Age Owner TLA Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * varlena.c
4 : * Functions for the variable-length built-in types.
5 : *
6 : * Portions Copyright (c) 1996-2023, PostgreSQL Global Development Group
7 : * Portions Copyright (c) 1994, Regents of the University of California
8 : *
9 : *
10 : * IDENTIFICATION
11 : * src/backend/utils/adt/varlena.c
12 : *
13 : *-------------------------------------------------------------------------
14 : */
15 : #include "postgres.h"
16 :
17 : #include <ctype.h>
18 : #include <limits.h>
19 :
20 : #include "access/detoast.h"
21 : #include "access/toast_compression.h"
22 : #include "catalog/pg_collation.h"
23 : #include "catalog/pg_type.h"
24 : #include "common/hashfn.h"
25 : #include "common/int.h"
26 : #include "common/unicode_norm.h"
27 : #include "funcapi.h"
28 : #include "lib/hyperloglog.h"
29 : #include "libpq/pqformat.h"
30 : #include "miscadmin.h"
31 : #include "nodes/execnodes.h"
32 : #include "parser/scansup.h"
33 : #include "port/pg_bswap.h"
34 : #include "regex/regex.h"
35 : #include "utils/builtins.h"
36 : #include "utils/bytea.h"
37 : #include "utils/guc.h"
38 : #include "utils/lsyscache.h"
39 : #include "utils/memutils.h"
40 : #include "utils/pg_locale.h"
41 : #include "utils/sortsupport.h"
42 : #include "utils/varlena.h"
43 :
44 :
45 : /* GUC variable */
46 : int bytea_output = BYTEA_OUTPUT_HEX;
47 :
48 : typedef struct varlena VarString;
49 :
50 : /*
51 : * State for text_position_* functions.
52 : */
53 : typedef struct
54 : {
55 : bool is_multibyte_char_in_char; /* need to check char boundaries? */
56 :
57 : char *str1; /* haystack string */
58 : char *str2; /* needle string */
59 : int len1; /* string lengths in bytes */
60 : int len2;
61 :
62 : /* Skip table for Boyer-Moore-Horspool search algorithm: */
63 : int skiptablemask; /* mask for ANDing with skiptable subscripts */
64 : int skiptable[256]; /* skip distance for given mismatched char */
65 :
66 : char *last_match; /* pointer to last match in 'str1' */
67 :
68 : /*
69 : * Sometimes we need to convert the byte position of a match to a
70 : * character position. These store the last position that was converted,
71 : * so that on the next call, we can continue from that point, rather than
72 : * count characters from the very beginning.
73 : */
74 : char *refpoint; /* pointer within original haystack string */
75 : int refpos; /* 0-based character offset of the same point */
76 : } TextPositionState;
77 :
78 : typedef struct
79 : {
80 : char *buf1; /* 1st string, or abbreviation original string
81 : * buf */
82 : char *buf2; /* 2nd string, or abbreviation strxfrm() buf */
83 : int buflen1; /* Allocated length of buf1 */
84 : int buflen2; /* Allocated length of buf2 */
85 : int last_len1; /* Length of last buf1 string/strxfrm() input */
86 : int last_len2; /* Length of last buf2 string/strxfrm() blob */
87 : int last_returned; /* Last comparison result (cache) */
88 : bool cache_blob; /* Does buf2 contain strxfrm() blob, etc? */
89 : bool collate_c;
90 : Oid typid; /* Actual datatype (text/bpchar/bytea/name) */
91 : hyperLogLogState abbr_card; /* Abbreviated key cardinality state */
92 : hyperLogLogState full_card; /* Full key cardinality state */
93 : double prop_card; /* Required cardinality proportion */
94 : pg_locale_t locale;
95 : } VarStringSortSupport;
96 :
97 : /*
98 : * Output data for split_text(): we output either to an array or a table.
99 : * tupstore and tupdesc must be set up in advance to output to a table.
100 : */
101 : typedef struct
102 : {
103 : ArrayBuildState *astate;
104 : Tuplestorestate *tupstore;
105 : TupleDesc tupdesc;
106 : } SplitTextOutputData;
107 :
108 : /*
109 : * This should be large enough that most strings will fit, but small enough
110 : * that we feel comfortable putting it on the stack
111 : */
112 : #define TEXTBUFLEN 1024
113 :
114 : #define DatumGetVarStringP(X) ((VarString *) PG_DETOAST_DATUM(X))
115 : #define DatumGetVarStringPP(X) ((VarString *) PG_DETOAST_DATUM_PACKED(X))
116 :
117 : static int varstrfastcmp_c(Datum x, Datum y, SortSupport ssup);
118 : static int bpcharfastcmp_c(Datum x, Datum y, SortSupport ssup);
119 : static int namefastcmp_c(Datum x, Datum y, SortSupport ssup);
120 : static int varlenafastcmp_locale(Datum x, Datum y, SortSupport ssup);
121 : static int namefastcmp_locale(Datum x, Datum y, SortSupport ssup);
122 : static int varstrfastcmp_locale(char *a1p, int len1, char *a2p, int len2, SortSupport ssup);
123 : static Datum varstr_abbrev_convert(Datum original, SortSupport ssup);
124 : static bool varstr_abbrev_abort(int memtupcount, SortSupport ssup);
125 : static int32 text_length(Datum str);
126 : static text *text_catenate(text *t1, text *t2);
127 : static text *text_substring(Datum str,
128 : int32 start,
129 : int32 length,
130 : bool length_not_specified);
131 : static text *text_overlay(text *t1, text *t2, int sp, int sl);
132 : static int text_position(text *t1, text *t2, Oid collid);
133 : static void text_position_setup(text *t1, text *t2, Oid collid, TextPositionState *state);
134 : static bool text_position_next(TextPositionState *state);
135 : static char *text_position_next_internal(char *start_ptr, TextPositionState *state);
136 : static char *text_position_get_match_ptr(TextPositionState *state);
137 : static int text_position_get_match_pos(TextPositionState *state);
138 : static void text_position_cleanup(TextPositionState *state);
139 : static void check_collation_set(Oid collid);
140 : static int text_cmp(text *arg1, text *arg2, Oid collid);
141 : static bytea *bytea_catenate(bytea *t1, bytea *t2);
142 : static bytea *bytea_substring(Datum str,
143 : int S,
144 : int L,
145 : bool length_not_specified);
146 : static bytea *bytea_overlay(bytea *t1, bytea *t2, int sp, int sl);
147 : static void appendStringInfoText(StringInfo str, const text *t);
148 : static bool split_text(FunctionCallInfo fcinfo, SplitTextOutputData *tstate);
149 : static void split_text_accum_result(SplitTextOutputData *tstate,
150 : text *field_value,
151 : text *null_string,
152 : Oid collation);
153 : static text *array_to_text_internal(FunctionCallInfo fcinfo, ArrayType *v,
154 : const char *fldsep, const char *null_string);
155 : static StringInfo makeStringAggState(FunctionCallInfo fcinfo);
156 : static bool text_format_parse_digits(const char **ptr, const char *end_ptr,
157 : int *value);
158 : static const char *text_format_parse_format(const char *start_ptr,
159 : const char *end_ptr,
160 : int *argpos, int *widthpos,
161 : int *flags, int *width);
162 : static void text_format_string_conversion(StringInfo buf, char conversion,
163 : FmgrInfo *typOutputInfo,
164 : Datum value, bool isNull,
165 : int flags, int width);
166 : static void text_format_append_string(StringInfo buf, const char *str,
167 : int flags, int width);
168 :
169 :
170 : /*****************************************************************************
171 : * CONVERSION ROUTINES EXPORTED FOR USE BY C CODE *
172 : *****************************************************************************/
173 :
174 : /*
175 : * cstring_to_text
5493 tgl 176 ECB : *
177 : * Create a text value from a null-terminated C string.
178 : *
179 : * The new text value is freshly palloc'd with a full-size VARHDR.
180 : */
181 : text *
5493 tgl 182 GIC 12430865 : cstring_to_text(const char *s)
183 : {
184 12430865 : return cstring_to_text_with_len(s, strlen(s));
185 : }
186 :
187 : /*
5493 tgl 188 ECB : * cstring_to_text_with_len
189 : *
190 : * Same as cstring_to_text except the caller specifies the string length;
191 : * the string need not be null_terminated.
192 : */
193 : text *
5493 tgl 194 GIC 14970995 : cstring_to_text_with_len(const char *s, int len)
5493 tgl 195 ECB : {
5493 tgl 196 GIC 14970995 : text *result = (text *) palloc(len + VARHDRSZ);
197 :
198 14970995 : SET_VARSIZE(result, len + VARHDRSZ);
199 14970995 : memcpy(VARDATA(result), s, len);
200 :
201 14970995 : return result;
202 : }
203 :
204 : /*
205 : * text_to_cstring
206 : *
207 : * Create a palloc'd, null-terminated C string from a text value.
208 : *
5493 tgl 209 ECB : * We support being passed a compressed or toasted text value.
210 : * This is a bit bogus since such values shouldn't really be referred to as
211 : * "text *", but it seems useful for robustness. If we didn't handle that
212 : * case here, we'd need another routine that did, anyway.
213 : */
214 : char *
5493 tgl 215 GIC 8999458 : text_to_cstring(const text *t)
5493 tgl 216 ECB : {
217 : /* must cast away the const, unfortunately */
1627 peter_e 218 CBC 8999458 : text *tunpacked = pg_detoast_datum_packed(unconstify(text *, t));
5493 tgl 219 GIC 8999458 : int len = VARSIZE_ANY_EXHDR(tunpacked);
5493 tgl 220 ECB : char *result;
221 :
5493 tgl 222 GIC 8999458 : result = (char *) palloc(len + 1);
5493 tgl 223 CBC 8999458 : memcpy(result, VARDATA_ANY(tunpacked), len);
5493 tgl 224 GIC 8999458 : result[len] = '\0';
225 :
226 8999458 : if (tunpacked != t)
227 64653 : pfree(tunpacked);
228 :
229 8999458 : return result;
230 : }
231 :
232 : /*
233 : * text_to_cstring_buffer
234 : *
235 : * Copy a text value into a caller-supplied buffer of size dst_len.
236 : *
237 : * The text string is truncated if necessary to fit. The result is
238 : * guaranteed null-terminated (unless dst_len == 0).
239 : *
5493 tgl 240 ECB : * We support being passed a compressed or toasted text value.
241 : * This is a bit bogus since such values shouldn't really be referred to as
242 : * "text *", but it seems useful for robustness. If we didn't handle that
243 : * case here, we'd need another routine that did, anyway.
244 : */
245 : void
5493 tgl 246 CBC 320 : text_to_cstring_buffer(const text *src, char *dst, size_t dst_len)
247 : {
5493 tgl 248 ECB : /* must cast away the const, unfortunately */
1627 peter_e 249 CBC 320 : text *srcunpacked = pg_detoast_datum_packed(unconstify(text *, src));
5493 tgl 250 320 : size_t src_len = VARSIZE_ANY_EXHDR(srcunpacked);
251 :
5493 tgl 252 GBC 320 : if (dst_len > 0)
5493 tgl 253 ECB : {
5493 tgl 254 CBC 320 : dst_len--;
5493 tgl 255 GIC 320 : if (dst_len >= src_len)
256 320 : dst_len = src_len;
2118 tgl 257 ECB : else /* ensure truncation is encoding-safe */
5493 tgl 258 UBC 0 : dst_len = pg_mbcliplen(VARDATA_ANY(srcunpacked), src_len, dst_len);
5493 tgl 259 CBC 320 : memcpy(dst, VARDATA_ANY(srcunpacked), dst_len);
5493 tgl 260 GIC 320 : dst[dst_len] = '\0';
261 : }
262 :
263 320 : if (srcunpacked != src)
5493 tgl 264 UIC 0 : pfree(srcunpacked);
5493 tgl 265 GIC 320 : }
266 :
267 :
268 : /*****************************************************************************
269 : * USER I/O ROUTINES *
270 : *****************************************************************************/
271 :
272 :
273 : #define VAL(CH) ((CH) - '0')
274 : #define DIG(VAL) ((VAL) + '0')
275 :
276 : /*
277 : * byteain - converts from printable representation of byte array
278 : *
279 : * Non-printable characters must be passed as '\nnn' (octal) and are
280 : * converted to internal form. '\' must be passed as '\\'.
281 : * ereport(ERROR, ...) if bad form.
9770 scrappy 282 ECB : *
283 : * BUGS:
6251 neilc 284 : * The input is scanned twice.
9345 bruce 285 : * The error checking of input is minimal.
286 : */
287 : Datum
8289 tgl 288 GIC 131417 : byteain(PG_FUNCTION_ARGS)
289 : {
290 131417 : char *inputText = PG_GETARG_CSTRING(0);
116 tgl 291 GNC 131417 : Node *escontext = fcinfo->context;
292 : char *tp;
9229 lockhart 293 ECB : char *rp;
294 : int bc;
8416 tgl 295 : bytea *result;
296 :
4996 297 : /* Recognize hex input */
4996 tgl 298 CBC 131417 : if (inputText[0] == '\\' && inputText[1] == 'x')
4996 tgl 299 ECB : {
4790 bruce 300 GIC 405 : size_t len = strlen(inputText);
4996 tgl 301 ECB :
598 michael 302 GIC 405 : bc = (len - 2) / 2 + VARHDRSZ; /* maximum possible length */
4996 tgl 303 CBC 405 : result = palloc(bc);
116 tgl 304 GNC 405 : bc = hex_decode_safe(inputText + 2, len - 2, VARDATA(result),
305 : escontext);
2118 tgl 306 GIC 399 : SET_VARSIZE(result, bc + VARHDRSZ); /* actual length */
307 :
4996 tgl 308 CBC 399 : PG_RETURN_BYTEA_P(result);
309 : }
4996 tgl 310 ECB :
311 : /* Else, it's the traditional escaped style */
4996 tgl 312 CBC 2176934 : for (bc = 0, tp = inputText; *tp != '\0'; bc++)
8289 tgl 313 ECB : {
7877 bruce 314 CBC 2045928 : if (tp[0] != '\\')
315 2045424 : tp++;
7836 316 504 : else if ((tp[0] == '\\') &&
317 504 : (tp[1] >= '0' && tp[1] <= '3') &&
318 498 : (tp[2] >= '0' && tp[2] <= '7') &&
7836 bruce 319 GBC 498 : (tp[3] >= '0' && tp[3] <= '7'))
7877 bruce 320 GIC 498 : tp += 4;
7836 321 6 : else if ((tp[0] == '\\') &&
322 6 : (tp[1] == '\\'))
7877 bruce 323 UIC 0 : tp += 2;
324 : else
7877 bruce 325 ECB : {
326 : /*
327 : * one backslash, not followed by another or ### valid octal
328 : */
116 tgl 329 GNC 6 : ereturn(escontext, (Datum) 0,
330 : (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
2272 alvherre 331 ECB : errmsg("invalid input syntax for type %s", "bytea")));
332 : }
8289 tgl 333 : }
334 :
4996 tgl 335 GIC 131006 : bc += VARHDRSZ;
5050 bruce 336 ECB :
4996 tgl 337 CBC 131006 : result = (bytea *) palloc(bc);
338 131006 : SET_VARSIZE(result, bc);
339 :
8289 340 131006 : tp = inputText;
7275 341 131006 : rp = VARDATA(result);
9345 bruce 342 2176913 : while (*tp != '\0')
8289 tgl 343 ECB : {
7877 bruce 344 CBC 2045907 : if (tp[0] != '\\')
9345 345 2045409 : *rp++ = *tp++;
7836 bruce 346 GIC 498 : else if ((tp[0] == '\\') &&
7836 bruce 347 CBC 498 : (tp[1] >= '0' && tp[1] <= '3') &&
348 498 : (tp[2] >= '0' && tp[2] <= '7') &&
349 498 : (tp[3] >= '0' && tp[3] <= '7'))
9345 bruce 350 ECB : {
4996 tgl 351 CBC 498 : bc = VAL(tp[1]);
4996 tgl 352 GIC 498 : bc <<= 3;
4996 tgl 353 CBC 498 : bc += VAL(tp[2]);
4996 tgl 354 GIC 498 : bc <<= 3;
4996 tgl 355 GBC 498 : *rp++ = bc + VAL(tp[3]);
5050 bruce 356 EUB :
7877 bruce 357 GIC 498 : tp += 4;
7877 bruce 358 EUB : }
7836 bruce 359 UBC 0 : else if ((tp[0] == '\\') &&
7836 bruce 360 UIC 0 : (tp[1] == '\\'))
361 : {
7877 362 0 : *rp++ = '\\';
363 0 : tp += 2;
364 : }
365 : else
7877 bruce 366 EUB : {
367 : /*
368 : * We should never get here. The first pass should not allow it.
369 : */
116 tgl 370 UNC 0 : ereturn(escontext, (Datum) 0,
371 : (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
2272 alvherre 372 ECB : errmsg("invalid input syntax for type %s", "bytea")));
373 : }
374 : }
375 :
8289 tgl 376 GIC 131006 : PG_RETURN_BYTEA_P(result);
377 : }
378 :
379 : /*
380 : * byteaout - converts to printable representation of byte array
381 : *
4996 tgl 382 ECB : * In the traditional escaped format, non-printable characters are
383 : * printed as '\nnn' (octal) and '\' as '\\'.
9770 scrappy 384 : */
385 : Datum
8289 tgl 386 GIC 6677 : byteaout(PG_FUNCTION_ARGS)
387 : {
5847 tgl 388 CBC 6677 : bytea *vlena = PG_GETARG_BYTEA_PP(0);
389 : char *result;
390 : char *rp;
4996 tgl 391 ECB :
4996 tgl 392 CBC 6677 : if (bytea_output == BYTEA_OUTPUT_HEX)
4996 tgl 393 ECB : {
394 : /* Print hex format */
598 michael 395 GIC 6486 : rp = result = palloc(VARSIZE_ANY_EXHDR(vlena) * 2 + 2 + 1);
4996 tgl 396 CBC 6486 : *rp++ = '\\';
4996 tgl 397 GIC 6486 : *rp++ = 'x';
598 michael 398 6486 : rp += hex_encode(VARDATA_ANY(vlena), VARSIZE_ANY_EXHDR(vlena), rp);
399 : }
4996 tgl 400 191 : else if (bytea_output == BYTEA_OUTPUT_ESCAPE)
401 : {
402 : /* Print traditional escaped format */
4790 bruce 403 ECB : char *vp;
1097 tgl 404 : uint64 len;
4790 bruce 405 : int i;
406 :
4790 bruce 407 CBC 191 : len = 1; /* empty string has 1 char */
4790 bruce 408 GBC 191 : vp = VARDATA_ANY(vlena);
4790 bruce 409 CBC 108825 : for (i = VARSIZE_ANY_EXHDR(vlena); i != 0; i--, vp++)
9345 bruce 410 ECB : {
4790 bruce 411 GIC 108634 : if (*vp == '\\')
4790 bruce 412 LBC 0 : len += 2;
4790 bruce 413 GIC 108634 : else if ((unsigned char) *vp < 0x20 || (unsigned char) *vp > 0x7e)
414 246 : len += 4;
415 : else
416 108388 : len++;
417 : }
418 :
419 : /*
1097 tgl 420 ECB : * In principle len can't overflow uint32 if the input fit in 1GB, but
1097 tgl 421 EUB : * for safety let's check rather than relying on palloc's internal
422 : * check.
423 : */
1097 tgl 424 CBC 191 : if (len > MaxAllocSize)
1097 tgl 425 UIC 0 : ereport(ERROR,
1097 tgl 426 ECB : (errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
427 : errmsg_internal("result of bytea output conversion is too large")));
4790 bruce 428 GIC 191 : rp = result = (char *) palloc(len);
1097 tgl 429 ECB :
4790 bruce 430 GIC 191 : vp = VARDATA_ANY(vlena);
4790 bruce 431 GBC 108825 : for (i = VARSIZE_ANY_EXHDR(vlena); i != 0; i--, vp++)
9345 bruce 432 EUB : {
4790 bruce 433 GIC 108634 : if (*vp == '\\')
4790 bruce 434 ECB : {
4790 bruce 435 LBC 0 : *rp++ = '\\';
4790 bruce 436 UIC 0 : *rp++ = '\\';
437 : }
4790 bruce 438 CBC 108634 : else if ((unsigned char) *vp < 0x20 || (unsigned char) *vp > 0x7e)
439 246 : {
4790 bruce 440 ECB : int val; /* holds unprintable chars */
441 :
4790 bruce 442 CBC 246 : val = *vp;
443 246 : rp[0] = '\\';
444 246 : rp[3] = DIG(val & 07);
445 246 : val >>= 3;
4790 bruce 446 GIC 246 : rp[2] = DIG(val & 07);
447 246 : val >>= 3;
4790 bruce 448 CBC 246 : rp[1] = DIG(val & 03);
4790 bruce 449 GIC 246 : rp += 4;
450 : }
451 : else
452 108388 : *rp++ = *vp;
9345 bruce 453 EUB : }
454 : }
455 : else
456 : {
4996 tgl 457 LBC 0 : elog(ERROR, "unrecognized bytea_output setting: %d",
4996 tgl 458 ECB : bytea_output);
459 : rp = result = NULL; /* keep compiler quiet */
460 : }
9345 bruce 461 GIC 6677 : *rp = '\0';
8289 tgl 462 6677 : PG_RETURN_CSTRING(result);
463 : }
464 :
7275 tgl 465 ECB : /*
466 : * bytearecv - converts external binary format to bytea
467 : */
468 : Datum
7275 tgl 469 GIC 506 : bytearecv(PG_FUNCTION_ARGS)
470 : {
7275 tgl 471 CBC 506 : StringInfo buf = (StringInfo) PG_GETARG_POINTER(0);
7275 tgl 472 ECB : bytea *result;
473 : int nbytes;
474 :
7275 tgl 475 CBC 506 : nbytes = buf->len - buf->cursor;
7275 tgl 476 GIC 506 : result = (bytea *) palloc(nbytes + VARHDRSZ);
5885 477 506 : SET_VARSIZE(result, nbytes + VARHDRSZ);
7275 478 506 : pq_copymsgbytes(buf, VARDATA(result), nbytes);
479 506 : PG_RETURN_BYTEA_P(result);
480 : }
481 :
482 : /*
483 : * byteasend - converts bytea to binary format
7275 tgl 484 ECB : *
485 : * This is a special case: just copy the input...
486 : */
487 : Datum
7275 tgl 488 CBC 2811 : byteasend(PG_FUNCTION_ARGS)
489 : {
7275 tgl 490 GIC 2811 : bytea *vlena = PG_GETARG_BYTEA_P_COPY(0);
491 :
7275 tgl 492 CBC 2811 : PG_RETURN_BYTEA_P(vlena);
493 : }
494 :
495 : Datum
4013 peter_e 496 46387 : bytea_string_agg_transfn(PG_FUNCTION_ARGS)
497 : {
498 : StringInfo state;
4125 rhaas 499 ECB :
4125 rhaas 500 GIC 46387 : state = PG_ARGISNULL(0) ? NULL : (StringInfo) PG_GETARG_POINTER(0);
4125 rhaas 501 ECB :
502 : /* Append the value unless null, preceding it with the delimiter. */
4125 rhaas 503 GIC 46387 : if (!PG_ARGISNULL(1))
504 : {
505 38887 : bytea *value = PG_GETARG_BYTEA_PP(1);
76 drowley 506 GNC 38887 : bool isfirst = false;
507 :
508 : /*
509 : * You might think we can just throw away the first delimiter, however
510 : * we must keep it as we may be a parallel worker doing partial
511 : * aggregation building a state to send to the main process. We need
512 : * to keep the delimiter of every aggregation so that the combine
513 : * function can properly join up the strings of two separately
514 : * partially aggregated results. The first delimiter is only stripped
515 : * off in the final function. To know how much to strip off the front
516 : * of the string, we store the length of the first delimiter in the
517 : * StringInfo's cursor field, which we don't otherwise need here.
518 : */
4125 rhaas 519 GIC 38887 : if (state == NULL)
520 : {
521 93 : state = makeStringAggState(fcinfo);
76 drowley 522 GNC 93 : isfirst = true;
523 : }
524 :
525 38887 : if (!PG_ARGISNULL(2))
526 : {
4013 peter_e 527 GIC 38881 : bytea *delim = PG_GETARG_BYTEA_PP(2);
528 :
76 drowley 529 GNC 38881 : appendBinaryStringInfo(state, VARDATA_ANY(delim),
530 38881 : VARSIZE_ANY_EXHDR(delim));
531 38881 : if (isfirst)
532 90 : state->cursor = VARSIZE_ANY_EXHDR(delim);
4013 peter_e 533 ECB : }
534 :
76 drowley 535 GNC 38887 : appendBinaryStringInfo(state, VARDATA_ANY(value),
536 38887 : VARSIZE_ANY_EXHDR(value));
4125 rhaas 537 ECB : }
538 :
539 : /*
4013 peter_e 540 : * The transition type for string_agg() is declared to be "internal",
541 : * which is a pass-by-value type the same size as a pointer.
4125 rhaas 542 : */
76 drowley 543 GNC 46387 : if (state)
544 46363 : PG_RETURN_POINTER(state);
545 24 : PG_RETURN_NULL();
4125 rhaas 546 ECB : }
547 :
548 : Datum
4013 peter_e 549 CBC 76 : bytea_string_agg_finalfn(PG_FUNCTION_ARGS)
550 : {
551 : StringInfo state;
4125 rhaas 552 ECB :
553 : /* cannot be called directly because of internal-type argument */
4125 rhaas 554 GIC 76 : Assert(AggCheckCallContext(fcinfo, NULL));
555 :
556 76 : state = PG_ARGISNULL(0) ? NULL : (StringInfo) PG_GETARG_POINTER(0);
557 :
558 76 : if (state != NULL)
559 : {
560 : /* As per comment in transfn, strip data before the cursor position */
4125 rhaas 561 ECB : bytea *result;
76 drowley 562 GNC 73 : int strippedlen = state->len - state->cursor;
4125 rhaas 563 ECB :
76 drowley 564 GNC 73 : result = (bytea *) palloc(strippedlen + VARHDRSZ);
565 73 : SET_VARSIZE(result, strippedlen + VARHDRSZ);
566 73 : memcpy(VARDATA(result), &state->data[state->cursor], strippedlen);
4125 rhaas 567 GIC 73 : PG_RETURN_BYTEA_P(result);
4125 rhaas 568 ECB : }
569 : else
4125 rhaas 570 GIC 3 : PG_RETURN_NULL();
571 : }
572 :
9770 scrappy 573 ECB : /*
574 : * textin - converts "..." to internal representation
575 : */
576 : Datum
8313 tgl 577 CBC 8942398 : textin(PG_FUNCTION_ARGS)
578 : {
8313 tgl 579 GIC 8942398 : char *inputText = PG_GETARG_CSTRING(0);
580 :
5493 tgl 581 CBC 8942398 : PG_RETURN_TEXT_P(cstring_to_text(inputText));
582 : }
9770 scrappy 583 ECB :
584 : /*
9345 bruce 585 : * textout - converts internal representation to "..."
9770 scrappy 586 : */
587 : Datum
8313 tgl 588 GIC 4169602 : textout(PG_FUNCTION_ARGS)
9770 scrappy 589 ECB : {
5493 tgl 590 GIC 4169602 : Datum txt = PG_GETARG_DATUM(0);
591 :
592 4169602 : PG_RETURN_CSTRING(TextDatumGetCString(txt));
593 : }
594 :
595 : /*
7275 tgl 596 ECB : * textrecv - converts external binary format to text
597 : */
598 : Datum
7275 tgl 599 GIC 53360 : textrecv(PG_FUNCTION_ARGS)
7275 tgl 600 ECB : {
7275 tgl 601 GIC 53360 : StringInfo buf = (StringInfo) PG_GETARG_POINTER(0);
602 : text *result;
603 : char *str;
604 : int nbytes;
605 :
606 53360 : str = pq_getmsgtext(buf, buf->len - buf->cursor, &nbytes);
6482 tgl 607 ECB :
5493 tgl 608 GIC 53360 : result = cstring_to_text_with_len(str, nbytes);
7275 tgl 609 CBC 53360 : pfree(str);
7275 tgl 610 GIC 53360 : PG_RETURN_TEXT_P(result);
7275 tgl 611 ECB : }
612 :
613 : /*
614 : * textsend - converts text to binary format
615 : */
616 : Datum
7275 tgl 617 GIC 34115 : textsend(PG_FUNCTION_ARGS)
7275 tgl 618 ECB : {
5847 tgl 619 GIC 34115 : text *t = PG_GETARG_TEXT_PP(0);
7275 tgl 620 ECB : StringInfoData buf;
621 :
7275 tgl 622 GIC 34115 : pq_begintypsend(&buf);
5847 623 34115 : pq_sendtext(&buf, VARDATA_ANY(t), VARSIZE_ANY_EXHDR(t));
7275 624 34115 : PG_RETURN_BYTEA_P(pq_endtypsend(&buf));
7275 tgl 625 ECB : }
626 :
9770 scrappy 627 :
7655 bruce 628 : /*
629 : * unknownin - converts "..." to internal representation
630 : */
631 : Datum
7655 bruce 632 UIC 0 : unknownin(PG_FUNCTION_ARGS)
633 : {
6523 tgl 634 0 : char *str = PG_GETARG_CSTRING(0);
635 :
6523 tgl 636 ECB : /* representation is same as cstring */
6523 tgl 637 UIC 0 : PG_RETURN_CSTRING(pstrdup(str));
7655 bruce 638 ECB : }
639 :
640 : /*
641 : * unknownout - converts internal representation to "..."
642 : */
643 : Datum
7655 bruce 644 GIC 340 : unknownout(PG_FUNCTION_ARGS)
645 : {
646 : /* representation is same as cstring */
6523 tgl 647 340 : char *str = PG_GETARG_CSTRING(0);
648 :
649 340 : PG_RETURN_CSTRING(pstrdup(str));
650 : }
7655 bruce 651 EUB :
652 : /*
7275 tgl 653 : * unknownrecv - converts external binary format to unknown
654 : */
655 : Datum
7275 tgl 656 UBC 0 : unknownrecv(PG_FUNCTION_ARGS)
657 : {
7275 tgl 658 UIC 0 : StringInfo buf = (StringInfo) PG_GETARG_POINTER(0);
659 : char *str;
660 : int nbytes;
661 :
6523 662 0 : str = pq_getmsgtext(buf, buf->len - buf->cursor, &nbytes);
6523 tgl 663 ECB : /* representation is same as cstring */
6523 tgl 664 UIC 0 : PG_RETURN_CSTRING(str);
665 : }
7275 tgl 666 ECB :
667 : /*
668 : * unknownsend - converts unknown to binary format
669 : */
670 : Datum
7275 tgl 671 UIC 0 : unknownsend(PG_FUNCTION_ARGS)
672 : {
673 : /* representation is same as cstring */
6523 674 0 : char *str = PG_GETARG_CSTRING(0);
6523 tgl 675 EUB : StringInfoData buf;
676 :
6523 tgl 677 UBC 0 : pq_begintypsend(&buf);
6523 tgl 678 UIC 0 : pq_sendtext(&buf, str, strlen(str));
679 0 : PG_RETURN_BYTEA_P(pq_endtypsend(&buf));
680 : }
7275 tgl 681 EUB :
682 :
9770 scrappy 683 : /* ========== PUBLIC ROUTINES ========== */
684 :
685 : /*
686 : * textlen -
687 : * returns the logical length of a text*
688 : * (which is less than the VARSIZE of the text*)
689 : */
8312 tgl 690 : Datum
8312 tgl 691 GIC 215278 : textlen(PG_FUNCTION_ARGS)
692 : {
7008 tgl 693 GBC 215278 : Datum str = PG_GETARG_DATUM(0);
694 :
695 : /* try to avoid decompressing argument */
696 215278 : PG_RETURN_INT32(text_length(str));
7535 bruce 697 EUB : }
8053 698 :
699 : /*
700 : * text_length -
701 : * Does the real work for textlen()
702 : *
703 : * This is broken out so it can be called directly by other string processing
704 : * functions. Note that the argument is passed as a Datum, to indicate that
705 : * it may still be in compressed form. We can avoid decompressing it at all
706 : * in some cases.
707 : */
708 : static int32
7535 bruce 709 GIC 215284 : text_length(Datum str)
7535 bruce 710 ECB : {
711 : /* fastpath when max encoding length is one */
7535 bruce 712 CBC 215284 : if (pg_database_encoding_max_length() == 1)
7535 bruce 713 GIC 10 : PG_RETURN_INT32(toast_raw_datum_size(str) - VARHDRSZ);
714 : else
7535 bruce 715 ECB : {
5847 tgl 716 GIC 215274 : text *t = DatumGetTextPP(str);
717 :
5624 bruce 718 215274 : PG_RETURN_INT32(pg_mbstrlen_with_len(VARDATA_ANY(t),
719 : VARSIZE_ANY_EXHDR(t)));
720 : }
721 : }
722 :
723 : /*
724 : * textoctetlen -
725 : * returns the physical length of a text*
726 : * (which is less than the VARSIZE of the text*)
727 : */
8312 tgl 728 ECB : Datum
8312 tgl 729 GIC 35 : textoctetlen(PG_FUNCTION_ARGS)
730 : {
7008 tgl 731 CBC 35 : Datum str = PG_GETARG_DATUM(0);
7008 tgl 732 ECB :
733 : /* We need not detoast the input at all */
7008 tgl 734 GIC 35 : PG_RETURN_INT32(toast_raw_datum_size(str) - VARHDRSZ);
8312 tgl 735 ECB : }
736 :
9770 scrappy 737 : /*
738 : * textcat -
739 : * takes two text* and returns a text* that is the concatenation of
740 : * the two.
741 : *
742 : * Rewritten by Sapa, sapa@hq.icb.chel.su. 8-Jul-96.
743 : * Updated by Thomas, Thomas.Lockhart@jpl.nasa.gov 1997-07-10.
744 : * Allocate space for output in all cases.
745 : * XXX - thomas 1997-07-10
746 : */
747 : Datum
8312 tgl 748 CBC 1335495 : textcat(PG_FUNCTION_ARGS)
749 : {
5847 750 1335495 : text *t1 = PG_GETARG_TEXT_PP(0);
5847 tgl 751 GIC 1335495 : text *t2 = PG_GETARG_TEXT_PP(1);
752 :
4822 tgl 753 CBC 1335495 : PG_RETURN_TEXT_P(text_catenate(t1, t2));
754 : }
755 :
756 : /*
757 : * text_catenate
758 : * Guts of textcat(), broken out so it can be used by other functions
759 : *
760 : * Arguments can be in short-header form, but not compressed or out-of-line
761 : */
762 : static text *
4822 tgl 763 GIC 1335535 : text_catenate(text *t1, text *t2)
764 : {
765 : text *result;
766 : int len1,
9344 bruce 767 ECB : len2,
768 : len;
8312 tgl 769 : char *ptr;
9770 scrappy 770 :
5847 tgl 771 GIC 1335535 : len1 = VARSIZE_ANY_EXHDR(t1);
4822 tgl 772 CBC 1335535 : len2 = VARSIZE_ANY_EXHDR(t2);
773 :
774 : /* paranoia ... probably should throw error instead? */
9345 bruce 775 GIC 1335535 : if (len1 < 0)
9345 bruce 776 UIC 0 : len1 = 0;
9345 bruce 777 GIC 1335535 : if (len2 < 0)
9345 bruce 778 UIC 0 : len2 = 0;
779 :
9101 lockhart 780 GIC 1335535 : len = len1 + len2 + VARHDRSZ;
8312 tgl 781 1335535 : result = (text *) palloc(len);
9101 lockhart 782 ECB :
783 : /* Set size of result string... */
5885 tgl 784 GIC 1335535 : SET_VARSIZE(result, len);
785 :
786 : /* Fill data field of result string... */
9345 bruce 787 1335535 : ptr = VARDATA(result);
9101 lockhart 788 1335535 : if (len1 > 0)
5847 tgl 789 1333907 : memcpy(ptr, VARDATA_ANY(t1), len1);
9101 lockhart 790 CBC 1335535 : if (len2 > 0)
5847 tgl 791 1335430 : memcpy(ptr + len1, VARDATA_ANY(t2), len2);
792 :
4822 tgl 793 GIC 1335535 : return result;
8312 tgl 794 ECB : }
9229 lockhart 795 EUB :
5996 tgl 796 ECB : /*
5996 tgl 797 EUB : * charlen_to_bytelen()
798 : * Compute the number of bytes occupied by n characters starting at *p
5996 tgl 799 ECB : *
800 : * It is caller's responsibility that there actually are n characters;
801 : * the string need not be null-terminated.
802 : */
803 : static int
5996 tgl 804 GIC 5457 : charlen_to_bytelen(const char *p, int n)
805 : {
5996 tgl 806 CBC 5457 : if (pg_database_encoding_max_length() == 1)
5996 tgl 807 ECB : {
808 : /* Optimization for single-byte encodings */
5996 tgl 809 LBC 0 : return n;
5996 tgl 810 ECB : }
811 : else
812 : {
813 : const char *s;
814 :
5996 tgl 815 GIC 2960165 : for (s = p; n > 0; n--)
816 2954708 : s += pg_mblen(s);
817 :
818 5457 : return s - p;
819 : }
820 : }
821 :
822 : /*
9229 lockhart 823 ECB : * text_substr()
824 : * Return a substring starting at the specified position.
825 : * - thomas 1997-12-31
826 : *
827 : * Input:
9173 bruce 828 EUB : * - string
829 : * - starting position (is one-based)
830 : * - string length
831 : *
832 : * If the starting position is zero or less, then return from the start of the string
833 : * adjusting the length to be consistent with the "negative start" per SQL.
9229 lockhart 834 ECB : * If the length is less than zero, return the remaining string.
835 : *
836 : * Added multibyte support.
9113 scrappy 837 : * - Tatsuo Ishii 1998-4-21
838 : * Changed behavior if starting position is less than one to conform to SQL behavior.
839 : * Formerly returned the entire string; now returns a portion.
840 : * - Thomas Lockhart 1998-12-10
841 : * Now uses faster TOAST-slicing interface
842 : * - John Gray 2002-02-22
843 : * Remove "#ifdef MULTIBYTE" and test for encoding_max_length instead. Change
844 : * behaviors conflicting with SQL to meet SQL (if E = S + L < S throw
845 : * error; if E < 1, return '', not entire string). Fixed MB related bug when
846 : * S > LC and < LC + 4 sometimes garbage characters are returned.
847 : * - Joe Conway 2002-08-10
848 : */
849 : Datum
8335 tgl 850 GIC 374965 : text_substr(PG_FUNCTION_ARGS)
851 : {
7535 bruce 852 374965 : PG_RETURN_TEXT_P(text_substring(PG_GETARG_DATUM(0),
853 : PG_GETARG_INT32(1),
854 : PG_GETARG_INT32(2),
855 : false));
856 : }
857 :
858 : /*
859 : * text_substr_no_len -
860 : * Wrapper to avoid opr_sanity failure due to
861 : * one function accepting a different number of args.
862 : */
863 : Datum
864 24 : text_substr_no_len(PG_FUNCTION_ARGS)
865 : {
866 24 : PG_RETURN_TEXT_P(text_substring(PG_GETARG_DATUM(0),
867 : PG_GETARG_INT32(1),
868 : -1, true));
7535 bruce 869 ECB : }
870 :
871 : /*
872 : * text_substring -
873 : * Does the real work for text_substr() and text_substr_no_len()
874 : *
875 : * This is broken out so it can be called directly by other string processing
876 : * functions. Note that the argument is passed as a Datum, to indicate that
877 : * it may still be in compressed/toasted form. We can avoid detoasting all
878 : * of it in some cases.
879 : *
880 : * The result is always a freshly palloc'd datum.
