1 | /*------------------------------------------------------------------------- |
2 | * |
3 | * arrayfuncs.c |
4 | * Support functions for arrays. |
5 | * |
6 | * Portions Copyright (c) 1996-2019, PostgreSQL Global Development Group |
7 | * Portions Copyright (c) 1994, Regents of the University of California |
8 | * |
9 | * |
10 | * IDENTIFICATION |
11 | * src/backend/utils/adt/arrayfuncs.c |
12 | * |
13 | *------------------------------------------------------------------------- |
14 | */ |
15 | #include "postgres.h" |
16 | |
17 | #include <ctype.h> |
18 | #include <math.h> |
19 | |
20 | #include "access/htup_details.h" |
21 | #include "catalog/pg_type.h" |
22 | #include "funcapi.h" |
23 | #include "libpq/pqformat.h" |
24 | #include "nodes/nodeFuncs.h" |
25 | #include "nodes/supportnodes.h" |
26 | #include "optimizer/optimizer.h" |
27 | #include "utils/array.h" |
28 | #include "utils/arrayaccess.h" |
29 | #include "utils/builtins.h" |
30 | #include "utils/datum.h" |
31 | #include "utils/lsyscache.h" |
32 | #include "utils/memutils.h" |
33 | #include "utils/selfuncs.h" |
34 | #include "utils/typcache.h" |
35 | |
36 | |
37 | /* |
38 | * GUC parameter |
39 | */ |
40 | bool Array_nulls = true; |
41 | |
42 | /* |
43 | * Local definitions |
44 | */ |
45 | #define ASSGN "=" |
46 | |
47 | #define AARR_FREE_IF_COPY(array,n) \ |
48 | do { \ |
49 | if (!VARATT_IS_EXPANDED_HEADER(array)) \ |
50 | PG_FREE_IF_COPY(array, n); \ |
51 | } while (0) |
52 | |
53 | typedef enum |
54 | { |
55 | ARRAY_NO_LEVEL, |
56 | ARRAY_LEVEL_STARTED, |
57 | ARRAY_ELEM_STARTED, |
58 | ARRAY_ELEM_COMPLETED, |
59 | ARRAY_QUOTED_ELEM_STARTED, |
60 | ARRAY_QUOTED_ELEM_COMPLETED, |
61 | ARRAY_ELEM_DELIMITED, |
62 | ARRAY_LEVEL_COMPLETED, |
63 | ARRAY_LEVEL_DELIMITED |
64 | } ArrayParseState; |
65 | |
66 | /* Working state for array_iterate() */ |
67 | typedef struct ArrayIteratorData |
68 | { |
69 | /* basic info about the array, set up during array_create_iterator() */ |
70 | ArrayType *arr; /* array we're iterating through */ |
71 | bits8 *nullbitmap; /* its null bitmap, if any */ |
72 | int nitems; /* total number of elements in array */ |
73 | int16 typlen; /* element type's length */ |
74 | bool typbyval; /* element type's byval property */ |
75 | char typalign; /* element type's align property */ |
76 | |
77 | /* information about the requested slice size */ |
78 | int slice_ndim; /* slice dimension, or 0 if not slicing */ |
79 | int slice_len; /* number of elements per slice */ |
80 | int *slice_dims; /* slice dims array */ |
81 | int *slice_lbound; /* slice lbound array */ |
82 | Datum *slice_values; /* workspace of length slice_len */ |
83 | bool *slice_nulls; /* workspace of length slice_len */ |
84 | |
85 | /* current position information, updated on each iteration */ |
86 | char *data_ptr; /* our current position in the array */ |
87 | int current_item; /* the item # we're at in the array */ |
88 | } ArrayIteratorData; |
89 | |
90 | static bool array_isspace(char ch); |
91 | static int ArrayCount(const char *str, int *dim, char typdelim); |
92 | static void ReadArrayStr(char *arrayStr, const char *origStr, |
93 | int nitems, int ndim, int *dim, |
94 | FmgrInfo *inputproc, Oid typioparam, int32 typmod, |
95 | char typdelim, |
96 | int typlen, bool typbyval, char typalign, |
97 | Datum *values, bool *nulls, |
98 | bool *hasnulls, int32 *nbytes); |
99 | static void ReadArrayBinary(StringInfo buf, int nitems, |
100 | FmgrInfo *receiveproc, Oid typioparam, int32 typmod, |
101 | int typlen, bool typbyval, char typalign, |
102 | Datum *values, bool *nulls, |
103 | bool *hasnulls, int32 *nbytes); |
104 | static Datum array_get_element_expanded(Datum arraydatum, |
105 | int nSubscripts, int *indx, |
106 | int arraytyplen, |
107 | int elmlen, bool elmbyval, char elmalign, |
108 | bool *isNull); |
109 | static Datum array_set_element_expanded(Datum arraydatum, |
110 | int nSubscripts, int *indx, |
111 | Datum dataValue, bool isNull, |
112 | int arraytyplen, |
113 | int elmlen, bool elmbyval, char elmalign); |
114 | static bool array_get_isnull(const bits8 *nullbitmap, int offset); |
115 | static void array_set_isnull(bits8 *nullbitmap, int offset, bool isNull); |
116 | static Datum ArrayCast(char *value, bool byval, int len); |
117 | static int ArrayCastAndSet(Datum src, |
118 | int typlen, bool typbyval, char typalign, |
119 | char *dest); |
120 | static char *array_seek(char *ptr, int offset, bits8 *nullbitmap, int nitems, |
121 | int typlen, bool typbyval, char typalign); |
122 | static int array_nelems_size(char *ptr, int offset, bits8 *nullbitmap, |
123 | int nitems, int typlen, bool typbyval, char typalign); |
124 | static int array_copy(char *destptr, int nitems, |
125 | char *srcptr, int offset, bits8 *nullbitmap, |
126 | int typlen, bool typbyval, char typalign); |
127 | static int array_slice_size(char *arraydataptr, bits8 *arraynullsptr, |
128 | int ndim, int *dim, int *lb, |
129 | int *st, int *endp, |
130 | int typlen, bool typbyval, char typalign); |
131 | static void array_extract_slice(ArrayType *newarray, |
132 | int ndim, int *dim, int *lb, |
133 | char *arraydataptr, bits8 *arraynullsptr, |
134 | int *st, int *endp, |
135 | int typlen, bool typbyval, char typalign); |
136 | static void array_insert_slice(ArrayType *destArray, ArrayType *origArray, |
137 | ArrayType *srcArray, |
138 | int ndim, int *dim, int *lb, |
139 | int *st, int *endp, |
140 | int typlen, bool typbyval, char typalign); |
141 | static int array_cmp(FunctionCallInfo fcinfo); |
142 | static ArrayType *create_array_envelope(int ndims, int *dimv, int *lbv, int nbytes, |
143 | Oid elmtype, int dataoffset); |
144 | static ArrayType *array_fill_internal(ArrayType *dims, ArrayType *lbs, |
145 | Datum value, bool isnull, Oid elmtype, |
146 | FunctionCallInfo fcinfo); |
147 | static ArrayType *array_replace_internal(ArrayType *array, |
148 | Datum search, bool search_isnull, |
149 | Datum replace, bool replace_isnull, |
150 | bool remove, Oid collation, |
151 | FunctionCallInfo fcinfo); |
152 | static int width_bucket_array_float8(Datum operand, ArrayType *thresholds); |
153 | static int width_bucket_array_fixed(Datum operand, |
154 | ArrayType *thresholds, |
155 | Oid collation, |
156 | TypeCacheEntry *typentry); |
157 | static int width_bucket_array_variable(Datum operand, |
158 | ArrayType *thresholds, |
159 | Oid collation, |
160 | TypeCacheEntry *typentry); |
161 | |
162 | |
163 | /* |
164 | * array_in : |
165 | * converts an array from the external format in "string" to |
166 | * its internal format. |
167 | * |
168 | * return value : |
169 | * the internal representation of the input array |
170 | */ |
171 | Datum |
172 | array_in(PG_FUNCTION_ARGS) |
173 | { |
174 | char *string = PG_GETARG_CSTRING(0); /* external form */ |
175 | Oid element_type = PG_GETARG_OID(1); /* type of an array |
176 | * element */ |
177 | int32 typmod = PG_GETARG_INT32(2); /* typmod for array elements */ |
178 | int typlen; |
179 | bool typbyval; |
180 | char typalign; |
181 | char typdelim; |
182 | Oid typioparam; |
183 | char *string_save, |
184 | *p; |
185 | int i, |
186 | nitems; |
187 | Datum *dataPtr; |
188 | bool *nullsPtr; |
189 | bool hasnulls; |
190 | int32 nbytes; |
191 | int32 dataoffset; |
192 | ArrayType *retval; |
193 | int ndim, |
194 | dim[MAXDIM], |
195 | lBound[MAXDIM]; |
196 | ArrayMetaState *; |
197 | |
198 | /* |
199 | * We arrange to look up info about element type, including its input |
200 | * conversion proc, only once per series of calls, assuming the element |
201 | * type doesn't change underneath us. |
202 | */ |
203 | my_extra = (ArrayMetaState *) fcinfo->flinfo->fn_extra; |
204 | if (my_extra == NULL) |
205 | { |
206 | fcinfo->flinfo->fn_extra = MemoryContextAlloc(fcinfo->flinfo->fn_mcxt, |
207 | sizeof(ArrayMetaState)); |
208 | my_extra = (ArrayMetaState *) fcinfo->flinfo->fn_extra; |
209 | my_extra->element_type = ~element_type; |
210 | } |
211 | |
212 | if (my_extra->element_type != element_type) |
213 | { |
214 | /* |
215 | * Get info about element type, including its input conversion proc |
216 | */ |
217 | get_type_io_data(element_type, IOFunc_input, |
218 | &my_extra->typlen, &my_extra->typbyval, |
219 | &my_extra->typalign, &my_extra->typdelim, |
220 | &my_extra->typioparam, &my_extra->typiofunc); |
221 | fmgr_info_cxt(my_extra->typiofunc, &my_extra->proc, |
222 | fcinfo->flinfo->fn_mcxt); |
223 | my_extra->element_type = element_type; |
224 | } |
225 | typlen = my_extra->typlen; |
226 | typbyval = my_extra->typbyval; |
227 | typalign = my_extra->typalign; |
228 | typdelim = my_extra->typdelim; |
229 | typioparam = my_extra->typioparam; |
230 | |
231 | /* Make a modifiable copy of the input */ |
232 | string_save = pstrdup(string); |
233 | |
234 | /* |
235 | * If the input string starts with dimension info, read and use that. |
236 | * Otherwise, we require the input to be in curly-brace style, and we |
237 | * prescan the input to determine dimensions. |
238 | * |
239 | * Dimension info takes the form of one or more [n] or [m:n] items. The |
240 | * outer loop iterates once per dimension item. |
241 | */ |
242 | p = string_save; |
243 | ndim = 0; |
244 | for (;;) |
245 | { |
246 | char *q; |
247 | int ub; |
248 | |
249 | /* |
250 | * Note: we currently allow whitespace between, but not within, |
251 | * dimension items. |
252 | */ |
253 | while (array_isspace(*p)) |
254 | p++; |
255 | if (*p != '[') |
256 | break; /* no more dimension items */ |
257 | p++; |
258 | if (ndim >= MAXDIM) |
259 | ereport(ERROR, |
260 | (errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED), |
261 | errmsg("number of array dimensions (%d) exceeds the maximum allowed (%d)" , |
262 | ndim + 1, MAXDIM))); |
263 | |
264 | for (q = p; isdigit((unsigned char) *q) || (*q == '-') || (*q == '+'); q++) |
265 | /* skip */ ; |
266 | if (q == p) /* no digits? */ |
267 | ereport(ERROR, |
268 | (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION), |
269 | errmsg("malformed array literal: \"%s\"" , string), |
270 | errdetail("\"[\" must introduce explicitly-specified array dimensions." ))); |
271 | |
272 | if (*q == ':') |
273 | { |
274 | /* [m:n] format */ |
275 | *q = '\0'; |
276 | lBound[ndim] = atoi(p); |
277 | p = q + 1; |
278 | for (q = p; isdigit((unsigned char) *q) || (*q == '-') || (*q == '+'); q++) |
279 | /* skip */ ; |
280 | if (q == p) /* no digits? */ |
281 | ereport(ERROR, |
282 | (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION), |
283 | errmsg("malformed array literal: \"%s\"" , string), |
284 | errdetail("Missing array dimension value." ))); |
285 | } |
286 | else |
287 | { |
288 | /* [n] format */ |
289 | lBound[ndim] = 1; |
290 | } |
291 | if (*q != ']') |
292 | ereport(ERROR, |
293 | (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION), |
294 | errmsg("malformed array literal: \"%s\"" , string), |
295 | errdetail("Missing \"%s\" after array dimensions." , |
296 | "]" ))); |
297 | |
298 | *q = '\0'; |
299 | ub = atoi(p); |
300 | p = q + 1; |
301 | if (ub < lBound[ndim]) |
302 | ereport(ERROR, |
303 | (errcode(ERRCODE_ARRAY_SUBSCRIPT_ERROR), |
304 | errmsg("upper bound cannot be less than lower bound" ))); |
305 | |
306 | dim[ndim] = ub - lBound[ndim] + 1; |
307 | ndim++; |
308 | } |
309 | |
310 | if (ndim == 0) |
311 | { |
312 | /* No array dimensions, so intuit dimensions from brace structure */ |
313 | if (*p != '{') |
314 | ereport(ERROR, |
315 | (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION), |
316 | errmsg("malformed array literal: \"%s\"" , string), |
317 | errdetail("Array value must start with \"{\" or dimension information." ))); |
318 | ndim = ArrayCount(p, dim, typdelim); |
319 | for (i = 0; i < ndim; i++) |
320 | lBound[i] = 1; |
321 | } |
322 | else |
323 | { |
324 | int ndim_braces, |
325 | dim_braces[MAXDIM]; |
326 | |
327 | /* If array dimensions are given, expect '=' operator */ |
328 | if (strncmp(p, ASSGN, strlen(ASSGN)) != 0) |
329 | ereport(ERROR, |
330 | (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION), |
331 | errmsg("malformed array literal: \"%s\"" , string), |
332 | errdetail("Missing \"%s\" after array dimensions." , |
333 | ASSGN))); |
334 | p += strlen(ASSGN); |
335 | while (array_isspace(*p)) |
336 | p++; |
337 | |
338 | /* |
339 | * intuit dimensions from brace structure -- it better match what we |
340 | * were given |
341 | */ |
342 | if (*p != '{') |
343 | ereport(ERROR, |
344 | (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION), |
345 | errmsg("malformed array literal: \"%s\"" , string), |
346 | errdetail("Array contents must start with \"{\"." ))); |
347 | ndim_braces = ArrayCount(p, dim_braces, typdelim); |
348 | if (ndim_braces != ndim) |
349 | ereport(ERROR, |
350 | (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION), |
351 | errmsg("malformed array literal: \"%s\"" , string), |
352 | errdetail("Specified array dimensions do not match array contents." ))); |
353 | for (i = 0; i < ndim; ++i) |
354 | { |
355 | if (dim[i] != dim_braces[i]) |
356 | ereport(ERROR, |
357 | (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION), |
358 | errmsg("malformed array literal: \"%s\"" , string), |
359 | errdetail("Specified array dimensions do not match array contents." ))); |
360 | } |
361 | } |
362 | |
363 | #ifdef ARRAYDEBUG |
364 | printf("array_in- ndim %d (" , ndim); |
365 | for (i = 0; i < ndim; i++) |
366 | { |
367 | printf(" %d" , dim[i]); |
368 | }; |
369 | printf(") for %s\n" , string); |
370 | #endif |
371 | |
372 | /* This checks for overflow of the array dimensions */ |
373 | nitems = ArrayGetNItems(ndim, dim); |
374 | /* Empty array? */ |
375 | if (nitems == 0) |
376 | PG_RETURN_ARRAYTYPE_P(construct_empty_array(element_type)); |
377 | |
378 | dataPtr = (Datum *) palloc(nitems * sizeof(Datum)); |
379 | nullsPtr = (bool *) palloc(nitems * sizeof(bool)); |
380 | ReadArrayStr(p, string, |
381 | nitems, ndim, dim, |
382 | &my_extra->proc, typioparam, typmod, |
383 | typdelim, |
384 | typlen, typbyval, typalign, |
385 | dataPtr, nullsPtr, |
386 | &hasnulls, &nbytes); |
387 | if (hasnulls) |
388 | { |
389 | dataoffset = ARR_OVERHEAD_WITHNULLS(ndim, nitems); |
390 | nbytes += dataoffset; |
391 | } |
392 | else |
393 | { |
394 | dataoffset = 0; /* marker for no null bitmap */ |
395 | nbytes += ARR_OVERHEAD_NONULLS(ndim); |
396 | } |
397 | retval = (ArrayType *) palloc0(nbytes); |
398 | SET_VARSIZE(retval, nbytes); |
399 | retval->ndim = ndim; |
400 | retval->dataoffset = dataoffset; |
401 | |
402 | /* |
403 | * This comes from the array's pg_type.typelem (which points to the base |
404 | * data type's pg_type.oid) and stores system oids in user tables. This |
405 | * oid must be preserved by binary upgrades. |
406 | */ |
407 | retval->elemtype = element_type; |
408 | memcpy(ARR_DIMS(retval), dim, ndim * sizeof(int)); |
409 | memcpy(ARR_LBOUND(retval), lBound, ndim * sizeof(int)); |
410 | |
411 | CopyArrayEls(retval, |
412 | dataPtr, nullsPtr, nitems, |
413 | typlen, typbyval, typalign, |
414 | true); |
415 | |
416 | pfree(dataPtr); |
417 | pfree(nullsPtr); |
418 | pfree(string_save); |
419 | |
420 | PG_RETURN_ARRAYTYPE_P(retval); |
421 | } |
422 | |
423 | /* |
424 | * array_isspace() --- a non-locale-dependent isspace() |
425 | * |
426 | * We used to use isspace() for parsing array values, but that has |
427 | * undesirable results: an array value might be silently interpreted |
428 | * differently depending on the locale setting. Now we just hard-wire |
429 | * the traditional ASCII definition of isspace(). |
430 | */ |
431 | static bool |
432 | array_isspace(char ch) |
433 | { |
434 | if (ch == ' ' || |
435 | ch == '\t' || |
436 | ch == '\n' || |
437 | ch == '\r' || |
438 | ch == '\v' || |
439 | ch == '\f') |
440 | return true; |
441 | return false; |
442 | } |
443 | |
444 | /* |
445 | * ArrayCount |
446 | * Determines the dimensions for an array string. |
447 | * |
448 | * Returns number of dimensions as function result. The axis lengths are |
449 | * returned in dim[], which must be of size MAXDIM. |
450 | */ |
451 | static int |
452 | ArrayCount(const char *str, int *dim, char typdelim) |
453 | { |
454 | int nest_level = 0, |
455 | i; |
456 | int ndim = 1, |
457 | temp[MAXDIM], |
458 | nelems[MAXDIM], |
459 | nelems_last[MAXDIM]; |
460 | bool in_quotes = false; |
461 | bool eoArray = false; |
462 | bool empty_array = true; |
463 | const char *ptr; |
464 | ArrayParseState parse_state = ARRAY_NO_LEVEL; |
465 | |
466 | for (i = 0; i < MAXDIM; ++i) |
467 | { |
468 | temp[i] = dim[i] = nelems_last[i] = 0; |
469 | nelems[i] = 1; |
470 | } |
471 | |
472 | ptr = str; |
473 | while (!eoArray) |
474 | { |
475 | bool itemdone = false; |
476 | |
477 | while (!itemdone) |
478 | { |
479 | if (parse_state == ARRAY_ELEM_STARTED || |
480 | parse_state == ARRAY_QUOTED_ELEM_STARTED) |
481 | empty_array = false; |
482 | |
483 | switch (*ptr) |
484 | { |
485 | case '\0': |
486 | /* Signal a premature end of the string */ |
487 | ereport(ERROR, |
488 | (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION), |
489 | errmsg("malformed array literal: \"%s\"" , str), |
490 | errdetail("Unexpected end of input." ))); |
491 | break; |
492 | case '\\': |
493 | |
494 | /* |
