Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * bytea.c
4 : * Functions for the bytea type.
5 : *
6 : * Portions Copyright (c) 2025-2026, PostgreSQL Global Development Group
7 : *
8 : *
9 : * IDENTIFICATION
10 : * src/backend/utils/adt/bytea.c
11 : *
12 : *-------------------------------------------------------------------------
13 : */
14 :
15 : #include "postgres.h"
16 :
17 : #include "access/detoast.h"
18 : #include "common/hashfn.h"
19 : #include "common/int.h"
20 : #include "fmgr.h"
21 : #include "lib/hyperloglog.h"
22 : #include "libpq/pqformat.h"
23 : #include "port/pg_bitutils.h"
24 : #include "port/pg_bswap.h"
25 : #include "utils/builtins.h"
26 : #include "utils/bytea.h"
27 : #include "utils/fmgrprotos.h"
28 : #include "utils/guc.h"
29 : #include "utils/memutils.h"
30 : #include "utils/sortsupport.h"
31 : #include "varatt.h"
32 :
33 : /* GUC variable */
34 : int bytea_output = BYTEA_OUTPUT_HEX;
35 :
36 : static bytea *bytea_catenate(bytea *t1, bytea *t2);
37 : static bytea *bytea_substring(Datum str, int S, int L,
38 : bool length_not_specified);
39 : static bytea *bytea_overlay(bytea *t1, bytea *t2, int sp, int sl);
40 :
41 : typedef struct
42 : {
43 : bool abbreviate; /* Should we abbreviate keys? */
44 : hyperLogLogState abbr_card; /* Abbreviated key cardinality state */
45 : hyperLogLogState full_card; /* Full key cardinality state */
46 : double prop_card; /* Required cardinality proportion */
47 : } ByteaSortSupport;
48 :
49 : /* Static function declarations for sort support */
50 : static int byteafastcmp(Datum x, Datum y, SortSupport ssup);
51 : static Datum bytea_abbrev_convert(Datum original, SortSupport ssup);
52 : static bool bytea_abbrev_abort(int memtupcount, SortSupport ssup);
53 :
54 : /*
55 : * bytea_catenate
56 : * Guts of byteacat(), broken out so it can be used by other functions
57 : *
58 : * Arguments can be in short-header form, but not compressed or out-of-line
59 : */
60 : static bytea *
61 779 : bytea_catenate(bytea *t1, bytea *t2)
62 : {
63 : bytea *result;
64 : int len1,
65 : len2,
66 : len;
67 : char *ptr;
68 :
69 779 : len1 = VARSIZE_ANY_EXHDR(t1);
70 779 : len2 = VARSIZE_ANY_EXHDR(t2);
71 :
72 : /* paranoia ... probably should throw error instead? */
73 779 : if (len1 < 0)
74 0 : len1 = 0;
75 779 : if (len2 < 0)
76 0 : len2 = 0;
77 :
78 779 : len = len1 + len2 + VARHDRSZ;
79 779 : result = (bytea *) palloc(len);
80 :
81 : /* Set size of result string... */
82 779 : SET_VARSIZE(result, len);
83 :
84 : /* Fill data field of result string... */
85 779 : ptr = VARDATA(result);
86 779 : if (len1 > 0)
87 779 : memcpy(ptr, VARDATA_ANY(t1), len1);
88 779 : if (len2 > 0)
89 770 : memcpy(ptr + len1, VARDATA_ANY(t2), len2);
90 :
91 779 : return result;
92 : }
93 :
94 : #define PG_STR_GET_BYTEA(str_) \
95 : DatumGetByteaPP(DirectFunctionCall1(byteain, CStringGetDatum(str_)))
96 :
97 : static bytea *
98 2101 : bytea_substring(Datum str,
99 : int S,
100 : int L,
101 : bool length_not_specified)
102 : {
103 : int32 S1; /* adjusted start position */
104 : int32 L1; /* adjusted substring length */
105 : int32 E; /* end position */
106 :
107 : /*
108 : * The logic here should generally match text_substring().
109 : */
110 2101 : S1 = Max(S, 1);
111 :
112 2101 : if (length_not_specified)
113 : {
114 : /*
115 : * Not passed a length - DatumGetByteaPSlice() grabs everything to the
116 : * end of the string if we pass it a negative value for length.
117 : */
118 1977 : L1 = -1;
119 : }
120 124 : else if (L < 0)
121 : {
122 : /* SQL99 says to throw an error for E < S, i.e., negative length */
123 6 : ereport(ERROR,
124 : (errcode(ERRCODE_SUBSTRING_ERROR),
125 : errmsg("negative substring length not allowed")));
126 : L1 = -1; /* silence stupider compilers */
127 : }
128 118 : else if (pg_add_s32_overflow(S, L, &E))
129 : {
130 : /*
131 : * L could be large enough for S + L to overflow, in which case the
132 : * substring must run to end of string.
133 : */
134 3 : L1 = -1;
135 : }
136 : else
137 : {
138 : /*
139 : * A zero or negative value for the end position can happen if the
140 : * start was negative or one. SQL99 says to return a zero-length
141 : * string.
142 : */
143 115 : if (E < 1)
144 0 : return PG_STR_GET_BYTEA("");
145 :
146 115 : L1 = E - S1;
147 : }
148 :
149 : /*
150 : * If the start position is past the end of the string, SQL99 says to
151 : * return a zero-length string -- DatumGetByteaPSlice() will do that for
152 : * us. We need only convert S1 to zero-based starting position.
153 : */
154 2095 : return DatumGetByteaPSlice(str, S1 - 1, L1);
155 : }
156 :
157 : static bytea *
158 9 : bytea_overlay(bytea *t1, bytea *t2, int sp, int sl)
159 : {
160 : bytea *result;
161 : bytea *s1;
162 : bytea *s2;
163 : int sp_pl_sl;
164 :
165 : /*
166 : * Check for possible integer-overflow cases. For negative sp, throw a
167 : * "substring length" error because that's what should be expected
168 : * according to the spec's definition of OVERLAY().
169 : */
170 9 : if (sp <= 0)
171 0 : ereport(ERROR,
172 : (errcode(ERRCODE_SUBSTRING_ERROR),
173 : errmsg("negative substring length not allowed")));
174 9 : if (pg_add_s32_overflow(sp, sl, &sp_pl_sl))
175 0 : ereport(ERROR,
176 : (errcode(ERRCODE_NUMERIC_VALUE_OUT_OF_RANGE),
177 : errmsg("integer out of range")));
178 :
179 9 : s1 = bytea_substring(PointerGetDatum(t1), 1, sp - 1, false);
180 9 : s2 = bytea_substring(PointerGetDatum(t1), sp_pl_sl, -1, true);
181 9 : result = bytea_catenate(s1, t2);
182 9 : result = bytea_catenate(result, s2);
183 :
184 9 : return result;
185 : }
186 :
187 : /*****************************************************************************
188 : * USER I/O ROUTINES *
189 : *****************************************************************************/
190 :
191 : #define VAL(CH) ((CH) - '0')
192 : #define DIG(VAL) ((VAL) + '0')
193 :
194 : /*
195 : * byteain - converts from printable representation of byte array
196 : *
197 : * Non-printable characters must be passed as '\nnn' (octal) and are
198 : * converted to internal form. '\' must be passed as '\\'.
