Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * datum.c
4 : * POSTGRES Datum (abstract data type) manipulation routines.
5 : *
6 : * Portions Copyright (c) 1996-2025, PostgreSQL Global Development Group
7 : * Portions Copyright (c) 1994, Regents of the University of California
8 : *
9 : *
10 : * IDENTIFICATION
11 : * src/backend/utils/adt/datum.c
12 : *
13 : *-------------------------------------------------------------------------
14 : */
15 :
16 : /*
17 : * In the implementation of these routines we assume the following:
18 : *
19 : * A) if a type is "byVal" then all the information is stored in the
20 : * Datum itself (i.e. no pointers involved!). In this case the
21 : * length of the type is always greater than zero and not more than
22 : * "sizeof(Datum)"
23 : *
24 : * B) if a type is not "byVal" and it has a fixed length (typlen > 0),
25 : * then the "Datum" always contains a pointer to a stream of bytes.
26 : * The number of significant bytes are always equal to the typlen.
27 : *
28 : * C) if a type is not "byVal" and has typlen == -1,
29 : * then the "Datum" always points to a "struct varlena".
30 : * This varlena structure has information about the actual length of this
31 : * particular instance of the type and about its value.
32 : *
33 : * D) if a type is not "byVal" and has typlen == -2,
34 : * then the "Datum" always points to a null-terminated C string.
35 : *
36 : * Note that we do not treat "toasted" datums specially; therefore what
37 : * will be copied or compared is the compressed data or toast reference.
38 : * An exception is made for datumCopy() of an expanded object, however,
39 : * because most callers expect to get a simple contiguous (and pfree'able)
40 : * result from datumCopy(). See also datumTransfer().
41 : */
42 :
43 : #include "postgres.h"
44 :
45 : #include "access/detoast.h"
46 : #include "common/hashfn.h"
47 : #include "fmgr.h"
48 : #include "utils/datum.h"
49 : #include "utils/expandeddatum.h"
50 : #include "utils/fmgrprotos.h"
51 :
52 :
53 : /*-------------------------------------------------------------------------
54 : * datumGetSize
55 : *
56 : * Find the "real" size of a datum, given the datum value,
57 : * whether it is a "by value", and the declared type length.
58 : * (For TOAST pointer datums, this is the size of the pointer datum.)
59 : *
60 : * This is essentially an out-of-line version of the att_addlength_datum()
61 : * macro in access/tupmacs.h. We do a tad more error checking though.
62 : *-------------------------------------------------------------------------
63 : */
64 : Size
65 11082052 : datumGetSize(Datum value, bool typByVal, int typLen)
66 : {
67 : Size size;
68 :
69 11082052 : if (typByVal)
70 : {
71 : /* Pass-by-value types are always fixed-length */
72 : Assert(typLen > 0 && typLen <= sizeof(Datum));
73 1570616 : size = (Size) typLen;
74 : }
75 : else
76 : {
77 9511436 : if (typLen > 0)
78 : {
79 : /* Fixed-length pass-by-ref type */
80 6685710 : size = (Size) typLen;
81 : }
82 2825726 : else if (typLen == -1)
83 : {
84 : /* It is a varlena datatype */
85 2406374 : struct varlena *s = (struct varlena *) DatumGetPointer(value);
86 :
87 2406374 : if (!PointerIsValid(s))
88 0 : ereport(ERROR,
89 : (errcode(ERRCODE_DATA_EXCEPTION),
90 : errmsg("invalid Datum pointer")));
91 :
92 2406374 : size = (Size) VARSIZE_ANY(s);
93 : }
94 419352 : else if (typLen == -2)
95 : {
96 : /* It is a cstring datatype */
97 419352 : char *s = (char *) DatumGetPointer(value);
98 :
99 419352 : if (!PointerIsValid(s))
100 0 : ereport(ERROR,
101 : (errcode(ERRCODE_DATA_EXCEPTION),
102 : errmsg("invalid Datum pointer")));
103 :
104 419352 : size = (Size) (strlen(s) + 1);
105 : }
106 : else
107 : {
108 0 : elog(ERROR, "invalid typLen: %d", typLen);
109 : size = 0; /* keep compiler quiet */
110 : }
111 : }
112 :
113 11082052 : return size;
114 : }
115 :
116 : /*-------------------------------------------------------------------------
117 : * datumCopy
118 : *
119 : * Make a copy of a non-NULL datum.
