1 /*-------------------------------------------------------------------------
4 * Functions for the variable-length built-in types.
6 * Portions Copyright (c) 1996-2022, PostgreSQL Global Development Group
7 * Portions Copyright (c) 1994, Regents of the University of California
11 * src/backend/utils/adt/varlena.c
13 *-------------------------------------------------------------------------
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"
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"
46 int bytea_output
= BYTEA_OUTPUT_HEX
;
48 typedef struct varlena unknown
;
49 typedef struct varlena VarString
;
52 * State for text_position_* functions.
56 bool is_multibyte_char_in_char
; /* need to check char boundaries? */
58 char *str1
; /* haystack string */
59 char *str2
; /* needle string */
60 int len1
; /* string lengths in bytes */
63 /* Skip table for Boyer-Moore-Horspool search algorithm: */
64 int skiptablemask
; /* mask for ANDing with skiptable subscripts */
65 int skiptable
[256]; /* skip distance for given mismatched char */
67 char *last_match
; /* pointer to last match in 'str1' */
70 * Sometimes we need to convert the byte position of a match to a
71 * character position. These store the last position that was converted,
72 * so that on the next call, we can continue from that point, rather than
73 * count characters from the very beginning.
75 char *refpoint
; /* pointer within original haystack string */
76 int refpos
; /* 0-based character offset of the same point */
81 char *buf1
; /* 1st string, or abbreviation original string
83 char *buf2
; /* 2nd string, or abbreviation strxfrm() buf */
84 int buflen1
; /* Allocated length of buf1 */
85 int buflen2
; /* Allocated length of buf2 */
86 int last_len1
; /* Length of last buf1 string/strxfrm() input */
87 int last_len2
; /* Length of last buf2 string/strxfrm() blob */
88 int last_returned
; /* Last comparison result (cache) */
89 bool cache_blob
; /* Does buf2 contain strxfrm() blob, etc? */
91 Oid typid
; /* Actual datatype (text/bpchar/bytea/name) */
92 hyperLogLogState abbr_card
; /* Abbreviated key cardinality state */
93 hyperLogLogState full_card
; /* Full key cardinality state */
94 double prop_card
; /* Required cardinality proportion */
96 } VarStringSortSupport
;
99 * Output data for split_text(): we output either to an array or a table.
100 * tupstore and tupdesc must be set up in advance to output to a table.
104 ArrayBuildState
*astate
;
105 Tuplestorestate
*tupstore
;
107 } SplitTextOutputData
;
110 * This should be large enough that most strings will fit, but small enough
111 * that we feel comfortable putting it on the stack
113 #define TEXTBUFLEN 1024
115 #define DatumGetUnknownP(X) ((unknown *) PG_DETOAST_DATUM(X))
116 #define DatumGetUnknownPCopy(X) ((unknown *) PG_DETOAST_DATUM_COPY(X))
117 #define PG_GETARG_UNKNOWN_P(n) DatumGetUnknownP(PG_GETARG_DATUM(n))
118 #define PG_GETARG_UNKNOWN_P_COPY(n) DatumGetUnknownPCopy(PG_GETARG_DATUM(n))
119 #define PG_RETURN_UNKNOWN_P(x) PG_RETURN_POINTER(x)
121 #define DatumGetVarStringP(X) ((VarString *) PG_DETOAST_DATUM(X))
122 #define DatumGetVarStringPP(X) ((VarString *) PG_DETOAST_DATUM_PACKED(X))
124 static int varstrfastcmp_c(Datum x
, Datum y
, SortSupport ssup
);
125 static int bpcharfastcmp_c(Datum x
, Datum y
, SortSupport ssup
);
126 static int namefastcmp_c(Datum x
, Datum y
, SortSupport ssup
);
127 static int varlenafastcmp_locale(Datum x
, Datum y
, SortSupport ssup
);
128 static int namefastcmp_locale(Datum x
, Datum y
, SortSupport ssup
);
129 static int varstrfastcmp_locale(char *a1p
, int len1
, char *a2p
, int len2
, SortSupport ssup
);
130 static Datum
varstr_abbrev_convert(Datum original
, SortSupport ssup
);
131 static bool varstr_abbrev_abort(int memtupcount
, SortSupport ssup
);
132 static int32
text_length(Datum str
);
133 static text
*text_catenate(text
*t1
, text
*t2
);
134 static text
*text_substring(Datum str
,
137 bool length_not_specified
);
138 static text
*text_overlay(text
*t1
, text
*t2
, int sp
, int sl
);
139 static int text_position(text
*t1
, text
*t2
, Oid collid
);
140 static void text_position_setup(text
*t1
, text
*t2
, Oid collid
, TextPositionState
*state
);
141 static bool text_position_next(TextPositionState
*state
);
142 static char *text_position_next_internal(char *start_ptr
, TextPositionState
*state
);
143 static char *text_position_get_match_ptr(TextPositionState
*state
);
144 static int text_position_get_match_pos(TextPositionState
*state
);
145 static void text_position_cleanup(TextPositionState
*state
);
146 static void check_collation_set(Oid collid
);
147 static int text_cmp(text
*arg1
, text
*arg2
, Oid collid
);
148 static bytea
*bytea_catenate(bytea
*t1
, bytea
*t2
);
149 static bytea
*bytea_substring(Datum str
,
152 bool length_not_specified
);
153 static bytea
*bytea_overlay(bytea
*t1
, bytea
*t2
, int sp
, int sl
);
154 static void appendStringInfoText(StringInfo str
, const text
*t
);
155 static bool split_text(FunctionCallInfo fcinfo
, SplitTextOutputData
*tstate
);
156 static void split_text_accum_result(SplitTextOutputData
*tstate
,
160 static text
*array_to_text_internal(FunctionCallInfo fcinfo
, ArrayType
*v
,
161 const char *fldsep
, const char *null_string
);
162 static StringInfo
makeStringAggState(FunctionCallInfo fcinfo
);
163 static bool text_format_parse_digits(const char **ptr
, const char *end_ptr
,
165 static const char *text_format_parse_format(const char *start_ptr
,
167 int *argpos
, int *widthpos
,
168 int *flags
, int *width
);
169 static void text_format_string_conversion(StringInfo buf
, char conversion
,
170 FmgrInfo
*typOutputInfo
,
171 Datum value
, bool isNull
,
172 int flags
, int width
);
173 static void text_format_append_string(StringInfo buf
, const char *str
,
174 int flags
, int width
);
177 /*****************************************************************************
178 * CONVERSION ROUTINES EXPORTED FOR USE BY C CODE *
179 *****************************************************************************/
184 * Create a text value from a null-terminated C string.
186 * The new text value is freshly palloc'd with a full-size VARHDR.
189 cstring_to_text(const char *s
)
191 return cstring_to_text_with_len(s
, strlen(s
));
195 * cstring_to_text_with_len
197 * Same as cstring_to_text except the caller specifies the string length;
198 * the string need not be null_terminated.
201 cstring_to_text_with_len(const char *s
, int len
)
203 text
*result
= (text
*) palloc(len
+ VARHDRSZ
);
205 SET_VARSIZE(result
, len
+ VARHDRSZ
);
206 memcpy(VARDATA(result
), s
, len
);
214 * Create a palloc'd, null-terminated C string from a text value.
216 * We support being passed a compressed or toasted text value.
217 * This is a bit bogus since such values shouldn't really be referred to as
218 * "text *", but it seems useful for robustness. If we didn't handle that
219 * case here, we'd need another routine that did, anyway.
222 text_to_cstring(const text
*t
)
224 /* must cast away the const, unfortunately */
225 text
*tunpacked
= pg_detoast_datum_packed(unconstify(text
*, t
));
226 int len
= VARSIZE_ANY_EXHDR(tunpacked
);
229 result
= (char *) palloc(len
+ 1);
230 memcpy(result
, VARDATA_ANY(tunpacked
), len
);
240 * text_to_cstring_buffer
242 * Copy a text value into a caller-supplied buffer of size dst_len.
244 * The text string is truncated if necessary to fit. The result is
245 * guaranteed null-terminated (unless dst_len == 0).
247 * We support being passed a compressed or toasted text value.
248 * This is a bit bogus since such values shouldn't really be referred to as
249 * "text *", but it seems useful for robustness. If we didn't handle that
250 * case here, we'd need another routine that did, anyway.
253 text_to_cstring_buffer(const text
*src
, char *dst
, size_t dst_len
)
255 /* must cast away the const, unfortunately */
256 text
*srcunpacked
= pg_detoast_datum_packed(unconstify(text
*, src
));
257 size_t src_len
= VARSIZE_ANY_EXHDR(srcunpacked
);
262 if (dst_len
>= src_len
)
264 else /* ensure truncation is encoding-safe */
265 dst_len
= pg_mbcliplen(VARDATA_ANY(srcunpacked
), src_len
, dst_len
);
266 memcpy(dst
, VARDATA_ANY(srcunpacked
), dst_len
);
270 if (srcunpacked
!= src
)
275 /*****************************************************************************
276 * USER I/O ROUTINES *
277 *****************************************************************************/
280 #define VAL(CH) ((CH) - '0')
281 #define DIG(VAL) ((VAL) + '0')
284 * byteain - converts from printable representation of byte array
286 * Non-printable characters must be passed as '\nnn' (octal) and are
287 * converted to internal form. '\' must be passed as '\\'.
288 * ereport(ERROR, ...) if bad form.
291 * The input is scanned twice.
292 * The error checking of input is minimal.
295 byteain(PG_FUNCTION_ARGS
)
297 char *inputText
= PG_GETARG_CSTRING(0);
298 Node
*escontext
= fcinfo
->context
;
304 /* Recognize hex input */
305 if (inputText
[0] == '\\' && inputText
[1] == 'x')
307 size_t len
= strlen(inputText
);
309 bc
= (len
- 2) / 2 + VARHDRSZ
; /* maximum possible length */
311 bc
= hex_decode_safe(inputText
+ 2, len
- 2, VARDATA(result
),
313 SET_VARSIZE(result
, bc
+ VARHDRSZ
); /* actual length */
315 PG_RETURN_BYTEA_P(result
);
318 /* Else, it's the traditional escaped style */
319 for (bc
= 0, tp
= inputText
; *tp
!= '\0'; bc
++)
323 else if ((tp
[0] == '\\') &&
324 (tp
[1] >= '0' && tp
[1] <= '3') &&
325 (tp
[2] >= '0' && tp
[2] <= '7') &&
326 (tp
[3] >= '0' && tp
[3] <= '7'))
328 else if ((tp
[0] == '\\') &&
334 * one backslash, not followed by another or ### valid octal
336 ereturn(escontext
, (Datum
) 0,
337 (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION
),
338 errmsg("invalid input syntax for type %s", "bytea")));
344 result
= (bytea
*) palloc(bc
);
345 SET_VARSIZE(result
, bc
);
348 rp
= VARDATA(result
);
353 else if ((tp
[0] == '\\') &&
354 (tp
[1] >= '0' && tp
[1] <= '3') &&
355 (tp
[2] >= '0' && tp
[2] <= '7') &&
356 (tp
[3] >= '0' && tp
[3] <= '7'))
362 *rp
++ = bc
+ VAL(tp
[3]);
366 else if ((tp
[0] == '\\') &&
375 * We should never get here. The first pass should not allow it.
377 ereturn(escontext
, (Datum
) 0,
378 (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION
),
379 errmsg("invalid input syntax for type %s", "bytea")));
383 PG_RETURN_BYTEA_P(result
);
387 * byteaout - converts to printable representation of byte array
389 * In the traditional escaped format, non-printable characters are
390 * printed as '\nnn' (octal) and '\' as '\\'.
393 byteaout(PG_FUNCTION_ARGS
)
395 bytea
*vlena
= PG_GETARG_BYTEA_PP(0);
399 if (bytea_output
== BYTEA_OUTPUT_HEX
)
401 /* Print hex format */
402 rp
= result
= palloc(VARSIZE_ANY_EXHDR(vlena
) * 2 + 2 + 1);
405 rp
+= hex_encode(VARDATA_ANY(vlena
), VARSIZE_ANY_EXHDR(vlena
), rp
);
407 else if (bytea_output
== BYTEA_OUTPUT_ESCAPE
)
409 /* Print traditional escaped format */
414 len
= 1; /* empty string has 1 char */
415 vp
= VARDATA_ANY(vlena
);
416 for (i
= VARSIZE_ANY_EXHDR(vlena
); i
!= 0; i
--, vp
++)
420 else if ((unsigned char) *vp
< 0x20 || (unsigned char) *vp
> 0x7e)
427 * In principle len can't overflow uint32 if the input fit in 1GB, but
428 * for safety let's check rather than relying on palloc's internal
431 if (len
> MaxAllocSize
)
433 (errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED
),
434 errmsg_internal("result of bytea output conversion is too large")));
435 rp
= result
= (char *) palloc(len
);
437 vp
= VARDATA_ANY(vlena
);
438 for (i
= VARSIZE_ANY_EXHDR(vlena
); i
!= 0; i
--, vp
++)
445 else if ((unsigned char) *vp
< 0x20 || (unsigned char) *vp
> 0x7e)
447 int val
; /* holds unprintable chars */
451 rp
[3] = DIG(val
& 07);
453 rp
[2] = DIG(val
& 07);
455 rp
[1] = DIG(val
& 03);
464 elog(ERROR
, "unrecognized bytea_output setting: %d",
466 rp
= result
= NULL
; /* keep compiler quiet */
469 PG_RETURN_CSTRING(result
);
473 * bytearecv - converts external binary format to bytea
476 bytearecv(PG_FUNCTION_ARGS
)
478 StringInfo buf
= (StringInfo
) PG_GETARG_POINTER(0);
482 nbytes
= buf
->len
- buf
->cursor
;
483 result
= (bytea
*) palloc(nbytes
+ VARHDRSZ
);
484 SET_VARSIZE(result
, nbytes
+ VARHDRSZ
);
485 pq_copymsgbytes(buf
, VARDATA(result
), nbytes
);
486 PG_RETURN_BYTEA_P(result
);
490 * byteasend - converts bytea to binary format
492 * This is a special case: just copy the input...
495 byteasend(PG_FUNCTION_ARGS
)
497 bytea
*vlena
= PG_GETARG_BYTEA_P_COPY(0);
499 PG_RETURN_BYTEA_P(vlena
);
503 bytea_string_agg_transfn(PG_FUNCTION_ARGS
)
507 state
= PG_ARGISNULL(0) ? NULL
: (StringInfo
) PG_GETARG_POINTER(0);
509 /* Append the value unless null. */
510 if (!PG_ARGISNULL(1))
512 bytea
*value
= PG_GETARG_BYTEA_PP(1);
514 /* On the first time through, we ignore the delimiter. */
516 state
= makeStringAggState(fcinfo
);
517 else if (!PG_ARGISNULL(2))
519 bytea
*delim
= PG_GETARG_BYTEA_PP(2);
521 appendBinaryStringInfo(state
, VARDATA_ANY(delim
), VARSIZE_ANY_EXHDR(delim
));
524 appendBinaryStringInfo(state
, VARDATA_ANY(value
), VARSIZE_ANY_EXHDR(value
));
528 * The transition type for string_agg() is declared to be "internal",
529 * which is a pass-by-value type the same size as a pointer.
531 PG_RETURN_POINTER(state
);
535 bytea_string_agg_finalfn(PG_FUNCTION_ARGS
)
539 /* cannot be called directly because of internal-type argument */
540 Assert(AggCheckCallContext(fcinfo
, NULL
));
542 state
= PG_ARGISNULL(0) ? NULL
: (StringInfo
) PG_GETARG_POINTER(0);
548 result
= (bytea
*) palloc(state
->len
+ VARHDRSZ
);
549 SET_VARSIZE(result
, state
->len
+ VARHDRSZ
);
550 memcpy(VARDATA(result
), state
->data
, state
->len
);
551 PG_RETURN_BYTEA_P(result
);
558 * textin - converts "..." to internal representation
561 textin(PG_FUNCTION_ARGS
)
563 char *inputText
= PG_GETARG_CSTRING(0);
565 PG_RETURN_TEXT_P(cstring_to_text(inputText
));
569 * textout - converts internal representation to "..."
572 textout(PG_FUNCTION_ARGS
)
574 Datum txt
= PG_GETARG_DATUM(0);
576 PG_RETURN_CSTRING(TextDatumGetCString(txt
));
580 * textrecv - converts external binary format to text
583 textrecv(PG_FUNCTION_ARGS
)
585 StringInfo buf
= (StringInfo
) PG_GETARG_POINTER(0);
590 str
= pq_getmsgtext(buf
, buf
->len
- buf
->cursor
, &nbytes
);
592 result
= cstring_to_text_with_len(str
, nbytes
);
594 PG_RETURN_TEXT_P(result
);
598 * textsend - converts text to binary format
601 textsend(PG_FUNCTION_ARGS
)
603 text
*t
= PG_GETARG_TEXT_PP(0);
606 pq_begintypsend(&buf
);
607 pq_sendtext(&buf
, VARDATA_ANY(t
), VARSIZE_ANY_EXHDR(t
));
608 PG_RETURN_BYTEA_P(pq_endtypsend(&buf
));
613 * unknownin - converts "..." to internal representation
616 unknownin(PG_FUNCTION_ARGS
)
618 char *str
= PG_GETARG_CSTRING(0);
620 /* representation is same as cstring */
621 PG_RETURN_CSTRING(pstrdup(str
));
625 * unknownout - converts internal representation to "..."
628 unknownout(PG_FUNCTION_ARGS
)
630 /* representation is same as cstring */
631 char *str
= PG_GETARG_CSTRING(0);
633 PG_RETURN_CSTRING(pstrdup(str
));
637 * unknownrecv - converts external binary format to unknown
640 unknownrecv(PG_FUNCTION_ARGS
)
642 StringInfo buf
= (StringInfo
) PG_GETARG_POINTER(0);
646 str
= pq_getmsgtext(buf
, buf
->len
- buf
->cursor
, &nbytes
);
647 /* representation is same as cstring */
648 PG_RETURN_CSTRING(str
);
652 * unknownsend - converts unknown to binary format
655 unknownsend(PG_FUNCTION_ARGS
)
657 /* representation is same as cstring */
658 char *str
= PG_GETARG_CSTRING(0);
661 pq_begintypsend(&buf
);
662 pq_sendtext(&buf
, str
, strlen(str
));
663 PG_RETURN_BYTEA_P(pq_endtypsend(&buf
));
667 /* ========== PUBLIC ROUTINES ========== */
671 * returns the logical length of a text*
672 * (which is less than the VARSIZE of the text*)
675 textlen(PG_FUNCTION_ARGS
)
677 Datum str
= PG_GETARG_DATUM(0);
679 /* try to avoid decompressing argument */
680 PG_RETURN_INT32(text_length(str
));
685 * Does the real work for textlen()
687 * This is broken out so it can be called directly by other string processing
688 * functions. Note that the argument is passed as a Datum, to indicate that
689 * it may still be in compressed form. We can avoid decompressing it at all
693 text_length(Datum str
)
695 /* fastpath when max encoding length is one */
696 if (pg_database_encoding_max_length() == 1)
697 PG_RETURN_INT32(toast_raw_datum_size(str
) - VARHDRSZ
);
700 text
*t
= DatumGetTextPP(str
);
702 PG_RETURN_INT32(pg_mbstrlen_with_len(VARDATA_ANY(t
),
703 VARSIZE_ANY_EXHDR(t
)));
709 * returns the physical length of a text*
710 * (which is less than the VARSIZE of the text*)
713 textoctetlen(PG_FUNCTION_ARGS
)
715 Datum str
= PG_GETARG_DATUM(0);
717 /* We need not detoast the input at all */
718 PG_RETURN_INT32(toast_raw_datum_size(str
) - VARHDRSZ
);
723 * takes two text* and returns a text* that is the concatenation of
726 * Rewritten by Sapa, sapa@hq.icb.chel.su. 8-Jul-96.
727 * Updated by Thomas, Thomas.Lockhart@jpl.nasa.gov 1997-07-10.
728 * Allocate space for output in all cases.
729 * XXX - thomas 1997-07-10
732 textcat(PG_FUNCTION_ARGS
)
734 text
*t1
= PG_GETARG_TEXT_PP(0);
735 text
*t2
= PG_GETARG_TEXT_PP(1);
737 PG_RETURN_TEXT_P(text_catenate(t1
, t2
));
742 * Guts of textcat(), broken out so it can be used by other functions
744 * Arguments can be in short-header form, but not compressed or out-of-line
747 text_catenate(text
*t1
, text
*t2
)
755 len1
= VARSIZE_ANY_EXHDR(t1
);
756 len2
= VARSIZE_ANY_EXHDR(t2
);
758 /* paranoia ... probably should throw error instead? */
764 len
= len1
+ len2
+ VARHDRSZ
;
765 result
= (text
*) palloc(len
);
767 /* Set size of result string... */
768 SET_VARSIZE(result
, len
);
770 /* Fill data field of result string... */
771 ptr
= VARDATA(result
);
773 memcpy(ptr
, VARDATA_ANY(t1
), len1
);
775 memcpy(ptr
+ len1
, VARDATA_ANY(t2
), len2
);
781 * charlen_to_bytelen()
782 * Compute the number of bytes occupied by n characters starting at *p
784 * It is caller's responsibility that there actually are n characters;
785 * the string need not be null-terminated.
788 charlen_to_bytelen(const char *p
, int n
)
790 if (pg_database_encoding_max_length() == 1)
792 /* Optimization for single-byte encodings */
799 for (s
= p
; n
> 0; n
--)
808 * Return a substring starting at the specified position.
809 * - thomas 1997-12-31
813 * - starting position (is one-based)
816 * If the starting position is zero or less, then return from the start of the string
817 * adjusting the length to be consistent with the "negative start" per SQL.
818 * If the length is less than zero, return the remaining string.
820 * Added multibyte support.
821 * - Tatsuo Ishii 1998-4-21
822 * Changed behavior if starting position is less than one to conform to SQL behavior.
823 * Formerly returned the entire string; now returns a portion.
824 * - Thomas Lockhart 1998-12-10
825 * Now uses faster TOAST-slicing interface
826 * - John Gray 2002-02-22
827 * Remove "#ifdef MULTIBYTE" and test for encoding_max_length instead. Change
828 * behaviors conflicting with SQL to meet SQL (if E = S + L < S throw
829 * error; if E < 1, return '', not entire string). Fixed MB related bug when
830 * S > LC and < LC + 4 sometimes garbage characters are returned.
831 * - Joe Conway 2002-08-10
834 text_substr(PG_FUNCTION_ARGS
)
836 PG_RETURN_TEXT_P(text_substring(PG_GETARG_DATUM(0),
843 * text_substr_no_len -
844 * Wrapper to avoid opr_sanity failure due to
845 * one function accepting a different number of args.
848 text_substr_no_len(PG_FUNCTION_ARGS
)
850 PG_RETURN_TEXT_P(text_substring(PG_GETARG_DATUM(0),
857 * Does the real work for text_substr() and text_substr_no_len()
859 * This is broken out so it can be called directly by other string processing
860 * functions. Note that the argument is passed as a Datum, to indicate that
861 * it may still be in compressed/toasted form. We can avoid detoasting all
862 * of it in some cases.