881 : */
882 : static text *
7535 bruce 883 CBC 394913 : text_substring(Datum str, int32 start, int32 length, bool length_not_specified)
884 : {
885 394913 : int32 eml = pg_database_encoding_max_length();
7522 bruce 886 GIC 394913 : int32 S = start; /* start position */
887 : int32 S1; /* adjusted start position */
888 : int32 L1; /* adjusted substring length */
889 : int32 E; /* end position */
890 :
891 : /*
892 : * SQL99 says S can be zero or negative, but we still must fetch from the
893 : * start of the string.
894 : */
825 tgl 895 394913 : S1 = Max(S, 1);
896 :
897 : /* life is easy if the encoding max length is 1 */
7535 bruce 898 394913 : if (eml == 1)
899 : {
2118 tgl 900 6 : if (length_not_specified) /* special case - get length to end of
901 : * string */
7535 bruce 902 LBC 0 : L1 = -1;
825 tgl 903 GIC 6 : else if (length < 0)
825 tgl 904 ECB : {
905 : /* SQL99 says to throw an error for E < S, i.e., negative length */
825 tgl 906 UIC 0 : ereport(ERROR,
907 : (errcode(ERRCODE_SUBSTRING_ERROR),
908 : errmsg("negative substring length not allowed")));
909 : L1 = -1; /* silence stupider compilers */
910 : }
825 tgl 911 GIC 6 : else if (pg_add_s32_overflow(S, length, &E))
912 : {
913 : /*
825 tgl 914 ECB : * L could be large enough for S + L to overflow, in which case
915 : * the substring must run to end of string.
916 : */
825 tgl 917 LBC 0 : L1 = -1;
918 : }
825 tgl 919 ECB : else
920 : {
7522 bruce 921 EUB : /*
6385 bruce 922 ECB : * A zero or negative value for the end position can happen if the
923 : * start was negative or one. SQL99 says to return a zero-length
924 : * string.
7535 bruce 925 EUB : */
7535 bruce 926 GIC 6 : if (E < 1)
5493 tgl 927 UIC 0 : return cstring_to_text("");
928 :
7535 bruce 929 GIC 6 : L1 = E - S1;
7535 bruce 930 ECB : }
931 :
932 : /*
933 : * If the start position is past the end of the string, SQL99 says to
934 : * return a zero-length string -- DatumGetTextPSlice() will do that
935 : * for us. We need only convert S1 to zero-based starting position.
7535 bruce 936 EUB : */
7535 bruce 937 GIC 6 : return DatumGetTextPSlice(str, S1 - 1, L1);
938 : }
939 394907 : else if (eml > 1)
940 : {
941 : /*
942 : * When encoding max length is > 1, we can't get LC without
943 : * detoasting, so we'll grab a conservatively large slice now and go
944 : * back later to do the right thing
7535 bruce 945 ECB : */
7535 bruce 946 EUB : int32 slice_start;
947 : int32 slice_size;
7535 bruce 948 ECB : int32 slice_strlen;
949 : text *slice;
950 : int32 E1;
951 : int32 i;
952 : char *p;
953 : char *s;
954 : text *ret;
955 :
956 : /*
957 : * We need to start at position zero because there is no way to know
6385 958 : * in advance which byte offset corresponds to the supplied start
959 : * position.
960 : */
7535 bruce 961 GIC 394907 : slice_start = 0;
962 :
2118 tgl 963 394907 : if (length_not_specified) /* special case - get length to end of
964 : * string */
7535 bruce 965 44 : slice_size = L1 = -1;
825 tgl 966 394863 : else if (length < 0)
967 : {
968 : /* SQL99 says to throw an error for E < S, i.e., negative length */
969 6 : ereport(ERROR,
970 : (errcode(ERRCODE_SUBSTRING_ERROR),
971 : errmsg("negative substring length not allowed")));
972 : slice_size = L1 = -1; /* silence stupider compilers */
973 : }
974 394857 : else if (pg_add_s32_overflow(S, length, &E))
975 : {
976 : /*
977 : * L could be large enough for S + L to overflow, in which case
978 : * the substring must run to end of string.
979 : */
825 tgl 980 CBC 3 : slice_size = L1 = -1;
981 : }
825 tgl 982 ECB : else
983 : {
7522 bruce 984 : /*
6385 985 : * A zero or negative value for the end position can happen if the
986 : * start was negative or one. SQL99 says to return a zero-length
987 : * string.
7535 988 : */
7535 bruce 989 GIC 394854 : if (E < 1)
5493 tgl 990 UIC 0 : return cstring_to_text("");
991 :
992 : /*
7522 bruce 993 ECB : * if E is past the end of the string, the tuple toaster will
994 : * truncate the length for us
995 : */
7535 bruce 996 GIC 394854 : L1 = E - S1;
997 :
998 : /*
6385 bruce 999 ECB : * Total slice size in bytes can't be any longer than the start
1000 : * position plus substring length times the encoding max length.
1001 : * If that overflows, we can just use -1.
1002 : */
825 tgl 1003 GIC 394854 : if (pg_mul_s32_overflow(E, eml, &slice_size))
1004 3 : slice_size = -1;
1005 : }
1006 :
1007 : /*
5624 bruce 1008 ECB : * If we're working with an untoasted source, no need to do an extra
5624 bruce 1009 EUB : * copying step.
1010 : */
5327 tgl 1011 GIC 394901 : if (VARATT_IS_COMPRESSED(DatumGetPointer(str)) ||
5475 1012 394874 : VARATT_IS_EXTERNAL(DatumGetPointer(str)))
5996 1013 162 : slice = DatumGetTextPSlice(str, slice_start, slice_size);
1014 : else
5996 tgl 1015 CBC 394739 : slice = (text *) DatumGetPointer(str);
1016 :
1017 : /* see if we got back an empty string */
5678 tgl 1018 GIC 394901 : if (VARSIZE_ANY_EXHDR(slice) == 0)
1019 : {
5996 tgl 1020 UIC 0 : if (slice != (text *) DatumGetPointer(str))
1021 0 : pfree(slice);
5493 tgl 1022 LBC 0 : return cstring_to_text("");
5996 tgl 1023 ECB : }
1024 :
1025 : /* Now we can get the actual length of the slice in MB characters */
5678 tgl 1026 GIC 394901 : slice_strlen = pg_mbstrlen_with_len(VARDATA_ANY(slice),
1027 394901 : VARSIZE_ANY_EXHDR(slice));
1028 :
1029 : /*
6385 bruce 1030 ECB : * Check that the start position wasn't > slice_strlen. If so, SQL99
1031 : * says to return a zero-length string.
7535 1032 : */
7535 bruce 1033 GIC 394901 : if (S1 > slice_strlen)
5996 tgl 1034 ECB : {
5996 tgl 1035 GIC 11 : if (slice != (text *) DatumGetPointer(str))
5996 tgl 1036 UIC 0 : pfree(slice);
5493 tgl 1037 CBC 11 : return cstring_to_text("");
1038 : }
9229 lockhart 1039 EUB :
7535 bruce 1040 : /*
6385 1041 : * Adjust L1 and E1 now that we know the slice string length. Again
1042 : * remember that S1 is one based, and slice_start is zero based.
1043 : */
7535 bruce 1044 GIC 394890 : if (L1 > -1)
7522 bruce 1045 CBC 394854 : E1 = Min(S1 + L1, slice_start + 1 + slice_strlen);
7535 bruce 1046 ECB : else
7535 bruce 1047 GIC 36 : E1 = slice_start + 1 + slice_strlen;
1048 :
1049 : /*
1050 : * Find the start position in the slice; remember S1 is not zero based
1051 : */
5678 tgl 1052 CBC 394890 : p = VARDATA_ANY(slice);
7535 bruce 1053 GIC 2805764 : for (i = 0; i < S1 - 1; i++)
7535 bruce 1054 CBC 2410874 : p += pg_mblen(p);
7535 bruce 1055 EUB :
7535 bruce 1056 ECB : /* hang onto a pointer to our start position */
7535 bruce 1057 GIC 394890 : s = p;
1058 :
1059 : /*
1060 : * Count the actual bytes used by the substring of the requested
1061 : * length.
1062 : */
7535 bruce 1063 CBC 4855113 : for (i = S1; i < E1; i++)
1064 4460223 : p += pg_mblen(p);
1065 :
1066 394890 : ret = (text *) palloc(VARHDRSZ + (p - s));
5885 tgl 1067 GIC 394890 : SET_VARSIZE(ret, VARHDRSZ + (p - s));
7535 bruce 1068 394890 : memcpy(VARDATA(ret), s, (p - s));
1069 :
5996 tgl 1070 394890 : if (slice != (text *) DatumGetPointer(str))
5996 tgl 1071 CBC 162 : pfree(slice);
5996 tgl 1072 ECB :
7535 bruce 1073 CBC 394890 : return ret;
1074 : }
1075 : else
7196 tgl 1076 LBC 0 : elog(ERROR, "invalid backend encoding: encoding max length < 1");
1077 :
1078 : /* not reached: suppress compiler warning */
1079 : return NULL;
1080 : }
1081 :
4822 tgl 1082 ECB : /*
1083 : * textoverlay
1084 : * Replace specified substring of first string with second
1085 : *
1086 : * The SQL standard defines OVERLAY() in terms of substring and concatenation.
1087 : * This code is a direct implementation of what the standard says.
1088 : */
1089 : Datum
4822 tgl 1090 CBC 14 : textoverlay(PG_FUNCTION_ARGS)
1091 : {
1092 14 : text *t1 = PG_GETARG_TEXT_PP(0);
4822 tgl 1093 GIC 14 : text *t2 = PG_GETARG_TEXT_PP(1);
2118 1094 14 : int sp = PG_GETARG_INT32(2); /* substring start position */
2118 tgl 1095 GBC 14 : int sl = PG_GETARG_INT32(3); /* substring length */
1096 :
4822 tgl 1097 GIC 14 : PG_RETURN_TEXT_P(text_overlay(t1, t2, sp, sl));
1098 : }
1099 :
1100 : Datum
1101 6 : textoverlay_no_len(PG_FUNCTION_ARGS)
1102 : {
1103 6 : text *t1 = PG_GETARG_TEXT_PP(0);
1104 6 : text *t2 = PG_GETARG_TEXT_PP(1);
2118 1105 6 : int sp = PG_GETARG_INT32(2); /* substring start position */
1106 : int sl;
1107 :
1108 6 : sl = text_length(PointerGetDatum(t2)); /* defaults to length(t2) */
4822 tgl 1109 CBC 6 : PG_RETURN_TEXT_P(text_overlay(t1, t2, sp, sl));
1110 : }
4822 tgl 1111 ECB :
1112 : static text *
4822 tgl 1113 CBC 20 : text_overlay(text *t1, text *t2, int sp, int sl)
4822 tgl 1114 ECB : {
1115 : text *result;
1116 : text *s1;
1117 : text *s2;
1118 : int sp_pl_sl;
1119 :
1120 : /*
1121 : * Check for possible integer-overflow cases. For negative sp, throw a
4790 bruce 1122 : * "substring length" error because that's what should be expected
1123 : * according to the spec's definition of OVERLAY().
4822 tgl 1124 : */
4822 tgl 1125 GIC 20 : if (sp <= 0)
4822 tgl 1126 UIC 0 : ereport(ERROR,
4822 tgl 1127 ECB : (errcode(ERRCODE_SUBSTRING_ERROR),
1128 : errmsg("negative substring length not allowed")));
1944 andres 1129 GIC 20 : if (pg_add_s32_overflow(sp, sl, &sp_pl_sl))
4822 tgl 1130 UIC 0 : ereport(ERROR,
1131 : (errcode(ERRCODE_NUMERIC_VALUE_OUT_OF_RANGE),
4822 tgl 1132 ECB : errmsg("integer out of range")));
1133 :
4790 bruce 1134 GIC 20 : s1 = text_substring(PointerGetDatum(t1), 1, sp - 1, false);
4822 tgl 1135 20 : s2 = text_substring(PointerGetDatum(t1), sp_pl_sl, -1, true);
1136 20 : result = text_catenate(s1, t2);
1137 20 : result = text_catenate(result, s2);
1138 :
1139 20 : return result;
1140 : }
1141 :
1142 : /*
1143 : * textpos -
9345 bruce 1144 ECB : * Return the position of the specified substring.
3641 peter_e 1145 EUB : * Implements the SQL POSITION() function.
1146 : * Ref: A Guide To The SQL Standard, Date & Darwen, 1997
1147 : * - thomas 1997-07-27
9385 lockhart 1148 ECB : */
8312 tgl 1149 EUB : Datum
8312 tgl 1150 GIC 53 : textpos(PG_FUNCTION_ARGS)
1151 : {
5678 1152 53 : text *str = PG_GETARG_TEXT_PP(0);
5678 tgl 1153 CBC 53 : text *search_str = PG_GETARG_TEXT_PP(1);
7008 tgl 1154 ECB :
1479 peter 1155 CBC 53 : PG_RETURN_INT32((int32) text_position(str, search_str, PG_GET_COLLATION()));
7535 bruce 1156 ECB : }
1157 :
1158 : /*
1159 : * text_position -
1160 : * Does the real work for textpos()
1161 : *
1162 : * Inputs:
1163 : * t1 - string to be searched
1164 : * t2 - pattern to match within t1
1165 : * Result:
1166 : * Character index of the first matched char, starting from 1,
1167 : * or 0 if no match.
1168 : *
1169 : * This is broken out so it can be called directly by other string processing
1170 : * functions.
1171 : */
6028 tgl 1172 : static int
1479 peter 1173 GIC 53 : text_position(text *t1, text *t2, Oid collid)
7535 bruce 1174 ECB : {
1175 : TextPositionState state;
1176 : int result;
1177 :
1178 : /* Empty needle always matches at position 1 */
1259 tgl 1179 GIC 53 : if (VARSIZE_ANY_EXHDR(t2) < 1)
1180 6 : return 1;
1181 :
1182 : /* Otherwise, can't match if haystack is shorter than needle */
1183 47 : if (VARSIZE_ANY_EXHDR(t1) < VARSIZE_ANY_EXHDR(t2))
1535 heikki.linnakangas 1184 11 : return 0;
1185 :
1479 peter 1186 36 : text_position_setup(t1, t2, collid, &state);
1535 heikki.linnakangas 1187 36 : if (!text_position_next(&state))
1188 12 : result = 0;
1189 : else
1190 24 : result = text_position_get_match_pos(&state);
6028 tgl 1191 36 : text_position_cleanup(&state);
6028 tgl 1192 CBC 36 : return result;
1193 : }
1194 :
1195 :
1196 : /*
1197 : * text_position_setup, text_position_next, text_position_cleanup -
6028 tgl 1198 ECB : * Component steps of text_position()
1199 : *
1200 : * These are broken out so that a string can be efficiently searched for
1201 : * multiple occurrences of the same pattern. text_position_next may be
1535 heikki.linnakangas 1202 : * called multiple times, and it advances to the next match on each call.
1203 : * text_position_get_match_ptr() and text_position_get_match_pos() return
1204 : * a pointer or 1-based character position of the last match, respectively.
1205 : *
1206 : * The "state" variable is normally just a local variable in the caller.
1207 : *
1208 : * NOTE: text_position_next skips over the matched portion. For example,
1209 : * searching for "xx" in "xxx" returns only one match, not two.
6028 tgl 1210 : */
9345 bruce 1211 :
1212 : static void
1479 peter 1213 GIC 1405 : text_position_setup(text *t1, text *t2, Oid collid, TextPositionState *state)
1214 : {
5678 tgl 1215 1405 : int len1 = VARSIZE_ANY_EXHDR(t1);
1216 1405 : int len2 = VARSIZE_ANY_EXHDR(t2);
1418 1217 1405 : pg_locale_t mylocale = 0;
1218 :
1479 peter 1219 1405 : check_collation_set(collid);
1220 :
444 1221 1405 : if (!lc_collate_is_c(collid))
1479 1222 140 : mylocale = pg_newlocale_from_collation(collid);
1223 :
45 jdavis 1224 GNC 1405 : if (!pg_locale_deterministic(mylocale))
1479 peter 1225 GIC 6 : ereport(ERROR,
1226 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1227 : errmsg("nondeterministic collations are not supported for substring searches")));
1228 :
1535 heikki.linnakangas 1229 1399 : Assert(len1 > 0);
1230 1399 : Assert(len2 > 0);
1231 :
1535 heikki.linnakangas 1232 ECB : /*
1233 : * Even with a multi-byte encoding, we perform the search using the raw
1234 : * byte sequence, ignoring multibyte issues. For UTF-8, that works fine,
1235 : * because in UTF-8 the byte sequence of one character cannot contain
1236 : * another character. For other multi-byte encodings, we do the search
1237 : * initially as a simple byte search, ignoring multibyte issues, but
1238 : * verify afterwards that the match we found is at a character boundary,
1239 : * and continue the search if it was a false match.
1240 : */
7008 tgl 1241 CBC 1399 : if (pg_database_encoding_max_length() == 1)
1535 heikki.linnakangas 1242 GIC 18 : state->is_multibyte_char_in_char = false;
1535 heikki.linnakangas 1243 CBC 1381 : else if (GetDatabaseEncoding() == PG_UTF8)
1244 1381 : state->is_multibyte_char_in_char = false;
1245 : else
1535 heikki.linnakangas 1246 UIC 0 : state->is_multibyte_char_in_char = true;
1247 :
1535 heikki.linnakangas 1248 CBC 1399 : state->str1 = VARDATA_ANY(t1);
1249 1399 : state->str2 = VARDATA_ANY(t2);
1535 heikki.linnakangas 1250 GIC 1399 : state->len1 = len1;
1251 1399 : state->len2 = len2;
1252 1399 : state->last_match = NULL;
1253 1399 : state->refpoint = state->str1;
1254 1399 : state->refpos = 0;
1255 :
1256 : /*
1257 : * Prepare the skip table for Boyer-Moore-Horspool searching. In these
1258 : * notes we use the terminology that the "haystack" is the string to be
1259 : * searched (t1) and the "needle" is the pattern being sought (t2).
5327 tgl 1260 ECB : *
1261 : * If the needle is empty or bigger than the haystack then there is no
3260 bruce 1262 : * point in wasting cycles initializing the table. We also choose not to
5050 1263 : * use B-M-H for needles of length 1, since the skip table can't possibly
1264 : * save anything in that case.
5327 tgl 1265 EUB : */
5327 tgl 1266 GIC 1399 : if (len1 >= len2 && len2 > 1)
5327 tgl 1267 ECB : {
5050 bruce 1268 CBC 1295 : int searchlength = len1 - len2;
5050 bruce 1269 ECB : int skiptablemask;
1270 : int last;
1271 : int i;
1535 heikki.linnakangas 1272 CBC 1295 : const char *str2 = state->str2;
5327 tgl 1273 ECB :
1274 : /*
1275 : * First we must determine how much of the skip table to use. The
1276 : * declaration of TextPositionState allows up to 256 elements, but for
1277 : * short search problems we don't really want to have to initialize so
1278 : * many elements --- it would take too long in comparison to the
1279 : * actual search time. So we choose a useful skip table size based on
1280 : * the haystack length minus the needle length. The closer the needle
1281 : * length is to the haystack length the less useful skipping becomes.
1282 : *
1283 : * Note: since we use bit-masking to select table elements, the skip
1284 : * table size MUST be a power of 2, and so the mask must be 2^N-1.
1285 : */
5327 tgl 1286 GIC 1295 : if (searchlength < 16)
5327 tgl 1287 CBC 27 : skiptablemask = 3;
5327 tgl 1288 GIC 1268 : else if (searchlength < 64)
1289 8 : skiptablemask = 7;
1290 1260 : else if (searchlength < 128)
5327 tgl 1291 CBC 7 : skiptablemask = 15;
5327 tgl 1292 GIC 1253 : else if (searchlength < 512)
1293 95 : skiptablemask = 31;
1294 1158 : else if (searchlength < 2048)
1295 1096 : skiptablemask = 63;
1296 62 : else if (searchlength < 4096)
1297 28 : skiptablemask = 127;
1298 : else
1299 34 : skiptablemask = 255;
1300 1295 : state->skiptablemask = skiptablemask;
1301 :
1302 : /*
1303 : * Initialize the skip table. We set all elements to the needle
1304 : * length, since this is the correct skip distance for any character
5327 tgl 1305 ECB : * not found in the needle.
1306 : */
5327 tgl 1307 CBC 87051 : for (i = 0; i <= skiptablemask; i++)
1308 85756 : state->skiptable[i] = len2;
5327 tgl 1309 ECB :
1310 : /*
3260 bruce 1311 : * Now examine the needle. For each character except the last one,
5327 tgl 1312 : * set the corresponding table element to the appropriate skip
1313 : * distance. Note that when two characters share the same skip table
5050 bruce 1314 : * entry, the one later in the needle must determine the skip
1315 : * distance.
5327 tgl 1316 : */
5327 tgl 1317 GIC 1295 : last = len2 - 1;
5327 tgl 1318 ECB :
1535 heikki.linnakangas 1319 CBC 16263 : for (i = 0; i < last; i++)
1535 heikki.linnakangas 1320 GIC 14968 : state->skiptable[(unsigned char) str2[i] & skiptablemask] = last - i;
1321 : }
6028 tgl 1322 1399 : }
1323 :
1324 : /*
1325 : * Advance to the next match, starting from the end of the previous match
1535 heikki.linnakangas 1326 ECB : * (or the beginning of the string, on first call). Returns true if a match
1327 : * is found.
1328 : *
1329 : * Note that this refuses to match an empty-string needle. Most callers
1330 : * will have handled that case specially and we'll never see it here.
1331 : */
1332 : static bool
1535 heikki.linnakangas 1333 GIC 4789 : text_position_next(TextPositionState *state)
1334 : {
5327 tgl 1335 4789 : int needle_len = state->len2;
1535 heikki.linnakangas 1336 ECB : char *start_ptr;
1337 : char *matchptr;
6028 tgl 1338 :
5327 tgl 1339 CBC 4789 : if (needle_len <= 0)
1535 heikki.linnakangas 1340 UIC 0 : return false; /* result for empty pattern */
6028 tgl 1341 ECB :
1342 : /* Start from the point right after the previous match. */
1535 heikki.linnakangas 1343 GIC 4789 : if (state->last_match)
1344 3384 : start_ptr = state->last_match + needle_len;
1345 : else
1346 1405 : start_ptr = state->str1;
1347 :
1348 4789 : retry:
1349 4789 : matchptr = text_position_next_internal(start_ptr, state);
1350 :
1351 4789 : if (!matchptr)
1535 heikki.linnakangas 1352 CBC 1369 : return false;
1353 :
1535 heikki.linnakangas 1354 ECB : /*
1355 : * Found a match for the byte sequence. If this is a multibyte encoding,
1356 : * where one character's byte sequence can appear inside a longer
1357 : * multi-byte character, we need to verify that the match was at a
1358 : * character boundary, not in the middle of a multi-byte character.
1535 heikki.linnakangas 1359 EUB : */
1535 heikki.linnakangas 1360 GIC 3420 : if (state->is_multibyte_char_in_char)
1361 : {
1535 heikki.linnakangas 1362 ECB : /* Walk one character at a time, until we reach the match. */
7008 tgl 1363 :
1364 : /* the search should never move backwards. */
1535 heikki.linnakangas 1365 LBC 0 : Assert(state->refpoint <= matchptr);
1366 :
1367 0 : while (state->refpoint < matchptr)
5327 tgl 1368 ECB : {
1369 : /* step to next character. */
1535 heikki.linnakangas 1370 LBC 0 : state->refpoint += pg_mblen(state->refpoint);
1371 0 : state->refpos++;
1372 :
1373 : /*
1374 : * If we stepped over the match's start position, then it was a
1375 : * false positive, where the byte sequence appeared in the middle
1376 : * of a multi-byte character. Skip it, and continue the search at
1377 : * the next character boundary.
1378 : */
1379 0 : if (state->refpoint > matchptr)
1380 : {
1535 heikki.linnakangas 1381 UIC 0 : start_ptr = state->refpoint;
1382 0 : goto retry;
1383 : }
5327 tgl 1384 EUB : }
1385 : }
1386 :
1535 heikki.linnakangas 1387 GIC 3420 : state->last_match = matchptr;
1388 3420 : return true;
1535 heikki.linnakangas 1389 EUB : }
5327 tgl 1390 :
1391 : /*
1392 : * Subroutine of text_position_next(). This searches for the raw byte
1393 : * sequence, ignoring any multi-byte encoding issues. Returns the first
1394 : * match starting at 'start_ptr', or NULL if no match is found.
1395 : */
1396 : static char *
1535 heikki.linnakangas 1397 GIC 4789 : text_position_next_internal(char *start_ptr, TextPositionState *state)
1535 heikki.linnakangas 1398 EUB : {
1535 heikki.linnakangas 1399 GIC 4789 : int haystack_len = state->len1;
1535 heikki.linnakangas 1400 GBC 4789 : int needle_len = state->len2;
1401 4789 : int skiptablemask = state->skiptablemask;
1535 heikki.linnakangas 1402 GIC 4789 : const char *haystack = state->str1;
1403 4789 : const char *needle = state->str2;
1404 4789 : const char *haystack_end = &haystack[haystack_len];
1405 : const char *hptr;
5050 bruce 1406 ECB :
1535 heikki.linnakangas 1407 CBC 4789 : Assert(start_ptr >= haystack && start_ptr <= haystack_end);
1408 :
1535 heikki.linnakangas 1409 GIC 4789 : if (needle_len == 1)
1410 : {
1411 : /* No point in using B-M-H for a one-character needle */
1412 377 : char nchar = *needle;
1413 :
1414 377 : hptr = start_ptr;
1415 2877 : while (hptr < haystack_end)
1535 heikki.linnakangas 1416 ECB : {
1535 heikki.linnakangas 1417 GIC 2794 : if (*hptr == nchar)
1535 heikki.linnakangas 1418 CBC 294 : return (char *) hptr;
1419 2500 : hptr++;
7535 bruce 1420 ECB : }
1421 : }
7008 tgl 1422 : else
7535 bruce 1423 : {
1535 heikki.linnakangas 1424 GIC 4412 : const char *needle_last = &needle[needle_len - 1];
1425 :
1535 heikki.linnakangas 1426 ECB : /* Start at startpos plus the length of the needle */
1535 heikki.linnakangas 1427 GIC 4412 : hptr = start_ptr + needle_len - 1;
1535 heikki.linnakangas 1428 CBC 116857 : while (hptr < haystack_end)
1429 : {
1430 : /* Match the needle scanning *backward* */
1535 heikki.linnakangas 1431 ECB : const char *nptr;
1432 : const char *p;
7008 tgl 1433 :
1535 heikki.linnakangas 1434 CBC 115571 : nptr = needle_last;
1535 heikki.linnakangas 1435 GIC 115571 : p = hptr;
1535 heikki.linnakangas 1436 CBC 160855 : while (*nptr == *p)
5327 tgl 1437 ECB : {
1535 heikki.linnakangas 1438 : /* Matched it all? If so, return 1-based position */
1535 heikki.linnakangas 1439 GIC 48410 : if (nptr == needle)
1440 3126 : return (char *) p;
1441 45284 : nptr--, p--;
1442 : }
1535 heikki.linnakangas 1443 ECB :
1444 : /*
1445 : * No match, so use the haystack char at hptr to decide how far to
1446 : * advance. If the needle had any occurrence of that character
1447 : * (or more precisely, one sharing the same skiptable entry)
1448 : * before its last character, then we advance far enough to align
1449 : * the last such needle character with that haystack position.
1450 : * Otherwise we can advance by the whole needle length.
1451 : */
1535 heikki.linnakangas 1452 GIC 112445 : hptr += state->skiptable[(unsigned char) *hptr & skiptablemask];
5327 tgl 1453 ECB : }
1535 heikki.linnakangas 1454 : }
5327 tgl 1455 :
1535 heikki.linnakangas 1456 GIC 1369 : return 0; /* not found */
1457 : }
5327 tgl 1458 ECB :
1535 heikki.linnakangas 1459 : /*
1460 : * Return a pointer to the current match.
1461 : *
1462 : * The returned pointer points into the original haystack string.
1463 : */
1464 : static char *
1535 heikki.linnakangas 1465 GIC 3381 : text_position_get_match_ptr(TextPositionState *state)
1466 : {
1467 3381 : return state->last_match;
1468 : }
1469 :
1470 : /*
1535 heikki.linnakangas 1471 ECB : * Return the offset of the current match.
1472 : *
1473 : * The offset is in characters, 1-based.
1474 : */
1475 : static int
1535 heikki.linnakangas 1476 GIC 24 : text_position_get_match_pos(TextPositionState *state)
1477 : {
1478 : /* Convert the byte position to char position. */
478 john.naylor 1479 48 : state->refpos += pg_mbstrlen_with_len(state->refpoint,
1480 24 : state->last_match - state->refpoint);
1481 24 : state->refpoint = state->last_match;
1482 24 : return state->refpos + 1;
1483 : }
9770 scrappy 1484 ECB :
1485 : /*
877 tgl 1486 : * Reset search state to the initial state installed by text_position_setup.
1487 : *
1488 : * The next call to text_position_next will search from the beginning
1489 : * of the string.
1490 : */
1491 : static void
877 tgl 1492 GIC 6 : text_position_reset(TextPositionState *state)
1493 : {
1494 6 : state->last_match = NULL;
877 tgl 1495 CBC 6 : state->refpoint = state->str1;
877 tgl 1496 GIC 6 : state->refpos = 0;
1497 6 : }
877 tgl 1498 ECB :
6028 1499 : static void
5624 bruce 1500 CBC 1399 : text_position_cleanup(TextPositionState *state)
6028 tgl 1501 ECB : {
1502 : /* no cleanup needed */
6028 tgl 1503 GIC 1399 : }
1504 :
1505 :
1506 : static void
1479 peter 1507 8518884 : check_collation_set(Oid collid)
1508 : {
1509 8518884 : if (!OidIsValid(collid))
1510 : {
1479 peter 1511 ECB : /*
1512 : * This typically means that the parser could not resolve a conflict
1513 : * of implicit collations, so report it that way.
1514 : */
1479 peter 1515 CBC 24 : ereport(ERROR,
1479 peter 1516 ECB : (errcode(ERRCODE_INDETERMINATE_COLLATION),
1517 : errmsg("could not determine which collation to use for string comparison"),
1518 : errhint("Use the COLLATE clause to set the collation explicitly.")));
1519 : }
1479 peter 1520 GIC 8518860 : }
1521 :
9063 bruce 1522 ECB : /* varstr_cmp()
1523 : * Comparison function for text strings with given lengths.
1524 : * Includes locale support, but must copy strings to temporary memory
1525 : * to allow null-termination for inputs to strcoll().
5099 heikki.linnakangas 1526 : * Returns an integer less than, equal to, or greater than zero, indicating
1527 : * whether arg1 is less than, equal to, or greater than arg2.
1296 tgl 1528 : *
1529 : * Note: many functions that depend on this are marked leakproof; therefore,
1530 : * avoid reporting the actual contents of the input when throwing errors.
1531 : * All errors herein should be things that can't happen except on corrupt
1532 : * data, anyway; otherwise we will have trouble with indexing strings that
1533 : * would cause them.
9496 scrappy 1534 : */
1535 : int
1986 peter_e 1536 GIC 4679892 : varstr_cmp(const char *arg1, int len1, const char *arg2, int len2, Oid collid)
1537 : {
1538 : int result;
9345 bruce 1539 ECB :
1479 peter 1540 GIC 4679892 : check_collation_set(collid);
1541 :
1542 : /*
1543 : * Unfortunately, there is no strncoll(), so in the non-C locale case we
1544 : * have to do some memory copying. This turns out to be significantly
1545 : * slower, so we optimize the case where LC_COLLATE is C. We also try to
1546 : * optimize relatively-short strings by avoiding palloc/pfree overhead.
1547 : */
4443 peter_e 1548 4679877 : if (lc_collate_is_c(collid))
1549 : {
4492 rhaas 1550 2394653 : result = memcmp(arg1, arg2, Min(len1, len2));
6435 tgl 1551 2394653 : if ((result == 0) && (len1 != len2))
1552 68787 : result = (len1 < len2) ? -1 : 1;
1553 : }
1554 : else
6435 tgl 1555 ECB : {
1556 : pg_locale_t mylocale;
1557 :
444 peter 1558 GIC 2285224 : mylocale = pg_newlocale_from_collation(collid);
1559 :
1560 : /*
1561 : * memcmp() can't tell us which of two unequal strings sorts first,
1562 : * but it's a cheap way to tell if they're equal. Testing shows that
3124 rhaas 1563 ECB : * memcmp() followed by strcoll() is only trivially slower than
1564 : * strcoll() by itself, so we don't lose much if this doesn't work out
1565 : * very often, and if it does - for example, because there are many
1566 : * equal strings in the input - then we win big by avoiding expensive
1567 : * collation-aware comparisons.
1568 : */
3124 rhaas 1569 GIC 2285224 : if (len1 == len2 && memcmp(arg1, arg2, len1) == 0)
1570 949522 : return 0;
1571 :
45 jdavis 1572 GNC 1335702 : result = pg_strncoll(arg1, len1, arg2, len2, mylocale);
1573 :
1574 : /* Break tie if necessary. */
1575 1335702 : if (result == 0 && pg_locale_deterministic(mylocale))
1576 : {
45 jdavis 1577 UNC 0 : result = memcmp(arg1, arg2, Min(len1, len2));
1578 0 : if ((result == 0) && (len1 != len2))
1579 0 : result = (len1 < len2) ? -1 : 1;
1580 : }
1581 : }
1582 :
8986 bruce 1583 CBC 3730355 : return result;
1584 : }
1585 :
1586 : /* text_cmp()
8011 tgl 1587 ECB : * Internal comparison function for text strings.
1588 : * Returns -1, 0 or 1
9496 scrappy 1589 : */
8949 bruce 1590 : static int
4443 peter_e 1591 GIC 3848460 : text_cmp(text *arg1, text *arg2, Oid collid)
1592 : {
8986 bruce 1593 ECB : char *a1p,
1594 : *a2p;
1595 : int len1,
1596 : len2;
1597 :
5847 tgl 1598 CBC 3848460 : a1p = VARDATA_ANY(arg1);
5847 tgl 1599 GIC 3848460 : a2p = VARDATA_ANY(arg2);
1600 :
1601 3848460 : len1 = VARSIZE_ANY_EXHDR(arg1);
5847 tgl 1602 CBC 3848460 : len2 = VARSIZE_ANY_EXHDR(arg2);
1603 :
4443 peter_e 1604 3848460 : return varstr_cmp(a1p, len1, a2p, len2, collid);
8312 tgl 1605 ECB : }
1606 :
1607 : /*
1608 : * Comparison functions for text strings.
1609 : *
8306 1610 : * Note: btree indexes need these routines not to leak memory; therefore,
1611 : * be careful to free working copies of toasted datums. Most places don't
1612 : * need to be so careful.
9063 bruce 1613 : */
1614 :
1615 : Datum
8011 tgl 1616 GIC 3610619 : texteq(PG_FUNCTION_ARGS)
8011 tgl 1617 ECB : {
1479 peter 1618 GIC 3610619 : Oid collid = PG_GET_COLLATION();
444 peter 1619 CBC 3610619 : bool locale_is_c = false;
332 tgl 1620 3610619 : pg_locale_t mylocale = 0;
1621 : bool result;
1622 :
1479 peter 1623 3610619 : check_collation_set(collid);
1624 :
444 1625 3610619 : if (lc_collate_is_c(collid))
1626 273132 : locale_is_c = true;
1627 : else
1628 3337487 : mylocale = pg_newlocale_from_collation(collid);
1629 :
45 jdavis 1630 GNC 3610619 : if (locale_is_c || pg_locale_deterministic(mylocale))
1479 peter 1631 GIC 3610377 : {
1479 peter 1632 CBC 3610377 : Datum arg1 = PG_GETARG_DATUM(0);
1479 peter 1633 GIC 3610377 : Datum arg2 = PG_GETARG_DATUM(1);
1479 peter 1634 ECB : Size len1,
1635 : len2;
1636 :
1637 : /*
1418 tgl 1638 : * Since we only care about equality or not-equality, we can avoid all
1639 : * the expense of strcoll() here, and just do bitwise comparison. In
1640 : * fact, we don't even have to do a bitwise comparison if we can show
1641 : * the lengths of the strings are unequal; which might save us from
1642 : * having to detoast one or both values.