495 | * An escape must be after a level start, after an element |
496 | * start, or after an element delimiter. In any case we |
497 | * now must be past an element start. |
498 | */ |
499 | if (parse_state != ARRAY_LEVEL_STARTED && |
500 | parse_state != ARRAY_ELEM_STARTED && |
501 | parse_state != ARRAY_QUOTED_ELEM_STARTED && |
502 | parse_state != ARRAY_ELEM_DELIMITED) |
503 | ereport(ERROR, |
504 | (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION), |
505 | errmsg("malformed array literal: \"%s\"" , str), |
506 | errdetail("Unexpected \"%c\" character." , |
507 | '\\'))); |
508 | if (parse_state != ARRAY_QUOTED_ELEM_STARTED) |
509 | parse_state = ARRAY_ELEM_STARTED; |
510 | /* skip the escaped character */ |
511 | if (*(ptr + 1)) |
512 | ptr++; |
513 | else |
514 | ereport(ERROR, |
515 | (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION), |
516 | errmsg("malformed array literal: \"%s\"" , str), |
517 | errdetail("Unexpected end of input." ))); |
518 | break; |
519 | case '"': |
520 | |
521 | /* |
522 | * A quote must be after a level start, after a quoted |
523 | * element start, or after an element delimiter. In any |
524 | * case we now must be past an element start. |
525 | */ |
526 | if (parse_state != ARRAY_LEVEL_STARTED && |
527 | parse_state != ARRAY_QUOTED_ELEM_STARTED && |
528 | parse_state != ARRAY_ELEM_DELIMITED) |
529 | ereport(ERROR, |
530 | (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION), |
531 | errmsg("malformed array literal: \"%s\"" , str), |
532 | errdetail("Unexpected array element." ))); |
533 | in_quotes = !in_quotes; |
534 | if (in_quotes) |
535 | parse_state = ARRAY_QUOTED_ELEM_STARTED; |
536 | else |
537 | parse_state = ARRAY_QUOTED_ELEM_COMPLETED; |
538 | break; |
539 | case '{': |
540 | if (!in_quotes) |
541 | { |
542 | /* |
543 | * A left brace can occur if no nesting has occurred |
544 | * yet, after a level start, or after a level |
545 | * delimiter. |
546 | */ |
547 | if (parse_state != ARRAY_NO_LEVEL && |
548 | parse_state != ARRAY_LEVEL_STARTED && |
549 | parse_state != ARRAY_LEVEL_DELIMITED) |
550 | ereport(ERROR, |
551 | (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION), |
552 | errmsg("malformed array literal: \"%s\"" , str), |
553 | errdetail("Unexpected \"%c\" character." , |
554 | '{'))); |
555 | parse_state = ARRAY_LEVEL_STARTED; |
556 | if (nest_level >= MAXDIM) |
557 | ereport(ERROR, |
558 | (errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED), |
559 | errmsg("number of array dimensions (%d) exceeds the maximum allowed (%d)" , |
560 | nest_level + 1, MAXDIM))); |
561 | temp[nest_level] = 0; |
562 | nest_level++; |
563 | if (ndim < nest_level) |
564 | ndim = nest_level; |
565 | } |
566 | break; |
567 | case '}': |
568 | if (!in_quotes) |
569 | { |
570 | /* |
571 | * A right brace can occur after an element start, an |
572 | * element completion, a quoted element completion, or |
573 | * a level completion. |
574 | */ |
575 | if (parse_state != ARRAY_ELEM_STARTED && |
576 | parse_state != ARRAY_ELEM_COMPLETED && |
577 | parse_state != ARRAY_QUOTED_ELEM_COMPLETED && |
578 | parse_state != ARRAY_LEVEL_COMPLETED && |
579 | !(nest_level == 1 && parse_state == ARRAY_LEVEL_STARTED)) |
580 | ereport(ERROR, |
581 | (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION), |
582 | errmsg("malformed array literal: \"%s\"" , str), |
583 | errdetail("Unexpected \"%c\" character." , |
584 | '}'))); |
585 | parse_state = ARRAY_LEVEL_COMPLETED; |
586 | if (nest_level == 0) |
587 | ereport(ERROR, |
588 | (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION), |
589 | errmsg("malformed array literal: \"%s\"" , str), |
590 | errdetail("Unmatched \"%c\" character." , '}'))); |
591 | nest_level--; |
592 | |
593 | if (nelems_last[nest_level] != 0 && |
594 | nelems[nest_level] != nelems_last[nest_level]) |
595 | ereport(ERROR, |
596 | (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION), |
597 | errmsg("malformed array literal: \"%s\"" , str), |
598 | errdetail("Multidimensional arrays must have " |
599 | "sub-arrays with matching " |
600 | "dimensions." ))); |
601 | nelems_last[nest_level] = nelems[nest_level]; |
602 | nelems[nest_level] = 1; |
603 | if (nest_level == 0) |
604 | eoArray = itemdone = true; |
605 | else |
606 | { |
607 | /* |
608 | * We don't set itemdone here; see comments in |
609 | * ReadArrayStr |
610 | */ |
611 | temp[nest_level - 1]++; |
612 | } |
613 | } |
614 | break; |
615 | default: |
616 | if (!in_quotes) |
617 | { |
618 | if (*ptr == typdelim) |
619 | { |
620 | /* |
621 | * Delimiters can occur after an element start, an |
622 | * element completion, a quoted element |
623 | * completion, or a level completion. |
624 | */ |
625 | if (parse_state != ARRAY_ELEM_STARTED && |
626 | parse_state != ARRAY_ELEM_COMPLETED && |
627 | parse_state != ARRAY_QUOTED_ELEM_COMPLETED && |
628 | parse_state != ARRAY_LEVEL_COMPLETED) |
629 | ereport(ERROR, |
630 | (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION), |
631 | errmsg("malformed array literal: \"%s\"" , str), |
632 | errdetail("Unexpected \"%c\" character." , |
633 | typdelim))); |
634 | if (parse_state == ARRAY_LEVEL_COMPLETED) |
635 | parse_state = ARRAY_LEVEL_DELIMITED; |
636 | else |
637 | parse_state = ARRAY_ELEM_DELIMITED; |
638 | itemdone = true; |
639 | nelems[nest_level - 1]++; |
640 | } |
641 | else if (!array_isspace(*ptr)) |
642 | { |
643 | /* |
644 | * Other non-space characters must be after a |
645 | * level start, after an element start, or after |
646 | * an element delimiter. In any case we now must |
647 | * be past an element start. |
648 | */ |
649 | if (parse_state != ARRAY_LEVEL_STARTED && |
650 | parse_state != ARRAY_ELEM_STARTED && |
651 | parse_state != ARRAY_ELEM_DELIMITED) |
652 | ereport(ERROR, |
653 | (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION), |
654 | errmsg("malformed array literal: \"%s\"" , str), |
655 | errdetail("Unexpected array element." ))); |
656 | parse_state = ARRAY_ELEM_STARTED; |
657 | } |
658 | } |
659 | break; |
660 | } |
661 | if (!itemdone) |
662 | ptr++; |
663 | } |
664 | temp[ndim - 1]++; |
665 | ptr++; |
666 | } |
667 | |
668 | /* only whitespace is allowed after the closing brace */ |
669 | while (*ptr) |
670 | { |
671 | if (!array_isspace(*ptr++)) |
672 | ereport(ERROR, |
673 | (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION), |
674 | errmsg("malformed array literal: \"%s\"" , str), |
675 | errdetail("Junk after closing right brace." ))); |
676 | } |
677 | |
678 | /* special case for an empty array */ |
679 | if (empty_array) |
680 | return 0; |
681 | |
682 | for (i = 0; i < ndim; ++i) |
683 | dim[i] = temp[i]; |
684 | |
685 | return ndim; |
686 | } |
687 | |
688 | /* |
689 | * ReadArrayStr : |
690 | * parses the array string pointed to by "arrayStr" and converts the values |
691 | * to internal format. Unspecified elements are initialized to nulls. |
692 | * The array dimensions must already have been determined. |
693 | * |
694 | * Inputs: |
695 | * arrayStr: the string to parse. |
696 | * CAUTION: the contents of "arrayStr" will be modified! |
697 | * origStr: the unmodified input string, used only in error messages. |
698 | * nitems: total number of array elements, as already determined. |
699 | * ndim: number of array dimensions |
700 | * dim[]: array axis lengths |
701 | * inputproc: type-specific input procedure for element datatype. |
702 | * typioparam, typmod: auxiliary values to pass to inputproc. |
703 | * typdelim: the value delimiter (type-specific). |
704 | * typlen, typbyval, typalign: storage parameters of element datatype. |
705 | * |
706 | * Outputs: |
707 | * values[]: filled with converted data values. |
708 | * nulls[]: filled with is-null markers. |
709 | * *hasnulls: set true iff there are any null elements. |
710 | * *nbytes: set to total size of data area needed (including alignment |
711 | * padding but not including array header overhead). |
712 | * |
713 | * Note that values[] and nulls[] are allocated by the caller, and must have |
714 | * nitems elements. |
715 | */ |
716 | static void |
717 | ReadArrayStr(char *arrayStr, |
718 | const char *origStr, |
719 | int nitems, |
720 | int ndim, |
721 | int *dim, |
722 | FmgrInfo *inputproc, |
723 | Oid typioparam, |
724 | int32 typmod, |
725 | char typdelim, |
726 | int typlen, |
727 | bool typbyval, |
728 | char typalign, |
729 | Datum *values, |
730 | bool *nulls, |
731 | bool *hasnulls, |
732 | int32 *nbytes) |
733 | { |
734 | int i, |
735 | nest_level = 0; |
736 | char *srcptr; |
737 | bool in_quotes = false; |
738 | bool eoArray = false; |
739 | bool hasnull; |
740 | int32 totbytes; |
741 | int indx[MAXDIM], |
742 | prod[MAXDIM]; |
743 | |
744 | mda_get_prod(ndim, dim, prod); |
745 | MemSet(indx, 0, sizeof(indx)); |
746 | |
747 | /* Initialize is-null markers to true */ |
748 | memset(nulls, true, nitems * sizeof(bool)); |
749 | |
750 | /* |
751 | * We have to remove " and \ characters to create a clean item value to |
752 | * pass to the datatype input routine. We overwrite each item value |
753 | * in-place within arrayStr to do this. srcptr is the current scan point, |
754 | * and dstptr is where we are copying to. |
755 | * |
756 | * We also want to suppress leading and trailing unquoted whitespace. We |
757 | * use the leadingspace flag to suppress leading space. Trailing space is |
758 | * tracked by using dstendptr to point to the last significant output |
759 | * character. |
760 | * |
761 | * The error checking in this routine is mostly pro-forma, since we expect |
762 | * that ArrayCount() already validated the string. So we don't bother |
763 | * with errdetail messages. |
764 | */ |
765 | srcptr = arrayStr; |
766 | while (!eoArray) |
767 | { |
768 | bool itemdone = false; |
769 | bool leadingspace = true; |
770 | bool hasquoting = false; |
771 | char *itemstart; |
772 | char *dstptr; |
773 | char *dstendptr; |
774 | |
775 | i = -1; |
776 | itemstart = dstptr = dstendptr = srcptr; |
777 | |
778 | while (!itemdone) |
779 | { |
780 | switch (*srcptr) |
781 | { |
782 | case '\0': |
783 | /* Signal a premature end of the string */ |
784 | ereport(ERROR, |
785 | (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION), |
786 | errmsg("malformed array literal: \"%s\"" , |
787 | origStr))); |
788 | break; |
789 | case '\\': |
790 | /* Skip backslash, copy next character as-is. */ |
791 | srcptr++; |
792 | if (*srcptr == '\0') |
793 | ereport(ERROR, |
794 | (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION), |
795 | errmsg("malformed array literal: \"%s\"" , |
796 | origStr))); |
797 | *dstptr++ = *srcptr++; |
798 | /* Treat the escaped character as non-whitespace */ |
799 | leadingspace = false; |
800 | dstendptr = dstptr; |
801 | hasquoting = true; /* can't be a NULL marker */ |
802 | break; |
803 | case '"': |
804 | in_quotes = !in_quotes; |
805 | if (in_quotes) |
806 | leadingspace = false; |
807 | else |
808 | { |
809 | /* |
810 | * Advance dstendptr when we exit in_quotes; this |
811 | * saves having to do it in all the other in_quotes |
812 | * cases. |
813 | */ |
814 | dstendptr = dstptr; |
815 | } |
816 | hasquoting = true; /* can't be a NULL marker */ |
817 | srcptr++; |
818 | break; |
819 | case '{': |
820 | if (!in_quotes) |
821 | { |
822 | if (nest_level >= ndim) |
823 | ereport(ERROR, |
824 | (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION), |
825 | errmsg("malformed array literal: \"%s\"" , |
826 | origStr))); |
827 | nest_level++; |
828 | indx[nest_level - 1] = 0; |
829 | srcptr++; |
830 | } |
831 | else |
832 | *dstptr++ = *srcptr++; |
833 | break; |
834 | case '}': |
835 | if (!in_quotes) |
836 | { |
837 | if (nest_level == 0) |
838 | ereport(ERROR, |
839 | (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION), |
840 | errmsg("malformed array literal: \"%s\"" , |
841 | origStr))); |
842 | if (i == -1) |
843 | i = ArrayGetOffset0(ndim, indx, prod); |
844 | indx[nest_level - 1] = 0; |
845 | nest_level--; |
846 | if (nest_level == 0) |
847 | eoArray = itemdone = true; |
848 | else |
849 | indx[nest_level - 1]++; |
850 | srcptr++; |
851 | } |
852 | else |
853 | *dstptr++ = *srcptr++; |
854 | break; |
855 | default: |
856 | if (in_quotes) |
857 | *dstptr++ = *srcptr++; |
858 | else if (*srcptr == typdelim) |
859 | { |
860 | if (i == -1) |
861 | i = ArrayGetOffset0(ndim, indx, prod); |
862 | itemdone = true; |
863 | indx[ndim - 1]++; |
864 | srcptr++; |
865 | } |
866 | else if (array_isspace(*srcptr)) |
867 | { |
868 | /* |
869 | * If leading space, drop it immediately. Else, copy |
870 | * but don't advance dstendptr. |
871 | */ |
872 | if (leadingspace) |
873 | srcptr++; |
874 | else |
875 | *dstptr++ = *srcptr++; |
876 | } |
877 | else |
878 | { |
879 | *dstptr++ = *srcptr++; |
880 | leadingspace = false; |
881 | dstendptr = dstptr; |
882 | } |
883 | break; |
884 | } |
885 | } |
886 | |
887 | Assert(dstptr < srcptr); |
888 | *dstendptr = '\0'; |
889 | |
890 | if (i < 0 || i >= nitems) |
891 | ereport(ERROR, |
892 | (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION), |
893 | errmsg("malformed array literal: \"%s\"" , |
894 | origStr))); |
895 | |
896 | if (Array_nulls && !hasquoting && |
897 | pg_strcasecmp(itemstart, "NULL" ) == 0) |
898 | { |
899 | /* it's a NULL item */ |
900 | values[i] = InputFunctionCall(inputproc, NULL, |
901 | typioparam, typmod); |
902 | nulls[i] = true; |
903 | } |
904 | else |
905 | { |
906 | values[i] = InputFunctionCall(inputproc, itemstart, |
907 | typioparam, typmod); |
908 | nulls[i] = false; |
909 | } |
910 | } |
911 | |
912 | /* |
913 | * Check for nulls, compute total data space needed |
914 | */ |
915 | hasnull = false; |
916 | totbytes = 0; |
917 | for (i = 0; i < nitems; i++) |
918 | { |
919 | if (nulls[i]) |
920 | hasnull = true; |
921 | else |
922 | { |
923 | /* let's just make sure data is not toasted */ |
924 | if (typlen == -1) |
925 | values[i] = PointerGetDatum(PG_DETOAST_DATUM(values[i])); |
926 | totbytes = att_addlength_datum(totbytes, typlen, values[i]); |
927 | totbytes = att_align_nominal(totbytes, typalign); |
928 | /* check for overflow of total request */ |
929 | if (!AllocSizeIsValid(totbytes)) |
930 | ereport(ERROR, |
931 | (errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED), |
932 | errmsg("array size exceeds the maximum allowed (%d)" , |
933 | (int) MaxAllocSize))); |
934 | } |
935 | } |
936 | *hasnulls = hasnull; |
937 | *nbytes = totbytes; |
938 | } |
939 | |
940 | |
941 | /* |
942 | * Copy data into an array object from a temporary array of Datums. |
943 | * |
944 | * array: array object (with header fields already filled in) |
945 | * values: array of Datums to be copied |
946 | * nulls: array of is-null flags (can be NULL if no nulls) |
947 | * nitems: number of Datums to be copied |
948 | * typbyval, typlen, typalign: info about element datatype |
949 | * freedata: if true and element type is pass-by-ref, pfree data values |
950 | * referenced by Datums after copying them. |
951 | * |
952 | * If the input data is of varlena type, the caller must have ensured that |
953 | * the values are not toasted. (Doing it here doesn't work since the |
954 | * caller has already allocated space for the array...) |
955 | */ |
956 | void |
957 | CopyArrayEls(ArrayType *array, |
958 | Datum *values, |
959 | bool *nulls, |
960 | int nitems, |
961 | int typlen, |
962 | bool typbyval, |
963 | char typalign, |
964 | bool freedata) |
965 | { |
966 | char *p = ARR_DATA_PTR(array); |
967 | bits8 *bitmap = ARR_NULLBITMAP(array); |
968 | int bitval = 0; |
969 | int bitmask = 1; |
970 | int i; |
971 | |
972 | if (typbyval) |
973 | freedata = false; |
974 | |
975 | for (i = 0; i < nitems; i++) |
976 | { |
977 | if (nulls && nulls[i]) |
978 | { |
979 | if (!bitmap) /* shouldn't happen */ |
980 | elog(ERROR, "null array element where not supported" ); |
981 | /* bitmap bit stays 0 */ |
982 | } |
983 | else |
984 | { |
985 | bitval |= bitmask; |
986 | p += ArrayCastAndSet(values[i], typlen, typbyval, typalign, p); |
987 | if (freedata) |
988 | pfree(DatumGetPointer(values[i])); |
989 | } |
990 | if (bitmap) |
991 | { |
992 | bitmask <<= 1; |
993 | if (bitmask == 0x100) |
994 | { |
995 | *bitmap++ = bitval; |
996 | bitval = 0; |
997 | bitmask = 1; |
998 | } |
999 | } |
1000 | } |
1001 | |
1002 | if (bitmap && bitmask != 1) |
1003 | *bitmap = bitval; |
1004 | } |
1005 | |
1006 | /* |
1007 | * array_out : |
1008 | * takes the internal representation of an array and returns a string |
1009 | * containing the array in its external format. |
1010 | */ |
1011 | Datum |
1012 | array_out(PG_FUNCTION_ARGS) |
1013 | { |
1014 | AnyArrayType *v = PG_GETARG_ANY_ARRAY_P(0); |
1015 | Oid element_type = AARR_ELEMTYPE(v); |
1016 | int typlen; |
1017 | bool typbyval; |
1018 | char typalign; |
1019 | char typdelim; |
1020 | char *p, |
1021 | *tmp, |
1022 | *retval, |
1023 | **values, |
1024 | dims_str[(MAXDIM * 33) + 2]; |
1025 | |
1026 | /* |
1027 | * 33 per dim since we assume 15 digits per number + ':' +'[]' |
1028 | * |
1029 | * +2 allows for assignment operator + trailing null |
1030 | */ |
1031 | bool *needquotes, |
1032 | needdims = false; |
1033 | size_t overall_length; |
1034 | int nitems, |
1035 | i, |
