(start_ptr,
4488 446 : pmatch[0].rm_so - data_pos);
4489 446 : appendBinaryStringInfo(&buf, start_ptr, chunk_len);
4490 :
4491 : /*
4492 : * Advance start_ptr over that text, to avoid multiple rescans of
4493 : * it if the replace_text contains multiple back-references.
4494 : */
4495 446 : start_ptr += chunk_len;
4496 446 : data_pos = pmatch[0].rm_so;
4497 : }
4498 :
4499 : /*
4500 : * Copy the replace_text. Process back references when the
4501 : * replace_text has escape characters.
4502 : */
4503 508 : if (have_escape)
4504 60 : appendStringInfoRegexpSubstr(&buf, replace_text, pmatch,
4505 : start_ptr, data_pos);
4506 : else
4507 448 : appendStringInfoText(&buf, replace_text);
4508 :
4509 : /* Advance start_ptr and data_pos over the matched text. */
4510 508 : start_ptr += charlen_to_bytelen(start_ptr,
4511 508 : pmatch[0].rm_eo - data_pos);
4512 508 : data_pos = pmatch[0].rm_eo;
4513 :
4514 : /*
4515 : * When global option is off, replace the first instance only.
4516 : */
4517 508 : if (!glob)
4518 170 : break;
4519 :
4520 : /*
4521 : * Advance search position. Normally we start the next search at the
4522 : * end of the previous match; but if the match was of zero length, we
4523 : * have to advance by one character, or we'd just find the same match
4524 : * again.
4525 : */
4526 338 : search_start = data_pos;
4527 338 : if (pmatch[0].rm_so == pmatch[0].rm_eo)
4528 8 : search_start++;
4529 : }
4530 :
4531 : /*
4532 : * Copy the text to the right of the last match.
4533 : */
4534 988 : if (data_pos < data_len)
4535 : {
4536 : int chunk_len;
4537 :
4538 894 : chunk_len = ((char *) src_text + VARSIZE_ANY(src_text)) - start_ptr;
4539 894 : appendBinaryStringInfo(&buf, start_ptr, chunk_len);
4540 : }
4541 :
4542 988 : ret_text = cstring_to_text_with_len(buf.data, buf.len);
4543 988 : pfree(buf.data);
4544 988 : pfree(data);
4545 :
4546 988 : return ret_text;
4547 : }
4548 :
4549 : /*
4550 : * split_text
4551 : * parse input string
4552 : * return ord item (1 based)
4553 : * based on provided field separator
4554 : */
4555 : Datum
4556 20 : split_text(PG_FUNCTION_ARGS)
4557 : {
4558 20 : text *inputstring = PG_GETARG_TEXT_PP(0);
4559 20 : text *fldsep = PG_GETARG_TEXT_PP(1);
4560 20 : int fldnum = PG_GETARG_INT32(2);
4561 : int inputstring_len;
4562 : int fldsep_len;
4563 : TextPositionState state;
4564 : char *start_ptr;
4565 : char *end_ptr;
4566 : text *result_text;
4567 : bool found;
4568 :
4569 : /* field number is 1 based */
4570 20 : if (fldnum < 1)
4571 4 : ereport(ERROR,
4572 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
4573 : errmsg("field position must be greater than zero")));
4574 :
4575 16 : inputstring_len = VARSIZE_ANY_EXHDR(inputstring);
4576 16 : fldsep_len = VARSIZE_ANY_EXHDR(fldsep);
4577 :
4578 : /* return empty string for empty input string */
4579 16 : if (inputstring_len < 1)
4580 0 : PG_RETURN_TEXT_P(cstring_to_text(""));
4581 :
4582 : /* empty field separator */
4583 16 : if (fldsep_len < 1)
4584 : {
4585 0 : text_position_cleanup(&state);
4586 : /* if first field, return input string, else empty string */
4587 0 : if (fldnum == 1)
4588 0 : PG_RETURN_TEXT_P(inputstring);
4589 : else
4590 0 : PG_RETURN_TEXT_P(cstring_to_text(""));
4591 : }
4592 :
4593 16 : text_position_setup(inputstring, fldsep, PG_GET_COLLATION(), &state);
4594 :
4595 : /* identify bounds of first field */
4596 16 : start_ptr = VARDATA_ANY(inputstring);
4597 16 : found = text_position_next(&state);
4598 :
4599 : /* special case if fldsep not found at all */
4600 16 : if (!found)
4601 : {
4602 0 : text_position_cleanup(&state);
4603 : /* if field 1 requested, return input string, else empty string */
4604 0 : if (fldnum == 1)
4605 0 : PG_RETURN_TEXT_P(inputstring);
4606 : else
4607 0 : PG_RETURN_TEXT_P(cstring_to_text(""));
4608 : }
4609 16 : end_ptr = text_position_get_match_ptr(&state);
4610 :
4611 44 : while (found && --fldnum > 0)
4612 : {
4613 : /* identify bounds of next field */
4614 12 : start_ptr = end_ptr + fldsep_len;
4615 12 : found = text_position_next(&state);
4616 12 : if (found)
4617 4 : end_ptr = text_position_get_match_ptr(&state);
4618 : }
4619 :
4620 16 : text_position_cleanup(&state);
4621 :
4622 16 : if (fldnum > 0)
4623 : {
4624 : /* N'th field separator not found */
4625 : /* if last field requested, return it, else empty string */
4626 8 : if (fldnum == 1)
4627 : {
4628 4 : int last_len = start_ptr - VARDATA_ANY(inputstring);
4629 :
4630 4 : result_text = cstring_to_text_with_len(start_ptr,
4631 : inputstring_len - last_len);
4632 : }
4633 : else
4634 4 : result_text = cstring_to_text("");
4635 : }
4636 : else
4637 : {
4638 : /* non-last field requested */
4639 8 : result_text = cstring_to_text_with_len(start_ptr, end_ptr - start_ptr);
4640 : }
4641 :
4642 16 : PG_RETURN_TEXT_P(result_text);
4643 : }
4644 :
4645 : /*
4646 : * Convenience function to return true when two text params are equal.
4647 : */
4648 : static bool
4649 56 : text_isequal(text *txt1, text *txt2, Oid collid)
4650 : {
4651 56 : return DatumGetBool(DirectFunctionCall2Coll(texteq,
4652 : collid,
4653 : PointerGetDatum(txt1),
4654 : PointerGetDatum(txt2)));
4655 : }
4656 :
4657 : /*
4658 : * text_to_array
4659 : * parse input string and return text array of elements,
4660 : * based on provided field separator
4661 : */
4662 : Datum
4663 56 : text_to_array(PG_FUNCTION_ARGS)
4664 : {
4665 56 : return text_to_array_internal(fcinfo);
4666 : }
4667 :
4668 : /*
4669 : * text_to_array_null
4670 : * parse input string and return text array of elements,
4671 : * based on provided field separator and null string
4672 : *
4673 : * This is a separate entry point only to prevent the regression tests from
4674 : * complaining about different argument sets for the same internal function.