864 * The result is always a freshly palloc'd datum.
867 text_substring(Datum str
, int32 start
, int32 length
, bool length_not_specified
)
869 int32 eml
= pg_database_encoding_max_length();
870 int32 S
= start
; /* start position */
871 int32 S1
; /* adjusted start position */
872 int32 L1
; /* adjusted substring length */
873 int32 E
; /* end position */
876 * SQL99 says S can be zero or negative, but we still must fetch from the
877 * start of the string.
881 /* life is easy if the encoding max length is 1 */
884 if (length_not_specified
) /* special case - get length to end of
889 /* SQL99 says to throw an error for E < S, i.e., negative length */
891 (errcode(ERRCODE_SUBSTRING_ERROR
),
892 errmsg("negative substring length not allowed")));
893 L1
= -1; /* silence stupider compilers */
895 else if (pg_add_s32_overflow(S
, length
, &E
))
898 * L could be large enough for S + L to overflow, in which case
899 * the substring must run to end of string.
906 * A zero or negative value for the end position can happen if the
907 * start was negative or one. SQL99 says to return a zero-length
911 return cstring_to_text("");
917 * If the start position is past the end of the string, SQL99 says to
918 * return a zero-length string -- DatumGetTextPSlice() will do that
919 * for us. We need only convert S1 to zero-based starting position.
921 return DatumGetTextPSlice(str
, S1
- 1, L1
);
926 * When encoding max length is > 1, we can't get LC without
927 * detoasting, so we'll grab a conservatively large slice now and go
928 * back later to do the right thing
941 * We need to start at position zero because there is no way to know
942 * in advance which byte offset corresponds to the supplied start
947 if (length_not_specified
) /* special case - get length to end of
949 slice_size
= L1
= -1;
952 /* SQL99 says to throw an error for E < S, i.e., negative length */
954 (errcode(ERRCODE_SUBSTRING_ERROR
),
955 errmsg("negative substring length not allowed")));
956 slice_size
= L1
= -1; /* silence stupider compilers */
958 else if (pg_add_s32_overflow(S
, length
, &E
))
961 * L could be large enough for S + L to overflow, in which case
962 * the substring must run to end of string.
964 slice_size
= L1
= -1;
969 * A zero or negative value for the end position can happen if the
970 * start was negative or one. SQL99 says to return a zero-length
974 return cstring_to_text("");
977 * if E is past the end of the string, the tuple toaster will
978 * truncate the length for us
983 * Total slice size in bytes can't be any longer than the start
984 * position plus substring length times the encoding max length.
985 * If that overflows, we can just use -1.
987 if (pg_mul_s32_overflow(E
, eml
, &slice_size
))
992 * If we're working with an untoasted source, no need to do an extra
995 if (VARATT_IS_COMPRESSED(DatumGetPointer(str
)) ||
996 VARATT_IS_EXTERNAL(DatumGetPointer(str
)))
997 slice
= DatumGetTextPSlice(str
, slice_start
, slice_size
);
999 slice
= (text
*) DatumGetPointer(str
);
1001 /* see if we got back an empty string */
1002 if (VARSIZE_ANY_EXHDR(slice
) == 0)
1004 if (slice
!= (text
*) DatumGetPointer(str
))
1006 return cstring_to_text("");
1009 /* Now we can get the actual length of the slice in MB characters */
1010 slice_strlen
= pg_mbstrlen_with_len(VARDATA_ANY(slice
),
1011 VARSIZE_ANY_EXHDR(slice
));
1014 * Check that the start position wasn't > slice_strlen. If so, SQL99
1015 * says to return a zero-length string.
1017 if (S1
> slice_strlen
)
1019 if (slice
!= (text
*) DatumGetPointer(str
))
1021 return cstring_to_text("");
1025 * Adjust L1 and E1 now that we know the slice string length. Again
1026 * remember that S1 is one based, and slice_start is zero based.
1029 E1
= Min(S1
+ L1
, slice_start
+ 1 + slice_strlen
);
1031 E1
= slice_start
+ 1 + slice_strlen
;
1034 * Find the start position in the slice; remember S1 is not zero based
1036 p
= VARDATA_ANY(slice
);
1037 for (i
= 0; i
< S1
- 1; i
++)
1040 /* hang onto a pointer to our start position */
1044 * Count the actual bytes used by the substring of the requested
1047 for (i
= S1
; i
< E1
; i
++)
1050 ret
= (text
*) palloc(VARHDRSZ
+ (p
- s
));
1051 SET_VARSIZE(ret
, VARHDRSZ
+ (p
- s
));
1052 memcpy(VARDATA(ret
), s
, (p
- s
));
1054 if (slice
!= (text
*) DatumGetPointer(str
))
1060 elog(ERROR
, "invalid backend encoding: encoding max length < 1");
1062 /* not reached: suppress compiler warning */
1068 * Replace specified substring of first string with second
1070 * The SQL standard defines OVERLAY() in terms of substring and concatenation.
1071 * This code is a direct implementation of what the standard says.
1074 textoverlay(PG_FUNCTION_ARGS
)
1076 text
*t1
= PG_GETARG_TEXT_PP(0);
1077 text
*t2
= PG_GETARG_TEXT_PP(1);
1078 int sp
= PG_GETARG_INT32(2); /* substring start position */
1079 int sl
= PG_GETARG_INT32(3); /* substring length */
1081 PG_RETURN_TEXT_P(text_overlay(t1
, t2
, sp
, sl
));
1085 textoverlay_no_len(PG_FUNCTION_ARGS
)
1087 text
*t1
= PG_GETARG_TEXT_PP(0);
1088 text
*t2
= PG_GETARG_TEXT_PP(1);
1089 int sp
= PG_GETARG_INT32(2); /* substring start position */
1092 sl
= text_length(PointerGetDatum(t2
)); /* defaults to length(t2) */
1093 PG_RETURN_TEXT_P(text_overlay(t1
, t2
, sp
, sl
));
1097 text_overlay(text
*t1
, text
*t2
, int sp
, int sl
)
1105 * Check for possible integer-overflow cases. For negative sp, throw a
1106 * "substring length" error because that's what should be expected
1107 * according to the spec's definition of OVERLAY().
1111 (errcode(ERRCODE_SUBSTRING_ERROR
),
1112 errmsg("negative substring length not allowed")));
1113 if (pg_add_s32_overflow(sp
, sl
, &sp_pl_sl
))
1115 (errcode(ERRCODE_NUMERIC_VALUE_OUT_OF_RANGE
),
1116 errmsg("integer out of range")));
1118 s1
= text_substring(PointerGetDatum(t1
), 1, sp
- 1, false);
1119 s2
= text_substring(PointerGetDatum(t1
), sp_pl_sl
, -1, true);
1120 result
= text_catenate(s1
, t2
);
1121 result
= text_catenate(result
, s2
);
1128 * Return the position of the specified substring.
1129 * Implements the SQL POSITION() function.
1130 * Ref: A Guide To The SQL Standard, Date & Darwen, 1997
1131 * - thomas 1997-07-27
1134 textpos(PG_FUNCTION_ARGS
)
1136 text
*str
= PG_GETARG_TEXT_PP(0);
1137 text
*search_str
= PG_GETARG_TEXT_PP(1);
1139 PG_RETURN_INT32((int32
) text_position(str
, search_str
, PG_GET_COLLATION()));
1144 * Does the real work for textpos()
1147 * t1 - string to be searched
1148 * t2 - pattern to match within t1
1150 * Character index of the first matched char, starting from 1,
1153 * This is broken out so it can be called directly by other string processing
1157 text_position(text
*t1
, text
*t2
, Oid collid
)
1159 TextPositionState state
;
1162 /* Empty needle always matches at position 1 */
1163 if (VARSIZE_ANY_EXHDR(t2
) < 1)
1166 /* Otherwise, can't match if haystack is shorter than needle */
1167 if (VARSIZE_ANY_EXHDR(t1
) < VARSIZE_ANY_EXHDR(t2
))
1170 text_position_setup(t1
, t2
, collid
, &state
);
1171 if (!text_position_next(&state
))
1174 result
= text_position_get_match_pos(&state
);
1175 text_position_cleanup(&state
);
1181 * text_position_setup, text_position_next, text_position_cleanup -
1182 * Component steps of text_position()
1184 * These are broken out so that a string can be efficiently searched for
1185 * multiple occurrences of the same pattern. text_position_next may be
1186 * called multiple times, and it advances to the next match on each call.
1187 * text_position_get_match_ptr() and text_position_get_match_pos() return
1188 * a pointer or 1-based character position of the last match, respectively.
1190 * The "state" variable is normally just a local variable in the caller.
1192 * NOTE: text_position_next skips over the matched portion. For example,
1193 * searching for "xx" in "xxx" returns only one match, not two.
1197 text_position_setup(text
*t1
, text
*t2
, Oid collid
, TextPositionState
*state
)
1199 int len1
= VARSIZE_ANY_EXHDR(t1
);
1200 int len2
= VARSIZE_ANY_EXHDR(t2
);
1201 pg_locale_t mylocale
= 0;
1203 check_collation_set(collid
);
1205 if (!lc_collate_is_c(collid
))
1206 mylocale
= pg_newlocale_from_collation(collid
);
1208 if (mylocale
&& !mylocale
->deterministic
)
1210 (errcode(ERRCODE_FEATURE_NOT_SUPPORTED
),
1211 errmsg("nondeterministic collations are not supported for substring searches")));
1217 * Even with a multi-byte encoding, we perform the search using the raw
1218 * byte sequence, ignoring multibyte issues. For UTF-8, that works fine,
1219 * because in UTF-8 the byte sequence of one character cannot contain
1220 * another character. For other multi-byte encodings, we do the search
1221 * initially as a simple byte search, ignoring multibyte issues, but
1222 * verify afterwards that the match we found is at a character boundary,
1223 * and continue the search if it was a false match.
1225 if (pg_database_encoding_max_length() == 1)
1226 state
->is_multibyte_char_in_char
= false;
1227 else if (GetDatabaseEncoding() == PG_UTF8
)
1228 state
->is_multibyte_char_in_char
= false;
1230 state
->is_multibyte_char_in_char
= true;
1232 state
->str1
= VARDATA_ANY(t1
);
1233 state
->str2
= VARDATA_ANY(t2
);
1236 state
->last_match
= NULL
;
1237 state
->refpoint
= state
->str1
;
1241 * Prepare the skip table for Boyer-Moore-Horspool searching. In these
1242 * notes we use the terminology that the "haystack" is the string to be
1243 * searched (t1) and the "needle" is the pattern being sought (t2).
1245 * If the needle is empty or bigger than the haystack then there is no
1246 * point in wasting cycles initializing the table. We also choose not to
1247 * use B-M-H for needles of length 1, since the skip table can't possibly
1248 * save anything in that case.
1250 if (len1
>= len2
&& len2
> 1)
1252 int searchlength
= len1
- len2
;
1256 const char *str2
= state
->str2
;
1259 * First we must determine how much of the skip table to use. The
1260 * declaration of TextPositionState allows up to 256 elements, but for
1261 * short search problems we don't really want to have to initialize so
1262 * many elements --- it would take too long in comparison to the
1263 * actual search time. So we choose a useful skip table size based on
1264 * the haystack length minus the needle length. The closer the needle
1265 * length is to the haystack length the less useful skipping becomes.
1267 * Note: since we use bit-masking to select table elements, the skip
1268 * table size MUST be a power of 2, and so the mask must be 2^N-1.
1270 if (searchlength
< 16)
1272 else if (searchlength
< 64)
1274 else if (searchlength
< 128)
1276 else if (searchlength
< 512)
1278 else if (searchlength
< 2048)
1280 else if (searchlength
< 4096)
1281 skiptablemask
= 127;
1283 skiptablemask
= 255;
1284 state
->skiptablemask
= skiptablemask
;
1287 * Initialize the skip table. We set all elements to the needle
1288 * length, since this is the correct skip distance for any character
1289 * not found in the needle.
1291 for (i
= 0; i
<= skiptablemask
; i
++)
1292 state
->skiptable
[i
] = len2
;
1295 * Now examine the needle. For each character except the last one,
1296 * set the corresponding table element to the appropriate skip
1297 * distance. Note that when two characters share the same skip table
1298 * entry, the one later in the needle must determine the skip
1303 for (i
= 0; i
< last
; i
++)
1304 state
->skiptable
[(unsigned char) str2
[i
] & skiptablemask
] = last
- i
;
1309 * Advance to the next match, starting from the end of the previous match
1310 * (or the beginning of the string, on first call). Returns true if a match
1313 * Note that this refuses to match an empty-string needle. Most callers
1314 * will have handled that case specially and we'll never see it here.
1317 text_position_next(TextPositionState
*state
)
1319 int needle_len
= state
->len2
;
1323 if (needle_len
<= 0)
1324 return false; /* result for empty pattern */
1326 /* Start from the point right after the previous match. */
1327 if (state
->last_match
)
1328 start_ptr
= state
->last_match
+ needle_len
;
1330 start_ptr
= state
->str1
;
1333 matchptr
= text_position_next_internal(start_ptr
, state
);
1339 * Found a match for the byte sequence. If this is a multibyte encoding,
1340 * where one character's byte sequence can appear inside a longer
1341 * multi-byte character, we need to verify that the match was at a
1342 * character boundary, not in the middle of a multi-byte character.
1344 if (state
->is_multibyte_char_in_char
)
1346 /* Walk one character at a time, until we reach the match. */
1348 /* the search should never move backwards. */
1349 Assert(state
->refpoint
<= matchptr
);
1351 while (state
->refpoint
< matchptr
)
1353 /* step to next character. */
1354 state
->refpoint
+= pg_mblen(state
->refpoint
);
1358 * If we stepped over the match's start position, then it was a
1359 * false positive, where the byte sequence appeared in the middle
1360 * of a multi-byte character. Skip it, and continue the search at
1361 * the next character boundary.
1363 if (state
->refpoint
> matchptr
)
1365 start_ptr
= state
->refpoint
;
1371 state
->last_match
= matchptr
;
1376 * Subroutine of text_position_next(). This searches for the raw byte
1377 * sequence, ignoring any multi-byte encoding issues. Returns the first
1378 * match starting at 'start_ptr', or NULL if no match is found.
1381 text_position_next_internal(char *start_ptr
, TextPositionState
*state
)
1383 int haystack_len
= state
->len1
;
1384 int needle_len
= state
->len2
;
1385 int skiptablemask
= state
->skiptablemask
;
1386 const char *haystack
= state
->str1
;
1387 const char *needle
= state
->str2
;
1388 const char *haystack_end
= &haystack
[haystack_len
];
1391 Assert(start_ptr
>= haystack
&& start_ptr
<= haystack_end
);
1393 if (needle_len
== 1)
1395 /* No point in using B-M-H for a one-character needle */
1396 char nchar
= *needle
;
1399 while (hptr
< haystack_end
)
1402 return (char *) hptr
;
1408 const char *needle_last
= &needle
[needle_len
- 1];
1410 /* Start at startpos plus the length of the needle */
1411 hptr
= start_ptr
+ needle_len
- 1;
1412 while (hptr
< haystack_end
)
1414 /* Match the needle scanning *backward* */
1422 /* Matched it all? If so, return 1-based position */
1429 * No match, so use the haystack char at hptr to decide how far to
1430 * advance. If the needle had any occurrence of that character
1431 * (or more precisely, one sharing the same skiptable entry)
1432 * before its last character, then we advance far enough to align
1433 * the last such needle character with that haystack position.
1434 * Otherwise we can advance by the whole needle length.
1436 hptr
+= state
->skiptable
[(unsigned char) *hptr
& skiptablemask
];
1440 return 0; /* not found */
1444 * Return a pointer to the current match.
1446 * The returned pointer points into the original haystack string.
1449 text_position_get_match_ptr(TextPositionState
*state
)
1451 return state
->last_match
;
1455 * Return the offset of the current match.
1457 * The offset is in characters, 1-based.
1460 text_position_get_match_pos(TextPositionState
*state
)
1462 /* Convert the byte position to char position. */
1463 state
->refpos
+= pg_mbstrlen_with_len(state
->refpoint
,
1464 state
->last_match
- state
->refpoint
);
1465 state
->refpoint
= state
->last_match
;
1466 return state
->refpos
+ 1;
1470 * Reset search state to the initial state installed by text_position_setup.
1472 * The next call to text_position_next will search from the beginning
1476 text_position_reset(TextPositionState
*state
)
1478 state
->last_match
= NULL
;
1479 state
->refpoint
= state
->str1
;
1484 text_position_cleanup(TextPositionState
*state
)
1486 /* no cleanup needed */
1491 check_collation_set(Oid collid
)
1493 if (!OidIsValid(collid
))
1496 * This typically means that the parser could not resolve a conflict
1497 * of implicit collations, so report it that way.
1500 (errcode(ERRCODE_INDETERMINATE_COLLATION
),
1501 errmsg("could not determine which collation to use for string comparison"),
1502 errhint("Use the COLLATE clause to set the collation explicitly.")));
1507 * Comparison function for text strings with given lengths.
1508 * Includes locale support, but must copy strings to temporary memory
1509 * to allow null-termination for inputs to strcoll().
1510 * Returns an integer less than, equal to, or greater than zero, indicating
1511 * whether arg1 is less than, equal to, or greater than arg2.
1513 * Note: many functions that depend on this are marked leakproof; therefore,
1514 * avoid reporting the actual contents of the input when throwing errors.
1515 * All errors herein should be things that can't happen except on corrupt
1516 * data, anyway; otherwise we will have trouble with indexing strings that
1520 varstr_cmp(const char *arg1
, int len1
, const char *arg2
, int len2
, Oid collid
)
1524 check_collation_set(collid
);
1527 * Unfortunately, there is no strncoll(), so in the non-C locale case we
1528 * have to do some memory copying. This turns out to be significantly
1529 * slower, so we optimize the case where LC_COLLATE is C. We also try to
1530 * optimize relatively-short strings by avoiding palloc/pfree overhead.
1532 if (lc_collate_is_c(collid
))
1534 result
= memcmp(arg1
, arg2
, Min(len1
, len2
));
1535 if ((result
== 0) && (len1
!= len2
))
1536 result
= (len1
< len2
) ? -1 : 1;
1540 char a1buf
[TEXTBUFLEN
];
1541 char a2buf
[TEXTBUFLEN
];
1544 pg_locale_t mylocale
;
1546 mylocale
= pg_newlocale_from_collation(collid
);
1549 * memcmp() can't tell us which of two unequal strings sorts first,
1550 * but it's a cheap way to tell if they're equal. Testing shows that
1551 * memcmp() followed by strcoll() is only trivially slower than
1552 * strcoll() by itself, so we don't lose much if this doesn't work out
1553 * very often, and if it does - for example, because there are many
1554 * equal strings in the input - then we win big by avoiding expensive
1555 * collation-aware comparisons.
1557 if (len1
== len2
&& memcmp(arg1
, arg2
, len1
) == 0)
1561 /* Win32 does not have UTF-8, so we need to map to UTF-16 */
1562 if (GetDatabaseEncoding() == PG_UTF8
1563 && (!mylocale
|| mylocale
->provider
== COLLPROVIDER_LIBC
))
1569 if (len1
>= TEXTBUFLEN
/ 2)
1571 a1len
= len1
* 2 + 2;
1572 a1p
= palloc(a1len
);
1579 if (len2
>= TEXTBUFLEN
/ 2)
1581 a2len
= len2
* 2 + 2;
1582 a2p
= palloc(a2len
);
1590 /* stupid Microsloth API does not work for zero-length input */
1595 r
= MultiByteToWideChar(CP_UTF8
, 0, arg1
, len1
,
1596 (LPWSTR
) a1p
, a1len
/ 2);
1599 (errmsg("could not convert string to UTF-16: error code %lu",
1602 ((LPWSTR
) a1p
)[r
] = 0;
1608 r
= MultiByteToWideChar(CP_UTF8
, 0, arg2
, len2
,
1609 (LPWSTR
) a2p
, a2len
/ 2);
1612 (errmsg("could not convert string to UTF-16: error code %lu",
1615 ((LPWSTR
) a2p
)[r
] = 0;
1618 #ifdef HAVE_LOCALE_T
1620 result
= wcscoll_l((LPWSTR
) a1p
, (LPWSTR
) a2p
, mylocale
->info
.lt
);
1623 result
= wcscoll((LPWSTR
) a1p
, (LPWSTR
) a2p
);
1624 if (result
== 2147483647) /* _NLSCMPERROR; missing from mingw
1627 (errmsg("could not compare Unicode strings: %m")));
1629 /* Break tie if necessary. */
1631 (!mylocale
|| mylocale
->deterministic
))
1633 result
= memcmp(arg1
, arg2
, Min(len1
, len2
));
1634 if ((result
== 0) && (len1
!= len2
))
1635 result
= (len1
< len2
) ? -1 : 1;
1647 if (len1
>= TEXTBUFLEN
)
1648 a1p
= (char *) palloc(len1
+ 1);
1651 if (len2
>= TEXTBUFLEN
)
1652 a2p
= (char *) palloc(len2
+ 1);
1656 memcpy(a1p
, arg1
, len1
);
1658 memcpy(a2p
, arg2
, len2
);
1663 if (mylocale
->provider
== COLLPROVIDER_ICU
)
1666 #ifdef HAVE_UCOL_STRCOLLUTF8
1667 if (GetDatabaseEncoding() == PG_UTF8
)
1671 status
= U_ZERO_ERROR
;
1672 result
= ucol_strcollUTF8(mylocale
->info
.icu
.ucol
,
1676 if (U_FAILURE(status
))
1678 (errmsg("collation failed: %s", u_errorName(status
))));
1688 ulen1
= icu_to_uchar(&uchar1
, arg1
, len1
);
1689 ulen2
= icu_to_uchar(&uchar2
, arg2
, len2
);
1691 result
= ucol_strcoll(mylocale
->info
.icu
.ucol
,
1698 #else /* not USE_ICU */
1699 /* shouldn't happen */
1700 elog(ERROR
, "unsupported collprovider: %c", mylocale
->provider
);
1701 #endif /* not USE_ICU */
1705 #ifdef HAVE_LOCALE_T
1706 result
= strcoll_l(a1p
, a2p
, mylocale
->info
.lt
);
1708 /* shouldn't happen */
1709 elog(ERROR
, "unsupported collprovider: %c", mylocale
->provider
);
1714 result
= strcoll(a1p
, a2p
);
1716 /* Break tie if necessary. */
1718 (!mylocale
|| mylocale
->deterministic
))
1719 result
= strcmp(a1p
, a2p
);
1731 * Internal comparison function for text strings.
1732 * Returns -1, 0 or 1
1735 text_cmp(text
*arg1
, text
*arg2
, Oid collid
)
1742 a1p
= VARDATA_ANY(arg1
);
1743 a2p
= VARDATA_ANY(arg2
);
1745 len1
= VARSIZE_ANY_EXHDR(arg1
);
1746 len2
= VARSIZE_ANY_EXHDR(arg2
);
1748 return varstr_cmp(a1p
, len1
, a2p
, len2
, collid
);
1752 * Comparison functions for text strings.