120 : *
121 : * If the datatype is pass-by-reference, memory is obtained with palloc().
122 : *
123 : * If the value is a reference to an expanded object, we flatten into memory
124 : * obtained with palloc(). We need to copy because one of the main uses of
125 : * this function is to copy a datum out of a transient memory context that's
126 : * about to be destroyed, and the expanded object is probably in a child
127 : * context that will also go away. Moreover, many callers assume that the
128 : * result is a single pfree-able chunk.
129 : *-------------------------------------------------------------------------
130 : */
131 : Datum
132 25177584 : datumCopy(Datum value, bool typByVal, int typLen)
133 : {
134 : Datum res;
135 :
136 25177584 : if (typByVal)
137 12734334 : res = value;
138 12443250 : else if (typLen == -1)
139 : {
140 : /* It is a varlena datatype */
141 6815210 : struct varlena *vl = (struct varlena *) DatumGetPointer(value);
142 :
143 6815210 : if (VARATT_IS_EXTERNAL_EXPANDED(vl))
144 944 : {
145 : /* Flatten into the caller's memory context */
146 944 : ExpandedObjectHeader *eoh = DatumGetEOHP(value);
147 : Size resultsize;
148 : char *resultptr;
149 :
150 944 : resultsize = EOH_get_flat_size(eoh);
151 944 : resultptr = (char *) palloc(resultsize);
152 944 : EOH_flatten_into(eoh, resultptr, resultsize);
153 944 : res = PointerGetDatum(resultptr);
154 : }
155 : else
156 : {
157 : /* Otherwise, just copy the varlena datum verbatim */
158 : Size realSize;
159 : char *resultptr;
160 :
161 6814266 : realSize = (Size) VARSIZE_ANY(vl);
162 6814266 : resultptr = (char *) palloc(realSize);
163 6814266 : memcpy(resultptr, vl, realSize);
164 6814266 : res = PointerGetDatum(resultptr);
165 : }
166 : }
167 : else
168 : {
169 : /* Pass by reference, but not varlena, so not toasted */
170 : Size realSize;
171 : char *resultptr;
172 :
173 5628040 : realSize = datumGetSize(value, typByVal, typLen);
174 :
175 5628040 : resultptr = (char *) palloc(realSize);
176 5628040 : memcpy(resultptr, DatumGetPointer(value), realSize);
177 5628040 : res = PointerGetDatum(resultptr);
178 : }
179 25177584 : return res;
180 : }
181 :
182 : /*-------------------------------------------------------------------------
183 : * datumTransfer
184 : *
185 : * Transfer a non-NULL datum into the current memory context.
186 : *
187 : * This is equivalent to datumCopy() except when the datum is a read-write
188 : * pointer to an expanded object. In that case we merely reparent the object
189 : * into the current context, and return its standard R/W pointer (in case the
190 : * given one is a transient pointer of shorter lifespan).
191 : *-------------------------------------------------------------------------
192 : */
193 : Datum
194 115962 : datumTransfer(Datum value, bool typByVal, int typLen)
195 : {
196 115962 : if (!typByVal && typLen == -1 &&
197 113304 : VARATT_IS_EXTERNAL_EXPANDED_RW(DatumGetPointer(value)))
198 3750 : value = TransferExpandedObject(value, CurrentMemoryContext);
199 : else
200 112212 : value = datumCopy(value, typByVal, typLen);
201 115962 : return value;
202 : }
203 :
204 : /*-------------------------------------------------------------------------
205 : * datumIsEqual
206 : *
207 : * Return true if two datums are equal, false otherwise
208 : *
209 : * NOTE: XXX!
210 : * We just compare the bytes of the two values, one by one.
211 : * This routine will return false if there are 2 different
212 : * representations of the same value (something along the lines
213 : * of say the representation of zero in one's complement arithmetic).
214 : * Also, it will probably not give the answer you want if either
215 : * datum has been "toasted".