1036 | j, |
1037 | k, |
1038 | indx[MAXDIM]; |
1039 | int ndim, |
1040 | *dims, |
1041 | *lb; |
1042 | array_iter iter; |
1043 | ArrayMetaState *; |
1044 | |
1045 | /* |
1046 | * We arrange to look up info about element type, including its output |
1047 | * conversion proc, only once per series of calls, assuming the element |
1048 | * type doesn't change underneath us. |
1049 | */ |
1050 | my_extra = (ArrayMetaState *) fcinfo->flinfo->fn_extra; |
1051 | if (my_extra == NULL) |
1052 | { |
1053 | fcinfo->flinfo->fn_extra = MemoryContextAlloc(fcinfo->flinfo->fn_mcxt, |
1054 | sizeof(ArrayMetaState)); |
1055 | my_extra = (ArrayMetaState *) fcinfo->flinfo->fn_extra; |
1056 | my_extra->element_type = ~element_type; |
1057 | } |
1058 | |
1059 | if (my_extra->element_type != element_type) |
1060 | { |
1061 | /* |
1062 | * Get info about element type, including its output conversion proc |
1063 | */ |
1064 | get_type_io_data(element_type, IOFunc_output, |
1065 | &my_extra->typlen, &my_extra->typbyval, |
1066 | &my_extra->typalign, &my_extra->typdelim, |
1067 | &my_extra->typioparam, &my_extra->typiofunc); |
1068 | fmgr_info_cxt(my_extra->typiofunc, &my_extra->proc, |
1069 | fcinfo->flinfo->fn_mcxt); |
1070 | my_extra->element_type = element_type; |
1071 | } |
1072 | typlen = my_extra->typlen; |
1073 | typbyval = my_extra->typbyval; |
1074 | typalign = my_extra->typalign; |
1075 | typdelim = my_extra->typdelim; |
1076 | |
1077 | ndim = AARR_NDIM(v); |
1078 | dims = AARR_DIMS(v); |
1079 | lb = AARR_LBOUND(v); |
1080 | nitems = ArrayGetNItems(ndim, dims); |
1081 | |
1082 | if (nitems == 0) |
1083 | { |
1084 | retval = pstrdup("{}" ); |
1085 | PG_RETURN_CSTRING(retval); |
1086 | } |
1087 | |
1088 | /* |
1089 | * we will need to add explicit dimensions if any dimension has a lower |
1090 | * bound other than one |
1091 | */ |
1092 | for (i = 0; i < ndim; i++) |
1093 | { |
1094 | if (lb[i] != 1) |
1095 | { |
1096 | needdims = true; |
1097 | break; |
1098 | } |
1099 | } |
1100 | |
1101 | /* |
1102 | * Convert all values to string form, count total space needed (including |
1103 | * any overhead such as escaping backslashes), and detect whether each |
1104 | * item needs double quotes. |
1105 | */ |
1106 | values = (char **) palloc(nitems * sizeof(char *)); |
1107 | needquotes = (bool *) palloc(nitems * sizeof(bool)); |
1108 | overall_length = 0; |
1109 | |
1110 | array_iter_setup(&iter, v); |
1111 | |
1112 | for (i = 0; i < nitems; i++) |
1113 | { |
1114 | Datum itemvalue; |
1115 | bool isnull; |
1116 | bool needquote; |
1117 | |
1118 | /* Get source element, checking for NULL */ |
1119 | itemvalue = array_iter_next(&iter, &isnull, i, |
1120 | typlen, typbyval, typalign); |
1121 | |
1122 | if (isnull) |
1123 | { |
1124 | values[i] = pstrdup("NULL" ); |
1125 | overall_length += 4; |
1126 | needquote = false; |
1127 | } |
1128 | else |
1129 | { |
1130 | values[i] = OutputFunctionCall(&my_extra->proc, itemvalue); |
1131 | |
1132 | /* count data plus backslashes; detect chars needing quotes */ |
1133 | if (values[i][0] == '\0') |
1134 | needquote = true; /* force quotes for empty string */ |
1135 | else if (pg_strcasecmp(values[i], "NULL" ) == 0) |
1136 | needquote = true; /* force quotes for literal NULL */ |
1137 | else |
1138 | needquote = false; |
1139 | |
1140 | for (tmp = values[i]; *tmp != '\0'; tmp++) |
1141 | { |
1142 | char ch = *tmp; |
1143 | |
1144 | overall_length += 1; |
1145 | if (ch == '"' || ch == '\\') |
1146 | { |
1147 | needquote = true; |
1148 | overall_length += 1; |
1149 | } |
1150 | else if (ch == '{' || ch == '}' || ch == typdelim || |
1151 | array_isspace(ch)) |
1152 | needquote = true; |
1153 | } |
1154 | } |
1155 | |
1156 | needquotes[i] = needquote; |
1157 | |
1158 | /* Count the pair of double quotes, if needed */ |
1159 | if (needquote) |
1160 | overall_length += 2; |
1161 | /* and the comma (or other typdelim delimiter) */ |
1162 | overall_length += 1; |
1163 | } |
1164 | |
1165 | /* |
1166 | * The very last array element doesn't have a typdelim delimiter after it, |
1167 | * but that's OK; that space is needed for the trailing '\0'. |
1168 | * |
1169 | * Now count total number of curly brace pairs in output string. |
1170 | */ |
1171 | for (i = j = 0, k = 1; i < ndim; i++) |
1172 | { |
1173 | j += k, k *= dims[i]; |
1174 | } |
1175 | overall_length += 2 * j; |
1176 | |
1177 | /* Format explicit dimensions if required */ |
1178 | dims_str[0] = '\0'; |
1179 | if (needdims) |
1180 | { |
1181 | char *ptr = dims_str; |
1182 | |
1183 | for (i = 0; i < ndim; i++) |
1184 | { |
1185 | sprintf(ptr, "[%d:%d]" , lb[i], lb[i] + dims[i] - 1); |
1186 | ptr += strlen(ptr); |
1187 | } |
1188 | *ptr++ = *ASSGN; |
1189 | *ptr = '\0'; |
1190 | overall_length += ptr - dims_str; |
1191 | } |
1192 | |
1193 | /* Now construct the output string */ |
1194 | retval = (char *) palloc(overall_length); |
1195 | p = retval; |
1196 | |
1197 | #define APPENDSTR(str) (strcpy(p, (str)), p += strlen(p)) |
1198 | #define APPENDCHAR(ch) (*p++ = (ch), *p = '\0') |
1199 | |
1200 | if (needdims) |
1201 | APPENDSTR(dims_str); |
1202 | APPENDCHAR('{'); |
1203 | for (i = 0; i < ndim; i++) |
1204 | indx[i] = 0; |
1205 | j = 0; |
1206 | k = 0; |
1207 | do |
1208 | { |
1209 | for (i = j; i < ndim - 1; i++) |
1210 | APPENDCHAR('{'); |
1211 | |
1212 | if (needquotes[k]) |
1213 | { |
1214 | APPENDCHAR('"'); |
1215 | for (tmp = values[k]; *tmp; tmp++) |
1216 | { |
1217 | char ch = *tmp; |
1218 | |
1219 | if (ch == '"' || ch == '\\') |
1220 | *p++ = '\\'; |
1221 | *p++ = ch; |
1222 | } |
1223 | *p = '\0'; |
1224 | APPENDCHAR('"'); |
1225 | } |
1226 | else |
1227 | APPENDSTR(values[k]); |
1228 | pfree(values[k++]); |
1229 | |
1230 | for (i = ndim - 1; i >= 0; i--) |
1231 | { |
1232 | if (++(indx[i]) < dims[i]) |
1233 | { |
1234 | APPENDCHAR(typdelim); |
1235 | break; |
1236 | } |
1237 | else |
1238 | { |
1239 | indx[i] = 0; |
1240 | APPENDCHAR('}'); |
1241 | } |
1242 | } |
1243 | j = i; |
1244 | } while (j != -1); |
1245 | |
1246 | #undef APPENDSTR |
1247 | #undef APPENDCHAR |
1248 | |
1249 | /* Assert that we calculated the string length accurately */ |
1250 | Assert(overall_length == (p - retval + 1)); |
1251 | |
1252 | pfree(values); |
1253 | pfree(needquotes); |
1254 | |
1255 | PG_RETURN_CSTRING(retval); |
1256 | } |
1257 | |
1258 | /* |
1259 | * array_recv : |
1260 | * converts an array from the external binary format to |
1261 | * its internal format. |
1262 | * |
1263 | * return value : |
1264 | * the internal representation of the input array |
1265 | */ |
1266 | Datum |
1267 | array_recv(PG_FUNCTION_ARGS) |
1268 | { |
1269 | StringInfo buf = (StringInfo) PG_GETARG_POINTER(0); |
1270 | Oid spec_element_type = PG_GETARG_OID(1); /* type of an array |
1271 | * element */ |
1272 | int32 typmod = PG_GETARG_INT32(2); /* typmod for array elements */ |
1273 | Oid element_type; |
1274 | int typlen; |
1275 | bool typbyval; |
1276 | char typalign; |
1277 | Oid typioparam; |
1278 | int i, |
1279 | nitems; |
1280 | Datum *dataPtr; |
1281 | bool *nullsPtr; |
1282 | bool hasnulls; |
1283 | int32 nbytes; |
1284 | int32 dataoffset; |
1285 | ArrayType *retval; |
1286 | int ndim, |
1287 | flags, |
1288 | dim[MAXDIM], |
1289 | lBound[MAXDIM]; |
1290 | ArrayMetaState *; |
1291 | |
1292 | /* Get the array header information */ |
1293 | ndim = pq_getmsgint(buf, 4); |
1294 | if (ndim < 0) /* we do allow zero-dimension arrays */ |
1295 | ereport(ERROR, |
1296 | (errcode(ERRCODE_INVALID_BINARY_REPRESENTATION), |
1297 | errmsg("invalid number of dimensions: %d" , ndim))); |
1298 | if (ndim > MAXDIM) |
1299 | ereport(ERROR, |
1300 | (errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED), |
1301 | errmsg("number of array dimensions (%d) exceeds the maximum allowed (%d)" , |
1302 | ndim, MAXDIM))); |
1303 | |
1304 | flags = pq_getmsgint(buf, 4); |
1305 | if (flags != 0 && flags != 1) |
1306 | ereport(ERROR, |
1307 | (errcode(ERRCODE_INVALID_BINARY_REPRESENTATION), |
1308 | errmsg("invalid array flags" ))); |
1309 | |
1310 | element_type = pq_getmsgint(buf, sizeof(Oid)); |
1311 | if (element_type != spec_element_type) |
1312 | { |
1313 | /* XXX Can we allow taking the input element type in any cases? */ |
1314 | ereport(ERROR, |
1315 | (errcode(ERRCODE_DATATYPE_MISMATCH), |
1316 | errmsg("wrong element type" ))); |
1317 | } |
1318 | |
1319 | for (i = 0; i < ndim; i++) |
1320 | { |
1321 | dim[i] = pq_getmsgint(buf, 4); |
1322 | lBound[i] = pq_getmsgint(buf, 4); |
1323 | |
1324 | /* |
1325 | * Check overflow of upper bound. (ArrayGetNItems() below checks that |
1326 | * dim[i] >= 0) |
1327 | */ |
1328 | if (dim[i] != 0) |
1329 | { |
1330 | int ub = lBound[i] + dim[i] - 1; |
1331 | |
1332 | if (lBound[i] > ub) |
1333 | ereport(ERROR, |
1334 | (errcode(ERRCODE_NUMERIC_VALUE_OUT_OF_RANGE), |
1335 | errmsg("integer out of range" ))); |
1336 | } |
1337 | } |
1338 | |
1339 | /* This checks for overflow of array dimensions */ |
1340 | nitems = ArrayGetNItems(ndim, dim); |
1341 | |
1342 | /* |
1343 | * We arrange to look up info about element type, including its receive |
1344 | * conversion proc, only once per series of calls, assuming the element |
1345 | * type doesn't change underneath us. |
1346 | */ |
1347 | my_extra = (ArrayMetaState *) fcinfo->flinfo->fn_extra; |
1348 | if (my_extra == NULL) |
1349 | { |
1350 | fcinfo->flinfo->fn_extra = MemoryContextAlloc(fcinfo->flinfo->fn_mcxt, |
1351 | sizeof(ArrayMetaState)); |
1352 | my_extra = (ArrayMetaState *) fcinfo->flinfo->fn_extra; |
1353 | my_extra->element_type = ~element_type; |
1354 | } |
1355 | |
1356 | if (my_extra->element_type != element_type) |
1357 | { |
1358 | /* Get info about element type, including its receive proc */ |
1359 | get_type_io_data(element_type, IOFunc_receive, |
1360 | &my_extra->typlen, &my_extra->typbyval, |
1361 | &my_extra->typalign, &my_extra->typdelim, |
1362 | &my_extra->typioparam, &my_extra->typiofunc); |
1363 | if (!OidIsValid(my_extra->typiofunc)) |
1364 | ereport(ERROR, |
1365 | (errcode(ERRCODE_UNDEFINED_FUNCTION), |
1366 | errmsg("no binary input function available for type %s" , |
1367 | format_type_be(element_type)))); |
1368 | fmgr_info_cxt(my_extra->typiofunc, &my_extra->proc, |
1369 | fcinfo->flinfo->fn_mcxt); |
1370 | my_extra->element_type = element_type; |
1371 | } |
1372 | |
1373 | if (nitems == 0) |
1374 | { |
1375 | /* Return empty array ... but not till we've validated element_type */ |
1376 | PG_RETURN_ARRAYTYPE_P(construct_empty_array(element_type)); |
1377 | } |
1378 | |
1379 | typlen = my_extra->typlen; |
1380 | typbyval = my_extra->typbyval; |
1381 | typalign = my_extra->typalign; |
1382 | typioparam = my_extra->typioparam; |
1383 | |
1384 | dataPtr = (Datum *) palloc(nitems * sizeof(Datum)); |
1385 | nullsPtr = (bool *) palloc(nitems * sizeof(bool)); |
1386 | ReadArrayBinary(buf, nitems, |
1387 | &my_extra->proc, typioparam, typmod, |
1388 | typlen, typbyval, typalign, |
1389 | dataPtr, nullsPtr, |
1390 | &hasnulls, &nbytes); |
1391 | if (hasnulls) |
1392 | { |
1393 | dataoffset = ARR_OVERHEAD_WITHNULLS(ndim, nitems); |
1394 | nbytes += dataoffset; |
1395 | } |
1396 | else |
1397 | { |
1398 | dataoffset = 0; /* marker for no null bitmap */ |
1399 | nbytes += ARR_OVERHEAD_NONULLS(ndim); |
1400 | } |
1401 | retval = (ArrayType *) palloc0(nbytes); |
1402 | SET_VARSIZE(retval, nbytes); |
1403 | retval->ndim = ndim; |
1404 | retval->dataoffset = dataoffset; |
1405 | retval->elemtype = element_type; |
1406 | memcpy(ARR_DIMS(retval), dim, ndim * sizeof(int)); |
1407 | memcpy(ARR_LBOUND(retval), lBound, ndim * sizeof(int)); |
1408 | |
1409 | CopyArrayEls(retval, |
1410 | dataPtr, nullsPtr, nitems, |
1411 | typlen, typbyval, typalign, |
1412 | true); |
1413 | |
1414 | pfree(dataPtr); |
1415 | pfree(nullsPtr); |
1416 | |
1417 | PG_RETURN_ARRAYTYPE_P(retval); |
1418 | } |
1419 | |
1420 | /* |
1421 | * ReadArrayBinary: |
1422 | * collect the data elements of an array being read in binary style. |
1423 | * |
1424 | * Inputs: |
1425 | * buf: the data buffer to read from. |
1426 | * nitems: total number of array elements (already read). |
1427 | * receiveproc: type-specific receive procedure for element datatype. |
1428 | * typioparam, typmod: auxiliary values to pass to receiveproc. |
1429 | * typlen, typbyval, typalign: storage parameters of element datatype. |
1430 | * |
1431 | * Outputs: |
1432 | * values[]: filled with converted data values. |
1433 | * nulls[]: filled with is-null markers. |
1434 | * *hasnulls: set true iff there are any null elements. |
1435 | * *nbytes: set to total size of data area needed (including alignment |
1436 | * padding but not including array header overhead). |
1437 | * |
1438 | * Note that values[] and nulls[] are allocated by the caller, and must have |
1439 | * nitems elements. |
1440 | */ |
1441 | static void |
1442 | ReadArrayBinary(StringInfo buf, |
1443 | int nitems, |
1444 | FmgrInfo *receiveproc, |
1445 | Oid typioparam, |
1446 | int32 typmod, |
1447 | int typlen, |
1448 | bool typbyval, |
1449 | char typalign, |
1450 | Datum *values, |
1451 | bool *nulls, |
1452 | bool *hasnulls, |
1453 | int32 *nbytes) |
1454 | { |
1455 | int i; |
1456 | bool hasnull; |
1457 | int32 totbytes; |
1458 | |
1459 | for (i = 0; i < nitems; i++) |
1460 | { |
1461 | int itemlen; |
1462 | StringInfoData elem_buf; |
1463 | char csave; |
1464 | |
1465 | /* Get and check the item length */ |
1466 | itemlen = pq_getmsgint(buf, 4); |
1467 | if (itemlen < -1 || itemlen > (buf->len - buf->cursor)) |
1468 | ereport(ERROR, |
1469 | (errcode(ERRCODE_INVALID_BINARY_REPRESENTATION), |
1470 | errmsg("insufficient data left in message" ))); |
1471 | |
1472 | if (itemlen == -1) |
1473 | { |
1474 | /* -1 length means NULL */ |
1475 | values[i] = ReceiveFunctionCall(receiveproc, NULL, |
1476 | typioparam, typmod); |
1477 | nulls[i] = true; |
1478 | continue; |
1479 | } |
1480 | |
1481 | /* |
1482 | * Rather than copying data around, we just set up a phony StringInfo |
1483 | * pointing to the correct portion of the input buffer. We assume we |
1484 | * can scribble on the input buffer so as to maintain the convention |
1485 | * that StringInfos have a trailing null. |
1486 | */ |
1487 | elem_buf.data = &buf->data[buf->cursor]; |
1488 | elem_buf.maxlen = itemlen + 1; |
1489 | elem_buf.len = itemlen; |
1490 | elem_buf.cursor = 0; |
1491 | |
1492 | buf->cursor += itemlen; |
1493 | |
1494 | csave = buf->data[buf->cursor]; |
1495 | buf->data[buf->cursor] = '\0'; |
1496 | |
1497 | /* Now call the element's receiveproc */ |
1498 | values[i] = ReceiveFunctionCall(receiveproc, &elem_buf, |
1499 | typioparam, typmod); |
1500 | nulls[i] = false; |
1501 | |
1502 | /* Trouble if it didn't eat the whole buffer */ |
1503 | if (elem_buf.cursor != itemlen) |
1504 | ereport(ERROR, |
1505 | (errcode(ERRCODE_INVALID_BINARY_REPRESENTATION), |
1506 | errmsg("improper binary format in array element %d" , |
1507 | i + 1))); |
1508 | |
1509 | buf->data[buf->cursor] = csave; |
1510 | } |
1511 | |
1512 | /* |
1513 | * Check for nulls, compute total data space needed |
1514 | */ |
1515 | hasnull = false; |
1516 | totbytes = 0; |
1517 | for (i = 0; i < nitems; i++) |
1518 | { |
1519 | if (nulls[i]) |
1520 | hasnull = true; |
1521 | else |
1522 | { |
1523 | /* let's just make sure data is not toasted */ |
1524 | if (typlen == -1) |
1525 | values[i] = PointerGetDatum(PG_DETOAST_DATUM(values[i])); |
1526 | totbytes = att_addlength_datum(totbytes, typlen, values[i]); |
1527 | totbytes = att_align_nominal(totbytes, typalign); |
1528 | /* check for overflow of total request */ |
1529 | if (!AllocSizeIsValid(totbytes)) |
1530 | ereport(ERROR, |
1531 | (errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED), |
1532 | errmsg("array size exceeds the maximum allowed (%d)" , |
1533 | (int) MaxAllocSize))); |
1534 | } |
1535 | } |
1536 | *hasnulls = hasnull; |
1537 | *nbytes = totbytes; |
1538 | } |
1539 | |
1540 | |
1541 | /* |
1542 | * array_send : |
1543 | * takes the internal representation of an array and returns a bytea |
1544 | * containing the array in its external binary format. |
1545 | */ |
1546 | Datum |
1547 | array_send(PG_FUNCTION_ARGS) |
1548 | { |
1549 | AnyArrayType *v = PG_GETARG_ANY_ARRAY_P(0); |
1550 | Oid element_type = AARR_ELEMTYPE(v); |
1551 | int typlen; |
1552 | bool typbyval; |
1553 | char typalign; |
1554 | int nitems, |
1555 | i; |
1556 | int ndim, |
1557 | *dim, |
1558 | *lb; |
1559 | StringInfoData buf; |
1560 | array_iter iter; |
1561 | ArrayMetaState *; |
1562 | |
1563 | /* |
1564 | * We arrange to look up info about element type, including its send |
1565 | * conversion proc, only once per series of calls, assuming the element |
1566 | * type doesn't change underneath us. |
1567 | */ |
1568 | my_extra = (ArrayMetaState *) fcinfo->flinfo->fn_extra; |
1569 | if (my_extra == NULL) |
1570 | { |
1571 | fcinfo->flinfo->fn_extra = MemoryContextAlloc(fcinfo->flinfo->fn_mcxt, |
1572 | sizeof(ArrayMetaState)); |
1573 | my_extra = (ArrayMetaState *) fcinfo->flinfo->fn_extra; |
1574 | my_extra->element_type = ~element_type; |
1575 | } |
1576 | |
1577 | if (my_extra->element_type != element_type) |
1578 | { |
1579 | /* Get info about element type, including its send proc */ |