1479 peter 1643 : */
1479 peter 1644 GIC 3610377 : len1 = toast_raw_datum_size(arg1);
1645 3610377 : len2 = toast_raw_datum_size(arg2);
1646 3610377 : if (len1 != len2)
1479 peter 1647 CBC 1435697 : result = false;
1648 : else
1479 peter 1649 ECB : {
1479 peter 1650 CBC 2174680 : text *targ1 = DatumGetTextPP(arg1);
1651 2174680 : text *targ2 = DatumGetTextPP(arg2);
1479 peter 1652 ECB :
1479 peter 1653 GIC 2174680 : result = (memcmp(VARDATA_ANY(targ1), VARDATA_ANY(targ2),
1654 : len1 - VARHDRSZ) == 0);
1655 :
1656 2174680 : PG_FREE_IF_COPY(targ1, 0);
1479 peter 1657 CBC 2174680 : PG_FREE_IF_COPY(targ2, 1);
1658 : }
1479 peter 1659 ECB : }
8011 tgl 1660 : else
1661 : {
1479 peter 1662 CBC 242 : text *arg1 = PG_GETARG_TEXT_PP(0);
1479 peter 1663 GBC 242 : text *arg2 = PG_GETARG_TEXT_PP(1);
1664 :
1479 peter 1665 GIC 242 : result = (text_cmp(arg1, arg2, collid) == 0);
1666 :
1479 peter 1667 CBC 242 : PG_FREE_IF_COPY(arg1, 0);
1668 242 : PG_FREE_IF_COPY(arg2, 1);
4464 tgl 1669 ECB : }
8011 tgl 1670 EUB :
8011 tgl 1671 GIC 3610619 : PG_RETURN_BOOL(result);
1672 : }
8011 tgl 1673 ECB :
1674 : Datum
8011 tgl 1675 GIC 9726 : textne(PG_FUNCTION_ARGS)
8011 tgl 1676 ECB : {
1479 peter 1677 CBC 9726 : Oid collid = PG_GET_COLLATION();
444 peter 1678 GIC 9726 : bool locale_is_c = false;
332 tgl 1679 CBC 9726 : pg_locale_t mylocale = 0;
8011 tgl 1680 ECB : bool result;
1681 :
1479 peter 1682 GIC 9726 : check_collation_set(collid);
1479 peter 1683 ECB :
444 peter 1684 GIC 9726 : if (lc_collate_is_c(collid))
1685 9 : locale_is_c = true;
1686 : else
444 peter 1687 CBC 9717 : mylocale = pg_newlocale_from_collation(collid);
1688 :
45 jdavis 1689 GNC 9726 : if (locale_is_c || pg_locale_deterministic(mylocale))
1479 peter 1690 CBC 9714 : {
1479 peter 1691 GIC 9714 : Datum arg1 = PG_GETARG_DATUM(0);
1692 9714 : Datum arg2 = PG_GETARG_DATUM(1);
1479 peter 1693 ECB : Size len1,
1694 : len2;
1695 :
1696 : /* See comment in texteq() */
1479 peter 1697 GIC 9714 : len1 = toast_raw_datum_size(arg1);
1479 peter 1698 CBC 9714 : len2 = toast_raw_datum_size(arg2);
1479 peter 1699 GIC 9714 : if (len1 != len2)
1700 917 : result = true;
1701 : else
1479 peter 1702 ECB : {
1479 peter 1703 GIC 8797 : text *targ1 = DatumGetTextPP(arg1);
1479 peter 1704 CBC 8797 : text *targ2 = DatumGetTextPP(arg2);
1479 peter 1705 ECB :
1479 peter 1706 GIC 8797 : result = (memcmp(VARDATA_ANY(targ1), VARDATA_ANY(targ2),
1707 : len1 - VARHDRSZ) != 0);
1479 peter 1708 ECB :
1479 peter 1709 GIC 8797 : PG_FREE_IF_COPY(targ1, 0);
1710 8797 : PG_FREE_IF_COPY(targ2, 1);
1479 peter 1711 ECB : }
1712 : }
8011 tgl 1713 : else
1714 : {
1479 peter 1715 CBC 12 : text *arg1 = PG_GETARG_TEXT_PP(0);
1479 peter 1716 GIC 12 : text *arg2 = PG_GETARG_TEXT_PP(1);
1717 :
1718 12 : result = (text_cmp(arg1, arg2, collid) != 0);
1719 :
1720 12 : PG_FREE_IF_COPY(arg1, 0);
1721 12 : PG_FREE_IF_COPY(arg2, 1);
1722 : }
1723 :
8011 tgl 1724 9726 : PG_RETURN_BOOL(result);
1725 : }
1726 :
1727 : Datum
8312 1728 62299 : text_lt(PG_FUNCTION_ARGS)
9063 bruce 1729 ECB : {
5847 tgl 1730 GIC 62299 : text *arg1 = PG_GETARG_TEXT_PP(0);
5847 tgl 1731 CBC 62299 : text *arg2 = PG_GETARG_TEXT_PP(1);
8306 tgl 1732 ECB : bool result;
1733 :
4443 peter_e 1734 CBC 62299 : result = (text_cmp(arg1, arg2, PG_GET_COLLATION()) < 0);
1735 :
8306 tgl 1736 62284 : PG_FREE_IF_COPY(arg1, 0);
8306 tgl 1737 GIC 62284 : PG_FREE_IF_COPY(arg2, 1);
1738 :
1739 62284 : PG_RETURN_BOOL(result);
1740 : }
1741 :
1742 : Datum
8312 1743 162176 : text_le(PG_FUNCTION_ARGS)
1744 : {
5847 1745 162176 : text *arg1 = PG_GETARG_TEXT_PP(0);
1746 162176 : text *arg2 = PG_GETARG_TEXT_PP(1);
1747 : bool result;
1748 :
4443 peter_e 1749 162176 : result = (text_cmp(arg1, arg2, PG_GET_COLLATION()) <= 0);
1750 :
8306 tgl 1751 CBC 162176 : PG_FREE_IF_COPY(arg1, 0);
8306 tgl 1752 GIC 162176 : PG_FREE_IF_COPY(arg2, 1);
9770 scrappy 1753 ECB :
8306 tgl 1754 CBC 162176 : PG_RETURN_BOOL(result);
8312 tgl 1755 ECB : }
1756 :
1757 : Datum
8312 tgl 1758 GIC 56829 : text_gt(PG_FUNCTION_ARGS)
9770 scrappy 1759 ECB : {
5847 tgl 1760 GIC 56829 : text *arg1 = PG_GETARG_TEXT_PP(0);
1761 56829 : text *arg2 = PG_GETARG_TEXT_PP(1);
8306 tgl 1762 ECB : bool result;
1763 :
4443 peter_e 1764 CBC 56829 : result = (text_cmp(arg1, arg2, PG_GET_COLLATION()) > 0);
1765 :
8306 tgl 1766 GIC 56829 : PG_FREE_IF_COPY(arg1, 0);
1767 56829 : PG_FREE_IF_COPY(arg2, 1);
1768 :
1769 56829 : PG_RETURN_BOOL(result);
1770 : }
1771 :
1772 : Datum
8312 tgl 1773 CBC 92446 : text_ge(PG_FUNCTION_ARGS)
1774 : {
5847 tgl 1775 GIC 92446 : text *arg1 = PG_GETARG_TEXT_PP(0);
1776 92446 : text *arg2 = PG_GETARG_TEXT_PP(1);
1777 : bool result;
8306 tgl 1778 ECB :
4443 peter_e 1779 GIC 92446 : result = (text_cmp(arg1, arg2, PG_GET_COLLATION()) >= 0);
8306 tgl 1780 EUB :
8306 tgl 1781 GIC 92446 : PG_FREE_IF_COPY(arg1, 0);
8306 tgl 1782 GBC 92446 : PG_FREE_IF_COPY(arg2, 1);
1783 :
8306 tgl 1784 GIC 92446 : PG_RETURN_BOOL(result);
9770 scrappy 1785 ECB : }
1786 :
1787 : Datum
1832 teodor 1788 GIC 18957 : text_starts_with(PG_FUNCTION_ARGS)
1789 : {
1790 18957 : Datum arg1 = PG_GETARG_DATUM(0);
1791 18957 : Datum arg2 = PG_GETARG_DATUM(1);
1479 peter 1792 18957 : Oid collid = PG_GET_COLLATION();
1418 tgl 1793 18957 : pg_locale_t mylocale = 0;
1794 : bool result;
1795 : Size len1,
1796 : len2;
1797 :
1479 peter 1798 18957 : check_collation_set(collid);
1799 :
444 1800 18957 : if (!lc_collate_is_c(collid))
1479 1801 18957 : mylocale = pg_newlocale_from_collation(collid);
1802 :
45 jdavis 1803 GNC 18957 : if (!pg_locale_deterministic(mylocale))
1479 peter 1804 LBC 0 : ereport(ERROR,
1805 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1806 : errmsg("nondeterministic collations are not supported for substring searches")));
1807 :
1832 teodor 1808 GIC 18957 : len1 = toast_raw_datum_size(arg1);
1809 18957 : len2 = toast_raw_datum_size(arg2);
1810 18957 : if (len2 > len1)
1832 teodor 1811 UIC 0 : result = false;
1812 : else
1832 teodor 1813 ECB : {
1468 sfrost 1814 GIC 18957 : text *targ1 = text_substring(arg1, 1, len2, false);
1832 teodor 1815 CBC 18957 : text *targ2 = DatumGetTextPP(arg2);
1832 teodor 1816 ECB :
1832 teodor 1817 CBC 18957 : result = (memcmp(VARDATA_ANY(targ1), VARDATA_ANY(targ2),
1818 18957 : VARSIZE_ANY_EXHDR(targ2)) == 0);
1832 teodor 1819 ECB :
1832 teodor 1820 GIC 18957 : PG_FREE_IF_COPY(targ1, 0);
1832 teodor 1821 CBC 18957 : PG_FREE_IF_COPY(targ2, 1);
1832 teodor 1822 ECB : }
1823 :
1832 teodor 1824 CBC 18957 : PG_RETURN_BOOL(result);
1832 teodor 1825 ECB : }
1826 :
1827 : Datum
8011 tgl 1828 GIC 3316584 : bttextcmp(PG_FUNCTION_ARGS)
1829 : {
5847 1830 3316584 : text *arg1 = PG_GETARG_TEXT_PP(0);
1831 3316584 : text *arg2 = PG_GETARG_TEXT_PP(1);
1832 : int32 result;
1833 :
4443 peter_e 1834 3316584 : result = text_cmp(arg1, arg2, PG_GET_COLLATION());
1835 :
8011 tgl 1836 3316584 : PG_FREE_IF_COPY(arg1, 0);
1837 3316584 : PG_FREE_IF_COPY(arg2, 1);
1838 :
1839 3316584 : PG_RETURN_INT32(result);
1840 : }
8011 tgl 1841 ECB :
3160 rhaas 1842 : Datum
3160 rhaas 1843 CBC 44359 : bttextsortsupport(PG_FUNCTION_ARGS)
3160 rhaas 1844 ECB : {
2878 bruce 1845 GIC 44359 : SortSupport ssup = (SortSupport) PG_GETARG_POINTER(0);
1846 44359 : Oid collid = ssup->ssup_collation;
1847 : MemoryContext oldcontext;
1848 :
3160 rhaas 1849 44359 : oldcontext = MemoryContextSwitchTo(ssup->ssup_cxt);
1850 :
2622 rhaas 1851 ECB : /* Use generic string SortSupport */
1572 tgl 1852 GIC 44359 : varstr_sortsupport(ssup, TEXTOID, collid);
3160 rhaas 1853 ECB :
3160 rhaas 1854 CBC 44350 : MemoryContextSwitchTo(oldcontext);
3160 rhaas 1855 ECB :
3160 rhaas 1856 CBC 44350 : PG_RETURN_VOID();
3160 rhaas 1857 ECB : }
1858 :
2622 1859 : /*
1860 : * Generic sortsupport interface for character type's operator classes.
1861 : * Includes locale support, and support for BpChar semantics (i.e. removing
1862 : * trailing spaces before comparison).
1863 : *
1864 : * Relies on the assumption that text, VarChar, BpChar, and bytea all have the
1865 : * same representation. Callers that always use the C collation (e.g.
1866 : * non-collatable type callers like bytea) may have NUL bytes in their strings;
1867 : * this will not work with any other collation, though.
1868 : */
1869 : void
1550 tgl 1870 CBC 83052 : varstr_sortsupport(SortSupport ssup, Oid typid, Oid collid)
3160 rhaas 1871 ECB : {
2878 bruce 1872 GIC 83052 : bool abbreviate = ssup->abbreviate;
1873 83052 : bool collate_c = false;
1874 : VarStringSortSupport *sss;
1875 83052 : pg_locale_t locale = 0;
1876 :
1479 peter 1877 83052 : check_collation_set(collid);
1479 peter 1878 ECB :
3160 rhaas 1879 : /*
1880 : * If possible, set ssup->comparator to a function which can be used to
2999 1881 : * directly compare two datums. If we can do this, we'll avoid the
2878 bruce 1882 : * overhead of a trip through the fmgr layer for every comparison, which
1883 : * can be substantial.
3002 rhaas 1884 : *
1572 tgl 1885 : * Most typically, we'll set the comparator to varlenafastcmp_locale,
1886 : * which uses strcoll() to perform comparisons. We use that for the
1887 : * BpChar case too, but type NAME uses namefastcmp_locale. However, if
1888 : * LC_COLLATE = C, we can make things quite a bit faster with
1889 : * varstrfastcmp_c, bpcharfastcmp_c, or namefastcmp_c, all of which use
1572 tgl 1890 EUB : * memcmp() rather than strcoll().
3160 rhaas 1891 ECB : */
2999 rhaas 1892 GBC 83043 : if (lc_collate_is_c(collid))
1893 : {
1550 tgl 1894 CBC 56902 : if (typid == BPCHAROID)
2622 rhaas 1895 GIC 11 : ssup->comparator = bpcharfastcmp_c;
1550 tgl 1896 56891 : else if (typid == NAMEOID)
1897 : {
1572 1898 38227 : ssup->comparator = namefastcmp_c;
1899 : /* Not supporting abbreviation with type NAME, for now */
1900 38227 : abbreviate = false;
1901 : }
1902 : else
1903 18664 : ssup->comparator = varstrfastcmp_c;
1904 :
2999 rhaas 1905 CBC 56902 : collate_c = true;
1906 : }
2999 rhaas 1907 ECB : else
3160 1908 : {
1909 : /*
1910 : * We need a collation-sensitive comparison. To make things faster,
1911 : * we'll figure out the collation based on the locale id and cache the
1912 : * result.
1913 : */
444 peter 1914 GIC 26141 : locale = pg_newlocale_from_collation(collid);
2023 peter_e 1915 ECB :
1916 : /*
1572 tgl 1917 : * We use varlenafastcmp_locale except for type NAME.
1918 : */
1550 tgl 1919 GIC 26141 : if (typid == NAMEOID)
1920 : {
1572 tgl 1921 UIC 0 : ssup->comparator = namefastcmp_locale;
1922 : /* Not supporting abbreviation with type NAME, for now */
1923 0 : abbreviate = false;
1572 tgl 1924 ECB : }
1925 : else
1572 tgl 1926 CBC 26141 : ssup->comparator = varlenafastcmp_locale;
3160 rhaas 1927 ECB : }
1928 :
3002 1929 : /*
1930 : * Unfortunately, it seems that abbreviation for non-C collations is
1931 : * broken on many common platforms; see pg_strxfrm_enabled().
2573 1932 : *
1933 : * Even apart from the risk of broken locales, it's possible that there
1934 : * are platforms where the use of abbreviated keys should be disabled at
1935 : * compile time. Having only 4 byte datums could make worst-case
1936 : * performance drastically more likely, for example. Moreover, macOS's
1937 : * strxfrm() implementation is known to not effectively concentrate a
1938 : * significant amount of entropy from the original string in earlier
1939 : * transformed blobs. It's possible that other supported platforms are
1940 : * similarly encumbered. So, if we ever get past disabling this
1941 : * categorically, we may still want or need to disable it for particular
1942 : * platforms.
3002 1943 : */
45 jdavis 1944 GNC 83043 : if (!collate_c && !pg_strxfrm_enabled(locale))
2573 rhaas 1945 CBC 105 : abbreviate = false;
1946 :
2999 rhaas 1947 ECB : /*
1948 : * If we're using abbreviated keys, or if we're using a locale-aware
1335 michael 1949 : * comparison, we need to initialize a VarStringSortSupport object. Both
2622 rhaas 1950 : * cases will make use of the temporary buffers we initialize here for
1951 : * scratch space (and to detect requirement for BpChar semantics from
1952 : * caller), and the abbreviation case requires additional state.
1953 : */
2999 rhaas 1954 GIC 83043 : if (abbreviate || !collate_c)
1955 : {
2617 tgl 1956 27973 : sss = palloc(sizeof(VarStringSortSupport));
2622 rhaas 1957 27973 : sss->buf1 = palloc(TEXTBUFLEN);
1958 27973 : sss->buflen1 = TEXTBUFLEN;
2622 rhaas 1959 GBC 27973 : sss->buf2 = palloc(TEXTBUFLEN);
2622 rhaas 1960 GIC 27973 : sss->buflen2 = TEXTBUFLEN;
2739 rhaas 1961 EUB : /* Start with invalid values */
2622 rhaas 1962 GBC 27973 : sss->last_len1 = -1;
2622 rhaas 1963 GIC 27973 : sss->last_len2 = -1;
2728 rhaas 1964 EUB : /* Initialize */
2622 rhaas 1965 GBC 27973 : sss->last_returned = 0;
2622 rhaas 1966 GIC 27973 : sss->locale = locale;
1967 :
1968 : /*
1969 : * To avoid somehow confusing a strxfrm() blob and an original string,
1970 : * constantly keep track of the variety of data that buf1 and buf2
1971 : * currently contain.
1972 : *
2728 rhaas 1973 ECB : * Comparisons may be interleaved with conversion calls. Frequently,
1974 : * conversions and comparisons are batched into two distinct phases,
1975 : * but the correctness of caching cannot hinge upon this. For
1976 : * comparison caching, buffer state is only trusted if cache_blob is
1977 : * found set to false, whereas strxfrm() caching only trusts the state
1978 : * when cache_blob is found set to true.
1979 : *
1980 : * Arbitrarily initialize cache_blob to true.
1981 : */
2622 rhaas 1982 GIC 27973 : sss->cache_blob = true;
1983 27973 : sss->collate_c = collate_c;
1550 tgl 1984 27973 : sss->typid = typid;
2622 rhaas 1985 27973 : ssup->ssup_extra = sss;
1986 :
1987 : /*
1988 : * If possible, plan to use the abbreviated keys optimization. The
1989 : * core code may switch back to authoritative comparator should
1990 : * abbreviation be aborted.
1991 : */
2999 1992 27973 : if (abbreviate)
1993 : {
2622 1994 24109 : sss->prop_card = 0.20;
2622 rhaas 1995 CBC 24109 : initHyperLogLog(&sss->abbr_card, 10);
2622 rhaas 1996 GIC 24109 : initHyperLogLog(&sss->full_card, 10);
2999 1997 24109 : ssup->abbrev_full_comparator = ssup->comparator;
372 john.naylor 1998 CBC 24109 : ssup->comparator = ssup_datum_unsigned_cmp;
2622 rhaas 1999 GIC 24109 : ssup->abbrev_converter = varstr_abbrev_convert;
2000 24109 : ssup->abbrev_abort = varstr_abbrev_abort;
2999 rhaas 2001 ECB : }
2002 : }
3160 rhaas 2003 GIC 83043 : }
2004 :
3160 rhaas 2005 ECB : /*
2006 : * sortsupport comparison func (for C locale case)
3160 rhaas 2007 EUB : */
2008 : static int
2622 rhaas 2009 GIC 63929703 : varstrfastcmp_c(Datum x, Datum y, SortSupport ssup)
3160 rhaas 2010 ECB : {
2617 tgl 2011 GIC 63929703 : VarString *arg1 = DatumGetVarStringPP(x);
2617 tgl 2012 CBC 63929703 : VarString *arg2 = DatumGetVarStringPP(y);
3160 rhaas 2013 ECB : char *a1p,
2014 : *a2p;
2015 : int len1,
2016 : len2,
2017 : result;
2018 :
3160 rhaas 2019 GIC 63929703 : a1p = VARDATA_ANY(arg1);
2020 63929703 : a2p = VARDATA_ANY(arg2);
2021 :
2022 63929703 : len1 = VARSIZE_ANY_EXHDR(arg1);
2023 63929703 : len2 = VARSIZE_ANY_EXHDR(arg2);
2024 :
2025 63929703 : result = memcmp(a1p, a2p, Min(len1, len2));
3160 rhaas 2026 CBC 63929703 : if ((result == 0) && (len1 != len2))
2027 1846743 : result = (len1 < len2) ? -1 : 1;
2028 :
3160 rhaas 2029 ECB : /* We can't afford to leak memory here. */
3160 rhaas 2030 CBC 63929703 : if (PointerGetDatum(arg1) != x)
3160 rhaas 2031 LBC 0 : pfree(arg1);
3160 rhaas 2032 CBC 63929703 : if (PointerGetDatum(arg2) != y)
3160 rhaas 2033 UIC 0 : pfree(arg2);
2034 :
3160 rhaas 2035 GIC 63929703 : return result;
2036 : }
2037 :
2038 : /*
2039 : * sortsupport comparison func (for BpChar C locale case)
2040 : *
2622 rhaas 2041 ECB : * BpChar outsources its sortsupport to this module. Specialization for the
2042 : * varstr_sortsupport BpChar case, modeled on
2043 : * internal_bpchar_pattern_compare().
2044 : */
2045 : static int
2622 rhaas 2046 GIC 8 : bpcharfastcmp_c(Datum x, Datum y, SortSupport ssup)
2622 rhaas 2047 ECB : {
2622 rhaas 2048 GIC 8 : BpChar *arg1 = DatumGetBpCharPP(x);
2049 8 : BpChar *arg2 = DatumGetBpCharPP(y);
2622 rhaas 2050 ECB : char *a1p,
2051 : *a2p;
2052 : int len1,
2053 : len2,
2054 : result;
2055 :
2622 rhaas 2056 CBC 8 : a1p = VARDATA_ANY(arg1);
2622 rhaas 2057 GBC 8 : a2p = VARDATA_ANY(arg2);
2058 :
2622 rhaas 2059 GIC 8 : len1 = bpchartruelen(a1p, VARSIZE_ANY_EXHDR(arg1));
2622 rhaas 2060 CBC 8 : len2 = bpchartruelen(a2p, VARSIZE_ANY_EXHDR(arg2));
2622 rhaas 2061 ECB :
2622 rhaas 2062 CBC 8 : result = memcmp(a1p, a2p, Min(len1, len2));
2622 rhaas 2063 GIC 8 : if ((result == 0) && (len1 != len2))
2622 rhaas 2064 UIC 0 : result = (len1 < len2) ? -1 : 1;
2065 :
2066 : /* We can't afford to leak memory here. */
2622 rhaas 2067 GIC 8 : if (PointerGetDatum(arg1) != x)
2622 rhaas 2068 UIC 0 : pfree(arg1);
2622 rhaas 2069 GIC 8 : if (PointerGetDatum(arg2) != y)
2622 rhaas 2070 UIC 0 : pfree(arg2);
2071 :
2622 rhaas 2072 GIC 8 : return result;
2622 rhaas 2073 ECB : }
2074 :
3160 2075 : /*
1572 tgl 2076 : * sortsupport comparison func (for NAME C locale case)
2077 : */
2078 : static int
1572 tgl 2079 GIC 63594147 : namefastcmp_c(Datum x, Datum y, SortSupport ssup)
2080 : {
2081 63594147 : Name arg1 = DatumGetName(x);
2082 63594147 : Name arg2 = DatumGetName(y);
2083 :
2084 63594147 : return strncmp(NameStr(*arg1), NameStr(*arg2), NAMEDATALEN);
2085 : }
1572 tgl 2086 ECB :
2087 : /*
2088 : * sortsupport comparison func (for locale case with all varlena types)
3160 rhaas 2089 : */
2090 : static int
1572 tgl 2091 GIC 18640117 : varlenafastcmp_locale(Datum x, Datum y, SortSupport ssup)
3160 rhaas 2092 ECB : {
2617 tgl 2093 CBC 18640117 : VarString *arg1 = DatumGetVarStringPP(x);
2617 tgl 2094 GIC 18640117 : VarString *arg2 = DatumGetVarStringPP(y);
2095 : char *a1p,
2096 : *a2p;
2097 : int len1,
2098 : len2,
2099 : result;
2100 :
3160 rhaas 2101 18640117 : a1p = VARDATA_ANY(arg1);
2102 18640117 : a2p = VARDATA_ANY(arg2);
2103 :
2104 18640117 : len1 = VARSIZE_ANY_EXHDR(arg1);
2105 18640117 : len2 = VARSIZE_ANY_EXHDR(arg2);
2106 :
1572 tgl 2107 18640117 : result = varstrfastcmp_locale(a1p, len1, a2p, len2, ssup);
2108 :
2109 : /* We can't afford to leak memory here. */
2110 18640117 : if (PointerGetDatum(arg1) != x)
2111 3 : pfree(arg1);
2112 18640117 : if (PointerGetDatum(arg2) != y)
2113 3 : pfree(arg2);
2114 :
2115 18640117 : return result;
2116 : }
2117 :
2118 : /*
2119 : * sortsupport comparison func (for locale case with NAME type)
2120 : */
2121 : static int
1572 tgl 2122 LBC 0 : namefastcmp_locale(Datum x, Datum y, SortSupport ssup)
1572 tgl 2123 ECB : {
1572 tgl 2124 UIC 0 : Name arg1 = DatumGetName(x);
2125 0 : Name arg2 = DatumGetName(y);
2126 :
2127 0 : return varstrfastcmp_locale(NameStr(*arg1), strlen(NameStr(*arg1)),
2128 0 : NameStr(*arg2), strlen(NameStr(*arg2)),
2129 : ssup);
2130 : }
2131 :
2132 : /*
2133 : * sortsupport comparison func for locale cases
1572 tgl 2134 ECB : */
2135 : static int
1572 tgl 2136 CBC 18640117 : varstrfastcmp_locale(char *a1p, int len1, char *a2p, int len2, SortSupport ssup)
1572 tgl 2137 ECB : {
1572 tgl 2138 GIC 18640117 : VarStringSortSupport *sss = (VarStringSortSupport *) ssup->ssup_extra;
2139 : int result;
2140 : bool arg1_match;
1572 tgl 2141 ECB :
3124 rhaas 2142 : /* Fast pre-check for equality, as discussed in varstr_cmp() */
3124 rhaas 2143 GIC 18640117 : if (len1 == len2 && memcmp(a1p, a2p, len1) == 0)
3124 rhaas 2144 ECB : {
2145 : /*
2739 2146 : * No change in buf1 or buf2 contents, so avoid changing last_len1 or
2147 : * last_len2. Existing contents of buffers might still be used by
2148 : * next call.
2622 2149 : *
2150 : * It's fine to allow the comparison of BpChar padding bytes here,
2151 : * even though that implies that the memcmp() will usually be
2152 : * performed for BpChar callers (though multibyte characters could
2153 : * still prevent that from occurring). The memcmp() is still very
2154 : * cheap, and BpChar's funny semantics have us remove trailing spaces
2617 tgl 2155 : * (not limited to padding), so we need make no distinction between
2156 : * padding space characters and "real" space characters.
2157 : */
1572 tgl 2158 CBC 6879216 : return 0;
2159 : }
3124 rhaas 2160 ECB :
1550 tgl 2161 GIC 11760901 : if (sss->typid == BPCHAROID)
3160 rhaas 2162 EUB : {
2163 : /* Get true number of bytes, ignoring trailing spaces */
2622 rhaas 2164 GBC 16919 : len1 = bpchartruelen(a1p, len1);
2622 rhaas 2165 GIC 16919 : len2 = bpchartruelen(a2p, len2);
2166 : }
2622 rhaas 2167 ECB :
2622 rhaas 2168 GIC 11760901 : if (len1 >= sss->buflen1)
3160 rhaas 2169 ECB : {
2622 rhaas 2170 UIC 0 : sss->buflen1 = Max(len1 + 1, Min(sss->buflen1 * 2, MaxAllocSize));
238 tgl 2171 0 : sss->buf1 = repalloc(sss->buf1, sss->buflen1);
2172 : }
2622 rhaas 2173 GIC 11760901 : if (len2 >= sss->buflen2)
2174 : {
2175 3 : sss->buflen2 = Max(len2 + 1, Min(sss->buflen2 * 2, MaxAllocSize));
238 tgl 2176 3 : sss->buf2 = repalloc(sss->buf2, sss->buflen2);
2177 : }
2178 :
2179 : /*
2180 : * We're likely to be asked to compare the same strings repeatedly, and
2181 : * memcmp() is so much cheaper than strcoll() that it pays to try to cache
2739 rhaas 2182 EUB : * comparisons, even though in general there is no reason to think that
2183 : * that will work out (every string datum may be unique). Caching does
2184 : * not slow things down measurably when it doesn't work out, and can speed
2185 : * things up by rather a lot when it does. In part, this is because the
2186 : * memcmp() compares data from cachelines that are needed in L1 cache even
2187 : * when the last comparison's result cannot be reused.
2188 : */
2739 rhaas 2189 GIC 11760901 : arg1_match = true;
2622 2190 11760901 : if (len1 != sss->last_len1 || memcmp(sss->buf1, a1p, len1) != 0)
2191 : {
2739 rhaas 2192 GBC 10502281 : arg1_match = false;
2622 rhaas 2193 GIC 10502281 : memcpy(sss->buf1, a1p, len1);
2622 rhaas 2194 GBC 10502281 : sss->buf1[len1] = '\0';
2622 rhaas 2195 GIC 10502281 : sss->last_len1 = len1;
2196 : }
2197 :
2198 : /*
2199 : * If we're comparing the same two strings as last time, we can return the
2200 : * same answer without calling strcoll() again. This is more likely than
2201 : * it seems (at least with moderate to low cardinality sets), because
2202 : * quicksort compares the same pivot against many values.
2203 : */
2204 11760901 : if (len2 != sss->last_len2 || memcmp(sss->buf2, a2p, len2) != 0)
2205 : {
2206 1944023 : memcpy(sss->buf2, a2p, len2);
2622 rhaas 2207 CBC 1944023 : sss->buf2[len2] = '\0';
2622 rhaas 2208 GIC 1944023 : sss->last_len2 = len2;
2209 : }
2210 9816878 : else if (arg1_match && !sss->cache_blob)
2211 : {
2212 : /* Use result cached following last actual strcoll() call */
1572 tgl 2213 1067944 : return sss->last_returned;
2214 : }
2215 :
45 jdavis 2216 GNC 10692957 : result = pg_strcoll(sss->buf1, sss->buf2, sss->locale);
2217 :
2218 : /* Break tie if necessary. */
2219 10692957 : if (result == 0 && pg_locale_deterministic(sss->locale))
2622 rhaas 2220 LBC 0 : result = strcmp(sss->buf1, sss->buf2);
2221 :
2222 : /* Cache result, perhaps saving an expensive strcoll() call next time */
2622 rhaas 2223 GIC 10692957 : sss->cache_blob = false;
2622 rhaas 2224 CBC 10692957 : sss->last_returned = result;
3160 rhaas 2225 GIC 10692957 : return result;
2226 : }
8011 tgl 2227 ECB :
3002 rhaas 2228 : /*
2229 : * Conversion routine for sortsupport. Converts original to abbreviated key
2622 2230 : * representation. Our encoding strategy is simple -- pack the first 8 bytes
2231 : * of a strxfrm() blob into a Datum (on little-endian machines, the 8 bytes are
2232 : * stored in reverse order), and treat it as an unsigned integer. When the "C"
2233 : * locale is used, or in case of bytea, just memcpy() from original instead.
2234 : */
2235 : static Datum
2622 rhaas 2236 GIC 577029 : varstr_abbrev_convert(Datum original, SortSupport ssup)
2237 : {
45 jdavis 2238 GNC 577029 : const size_t max_prefix_bytes = sizeof(Datum);
2617 tgl 2239 CBC 577029 : VarStringSortSupport *sss = (VarStringSortSupport *) ssup->ssup_extra;
2617 tgl 2240 GBC 577029 : VarString *authoritative = DatumGetVarStringPP(original);
2617 tgl 2241 GIC 577029 : char *authoritative_data = VARDATA_ANY(authoritative);
3002 rhaas 2242 ECB :
3002 rhaas 2243 EUB : /* working state */
2244 : Datum res;
2245 : char *pres;
2246 : int len;
2247 : uint32 hash;
2248 :
3002 rhaas 2249 GIC 577029 : pres = (char *) &res;
2250 : /* memset(), so any non-overwritten bytes are NUL */
45 jdavis 2251 GNC 577029 : memset(pres, 0, max_prefix_bytes);
3002 rhaas 2252 GIC 577029 : len = VARSIZE_ANY_EXHDR(authoritative);
3002 rhaas 2253 EUB :
2254 : /* Get number of bytes, ignoring trailing spaces */
1550 tgl 2255 GBC 577029 : if (sss->typid == BPCHAROID)
2622 rhaas 2256 GIC 1296 : len = bpchartruelen(authoritative_data, len);
2257 :
2258 : /*
2259 : * If we're using the C collation, use memcpy(), rather than strxfrm(), to
2260 : * abbreviate keys. The full comparator for the C locale is always
2261 : * memcmp(). It would be incorrect to allow bytea callers (callers that
2262 : * always force the C collation -- bytea isn't a collatable type, but this
2263 : * approach is convenient) to use strxfrm(). This is because bytea
2264 : * strings may contain NUL bytes. Besides, this should be faster, too.
2265 : *
2266 : * More generally, it's okay that bytea callers can have NUL bytes in
2267 : * strings because abbreviated cmp need not make a distinction between
2268 : * terminating NUL bytes, and NUL bytes representing actual NULs in the
2269 : * authoritative representation. Hopefully a comparison at or past one
2270 : * abbreviated key's terminating NUL byte will resolve the comparison
2271 : * without consulting the authoritative representation; specifically, some
2272 : * later non-NUL byte in the longer string can resolve the comparison
2273 : * against a subsequent terminating NUL in the shorter string. There will
2274 : * usually be what is effectively a "length-wise" resolution there and
2275 : * then.
2276 : *
2277 : * If that doesn't work out -- if all bytes in the longer string
2617 tgl 2278 ECB : * positioned at or past the offset of the smaller string's (first)
2279 : * terminating NUL are actually representative of NUL bytes in the
2280 : * authoritative binary string (perhaps with some *terminating* NUL bytes
2281 : * towards the end of the longer string iff it happens to still be small)
2282 : * -- then an authoritative tie-breaker will happen, and do the right
2283 : * thing: explicitly consider string length.
2284 : */
2622 rhaas 2285 GIC 577029 : if (sss->collate_c)
45 jdavis 2286 GNC 257041 : memcpy(pres, authoritative_data, Min(len, max_prefix_bytes));
2287 : else
2288 : {
2289 : Size bsize;
2290 :
2291 : /*
2292 : * We're not using the C collation, so fall back on strxfrm or ICU
2293 : * analogs.
2294 : */
2295 :
2296 : /* By convention, we use buffer 1 to store and NUL-terminate */
2622 rhaas 2297 GIC 319988 : if (len >= sss->buflen1)
2298 : {
2299 12 : sss->buflen1 = Max(len + 1, Min(sss->buflen1 * 2, MaxAllocSize));
238 tgl 2300 12 : sss->buf1 = repalloc(sss->buf1, sss->buflen1);
2999 rhaas 2301 ECB : }
3002 2302 :
2303 : /* Might be able to reuse strxfrm() blob from last call */
2622 rhaas 2304 CBC 319988 : if (sss->last_len1 == len && sss->cache_blob &&
2622 rhaas 2305 GIC 312042 : memcmp(sss->buf1, authoritative_data, len) == 0)
2306 : {
45 jdavis 2307 GNC 153194 : memcpy(pres, sss->buf2, Min(max_prefix_bytes, sss->last_len2));
2308 : /* No change affecting cardinality, so no hashing required */
2739 rhaas 2309 GIC 153194 : goto done;
2310 : }
2311 :
2622 2312 166794 : memcpy(sss->buf1, authoritative_data, len);
2313 :
2314 : /*
2315 : * pg_strxfrm() and pg_strxfrm_prefix expect NUL-terminated
2316 : * strings.
2317 : */
2318 166794 : sss->buf1[len] = '\0';
2622 rhaas 2319 CBC 166794 : sss->last_len1 = len;
3002 rhaas 2320 EUB :
45 jdavis 2321 GNC 166794 : if (pg_strxfrm_prefix_enabled(sss->locale))
2322 : {
2323 166794 : if (sss->buflen2 < max_prefix_bytes)
2208 peter_e 2324 ECB : {
45 jdavis 2325 UNC 0 : sss->buflen2 = Max(max_prefix_bytes,
2326 : Min(sss->buflen2 * 2, MaxAllocSize));
2327 0 : sss->buf2 = repalloc(sss->buf2, sss->buflen2);
2208 peter_e 2328 ECB : }
2329 :
45 jdavis 2330 GNC 166794 : bsize = pg_strxfrm_prefix(sss->buf2, sss->buf1,
2331 : max_prefix_bytes, sss->locale);
15 jdavis 2332 CBC 166794 : sss->last_len2 = bsize;
2333 : }
2334 : else
2335 : {
2999 rhaas 2336 ECB : /*
2337 : * Loop: Call pg_strxfrm(), possibly enlarge buffer, and try
2338 : * again. The pg_strxfrm() function leaves the result buffer
2339 : * content undefined if the result did not fit, so we need to
2340 : * retry until everything fits, even though we only need the first
2341 : * few bytes in the end.
2342 : */
2343 : for (;;)
2344 : {
45 jdavis 2345 UNC 0 : bsize = pg_strxfrm(sss->buf2, sss->buf1, sss->buflen2,
2346 : sss->locale);
2347 :
2348 0 : sss->last_len2 = bsize;
2349 0 : if (bsize < sss->buflen2)
2350 0 : break;
2351 :
2352 : /*
2353 : * Grow buffer and retry.
2354 : */
2355 0 : sss->buflen2 = Max(bsize + 1,
2356 : Min(sss->buflen2 * 2, MaxAllocSize));
2357 0 : sss->buf2 = repalloc(sss->buf2, sss->buflen2);
2358 : }
2999 rhaas 2359 ECB : }
2360 :
2361 : /*
2362 : * Every Datum byte is always compared. This is safe because the
2363 : * strxfrm() blob is itself NUL terminated, leaving no danger of
2364 : * misinterpreting any NUL bytes not intended to be interpreted as
2365 : * logically representing termination.
2622 2366 : *
2367 : * (Actually, even if there were NUL bytes in the blob it would be
2368 : * okay. See remarks on bytea case above.)
2369 : */
45 jdavis 2370 GNC 166794 : memcpy(pres, sss->buf2, Min(max_prefix_bytes, bsize));
3002 rhaas 2371 ECB : }
2372 :
2373 : /*
2374 : * Maintain approximate cardinality of both abbreviated keys and original,
2375 : * authoritative keys using HyperLogLog. Used as cheap insurance against
2878 bruce 2376 : * the worst case, where we do many string transformations for no saving
2377 : * in full strcoll()-based comparisons. These statistics are used by
2622 rhaas 2378 : * varstr_abbrev_abort().
3002 2379 : *
2380 : * First, Hash key proper, or a significant fraction of it. Mix in length
2381 : * in order to compensate for cases where differences are past
2998 2382 : * PG_CACHE_LINE_SIZE bytes, so as to limit the overhead of hashing.
3002 2383 : */
2929 rhaas 2384 GIC 423835 : hash = DatumGetUInt32(hash_any((unsigned char *) authoritative_data,
2385 : Min(len, PG_CACHE_LINE_SIZE)));
3002 rhaas 2386 ECB :
3002 rhaas 2387 GIC 423835 : if (len > PG_CACHE_LINE_SIZE)
3002 rhaas 2388 CBC 23 : hash ^= DatumGetUInt32(hash_uint32((uint32) len));
2389 :
2622 rhaas 2390 GIC 423835 : addHyperLogLog(&sss->full_card, hash);
2391 :
3002 rhaas 2392 ECB : /* Hash abbreviated key */
2393 : #if SIZEOF_DATUM == 8
2394 : {
2878 bruce 2395 : uint32 lohalf,
2396 : hihalf;
3002 rhaas 2397 :
3002 rhaas 2398 CBC 423835 : lohalf = (uint32) res;
3002 rhaas 2399 GIC 423835 : hihalf = (uint32) (res >> 32);
2929 2400 423835 : hash = DatumGetUInt32(hash_uint32(lohalf ^ hihalf));
3002 rhaas 2401 ECB : }
2402 : #else /* SIZEOF_DATUM != 8 */
2929 2403 : hash = DatumGetUInt32(hash_uint32((uint32) res));
3002 2404 : #endif
3002 rhaas 2405 EUB :
2622 rhaas 2406 GIC 423835 : addHyperLogLog(&sss->abbr_card, hash);
3002 rhaas 2407 ECB :
2728 2408 : /* Cache result, perhaps saving an expensive strxfrm() call next time */
2622 rhaas 2409 GIC 423835 : sss->cache_blob = true;
2739 2410 577029 : done:
2617 tgl 2411 ECB :
2412 : /*
2739 rhaas 2413 : * Byteswap on little-endian machines.
2414 : *
2415 : * This is needed so that ssup_datum_unsigned_cmp() (an unsigned integer
2416 : * 3-way comparator) works correctly on all platforms. If we didn't do
332 tgl 2417 : * this, the comparator would have to call memcmp() with a pair of
2418 : * pointers to the first byte of each abbreviated key, which is slower.
2739 rhaas 2419 : */
2739 rhaas 2420 CBC 577029 : res = DatumBigEndianToNative(res);
2739 rhaas 2421 ECB :
2841 2422 : /* Don't leak memory here */
2841 rhaas 2423 CBC 577029 : if (PointerGetDatum(authoritative) != original)
2841 rhaas 2424 GIC 6 : pfree(authoritative);
2425 :
3002 rhaas 2426 CBC 577029 : return res;
2427 : }
3002 rhaas 2428 ECB :
3002 rhaas 2429 EUB : /*
2430 : * Callback for estimating effectiveness of abbreviated key optimization, using
2431 : * heuristic rules. Returns value indicating if the abbreviation optimization
3002 rhaas 2432 ECB : * should be aborted, based on its projected effectiveness.
2433 : */
2434 : static bool
2622 rhaas 2435 GIC 1869 : varstr_abbrev_abort(int memtupcount, SortSupport ssup)
3002 rhaas 2436 ECB : {
2617 tgl 2437 GIC 1869 : VarStringSortSupport *sss = (VarStringSortSupport *) ssup->ssup_extra;
2878 bruce 2438 ECB : double abbrev_distinct,
2439 : key_distinct;
2440 :
3002 rhaas 2441 GIC 1869 : Assert(ssup->abbreviate);
3002 rhaas 2442 ECB :
2443 : /* Have a little patience */
2928 rhaas 2444 CBC 1869 : if (memtupcount < 100)
3002 2445 1147 : return false;
2446 :
2622 rhaas 2447 GIC 722 : abbrev_distinct = estimateHyperLogLog(&sss->abbr_card);
2622 rhaas 2448 CBC 722 : key_distinct = estimateHyperLogLog(&sss->full_card);
3002 rhaas 2449 ECB :
2450 : /*
2451 : * Clamp cardinality estimates to at least one distinct value. While
2878 bruce 2452 : * NULLs are generally disregarded, if only NULL values were seen so far,
2453 : * that might misrepresent costs if we failed to clamp.