216 : *
217 : * Do not try to make this any smarter than it currently is with respect
218 : * to "toasted" datums, because some of the callers could be working in the
219 : * context of an aborted transaction.
220 : *-------------------------------------------------------------------------
221 : */
222 : bool
223 3786200 : datumIsEqual(Datum value1, Datum value2, bool typByVal, int typLen)
224 : {
225 : bool res;
226 :
227 3786200 : if (typByVal)
228 : {
229 : /*
230 : * just compare the two datums. NOTE: just comparing "len" bytes will
231 : * not do the work, because we do not know how these bytes are aligned
232 : * inside the "Datum". We assume instead that any given datatype is
233 : * consistent about how it fills extraneous bits in the Datum.
234 : */
235 2670380 : res = (value1 == value2);
236 : }
237 : else
238 : {
239 : Size size1,
240 : size2;
241 : char *s1,
242 : *s2;
243 :
244 : /*
245 : * Compare the bytes pointed by the pointers stored in the datums.
246 : */
247 1115820 : size1 = datumGetSize(value1, typByVal, typLen);
248 1115820 : size2 = datumGetSize(value2, typByVal, typLen);
249 1115820 : if (size1 != size2)
250 13534 : return false;
251 1102286 : s1 = (char *) DatumGetPointer(value1);
252 1102286 : s2 = (char *) DatumGetPointer(value2);
253 1102286 : res = (memcmp(s1, s2, size1) == 0);
254 : }
255 3772666 : return res;
256 : }
257 :
258 : /*-------------------------------------------------------------------------
259 : * datum_image_eq
260 : *
261 : * Compares two datums for identical contents, based on byte images. Return
262 : * true if the two datums are equal, false otherwise.
263 : *-------------------------------------------------------------------------
264 : */
265 : bool
266 20661610 : datum_image_eq(Datum value1, Datum value2, bool typByVal, int typLen)
267 : {
268 : Size len1,
269 : len2;
270 20661610 : bool result = true;
271 :
272 20661610 : if (typByVal)
273 : {
274 17320378 : result = (value1 == value2);
275 : }
276 3341232 : else if (typLen > 0)
277 : {
278 2401080 : result = (memcmp(DatumGetPointer(value1),
279 2401080 : DatumGetPointer(value2),
280 : typLen) == 0);
281 : }
282 940152 : else if (typLen == -1)
283 : {
284 395758 : len1 = toast_raw_datum_size(value1);
285 395758 : len2 = toast_raw_datum_size(value2);
286 : /* No need to de-toast if lengths don't match. */
287 395758 : if (len1 != len2)
288 27988 : result = false;
289 : else
290 : {
291 : struct varlena *arg1val;
292 : struct varlena *arg2val;
293 :
294 367770 : arg1val = PG_DETOAST_DATUM_PACKED(value1);
295 367770 : arg2val = PG_DETOAST_DATUM_PACKED(value2);
296 :
297 367770 : result = (memcmp(VARDATA_ANY(arg1val),
298 367770 : VARDATA_ANY(arg2val),
299 : len1 - VARHDRSZ) == 0);
300 :
301 : /* Only free memory if it's a copy made here. */
302 367770 : if ((Pointer) arg1val != (Pointer) value1)
303 6 : pfree(arg1val);
304 367770 : if ((Pointer) arg2val != (Pointer) value2)
305 6 : pfree(arg2val);
306 : }
307 : }
308 544394 : else if (typLen == -2)
309 : {
310 : char *s1,
311 : *s2;
312 :
313 : /* Compare cstring datums */
314 544394 : s1 = DatumGetCString(value1);
315 544394 : s2 = DatumGetCString(value2);
316 544394 : len1 = strlen(s1) + 1;
317 544394 : len2 = strlen(s2) + 1;
318 544394 : if (len1 != len2)
319 295332 : return false;
320 249062 : result = (memcmp(s1, s2, len1) == 0);
321 : }
322 : else
323 0 : elog(ERROR, "unexpected typLen: %d", typLen);
324 :
325 20366278 : return result;
326 : }
327 :
328 : /*-------------------------------------------------------------------------
329 : * datum_image_hash
330 : *
331 : * Generate a hash value based on the binary representation of 'value'. Most
332 : * use cases will want to use the hash function specific to the Datum's type,
333 : * however, some corner cases require generating a hash value based on the
334 : * actual bits rather than the logical value.