1580 | get_type_io_data(element_type, IOFunc_send, |
1581 | &my_extra->typlen, &my_extra->typbyval, |
1582 | &my_extra->typalign, &my_extra->typdelim, |
1583 | &my_extra->typioparam, &my_extra->typiofunc); |
1584 | if (!OidIsValid(my_extra->typiofunc)) |
1585 | ereport(ERROR, |
1586 | (errcode(ERRCODE_UNDEFINED_FUNCTION), |
1587 | errmsg("no binary output function available for type %s" , |
1588 | format_type_be(element_type)))); |
1589 | fmgr_info_cxt(my_extra->typiofunc, &my_extra->proc, |
1590 | fcinfo->flinfo->fn_mcxt); |
1591 | my_extra->element_type = element_type; |
1592 | } |
1593 | typlen = my_extra->typlen; |
1594 | typbyval = my_extra->typbyval; |
1595 | typalign = my_extra->typalign; |
1596 | |
1597 | ndim = AARR_NDIM(v); |
1598 | dim = AARR_DIMS(v); |
1599 | lb = AARR_LBOUND(v); |
1600 | nitems = ArrayGetNItems(ndim, dim); |
1601 | |
1602 | pq_begintypsend(&buf); |
1603 | |
1604 | /* Send the array header information */ |
1605 | pq_sendint32(&buf, ndim); |
1606 | pq_sendint32(&buf, AARR_HASNULL(v) ? 1 : 0); |
1607 | pq_sendint32(&buf, element_type); |
1608 | for (i = 0; i < ndim; i++) |
1609 | { |
1610 | pq_sendint32(&buf, dim[i]); |
1611 | pq_sendint32(&buf, lb[i]); |
1612 | } |
1613 | |
1614 | /* Send the array elements using the element's own sendproc */ |
1615 | array_iter_setup(&iter, v); |
1616 | |
1617 | for (i = 0; i < nitems; i++) |
1618 | { |
1619 | Datum itemvalue; |
1620 | bool isnull; |
1621 | |
1622 | /* Get source element, checking for NULL */ |
1623 | itemvalue = array_iter_next(&iter, &isnull, i, |
1624 | typlen, typbyval, typalign); |
1625 | |
1626 | if (isnull) |
1627 | { |
1628 | /* -1 length means a NULL */ |
1629 | pq_sendint32(&buf, -1); |
1630 | } |
1631 | else |
1632 | { |
1633 | bytea *outputbytes; |
1634 | |
1635 | outputbytes = SendFunctionCall(&my_extra->proc, itemvalue); |
1636 | pq_sendint32(&buf, VARSIZE(outputbytes) - VARHDRSZ); |
1637 | pq_sendbytes(&buf, VARDATA(outputbytes), |
1638 | VARSIZE(outputbytes) - VARHDRSZ); |
1639 | pfree(outputbytes); |
1640 | } |
1641 | } |
1642 | |
1643 | PG_RETURN_BYTEA_P(pq_endtypsend(&buf)); |
1644 | } |
1645 | |
1646 | /* |
1647 | * array_ndims : |
1648 | * returns the number of dimensions of the array pointed to by "v" |
1649 | */ |
1650 | Datum |
1651 | array_ndims(PG_FUNCTION_ARGS) |
1652 | { |
1653 | AnyArrayType *v = PG_GETARG_ANY_ARRAY_P(0); |
1654 | |
1655 | /* Sanity check: does it look like an array at all? */ |
1656 | if (AARR_NDIM(v) <= 0 || AARR_NDIM(v) > MAXDIM) |
1657 | PG_RETURN_NULL(); |
1658 | |
1659 | PG_RETURN_INT32(AARR_NDIM(v)); |
1660 | } |
1661 | |
1662 | /* |
1663 | * array_dims : |
1664 | * returns the dimensions of the array pointed to by "v", as a "text" |
1665 | */ |
1666 | Datum |
1667 | array_dims(PG_FUNCTION_ARGS) |
1668 | { |
1669 | AnyArrayType *v = PG_GETARG_ANY_ARRAY_P(0); |
1670 | char *p; |
1671 | int i; |
1672 | int *dimv, |
1673 | *lb; |
1674 | |
1675 | /* |
1676 | * 33 since we assume 15 digits per number + ':' +'[]' |
1677 | * |
1678 | * +1 for trailing null |
1679 | */ |
1680 | char buf[MAXDIM * 33 + 1]; |
1681 | |
1682 | /* Sanity check: does it look like an array at all? */ |
1683 | if (AARR_NDIM(v) <= 0 || AARR_NDIM(v) > MAXDIM) |
1684 | PG_RETURN_NULL(); |
1685 | |
1686 | dimv = AARR_DIMS(v); |
1687 | lb = AARR_LBOUND(v); |
1688 | |
1689 | p = buf; |
1690 | for (i = 0; i < AARR_NDIM(v); i++) |
1691 | { |
1692 | sprintf(p, "[%d:%d]" , lb[i], dimv[i] + lb[i] - 1); |
1693 | p += strlen(p); |
1694 | } |
1695 | |
1696 | PG_RETURN_TEXT_P(cstring_to_text(buf)); |
1697 | } |
1698 | |
1699 | /* |
1700 | * array_lower : |
1701 | * returns the lower dimension, of the DIM requested, for |
1702 | * the array pointed to by "v", as an int4 |
1703 | */ |
1704 | Datum |
1705 | array_lower(PG_FUNCTION_ARGS) |
1706 | { |
1707 | AnyArrayType *v = PG_GETARG_ANY_ARRAY_P(0); |
1708 | int reqdim = PG_GETARG_INT32(1); |
1709 | int *lb; |
1710 | int result; |
1711 | |
1712 | /* Sanity check: does it look like an array at all? */ |
1713 | if (AARR_NDIM(v) <= 0 || AARR_NDIM(v) > MAXDIM) |
1714 | PG_RETURN_NULL(); |
1715 | |
1716 | /* Sanity check: was the requested dim valid */ |
1717 | if (reqdim <= 0 || reqdim > AARR_NDIM(v)) |
1718 | PG_RETURN_NULL(); |
1719 | |
1720 | lb = AARR_LBOUND(v); |
1721 | result = lb[reqdim - 1]; |
1722 | |
1723 | PG_RETURN_INT32(result); |
1724 | } |
1725 | |
1726 | /* |
1727 | * array_upper : |
1728 | * returns the upper dimension, of the DIM requested, for |
1729 | * the array pointed to by "v", as an int4 |
1730 | */ |
1731 | Datum |
1732 | array_upper(PG_FUNCTION_ARGS) |
1733 | { |
1734 | AnyArrayType *v = PG_GETARG_ANY_ARRAY_P(0); |
1735 | int reqdim = PG_GETARG_INT32(1); |
1736 | int *dimv, |
1737 | *lb; |
1738 | int result; |
1739 | |
1740 | /* Sanity check: does it look like an array at all? */ |
1741 | if (AARR_NDIM(v) <= 0 || AARR_NDIM(v) > MAXDIM) |
1742 | PG_RETURN_NULL(); |
1743 | |
1744 | /* Sanity check: was the requested dim valid */ |
1745 | if (reqdim <= 0 || reqdim > AARR_NDIM(v)) |
1746 | PG_RETURN_NULL(); |
1747 | |
1748 | lb = AARR_LBOUND(v); |
1749 | dimv = AARR_DIMS(v); |
1750 | |
1751 | result = dimv[reqdim - 1] + lb[reqdim - 1] - 1; |
1752 | |
1753 | PG_RETURN_INT32(result); |
1754 | } |
1755 | |
1756 | /* |
1757 | * array_length : |
1758 | * returns the length, of the dimension requested, for |
1759 | * the array pointed to by "v", as an int4 |
1760 | */ |
1761 | Datum |
1762 | array_length(PG_FUNCTION_ARGS) |
1763 | { |
1764 | AnyArrayType *v = PG_GETARG_ANY_ARRAY_P(0); |
1765 | int reqdim = PG_GETARG_INT32(1); |
1766 | int *dimv; |
1767 | int result; |
1768 | |
1769 | /* Sanity check: does it look like an array at all? */ |
1770 | if (AARR_NDIM(v) <= 0 || AARR_NDIM(v) > MAXDIM) |
1771 | PG_RETURN_NULL(); |
1772 | |
1773 | /* Sanity check: was the requested dim valid */ |
1774 | if (reqdim <= 0 || reqdim > AARR_NDIM(v)) |
1775 | PG_RETURN_NULL(); |
1776 | |
1777 | dimv = AARR_DIMS(v); |
1778 | |
1779 | result = dimv[reqdim - 1]; |
1780 | |
1781 | PG_RETURN_INT32(result); |
1782 | } |
1783 | |
1784 | /* |
1785 | * array_cardinality: |
1786 | * returns the total number of elements in an array |
1787 | */ |
1788 | Datum |
1789 | array_cardinality(PG_FUNCTION_ARGS) |
1790 | { |
1791 | AnyArrayType *v = PG_GETARG_ANY_ARRAY_P(0); |
1792 | |
1793 | PG_RETURN_INT32(ArrayGetNItems(AARR_NDIM(v), AARR_DIMS(v))); |
1794 | } |
1795 | |
1796 | |
1797 | /* |
1798 | * array_get_element : |
1799 | * This routine takes an array datum and a subscript array and returns |
1800 | * the referenced item as a Datum. Note that for a pass-by-reference |
1801 | * datatype, the returned Datum is a pointer into the array object. |
1802 | * |
1803 | * This handles both ordinary varlena arrays and fixed-length arrays. |
1804 | * |
1805 | * Inputs: |
1806 | * arraydatum: the array object (mustn't be NULL) |
1807 | * nSubscripts: number of subscripts supplied |
1808 | * indx[]: the subscript values |
1809 | * arraytyplen: pg_type.typlen for the array type |
1810 | * elmlen: pg_type.typlen for the array's element type |
1811 | * elmbyval: pg_type.typbyval for the array's element type |
1812 | * elmalign: pg_type.typalign for the array's element type |
1813 | * |
1814 | * Outputs: |
1815 | * The return value is the element Datum. |
1816 | * *isNull is set to indicate whether the element is NULL. |
1817 | */ |
1818 | Datum |
1819 | array_get_element(Datum arraydatum, |
1820 | int nSubscripts, |
1821 | int *indx, |
1822 | int arraytyplen, |
1823 | int elmlen, |
1824 | bool elmbyval, |
1825 | char elmalign, |
1826 | bool *isNull) |
1827 | { |
1828 | int i, |
1829 | ndim, |
1830 | *dim, |
1831 | *lb, |
1832 | offset, |
1833 | fixedDim[1], |
1834 | fixedLb[1]; |
1835 | char *arraydataptr, |
1836 | *retptr; |
1837 | bits8 *arraynullsptr; |
1838 | |
1839 | if (arraytyplen > 0) |
1840 | { |
1841 | /* |
1842 | * fixed-length arrays -- these are assumed to be 1-d, 0-based |
1843 | */ |
1844 | ndim = 1; |
1845 | fixedDim[0] = arraytyplen / elmlen; |
1846 | fixedLb[0] = 0; |
1847 | dim = fixedDim; |
1848 | lb = fixedLb; |
1849 | arraydataptr = (char *) DatumGetPointer(arraydatum); |
1850 | arraynullsptr = NULL; |
1851 | } |
1852 | else if (VARATT_IS_EXTERNAL_EXPANDED(DatumGetPointer(arraydatum))) |
1853 | { |
1854 | /* expanded array: let's do this in a separate function */ |
1855 | return array_get_element_expanded(arraydatum, |
1856 | nSubscripts, |
1857 | indx, |
1858 | arraytyplen, |
1859 | elmlen, |
1860 | elmbyval, |
1861 | elmalign, |
1862 | isNull); |
1863 | } |
1864 | else |
1865 | { |
1866 | /* detoast array if necessary, producing normal varlena input */ |
1867 | ArrayType *array = DatumGetArrayTypeP(arraydatum); |
1868 | |
1869 | ndim = ARR_NDIM(array); |
1870 | dim = ARR_DIMS(array); |
1871 | lb = ARR_LBOUND(array); |
1872 | arraydataptr = ARR_DATA_PTR(array); |
1873 | arraynullsptr = ARR_NULLBITMAP(array); |
1874 | } |
1875 | |
1876 | /* |
1877 | * Return NULL for invalid subscript |
1878 | */ |
1879 | if (ndim != nSubscripts || ndim <= 0 || ndim > MAXDIM) |
1880 | { |
1881 | *isNull = true; |
1882 | return (Datum) 0; |
1883 | } |
1884 | for (i = 0; i < ndim; i++) |
1885 | { |
1886 | if (indx[i] < lb[i] || indx[i] >= (dim[i] + lb[i])) |
1887 | { |
1888 | *isNull = true; |
1889 | return (Datum) 0; |
1890 | } |
1891 | } |
1892 | |
1893 | /* |
1894 | * Calculate the element number |
1895 | */ |
1896 | offset = ArrayGetOffset(nSubscripts, dim, lb, indx); |
1897 | |
1898 | /* |
1899 | * Check for NULL array element |
1900 | */ |
1901 | if (array_get_isnull(arraynullsptr, offset)) |
1902 | { |
1903 | *isNull = true; |
1904 | return (Datum) 0; |
1905 | } |
1906 | |
1907 | /* |
1908 | * OK, get the element |
1909 | */ |
1910 | *isNull = false; |
1911 | retptr = array_seek(arraydataptr, 0, arraynullsptr, offset, |
1912 | elmlen, elmbyval, elmalign); |
1913 | return ArrayCast(retptr, elmbyval, elmlen); |
1914 | } |
1915 | |
1916 | /* |
1917 | * Implementation of array_get_element() for an expanded array |
1918 | */ |
1919 | static Datum |
1920 | array_get_element_expanded(Datum arraydatum, |
1921 | int nSubscripts, int *indx, |
1922 | int arraytyplen, |
1923 | int elmlen, bool elmbyval, char elmalign, |
1924 | bool *isNull) |
1925 | { |
1926 | ExpandedArrayHeader *eah; |
1927 | int i, |
1928 | ndim, |
1929 | *dim, |
1930 | *lb, |
1931 | offset; |
1932 | Datum *dvalues; |
1933 | bool *dnulls; |
1934 | |
1935 | eah = (ExpandedArrayHeader *) DatumGetEOHP(arraydatum); |
1936 | Assert(eah->ea_magic == EA_MAGIC); |
1937 | |
1938 | /* sanity-check caller's info against object */ |
1939 | Assert(arraytyplen == -1); |
1940 | Assert(elmlen == eah->typlen); |
1941 | Assert(elmbyval == eah->typbyval); |
1942 | Assert(elmalign == eah->typalign); |
1943 | |
1944 | ndim = eah->ndims; |
1945 | dim = eah->dims; |
1946 | lb = eah->lbound; |
1947 | |
1948 | /* |
1949 | * Return NULL for invalid subscript |
1950 | */ |
1951 | if (ndim != nSubscripts || ndim <= 0 || ndim > MAXDIM) |
1952 | { |
1953 | *isNull = true; |
1954 | return (Datum) 0; |
1955 | } |
1956 | for (i = 0; i < ndim; i++) |
1957 | { |
1958 | if (indx[i] < lb[i] || indx[i] >= (dim[i] + lb[i])) |
1959 | { |
1960 | *isNull = true; |
1961 | return (Datum) 0; |
1962 | } |
1963 | } |
1964 | |
1965 | /* |
1966 | * Calculate the element number |
1967 | */ |
1968 | offset = ArrayGetOffset(nSubscripts, dim, lb, indx); |
1969 | |
1970 | /* |
1971 | * Deconstruct array if we didn't already. Note that we apply this even |
1972 | * if the input is nominally read-only: it should be safe enough. |
1973 | */ |
1974 | deconstruct_expanded_array(eah); |
1975 | |
1976 | dvalues = eah->dvalues; |
1977 | dnulls = eah->dnulls; |
1978 | |
1979 | /* |
1980 | * Check for NULL array element |
1981 | */ |
1982 | if (dnulls && dnulls[offset]) |
1983 | { |
1984 | *isNull = true; |
1985 | return (Datum) 0; |
1986 | } |
1987 | |
1988 | /* |
1989 | * OK, get the element. It's OK to return a pass-by-ref value as a |
1990 | * pointer into the expanded array, for the same reason that regular |
1991 | * array_get_element can return a pointer into flat arrays: the value is |
1992 | * assumed not to change for as long as the Datum reference can exist. |
1993 | */ |
1994 | *isNull = false; |
1995 | return dvalues[offset]; |
1996 | } |
1997 | |
1998 | /* |
1999 | * array_get_slice : |
2000 | * This routine takes an array and a range of indices (upperIndex and |
2001 | * lowerIndx), creates a new array structure for the referred elements |
2002 | * and returns a pointer to it. |
2003 | * |
2004 | * This handles both ordinary varlena arrays and fixed-length arrays. |
2005 | * |
2006 | * Inputs: |
2007 | * arraydatum: the array object (mustn't be NULL) |
2008 | * nSubscripts: number of subscripts supplied (must be same for upper/lower) |
2009 | * upperIndx[]: the upper subscript values |
2010 | * lowerIndx[]: the lower subscript values |
2011 | * upperProvided[]: true for provided upper subscript values |
2012 | * lowerProvided[]: true for provided lower subscript values |
2013 | * arraytyplen: pg_type.typlen for the array type |
2014 | * elmlen: pg_type.typlen for the array's element type |
2015 | * elmbyval: pg_type.typbyval for the array's element type |
2016 | * elmalign: pg_type.typalign for the array's element type |
2017 | * |
2018 | * Outputs: |
2019 | * The return value is the new array Datum (it's never NULL) |
2020 | * |
2021 | * Omitted upper and lower subscript values are replaced by the corresponding |
2022 | * array bound. |
2023 | * |
2024 | * NOTE: we assume it is OK to scribble on the provided subscript arrays |
2025 | * lowerIndx[] and upperIndx[]. These are generally just temporaries. |
2026 | */ |
2027 | Datum |
2028 | array_get_slice(Datum arraydatum, |
2029 | int nSubscripts, |
2030 | int *upperIndx, |
2031 | int *lowerIndx, |
2032 | bool *upperProvided, |
2033 | bool *lowerProvided, |
2034 | int arraytyplen, |
2035 | int elmlen, |
2036 | bool elmbyval, |
2037 | char elmalign) |
2038 | { |
2039 | ArrayType *array; |
2040 | ArrayType *newarray; |
2041 | int i, |
2042 | ndim, |
2043 | *dim, |
2044 | *lb, |
2045 | *newlb; |
2046 | int fixedDim[1], |
2047 | fixedLb[1]; |
2048 | Oid elemtype; |
2049 | char *arraydataptr; |
2050 | bits8 *arraynullsptr; |
2051 | int32 dataoffset; |
2052 | int bytes, |
2053 | span[MAXDIM]; |
2054 | |
2055 | if (arraytyplen > 0) |
2056 | { |
2057 | /* |
2058 | * fixed-length arrays -- currently, cannot slice these because parser |
2059 | * labels output as being of the fixed-length array type! Code below |
2060 | * shows how we could support it if the parser were changed to label |
2061 | * output as a suitable varlena array type. |
2062 | */ |
2063 | ereport(ERROR, |
2064 | (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), |
2065 | errmsg("slices of fixed-length arrays not implemented" ))); |
2066 | |
2067 | /* |
2068 | * fixed-length arrays -- these are assumed to be 1-d, 0-based |
2069 | * |
2070 | * XXX where would we get the correct ELEMTYPE from? |
2071 | */ |
2072 | ndim = 1; |
2073 | fixedDim[0] = arraytyplen / elmlen; |
2074 | fixedLb[0] = 0; |
2075 | dim = fixedDim; |
2076 | lb = fixedLb; |
2077 | elemtype = InvalidOid; /* XXX */ |
2078 | arraydataptr = (char *) DatumGetPointer(arraydatum); |
2079 | arraynullsptr = NULL; |
2080 | } |
2081 | else |
2082 | { |
2083 | /* detoast input array if necessary */ |
2084 | array = DatumGetArrayTypeP(arraydatum); |
2085 | |
2086 | ndim = ARR_NDIM(array); |
2087 | dim = ARR_DIMS(array); |
2088 | lb = ARR_LBOUND(array); |
2089 | elemtype = ARR_ELEMTYPE(array); |
2090 | arraydataptr = ARR_DATA_PTR(array); |
2091 | arraynullsptr = ARR_NULLBITMAP(array); |
2092 | } |
2093 | |
2094 | /* |
2095 | * Check provided subscripts. A slice exceeding the current array limits |
2096 | * is silently truncated to the array limits. If we end up with an empty |
2097 | * slice, return an empty array. |
2098 | */ |
2099 | if (ndim < nSubscripts || ndim <= 0 || ndim > MAXDIM) |
2100 | return PointerGetDatum(construct_empty_array(elemtype)); |
2101 | |
2102 | for (i = 0; i < nSubscripts; i++) |
2103 | { |
2104 | if (!lowerProvided[i] || lowerIndx[i] < lb[i]) |
2105 | lowerIndx[i] = lb[i]; |
2106 | if (!upperProvided[i] || upperIndx[i] >= (dim[i] + lb[i])) |
2107 | upperIndx[i] = dim[i] + lb[i] - 1; |
2108 | if (lowerIndx[i] > upperIndx[i]) |
2109 | return PointerGetDatum(construct_empty_array(elemtype)); |
2110 | } |
2111 | /* fill any missing subscript positions with full array range */ |
2112 | for (; i < ndim; i++) |
2113 | { |
2114 | lowerIndx[i] = lb[i]; |
2115 | upperIndx[i] = dim[i] + lb[i] - 1; |
2116 | if (lowerIndx[i] > upperIndx[i]) |
2117 | return PointerGetDatum(construct_empty_array(elemtype)); |
2118 | } |
2119 | |
2120 | mda_get_range(ndim, span, lowerIndx, upperIndx); |
2121 | |
2122 | bytes = array_slice_size(arraydataptr, arraynullsptr, |
2123 | ndim, dim, lb, |
2124 | lowerIndx, upperIndx, |
2125 | elmlen, elmbyval, elmalign); |
2126 | |
2127 | /* |
2128 | * Currently, we put a null bitmap in the result if the source has one; |
2129 | * could be smarter ... |
2130 | */ |
2131 | if (arraynullsptr) |
2132 | { |
2133 | dataoffset = ARR_OVERHEAD_WITHNULLS(ndim, ArrayGetNItems(ndim, span)); |
