Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * datum.c
4 : * POSTGRES Datum (abstract data type) manipulation routines.
5 : *
6 : * Portions Copyright (c) 1996-2026, 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 "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 5703037 : datumGetSize(Datum value, bool typByVal, int typLen)
66 : {
67 : Size size;
68 :
69 5703037 : if (typByVal)
70 : {
71 : /* Pass-by-value types are always fixed-length */
72 : Assert(typLen > 0 && typLen <= sizeof(Datum));
73 837393 : size = (Size) typLen;
74 : }
75 : else
76 : {
77 4865644 : if (typLen > 0)
78 : {
79 : /* Fixed-length pass-by-ref type */
80 3387096 : size = (Size) typLen;
81 : }
82 1478548 : else if (typLen == -1)
83 : {
84 : /* It is a varlena datatype */
85 1268140 : varlena *s = (varlena *) DatumGetPointer(value);
86 :
87 1268140 : if (!s)
88 0 : ereport(ERROR,
89 : (errcode(ERRCODE_DATA_EXCEPTION),
90 : errmsg("invalid Datum pointer")));
91 :
92 1268140 : size = (Size) VARSIZE_ANY(s);
93 : }
94 210408 : else if (typLen == -2)
95 : {
96 : /* It is a cstring datatype */
97 210408 : char *s = (char *) DatumGetPointer(value);
98 :
99 210408 : if (!s)
100 0 : ereport(ERROR,
101 : (errcode(ERRCODE_DATA_EXCEPTION),
102 : errmsg("invalid Datum pointer")));
103 :
104 210408 : 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 5703037 : 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 12923904 : datumCopy(Datum value, bool typByVal, int typLen)
133 : {
134 : Datum res;
135 :
136 12923904 : if (typByVal)
137 6588633 : res = value;
138 6335271 : else if (typLen == -1)
139 : {
140 : /* It is a varlena datatype */
141 3491346 : varlena *vl = (varlena *) DatumGetPointer(value);
142 :
143 3491346 : if (VARATT_IS_EXTERNAL_EXPANDED(vl))
144 : {
145 : /* Flatten into the caller's memory context */
146 472 : ExpandedObjectHeader *eoh = DatumGetEOHP(value);
147 : Size resultsize;
148 : char *resultptr;
149 :
150 472 : resultsize = EOH_get_flat_size(eoh);
151 472 : resultptr = (char *) palloc(resultsize);
152 472 : EOH_flatten_into(eoh, resultptr, resultsize);
153 472 : res = PointerGetDatum(resultptr);
154 : }
155 : else
156 : {
157 : /* Otherwise, just copy the varlena datum verbatim */
158 : Size realSize;
159 : char *resultptr;
160 :
161 3490874 : realSize = (Size) VARSIZE_ANY(vl);
162 3490874 : resultptr = (char *) palloc(realSize);
163 3490874 : memcpy(resultptr, vl, realSize);
164 3490874 : 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 2843925 : realSize = datumGetSize(value, typByVal, typLen);
174 :
175 2843925 : resultptr = (char *) palloc(realSize);
176 2843925 : memcpy(resultptr, DatumGetPointer(value), realSize);
177 2843925 : res = PointerGetDatum(resultptr);
178 : }
179 12923904 : 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 59528 : datumTransfer(Datum value, bool typByVal, int typLen)
195 : {
196 117273 : if (!typByVal && typLen == -1 &&
197 57745 : VARATT_IS_EXTERNAL_EXPANDED_RW(DatumGetPointer(value)))
198 1996 : value = TransferExpandedObject(value, CurrentMemoryContext);
199 : else
200 57532 : value = datumCopy(value, typByVal, typLen);
201 59528 : 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 1998648 : datumIsEqual(Datum value1, Datum value2, bool typByVal, int typLen)
224 : {
225 : bool res;
226 :
227 1998648 : 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 1411298 : 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 587350 : size1 = datumGetSize(value1, typByVal, typLen);
248 587350 : size2 = datumGetSize(value2, typByVal, typLen);
249 587350 : if (size1 != size2)
250 7398 : return false;
251 579952 : s1 = (char *) DatumGetPointer(value1);
252 579952 : s2 = (char *) DatumGetPointer(value2);
253 579952 : res = (memcmp(s1, s2, size1) == 0);
254 : }