3002 rhaas 2454 : */
3002 rhaas 2455 GIC 722 : if (abbrev_distinct <= 1.0)
3002 rhaas 2456 UIC 0 : abbrev_distinct = 1.0;
2457 :
3002 rhaas 2458 GBC 722 : if (key_distinct <= 1.0)
3002 rhaas 2459 UIC 0 : key_distinct = 1.0;
3002 rhaas 2460 EUB :
2461 : /*
2462 : * In the worst case all abbreviated keys are identical, while at the same
2463 : * time there are differences within full key strings not captured in
2464 : * abbreviations.
2465 : */
2466 : #ifdef TRACE_SORT
2924 rhaas 2467 GIC 722 : if (trace_sort)
3002 rhaas 2468 EUB : {
2878 bruce 2469 UIC 0 : double norm_abbrev_card = abbrev_distinct / (double) memtupcount;
3002 rhaas 2470 EUB :
2622 rhaas 2471 UIC 0 : elog(LOG, "varstr_abbrev: abbrev_distinct after %d: %f "
2472 : "(key_distinct: %f, norm_abbrev_card: %f, prop_card: %f)",
2473 : memtupcount, abbrev_distinct, key_distinct, norm_abbrev_card,
2474 : sss->prop_card);
2475 : }
2476 : #endif
2477 :
2478 : /*
2479 : * If the number of distinct abbreviated keys approximately matches the
3002 rhaas 2480 ECB : * number of distinct authoritative original keys, that's reason enough to
2481 : * proceed. We can win even with a very low cardinality set if most
2482 : * tie-breakers only memcmp(). This is by far the most important
2483 : * consideration.
2484 : *
2485 : * While comparisons that are resolved at the abbreviated key level are
3002 rhaas 2486 EUB : * considerably cheaper than tie-breakers resolved with memcmp(), both of
2487 : * those two outcomes are so much cheaper than a full strcoll() once
2488 : * sorting is underway that it doesn't seem worth it to weigh abbreviated
2489 : * cardinality against the overall size of the set in order to more
2490 : * accurately model costs. Assume that an abbreviated comparison, and an
2491 : * abbreviated comparison with a cheap memcmp()-based authoritative
2492 : * resolution are equivalent.
2493 : */
2622 rhaas 2494 GBC 722 : if (abbrev_distinct > key_distinct * sss->prop_card)
2495 : {
2496 : /*
2497 : * When we have exceeded 10,000 tuples, decay required cardinality
2928 rhaas 2498 ECB : * aggressively for next call.
2499 : *
2878 bruce 2500 : * This is useful because the number of comparisons required on
2501 : * average increases at a linearithmic rate, and at roughly 10,000
2502 : * tuples that factor will start to dominate over the linear costs of
2503 : * string transformation (this is a conservative estimate). The decay
2878 bruce 2504 EUB : * rate is chosen to be a little less aggressive than halving -- which
2505 : * (since we're called at points at which memtupcount has doubled)
2506 : * would never see the cost model actually abort past the first call
2507 : * following a decay. This decay rate is mostly a precaution against
2508 : * a sudden, violent swing in how well abbreviated cardinality tracks
2509 : * full key cardinality. The decay also serves to prevent a marginal
2510 : * case from being aborted too late, when too much has already been
2511 : * invested in string transformation.
2928 rhaas 2512 : *
2513 : * It's possible for sets of several million distinct strings with
2514 : * mere tens of thousands of distinct abbreviated keys to still
2515 : * benefit very significantly. This will generally occur provided
2878 bruce 2516 : * each abbreviated key is a proxy for a roughly uniform number of the
2517 : * set's full keys. If it isn't so, we hope to catch that early and
2518 : * abort. If it isn't caught early, by the time the problem is
2519 : * apparent it's probably not worth aborting.
2520 : */
2928 rhaas 2521 GIC 693 : if (memtupcount > 10000)
2622 rhaas 2522 GBC 3 : sss->prop_card *= 0.65;
2523 :
3002 2524 693 : return false;
2525 : }
2526 :
2527 : /*
2528 : * Abort abbreviation strategy.
2529 : *
2530 : * The worst case, where all abbreviated keys are identical while all
2531 : * original strings differ will typically only see a regression of about
2532 : * 10% in execution time for small to medium sized lists of strings.
2533 : * Whereas on modern CPUs where cache stalls are the dominant cost, we can
2534 : * often expect very large improvements, particularly with sets of strings
2535 : * of moderately high to high abbreviated cardinality. There is little to
2536 : * lose but much to gain, which our strategy reflects.
2537 : */
2538 : #ifdef TRACE_SORT
2924 rhaas 2539 CBC 29 : if (trace_sort)
2622 rhaas 2540 UIC 0 : elog(LOG, "varstr_abbrev: aborted abbreviation at %d "
2541 : "(abbrev_distinct: %f, key_distinct: %f, prop_card: %f)",
2542 : memtupcount, abbrev_distinct, key_distinct, sss->prop_card);
2543 : #endif
2544 :
3002 rhaas 2545 CBC 29 : return true;
3002 rhaas 2546 ECB : }
2547 :
1138 pg 2548 : /*
2549 : * Generic equalimage support function for character type's operator classes.
2550 : * Disables the use of deduplication with nondeterministic collations.
2551 : */
1138 pg 2552 EUB : Datum
1138 pg 2553 CBC 13033 : btvarstrequalimage(PG_FUNCTION_ARGS)
1138 pg 2554 ECB : {
2555 : /* Oid opcintype = PG_GETARG_OID(0); */
1138 pg 2556 CBC 13033 : Oid collid = PG_GET_COLLATION();
2557 :
1138 pg 2558 GIC 13033 : check_collation_set(collid);
2559 :
2560 13033 : if (lc_collate_is_c(collid) ||
1138 pg 2561 CBC 25 : collid == DEFAULT_COLLATION_OID ||
1138 pg 2562 GIC 25 : get_collation_isdeterministic(collid))
1138 pg 2563 CBC 13023 : PG_RETURN_BOOL(true);
1138 pg 2564 ECB : else
1138 pg 2565 GIC 10 : PG_RETURN_BOOL(false);
2566 : }
1138 pg 2567 ECB :
2568 : Datum
8312 tgl 2569 CBC 114807 : text_larger(PG_FUNCTION_ARGS)
8888 lockhart 2570 ECB : {
5847 tgl 2571 GIC 114807 : text *arg1 = PG_GETARG_TEXT_PP(0);
5847 tgl 2572 CBC 114807 : text *arg2 = PG_GETARG_TEXT_PP(1);
2573 : text *result;
2574 :
4443 peter_e 2575 GIC 114807 : result = ((text_cmp(arg1, arg2, PG_GET_COLLATION()) > 0) ? arg1 : arg2);
2576 :
8312 tgl 2577 CBC 114807 : PG_RETURN_TEXT_P(result);
2578 : }
8888 lockhart 2579 ECB :
8312 tgl 2580 : Datum
8312 tgl 2581 GIC 43065 : text_smaller(PG_FUNCTION_ARGS)
2582 : {
5847 tgl 2583 CBC 43065 : text *arg1 = PG_GETARG_TEXT_PP(0);
5847 tgl 2584 GIC 43065 : text *arg2 = PG_GETARG_TEXT_PP(1);
8720 bruce 2585 ECB : text *result;
8888 lockhart 2586 :
4443 peter_e 2587 GIC 43065 : result = ((text_cmp(arg1, arg2, PG_GET_COLLATION()) < 0) ? arg1 : arg2);
8888 lockhart 2588 ECB :
8312 tgl 2589 GIC 43065 : PG_RETURN_TEXT_P(result);
2590 : }
2591 :
2592 :
1572 tgl 2593 ECB : /*
2594 : * Cross-type comparison functions for types text and name.
2595 : */
2596 :
2597 : Datum
1572 tgl 2598 GIC 98804 : nameeqtext(PG_FUNCTION_ARGS)
1572 tgl 2599 ECB : {
1572 tgl 2600 GIC 98804 : Name arg1 = PG_GETARG_NAME(0);
1572 tgl 2601 CBC 98804 : text *arg2 = PG_GETARG_TEXT_PP(1);
2602 98804 : size_t len1 = strlen(NameStr(*arg1));
1572 tgl 2603 GIC 98804 : size_t len2 = VARSIZE_ANY_EXHDR(arg2);
1479 peter 2604 CBC 98804 : Oid collid = PG_GET_COLLATION();
2605 : bool result;
2606 :
1479 peter 2607 GIC 98804 : check_collation_set(collid);
2608 :
1479 peter 2609 CBC 98804 : if (collid == C_COLLATION_OID)
1479 peter 2610 GIC 148014 : result = (len1 == len2 &&
1479 peter 2611 CBC 65114 : memcmp(NameStr(*arg1), VARDATA_ANY(arg2), len1) == 0);
1479 peter 2612 ECB : else
1479 peter 2613 GIC 15904 : result = (varstr_cmp(NameStr(*arg1), len1,
2614 15904 : VARDATA_ANY(arg2), len2,
1479 peter 2615 ECB : collid) == 0);
2616 :
1572 tgl 2617 CBC 98804 : PG_FREE_IF_COPY(arg2, 1);
1572 tgl 2618 ECB :
1572 tgl 2619 GIC 98804 : PG_RETURN_BOOL(result);
1572 tgl 2620 ECB : }
2621 :
2622 : Datum
1572 tgl 2623 GIC 3369 : texteqname(PG_FUNCTION_ARGS)
2624 : {
1572 tgl 2625 CBC 3369 : text *arg1 = PG_GETARG_TEXT_PP(0);
1572 tgl 2626 GIC 3369 : Name arg2 = PG_GETARG_NAME(1);
1572 tgl 2627 CBC 3369 : size_t len1 = VARSIZE_ANY_EXHDR(arg1);
2628 3369 : size_t len2 = strlen(NameStr(*arg2));
1479 peter 2629 GIC 3369 : Oid collid = PG_GET_COLLATION();
2630 : bool result;
1572 tgl 2631 ECB :
1479 peter 2632 GIC 3369 : check_collation_set(collid);
1479 peter 2633 ECB :
1479 peter 2634 CBC 3369 : if (collid == C_COLLATION_OID)
1479 peter 2635 GIC 282 : result = (len1 == len2 &&
1479 peter 2636 CBC 90 : memcmp(VARDATA_ANY(arg1), NameStr(*arg2), len1) == 0);
2637 : else
1479 peter 2638 GIC 3177 : result = (varstr_cmp(VARDATA_ANY(arg1), len1,
2639 3177 : NameStr(*arg2), len2,
2640 : collid) == 0);
1572 tgl 2641 ECB :
1572 tgl 2642 GIC 3369 : PG_FREE_IF_COPY(arg1, 0);
1572 tgl 2643 ECB :
1572 tgl 2644 GIC 3369 : PG_RETURN_BOOL(result);
2645 : }
1572 tgl 2646 ECB :
2647 : Datum
1572 tgl 2648 GIC 18 : namenetext(PG_FUNCTION_ARGS)
1572 tgl 2649 ECB : {
1572 tgl 2650 GIC 18 : Name arg1 = PG_GETARG_NAME(0);
1572 tgl 2651 CBC 18 : text *arg2 = PG_GETARG_TEXT_PP(1);
1572 tgl 2652 GIC 18 : size_t len1 = strlen(NameStr(*arg1));
1572 tgl 2653 CBC 18 : size_t len2 = VARSIZE_ANY_EXHDR(arg2);
1479 peter 2654 GIC 18 : Oid collid = PG_GET_COLLATION();
2655 : bool result;
2656 :
2657 18 : check_collation_set(collid);
2658 :
2659 18 : if (collid == C_COLLATION_OID)
2660 9 : result = !(len1 == len2 &&
1479 peter 2661 UIC 0 : memcmp(NameStr(*arg1), VARDATA_ANY(arg2), len1) == 0);
2662 : else
1479 peter 2663 GIC 9 : result = !(varstr_cmp(NameStr(*arg1), len1,
1479 peter 2664 CBC 9 : VARDATA_ANY(arg2), len2,
2665 : collid) == 0);
1572 tgl 2666 ECB :
1572 tgl 2667 GIC 18 : PG_FREE_IF_COPY(arg2, 1);
2668 :
1572 tgl 2669 CBC 18 : PG_RETURN_BOOL(result);
2670 : }
2671 :
2672 : Datum
1572 tgl 2673 GIC 9 : textnename(PG_FUNCTION_ARGS)
2674 : {
2675 9 : text *arg1 = PG_GETARG_TEXT_PP(0);
2676 9 : Name arg2 = PG_GETARG_NAME(1);
2677 9 : size_t len1 = VARSIZE_ANY_EXHDR(arg1);
2678 9 : size_t len2 = strlen(NameStr(*arg2));
1479 peter 2679 9 : Oid collid = PG_GET_COLLATION();
1572 tgl 2680 ECB : bool result;
2681 :
1479 peter 2682 CBC 9 : check_collation_set(collid);
1479 peter 2683 ECB :
1479 peter 2684 GIC 9 : if (collid == C_COLLATION_OID)
1479 peter 2685 LBC 0 : result = !(len1 == len2 &&
1479 peter 2686 UIC 0 : memcmp(VARDATA_ANY(arg1), NameStr(*arg2), len1) == 0);
2687 : else
1479 peter 2688 GIC 9 : result = !(varstr_cmp(VARDATA_ANY(arg1), len1,
2689 9 : NameStr(*arg2), len2,
2690 : collid) == 0);
2691 :
1572 tgl 2692 9 : PG_FREE_IF_COPY(arg1, 0);
2693 :
2694 9 : PG_RETURN_BOOL(result);
1572 tgl 2695 ECB : }
2696 :
2697 : Datum
1572 tgl 2698 GIC 81923 : btnametextcmp(PG_FUNCTION_ARGS)
2699 : {
2700 81923 : Name arg1 = PG_GETARG_NAME(0);
2701 81923 : text *arg2 = PG_GETARG_TEXT_PP(1);
2702 : int32 result;
1572 tgl 2703 ECB :
1572 tgl 2704 CBC 163846 : result = varstr_cmp(NameStr(*arg1), strlen(NameStr(*arg1)),
1572 tgl 2705 GIC 163846 : VARDATA_ANY(arg2), VARSIZE_ANY_EXHDR(arg2),
2706 : PG_GET_COLLATION());
1572 tgl 2707 ECB :
1572 tgl 2708 GBC 81923 : PG_FREE_IF_COPY(arg2, 1);
1572 tgl 2709 ECB :
1572 tgl 2710 GBC 81923 : PG_RETURN_INT32(result);
2711 : }
1572 tgl 2712 ECB :
2713 : Datum
1572 tgl 2714 UIC 0 : bttextnamecmp(PG_FUNCTION_ARGS)
2715 : {
1572 tgl 2716 LBC 0 : text *arg1 = PG_GETARG_TEXT_PP(0);
1572 tgl 2717 UIC 0 : Name arg2 = PG_GETARG_NAME(1);
2718 : int32 result;
1572 tgl 2719 ECB :
1572 tgl 2720 LBC 0 : result = varstr_cmp(VARDATA_ANY(arg1), VARSIZE_ANY_EXHDR(arg1),
2721 0 : NameStr(*arg2), strlen(NameStr(*arg2)),
1572 tgl 2722 ECB : PG_GET_COLLATION());
2723 :
1572 tgl 2724 UIC 0 : PG_FREE_IF_COPY(arg1, 0);
1572 tgl 2725 ECB :
1572 tgl 2726 UIC 0 : PG_RETURN_INT32(result);
2727 : }
2728 :
2729 : #define CmpCall(cmpfunc) \
2730 : DatumGetInt32(DirectFunctionCall2Coll(cmpfunc, \
2731 : PG_GET_COLLATION(), \
2732 : PG_GETARG_DATUM(0), \
2733 : PG_GETARG_DATUM(1)))
2734 :
2735 : Datum
1572 tgl 2736 GIC 26815 : namelttext(PG_FUNCTION_ARGS)
2737 : {
2738 26815 : PG_RETURN_BOOL(CmpCall(btnametextcmp) < 0);
2739 : }
2740 :
2741 : Datum
1572 tgl 2742 UIC 0 : nameletext(PG_FUNCTION_ARGS)
2743 : {
2744 0 : PG_RETURN_BOOL(CmpCall(btnametextcmp) <= 0);
2745 : }
2746 :
1572 tgl 2747 ECB : Datum
1572 tgl 2748 UIC 0 : namegttext(PG_FUNCTION_ARGS)
1572 tgl 2749 ECB : {
1572 tgl 2750 UIC 0 : PG_RETURN_BOOL(CmpCall(btnametextcmp) > 0);
2751 : }
2752 :
2753 : Datum
1572 tgl 2754 GIC 25974 : namegetext(PG_FUNCTION_ARGS)
2755 : {
2756 25974 : PG_RETURN_BOOL(CmpCall(btnametextcmp) >= 0);
2757 : }
2758 :
2759 : Datum
1572 tgl 2760 UIC 0 : textltname(PG_FUNCTION_ARGS)
1572 tgl 2761 ECB : {
1572 tgl 2762 UIC 0 : PG_RETURN_BOOL(CmpCall(bttextnamecmp) < 0);
1572 tgl 2763 ECB : }
2764 :
2765 : Datum
1572 tgl 2766 UIC 0 : textlename(PG_FUNCTION_ARGS)
2767 : {
2768 0 : PG_RETURN_BOOL(CmpCall(bttextnamecmp) <= 0);
2769 : }
1572 tgl 2770 ECB :
2771 : Datum
1572 tgl 2772 UIC 0 : textgtname(PG_FUNCTION_ARGS)
2773 : {
2774 0 : PG_RETURN_BOOL(CmpCall(bttextnamecmp) > 0);
2775 : }
2776 :
2777 : Datum
2778 0 : textgename(PG_FUNCTION_ARGS)
2779 : {
2780 0 : PG_RETURN_BOOL(CmpCall(bttextnamecmp) >= 0);
2781 : }
1572 tgl 2782 ECB :
2783 : #undef CmpCall
2784 :
2785 :
2786 : /*
2787 : * The following operators support character-by-character comparison
2788 : * of text datums, to allow building indexes suitable for LIKE clauses.
2789 : * Note that the regular texteq/textne comparison operators, and regular
2622 rhaas 2790 : * support functions 1 and 2 with "C" collation are assumed to be
2791 : * compatible with these!
7269 peter_e 2792 : */
2793 :
2794 : static int
1296 tgl 2795 CBC 76040 : internal_text_pattern_compare(text *arg1, text *arg2)
2796 : {
2797 : int result;
2798 : int len1,
2799 : len2;
5430 tgl 2800 ECB :
5430 tgl 2801 GIC 76040 : len1 = VARSIZE_ANY_EXHDR(arg1);
2802 76040 : len2 = VARSIZE_ANY_EXHDR(arg2);
2803 :
4492 rhaas 2804 76040 : result = memcmp(VARDATA_ANY(arg1), VARDATA_ANY(arg2), Min(len1, len2));
7269 peter_e 2805 76040 : if (result != 0)
7269 peter_e 2806 CBC 76013 : return result;
5430 tgl 2807 GIC 27 : else if (len1 < len2)
7269 peter_e 2808 UIC 0 : return -1;
5430 tgl 2809 GIC 27 : else if (len1 > len2)
7269 peter_e 2810 9 : return 1;
2811 : else
2812 18 : return 0;
2813 : }
2814 :
7269 peter_e 2815 ECB :
7269 peter_e 2816 EUB : Datum
7269 peter_e 2817 GIC 19769 : text_pattern_lt(PG_FUNCTION_ARGS)
7269 peter_e 2818 ECB : {
5847 tgl 2819 GIC 19769 : text *arg1 = PG_GETARG_TEXT_PP(0);
2820 19769 : text *arg2 = PG_GETARG_TEXT_PP(1);
2821 : int result;
2822 :
1296 2823 19769 : result = internal_text_pattern_compare(arg1, arg2);
2824 :
7269 peter_e 2825 19769 : PG_FREE_IF_COPY(arg1, 0);
7269 peter_e 2826 CBC 19769 : PG_FREE_IF_COPY(arg2, 1);
2827 :
7269 peter_e 2828 GIC 19769 : PG_RETURN_BOOL(result < 0);
2829 : }
2830 :
2831 :
2832 : Datum
2833 18755 : text_pattern_le(PG_FUNCTION_ARGS)
2834 : {
5847 tgl 2835 18755 : text *arg1 = PG_GETARG_TEXT_PP(0);
2836 18755 : text *arg2 = PG_GETARG_TEXT_PP(1);
7269 peter_e 2837 ECB : int result;
2838 :
1296 tgl 2839 CBC 18755 : result = internal_text_pattern_compare(arg1, arg2);
7269 peter_e 2840 ECB :
7269 peter_e 2841 CBC 18755 : PG_FREE_IF_COPY(arg1, 0);
2842 18755 : PG_FREE_IF_COPY(arg2, 1);
2843 :
2844 18755 : PG_RETURN_BOOL(result <= 0);
2845 : }
2846 :
2847 :
7269 peter_e 2848 ECB : Datum
7269 peter_e 2849 GIC 18755 : text_pattern_ge(PG_FUNCTION_ARGS)
7269 peter_e 2850 ECB : {
5847 tgl 2851 CBC 18755 : text *arg1 = PG_GETARG_TEXT_PP(0);
2852 18755 : text *arg2 = PG_GETARG_TEXT_PP(1);
2853 : int result;
2854 :
1296 2855 18755 : result = internal_text_pattern_compare(arg1, arg2);
7269 peter_e 2856 ECB :
7269 peter_e 2857 GIC 18755 : PG_FREE_IF_COPY(arg1, 0);
2858 18755 : PG_FREE_IF_COPY(arg2, 1);
2859 :
7269 peter_e 2860 CBC 18755 : PG_RETURN_BOOL(result >= 0);
2861 : }
2862 :
2863 :
2864 : Datum
7269 peter_e 2865 GIC 18755 : text_pattern_gt(PG_FUNCTION_ARGS)
2866 : {
5847 tgl 2867 18755 : text *arg1 = PG_GETARG_TEXT_PP(0);
2868 18755 : text *arg2 = PG_GETARG_TEXT_PP(1);
2869 : int result;
2870 :
1296 2871 18755 : result = internal_text_pattern_compare(arg1, arg2);
7269 peter_e 2872 ECB :
7269 peter_e 2873 GBC 18755 : PG_FREE_IF_COPY(arg1, 0);
7269 peter_e 2874 GIC 18755 : PG_FREE_IF_COPY(arg2, 1);
2875 :
7269 peter_e 2876 CBC 18755 : PG_RETURN_BOOL(result > 0);
7269 peter_e 2877 EUB : }
2878 :
2879 :
2880 : Datum
7269 peter_e 2881 CBC 6 : bttext_pattern_cmp(PG_FUNCTION_ARGS)
7269 peter_e 2882 ECB : {
5847 tgl 2883 CBC 6 : text *arg1 = PG_GETARG_TEXT_PP(0);
2884 6 : text *arg2 = PG_GETARG_TEXT_PP(1);
2885 : int result;
7269 peter_e 2886 ECB :
1296 tgl 2887 GIC 6 : result = internal_text_pattern_compare(arg1, arg2);
2888 :
7269 peter_e 2889 6 : PG_FREE_IF_COPY(arg1, 0);
2890 6 : PG_FREE_IF_COPY(arg2, 1);
2891 :
2892 6 : PG_RETURN_INT32(result);
7269 peter_e 2893 ECB : }
2894 :
2895 :
2896 : Datum
2622 rhaas 2897 CBC 58 : bttext_pattern_sortsupport(PG_FUNCTION_ARGS)
2898 : {
2622 rhaas 2899 GIC 58 : SortSupport ssup = (SortSupport) PG_GETARG_POINTER(0);
2900 : MemoryContext oldcontext;
2901 :
2902 58 : oldcontext = MemoryContextSwitchTo(ssup->ssup_cxt);
2903 :
2904 : /* Use generic string SortSupport, forcing "C" collation */
1572 tgl 2905 58 : varstr_sortsupport(ssup, TEXTOID, C_COLLATION_OID);
2906 :
2622 rhaas 2907 GBC 58 : MemoryContextSwitchTo(oldcontext);
2908 :
2909 58 : PG_RETURN_VOID();
2622 rhaas 2910 EUB : }
2911 :
2912 :
2913 : /*-------------------------------------------------------------
2914 : * byteaoctetlen
2915 : *
2916 : * get the number of bytes contained in an instance of type 'bytea'
2917 : *-------------------------------------------------------------
2918 : */
8289 tgl 2919 : Datum
8289 tgl 2920 GBC 157 : byteaoctetlen(PG_FUNCTION_ARGS)
2921 : {
7008 2922 157 : Datum str = PG_GETARG_DATUM(0);
7008 tgl 2923 EUB :
2924 : /* We need not detoast the input at all */
7008 tgl 2925 GBC 157 : PG_RETURN_INT32(toast_raw_datum_size(str) - VARHDRSZ);
9770 scrappy 2926 EUB : }
2927 :
7877 bruce 2928 : /*
2929 : * byteacat -
2930 : * takes two bytea* and returns a bytea* that is the concatenation of
2931 : * the two.
2932 : *
2933 : * Cloned from textcat and modified as required.
2934 : */
2935 : Datum
7877 bruce 2936 GIC 760 : byteacat(PG_FUNCTION_ARGS)
7877 bruce 2937 EUB : {
5847 tgl 2938 GIC 760 : bytea *t1 = PG_GETARG_BYTEA_PP(0);
2939 760 : bytea *t2 = PG_GETARG_BYTEA_PP(1);
4822 tgl 2940 EUB :
4822 tgl 2941 GIC 760 : PG_RETURN_BYTEA_P(bytea_catenate(t1, t2));
2942 : }
2943 :
2944 : /*
2945 : * bytea_catenate
2946 : * Guts of byteacat(), broken out so it can be used by other functions
2947 : *
2948 : * Arguments can be in short-header form, but not compressed or out-of-line
2949 : */
2950 : static bytea *
4822 tgl 2951 CBC 778 : bytea_catenate(bytea *t1, bytea *t2)
2952 : {
4822 tgl 2953 ECB : bytea *result;
7877 bruce 2954 : int len1,
2955 : len2,
2956 : len;
2957 : char *ptr;
2958 :
5847 tgl 2959 GIC 778 : len1 = VARSIZE_ANY_EXHDR(t1);
4822 tgl 2960 CBC 778 : len2 = VARSIZE_ANY_EXHDR(t2);
4822 tgl 2961 ECB :
2962 : /* paranoia ... probably should throw error instead? */
7877 bruce 2963 GIC 778 : if (len1 < 0)
7877 bruce 2964 UIC 0 : len1 = 0;
7877 bruce 2965 GIC 778 : if (len2 < 0)
7877 bruce 2966 LBC 0 : len2 = 0;
2967 :
7877 bruce 2968 CBC 778 : len = len1 + len2 + VARHDRSZ;
7877 bruce 2969 GIC 778 : result = (bytea *) palloc(len);
2970 :
2971 : /* Set size of result string... */
5885 tgl 2972 778 : SET_VARSIZE(result, len);
2973 :
2974 : /* Fill data field of result string... */
7877 bruce 2975 778 : ptr = VARDATA(result);
2976 778 : if (len1 > 0)
5847 tgl 2977 778 : memcpy(ptr, VARDATA_ANY(t1), len1);
7877 bruce 2978 778 : if (len2 > 0)
5847 tgl 2979 769 : memcpy(ptr + len1, VARDATA_ANY(t2), len2);
7877 bruce 2980 ECB :
4822 tgl 2981 GIC 778 : return result;
7877 bruce 2982 ECB : }
2983 :
2984 : #define PG_STR_GET_BYTEA(str_) \
2985 : DatumGetByteaPP(DirectFunctionCall1(byteain, CStringGetDatum(str_)))
2986 :
2987 : /*
2988 : * bytea_substr()
2989 : * Return a substring starting at the specified position.
2990 : * Cloned from text_substr and modified as required.
2991 : *
2992 : * Input:
2993 : * - string
2994 : * - starting position (is one-based)
2995 : * - string length (optional)
2996 : *
2997 : * If the starting position is zero or less, then return from the start of the string
3641 peter_e 2998 : * adjusting the length to be consistent with the "negative start" per SQL.
7535 bruce 2999 : * If the length is less than zero, an ERROR is thrown. If no third argument
3000 : * (length) is provided, the length to the end of the string is assumed.
7877 3001 : */
3002 : Datum
7877 bruce 3003 CBC 43 : bytea_substr(PG_FUNCTION_ARGS)
7877 bruce 3004 ECB : {
4822 tgl 3005 GIC 43 : PG_RETURN_BYTEA_P(bytea_substring(PG_GETARG_DATUM(0),
4822 tgl 3006 EUB : PG_GETARG_INT32(1),
3007 : PG_GETARG_INT32(2),
3008 : false));
3009 : }
3010 :
3011 : /*
3012 : * bytea_substr_no_len -
3013 : * Wrapper to avoid opr_sanity failure due to
3014 : * one function accepting a different number of args.
3015 : */
3016 : Datum
4822 tgl 3017 GIC 1950 : bytea_substr_no_len(PG_FUNCTION_ARGS)
4822 tgl 3018 ECB : {
4822 tgl 3019 GIC 1950 : PG_RETURN_BYTEA_P(bytea_substring(PG_GETARG_DATUM(0),
4822 tgl 3020 ECB : PG_GETARG_INT32(1),
3021 : -1,
3022 : true));
3023 : }
3024 :
3025 : static bytea *
4822 tgl 3026 GIC 2011 : bytea_substring(Datum str,
4822 tgl 3027 ECB : int S,
3028 : int L,
3029 : bool length_not_specified)
3030 : {
3031 : int32 S1; /* adjusted start position */
3032 : int32 L1; /* adjusted substring length */
3033 : int32 E; /* end position */
3034 :
3035 : /*
825 3036 : * The logic here should generally match text_substring().
3037 : */
7535 bruce 3038 CBC 2011 : S1 = Max(S, 1);
3039 :
4822 tgl 3040 GIC 2011 : if (length_not_specified)
3041 : {
3042 : /*
3043 : * Not passed a length - DatumGetByteaPSlice() grabs everything to the
3044 : * end of the string if we pass it a negative value for length.
3045 : */
7535 bruce 3046 1959 : L1 = -1;
3047 : }
825 tgl 3048 52 : else if (L < 0)
3049 : {
825 tgl 3050 ECB : /* SQL99 says to throw an error for E < S, i.e., negative length */
825 tgl 3051 GIC 6 : ereport(ERROR,
825 tgl 3052 ECB : (errcode(ERRCODE_SUBSTRING_ERROR),
3053 : errmsg("negative substring length not allowed")));
3054 : L1 = -1; /* silence stupider compilers */
3055 : }
825 tgl 3056 GIC 46 : else if (pg_add_s32_overflow(S, L, &E))
3057 : {
3058 : /*
3059 : * L could be large enough for S + L to overflow, in which case the
3060 : * substring must run to end of string.
7535 bruce 3061 ECB : */
825 tgl 3062 GIC 3 : L1 = -1;
825 tgl 3063 ECB : }
3064 : else
3065 : {
3066 : /*
3067 : * A zero or negative value for the end position can happen if the
3068 : * start was negative or one. SQL99 says to return a zero-length
3069 : * string.
7535 bruce 3070 : */
7535 bruce 3071 CBC 43 : if (E < 1)
4822 tgl 3072 UIC 0 : return PG_STR_GET_BYTEA("");
3073 :
7535 bruce 3074 GIC 43 : L1 = E - S1;
3075 : }
7535 bruce 3076 ECB :
7522 bruce 3077 EUB : /*
3078 : * If the start position is past the end of the string, SQL99 says to
3079 : * return a zero-length string -- DatumGetByteaPSlice() will do that for
3080 : * us. We need only convert S1 to zero-based starting position.
3081 : */
4822 tgl 3082 GIC 2005 : return DatumGetByteaPSlice(str, S1 - 1, L1);
3083 : }
7535 bruce 3084 ECB :
3085 : /*
4822 tgl 3086 : * byteaoverlay
3087 : * Replace specified substring of first string with second
3088 : *
4822 tgl 3089 EUB : * The SQL standard defines OVERLAY() in terms of substring and concatenation.
3090 : * This code is a direct implementation of what the standard says.
7535 bruce 3091 ECB : */
3092 : Datum
4822 tgl 3093 CBC 3 : byteaoverlay(PG_FUNCTION_ARGS)
3094 : {
4822 tgl 3095 GIC 3 : bytea *t1 = PG_GETARG_BYTEA_PP(0);
3096 3 : bytea *t2 = PG_GETARG_BYTEA_PP(1);
2118 3097 3 : int sp = PG_GETARG_INT32(2); /* substring start position */
3098 3 : int sl = PG_GETARG_INT32(3); /* substring length */
3099 :
4822 3100 3 : PG_RETURN_BYTEA_P(bytea_overlay(t1, t2, sp, sl));
4822 tgl 3101 ECB : }
3102 :
3103 : Datum
4822 tgl 3104 GIC 6 : byteaoverlay_no_len(PG_FUNCTION_ARGS)
3105 : {
3106 6 : bytea *t1 = PG_GETARG_BYTEA_PP(0);
4822 tgl 3107 CBC 6 : bytea *t2 = PG_GETARG_BYTEA_PP(1);
2118 tgl 3108 GIC 6 : int sp = PG_GETARG_INT32(2); /* substring start position */
3109 : int sl;
4822 tgl 3110 ECB :
4790 bruce 3111 CBC 6 : sl = VARSIZE_ANY_EXHDR(t2); /* defaults to length(t2) */
4822 tgl 3112 GIC 6 : PG_RETURN_BYTEA_P(bytea_overlay(t1, t2, sp, sl));
3113 : }
4822 tgl 3114 ECB :
3115 : static bytea *
4822 tgl 3116 GIC 9 : bytea_overlay(bytea *t1, bytea *t2, int sp, int sl)
4822 tgl 3117 ECB : {
3118 : bytea *result;
3119 : bytea *s1;
3120 : bytea *s2;
3121 : int sp_pl_sl;
3122 :
3123 : /*
4790 bruce 3124 : * Check for possible integer-overflow cases. For negative sp, throw a
3125 : * "substring length" error because that's what should be expected
3126 : * according to the spec's definition of OVERLAY().
3127 : */
4822 tgl 3128 CBC 9 : if (sp <= 0)
4822 tgl 3129 UIC 0 : ereport(ERROR,
3130 : (errcode(ERRCODE_SUBSTRING_ERROR),
3131 : errmsg("negative substring length not allowed")));
1944 andres 3132 GIC 9 : if (pg_add_s32_overflow(sp, sl, &sp_pl_sl))
4822 tgl 3133 UIC 0 : ereport(ERROR,
3134 : (errcode(ERRCODE_NUMERIC_VALUE_OUT_OF_RANGE),
3135 : errmsg("integer out of range")));
3136 :
4790 bruce 3137 GIC 9 : s1 = bytea_substring(PointerGetDatum(t1), 1, sp - 1, false);
4822 tgl 3138 9 : s2 = bytea_substring(PointerGetDatum(t1), sp_pl_sl, -1, true);
3139 9 : result = bytea_catenate(s1, t2);
3140 9 : result = bytea_catenate(result, s2);
4822 tgl 3141 ECB :
4822 tgl 3142 GIC 9 : return result;
3143 : }
7877 bruce 3144 ECB :
3145 : /*
3146 : * bit_count
3147 : */
3148 : Datum
747 peter 3149 GIC 3 : bytea_bit_count(PG_FUNCTION_ARGS)
747 peter 3150 ECB : {
747 peter 3151 GIC 3 : bytea *t1 = PG_GETARG_BYTEA_PP(0);
747 peter 3152 ECB :
747 peter 3153 GBC 3 : PG_RETURN_INT64(pg_popcount(VARDATA_ANY(t1), VARSIZE_ANY_EXHDR(t1)));
3154 : }
3155 :
3156 : /*
7877 bruce 3157 ECB : * byteapos -
7877 bruce 3158 EUB : * Return the position of the specified substring.
3159 : * Implements the SQL POSITION() function.
3160 : * Cloned from textpos and modified as required.
3161 : */
7877 bruce 3162 ECB : Datum
7877 bruce 3163 UIC 0 : byteapos(PG_FUNCTION_ARGS)
7877 bruce 3164 ECB : {
5847 tgl 3165 UIC 0 : bytea *t1 = PG_GETARG_BYTEA_PP(0);
5847 tgl 3166 LBC 0 : bytea *t2 = PG_GETARG_BYTEA_PP(1);
3167 : int pos;
3168 : int px,
7877 bruce 3169 ECB : p;
3170 : int len1,
3171 : len2;
7836 3172 : char *p1,
3173 : *p2;
3174 :
5847 tgl 3175 UIC 0 : len1 = VARSIZE_ANY_EXHDR(t1);
3176 0 : len2 = VARSIZE_ANY_EXHDR(t2);
3177 :
3178 0 : if (len2 <= 0)
3179 0 : PG_RETURN_INT32(1); /* result for empty pattern */
3180 :
3181 0 : p1 = VARDATA_ANY(t1);
3182 0 : p2 = VARDATA_ANY(t2);
3183 :
7877 bruce 3184 0 : pos = 0;
3185 0 : px = (len1 - len2);
3186 0 : for (p = 0; p <= px; p++)
3187 : {
3188 0 : if ((*p2 == *p1) && (memcmp(p1, p2, len2) == 0))
3189 : {
3190 0 : pos = p + 1;
3191 0 : break;
3192 : };
3193 0 : p1++;
3194 : };
3195 :
3196 0 : PG_RETURN_INT32(pos);
3197 : }
3198 :
9770 scrappy 3199 ECB : /*-------------------------------------------------------------
3200 : * byteaGetByte
3201 : *
3202 : * this routine treats "bytea" as an array of bytes.
8335 tgl 3203 : * It returns the Nth byte (a number between 0 and 255).
3204 : *-------------------------------------------------------------
9770 scrappy 3205 : */
3206 : Datum
8335 tgl 3207 CBC 30 : byteaGetByte(PG_FUNCTION_ARGS)
9770 scrappy 3208 EUB : {
5847 tgl 3209 GIC 30 : bytea *v = PG_GETARG_BYTEA_PP(0);
8335 tgl 3210 CBC 30 : int32 n = PG_GETARG_INT32(1);
9344 bruce 3211 ECB : int len;
3212 : int byte;
3213 :
5847 tgl 3214 GIC 30 : len = VARSIZE_ANY_EXHDR(v);
3215 :
8416 3216 30 : if (n < 0 || n >= len)
7196 3217 3 : ereport(ERROR,
3218 : (errcode(ERRCODE_ARRAY_SUBSCRIPT_ERROR),
7196 tgl 3219 ECB : errmsg("index %d out of valid range, 0..%d",
3220 : n, len - 1)));
3221 :
5847 tgl 3222 CBC 27 : byte = ((unsigned char *) VARDATA_ANY(v))[n];
3223 :
8335 tgl 3224 GIC 27 : PG_RETURN_INT32(byte);
9770 scrappy 3225 ECB : }
3226 :
9770 scrappy 3227 EUB : /*-------------------------------------------------------------
9770 scrappy 3228 ECB : * byteaGetBit
3229 : *
3230 : * This routine treats a "bytea" type like an array of bits.
3231 : * It returns the value of the Nth bit (0 or 1).