2134 | bytes += dataoffset; |
2135 | } |
2136 | else |
2137 | { |
2138 | dataoffset = 0; /* marker for no null bitmap */ |
2139 | bytes += ARR_OVERHEAD_NONULLS(ndim); |
2140 | } |
2141 | |
2142 | newarray = (ArrayType *) palloc0(bytes); |
2143 | SET_VARSIZE(newarray, bytes); |
2144 | newarray->ndim = ndim; |
2145 | newarray->dataoffset = dataoffset; |
2146 | newarray->elemtype = elemtype; |
2147 | memcpy(ARR_DIMS(newarray), span, ndim * sizeof(int)); |
2148 | |
2149 | /* |
2150 | * Lower bounds of the new array are set to 1. Formerly (before 7.3) we |
2151 | * copied the given lowerIndx values ... but that seems confusing. |
2152 | */ |
2153 | newlb = ARR_LBOUND(newarray); |
2154 | for (i = 0; i < ndim; i++) |
2155 | newlb[i] = 1; |
2156 | |
2157 | array_extract_slice(newarray, |
2158 | ndim, dim, lb, |
2159 | arraydataptr, arraynullsptr, |
2160 | lowerIndx, upperIndx, |
2161 | elmlen, elmbyval, elmalign); |
2162 | |
2163 | return PointerGetDatum(newarray); |
2164 | } |
2165 | |
2166 | /* |
2167 | * array_set_element : |
2168 | * This routine sets the value of one array element (specified by |
2169 | * a subscript array) to a new value specified by "dataValue". |
2170 | * |
2171 | * This handles both ordinary varlena arrays and fixed-length arrays. |
2172 | * |
2173 | * Inputs: |
2174 | * arraydatum: the initial array object (mustn't be NULL) |
2175 | * nSubscripts: number of subscripts supplied |
2176 | * indx[]: the subscript values |
2177 | * dataValue: the datum to be inserted at the given position |
2178 | * isNull: whether dataValue is NULL |
2179 | * arraytyplen: pg_type.typlen for the array type |
2180 | * elmlen: pg_type.typlen for the array's element type |
2181 | * elmbyval: pg_type.typbyval for the array's element type |
2182 | * elmalign: pg_type.typalign for the array's element type |
2183 | * |
2184 | * Result: |
2185 | * A new array is returned, just like the old except for the one |
2186 | * modified entry. The original array object is not changed, |
2187 | * unless what is passed is a read-write reference to an expanded |
2188 | * array object; in that case the expanded array is updated in-place. |
2189 | * |
2190 | * For one-dimensional arrays only, we allow the array to be extended |
2191 | * by assigning to a position outside the existing subscript range; any |
2192 | * positions between the existing elements and the new one are set to NULLs. |
2193 | * (XXX TODO: allow a corresponding behavior for multidimensional arrays) |
2194 | * |
2195 | * NOTE: For assignments, we throw an error for invalid subscripts etc, |
2196 | * rather than returning a NULL as the fetch operations do. |
2197 | */ |
2198 | Datum |
2199 | array_set_element(Datum arraydatum, |
2200 | int nSubscripts, |
2201 | int *indx, |
2202 | Datum dataValue, |
2203 | bool isNull, |
2204 | int arraytyplen, |
2205 | int elmlen, |
2206 | bool elmbyval, |
2207 | char elmalign) |
2208 | { |
2209 | ArrayType *array; |
2210 | ArrayType *newarray; |
2211 | int i, |
2212 | ndim, |
2213 | dim[MAXDIM], |
2214 | lb[MAXDIM], |
2215 | offset; |
2216 | char *elt_ptr; |
2217 | bool newhasnulls; |
2218 | bits8 *oldnullbitmap; |
2219 | int oldnitems, |
2220 | newnitems, |
2221 | olddatasize, |
2222 | newsize, |
2223 | olditemlen, |
2224 | newitemlen, |
2225 | overheadlen, |
2226 | oldoverheadlen, |
2227 | addedbefore, |
2228 | addedafter, |
2229 | lenbefore, |
2230 | lenafter; |
2231 | |
2232 | if (arraytyplen > 0) |
2233 | { |
2234 | /* |
2235 | * fixed-length arrays -- these are assumed to be 1-d, 0-based. We |
2236 | * cannot extend them, either. |
2237 | */ |
2238 | char *resultarray; |
2239 | |
2240 | if (nSubscripts != 1) |
2241 | ereport(ERROR, |
2242 | (errcode(ERRCODE_ARRAY_SUBSCRIPT_ERROR), |
2243 | errmsg("wrong number of array subscripts" ))); |
2244 | |
2245 | if (indx[0] < 0 || indx[0] * elmlen >= arraytyplen) |
2246 | ereport(ERROR, |
2247 | (errcode(ERRCODE_ARRAY_SUBSCRIPT_ERROR), |
2248 | errmsg("array subscript out of range" ))); |
2249 | |
2250 | if (isNull) |
2251 | ereport(ERROR, |
2252 | (errcode(ERRCODE_NULL_VALUE_NOT_ALLOWED), |
2253 | errmsg("cannot assign null value to an element of a fixed-length array" ))); |
2254 | |
2255 | resultarray = (char *) palloc(arraytyplen); |
2256 | memcpy(resultarray, DatumGetPointer(arraydatum), arraytyplen); |
2257 | elt_ptr = (char *) resultarray + indx[0] * elmlen; |
2258 | ArrayCastAndSet(dataValue, elmlen, elmbyval, elmalign, elt_ptr); |
2259 | return PointerGetDatum(resultarray); |
2260 | } |
2261 | |
2262 | if (nSubscripts <= 0 || nSubscripts > MAXDIM) |
2263 | ereport(ERROR, |
2264 | (errcode(ERRCODE_ARRAY_SUBSCRIPT_ERROR), |
2265 | errmsg("wrong number of array subscripts" ))); |
2266 | |
2267 | /* make sure item to be inserted is not toasted */ |
2268 | if (elmlen == -1 && !isNull) |
2269 | dataValue = PointerGetDatum(PG_DETOAST_DATUM(dataValue)); |
2270 | |
2271 | if (VARATT_IS_EXTERNAL_EXPANDED(DatumGetPointer(arraydatum))) |
2272 | { |
2273 | /* expanded array: let's do this in a separate function */ |
2274 | return array_set_element_expanded(arraydatum, |
2275 | nSubscripts, |
2276 | indx, |
2277 | dataValue, |
2278 | isNull, |
2279 | arraytyplen, |
2280 | elmlen, |
2281 | elmbyval, |
2282 | elmalign); |
2283 | } |
2284 | |
2285 | /* detoast input array if necessary */ |
2286 | array = DatumGetArrayTypeP(arraydatum); |
2287 | |
2288 | ndim = ARR_NDIM(array); |
2289 | |
2290 | /* |
2291 | * if number of dims is zero, i.e. an empty array, create an array with |
2292 | * nSubscripts dimensions, and set the lower bounds to the supplied |
2293 | * subscripts |
2294 | */ |
2295 | if (ndim == 0) |
2296 | { |
2297 | Oid elmtype = ARR_ELEMTYPE(array); |
2298 | |
2299 | for (i = 0; i < nSubscripts; i++) |
2300 | { |
2301 | dim[i] = 1; |
2302 | lb[i] = indx[i]; |
2303 | } |
2304 | |
2305 | return PointerGetDatum(construct_md_array(&dataValue, &isNull, |
2306 | nSubscripts, dim, lb, |
2307 | elmtype, |
2308 | elmlen, elmbyval, elmalign)); |
2309 | } |
2310 | |
2311 | if (ndim != nSubscripts) |
2312 | ereport(ERROR, |
2313 | (errcode(ERRCODE_ARRAY_SUBSCRIPT_ERROR), |
2314 | errmsg("wrong number of array subscripts" ))); |
2315 | |
2316 | /* copy dim/lb since we may modify them */ |
2317 | memcpy(dim, ARR_DIMS(array), ndim * sizeof(int)); |
2318 | memcpy(lb, ARR_LBOUND(array), ndim * sizeof(int)); |
2319 | |
2320 | newhasnulls = (ARR_HASNULL(array) || isNull); |
2321 | addedbefore = addedafter = 0; |
2322 | |
2323 | /* |
2324 | * Check subscripts |
2325 | */ |
2326 | if (ndim == 1) |
2327 | { |
2328 | if (indx[0] < lb[0]) |
2329 | { |
2330 | addedbefore = lb[0] - indx[0]; |
2331 | dim[0] += addedbefore; |
2332 | lb[0] = indx[0]; |
2333 | if (addedbefore > 1) |
2334 | newhasnulls = true; /* will insert nulls */ |
2335 | } |
2336 | if (indx[0] >= (dim[0] + lb[0])) |
2337 | { |
2338 | addedafter = indx[0] - (dim[0] + lb[0]) + 1; |
2339 | dim[0] += addedafter; |
2340 | if (addedafter > 1) |
2341 | newhasnulls = true; /* will insert nulls */ |
2342 | } |
2343 | } |
2344 | else |
2345 | { |
2346 | /* |
2347 | * XXX currently we do not support extending multi-dimensional arrays |
2348 | * during assignment |
2349 | */ |
2350 | for (i = 0; i < ndim; i++) |
2351 | { |
2352 | if (indx[i] < lb[i] || |
2353 | indx[i] >= (dim[i] + lb[i])) |
2354 | ereport(ERROR, |
2355 | (errcode(ERRCODE_ARRAY_SUBSCRIPT_ERROR), |
2356 | errmsg("array subscript out of range" ))); |
2357 | } |
2358 | } |
2359 | |
2360 | /* |
2361 | * Compute sizes of items and areas to copy |
2362 | */ |
2363 | newnitems = ArrayGetNItems(ndim, dim); |
2364 | if (newhasnulls) |
2365 | overheadlen = ARR_OVERHEAD_WITHNULLS(ndim, newnitems); |
2366 | else |
2367 | overheadlen = ARR_OVERHEAD_NONULLS(ndim); |
2368 | oldnitems = ArrayGetNItems(ndim, ARR_DIMS(array)); |
2369 | oldnullbitmap = ARR_NULLBITMAP(array); |
2370 | oldoverheadlen = ARR_DATA_OFFSET(array); |
2371 | olddatasize = ARR_SIZE(array) - oldoverheadlen; |
2372 | if (addedbefore) |
2373 | { |
2374 | offset = 0; |
2375 | lenbefore = 0; |
2376 | olditemlen = 0; |
2377 | lenafter = olddatasize; |
2378 | } |
2379 | else if (addedafter) |
2380 | { |
2381 | offset = oldnitems; |
2382 | lenbefore = olddatasize; |
2383 | olditemlen = 0; |
2384 | lenafter = 0; |
2385 | } |
2386 | else |
2387 | { |
2388 | offset = ArrayGetOffset(nSubscripts, dim, lb, indx); |
2389 | elt_ptr = array_seek(ARR_DATA_PTR(array), 0, oldnullbitmap, offset, |
2390 | elmlen, elmbyval, elmalign); |
2391 | lenbefore = (int) (elt_ptr - ARR_DATA_PTR(array)); |
2392 | if (array_get_isnull(oldnullbitmap, offset)) |
2393 | olditemlen = 0; |
2394 | else |
2395 | { |
2396 | olditemlen = att_addlength_pointer(0, elmlen, elt_ptr); |
2397 | olditemlen = att_align_nominal(olditemlen, elmalign); |
2398 | } |
2399 | lenafter = (int) (olddatasize - lenbefore - olditemlen); |
2400 | } |
2401 | |
2402 | if (isNull) |
2403 | newitemlen = 0; |
2404 | else |
2405 | { |
2406 | newitemlen = att_addlength_datum(0, elmlen, dataValue); |
2407 | newitemlen = att_align_nominal(newitemlen, elmalign); |
2408 | } |
2409 | |
2410 | newsize = overheadlen + lenbefore + newitemlen + lenafter; |
2411 | |
2412 | /* |
2413 | * OK, create the new array and fill in header/dimensions |
2414 | */ |
2415 | newarray = (ArrayType *) palloc0(newsize); |
2416 | SET_VARSIZE(newarray, newsize); |
2417 | newarray->ndim = ndim; |
2418 | newarray->dataoffset = newhasnulls ? overheadlen : 0; |
2419 | newarray->elemtype = ARR_ELEMTYPE(array); |
2420 | memcpy(ARR_DIMS(newarray), dim, ndim * sizeof(int)); |
2421 | memcpy(ARR_LBOUND(newarray), lb, ndim * sizeof(int)); |
2422 | |
2423 | /* |
2424 | * Fill in data |
2425 | */ |
2426 | memcpy((char *) newarray + overheadlen, |
2427 | (char *) array + oldoverheadlen, |
2428 | lenbefore); |
2429 | if (!isNull) |
2430 | ArrayCastAndSet(dataValue, elmlen, elmbyval, elmalign, |
2431 | (char *) newarray + overheadlen + lenbefore); |
2432 | memcpy((char *) newarray + overheadlen + lenbefore + newitemlen, |
2433 | (char *) array + oldoverheadlen + lenbefore + olditemlen, |
2434 | lenafter); |
2435 | |
2436 | /* |
2437 | * Fill in nulls bitmap if needed |
2438 | * |
2439 | * Note: it's possible we just replaced the last NULL with a non-NULL, and |
2440 | * could get rid of the bitmap. Seems not worth testing for though. |
2441 | */ |
2442 | if (newhasnulls) |
2443 | { |
2444 | bits8 *newnullbitmap = ARR_NULLBITMAP(newarray); |
2445 | |
2446 | /* Zero the bitmap to take care of marking inserted positions null */ |
2447 | MemSet(newnullbitmap, 0, (newnitems + 7) / 8); |
2448 | /* Fix the inserted value */ |
2449 | if (addedafter) |
2450 | array_set_isnull(newnullbitmap, newnitems - 1, isNull); |
2451 | else |
2452 | array_set_isnull(newnullbitmap, offset, isNull); |
2453 | /* Fix the copied range(s) */ |
2454 | if (addedbefore) |
2455 | array_bitmap_copy(newnullbitmap, addedbefore, |
2456 | oldnullbitmap, 0, |
2457 | oldnitems); |
2458 | else |
2459 | { |
2460 | array_bitmap_copy(newnullbitmap, 0, |
2461 | oldnullbitmap, 0, |
2462 | offset); |
2463 | if (addedafter == 0) |
2464 | array_bitmap_copy(newnullbitmap, offset + 1, |
2465 | oldnullbitmap, offset + 1, |
2466 | oldnitems - offset - 1); |
2467 | } |
2468 | } |
2469 | |
2470 | return PointerGetDatum(newarray); |
2471 | } |
2472 | |
2473 | /* |
2474 | * Implementation of array_set_element() for an expanded array |
2475 | * |
2476 | * Note: as with any operation on a read/write expanded object, we must |
2477 | * take pains not to leave the object in a corrupt state if we fail partway |
2478 | * through. |
2479 | */ |
2480 | static Datum |
2481 | array_set_element_expanded(Datum arraydatum, |
2482 | int nSubscripts, int *indx, |
2483 | Datum dataValue, bool isNull, |
2484 | int arraytyplen, |
2485 | int elmlen, bool elmbyval, char elmalign) |
2486 | { |
2487 | ExpandedArrayHeader *eah; |
2488 | Datum *dvalues; |
2489 | bool *dnulls; |
2490 | int i, |
2491 | ndim, |
2492 | dim[MAXDIM], |
2493 | lb[MAXDIM], |
2494 | offset; |
2495 | bool dimschanged, |
2496 | newhasnulls; |
2497 | int addedbefore, |
2498 | addedafter; |
2499 | char *oldValue; |
2500 | |
2501 | /* Convert to R/W object if not so already */ |
2502 | eah = DatumGetExpandedArray(arraydatum); |
2503 | |
2504 | /* Sanity-check caller's info against object; we don't use it otherwise */ |
2505 | Assert(arraytyplen == -1); |
2506 | Assert(elmlen == eah->typlen); |
2507 | Assert(elmbyval == eah->typbyval); |
2508 | Assert(elmalign == eah->typalign); |
2509 | |
2510 | /* |
2511 | * Copy dimension info into local storage. This allows us to modify the |
2512 | * dimensions if needed, while not messing up the expanded value if we |
2513 | * fail partway through. |
2514 | */ |
2515 | ndim = eah->ndims; |
2516 | Assert(ndim >= 0 && ndim <= MAXDIM); |
2517 | memcpy(dim, eah->dims, ndim * sizeof(int)); |
2518 | memcpy(lb, eah->lbound, ndim * sizeof(int)); |
2519 | dimschanged = false; |
2520 | |
2521 | /* |
2522 | * if number of dims is zero, i.e. an empty array, create an array with |
2523 | * nSubscripts dimensions, and set the lower bounds to the supplied |
2524 | * subscripts. |
2525 | */ |
2526 | if (ndim == 0) |
2527 | { |
2528 | /* |
2529 | * Allocate adequate space for new dimension info. This is harmless |
2530 | * if we fail later. |
2531 | */ |
2532 | Assert(nSubscripts > 0 && nSubscripts <= MAXDIM); |
2533 | eah->dims = (int *) MemoryContextAllocZero(eah->hdr.eoh_context, |
2534 | nSubscripts * sizeof(int)); |
2535 | eah->lbound = (int *) MemoryContextAllocZero(eah->hdr.eoh_context, |
2536 | nSubscripts * sizeof(int)); |
2537 | |
2538 | /* Update local copies of dimension info */ |
2539 | ndim = nSubscripts; |
2540 | for (i = 0; i < nSubscripts; i++) |
2541 | { |
2542 | dim[i] = 0; |
2543 | lb[i] = indx[i]; |
2544 | } |
2545 | dimschanged = true; |
2546 | } |
2547 | else if (ndim != nSubscripts) |
2548 | ereport(ERROR, |
2549 | (errcode(ERRCODE_ARRAY_SUBSCRIPT_ERROR), |
2550 | errmsg("wrong number of array subscripts" ))); |
2551 | |
2552 | /* |
2553 | * Deconstruct array if we didn't already. (Someday maybe add a special |
2554 | * case path for fixed-length, no-nulls cases, where we can overwrite an |
2555 | * element in place without ever deconstructing. But today is not that |
2556 | * day.) |
2557 | */ |
2558 | deconstruct_expanded_array(eah); |
2559 | |
2560 | /* |
2561 | * Copy new element into array's context, if needed (we assume it's |
2562 | * already detoasted, so no junk should be created). If we fail further |
2563 | * down, this memory is leaked, but that's reasonably harmless. |
2564 | */ |
2565 | if (!eah->typbyval && !isNull) |
2566 | { |
2567 | MemoryContext oldcxt = MemoryContextSwitchTo(eah->hdr.eoh_context); |
2568 | |
2569 | dataValue = datumCopy(dataValue, false, eah->typlen); |
2570 | MemoryContextSwitchTo(oldcxt); |
2571 | } |
2572 | |
2573 | dvalues = eah->dvalues; |
2574 | dnulls = eah->dnulls; |
2575 | |
2576 | newhasnulls = ((dnulls != NULL) || isNull); |
2577 | addedbefore = addedafter = 0; |
2578 | |
2579 | /* |
2580 | * Check subscripts (this logic matches original array_set_element) |
2581 | */ |
2582 | if (ndim == 1) |
2583 | { |
2584 | if (indx[0] < lb[0]) |
2585 | { |
2586 | addedbefore = lb[0] - indx[0]; |
2587 | dim[0] += addedbefore; |
2588 | lb[0] = indx[0]; |
2589 | dimschanged = true; |
2590 | if (addedbefore > 1) |
2591 | newhasnulls = true; /* will insert nulls */ |
2592 | } |
2593 | if (indx[0] >= (dim[0] + lb[0])) |
2594 | { |
2595 | addedafter = indx[0] - (dim[0] + lb[0]) + 1; |
2596 | dim[0] += addedafter; |
2597 | dimschanged = true; |
2598 | if (addedafter > 1) |
2599 | newhasnulls = true; /* will insert nulls */ |
2600 | } |
2601 | } |
2602 | else |
2603 | { |
2604 | /* |
2605 | * XXX currently we do not support extending multi-dimensional arrays |
2606 | * during assignment |
2607 | */ |
2608 | for (i = 0; i < ndim; i++) |
2609 | { |
2610 | if (indx[i] < lb[i] || |
2611 | indx[i] >= (dim[i] + lb[i])) |
2612 | ereport(ERROR, |
2613 | (errcode(ERRCODE_ARRAY_SUBSCRIPT_ERROR), |
2614 | errmsg("array subscript out of range" ))); |
2615 | } |
2616 | } |
2617 | |
2618 | /* Now we can calculate linear offset of target item in array */ |
2619 | offset = ArrayGetOffset(nSubscripts, dim, lb, indx); |
2620 | |
2621 | /* Physically enlarge existing dvalues/dnulls arrays if needed */ |
2622 | if (dim[0] > eah->dvalueslen) |
2623 | { |
2624 | /* We want some extra space if we're enlarging */ |
2625 | int newlen = dim[0] + dim[0] / 8; |
2626 | |
2627 | newlen = Max(newlen, dim[0]); /* integer overflow guard */ |
2628 | eah->dvalues = dvalues = (Datum *) |
2629 | repalloc(dvalues, newlen * sizeof(Datum)); |
2630 | if (dnulls) |
2631 | eah->dnulls = dnulls = (bool *) |
2632 | repalloc(dnulls, newlen * sizeof(bool)); |
2633 | eah->dvalueslen = newlen; |
2634 | } |
2635 | |
2636 | /* |
2637 | * If we need a nulls bitmap and don't already have one, create it, being |
2638 | * sure to mark all existing entries as not null. |
2639 | */ |
2640 | if (newhasnulls && dnulls == NULL) |
2641 | eah->dnulls = dnulls = (bool *) |
2642 | MemoryContextAllocZero(eah->hdr.eoh_context, |
2643 | eah->dvalueslen * sizeof(bool)); |
2644 | |
2645 | /* |