1754 * Note: btree indexes need these routines not to leak memory; therefore,
1755 * be careful to free working copies of toasted datums. Most places don't
1756 * need to be so careful.
1760 texteq(PG_FUNCTION_ARGS
)
1762 Oid collid
= PG_GET_COLLATION();
1763 bool locale_is_c
= false;
1764 pg_locale_t mylocale
= 0;
1767 check_collation_set(collid
);
1769 if (lc_collate_is_c(collid
))
1772 mylocale
= pg_newlocale_from_collation(collid
);
1774 if (locale_is_c
|| !mylocale
|| mylocale
->deterministic
)
1776 Datum arg1
= PG_GETARG_DATUM(0);
1777 Datum arg2
= PG_GETARG_DATUM(1);
1782 * Since we only care about equality or not-equality, we can avoid all
1783 * the expense of strcoll() here, and just do bitwise comparison. In
1784 * fact, we don't even have to do a bitwise comparison if we can show
1785 * the lengths of the strings are unequal; which might save us from
1786 * having to detoast one or both values.
1788 len1
= toast_raw_datum_size(arg1
);
1789 len2
= toast_raw_datum_size(arg2
);
1794 text
*targ1
= DatumGetTextPP(arg1
);
1795 text
*targ2
= DatumGetTextPP(arg2
);
1797 result
= (memcmp(VARDATA_ANY(targ1
), VARDATA_ANY(targ2
),
1798 len1
- VARHDRSZ
) == 0);
1800 PG_FREE_IF_COPY(targ1
, 0);
1801 PG_FREE_IF_COPY(targ2
, 1);
1806 text
*arg1
= PG_GETARG_TEXT_PP(0);
1807 text
*arg2
= PG_GETARG_TEXT_PP(1);
1809 result
= (text_cmp(arg1
, arg2
, collid
) == 0);
1811 PG_FREE_IF_COPY(arg1
, 0);
1812 PG_FREE_IF_COPY(arg2
, 1);
1815 PG_RETURN_BOOL(result
);
1819 textne(PG_FUNCTION_ARGS
)
1821 Oid collid
= PG_GET_COLLATION();
1822 bool locale_is_c
= false;
1823 pg_locale_t mylocale
= 0;
1826 check_collation_set(collid
);
1828 if (lc_collate_is_c(collid
))
1831 mylocale
= pg_newlocale_from_collation(collid
);
1833 if (locale_is_c
|| !mylocale
|| mylocale
->deterministic
)
1835 Datum arg1
= PG_GETARG_DATUM(0);
1836 Datum arg2
= PG_GETARG_DATUM(1);
1840 /* See comment in texteq() */
1841 len1
= toast_raw_datum_size(arg1
);
1842 len2
= toast_raw_datum_size(arg2
);
1847 text
*targ1
= DatumGetTextPP(arg1
);
1848 text
*targ2
= DatumGetTextPP(arg2
);
1850 result
= (memcmp(VARDATA_ANY(targ1
), VARDATA_ANY(targ2
),
1851 len1
- VARHDRSZ
) != 0);
1853 PG_FREE_IF_COPY(targ1
, 0);
1854 PG_FREE_IF_COPY(targ2
, 1);
1859 text
*arg1
= PG_GETARG_TEXT_PP(0);
1860 text
*arg2
= PG_GETARG_TEXT_PP(1);
1862 result
= (text_cmp(arg1
, arg2
, collid
) != 0);
1864 PG_FREE_IF_COPY(arg1
, 0);
1865 PG_FREE_IF_COPY(arg2
, 1);
1868 PG_RETURN_BOOL(result
);
1872 text_lt(PG_FUNCTION_ARGS
)
1874 text
*arg1
= PG_GETARG_TEXT_PP(0);
1875 text
*arg2
= PG_GETARG_TEXT_PP(1);
1878 result
= (text_cmp(arg1
, arg2
, PG_GET_COLLATION()) < 0);
1880 PG_FREE_IF_COPY(arg1
, 0);
1881 PG_FREE_IF_COPY(arg2
, 1);
1883 PG_RETURN_BOOL(result
);
1887 text_le(PG_FUNCTION_ARGS
)
1889 text
*arg1
= PG_GETARG_TEXT_PP(0);
1890 text
*arg2
= PG_GETARG_TEXT_PP(1);
1893 result
= (text_cmp(arg1
, arg2
, PG_GET_COLLATION()) <= 0);
1895 PG_FREE_IF_COPY(arg1
, 0);
1896 PG_FREE_IF_COPY(arg2
, 1);
1898 PG_RETURN_BOOL(result
);
1902 text_gt(PG_FUNCTION_ARGS
)
1904 text
*arg1
= PG_GETARG_TEXT_PP(0);
1905 text
*arg2
= PG_GETARG_TEXT_PP(1);
1908 result
= (text_cmp(arg1
, arg2
, PG_GET_COLLATION()) > 0);
1910 PG_FREE_IF_COPY(arg1
, 0);
1911 PG_FREE_IF_COPY(arg2
, 1);
1913 PG_RETURN_BOOL(result
);
1917 text_ge(PG_FUNCTION_ARGS
)
1919 text
*arg1
= PG_GETARG_TEXT_PP(0);
1920 text
*arg2
= PG_GETARG_TEXT_PP(1);
1923 result
= (text_cmp(arg1
, arg2
, PG_GET_COLLATION()) >= 0);
1925 PG_FREE_IF_COPY(arg1
, 0);
1926 PG_FREE_IF_COPY(arg2
, 1);
1928 PG_RETURN_BOOL(result
);
1932 text_starts_with(PG_FUNCTION_ARGS
)
1934 Datum arg1
= PG_GETARG_DATUM(0);
1935 Datum arg2
= PG_GETARG_DATUM(1);
1936 Oid collid
= PG_GET_COLLATION();
1937 pg_locale_t mylocale
= 0;
1942 check_collation_set(collid
);
1944 if (!lc_collate_is_c(collid
))
1945 mylocale
= pg_newlocale_from_collation(collid
);
1947 if (mylocale
&& !mylocale
->deterministic
)
1949 (errcode(ERRCODE_FEATURE_NOT_SUPPORTED
),
1950 errmsg("nondeterministic collations are not supported for substring searches")));
1952 len1
= toast_raw_datum_size(arg1
);
1953 len2
= toast_raw_datum_size(arg2
);
1958 text
*targ1
= text_substring(arg1
, 1, len2
, false);
1959 text
*targ2
= DatumGetTextPP(arg2
);
1961 result
= (memcmp(VARDATA_ANY(targ1
), VARDATA_ANY(targ2
),
1962 VARSIZE_ANY_EXHDR(targ2
)) == 0);
1964 PG_FREE_IF_COPY(targ1
, 0);
1965 PG_FREE_IF_COPY(targ2
, 1);
1968 PG_RETURN_BOOL(result
);
1972 bttextcmp(PG_FUNCTION_ARGS
)
1974 text
*arg1
= PG_GETARG_TEXT_PP(0);
1975 text
*arg2
= PG_GETARG_TEXT_PP(1);
1978 result
= text_cmp(arg1
, arg2
, PG_GET_COLLATION());
1980 PG_FREE_IF_COPY(arg1
, 0);
1981 PG_FREE_IF_COPY(arg2
, 1);
1983 PG_RETURN_INT32(result
);
1987 bttextsortsupport(PG_FUNCTION_ARGS
)
1989 SortSupport ssup
= (SortSupport
) PG_GETARG_POINTER(0);
1990 Oid collid
= ssup
->ssup_collation
;
1991 MemoryContext oldcontext
;
1993 oldcontext
= MemoryContextSwitchTo(ssup
->ssup_cxt
);
1995 /* Use generic string SortSupport */
1996 varstr_sortsupport(ssup
, TEXTOID
, collid
);
1998 MemoryContextSwitchTo(oldcontext
);
2004 * Generic sortsupport interface for character type's operator classes.
2005 * Includes locale support, and support for BpChar semantics (i.e. removing
2006 * trailing spaces before comparison).
2008 * Relies on the assumption that text, VarChar, BpChar, and bytea all have the
2009 * same representation. Callers that always use the C collation (e.g.
2010 * non-collatable type callers like bytea) may have NUL bytes in their strings;
2011 * this will not work with any other collation, though.
2014 varstr_sortsupport(SortSupport ssup
, Oid typid
, Oid collid
)
2016 bool abbreviate
= ssup
->abbreviate
;
2017 bool collate_c
= false;
2018 VarStringSortSupport
*sss
;
2019 pg_locale_t locale
= 0;
2021 check_collation_set(collid
);
2024 * If possible, set ssup->comparator to a function which can be used to
2025 * directly compare two datums. If we can do this, we'll avoid the
2026 * overhead of a trip through the fmgr layer for every comparison, which
2027 * can be substantial.
2029 * Most typically, we'll set the comparator to varlenafastcmp_locale,
2030 * which uses strcoll() to perform comparisons. We use that for the
2031 * BpChar case too, but type NAME uses namefastcmp_locale. However, if
2032 * LC_COLLATE = C, we can make things quite a bit faster with
2033 * varstrfastcmp_c, bpcharfastcmp_c, or namefastcmp_c, all of which use
2034 * memcmp() rather than strcoll().
2036 if (lc_collate_is_c(collid
))
2038 if (typid
== BPCHAROID
)
2039 ssup
->comparator
= bpcharfastcmp_c
;
2040 else if (typid
== NAMEOID
)
2042 ssup
->comparator
= namefastcmp_c
;
2043 /* Not supporting abbreviation with type NAME, for now */
2047 ssup
->comparator
= varstrfastcmp_c
;
2054 * We need a collation-sensitive comparison. To make things faster,
2055 * we'll figure out the collation based on the locale id and cache the
2058 locale
= pg_newlocale_from_collation(collid
);
2061 * There is a further exception on Windows. When the database
2062 * encoding is UTF-8 and we are not using the C collation, complex
2063 * hacks are required. We don't currently have a comparator that
2064 * handles that case, so we fall back on the slow method of having the
2065 * sort code invoke bttextcmp() (in the case of text) via the fmgr
2066 * trampoline. ICU locales work just the same on Windows, however.
2069 if (GetDatabaseEncoding() == PG_UTF8
&&
2070 !(locale
&& locale
->provider
== COLLPROVIDER_ICU
))
2075 * We use varlenafastcmp_locale except for type NAME.
2077 if (typid
== NAMEOID
)
2079 ssup
->comparator
= namefastcmp_locale
;
2080 /* Not supporting abbreviation with type NAME, for now */
2084 ssup
->comparator
= varlenafastcmp_locale
;
2088 * Unfortunately, it seems that abbreviation for non-C collations is
2089 * broken on many common platforms; testing of multiple versions of glibc
2090 * reveals that, for many locales, strcoll() and strxfrm() do not return
2091 * consistent results, which is fatal to this optimization. While no
2092 * other libc other than Cygwin has so far been shown to have a problem,
2093 * we take the conservative course of action for right now and disable
2094 * this categorically. (Users who are certain this isn't a problem on
2095 * their system can define TRUST_STRXFRM.)
2097 * Even apart from the risk of broken locales, it's possible that there
2098 * are platforms where the use of abbreviated keys should be disabled at
2099 * compile time. Having only 4 byte datums could make worst-case
2100 * performance drastically more likely, for example. Moreover, macOS's
2101 * strxfrm() implementation is known to not effectively concentrate a
2102 * significant amount of entropy from the original string in earlier
2103 * transformed blobs. It's possible that other supported platforms are
2104 * similarly encumbered. So, if we ever get past disabling this
2105 * categorically, we may still want or need to disable it for particular
2108 #ifndef TRUST_STRXFRM
2109 if (!collate_c
&& !(locale
&& locale
->provider
== COLLPROVIDER_ICU
))
2114 * If we're using abbreviated keys, or if we're using a locale-aware
2115 * comparison, we need to initialize a VarStringSortSupport object. Both
2116 * cases will make use of the temporary buffers we initialize here for
2117 * scratch space (and to detect requirement for BpChar semantics from
2118 * caller), and the abbreviation case requires additional state.
2120 if (abbreviate
|| !collate_c
)
2122 sss
= palloc(sizeof(VarStringSortSupport
));
2123 sss
->buf1
= palloc(TEXTBUFLEN
);
2124 sss
->buflen1
= TEXTBUFLEN
;
2125 sss
->buf2
= palloc(TEXTBUFLEN
);
2126 sss
->buflen2
= TEXTBUFLEN
;
2127 /* Start with invalid values */
2128 sss
->last_len1
= -1;
2129 sss
->last_len2
= -1;
2131 sss
->last_returned
= 0;
2132 sss
->locale
= locale
;
2135 * To avoid somehow confusing a strxfrm() blob and an original string,
2136 * constantly keep track of the variety of data that buf1 and buf2
2137 * currently contain.
2139 * Comparisons may be interleaved with conversion calls. Frequently,
2140 * conversions and comparisons are batched into two distinct phases,
2141 * but the correctness of caching cannot hinge upon this. For
2142 * comparison caching, buffer state is only trusted if cache_blob is
2143 * found set to false, whereas strxfrm() caching only trusts the state
2144 * when cache_blob is found set to true.
2146 * Arbitrarily initialize cache_blob to true.
2148 sss
->cache_blob
= true;
2149 sss
->collate_c
= collate_c
;
2151 ssup
->ssup_extra
= sss
;
2154 * If possible, plan to use the abbreviated keys optimization. The
2155 * core code may switch back to authoritative comparator should
2156 * abbreviation be aborted.
2160 sss
->prop_card
= 0.20;
2161 initHyperLogLog(&sss
->abbr_card
, 10);
2162 initHyperLogLog(&sss
->full_card
, 10);
2163 ssup
->abbrev_full_comparator
= ssup
->comparator
;
2164 ssup
->comparator
= ssup_datum_unsigned_cmp
;
2165 ssup
->abbrev_converter
= varstr_abbrev_convert
;
2166 ssup
->abbrev_abort
= varstr_abbrev_abort
;
2172 * sortsupport comparison func (for C locale case)
2175 varstrfastcmp_c(Datum x
, Datum y
, SortSupport ssup
)
2177 VarString
*arg1
= DatumGetVarStringPP(x
);
2178 VarString
*arg2
= DatumGetVarStringPP(y
);
2185 a1p
= VARDATA_ANY(arg1
);
2186 a2p
= VARDATA_ANY(arg2
);
2188 len1
= VARSIZE_ANY_EXHDR(arg1
);
2189 len2
= VARSIZE_ANY_EXHDR(arg2
);
2191 result
= memcmp(a1p
, a2p
, Min(len1
, len2
));
2192 if ((result
== 0) && (len1
!= len2
))
2193 result
= (len1
< len2
) ? -1 : 1;
2195 /* We can't afford to leak memory here. */
2196 if (PointerGetDatum(arg1
) != x
)
2198 if (PointerGetDatum(arg2
) != y
)
2205 * sortsupport comparison func (for BpChar C locale case)
2207 * BpChar outsources its sortsupport to this module. Specialization for the
2208 * varstr_sortsupport BpChar case, modeled on
2209 * internal_bpchar_pattern_compare().
2212 bpcharfastcmp_c(Datum x
, Datum y
, SortSupport ssup
)
2214 BpChar
*arg1
= DatumGetBpCharPP(x
);
2215 BpChar
*arg2
= DatumGetBpCharPP(y
);
2222 a1p
= VARDATA_ANY(arg1
);
2223 a2p
= VARDATA_ANY(arg2
);
2225 len1
= bpchartruelen(a1p
, VARSIZE_ANY_EXHDR(arg1
));
2226 len2
= bpchartruelen(a2p
, VARSIZE_ANY_EXHDR(arg2
));
2228 result
= memcmp(a1p
, a2p
, Min(len1
, len2
));
2229 if ((result
== 0) && (len1
!= len2
))
2230 result
= (len1
< len2
) ? -1 : 1;
2232 /* We can't afford to leak memory here. */
2233 if (PointerGetDatum(arg1
) != x
)
2235 if (PointerGetDatum(arg2
) != y
)
2242 * sortsupport comparison func (for NAME C locale case)
2245 namefastcmp_c(Datum x
, Datum y
, SortSupport ssup
)
2247 Name arg1
= DatumGetName(x
);
2248 Name arg2
= DatumGetName(y
);
2250 return strncmp(NameStr(*arg1
), NameStr(*arg2
), NAMEDATALEN
);
2254 * sortsupport comparison func (for locale case with all varlena types)
2257 varlenafastcmp_locale(Datum x
, Datum y
, SortSupport ssup
)
2259 VarString
*arg1
= DatumGetVarStringPP(x
);
2260 VarString
*arg2
= DatumGetVarStringPP(y
);
2267 a1p
= VARDATA_ANY(arg1
);
2268 a2p
= VARDATA_ANY(arg2
);
2270 len1
= VARSIZE_ANY_EXHDR(arg1
);
2271 len2
= VARSIZE_ANY_EXHDR(arg2
);
2273 result
= varstrfastcmp_locale(a1p
, len1
, a2p
, len2
, ssup
);
2275 /* We can't afford to leak memory here. */
2276 if (PointerGetDatum(arg1
) != x
)
2278 if (PointerGetDatum(arg2
) != y
)
2285 * sortsupport comparison func (for locale case with NAME type)
2288 namefastcmp_locale(Datum x
, Datum y
, SortSupport ssup
)
2290 Name arg1
= DatumGetName(x
);
2291 Name arg2
= DatumGetName(y
);
2293 return varstrfastcmp_locale(NameStr(*arg1
), strlen(NameStr(*arg1
)),
2294 NameStr(*arg2
), strlen(NameStr(*arg2
)),
2299 * sortsupport comparison func for locale cases
2302 varstrfastcmp_locale(char *a1p
, int len1
, char *a2p
, int len2
, SortSupport ssup
)
2304 VarStringSortSupport
*sss
= (VarStringSortSupport
*) ssup
->ssup_extra
;
2308 /* Fast pre-check for equality, as discussed in varstr_cmp() */
2309 if (len1
== len2
&& memcmp(a1p
, a2p
, len1
) == 0)
2312 * No change in buf1 or buf2 contents, so avoid changing last_len1 or
2313 * last_len2. Existing contents of buffers might still be used by
2316 * It's fine to allow the comparison of BpChar padding bytes here,
2317 * even though that implies that the memcmp() will usually be
2318 * performed for BpChar callers (though multibyte characters could
2319 * still prevent that from occurring). The memcmp() is still very
2320 * cheap, and BpChar's funny semantics have us remove trailing spaces
2321 * (not limited to padding), so we need make no distinction between
2322 * padding space characters and "real" space characters.
2327 if (sss
->typid
== BPCHAROID
)
2329 /* Get true number of bytes, ignoring trailing spaces */
2330 len1
= bpchartruelen(a1p
, len1
);
2331 len2
= bpchartruelen(a2p
, len2
);
2334 if (len1
>= sss
->buflen1
)
2336 sss
->buflen1
= Max(len1
+ 1, Min(sss
->buflen1
* 2, MaxAllocSize
));
2337 sss
->buf1
= repalloc(sss
->buf1
, sss
->buflen1
);
2339 if (len2
>= sss
->buflen2
)
2341 sss
->buflen2
= Max(len2
+ 1, Min(sss
->buflen2
* 2, MaxAllocSize
));
2342 sss
->buf2
= repalloc(sss
->buf2
, sss
->buflen2
);
2346 * We're likely to be asked to compare the same strings repeatedly, and
2347 * memcmp() is so much cheaper than strcoll() that it pays to try to cache
2348 * comparisons, even though in general there is no reason to think that
2349 * that will work out (every string datum may be unique). Caching does
2350 * not slow things down measurably when it doesn't work out, and can speed
2351 * things up by rather a lot when it does. In part, this is because the
2352 * memcmp() compares data from cachelines that are needed in L1 cache even
2353 * when the last comparison's result cannot be reused.
2356 if (len1
!= sss
->last_len1
|| memcmp(sss
->buf1
, a1p
, len1
) != 0)
2359 memcpy(sss
->buf1
, a1p
, len1
);
2360 sss
->buf1
[len1
] = '\0';
2361 sss
->last_len1
= len1
;
2365 * If we're comparing the same two strings as last time, we can return the
2366 * same answer without calling strcoll() again. This is more likely than
2367 * it seems (at least with moderate to low cardinality sets), because
2368 * quicksort compares the same pivot against many values.
2370 if (len2
!= sss
->last_len2
|| memcmp(sss
->buf2
, a2p
, len2
) != 0)
2372 memcpy(sss
->buf2
, a2p
, len2
);
2373 sss
->buf2
[len2
] = '\0';
2374 sss
->last_len2
= len2
;
2376 else if (arg1_match
&& !sss
->cache_blob
)
2378 /* Use result cached following last actual strcoll() call */
2379 return sss
->last_returned
;
2384 if (sss
->locale
->provider
== COLLPROVIDER_ICU
)
2387 #ifdef HAVE_UCOL_STRCOLLUTF8
2388 if (GetDatabaseEncoding() == PG_UTF8
)
2392 status
= U_ZERO_ERROR
;
2393 result
= ucol_strcollUTF8(sss
->locale
->info
.icu
.ucol
,
2397 if (U_FAILURE(status
))
2399 (errmsg("collation failed: %s", u_errorName(status
))));
2409 ulen1
= icu_to_uchar(&uchar1
, a1p
, len1
);
2410 ulen2
= icu_to_uchar(&uchar2
, a2p
, len2
);
2412 result
= ucol_strcoll(sss
->locale
->info
.icu
.ucol
,
2419 #else /* not USE_ICU */
2420 /* shouldn't happen */
2421 elog(ERROR
, "unsupported collprovider: %c", sss
->locale
->provider
);
2422 #endif /* not USE_ICU */
2426 #ifdef HAVE_LOCALE_T
2427 result
= strcoll_l(sss
->buf1
, sss
->buf2
, sss
->locale
->info
.lt
);
2429 /* shouldn't happen */
2430 elog(ERROR
, "unsupported collprovider: %c", sss
->locale
->provider
);
2435 result
= strcoll(sss
->buf1
, sss
->buf2
);
2437 /* Break tie if necessary. */
2439 (!sss
->locale
|| sss
->locale
->deterministic
))
2440 result
= strcmp(sss
->buf1
, sss
->buf2
);
2442 /* Cache result, perhaps saving an expensive strcoll() call next time */
2443 sss
->cache_blob
= false;
2444 sss
->last_returned
= result
;
2449 * Conversion routine for sortsupport. Converts original to abbreviated key
2450 * representation. Our encoding strategy is simple -- pack the first 8 bytes
2451 * of a strxfrm() blob into a Datum (on little-endian machines, the 8 bytes are
2452 * stored in reverse order), and treat it as an unsigned integer. When the "C"
2453 * locale is used, or in case of bytea, just memcpy() from original instead.