4675 : */
4676 : Datum
4677 16 : text_to_array_null(PG_FUNCTION_ARGS)
4678 : {
4679 16 : return text_to_array_internal(fcinfo);
4680 : }
4681 :
4682 : /*
4683 : * common code for text_to_array and text_to_array_null functions
4684 : *
4685 : * These are not strict so we have to test for null inputs explicitly.
4686 : */
4687 : static Datum
4688 72 : text_to_array_internal(PG_FUNCTION_ARGS)
4689 : {
4690 : text *inputstring;
4691 : text *fldsep;
4692 : text *null_string;
4693 : int inputstring_len;
4694 : int fldsep_len;
4695 : char *start_ptr;
4696 : text *result_text;
4697 : bool is_null;
4698 72 : ArrayBuildState *astate = NULL;
4699 :
4700 : /* when input string is NULL, then result is NULL too */
4701 72 : if (PG_ARGISNULL(0))
4702 4 : PG_RETURN_NULL();
4703 :
4704 68 : inputstring = PG_GETARG_TEXT_PP(0);
4705 :
4706 : /* fldsep can be NULL */
4707 68 : if (!PG_ARGISNULL(1))
4708 64 : fldsep = PG_GETARG_TEXT_PP(1);
4709 : else
4710 4 : fldsep = NULL;
4711 :
4712 : /* null_string can be NULL or omitted */
4713 68 : if (PG_NARGS() > 2 && !PG_ARGISNULL(2))
4714 16 : null_string = PG_GETARG_TEXT_PP(2);
4715 : else
4716 52 : null_string = NULL;
4717 :
4718 68 : if (fldsep != NULL)
4719 : {
4720 : /*
4721 : * Normal case with non-null fldsep. Use the text_position machinery
4722 : * to search for occurrences of fldsep.
4723 : */
4724 : TextPositionState state;
4725 :
4726 64 : inputstring_len = VARSIZE_ANY_EXHDR(inputstring);
4727 64 : fldsep_len = VARSIZE_ANY_EXHDR(fldsep);
4728 :
4729 : /* return empty array for empty input string */
4730 64 : if (inputstring_len < 1)
4731 28 : PG_RETURN_ARRAYTYPE_P(construct_empty_array(TEXTOID));
4732 :
4733 : /*
4734 : * empty field separator: return the input string as a one-element
4735 : * array
4736 : */
4737 60 : if (fldsep_len < 1)
4738 : {
4739 : Datum elems[1];
4740 : bool nulls[1];
4741 : int dims[1];
4742 : int lbs[1];
4743 :
4744 : /* single element can be a NULL too */
4745 20 : is_null = null_string ? text_isequal(inputstring, null_string, PG_GET_COLLATION()) : false;
4746 :
4747 20 : elems[0] = PointerGetDatum(inputstring);
4748 20 : nulls[0] = is_null;
4749 20 : dims[0] = 1;
4750 20 : lbs[0] = 1;
4751 : /* XXX: this hardcodes assumptions about the text type */
4752 20 : PG_RETURN_ARRAYTYPE_P(construct_md_array(elems, nulls,
4753 : 1, dims, lbs,
4754 : TEXTOID, -1, false, 'i'));
4755 : }
4756 :
4757 40 : text_position_setup(inputstring, fldsep, PG_GET_COLLATION(), &state);
4758 :
4759 40 : start_ptr = VARDATA_ANY(inputstring);
4760 :
4761 : for (;;)
4762 208 : {
4763 : bool found;
4764 : char *end_ptr;
4765 : int chunk_len;
4766 :
4767 248 : CHECK_FOR_INTERRUPTS();
4768 :
4769 248 : found = text_position_next(&state);
4770 248 : if (!found)
4771 : {
4772 : /* fetch last field */
4773 40 : chunk_len = ((char *) inputstring + VARSIZE_ANY(inputstring)) - start_ptr;
4774 40 : end_ptr = NULL; /* not used, but some compilers complain */
4775 : }
4776 : else
4777 : {
4778 : /* fetch non-last field */
4779 208 : end_ptr = text_position_get_match_ptr(&state);
4780 208 : chunk_len = end_ptr - start_ptr;
4781 : }
4782 :
4783 : /* must build a temp text datum to pass to accumArrayResult */
4784 248 : result_text = cstring_to_text_with_len(start_ptr, chunk_len);
4785 248 : is_null = null_string ? text_isequal(result_text, null_string, PG_GET_COLLATION()) : false;
4786 :
4787 : /* stash away this field */
4788 248 : astate = accumArrayResult(astate,
4789 : PointerGetDatum(result_text),
4790 : is_null,
4791 : TEXTOID,
4792 : CurrentMemoryContext);
4793 :
4794 248 : pfree(result_text);
4795 :
4796 248 : if (!found)
4797 40 : break;
4798 :
4799 208 : start_ptr = end_ptr + fldsep_len;
4800 : }
4801 :
4802 40 : text_position_cleanup(&state);
4803 : }
4804 : else
4805 : {
4806 : /*
4807 : * When fldsep is NULL, each character in the inputstring becomes an
4808 : * element in the result array. The separator is effectively the
4809 : * space between characters.
4810 : */
4811 4 : inputstring_len = VARSIZE_ANY_EXHDR(inputstring);
4812 :
4813 : /* return empty array for empty input string */
4814 4 : if (inputstring_len < 1)
4815 0 : PG_RETURN_ARRAYTYPE_P(construct_empty_array(TEXTOID));
4816 :
4817 4 : start_ptr = VARDATA_ANY(inputstring);
4818 :
4819 28 : while (inputstring_len > 0)
4820 : {
4821 20 : int chunk_len = pg_mblen(start_ptr);
4822 :
4823 20 : CHECK_FOR_INTERRUPTS();
4824 :
4825 : /* must build a temp text datum to pass to accumArrayResult */
4826 20 : result_text = cstring_to_text_with_len(start_ptr, chunk_len);
4827 20 : is_null = null_string ? text_isequal(result_text, null_string, PG_GET_COLLATION()) : false;
4828 :
4829 : /* stash away this field */
4830 20 : astate = accumArrayResult(astate,
4831 : PointerGetDatum(result_text),
4832 : is_null,
4833 : TEXTOID,
4834 : CurrentMemoryContext);
4835 :
4836 20 : pfree(result_text);
4837 :
4838 20 : start_ptr += chunk_len;
4839 20 : inputstring_len -= chunk_len;
4840 : }
4841 : }
4842 :
4843 44 : PG_RETURN_ARRAYTYPE_P(makeArrayResult(astate,
4844 : CurrentMemoryContext));
4845 : }
4846 :
4847 : /*
4848 : * array_to_text
4849 : * concatenate Cstring representation of input array elements
4850 : * using provided field separator
4851 : */
4852 : Datum
4853 34592 : array_to_text(PG_FUNCTION_ARGS)
4854 : {
4855 34592 : ArrayType *v = PG_GETARG_ARRAYTYPE_P(0);
4856 34592 : char *fldsep = text_to_cstring(PG_GETARG_TEXT_PP(1));
4857 :
4858 34592 : PG_RETURN_TEXT_P(array_to_text_internal(fcinfo, v, fldsep, NULL));
4859 : }
4860 :
4861 : /*
4862 : * array_to_text_null
4863 : * concatenate Cstring representation of input array elements
4864 : * using provided field separator and null string
4865 : *
4866 : * This version is not strict so we have to test for null inputs explicitly.