255 1991250 : 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 10482279 : datum_image_eq(Datum value1, Datum value2, bool typByVal, int typLen)
267 : {
268 : Size len1,
269 : len2;
270 10482279 : bool result = true;
271 :
272 10482279 : if (typByVal)
273 : {
274 8862165 : result = (value1 == value2);
275 : }
276 1620114 : else if (typLen > 0)
277 : {
278 1200540 : result = (memcmp(DatumGetPointer(value1),
279 1200540 : DatumGetPointer(value2),
280 : typLen) == 0);
281 : }
282 419574 : else if (typLen == -1)
283 : {
284 164590 : len1 = toast_raw_datum_size(value1);
285 164590 : len2 = toast_raw_datum_size(value2);
286 : /* No need to de-toast if lengths don't match. */
287 164590 : if (len1 != len2)
288 11426 : result = false;
289 : else
290 : {
291 : varlena *arg1val;
292 : varlena *arg2val;
293 :
294 153164 : arg1val = PG_DETOAST_DATUM_PACKED(value1);
295 153164 : arg2val = PG_DETOAST_DATUM_PACKED(value2);
296 :
297 153164 : result = (memcmp(VARDATA_ANY(arg1val),
298 153164 : VARDATA_ANY(arg2val),
299 : len1 - VARHDRSZ) == 0);
300 :
301 : /* Only free memory if it's a copy made here. */
302 153164 : if (arg1val != DatumGetPointer(value1))
303 3 : pfree(arg1val);
304 153164 : if (arg2val != DatumGetPointer(value2))
305 3 : pfree(arg2val);
306 : }
307 : }
308 254984 : else if (typLen == -2)
309 : {
310 : char *s1,
311 : *s2;
312 :
313 : /* Compare cstring datums */
314 254984 : s1 = DatumGetCString(value1);
315 254984 : s2 = DatumGetCString(value2);
316 254984 : len1 = strlen(s1) + 1;
317 254984 : len2 = strlen(s2) + 1;
318 254984 : if (len1 != len2)
319 140293 : return false;
320 114691 : result = (memcmp(s1, s2, len1) == 0);
321 : }
322 : else
323 0 : elog(ERROR, "unexpected typLen: %d", typLen);
324 :
325 10341986 : 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 60394 : datum_image_hash(Datum value, bool typByVal, int typLen)
339 : {
340 : Size len;
341 : uint32 result;
342 :
343 60394 : if (typByVal)
344 60292 : result = hash_bytes((unsigned char *) &value, sizeof(Datum));
345 102 : else if (typLen > 0)
346 18 : result = hash_bytes((unsigned char *) DatumGetPointer(value), typLen);
347 84 : else if (typLen == -1)
348 : {
349 : varlena *val;
350 :
351 84 : len = toast_raw_datum_size(value);
352 :
353 84 : val = PG_DETOAST_DATUM_PACKED(value);
354 :
355 84 : result = hash_bytes((unsigned char *) VARDATA_ANY(val), len - VARHDRSZ);
356 :
357 : /* Only free memory if it's a copy made here. */
358 84 : if (val != DatumGetPointer(value))
359 6 : 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 60394 : 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 44917 : btequalimage(PG_FUNCTION_ARGS)
398 : {
399 : #ifdef NOT_USED
400 : Oid opcintype = PG_GETARG_OID(0);
401 : #endif
402 :
403 44917 : PG_RETURN_BOOL(true);
404 : }
405 :
406 : /*-------------------------------------------------------------------------
407 : * datumEstimateSpace
408 : *
409 : * Compute the amount of space that datumSerialize will require for a
410 : * particular Datum.
411 : *-------------------------------------------------------------------------
412 : */
413 : Size
414 75 : datumEstimateSpace(Datum value, bool isnull, bool typByVal, int typLen)
415 : {
416 75 : Size sz = sizeof(int);
417 :
418 75 : if (!isnull)
419 : {
420 : /* no need to use add_size, can't overflow */
421 75 : if (typByVal)
422 66 : sz += sizeof(Datum);
423 18 : else if (typLen == -1 &&
424 9 : VARATT_IS_EXTERNAL_EXPANDED(DatumGetPointer(value)))
425 : {
426 : /* Expanded objects need to be flattened, see comment below */
427 3 : sz += EOH_get_flat_size(DatumGetEOHP(value));
428 : }
429 : else
430 6 : sz += datumGetSize(value, typByVal, typLen);
431 : }
432 :
433 75 : return sz;
434 : }
435 :
436 : /*-------------------------------------------------------------------------
437 : * datumSerialize
438 : *
439 : * Serialize a possibly-NULL datum into caller-provided storage.