2456 varstr_abbrev_convert(Datum original
, SortSupport ssup
)
2458 VarStringSortSupport
*sss
= (VarStringSortSupport
*) ssup
->ssup_extra
;
2459 VarString
*authoritative
= DatumGetVarStringPP(original
);
2460 char *authoritative_data
= VARDATA_ANY(authoritative
);
2468 pres
= (char *) &res
;
2469 /* memset(), so any non-overwritten bytes are NUL */
2470 memset(pres
, 0, sizeof(Datum
));
2471 len
= VARSIZE_ANY_EXHDR(authoritative
);
2473 /* Get number of bytes, ignoring trailing spaces */
2474 if (sss
->typid
== BPCHAROID
)
2475 len
= bpchartruelen(authoritative_data
, len
);
2478 * If we're using the C collation, use memcpy(), rather than strxfrm(), to
2479 * abbreviate keys. The full comparator for the C locale is always
2480 * memcmp(). It would be incorrect to allow bytea callers (callers that
2481 * always force the C collation -- bytea isn't a collatable type, but this
2482 * approach is convenient) to use strxfrm(). This is because bytea
2483 * strings may contain NUL bytes. Besides, this should be faster, too.
2485 * More generally, it's okay that bytea callers can have NUL bytes in
2486 * strings because abbreviated cmp need not make a distinction between
2487 * terminating NUL bytes, and NUL bytes representing actual NULs in the
2488 * authoritative representation. Hopefully a comparison at or past one
2489 * abbreviated key's terminating NUL byte will resolve the comparison
2490 * without consulting the authoritative representation; specifically, some
2491 * later non-NUL byte in the longer string can resolve the comparison
2492 * against a subsequent terminating NUL in the shorter string. There will
2493 * usually be what is effectively a "length-wise" resolution there and
2496 * If that doesn't work out -- if all bytes in the longer string
2497 * positioned at or past the offset of the smaller string's (first)
2498 * terminating NUL are actually representative of NUL bytes in the
2499 * authoritative binary string (perhaps with some *terminating* NUL bytes
2500 * towards the end of the longer string iff it happens to still be small)
2501 * -- then an authoritative tie-breaker will happen, and do the right
2502 * thing: explicitly consider string length.
2505 memcpy(pres
, authoritative_data
, Min(len
, sizeof(Datum
)));
2511 UChar
*uchar
= NULL
;
2515 * We're not using the C collation, so fall back on strxfrm or ICU
2519 /* By convention, we use buffer 1 to store and NUL-terminate */
2520 if (len
>= sss
->buflen1
)
2522 sss
->buflen1
= Max(len
+ 1, Min(sss
->buflen1
* 2, MaxAllocSize
));
2523 sss
->buf1
= repalloc(sss
->buf1
, sss
->buflen1
);
2526 /* Might be able to reuse strxfrm() blob from last call */
2527 if (sss
->last_len1
== len
&& sss
->cache_blob
&&
2528 memcmp(sss
->buf1
, authoritative_data
, len
) == 0)
2530 memcpy(pres
, sss
->buf2
, Min(sizeof(Datum
), sss
->last_len2
));
2531 /* No change affecting cardinality, so no hashing required */
2535 memcpy(sss
->buf1
, authoritative_data
, len
);
2538 * Just like strcoll(), strxfrm() expects a NUL-terminated string. Not
2539 * necessary for ICU, but doesn't hurt.
2541 sss
->buf1
[len
] = '\0';
2542 sss
->last_len1
= len
;
2545 /* When using ICU and not UTF8, convert string to UChar. */
2546 if (sss
->locale
&& sss
->locale
->provider
== COLLPROVIDER_ICU
&&
2547 GetDatabaseEncoding() != PG_UTF8
)
2548 ulen
= icu_to_uchar(&uchar
, sss
->buf1
, len
);
2552 * Loop: Call strxfrm() or ucol_getSortKey(), possibly enlarge buffer,
2553 * and try again. Both of these functions have the result buffer
2554 * content undefined if the result did not fit, so we need to retry
2555 * until everything fits, even though we only need the first few bytes
2556 * in the end. When using ucol_nextSortKeyPart(), however, we only
2557 * ask for as many bytes as we actually need.
2562 if (sss
->locale
&& sss
->locale
->provider
== COLLPROVIDER_ICU
)
2565 * When using UTF8, use the iteration interface so we only
2566 * need to produce as many bytes as we actually need.
2568 if (GetDatabaseEncoding() == PG_UTF8
)
2574 uiter_setUTF8(&iter
, sss
->buf1
, len
);
2575 state
[0] = state
[1] = 0; /* won't need that again */
2576 status
= U_ZERO_ERROR
;
2577 bsize
= ucol_nextSortKeyPart(sss
->locale
->info
.icu
.ucol
,
2580 (uint8_t *) sss
->buf2
,
2581 Min(sizeof(Datum
), sss
->buflen2
),
2583 if (U_FAILURE(status
))
2585 (errmsg("sort key generation failed: %s",
2586 u_errorName(status
))));
2589 bsize
= ucol_getSortKey(sss
->locale
->info
.icu
.ucol
,
2591 (uint8_t *) sss
->buf2
, sss
->buflen2
);
2595 #ifdef HAVE_LOCALE_T
2596 if (sss
->locale
&& sss
->locale
->provider
== COLLPROVIDER_LIBC
)
2597 bsize
= strxfrm_l(sss
->buf2
, sss
->buf1
,
2598 sss
->buflen2
, sss
->locale
->info
.lt
);
2601 bsize
= strxfrm(sss
->buf2
, sss
->buf1
, sss
->buflen2
);
2603 sss
->last_len2
= bsize
;
2604 if (bsize
< sss
->buflen2
)
2608 * Grow buffer and retry.
2610 sss
->buflen2
= Max(bsize
+ 1,
2611 Min(sss
->buflen2
* 2, MaxAllocSize
));
2612 sss
->buf2
= repalloc(sss
->buf2
, sss
->buflen2
);
2616 * Every Datum byte is always compared. This is safe because the
2617 * strxfrm() blob is itself NUL terminated, leaving no danger of
2618 * misinterpreting any NUL bytes not intended to be interpreted as
2619 * logically representing termination.
2621 * (Actually, even if there were NUL bytes in the blob it would be
2622 * okay. See remarks on bytea case above.)
2624 memcpy(pres
, sss
->buf2
, Min(sizeof(Datum
), bsize
));
2633 * Maintain approximate cardinality of both abbreviated keys and original,
2634 * authoritative keys using HyperLogLog. Used as cheap insurance against
2635 * the worst case, where we do many string transformations for no saving
2636 * in full strcoll()-based comparisons. These statistics are used by
2637 * varstr_abbrev_abort().
2639 * First, Hash key proper, or a significant fraction of it. Mix in length
2640 * in order to compensate for cases where differences are past
2641 * PG_CACHE_LINE_SIZE bytes, so as to limit the overhead of hashing.
2643 hash
= DatumGetUInt32(hash_any((unsigned char *) authoritative_data
,
2644 Min(len
, PG_CACHE_LINE_SIZE
)));
2646 if (len
> PG_CACHE_LINE_SIZE
)
2647 hash
^= DatumGetUInt32(hash_uint32((uint32
) len
));
2649 addHyperLogLog(&sss
->full_card
, hash
);
2651 /* Hash abbreviated key */
2652 #if SIZEOF_DATUM == 8
2657 lohalf
= (uint32
) res
;
2658 hihalf
= (uint32
) (res
>> 32);
2659 hash
= DatumGetUInt32(hash_uint32(lohalf
^ hihalf
));
2661 #else /* SIZEOF_DATUM != 8 */
2662 hash
= DatumGetUInt32(hash_uint32((uint32
) res
));
2665 addHyperLogLog(&sss
->abbr_card
, hash
);
2667 /* Cache result, perhaps saving an expensive strxfrm() call next time */
2668 sss
->cache_blob
= true;
2672 * Byteswap on little-endian machines.
2674 * This is needed so that ssup_datum_unsigned_cmp() (an unsigned integer
2675 * 3-way comparator) works correctly on all platforms. If we didn't do
2676 * this, the comparator would have to call memcmp() with a pair of
2677 * pointers to the first byte of each abbreviated key, which is slower.
2679 res
= DatumBigEndianToNative(res
);
2681 /* Don't leak memory here */
2682 if (PointerGetDatum(authoritative
) != original
)
2683 pfree(authoritative
);
2689 * Callback for estimating effectiveness of abbreviated key optimization, using
2690 * heuristic rules. Returns value indicating if the abbreviation optimization
2691 * should be aborted, based on its projected effectiveness.
2694 varstr_abbrev_abort(int memtupcount
, SortSupport ssup
)
2696 VarStringSortSupport
*sss
= (VarStringSortSupport
*) ssup
->ssup_extra
;
2697 double abbrev_distinct
,
2700 Assert(ssup
->abbreviate
);
2702 /* Have a little patience */
2703 if (memtupcount
< 100)
2706 abbrev_distinct
= estimateHyperLogLog(&sss
->abbr_card
);
2707 key_distinct
= estimateHyperLogLog(&sss
->full_card
);
2710 * Clamp cardinality estimates to at least one distinct value. While
2711 * NULLs are generally disregarded, if only NULL values were seen so far,
2712 * that might misrepresent costs if we failed to clamp.
2714 if (abbrev_distinct
<= 1.0)
2715 abbrev_distinct
= 1.0;
2717 if (key_distinct
<= 1.0)
2721 * In the worst case all abbreviated keys are identical, while at the same
2722 * time there are differences within full key strings not captured in
2728 double norm_abbrev_card
= abbrev_distinct
/ (double) memtupcount
;
2730 elog(LOG
, "varstr_abbrev: abbrev_distinct after %d: %f "
2731 "(key_distinct: %f, norm_abbrev_card: %f, prop_card: %f)",
2732 memtupcount
, abbrev_distinct
, key_distinct
, norm_abbrev_card
,
2738 * If the number of distinct abbreviated keys approximately matches the
2739 * number of distinct authoritative original keys, that's reason enough to
2740 * proceed. We can win even with a very low cardinality set if most
2741 * tie-breakers only memcmp(). This is by far the most important
2744 * While comparisons that are resolved at the abbreviated key level are
2745 * considerably cheaper than tie-breakers resolved with memcmp(), both of
2746 * those two outcomes are so much cheaper than a full strcoll() once
2747 * sorting is underway that it doesn't seem worth it to weigh abbreviated
2748 * cardinality against the overall size of the set in order to more
2749 * accurately model costs. Assume that an abbreviated comparison, and an
2750 * abbreviated comparison with a cheap memcmp()-based authoritative
2751 * resolution are equivalent.
2753 if (abbrev_distinct
> key_distinct
* sss
->prop_card
)
2756 * When we have exceeded 10,000 tuples, decay required cardinality
2757 * aggressively for next call.
2759 * This is useful because the number of comparisons required on
2760 * average increases at a linearithmic rate, and at roughly 10,000
2761 * tuples that factor will start to dominate over the linear costs of
2762 * string transformation (this is a conservative estimate). The decay
2763 * rate is chosen to be a little less aggressive than halving -- which
2764 * (since we're called at points at which memtupcount has doubled)
2765 * would never see the cost model actually abort past the first call
2766 * following a decay. This decay rate is mostly a precaution against
2767 * a sudden, violent swing in how well abbreviated cardinality tracks
2768 * full key cardinality. The decay also serves to prevent a marginal
2769 * case from being aborted too late, when too much has already been
2770 * invested in string transformation.
2772 * It's possible for sets of several million distinct strings with
2773 * mere tens of thousands of distinct abbreviated keys to still
2774 * benefit very significantly. This will generally occur provided
2775 * each abbreviated key is a proxy for a roughly uniform number of the
2776 * set's full keys. If it isn't so, we hope to catch that early and
2777 * abort. If it isn't caught early, by the time the problem is
2778 * apparent it's probably not worth aborting.
2780 if (memtupcount
> 10000)
2781 sss
->prop_card
*= 0.65;
2787 * Abort abbreviation strategy.
2789 * The worst case, where all abbreviated keys are identical while all
2790 * original strings differ will typically only see a regression of about
2791 * 10% in execution time for small to medium sized lists of strings.
2792 * Whereas on modern CPUs where cache stalls are the dominant cost, we can
2793 * often expect very large improvements, particularly with sets of strings
2794 * of moderately high to high abbreviated cardinality. There is little to
2795 * lose but much to gain, which our strategy reflects.
2799 elog(LOG
, "varstr_abbrev: aborted abbreviation at %d "
2800 "(abbrev_distinct: %f, key_distinct: %f, prop_card: %f)",
2801 memtupcount
, abbrev_distinct
, key_distinct
, sss
->prop_card
);
2808 * Generic equalimage support function for character type's operator classes.
2809 * Disables the use of deduplication with nondeterministic collations.
2812 btvarstrequalimage(PG_FUNCTION_ARGS
)
2814 /* Oid opcintype = PG_GETARG_OID(0); */
2815 Oid collid
= PG_GET_COLLATION();
2817 check_collation_set(collid
);
2819 if (lc_collate_is_c(collid
) ||
2820 collid
== DEFAULT_COLLATION_OID
||
2821 get_collation_isdeterministic(collid
))
2822 PG_RETURN_BOOL(true);
2824 PG_RETURN_BOOL(false);
2828 text_larger(PG_FUNCTION_ARGS
)
2830 text
*arg1
= PG_GETARG_TEXT_PP(0);
2831 text
*arg2
= PG_GETARG_TEXT_PP(1);
2834 result
= ((text_cmp(arg1
, arg2
, PG_GET_COLLATION()) > 0) ? arg1
: arg2
);
2836 PG_RETURN_TEXT_P(result
);
2840 text_smaller(PG_FUNCTION_ARGS
)
2842 text
*arg1
= PG_GETARG_TEXT_PP(0);
2843 text
*arg2
= PG_GETARG_TEXT_PP(1);
2846 result
= ((text_cmp(arg1
, arg2
, PG_GET_COLLATION()) < 0) ? arg1
: arg2
);
2848 PG_RETURN_TEXT_P(result
);
2853 * Cross-type comparison functions for types text and name.
2857 nameeqtext(PG_FUNCTION_ARGS
)
2859 Name arg1
= PG_GETARG_NAME(0);
2860 text
*arg2
= PG_GETARG_TEXT_PP(1);
2861 size_t len1
= strlen(NameStr(*arg1
));
2862 size_t len2
= VARSIZE_ANY_EXHDR(arg2
);
2863 Oid collid
= PG_GET_COLLATION();
2866 check_collation_set(collid
);
2868 if (collid
== C_COLLATION_OID
)
2869 result
= (len1
== len2
&&
2870 memcmp(NameStr(*arg1
), VARDATA_ANY(arg2
), len1
) == 0);
2872 result
= (varstr_cmp(NameStr(*arg1
), len1
,
2873 VARDATA_ANY(arg2
), len2
,
2876 PG_FREE_IF_COPY(arg2
, 1);
2878 PG_RETURN_BOOL(result
);
2882 texteqname(PG_FUNCTION_ARGS
)
2884 text
*arg1
= PG_GETARG_TEXT_PP(0);
2885 Name arg2
= PG_GETARG_NAME(1);
2886 size_t len1
= VARSIZE_ANY_EXHDR(arg1
);
2887 size_t len2
= strlen(NameStr(*arg2
));
2888 Oid collid
= PG_GET_COLLATION();
2891 check_collation_set(collid
);
2893 if (collid
== C_COLLATION_OID
)
2894 result
= (len1
== len2
&&
2895 memcmp(VARDATA_ANY(arg1
), NameStr(*arg2
), len1
) == 0);
2897 result
= (varstr_cmp(VARDATA_ANY(arg1
), len1
,
2898 NameStr(*arg2
), len2
,
2901 PG_FREE_IF_COPY(arg1
, 0);
2903 PG_RETURN_BOOL(result
);
2907 namenetext(PG_FUNCTION_ARGS
)
2909 Name arg1
= PG_GETARG_NAME(0);
2910 text
*arg2
= PG_GETARG_TEXT_PP(1);
2911 size_t len1
= strlen(NameStr(*arg1
));
2912 size_t len2
= VARSIZE_ANY_EXHDR(arg2
);
2913 Oid collid
= PG_GET_COLLATION();
2916 check_collation_set(collid
);
2918 if (collid
== C_COLLATION_OID
)
2919 result
= !(len1
== len2
&&
2920 memcmp(NameStr(*arg1
), VARDATA_ANY(arg2
), len1
) == 0);
2922 result
= !(varstr_cmp(NameStr(*arg1
), len1
,
2923 VARDATA_ANY(arg2
), len2
,
2926 PG_FREE_IF_COPY(arg2
, 1);
2928 PG_RETURN_BOOL(result
);
2932 textnename(PG_FUNCTION_ARGS
)
2934 text
*arg1
= PG_GETARG_TEXT_PP(0);
2935 Name arg2
= PG_GETARG_NAME(1);
2936 size_t len1
= VARSIZE_ANY_EXHDR(arg1
);
2937 size_t len2
= strlen(NameStr(*arg2
));
2938 Oid collid
= PG_GET_COLLATION();
2941 check_collation_set(collid
);
2943 if (collid
== C_COLLATION_OID
)
2944 result
= !(len1
== len2
&&
2945 memcmp(VARDATA_ANY(arg1
), NameStr(*arg2
), len1
) == 0);
2947 result
= !(varstr_cmp(VARDATA_ANY(arg1
), len1
,
2948 NameStr(*arg2
), len2
,
2951 PG_FREE_IF_COPY(arg1
, 0);
2953 PG_RETURN_BOOL(result
);
2957 btnametextcmp(PG_FUNCTION_ARGS
)
2959 Name arg1
= PG_GETARG_NAME(0);
2960 text
*arg2
= PG_GETARG_TEXT_PP(1);
2963 result
= varstr_cmp(NameStr(*arg1
), strlen(NameStr(*arg1
)),
2964 VARDATA_ANY(arg2
), VARSIZE_ANY_EXHDR(arg2
),
2965 PG_GET_COLLATION());
2967 PG_FREE_IF_COPY(arg2
, 1);
2969 PG_RETURN_INT32(result
);
2973 bttextnamecmp(PG_FUNCTION_ARGS
)
2975 text
*arg1
= PG_GETARG_TEXT_PP(0);
2976 Name arg2
= PG_GETARG_NAME(1);
2979 result
= varstr_cmp(VARDATA_ANY(arg1
), VARSIZE_ANY_EXHDR(arg1
),
2980 NameStr(*arg2
), strlen(NameStr(*arg2
)),
2981 PG_GET_COLLATION());
2983 PG_FREE_IF_COPY(arg1
, 0);
2985 PG_RETURN_INT32(result
);
2988 #define CmpCall(cmpfunc) \
2989 DatumGetInt32(DirectFunctionCall2Coll(cmpfunc, \
2990 PG_GET_COLLATION(), \
2991 PG_GETARG_DATUM(0), \
2992 PG_GETARG_DATUM(1)))
2995 namelttext(PG_FUNCTION_ARGS
)
2997 PG_RETURN_BOOL(CmpCall(btnametextcmp
) < 0);
3001 nameletext(PG_FUNCTION_ARGS
)
3003 PG_RETURN_BOOL(CmpCall(btnametextcmp
) <= 0);
3007 namegttext(PG_FUNCTION_ARGS
)
3009 PG_RETURN_BOOL(CmpCall(btnametextcmp
) > 0);
3013 namegetext(PG_FUNCTION_ARGS
)
3015 PG_RETURN_BOOL(CmpCall(btnametextcmp
) >= 0);
3019 textltname(PG_FUNCTION_ARGS
)
3021 PG_RETURN_BOOL(CmpCall(bttextnamecmp
) < 0);
3025 textlename(PG_FUNCTION_ARGS
)
3027 PG_RETURN_BOOL(CmpCall(bttextnamecmp
) <= 0);
3031 textgtname(PG_FUNCTION_ARGS
)
3033 PG_RETURN_BOOL(CmpCall(bttextnamecmp
) > 0);
3037 textgename(PG_FUNCTION_ARGS
)
3039 PG_RETURN_BOOL(CmpCall(bttextnamecmp
) >= 0);
3046 * The following operators support character-by-character comparison
3047 * of text datums, to allow building indexes suitable for LIKE clauses.
3048 * Note that the regular texteq/textne comparison operators, and regular
3049 * support functions 1 and 2 with "C" collation are assumed to be
3050 * compatible with these!
3054 internal_text_pattern_compare(text
*arg1
, text
*arg2
)
3060 len1
= VARSIZE_ANY_EXHDR(arg1
);
3061 len2
= VARSIZE_ANY_EXHDR(arg2
);
3063 result
= memcmp(VARDATA_ANY(arg1
), VARDATA_ANY(arg2
), Min(len1
, len2
));
3066 else if (len1
< len2
)
3068 else if (len1
> len2
)
3076 text_pattern_lt(PG_FUNCTION_ARGS
)
3078 text
*arg1
= PG_GETARG_TEXT_PP(0);
3079 text
*arg2
= PG_GETARG_TEXT_PP(1);
3082 result
= internal_text_pattern_compare(arg1
, arg2
);
3084 PG_FREE_IF_COPY(arg1
, 0);
3085 PG_FREE_IF_COPY(arg2
, 1);
3087 PG_RETURN_BOOL(result
< 0);
3092 text_pattern_le(PG_FUNCTION_ARGS
)
3094 text
*arg1
= PG_GETARG_TEXT_PP(0);
3095 text
*arg2
= PG_GETARG_TEXT_PP(1);
3098 result
= internal_text_pattern_compare(arg1
, arg2
);
3100 PG_FREE_IF_COPY(arg1
, 0);
3101 PG_FREE_IF_COPY(arg2
, 1);
3103 PG_RETURN_BOOL(result
<= 0);
3108 text_pattern_ge(PG_FUNCTION_ARGS
)
3110 text
*arg1
= PG_GETARG_TEXT_PP(0);
3111 text
*arg2
= PG_GETARG_TEXT_PP(1);
3114 result
= internal_text_pattern_compare(arg1
, arg2
);
3116 PG_FREE_IF_COPY(arg1
, 0);
3117 PG_FREE_IF_COPY(arg2
, 1);
3119 PG_RETURN_BOOL(result
>= 0);
3124 text_pattern_gt(PG_FUNCTION_ARGS
)
3126 text
*arg1
= PG_GETARG_TEXT_PP(0);
3127 text
*arg2
= PG_GETARG_TEXT_PP(1);
3130 result
= internal_text_pattern_compare(arg1
, arg2
);
3132 PG_FREE_IF_COPY(arg1
, 0);
3133 PG_FREE_IF_COPY(arg2
, 1);
3135 PG_RETURN_BOOL(result
> 0);
3140 bttext_pattern_cmp(PG_FUNCTION_ARGS
)
3142 text
*arg1
= PG_GETARG_TEXT_PP(0);
3143 text
*arg2
= PG_GETARG_TEXT_PP(1);
3146 result
= internal_text_pattern_compare(arg1
, arg2
);
3148 PG_FREE_IF_COPY(arg1
, 0);
3149 PG_FREE_IF_COPY(arg2
, 1);
3151 PG_RETURN_INT32(result
);
3156 bttext_pattern_sortsupport(PG_FUNCTION_ARGS
)
3158 SortSupport ssup
= (SortSupport
) PG_GETARG_POINTER(0);
3159 MemoryContext oldcontext
;
3161 oldcontext
= MemoryContextSwitchTo(ssup
->ssup_cxt
);
3163 /* Use generic string SortSupport, forcing "C" collation */
3164 varstr_sortsupport(ssup
, TEXTOID
, C_COLLATION_OID
);
3166 MemoryContextSwitchTo(oldcontext
);
3172 /*-------------------------------------------------------------
3175 * get the number of bytes contained in an instance of type 'bytea'
3176 *-------------------------------------------------------------
3179 byteaoctetlen(PG_FUNCTION_ARGS
)
3181 Datum str
= PG_GETARG_DATUM(0);
3183 /* We need not detoast the input at all */
3184 PG_RETURN_INT32(toast_raw_datum_size(str
) - VARHDRSZ
);
3189 * takes two bytea* and returns a bytea* that is the concatenation of
3192 * Cloned from textcat and modified as required.