3232 : *
3233 : *-------------------------------------------------------------
3234 : */
8335 tgl 3235 : Datum
8335 tgl 3236 GIC 6 : byteaGetBit(PG_FUNCTION_ARGS)
3237 : {
5847 3238 6 : bytea *v = PG_GETARG_BYTEA_PP(0);
1097 3239 6 : int64 n = PG_GETARG_INT64(1);
3240 : int byteNo,
3241 : bitNo;
3242 : int len;
4382 bruce 3243 ECB : int byte;
9345 3244 :
5847 tgl 3245 CBC 6 : len = VARSIZE_ANY_EXHDR(v);
8416 tgl 3246 ECB :
1097 tgl 3247 CBC 6 : if (n < 0 || n >= (int64) len * 8)
7196 3248 3 : ereport(ERROR,
7196 tgl 3249 EUB : (errcode(ERRCODE_ARRAY_SUBSCRIPT_ERROR),
3250 : errmsg("index %lld out of valid range, 0..%lld",
3251 : (long long) n, (long long) len * 8 - 1)));
3252 :
3253 : /* n/8 is now known < len, so safe to cast to int */
1097 tgl 3254 GIC 3 : byteNo = (int) (n / 8);
3255 3 : bitNo = (int) (n % 8);
3256 :
5847 3257 3 : byte = ((unsigned char *) VARDATA_ANY(v))[byteNo];
3258 :
4382 bruce 3259 3 : if (byte & (1 << bitNo))
8335 tgl 3260 CBC 3 : PG_RETURN_INT32(1);
9345 bruce 3261 ECB : else
8335 tgl 3262 LBC 0 : PG_RETURN_INT32(0);
9770 scrappy 3263 ECB : }
9345 bruce 3264 :
3265 : /*-------------------------------------------------------------
3266 : * byteaSetByte
9770 scrappy 3267 : *
3268 : * Given an instance of type 'bytea' creates a new one with
3269 : * the Nth byte set to the given value.
3270 : *
3271 : *-------------------------------------------------------------
3272 : */
8335 tgl 3273 : Datum
8335 tgl 3274 CBC 6 : byteaSetByte(PG_FUNCTION_ARGS)
3275 : {
2219 noah 3276 GIC 6 : bytea *res = PG_GETARG_BYTEA_P_COPY(0);
8335 tgl 3277 CBC 6 : int32 n = PG_GETARG_INT32(1);
3278 6 : int32 newByte = PG_GETARG_INT32(2);
3279 : int len;
9345 bruce 3280 ECB :
2219 noah 3281 GIC 6 : len = VARSIZE(res) - VARHDRSZ;
3282 :
8416 tgl 3283 CBC 6 : if (n < 0 || n >= len)
7196 tgl 3284 GIC 3 : ereport(ERROR,
3285 : (errcode(ERRCODE_ARRAY_SUBSCRIPT_ERROR),
7196 tgl 3286 ECB : errmsg("index %d out of valid range, 0..%d",
3287 : n, len - 1)));
3288 :
3289 : /*
3290 : * Now set the byte.
9345 bruce 3291 : */
8416 tgl 3292 GIC 3 : ((unsigned char *) VARDATA(res))[n] = newByte;
3293 :
8335 tgl 3294 CBC 3 : PG_RETURN_BYTEA_P(res);
3295 : }
9770 scrappy 3296 ECB :
3297 : /*-------------------------------------------------------------
3298 : * byteaSetBit
3299 : *
3300 : * Given an instance of type 'bytea' creates a new one with
3301 : * the Nth bit set to the given value.
3302 : *
3303 : *-------------------------------------------------------------
3304 : */
3305 : Datum
8335 tgl 3306 GIC 6 : byteaSetBit(PG_FUNCTION_ARGS)
3307 : {
2219 noah 3308 6 : bytea *res = PG_GETARG_BYTEA_P_COPY(0);
1097 tgl 3309 6 : int64 n = PG_GETARG_INT64(1);
8335 3310 6 : int32 newBit = PG_GETARG_INT32(2);
3311 : int len;
3312 : int oldByte,
3313 : newByte;
3314 : int byteNo,
3315 : bitNo;
3316 :
2219 noah 3317 6 : len = VARSIZE(res) - VARHDRSZ;
3318 :
1097 tgl 3319 6 : if (n < 0 || n >= (int64) len * 8)
7196 3320 3 : ereport(ERROR,
3321 : (errcode(ERRCODE_ARRAY_SUBSCRIPT_ERROR),
3322 : errmsg("index %lld out of valid range, 0..%lld",
3323 : (long long) n, (long long) len * 8 - 1)));
3324 :
3325 : /* n/8 is now known < len, so safe to cast to int */
1097 tgl 3326 CBC 3 : byteNo = (int) (n / 8);
1097 tgl 3327 GIC 3 : bitNo = (int) (n % 8);
3328 :
9345 bruce 3329 ECB : /*
3330 : * sanity check!
3331 : */
9345 bruce 3332 CBC 3 : if (newBit != 0 && newBit != 1)
7196 tgl 3333 UIC 0 : ereport(ERROR,
7196 tgl 3334 ECB : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
7196 tgl 3335 EUB : errmsg("new bit must be 0 or 1")));
3336 :
9345 bruce 3337 ECB : /*
8335 tgl 3338 : * Update the byte.
3339 : */
8335 tgl 3340 GIC 3 : oldByte = ((unsigned char *) VARDATA(res))[byteNo];
3341 :
9345 bruce 3342 3 : if (newBit == 0)
3343 3 : newByte = oldByte & (~(1 << bitNo));
3344 : else
9345 bruce 3345 UIC 0 : newByte = oldByte | (1 << bitNo);
9345 bruce 3346 ECB :
8335 tgl 3347 GIC 3 : ((unsigned char *) VARDATA(res))[byteNo] = newByte;
3348 :
8335 tgl 3349 GBC 3 : PG_RETURN_BYTEA_P(res);
3350 : }
3351 :
9081 lockhart 3352 EUB :
3353 : /* text_name()
8312 tgl 3354 : * Converts a text type to a Name type.
9081 lockhart 3355 : */
8312 tgl 3356 : Datum
8312 tgl 3357 GIC 15263 : text_name(PG_FUNCTION_ARGS)
9081 lockhart 3358 EUB : {
5847 tgl 3359 GBC 15263 : text *s = PG_GETARG_TEXT_PP(0);
3360 : Name result;
3361 : int len;
9081 lockhart 3362 EUB :
5847 tgl 3363 GIC 15263 : len = VARSIZE_ANY_EXHDR(s);
3364 :
3365 : /* Truncate oversize input */
8312 3366 15263 : if (len >= NAMEDATALEN)
3971 tgl 3367 CBC 3 : len = pg_mbcliplen(VARDATA_ANY(s), len, NAMEDATALEN - 1);
9081 lockhart 3368 ECB :
3369 : /* We use palloc0 here to ensure result is zero-padded */
3971 tgl 3370 GIC 15263 : result = (Name) palloc0(NAMEDATALEN);
5847 tgl 3371 CBC 15263 : memcpy(NameStr(*result), VARDATA_ANY(s), len);
9081 lockhart 3372 ECB :
8312 tgl 3373 CBC 15263 : PG_RETURN_NAME(result);
3374 : }
9081 lockhart 3375 ECB :
9081 lockhart 3376 EUB : /* name_text()
3377 : * Converts a Name type to a text type.
3378 : */
8312 tgl 3379 ECB : Datum
8312 tgl 3380 GBC 525697 : name_text(PG_FUNCTION_ARGS)
3381 : {
8312 tgl 3382 CBC 525697 : Name s = PG_GETARG_NAME(0);
3383 :
5493 tgl 3384 GBC 525697 : PG_RETURN_TEXT_P(cstring_to_text(NameStr(*s)));
8312 tgl 3385 EUB : }
7909 3386 :
3387 :
3388 : /*
7680 tgl 3389 ECB : * textToQualifiedNameList - convert a text object to list of names
3390 : *
3391 : * This implements the input parsing needed by nextval() and other
7680 tgl 3392 EUB : * functions that take a text parameter representing a qualified name.
3393 : * We split the name at dots, downcase if not double-quoted, and
3394 : * truncate names if they're too long.
7680 tgl 3395 ECB : */
3396 : List *
6526 neilc 3397 GIC 685 : textToQualifiedNameList(text *textval)
7680 tgl 3398 ECB : {
7680 tgl 3399 EUB : char *rawname;
7680 tgl 3400 GIC 685 : List *result = NIL;
3401 : List *namelist;
3402 : ListCell *l;
3403 :
7680 tgl 3404 ECB : /* Convert to C string (handles possible detoasting). */
3405 : /* Note we rely on being able to modify rawname below. */
5493 tgl 3406 CBC 685 : rawname = text_to_cstring(textval);
3407 :
7678 tgl 3408 GIC 685 : if (!SplitIdentifierString(rawname, '.', &namelist))
7196 tgl 3409 LBC 0 : ereport(ERROR,
3410 : (errcode(ERRCODE_INVALID_NAME),
7196 tgl 3411 ECB : errmsg("invalid name syntax")));
3412 :
7678 tgl 3413 GIC 685 : if (namelist == NIL)
7196 tgl 3414 UIC 0 : ereport(ERROR,
3415 : (errcode(ERRCODE_INVALID_NAME),
3416 : errmsg("invalid name syntax")));
3417 :
7678 tgl 3418 GIC 1425 : foreach(l, namelist)
3419 : {
7522 bruce 3420 740 : char *curname = (char *) lfirst(l);
3421 :
7678 tgl 3422 740 : result = lappend(result, makeString(pstrdup(curname)));
3423 : }
3424 :
3425 685 : pfree(rawname);
6888 neilc 3426 685 : list_free(namelist);
3427 :
7678 tgl 3428 685 : return result;
3429 : }
3430 :
3431 : /*
3432 : * SplitIdentifierString --- parse a string containing identifiers
3433 : *
3434 : * This is the guts of textToQualifiedNameList, and is exported for use in
3435 : * other situations such as parsing GUC variables. In the GUC case, it's
3436 : * important to avoid memory leaks, so the API is designed to minimize the
3437 : * amount of stuff that needs to be allocated and freed.
3438 : *
3439 : * Inputs:
3440 : * rawstring: the input string; must be overwritable! On return, it's
3441 : * been modified to contain the separated identifiers.
3442 : * separator: the separator punctuation expected between identifiers
3443 : * (typically '.' or ','). Whitespace may also appear around
3444 : * identifiers.
3445 : * Outputs:
3446 : * namelist: filled with a palloc'd list of pointers to identifiers within
6414 neilc 3447 ECB : * rawstring. Caller should list_free() this even on error return.
3448 : *
3449 : * Returns true if okay, false if there is a syntax error in the string.
7678 tgl 3450 : *
3451 : * Note that an empty string is considered okay here, though not in
3452 : * textToQualifiedNameList.
3453 : */
3454 : bool
7678 tgl 3455 CBC 72651 : SplitIdentifierString(char *rawstring, char separator,
7678 tgl 3456 EUB : List **namelist)
3457 : {
7678 tgl 3458 CBC 72651 : char *nextp = rawstring;
3459 72651 : bool done = false;
3460 :
7678 tgl 3461 GIC 72651 : *namelist = NIL;
3462 :
2146 3463 72651 : while (scanner_isspace(*nextp))
7678 tgl 3464 UIC 0 : nextp++; /* skip leading whitespace */
3465 :
7678 tgl 3466 GIC 72651 : if (*nextp == '\0')
7678 tgl 3467 CBC 9918 : return true; /* allow empty string */
3468 :
3469 : /* At the top of the loop, we are at start of a new identifier. */
7680 tgl 3470 ECB : do
3471 : {
3472 : char *curname;
3473 : char *endp;
3474 :
2665 peter_e 3475 GBC 99650 : if (*nextp == '"')
7680 tgl 3476 ECB : {
3477 : /* Quoted name --- collapse quote-quote pairs, no downcasing */
7680 tgl 3478 GIC 15347 : curname = nextp + 1;
7680 tgl 3479 ECB : for (;;)
3480 : {
2665 peter_e 3481 GIC 15349 : endp = strchr(nextp + 1, '"');
7680 tgl 3482 15348 : if (endp == NULL)
2118 tgl 3483 LBC 0 : return false; /* mismatched quotes */
2665 peter_e 3484 GIC 15348 : if (endp[1] != '"')
7680 tgl 3485 15347 : break; /* found end of quoted name */
3486 : /* Collapse adjacent quotes into one quote, and look again */
7522 bruce 3487 1 : memmove(endp, endp + 1, strlen(endp));
7680 tgl 3488 CBC 1 : nextp = endp;
7680 tgl 3489 ECB : }
7678 3490 : /* endp now points at the terminating quote */
7680 tgl 3491 CBC 15347 : nextp = endp + 1;
7680 tgl 3492 ECB : }
3493 : else
7680 tgl 3494 EUB : {
3495 : /* Unquoted name --- extends to separator or whitespace */
3496 : char *downname;
6987 tgl 3497 ECB : int len;
6987 tgl 3498 EUB :
7680 tgl 3499 GIC 84303 : curname = nextp;
7678 tgl 3500 CBC 727631 : while (*nextp && *nextp != separator &&
2146 tgl 3501 GIC 643329 : !scanner_isspace(*nextp))
7678 tgl 3502 CBC 643328 : nextp++;
3503 84303 : endp = nextp;
3504 84303 : if (curname == nextp)
7678 tgl 3505 UIC 0 : return false; /* empty unquoted name not allowed */
3506 :
6987 tgl 3507 ECB : /*
6385 bruce 3508 : * Downcase the identifier, using same code as main lexer does.
3509 : *
6987 tgl 3510 EUB : * XXX because we want to overwrite the input in-place, we cannot
3511 : * support a downcasing transformation that increases the string
3512 : * length. This is not a problem given the current implementation
6385 bruce 3513 ECB : * of downcase_truncate_identifier, but we'll probably have to do
3514 : * something about this someday.
3515 : */
6987 tgl 3516 GIC 84303 : len = endp - curname;
3517 84303 : downname = downcase_truncate_identifier(curname, len, false);
6987 tgl 3518 CBC 84303 : Assert(strlen(downname) <= len);
2997 tgl 3519 GIC 84303 : strncpy(curname, downname, len); /* strncpy is required here */
6987 3520 84303 : pfree(downname);
7680 tgl 3521 ECB : }
3522 :
2146 tgl 3523 CBC 99651 : while (scanner_isspace(*nextp))
7678 tgl 3524 GIC 1 : nextp++; /* skip trailing whitespace */
3525 :
3526 99650 : if (*nextp == separator)
3527 : {
3528 36917 : nextp++;
2146 3529 59216 : while (scanner_isspace(*nextp))
7678 3530 22299 : nextp++; /* skip leading whitespace for next */
3531 : /* we expect another name, so done remains false */
3532 : }
3533 62733 : else if (*nextp == '\0')
3534 62732 : done = true;
3535 : else
7678 tgl 3536 CBC 1 : return false; /* invalid syntax */
3537 :
7678 tgl 3538 ECB : /* Now safe to overwrite separator with a null */
7678 tgl 3539 CBC 99649 : *endp = '\0';
3540 :
3541 : /* Truncate name if it's overlength */
6987 tgl 3542 GIC 99649 : truncate_identifier(curname, strlen(curname), false);
3543 :
3544 : /*
3545 : * Finished isolating current name --- add it to list
3546 : */
7678 3547 99649 : *namelist = lappend(*namelist, curname);
7680 tgl 3548 ECB :
7678 3549 : /* Loop back if we didn't reach end of string */
7678 tgl 3550 CBC 99649 : } while (!done);
7680 tgl 3551 ECB :
7678 tgl 3552 GIC 62732 : return true;
3553 : }
7680 tgl 3554 ECB :
3555 :
3556 : /*
2119 3557 : * SplitDirectoriesString --- parse a string containing file/directory names
3558 : *
3559 : * This works fine on file names too; the function name is historical.
3894 3560 : *
3561 : * This is similar to SplitIdentifierString, except that the parsing
3562 : * rules are meant to handle pathnames instead of identifiers: there is
3563 : * no downcasing, embedded spaces are allowed, the max length is MAXPGPATH-1,
3867 3564 : * and we apply canonicalize_path() to each extracted string. Because of the
3565 : * last, the returned strings are separately palloc'd rather than being
3566 : * pointers into rawstring --- but we still scribble on rawstring.
3567 : *
3894 3568 : * Inputs:
3569 : * rawstring: the input string; must be modifiable!
3570 : * separator: the separator punctuation expected between directories
3260 bruce 3571 : * (typically ',' or ';'). Whitespace may also appear around
3572 : * directories.
3573 : * Outputs:
3574 : * namelist: filled with a palloc'd list of directory names.
3575 : * Caller should list_free_deep() this even on error return.
3576 : *
3577 : * Returns true if okay, false if there is a syntax error in the string.
3578 : *
3579 : * Note that an empty string is considered okay here.
3894 tgl 3580 : */
3581 : bool
3894 tgl 3582 CBC 621 : SplitDirectoriesString(char *rawstring, char separator,
3894 tgl 3583 EUB : List **namelist)
3584 : {
3894 tgl 3585 GIC 621 : char *nextp = rawstring;
3894 tgl 3586 CBC 621 : bool done = false;
3894 tgl 3587 ECB :
3894 tgl 3588 GIC 621 : *namelist = NIL;
3894 tgl 3589 ECB :
2146 tgl 3590 GIC 621 : while (scanner_isspace(*nextp))
3894 tgl 3591 UIC 0 : nextp++; /* skip leading whitespace */
3894 tgl 3592 ECB :
3894 tgl 3593 CBC 621 : if (*nextp == '\0')
3894 tgl 3594 GIC 3 : return true; /* allow empty string */
3595 :
3894 tgl 3596 ECB : /* At the top of the loop, we are at start of a new directory. */
3597 : do
3598 : {
3599 : char *curname;
3600 : char *endp;
3601 :
2665 peter_e 3602 CBC 618 : if (*nextp == '"')
3894 tgl 3603 ECB : {
3604 : /* Quoted name --- collapse quote-quote pairs */
3894 tgl 3605 UIC 0 : curname = nextp + 1;
3606 : for (;;)
3607 : {
2665 peter_e 3608 LBC 0 : endp = strchr(nextp + 1, '"');
3894 tgl 3609 0 : if (endp == NULL)
2118 tgl 3610 UIC 0 : return false; /* mismatched quotes */
2665 peter_e 3611 LBC 0 : if (endp[1] != '"')
3894 tgl 3612 UIC 0 : break; /* found end of quoted name */
3894 tgl 3613 ECB : /* Collapse adjacent quotes into one quote, and look again */
3894 tgl 3614 LBC 0 : memmove(endp, endp + 1, strlen(endp));
3894 tgl 3615 UIC 0 : nextp = endp;
3894 tgl 3616 ECB : }
3617 : /* endp now points at the terminating quote */
3894 tgl 3618 UIC 0 : nextp = endp + 1;
3619 : }
3894 tgl 3620 ECB : else
3621 : {
3867 3622 : /* Unquoted name --- extends to separator or end of string */
3867 tgl 3623 CBC 618 : curname = endp = nextp;
3867 tgl 3624 GIC 10456 : while (*nextp && *nextp != separator)
3625 : {
3626 : /* trailing whitespace should not be included in name */
2146 3627 9838 : if (!scanner_isspace(*nextp))
3867 tgl 3628 CBC 9838 : endp = nextp + 1;
3894 3629 9838 : nextp++;
3630 : }
3867 3631 618 : if (curname == endp)
3894 tgl 3632 UIC 0 : return false; /* empty unquoted name not allowed */
3894 tgl 3633 ECB : }
3634 :
2146 tgl 3635 GIC 618 : while (scanner_isspace(*nextp))
3894 tgl 3636 LBC 0 : nextp++; /* skip trailing whitespace */
3637 :
3894 tgl 3638 GIC 618 : if (*nextp == separator)
3639 : {
3894 tgl 3640 LBC 0 : nextp++;
2146 tgl 3641 UIC 0 : while (scanner_isspace(*nextp))
3894 tgl 3642 LBC 0 : nextp++; /* skip leading whitespace for next */
3894 tgl 3643 ECB : /* we expect another name, so done remains false */
3644 : }
3894 tgl 3645 GIC 618 : else if (*nextp == '\0')
3646 618 : done = true;
3647 : else
3894 tgl 3648 LBC 0 : return false; /* invalid syntax */
3894 tgl 3649 ECB :
3650 : /* Now safe to overwrite separator with a null */
3894 tgl 3651 CBC 618 : *endp = '\0';
3652 :
3894 tgl 3653 ECB : /* Truncate path if it's overlength */
3894 tgl 3654 CBC 618 : if (strlen(curname) >= MAXPGPATH)
3894 tgl 3655 UIC 0 : curname[MAXPGPATH - 1] = '\0';
3894 tgl 3656 ECB :
3657 : /*
3658 : * Finished isolating current name --- add it to list
3659 : */
3894 tgl 3660 CBC 618 : curname = pstrdup(curname);
3894 tgl 3661 GIC 618 : canonicalize_path(curname);
3894 tgl 3662 CBC 618 : *namelist = lappend(*namelist, curname);
3894 tgl 3663 ECB :
3664 : /* Loop back if we didn't reach end of string */
3894 tgl 3665 GIC 618 : } while (!done);
3666 :
3667 618 : return true;
3894 tgl 3668 ECB : }
3669 :
3670 :
1713 3671 : /*
3672 : * SplitGUCList --- parse a string containing identifiers or file names
3673 : *
3674 : * This is used to split the value of a GUC_LIST_QUOTE GUC variable, without
3675 : * presuming whether the elements will be taken as identifiers or file names.
3676 : * We assume the input has already been through flatten_set_variable_args(),
3677 : * so that we need never downcase (if appropriate, that was done already).
3678 : * Nor do we ever truncate, since we don't know the correct max length.
3679 : * We disallow embedded whitespace for simplicity (it shouldn't matter,
3680 : * because any embedded whitespace should have led to double-quoting).
3681 : * Otherwise the API is identical to SplitIdentifierString.
3682 : *
3683 : * XXX it's annoying to have so many copies of this string-splitting logic.
3684 : * However, it's not clear that having one function with a bunch of option
3685 : * flags would be much better.
3686 : *
3687 : * XXX there is a version of this function in src/bin/pg_dump/dumputils.c.
3688 : * Be sure to update that if you have to change this.
3689 : *
3690 : * Inputs:
3691 : * rawstring: the input string; must be overwritable! On return, it's
3692 : * been modified to contain the separated identifiers.
3693 : * separator: the separator punctuation expected between identifiers
3694 : * (typically '.' or ','). Whitespace may also appear around
3695 : * identifiers.
3696 : * Outputs:
3697 : * namelist: filled with a palloc'd list of pointers to identifiers within
3698 : * rawstring. Caller should list_free() this even on error return.
3699 : *
3700 : * Returns true if okay, false if there is a syntax error in the string.
3701 : */
3702 : bool
1713 tgl 3703 GIC 2458 : SplitGUCList(char *rawstring, char separator,
1713 tgl 3704 ECB : List **namelist)
3705 : {
1713 tgl 3706 GIC 2458 : char *nextp = rawstring;
1713 tgl 3707 CBC 2458 : bool done = false;
3708 :
1713 tgl 3709 GIC 2458 : *namelist = NIL;
1713 tgl 3710 ECB :
1713 tgl 3711 GIC 2458 : while (scanner_isspace(*nextp))
1713 tgl 3712 LBC 0 : nextp++; /* skip leading whitespace */
3713 :
1713 tgl 3714 CBC 2458 : if (*nextp == '\0')
1713 tgl 3715 GIC 2416 : return true; /* allow empty string */
3716 :
3717 : /* At the top of the loop, we are at start of a new identifier. */
3718 : do
3719 : {
3720 : char *curname;
3721 : char *endp;
3722 :
3723 55 : if (*nextp == '"')
1713 tgl 3724 ECB : {
3725 : /* Quoted name --- collapse quote-quote pairs */
1713 tgl 3726 CBC 12 : curname = nextp + 1;
1713 tgl 3727 ECB : for (;;)
3728 : {
1713 tgl 3729 GIC 18 : endp = strchr(nextp + 1, '"');
3730 15 : if (endp == NULL)
1713 tgl 3731 UIC 0 : return false; /* mismatched quotes */
1713 tgl 3732 GIC 15 : if (endp[1] != '"')
3733 12 : break; /* found end of quoted name */
3734 : /* Collapse adjacent quotes into one quote, and look again */
3735 3 : memmove(endp, endp + 1, strlen(endp));
3736 3 : nextp = endp;
3737 : }
1713 tgl 3738 ECB : /* endp now points at the terminating quote */
1713 tgl 3739 GIC 12 : nextp = endp + 1;
1713 tgl 3740 ECB : }
3741 : else
3742 : {
3743 : /* Unquoted name --- extends to separator or whitespace */
1713 tgl 3744 GIC 43 : curname = nextp;
3745 409 : while (*nextp && *nextp != separator &&
3746 366 : !scanner_isspace(*nextp))
3747 366 : nextp++;
3748 43 : endp = nextp;
3749 43 : if (curname == nextp)
1713 tgl 3750 UIC 0 : return false; /* empty unquoted name not allowed */
3751 : }
3752 :
1713 tgl 3753 CBC 55 : while (scanner_isspace(*nextp))
1713 tgl 3754 LBC 0 : nextp++; /* skip trailing whitespace */
3755 :
1713 tgl 3756 GIC 55 : if (*nextp == separator)
1713 tgl 3757 ECB : {
1713 tgl 3758 GIC 13 : nextp++;
1713 tgl 3759 GBC 22 : while (scanner_isspace(*nextp))
1713 tgl 3760 GIC 9 : nextp++; /* skip leading whitespace for next */
3761 : /* we expect another name, so done remains false */
1713 tgl 3762 ECB : }
1713 tgl 3763 GIC 42 : else if (*nextp == '\0')
1713 tgl 3764 CBC 42 : done = true;
3765 : else
1713 tgl 3766 UIC 0 : return false; /* invalid syntax */
1713 tgl 3767 ECB :
3768 : /* Now safe to overwrite separator with a null */
1713 tgl 3769 CBC 55 : *endp = '\0';
1713 tgl 3770 ECB :
3771 : /*
3772 : * Finished isolating current name --- add it to list
3773 : */
1713 tgl 3774 GIC 55 : *namelist = lappend(*namelist, curname);
1713 tgl 3775 ECB :
3776 : /* Loop back if we didn't reach end of string */
1713 tgl 3777 GIC 55 : } while (!done);
3778 :
1713 tgl 3779 CBC 42 : return true;
3780 : }
3781 :
1713 tgl 3782 ECB :
7909 3783 : /*****************************************************************************
3784 : * Comparison Functions used for bytea
3785 : *
3786 : * Note: btree indexes need these routines not to leak memory; therefore,
3787 : * be careful to free working copies of toasted datums. Most places don't
3788 : * need to be so careful.
3789 : *****************************************************************************/
3790 :
3791 : Datum
7909 tgl 3792 GIC 5188 : byteaeq(PG_FUNCTION_ARGS)
7909 tgl 3793 ECB : {
4464 tgl 3794 GIC 5188 : Datum arg1 = PG_GETARG_DATUM(0);
3795 5188 : Datum arg2 = PG_GETARG_DATUM(1);
7909 tgl 3796 ECB : bool result;
4464 3797 : Size len1,
3798 : len2;
7909 3799 :
3800 : /*
4464 3801 : * We can use a fast path for unequal lengths, which might save us from
3802 : * having to detoast one or both values.
3803 : */
4464 tgl 3804 CBC 5188 : len1 = toast_raw_datum_size(arg1);
4464 tgl 3805 GIC 5188 : len2 = toast_raw_datum_size(arg2);
7909 3806 5188 : if (len1 != len2)
3807 2154 : result = false;
3808 : else
3809 : {
4464 3810 3034 : bytea *barg1 = DatumGetByteaPP(arg1);
3811 3034 : bytea *barg2 = DatumGetByteaPP(arg2);
3812 :
3813 3034 : result = (memcmp(VARDATA_ANY(barg1), VARDATA_ANY(barg2),
3814 : len1 - VARHDRSZ) == 0);
4464 tgl 3815 ECB :
4464 tgl 3816 GIC 3034 : PG_FREE_IF_COPY(barg1, 0);
4464 tgl 3817 CBC 3034 : PG_FREE_IF_COPY(barg2, 1);
4464 tgl 3818 ECB : }
7909 3819 :
7909 tgl 3820 GIC 5188 : PG_RETURN_BOOL(result);
7909 tgl 3821 ECB : }
3822 :
3823 : Datum
7909 tgl 3824 CBC 384 : byteane(PG_FUNCTION_ARGS)
7909 tgl 3825 ECB : {
4464 tgl 3826 CBC 384 : Datum arg1 = PG_GETARG_DATUM(0);
3827 384 : Datum arg2 = PG_GETARG_DATUM(1);
3828 : bool result;
4464 tgl 3829 ECB : Size len1,
3830 : len2;
7909 3831 :
4464 3832 : /*
3833 : * We can use a fast path for unequal lengths, which might save us from
3834 : * having to detoast one or both values.
3835 : */
4464 tgl 3836 GIC 384 : len1 = toast_raw_datum_size(arg1);
4464 tgl 3837 CBC 384 : len2 = toast_raw_datum_size(arg2);
7909 tgl 3838 GIC 384 : if (len1 != len2)
7909 tgl 3839 UIC 0 : result = true;
3840 : else
3841 : {
4464 tgl 3842 GIC 384 : bytea *barg1 = DatumGetByteaPP(arg1);
3843 384 : bytea *barg2 = DatumGetByteaPP(arg2);
3844 :
3845 384 : result = (memcmp(VARDATA_ANY(barg1), VARDATA_ANY(barg2),
3846 : len1 - VARHDRSZ) != 0);
3847 :
4464 tgl 3848 CBC 384 : PG_FREE_IF_COPY(barg1, 0);
4464 tgl 3849 GIC 384 : PG_FREE_IF_COPY(barg2, 1);
3850 : }
3851 :
7909 tgl 3852 CBC 384 : PG_RETURN_BOOL(result);
7909 tgl 3853 ECB : }
3854 :
3855 : Datum
7909 tgl 3856 GIC 4158 : bytealt(PG_FUNCTION_ARGS)
7909 tgl 3857 ECB : {
5847 tgl 3858 GIC 4158 : bytea *arg1 = PG_GETARG_BYTEA_PP(0);
3859 4158 : bytea *arg2 = PG_GETARG_BYTEA_PP(1);
3860 : int len1,
3861 : len2;
7909 tgl 3862 ECB : int cmp;
3863 :
5847 tgl 3864 CBC 4158 : len1 = VARSIZE_ANY_EXHDR(arg1);
5847 tgl 3865 GIC 4158 : len2 = VARSIZE_ANY_EXHDR(arg2);
3866 :
5847 tgl 3867 CBC 4158 : cmp = memcmp(VARDATA_ANY(arg1), VARDATA_ANY(arg2), Min(len1, len2));
7909 tgl 3868 ECB :
7909 tgl 3869 GIC 4158 : PG_FREE_IF_COPY(arg1, 0);
3870 4158 : PG_FREE_IF_COPY(arg2, 1);
7909 tgl 3871 ECB :
7909 tgl 3872 CBC 4158 : PG_RETURN_BOOL((cmp < 0) || ((cmp == 0) && (len1 < len2)));
7909 tgl 3873 ECB : }
3874 :
3875 : Datum
7909 tgl 3876 GIC 3178 : byteale(PG_FUNCTION_ARGS)
3877 : {
5847 tgl 3878 CBC 3178 : bytea *arg1 = PG_GETARG_BYTEA_PP(0);
3879 3178 : bytea *arg2 = PG_GETARG_BYTEA_PP(1);
3880 : int len1,
3881 : len2;
7909 tgl 3882 ECB : int cmp;
3883 :
5847 tgl 3884 GIC 3178 : len1 = VARSIZE_ANY_EXHDR(arg1);
5847 tgl 3885 CBC 3178 : len2 = VARSIZE_ANY_EXHDR(arg2);
3886 :
3887 3178 : cmp = memcmp(VARDATA_ANY(arg1), VARDATA_ANY(arg2), Min(len1, len2));
7909 tgl 3888 ECB :
7909 tgl 3889 CBC 3178 : PG_FREE_IF_COPY(arg1, 0);
7909 tgl 3890 GIC 3178 : PG_FREE_IF_COPY(arg2, 1);
7909 tgl 3891 ECB :
7909 tgl 3892 GIC 3178 : PG_RETURN_BOOL((cmp < 0) || ((cmp == 0) && (len1 <= len2)));
3893 : }
7909 tgl 3894 ECB :
3895 : Datum
7909 tgl 3896 CBC 3114 : byteagt(PG_FUNCTION_ARGS)
3897 : {
5847 3898 3114 : bytea *arg1 = PG_GETARG_BYTEA_PP(0);
5847 tgl 3899 GIC 3114 : bytea *arg2 = PG_GETARG_BYTEA_PP(1);
3900 : int len1,
7909 tgl 3901 ECB : len2;
3902 : int cmp;
3903 :
5847 tgl 3904 GIC 3114 : len1 = VARSIZE_ANY_EXHDR(arg1);
3905 3114 : len2 = VARSIZE_ANY_EXHDR(arg2);
3906 :
3907 3114 : cmp = memcmp(VARDATA_ANY(arg1), VARDATA_ANY(arg2), Min(len1, len2));
3908 :
7909 3909 3114 : PG_FREE_IF_COPY(arg1, 0);
3910 3114 : PG_FREE_IF_COPY(arg2, 1);
3911 :
7909 tgl 3912 CBC 3114 : PG_RETURN_BOOL((cmp > 0) || ((cmp == 0) && (len1 > len2)));
7909 tgl 3913 ECB : }
3914 :
3915 : Datum
7909 tgl 3916 CBC 2505 : byteage(PG_FUNCTION_ARGS)
3917 : {
5847 tgl 3918 GIC 2505 : bytea *arg1 = PG_GETARG_BYTEA_PP(0);
3919 2505 : bytea *arg2 = PG_GETARG_BYTEA_PP(1);
3920 : int len1,
3921 : len2;
3922 : int cmp;
3923 :
3924 2505 : len1 = VARSIZE_ANY_EXHDR(arg1);
5847 tgl 3925 CBC 2505 : len2 = VARSIZE_ANY_EXHDR(arg2);
7909 tgl 3926 ECB :
5847 tgl 3927 CBC 2505 : cmp = memcmp(VARDATA_ANY(arg1), VARDATA_ANY(arg2), Min(len1, len2));
7909 tgl 3928 ECB :
7909 tgl 3929 CBC 2505 : PG_FREE_IF_COPY(arg1, 0);
7909 tgl 3930 GIC 2505 : PG_FREE_IF_COPY(arg2, 1);
3931 :
7909 tgl 3932 CBC 2505 : PG_RETURN_BOOL((cmp > 0) || ((cmp == 0) && (len1 >= len2)));
3933 : }
3934 :
3935 : Datum
7909 tgl 3936 GIC 43878 : byteacmp(PG_FUNCTION_ARGS)
3937 : {
5847 3938 43878 : bytea *arg1 = PG_GETARG_BYTEA_PP(0);
3939 43878 : bytea *arg2 = PG_GETARG_BYTEA_PP(1);
3940 : int len1,
3941 : len2;
3942 : int cmp;
3943 :
3944 43878 : len1 = VARSIZE_ANY_EXHDR(arg1);
3945 43878 : len2 = VARSIZE_ANY_EXHDR(arg2);
3946 :
3947 43878 : cmp = memcmp(VARDATA_ANY(arg1), VARDATA_ANY(arg2), Min(len1, len2));
7909 tgl 3948 CBC 43878 : if ((cmp == 0) && (len1 != len2))
7909 tgl 3949 GIC 7326 : cmp = (len1 < len2) ? -1 : 1;
3950 :
3951 43878 : PG_FREE_IF_COPY(arg1, 0);
3952 43878 : PG_FREE_IF_COPY(arg2, 1);
3953 :
3954 43878 : PG_RETURN_INT32(cmp);
7909 tgl 3955 ECB : }
7535 bruce 3956 :
3957 : Datum
2622 rhaas 3958 GIC 13 : bytea_sortsupport(PG_FUNCTION_ARGS)
2622 rhaas 3959 ECB : {
2622 rhaas 3960 GIC 13 : SortSupport ssup = (SortSupport) PG_GETARG_POINTER(0);
3961 : MemoryContext oldcontext;
3962 :
3963 13 : oldcontext = MemoryContextSwitchTo(ssup->ssup_cxt);
3964 :
3965 : /* Use generic string SortSupport, forcing "C" collation */
1572 tgl 3966 CBC 13 : varstr_sortsupport(ssup, BYTEAOID, C_COLLATION_OID);
3967 :
2622 rhaas 3968 GIC 13 : MemoryContextSwitchTo(oldcontext);
2622 rhaas 3969 ECB :
2622 rhaas 3970 CBC 13 : PG_RETURN_VOID();
3971 : }
3972 :
6488 bruce 3973 ECB : /*
3974 : * appendStringInfoText
3975 : *
3976 : * Append a text to str.
3977 : * Like appendStringInfoString(str, text_to_cstring(t)) but faster.
3978 : */
3979 : static void
6488 bruce 3980 CBC 832552 : appendStringInfoText(StringInfo str, const text *t)
3981 : {
5678 tgl 3982 GIC 832552 : appendBinaryStringInfo(str, VARDATA_ANY(t), VARSIZE_ANY_EXHDR(t));
6488 bruce 3983 832552 : }
6488 bruce 3984 ECB :
3985 : /*
3986 : * replace_text
7335 tgl 3987 : * replace all occurrences of 'old_sub_str' in 'orig_str'
7535 bruce 3988 : * with 'new_sub_str' to form 'new_str'
3989 : *
3990 : * returns 'orig_str' if 'old_sub_str' == '' or 'orig_str' == ''
3991 : * otherwise returns 'new_str'
3992 : */
3993 : Datum
7535 bruce 3994 CBC 1275 : replace_text(PG_FUNCTION_ARGS)
3995 : {
5678 tgl 3996 1275 : text *src_text = PG_GETARG_TEXT_PP(0);
5678 tgl 3997 GIC 1275 : text *from_sub_text = PG_GETARG_TEXT_PP(1);
3998 1275 : text *to_sub_text = PG_GETARG_TEXT_PP(2);
3999 : int src_text_len;
4000 : int from_sub_text_len;
4001 : TextPositionState state;
4002 : text *ret_text;
4003 : int chunk_len;
4004 : char *curr_ptr;
5996 tgl 4005 ECB : char *start_ptr;
6248 neilc 4006 : StringInfoData str;
4007 : bool found;
7535 bruce 4008 :
1535 heikki.linnakangas 4009 GIC 1275 : src_text_len = VARSIZE_ANY_EXHDR(src_text);
4010 1275 : from_sub_text_len = VARSIZE_ANY_EXHDR(from_sub_text);
4011 :
5743 tgl 4012 EUB : /* Return unmodified source string if empty source or pattern */
5743 tgl 4013 GBC 1275 : if (src_text_len < 1 || from_sub_text_len < 1)
4014 : {
5743 tgl 4015 UIC 0 : PG_RETURN_TEXT_P(src_text);
4016 : }
4017 :
1479 peter 4018 GIC 1275 : text_position_setup(src_text, from_sub_text, PG_GET_COLLATION(), &state);
4019 :
1535 heikki.linnakangas 4020 1275 : found = text_position_next(&state);
6488 bruce 4021 ECB :
4022 : /* When the from_sub_text is not found, there is nothing to do. */
1535 heikki.linnakangas 4023 GIC 1275 : if (!found)
4024 : {
6028 tgl 4025 390 : text_position_cleanup(&state);
6488 bruce 4026 390 : PG_RETURN_TEXT_P(src_text);
4027 : }
1535 heikki.linnakangas 4028 885 : curr_ptr = text_position_get_match_ptr(&state);
5678 tgl 4029 CBC 885 : start_ptr = VARDATA_ANY(src_text);
5996 tgl 4030 ECB :
6248 neilc 4031 GBC 885 : initStringInfo(&str);
7535 bruce 4032 ECB :
4033 : do
4034 : {
5743 tgl 4035 GIC 3119 : CHECK_FOR_INTERRUPTS();
4036 :
4037 : /* copy the data skipped over by last text_position_next() */
1535 heikki.linnakangas 4038 3119 : chunk_len = curr_ptr - start_ptr;
5996 tgl 4039 CBC 3119 : appendBinaryStringInfo(&str, start_ptr, chunk_len);
4040 :
6248 neilc 4041 GIC 3119 : appendStringInfoText(&str, to_sub_text);
4042 :
1535 heikki.linnakangas 4043 CBC 3119 : start_ptr = curr_ptr + from_sub_text_len;
5996 tgl 4044 ECB :
1535 heikki.linnakangas 4045 CBC 3119 : found = text_position_next(&state);
1535 heikki.linnakangas 4046 GIC 3119 : if (found)
4047 2234 : curr_ptr = text_position_get_match_ptr(&state);
4048 : }
4049 3119 : while (found);
4050 :
5996 tgl 4051 ECB : /* copy trailing data */
5678 tgl 4052 CBC 885 : chunk_len = ((char *) src_text + VARSIZE_ANY(src_text)) - start_ptr;
5996 tgl 4053 GIC 885 : appendBinaryStringInfo(&str, start_ptr, chunk_len);
4054 :
6028 4055 885 : text_position_cleanup(&state);
4056 :
5493 4057 885 : ret_text = cstring_to_text_with_len(str.data, str.len);
6248 neilc 4058 CBC 885 : pfree(str.data);
7535 bruce 4059 ECB :
7535 bruce 4060 GIC 885 : PG_RETURN_TEXT_P(ret_text);
4061 : }
7535 bruce 4062 ECB :
4063 : /*
4064 : * check_replace_text_has_escape
6382 tgl 4065 : *
608 4066 : * Returns 0 if text contains no backslashes that need processing.