4867 : */
4868 : Datum
4869 8 : array_to_text_null(PG_FUNCTION_ARGS)
4870 : {
4871 : ArrayType *v;
4872 : char *fldsep;
4873 : char *null_string;
4874 :
4875 : /* returns NULL when first or second parameter is NULL */
4876 8 : if (PG_ARGISNULL(0) || PG_ARGISNULL(1))
4877 0 : PG_RETURN_NULL();
4878 :
4879 8 : v = PG_GETARG_ARRAYTYPE_P(0);
4880 8 : fldsep = text_to_cstring(PG_GETARG_TEXT_PP(1));
4881 :
4882 : /* NULL null string is passed through as a null pointer */
4883 8 : if (!PG_ARGISNULL(2))
4884 4 : null_string = text_to_cstring(PG_GETARG_TEXT_PP(2));
4885 : else
4886 4 : null_string = NULL;
4887 :
4888 8 : PG_RETURN_TEXT_P(array_to_text_internal(fcinfo, v, fldsep, null_string));
4889 : }
4890 :
4891 : /*
4892 : * common code for array_to_text and array_to_text_null functions
4893 : */
4894 : static text *
4895 34612 : array_to_text_internal(FunctionCallInfo fcinfo, ArrayType *v,
4896 : const char *fldsep, const char *null_string)
4897 : {
4898 : text *result;
4899 : int nitems,
4900 : *dims,
4901 : ndims;
4902 : Oid element_type;
4903 : int typlen;
4904 : bool typbyval;
4905 : char typalign;
4906 : StringInfoData buf;
4907 34612 : bool printed = false;
4908 : char *p;
4909 : bits8 *bitmap;
4910 : int bitmask;
4911 : int i;
4912 : ArrayMetaState *my_extra;
4913 :
4914 34612 : ndims = ARR_NDIM(v);
4915 34612 : dims = ARR_DIMS(v);
4916 34612 : nitems = ArrayGetNItems(ndims, dims);
4917 :
4918 : /* if there are no elements, return an empty string */
4919 34612 : if (nitems == 0)
4920 20280 : return cstring_to_text_with_len("", 0);
4921 :
4922 14332 : element_type = ARR_ELEMTYPE(v);
4923 14332 : initStringInfo(&buf);
4924 :
4925 : /*
4926 : * We arrange to look up info about element type, including its output
4927 : * conversion proc, only once per series of calls, assuming the element
4928 : * type doesn't change underneath us.
4929 : */
4930 14332 : my_extra = (ArrayMetaState *) fcinfo->flinfo->fn_extra;
4931 14332 : if (my_extra == NULL)
4932 : {
4933 920 : fcinfo->flinfo->fn_extra = MemoryContextAlloc(fcinfo->flinfo->fn_mcxt,
4934 : sizeof(ArrayMetaState));
4935 920 : my_extra = (ArrayMetaState *) fcinfo->flinfo->fn_extra;
4936 920 : my_extra->element_type = ~element_type;
4937 : }
4938 :
4939 14332 : if (my_extra->element_type != element_type)
4940 : {
4941 : /*
4942 : * Get info about element type, including its output conversion proc
4943 : */
4944 920 : get_type_io_data(element_type, IOFunc_output,
4945 : &my_extra->typlen, &my_extra->typbyval,
4946 : &my_extra->typalign, &my_extra->typdelim,
4947 : &my_extra->typioparam, &my_extra->typiofunc);
4948 920 : fmgr_info_cxt(my_extra->typiofunc, &my_extra->proc,
4949 920 : fcinfo->flinfo->fn_mcxt);
4950 920 : my_extra->element_type = element_type;
4951 : }
4952 14332 : typlen = my_extra->typlen;
4953 14332 : typbyval = my_extra->typbyval;
4954 14332 : typalign = my_extra->typalign;
4955 :
4956 14332 : p = ARR_DATA_PTR(v);
4957 14332 : bitmap = ARR_NULLBITMAP(v);
4958 14332 : bitmask = 1;
4959 :
4960 49034 : for (i = 0; i < nitems; i++)
4961 : {
4962 : Datum itemvalue;
4963 : char *value;
4964 :
4965 : /* Get source element, checking for NULL */
4966 34702 : if (bitmap && (*bitmap & bitmask) == 0)
4967 : {
4968 : /* if null_string is NULL, we just ignore null elements */
4969 24 : if (null_string != NULL)
4970 : {
4971 4 : if (printed)
4972 4 : appendStringInfo(&buf, "%s%s", fldsep, null_string);
4973 : else
4974 0 : appendStringInfoString(&buf, null_string);
4975 4 : printed = true;
4976 : }
4977 : }
4978 : else
4979 : {
4980 34690 : itemvalue = fetch_att(p, typbyval, typlen);
4981 :
4982 34690 : value = OutputFunctionCall(&my_extra->proc, itemvalue);
4983 :
4984 34690 : if (printed)
4985 20358 : appendStringInfo(&buf, "%s%s", fldsep, value);
4986 : else
4987 14332 : appendStringInfoString(&buf, value);
4988 34690 : printed = true;
4989 :
4990 34690 : p = att_addlength_pointer(p, typlen, p);
4991 34690 : p = (char *) att_align_nominal(p, typalign);
4992 : }
4993 :
4994 : /* advance bitmap pointer if any */
4995 34702 : if (bitmap)
4996 : {
4997 72 : bitmask <<= 1;
4998 72 : if (bitmask == 0x100)
4999 : {
5000 0 : bitmap++;
5001 0 : bitmask = 1;
5002 : }
5003 : }
5004 : }
5005 :
5006 14332 : result = cstring_to_text_with_len(buf.data, buf.len);
5007 14332 : pfree(buf.data);
5008 :
5009 14332 : return result;
5010 : }
5011 :
5012 : #define HEXBASE 16
5013 : /*
5014 : * Convert an int32 to a string containing a base 16 (hex) representation of
5015 : * the number.
5016 : */
5017 : Datum
5018 1324 : to_hex32(PG_FUNCTION_ARGS)
5019 : {
5020 1324 : uint32 value = (uint32) PG_GETARG_INT32(0);
5021 : char *ptr;
5022 1324 : const char *digits = "0123456789abcdef";
5023 : char buf[32]; /* bigger than needed, but reasonable */
5024 :
5025 1324 : ptr = buf + sizeof(buf) - 1;
5026 1324 : *ptr = '\0';
5027 :
5028 : do
5029 : {
5030 2472 : *--ptr = digits[value % HEXBASE];
5031 2472 : value /= HEXBASE;
5032 2472 : } while (ptr > buf && value);
5033 :
5034 1324 : PG_RETURN_TEXT_P(cstring_to_text(ptr));
5035 : }
5036 :
5037 : /*
5038 : * Convert an int64 to a string containing a base 16 (hex) representation of
5039 : * the number.