3195 byteacat(PG_FUNCTION_ARGS
)
3197 bytea
*t1
= PG_GETARG_BYTEA_PP(0);
3198 bytea
*t2
= PG_GETARG_BYTEA_PP(1);
3200 PG_RETURN_BYTEA_P(bytea_catenate(t1
, t2
));
3205 * Guts of byteacat(), broken out so it can be used by other functions
3207 * Arguments can be in short-header form, but not compressed or out-of-line
3210 bytea_catenate(bytea
*t1
, bytea
*t2
)
3218 len1
= VARSIZE_ANY_EXHDR(t1
);
3219 len2
= VARSIZE_ANY_EXHDR(t2
);
3221 /* paranoia ... probably should throw error instead? */
3227 len
= len1
+ len2
+ VARHDRSZ
;
3228 result
= (bytea
*) palloc(len
);
3230 /* Set size of result string... */
3231 SET_VARSIZE(result
, len
);
3233 /* Fill data field of result string... */
3234 ptr
= VARDATA(result
);
3236 memcpy(ptr
, VARDATA_ANY(t1
), len1
);
3238 memcpy(ptr
+ len1
, VARDATA_ANY(t2
), len2
);
3243 #define PG_STR_GET_BYTEA(str_) \
3244 DatumGetByteaPP(DirectFunctionCall1(byteain, CStringGetDatum(str_)))
3248 * Return a substring starting at the specified position.
3249 * Cloned from text_substr and modified as required.
3253 * - starting position (is one-based)
3254 * - string length (optional)
3256 * If the starting position is zero or less, then return from the start of the string
3257 * adjusting the length to be consistent with the "negative start" per SQL.
3258 * If the length is less than zero, an ERROR is thrown. If no third argument
3259 * (length) is provided, the length to the end of the string is assumed.
3262 bytea_substr(PG_FUNCTION_ARGS
)
3264 PG_RETURN_BYTEA_P(bytea_substring(PG_GETARG_DATUM(0),
3271 * bytea_substr_no_len -
3272 * Wrapper to avoid opr_sanity failure due to
3273 * one function accepting a different number of args.
3276 bytea_substr_no_len(PG_FUNCTION_ARGS
)
3278 PG_RETURN_BYTEA_P(bytea_substring(PG_GETARG_DATUM(0),
3285 bytea_substring(Datum str
,
3288 bool length_not_specified
)
3290 int32 S1
; /* adjusted start position */
3291 int32 L1
; /* adjusted substring length */
3292 int32 E
; /* end position */
3295 * The logic here should generally match text_substring().
3299 if (length_not_specified
)
3302 * Not passed a length - DatumGetByteaPSlice() grabs everything to the
3303 * end of the string if we pass it a negative value for length.
3309 /* SQL99 says to throw an error for E < S, i.e., negative length */
3311 (errcode(ERRCODE_SUBSTRING_ERROR
),
3312 errmsg("negative substring length not allowed")));
3313 L1
= -1; /* silence stupider compilers */
3315 else if (pg_add_s32_overflow(S
, L
, &E
))
3318 * L could be large enough for S + L to overflow, in which case the
3319 * substring must run to end of string.
3326 * A zero or negative value for the end position can happen if the
3327 * start was negative or one. SQL99 says to return a zero-length
3331 return PG_STR_GET_BYTEA("");
3337 * If the start position is past the end of the string, SQL99 says to
3338 * return a zero-length string -- DatumGetByteaPSlice() will do that for
3339 * us. We need only convert S1 to zero-based starting position.
3341 return DatumGetByteaPSlice(str
, S1
- 1, L1
);
3346 * Replace specified substring of first string with second
3348 * The SQL standard defines OVERLAY() in terms of substring and concatenation.
3349 * This code is a direct implementation of what the standard says.
3352 byteaoverlay(PG_FUNCTION_ARGS
)
3354 bytea
*t1
= PG_GETARG_BYTEA_PP(0);
3355 bytea
*t2
= PG_GETARG_BYTEA_PP(1);
3356 int sp
= PG_GETARG_INT32(2); /* substring start position */
3357 int sl
= PG_GETARG_INT32(3); /* substring length */
3359 PG_RETURN_BYTEA_P(bytea_overlay(t1
, t2
, sp
, sl
));
3363 byteaoverlay_no_len(PG_FUNCTION_ARGS
)
3365 bytea
*t1
= PG_GETARG_BYTEA_PP(0);
3366 bytea
*t2
= PG_GETARG_BYTEA_PP(1);
3367 int sp
= PG_GETARG_INT32(2); /* substring start position */
3370 sl
= VARSIZE_ANY_EXHDR(t2
); /* defaults to length(t2) */
3371 PG_RETURN_BYTEA_P(bytea_overlay(t1
, t2
, sp
, sl
));
3375 bytea_overlay(bytea
*t1
, bytea
*t2
, int sp
, int sl
)
3383 * Check for possible integer-overflow cases. For negative sp, throw a
3384 * "substring length" error because that's what should be expected
3385 * according to the spec's definition of OVERLAY().
3389 (errcode(ERRCODE_SUBSTRING_ERROR
),
3390 errmsg("negative substring length not allowed")));
3391 if (pg_add_s32_overflow(sp
, sl
, &sp_pl_sl
))
3393 (errcode(ERRCODE_NUMERIC_VALUE_OUT_OF_RANGE
),
3394 errmsg("integer out of range")));
3396 s1
= bytea_substring(PointerGetDatum(t1
), 1, sp
- 1, false);
3397 s2
= bytea_substring(PointerGetDatum(t1
), sp_pl_sl
, -1, true);
3398 result
= bytea_catenate(s1
, t2
);
3399 result
= bytea_catenate(result
, s2
);
3408 bytea_bit_count(PG_FUNCTION_ARGS
)
3410 bytea
*t1
= PG_GETARG_BYTEA_PP(0);
3412 PG_RETURN_INT64(pg_popcount(VARDATA_ANY(t1
), VARSIZE_ANY_EXHDR(t1
)));
3417 * Return the position of the specified substring.
3418 * Implements the SQL POSITION() function.
3419 * Cloned from textpos and modified as required.
3422 byteapos(PG_FUNCTION_ARGS
)
3424 bytea
*t1
= PG_GETARG_BYTEA_PP(0);
3425 bytea
*t2
= PG_GETARG_BYTEA_PP(1);
3434 len1
= VARSIZE_ANY_EXHDR(t1
);
3435 len2
= VARSIZE_ANY_EXHDR(t2
);
3438 PG_RETURN_INT32(1); /* result for empty pattern */
3440 p1
= VARDATA_ANY(t1
);
3441 p2
= VARDATA_ANY(t2
);
3445 for (p
= 0; p
<= px
; p
++)
3447 if ((*p2
== *p1
) && (memcmp(p1
, p2
, len2
) == 0))
3455 PG_RETURN_INT32(pos
);
3458 /*-------------------------------------------------------------
3461 * this routine treats "bytea" as an array of bytes.
3462 * It returns the Nth byte (a number between 0 and 255).
3463 *-------------------------------------------------------------
3466 byteaGetByte(PG_FUNCTION_ARGS
)
3468 bytea
*v
= PG_GETARG_BYTEA_PP(0);
3469 int32 n
= PG_GETARG_INT32(1);
3473 len
= VARSIZE_ANY_EXHDR(v
);
3475 if (n
< 0 || n
>= len
)
3477 (errcode(ERRCODE_ARRAY_SUBSCRIPT_ERROR
),
3478 errmsg("index %d out of valid range, 0..%d",
3481 byte
= ((unsigned char *) VARDATA_ANY(v
))[n
];
3483 PG_RETURN_INT32(byte
);
3486 /*-------------------------------------------------------------
3489 * This routine treats a "bytea" type like an array of bits.
3490 * It returns the value of the Nth bit (0 or 1).
3492 *-------------------------------------------------------------
3495 byteaGetBit(PG_FUNCTION_ARGS
)
3497 bytea
*v
= PG_GETARG_BYTEA_PP(0);
3498 int64 n
= PG_GETARG_INT64(1);
3504 len
= VARSIZE_ANY_EXHDR(v
);
3506 if (n
< 0 || n
>= (int64
) len
* 8)
3508 (errcode(ERRCODE_ARRAY_SUBSCRIPT_ERROR
),
3509 errmsg("index %lld out of valid range, 0..%lld",
3510 (long long) n
, (long long) len
* 8 - 1)));
3512 /* n/8 is now known < len, so safe to cast to int */
3513 byteNo
= (int) (n
/ 8);
3514 bitNo
= (int) (n
% 8);
3516 byte
= ((unsigned char *) VARDATA_ANY(v
))[byteNo
];
3518 if (byte
& (1 << bitNo
))
3524 /*-------------------------------------------------------------
3527 * Given an instance of type 'bytea' creates a new one with
3528 * the Nth byte set to the given value.
3530 *-------------------------------------------------------------
3533 byteaSetByte(PG_FUNCTION_ARGS
)
3535 bytea
*res
= PG_GETARG_BYTEA_P_COPY(0);
3536 int32 n
= PG_GETARG_INT32(1);
3537 int32 newByte
= PG_GETARG_INT32(2);
3540 len
= VARSIZE(res
) - VARHDRSZ
;
3542 if (n
< 0 || n
>= len
)
3544 (errcode(ERRCODE_ARRAY_SUBSCRIPT_ERROR
),
3545 errmsg("index %d out of valid range, 0..%d",
3551 ((unsigned char *) VARDATA(res
))[n
] = newByte
;
3553 PG_RETURN_BYTEA_P(res
);
3556 /*-------------------------------------------------------------
3559 * Given an instance of type 'bytea' creates a new one with
3560 * the Nth bit set to the given value.
3562 *-------------------------------------------------------------
3565 byteaSetBit(PG_FUNCTION_ARGS
)
3567 bytea
*res
= PG_GETARG_BYTEA_P_COPY(0);
3568 int64 n
= PG_GETARG_INT64(1);
3569 int32 newBit
= PG_GETARG_INT32(2);
3576 len
= VARSIZE(res
) - VARHDRSZ
;
3578 if (n
< 0 || n
>= (int64
) len
* 8)
3580 (errcode(ERRCODE_ARRAY_SUBSCRIPT_ERROR
),
3581 errmsg("index %lld out of valid range, 0..%lld",
3582 (long long) n
, (long long) len
* 8 - 1)));
3584 /* n/8 is now known < len, so safe to cast to int */
3585 byteNo
= (int) (n
/ 8);
3586 bitNo
= (int) (n
% 8);
3591 if (newBit
!= 0 && newBit
!= 1)
3593 (errcode(ERRCODE_INVALID_PARAMETER_VALUE
),
3594 errmsg("new bit must be 0 or 1")));
3599 oldByte
= ((unsigned char *) VARDATA(res
))[byteNo
];
3602 newByte
= oldByte
& (~(1 << bitNo
));
3604 newByte
= oldByte
| (1 << bitNo
);
3606 ((unsigned char *) VARDATA(res
))[byteNo
] = newByte
;
3608 PG_RETURN_BYTEA_P(res
);
3613 * Converts a text type to a Name type.
3616 text_name(PG_FUNCTION_ARGS
)
3618 text
*s
= PG_GETARG_TEXT_PP(0);
3622 len
= VARSIZE_ANY_EXHDR(s
);
3624 /* Truncate oversize input */
3625 if (len
>= NAMEDATALEN
)
3626 len
= pg_mbcliplen(VARDATA_ANY(s
), len
, NAMEDATALEN
- 1);
3628 /* We use palloc0 here to ensure result is zero-padded */
3629 result
= (Name
) palloc0(NAMEDATALEN
);
3630 memcpy(NameStr(*result
), VARDATA_ANY(s
), len
);
3632 PG_RETURN_NAME(result
);
3636 * Converts a Name type to a text type.
3639 name_text(PG_FUNCTION_ARGS
)
3641 Name s
= PG_GETARG_NAME(0);
3643 PG_RETURN_TEXT_P(cstring_to_text(NameStr(*s
)));
3648 * textToQualifiedNameList - convert a text object to list of names
3650 * This implements the input parsing needed by nextval() and other
3651 * functions that take a text parameter representing a qualified name.
3652 * We split the name at dots, downcase if not double-quoted, and
3653 * truncate names if they're too long.
3656 textToQualifiedNameList(text
*textval
)
3663 /* Convert to C string (handles possible detoasting). */
3664 /* Note we rely on being able to modify rawname below. */
3665 rawname
= text_to_cstring(textval
);
3667 if (!SplitIdentifierString(rawname
, '.', &namelist
))
3669 (errcode(ERRCODE_INVALID_NAME
),
3670 errmsg("invalid name syntax")));
3672 if (namelist
== NIL
)
3674 (errcode(ERRCODE_INVALID_NAME
),
3675 errmsg("invalid name syntax")));
3677 foreach(l
, namelist
)
3679 char *curname
= (char *) lfirst(l
);
3681 result
= lappend(result
, makeString(pstrdup(curname
)));
3685 list_free(namelist
);
3691 * SplitIdentifierString --- parse a string containing identifiers
3693 * This is the guts of textToQualifiedNameList, and is exported for use in
3694 * other situations such as parsing GUC variables. In the GUC case, it's
3695 * important to avoid memory leaks, so the API is designed to minimize the
3696 * amount of stuff that needs to be allocated and freed.
3699 * rawstring: the input string; must be overwritable! On return, it's
3700 * been modified to contain the separated identifiers.
3701 * separator: the separator punctuation expected between identifiers
3702 * (typically '.' or ','). Whitespace may also appear around
3705 * namelist: filled with a palloc'd list of pointers to identifiers within
3706 * rawstring. Caller should list_free() this even on error return.
3708 * Returns true if okay, false if there is a syntax error in the string.
3710 * Note that an empty string is considered okay here, though not in
3711 * textToQualifiedNameList.
3714 SplitIdentifierString(char *rawstring
, char separator
,
3717 char *nextp
= rawstring
;
3722 while (scanner_isspace(*nextp
))
3723 nextp
++; /* skip leading whitespace */
3726 return true; /* allow empty string */
3728 /* At the top of the loop, we are at start of a new identifier. */
3736 /* Quoted name --- collapse quote-quote pairs, no downcasing */
3737 curname
= nextp
+ 1;
3740 endp
= strchr(nextp
+ 1, '"');
3742 return false; /* mismatched quotes */
3744 break; /* found end of quoted name */
3745 /* Collapse adjacent quotes into one quote, and look again */
3746 memmove(endp
, endp
+ 1, strlen(endp
));
3749 /* endp now points at the terminating quote */
3754 /* Unquoted name --- extends to separator or whitespace */
3759 while (*nextp
&& *nextp
!= separator
&&
3760 !scanner_isspace(*nextp
))
3763 if (curname
== nextp
)
3764 return false; /* empty unquoted name not allowed */
3767 * Downcase the identifier, using same code as main lexer does.
3769 * XXX because we want to overwrite the input in-place, we cannot
3770 * support a downcasing transformation that increases the string
3771 * length. This is not a problem given the current implementation
3772 * of downcase_truncate_identifier, but we'll probably have to do
3773 * something about this someday.
3775 len
= endp
- curname
;
3776 downname
= downcase_truncate_identifier(curname
, len
, false);
3777 Assert(strlen(downname
) <= len
);
3778 strncpy(curname
, downname
, len
); /* strncpy is required here */
3782 while (scanner_isspace(*nextp
))
3783 nextp
++; /* skip trailing whitespace */
3785 if (*nextp
== separator
)
3788 while (scanner_isspace(*nextp
))
3789 nextp
++; /* skip leading whitespace for next */
3790 /* we expect another name, so done remains false */
3792 else if (*nextp
== '\0')
3795 return false; /* invalid syntax */
3797 /* Now safe to overwrite separator with a null */
3800 /* Truncate name if it's overlength */
3801 truncate_identifier(curname
, strlen(curname
), false);
3804 * Finished isolating current name --- add it to list
3806 *namelist
= lappend(*namelist
, curname
);
3808 /* Loop back if we didn't reach end of string */
3816 * SplitDirectoriesString --- parse a string containing file/directory names
3818 * This works fine on file names too; the function name is historical.
3820 * This is similar to SplitIdentifierString, except that the parsing
3821 * rules are meant to handle pathnames instead of identifiers: there is
3822 * no downcasing, embedded spaces are allowed, the max length is MAXPGPATH-1,
3823 * and we apply canonicalize_path() to each extracted string. Because of the
3824 * last, the returned strings are separately palloc'd rather than being
3825 * pointers into rawstring --- but we still scribble on rawstring.
3828 * rawstring: the input string; must be modifiable!
3829 * separator: the separator punctuation expected between directories
3830 * (typically ',' or ';'). Whitespace may also appear around
3833 * namelist: filled with a palloc'd list of directory names.
3834 * Caller should list_free_deep() this even on error return.
3836 * Returns true if okay, false if there is a syntax error in the string.
3838 * Note that an empty string is considered okay here.
3841 SplitDirectoriesString(char *rawstring
, char separator
,
3844 char *nextp
= rawstring
;
3849 while (scanner_isspace(*nextp
))
3850 nextp
++; /* skip leading whitespace */
3853 return true; /* allow empty string */
3855 /* At the top of the loop, we are at start of a new directory. */
3863 /* Quoted name --- collapse quote-quote pairs */
3864 curname
= nextp
+ 1;
3867 endp
= strchr(nextp
+ 1, '"');
3869 return false; /* mismatched quotes */
3871 break; /* found end of quoted name */
3872 /* Collapse adjacent quotes into one quote, and look again */
3873 memmove(endp
, endp
+ 1, strlen(endp
));
3876 /* endp now points at the terminating quote */
3881 /* Unquoted name --- extends to separator or end of string */
3882 curname
= endp
= nextp
;
3883 while (*nextp
&& *nextp
!= separator
)
3885 /* trailing whitespace should not be included in name */
3886 if (!scanner_isspace(*nextp
))
3890 if (curname
== endp
)
3891 return false; /* empty unquoted name not allowed */
3894 while (scanner_isspace(*nextp
))
3895 nextp
++; /* skip trailing whitespace */
3897 if (*nextp
== separator
)
3900 while (scanner_isspace(*nextp
))
3901 nextp
++; /* skip leading whitespace for next */
3902 /* we expect another name, so done remains false */
3904 else if (*nextp
== '\0')
3907 return false; /* invalid syntax */
3909 /* Now safe to overwrite separator with a null */
3912 /* Truncate path if it's overlength */
3913 if (strlen(curname
) >= MAXPGPATH
)
3914 curname
[MAXPGPATH
- 1] = '\0';
3917 * Finished isolating current name --- add it to list
3919 curname
= pstrdup(curname
);
3920 canonicalize_path(curname
);
3921 *namelist
= lappend(*namelist
, curname
);
3923 /* Loop back if we didn't reach end of string */
3931 * SplitGUCList --- parse a string containing identifiers or file names
3933 * This is used to split the value of a GUC_LIST_QUOTE GUC variable, without
3934 * presuming whether the elements will be taken as identifiers or file names.
3935 * We assume the input has already been through flatten_set_variable_args(),
3936 * so that we need never downcase (if appropriate, that was done already).
3937 * Nor do we ever truncate, since we don't know the correct max length.
3938 * We disallow embedded whitespace for simplicity (it shouldn't matter,
3939 * because any embedded whitespace should have led to double-quoting).
3940 * Otherwise the API is identical to SplitIdentifierString.
3942 * XXX it's annoying to have so many copies of this string-splitting logic.
3943 * However, it's not clear that having one function with a bunch of option
3944 * flags would be much better.
3946 * XXX there is a version of this function in src/bin/pg_dump/dumputils.c.
3947 * Be sure to update that if you have to change this.
3950 * rawstring: the input string; must be overwritable! On return, it's
3951 * been modified to contain the separated identifiers.
3952 * separator: the separator punctuation expected between identifiers
3953 * (typically '.' or ','). Whitespace may also appear around
3956 * namelist: filled with a palloc'd list of pointers to identifiers within
3957 * rawstring. Caller should list_free() this even on error return.
3959 * Returns true if okay, false if there is a syntax error in the string.
3962 SplitGUCList(char *rawstring
, char separator
,
3965 char *nextp
= rawstring
;
3970 while (scanner_isspace(*nextp
))
3971 nextp
++; /* skip leading whitespace */
3974 return true; /* allow empty string */
3976 /* At the top of the loop, we are at start of a new identifier. */
3984 /* Quoted name --- collapse quote-quote pairs */
3985 curname
= nextp
+ 1;
3988 endp
= strchr(nextp
+ 1, '"');
3990 return false; /* mismatched quotes */
3992 break; /* found end of quoted name */
3993 /* Collapse adjacent quotes into one quote, and look again */
3994 memmove(endp
, endp
+ 1, strlen(endp
));
3997 /* endp now points at the terminating quote */
4002 /* Unquoted name --- extends to separator or whitespace */
4004 while (*nextp
&& *nextp
!= separator
&&
4005 !scanner_isspace(*nextp
))
4008 if (curname
== nextp
)
4009 return false; /* empty unquoted name not allowed */
4012 while (scanner_isspace(*nextp
))
4013 nextp
++; /* skip trailing whitespace */
4015 if (*nextp
== separator
)
4018 while (scanner_isspace(*nextp
))
4019 nextp
++; /* skip leading whitespace for next */
4020 /* we expect another name, so done remains false */
4022 else if (*nextp
== '\0')
4025 return false; /* invalid syntax */
4027 /* Now safe to overwrite separator with a null */
4031 * Finished isolating current name --- add it to list
4033 *namelist
= lappend(*namelist
, curname
);
4035 /* Loop back if we didn't reach end of string */
4042 /*****************************************************************************
4043 * Comparison Functions used for bytea
4045 * Note: btree indexes need these routines not to leak memory; therefore,
4046 * be careful to free working copies of toasted datums. Most places don't
4047 * need to be so careful.
4048 *****************************************************************************/
4051 byteaeq(PG_FUNCTION_ARGS
)
4053 Datum arg1
= PG_GETARG_DATUM(0);
4054 Datum arg2
= PG_GETARG_DATUM(1);
4060 * We can use a fast path for unequal lengths, which might save us from
4061 * having to detoast one or both values.