199 : */
200 : Datum
201 693941 : byteain(PG_FUNCTION_ARGS)
202 : {
203 693941 : char *inputText = PG_GETARG_CSTRING(0);
204 693941 : Node *escontext = fcinfo->context;
205 693941 : size_t len = strlen(inputText);
206 : size_t bc;
207 : char *tp;
208 : char *rp;
209 : bytea *result;
210 :
211 : /* Recognize hex input */
212 693941 : if (inputText[0] == '\\' && inputText[1] == 'x')
213 : {
214 56110 : bc = (len - 2) / 2 + VARHDRSZ; /* maximum possible length */
215 56110 : result = palloc(bc);
216 56110 : bc = hex_decode_safe(inputText + 2, len - 2, VARDATA(result),
217 : escontext);
218 56080 : SET_VARSIZE(result, bc + VARHDRSZ); /* actual length */
219 :
220 56080 : PG_RETURN_BYTEA_P(result);
221 : }
222 :
223 : /* Else, it's the traditional escaped style */
224 637831 : result = (bytea *) palloc(len + VARHDRSZ); /* maximum possible length */
225 :
226 637831 : tp = inputText;
227 637831 : rp = VARDATA(result);
228 4787485 : while (*tp != '\0')
229 : {
230 4149660 : if (tp[0] != '\\')
231 4149136 : *rp++ = *tp++;
232 524 : else if ((tp[1] >= '0' && tp[1] <= '3') &&
233 512 : (tp[2] >= '0' && tp[2] <= '7') &&
234 512 : (tp[3] >= '0' && tp[3] <= '7'))
235 512 : {
236 : int v;
237 :
238 512 : v = VAL(tp[1]);
239 512 : v <<= 3;
240 512 : v += VAL(tp[2]);
241 512 : v <<= 3;
242 512 : *rp++ = v + VAL(tp[3]);
243 :
244 512 : tp += 4;
245 : }
246 12 : else if (tp[1] == '\\')
247 : {
248 6 : *rp++ = '\\';
249 6 : tp += 2;
250 : }
251 : else
252 : {
253 : /*
254 : * one backslash, not followed by another or ### valid octal
255 : */
256 6 : ereturn(escontext, (Datum) 0,
257 : (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
258 : errmsg("invalid input syntax for type %s", "bytea")));
259 : }
260 : }
261 :
262 637825 : bc = rp - VARDATA(result); /* actual length */
263 637825 : SET_VARSIZE(result, bc + VARHDRSZ);
264 :
265 637825 : PG_RETURN_BYTEA_P(result);
266 : }
267 :
268 : /*
269 : * byteaout - converts to printable representation of byte array
270 : *
271 : * In the traditional escaped format, non-printable characters are
272 : * printed as '\nnn' (octal) and '\' as '\\'.
273 : */
274 : Datum
275 280631 : byteaout(PG_FUNCTION_ARGS)
276 : {
277 280631 : bytea *vlena = PG_GETARG_BYTEA_PP(0);
278 : char *result;
279 : char *rp;
280 :
281 280631 : if (bytea_output == BYTEA_OUTPUT_HEX)
282 : {
283 : /* Print hex format */
284 280436 : rp = result = palloc(VARSIZE_ANY_EXHDR(vlena) * 2 + 2 + 1);
285 280436 : *rp++ = '\\';
286 280436 : *rp++ = 'x';
287 280436 : rp += hex_encode(VARDATA_ANY(vlena), VARSIZE_ANY_EXHDR(vlena), rp);
288 : }
289 195 : else if (bytea_output == BYTEA_OUTPUT_ESCAPE)
290 : {
291 : /* Print traditional escaped format */
292 : char *vp;
293 : uint64 len;
294 : int i;
295 :
296 195 : len = 1; /* empty string has 1 char */
297 195 : vp = VARDATA_ANY(vlena);
298 108860 : for (i = VARSIZE_ANY_EXHDR(vlena); i != 0; i--, vp++)
299 : {
300 108665 : if (*vp == '\\')
301 3 : len += 2;
302 108662 : else if ((unsigned char) *vp < 0x20 || (unsigned char) *vp > 0x7e)
303 249 : len += 4;
304 : else
305 108413 : len++;
306 : }
307 :
308 : /*
309 : * In principle len can't overflow uint32 if the input fit in 1GB, but
310 : * for safety let's check rather than relying on palloc's internal
311 : * check.
312 : */
313 195 : if (len > MaxAllocSize)
314 0 : ereport(ERROR,
315 : (errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
316 : errmsg_internal("result of bytea output conversion is too large")));
317 195 : rp = result = (char *) palloc(len);
318 :
319 195 : vp = VARDATA_ANY(vlena);
320 108860 : for (i = VARSIZE_ANY_EXHDR(vlena); i != 0; i--, vp++)
321 : {
322 108665 : if (*vp == '\\')
323 : {
324 3 : *rp++ = '\\';
325 3 : *rp++ = '\\';
326 : }
327 108662 : else if ((unsigned char) *vp < 0x20 || (unsigned char) *vp > 0x7e)
328 249 : {
329 : int val; /* holds unprintable chars */
330 :
331 249 : val = *vp;
332 249 : rp[0] = '\\';
333 249 : rp[3] = DIG(val & 07);
334 249 : val >>= 3;
335 249 : rp[2] = DIG(val & 07);
336 249 : val >>= 3;
337 249 : rp[1] = DIG(val & 03);
338 249 : rp += 4;
339 : }
340 : else
341 108413 : *rp++ = *vp;
342 : }
343 : }
344 : else
345 : {
346 0 : elog(ERROR, "unrecognized \"bytea_output\" setting: %d",
347 : bytea_output);
348 : rp = result = NULL; /* keep compiler quiet */
349 : }
350 280631 : *rp = '\0';
351 280631 : PG_RETURN_CSTRING(result);
352 : }
353 :
354 : /*
355 : * bytearecv - converts external binary format to bytea
356 : */
357 : Datum
358 53855 : bytearecv(PG_FUNCTION_ARGS)
359 : {
360 53855 : StringInfo buf = (StringInfo) PG_GETARG_POINTER(0);
361 : bytea *result;
362 : int nbytes;
363 :
364 53855 : nbytes = buf->len - buf->cursor;
365 53855 : result = (bytea *) palloc(nbytes + VARHDRSZ);
366 53855 : SET_VARSIZE(result, nbytes + VARHDRSZ);
367 53855 : pq_copymsgbytes(buf, VARDATA(result), nbytes);
368 53855 : PG_RETURN_BYTEA_P(result);
369 : }
370 :
371 : /*
372 : * byteasend - converts bytea to binary format
373 : *
374 : * This is a special case: just copy the input...