5040 : */
5041 : Datum
5042 4 : to_hex64(PG_FUNCTION_ARGS)
5043 : {
5044 4 : uint64 value = (uint64) PG_GETARG_INT64(0);
5045 : char *ptr;
5046 4 : const char *digits = "0123456789abcdef";
5047 : char buf[32]; /* bigger than needed, but reasonable */
5048 :
5049 4 : ptr = buf + sizeof(buf) - 1;
5050 4 : *ptr = '\0';
5051 :
5052 : do
5053 : {
5054 32 : *--ptr = digits[value % HEXBASE];
5055 32 : value /= HEXBASE;
5056 32 : } while (ptr > buf && value);
5057 :
5058 4 : PG_RETURN_TEXT_P(cstring_to_text(ptr));
5059 : }
5060 :
5061 : /*
5062 : * Return the size of a datum, possibly compressed
5063 : *
5064 : * Works on any data type
5065 : */
5066 : Datum
5067 102 : pg_column_size(PG_FUNCTION_ARGS)
5068 : {
5069 102 : Datum value = PG_GETARG_DATUM(0);
5070 : int32 result;
5071 : int typlen;
5072 :
5073 : /* On first call, get the input type's typlen, and save at *fn_extra */
5074 102 : if (fcinfo->flinfo->fn_extra == NULL)
5075 : {
5076 : /* Lookup the datatype of the supplied argument */
5077 102 : Oid argtypeid = get_fn_expr_argtype(fcinfo->flinfo, 0);
5078 :
5079 102 : typlen = get_typlen(argtypeid);
5080 102 : if (typlen == 0) /* should not happen */
5081 0 : elog(ERROR, "cache lookup failed for type %u", argtypeid);
5082 :
5083 102 : fcinfo->flinfo->fn_extra = MemoryContextAlloc(fcinfo->flinfo->fn_mcxt,
5084 : sizeof(int));
5085 102 : *((int *) fcinfo->flinfo->fn_extra) = typlen;
5086 : }
5087 : else
5088 0 : typlen = *((int *) fcinfo->flinfo->fn_extra);
5089 :
5090 102 : if (typlen == -1)
5091 : {
5092 : /* varlena type, possibly toasted */
5093 102 : result = toast_datum_size(value);
5094 : }
5095 0 : else if (typlen == -2)
5096 : {
5097 : /* cstring */
5098 0 : result = strlen(DatumGetCString(value)) + 1;
5099 : }
5100 : else
5101 : {
5102 : /* ordinary fixed-width type */
5103 0 : result = typlen;
5104 : }
5105 :
5106 102 : PG_RETURN_INT32(result);
5107 : }
5108 :
5109 : /*
5110 : * string_agg - Concatenates values and returns string.
5111 : *
5112 : * Syntax: string_agg(value text, delimiter text) RETURNS text
5113 : *
5114 : * Note: Any NULL values are ignored. The first-call delimiter isn't
5115 : * actually used at all, and on subsequent calls the delimiter precedes
5116 : * the associated value.
5117 : */
5118 :
5119 : /* subroutine to initialize state */
5120 : static StringInfo
5121 578 : makeStringAggState(FunctionCallInfo fcinfo)
5122 : {
5123 : StringInfo state;
5124 : MemoryContext aggcontext;
5125 : MemoryContext oldcontext;
5126 :
5127 578 : if (!AggCheckCallContext(fcinfo, &aggcontext))
5128 : {
5129 : /* cannot be called directly because of internal-type argument */
5130 0 : elog(ERROR, "string_agg_transfn called in non-aggregate context");
5131 : }
5132 :
5133 : /*
5134 : * Create state in aggregate context. It'll stay there across subsequent
5135 : * calls.
5136 : */
5137 578 : oldcontext = MemoryContextSwitchTo(aggcontext);
5138 578 : state = makeStringInfo();
5139 578 : MemoryContextSwitchTo(oldcontext);
5140 :
5141 578 : return state;
5142 : }
5143 :
5144 : Datum
5145 614166 : string_agg_transfn(PG_FUNCTION_ARGS)
5146 : {
5147 : StringInfo state;
5148 :
5149 614166 : state = PG_ARGISNULL(0) ? NULL : (StringInfo) PG_GETARG_POINTER(0);
5150 :
5151 : /* Append the value unless null. */
5152 614166 : if (!PG_ARGISNULL(1))
5153 : {
5154 : /* On the first time through, we ignore the delimiter. */
5155 614134 : if (state == NULL)
5156 562 : state = makeStringAggState(fcinfo);
5157 613572 : else if (!PG_ARGISNULL(2))
5158 613572 : appendStringInfoText(state, PG_GETARG_TEXT_PP(2)); /* delimiter */
5159 :
5160 614134 : appendStringInfoText(state, PG_GETARG_TEXT_PP(1)); /* value */
5161 : }
5162 :
5163 : /*
5164 : * The transition type for string_agg() is declared to be "internal",
5165 : * which is a pass-by-value type the same size as a pointer.
5166 : */
5167 614166 : PG_RETURN_POINTER(state);
5168 : }
5169 :
5170 : Datum
5171 602 : string_agg_finalfn(PG_FUNCTION_ARGS)
5172 : {
5173 : StringInfo state;
5174 :
5175 : /* cannot be called directly because of internal-type argument */
5176 : Assert(AggCheckCallContext(fcinfo, NULL));
5177 :
5178 602 : state = PG_ARGISNULL(0) ? NULL : (StringInfo) PG_GETARG_POINTER(0);
5179 :
5180 602 : if (state != NULL)
5181 562 : PG_RETURN_TEXT_P(cstring_to_text_with_len(state->data, state->len));
5182 : else
5183 40 : PG_RETURN_NULL();
5184 : }
5185 :
5186 : /*
5187 : * Prepare cache with fmgr info for the output functions of the datatypes of
5188 : * the arguments of a concat-like function, beginning with argument "argidx".
5189 : * (Arguments before that will have corresponding slots in the resulting
5190 : * FmgrInfo array, but we don't fill those slots.)
5191 : */
5192 : static FmgrInfo *
5193 24 : build_concat_foutcache(FunctionCallInfo fcinfo, int argidx)
5194 : {
5195 : FmgrInfo *foutcache;
5196 : int i;
5197 :
5198 : /* We keep the info in fn_mcxt so it survives across calls */
5199 24 : foutcache = (FmgrInfo *) MemoryContextAlloc(fcinfo->flinfo->fn_mcxt,
5200 24 : PG_NARGS() * sizeof(FmgrInfo));
5201 :
5202 120 : for (i = argidx; i < PG_NARGS(); i++)
5203 : {
5204 : Oid valtype;
5205 : Oid typOutput;
5206 : bool typIsVarlena;
5207 :
5208 96 : valtype = get_fn_expr_argtype(fcinfo->flinfo, i);
5209 96 : if (!OidIsValid(valtype))
5210 0 : elog(ERROR, "could not determine data type of concat() input");
5211 :
5212 96 : getTypeOutputInfo(valtype, &typOutput, &typIsVarlena);
5213 96 : fmgr_info_cxt(typOutput, &foutcache[i], fcinfo->flinfo->fn_mcxt);
5214 : }
5215 :
5216 24 : fcinfo->flinfo->fn_extra = foutcache;
5217 :
5218 24 : return foutcache;
5219 : }
5220 :
5221 : /*
5222 : * Implementation of both concat() and concat_ws().