4063 len1
= toast_raw_datum_size(arg1
);
4064 len2
= toast_raw_datum_size(arg2
);
4069 bytea
*barg1
= DatumGetByteaPP(arg1
);
4070 bytea
*barg2
= DatumGetByteaPP(arg2
);
4072 result
= (memcmp(VARDATA_ANY(barg1
), VARDATA_ANY(barg2
),
4073 len1
- VARHDRSZ
) == 0);
4075 PG_FREE_IF_COPY(barg1
, 0);
4076 PG_FREE_IF_COPY(barg2
, 1);
4079 PG_RETURN_BOOL(result
);
4083 byteane(PG_FUNCTION_ARGS
)
4085 Datum arg1
= PG_GETARG_DATUM(0);
4086 Datum arg2
= PG_GETARG_DATUM(1);
4092 * We can use a fast path for unequal lengths, which might save us from
4093 * having to detoast one or both values.
4095 len1
= toast_raw_datum_size(arg1
);
4096 len2
= toast_raw_datum_size(arg2
);
4101 bytea
*barg1
= DatumGetByteaPP(arg1
);
4102 bytea
*barg2
= DatumGetByteaPP(arg2
);
4104 result
= (memcmp(VARDATA_ANY(barg1
), VARDATA_ANY(barg2
),
4105 len1
- VARHDRSZ
) != 0);
4107 PG_FREE_IF_COPY(barg1
, 0);
4108 PG_FREE_IF_COPY(barg2
, 1);
4111 PG_RETURN_BOOL(result
);
4115 bytealt(PG_FUNCTION_ARGS
)
4117 bytea
*arg1
= PG_GETARG_BYTEA_PP(0);
4118 bytea
*arg2
= PG_GETARG_BYTEA_PP(1);
4123 len1
= VARSIZE_ANY_EXHDR(arg1
);
4124 len2
= VARSIZE_ANY_EXHDR(arg2
);
4126 cmp
= memcmp(VARDATA_ANY(arg1
), VARDATA_ANY(arg2
), Min(len1
, len2
));
4128 PG_FREE_IF_COPY(arg1
, 0);
4129 PG_FREE_IF_COPY(arg2
, 1);
4131 PG_RETURN_BOOL((cmp
< 0) || ((cmp
== 0) && (len1
< len2
)));
4135 byteale(PG_FUNCTION_ARGS
)
4137 bytea
*arg1
= PG_GETARG_BYTEA_PP(0);
4138 bytea
*arg2
= PG_GETARG_BYTEA_PP(1);
4143 len1
= VARSIZE_ANY_EXHDR(arg1
);
4144 len2
= VARSIZE_ANY_EXHDR(arg2
);
4146 cmp
= memcmp(VARDATA_ANY(arg1
), VARDATA_ANY(arg2
), Min(len1
, len2
));
4148 PG_FREE_IF_COPY(arg1
, 0);
4149 PG_FREE_IF_COPY(arg2
, 1);
4151 PG_RETURN_BOOL((cmp
< 0) || ((cmp
== 0) && (len1
<= len2
)));
4155 byteagt(PG_FUNCTION_ARGS
)
4157 bytea
*arg1
= PG_GETARG_BYTEA_PP(0);
4158 bytea
*arg2
= PG_GETARG_BYTEA_PP(1);
4163 len1
= VARSIZE_ANY_EXHDR(arg1
);
4164 len2
= VARSIZE_ANY_EXHDR(arg2
);
4166 cmp
= memcmp(VARDATA_ANY(arg1
), VARDATA_ANY(arg2
), Min(len1
, len2
));
4168 PG_FREE_IF_COPY(arg1
, 0);
4169 PG_FREE_IF_COPY(arg2
, 1);
4171 PG_RETURN_BOOL((cmp
> 0) || ((cmp
== 0) && (len1
> len2
)));
4175 byteage(PG_FUNCTION_ARGS
)
4177 bytea
*arg1
= PG_GETARG_BYTEA_PP(0);
4178 bytea
*arg2
= PG_GETARG_BYTEA_PP(1);
4183 len1
= VARSIZE_ANY_EXHDR(arg1
);
4184 len2
= VARSIZE_ANY_EXHDR(arg2
);
4186 cmp
= memcmp(VARDATA_ANY(arg1
), VARDATA_ANY(arg2
), Min(len1
, len2
));
4188 PG_FREE_IF_COPY(arg1
, 0);
4189 PG_FREE_IF_COPY(arg2
, 1);
4191 PG_RETURN_BOOL((cmp
> 0) || ((cmp
== 0) && (len1
>= len2
)));
4195 byteacmp(PG_FUNCTION_ARGS
)
4197 bytea
*arg1
= PG_GETARG_BYTEA_PP(0);
4198 bytea
*arg2
= PG_GETARG_BYTEA_PP(1);
4203 len1
= VARSIZE_ANY_EXHDR(arg1
);
4204 len2
= VARSIZE_ANY_EXHDR(arg2
);
4206 cmp
= memcmp(VARDATA_ANY(arg1
), VARDATA_ANY(arg2
), Min(len1
, len2
));
4207 if ((cmp
== 0) && (len1
!= len2
))
4208 cmp
= (len1
< len2
) ? -1 : 1;
4210 PG_FREE_IF_COPY(arg1
, 0);
4211 PG_FREE_IF_COPY(arg2
, 1);
4213 PG_RETURN_INT32(cmp
);
4217 bytea_sortsupport(PG_FUNCTION_ARGS
)
4219 SortSupport ssup
= (SortSupport
) PG_GETARG_POINTER(0);
4220 MemoryContext oldcontext
;
4222 oldcontext
= MemoryContextSwitchTo(ssup
->ssup_cxt
);
4224 /* Use generic string SortSupport, forcing "C" collation */
4225 varstr_sortsupport(ssup
, BYTEAOID
, C_COLLATION_OID
);
4227 MemoryContextSwitchTo(oldcontext
);
4233 * appendStringInfoText
4235 * Append a text to str.
4236 * Like appendStringInfoString(str, text_to_cstring(t)) but faster.
4239 appendStringInfoText(StringInfo str
, const text
*t
)
4241 appendBinaryStringInfo(str
, VARDATA_ANY(t
), VARSIZE_ANY_EXHDR(t
));
4246 * replace all occurrences of 'old_sub_str' in 'orig_str'
4247 * with 'new_sub_str' to form 'new_str'
4249 * returns 'orig_str' if 'old_sub_str' == '' or 'orig_str' == ''
4250 * otherwise returns 'new_str'
4253 replace_text(PG_FUNCTION_ARGS
)
4255 text
*src_text
= PG_GETARG_TEXT_PP(0);
4256 text
*from_sub_text
= PG_GETARG_TEXT_PP(1);
4257 text
*to_sub_text
= PG_GETARG_TEXT_PP(2);
4259 int from_sub_text_len
;
4260 TextPositionState state
;
4268 src_text_len
= VARSIZE_ANY_EXHDR(src_text
);
4269 from_sub_text_len
= VARSIZE_ANY_EXHDR(from_sub_text
);
4271 /* Return unmodified source string if empty source or pattern */
4272 if (src_text_len
< 1 || from_sub_text_len
< 1)
4274 PG_RETURN_TEXT_P(src_text
);
4277 text_position_setup(src_text
, from_sub_text
, PG_GET_COLLATION(), &state
);
4279 found
= text_position_next(&state
);
4281 /* When the from_sub_text is not found, there is nothing to do. */
4284 text_position_cleanup(&state
);
4285 PG_RETURN_TEXT_P(src_text
);
4287 curr_ptr
= text_position_get_match_ptr(&state
);
4288 start_ptr
= VARDATA_ANY(src_text
);
4290 initStringInfo(&str
);
4294 CHECK_FOR_INTERRUPTS();
4296 /* copy the data skipped over by last text_position_next() */
4297 chunk_len
= curr_ptr
- start_ptr
;
4298 appendBinaryStringInfo(&str
, start_ptr
, chunk_len
);
4300 appendStringInfoText(&str
, to_sub_text
);
4302 start_ptr
= curr_ptr
+ from_sub_text_len
;
4304 found
= text_position_next(&state
);
4306 curr_ptr
= text_position_get_match_ptr(&state
);
4310 /* copy trailing data */
4311 chunk_len
= ((char *) src_text
+ VARSIZE_ANY(src_text
)) - start_ptr
;
4312 appendBinaryStringInfo(&str
, start_ptr
, chunk_len
);
4314 text_position_cleanup(&state
);
4316 ret_text
= cstring_to_text_with_len(str
.data
, str
.len
);
4319 PG_RETURN_TEXT_P(ret_text
);
4323 * check_replace_text_has_escape
4325 * Returns 0 if text contains no backslashes that need processing.
4326 * Returns 1 if text contains backslashes, but not regexp submatch specifiers.
4327 * Returns 2 if text contains regexp submatch specifiers (\1 .. \9).
4330 check_replace_text_has_escape(const text
*replace_text
)
4333 const char *p
= VARDATA_ANY(replace_text
);
4334 const char *p_end
= p
+ VARSIZE_ANY_EXHDR(replace_text
);
4338 /* Find next escape char, if any. */
4339 p
= memchr(p
, '\\', p_end
- p
);
4343 /* Note: a backslash at the end doesn't require extra processing. */
4346 if (*p
>= '1' && *p
<= '9')
4347 return 2; /* Found a submatch specifier, so done */
4348 result
= 1; /* Found some other sequence, keep looking */
4356 * appendStringInfoRegexpSubstr
4358 * Append replace_text to str, substituting regexp back references for
4359 * \n escapes. start_ptr is the start of the match in the source string,
4360 * at logical character position data_pos.
4363 appendStringInfoRegexpSubstr(StringInfo str
, text
*replace_text
,
4365 char *start_ptr
, int data_pos
)
4367 const char *p
= VARDATA_ANY(replace_text
);
4368 const char *p_end
= p
+ VARSIZE_ANY_EXHDR(replace_text
);
4372 const char *chunk_start
= p
;
4376 /* Find next escape char, if any. */
4377 p
= memchr(p
, '\\', p_end
- p
);
4381 /* Copy the text we just scanned over, if any. */
4382 if (p
> chunk_start
)
4383 appendBinaryStringInfo(str
, chunk_start
, p
- chunk_start
);
4385 /* Done if at end of string, else advance over escape char. */
4392 /* Escape at very end of input. Treat same as unexpected char */
4393 appendStringInfoChar(str
, '\\');
4397 if (*p
>= '1' && *p
<= '9')
4399 /* Use the back reference of regexp. */
4402 so
= pmatch
[idx
].rm_so
;
4403 eo
= pmatch
[idx
].rm_eo
;
4408 /* Use the entire matched string. */
4409 so
= pmatch
[0].rm_so
;
4410 eo
= pmatch
[0].rm_eo
;
4413 else if (*p
== '\\')
4415 /* \\ means transfer one \ to output. */
4416 appendStringInfoChar(str
, '\\');
4423 * If escape char is not followed by any expected char, just treat
4424 * it as ordinary data to copy. (XXX would it be better to throw
4427 appendStringInfoChar(str
, '\\');
4431 if (so
>= 0 && eo
>= 0)
4434 * Copy the text that is back reference of regexp. Note so and eo
4435 * are counted in characters not bytes.
4440 Assert(so
>= data_pos
);
4441 chunk_start
= start_ptr
;
4442 chunk_start
+= charlen_to_bytelen(chunk_start
, so
- data_pos
);
4443 chunk_len
= charlen_to_bytelen(chunk_start
, eo
- so
);
4444 appendBinaryStringInfo(str
, chunk_start
, chunk_len
);
4450 * replace_text_regexp
4452 * replace substring(s) in src_text that match pattern with replace_text.
4453 * The replace_text can contain backslash markers to substitute
4454 * (parts of) the matched text.
4456 * cflags: regexp compile flags.
4457 * collation: collation to use.
4458 * search_start: the character (not byte) offset in src_text at which to
4460 * n: if 0, replace all matches; if > 0, replace only the N'th match.
4463 replace_text_regexp(text
*src_text
, text
*pattern_text
,
4465 int cflags
, Oid collation
,
4466 int search_start
, int n
)
4470 int src_text_len
= VARSIZE_ANY_EXHDR(src_text
);
4473 regmatch_t pmatch
[10]; /* main match, plus \1 to \9 */
4474 int nmatch
= lengthof(pmatch
);
4481 initStringInfo(&buf
);
4483 /* Convert data string to wide characters. */
4484 data
= (pg_wchar
*) palloc((src_text_len
+ 1) * sizeof(pg_wchar
));
4485 data_len
= pg_mb2wchar_with_len(VARDATA_ANY(src_text
), data
, src_text_len
);
4487 /* Check whether replace_text has escapes, especially regexp submatches. */
4488 escape_status
= check_replace_text_has_escape(replace_text
);
4490 /* If no regexp submatches, we can use REG_NOSUB. */
4491 if (escape_status
< 2)
4493 cflags
|= REG_NOSUB
;
4494 /* Also tell pg_regexec we only want the whole-match location. */
4498 /* Prepare the regexp. */
4499 re
= RE_compile_and_cache(pattern_text
, cflags
, collation
);
4501 /* start_ptr points to the data_pos'th character of src_text */
4502 start_ptr
= (char *) VARDATA_ANY(src_text
);
4505 while (search_start
<= data_len
)
4509 CHECK_FOR_INTERRUPTS();
4511 regexec_result
= pg_regexec(re
,
4515 NULL
, /* no details */
4520 if (regexec_result
== REG_NOMATCH
)
4523 if (regexec_result
!= REG_OKAY
)
4527 CHECK_FOR_INTERRUPTS();
4528 pg_regerror(regexec_result
, re
, errMsg
, sizeof(errMsg
));
4530 (errcode(ERRCODE_INVALID_REGULAR_EXPRESSION
),
4531 errmsg("regular expression failed: %s", errMsg
)));
4535 * Count matches, and decide whether to replace this match.
4538 if (n
> 0 && nmatches
!= n
)
4541 * No, so advance search_start, but not start_ptr/data_pos. (Thus,
4542 * we treat the matched text as if it weren't matched, and copy it
4543 * to the output later.)
4545 search_start
= pmatch
[0].rm_eo
;
4546 if (pmatch
[0].rm_so
== pmatch
[0].rm_eo
)
4552 * Copy the text to the left of the match position. Note we are given
4553 * character not byte indexes.
4555 if (pmatch
[0].rm_so
- data_pos
> 0)
4559 chunk_len
= charlen_to_bytelen(start_ptr
,
4560 pmatch
[0].rm_so
- data_pos
);
4561 appendBinaryStringInfo(&buf
, start_ptr
, chunk_len
);
4564 * Advance start_ptr over that text, to avoid multiple rescans of
4565 * it if the replace_text contains multiple back-references.
4567 start_ptr
+= chunk_len
;
4568 data_pos
= pmatch
[0].rm_so
;
4572 * Copy the replace_text, processing escapes if any are present.
4574 if (escape_status
> 0)
4575 appendStringInfoRegexpSubstr(&buf
, replace_text
, pmatch
,
4576 start_ptr
, data_pos
);
4578 appendStringInfoText(&buf
, replace_text
);
4580 /* Advance start_ptr and data_pos over the matched text. */
4581 start_ptr
+= charlen_to_bytelen(start_ptr
,
4582 pmatch
[0].rm_eo
- data_pos
);
4583 data_pos
= pmatch
[0].rm_eo
;
4586 * If we only want to replace one occurrence, we're done.
4592 * Advance search position. Normally we start the next search at the
4593 * end of the previous match; but if the match was of zero length, we
4594 * have to advance by one character, or we'd just find the same match
4597 search_start
= data_pos
;
4598 if (pmatch
[0].rm_so
== pmatch
[0].rm_eo
)
4603 * Copy the text to the right of the last match.
4605 if (data_pos
< data_len
)
4609 chunk_len
= ((char *) src_text
+ VARSIZE_ANY(src_text
)) - start_ptr
;
4610 appendBinaryStringInfo(&buf
, start_ptr
, chunk_len
);
4613 ret_text
= cstring_to_text_with_len(buf
.data
, buf
.len
);
4622 * parse input string based on provided field separator
4623 * return N'th item (1 based, negative counts from end)
4626 split_part(PG_FUNCTION_ARGS
)
4628 text
*inputstring
= PG_GETARG_TEXT_PP(0);
4629 text
*fldsep
= PG_GETARG_TEXT_PP(1);
4630 int fldnum
= PG_GETARG_INT32(2);
4631 int inputstring_len
;
4633 TextPositionState state
;
4639 /* field number is 1 based */
4642 (errcode(ERRCODE_INVALID_PARAMETER_VALUE
),
4643 errmsg("field position must not be zero")));
4645 inputstring_len
= VARSIZE_ANY_EXHDR(inputstring
);
4646 fldsep_len
= VARSIZE_ANY_EXHDR(fldsep
);
4648 /* return empty string for empty input string */
4649 if (inputstring_len
< 1)
4650 PG_RETURN_TEXT_P(cstring_to_text(""));
4652 /* handle empty field separator */
4655 /* if first or last field, return input string, else empty string */
4656 if (fldnum
== 1 || fldnum
== -1)
4657 PG_RETURN_TEXT_P(inputstring
);
4659 PG_RETURN_TEXT_P(cstring_to_text(""));
4662 /* find the first field separator */
4663 text_position_setup(inputstring
, fldsep
, PG_GET_COLLATION(), &state
);
4665 found
= text_position_next(&state
);
4667 /* special case if fldsep not found at all */
4670 text_position_cleanup(&state
);
4671 /* if first or last field, return input string, else empty string */
4672 if (fldnum
== 1 || fldnum
== -1)
4673 PG_RETURN_TEXT_P(inputstring
);
4675 PG_RETURN_TEXT_P(cstring_to_text(""));
4679 * take care of a negative field number (i.e. count from the right) by
4680 * converting to a positive field number; we need total number of fields
4684 /* we found a fldsep, so there are at least two fields */
4687 while (text_position_next(&state
))
4690 /* special case of last field does not require an extra pass */
4693 start_ptr
= text_position_get_match_ptr(&state
) + fldsep_len
;
4694 end_ptr
= VARDATA_ANY(inputstring
) + inputstring_len
;
4695 text_position_cleanup(&state
);
4696 PG_RETURN_TEXT_P(cstring_to_text_with_len(start_ptr
,
4697 end_ptr
- start_ptr
));
4700 /* else, convert fldnum to positive notation */
4701 fldnum
+= numfields
+ 1;
4703 /* if nonexistent field, return empty string */
4706 text_position_cleanup(&state
);
4707 PG_RETURN_TEXT_P(cstring_to_text(""));
4710 /* reset to pointing at first match, but now with positive fldnum */
4711 text_position_reset(&state
);
4712 found
= text_position_next(&state
);
4716 /* identify bounds of first field */
4717 start_ptr
= VARDATA_ANY(inputstring
);
4718 end_ptr
= text_position_get_match_ptr(&state
);
4720 while (found
&& --fldnum
> 0)
4722 /* identify bounds of next field */
4723 start_ptr
= end_ptr
+ fldsep_len
;
4724 found
= text_position_next(&state
);
4726 end_ptr
= text_position_get_match_ptr(&state
);
4729 text_position_cleanup(&state
);
4733 /* N'th field separator not found */
4734 /* if last field requested, return it, else empty string */
4737 int last_len
= start_ptr
- VARDATA_ANY(inputstring
);
4739 result_text
= cstring_to_text_with_len(start_ptr
,
4740 inputstring_len
- last_len
);
4743 result_text
= cstring_to_text("");
4747 /* non-last field requested */
4748 result_text
= cstring_to_text_with_len(start_ptr
, end_ptr
- start_ptr
);
4751 PG_RETURN_TEXT_P(result_text
);
4755 * Convenience function to return true when two text params are equal.
4758 text_isequal(text
*txt1
, text
*txt2
, Oid collid
)
4760 return DatumGetBool(DirectFunctionCall2Coll(texteq
,
4762 PointerGetDatum(txt1
),
4763 PointerGetDatum(txt2
)));
4768 * parse input string and return text array of elements,
4769 * based on provided field separator
4772 text_to_array(PG_FUNCTION_ARGS
)
4774 SplitTextOutputData tstate
;
4776 /* For array output, tstate should start as all zeroes */
4777 memset(&tstate
, 0, sizeof(tstate
));
4779 if (!split_text(fcinfo
, &tstate
))
4782 if (tstate
.astate
== NULL
)
4783 PG_RETURN_ARRAYTYPE_P(construct_empty_array(TEXTOID
));
4785 PG_RETURN_DATUM(makeArrayResult(tstate
.astate
,
4786 CurrentMemoryContext
));
4790 * text_to_array_null
4791 * parse input string and return text array of elements,
4792 * based on provided field separator and null string
4794 * This is a separate entry point only to prevent the regression tests from
4795 * complaining about different argument sets for the same internal function.
4798 text_to_array_null(PG_FUNCTION_ARGS
)
4800 return text_to_array(fcinfo
);
4805 * parse input string and return table of elements,
4806 * based on provided field separator
4809 text_to_table(PG_FUNCTION_ARGS
)
4811 ReturnSetInfo
*rsi
= (ReturnSetInfo
*) fcinfo
->resultinfo
;
4812 SplitTextOutputData tstate
;
4814 tstate
.astate
= NULL
;
4815 InitMaterializedSRF(fcinfo
, MAT_SRF_USE_EXPECTED_DESC
);
4816 tstate
.tupstore
= rsi
->setResult
;
4817 tstate
.tupdesc
= rsi
->setDesc
;
4819 (void) split_text(fcinfo
, &tstate
);
4825 * text_to_table_null
4826 * parse input string and return table of elements,
4827 * based on provided field separator and null string
4829 * This is a separate entry point only to prevent the regression tests from
4830 * complaining about different argument sets for the same internal function.
4833 text_to_table_null(PG_FUNCTION_ARGS
)
4835 return text_to_table(fcinfo
);
4839 * Common code for text_to_array, text_to_array_null, text_to_table
4840 * and text_to_table_null functions.
4842 * These are not strict so we have to test for null inputs explicitly.
4843 * Returns false if result is to be null, else returns true.
4845 * Note that if the result is valid but empty (zero elements), we return
4846 * without changing *tstate --- caller must handle that case, too.
4849 split_text(FunctionCallInfo fcinfo
, SplitTextOutputData
*tstate
)
4854 Oid collation
= PG_GET_COLLATION();
4855 int inputstring_len
;
4860 /* when input string is NULL, then result is NULL too */
4861 if (PG_ARGISNULL(0))
4864 inputstring
= PG_GETARG_TEXT_PP(0);
4866 /* fldsep can be NULL */
4867 if (!PG_ARGISNULL(1))
4868 fldsep
= PG_GETARG_TEXT_PP(1);
4872 /* null_string can be NULL or omitted */
4873 if (PG_NARGS() > 2 && !PG_ARGISNULL(2))
4874 null_string
= PG_GETARG_TEXT_PP(2);
4881 * Normal case with non-null fldsep. Use the text_position machinery
4882 * to search for occurrences of fldsep.