375 : */
376 : Datum
377 34397 : byteasend(PG_FUNCTION_ARGS)
378 : {
379 34397 : bytea *vlena = PG_GETARG_BYTEA_P_COPY(0);
380 :
381 34397 : PG_RETURN_BYTEA_P(vlena);
382 : }
383 :
384 : Datum
385 129387 : bytea_string_agg_transfn(PG_FUNCTION_ARGS)
386 : {
387 : StringInfo state;
388 :
389 129387 : state = PG_ARGISNULL(0) ? NULL : (StringInfo) PG_GETARG_POINTER(0);
390 :
391 : /* Append the value unless null, preceding it with the delimiter. */
392 129387 : if (!PG_ARGISNULL(1))
393 : {
394 121887 : bytea *value = PG_GETARG_BYTEA_PP(1);
395 121887 : bool isfirst = false;
396 :
397 : /*
398 : * You might think we can just throw away the first delimiter, however
399 : * we must keep it as we may be a parallel worker doing partial
400 : * aggregation building a state to send to the main process. We need
401 : * to keep the delimiter of every aggregation so that the combine
402 : * function can properly join up the strings of two separately
403 : * partially aggregated results. The first delimiter is only stripped
404 : * off in the final function. To know how much to strip off the front
405 : * of the string, we store the length of the first delimiter in the
406 : * StringInfo's cursor field, which we don't otherwise need here.
407 : */
408 121887 : if (state == NULL)
409 : {
410 : MemoryContext aggcontext;
411 : MemoryContext oldcontext;
412 :
413 104 : if (!AggCheckCallContext(fcinfo, &aggcontext))
414 : {
415 : /* cannot be called directly because of internal-type argument */
416 0 : elog(ERROR, "bytea_string_agg_transfn called in non-aggregate context");
417 : }
418 :
419 : /*
420 : * Create state in aggregate context. It'll stay there across
421 : * subsequent calls.
422 : */
423 104 : oldcontext = MemoryContextSwitchTo(aggcontext);
424 104 : state = makeStringInfo();
425 104 : MemoryContextSwitchTo(oldcontext);
426 :
427 104 : isfirst = true;
428 : }
429 :
430 121887 : if (!PG_ARGISNULL(2))
431 : {
432 121881 : bytea *delim = PG_GETARG_BYTEA_PP(2);
433 :
434 121881 : appendBinaryStringInfo(state, VARDATA_ANY(delim),
435 121881 : VARSIZE_ANY_EXHDR(delim));
436 121881 : if (isfirst)
437 101 : state->cursor = VARSIZE_ANY_EXHDR(delim);
438 : }
439 :
440 121887 : appendBinaryStringInfo(state, VARDATA_ANY(value),
441 121887 : VARSIZE_ANY_EXHDR(value));
442 : }
443 :
444 : /*
445 : * The transition type for string_agg() is declared to be "internal",
446 : * which is a pass-by-value type the same size as a pointer.
447 : */
448 129387 : if (state)
449 129363 : PG_RETURN_POINTER(state);
450 24 : PG_RETURN_NULL();
451 : }
452 :
453 : Datum
454 77 : bytea_string_agg_finalfn(PG_FUNCTION_ARGS)
455 : {
456 : StringInfo state;
457 :
458 : /* cannot be called directly because of internal-type argument */
459 : Assert(AggCheckCallContext(fcinfo, NULL));
460 :
461 77 : state = PG_ARGISNULL(0) ? NULL : (StringInfo) PG_GETARG_POINTER(0);
462 :
463 77 : if (state != NULL)
464 : {
465 : /* As per comment in transfn, strip data before the cursor position */
466 : bytea *result;
467 74 : int strippedlen = state->len - state->cursor;
468 :
469 74 : result = (bytea *) palloc(strippedlen + VARHDRSZ);
470 74 : SET_VARSIZE(result, strippedlen + VARHDRSZ);
471 74 : memcpy(VARDATA(result), &state->data[state->cursor], strippedlen);
472 74 : PG_RETURN_BYTEA_P(result);
473 : }
474 : else
475 3 : PG_RETURN_NULL();
476 : }
477 :
478 : /*-------------------------------------------------------------
479 : * byteaoctetlen
480 : *
481 : * get the number of bytes contained in an instance of type 'bytea'
482 : *-------------------------------------------------------------
483 : */
484 : Datum
485 607 : byteaoctetlen(PG_FUNCTION_ARGS)
486 : {
487 607 : Datum str = PG_GETARG_DATUM(0);
488 :
489 : /* We need not detoast the input at all */
490 607 : PG_RETURN_INT32(toast_raw_datum_size(str) - VARHDRSZ);
491 : }
492 :
493 : /*
494 : * byteacat -
495 : * takes two bytea* and returns a bytea* that is the concatenation of
496 : * the two.
497 : *
498 : * Cloned from textcat and modified as required.
499 : */
500 : Datum
501 761 : byteacat(PG_FUNCTION_ARGS)
502 : {
503 761 : bytea *t1 = PG_GETARG_BYTEA_PP(0);
504 761 : bytea *t2 = PG_GETARG_BYTEA_PP(1);
505 :
506 761 : PG_RETURN_BYTEA_P(bytea_catenate(t1, t2));
507 : }
508 :
509 : /*
510 : * byteaoverlay
511 : * Replace specified substring of first string with second
512 : *
513 : * The SQL standard defines OVERLAY() in terms of substring and concatenation.
514 : * This code is a direct implementation of what the standard says.
515 : */
516 : Datum
517 3 : byteaoverlay(PG_FUNCTION_ARGS)
518 : {
519 3 : bytea *t1 = PG_GETARG_BYTEA_PP(0);
520 3 : bytea *t2 = PG_GETARG_BYTEA_PP(1);
521 3 : int sp = PG_GETARG_INT32(2); /* substring start position */
522 3 : int sl = PG_GETARG_INT32(3); /* substring length */
523 :
524 3 : PG_RETURN_BYTEA_P(bytea_overlay(t1, t2, sp, sl));
525 : }
526 :
527 : Datum
528 6 : byteaoverlay_no_len(PG_FUNCTION_ARGS)
529 : {
530 6 : bytea *t1 = PG_GETARG_BYTEA_PP(0);
531 6 : bytea *t2 = PG_GETARG_BYTEA_PP(1);
532 6 : int sp = PG_GETARG_INT32(2); /* substring start position */
533 : int sl;
534 :
535 6 : sl = VARSIZE_ANY_EXHDR(t2); /* defaults to length(t2) */
536 6 : PG_RETURN_BYTEA_P(bytea_overlay(t1, t2, sp, sl));
537 : }
538 :
539 : /*
540 : * bytea_substr()
541 : * Return a substring starting at the specified position.