335 : *-------------------------------------------------------------------------
336 : */
337 : uint32
338 120716 : datum_image_hash(Datum value, bool typByVal, int typLen)
339 : {
340 : Size len;
341 : uint32 result;
342 :
343 120716 : if (typByVal)
344 120512 : result = hash_bytes((unsigned char *) &value, sizeof(Datum));
345 204 : else if (typLen > 0)
346 36 : result = hash_bytes((unsigned char *) DatumGetPointer(value), typLen);
347 168 : else if (typLen == -1)
348 : {
349 : struct varlena *val;
350 :
351 168 : len = toast_raw_datum_size(value);
352 :
353 168 : val = PG_DETOAST_DATUM_PACKED(value);
354 :
355 168 : result = hash_bytes((unsigned char *) VARDATA_ANY(val), len - VARHDRSZ);
356 :
357 : /* Only free memory if it's a copy made here. */
358 168 : if ((Pointer) val != (Pointer) value)
359 12 : pfree(val);
360 : }
361 0 : else if (typLen == -2)
362 : {
363 : char *s;
364 :
365 0 : s = DatumGetCString(value);
366 0 : len = strlen(s) + 1;
367 :
368 0 : result = hash_bytes((unsigned char *) s, len);
369 : }
370 : else
371 : {
372 0 : elog(ERROR, "unexpected typLen: %d", typLen);
373 : result = 0; /* keep compiler quiet */
374 : }
375 :
376 120716 : return result;
377 : }
378 :
379 : /*-------------------------------------------------------------------------
380 : * btequalimage
381 : *
382 : * Generic "equalimage" support function.
383 : *
384 : * B-Tree operator classes whose equality function could safely be replaced by
385 : * datum_image_eq() in all cases can use this as their "equalimage" support
386 : * function.
387 : *
388 : * Currently, we unconditionally assume that any B-Tree operator class that
389 : * registers btequalimage as its support function 4 must be able to safely use
390 : * optimizations like deduplication (i.e. we return true unconditionally). If
391 : * it ever proved necessary to rescind support for an operator class, we could
392 : * do that in a targeted fashion by doing something with the opcintype
393 : * argument.
394 : *-------------------------------------------------------------------------
395 : */
396 : Datum
397 89336 : btequalimage(PG_FUNCTION_ARGS)
398 : {
399 : /* Oid opcintype = PG_GETARG_OID(0); */
400 :
401 89336 : PG_RETURN_BOOL(true);
402 : }
403 :
404 : /*-------------------------------------------------------------------------
405 : * datumEstimateSpace
406 : *
407 : * Compute the amount of space that datumSerialize will require for a
408 : * particular Datum.
409 : *-------------------------------------------------------------------------
410 : */
411 : Size
412 150 : datumEstimateSpace(Datum value, bool isnull, bool typByVal, int typLen)
413 : {
414 150 : Size sz = sizeof(int);
415 :
416 150 : if (!isnull)
417 : {
418 : /* no need to use add_size, can't overflow */
419 150 : if (typByVal)
420 132 : sz += sizeof(Datum);
421 18 : else if (typLen == -1 &&
422 18 : VARATT_IS_EXTERNAL_EXPANDED(DatumGetPointer(value)))
423 : {
424 : /* Expanded objects need to be flattened, see comment below */
425 6 : sz += EOH_get_flat_size(DatumGetEOHP(value));
426 : }
427 : else
428 12 : sz += datumGetSize(value, typByVal, typLen);
429 : }
430 :
431 150 : return sz;
432 : }
433 :
434 : /*-------------------------------------------------------------------------
435 : * datumSerialize
436 : *
437 : * Serialize a possibly-NULL datum into caller-provided storage.
438 : *
439 : * Note: "expanded" objects are flattened so as to produce a self-contained
440 : * representation, but other sorts of toast pointers are transferred as-is.