4884 TextPositionState state
;
4886 inputstring_len
= VARSIZE_ANY_EXHDR(inputstring
);
4887 fldsep_len
= VARSIZE_ANY_EXHDR(fldsep
);
4889 /* return empty set for empty input string */
4890 if (inputstring_len
< 1)
4893 /* empty field separator: return input string as a one-element set */
4896 split_text_accum_result(tstate
, inputstring
,
4897 null_string
, collation
);
4901 text_position_setup(inputstring
, fldsep
, collation
, &state
);
4903 start_ptr
= VARDATA_ANY(inputstring
);
4911 CHECK_FOR_INTERRUPTS();
4913 found
= text_position_next(&state
);
4916 /* fetch last field */
4917 chunk_len
= ((char *) inputstring
+ VARSIZE_ANY(inputstring
)) - start_ptr
;
4918 end_ptr
= NULL
; /* not used, but some compilers complain */
4922 /* fetch non-last field */
4923 end_ptr
= text_position_get_match_ptr(&state
);
4924 chunk_len
= end_ptr
- start_ptr
;
4927 /* build a temp text datum to pass to split_text_accum_result */
4928 result_text
= cstring_to_text_with_len(start_ptr
, chunk_len
);
4930 /* stash away this field */
4931 split_text_accum_result(tstate
, result_text
,
4932 null_string
, collation
);
4939 start_ptr
= end_ptr
+ fldsep_len
;
4942 text_position_cleanup(&state
);
4947 * When fldsep is NULL, each character in the input string becomes a
4948 * separate element in the result set. The separator is effectively
4949 * the space between characters.
4951 inputstring_len
= VARSIZE_ANY_EXHDR(inputstring
);
4953 start_ptr
= VARDATA_ANY(inputstring
);
4955 while (inputstring_len
> 0)
4957 int chunk_len
= pg_mblen(start_ptr
);
4959 CHECK_FOR_INTERRUPTS();
4961 /* build a temp text datum to pass to split_text_accum_result */
4962 result_text
= cstring_to_text_with_len(start_ptr
, chunk_len
);
4964 /* stash away this field */
4965 split_text_accum_result(tstate
, result_text
,
4966 null_string
, collation
);
4970 start_ptr
+= chunk_len
;
4971 inputstring_len
-= chunk_len
;
4979 * Add text item to result set (table or array).
4981 * This is also responsible for checking to see if the item matches
4982 * the null_string, in which case we should emit NULL instead.
4985 split_text_accum_result(SplitTextOutputData
*tstate
,
4990 bool is_null
= false;
4992 if (null_string
&& text_isequal(field_value
, null_string
, collation
))
4995 if (tstate
->tupstore
)
5000 values
[0] = PointerGetDatum(field_value
);
5003 tuplestore_putvalues(tstate
->tupstore
,
5010 tstate
->astate
= accumArrayResult(tstate
->astate
,
5011 PointerGetDatum(field_value
),
5014 CurrentMemoryContext
);
5020 * concatenate Cstring representation of input array elements
5021 * using provided field separator
5024 array_to_text(PG_FUNCTION_ARGS
)
5026 ArrayType
*v
= PG_GETARG_ARRAYTYPE_P(0);
5027 char *fldsep
= text_to_cstring(PG_GETARG_TEXT_PP(1));
5029 PG_RETURN_TEXT_P(array_to_text_internal(fcinfo
, v
, fldsep
, NULL
));
5033 * array_to_text_null
5034 * concatenate Cstring representation of input array elements
5035 * using provided field separator and null string
5037 * This version is not strict so we have to test for null inputs explicitly.
5040 array_to_text_null(PG_FUNCTION_ARGS
)
5046 /* returns NULL when first or second parameter is NULL */
5047 if (PG_ARGISNULL(0) || PG_ARGISNULL(1))
5050 v
= PG_GETARG_ARRAYTYPE_P(0);
5051 fldsep
= text_to_cstring(PG_GETARG_TEXT_PP(1));
5053 /* NULL null string is passed through as a null pointer */
5054 if (!PG_ARGISNULL(2))
5055 null_string
= text_to_cstring(PG_GETARG_TEXT_PP(2));
5059 PG_RETURN_TEXT_P(array_to_text_internal(fcinfo
, v
, fldsep
, null_string
));
5063 * common code for array_to_text and array_to_text_null functions
5066 array_to_text_internal(FunctionCallInfo fcinfo
, ArrayType
*v
,
5067 const char *fldsep
, const char *null_string
)
5078 bool printed
= false;
5083 ArrayMetaState
*my_extra
;
5085 ndims
= ARR_NDIM(v
);
5087 nitems
= ArrayGetNItems(ndims
, dims
);
5089 /* if there are no elements, return an empty string */
5091 return cstring_to_text_with_len("", 0);
5093 element_type
= ARR_ELEMTYPE(v
);
5094 initStringInfo(&buf
);
5097 * We arrange to look up info about element type, including its output
5098 * conversion proc, only once per series of calls, assuming the element
5099 * type doesn't change underneath us.
5101 my_extra
= (ArrayMetaState
*) fcinfo
->flinfo
->fn_extra
;
5102 if (my_extra
== NULL
)
5104 fcinfo
->flinfo
->fn_extra
= MemoryContextAlloc(fcinfo
->flinfo
->fn_mcxt
,
5105 sizeof(ArrayMetaState
));
5106 my_extra
= (ArrayMetaState
*) fcinfo
->flinfo
->fn_extra
;
5107 my_extra
->element_type
= ~element_type
;
5110 if (my_extra
->element_type
!= element_type
)
5113 * Get info about element type, including its output conversion proc
5115 get_type_io_data(element_type
, IOFunc_output
,
5116 &my_extra
->typlen
, &my_extra
->typbyval
,
5117 &my_extra
->typalign
, &my_extra
->typdelim
,
5118 &my_extra
->typioparam
, &my_extra
->typiofunc
);
5119 fmgr_info_cxt(my_extra
->typiofunc
, &my_extra
->proc
,
5120 fcinfo
->flinfo
->fn_mcxt
);
5121 my_extra
->element_type
= element_type
;
5123 typlen
= my_extra
->typlen
;
5124 typbyval
= my_extra
->typbyval
;
5125 typalign
= my_extra
->typalign
;
5127 p
= ARR_DATA_PTR(v
);
5128 bitmap
= ARR_NULLBITMAP(v
);
5131 for (i
= 0; i
< nitems
; i
++)
5136 /* Get source element, checking for NULL */
5137 if (bitmap
&& (*bitmap
& bitmask
) == 0)
5139 /* if null_string is NULL, we just ignore null elements */
5140 if (null_string
!= NULL
)
5143 appendStringInfo(&buf
, "%s%s", fldsep
, null_string
);
5145 appendStringInfoString(&buf
, null_string
);
5151 itemvalue
= fetch_att(p
, typbyval
, typlen
);
5153 value
= OutputFunctionCall(&my_extra
->proc
, itemvalue
);
5156 appendStringInfo(&buf
, "%s%s", fldsep
, value
);
5158 appendStringInfoString(&buf
, value
);
5161 p
= att_addlength_pointer(p
, typlen
, p
);
5162 p
= (char *) att_align_nominal(p
, typalign
);
5165 /* advance bitmap pointer if any */
5169 if (bitmask
== 0x100)
5177 result
= cstring_to_text_with_len(buf
.data
, buf
.len
);
5185 * Convert an int32 to a string containing a base 16 (hex) representation of
5189 to_hex32(PG_FUNCTION_ARGS
)
5191 uint32 value
= (uint32
) PG_GETARG_INT32(0);
5193 const char *digits
= "0123456789abcdef";
5194 char buf
[32]; /* bigger than needed, but reasonable */
5196 ptr
= buf
+ sizeof(buf
) - 1;
5201 *--ptr
= digits
[value
% HEXBASE
];
5203 } while (ptr
> buf
&& value
);
5205 PG_RETURN_TEXT_P(cstring_to_text(ptr
));
5209 * Convert an int64 to a string containing a base 16 (hex) representation of
5213 to_hex64(PG_FUNCTION_ARGS
)
5215 uint64 value
= (uint64
) PG_GETARG_INT64(0);
5217 const char *digits
= "0123456789abcdef";
5218 char buf
[32]; /* bigger than needed, but reasonable */
5220 ptr
= buf
+ sizeof(buf
) - 1;
5225 *--ptr
= digits
[value
% HEXBASE
];
5227 } while (ptr
> buf
&& value
);
5229 PG_RETURN_TEXT_P(cstring_to_text(ptr
));
5233 * Return the size of a datum, possibly compressed
5235 * Works on any data type
5238 pg_column_size(PG_FUNCTION_ARGS
)
5240 Datum value
= PG_GETARG_DATUM(0);
5244 /* On first call, get the input type's typlen, and save at *fn_extra */
5245 if (fcinfo
->flinfo
->fn_extra
== NULL
)
5247 /* Lookup the datatype of the supplied argument */
5248 Oid argtypeid
= get_fn_expr_argtype(fcinfo
->flinfo
, 0);
5250 typlen
= get_typlen(argtypeid
);
5251 if (typlen
== 0) /* should not happen */
5252 elog(ERROR
, "cache lookup failed for type %u", argtypeid
);
5254 fcinfo
->flinfo
->fn_extra
= MemoryContextAlloc(fcinfo
->flinfo
->fn_mcxt
,
5256 *((int *) fcinfo
->flinfo
->fn_extra
) = typlen
;
5259 typlen
= *((int *) fcinfo
->flinfo
->fn_extra
);
5263 /* varlena type, possibly toasted */
5264 result
= toast_datum_size(value
);
5266 else if (typlen
== -2)
5269 result
= strlen(DatumGetCString(value
)) + 1;
5273 /* ordinary fixed-width type */
5277 PG_RETURN_INT32(result
);
5281 * Return the compression method stored in the compressed attribute. Return
5282 * NULL for non varlena type or uncompressed data.
5285 pg_column_compression(PG_FUNCTION_ARGS
)
5289 ToastCompressionId cmid
;
5291 /* On first call, get the input type's typlen, and save at *fn_extra */
5292 if (fcinfo
->flinfo
->fn_extra
== NULL
)
5294 /* Lookup the datatype of the supplied argument */
5295 Oid argtypeid
= get_fn_expr_argtype(fcinfo
->flinfo
, 0);
5297 typlen
= get_typlen(argtypeid
);
5298 if (typlen
== 0) /* should not happen */
5299 elog(ERROR
, "cache lookup failed for type %u", argtypeid
);
5301 fcinfo
->flinfo
->fn_extra
= MemoryContextAlloc(fcinfo
->flinfo
->fn_mcxt
,
5303 *((int *) fcinfo
->flinfo
->fn_extra
) = typlen
;
5306 typlen
= *((int *) fcinfo
->flinfo
->fn_extra
);
5311 /* get the compression method id stored in the compressed varlena */
5312 cmid
= toast_get_compression_id((struct varlena
*)
5313 DatumGetPointer(PG_GETARG_DATUM(0)));
5314 if (cmid
== TOAST_INVALID_COMPRESSION_ID
)
5317 /* convert compression method id to compression method name */
5320 case TOAST_PGLZ_COMPRESSION_ID
:
5323 case TOAST_LZ4_COMPRESSION_ID
:
5327 elog(ERROR
, "invalid compression method id %d", cmid
);
5330 PG_RETURN_TEXT_P(cstring_to_text(result
));
5334 * string_agg - Concatenates values and returns string.
5336 * Syntax: string_agg(value text, delimiter text) RETURNS text
5338 * Note: Any NULL values are ignored. The first-call delimiter isn't
5339 * actually used at all, and on subsequent calls the delimiter precedes
5340 * the associated value.
5343 /* subroutine to initialize state */
5345 makeStringAggState(FunctionCallInfo fcinfo
)
5348 MemoryContext aggcontext
;
5349 MemoryContext oldcontext
;
5351 if (!AggCheckCallContext(fcinfo
, &aggcontext
))
5353 /* cannot be called directly because of internal-type argument */
5354 elog(ERROR
, "string_agg_transfn called in non-aggregate context");
5358 * Create state in aggregate context. It'll stay there across subsequent
5361 oldcontext
= MemoryContextSwitchTo(aggcontext
);
5362 state
= makeStringInfo();
5363 MemoryContextSwitchTo(oldcontext
);
5369 string_agg_transfn(PG_FUNCTION_ARGS
)
5373 state
= PG_ARGISNULL(0) ? NULL
: (StringInfo
) PG_GETARG_POINTER(0);
5375 /* Append the value unless null. */
5376 if (!PG_ARGISNULL(1))
5378 /* On the first time through, we ignore the delimiter. */
5380 state
= makeStringAggState(fcinfo
);
5381 else if (!PG_ARGISNULL(2))
5382 appendStringInfoText(state
, PG_GETARG_TEXT_PP(2)); /* delimiter */
5384 appendStringInfoText(state
, PG_GETARG_TEXT_PP(1)); /* value */
5388 * The transition type for string_agg() is declared to be "internal",
5389 * which is a pass-by-value type the same size as a pointer.
5391 PG_RETURN_POINTER(state
);
5395 string_agg_finalfn(PG_FUNCTION_ARGS
)
5399 /* cannot be called directly because of internal-type argument */
5400 Assert(AggCheckCallContext(fcinfo
, NULL
));
5402 state
= PG_ARGISNULL(0) ? NULL
: (StringInfo
) PG_GETARG_POINTER(0);
5405 PG_RETURN_TEXT_P(cstring_to_text_with_len(state
->data
, state
->len
));
5411 * Prepare cache with fmgr info for the output functions of the datatypes of
5412 * the arguments of a concat-like function, beginning with argument "argidx".
5413 * (Arguments before that will have corresponding slots in the resulting
5414 * FmgrInfo array, but we don't fill those slots.)
5417 build_concat_foutcache(FunctionCallInfo fcinfo
, int argidx
)
5419 FmgrInfo
*foutcache
;
5422 /* We keep the info in fn_mcxt so it survives across calls */
5423 foutcache
= (FmgrInfo
*) MemoryContextAlloc(fcinfo
->flinfo
->fn_mcxt
,
5424 PG_NARGS() * sizeof(FmgrInfo
));
5426 for (i
= argidx
; i
< PG_NARGS(); i
++)
5432 valtype
= get_fn_expr_argtype(fcinfo
->flinfo
, i
);
5433 if (!OidIsValid(valtype
))
5434 elog(ERROR
, "could not determine data type of concat() input");
5436 getTypeOutputInfo(valtype
, &typOutput
, &typIsVarlena
);
5437 fmgr_info_cxt(typOutput
, &foutcache
[i
], fcinfo
->flinfo
->fn_mcxt
);
5440 fcinfo
->flinfo
->fn_extra
= foutcache
;
5446 * Implementation of both concat() and concat_ws().
5448 * sepstr is the separator string to place between values.
5449 * argidx identifies the first argument to concatenate (counting from zero);
5450 * note that this must be constant across any one series of calls.
5452 * Returns NULL if result should be NULL, else text value.
5455 concat_internal(const char *sepstr
, int argidx
,
5456 FunctionCallInfo fcinfo
)
5460 FmgrInfo
*foutcache
;
5461 bool first_arg
= true;
5465 * concat(VARIADIC some-array) is essentially equivalent to
5466 * array_to_text(), ie concat the array elements with the given separator.
5467 * So we just pass the case off to that code.
5469 if (get_fn_expr_variadic(fcinfo
->flinfo
))
5473 /* Should have just the one argument */
5474 Assert(argidx
== PG_NARGS() - 1);
5476 /* concat(VARIADIC NULL) is defined as NULL */
5477 if (PG_ARGISNULL(argidx
))
5481 * Non-null argument had better be an array. We assume that any call
5482 * context that could let get_fn_expr_variadic return true will have
5483 * checked that a VARIADIC-labeled parameter actually is an array. So
5484 * it should be okay to just Assert that it's an array rather than
5485 * doing a full-fledged error check.
5487 Assert(OidIsValid(get_base_element_type(get_fn_expr_argtype(fcinfo
->flinfo
, argidx
))));
5489 /* OK, safe to fetch the array value */
5490 arr
= PG_GETARG_ARRAYTYPE_P(argidx
);
5493 * And serialize the array. We tell array_to_text to ignore null
5494 * elements, which matches the behavior of the loop below.
5496 return array_to_text_internal(fcinfo
, arr
, sepstr
, NULL
);
5499 /* Normal case without explicit VARIADIC marker */
5500 initStringInfo(&str
);
5502 /* Get output function info, building it if first time through */
5503 foutcache
= (FmgrInfo
*) fcinfo
->flinfo
->fn_extra
;
5504 if (foutcache
== NULL
)
5505 foutcache
= build_concat_foutcache(fcinfo
, argidx
);
5507 for (i
= argidx
; i
< PG_NARGS(); i
++)
5509 if (!PG_ARGISNULL(i
))
5511 Datum value
= PG_GETARG_DATUM(i
);
5513 /* add separator if appropriate */
5517 appendStringInfoString(&str
, sepstr
);
5519 /* call the appropriate type output function, append the result */
5520 appendStringInfoString(&str
,
5521 OutputFunctionCall(&foutcache
[i
], value
));
5525 result
= cstring_to_text_with_len(str
.data
, str
.len
);
5532 * Concatenate all arguments. NULL arguments are ignored.
5535 text_concat(PG_FUNCTION_ARGS
)
5539 result
= concat_internal("", 0, fcinfo
);
5542 PG_RETURN_TEXT_P(result
);
5546 * Concatenate all but first argument value with separators. The first
5547 * parameter is used as the separator. NULL arguments are ignored.
5550 text_concat_ws(PG_FUNCTION_ARGS
)
5555 /* return NULL when separator is NULL */
5556 if (PG_ARGISNULL(0))
5558 sep
= text_to_cstring(PG_GETARG_TEXT_PP(0));
5560 result
= concat_internal(sep
, 1, fcinfo
);
5563 PG_RETURN_TEXT_P(result
);
5567 * Return first n characters in the string. When n is negative,
5568 * return all but last |n| characters.
5571 text_left(PG_FUNCTION_ARGS
)
5573 int n
= PG_GETARG_INT32(1);
5577 text
*str
= PG_GETARG_TEXT_PP(0);
5578 const char *p
= VARDATA_ANY(str
);
5579 int len
= VARSIZE_ANY_EXHDR(str
);
5582 n
= pg_mbstrlen_with_len(p
, len
) + n
;
5583 rlen
= pg_mbcharcliplen(p
, len
, n
);
5584 PG_RETURN_TEXT_P(cstring_to_text_with_len(p
, rlen
));
5587 PG_RETURN_TEXT_P(text_substring(PG_GETARG_DATUM(0), 1, n
, false));
5591 * Return last n characters in the string. When n is negative,
5592 * return all but first |n| characters.
5595 text_right(PG_FUNCTION_ARGS
)
5597 text
*str
= PG_GETARG_TEXT_PP(0);
5598 const char *p
= VARDATA_ANY(str
);
5599 int len
= VARSIZE_ANY_EXHDR(str
);
5600 int n
= PG_GETARG_INT32(1);
5606 n
= pg_mbstrlen_with_len(p
, len
) - n
;
5607 off
= pg_mbcharcliplen(p
, len
, n
);
5609 PG_RETURN_TEXT_P(cstring_to_text_with_len(p
+ off
, len
- off
));
5613 * Return reversed string
5616 text_reverse(PG_FUNCTION_ARGS
)
5618 text
*str
= PG_GETARG_TEXT_PP(0);
5619 const char *p
= VARDATA_ANY(str
);
5620 int len
= VARSIZE_ANY_EXHDR(str
);
5621 const char *endp
= p
+ len
;
5625 result
= palloc(len
+ VARHDRSZ
);
5626 dst
= (char *) VARDATA(result
) + len
;
5627 SET_VARSIZE(result
, len
+ VARHDRSZ
);
5629 if (pg_database_encoding_max_length() > 1)
5631 /* multibyte version */
5644 /* single byte version */
5649 PG_RETURN_TEXT_P(result
);
5654 * Support macros for text_format()
5656 #define TEXT_FORMAT_FLAG_MINUS 0x0001 /* is minus flag present? */
5658 #define ADVANCE_PARSE_POINTER(ptr,end_ptr) \
5660 if (++(ptr) >= (end_ptr)) \
5662 (errcode(ERRCODE_INVALID_PARAMETER_VALUE), \
5663 errmsg("unterminated format() type specifier"), \
5664 errhint("For a single \"%%\" use \"%%%%\"."))); \
5668 * Returns a formatted string
5671 text_format(PG_FUNCTION_ARGS
)
5676 const char *start_ptr
;
5677 const char *end_ptr
;
5682 Datum
*elements
= NULL
;
5684 Oid element_type
= InvalidOid
;
5685 Oid prev_type
= InvalidOid
;
5686 Oid prev_width_type
= InvalidOid
;
5687 FmgrInfo typoutputfinfo
;
5688 FmgrInfo typoutputinfo_width
;
5690 /* When format string is null, immediately return null */
5691 if (PG_ARGISNULL(0))
5694 /* If argument is marked VARIADIC, expand array into elements */
5695 if (get_fn_expr_variadic(fcinfo
->flinfo
))
5703 /* Should have just the one argument */
5704 Assert(PG_NARGS() == 2);
5706 /* If argument is NULL, we treat it as zero-length array */
5707 if (PG_ARGISNULL(1))
5712 * Non-null argument had better be an array. We assume that any
5713 * call context that could let get_fn_expr_variadic return true
5714 * will have checked that a VARIADIC-labeled parameter actually is
5715 * an array. So it should be okay to just Assert that it's an
5716 * array rather than doing a full-fledged error check.
5718 Assert(OidIsValid(get_base_element_type(get_fn_expr_argtype(fcinfo
->flinfo
, 1))));
5720 /* OK, safe to fetch the array value */
5721 arr
= PG_GETARG_ARRAYTYPE_P(1);
5723 /* Get info about array element type */
5724 element_type
= ARR_ELEMTYPE(arr
);
5725 get_typlenbyvalalign(element_type
,
5726 &elmlen
, &elmbyval
, &elmalign
);
5728 /* Extract all array elements */
5729 deconstruct_array(arr
, element_type
, elmlen
, elmbyval
, elmalign
,
5730 &elements
, &nulls
, &nitems
);
5734 funcvariadic
= true;
5738 /* Non-variadic case, we'll process the arguments individually */
5740 funcvariadic
= false;
5743 /* Setup for main loop. */
5744 fmt
= PG_GETARG_TEXT_PP(0);
5745 start_ptr
= VARDATA_ANY(fmt
);
5746 end_ptr
= start_ptr
+ VARSIZE_ANY_EXHDR(fmt
);
5747 initStringInfo(&str
);
5748 arg
= 1; /* next argument position to print */
5750 /* Scan format string, looking for conversion specifiers. */
5751 for (cp
= start_ptr
; cp
< end_ptr
; cp
++)
5762 * If it's not the start of a conversion specifier, just copy it to
5763 * the output buffer.