542 : * Cloned from text_substr and modified as required.
543 : *
544 : * Input:
545 : * - string
546 : * - starting position (is one-based)
547 : * - string length (optional)
548 : *
549 : * If the starting position is zero or less, then return from the start of the string
550 : * adjusting the length to be consistent with the "negative start" per SQL.
551 : * If the length is less than zero, an ERROR is thrown. If no third argument
552 : * (length) is provided, the length to the end of the string is assumed.
553 : */
554 : Datum
555 115 : bytea_substr(PG_FUNCTION_ARGS)
556 : {
557 115 : PG_RETURN_BYTEA_P(bytea_substring(PG_GETARG_DATUM(0),
558 : PG_GETARG_INT32(1),
559 : PG_GETARG_INT32(2),
560 : false));
561 : }
562 :
563 : /*
564 : * bytea_substr_no_len -
565 : * Wrapper to avoid opr_sanity failure due to
566 : * one function accepting a different number of args.
567 : */
568 : Datum
569 1968 : bytea_substr_no_len(PG_FUNCTION_ARGS)
570 : {
571 1968 : PG_RETURN_BYTEA_P(bytea_substring(PG_GETARG_DATUM(0),
572 : PG_GETARG_INT32(1),
573 : -1,
574 : true));
575 : }
576 :
577 : /*
578 : * bit_count
579 : */
580 : Datum
581 3 : bytea_bit_count(PG_FUNCTION_ARGS)
582 : {
583 3 : bytea *t1 = PG_GETARG_BYTEA_PP(0);
584 :
585 3 : PG_RETURN_INT64(pg_popcount(VARDATA_ANY(t1), VARSIZE_ANY_EXHDR(t1)));
586 : }
587 :
588 : /*
589 : * byteapos -
590 : * Return the position of the specified substring.
591 : * Implements the SQL POSITION() function.
592 : * Cloned from textpos and modified as required.
593 : */
594 : Datum
595 15 : byteapos(PG_FUNCTION_ARGS)
596 : {
597 15 : bytea *t1 = PG_GETARG_BYTEA_PP(0);
598 15 : bytea *t2 = PG_GETARG_BYTEA_PP(1);
599 : int pos;
600 : int px,
601 : p;
602 : int len1,
603 : len2;
604 : char *p1,
605 : *p2;
606 :
607 15 : len1 = VARSIZE_ANY_EXHDR(t1);
608 15 : len2 = VARSIZE_ANY_EXHDR(t2);
609 :
610 15 : if (len2 <= 0)
611 3 : PG_RETURN_INT32(1); /* result for empty pattern */
612 :
613 12 : p1 = VARDATA_ANY(t1);
614 12 : p2 = VARDATA_ANY(t2);
615 :
616 12 : pos = 0;
617 12 : px = (len1 - len2);
618 27 : for (p = 0; p <= px; p++)
619 : {
620 21 : if ((*p2 == *p1) && (memcmp(p1, p2, len2) == 0))
621 : {
622 6 : pos = p + 1;
623 6 : break;
624 : };
625 15 : p1++;
626 : };
627 :
628 12 : PG_RETURN_INT32(pos);
629 : }
630 :
631 : /*-------------------------------------------------------------
632 : * byteaGetByte
633 : *
634 : * this routine treats "bytea" as an array of bytes.
635 : * It returns the Nth byte (a number between 0 and 255).
636 : *-------------------------------------------------------------
637 : */
638 : Datum
639 38 : byteaGetByte(PG_FUNCTION_ARGS)
640 : {
641 38 : bytea *v = PG_GETARG_BYTEA_PP(0);
642 38 : int32 n = PG_GETARG_INT32(1);
643 : int len;
644 : int byte;
645 :
646 38 : len = VARSIZE_ANY_EXHDR(v);
647 :
648 38 : if (n < 0 || n >= len)
649 3 : ereport(ERROR,
650 : (errcode(ERRCODE_ARRAY_SUBSCRIPT_ERROR),
651 : errmsg("index %d out of valid range, 0..%d",
652 : n, len - 1)));
653 :
654 35 : byte = ((unsigned char *) VARDATA_ANY(v))[n];
655 :
656 35 : PG_RETURN_INT32(byte);
657 : }
658 :
659 : /*-------------------------------------------------------------
660 : * byteaGetBit
661 : *
662 : * This routine treats a "bytea" type like an array of bits.
663 : * It returns the value of the Nth bit (0 or 1).
664 : *
665 : *-------------------------------------------------------------
666 : */
667 : Datum
668 6 : byteaGetBit(PG_FUNCTION_ARGS)
669 : {
670 6 : bytea *v = PG_GETARG_BYTEA_PP(0);
671 6 : int64 n = PG_GETARG_INT64(1);
672 : int byteNo,
673 : bitNo;
674 : int len;
675 : int byte;
676 :
677 6 : len = VARSIZE_ANY_EXHDR(v);
678 :
679 6 : if (n < 0 || n >= (int64) len * 8)
680 3 : ereport(ERROR,
681 : (errcode(ERRCODE_ARRAY_SUBSCRIPT_ERROR),
682 : errmsg("index %" PRId64 " out of valid range, 0..%" PRId64,
683 : n, (int64) len * 8 - 1)));
684 :
685 : /* n/8 is now known < len, so safe to cast to int */
686 3 : byteNo = (int) (n / 8);
687 3 : bitNo = (int) (n % 8);
688 :
689 3 : byte = ((unsigned char *) VARDATA_ANY(v))[byteNo];
690 :
691 3 : if (byte & (1 << bitNo))
692 3 : PG_RETURN_INT32(1);
693 : else
694 0 : PG_RETURN_INT32(0);
695 : }
696 :
697 : /*-------------------------------------------------------------
698 : * byteaSetByte
699 : *
700 : * Given an instance of type 'bytea' creates a new one with
701 : * the Nth byte set to the given value.