441 : * This is because the intended use of this function is to pass the value
442 : * to another process within the same database server. The other process
443 : * could not access an "expanded" object within this process's memory, but
444 : * we assume it can dereference the same TOAST pointers this one can.
445 : *
446 : * The format is as follows: first, we write a 4-byte header word, which
447 : * is either the length of a pass-by-reference datum, -1 for a
448 : * pass-by-value datum, or -2 for a NULL. If the value is NULL, nothing
449 : * further is written. If it is pass-by-value, sizeof(Datum) bytes
450 : * follow. Otherwise, the number of bytes indicated by the header word
451 : * follow. The caller is responsible for ensuring that there is enough
452 : * storage to store the number of bytes that will be written; use
453 : * datumEstimateSpace() to find out how many will be needed.
454 : * *start_address is updated to point to the byte immediately following
455 : * those written.
456 : *-------------------------------------------------------------------------
457 : */
458 : void
459 114 : datumSerialize(Datum value, bool isnull, bool typByVal, int typLen,
460 : char **start_address)
461 : {
462 114 : ExpandedObjectHeader *eoh = NULL;
463 : int header;
464 :
465 : /* Write header word. */
466 114 : if (isnull)
467 0 : header = -2;
468 114 : else if (typByVal)
469 96 : header = -1;
470 18 : else if (typLen == -1 &&
471 18 : VARATT_IS_EXTERNAL_EXPANDED(DatumGetPointer(value)))
472 : {
473 6 : eoh = DatumGetEOHP(value);
474 6 : header = EOH_get_flat_size(eoh);
475 : }
476 : else
477 12 : header = datumGetSize(value, typByVal, typLen);
478 114 : memcpy(*start_address, &header, sizeof(int));
479 114 : *start_address += sizeof(int);
480 :
481 : /* If not null, write payload bytes. */
482 114 : if (!isnull)
483 : {
484 114 : if (typByVal)
485 : {
486 96 : memcpy(*start_address, &value, sizeof(Datum));
487 96 : *start_address += sizeof(Datum);
488 : }
489 18 : else if (eoh)
490 : {
491 : char *tmp;
492 :
493 : /*
494 : * EOH_flatten_into expects the target address to be maxaligned,
495 : * so we can't store directly to *start_address.
496 : */
497 6 : tmp = (char *) palloc(header);
498 6 : EOH_flatten_into(eoh, tmp, header);
499 6 : memcpy(*start_address, tmp, header);
500 6 : *start_address += header;
501 :
502 : /* be tidy. */
503 6 : pfree(tmp);
504 : }
505 : else
506 : {
507 12 : memcpy(*start_address, DatumGetPointer(value), header);
508 12 : *start_address += header;
509 : }
510 : }
511 114 : }
512 :
513 : /*-------------------------------------------------------------------------
514 : * datumRestore
515 : *
516 : * Restore a possibly-NULL datum previously serialized by datumSerialize.
517 : * *start_address is updated according to the number of bytes consumed.
518 : *-------------------------------------------------------------------------
519 : */
520 : Datum
521 276 : datumRestore(char **start_address, bool *isnull)
522 : {
523 : int header;
524 : void *d;
525 :
526 : /* Read header word. */
527 276 : memcpy(&header, *start_address, sizeof(int));
528 276 : *start_address += sizeof(int);
529 :
530 : /* If this datum is NULL, we can stop here. */
531 276 : if (header == -2)
532 : {
533 0 : *isnull = true;
534 0 : return (Datum) 0;
535 : }
536 :
537 : /* OK, datum is not null. */
538 276 : *isnull = false;
539 :
540 : /* If this datum is pass-by-value, sizeof(Datum) bytes follow. */
541 276 : if (header == -1)
542 : {
543 : Datum val;
544 :
545 216 : memcpy(&val, *start_address, sizeof(Datum));
546 216 : *start_address += sizeof(Datum);
547 216 : return val;
548 : }
549 :
550 : /* Pass-by-reference case; copy indicated number of bytes. */
551 : Assert(header > 0);
552 60 : d = palloc(header);
553 60 : memcpy(d, *start_address, header);
554 60 : *start_address += header;
555 60 : return PointerGetDatum(d);
556 : }
|