5767 appendStringInfoCharMacro(&str
, *cp
);
5771 ADVANCE_PARSE_POINTER(cp
, end_ptr
);
5773 /* Easy case: %% outputs a single % */
5776 appendStringInfoCharMacro(&str
, *cp
);
5780 /* Parse the optional portions of the format specifier */
5781 cp
= text_format_parse_format(cp
, end_ptr
,
5786 * Next we should see the main conversion specifier. Whether or not
5787 * an argument position was present, it's known that at least one
5788 * character remains in the string at this point. Experience suggests
5789 * that it's worth checking that that character is one of the expected
5790 * ones before we try to fetch arguments, so as to produce the least
5791 * confusing response to a mis-formatted specifier.
5793 if (strchr("sIL", *cp
) == NULL
)
5795 (errcode(ERRCODE_INVALID_PARAMETER_VALUE
),
5796 errmsg("unrecognized format() type specifier \"%.*s\"",
5798 errhint("For a single \"%%\" use \"%%%%\".")));
5800 /* If indirect width was specified, get its value */
5803 /* Collect the specified or next argument position */
5808 (errcode(ERRCODE_INVALID_PARAMETER_VALUE
),
5809 errmsg("too few arguments for format()")));
5811 /* Get the value and type of the selected argument */
5814 value
= PG_GETARG_DATUM(arg
);
5815 isNull
= PG_ARGISNULL(arg
);
5816 typid
= get_fn_expr_argtype(fcinfo
->flinfo
, arg
);
5820 value
= elements
[arg
- 1];
5821 isNull
= nulls
[arg
- 1];
5822 typid
= element_type
;
5824 if (!OidIsValid(typid
))
5825 elog(ERROR
, "could not determine data type of format() input");
5829 /* We can treat NULL width the same as zero */
5832 else if (typid
== INT4OID
)
5833 width
= DatumGetInt32(value
);
5834 else if (typid
== INT2OID
)
5835 width
= DatumGetInt16(value
);
5838 /* For less-usual datatypes, convert to text then to int */
5841 if (typid
!= prev_width_type
)
5846 getTypeOutputInfo(typid
, &typoutputfunc
, &typIsVarlena
);
5847 fmgr_info(typoutputfunc
, &typoutputinfo_width
);
5848 prev_width_type
= typid
;
5851 str
= OutputFunctionCall(&typoutputinfo_width
, value
);
5853 /* pg_strtoint32 will complain about bad data or overflow */
5854 width
= pg_strtoint32(str
);
5860 /* Collect the specified or next argument position */
5865 (errcode(ERRCODE_INVALID_PARAMETER_VALUE
),
5866 errmsg("too few arguments for format()")));
5868 /* Get the value and type of the selected argument */
5871 value
= PG_GETARG_DATUM(arg
);
5872 isNull
= PG_ARGISNULL(arg
);
5873 typid
= get_fn_expr_argtype(fcinfo
->flinfo
, arg
);
5877 value
= elements
[arg
- 1];
5878 isNull
= nulls
[arg
- 1];
5879 typid
= element_type
;
5881 if (!OidIsValid(typid
))
5882 elog(ERROR
, "could not determine data type of format() input");
5887 * Get the appropriate typOutput function, reusing previous one if
5888 * same type as previous argument. That's particularly useful in the
5889 * variadic-array case, but often saves work even for ordinary calls.
5891 if (typid
!= prev_type
)
5896 getTypeOutputInfo(typid
, &typoutputfunc
, &typIsVarlena
);
5897 fmgr_info(typoutputfunc
, &typoutputfinfo
);
5902 * And now we can format the value.
5909 text_format_string_conversion(&str
, *cp
, &typoutputfinfo
,
5914 /* should not get here, because of previous check */
5916 (errcode(ERRCODE_INVALID_PARAMETER_VALUE
),
5917 errmsg("unrecognized format() type specifier \"%.*s\"",
5919 errhint("For a single \"%%\" use \"%%%%\".")));
5924 /* Don't need deconstruct_array results anymore. */
5925 if (elements
!= NULL
)
5930 /* Generate results. */
5931 result
= cstring_to_text_with_len(str
.data
, str
.len
);
5934 PG_RETURN_TEXT_P(result
);
5938 * Parse contiguous digits as a decimal number.
5940 * Returns true if some digits could be parsed.
5941 * The value is returned into *value, and *ptr is advanced to the next
5942 * character to be parsed.
5944 * Note parsing invariant: at least one character is known available before
5945 * string end (end_ptr) at entry, and this is still true at exit.
5948 text_format_parse_digits(const char **ptr
, const char *end_ptr
, int *value
)
5951 const char *cp
= *ptr
;
5954 while (*cp
>= '0' && *cp
<= '9')
5956 int8 digit
= (*cp
- '0');
5958 if (unlikely(pg_mul_s32_overflow(val
, 10, &val
)) ||
5959 unlikely(pg_add_s32_overflow(val
, digit
, &val
)))
5961 (errcode(ERRCODE_NUMERIC_VALUE_OUT_OF_RANGE
),
5962 errmsg("number is out of range")));
5963 ADVANCE_PARSE_POINTER(cp
, end_ptr
);
5974 * Parse a format specifier (generally following the SUS printf spec).
5976 * We have already advanced over the initial '%', and we are looking for
5977 * [argpos][flags][width]type (but the type character is not consumed here).
5979 * Inputs are start_ptr (the position after '%') and end_ptr (string end + 1).
5980 * Output parameters:
5981 * argpos: argument position for value to be printed. -1 means unspecified.
5982 * widthpos: argument position for width. Zero means the argument position
5983 * was unspecified (ie, take the next arg) and -1 means no width
5984 * argument (width was omitted or specified as a constant).
5985 * flags: bitmask of flags.
5986 * width: directly-specified width value. Zero means the width was omitted
5987 * (note it's not necessary to distinguish this case from an explicit
5988 * zero width value).
5990 * The function result is the next character position to be parsed, ie, the
5991 * location where the type character is/should be.
5993 * Note parsing invariant: at least one character is known available before
5994 * string end (end_ptr) at entry, and this is still true at exit.
5997 text_format_parse_format(const char *start_ptr
, const char *end_ptr
,
5998 int *argpos
, int *widthpos
,
5999 int *flags
, int *width
)
6001 const char *cp
= start_ptr
;
6004 /* set defaults for output parameters */
6010 /* try to identify first number */
6011 if (text_format_parse_digits(&cp
, end_ptr
, &n
))
6015 /* Must be just a width and a type, so we're done */
6019 /* The number was argument position */
6021 /* Explicit 0 for argument index is immediately refused */
6024 (errcode(ERRCODE_INVALID_PARAMETER_VALUE
),
6025 errmsg("format specifies argument 0, but arguments are numbered from 1")));
6026 ADVANCE_PARSE_POINTER(cp
, end_ptr
);
6029 /* Handle flags (only minus is supported now) */
6032 *flags
|= TEXT_FORMAT_FLAG_MINUS
;
6033 ADVANCE_PARSE_POINTER(cp
, end_ptr
);
6038 /* Handle indirect width */
6039 ADVANCE_PARSE_POINTER(cp
, end_ptr
);
6040 if (text_format_parse_digits(&cp
, end_ptr
, &n
))
6042 /* number in this position must be closed by $ */
6045 (errcode(ERRCODE_INVALID_PARAMETER_VALUE
),
6046 errmsg("width argument position must be ended by \"$\"")));
6047 /* The number was width argument position */
6049 /* Explicit 0 for argument index is immediately refused */
6052 (errcode(ERRCODE_INVALID_PARAMETER_VALUE
),
6053 errmsg("format specifies argument 0, but arguments are numbered from 1")));
6054 ADVANCE_PARSE_POINTER(cp
, end_ptr
);
6057 *widthpos
= 0; /* width's argument position is unspecified */
6061 /* Check for direct width specification */
6062 if (text_format_parse_digits(&cp
, end_ptr
, &n
))
6066 /* cp should now be pointing at type character */
6071 * Format a %s, %I, or %L conversion
6074 text_format_string_conversion(StringInfo buf
, char conversion
,
6075 FmgrInfo
*typOutputInfo
,
6076 Datum value
, bool isNull
,
6077 int flags
, int width
)
6081 /* Handle NULL arguments before trying to stringify the value. */
6084 if (conversion
== 's')
6085 text_format_append_string(buf
, "", flags
, width
);
6086 else if (conversion
== 'L')
6087 text_format_append_string(buf
, "NULL", flags
, width
);
6088 else if (conversion
== 'I')
6090 (errcode(ERRCODE_NULL_VALUE_NOT_ALLOWED
),
6091 errmsg("null values cannot be formatted as an SQL identifier")));
6096 str
= OutputFunctionCall(typOutputInfo
, value
);
6099 if (conversion
== 'I')
6101 /* quote_identifier may or may not allocate a new string. */
6102 text_format_append_string(buf
, quote_identifier(str
), flags
, width
);
6104 else if (conversion
== 'L')
6106 char *qstr
= quote_literal_cstr(str
);
6108 text_format_append_string(buf
, qstr
, flags
, width
);
6109 /* quote_literal_cstr() always allocates a new string */
6113 text_format_append_string(buf
, str
, flags
, width
);
6120 * Append str to buf, padding as directed by flags/width
6123 text_format_append_string(StringInfo buf
, const char *str
,
6124 int flags
, int width
)
6126 bool align_to_left
= false;
6129 /* fast path for typical easy case */
6132 appendStringInfoString(buf
, str
);
6138 /* Negative width: implicit '-' flag, then take absolute value */
6139 align_to_left
= true;
6140 /* -INT_MIN is undefined */
6141 if (width
<= INT_MIN
)
6143 (errcode(ERRCODE_NUMERIC_VALUE_OUT_OF_RANGE
),
6144 errmsg("number is out of range")));
6147 else if (flags
& TEXT_FORMAT_FLAG_MINUS
)
6148 align_to_left
= true;
6150 len
= pg_mbstrlen(str
);
6154 appendStringInfoString(buf
, str
);
6156 appendStringInfoSpaces(buf
, width
- len
);
6162 appendStringInfoSpaces(buf
, width
- len
);
6163 appendStringInfoString(buf
, str
);
6168 * text_format_nv - nonvariadic wrapper for text_format function.
6170 * note: this wrapper is necessary to pass the sanity check in opr_sanity,
6171 * which checks that all built-in functions that share the implementing C
6172 * function take the same number of arguments.
6175 text_format_nv(PG_FUNCTION_ARGS
)
6177 return text_format(fcinfo
);
6181 * Helper function for Levenshtein distance functions. Faster than memcmp(),
6182 * for this use case.
6185 rest_of_char_same(const char *s1
, const char *s2
, int len
)
6190 if (s1
[len
] != s2
[len
])
6196 /* Expand each Levenshtein distance variant */
6197 #include "levenshtein.c"
6198 #define LEVENSHTEIN_LESS_EQUAL
6199 #include "levenshtein.c"
6203 * The following *ClosestMatch() functions can be used to determine whether a
6204 * user-provided string resembles any known valid values, which is useful for
6205 * providing hints in log messages, among other things. Use these functions
6208 * initClosestMatch(&state, source_string, max_distance);
6210 * for (int i = 0; i < num_valid_strings; i++)
6211 * updateClosestMatch(&state, valid_strings[i]);
6213 * closestMatch = getClosestMatch(&state);
6217 * Initialize the given state with the source string and maximum Levenshtein
6218 * distance to consider.
6221 initClosestMatch(ClosestMatchState
*state
, const char *source
, int max_d
)
6226 state
->source
= source
;
6228 state
->max_d
= max_d
;
6229 state
->match
= NULL
;
6233 * If the candidate string is a closer match than the current one saved (or
6234 * there is no match saved), save it as the closest match.
6236 * If the source or candidate string is NULL, empty, or too long, this function
6237 * takes no action. Likewise, if the Levenshtein distance exceeds the maximum
6238 * allowed or more than half the characters are different, no action is taken.
6241 updateClosestMatch(ClosestMatchState
*state
, const char *candidate
)
6247 if (state
->source
== NULL
|| state
->source
[0] == '\0' ||
6248 candidate
== NULL
|| candidate
[0] == '\0')
6252 * To avoid ERROR-ing, we check the lengths here instead of setting
6253 * 'trusted' to false in the call to varstr_levenshtein_less_equal().
6255 if (strlen(state
->source
) > MAX_LEVENSHTEIN_STRLEN
||
6256 strlen(candidate
) > MAX_LEVENSHTEIN_STRLEN
)
6259 dist
= varstr_levenshtein_less_equal(state
->source
, strlen(state
->source
),
6260 candidate
, strlen(candidate
), 1, 1, 1,
6261 state
->max_d
, true);
6262 if (dist
<= state
->max_d
&&
6263 dist
<= strlen(state
->source
) / 2 &&
6264 (state
->min_d
== -1 || dist
< state
->min_d
))
6266 state
->min_d
= dist
;
6267 state
->match
= candidate
;
6272 * Return the closest match. If no suitable candidates were provided via
6273 * updateClosestMatch(), return NULL.
6276 getClosestMatch(ClosestMatchState
*state
)
6280 return state
->match
;
6288 static UnicodeNormalizationForm
6289 unicode_norm_form_from_string(const char *formstr
)
6291 UnicodeNormalizationForm form
= -1;
6294 * Might as well check this while we're here.
6296 if (GetDatabaseEncoding() != PG_UTF8
)
6298 (errcode(ERRCODE_SYNTAX_ERROR
),
6299 errmsg("Unicode normalization can only be performed if server encoding is UTF8")));
6301 if (pg_strcasecmp(formstr
, "NFC") == 0)
6303 else if (pg_strcasecmp(formstr
, "NFD") == 0)
6305 else if (pg_strcasecmp(formstr
, "NFKC") == 0)
6306 form
= UNICODE_NFKC
;
6307 else if (pg_strcasecmp(formstr
, "NFKD") == 0)
6308 form
= UNICODE_NFKD
;
6311 (errcode(ERRCODE_INVALID_PARAMETER_VALUE
),
6312 errmsg("invalid normalization form: %s", formstr
)));
6318 unicode_normalize_func(PG_FUNCTION_ARGS
)
6320 text
*input
= PG_GETARG_TEXT_PP(0);
6321 char *formstr
= text_to_cstring(PG_GETARG_TEXT_PP(1));
6322 UnicodeNormalizationForm form
;
6324 pg_wchar
*input_chars
;
6325 pg_wchar
*output_chars
;
6330 form
= unicode_norm_form_from_string(formstr
);
6332 /* convert to pg_wchar */
6333 size
= pg_mbstrlen_with_len(VARDATA_ANY(input
), VARSIZE_ANY_EXHDR(input
));
6334 input_chars
= palloc((size
+ 1) * sizeof(pg_wchar
));
6335 p
= (unsigned char *) VARDATA_ANY(input
);
6336 for (i
= 0; i
< size
; i
++)
6338 input_chars
[i
] = utf8_to_unicode(p
);
6339 p
+= pg_utf_mblen(p
);
6341 input_chars
[i
] = (pg_wchar
) '\0';
6342 Assert((char *) p
== VARDATA_ANY(input
) + VARSIZE_ANY_EXHDR(input
));
6345 output_chars
= unicode_normalize(form
, input_chars
);
6347 /* convert back to UTF-8 string */
6349 for (pg_wchar
*wp
= output_chars
; *wp
; wp
++)
6351 unsigned char buf
[4];
6353 unicode_to_utf8(*wp
, buf
);
6354 size
+= pg_utf_mblen(buf
);
6357 result
= palloc(size
+ VARHDRSZ
);
6358 SET_VARSIZE(result
, size
+ VARHDRSZ
);
6360 p
= (unsigned char *) VARDATA_ANY(result
);
6361 for (pg_wchar
*wp
= output_chars
; *wp
; wp
++)
6363 unicode_to_utf8(*wp
, p
);
6364 p
+= pg_utf_mblen(p
);
6366 Assert((char *) p
== (char *) result
+ size
+ VARHDRSZ
);
6368 PG_RETURN_TEXT_P(result
);
6372 * Check whether the string is in the specified Unicode normalization form.
6374 * This is done by converting the string to the specified normal form and then
6375 * comparing that to the original string. To speed that up, we also apply the
6376 * "quick check" algorithm specified in UAX #15, which can give a yes or no
6377 * answer for many strings by just scanning the string once.
6379 * This function should generally be optimized for the case where the string
6380 * is in fact normalized. In that case, we'll end up looking at the entire
6381 * string, so it's probably not worth doing any incremental conversion etc.
6384 unicode_is_normalized(PG_FUNCTION_ARGS
)
6386 text
*input
= PG_GETARG_TEXT_PP(0);
6387 char *formstr
= text_to_cstring(PG_GETARG_TEXT_PP(1));
6388 UnicodeNormalizationForm form
;
6390 pg_wchar
*input_chars
;
6391 pg_wchar
*output_chars
;
6394 UnicodeNormalizationQC quickcheck
;
6398 form
= unicode_norm_form_from_string(formstr
);
6400 /* convert to pg_wchar */
6401 size
= pg_mbstrlen_with_len(VARDATA_ANY(input
), VARSIZE_ANY_EXHDR(input
));
6402 input_chars
= palloc((size
+ 1) * sizeof(pg_wchar
));
6403 p
= (unsigned char *) VARDATA_ANY(input
);
6404 for (i
= 0; i
< size
; i
++)
6406 input_chars
[i
] = utf8_to_unicode(p
);
6407 p
+= pg_utf_mblen(p
);
6409 input_chars
[i
] = (pg_wchar
) '\0';
6410 Assert((char *) p
== VARDATA_ANY(input
) + VARSIZE_ANY_EXHDR(input
));
6412 /* quick check (see UAX #15) */
6413 quickcheck
= unicode_is_normalized_quickcheck(form
, input_chars
);
6414 if (quickcheck
== UNICODE_NORM_QC_YES
)
6415 PG_RETURN_BOOL(true);
6416 else if (quickcheck
== UNICODE_NORM_QC_NO
)
6417 PG_RETURN_BOOL(false);
6419 /* normalize and compare with original */
6420 output_chars
= unicode_normalize(form
, input_chars
);
6423 for (pg_wchar
*wp
= output_chars
; *wp
; wp
++)
6426 result
= (size
== output_size
) &&
6427 (memcmp(input_chars
, output_chars
, size
* sizeof(pg_wchar
)) == 0);
6429 PG_RETURN_BOOL(result
);
6433 * Check if first n chars are hexadecimal digits
6436 isxdigits_n(const char *instr
, size_t n
)
6438 for (size_t i
= 0; i
< n
; i
++)
6439 if (!isxdigit((unsigned char) instr
[i
]))
6446 hexval(unsigned char c
)
6448 if (c
>= '0' && c
<= '9')
6450 if (c
>= 'a' && c
<= 'f')
6451 return c
- 'a' + 0xA;
6452 if (c
>= 'A' && c
<= 'F')
6453 return c
- 'A' + 0xA;
6454 elog(ERROR
, "invalid hexadecimal digit");
6455 return 0; /* not reached */
6459 * Translate string with hexadecimal digits to number
6462 hexval_n(const char *instr
, size_t n
)
6464 unsigned int result
= 0;
6466 for (size_t i
= 0; i
< n
; i
++)
6467 result
+= hexval(instr
[i
]) << (4 * (n
- i
- 1));
6473 * Replaces Unicode escape sequences by Unicode characters
6476 unistr(PG_FUNCTION_ARGS
)
6478 text
*input_text
= PG_GETARG_TEXT_PP(0);
6483 pg_wchar pair_first
= 0;
6484 char cbuf
[MAX_UNICODE_EQUIVALENT_STRING
+ 1];
6486 instr
= VARDATA_ANY(input_text
);
6487 len
= VARSIZE_ANY_EXHDR(input_text
);
6489 initStringInfo(&str
);
6493 if (instr
[0] == '\\')
6500 appendStringInfoChar(&str
, '\\');
6504 else if ((len
>= 5 && isxdigits_n(instr
+ 1, 4)) ||
6505 (len
>= 6 && instr
[1] == 'u' && isxdigits_n(instr
+ 2, 4)))
6508 int offset
= instr
[1] == 'u' ? 2 : 1;
6510 unicode
= hexval_n(instr
+ offset
, 4);
6512 if (!is_valid_unicode_codepoint(unicode
))
6514 errcode(ERRCODE_INVALID_PARAMETER_VALUE
),
6515 errmsg("invalid Unicode code point: %04X", unicode
));
6519 if (is_utf16_surrogate_second(unicode
))
6521 unicode
= surrogate_pair_to_codepoint(pair_first
, unicode
);
6527 else if (is_utf16_surrogate_second(unicode
))
6530 if (is_utf16_surrogate_first(unicode
))
6531 pair_first
= unicode
;
6534 pg_unicode_to_server(unicode
, (unsigned char *) cbuf
);
6535 appendStringInfoString(&str
, cbuf
);
6538 instr
+= 4 + offset
;
6541 else if (len
>= 8 && instr
[1] == '+' && isxdigits_n(instr
+ 2, 6))
6545 unicode
= hexval_n(instr
+ 2, 6);
6547 if (!is_valid_unicode_codepoint(unicode
))
6549 errcode(ERRCODE_INVALID_PARAMETER_VALUE
),
6550 errmsg("invalid Unicode code point: %04X", unicode
));
6554 if (is_utf16_surrogate_second(unicode
))
6556 unicode
= surrogate_pair_to_codepoint(pair_first
, unicode
);
6562 else if (is_utf16_surrogate_second(unicode
))
6565 if (is_utf16_surrogate_first(unicode
))
6566 pair_first
= unicode
;
6569 pg_unicode_to_server(unicode
, (unsigned char *) cbuf
);
6570 appendStringInfoString(&str
, cbuf
);
6576 else if (len
>= 10 && instr
[1] == 'U' && isxdigits_n(instr
+ 2, 8))
6580 unicode
= hexval_n(instr
+ 2, 8);
6582 if (!is_valid_unicode_codepoint(unicode
))
6584 errcode(ERRCODE_INVALID_PARAMETER_VALUE
),
6585 errmsg("invalid Unicode code point: %04X", unicode
));
6589 if (is_utf16_surrogate_second(unicode
))
6591 unicode
= surrogate_pair_to_codepoint(pair_first
, unicode
);
6597 else if (is_utf16_surrogate_second(unicode
))
6600 if (is_utf16_surrogate_first(unicode
))
6601 pair_first
= unicode
;
6604 pg_unicode_to_server(unicode
, (unsigned char *) cbuf
);
6605 appendStringInfoString(&str
, cbuf
);
6613 (errcode(ERRCODE_SYNTAX_ERROR
),
6614 errmsg("invalid Unicode escape"),
6615 errhint("Unicode escapes must be \\XXXX, \\+XXXXXX, \\uXXXX, or \\UXXXXXXXX.")));
6622 appendStringInfoChar(&str
, *instr
++);
6627 /* unfinished surrogate pair? */
6631 result
= cstring_to_text_with_len(str
.data
, str
.len
);
6634 PG_RETURN_TEXT_P(result
);
6638 (errcode(ERRCODE_SYNTAX_ERROR
),
6639 errmsg("invalid Unicode surrogate pair")));
6640 PG_RETURN_NULL(); /* keep compiler quiet */