702 : *
703 : *-------------------------------------------------------------
704 : */
705 : Datum
706 6 : byteaSetByte(PG_FUNCTION_ARGS)
707 : {
708 6 : bytea *res = PG_GETARG_BYTEA_P_COPY(0);
709 6 : int32 n = PG_GETARG_INT32(1);
710 6 : int32 newByte = PG_GETARG_INT32(2);
711 : int len;
712 :
713 6 : len = VARSIZE(res) - VARHDRSZ;
714 :
715 6 : if (n < 0 || n >= len)
716 3 : ereport(ERROR,
717 : (errcode(ERRCODE_ARRAY_SUBSCRIPT_ERROR),
718 : errmsg("index %d out of valid range, 0..%d",
719 : n, len - 1)));
720 :
721 : /*
722 : * Now set the byte.
723 : */
724 3 : ((unsigned char *) VARDATA(res))[n] = newByte;
725 :
726 3 : PG_RETURN_BYTEA_P(res);
727 : }
728 :
729 : /*-------------------------------------------------------------
730 : * byteaSetBit
731 : *
732 : * Given an instance of type 'bytea' creates a new one with
733 : * the Nth bit set to the given value.
734 : *
735 : *-------------------------------------------------------------
736 : */
737 : Datum
738 6 : byteaSetBit(PG_FUNCTION_ARGS)
739 : {
740 6 : bytea *res = PG_GETARG_BYTEA_P_COPY(0);
741 6 : int64 n = PG_GETARG_INT64(1);
742 6 : int32 newBit = PG_GETARG_INT32(2);
743 : int len;
744 : int oldByte,
745 : newByte;
746 : int byteNo,
747 : bitNo;
748 :
749 6 : len = VARSIZE(res) - VARHDRSZ;
750 :
751 6 : if (n < 0 || n >= (int64) len * 8)
752 3 : ereport(ERROR,
753 : (errcode(ERRCODE_ARRAY_SUBSCRIPT_ERROR),
754 : errmsg("index %" PRId64 " out of valid range, 0..%" PRId64,
755 : n, (int64) len * 8 - 1)));
756 :
757 : /* n/8 is now known < len, so safe to cast to int */
758 3 : byteNo = (int) (n / 8);
759 3 : bitNo = (int) (n % 8);
760 :
761 : /*
762 : * sanity check!
763 : */
764 3 : if (newBit != 0 && newBit != 1)
765 0 : ereport(ERROR,
766 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
767 : errmsg("new bit must be 0 or 1")));
768 :
769 : /*
770 : * Update the byte.
771 : */
772 3 : oldByte = ((unsigned char *) VARDATA(res))[byteNo];
773 :
774 3 : if (newBit == 0)
775 3 : newByte = oldByte & (~(1 << bitNo));
776 : else
777 0 : newByte = oldByte | (1 << bitNo);
778 :
779 3 : ((unsigned char *) VARDATA(res))[byteNo] = newByte;
780 :
781 3 : PG_RETURN_BYTEA_P(res);
782 : }
783 :
784 : /*
785 : * Return reversed bytea
786 : */
787 : Datum
788 9 : bytea_reverse(PG_FUNCTION_ARGS)
789 : {
790 9 : bytea *v = PG_GETARG_BYTEA_PP(0);
791 9 : const char *p = VARDATA_ANY(v);
792 9 : int len = VARSIZE_ANY_EXHDR(v);
793 9 : const char *endp = p + len;
794 9 : bytea *result = palloc(len + VARHDRSZ);
795 9 : char *dst = (char *) VARDATA(result) + len;
796 :
797 9 : SET_VARSIZE(result, len + VARHDRSZ);
798 :
799 18 : while (p < endp)
800 9 : *(--dst) = *p++;
801 :
802 9 : PG_RETURN_BYTEA_P(result);
803 : }
804 :
805 :
806 : /*****************************************************************************
807 : * Comparison Functions used for bytea
808 : *
809 : * Note: btree indexes need these routines not to leak memory; therefore,
810 : * be careful to free working copies of toasted datums. Most places don't
811 : * need to be so careful.
812 : *****************************************************************************/
813 :
814 : Datum
815 5479 : byteaeq(PG_FUNCTION_ARGS)
816 : {
817 5479 : Datum arg1 = PG_GETARG_DATUM(0);
818 5479 : Datum arg2 = PG_GETARG_DATUM(1);
819 : bool result;
820 : Size len1,
821 : len2;
822 :
823 : /*
824 : * We can use a fast path for unequal lengths, which might save us from
825 : * having to detoast one or both values.
826 : */
827 5479 : len1 = toast_raw_datum_size(arg1);
828 5479 : len2 = toast_raw_datum_size(arg2);
829 5479 : if (len1 != len2)
830 2305 : result = false;
831 : else
832 : {
833 3174 : bytea *barg1 = DatumGetByteaPP(arg1);
834 3174 : bytea *barg2 = DatumGetByteaPP(arg2);
835 :
836 3174 : result = (memcmp(VARDATA_ANY(barg1), VARDATA_ANY(barg2),
837 : len1 - VARHDRSZ) == 0);
838 :
839 3174 : PG_FREE_IF_COPY(barg1, 0);
840 3174 : PG_FREE_IF_COPY(barg2, 1);
841 : }
842 :
843 5479 : PG_RETURN_BOOL(result);
844 : }
845 :
846 : Datum
847 384 : byteane(PG_FUNCTION_ARGS)
848 : {
849 384 : Datum arg1 = PG_GETARG_DATUM(0);
850 384 : Datum arg2 = PG_GETARG_DATUM(1);
851 : bool result;
852 : Size len1,
853 : len2;
854 :
855 : /*
856 : * We can use a fast path for unequal lengths, which might save us from
857 : * having to detoast one or both values.