5223 : *
5224 : * sepstr is the separator string to place between values.
5225 : * argidx identifies the first argument to concatenate (counting from zero);
5226 : * note that this must be constant across any one series of calls.
5227 : *
5228 : * Returns NULL if result should be NULL, else text value.
5229 : */
5230 : static text *
5231 44 : concat_internal(const char *sepstr, int argidx,
5232 : FunctionCallInfo fcinfo)
5233 : {
5234 : text *result;
5235 : StringInfoData str;
5236 : FmgrInfo *foutcache;
5237 44 : bool first_arg = true;
5238 : int i;
5239 :
5240 : /*
5241 : * concat(VARIADIC some-array) is essentially equivalent to
5242 : * array_to_text(), ie concat the array elements with the given separator.
5243 : * So we just pass the case off to that code.
5244 : */
5245 44 : if (get_fn_expr_variadic(fcinfo->flinfo))
5246 : {
5247 : ArrayType *arr;
5248 :
5249 : /* Should have just the one argument */
5250 : Assert(argidx == PG_NARGS() - 1);
5251 :
5252 : /* concat(VARIADIC NULL) is defined as NULL */
5253 20 : if (PG_ARGISNULL(argidx))
5254 8 : return NULL;
5255 :
5256 : /*
5257 : * Non-null argument had better be an array. We assume that any call
5258 : * context that could let get_fn_expr_variadic return true will have
5259 : * checked that a VARIADIC-labeled parameter actually is an array. So
5260 : * it should be okay to just Assert that it's an array rather than
5261 : * doing a full-fledged error check.
5262 : */
5263 : Assert(OidIsValid(get_base_element_type(get_fn_expr_argtype(fcinfo->flinfo, argidx))));
5264 :
5265 : /* OK, safe to fetch the array value */
5266 12 : arr = PG_GETARG_ARRAYTYPE_P(argidx);
5267 :
5268 : /*
5269 : * And serialize the array. We tell array_to_text to ignore null
5270 : * elements, which matches the behavior of the loop below.
5271 : */
5272 12 : return array_to_text_internal(fcinfo, arr, sepstr, NULL);
5273 : }
5274 :
5275 : /* Normal case without explicit VARIADIC marker */
5276 24 : initStringInfo(&str);
5277 :
5278 : /* Get output function info, building it if first time through */
5279 24 : foutcache = (FmgrInfo *) fcinfo->flinfo->fn_extra;
5280 24 : if (foutcache == NULL)
5281 24 : foutcache = build_concat_foutcache(fcinfo, argidx);
5282 :
5283 120 : for (i = argidx; i < PG_NARGS(); i++)
5284 : {
5285 96 : if (!PG_ARGISNULL(i))
5286 : {
5287 88 : Datum value = PG_GETARG_DATUM(i);
5288 :
5289 : /* add separator if appropriate */
5290 88 : if (first_arg)
5291 24 : first_arg = false;
5292 : else
5293 64 : appendStringInfoString(&str, sepstr);
5294 :
5295 : /* call the appropriate type output function, append the result */
5296 88 : appendStringInfoString(&str,
5297 88 : OutputFunctionCall(&foutcache[i], value));
5298 : }
5299 : }
5300 :
5301 24 : result = cstring_to_text_with_len(str.data, str.len);
5302 24 : pfree(str.data);
5303 :
5304 24 : return result;
5305 : }
5306 :
5307 : /*
5308 : * Concatenate all arguments. NULL arguments are ignored.
5309 : */
5310 : Datum
5311 20 : text_concat(PG_FUNCTION_ARGS)
5312 : {
5313 : text *result;
5314 :
5315 20 : result = concat_internal("", 0, fcinfo);
5316 20 : if (result == NULL)
5317 4 : PG_RETURN_NULL();
5318 16 : PG_RETURN_TEXT_P(result);
5319 : }
5320 :
5321 : /*
5322 : * Concatenate all but first argument value with separators. The first
5323 : * parameter is used as the separator. NULL arguments are ignored.
5324 : */
5325 : Datum
5326 28 : text_concat_ws(PG_FUNCTION_ARGS)
5327 : {
5328 : char *sep;
5329 : text *result;
5330 :
5331 : /* return NULL when separator is NULL */
5332 28 : if (PG_ARGISNULL(0))
5333 4 : PG_RETURN_NULL();
5334 24 : sep = text_to_cstring(PG_GETARG_TEXT_PP(0));
5335 :
5336 24 : result = concat_internal(sep, 1, fcinfo);
5337 24 : if (result == NULL)
5338 4 : PG_RETURN_NULL();
5339 20 : PG_RETURN_TEXT_P(result);
5340 : }
5341 :
5342 : /*
5343 : * Return first n characters in the string. When n is negative,
5344 : * return all but last |n| characters.
5345 : */
5346 : Datum
5347 44 : text_left(PG_FUNCTION_ARGS)
5348 : {
5349 44 : int n = PG_GETARG_INT32(1);
5350 :
5351 44 : if (n < 0)
5352 : {
5353 20 : text *str = PG_GETARG_TEXT_PP(0);
5354 20 : const char *p = VARDATA_ANY(str);
5355 20 : int len = VARSIZE_ANY_EXHDR(str);
5356 : int rlen;
5357 :
5358 20 : n = pg_mbstrlen_with_len(p, len) + n;
5359 20 : rlen = pg_mbcharcliplen(p, len, n);
5360 20 : PG_RETURN_TEXT_P(cstring_to_text_with_len(p, rlen));
5361 : }
5362 : else
5363 24 : PG_RETURN_TEXT_P(text_substring(PG_GETARG_DATUM(0), 1, n, false));
5364 : }
5365 :
5366 : /*
5367 : * Return last n characters in the string. When n is negative,
5368 : * return all but first |n| characters.