4067 : * Returns 1 if text contains backslashes, but not regexp submatch specifiers.
4068 : * Returns 2 if text contains regexp submatch specifiers (\1 .. \9).
4069 : */
4070 : static int
608 tgl 4071 GIC 5263 : check_replace_text_has_escape(const text *replace_text)
6482 bruce 4072 ECB : {
608 tgl 4073 CBC 5263 : int result = 0;
5678 tgl 4074 GIC 5263 : const char *p = VARDATA_ANY(replace_text);
4075 5263 : const char *p_end = p + VARSIZE_ANY_EXHDR(replace_text);
4076 :
608 4077 10548 : while (p < p_end)
4078 : {
4079 : /* Find next escape char, if any. */
4080 4635 : p = memchr(p, '\\', p_end - p);
608 tgl 4081 CBC 4635 : if (p == NULL)
4082 4296 : break;
4083 339 : p++;
4084 : /* Note: a backslash at the end doesn't require extra processing. */
608 tgl 4085 GIC 339 : if (p < p_end)
4086 : {
4087 339 : if (*p >= '1' && *p <= '9')
4088 317 : return 2; /* Found a submatch specifier, so done */
608 tgl 4089 CBC 22 : result = 1; /* Found some other sequence, keep looking */
608 tgl 4090 GIC 22 : p++;
4091 : }
4092 : }
608 tgl 4093 CBC 4946 : return result;
6482 bruce 4094 ECB : }
4095 :
4096 : /*
4097 : * appendStringInfoRegexpSubstr
6382 tgl 4098 : *
4099 : * Append replace_text to str, substituting regexp back references for
4100 : * \n escapes. start_ptr is the start of the match in the source string,
5996 4101 : * at logical character position data_pos.
4102 : */
4103 : static void
6482 bruce 4104 GIC 106 : appendStringInfoRegexpSubstr(StringInfo str, text *replace_text,
4105 : regmatch_t *pmatch,
4106 : char *start_ptr, int data_pos)
4107 : {
5678 tgl 4108 106 : const char *p = VARDATA_ANY(replace_text);
4109 106 : const char *p_end = p + VARSIZE_ANY_EXHDR(replace_text);
6482 bruce 4110 ECB :
608 tgl 4111 GIC 263 : while (p < p_end)
6482 bruce 4112 ECB : {
6382 tgl 4113 CBC 235 : const char *chunk_start = p;
6382 tgl 4114 ECB : int so;
4115 : int eo;
4116 :
4117 : /* Find next escape char, if any. */
608 tgl 4118 GIC 235 : p = memchr(p, '\\', p_end - p);
4119 235 : if (p == NULL)
4120 75 : p = p_end;
4121 :
4122 : /* Copy the text we just scanned over, if any. */
6382 4123 235 : if (p > chunk_start)
6382 tgl 4124 CBC 147 : appendBinaryStringInfo(str, chunk_start, p - chunk_start);
6482 bruce 4125 ECB :
4126 : /* Done if at end of string, else advance over escape char. */
6382 tgl 4127 GIC 235 : if (p >= p_end)
6482 bruce 4128 75 : break;
6482 bruce 4129 CBC 160 : p++;
6382 tgl 4130 ECB :
6382 tgl 4131 GIC 160 : if (p >= p_end)
4132 : {
6382 tgl 4133 ECB : /* Escape at very end of input. Treat same as unexpected char */
6382 tgl 4134 CBC 3 : appendStringInfoChar(str, '\\');
6382 tgl 4135 GIC 3 : break;
4136 : }
6482 bruce 4137 ECB :
6482 bruce 4138 GIC 157 : if (*p >= '1' && *p <= '9')
4139 127 : {
6482 bruce 4140 ECB : /* Use the back reference of regexp. */
6385 bruce 4141 CBC 127 : int idx = *p - '0';
4142 :
6482 4143 127 : so = pmatch[idx].rm_so;
6482 bruce 4144 GIC 127 : eo = pmatch[idx].rm_eo;
4145 127 : p++;
4146 : }
6482 bruce 4147 CBC 30 : else if (*p == '&')
4148 : {
6482 bruce 4149 ECB : /* Use the entire matched string. */
6482 bruce 4150 GIC 9 : so = pmatch[0].rm_so;
4151 9 : eo = pmatch[0].rm_eo;
6482 bruce 4152 CBC 9 : p++;
4153 : }
6382 tgl 4154 21 : else if (*p == '\\')
4155 : {
6382 tgl 4156 ECB : /* \\ means transfer one \ to output. */
6382 tgl 4157 CBC 18 : appendStringInfoChar(str, '\\');
6382 tgl 4158 GIC 18 : p++;
6382 tgl 4159 CBC 18 : continue;
4160 : }
4161 : else
4162 : {
4163 : /*
4164 : * If escape char is not followed by any expected char, just treat
4165 : * it as ordinary data to copy. (XXX would it be better to throw
6347 bruce 4166 ECB : * an error?)
4167 : */
6382 tgl 4168 GIC 3 : appendStringInfoChar(str, '\\');
6382 tgl 4169 CBC 3 : continue;
4170 : }
6482 bruce 4171 ECB :
608 tgl 4172 CBC 136 : if (so >= 0 && eo >= 0)
4173 : {
4174 : /*
3260 bruce 4175 ECB : * Copy the text that is back reference of regexp. Note so and eo
4176 : * are counted in characters not bytes.
6382 tgl 4177 : */
5996 4178 : char *chunk_start;
4179 : int chunk_len;
4180 :
5996 tgl 4181 GIC 136 : Assert(so >= data_pos);
4182 136 : chunk_start = start_ptr;
4183 136 : chunk_start += charlen_to_bytelen(chunk_start, so - data_pos);
4184 136 : chunk_len = charlen_to_bytelen(chunk_start, eo - so);
5996 tgl 4185 CBC 136 : appendBinaryStringInfo(str, chunk_start, chunk_len);
4186 : }
4187 : }
6482 bruce 4188 106 : }
4189 :
6482 bruce 4190 ECB : /*
4191 : * replace_text_regexp
4192 : *
4193 : * replace substring(s) in src_text that match pattern with replace_text.
4194 : * The replace_text can contain backslash markers to substitute
608 tgl 4195 : * (parts of) the matched text.
614 4196 : *
608 4197 : * cflags: regexp compile flags.
4198 : * collation: collation to use.
4199 : * search_start: the character (not byte) offset in src_text at which to
4200 : * begin searching.
614 4201 : * n: if 0, replace all matches; if > 0, replace only the N'th match.
6482 bruce 4202 : */
4203 : text *
608 tgl 4204 CBC 5263 : replace_text_regexp(text *src_text, text *pattern_text,
4205 : text *replace_text,
4206 : int cflags, Oid collation,
614 tgl 4207 ECB : int search_start, int n)
6482 bruce 4208 : {
4209 : text *ret_text;
608 tgl 4210 : regex_t *re;
5678 tgl 4211 GIC 5263 : int src_text_len = VARSIZE_ANY_EXHDR(src_text);
614 4212 5263 : int nmatches = 0;
6031 bruce 4213 ECB : StringInfoData buf;
4214 : regmatch_t pmatch[10]; /* main match, plus \1 to \9 */
608 tgl 4215 CBC 5263 : int nmatch = lengthof(pmatch);
4216 : pg_wchar *data;
4217 : size_t data_len;
4218 : int data_pos;
5996 tgl 4219 ECB : char *start_ptr;
4220 : int escape_status;
6482 bruce 4221 :
6248 neilc 4222 GIC 5263 : initStringInfo(&buf);
6248 neilc 4223 ECB :
4224 : /* Convert data string to wide characters. */
6482 bruce 4225 GIC 5263 : data = (pg_wchar *) palloc((src_text_len + 1) * sizeof(pg_wchar));
5678 tgl 4226 5263 : data_len = pg_mb2wchar_with_len(VARDATA_ANY(src_text), data, src_text_len);
6482 bruce 4227 ECB :
4228 : /* Check whether replace_text has escapes, especially regexp submatches. */
608 tgl 4229 GIC 5263 : escape_status = check_replace_text_has_escape(replace_text);
4230 :
4231 : /* If no regexp submatches, we can use REG_NOSUB. */
608 tgl 4232 CBC 5263 : if (escape_status < 2)
4233 : {
608 tgl 4234 GIC 4946 : cflags |= REG_NOSUB;
608 tgl 4235 ECB : /* Also tell pg_regexec we only want the whole-match location. */
608 tgl 4236 GIC 4946 : nmatch = 1;
4237 : }
4238 :
4239 : /* Prepare the regexp. */
4240 5263 : re = RE_compile_and_cache(pattern_text, cflags, collation);
4241 :
5996 tgl 4242 ECB : /* start_ptr points to the data_pos'th character of src_text */
5678 tgl 4243 GIC 5263 : start_ptr = (char *) VARDATA_ANY(src_text);
5996 tgl 4244 CBC 5263 : data_pos = 0;
4245 :
5996 tgl 4246 GIC 7251 : while (search_start <= data_len)
4247 : {
4248 : int regexec_result;
4249 :
4250 7248 : CHECK_FOR_INTERRUPTS();
4251 :
6482 bruce 4252 7248 : regexec_result = pg_regexec(re,
4253 : data,
4254 : data_len,
4255 : search_start,
2118 tgl 4256 ECB : NULL, /* no details */
4257 : nmatch,
4258 : pmatch,
4259 : 0);
4260 :
6248 neilc 4261 CBC 7248 : if (regexec_result == REG_NOMATCH)
6248 neilc 4262 GIC 4583 : break;
6248 neilc 4263 ECB :
6248 neilc 4264 CBC 2665 : if (regexec_result != REG_OKAY)
4265 : {
6385 bruce 4266 ECB : char errMsg[100];
6482 4267 :
6482 bruce 4268 LBC 0 : pg_regerror(regexec_result, re, errMsg, sizeof(errMsg));
6482 bruce 4269 UIC 0 : ereport(ERROR,
4270 : (errcode(ERRCODE_INVALID_REGULAR_EXPRESSION),
4271 : errmsg("regular expression failed: %s", errMsg)));
4272 : }
4273 :
4274 : /*
4275 : * Count matches, and decide whether to replace this match.
4276 : */
614 tgl 4277 GIC 2665 : nmatches++;
4278 2665 : if (n > 0 && nmatches != n)
4279 : {
4280 : /*
614 tgl 4281 ECB : * No, so advance search_start, but not start_ptr/data_pos. (Thus,
4282 : * we treat the matched text as if it weren't matched, and copy it
4283 : * to the output later.)
4284 : */
614 tgl 4285 GIC 30 : search_start = pmatch[0].rm_eo;
4286 30 : if (pmatch[0].rm_so == pmatch[0].rm_eo)
614 tgl 4287 UIC 0 : search_start++;
614 tgl 4288 GIC 30 : continue;
4289 : }
4290 :
4291 : /*
5624 bruce 4292 ECB : * Copy the text to the left of the match position. Note we are given
4293 : * character not byte indexes.
6385 4294 : */
6482 bruce 4295 GIC 2635 : if (pmatch[0].rm_so - data_pos > 0)
4296 : {
5996 tgl 4297 ECB : int chunk_len;
4298 :
5996 tgl 4299 CBC 2550 : chunk_len = charlen_to_bytelen(start_ptr,
4300 2550 : pmatch[0].rm_so - data_pos);
5996 tgl 4301 GIC 2550 : appendBinaryStringInfo(&buf, start_ptr, chunk_len);
5624 bruce 4302 ECB :
4303 : /*
4304 : * Advance start_ptr over that text, to avoid multiple rescans of
4305 : * it if the replace_text contains multiple back-references.
4306 : */
5996 tgl 4307 GIC 2550 : start_ptr += chunk_len;
4308 2550 : data_pos = pmatch[0].rm_so;
4309 : }
4310 :
4311 : /*
4312 : * Copy the replace_text, processing escapes if any are present.
4313 : */
608 4314 2635 : if (escape_status > 0)
5996 4315 106 : appendStringInfoRegexpSubstr(&buf, replace_text, pmatch,
5996 tgl 4316 ECB : start_ptr, data_pos);
4317 : else
6248 neilc 4318 CBC 2529 : appendStringInfoText(&buf, replace_text);
4319 :
4320 : /* Advance start_ptr and data_pos over the matched text. */
5996 tgl 4321 GIC 5270 : start_ptr += charlen_to_bytelen(start_ptr,
4322 2635 : pmatch[0].rm_eo - data_pos);
4323 2635 : data_pos = pmatch[0].rm_eo;
4324 :
4325 : /*
4326 : * If we only want to replace one occurrence, we're done.
4327 : */
614 4328 2635 : if (n > 0)
6482 bruce 4329 677 : break;
4330 :
4331 : /*
3539 tgl 4332 ECB : * Advance search position. Normally we start the next search at the
4333 : * end of the previous match; but if the match was of zero length, we
4334 : * have to advance by one character, or we'd just find the same match
4335 : * again.
4336 : */
5996 tgl 4337 CBC 1958 : search_start = data_pos;
6482 bruce 4338 GIC 1958 : if (pmatch[0].rm_so == pmatch[0].rm_eo)
4339 6 : search_start++;
4340 : }
4341 :
4342 : /*
4343 : * Copy the text to the right of the last match.
6482 bruce 4344 ECB : */
6482 bruce 4345 CBC 5263 : if (data_pos < data_len)
4346 : {
5996 tgl 4347 ECB : int chunk_len;
4348 :
5678 tgl 4349 GIC 5032 : chunk_len = ((char *) src_text + VARSIZE_ANY(src_text)) - start_ptr;
5996 tgl 4350 CBC 5032 : appendBinaryStringInfo(&buf, start_ptr, chunk_len);
6482 bruce 4351 ECB : }
4352 :
5493 tgl 4353 CBC 5263 : ret_text = cstring_to_text_with_len(buf.data, buf.len);
6248 neilc 4354 GIC 5263 : pfree(buf.data);
6482 bruce 4355 5263 : pfree(data);
6482 bruce 4356 ECB :
6382 tgl 4357 CBC 5263 : return ret_text;
4358 : }
6482 bruce 4359 ECB :
4360 : /*
949 tgl 4361 : * split_part
4362 : * parse input string based on provided field separator
4363 : * return N'th item (1 based, negative counts from end)
4364 : */
4365 : Datum
949 tgl 4366 GIC 51 : split_part(PG_FUNCTION_ARGS)
4367 : {
5678 4368 51 : text *inputstring = PG_GETARG_TEXT_PP(0);
5678 tgl 4369 CBC 51 : text *fldsep = PG_GETARG_TEXT_PP(1);
7535 bruce 4370 51 : int fldnum = PG_GETARG_INT32(2);
4371 : int inputstring_len;
4372 : int fldsep_len;
6028 tgl 4373 ECB : TextPositionState state;
1535 heikki.linnakangas 4374 : char *start_ptr;
4375 : char *end_ptr;
4376 : text *result_text;
4377 : bool found;
4378 :
7008 tgl 4379 : /* field number is 1 based */
877 tgl 4380 GIC 51 : if (fldnum == 0)
7008 tgl 4381 CBC 3 : ereport(ERROR,
4382 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
4383 : errmsg("field position must not be zero")));
7008 tgl 4384 ECB :
1535 heikki.linnakangas 4385 GIC 48 : inputstring_len = VARSIZE_ANY_EXHDR(inputstring);
1535 heikki.linnakangas 4386 CBC 48 : fldsep_len = VARSIZE_ANY_EXHDR(fldsep);
4387 :
4388 : /* return empty string for empty input string */
7535 bruce 4389 48 : if (inputstring_len < 1)
5493 tgl 4390 GIC 6 : PG_RETURN_TEXT_P(cstring_to_text(""));
4391 :
4392 : /* handle empty field separator */
7535 bruce 4393 42 : if (fldsep_len < 1)
7535 bruce 4394 ECB : {
4395 : /* if first or last field, return input string, else empty string */
877 tgl 4396 CBC 12 : if (fldnum == 1 || fldnum == -1)
7535 bruce 4397 6 : PG_RETURN_TEXT_P(inputstring);
4398 : else
5493 tgl 4399 GIC 6 : PG_RETURN_TEXT_P(cstring_to_text(""));
7535 bruce 4400 ECB : }
4401 :
4402 : /* find the first field separator */
1479 peter 4403 GIC 30 : text_position_setup(inputstring, fldsep, PG_GET_COLLATION(), &state);
4404 :
1535 heikki.linnakangas 4405 30 : found = text_position_next(&state);
7535 bruce 4406 ECB :
6028 tgl 4407 : /* special case if fldsep not found at all */
1535 heikki.linnakangas 4408 GIC 30 : if (!found)
4409 : {
6028 tgl 4410 6 : text_position_cleanup(&state);
877 tgl 4411 ECB : /* if first or last field, return input string, else empty string */
877 tgl 4412 GIC 6 : if (fldnum == 1 || fldnum == -1)
7535 bruce 4413 3 : PG_RETURN_TEXT_P(inputstring);
7188 bruce 4414 ECB : else
5493 tgl 4415 GIC 3 : PG_RETURN_TEXT_P(cstring_to_text(""));
4416 : }
877 tgl 4417 ECB :
4418 : /*
4419 : * take care of a negative field number (i.e. count from the right) by
4420 : * converting to a positive field number; we need total number of fields
4421 : */
877 tgl 4422 CBC 24 : if (fldnum < 0)
4423 : {
4424 : /* we found a fldsep, so there are at least two fields */
4425 12 : int numfields = 2;
4426 :
877 tgl 4427 GIC 18 : while (text_position_next(&state))
4428 6 : numfields++;
4429 :
4430 : /* special case of last field does not require an extra pass */
4431 12 : if (fldnum == -1)
4432 : {
4433 3 : start_ptr = text_position_get_match_ptr(&state) + fldsep_len;
877 tgl 4434 CBC 3 : end_ptr = VARDATA_ANY(inputstring) + inputstring_len;
877 tgl 4435 GIC 3 : text_position_cleanup(&state);
877 tgl 4436 CBC 3 : PG_RETURN_TEXT_P(cstring_to_text_with_len(start_ptr,
4437 : end_ptr - start_ptr));
877 tgl 4438 ECB : }
4439 :
4440 : /* else, convert fldnum to positive notation */
877 tgl 4441 GIC 9 : fldnum += numfields + 1;
877 tgl 4442 ECB :
4443 : /* if nonexistent field, return empty string */
877 tgl 4444 GIC 9 : if (fldnum <= 0)
877 tgl 4445 ECB : {
877 tgl 4446 GIC 3 : text_position_cleanup(&state);
4447 3 : PG_RETURN_TEXT_P(cstring_to_text(""));
877 tgl 4448 ECB : }
4449 :
4450 : /* reset to pointing at first match, but now with positive fldnum */
877 tgl 4451 CBC 6 : text_position_reset(&state);
877 tgl 4452 GIC 6 : found = text_position_next(&state);
877 tgl 4453 CBC 6 : Assert(found);
877 tgl 4454 ECB : }
4455 :
4456 : /* identify bounds of first field */
877 tgl 4457 GIC 18 : start_ptr = VARDATA_ANY(inputstring);
1535 heikki.linnakangas 4458 CBC 18 : end_ptr = text_position_get_match_ptr(&state);
4459 :
1535 heikki.linnakangas 4460 GIC 33 : while (found && --fldnum > 0)
4461 : {
4462 : /* identify bounds of next field */
4463 15 : start_ptr = end_ptr + fldsep_len;
4464 15 : found = text_position_next(&state);
4465 15 : if (found)
4466 9 : end_ptr = text_position_get_match_ptr(&state);
4467 : }
6028 tgl 4468 ECB :
6028 tgl 4469 GIC 18 : text_position_cleanup(&state);
4470 :
4471 18 : if (fldnum > 0)
4472 : {
6028 tgl 4473 ECB : /* N'th field separator not found */
4474 : /* if last field requested, return it, else empty string */
6028 tgl 4475 CBC 6 : if (fldnum == 1)
1535 heikki.linnakangas 4476 ECB : {
1535 heikki.linnakangas 4477 GIC 3 : int last_len = start_ptr - VARDATA_ANY(inputstring);
1535 heikki.linnakangas 4478 ECB :
1535 heikki.linnakangas 4479 GIC 3 : result_text = cstring_to_text_with_len(start_ptr,
4480 : inputstring_len - last_len);
4481 : }
4482 : else
5493 tgl 4483 CBC 3 : result_text = cstring_to_text("");
7535 bruce 4484 ECB : }
4485 : else
4486 : {
4487 : /* non-last field requested */
1535 heikki.linnakangas 4488 GIC 12 : result_text = cstring_to_text_with_len(start_ptr, end_ptr - start_ptr);
4489 : }
4490 :
6028 tgl 4491 18 : PG_RETURN_TEXT_P(result_text);
4492 : }
7535 bruce 4493 ECB :
4494 : /*
4495 : * Convenience function to return true when two text params are equal.
4496 : */
4497 : static bool
1479 peter 4498 GIC 174 : text_isequal(text *txt1, text *txt2, Oid collid)
4625 tgl 4499 ECB : {
1479 peter 4500 GIC 174 : return DatumGetBool(DirectFunctionCall2Coll(texteq,
4501 : collid,
4502 : PointerGetDatum(txt1),
4503 : PointerGetDatum(txt2)));
4504 : }
4505 :
4506 : /*
7226 tgl 4507 ECB : * text_to_array
4508 : * parse input string and return text array of elements,
4509 : * based on provided field separator
4510 : */
4511 : Datum
7226 tgl 4512 CBC 73 : text_to_array(PG_FUNCTION_ARGS)
4513 : {
4514 : SplitTextOutputData tstate;
4515 :
4516 : /* For array output, tstate should start as all zeroes */
949 tgl 4517 GIC 73 : memset(&tstate, 0, sizeof(tstate));
4518 :
4519 73 : if (!split_text(fcinfo, &tstate))
4520 3 : PG_RETURN_NULL();
4521 :
4522 64 : if (tstate.astate == NULL)
949 tgl 4523 CBC 3 : PG_RETURN_ARRAYTYPE_P(construct_empty_array(TEXTOID));
4524 :
224 peter 4525 GNC 61 : PG_RETURN_DATUM(makeArrayResult(tstate.astate,
4526 : CurrentMemoryContext));
4527 : }
4528 :
4529 : /*
4625 tgl 4530 ECB : * text_to_array_null
4625 tgl 4531 EUB : * parse input string and return text array of elements,
4532 : * based on provided field separator and null string
4625 tgl 4533 ECB : *
4534 : * This is a separate entry point only to prevent the regression tests from
4535 : * complaining about different argument sets for the same internal function.
4536 : */
4537 : Datum
4625 tgl 4538 CBC 30 : text_to_array_null(PG_FUNCTION_ARGS)
4539 : {
949 4540 30 : return text_to_array(fcinfo);
4541 : }
949 tgl 4542 ECB :
4543 : /*
4544 : * text_to_table
4545 : * parse input string and return table of elements,
4546 : * based on provided field separator
4547 : */
4548 : Datum
949 tgl 4549 CBC 42 : text_to_table(PG_FUNCTION_ARGS)
4550 : {
949 tgl 4551 GIC 42 : ReturnSetInfo *rsi = (ReturnSetInfo *) fcinfo->resultinfo;
4552 : SplitTextOutputData tstate;
4553 :
4554 42 : tstate.astate = NULL;
173 michael 4555 42 : InitMaterializedSRF(fcinfo, MAT_SRF_USE_EXPECTED_DESC);
398 4556 42 : tstate.tupstore = rsi->setResult;
4557 42 : tstate.tupdesc = rsi->setDesc;
4558 :
949 tgl 4559 42 : (void) split_text(fcinfo, &tstate);
4560 :
949 tgl 4561 CBC 42 : return (Datum) 0;
4562 : }
4563 :
4564 : /*
4565 : * text_to_table_null
4566 : * parse input string and return table of elements,
4567 : * based on provided field separator and null string
949 tgl 4568 ECB : *
4569 : * This is a separate entry point only to prevent the regression tests from
4570 : * complaining about different argument sets for the same internal function.
4571 : */
4572 : Datum
949 tgl 4573 CBC 12 : text_to_table_null(PG_FUNCTION_ARGS)
949 tgl 4574 ECB : {
949 tgl 4575 GIC 12 : return text_to_table(fcinfo);
4625 tgl 4576 ECB : }
4577 :
4578 : /*
4579 : * Common code for text_to_array, text_to_array_null, text_to_table
4580 : * and text_to_table_null functions.
4581 : *
4582 : * These are not strict so we have to test for null inputs explicitly.
4583 : * Returns false if result is to be null, else returns true.
949 4584 : *
4585 : * Note that if the result is valid but empty (zero elements), we return
4586 : * without changing *tstate --- caller must handle that case, too.
4625 4587 : */
4588 : static bool
949 tgl 4589 CBC 115 : split_text(FunctionCallInfo fcinfo, SplitTextOutputData *tstate)
4625 tgl 4590 ECB : {
4591 : text *inputstring;
4592 : text *fldsep;
4593 : text *null_string;
949 tgl 4594 GIC 115 : Oid collation = PG_GET_COLLATION();
4595 : int inputstring_len;
4596 : int fldsep_len;
4597 : char *start_ptr;
7008 tgl 4598 ECB : text *result_text;
4599 :
4600 : /* when input string is NULL, then result is NULL too */
4625 tgl 4601 GIC 115 : if (PG_ARGISNULL(0))
949 tgl 4602 CBC 6 : return false;
7226 tgl 4603 ECB :
4625 tgl 4604 CBC 109 : inputstring = PG_GETARG_TEXT_PP(0);
4605 :
4625 tgl 4606 ECB : /* fldsep can be NULL */
4625 tgl 4607 CBC 109 : if (!PG_ARGISNULL(1))
4608 94 : fldsep = PG_GETARG_TEXT_PP(1);
4609 : else
4610 15 : fldsep = NULL;
4625 tgl 4611 ECB :
4612 : /* null_string can be NULL or omitted */
4625 tgl 4613 GIC 109 : if (PG_NARGS() > 2 && !PG_ARGISNULL(2))
4625 tgl 4614 CBC 42 : null_string = PG_GETARG_TEXT_PP(2);
4615 : else
4625 tgl 4616 GIC 67 : null_string = NULL;
4617 :
4618 109 : if (fldsep != NULL)
4619 : {
4625 tgl 4620 ECB : /*
4621 : * Normal case with non-null fldsep. Use the text_position machinery
4622 : * to search for occurrences of fldsep.
4623 : */
4624 : TextPositionState state;
4520 peter_e 4625 :
1535 heikki.linnakangas 4626 CBC 94 : inputstring_len = VARSIZE_ANY_EXHDR(inputstring);
1535 heikki.linnakangas 4627 GIC 94 : fldsep_len = VARSIZE_ANY_EXHDR(fldsep);
7226 tgl 4628 EUB :
949 tgl 4629 ECB : /* return empty set for empty input string */
4625 tgl 4630 GIC 94 : if (inputstring_len < 1)
949 4631 30 : return true;
4632 :
4633 : /* empty field separator: return input string as a one-element set */
4625 tgl 4634 CBC 88 : if (fldsep_len < 1)
4635 : {
949 4636 24 : split_text_accum_result(tstate, inputstring,
4637 : null_string, collation);
4638 24 : return true;
7226 tgl 4639 ECB : }
4640 :
949 tgl 4641 CBC 64 : text_position_setup(inputstring, fldsep, collation, &state);
1535 heikki.linnakangas 4642 ECB :
4625 tgl 4643 GIC 58 : start_ptr = VARDATA_ANY(inputstring);
4625 tgl 4644 ECB :
1535 4645 : for (;;)
4625 tgl 4646 GIC 232 : {
4647 : bool found;
4648 : char *end_ptr;
1535 tgl 4649 ECB : int chunk_len;
4650 :
1535 heikki.linnakangas 4651 CBC 290 : CHECK_FOR_INTERRUPTS();
5996 tgl 4652 ECB :
1535 heikki.linnakangas 4653 GIC 290 : found = text_position_next(&state);
1535 heikki.linnakangas 4654 GBC 290 : if (!found)
4625 tgl 4655 EUB : {
4656 : /* fetch last field */
4625 tgl 4657 GIC 58 : chunk_len = ((char *) inputstring + VARSIZE_ANY(inputstring)) - start_ptr;
1535 4658 58 : end_ptr = NULL; /* not used, but some compilers complain */
4659 : }
4625 tgl 4660 ECB : else
4661 : {
4662 : /* fetch non-last field */
1535 heikki.linnakangas 4663 CBC 232 : end_ptr = text_position_get_match_ptr(&state);
1535 heikki.linnakangas 4664 GIC 232 : chunk_len = end_ptr - start_ptr;
4665 : }
4666 :
4667 : /* build a temp text datum to pass to split_text_accum_result */
4625 tgl 4668 290 : result_text = cstring_to_text_with_len(start_ptr, chunk_len);
4669 :
4670 : /* stash away this field */
949 4671 290 : split_text_accum_result(tstate, result_text,
949 tgl 4672 ECB : null_string, collation);
4673 :
4625 tgl 4674 CBC 290 : pfree(result_text);
4675 :
1535 heikki.linnakangas 4676 290 : if (!found)
4625 tgl 4677 GIC 58 : break;
4678 :
1535 heikki.linnakangas 4679 CBC 232 : start_ptr = end_ptr + fldsep_len;
4625 tgl 4680 ECB : }
4681 :
4625 tgl 4682 GIC 58 : text_position_cleanup(&state);
4683 : }
4625 tgl 4684 ECB : else
4685 : {
4520 peter_e 4686 : /*
4687 : * When fldsep is NULL, each character in the input string becomes a
949 tgl 4688 : * separate element in the result set. The separator is effectively
4689 : * the space between characters.
4690 : */
4625 tgl 4691 GIC 15 : inputstring_len = VARSIZE_ANY_EXHDR(inputstring);
4692 :
4693 15 : start_ptr = VARDATA_ANY(inputstring);
4694 :
4695 126 : while (inputstring_len > 0)
4625 tgl 4696 ECB : {
4382 bruce 4697 GIC 111 : int chunk_len = pg_mblen(start_ptr);
7226 tgl 4698 ECB :
4625 tgl 4699 GIC 111 : CHECK_FOR_INTERRUPTS();
4625 tgl 4700 ECB :
4701 : /* build a temp text datum to pass to split_text_accum_result */
4625 tgl 4702 GIC 111 : result_text = cstring_to_text_with_len(start_ptr, chunk_len);
4520 peter_e 4703 ECB :
4625 tgl 4704 : /* stash away this field */
949 tgl 4705 GIC 111 : split_text_accum_result(tstate, result_text,
4706 : null_string, collation);
4707 :
4625 tgl 4708 CBC 111 : pfree(result_text);
4625 tgl 4709 ECB :
4625 tgl 4710 CBC 111 : start_ptr += chunk_len;
4625 tgl 4711 GIC 111 : inputstring_len -= chunk_len;
4625 tgl 4712 ECB : }
4713 : }
4714 :
949 tgl 4715 GIC 73 : return true;
4716 : }
4717 :
4718 : /*
4719 : * Add text item to result set (table or array).
4720 : *
949 tgl 4721 ECB : * This is also responsible for checking to see if the item matches
4722 : * the null_string, in which case we should emit NULL instead.
4723 : */
4724 : static void
949 tgl 4725 GIC 425 : split_text_accum_result(SplitTextOutputData *tstate,
4726 : text *field_value,
4727 : text *null_string,
949 tgl 4728 ECB : Oid collation)
4729 : {
949 tgl 4730 GIC 425 : bool is_null = false;
949 tgl 4731 ECB :
949 tgl 4732 GIC 425 : if (null_string && text_isequal(field_value, null_string, collation))
949 tgl 4733 CBC 30 : is_null = true;
949 tgl 4734 ECB :
949 tgl 4735 GBC 425 : if (tstate->tupstore)
4736 : {
949 tgl 4737 ECB : Datum values[1];
4738 : bool nulls[1];
4739 :
949 tgl 4740 GIC 114 : values[0] = PointerGetDatum(field_value);
4741 114 : nulls[0] = is_null;
949 tgl 4742 EUB :
949 tgl 4743 GIC 114 : tuplestore_putvalues(tstate->tupstore,
949 tgl 4744 ECB : tstate->tupdesc,
4745 : values,
4746 : nulls);
4747 : }
4748 : else
949 tgl 4749 EUB : {
949 tgl 4750 GIC 311 : tstate->astate = accumArrayResult(tstate->astate,
4751 : PointerGetDatum(field_value),
949 tgl 4752 EUB : is_null,
4753 : TEXTOID,
4754 : CurrentMemoryContext);
4755 : }
7226 tgl 4756 GIC 425 : }
7226 tgl 4757 EUB :
4758 : /*
4759 : * array_to_text
7226 tgl 4760 ECB : * concatenate Cstring representation of input array elements
4761 : * using provided field separator
4762 : */
4763 : Datum
7226 tgl 4764 GIC 30154 : array_to_text(PG_FUNCTION_ARGS)
4765 : {
4766 30154 : ArrayType *v = PG_GETARG_ARRAYTYPE_P(0);
5493 4767 30154 : char *fldsep = text_to_cstring(PG_GETARG_TEXT_PP(1));
4625 tgl 4768 ECB :
4625 tgl 4769 GIC 30154 : PG_RETURN_TEXT_P(array_to_text_internal(fcinfo, v, fldsep, NULL));
4770 : }
4771 :
4772 : /*
4773 : * array_to_text_null
4774 : * concatenate Cstring representation of input array elements
4625 tgl 4775 ECB : * using provided field separator and null string
4776 : *
4777 : * This version is not strict so we have to test for null inputs explicitly.
4778 : */
4779 : Datum
4625 tgl 4780 CBC 6 : array_to_text_null(PG_FUNCTION_ARGS)
4625 tgl 4781 ECB : {
4625 tgl 4782 EUB : ArrayType *v;
4783 : char *fldsep;
4625 tgl 4784 ECB : char *null_string;
4785 :
4786 : /* returns NULL when first or second parameter is NULL */
4625 tgl 4787 GIC 6 : if (PG_ARGISNULL(0) || PG_ARGISNULL(1))
4625 tgl 4788 UIC 0 : PG_RETURN_NULL();
4520 peter_e 4789 ECB :
4625 tgl 4790 GIC 6 : v = PG_GETARG_ARRAYTYPE_P(0);
4625 tgl 4791 CBC 6 : fldsep = text_to_cstring(PG_GETARG_TEXT_PP(1));
4625 tgl 4792 EUB :
4793 : /* NULL null string is passed through as a null pointer */
4625 tgl 4794 GIC 6 : if (!PG_ARGISNULL(2))
4625 tgl 4795 CBC 3 : null_string = text_to_cstring(PG_GETARG_TEXT_PP(2));
4625 tgl 4796 ECB : else
4625 tgl 4797 CBC 3 : null_string = NULL;
4625 tgl 4798 ECB :
4625 tgl 4799 GIC 6 : PG_RETURN_TEXT_P(array_to_text_internal(fcinfo, v, fldsep, null_string));
4800 : }
4625 tgl 4801 ECB :
4802 : /*
4803 : * common code for array_to_text and array_to_text_null functions
4804 : */
4805 : static text *
4625 tgl 4806 CBC 30169 : array_to_text_internal(FunctionCallInfo fcinfo, ArrayType *v,
3726 tgl 4807 ECB : const char *fldsep, const char *null_string)
4625 4808 : {
4625 tgl 4809 EUB : text *result;
7188 bruce 4810 : int nitems,
4811 : *dims,
4812 : ndims;
7226 tgl 4813 ECB : Oid element_type;
4814 : int typlen;
4815 : bool typbyval;
4816 : char typalign;
4817 : StringInfoData buf;
6351 tgl 4818 GIC 30169 : bool printed = false;
4819 : char *p;
4820 : bits8 *bitmap;
4821 : int bitmask;
4822 : int i;
4823 : ArrayMetaState *my_extra;
4824 :
7226 4825 30169 : ndims = ARR_NDIM(v);
4826 30169 : dims = ARR_DIMS(v);
4827 30169 : nitems = ArrayGetNItems(ndims, dims);
7226 tgl 4828 ECB :
4829 : /* if there are no elements, return an empty string */
7226 tgl 4830 GIC 30169 : if (nitems == 0)
4625 4831 18246 : return cstring_to_text_with_len("", 0);
4832 :
7226 4833 11923 : element_type = ARR_ELEMTYPE(v);
6248 neilc 4834 CBC 11923 : initStringInfo(&buf);
4835 :
4836 : /*
7226 tgl 4837 EUB : * We arrange to look up info about element type, including its output
4838 : * conversion proc, only once per series of calls, assuming the element
4839 : * type doesn't change underneath us.