858 : */
859 384 : len1 = toast_raw_datum_size(arg1);
860 384 : len2 = toast_raw_datum_size(arg2);
861 384 : if (len1 != len2)
862 0 : result = true;
863 : else
864 : {
865 384 : bytea *barg1 = DatumGetByteaPP(arg1);
866 384 : bytea *barg2 = DatumGetByteaPP(arg2);
867 :
868 384 : result = (memcmp(VARDATA_ANY(barg1), VARDATA_ANY(barg2),
869 : len1 - VARHDRSZ) != 0);
870 :
871 384 : PG_FREE_IF_COPY(barg1, 0);
872 384 : PG_FREE_IF_COPY(barg2, 1);
873 : }
874 :
875 384 : PG_RETURN_BOOL(result);
876 : }
877 :
878 : Datum
879 4126 : bytealt(PG_FUNCTION_ARGS)
880 : {
881 4126 : bytea *arg1 = PG_GETARG_BYTEA_PP(0);
882 4126 : bytea *arg2 = PG_GETARG_BYTEA_PP(1);
883 : int len1,
884 : len2;
885 : int cmp;
886 :
887 4126 : len1 = VARSIZE_ANY_EXHDR(arg1);
888 4126 : len2 = VARSIZE_ANY_EXHDR(arg2);
889 :
890 4126 : cmp = memcmp(VARDATA_ANY(arg1), VARDATA_ANY(arg2), Min(len1, len2));
891 :
892 4126 : PG_FREE_IF_COPY(arg1, 0);
893 4126 : PG_FREE_IF_COPY(arg2, 1);
894 :
895 4126 : PG_RETURN_BOOL((cmp < 0) || ((cmp == 0) && (len1 < len2)));
896 : }
897 :
898 : Datum
899 3178 : byteale(PG_FUNCTION_ARGS)
900 : {
901 3178 : bytea *arg1 = PG_GETARG_BYTEA_PP(0);
902 3178 : bytea *arg2 = PG_GETARG_BYTEA_PP(1);
903 : int len1,
904 : len2;
905 : int cmp;
906 :
907 3178 : len1 = VARSIZE_ANY_EXHDR(arg1);
908 3178 : len2 = VARSIZE_ANY_EXHDR(arg2);
909 :
910 3178 : cmp = memcmp(VARDATA_ANY(arg1), VARDATA_ANY(arg2), Min(len1, len2));
911 :
912 3178 : PG_FREE_IF_COPY(arg1, 0);
913 3178 : PG_FREE_IF_COPY(arg2, 1);
914 :
915 3178 : PG_RETURN_BOOL((cmp < 0) || ((cmp == 0) && (len1 <= len2)));
916 : }
917 :
918 : Datum
919 3136 : byteagt(PG_FUNCTION_ARGS)
920 : {
921 3136 : bytea *arg1 = PG_GETARG_BYTEA_PP(0);
922 3136 : bytea *arg2 = PG_GETARG_BYTEA_PP(1);
923 : int len1,
924 : len2;
925 : int cmp;
926 :
927 3136 : len1 = VARSIZE_ANY_EXHDR(arg1);
928 3136 : len2 = VARSIZE_ANY_EXHDR(arg2);
929 :
930 3136 : cmp = memcmp(VARDATA_ANY(arg1), VARDATA_ANY(arg2), Min(len1, len2));
931 :
932 3136 : PG_FREE_IF_COPY(arg1, 0);
933 3136 : PG_FREE_IF_COPY(arg2, 1);
934 :
935 3136 : PG_RETURN_BOOL((cmp > 0) || ((cmp == 0) && (len1 > len2)));
936 : }
937 :
938 : Datum
939 2527 : byteage(PG_FUNCTION_ARGS)
940 : {
941 2527 : bytea *arg1 = PG_GETARG_BYTEA_PP(0);
942 2527 : bytea *arg2 = PG_GETARG_BYTEA_PP(1);
943 : int len1,
944 : len2;
945 : int cmp;
946 :
947 2527 : len1 = VARSIZE_ANY_EXHDR(arg1);
948 2527 : len2 = VARSIZE_ANY_EXHDR(arg2);
949 :
950 2527 : cmp = memcmp(VARDATA_ANY(arg1), VARDATA_ANY(arg2), Min(len1, len2));
951 :
952 2527 : PG_FREE_IF_COPY(arg1, 0);
953 2527 : PG_FREE_IF_COPY(arg2, 1);
954 :
955 2527 : PG_RETURN_BOOL((cmp > 0) || ((cmp == 0) && (len1 >= len2)));
956 : }
957 :
958 : Datum
959 47296 : byteacmp(PG_FUNCTION_ARGS)
960 : {
961 47296 : bytea *arg1 = PG_GETARG_BYTEA_PP(0);
962 47296 : bytea *arg2 = PG_GETARG_BYTEA_PP(1);
963 : int len1,
964 : len2;
965 : int cmp;
966 :
967 47296 : len1 = VARSIZE_ANY_EXHDR(arg1);
968 47296 : len2 = VARSIZE_ANY_EXHDR(arg2);
969 :
970 47296 : cmp = memcmp(VARDATA_ANY(arg1), VARDATA_ANY(arg2), Min(len1, len2));
971 47296 : if ((cmp == 0) && (len1 != len2))
972 1700 : cmp = (len1 < len2) ? -1 : 1;
973 :
974 47296 : PG_FREE_IF_COPY(arg1, 0);
975 47296 : PG_FREE_IF_COPY(arg2, 1);
976 :
977 47296 : PG_RETURN_INT32(cmp);
978 : }
979 :
980 : Datum
981 12 : bytea_larger(PG_FUNCTION_ARGS)
982 : {
983 12 : bytea *arg1 = PG_GETARG_BYTEA_PP(0);
984 12 : bytea *arg2 = PG_GETARG_BYTEA_PP(1);
985 : bytea *result;
986 : int len1,
987 : len2;
988 : int cmp;
989 :
990 12 : len1 = VARSIZE_ANY_EXHDR(arg1);
991 12 : len2 = VARSIZE_ANY_EXHDR(arg2);
992 :
993 12 : cmp = memcmp(VARDATA_ANY(arg1), VARDATA_ANY(arg2), Min(len1, len2));
994 12 : result = ((cmp > 0) || ((cmp == 0) && (len1 > len2)) ? arg1 : arg2);
995 :
996 12 : PG_RETURN_BYTEA_P(result);
997 : }
998 :
999 : Datum
1000 12 : bytea_smaller(PG_FUNCTION_ARGS)
1001 : {
1002 12 : bytea *arg1 = PG_GETARG_BYTEA_PP(0);
1003 12 : bytea *arg2 = PG_GETARG_BYTEA_PP(1);
1004 : bytea *result;
1005 : int len1,
1006 : len2;
1007 : int cmp;
1008 :
1009 12 : len1 = VARSIZE_ANY_EXHDR(arg1);
1010 12 : len2 = VARSIZE_ANY_EXHDR(arg2);
1011 :
1012 12 : cmp = memcmp(VARDATA_ANY(arg1), VARDATA_ANY(arg2), Min(len1, len2));
1013 12 : result = ((cmp < 0) || ((cmp == 0) && (len1 < len2)) ? arg1 : arg2);
1014 :
1015 12 : PG_RETURN_BYTEA_P(result);
1016 : }
1017 :
1018 : /*
1019 : * sortsupport comparison func
1020 : */
1021 : static int
1022 10132 : byteafastcmp(Datum x, Datum y, SortSupport ssup)
1023 : {
1024 10132 : bytea *arg1 = DatumGetByteaPP(x);
1025 10132 : bytea *arg2 = DatumGetByteaPP(y);
1026 : char *a1p,
1027 : *a2p;
1028 : int len1,
1029 : len2,
1030 : result;
1031 :
1032 10132 : a1p = VARDATA_ANY(arg1);
1033 10132 : a2p = VARDATA_ANY(arg2);
1034 :
1035 10132 : len1 = VARSIZE_ANY_EXHDR(arg1);
1036 10132 : len2 = VARSIZE_ANY_EXHDR(arg2);
1037 :
1038 10132 : result = memcmp(a1p, a2p, Min(len1, len2));
1039 10132 : if ((result == 0) && (len1 != len2))
1040 415 : result = (len1 < len2) ? -1 : 1;
1041 :
1042 : /* We can't afford to leak memory here. */
1043 10132 : if (PointerGetDatum(arg1) != x)
1044 0 : pfree(arg1);
1045 10132 : if (PointerGetDatum(arg2) != y)
1046 0 : pfree(arg2);
1047 :
1048 10132 : return result;
1049 : }
1050 :
1051 : /*
1052 : * Conversion routine for sortsupport. Converts original to abbreviated key
1053 : * representation. Our encoding strategy is simple -- pack the first 8 bytes
1054 : * of the bytea data into a Datum (on little-endian machines, the bytes are
1055 : * stored in reverse order), and treat it as an unsigned integer.