5369 : */
5370 : Datum
5371 44 : text_right(PG_FUNCTION_ARGS)
5372 : {
5373 44 : text *str = PG_GETARG_TEXT_PP(0);
5374 44 : const char *p = VARDATA_ANY(str);
5375 44 : int len = VARSIZE_ANY_EXHDR(str);
5376 44 : int n = PG_GETARG_INT32(1);
5377 : int off;
5378 :
5379 44 : if (n < 0)
5380 20 : n = -n;
5381 : else
5382 24 : n = pg_mbstrlen_with_len(p, len) - n;
5383 44 : off = pg_mbcharcliplen(p, len, n);
5384 :
5385 44 : PG_RETURN_TEXT_P(cstring_to_text_with_len(p + off, len - off));
5386 : }
5387 :
5388 : /*
5389 : * Return reversed string
5390 : */
5391 : Datum
5392 4 : text_reverse(PG_FUNCTION_ARGS)
5393 : {
5394 4 : text *str = PG_GETARG_TEXT_PP(0);
5395 4 : const char *p = VARDATA_ANY(str);
5396 4 : int len = VARSIZE_ANY_EXHDR(str);
5397 4 : const char *endp = p + len;
5398 : text *result;
5399 : char *dst;
5400 :
5401 4 : result = palloc(len + VARHDRSZ);
5402 4 : dst = (char *) VARDATA(result) + len;
5403 4 : SET_VARSIZE(result, len + VARHDRSZ);
5404 :
5405 4 : if (pg_database_encoding_max_length() > 1)
5406 : {
5407 : /* multibyte version */
5408 28 : while (p < endp)
5409 : {
5410 : int sz;
5411 :
5412 20 : sz = pg_mblen(p);
5413 20 : dst -= sz;
5414 20 : memcpy(dst, p, sz);
5415 20 : p += sz;
5416 : }
5417 : }
5418 : else
5419 : {
5420 : /* single byte version */
5421 0 : while (p < endp)
5422 0 : *(--dst) = *p++;
5423 : }
5424 :
5425 4 : PG_RETURN_TEXT_P(result);
5426 : }
5427 :
5428 :
5429 : /*
5430 : * Support macros for text_format()
5431 : */
5432 : #define TEXT_FORMAT_FLAG_MINUS 0x0001 /* is minus flag present? */
5433 :
5434 : #define ADVANCE_PARSE_POINTER(ptr,end_ptr) \
5435 : do { \
5436 : if (++(ptr) >= (end_ptr)) \
5437 : ereport(ERROR, \
5438 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE), \
5439 : errmsg("unterminated format() type specifier"), \
5440 : errhint("For a single \"%%\" use \"%%%%\"."))); \
5441 : } while (0)
5442 :
5443 : /*
5444 : * Returns a formatted string
5445 : */
5446 : Datum
5447 7500 : text_format(PG_FUNCTION_ARGS)
5448 : {
5449 : text *fmt;
5450 : StringInfoData str;
5451 : const char *cp;
5452 : const char *start_ptr;
5453 : const char *end_ptr;
5454 : text *result;
5455 : int arg;
5456 : bool funcvariadic;
5457 : int nargs;
5458 7500 : Datum *elements = NULL;
5459 7500 : bool *nulls = NULL;
5460 7500 : Oid element_type = InvalidOid;
5461 7500 : Oid prev_type = InvalidOid;
5462 7500 : Oid prev_width_type = InvalidOid;
5463 : FmgrInfo typoutputfinfo;
5464 : FmgrInfo typoutputinfo_width;
5465 :
5466 : /* When format string is null, immediately return null */
5467 7500 : if (PG_ARGISNULL(0))
5468 4 : PG_RETURN_NULL();
5469 :
5470 : /* If argument is marked VARIADIC, expand array into elements */
5471 7496 : if (get_fn_expr_variadic(fcinfo->flinfo))
5472 : {
5473 : ArrayType *arr;
5474 : int16 elmlen;
5475 : bool elmbyval;
5476 : char elmalign;
5477 : int nitems;
5478 :
5479 : /* Should have just the one argument */
5480 : Assert(PG_NARGS() == 2);
5481 :
5482 : /* If argument is NULL, we treat it as zero-length array */
5483 32 : if (PG_ARGISNULL(1))
5484 4 : nitems = 0;
5485 : else
5486 : {
5487 : /*
5488 : * Non-null argument had better be an array. We assume that any
5489 : * call context that could let get_fn_expr_variadic return true
5490 : * will have checked that a VARIADIC-labeled parameter actually is
5491 : * an array. So it should be okay to just Assert that it's an
5492 : * array rather than doing a full-fledged error check.
5493 : */
5494 : Assert(OidIsValid(get_base_element_type(get_fn_expr_argtype(fcinfo->flinfo, 1))));
5495 :
5496 : /* OK, safe to fetch the array value */
5497 28 : arr = PG_GETARG_ARRAYTYPE_P(1);
5498 :
5499 : /* Get info about array element type */
5500 28 : element_type = ARR_ELEMTYPE(arr);
5501 28 : get_typlenbyvalalign(element_type,
5502 : &elmlen, &elmbyval, &elmalign);
5503 :
5504 : /* Extract all array elements */
5505 28 : deconstruct_array(arr, element_type, elmlen, elmbyval, elmalign,
5506 : &elements, &nulls, &nitems);
5507 : }
5508 :
5509 32 : nargs = nitems + 1;
5510 32 : funcvariadic = true;
5511 : }
5512 : else
5513 : {
5514 : /* Non-variadic case, we'll process the arguments individually */
5515 7464 : nargs = PG_NARGS();
5516 7464 : funcvariadic = false;
5517 : }
5518 :
5519 : /* Setup for main loop. */
5520 7496 : fmt = PG_GETARG_TEXT_PP(0);
5521 7496 : start_ptr = VARDATA_ANY(fmt);
5522 7496 : end_ptr = start_ptr + VARSIZE_ANY_EXHDR(fmt);
5523 7496 : initStringInfo(&str);
5524 7496 : arg = 1; /* next argument position to print */
5525 :
5526 : /* Scan format string, looking for conversion specifiers. */
5527 238866 : for (cp = start_ptr; cp < end_ptr; cp++)
5528 : {
5529 : int argpos;
5530 : int widthpos;
5531 : int flags;
5532 : int width;
5533 : Datum value;
5534 : bool isNull;
5535 : Oid typid;
5536 :
5537 : /*
5538 : * If it's not the start of a conversion specifier, just copy it to
5539 : * the output buffer.
5540 : */
5541 231410 : if (*cp != '%')
5542 : {
5543 214666 : appendStringInfoCharMacro(&str, *cp);
5544 429344 : continue;
5545 : }
5546 :
5547 16744 : ADVANCE_PARSE_POINTER(cp, end_ptr);
5548 :
5549 : /* Easy case: %% outputs a single % */
5550 16744 : if (*cp == '%')
5551 : {
5552 12 : appendStringInfoCharMacro(&str, *cp);
5553 12 : continue;
5554 : }
5555 :
5556 : /* Parse the optional portions of the format specifier */
5557 16732 : cp = text_format_parse_format(cp, end_ptr,
5558 : &argpos, &widthpos,
5559 : &flags, &width);
5560 :
5561 : /*
5562 : * Next we should see the main conversion specifier. Whether or not
5563 : * an argument position was present, it's known that at least one
5564 : * character remains in the string at this point. Experience suggests
5565 : * that it's worth checking that that character is one of the expected
5566 : * ones before we try to fetch arguments, so as to produce the least
5567 : * confusing response to a mis-formatted specifier.