4840 : */
7226 tgl 4841 GIC 11923 : my_extra = (ArrayMetaState *) fcinfo->flinfo->fn_extra;
4842 11923 : if (my_extra == NULL)
4843 : {
7226 tgl 4844 CBC 695 : fcinfo->flinfo->fn_extra = MemoryContextAlloc(fcinfo->flinfo->fn_mcxt,
6385 bruce 4845 ECB : sizeof(ArrayMetaState));
7226 tgl 4846 CBC 695 : my_extra = (ArrayMetaState *) fcinfo->flinfo->fn_extra;
6351 tgl 4847 GIC 695 : my_extra->element_type = ~element_type;
7226 tgl 4848 ECB : }
4849 :
7226 tgl 4850 GIC 11923 : if (my_extra->element_type != element_type)
4851 : {
7188 bruce 4852 ECB : /*
4853 : * Get info about element type, including its output conversion proc
4854 : */
7226 tgl 4855 GIC 695 : get_type_io_data(element_type, IOFunc_output,
7226 tgl 4856 ECB : &my_extra->typlen, &my_extra->typbyval,
4857 : &my_extra->typalign, &my_extra->typdelim,
4858 : &my_extra->typioparam, &my_extra->typiofunc);
7226 tgl 4859 CBC 695 : fmgr_info_cxt(my_extra->typiofunc, &my_extra->proc,
7226 tgl 4860 GIC 695 : fcinfo->flinfo->fn_mcxt);
7226 tgl 4861 CBC 695 : my_extra->element_type = element_type;
7226 tgl 4862 ECB : }
7226 tgl 4863 GIC 11923 : typlen = my_extra->typlen;
4864 11923 : typbyval = my_extra->typbyval;
4865 11923 : typalign = my_extra->typalign;
4866 :
6351 4867 11923 : p = ARR_DATA_PTR(v);
4868 11923 : bitmap = ARR_NULLBITMAP(v);
4869 11923 : bitmask = 1;
4870 :
7226 4871 40628 : for (i = 0; i < nitems; i++)
4872 : {
4873 : Datum itemvalue;
4874 : char *value;
7226 tgl 4875 ECB :
4876 : /* Get source element, checking for NULL */
6351 tgl 4877 CBC 28705 : if (bitmap && (*bitmap & bitmask) == 0)
6351 tgl 4878 ECB : {
4879 : /* if null_string is NULL, we just ignore null elements */
4625 tgl 4880 GIC 9 : if (null_string != NULL)
4625 tgl 4881 ECB : {
4625 tgl 4882 GIC 3 : if (printed)
4625 tgl 4883 CBC 3 : appendStringInfo(&buf, "%s%s", fldsep, null_string);
4884 : else
4625 tgl 4885 LBC 0 : appendStringInfoString(&buf, null_string);
4625 tgl 4886 CBC 3 : printed = true;
4625 tgl 4887 ECB : }
4888 : }
4889 : else
6351 4890 : {
6351 tgl 4891 GIC 28696 : itemvalue = fetch_att(p, typbyval, typlen);
4892 :
6214 4893 28696 : value = OutputFunctionCall(&my_extra->proc, itemvalue);
4894 :
6351 4895 28696 : if (printed)
6248 neilc 4896 16773 : appendStringInfo(&buf, "%s%s", fldsep, value);
6351 tgl 4897 ECB : else
6248 neilc 4898 CBC 11923 : appendStringInfoString(&buf, value);
6351 tgl 4899 28696 : printed = true;
4900 :
5847 tgl 4901 GIC 28696 : p = att_addlength_pointer(p, typlen, p);
4902 28696 : p = (char *) att_align_nominal(p, typalign);
4903 : }
4904 :
4905 : /* advance bitmap pointer if any */
6351 4906 28705 : if (bitmap)
6351 tgl 4907 ECB : {
6351 tgl 4908 GIC 54 : bitmask <<= 1;
4909 54 : if (bitmask == 0x100)
4910 : {
6351 tgl 4911 UIC 0 : bitmap++;
4912 0 : bitmask = 1;
6351 tgl 4913 ECB : }
6351 tgl 4914 EUB : }
4915 : }
4520 peter_e 4916 ECB :
4625 tgl 4917 CBC 11923 : result = cstring_to_text_with_len(buf.data, buf.len);
4625 tgl 4918 GIC 11923 : pfree(buf.data);
7226 tgl 4919 ECB :
4625 tgl 4920 GIC 11923 : return result;
4921 : }
4922 :
4923 : #define HEXBASE 16
4924 : /*
2881 heikki.linnakangas 4925 EUB : * Convert an int32 to a string containing a base 16 (hex) representation of
7535 bruce 4926 : * the number.
4927 : */
4928 : Datum
7535 bruce 4929 GIC 19342 : to_hex32(PG_FUNCTION_ARGS)
7535 bruce 4930 ECB : {
7051 tgl 4931 GIC 19342 : uint32 value = (uint32) PG_GETARG_INT32(0);
4932 : char *ptr;
4933 19342 : const char *digits = "0123456789abcdef";
4934 : char buf[32]; /* bigger than needed, but reasonable */
7535 bruce 4935 ECB :
7529 bruce 4936 CBC 19342 : ptr = buf + sizeof(buf) - 1;
7535 4937 19342 : *ptr = '\0';
7535 bruce 4938 ECB :
4939 : do
4940 : {
7535 bruce 4941 CBC 37279 : *--ptr = digits[value % HEXBASE];
7535 bruce 4942 GIC 37279 : value /= HEXBASE;
4943 37279 : } while (ptr > buf && value);
7535 bruce 4944 ECB :
5493 tgl 4945 GIC 19342 : PG_RETURN_TEXT_P(cstring_to_text(ptr));
4946 : }
7535 bruce 4947 ECB :
4948 : /*
4949 : * Convert an int64 to a string containing a base 16 (hex) representation of
4950 : * the number.
4951 : */
4952 : Datum
7535 bruce 4953 GIC 3 : to_hex64(PG_FUNCTION_ARGS)
4954 : {
7051 tgl 4955 3 : uint64 value = (uint64) PG_GETARG_INT64(0);
4956 : char *ptr;
7051 tgl 4957 CBC 3 : const char *digits = "0123456789abcdef";
4958 : char buf[32]; /* bigger than needed, but reasonable */
4959 :
7529 bruce 4960 GIC 3 : ptr = buf + sizeof(buf) - 1;
7535 4961 3 : *ptr = '\0';
4962 :
4963 : do
7535 bruce 4964 ECB : {
7535 bruce 4965 GIC 24 : *--ptr = digits[value % HEXBASE];
7535 bruce 4966 CBC 24 : value /= HEXBASE;
7535 bruce 4967 GIC 24 : } while (ptr > buf && value);
7535 bruce 4968 ECB :
5493 tgl 4969 GIC 3 : PG_RETURN_TEXT_P(cstring_to_text(ptr));
4970 : }
7429 bruce 4971 ECB :
4972 : /*
4973 : * Return the size of a datum, possibly compressed
6459 tgl 4974 : *
4975 : * Works on any data type
6486 bruce 4976 : */
4977 : Datum
6486 bruce 4978 CBC 61 : pg_column_size(PG_FUNCTION_ARGS)
4979 : {
6459 tgl 4980 GIC 61 : Datum value = PG_GETARG_DATUM(0);
4981 : int32 result;
4982 : int typlen;
4983 :
4984 : /* On first call, get the input type's typlen, and save at *fn_extra */
4985 61 : if (fcinfo->flinfo->fn_extra == NULL)
4986 : {
4987 : /* Lookup the datatype of the supplied argument */
6385 bruce 4988 CBC 61 : Oid argtypeid = get_fn_expr_argtype(fcinfo->flinfo, 0);
4989 :
6459 tgl 4990 GIC 61 : typlen = get_typlen(argtypeid);
4991 61 : if (typlen == 0) /* should not happen */
6485 bruce 4992 UIC 0 : elog(ERROR, "cache lookup failed for type %u", argtypeid);
4993 :
6486 bruce 4994 GIC 61 : fcinfo->flinfo->fn_extra = MemoryContextAlloc(fcinfo->flinfo->fn_mcxt,
4995 : sizeof(int));
6459 tgl 4996 61 : *((int *) fcinfo->flinfo->fn_extra) = typlen;
6486 bruce 4997 ECB : }
4998 : else
6459 tgl 4999 LBC 0 : typlen = *((int *) fcinfo->flinfo->fn_extra);
5000 :
6459 tgl 5001 GIC 61 : if (typlen == -1)
5002 : {
5003 : /* varlena type, possibly toasted */
5004 61 : result = toast_datum_size(value);
6459 tgl 5005 ECB : }
6459 tgl 5006 LBC 0 : else if (typlen == -2)
6459 tgl 5007 ECB : {
5008 : /* cstring */
6459 tgl 5009 LBC 0 : result = strlen(DatumGetCString(value)) + 1;
5010 : }
5011 : else
6486 bruce 5012 ECB : {
5013 : /* ordinary fixed-width type */
6459 tgl 5014 UIC 0 : result = typlen;
6486 bruce 5015 ECB : }
6459 tgl 5016 :
6459 tgl 5017 CBC 61 : PG_RETURN_INT32(result);
5018 : }
4815 itagaki.takahiro 5019 ECB :
751 rhaas 5020 : /*
5021 : * Return the compression method stored in the compressed attribute. Return
5022 : * NULL for non varlena type or uncompressed data.
5023 : */
5024 : Datum
751 rhaas 5025 GIC 81 : pg_column_compression(PG_FUNCTION_ARGS)
751 rhaas 5026 ECB : {
5027 : int typlen;
5028 : char *result;
5029 : ToastCompressionId cmid;
5030 :
5031 : /* On first call, get the input type's typlen, and save at *fn_extra */
751 rhaas 5032 GIC 81 : if (fcinfo->flinfo->fn_extra == NULL)
751 rhaas 5033 ECB : {
5034 : /* Lookup the datatype of the supplied argument */
751 rhaas 5035 CBC 54 : Oid argtypeid = get_fn_expr_argtype(fcinfo->flinfo, 0);
5036 :
751 rhaas 5037 GIC 54 : typlen = get_typlen(argtypeid);
751 rhaas 5038 CBC 54 : if (typlen == 0) /* should not happen */
751 rhaas 5039 UIC 0 : elog(ERROR, "cache lookup failed for type %u", argtypeid);
5040 :
751 rhaas 5041 GIC 54 : fcinfo->flinfo->fn_extra = MemoryContextAlloc(fcinfo->flinfo->fn_mcxt,
751 rhaas 5042 ECB : sizeof(int));
751 rhaas 5043 GIC 54 : *((int *) fcinfo->flinfo->fn_extra) = typlen;
5044 : }
5045 : else
5046 27 : typlen = *((int *) fcinfo->flinfo->fn_extra);
5047 :
5048 81 : if (typlen != -1)
751 rhaas 5049 UIC 0 : PG_RETURN_NULL();
5050 :
5051 : /* get the compression method id stored in the compressed varlena */
751 rhaas 5052 CBC 81 : cmid = toast_get_compression_id((struct varlena *)
751 rhaas 5053 GIC 81 : DatumGetPointer(PG_GETARG_DATUM(0)));
5054 81 : if (cmid == TOAST_INVALID_COMPRESSION_ID)
5055 3 : PG_RETURN_NULL();
5056 :
5057 : /* convert compression method id to compression method name */
751 rhaas 5058 CBC 78 : switch (cmid)
751 rhaas 5059 ECB : {
751 rhaas 5060 GIC 33 : case TOAST_PGLZ_COMPRESSION_ID:
751 rhaas 5061 CBC 33 : result = "pglz";
751 rhaas 5062 GIC 33 : break;
5063 45 : case TOAST_LZ4_COMPRESSION_ID:
5064 45 : result = "lz4";
5065 45 : break;
751 rhaas 5066 UIC 0 : default:
751 rhaas 5067 LBC 0 : elog(ERROR, "invalid compression method id %d", cmid);
751 rhaas 5068 ECB : }
751 rhaas 5069 EUB :
751 rhaas 5070 GIC 78 : PG_RETURN_TEXT_P(cstring_to_text(result));
751 rhaas 5071 ECB : }
5072 :
5073 : /*
5074 : * string_agg - Concatenates values and returns string.
4815 itagaki.takahiro 5075 : *
5076 : * Syntax: string_agg(value text, delimiter text) RETURNS text
5077 : *
5078 : * Note: Any NULL values are ignored. The first-call delimiter isn't
5079 : * actually used at all, and on subsequent calls the delimiter precedes
5080 : * the associated value.
5081 : */
5082 :
5083 : /* subroutine to initialize state */
5084 : static StringInfo
4808 tgl 5085 GIC 1026 : makeStringAggState(FunctionCallInfo fcinfo)
5086 : {
5087 : StringInfo state;
5088 : MemoryContext aggcontext;
5089 : MemoryContext oldcontext;
4815 itagaki.takahiro 5090 ECB :
4808 tgl 5091 GIC 1026 : if (!AggCheckCallContext(fcinfo, &aggcontext))
5092 : {
5093 : /* cannot be called directly because of internal-type argument */
4815 itagaki.takahiro 5094 UIC 0 : elog(ERROR, "string_agg_transfn called in non-aggregate context");
5095 : }
4815 itagaki.takahiro 5096 ECB :
5097 : /*
5098 : * Create state in aggregate context. It'll stay there across subsequent
5099 : * calls.
5100 : */
4815 itagaki.takahiro 5101 GIC 1026 : oldcontext = MemoryContextSwitchTo(aggcontext);
5102 1026 : state = makeStringInfo();
5103 1026 : MemoryContextSwitchTo(oldcontext);
4815 itagaki.takahiro 5104 ECB :
4815 itagaki.takahiro 5105 GIC 1026 : return state;
5106 : }
5107 :
5108 : Datum
4815 itagaki.takahiro 5109 CBC 420976 : string_agg_transfn(PG_FUNCTION_ARGS)
5110 : {
5111 : StringInfo state;
4815 itagaki.takahiro 5112 ECB :
4815 itagaki.takahiro 5113 CBC 420976 : state = PG_ARGISNULL(0) ? NULL : (StringInfo) PG_GETARG_POINTER(0);
5114 :
5115 : /* Append the value unless null, preceding it with the delimiter. */
4815 itagaki.takahiro 5116 GIC 420976 : if (!PG_ARGISNULL(1))
5117 : {
76 drowley 5118 GNC 413452 : text *value = PG_GETARG_TEXT_PP(1);
5119 413452 : bool isfirst = false;
5120 :
5121 : /*
5122 : * You might think we can just throw away the first delimiter, however
5123 : * we must keep it as we may be a parallel worker doing partial
5124 : * aggregation building a state to send to the main process. We need
5125 : * to keep the delimiter of every aggregation so that the combine
5126 : * function can properly join up the strings of two separately
5127 : * partially aggregated results. The first delimiter is only stripped
5128 : * off in the final function. To know how much to strip off the front
5129 : * of the string, we store the length of the first delimiter in the
5130 : * StringInfo's cursor field, which we don't otherwise need here.
5131 : */
4815 itagaki.takahiro 5132 413452 : if (state == NULL)
5133 : {
4808 tgl 5134 773 : state = makeStringAggState(fcinfo);
76 drowley 5135 773 : isfirst = true;
5136 : }
5137 :
5138 413452 : if (!PG_ARGISNULL(2))
5139 : {
5140 413452 : text *delim = PG_GETARG_TEXT_PP(2);
5141 :
5142 413452 : appendStringInfoText(state, delim);
5143 413452 : if (isfirst)
5144 773 : state->cursor = VARSIZE_ANY_EXHDR(delim);
5145 : }
5146 :
5147 413452 : appendStringInfoText(state, value);
4815 itagaki.takahiro 5148 ECB : }
5149 :
5150 : /*
5151 : * The transition type for string_agg() is declared to be "internal",
5152 : * which is a pass-by-value type the same size as a pointer.
5153 : */
76 drowley 5154 GNC 420976 : if (state)
5155 420931 : PG_RETURN_POINTER(state);
5156 45 : PG_RETURN_NULL();
5157 : }
5158 :
5159 : /*
5160 : * string_agg_combine
5161 : * Aggregate combine function for string_agg(text) and string_agg(bytea)
5162 : */
5163 : Datum
5164 100 : string_agg_combine(PG_FUNCTION_ARGS)
5165 : {
5166 : StringInfo state1;
5167 : StringInfo state2;
5168 : MemoryContext agg_context;
5169 :
5170 100 : if (!AggCheckCallContext(fcinfo, &agg_context))
76 drowley 5171 UNC 0 : elog(ERROR, "aggregate function called in non-aggregate context");
5172 :
76 drowley 5173 GNC 100 : state1 = PG_ARGISNULL(0) ? NULL : (StringInfo) PG_GETARG_POINTER(0);
5174 100 : state2 = PG_ARGISNULL(1) ? NULL : (StringInfo) PG_GETARG_POINTER(1);
5175 :
5176 100 : if (state2 == NULL)
5177 : {
5178 : /*
5179 : * NULL state2 is easy, just return state1, which we know is already
5180 : * in the agg_context
5181 : */
76 drowley 5182 UNC 0 : if (state1 == NULL)
5183 0 : PG_RETURN_NULL();
5184 0 : PG_RETURN_POINTER(state1);
5185 : }
5186 :
76 drowley 5187 GNC 100 : if (state1 == NULL)
5188 : {
5189 : /* We must copy state2's data into the agg_context */
5190 : MemoryContext old_context;
5191 :
5192 60 : old_context = MemoryContextSwitchTo(agg_context);
5193 60 : state1 = makeStringAggState(fcinfo);
5194 60 : appendBinaryStringInfo(state1, state2->data, state2->len);
5195 60 : state1->cursor = state2->cursor;
5196 60 : MemoryContextSwitchTo(old_context);
5197 : }
5198 40 : else if (state2->len > 0)
5199 : {
5200 : /* Combine ... state1->cursor does not change in this case */
5201 40 : appendBinaryStringInfo(state1, state2->data, state2->len);
5202 : }
5203 :
5204 100 : PG_RETURN_POINTER(state1);
5205 : }
5206 :
5207 : /*
5208 : * string_agg_serialize
5209 : * Aggregate serialize function for string_agg(text) and string_agg(bytea)
5210 : *
5211 : * This is strict, so we need not handle NULL input
5212 : */
5213 : Datum
5214 100 : string_agg_serialize(PG_FUNCTION_ARGS)
5215 : {
5216 : StringInfo state;
5217 : StringInfoData buf;
5218 : bytea *result;
5219 :
5220 : /* cannot be called directly because of internal-type argument */
5221 100 : Assert(AggCheckCallContext(fcinfo, NULL));
5222 :
5223 100 : state = (StringInfo) PG_GETARG_POINTER(0);
5224 :
5225 100 : pq_begintypsend(&buf);
5226 :
5227 : /* cursor */
5228 100 : pq_sendint(&buf, state->cursor, 4);
5229 :
5230 : /* data */
5231 100 : pq_sendbytes(&buf, state->data, state->len);
5232 :
5233 100 : result = pq_endtypsend(&buf);
5234 :
5235 100 : PG_RETURN_BYTEA_P(result);
5236 : }
5237 :
5238 : /*
5239 : * string_agg_deserialize
5240 : * Aggregate deserial function for string_agg(text) and string_agg(bytea)
5241 : *
5242 : * This is strict, so we need not handle NULL input
5243 : */
5244 : Datum
5245 100 : string_agg_deserialize(PG_FUNCTION_ARGS)
5246 : {
5247 : bytea *sstate;
5248 : StringInfo result;
5249 : StringInfoData buf;
5250 : char *data;
5251 : int datalen;
5252 :
5253 : /* cannot be called directly because of internal-type argument */
5254 100 : Assert(AggCheckCallContext(fcinfo, NULL));
5255 :
5256 100 : sstate = PG_GETARG_BYTEA_PP(0);
5257 :
5258 : /*
5259 : * Copy the bytea into a StringInfo so that we can "receive" it using the
5260 : * standard recv-function infrastructure.
5261 : */
5262 100 : initStringInfo(&buf);
5263 200 : appendBinaryStringInfo(&buf,
5264 200 : VARDATA_ANY(sstate), VARSIZE_ANY_EXHDR(sstate));
5265 :
5266 100 : result = makeStringAggState(fcinfo);
5267 :
5268 : /* cursor */
5269 100 : result->cursor = pq_getmsgint(&buf, 4);
5270 :
5271 : /* data */
5272 100 : datalen = VARSIZE_ANY_EXHDR(sstate) - 4;
5273 100 : data = (char *) pq_getmsgbytes(&buf, datalen);
5274 100 : appendBinaryStringInfo(result, data, datalen);
5275 :
5276 100 : pq_getmsgend(&buf);
5277 100 : pfree(buf.data);
5278 :
5279 100 : PG_RETURN_POINTER(result);
5280 : }
5281 :
5282 : Datum
4815 itagaki.takahiro 5283 CBC 789 : string_agg_finalfn(PG_FUNCTION_ARGS)
5284 : {
5285 : StringInfo state;
4815 itagaki.takahiro 5286 ECB :
5287 : /* cannot be called directly because of internal-type argument */
4808 tgl 5288 CBC 789 : Assert(AggCheckCallContext(fcinfo, NULL));
5289 :
5290 789 : state = PG_ARGISNULL(0) ? NULL : (StringInfo) PG_GETARG_POINTER(0);
5291 :
4815 itagaki.takahiro 5292 789 : if (state != NULL)
5293 : {
5294 : /* As per comment in transfn, strip data before the cursor position */
76 drowley 5295 GNC 753 : PG_RETURN_TEXT_P(cstring_to_text_with_len(&state->data[state->cursor],
5296 : state->len - state->cursor));
5297 : }
4815 itagaki.takahiro 5298 ECB : else
4815 itagaki.takahiro 5299 GIC 36 : PG_RETURN_NULL();
5300 : }
4611 itagaki.takahiro 5301 ECB :
2028 tgl 5302 : /*
5303 : * Prepare cache with fmgr info for the output functions of the datatypes of
5304 : * the arguments of a concat-like function, beginning with argument "argidx".
5305 : * (Arguments before that will have corresponding slots in the resulting
5306 : * FmgrInfo array, but we don't fill those slots.)
5307 : */
5308 : static FmgrInfo *
2028 tgl 5309 GIC 20 : build_concat_foutcache(FunctionCallInfo fcinfo, int argidx)
5310 : {
5311 : FmgrInfo *foutcache;
2028 tgl 5312 ECB : int i;
5313 :
5314 : /* We keep the info in fn_mcxt so it survives across calls */
2028 tgl 5315 CBC 20 : foutcache = (FmgrInfo *) MemoryContextAlloc(fcinfo->flinfo->fn_mcxt,
2028 tgl 5316 GIC 20 : PG_NARGS() * sizeof(FmgrInfo));
5317 :
5318 98 : for (i = argidx; i < PG_NARGS(); i++)
5319 : {
5320 : Oid valtype;
5321 : Oid typOutput;
2028 tgl 5322 ECB : bool typIsVarlena;
5323 :
2028 tgl 5324 GIC 78 : valtype = get_fn_expr_argtype(fcinfo->flinfo, i);
5325 78 : if (!OidIsValid(valtype))
2028 tgl 5326 LBC 0 : elog(ERROR, "could not determine data type of concat() input");
2028 tgl 5327 ECB :
2028 tgl 5328 CBC 78 : getTypeOutputInfo(valtype, &typOutput, &typIsVarlena);
5329 78 : fmgr_info_cxt(typOutput, &foutcache[i], fcinfo->flinfo->fn_mcxt);
5330 : }
5331 :
2028 tgl 5332 GIC 20 : fcinfo->flinfo->fn_extra = foutcache;
5333 :
5334 20 : return foutcache;
5335 : }
5336 :
4241 tgl 5337 ECB : /*
5338 : * Implementation of both concat() and concat_ws().
5339 : *
5340 : * sepstr is the separator string to place between values.
5341 : * argidx identifies the first argument to concatenate (counting from zero);
5342 : * note that this must be constant across any one series of calls.
2028 5343 : *
3726 5344 : * Returns NULL if result should be NULL, else text value.
4241 5345 : */
5346 : static text *
3726 tgl 5347 CBC 36 : concat_internal(const char *sepstr, int argidx,
4241 tgl 5348 ECB : FunctionCallInfo fcinfo)
4611 itagaki.takahiro 5349 : {
4382 bruce 5350 : text *result;
5351 : StringInfoData str;
5352 : FmgrInfo *foutcache;
4241 tgl 5353 GIC 36 : bool first_arg = true;
5354 : int i;
5355 :
5356 : /*
5357 : * concat(VARIADIC some-array) is essentially equivalent to
3726 tgl 5358 ECB : * array_to_text(), ie concat the array elements with the given separator.
5359 : * So we just pass the case off to that code.
5360 : */
3726 tgl 5361 GIC 36 : if (get_fn_expr_variadic(fcinfo->flinfo))
3726 tgl 5362 ECB : {
5363 : ArrayType *arr;
5364 :
5365 : /* Should have just the one argument */
3726 tgl 5366 CBC 15 : Assert(argidx == PG_NARGS() - 1);
5367 :
5368 : /* concat(VARIADIC NULL) is defined as NULL */
5369 15 : if (PG_ARGISNULL(argidx))
5370 6 : return NULL;
3726 tgl 5371 ECB :
5372 : /*
5373 : * Non-null argument had better be an array. We assume that any call
3293 5374 : * context that could let get_fn_expr_variadic return true will have
5375 : * checked that a VARIADIC-labeled parameter actually is an array. So
5376 : * it should be okay to just Assert that it's an array rather than
5377 : * doing a full-fledged error check.
5378 : */
3293 tgl 5379 GIC 9 : Assert(OidIsValid(get_base_element_type(get_fn_expr_argtype(fcinfo->flinfo, argidx))));
5380 :
5381 : /* OK, safe to fetch the array value */
3726 tgl 5382 CBC 9 : arr = PG_GETARG_ARRAYTYPE_P(argidx);
5383 :
3726 tgl 5384 ECB : /*
5385 : * And serialize the array. We tell array_to_text to ignore null
5386 : * elements, which matches the behavior of the loop below.
5387 : */
3726 tgl 5388 GIC 9 : return array_to_text_internal(fcinfo, arr, sepstr, NULL);
5389 : }
3726 tgl 5390 ECB :
5391 : /* Normal case without explicit VARIADIC marker */
4611 itagaki.takahiro 5392 GIC 21 : initStringInfo(&str);
4611 itagaki.takahiro 5393 ECB :
2028 tgl 5394 : /* Get output function info, building it if first time through */
2028 tgl 5395 GIC 21 : foutcache = (FmgrInfo *) fcinfo->flinfo->fn_extra;
2028 tgl 5396 CBC 21 : if (foutcache == NULL)
2028 tgl 5397 GIC 20 : foutcache = build_concat_foutcache(fcinfo, argidx);
5398 :
4611 itagaki.takahiro 5399 102 : for (i = argidx; i < PG_NARGS(); i++)
5400 : {
5401 81 : if (!PG_ARGISNULL(i))
5402 : {
4241 tgl 5403 CBC 75 : Datum value = PG_GETARG_DATUM(i);
5404 :
4241 tgl 5405 ECB : /* add separator if appropriate */
4241 tgl 5406 CBC 75 : if (first_arg)
5407 21 : first_arg = false;
4241 tgl 5408 ECB : else
3726 tgl 5409 GIC 54 : appendStringInfoString(&str, sepstr);
5410 :
5411 : /* call the appropriate type output function, append the result */
4611 itagaki.takahiro 5412 CBC 75 : appendStringInfoString(&str,
2028 tgl 5413 75 : OutputFunctionCall(&foutcache[i], value));
4611 itagaki.takahiro 5414 ECB : }
5415 : }
5416 :
4611 itagaki.takahiro 5417 GIC 21 : result = cstring_to_text_with_len(str.data, str.len);
5418 21 : pfree(str.data);
4611 itagaki.takahiro 5419 ECB :
4611 itagaki.takahiro 5420 GIC 21 : return result;
5421 : }
5422 :
4611 itagaki.takahiro 5423 ECB : /*
5424 : * Concatenate all arguments. NULL arguments are ignored.
5425 : */
5426 : Datum
4611 itagaki.takahiro 5427 GIC 18 : text_concat(PG_FUNCTION_ARGS)
5428 : {
5429 : text *result;
5430 :
3726 tgl 5431 18 : result = concat_internal("", 0, fcinfo);
3726 tgl 5432 GBC 18 : if (result == NULL)
5433 3 : PG_RETURN_NULL();
3726 tgl 5434 GIC 15 : PG_RETURN_TEXT_P(result);
5435 : }
4611 itagaki.takahiro 5436 ECB :
5437 : /*
5438 : * Concatenate all but first argument value with separators. The first
5439 : * parameter is used as the separator. NULL arguments are ignored.
5440 : */
5441 : Datum
4611 itagaki.takahiro 5442 GIC 21 : text_concat_ws(PG_FUNCTION_ARGS)
5443 : {
5444 : char *sep;
5445 : text *result;
5446 :
5447 : /* return NULL when separator is NULL */
5448 21 : if (PG_ARGISNULL(0))
5449 3 : PG_RETURN_NULL();
3726 tgl 5450 18 : sep = text_to_cstring(PG_GETARG_TEXT_PP(0));
5451 :
5452 18 : result = concat_internal(sep, 1, fcinfo);
5453 18 : if (result == NULL)
5454 3 : PG_RETURN_NULL();
5455 15 : PG_RETURN_TEXT_P(result);
5456 : }
5457 :
4611 itagaki.takahiro 5458 ECB : /*
5459 : * Return first n characters in the string. When n is negative,
5460 : * return all but last |n| characters.
5461 : */
5462 : Datum
4611 itagaki.takahiro 5463 GIC 942 : text_left(PG_FUNCTION_ARGS)
5464 : {
1418 tgl 5465 942 : int n = PG_GETARG_INT32(1);
5466 :
4611 itagaki.takahiro 5467 942 : if (n < 0)
5468 : {
1468 sfrost 5469 CBC 15 : text *str = PG_GETARG_TEXT_PP(0);
5470 15 : const char *p = VARDATA_ANY(str);
5471 15 : int len = VARSIZE_ANY_EXHDR(str);
1468 sfrost 5472 ECB : int rlen;
4611 itagaki.takahiro 5473 :
1468 sfrost 5474 GIC 15 : n = pg_mbstrlen_with_len(p, len) + n;
5475 15 : rlen = pg_mbcharcliplen(p, len, n);
5476 15 : PG_RETURN_TEXT_P(cstring_to_text_with_len(p, rlen));
5477 : }
1468 sfrost 5478 ECB : else
1468 sfrost 5479 CBC 927 : PG_RETURN_TEXT_P(text_substring(PG_GETARG_DATUM(0), 1, n, false));
5480 : }
5481 :
4611 itagaki.takahiro 5482 ECB : /*
5483 : * Return last n characters in the string. When n is negative,
5484 : * return all but first |n| characters.
5485 : */
5486 : Datum
4611 itagaki.takahiro 5487 GIC 33 : text_right(PG_FUNCTION_ARGS)
5488 : {
5489 33 : text *str = PG_GETARG_TEXT_PP(0);
5490 33 : const char *p = VARDATA_ANY(str);
4611 itagaki.takahiro 5491 CBC 33 : int len = VARSIZE_ANY_EXHDR(str);
4611 itagaki.takahiro 5492 GIC 33 : int n = PG_GETARG_INT32(1);
5493 : int off;
4611 itagaki.takahiro 5494 ECB :
4611 itagaki.takahiro 5495 CBC 33 : if (n < 0)
4611 itagaki.takahiro 5496 GIC 15 : n = -n;
5497 : else
5498 18 : n = pg_mbstrlen_with_len(p, len) - n;
5499 33 : off = pg_mbcharcliplen(p, len, n);
5500 :
5501 33 : PG_RETURN_TEXT_P(cstring_to_text_with_len(p + off, len - off));
5502 : }
5503 :
5504 : /*
4611 itagaki.takahiro 5505 ECB : * Return reversed string
5506 : */
5507 : Datum
4611 itagaki.takahiro 5508 CBC 3 : text_reverse(PG_FUNCTION_ARGS)
5509 : {
4382 bruce 5510 GIC 3 : text *str = PG_GETARG_TEXT_PP(0);
4382 bruce 5511 CBC 3 : const char *p = VARDATA_ANY(str);
5512 3 : int len = VARSIZE_ANY_EXHDR(str);
4382 bruce 5513 GIC 3 : const char *endp = p + len;
5514 : text *result;
5515 : char *dst;
4611 itagaki.takahiro 5516 ECB :
4611 itagaki.takahiro 5517 GIC 3 : result = palloc(len + VARHDRSZ);
4382 bruce 5518 3 : dst = (char *) VARDATA(result) + len;
4611 itagaki.takahiro 5519 3 : SET_VARSIZE(result, len + VARHDRSZ);
4611 itagaki.takahiro 5520 ECB :
4611 itagaki.takahiro 5521 CBC 3 : if (pg_database_encoding_max_length() > 1)
5522 : {
5523 : /* multibyte version */
4611 itagaki.takahiro 5524 GIC 18 : while (p < endp)
5525 : {
4382 bruce 5526 ECB : int sz;
4611 itagaki.takahiro 5527 :
4611 itagaki.takahiro 5528 GIC 15 : sz = pg_mblen(p);
5529 15 : dst -= sz;
5530 15 : memcpy(dst, p, sz);
4611 itagaki.takahiro 5531 CBC 15 : p += sz;
4611 itagaki.takahiro 5532 ECB : }
5533 : }
5534 : else
5535 : {
5536 : /* single byte version */
4611 itagaki.takahiro 5537 UIC 0 : while (p < endp)
4611 itagaki.takahiro 5538 LBC 0 : *(--dst) = *p++;
5539 : }
5540 :
4611 itagaki.takahiro 5541 GIC 3 : PG_RETURN_TEXT_P(result);
5542 : }
5543 :
5544 :
5545 : /*
5546 : * Support macros for text_format()
5547 : */
5548 : #define TEXT_FORMAT_FLAG_MINUS 0x0001 /* is minus flag present? */
5549 :
5550 : #define ADVANCE_PARSE_POINTER(ptr,end_ptr) \
5551 : do { \
3678 tgl 5552 ECB : if (++(ptr) >= (end_ptr)) \
5553 : ereport(ERROR, \
5554 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE), \
2614 teodor 5555 : errmsg("unterminated format() type specifier"), \
5556 : errhint("For a single \"%%\" use \"%%%%\"."))); \
5557 : } while (0)
3678 tgl 5558 :
5559 : /*
5560 : * Returns a formatted string
4523 rhaas 5561 : */
5562 : Datum
4523 rhaas 5563 CBC 12534 : text_format(PG_FUNCTION_ARGS)
4523 rhaas 5564 ECB : {
5565 : text *fmt;
5566 : StringInfoData str;
5567 : const char *cp;
5568 : const char *start_ptr;
5569 : const char *end_ptr;
5570 : text *result;
5571 : int arg;
5572 : bool funcvariadic;
5573 : int nargs;
3726 tgl 5574 GIC 12534 : Datum *elements = NULL;
5575 12534 : bool *nulls = NULL;
5576 12534 : Oid element_type = InvalidOid;
5577 12534 : Oid prev_type = InvalidOid;
3678 5578 12534 : Oid prev_width_type = InvalidOid;
5579 : FmgrInfo typoutputfinfo;
3678 tgl 5580 ECB : FmgrInfo typoutputinfo_width;
4523 rhaas 5581 :
5582 : /* When format string is null, immediately return null */
4523 rhaas 5583 GIC 12534 : if (PG_ARGISNULL(0))
5584 3 : PG_RETURN_NULL();
5585 :
5586 : /* If argument is marked VARIADIC, expand array into elements */
3726 tgl 5587 12531 : if (get_fn_expr_variadic(fcinfo->flinfo))
3726 tgl 5588 ECB : {
5589 : ArrayType *arr;
5590 : int16 elmlen;
5591 : bool elmbyval;
5592 : char elmalign;
5593 : int nitems;
3726 tgl 5594 EUB :
5595 : /* Should have just the one argument */
3726 tgl 5596 GIC 24 : Assert(PG_NARGS() == 2);
5597 :
5598 : /* If argument is NULL, we treat it as zero-length array */
3726 tgl 5599 CBC 24 : if (PG_ARGISNULL(1))
3726 tgl 5600 GIC 3 : nitems = 0;
3726 tgl 5601 ECB : else
5602 : {
5603 : /*
5604 : * Non-null argument had better be an array. We assume that any
5605 : * call context that could let get_fn_expr_variadic return true
5606 : * will have checked that a VARIADIC-labeled parameter actually is
3293 tgl 5607 EUB : * an array. So it should be okay to just Assert that it's an
5608 : * array rather than doing a full-fledged error check.
3726 5609 : */
3293 tgl 5610 GIC 21 : Assert(OidIsValid(get_base_element_type(get_fn_expr_argtype(fcinfo->flinfo, 1))));
3726 tgl 5611 ECB :
3293 tgl 5612 EUB : /* OK, safe to fetch the array value */
3726 tgl 5613 GIC 21 : arr = PG_GETARG_ARRAYTYPE_P(1);
3726 tgl 5614 ECB :
5615 : /* Get info about array element type */
3726 tgl 5616 GIC 21 : element_type = ARR_ELEMTYPE(arr);
3726 tgl 5617 CBC 21 : get_typlenbyvalalign(element_type,
3726 tgl 5618 ECB : &elmlen, &elmbyval, &elmalign);
5619 :
5620 : /* Extract all array elements */
3726 tgl 5621 GBC 21 : deconstruct_array(arr, element_type, elmlen, elmbyval, elmalign,
3726 tgl 5622 EUB : &elements, &nulls, &nitems);
5623 : }
5624 :
3726 tgl 5625 GIC 24 : nargs = nitems + 1;
5626 24 : funcvariadic = true;
5627 : }
3726 tgl 5628 EUB : else
5629 : {
5630 : /* Non-variadic case, we'll process the arguments individually */
3726 tgl 5631 GIC 12507 : nargs = PG_NARGS();
5632 12507 : funcvariadic = false;
3726 tgl 5633 EUB : }
5634 :
4523 rhaas 5635 : /* Setup for main loop. */
4523 rhaas 5636 GIC 12531 : fmt = PG_GETARG_TEXT_PP(0);
5637 12531 : start_ptr = VARDATA_ANY(fmt);
4523 rhaas 5638 GBC 12531 : end_ptr = start_ptr + VARSIZE_ANY_EXHDR(fmt);
4523 rhaas 5639 GIC 12531 : initStringInfo(&str);
3678 tgl 5640 12531 : arg = 1; /* next argument position to print */
4523 rhaas 5641 EUB :
5642 : /* Scan format string, looking for conversion specifiers. */
4523 rhaas 5643 GBC 363849 : for (cp = start_ptr; cp < end_ptr; cp++)
5644 : {
5645 : int argpos;
5646 : int widthpos;
5647 : int flags;
3678 tgl 5648 ECB : int width;
4382 bruce 5649 : Datum value;
5650 : bool isNull;
5651 : Oid typid;
5652 :
5653 : /*
5654 : * If it's not the start of a conversion specifier, just copy it to
5655 : * the output buffer.
4523 rhaas 5656 : */
4523 rhaas 5657 GIC 351348 : if (*cp != '%')
4523 rhaas 5658 ECB : {
4523 rhaas 5659 CBC 321807 : appendStringInfoCharMacro(&str, *cp);
5660 321816 : continue;
5661 : }
5662 :
3678 tgl 5663 GIC 29541 : ADVANCE_PARSE_POINTER(cp, end_ptr);
4523 rhaas 5664 ECB :
5665 : /* Easy case: %% outputs a single % */
4523 rhaas 5666 CBC 29541 : if (*cp == '%')
5667 : {
5668 9 : appendStringInfoCharMacro(&str, *cp);
4523 rhaas 5669 GBC 9 : continue;
5670 : }
4523 rhaas 5671 ECB :
5672 : /* Parse the optional portions of the format specifier */
3678 tgl 5673 GIC 29532 : cp = text_format_parse_format(cp, end_ptr,
5674 : &argpos, &widthpos,
5675 : &flags, &width);
5676 :
5677 : /*
3678 tgl 5678 ECB : * Next we should see the main conversion specifier. Whether or not
5679 : * an argument position was present, it's known that at least one
5680 : * character remains in the string at this point. Experience suggests
5681 : * that it's worth checking that that character is one of the expected
5682 : * ones before we try to fetch arguments, so as to produce the least
5683 : * confusing response to a mis-formatted specifier.