1056 : */
1057 : static Datum
1058 12 : bytea_abbrev_convert(Datum original, SortSupport ssup)
1059 : {
1060 12 : const size_t max_prefix_bytes = sizeof(Datum);
1061 12 : ByteaSortSupport *bss = (ByteaSortSupport *) ssup->ssup_extra;
1062 12 : bytea *authoritative = DatumGetByteaPP(original);
1063 12 : char *authoritative_data = VARDATA_ANY(authoritative);
1064 : Datum res;
1065 : char *pres;
1066 : int len;
1067 : uint32 hash;
1068 :
1069 12 : pres = (char *) &res;
1070 :
1071 : /* memset(), so any non-overwritten bytes are NUL */
1072 12 : memset(pres, 0, max_prefix_bytes);
1073 12 : len = VARSIZE_ANY_EXHDR(authoritative);
1074 :
1075 : /*
1076 : * Short byteas will have terminating NUL bytes in the abbreviated datum.
1077 : * Abbreviated comparison need not make a distinction between these NUL
1078 : * bytes, and NUL bytes representing actual NULs in the authoritative
1079 : * representation.
1080 : *
1081 : * Hopefully a comparison at or past one abbreviated key's terminating NUL
1082 : * byte will resolve the comparison without consulting the authoritative
1083 : * representation; specifically, some later non-NUL byte in the longer
1084 : * bytea can resolve the comparison against a subsequent terminating NUL
1085 : * in the shorter bytea. There will usually be what is effectively a
1086 : * "length-wise" resolution there and then.
1087 : *
1088 : * If that doesn't work out -- if all bytes in the longer bytea positioned
1089 : * at or past the offset of the smaller bytea (first) terminating NUL are
1090 : * actually representative of NUL bytes in the authoritative binary bytea
1091 : * (perhaps with some *terminating* NUL bytes towards the end of the
1092 : * longer bytea iff it happens to still be small) -- then an authoritative
1093 : * tie-breaker will happen, and do the right thing: explicitly consider
1094 : * bytea length.
1095 : */
1096 12 : memcpy(pres, authoritative_data, Min(len, max_prefix_bytes));
1097 :
1098 : /*
1099 : * Maintain approximate cardinality of both abbreviated keys and original,
1100 : * authoritative keys using HyperLogLog. Used as cheap insurance against
1101 : * the worst case, where we do many string abbreviations for no saving in
1102 : * full memcmp()-based comparisons. These statistics are used by
1103 : * bytea_abbrev_abort().
1104 : *
1105 : * First, Hash key proper, or a significant fraction of it. Mix in length
1106 : * in order to compensate for cases where differences are past
1107 : * PG_CACHE_LINE_SIZE bytes, so as to limit the overhead of hashing.
1108 : */
1109 12 : hash = DatumGetUInt32(hash_any((unsigned char *) authoritative_data,
1110 : Min(len, PG_CACHE_LINE_SIZE)));
1111 :
1112 12 : if (len > PG_CACHE_LINE_SIZE)
1113 0 : hash ^= DatumGetUInt32(hash_uint32((uint32) len));
1114 :
1115 12 : addHyperLogLog(&bss->full_card, hash);
1116 :
1117 : /* Hash abbreviated key */
1118 : {
1119 : uint32 tmp;
1120 :
1121 12 : tmp = DatumGetUInt32(res) ^ (uint32) (DatumGetUInt64(res) >> 32);
1122 12 : hash = DatumGetUInt32(hash_uint32(tmp));
1123 : }
1124 :
1125 12 : addHyperLogLog(&bss->abbr_card, hash);
1126 :
1127 : /*
1128 : * Byteswap on little-endian machines.
1129 : *
1130 : * This is needed so that ssup_datum_unsigned_cmp() works correctly on all
1131 : * platforms.
1132 : */
1133 12 : res = DatumBigEndianToNative(res);
1134 :
1135 : /* Don't leak memory here */
1136 12 : if (PointerGetDatum(authoritative) != original)
1137 0 : pfree(authoritative);
1138 :
1139 12 : return res;
1140 : }
1141 :
1142 : /*
1143 : * Callback for estimating effectiveness of abbreviated key optimization, using
1144 : * heuristic rules. Returns value indicating if the abbreviation optimization
1145 : * should be aborted, based on its projected effectiveness.
1146 : *
1147 : * This is based on varstr_abbrev_abort(), but some comments have been elided
1148 : * for brevity. See there for more details.
1149 : */
1150 : static bool
1151 0 : bytea_abbrev_abort(int memtupcount, SortSupport ssup)
1152 : {
1153 0 : ByteaSortSupport *bss = (ByteaSortSupport *) ssup->ssup_extra;
1154 : double abbrev_distinct,
1155 : key_distinct;
1156 :
1157 : Assert(ssup->abbreviate);
1158 :
1159 : /* Have a little patience */
1160 0 : if (memtupcount < 100)
1161 0 : return false;
1162 :
1163 0 : abbrev_distinct = estimateHyperLogLog(&bss->abbr_card);
1164 0 : key_distinct = estimateHyperLogLog(&bss->full_card);
1165 :
1166 : /*
1167 : * Clamp cardinality estimates to at least one distinct value. While
1168 : * NULLs are generally disregarded, if only NULL values were seen so far,
1169 : * that might misrepresent costs if we failed to clamp.