2646 | * We now have all the needed space allocated, so we're ready to make |
2647 | * irreversible changes. Be very wary of allowing failure below here. |
2648 | */ |
2649 | |
2650 | /* Flattened value will no longer represent array accurately */ |
2651 | eah->fvalue = NULL; |
2652 | /* And we don't know the flattened size either */ |
2653 | eah->flat_size = 0; |
2654 | |
2655 | /* Update dimensionality info if needed */ |
2656 | if (dimschanged) |
2657 | { |
2658 | eah->ndims = ndim; |
2659 | memcpy(eah->dims, dim, ndim * sizeof(int)); |
2660 | memcpy(eah->lbound, lb, ndim * sizeof(int)); |
2661 | } |
2662 | |
2663 | /* Reposition items if needed, and fill addedbefore items with nulls */ |
2664 | if (addedbefore > 0) |
2665 | { |
2666 | memmove(dvalues + addedbefore, dvalues, eah->nelems * sizeof(Datum)); |
2667 | for (i = 0; i < addedbefore; i++) |
2668 | dvalues[i] = (Datum) 0; |
2669 | if (dnulls) |
2670 | { |
2671 | memmove(dnulls + addedbefore, dnulls, eah->nelems * sizeof(bool)); |
2672 | for (i = 0; i < addedbefore; i++) |
2673 | dnulls[i] = true; |
2674 | } |
2675 | eah->nelems += addedbefore; |
2676 | } |
2677 | |
2678 | /* fill addedafter items with nulls */ |
2679 | if (addedafter > 0) |
2680 | { |
2681 | for (i = 0; i < addedafter; i++) |
2682 | dvalues[eah->nelems + i] = (Datum) 0; |
2683 | if (dnulls) |
2684 | { |
2685 | for (i = 0; i < addedafter; i++) |
2686 | dnulls[eah->nelems + i] = true; |
2687 | } |
2688 | eah->nelems += addedafter; |
2689 | } |
2690 | |
2691 | /* Grab old element value for pfree'ing, if needed. */ |
2692 | if (!eah->typbyval && (dnulls == NULL || !dnulls[offset])) |
2693 | oldValue = (char *) DatumGetPointer(dvalues[offset]); |
2694 | else |
2695 | oldValue = NULL; |
2696 | |
2697 | /* And finally we can insert the new element. */ |
2698 | dvalues[offset] = dataValue; |
2699 | if (dnulls) |
2700 | dnulls[offset] = isNull; |
2701 | |
2702 | /* |
2703 | * Free old element if needed; this keeps repeated element replacements |
2704 | * from bloating the array's storage. If the pfree somehow fails, it |
2705 | * won't corrupt the array. |
2706 | */ |
2707 | if (oldValue) |
2708 | { |
2709 | /* Don't try to pfree a part of the original flat array */ |
2710 | if (oldValue < eah->fstartptr || oldValue >= eah->fendptr) |
2711 | pfree(oldValue); |
2712 | } |
2713 | |
2714 | /* Done, return standard TOAST pointer for object */ |
2715 | return EOHPGetRWDatum(&eah->hdr); |
2716 | } |
2717 | |
2718 | /* |
2719 | * array_set_slice : |
2720 | * This routine sets the value of a range of array locations (specified |
2721 | * by upper and lower subscript values) to new values passed as |
2722 | * another array. |
2723 | * |
2724 | * This handles both ordinary varlena arrays and fixed-length arrays. |
2725 | * |
2726 | * Inputs: |
2727 | * arraydatum: the initial array object (mustn't be NULL) |
2728 | * nSubscripts: number of subscripts supplied (must be same for upper/lower) |
2729 | * upperIndx[]: the upper subscript values |
2730 | * lowerIndx[]: the lower subscript values |
2731 | * upperProvided[]: true for provided upper subscript values |
2732 | * lowerProvided[]: true for provided lower subscript values |
2733 | * srcArrayDatum: the source for the inserted values |
2734 | * isNull: indicates whether srcArrayDatum is NULL |
2735 | * arraytyplen: pg_type.typlen for the array type |
2736 | * elmlen: pg_type.typlen for the array's element type |
2737 | * elmbyval: pg_type.typbyval for the array's element type |
2738 | * elmalign: pg_type.typalign for the array's element type |
2739 | * |
2740 | * Result: |
2741 | * A new array is returned, just like the old except for the |
2742 | * modified range. The original array object is not changed. |
2743 | * |
2744 | * Omitted upper and lower subscript values are replaced by the corresponding |
2745 | * array bound. |
2746 | * |
2747 | * For one-dimensional arrays only, we allow the array to be extended |
2748 | * by assigning to positions outside the existing subscript range; any |
2749 | * positions between the existing elements and the new ones are set to NULLs. |
2750 | * (XXX TODO: allow a corresponding behavior for multidimensional arrays) |
2751 | * |
2752 | * NOTE: we assume it is OK to scribble on the provided index arrays |
2753 | * lowerIndx[] and upperIndx[]. These are generally just temporaries. |
2754 | * |
2755 | * NOTE: For assignments, we throw an error for silly subscripts etc, |
2756 | * rather than returning a NULL or empty array as the fetch operations do. |
2757 | */ |
2758 | Datum |
2759 | array_set_slice(Datum arraydatum, |
2760 | int nSubscripts, |
2761 | int *upperIndx, |
2762 | int *lowerIndx, |
2763 | bool *upperProvided, |
2764 | bool *lowerProvided, |
2765 | Datum srcArrayDatum, |
2766 | bool isNull, |
2767 | int arraytyplen, |
2768 | int elmlen, |
2769 | bool elmbyval, |
2770 | char elmalign) |
2771 | { |
2772 | ArrayType *array; |
2773 | ArrayType *srcArray; |
2774 | ArrayType *newarray; |
2775 | int i, |
2776 | ndim, |
2777 | dim[MAXDIM], |
2778 | lb[MAXDIM], |
2779 | span[MAXDIM]; |
2780 | bool newhasnulls; |
2781 | int nitems, |
2782 | nsrcitems, |
2783 | olddatasize, |
2784 | newsize, |
2785 | olditemsize, |
2786 | newitemsize, |
2787 | overheadlen, |
2788 | oldoverheadlen, |
2789 | addedbefore, |
2790 | addedafter, |
2791 | lenbefore, |
2792 | lenafter, |
2793 | itemsbefore, |
2794 | itemsafter, |
2795 | nolditems; |
2796 | |
2797 | /* Currently, assignment from a NULL source array is a no-op */ |
2798 | if (isNull) |
2799 | return arraydatum; |
2800 | |
2801 | if (arraytyplen > 0) |
2802 | { |
2803 | /* |
2804 | * fixed-length arrays -- not got round to doing this... |
2805 | */ |
2806 | ereport(ERROR, |
2807 | (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), |
2808 | errmsg("updates on slices of fixed-length arrays not implemented" ))); |
2809 | } |
2810 | |
2811 | /* detoast arrays if necessary */ |
2812 | array = DatumGetArrayTypeP(arraydatum); |
2813 | srcArray = DatumGetArrayTypeP(srcArrayDatum); |
2814 | |
2815 | /* note: we assume srcArray contains no toasted elements */ |
2816 | |
2817 | ndim = ARR_NDIM(array); |
2818 | |
2819 | /* |
2820 | * if number of dims is zero, i.e. an empty array, create an array with |
2821 | * nSubscripts dimensions, and set the upper and lower bounds to the |
2822 | * supplied subscripts |
2823 | */ |
2824 | if (ndim == 0) |
2825 | { |
2826 | Datum *dvalues; |
2827 | bool *dnulls; |
2828 | int nelems; |
2829 | Oid elmtype = ARR_ELEMTYPE(array); |
2830 | |
2831 | deconstruct_array(srcArray, elmtype, elmlen, elmbyval, elmalign, |
2832 | &dvalues, &dnulls, &nelems); |
2833 | |
2834 | for (i = 0; i < nSubscripts; i++) |
2835 | { |
2836 | if (!upperProvided[i] || !lowerProvided[i]) |
2837 | ereport(ERROR, |
2838 | (errcode(ERRCODE_ARRAY_SUBSCRIPT_ERROR), |
2839 | errmsg("array slice subscript must provide both boundaries" ), |
2840 | errdetail("When assigning to a slice of an empty array value," |
2841 | " slice boundaries must be fully specified." ))); |
2842 | |
2843 | dim[i] = 1 + upperIndx[i] - lowerIndx[i]; |
2844 | lb[i] = lowerIndx[i]; |
2845 | } |
2846 | |
2847 | /* complain if too few source items; we ignore extras, however */ |
2848 | if (nelems < ArrayGetNItems(nSubscripts, dim)) |
2849 | ereport(ERROR, |
2850 | (errcode(ERRCODE_ARRAY_SUBSCRIPT_ERROR), |
2851 | errmsg("source array too small" ))); |
2852 | |
2853 | return PointerGetDatum(construct_md_array(dvalues, dnulls, nSubscripts, |
2854 | dim, lb, elmtype, |
2855 | elmlen, elmbyval, elmalign)); |
2856 | } |
2857 | |
2858 | if (ndim < nSubscripts || ndim <= 0 || ndim > MAXDIM) |
2859 | ereport(ERROR, |
2860 | (errcode(ERRCODE_ARRAY_SUBSCRIPT_ERROR), |
2861 | errmsg("wrong number of array subscripts" ))); |
2862 | |
2863 | /* copy dim/lb since we may modify them */ |
2864 | memcpy(dim, ARR_DIMS(array), ndim * sizeof(int)); |
2865 | memcpy(lb, ARR_LBOUND(array), ndim * sizeof(int)); |
2866 | |
2867 | newhasnulls = (ARR_HASNULL(array) || ARR_HASNULL(srcArray)); |
2868 | addedbefore = addedafter = 0; |
2869 | |
2870 | /* |
2871 | * Check subscripts |
2872 | */ |
2873 | if (ndim == 1) |
2874 | { |
2875 | Assert(nSubscripts == 1); |
2876 | if (!lowerProvided[0]) |
2877 | lowerIndx[0] = lb[0]; |
2878 | if (!upperProvided[0]) |
2879 | upperIndx[0] = dim[0] + lb[0] - 1; |
2880 | if (lowerIndx[0] > upperIndx[0]) |
2881 | ereport(ERROR, |
2882 | (errcode(ERRCODE_ARRAY_SUBSCRIPT_ERROR), |
2883 | errmsg("upper bound cannot be less than lower bound" ))); |
2884 | if (lowerIndx[0] < lb[0]) |
2885 | { |
2886 | if (upperIndx[0] < lb[0] - 1) |
2887 | newhasnulls = true; /* will insert nulls */ |
2888 | addedbefore = lb[0] - lowerIndx[0]; |
2889 | dim[0] += addedbefore; |
2890 | lb[0] = lowerIndx[0]; |
2891 | } |
2892 | if (upperIndx[0] >= (dim[0] + lb[0])) |
2893 | { |
2894 | if (lowerIndx[0] > (dim[0] + lb[0])) |
2895 | newhasnulls = true; /* will insert nulls */ |
2896 | addedafter = upperIndx[0] - (dim[0] + lb[0]) + 1; |
2897 | dim[0] += addedafter; |
2898 | } |
2899 | } |
2900 | else |
2901 | { |
2902 | /* |
2903 | * XXX currently we do not support extending multi-dimensional arrays |
2904 | * during assignment |
2905 | */ |
2906 | for (i = 0; i < nSubscripts; i++) |
2907 | { |
2908 | if (!lowerProvided[i]) |
2909 | lowerIndx[i] = lb[i]; |
2910 | if (!upperProvided[i]) |
2911 | upperIndx[i] = dim[i] + lb[i] - 1; |
2912 | if (lowerIndx[i] > upperIndx[i]) |
2913 | ereport(ERROR, |
2914 | (errcode(ERRCODE_ARRAY_SUBSCRIPT_ERROR), |
2915 | errmsg("upper bound cannot be less than lower bound" ))); |
2916 | if (lowerIndx[i] < lb[i] || |
2917 | upperIndx[i] >= (dim[i] + lb[i])) |
2918 | ereport(ERROR, |
2919 | (errcode(ERRCODE_ARRAY_SUBSCRIPT_ERROR), |
2920 | errmsg("array subscript out of range" ))); |
2921 | } |
2922 | /* fill any missing subscript positions with full array range */ |
2923 | for (; i < ndim; i++) |
2924 | { |
2925 | lowerIndx[i] = lb[i]; |
2926 | upperIndx[i] = dim[i] + lb[i] - 1; |
2927 | if (lowerIndx[i] > upperIndx[i]) |
2928 | ereport(ERROR, |
2929 | (errcode(ERRCODE_ARRAY_SUBSCRIPT_ERROR), |
2930 | errmsg("upper bound cannot be less than lower bound" ))); |
2931 | } |
2932 | } |
2933 | |
2934 | /* Do this mainly to check for overflow */ |
2935 | nitems = ArrayGetNItems(ndim, dim); |
2936 | |
2937 | /* |
2938 | * Make sure source array has enough entries. Note we ignore the shape of |
2939 | * the source array and just read entries serially. |
2940 | */ |
2941 | mda_get_range(ndim, span, lowerIndx, upperIndx); |
2942 | nsrcitems = ArrayGetNItems(ndim, span); |
2943 | if (nsrcitems > ArrayGetNItems(ARR_NDIM(srcArray), ARR_DIMS(srcArray))) |
2944 | ereport(ERROR, |
2945 | (errcode(ERRCODE_ARRAY_SUBSCRIPT_ERROR), |
2946 | errmsg("source array too small" ))); |
2947 | |
2948 | /* |
2949 | * Compute space occupied by new entries, space occupied by replaced |
2950 | * entries, and required space for new array. |
2951 | */ |
2952 | if (newhasnulls) |
2953 | overheadlen = ARR_OVERHEAD_WITHNULLS(ndim, nitems); |
2954 | else |
2955 | overheadlen = ARR_OVERHEAD_NONULLS(ndim); |
2956 | newitemsize = array_nelems_size(ARR_DATA_PTR(srcArray), 0, |
2957 | ARR_NULLBITMAP(srcArray), nsrcitems, |
2958 | elmlen, elmbyval, elmalign); |
2959 | oldoverheadlen = ARR_DATA_OFFSET(array); |
2960 | olddatasize = ARR_SIZE(array) - oldoverheadlen; |
2961 | if (ndim > 1) |
2962 | { |
2963 | /* |
2964 | * here we do not need to cope with extension of the array; it would |
2965 | * be a lot more complicated if we had to do so... |
2966 | */ |
2967 | olditemsize = array_slice_size(ARR_DATA_PTR(array), |
2968 | ARR_NULLBITMAP(array), |
2969 | ndim, dim, lb, |
2970 | lowerIndx, upperIndx, |
2971 | elmlen, elmbyval, elmalign); |
2972 | lenbefore = lenafter = 0; /* keep compiler quiet */ |
2973 | itemsbefore = itemsafter = nolditems = 0; |
2974 | } |
2975 | else |
2976 | { |
2977 | /* |
2978 | * here we must allow for possibility of slice larger than orig array |
2979 | * and/or not adjacent to orig array subscripts |
2980 | */ |
2981 | int oldlb = ARR_LBOUND(array)[0]; |
2982 | int oldub = oldlb + ARR_DIMS(array)[0] - 1; |
2983 | int slicelb = Max(oldlb, lowerIndx[0]); |
2984 | int sliceub = Min(oldub, upperIndx[0]); |
2985 | char *oldarraydata = ARR_DATA_PTR(array); |
2986 | bits8 *oldarraybitmap = ARR_NULLBITMAP(array); |
2987 | |
2988 | /* count/size of old array entries that will go before the slice */ |
2989 | itemsbefore = Min(slicelb, oldub + 1) - oldlb; |
2990 | lenbefore = array_nelems_size(oldarraydata, 0, oldarraybitmap, |
2991 | itemsbefore, |
2992 | elmlen, elmbyval, elmalign); |
2993 | /* count/size of old array entries that will be replaced by slice */ |
2994 | if (slicelb > sliceub) |
2995 | { |
2996 | nolditems = 0; |
2997 | olditemsize = 0; |
2998 | } |
2999 | else |
3000 | { |
3001 | nolditems = sliceub - slicelb + 1; |
3002 | olditemsize = array_nelems_size(oldarraydata + lenbefore, |
3003 | itemsbefore, oldarraybitmap, |
3004 | nolditems, |
3005 | elmlen, elmbyval, elmalign); |
3006 | } |
3007 | /* count/size of old array entries that will go after the slice */ |
3008 | itemsafter = oldub + 1 - Max(sliceub + 1, oldlb); |
3009 | lenafter = olddatasize - lenbefore - olditemsize; |
3010 | } |
3011 | |
3012 | newsize = overheadlen + olddatasize - olditemsize + newitemsize; |
3013 | |
3014 | newarray = (ArrayType *) palloc0(newsize); |
3015 | SET_VARSIZE(newarray, newsize); |
3016 | newarray->ndim = ndim; |
3017 | newarray->dataoffset = newhasnulls ? overheadlen : 0; |
3018 | newarray->elemtype = ARR_ELEMTYPE(array); |
3019 | memcpy(ARR_DIMS(newarray), dim, ndim * sizeof(int)); |
3020 | memcpy(ARR_LBOUND(newarray), lb, ndim * sizeof(int)); |
3021 | |
3022 | if (ndim > 1) |
3023 | { |
3024 | /* |
3025 | * here we do not need to cope with extension of the array; it would |
3026 | * be a lot more complicated if we had to do so... |
3027 | */ |
3028 | array_insert_slice(newarray, array, srcArray, |
3029 | ndim, dim, lb, |
3030 | lowerIndx, upperIndx, |
3031 | elmlen, elmbyval, elmalign); |
3032 | } |
3033 | else |
3034 | { |
3035 | /* fill in data */ |
3036 | memcpy((char *) newarray + overheadlen, |
3037 | (char *) array + oldoverheadlen, |
3038 | lenbefore); |
3039 | memcpy((char *) newarray + overheadlen + lenbefore, |
3040 | ARR_DATA_PTR(srcArray), |
3041 | newitemsize); |
3042 | memcpy((char *) newarray + overheadlen + lenbefore + newitemsize, |
3043 | (char *) array + oldoverheadlen + lenbefore + olditemsize, |
3044 | lenafter); |
3045 | /* fill in nulls bitmap if needed */ |
3046 | if (newhasnulls) |
3047 | { |
3048 | bits8 *newnullbitmap = ARR_NULLBITMAP(newarray); |
3049 | bits8 *oldnullbitmap = ARR_NULLBITMAP(array); |
3050 | |
3051 | /* Zero the bitmap to handle marking inserted positions null */ |
3052 | MemSet(newnullbitmap, 0, (nitems + 7) / 8); |
3053 | array_bitmap_copy(newnullbitmap, addedbefore, |
3054 | oldnullbitmap, 0, |
3055 | itemsbefore); |
3056 | array_bitmap_copy(newnullbitmap, lowerIndx[0] - lb[0], |
3057 | ARR_NULLBITMAP(srcArray), 0, |
3058 | nsrcitems); |
3059 | array_bitmap_copy(newnullbitmap, addedbefore + itemsbefore + nolditems, |
3060 | oldnullbitmap, itemsbefore + nolditems, |
3061 | itemsafter); |
3062 | } |
3063 | } |
3064 | |
3065 | return PointerGetDatum(newarray); |
3066 | } |
3067 | |
3068 | /* |
3069 | * array_ref : backwards compatibility wrapper for array_get_element |
3070 | * |
3071 | * This only works for detoasted/flattened varlena arrays, since the array |
3072 | * argument is declared as "ArrayType *". However there's enough code like |
3073 | * that to justify preserving this API. |
3074 | */ |
3075 | Datum |
3076 | array_ref(ArrayType *array, int nSubscripts, int *indx, |
3077 | int arraytyplen, int elmlen, bool elmbyval, char elmalign, |
3078 | bool *isNull) |
3079 | { |
3080 | return array_get_element(PointerGetDatum(array), nSubscripts, indx, |
3081 | arraytyplen, elmlen, elmbyval, elmalign, |
3082 | isNull); |
3083 | } |
3084 | |
3085 | /* |
3086 | * array_set : backwards compatibility wrapper for array_set_element |
3087 | * |
3088 | * This only works for detoasted/flattened varlena arrays, since the array |
3089 | * argument and result are declared as "ArrayType *". However there's enough |
3090 | * code like that to justify preserving this API. |
3091 | */ |
3092 | ArrayType * |
3093 | array_set(ArrayType *array, int nSubscripts, int *indx, |
3094 | Datum dataValue, bool isNull, |
3095 | int arraytyplen, int elmlen, bool elmbyval, char elmalign) |
3096 | { |
3097 | return DatumGetArrayTypeP(array_set_element(PointerGetDatum(array), |
3098 | nSubscripts, indx, |
3099 | dataValue, isNull, |
3100 | arraytyplen, |
3101 | elmlen, elmbyval, elmalign)); |
3102 | } |
3103 | |
3104 | /* |
3105 | * array_map() |
3106 | * |
3107 | * Map an array through an arbitrary expression. Return a new array with |
3108 | * the same dimensions and each source element transformed by the given, |
3109 | * already-compiled expression. Each source element is placed in the |
3110 | * innermost_caseval/innermost_casenull fields of the ExprState. |
3111 | * |
3112 | * Parameters are: |
3113 | * * arrayd: Datum representing array argument. |
3114 | * * exprstate: ExprState representing the per-element transformation. |