5568 : */
5569 16716 : if (strchr("sIL", *cp) == NULL)
5570 4 : ereport(ERROR,
5571 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
5572 : errmsg("unrecognized format() type specifier \"%c\"",
5573 : *cp),
5574 : errhint("For a single \"%%\" use \"%%%%\".")));
5575 :
5576 : /* If indirect width was specified, get its value */
5577 16712 : if (widthpos >= 0)
5578 : {
5579 : /* Collect the specified or next argument position */
5580 28 : if (widthpos > 0)
5581 24 : arg = widthpos;
5582 28 : if (arg >= nargs)
5583 0 : ereport(ERROR,
5584 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
5585 : errmsg("too few arguments for format()")));
5586 :
5587 : /* Get the value and type of the selected argument */
5588 28 : if (!funcvariadic)
5589 : {
5590 28 : value = PG_GETARG_DATUM(arg);
5591 28 : isNull = PG_ARGISNULL(arg);
5592 28 : typid = get_fn_expr_argtype(fcinfo->flinfo, arg);
5593 : }
5594 : else
5595 : {
5596 0 : value = elements[arg - 1];
5597 0 : isNull = nulls[arg - 1];
5598 0 : typid = element_type;
5599 : }
5600 28 : if (!OidIsValid(typid))
5601 0 : elog(ERROR, "could not determine data type of format() input");
5602 :
5603 28 : arg++;
5604 :
5605 : /* We can treat NULL width the same as zero */
5606 28 : if (isNull)
5607 4 : width = 0;
5608 24 : else if (typid == INT4OID)
5609 24 : width = DatumGetInt32(value);
5610 0 : else if (typid == INT2OID)
5611 0 : width = DatumGetInt16(value);
5612 : else
5613 : {
5614 : /* For less-usual datatypes, convert to text then to int */
5615 : char *str;
5616 :
5617 0 : if (typid != prev_width_type)
5618 : {
5619 : Oid typoutputfunc;
5620 : bool typIsVarlena;
5621 :
5622 0 : getTypeOutputInfo(typid, &typoutputfunc, &typIsVarlena);
5623 0 : fmgr_info(typoutputfunc, &typoutputinfo_width);
5624 0 : prev_width_type = typid;
5625 : }
5626 :
5627 0 : str = OutputFunctionCall(&typoutputinfo_width, value);
5628 :
5629 : /* pg_strtoint32 will complain about bad data or overflow */
5630 0 : width = pg_strtoint32(str);
5631 :
5632 0 : pfree(str);
5633 : }
5634 : }
5635 :
5636 : /* Collect the specified or next argument position */
5637 16712 : if (argpos > 0)
5638 88 : arg = argpos;
5639 16712 : if (arg >= nargs)
5640 16 : ereport(ERROR,
5641 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
5642 : errmsg("too few arguments for format()")));
5643 :
5644 : /* Get the value and type of the selected argument */
5645 16696 : if (!funcvariadic)
5646 : {
5647 15848 : value = PG_GETARG_DATUM(arg);
5648 15848 : isNull = PG_ARGISNULL(arg);
5649 15848 : typid = get_fn_expr_argtype(fcinfo->flinfo, arg);
5650 : }
5651 : else
5652 : {
5653 848 : value = elements[arg - 1];
5654 848 : isNull = nulls[arg - 1];
5655 848 : typid = element_type;
5656 : }
5657 16696 : if (!OidIsValid(typid))
5658 0 : elog(ERROR, "could not determine data type of format() input");
5659 :
5660 16696 : arg++;
5661 :
5662 : /*
5663 : * Get the appropriate typOutput function, reusing previous one if
5664 : * same type as previous argument. That's particularly useful in the
5665 : * variadic-array case, but often saves work even for ordinary calls.
5666 : */
5667 16696 : if (typid != prev_type)
5668 : {
5669 : Oid typoutputfunc;
5670 : bool typIsVarlena;
5671 :
5672 8484 : getTypeOutputInfo(typid, &typoutputfunc, &typIsVarlena);
5673 8484 : fmgr_info(typoutputfunc, &typoutputfinfo);
5674 8484 : prev_type = typid;
5675 : }
5676 :
5677 : /*
5678 : * And now we can format the value.
5679 : */
5680 16696 : switch (*cp)
5681 : {
5682 : case 's':
5683 : case 'I':
5684 : case 'L':
5685 16696 : text_format_string_conversion(&str, *cp, &typoutputfinfo,
5686 : value, isNull,
5687 : flags, width);
5688 16692 : break;
5689 : default:
5690 : /* should not get here, because of previous check */
5691 0 : ereport(ERROR,
5692 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
5693 : errmsg("unrecognized format() type specifier \"%c\"",
5694 : *cp),
5695 : errhint("For a single \"%%\" use \"%%%%\".")));
5696 : break;
5697 : }
5698 : }
5699 :
5700 : /* Don't need deconstruct_array results anymore. */
5701 7456 : if (elements != NULL)
5702 28 : pfree(elements);
5703 7456 : if (nulls != NULL)
5704 28 : pfree(nulls);
5705 :
5706 : /* Generate results. */
5707 7456 : result = cstring_to_text_with_len(str.data, str.len);
5708 7456 : pfree(str.data);
5709 :
5710 7456 : PG_RETURN_TEXT_P(result);
5711 : }
5712 :
5713 : /*
5714 : * Parse contiguous digits as a decimal number.
5715 : *
5716 : * Returns true if some digits could be parsed.
5717 : * The value is returned into *value, and *ptr is advanced to the next
5718 : * character to be parsed.
5719 : *
5720 : * Note parsing invariant: at least one character is known available before
5721 : * string end (end_ptr) at entry, and this is still true at exit.
5722 : */
5723 : static bool
5724 33440 : text_format_parse_digits(const char **ptr, const char *end_ptr, int *value)
5725 : {
5726 33440 : bool found = false;
5727 33440 : const char *cp = *ptr;
5728 33440 : int val = 0;
5729 :
5730 67088 : while (*cp >= '0' && *cp <= '9')
5731 : {
5732 212 : int8 digit = (*cp - '0');
5733 :
5734 424 : if (unlikely(pg_mul_s32_overflow(val, 10, &val)) ||
5735 212 : unlikely(pg_add_s32_overflow(val, digit, &val)))
5736 0 : ereport(ERROR,
5737 : (errcode(ERRCODE_NUMERIC_VALUE_OUT_OF_RANGE),
5738 : errmsg("number is out of range")));
5739 212 : ADVANCE_PARSE_POINTER(cp, end_ptr);
5740 208 : found = true;
5741 : }
5742 :
5743 33436 : *ptr = cp;
5744 33436 : *value = val;
5745 :
5746 33436 : return found;
5747 : }
5748 :
5749 : /*
5750 : * Parse a format specifier (generally following the SUS printf spec).
5751 : *
5752 : * We have already advanced over the initial '%', and we are looking for
5753 : * [argpos][flags][width]type (but the type character is not consumed here).
5754 : *
5755 : * Inputs are start_ptr (the position after '%') and end_ptr (string end + 1).
5756 : * Output parameters:
5757 : * argpos: argument position for value to be printed. -1 means unspecified.
5758 : * widthpos: argument position for width. Zero means the argument position
5759 : * was unspecified (ie, take the next arg) and -1 means no width
5760 : * argument (width was omitted or specified as a constant).
5761 : * flags: bitmask of flags.
5762 : * width: directly-specified width value. Zero means the width was omitted
5763 : * (note it's not necessary to distinguish this case from an explicit
5764 : * zero width value).
5765 : *
5766 : * The function result is the next character position to be parsed, ie, the
5767 : * location where the type character is/should be.
5768 : *
5769 : * Note parsing invariant: at least one character is known available before
5770 : * string end (end_ptr) at entry, and this is still true at exit.