4523 rhaas 5684 : */
3678 tgl 5685 CBC 29520 : if (strchr("sIL", *cp) == NULL)
3678 tgl 5686 GIC 3 : ereport(ERROR,
5687 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
5688 : errmsg("unrecognized format() type specifier \"%.*s\"",
5689 : pg_mblen(cp), cp),
5690 : errhint("For a single \"%%\" use \"%%%%\".")));
3678 tgl 5691 ECB :
5692 : /* If indirect width was specified, get its value */
3678 tgl 5693 CBC 29517 : if (widthpos >= 0)
5694 : {
5695 : /* Collect the specified or next argument position */
5696 21 : if (widthpos > 0)
3678 tgl 5697 GIC 18 : arg = widthpos;
5698 21 : if (arg >= nargs)
4339 heikki.linnakangas 5699 LBC 0 : ereport(ERROR,
3678 tgl 5700 EUB : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
5701 : errmsg("too few arguments for format()")));
4523 rhaas 5702 :
5703 : /* Get the value and type of the selected argument */
3678 tgl 5704 GIC 21 : if (!funcvariadic)
5705 : {
5706 21 : value = PG_GETARG_DATUM(arg);
5707 21 : isNull = PG_ARGISNULL(arg);
5708 21 : typid = get_fn_expr_argtype(fcinfo->flinfo, arg);
5709 : }
5710 : else
5711 : {
3678 tgl 5712 LBC 0 : value = elements[arg - 1];
5713 0 : isNull = nulls[arg - 1];
5714 0 : typid = element_type;
3678 tgl 5715 ECB : }
3678 tgl 5716 GIC 21 : if (!OidIsValid(typid))
3678 tgl 5717 UIC 0 : elog(ERROR, "could not determine data type of format() input");
4339 heikki.linnakangas 5718 ECB :
3678 tgl 5719 CBC 21 : arg++;
5720 :
3678 tgl 5721 ECB : /* We can treat NULL width the same as zero */
3678 tgl 5722 GIC 21 : if (isNull)
5723 3 : width = 0;
5724 18 : else if (typid == INT4OID)
5725 18 : width = DatumGetInt32(value);
3678 tgl 5726 UIC 0 : else if (typid == INT2OID)
5727 0 : width = DatumGetInt16(value);
5728 : else
5729 : {
5730 : /* For less-usual datatypes, convert to text then to int */
5731 : char *str;
5732 :
5733 0 : if (typid != prev_width_type)
5734 : {
3678 tgl 5735 ECB : Oid typoutputfunc;
5736 : bool typIsVarlena;
5737 :
3678 tgl 5738 LBC 0 : getTypeOutputInfo(typid, &typoutputfunc, &typIsVarlena);
5739 0 : fmgr_info(typoutputfunc, &typoutputinfo_width);
3678 tgl 5740 UIC 0 : prev_width_type = typid;
3678 tgl 5741 ECB : }
5742 :
3678 tgl 5743 LBC 0 : str = OutputFunctionCall(&typoutputinfo_width, value);
5744 :
1722 andres 5745 ECB : /* pg_strtoint32 will complain about bad data or overflow */
1722 andres 5746 LBC 0 : width = pg_strtoint32(str);
3678 tgl 5747 EUB :
3678 tgl 5748 UIC 0 : pfree(str);
5749 : }
4523 rhaas 5750 ECB : }
5751 :
5752 : /* Collect the specified or next argument position */
3678 tgl 5753 GIC 29517 : if (argpos > 0)
3678 tgl 5754 CBC 66 : arg = argpos;
5755 29517 : if (arg >= nargs)
4523 rhaas 5756 GIC 12 : ereport(ERROR,
4523 rhaas 5757 ECB : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
5758 : errmsg("too few arguments for format()")));
5759 :
5760 : /* Get the value and type of the selected argument */
3726 tgl 5761 GIC 29505 : if (!funcvariadic)
5762 : {
5763 28869 : value = PG_GETARG_DATUM(arg);
5764 28869 : isNull = PG_ARGISNULL(arg);
5765 28869 : typid = get_fn_expr_argtype(fcinfo->flinfo, arg);
5766 : }
5767 : else
5768 : {
5769 636 : value = elements[arg - 1];
5770 636 : isNull = nulls[arg - 1];
5771 636 : typid = element_type;
5772 : }
5773 29505 : if (!OidIsValid(typid))
3726 tgl 5774 UIC 0 : elog(ERROR, "could not determine data type of format() input");
5775 :
3678 tgl 5776 GIC 29505 : arg++;
5777 :
5778 : /*
5779 : * Get the appropriate typOutput function, reusing previous one if
5780 : * same type as previous argument. That's particularly useful in the
5781 : * variadic-array case, but often saves work even for ordinary calls.
5782 : */
3726 5783 29505 : if (typid != prev_type)
3726 tgl 5784 ECB : {
5785 : Oid typoutputfunc;
5786 : bool typIsVarlena;
5787 :
3726 tgl 5788 CBC 13980 : getTypeOutputInfo(typid, &typoutputfunc, &typIsVarlena);
3726 tgl 5789 GIC 13980 : fmgr_info(typoutputfunc, &typoutputfinfo);
5790 13980 : prev_type = typid;
5791 : }
3726 tgl 5792 ECB :
4523 rhaas 5793 : /*
3678 tgl 5794 : * And now we can format the value.
4523 rhaas 5795 : */
4523 rhaas 5796 GIC 29505 : switch (*cp)
5797 : {
4523 rhaas 5798 CBC 29505 : case 's':
5799 : case 'I':
4523 rhaas 5800 ECB : case 'L':
3726 tgl 5801 GIC 29505 : text_format_string_conversion(&str, *cp, &typoutputfinfo,
5802 : value, isNull,
3678 tgl 5803 ECB : flags, width);
4523 rhaas 5804 CBC 29502 : break;
4523 rhaas 5805 UIC 0 : default:
5806 : /* should not get here, because of previous check */
4523 rhaas 5807 LBC 0 : ereport(ERROR,
5808 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1014 tgl 5809 ECB : errmsg("unrecognized format() type specifier \"%.*s\"",
5810 : pg_mblen(cp), cp),
5811 : errhint("For a single \"%%\" use \"%%%%\".")));
5812 : break;
4523 rhaas 5813 : }
5814 : }
5815 :
5816 : /* Don't need deconstruct_array results anymore. */
3726 tgl 5817 CBC 12501 : if (elements != NULL)
3726 tgl 5818 GIC 21 : pfree(elements);
3726 tgl 5819 CBC 12501 : if (nulls != NULL)
5820 21 : pfree(nulls);
5821 :
5822 : /* Generate results. */
4523 rhaas 5823 12501 : result = cstring_to_text_with_len(str.data, str.len);
4523 rhaas 5824 GIC 12501 : pfree(str.data);
5825 :
4523 rhaas 5826 CBC 12501 : PG_RETURN_TEXT_P(result);
4523 rhaas 5827 ECB : }
5828 :
5829 : /*
3678 tgl 5830 : * Parse contiguous digits as a decimal number.
3678 tgl 5831 EUB : *
5832 : * Returns true if some digits could be parsed.
5833 : * The value is returned into *value, and *ptr is advanced to the next
5834 : * character to be parsed.
3678 tgl 5835 ECB : *
5836 : * Note parsing invariant: at least one character is known available before
5837 : * string end (end_ptr) at entry, and this is still true at exit.
5838 : */
5839 : static bool
3678 tgl 5840 GIC 59046 : text_format_parse_digits(const char **ptr, const char *end_ptr, int *value)
3678 tgl 5841 ECB : {
3678 tgl 5842 GIC 59046 : bool found = false;
5843 59046 : const char *cp = *ptr;
3678 tgl 5844 CBC 59046 : int val = 0;
5845 :
3678 tgl 5846 GIC 59202 : while (*cp >= '0' && *cp <= '9')
5847 : {
1944 andres 5848 159 : int8 digit = (*cp - '0');
3678 tgl 5849 ECB :
1944 andres 5850 CBC 159 : if (unlikely(pg_mul_s32_overflow(val, 10, &val)) ||
1944 andres 5851 GIC 159 : unlikely(pg_add_s32_overflow(val, digit, &val)))
3678 tgl 5852 UIC 0 : ereport(ERROR,
5853 : (errcode(ERRCODE_NUMERIC_VALUE_OUT_OF_RANGE),
3678 tgl 5854 ECB : errmsg("number is out of range")));
3678 tgl 5855 GIC 159 : ADVANCE_PARSE_POINTER(cp, end_ptr);
5856 156 : found = true;
5857 : }
5858 :
5859 59043 : *ptr = cp;
5860 59043 : *value = val;
3678 tgl 5861 ECB :
3678 tgl 5862 GIC 59043 : return found;
5863 : }
5864 :
5865 : /*
5866 : * Parse a format specifier (generally following the SUS printf spec).
5867 : *
5868 : * We have already advanced over the initial '%', and we are looking for
3678 tgl 5869 ECB : * [argpos][flags][width]type (but the type character is not consumed here).
5870 : *
5871 : * Inputs are start_ptr (the position after '%') and end_ptr (string end + 1).
5872 : * Output parameters:
3260 bruce 5873 : * argpos: argument position for value to be printed. -1 means unspecified.
5874 : * widthpos: argument position for width. Zero means the argument position
3678 tgl 5875 : * was unspecified (ie, take the next arg) and -1 means no width
5876 : * argument (width was omitted or specified as a constant).
5877 : * flags: bitmask of flags.
5878 : * width: directly-specified width value. Zero means the width was omitted
5879 : * (note it's not necessary to distinguish this case from an explicit
5880 : * zero width value).
5881 : *
5882 : * The function result is the next character position to be parsed, ie, the
5883 : * location where the type character is/should be.
5884 : *
5885 : * Note parsing invariant: at least one character is known available before
5886 : * string end (end_ptr) at entry, and this is still true at exit.
5887 : */
5888 : static const char *
3678 tgl 5889 CBC 29532 : text_format_parse_format(const char *start_ptr, const char *end_ptr,
5890 : int *argpos, int *widthpos,
3678 tgl 5891 ECB : int *flags, int *width)
5892 : {
3678 tgl 5893 CBC 29532 : const char *cp = start_ptr;
5894 : int n;
3678 tgl 5895 ECB :
5896 : /* set defaults for output parameters */
3678 tgl 5897 CBC 29532 : *argpos = -1;
3678 tgl 5898 GIC 29532 : *widthpos = -1;
5899 29532 : *flags = 0;
3678 tgl 5900 CBC 29532 : *width = 0;
5901 :
5902 : /* try to identify first number */
5903 29532 : if (text_format_parse_digits(&cp, end_ptr, &n))
5904 : {
3678 tgl 5905 GIC 87 : if (*cp != '$')
5906 : {
5907 : /* Must be just a width and a type, so we're done */
5908 12 : *width = n;
5909 12 : return cp;
3678 tgl 5910 ECB : }
5911 : /* The number was argument position */
3678 tgl 5912 GIC 75 : *argpos = n;
3678 tgl 5913 ECB : /* Explicit 0 for argument index is immediately refused */
3678 tgl 5914 GIC 75 : if (n == 0)
5915 3 : ereport(ERROR,
5916 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
3678 tgl 5917 ECB : errmsg("format specifies argument 0, but arguments are numbered from 1")));
3678 tgl 5918 GIC 72 : ADVANCE_PARSE_POINTER(cp, end_ptr);
3678 tgl 5919 ECB : }
5920 :
5921 : /* Handle flags (only minus is supported now) */
3678 tgl 5922 GIC 29529 : while (*cp == '-')
3678 tgl 5923 ECB : {
3678 tgl 5924 GIC 15 : *flags |= TEXT_FORMAT_FLAG_MINUS;
5925 15 : ADVANCE_PARSE_POINTER(cp, end_ptr);
3678 tgl 5926 ECB : }
5927 :
3678 tgl 5928 CBC 29514 : if (*cp == '*')
3678 tgl 5929 EUB : {
5930 : /* Handle indirect width */
3678 tgl 5931 GIC 24 : ADVANCE_PARSE_POINTER(cp, end_ptr);
3678 tgl 5932 CBC 24 : if (text_format_parse_digits(&cp, end_ptr, &n))
5933 : {
3678 tgl 5934 ECB : /* number in this position must be closed by $ */
3678 tgl 5935 CBC 21 : if (*cp != '$')
3678 tgl 5936 UIC 0 : ereport(ERROR,
3678 tgl 5937 ECB : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
2118 5938 : errmsg("width argument position must be ended by \"$\"")));
5939 : /* The number was width argument position */
3678 tgl 5940 GIC 21 : *widthpos = n;
3678 tgl 5941 ECB : /* Explicit 0 for argument index is immediately refused */
3678 tgl 5942 CBC 21 : if (n == 0)
5943 3 : ereport(ERROR,
5944 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
5945 : errmsg("format specifies argument 0, but arguments are numbered from 1")));
3678 tgl 5946 GIC 18 : ADVANCE_PARSE_POINTER(cp, end_ptr);
5947 : }
3678 tgl 5948 ECB : else
3678 tgl 5949 CBC 3 : *widthpos = 0; /* width's argument position is unspecified */
3678 tgl 5950 ECB : }
5951 : else
5952 : {
5953 : /* Check for direct width specification */
3678 tgl 5954 GIC 29490 : if (text_format_parse_digits(&cp, end_ptr, &n))
5955 15 : *width = n;
5956 : }
5957 :
5958 : /* cp should now be pointing at type character */
5959 29508 : return cp;
5960 : }
5961 :
3678 tgl 5962 ECB : /*
5963 : * Format a %s, %I, or %L conversion
5964 : */
5965 : static void
4523 rhaas 5966 GIC 29505 : text_format_string_conversion(StringInfo buf, char conversion,
5967 : FmgrInfo *typOutputInfo,
5968 : Datum value, bool isNull,
5969 : int flags, int width)
5970 : {
5971 : char *str;
4523 rhaas 5972 EUB :
5973 : /* Handle NULL arguments before trying to stringify the value. */
4523 rhaas 5974 GBC 29505 : if (isNull)
5975 : {
3678 tgl 5976 153 : if (conversion == 's')
5977 117 : text_format_append_string(buf, "", flags, width);
5978 36 : else if (conversion == 'L')
3678 tgl 5979 GIC 33 : text_format_append_string(buf, "NULL", flags, width);
4523 rhaas 5980 GBC 3 : else if (conversion == 'I')
4382 bruce 5981 GIC 3 : ereport(ERROR,
5982 : (errcode(ERRCODE_NULL_VALUE_NOT_ALLOWED),
5983 : errmsg("null values cannot be formatted as an SQL identifier")));
4523 rhaas 5984 150 : return;
5985 : }
5986 :
5987 : /* Stringify. */
3726 tgl 5988 29352 : str = OutputFunctionCall(typOutputInfo, value);
5989 :
5990 : /* Escape. */
4523 rhaas 5991 29352 : if (conversion == 'I')
5992 : {
5993 : /* quote_identifier may or may not allocate a new string. */
3678 tgl 5994 1553 : text_format_append_string(buf, quote_identifier(str), flags, width);
5995 : }
4523 rhaas 5996 27799 : else if (conversion == 'L')
5997 : {
4382 bruce 5998 1289 : char *qstr = quote_literal_cstr(str);
5999 :
3678 tgl 6000 1289 : text_format_append_string(buf, qstr, flags, width);
6001 : /* quote_literal_cstr() always allocates a new string */
4523 rhaas 6002 1289 : pfree(qstr);
6003 : }
6004 : else
3678 tgl 6005 26510 : text_format_append_string(buf, str, flags, width);
6006 :
6007 : /* Cleanup. */
4523 rhaas 6008 CBC 29352 : pfree(str);
6009 : }
4523 rhaas 6010 ECB :
3678 tgl 6011 : /*
6012 : * Append str to buf, padding as directed by flags/width
6013 : */
6014 : static void
3678 tgl 6015 CBC 29502 : text_format_append_string(StringInfo buf, const char *str,
3678 tgl 6016 ECB : int flags, int width)
6017 : {
3678 tgl 6018 GIC 29502 : bool align_to_left = false;
6019 : int len;
6020 :
6021 : /* fast path for typical easy case */
6022 29502 : if (width == 0)
6023 : {
6024 29460 : appendStringInfoString(buf, str);
6025 29460 : return;
6026 : }
6027 :
3678 tgl 6028 CBC 42 : if (width < 0)
6029 : {
6030 : /* Negative width: implicit '-' flag, then take absolute value */
3678 tgl 6031 GIC 3 : align_to_left = true;
3678 tgl 6032 ECB : /* -INT_MIN is undefined */
3678 tgl 6033 GIC 3 : if (width <= INT_MIN)
3678 tgl 6034 LBC 0 : ereport(ERROR,
3678 tgl 6035 ECB : (errcode(ERRCODE_NUMERIC_VALUE_OUT_OF_RANGE),
3678 tgl 6036 EUB : errmsg("number is out of range")));
3678 tgl 6037 GIC 3 : width = -width;
6038 : }
6039 39 : else if (flags & TEXT_FORMAT_FLAG_MINUS)
6040 12 : align_to_left = true;
6041 :
3678 tgl 6042 CBC 42 : len = pg_mbstrlen(str);
6043 42 : if (align_to_left)
3678 tgl 6044 EUB : {
6045 : /* left justify */
3678 tgl 6046 CBC 15 : appendStringInfoString(buf, str);
6047 15 : if (len < width)
3678 tgl 6048 GIC 15 : appendStringInfoSpaces(buf, width - len);
3678 tgl 6049 ECB : }
6050 : else
6051 : {
6052 : /* right justify */
3678 tgl 6053 CBC 27 : if (len < width)
6054 27 : appendStringInfoSpaces(buf, width - len);
3678 tgl 6055 GIC 27 : appendStringInfoString(buf, str);
6056 : }
6057 : }
6058 :
6059 : /*
6060 : * text_format_nv - nonvariadic wrapper for text_format function.
6061 : *
6062 : * note: this wrapper is necessary to pass the sanity check in opr_sanity,
4339 heikki.linnakangas 6063 ECB : * which checks that all built-in functions that share the implementing C
6064 : * function take the same number of arguments.
4523 rhaas 6065 : */
6066 : Datum
4523 rhaas 6067 CBC 15 : text_format_nv(PG_FUNCTION_ARGS)
6068 : {
4523 rhaas 6069 GIC 15 : return text_format(fcinfo);
6070 : }
6071 :
6072 : /*
6073 : * Helper function for Levenshtein distance functions. Faster than memcmp(),
6074 : * for this use case.
6075 : */
3069 rhaas 6076 ECB : static inline bool
3069 rhaas 6077 UIC 0 : rest_of_char_same(const char *s1, const char *s2, int len)
3069 rhaas 6078 ECB : {
3069 rhaas 6079 UIC 0 : while (len > 0)
6080 : {
6081 0 : len--;
6082 0 : if (s1[len] != s2[len])
3069 rhaas 6083 LBC 0 : return false;
3069 rhaas 6084 EUB : }
3069 rhaas 6085 UIC 0 : return true;
6086 : }
6087 :
3069 rhaas 6088 ECB : /* Expand each Levenshtein distance variant */
6089 : #include "levenshtein.c"
6090 : #define LEVENSHTEIN_LESS_EQUAL
6091 : #include "levenshtein.c"
1109 peter 6092 :
6093 :
6094 : /*
6095 : * The following *ClosestMatch() functions can be used to determine whether a
6096 : * user-provided string resembles any known valid values, which is useful for
6097 : * providing hints in log messages, among other things. Use these functions
6098 : * like so:
6099 : *
6100 : * initClosestMatch(&state, source_string, max_distance);
6101 : *
6102 : * for (int i = 0; i < num_valid_strings; i++)
6103 : * updateClosestMatch(&state, valid_strings[i]);
6104 : *
6105 : * closestMatch = getClosestMatch(&state);
6106 : */
6107 :
6108 : /*
6109 : * Initialize the given state with the source string and maximum Levenshtein
6110 : * distance to consider.
6111 : */
6112 : void
205 peter 6113 GNC 28 : initClosestMatch(ClosestMatchState *state, const char *source, int max_d)
6114 : {
6115 28 : Assert(state);
6116 28 : Assert(max_d >= 0);
6117 :
6118 28 : state->source = source;
6119 28 : state->min_d = -1;
6120 28 : state->max_d = max_d;
6121 28 : state->match = NULL;
6122 28 : }
6123 :
6124 : /*
6125 : * If the candidate string is a closer match than the current one saved (or
6126 : * there is no match saved), save it as the closest match.
6127 : *
6128 : * If the source or candidate string is NULL, empty, or too long, this function
6129 : * takes no action. Likewise, if the Levenshtein distance exceeds the maximum
6130 : * allowed or more than half the characters are different, no action is taken.
6131 : */
6132 : void
6133 159 : updateClosestMatch(ClosestMatchState *state, const char *candidate)
6134 : {
6135 : int dist;
6136 :
6137 159 : Assert(state);
6138 :
6139 159 : if (state->source == NULL || state->source[0] == '\0' ||
6140 159 : candidate == NULL || candidate[0] == '\0')
205 peter 6141 UNC 0 : return;
6142 :
6143 : /*
6144 : * To avoid ERROR-ing, we check the lengths here instead of setting
6145 : * 'trusted' to false in the call to varstr_levenshtein_less_equal().
6146 : */
205 peter 6147 GNC 159 : if (strlen(state->source) > MAX_LEVENSHTEIN_STRLEN ||
6148 159 : strlen(candidate) > MAX_LEVENSHTEIN_STRLEN)
205 peter 6149 UNC 0 : return;
6150 :
205 peter 6151 GNC 159 : dist = varstr_levenshtein_less_equal(state->source, strlen(state->source),
6152 159 : candidate, strlen(candidate), 1, 1, 1,
6153 : state->max_d, true);
6154 159 : if (dist <= state->max_d &&
6155 28 : dist <= strlen(state->source) / 2 &&
6156 7 : (state->min_d == -1 || dist < state->min_d))
6157 : {
6158 7 : state->min_d = dist;
6159 7 : state->match = candidate;
6160 : }
6161 : }
6162 :
6163 : /*
6164 : * Return the closest match. If no suitable candidates were provided via
6165 : * updateClosestMatch(), return NULL.
6166 : */
6167 : const char *
6168 28 : getClosestMatch(ClosestMatchState *state)
6169 : {
6170 28 : Assert(state);
6171 :
6172 28 : return state->match;
6173 : }
6174 :
6175 :
1109 peter 6176 ECB : /*
6177 : * Unicode support
6178 : */
6179 :
6180 : static UnicodeNormalizationForm
1109 peter 6181 GIC 93 : unicode_norm_form_from_string(const char *formstr)
6182 : {
1109 peter 6183 CBC 93 : UnicodeNormalizationForm form = -1;
6184 :
6185 : /*
6186 : * Might as well check this while we're here.
1109 peter 6187 ECB : */
1109 peter 6188 GIC 93 : if (GetDatabaseEncoding() != PG_UTF8)
1109 peter 6189 LBC 0 : ereport(ERROR,
1109 peter 6190 ECB : (errcode(ERRCODE_SYNTAX_ERROR),
6191 : errmsg("Unicode normalization can only be performed if server encoding is UTF8")));
6192 :
1109 peter 6193 GIC 93 : if (pg_strcasecmp(formstr, "NFC") == 0)
6194 33 : form = UNICODE_NFC;
6195 60 : else if (pg_strcasecmp(formstr, "NFD") == 0)
6196 18 : form = UNICODE_NFD;
6197 42 : else if (pg_strcasecmp(formstr, "NFKC") == 0)
6198 18 : form = UNICODE_NFKC;
1109 peter 6199 CBC 24 : else if (pg_strcasecmp(formstr, "NFKD") == 0)
1109 peter 6200 GIC 18 : form = UNICODE_NFKD;
6201 : else
1109 peter 6202 CBC 6 : ereport(ERROR,
1109 peter 6203 ECB : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
6204 : errmsg("invalid normalization form: %s", formstr)));
6205 :
1109 peter 6206 GIC 87 : return form;
1109 peter 6207 ECB : }
6208 :
6209 : Datum
1109 peter 6210 CBC 24 : unicode_normalize_func(PG_FUNCTION_ARGS)
1109 peter 6211 ECB : {
1109 peter 6212 GIC 24 : text *input = PG_GETARG_TEXT_PP(0);
6213 24 : char *formstr = text_to_cstring(PG_GETARG_TEXT_PP(1));
1109 peter 6214 ECB : UnicodeNormalizationForm form;
6215 : int size;
6216 : pg_wchar *input_chars;
6217 : pg_wchar *output_chars;
6218 : unsigned char *p;
6219 : text *result;
6220 : int i;
6221 :
1109 peter 6222 CBC 24 : form = unicode_norm_form_from_string(formstr);
1109 peter 6223 ECB :
6224 : /* convert to pg_wchar */
1109 peter 6225 GIC 21 : size = pg_mbstrlen_with_len(VARDATA_ANY(input), VARSIZE_ANY_EXHDR(input));
1109 peter 6226 CBC 21 : input_chars = palloc((size + 1) * sizeof(pg_wchar));
6227 21 : p = (unsigned char *) VARDATA_ANY(input);
1109 peter 6228 GIC 84 : for (i = 0; i < size; i++)
1109 peter 6229 ECB : {
1109 peter 6230 CBC 63 : input_chars[i] = utf8_to_unicode(p);
1109 peter 6231 GIC 63 : p += pg_utf_mblen(p);
1109 peter 6232 ECB : }
1109 peter 6233 CBC 21 : input_chars[i] = (pg_wchar) '\0';
1109 peter 6234 GIC 21 : Assert((char *) p == VARDATA_ANY(input) + VARSIZE_ANY_EXHDR(input));
1109 peter 6235 ECB :
6236 : /* action */
1109 peter 6237 CBC 21 : output_chars = unicode_normalize(form, input_chars);
6238 :
6239 : /* convert back to UTF-8 string */
1109 peter 6240 GIC 21 : size = 0;
6241 81 : for (pg_wchar *wp = output_chars; *wp; wp++)
6242 : {
6243 : unsigned char buf[4];
6244 :
6245 60 : unicode_to_utf8(*wp, buf);
6246 60 : size += pg_utf_mblen(buf);
6247 : }
6248 :
6249 21 : result = palloc(size + VARHDRSZ);
6250 21 : SET_VARSIZE(result, size + VARHDRSZ);
6251 :
6252 21 : p = (unsigned char *) VARDATA_ANY(result);
1109 peter 6253 CBC 81 : for (pg_wchar *wp = output_chars; *wp; wp++)
6254 : {
6255 60 : unicode_to_utf8(*wp, p);
6256 60 : p += pg_utf_mblen(p);
6257 : }
1109 peter 6258 GIC 21 : Assert((char *) p == (char *) result + size + VARHDRSZ);
6259 :
6260 21 : PG_RETURN_TEXT_P(result);
6261 : }
6262 :
6263 : /*
6264 : * Check whether the string is in the specified Unicode normalization form.
6265 : *
6266 : * This is done by converting the string to the specified normal form and then
1109 peter 6267 ECB : * comparing that to the original string. To speed that up, we also apply the
6268 : * "quick check" algorithm specified in UAX #15, which can give a yes or no
6269 : * answer for many strings by just scanning the string once.
6270 : *
6271 : * This function should generally be optimized for the case where the string
6272 : * is in fact normalized. In that case, we'll end up looking at the entire
6273 : * string, so it's probably not worth doing any incremental conversion etc.
6274 : */
6275 : Datum
1109 peter 6276 CBC 69 : unicode_is_normalized(PG_FUNCTION_ARGS)
6277 : {
6278 69 : text *input = PG_GETARG_TEXT_PP(0);
6279 69 : char *formstr = text_to_cstring(PG_GETARG_TEXT_PP(1));
6280 : UnicodeNormalizationForm form;
6281 : int size;
1109 peter 6282 ECB : pg_wchar *input_chars;
6283 : pg_wchar *output_chars;
6284 : unsigned char *p;
6285 : int i;
6286 : UnicodeNormalizationQC quickcheck;
6287 : int output_size;
6288 : bool result;
6289 :
1109 peter 6290 GIC 69 : form = unicode_norm_form_from_string(formstr);
1109 peter 6291 ECB :
6292 : /* convert to pg_wchar */
1109 peter 6293 CBC 66 : size = pg_mbstrlen_with_len(VARDATA_ANY(input), VARSIZE_ANY_EXHDR(input));
1109 peter 6294 GIC 66 : input_chars = palloc((size + 1) * sizeof(pg_wchar));
1109 peter 6295 CBC 66 : p = (unsigned char *) VARDATA_ANY(input);
6296 252 : for (i = 0; i < size; i++)
6297 : {
6298 186 : input_chars[i] = utf8_to_unicode(p);
1109 peter 6299 GIC 186 : p += pg_utf_mblen(p);
6300 : }
6301 66 : input_chars[i] = (pg_wchar) '\0';
6302 66 : Assert((char *) p == VARDATA_ANY(input) + VARSIZE_ANY_EXHDR(input));
6303 :
6304 : /* quick check (see UAX #15) */
1109 peter 6305 CBC 66 : quickcheck = unicode_is_normalized_quickcheck(form, input_chars);
1109 peter 6306 GIC 66 : if (quickcheck == UNICODE_NORM_QC_YES)
1109 peter 6307 CBC 21 : PG_RETURN_BOOL(true);
6308 45 : else if (quickcheck == UNICODE_NORM_QC_NO)
6309 6 : PG_RETURN_BOOL(false);
6310 :
1109 peter 6311 ECB : /* normalize and compare with original */
1109 peter 6312 GIC 39 : output_chars = unicode_normalize(form, input_chars);
6313 :
6314 39 : output_size = 0;
1109 peter 6315 CBC 162 : for (pg_wchar *wp = output_chars; *wp; wp++)
1109 peter 6316 GIC 123 : output_size++;
1109 peter 6317 ECB :
1109 peter 6318 CBC 57 : result = (size == output_size) &&
6319 18 : (memcmp(input_chars, output_chars, size * sizeof(pg_wchar)) == 0);
1109 peter 6320 ECB :
1109 peter 6321 CBC 39 : PG_RETURN_BOOL(result);
1109 peter 6322 ECB : }
742 peter 6323 EUB :
6324 : /*
6325 : * Check if first n chars are hexadecimal digits
6326 : */
6327 : static bool
742 peter 6328 GIC 78 : isxdigits_n(const char *instr, size_t n)
6329 : {
6330 330 : for (size_t i = 0; i < n; i++)
742 peter 6331 CBC 285 : if (!isxdigit((unsigned char) instr[i]))
742 peter 6332 GIC 33 : return false;
742 peter 6333 ECB :
742 peter 6334 GIC 45 : return true;
742 peter 6335 ECB : }
6336 :
6337 : static unsigned int
742 peter 6338 CBC 252 : hexval(unsigned char c)
6339 : {
742 peter 6340 GIC 252 : if (c >= '0' && c <= '9')
6341 192 : return c - '0';
6342 60 : if (c >= 'a' && c <= 'f')
6343 30 : return c - 'a' + 0xA;
6344 30 : if (c >= 'A' && c <= 'F')
742 peter 6345 CBC 30 : return c - 'A' + 0xA;
742 peter 6346 UIC 0 : elog(ERROR, "invalid hexadecimal digit");
742 peter 6347 ECB : return 0; /* not reached */
6348 : }
6349 :
6350 : /*
6351 : * Translate string with hexadecimal digits to number
6352 : */
6353 : static unsigned int
742 peter 6354 GIC 45 : hexval_n(const char *instr, size_t n)
742 peter 6355 ECB : {
742 peter 6356 CBC 45 : unsigned int result = 0;
6357 :
6358 297 : for (size_t i = 0; i < n; i++)
742 peter 6359 GIC 252 : result += hexval(instr[i]) << (4 * (n - i - 1));
742 peter 6360 ECB :
742 peter 6361 GIC 45 : return result;
742 peter 6362 ECB : }
6363 :
6364 : /*
6365 : * Replaces Unicode escape sequences by Unicode characters
6366 : */
6367 : Datum
742 peter 6368 GBC 33 : unistr(PG_FUNCTION_ARGS)
742 peter 6369 ECB : {
742 peter 6370 CBC 33 : text *input_text = PG_GETARG_TEXT_PP(0);
742 peter 6371 ECB : char *instr;
6372 : int len;
6373 : StringInfoData str;
6374 : text *result;
742 peter 6375 CBC 33 : pg_wchar pair_first = 0;
6376 : char cbuf[MAX_UNICODE_EQUIVALENT_STRING + 1];
742 peter 6377 ECB :
742 peter 6378 GIC 33 : instr = VARDATA_ANY(input_text);
742 peter 6379 CBC 33 : len = VARSIZE_ANY_EXHDR(input_text);
6380 :
6381 33 : initStringInfo(&str);
742 peter 6382 EUB :
742 peter 6383 GIC 255 : while (len > 0)
6384 : {
6385 243 : if (instr[0] == '\\')
742 peter 6386 ECB : {
742 peter 6387 GIC 51 : if (len >= 2 &&
742 peter 6388 CBC 51 : instr[1] == '\\')
6389 : {
742 peter 6390 GBC 3 : if (pair_first)
742 peter 6391 UBC 0 : goto invalid_pair;
742 peter 6392 GIC 3 : appendStringInfoChar(&str, '\\');
6393 3 : instr += 2;
742 peter 6394 CBC 3 : len -= 2;
6395 : }
6396 48 : else if ((len >= 5 && isxdigits_n(instr + 1, 4)) ||
742 peter 6397 GBC 33 : (len >= 6 && instr[1] == 'u' && isxdigits_n(instr + 2, 4)))
742 peter 6398 GIC 15 : {
742 peter 6399 ECB : pg_wchar unicode;
742 peter 6400 CBC 21 : int offset = instr[1] == 'u' ? 2 : 1;
6401 :
742 peter 6402 GIC 21 : unicode = hexval_n(instr + offset, 4);
742 peter 6403 ECB :
742 peter 6404 CBC 21 : if (!is_valid_unicode_codepoint(unicode))
742 peter 6405 UIC 0 : ereport(ERROR,
6406 : errcode(ERRCODE_INVALID_PARAMETER_VALUE),
742 peter 6407 ECB : errmsg("invalid Unicode code point: %04X", unicode));
6408 :
742 peter 6409 GIC 21 : if (pair_first)
742 peter 6410 ECB : {
742 peter 6411 CBC 6 : if (is_utf16_surrogate_second(unicode))
6412 : {
742 peter 6413 UIC 0 : unicode = surrogate_pair_to_codepoint(pair_first, unicode);
742 peter 6414 LBC 0 : pair_first = 0;
6415 : }
742 peter 6416 ECB : else
742 peter 6417 CBC 6 : goto invalid_pair;
6418 : }
742 peter 6419 GIC 15 : else if (is_utf16_surrogate_second(unicode))
742 peter 6420 UIC 0 : goto invalid_pair;
742 peter 6421 ECB :
742 peter 6422 GIC 15 : if (is_utf16_surrogate_first(unicode))
742 peter 6423 CBC 9 : pair_first = unicode;
6424 : else
742 peter 6425 EUB : {
742 peter 6426 GBC 6 : pg_unicode_to_server(unicode, (unsigned char *) cbuf);
742 peter 6427 GIC 6 : appendStringInfoString(&str, cbuf);
6428 : }
742 peter 6429 ECB :
742 peter 6430 GIC 15 : instr += 4 + offset;
742 peter 6431 CBC 15 : len -= 4 + offset;
742 peter 6432 EUB : }
742 peter 6433 GIC 27 : else if (len >= 8 && instr[1] == '+' && isxdigits_n(instr + 2, 6))
742 peter 6434 CBC 6 : {
742 peter 6435 ECB : pg_wchar unicode;
6436 :
742 peter 6437 GIC 12 : unicode = hexval_n(instr + 2, 6);
742 peter 6438 ECB :
742 peter 6439 CBC 12 : if (!is_valid_unicode_codepoint(unicode))
742 peter 6440 GIC 3 : ereport(ERROR,
6441 : errcode(ERRCODE_INVALID_PARAMETER_VALUE),
742 peter 6442 ECB : errmsg("invalid Unicode code point: %04X", unicode));
6443 :
742 peter 6444 GIC 9 : if (pair_first)
742 peter 6445 ECB : {
742 peter 6446 CBC 3 : if (is_utf16_surrogate_second(unicode))
6447 : {
742 peter 6448 UIC 0 : unicode = surrogate_pair_to_codepoint(pair_first, unicode);
742 peter 6449 LBC 0 : pair_first = 0;
6450 : }
742 peter 6451 ECB : else
742 peter 6452 CBC 3 : goto invalid_pair;
6453 : }
742 peter 6454 GIC 6 : else if (is_utf16_surrogate_second(unicode))
742 peter 6455 UIC 0 : goto invalid_pair;
742 peter 6456 ECB :
742 peter 6457 GIC 6 : if (is_utf16_surrogate_first(unicode))
742 peter 6458 CBC 3 : pair_first = unicode;
6459 : else
742 peter 6460 EUB : {
742 peter 6461 GBC 3 : pg_unicode_to_server(unicode, (unsigned char *) cbuf);
742 peter 6462 GIC 3 : appendStringInfoString(&str, cbuf);
6463 : }
742 peter 6464 ECB :
742 peter 6465 GIC 6 : instr += 8;
742 peter 6466 CBC 6 : len -= 8;
742 peter 6467 EUB : }
742 peter 6468 GIC 15 : else if (len >= 10 && instr[1] == 'U' && isxdigits_n(instr + 2, 8))
742 peter 6469 CBC 6 : {
742 peter 6470 ECB : pg_wchar unicode;
6471 :
742 peter 6472 GIC 12 : unicode = hexval_n(instr + 2, 8);
742 peter 6473 ECB :
742 peter 6474 CBC 12 : if (!is_valid_unicode_codepoint(unicode))
742 peter 6475 GIC 3 : ereport(ERROR,
6476 : errcode(ERRCODE_INVALID_PARAMETER_VALUE),
742 peter 6477 ECB : errmsg("invalid Unicode code point: %04X", unicode));
6478 :
742 peter 6479 GIC 9 : if (pair_first)
6480 : {
742 peter 6481 CBC 3 : if (is_utf16_surrogate_second(unicode))
6482 : {
742 peter 6483 UIC 0 : unicode = surrogate_pair_to_codepoint(pair_first, unicode);
6484 0 : pair_first = 0;
6485 : }
6486 : else
742 peter 6487 GIC 3 : goto invalid_pair;
742 peter 6488 ECB : }
742 peter 6489 GBC 6 : else if (is_utf16_surrogate_second(unicode))
742 peter 6490 UIC 0 : goto invalid_pair;
742 peter 6491 ECB :
742 peter 6492 CBC 6 : if (is_utf16_surrogate_first(unicode))
742 peter 6493 GIC 3 : pair_first = unicode;
6494 : else
6495 : {
6496 3 : pg_unicode_to_server(unicode, (unsigned char *) cbuf);
742 peter 6497 CBC 3 : appendStringInfoString(&str, cbuf);
742 peter 6498 ECB : }
6499 :
742 peter 6500 CBC 6 : instr += 10;
6501 6 : len -= 10;
6502 : }
742 peter 6503 ECB : else
742 peter 6504 GIC 3 : ereport(ERROR,
742 peter 6505 ECB : (errcode(ERRCODE_SYNTAX_ERROR),
6506 : errmsg("invalid Unicode escape"),
6507 : errhint("Unicode escapes must be \\XXXX, \\+XXXXXX, \\uXXXX, or \\UXXXXXXXX.")));
6508 : }
6509 : else
6510 : {
742 peter 6511 GIC 192 : if (pair_first)
742 peter 6512 UIC 0 : goto invalid_pair;
6513 :
742 peter 6514 GIC 192 : appendStringInfoChar(&str, *instr++);
6515 192 : len--;
6516 : }
6517 : }
6518 :
6519 : /* unfinished surrogate pair? */
6520 12 : if (pair_first)
6521 3 : goto invalid_pair;
6522 :
6523 9 : result = cstring_to_text_with_len(str.data, str.len);
6524 9 : pfree(str.data);
6525 :
6526 9 : PG_RETURN_TEXT_P(result);
6527 :
6528 15 : invalid_pair:
6529 15 : ereport(ERROR,
6530 : (errcode(ERRCODE_SYNTAX_ERROR),
6531 : errmsg("invalid Unicode surrogate pair")));
6532 : PG_RETURN_NULL(); /* keep compiler quiet */
6533 : }
|