3115 | * * econtext: context for expression evaluation. |
3116 | * * retType: OID of element type of output array. This must be the same as, |
3117 | * or binary-compatible with, the result type of the expression. It might |
3118 | * be different from the input array's element type. |
3119 | * * amstate: workspace for array_map. Must be zeroed by caller before |
3120 | * first call, and not touched after that. |
3121 | * |
3122 | * It is legitimate to pass a freshly-zeroed ArrayMapState on each call, |
3123 | * but better performance can be had if the state can be preserved across |
3124 | * a series of calls. |
3125 | * |
3126 | * NB: caller must assure that input array is not NULL. NULL elements in |
3127 | * the array are OK however. |
3128 | * NB: caller should be running in econtext's per-tuple memory context. |
3129 | */ |
3130 | Datum |
3131 | array_map(Datum arrayd, |
3132 | ExprState *exprstate, ExprContext *econtext, |
3133 | Oid retType, ArrayMapState *amstate) |
3134 | { |
3135 | AnyArrayType *v = DatumGetAnyArrayP(arrayd); |
3136 | ArrayType *result; |
3137 | Datum *values; |
3138 | bool *nulls; |
3139 | int *dim; |
3140 | int ndim; |
3141 | int nitems; |
3142 | int i; |
3143 | int32 nbytes = 0; |
3144 | int32 dataoffset; |
3145 | bool hasnulls; |
3146 | Oid inpType; |
3147 | int inp_typlen; |
3148 | bool inp_typbyval; |
3149 | char inp_typalign; |
3150 | int typlen; |
3151 | bool typbyval; |
3152 | char typalign; |
3153 | array_iter iter; |
3154 | ArrayMetaState *; |
3155 | ArrayMetaState *; |
3156 | Datum *transform_source = exprstate->innermost_caseval; |
3157 | bool *transform_source_isnull = exprstate->innermost_casenull; |
3158 | |
3159 | inpType = AARR_ELEMTYPE(v); |
3160 | ndim = AARR_NDIM(v); |
3161 | dim = AARR_DIMS(v); |
3162 | nitems = ArrayGetNItems(ndim, dim); |
3163 | |
3164 | /* Check for empty array */ |
3165 | if (nitems <= 0) |
3166 | { |
3167 | /* Return empty array */ |
3168 | return PointerGetDatum(construct_empty_array(retType)); |
3169 | } |
3170 | |
3171 | /* |
3172 | * We arrange to look up info about input and return element types only |
3173 | * once per series of calls, assuming the element type doesn't change |
3174 | * underneath us. |
3175 | */ |
3176 | inp_extra = &amstate->inp_extra; |
3177 | ret_extra = &amstate->ret_extra; |
3178 | |
3179 | if (inp_extra->element_type != inpType) |
3180 | { |
3181 | get_typlenbyvalalign(inpType, |
3182 | &inp_extra->typlen, |
3183 | &inp_extra->typbyval, |
3184 | &inp_extra->typalign); |
3185 | inp_extra->element_type = inpType; |
3186 | } |
3187 | inp_typlen = inp_extra->typlen; |
3188 | inp_typbyval = inp_extra->typbyval; |
3189 | inp_typalign = inp_extra->typalign; |
3190 | |
3191 | if (ret_extra->element_type != retType) |
3192 | { |
3193 | get_typlenbyvalalign(retType, |
3194 | &ret_extra->typlen, |
3195 | &ret_extra->typbyval, |
3196 | &ret_extra->typalign); |
3197 | ret_extra->element_type = retType; |
3198 | } |
3199 | typlen = ret_extra->typlen; |
3200 | typbyval = ret_extra->typbyval; |
3201 | typalign = ret_extra->typalign; |
3202 | |
3203 | /* Allocate temporary arrays for new values */ |
3204 | values = (Datum *) palloc(nitems * sizeof(Datum)); |
3205 | nulls = (bool *) palloc(nitems * sizeof(bool)); |
3206 | |
3207 | /* Loop over source data */ |
3208 | array_iter_setup(&iter, v); |
3209 | hasnulls = false; |
3210 | |
3211 | for (i = 0; i < nitems; i++) |
3212 | { |
3213 | /* Get source element, checking for NULL */ |
3214 | *transform_source = |
3215 | array_iter_next(&iter, transform_source_isnull, i, |
3216 | inp_typlen, inp_typbyval, inp_typalign); |
3217 | |
3218 | /* Apply the given expression to source element */ |
3219 | values[i] = ExecEvalExpr(exprstate, econtext, &nulls[i]); |
3220 | |
3221 | if (nulls[i]) |
3222 | hasnulls = true; |
3223 | else |
3224 | { |
3225 | /* Ensure data is not toasted */ |
3226 | if (typlen == -1) |
3227 | values[i] = PointerGetDatum(PG_DETOAST_DATUM(values[i])); |
3228 | /* Update total result size */ |
3229 | nbytes = att_addlength_datum(nbytes, typlen, values[i]); |
3230 | nbytes = att_align_nominal(nbytes, typalign); |
3231 | /* check for overflow of total request */ |
3232 | if (!AllocSizeIsValid(nbytes)) |
3233 | ereport(ERROR, |
3234 | (errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED), |
3235 | errmsg("array size exceeds the maximum allowed (%d)" , |
3236 | (int) MaxAllocSize))); |
3237 | } |
3238 | } |
3239 | |
3240 | /* Allocate and fill the result array */ |
3241 | if (hasnulls) |
3242 | { |
3243 | dataoffset = ARR_OVERHEAD_WITHNULLS(ndim, nitems); |
3244 | nbytes += dataoffset; |
3245 | } |
3246 | else |
3247 | { |
3248 | dataoffset = 0; /* marker for no null bitmap */ |
3249 | nbytes += ARR_OVERHEAD_NONULLS(ndim); |
3250 | } |
3251 | result = (ArrayType *) palloc0(nbytes); |
3252 | SET_VARSIZE(result, nbytes); |
3253 | result->ndim = ndim; |
3254 | result->dataoffset = dataoffset; |
3255 | result->elemtype = retType; |
3256 | memcpy(ARR_DIMS(result), AARR_DIMS(v), ndim * sizeof(int)); |
3257 | memcpy(ARR_LBOUND(result), AARR_LBOUND(v), ndim * sizeof(int)); |
3258 | |
3259 | CopyArrayEls(result, |
3260 | values, nulls, nitems, |
3261 | typlen, typbyval, typalign, |
3262 | false); |
3263 | |
3264 | /* |
3265 | * Note: do not risk trying to pfree the results of the called expression |
3266 | */ |
3267 | pfree(values); |
3268 | pfree(nulls); |
3269 | |
3270 | return PointerGetDatum(result); |
3271 | } |
3272 | |
3273 | /* |
3274 | * construct_array --- simple method for constructing an array object |
3275 | * |
3276 | * elems: array of Datum items to become the array contents |
3277 | * (NULL element values are not supported). |
3278 | * nelems: number of items |
3279 | * elmtype, elmlen, elmbyval, elmalign: info for the datatype of the items |
3280 | * |
3281 | * A palloc'd 1-D array object is constructed and returned. Note that |
3282 | * elem values will be copied into the object even if pass-by-ref type. |
3283 | * Also note the result will be 0-D not 1-D if nelems = 0. |
3284 | * |
3285 | * NOTE: it would be cleaner to look up the elmlen/elmbval/elmalign info |
3286 | * from the system catalogs, given the elmtype. However, the caller is |
3287 | * in a better position to cache this info across multiple uses, or even |
3288 | * to hard-wire values if the element type is hard-wired. |
3289 | */ |
3290 | ArrayType * |
3291 | construct_array(Datum *elems, int nelems, |
3292 | Oid elmtype, |
3293 | int elmlen, bool elmbyval, char elmalign) |
3294 | { |
3295 | int dims[1]; |
3296 | int lbs[1]; |
3297 | |
3298 | dims[0] = nelems; |
3299 | lbs[0] = 1; |
3300 | |
3301 | return construct_md_array(elems, NULL, 1, dims, lbs, |
3302 | elmtype, elmlen, elmbyval, elmalign); |
3303 | } |
3304 | |
3305 | /* |
3306 | * construct_md_array --- simple method for constructing an array object |
3307 | * with arbitrary dimensions and possible NULLs |
3308 | * |
3309 | * elems: array of Datum items to become the array contents |
3310 | * nulls: array of is-null flags (can be NULL if no nulls) |
3311 | * ndims: number of dimensions |
3312 | * dims: integer array with size of each dimension |
3313 | * lbs: integer array with lower bound of each dimension |
3314 | * elmtype, elmlen, elmbyval, elmalign: info for the datatype of the items |
3315 | * |
3316 | * A palloc'd ndims-D array object is constructed and returned. Note that |
3317 | * elem values will be copied into the object even if pass-by-ref type. |
3318 | * Also note the result will be 0-D not ndims-D if any dims[i] = 0. |
3319 | * |
3320 | * NOTE: it would be cleaner to look up the elmlen/elmbval/elmalign info |
3321 | * from the system catalogs, given the elmtype. However, the caller is |
3322 | * in a better position to cache this info across multiple uses, or even |
3323 | * to hard-wire values if the element type is hard-wired. |
3324 | */ |
3325 | ArrayType * |
3326 | construct_md_array(Datum *elems, |
3327 | bool *nulls, |
3328 | int ndims, |
3329 | int *dims, |
3330 | int *lbs, |
3331 | Oid elmtype, int elmlen, bool elmbyval, char elmalign) |
3332 | { |
3333 | ArrayType *result; |
3334 | bool hasnulls; |
3335 | int32 nbytes; |
3336 | int32 dataoffset; |
3337 | int i; |
3338 | int nelems; |
3339 | |
3340 | if (ndims < 0) /* we do allow zero-dimension arrays */ |
3341 | ereport(ERROR, |
3342 | (errcode(ERRCODE_INVALID_PARAMETER_VALUE), |
3343 | errmsg("invalid number of dimensions: %d" , ndims))); |
3344 | if (ndims > MAXDIM) |
3345 | ereport(ERROR, |
3346 | (errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED), |
3347 | errmsg("number of array dimensions (%d) exceeds the maximum allowed (%d)" , |
3348 | ndims, MAXDIM))); |
3349 | |
3350 | nelems = ArrayGetNItems(ndims, dims); |
3351 | |
3352 | /* if ndims <= 0 or any dims[i] == 0, return empty array */ |
3353 | if (nelems <= 0) |
3354 | return construct_empty_array(elmtype); |
3355 | |
3356 | /* compute required space */ |
3357 | nbytes = 0; |
3358 | hasnulls = false; |
3359 | for (i = 0; i < nelems; i++) |
3360 | { |
3361 | if (nulls && nulls[i]) |
3362 | { |
3363 | hasnulls = true; |
3364 | continue; |
3365 | } |
3366 | /* make sure data is not toasted */ |
3367 | if (elmlen == -1) |
3368 | elems[i] = PointerGetDatum(PG_DETOAST_DATUM(elems[i])); |
3369 | nbytes = att_addlength_datum(nbytes, elmlen, elems[i]); |
3370 | nbytes = att_align_nominal(nbytes, elmalign); |
3371 | /* check for overflow of total request */ |
3372 | if (!AllocSizeIsValid(nbytes)) |
3373 | ereport(ERROR, |
3374 | (errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED), |
3375 | errmsg("array size exceeds the maximum allowed (%d)" , |
3376 | (int) MaxAllocSize))); |
3377 | } |
3378 | |
3379 | /* Allocate and initialize result array */ |
3380 | if (hasnulls) |
3381 | { |
3382 | dataoffset = ARR_OVERHEAD_WITHNULLS(ndims, nelems); |
3383 | nbytes += dataoffset; |
3384 | } |
3385 | else |
3386 | { |
3387 | dataoffset = 0; /* marker for no null bitmap */ |
3388 | nbytes += ARR_OVERHEAD_NONULLS(ndims); |
3389 | } |
3390 | result = (ArrayType *) palloc0(nbytes); |
3391 | SET_VARSIZE(result, nbytes); |
3392 | result->ndim = ndims; |
3393 | result->dataoffset = dataoffset; |
3394 | result->elemtype = elmtype; |
3395 | memcpy(ARR_DIMS(result), dims, ndims * sizeof(int)); |
3396 | memcpy(ARR_LBOUND(result), lbs, ndims * sizeof(int)); |
3397 | |
3398 | CopyArrayEls(result, |
3399 | elems, nulls, nelems, |
3400 | elmlen, elmbyval, elmalign, |
3401 | false); |
3402 | |
3403 | return result; |
3404 | } |
3405 | |
3406 | /* |
3407 | * construct_empty_array --- make a zero-dimensional array of given type |
3408 | */ |
3409 | ArrayType * |
3410 | construct_empty_array(Oid elmtype) |
3411 | { |
3412 | ArrayType *result; |
3413 | |
3414 | result = (ArrayType *) palloc0(sizeof(ArrayType)); |
3415 | SET_VARSIZE(result, sizeof(ArrayType)); |
3416 | result->ndim = 0; |
3417 | result->dataoffset = 0; |
3418 | result->elemtype = elmtype; |
3419 | return result; |
3420 | } |
3421 | |
3422 | /* |
3423 | * construct_empty_expanded_array: make an empty expanded array |
3424 | * given only type information. (metacache can be NULL if not needed.) |
3425 | */ |
3426 | ExpandedArrayHeader * |
3427 | construct_empty_expanded_array(Oid element_type, |
3428 | MemoryContext parentcontext, |
3429 | ArrayMetaState *metacache) |
3430 | { |
3431 | ArrayType *array = construct_empty_array(element_type); |
3432 | Datum d; |
3433 | |
3434 | d = expand_array(PointerGetDatum(array), parentcontext, metacache); |
3435 | pfree(array); |
3436 | return (ExpandedArrayHeader *) DatumGetEOHP(d); |
3437 | } |
3438 | |
3439 | /* |
3440 | * deconstruct_array --- simple method for extracting data from an array |
3441 | * |
3442 | * array: array object to examine (must not be NULL) |
3443 | * elmtype, elmlen, elmbyval, elmalign: info for the datatype of the items |
3444 | * elemsp: return value, set to point to palloc'd array of Datum values |
3445 | * nullsp: return value, set to point to palloc'd array of isnull markers |
3446 | * nelemsp: return value, set to number of extracted values |
3447 | * |
3448 | * The caller may pass nullsp == NULL if it does not support NULLs in the |
3449 | * array. Note that this produces a very uninformative error message, |
3450 | * so do it only in cases where a NULL is really not expected. |
3451 | * |
3452 | * If array elements are pass-by-ref data type, the returned Datums will |
3453 | * be pointers into the array object. |
3454 | * |
3455 | * NOTE: it would be cleaner to look up the elmlen/elmbval/elmalign info |
3456 | * from the system catalogs, given the elmtype. However, in most current |
3457 | * uses the type is hard-wired into the caller and so we can save a lookup |
3458 | * cycle by hard-wiring the type info as well. |
3459 | */ |
3460 | void |
3461 | deconstruct_array(ArrayType *array, |
3462 | Oid elmtype, |
3463 | int elmlen, bool elmbyval, char elmalign, |
3464 | Datum **elemsp, bool **nullsp, int *nelemsp) |
3465 | { |
3466 | Datum *elems; |
3467 | bool *nulls; |
3468 | int nelems; |
3469 | char *p; |
3470 | bits8 *bitmap; |
3471 | int bitmask; |
3472 | int i; |
3473 | |
3474 | Assert(ARR_ELEMTYPE(array) == elmtype); |
3475 | |
3476 | nelems = ArrayGetNItems(ARR_NDIM(array), ARR_DIMS(array)); |
3477 | *elemsp = elems = (Datum *) palloc(nelems * sizeof(Datum)); |
3478 | if (nullsp) |
3479 | *nullsp = nulls = (bool *) palloc0(nelems * sizeof(bool)); |
3480 | else |
3481 | nulls = NULL; |
3482 | *nelemsp = nelems; |
3483 | |
3484 | p = ARR_DATA_PTR(array); |
3485 | bitmap = ARR_NULLBITMAP(array); |
3486 | bitmask = 1; |
3487 | |
3488 | for (i = 0; i < nelems; i++) |
3489 | { |
3490 | /* Get source element, checking for NULL */ |
3491 | if (bitmap && (*bitmap & bitmask) == 0) |
3492 | { |
3493 | elems[i] = (Datum) 0; |
3494 | if (nulls) |
3495 | nulls[i] = true; |
3496 | else |
3497 | ereport(ERROR, |
3498 | (errcode(ERRCODE_NULL_VALUE_NOT_ALLOWED), |
3499 | errmsg("null array element not allowed in this context" ))); |
3500 | } |
3501 | else |
3502 | { |
3503 | elems[i] = fetch_att(p, elmbyval, elmlen); |
3504 | p = att_addlength_pointer(p, elmlen, p); |
3505 | p = (char *) att_align_nominal(p, elmalign); |
3506 | } |
3507 | |
3508 | /* advance bitmap pointer if any */ |
3509 | if (bitmap) |
3510 | { |
3511 | bitmask <<= 1; |
3512 | if (bitmask == 0x100) |
3513 | { |
3514 | bitmap++; |
3515 | bitmask = 1; |
3516 | } |
3517 | } |
3518 | } |
3519 | } |
3520 | |
3521 | /* |
3522 | * array_contains_nulls --- detect whether an array has any null elements |
3523 | * |
3524 | * This gives an accurate answer, whereas testing ARR_HASNULL only tells |
3525 | * if the array *might* contain a null. |
3526 | */ |
3527 | bool |
3528 | array_contains_nulls(ArrayType *array) |
3529 | { |
3530 | int nelems; |
3531 | bits8 *bitmap; |
3532 | int bitmask; |
3533 | |
3534 | /* Easy answer if there's no null bitmap */ |
3535 | if (!ARR_HASNULL(array)) |
3536 | return false; |
3537 | |
3538 | nelems = ArrayGetNItems(ARR_NDIM(array), ARR_DIMS(array)); |
3539 | |
3540 | bitmap = ARR_NULLBITMAP(array); |
3541 | |
3542 | /* check whole bytes of the bitmap byte-at-a-time */ |
3543 | while (nelems >= 8) |
3544 | { |
3545 | if (*bitmap != 0xFF) |
3546 | return true; |
3547 | bitmap++; |
3548 | nelems -= 8; |
3549 | } |
3550 | |
3551 | /* check last partial byte */ |
3552 | bitmask = 1; |
3553 | while (nelems > 0) |
3554 | { |
3555 | if ((*bitmap & bitmask) == 0) |
3556 | return true; |
3557 | bitmask <<= 1; |
3558 | nelems--; |
3559 | } |
3560 | |
3561 | return false; |
3562 | } |
3563 | |
3564 | |
3565 | /* |
3566 | * array_eq : |
3567 | * compares two arrays for equality |
3568 | * result : |
3569 | * returns true if the arrays are equal, false otherwise. |
3570 | * |
3571 | * Note: we do not use array_cmp here, since equality may be meaningful in |
3572 | * datatypes that don't have a total ordering (and hence no btree support). |
3573 | */ |
3574 | Datum |
3575 | array_eq(PG_FUNCTION_ARGS) |
3576 | { |
3577 | LOCAL_FCINFO(locfcinfo, 2); |
3578 | AnyArrayType *array1 = PG_GETARG_ANY_ARRAY_P(0); |
3579 | AnyArrayType *array2 = PG_GETARG_ANY_ARRAY_P(1); |
3580 | Oid collation = PG_GET_COLLATION(); |
3581 | int ndims1 = AARR_NDIM(array1); |
3582 | int ndims2 = AARR_NDIM(array2); |
3583 | int *dims1 = AARR_DIMS(array1); |
3584 | int *dims2 = AARR_DIMS(array2); |
3585 | int *lbs1 = AARR_LBOUND(array1); |
3586 | int *lbs2 = AARR_LBOUND(array2); |
3587 | Oid element_type = AARR_ELEMTYPE(array1); |
3588 | bool result = true; |
3589 | int nitems; |
3590 | TypeCacheEntry *typentry; |
3591 | int typlen; |
3592 | bool typbyval; |
3593 | char typalign; |
3594 | array_iter it1; |
3595 | array_iter it2; |
3596 | int i; |
3597 | |
3598 | if (element_type != AARR_ELEMTYPE(array2)) |
3599 | ereport(ERROR, |
3600 | (errcode(ERRCODE_DATATYPE_MISMATCH), |
3601 | errmsg("cannot compare arrays of different element types" ))); |
3602 | |
3603 | /* fast path if the arrays do not have the same dimensionality */ |
3604 | if (ndims1 != ndims2 || |
3605 | memcmp(dims1, dims2, ndims1 * sizeof(int)) != 0 || |
3606 | memcmp(lbs1, lbs2, ndims1 * sizeof(int)) != 0) |
3607 | result = false; |
3608 | else |
3609 | { |
3610 | /* |