5771 : */
5772 : static const char *
5773 16732 : text_format_parse_format(const char *start_ptr, const char *end_ptr,
5774 : int *argpos, int *widthpos,
5775 : int *flags, int *width)
5776 : {
5777 16732 : const char *cp = start_ptr;
5778 : int n;
5779 :
5780 : /* set defaults for output parameters */
5781 16732 : *argpos = -1;
5782 16732 : *widthpos = -1;
5783 16732 : *flags = 0;
5784 16732 : *width = 0;
5785 :
5786 : /* try to identify first number */
5787 16732 : if (text_format_parse_digits(&cp, end_ptr, &n))
5788 : {
5789 116 : if (*cp != '$')
5790 : {
5791 : /* Must be just a width and a type, so we're done */
5792 16 : *width = n;
5793 16 : return cp;
5794 : }
5795 : /* The number was argument position */
5796 100 : *argpos = n;
5797 : /* Explicit 0 for argument index is immediately refused */
5798 100 : if (n == 0)
5799 4 : ereport(ERROR,
5800 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
5801 : errmsg("format specifies argument 0, but arguments are numbered from 1")));
5802 96 : ADVANCE_PARSE_POINTER(cp, end_ptr);
5803 : }
5804 :
5805 : /* Handle flags (only minus is supported now) */
5806 33436 : while (*cp == '-')
5807 : {
5808 20 : *flags |= TEXT_FORMAT_FLAG_MINUS;
5809 20 : ADVANCE_PARSE_POINTER(cp, end_ptr);
5810 : }
5811 :
5812 16708 : if (*cp == '*')
5813 : {
5814 : /* Handle indirect width */
5815 32 : ADVANCE_PARSE_POINTER(cp, end_ptr);
5816 32 : if (text_format_parse_digits(&cp, end_ptr, &n))
5817 : {
5818 : /* number in this position must be closed by $ */
5819 28 : if (*cp != '$')
5820 0 : ereport(ERROR,
5821 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
5822 : errmsg("width argument position must be ended by \"$\"")));
5823 : /* The number was width argument position */
5824 28 : *widthpos = n;
5825 : /* Explicit 0 for argument index is immediately refused */
5826 28 : if (n == 0)
5827 4 : ereport(ERROR,
5828 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
5829 : errmsg("format specifies argument 0, but arguments are numbered from 1")));
5830 24 : ADVANCE_PARSE_POINTER(cp, end_ptr);
5831 : }
5832 : else
5833 4 : *widthpos = 0; /* width's argument position is unspecified */
5834 : }
5835 : else
5836 : {
5837 : /* Check for direct width specification */
5838 16676 : if (text_format_parse_digits(&cp, end_ptr, &n))
5839 20 : *width = n;
5840 : }
5841 :
5842 : /* cp should now be pointing at type character */
5843 16700 : return cp;
5844 : }
5845 :
5846 : /*
5847 : * Format a %s, %I, or %L conversion
5848 : */
5849 : static void
5850 16696 : text_format_string_conversion(StringInfo buf, char conversion,
5851 : FmgrInfo *typOutputInfo,
5852 : Datum value, bool isNull,
5853 : int flags, int width)
5854 : {
5855 : char *str;
5856 :
5857 : /* Handle NULL arguments before trying to stringify the value. */
5858 16696 : if (isNull)
5859 : {
5860 44 : if (conversion == 's')
5861 12 : text_format_append_string(buf, "", flags, width);
5862 32 : else if (conversion == 'L')
5863 28 : text_format_append_string(buf, "NULL", flags, width);
5864 4 : else if (conversion == 'I')
5865 4 : ereport(ERROR,
5866 : (errcode(ERRCODE_NULL_VALUE_NOT_ALLOWED),
5867 : errmsg("null values cannot be formatted as an SQL identifier")));
5868 40 : return;
5869 : }
5870 :
5871 : /* Stringify. */
5872 16652 : str = OutputFunctionCall(typOutputInfo, value);
5873 :
5874 : /* Escape. */
5875 16652 : if (conversion == 'I')
5876 : {
5877 : /* quote_identifier may or may not allocate a new string. */
5878 1104 : text_format_append_string(buf, quote_identifier(str), flags, width);
5879 : }
5880 15548 : else if (conversion == 'L')
5881 : {
5882 1018 : char *qstr = quote_literal_cstr(str);
5883 :
5884 1018 : text_format_append_string(buf, qstr, flags, width);
5885 : /* quote_literal_cstr() always allocates a new string */
5886 1018 : pfree(qstr);
5887 : }
5888 : else
5889 14530 : text_format_append_string(buf, str, flags, width);
5890 :
5891 : /* Cleanup. */
5892 16652 : pfree(str);
5893 : }
5894 :
5895 : /*
5896 : * Append str to buf, padding as directed by flags/width
5897 : */
5898 : static void
5899 16692 : text_format_append_string(StringInfo buf, const char *str,
5900 : int flags, int width)
5901 : {
5902 16692 : bool align_to_left = false;
5903 : int len;
5904 :
5905 : /* fast path for typical easy case */
5906 16692 : if (width == 0)
5907 : {
5908 16636 : appendStringInfoString(buf, str);
5909 16636 : return;
5910 : }
5911 :
5912 56 : if (width < 0)
5913 : {
5914 : /* Negative width: implicit '-' flag, then take absolute value */
5915 4 : align_to_left = true;
5916 : /* -INT_MIN is undefined */
5917 4 : if (width <= INT_MIN)
5918 0 : ereport(ERROR,
5919 : (errcode(ERRCODE_NUMERIC_VALUE_OUT_OF_RANGE),
5920 : errmsg("number is out of range")));
5921 4 : width = -width;
5922 : }
5923 52 : else if (flags & TEXT_FORMAT_FLAG_MINUS)
5924 16 : align_to_left = true;
5925 :
5926 56 : len = pg_mbstrlen(str);
5927 56 : if (align_to_left)
5928 : {
5929 : /* left justify */
5930 20 : appendStringInfoString(buf, str);
5931 20 : if (len < width)
5932 20 : appendStringInfoSpaces(buf, width - len);
5933 : }
5934 : else
5935 : {
5936 : /* right justify */
5937 36 : if (len < width)
5938 36 : appendStringInfoSpaces(buf, width - len);
5939 36 : appendStringInfoString(buf, str);
5940 : }
5941 : }
5942 :
5943 : /*
5944 : * text_format_nv - nonvariadic wrapper for text_format function.
5945 : *
5946 : * note: this wrapper is necessary to pass the sanity check in opr_sanity,
5947 : * which checks that all built-in functions that share the implementing C
5948 : * function take the same number of arguments.
5949 : */
5950 : Datum
5951 20 : text_format_nv(PG_FUNCTION_ARGS)
5952 : {
5953 20 : return text_format(fcinfo);
5954 : }
5955 :
5956 : /*
5957 : * Helper function for Levenshtein distance functions. Faster than memcmp(),
5958 : * for this use case.
5959 : */
5960 : static inline bool
5961 0 : rest_of_char_same(const char *s1, const char *s2, int len)
5962 : {
5963 0 : while (len > 0)
5964 : {
5965 0 : len--;
5966 0 : if (s1[len] != s2[len])
5967 0 : return false;
5968 : }
5969 0 : return true;
5970 : }
5971 :
5972 : /* Expand each Levenshtein distance variant */
5973 : #include "levenshtein.c"
5974 : #define LEVENSHTEIN_LESS_EQUAL
5975 : #include "levenshtein.c"
|