1170 : */
1171 0 : if (abbrev_distinct < 1.0)
1172 0 : abbrev_distinct = 1.0;
1173 :
1174 0 : if (key_distinct < 1.0)
1175 0 : key_distinct = 1.0;
1176 :
1177 0 : if (trace_sort)
1178 : {
1179 0 : double norm_abbrev_card = abbrev_distinct / (double) memtupcount;
1180 :
1181 0 : elog(LOG, "bytea_abbrev: abbrev_distinct after %d: %f "
1182 : "(key_distinct: %f, norm_abbrev_card: %f, prop_card: %f)",
1183 : memtupcount, abbrev_distinct, key_distinct, norm_abbrev_card,
1184 : bss->prop_card);
1185 : }
1186 :
1187 : /*
1188 : * If the number of distinct abbreviated keys approximately matches the
1189 : * number of distinct original keys, continue with abbreviation.
1190 : */
1191 0 : if (abbrev_distinct > key_distinct * bss->prop_card)
1192 : {
1193 : /*
1194 : * Decay required cardinality aggressively after 10,000 tuples.
1195 : */
1196 0 : if (memtupcount > 10000)
1197 0 : bss->prop_card *= 0.65;
1198 :
1199 0 : return false;
1200 : }
1201 :
1202 : /*
1203 : * Abort abbreviation strategy.
1204 : */
1205 0 : if (trace_sort)
1206 0 : elog(LOG, "bytea_abbrev: aborted abbreviation at %d "
1207 : "(abbrev_distinct: %f, key_distinct: %f, prop_card: %f)",
1208 : memtupcount, abbrev_distinct, key_distinct, bss->prop_card);
1209 :
1210 0 : return true;
1211 : }
1212 :
1213 : Datum
1214 72 : bytea_sortsupport(PG_FUNCTION_ARGS)
1215 : {
1216 72 : SortSupport ssup = (SortSupport) PG_GETARG_POINTER(0);
1217 : MemoryContext oldcontext;
1218 :
1219 72 : oldcontext = MemoryContextSwitchTo(ssup->ssup_cxt);
1220 :
1221 72 : ssup->comparator = byteafastcmp;
1222 :
1223 : /*
1224 : * Set up abbreviation support if requested.
1225 : */
1226 72 : if (ssup->abbreviate)
1227 : {
1228 : ByteaSortSupport *bss;
1229 :
1230 10 : bss = palloc_object(ByteaSortSupport);
1231 10 : bss->abbreviate = true;
1232 10 : bss->prop_card = 0.20;
1233 10 : initHyperLogLog(&bss->abbr_card, 10);
1234 10 : initHyperLogLog(&bss->full_card, 10);
1235 :
1236 10 : ssup->ssup_extra = bss;
1237 10 : ssup->abbrev_full_comparator = ssup->comparator;
1238 10 : ssup->comparator = ssup_datum_unsigned_cmp;
1239 10 : ssup->abbrev_converter = bytea_abbrev_convert;
1240 10 : ssup->abbrev_abort = bytea_abbrev_abort;
1241 : }
1242 :
1243 72 : MemoryContextSwitchTo(oldcontext);
1244 :
1245 72 : PG_RETURN_VOID();
1246 : }
1247 :
1248 : /* Cast bytea -> int2 */
1249 : Datum
1250 18 : bytea_int2(PG_FUNCTION_ARGS)
1251 : {
1252 18 : bytea *v = PG_GETARG_BYTEA_PP(0);
1253 18 : int len = VARSIZE_ANY_EXHDR(v);
1254 : uint16 result;
1255 :
1256 : /* Check that the byte array is not too long */
1257 18 : if (len > sizeof(result))
1258 3 : ereport(ERROR,
1259 : errcode(ERRCODE_NUMERIC_VALUE_OUT_OF_RANGE),
1260 : errmsg("smallint out of range"));
1261 :
1262 : /* Convert it to an integer; most significant bytes come first */
1263 15 : result = 0;
1264 36 : for (int i = 0; i < len; i++)
1265 : {
1266 21 : result <<= BITS_PER_BYTE;
1267 21 : result |= ((unsigned char *) VARDATA_ANY(v))[i];
1268 : }
1269 :
1270 15 : PG_RETURN_INT16(result);
1271 : }
1272 :
1273 : /* Cast bytea -> int4 */
1274 : Datum
1275 18 : bytea_int4(PG_FUNCTION_ARGS)
1276 : {
1277 18 : bytea *v = PG_GETARG_BYTEA_PP(0);
1278 18 : int len = VARSIZE_ANY_EXHDR(v);
1279 : uint32 result;
1280 :
1281 : /* Check that the byte array is not too long */
1282 18 : if (len > sizeof(result))
1283 3 : ereport(ERROR,
1284 : errcode(ERRCODE_NUMERIC_VALUE_OUT_OF_RANGE),
1285 : errmsg("integer out of range"));
1286 :
1287 : /* Convert it to an integer; most significant bytes come first */
1288 15 : result = 0;
1289 54 : for (int i = 0; i < len; i++)
1290 : {
1291 39 : result <<= BITS_PER_BYTE;
1292 39 : result |= ((unsigned char *) VARDATA_ANY(v))[i];
1293 : }
1294 :
1295 15 : PG_RETURN_INT32(result);
1296 : }
1297 :
1298 : /* Cast bytea -> int8 */
1299 : Datum
1300 18 : bytea_int8(PG_FUNCTION_ARGS)
1301 : {
1302 18 : bytea *v = PG_GETARG_BYTEA_PP(0);
1303 18 : int len = VARSIZE_ANY_EXHDR(v);
1304 : uint64 result;
1305 :
1306 : /* Check that the byte array is not too long */
1307 18 : if (len > sizeof(result))
1308 3 : ereport(ERROR,
1309 : errcode(ERRCODE_NUMERIC_VALUE_OUT_OF_RANGE),
1310 : errmsg("bigint out of range"));
1311 :
1312 : /* Convert it to an integer; most significant bytes come first */
1313 15 : result = 0;
1314 90 : for (int i = 0; i < len; i++)
1315 : {
1316 75 : result <<= BITS_PER_BYTE;
1317 75 : result |= ((unsigned char *) VARDATA_ANY(v))[i];
1318 : }
1319 :
1320 15 : PG_RETURN_INT64(result);
1321 : }
1322 :
1323 : /* Cast int2 -> bytea; can just use int2send() */
1324 : Datum
1325 6 : int2_bytea(PG_FUNCTION_ARGS)
1326 : {
1327 6 : return int2send(fcinfo);
1328 : }
1329 :
1330 : /* Cast int4 -> bytea; can just use int4send() */
1331 : Datum
1332 20486 : int4_bytea(PG_FUNCTION_ARGS)
1333 : {
1334 20486 : return int4send(fcinfo);
1335 : }
1336 :
1337 : /* Cast int8 -> bytea; can just use int8send() */
1338 : Datum
1339 6 : int8_bytea(PG_FUNCTION_ARGS)
1340 : {
1341 6 : return int8send(fcinfo);
1342 : }
|