440 : *
441 : * Note: "expanded" objects are flattened so as to produce a self-contained
442 : * representation, but other sorts of toast pointers are transferred as-is.
443 : * This is because the intended use of this function is to pass the value
444 : * to another process within the same database server. The other process
445 : * could not access an "expanded" object within this process's memory, but
446 : * we assume it can dereference the same TOAST pointers this one can.
447 : *
448 : * The format is as follows: first, we write a 4-byte header word, which
449 : * is either the length of a pass-by-reference datum, -1 for a
450 : * pass-by-value datum, or -2 for a NULL. If the value is NULL, nothing
451 : * further is written. If it is pass-by-value, sizeof(Datum) bytes
452 : * follow. Otherwise, the number of bytes indicated by the header word
453 : * follow. The caller is responsible for ensuring that there is enough
454 : * storage to store the number of bytes that will be written; use
455 : * datumEstimateSpace() to find out how many will be needed.
456 : * *start_address is updated to point to the byte immediately following
457 : * those written.
458 : *-------------------------------------------------------------------------
459 : */
460 : void
461 57 : datumSerialize(Datum value, bool isnull, bool typByVal, int typLen,
462 : char **start_address)
463 : {
464 57 : ExpandedObjectHeader *eoh = NULL;
465 : int header;
466 :
467 : /* Write header word. */
468 57 : if (isnull)
469 0 : header = -2;
470 57 : else if (typByVal)
471 48 : header = -1;
472 18 : else if (typLen == -1 &&
473 9 : VARATT_IS_EXTERNAL_EXPANDED(DatumGetPointer(value)))
474 : {
475 3 : eoh = DatumGetEOHP(value);
476 3 : header = EOH_get_flat_size(eoh);
477 : }
478 : else
479 6 : header = datumGetSize(value, typByVal, typLen);
480 57 : memcpy(*start_address, &header, sizeof(int));
481 57 : *start_address += sizeof(int);
482 :
483 : /* If not null, write payload bytes. */
484 57 : if (!isnull)
485 : {
486 57 : if (typByVal)
487 : {
488 48 : memcpy(*start_address, &value, sizeof(Datum));
489 48 : *start_address += sizeof(Datum);
490 : }
491 9 : else if (eoh)
492 : {
493 : char *tmp;
494 :
495 : /*
496 : * EOH_flatten_into expects the target address to be maxaligned,
497 : * so we can't store directly to *start_address.
498 : */
499 3 : tmp = (char *) palloc(header);
500 3 : EOH_flatten_into(eoh, tmp, header);
501 3 : memcpy(*start_address, tmp, header);
502 3 : *start_address += header;
503 :
504 : /* be tidy. */
505 3 : pfree(tmp);
506 : }
507 : else
508 : {
509 6 : memcpy(*start_address, DatumGetPointer(value), header);
510 6 : *start_address += header;
511 : }
512 : }
513 57 : }
514 :
515 : /*-------------------------------------------------------------------------
516 : * datumRestore
517 : *
518 : * Restore a possibly-NULL datum previously serialized by datumSerialize.
519 : * *start_address is updated according to the number of bytes consumed.
520 : *-------------------------------------------------------------------------
521 : */
522 : Datum
523 138 : datumRestore(char **start_address, bool *isnull)
524 : {
525 : int header;
526 : void *d;
527 :
528 : /* Read header word. */
529 138 : memcpy(&header, *start_address, sizeof(int));
530 138 : *start_address += sizeof(int);
531 :
532 : /* If this datum is NULL, we can stop here. */
533 138 : if (header == -2)
534 : {
535 0 : *isnull = true;
536 0 : return (Datum) 0;
537 : }
538 :
539 : /* OK, datum is not null. */
540 138 : *isnull = false;
541 :
542 : /* If this datum is pass-by-value, sizeof(Datum) bytes follow. */
543 138 : if (header == -1)
544 : {
545 : Datum val;
546 :
547 108 : memcpy(&val, *start_address, sizeof(Datum));
548 108 : *start_address += sizeof(Datum);
549 108 : return val;
550 : }
551 :
552 : /* Pass-by-reference case; copy indicated number of bytes. */
553 : Assert(header > 0);
554 30 : d = palloc(header);
555 30 : memcpy(d, *start_address, header);
556 30 : *start_address += header;
557 30 : return PointerGetDatum(d);
558 : }
|