3611 | * We arrange to look up the equality function only once per series of |
3612 | * calls, assuming the element type doesn't change underneath us. The |
3613 | * typcache is used so that we have no memory leakage when being used |
3614 | * as an index support function. |
3615 | */ |
3616 | typentry = (TypeCacheEntry *) fcinfo->flinfo->fn_extra; |
3617 | if (typentry == NULL || |
3618 | typentry->type_id != element_type) |
3619 | { |
3620 | typentry = lookup_type_cache(element_type, |
3621 | TYPECACHE_EQ_OPR_FINFO); |
3622 | if (!OidIsValid(typentry->eq_opr_finfo.fn_oid)) |
3623 | ereport(ERROR, |
3624 | (errcode(ERRCODE_UNDEFINED_FUNCTION), |
3625 | errmsg("could not identify an equality operator for type %s" , |
3626 | format_type_be(element_type)))); |
3627 | fcinfo->flinfo->fn_extra = (void *) typentry; |
3628 | } |
3629 | typlen = typentry->typlen; |
3630 | typbyval = typentry->typbyval; |
3631 | typalign = typentry->typalign; |
3632 | |
3633 | /* |
3634 | * apply the operator to each pair of array elements. |
3635 | */ |
3636 | InitFunctionCallInfoData(*locfcinfo, &typentry->eq_opr_finfo, 2, |
3637 | collation, NULL, NULL); |
3638 | |
3639 | /* Loop over source data */ |
3640 | nitems = ArrayGetNItems(ndims1, dims1); |
3641 | array_iter_setup(&it1, array1); |
3642 | array_iter_setup(&it2, array2); |
3643 | |
3644 | for (i = 0; i < nitems; i++) |
3645 | { |
3646 | Datum elt1; |
3647 | Datum elt2; |
3648 | bool isnull1; |
3649 | bool isnull2; |
3650 | bool oprresult; |
3651 | |
3652 | /* Get elements, checking for NULL */ |
3653 | elt1 = array_iter_next(&it1, &isnull1, i, |
3654 | typlen, typbyval, typalign); |
3655 | elt2 = array_iter_next(&it2, &isnull2, i, |
3656 | typlen, typbyval, typalign); |
3657 | |
3658 | /* |
3659 | * We consider two NULLs equal; NULL and not-NULL are unequal. |
3660 | */ |
3661 | if (isnull1 && isnull2) |
3662 | continue; |
3663 | if (isnull1 || isnull2) |
3664 | { |
3665 | result = false; |
3666 | break; |
3667 | } |
3668 | |
3669 | /* |
3670 | * Apply the operator to the element pair |
3671 | */ |
3672 | locfcinfo->args[0].value = elt1; |
3673 | locfcinfo->args[0].isnull = false; |
3674 | locfcinfo->args[1].value = elt2; |
3675 | locfcinfo->args[1].isnull = false; |
3676 | locfcinfo->isnull = false; |
3677 | oprresult = DatumGetBool(FunctionCallInvoke(locfcinfo)); |
3678 | if (!oprresult) |
3679 | { |
3680 | result = false; |
3681 | break; |
3682 | } |
3683 | } |
3684 | } |
3685 | |
3686 | /* Avoid leaking memory when handed toasted input. */ |
3687 | AARR_FREE_IF_COPY(array1, 0); |
3688 | AARR_FREE_IF_COPY(array2, 1); |
3689 | |
3690 | PG_RETURN_BOOL(result); |
3691 | } |
3692 | |
3693 | |
3694 | /*----------------------------------------------------------------------------- |
3695 | * array-array bool operators: |
3696 | * Given two arrays, iterate comparison operators |
3697 | * over the array. Uses logic similar to text comparison |
3698 | * functions, except element-by-element instead of |
3699 | * character-by-character. |
3700 | *---------------------------------------------------------------------------- |
3701 | */ |
3702 | |
3703 | Datum |
3704 | array_ne(PG_FUNCTION_ARGS) |
3705 | { |
3706 | PG_RETURN_BOOL(!DatumGetBool(array_eq(fcinfo))); |
3707 | } |
3708 | |
3709 | Datum |
3710 | array_lt(PG_FUNCTION_ARGS) |
3711 | { |
3712 | PG_RETURN_BOOL(array_cmp(fcinfo) < 0); |
3713 | } |
3714 | |
3715 | Datum |
3716 | array_gt(PG_FUNCTION_ARGS) |
3717 | { |
3718 | PG_RETURN_BOOL(array_cmp(fcinfo) > 0); |
3719 | } |
3720 | |
3721 | Datum |
3722 | array_le(PG_FUNCTION_ARGS) |
3723 | { |
3724 | PG_RETURN_BOOL(array_cmp(fcinfo) <= 0); |
3725 | } |
3726 | |
3727 | Datum |
3728 | array_ge(PG_FUNCTION_ARGS) |
3729 | { |
3730 | PG_RETURN_BOOL(array_cmp(fcinfo) >= 0); |
3731 | } |
3732 | |
3733 | Datum |
3734 | btarraycmp(PG_FUNCTION_ARGS) |
3735 | { |
3736 | PG_RETURN_INT32(array_cmp(fcinfo)); |
3737 | } |
3738 | |
3739 | /* |
3740 | * array_cmp() |
3741 | * Internal comparison function for arrays. |
3742 | * |
3743 | * Returns -1, 0 or 1 |
3744 | */ |
3745 | static int |
3746 | array_cmp(FunctionCallInfo fcinfo) |
3747 | { |
3748 | LOCAL_FCINFO(locfcinfo, 2); |
3749 | AnyArrayType *array1 = PG_GETARG_ANY_ARRAY_P(0); |
3750 | AnyArrayType *array2 = PG_GETARG_ANY_ARRAY_P(1); |
3751 | Oid collation = PG_GET_COLLATION(); |
3752 | int ndims1 = AARR_NDIM(array1); |
3753 | int ndims2 = AARR_NDIM(array2); |
3754 | int *dims1 = AARR_DIMS(array1); |
3755 | int *dims2 = AARR_DIMS(array2); |
3756 | int nitems1 = ArrayGetNItems(ndims1, dims1); |
3757 | int nitems2 = ArrayGetNItems(ndims2, dims2); |
3758 | Oid element_type = AARR_ELEMTYPE(array1); |
3759 | int result = 0; |
3760 | TypeCacheEntry *typentry; |
3761 | int typlen; |
3762 | bool typbyval; |
3763 | char typalign; |
3764 | int min_nitems; |
3765 | array_iter it1; |
3766 | array_iter it2; |
3767 | int i; |
3768 | |
3769 | if (element_type != AARR_ELEMTYPE(array2)) |
3770 | ereport(ERROR, |
3771 | (errcode(ERRCODE_DATATYPE_MISMATCH), |
3772 | errmsg("cannot compare arrays of different element types" ))); |
3773 | |
3774 | /* |
3775 | * We arrange to look up the comparison function only once per series of |
3776 | * calls, assuming the element type doesn't change underneath us. The |
3777 | * typcache is used so that we have no memory leakage when being used as |
3778 | * an index support function. |
3779 | */ |
3780 | typentry = (TypeCacheEntry *) fcinfo->flinfo->fn_extra; |
3781 | if (typentry == NULL || |
3782 | typentry->type_id != element_type) |
3783 | { |
3784 | typentry = lookup_type_cache(element_type, |
3785 | TYPECACHE_CMP_PROC_FINFO); |
3786 | if (!OidIsValid(typentry->cmp_proc_finfo.fn_oid)) |
3787 | ereport(ERROR, |
3788 | (errcode(ERRCODE_UNDEFINED_FUNCTION), |
3789 | errmsg("could not identify a comparison function for type %s" , |
3790 | format_type_be(element_type)))); |
3791 | fcinfo->flinfo->fn_extra = (void *) typentry; |
3792 | } |
3793 | typlen = typentry->typlen; |
3794 | typbyval = typentry->typbyval; |
3795 | typalign = typentry->typalign; |
3796 | |
3797 | /* |
3798 | * apply the operator to each pair of array elements. |
3799 | */ |
3800 | InitFunctionCallInfoData(*locfcinfo, &typentry->cmp_proc_finfo, 2, |
3801 | collation, NULL, NULL); |
3802 | |
3803 | /* Loop over source data */ |
3804 | min_nitems = Min(nitems1, nitems2); |
3805 | array_iter_setup(&it1, array1); |
3806 | array_iter_setup(&it2, array2); |
3807 | |
3808 | for (i = 0; i < min_nitems; i++) |
3809 | { |
3810 | Datum elt1; |
3811 | Datum elt2; |
3812 | bool isnull1; |
3813 | bool isnull2; |
3814 | int32 cmpresult; |
3815 | |
3816 | /* Get elements, checking for NULL */ |
3817 | elt1 = array_iter_next(&it1, &isnull1, i, typlen, typbyval, typalign); |
3818 | elt2 = array_iter_next(&it2, &isnull2, i, typlen, typbyval, typalign); |
3819 | |
3820 | /* |
3821 | * We consider two NULLs equal; NULL > not-NULL. |
3822 | */ |
3823 | if (isnull1 && isnull2) |
3824 | continue; |
3825 | if (isnull1) |
3826 | { |
3827 | /* arg1 is greater than arg2 */ |
3828 | result = 1; |
3829 | break; |
3830 | } |
3831 | if (isnull2) |
3832 | { |
3833 | /* arg1 is less than arg2 */ |
3834 | result = -1; |
3835 | break; |
3836 | } |
3837 | |
3838 | /* Compare the pair of elements */ |
3839 | locfcinfo->args[0].value = elt1; |
3840 | locfcinfo->args[0].isnull = false; |
3841 | locfcinfo->args[1].value = elt2; |
3842 | locfcinfo->args[1].isnull = false; |
3843 | locfcinfo->isnull = false; |
3844 | cmpresult = DatumGetInt32(FunctionCallInvoke(locfcinfo)); |
3845 | |
3846 | if (cmpresult == 0) |
3847 | continue; /* equal */ |
3848 | |
3849 | if (cmpresult < 0) |
3850 | { |
3851 | /* arg1 is less than arg2 */ |
3852 | result = -1; |
3853 | break; |
3854 | } |
3855 | else |
3856 | { |
3857 | /* arg1 is greater than arg2 */ |
3858 | result = 1; |
3859 | break; |
3860 | } |
3861 | } |
3862 | |
3863 | /* |
3864 | * If arrays contain same data (up to end of shorter one), apply |
3865 | * additional rules to sort by dimensionality. The relative significance |
3866 | * of the different bits of information is historical; mainly we just care |
3867 | * that we don't say "equal" for arrays of different dimensionality. |
3868 | */ |
3869 | if (result == 0) |
3870 | { |
3871 | if (nitems1 != nitems2) |
3872 | result = (nitems1 < nitems2) ? -1 : 1; |
3873 | else if (ndims1 != ndims2) |
3874 | result = (ndims1 < ndims2) ? -1 : 1; |
3875 | else |
3876 | { |
3877 | for (i = 0; i < ndims1; i++) |
3878 | { |
3879 | if (dims1[i] != dims2[i]) |
3880 | { |
3881 | result = (dims1[i] < dims2[i]) ? -1 : 1; |
3882 | break; |
3883 | } |
3884 | } |
3885 | if (result == 0) |
3886 | { |
3887 | int *lbound1 = AARR_LBOUND(array1); |
3888 | int *lbound2 = AARR_LBOUND(array2); |
3889 | |
3890 | for (i = 0; i < ndims1; i++) |
3891 | { |
3892 | if (lbound1[i] != lbound2[i]) |
3893 | { |
3894 | result = (lbound1[i] < lbound2[i]) ? -1 : 1; |
3895 | break; |
3896 | } |
3897 | } |
3898 | } |
3899 | } |
3900 | } |
3901 | |
3902 | /* Avoid leaking memory when handed toasted input. */ |
3903 | AARR_FREE_IF_COPY(array1, 0); |
3904 | AARR_FREE_IF_COPY(array2, 1); |
3905 | |
3906 | return result; |
3907 | } |
3908 | |
3909 | |
3910 | /*----------------------------------------------------------------------------- |
3911 | * array hashing |
3912 | * Hash the elements and combine the results. |
3913 | *---------------------------------------------------------------------------- |
3914 | */ |
3915 | |
3916 | Datum |
3917 | hash_array(PG_FUNCTION_ARGS) |
3918 | { |
3919 | LOCAL_FCINFO(locfcinfo, 1); |
3920 | AnyArrayType *array = PG_GETARG_ANY_ARRAY_P(0); |
3921 | int ndims = AARR_NDIM(array); |
3922 | int *dims = AARR_DIMS(array); |
3923 | Oid element_type = AARR_ELEMTYPE(array); |
3924 | uint32 result = 1; |
3925 | int nitems; |
3926 | TypeCacheEntry *typentry; |
3927 | int typlen; |
3928 | bool typbyval; |
3929 | char typalign; |
3930 | int i; |
3931 | array_iter iter; |
3932 | |
3933 | /* |
3934 | * We arrange to look up the hash function only once per series of calls, |
3935 | * assuming the element type doesn't change underneath us. The typcache |
3936 | * is used so that we have no memory leakage when being used as an index |
3937 | * support function. |
3938 | */ |
3939 | typentry = (TypeCacheEntry *) fcinfo->flinfo->fn_extra; |
3940 | if (typentry == NULL || |
3941 | typentry->type_id != element_type) |
3942 | { |
3943 | typentry = lookup_type_cache(element_type, |
3944 | TYPECACHE_HASH_PROC_FINFO); |
3945 | if (!OidIsValid(typentry->hash_proc_finfo.fn_oid)) |
3946 | ereport(ERROR, |
3947 | (errcode(ERRCODE_UNDEFINED_FUNCTION), |
3948 | errmsg("could not identify a hash function for type %s" , |
3949 | format_type_be(element_type)))); |
3950 | fcinfo->flinfo->fn_extra = (void *) typentry; |
3951 | } |
3952 | typlen = typentry->typlen; |
3953 | typbyval = typentry->typbyval; |
3954 | typalign = typentry->typalign; |
3955 | |
3956 | /* |
3957 | * apply the hash function to each array element. |
3958 | */ |
3959 | InitFunctionCallInfoData(*locfcinfo, &typentry->hash_proc_finfo, 1, |
3960 | PG_GET_COLLATION(), NULL, NULL); |
3961 | |
3962 | /* Loop over source data */ |
3963 | nitems = ArrayGetNItems(ndims, dims); |
3964 | array_iter_setup(&iter, array); |
3965 | |
3966 | for (i = 0; i < nitems; i++) |
3967 | { |
3968 | Datum elt; |
3969 | bool isnull; |
3970 | uint32 elthash; |
3971 | |
3972 | /* Get element, checking for NULL */ |
3973 | elt = array_iter_next(&iter, &isnull, i, typlen, typbyval, typalign); |
3974 | |
3975 | if (isnull) |
3976 | { |
3977 | /* Treat nulls as having hashvalue 0 */ |
3978 | elthash = 0; |
3979 | } |
3980 | else |
3981 | { |
3982 | /* Apply the hash function */ |
3983 | locfcinfo->args[0].value = elt; |
3984 | locfcinfo->args[0].isnull = false; |
3985 | locfcinfo->isnull = false; |
3986 | elthash = DatumGetUInt32(FunctionCallInvoke(locfcinfo)); |
3987 | } |
3988 | |
3989 | /* |
3990 | * Combine hash values of successive elements by multiplying the |
3991 | * current value by 31 and adding on the new element's hash value. |
3992 | * |
3993 | * The result is a sum in which each element's hash value is |
3994 | * multiplied by a different power of 31. This is modulo 2^32 |
3995 | * arithmetic, and the powers of 31 modulo 2^32 form a cyclic group of |
3996 | * order 2^27. So for arrays of up to 2^27 elements, each element's |
3997 | * hash value is multiplied by a different (odd) number, resulting in |
3998 | * a good mixing of all the elements' hash values. |
3999 | */ |
4000 | result = (result << 5) - result + elthash; |
4001 | } |
4002 | |
4003 | /* Avoid leaking memory when handed toasted input. */ |
4004 | AARR_FREE_IF_COPY(array, 0); |
4005 | |
4006 | PG_RETURN_UINT32(result); |
4007 | } |
4008 | |
4009 | /* |
4010 | * Returns 64-bit value by hashing a value to a 64-bit value, with a seed. |
4011 | * Otherwise, similar to hash_array. |
4012 | */ |
4013 | Datum |
4014 | hash_array_extended(PG_FUNCTION_ARGS) |
4015 | { |
4016 | LOCAL_FCINFO(locfcinfo, 2); |
4017 | AnyArrayType *array = PG_GETARG_ANY_ARRAY_P(0); |
4018 | uint64 seed = PG_GETARG_INT64(1); |
4019 | int ndims = AARR_NDIM(array); |
4020 | int *dims = AARR_DIMS(array); |
4021 | Oid element_type = AARR_ELEMTYPE(array); |
4022 | uint64 result = 1; |
4023 | int nitems; |
4024 | TypeCacheEntry *typentry; |
4025 | int typlen; |
4026 | bool typbyval; |
4027 | char typalign; |
4028 | int i; |
4029 | array_iter iter; |
4030 | |
4031 | typentry = (TypeCacheEntry *) fcinfo->flinfo->fn_extra; |
4032 | if (typentry == NULL || |
4033 | typentry->type_id != element_type) |
4034 | { |
4035 | typentry = lookup_type_cache(element_type, |
4036 | TYPECACHE_HASH_EXTENDED_PROC_FINFO); |
4037 | if (!OidIsValid(typentry->hash_extended_proc_finfo.fn_oid)) |
4038 | ereport(ERROR, |
4039 | (errcode(ERRCODE_UNDEFINED_FUNCTION), |
4040 | errmsg("could not identify an extended hash function for type %s" , |
4041 | format_type_be(element_type)))); |
4042 | fcinfo->flinfo->fn_extra = (void *) typentry; |
4043 | } |
4044 | typlen = typentry->typlen; |
4045 | typbyval = typentry->typbyval; |
4046 | typalign = typentry->typalign; |
4047 | |
4048 | InitFunctionCallInfoData(*locfcinfo, &typentry->hash_extended_proc_finfo, 2, |
4049 | InvalidOid, NULL, NULL); |
4050 | |
4051 | /* Loop over source data */ |
4052 | nitems = ArrayGetNItems(ndims, dims); |
4053 | array_iter_setup(&iter, array); |
4054 | |
4055 | for (i = 0; i < nitems; i++) |
4056 | { |
4057 | Datum elt; |
4058 | bool isnull; |
4059 | uint64 elthash; |
4060 | |
4061 | /* Get element, checking for NULL */ |
4062 | elt = array_iter_next(&iter, &isnull, i, typlen, typbyval, typalign); |
4063 | |
4064 | if (isnull) |
4065 | { |
4066 | elthash = 0; |
4067 | } |
4068 | else |
4069 | { |
4070 | /* Apply the hash function */ |
4071 | locfcinfo->args[0].value = elt; |
4072 | locfcinfo->args[0].isnull = false; |
4073 | locfcinfo->args[1].value = Int64GetDatum(seed); |
4074 | locfcinfo->args[1].isnull = false; |
4075 | elthash = DatumGetUInt64(FunctionCallInvoke(locfcinfo)); |
4076 | } |
4077 | |
4078 | result = (result << 5) - result + elthash; |
4079 | } |
4080 | |
4081 | AARR_FREE_IF_COPY(array, 0); |
4082 | |
4083 | PG_RETURN_UINT64(result); |
4084 | } |
4085 | |
4086 | |
4087 | /*----------------------------------------------------------------------------- |
4088 | * array overlap/containment comparisons |
4089 | * These use the same methods of comparing array elements as array_eq. |
4090 | * We consider only the elements of the arrays, ignoring dimensionality. |
4091 | *---------------------------------------------------------------------------- |
4092 | */ |
4093 | |
4094 | /* |
4095 | * array_contain_compare : |
4096 | * compares two arrays for overlap/containment |
4097 | * |
4098 | * When matchall is true, return true if all members of array1 are in array2. |
4099 | * When matchall is false, return true if any members of array1 are in array2. |
4100 | */ |
4101 | static bool |
4102 | array_contain_compare(AnyArrayType *array1, AnyArrayType *array2, Oid collation, |
4103 | bool matchall, void **) |
4104 | { |
4105 | LOCAL_FCINFO(locfcinfo, 2); |
4106 | bool result = matchall; |
4107 | Oid element_type = AARR_ELEMTYPE(array1); |
4108 | TypeCacheEntry *typentry; |
4109 | int nelems1; |
4110 | Datum *values2; |
4111 | bool *nulls2; |
4112 | int nelems2; |
4113 | int typlen; |
4114 | bool typbyval; |
4115 | char typalign; |
4116 | int i; |
4117 | int j; |
4118 | array_iter it1; |
4119 | |
4120 | if (element_type != AARR_ELEMTYPE(array2)) |
4121 | ereport(ERROR, |
4122 | (errcode(ERRCODE_DATATYPE_MISMATCH), |
4123 | errmsg("cannot compare arrays of different element types" ))); |
4124 | |
4125 | /* |
4126 | * We arrange to look up the equality function only once per series of |
4127 | * calls, assuming the element type doesn't change underneath us. The |
4128 | * typcache is used so that we have no memory leakage when being used as |
4129 | * an index support function. |
4130 | */ |
4131 | typentry = (TypeCacheEntry *) *fn_extra; |
4132 | if (typentry == NULL || |
4133 | typentry->type_id != element_type) |
4134 | { |
4135 | typentry = lookup_type_cache(element_type, |
4136 | TYPECACHE_EQ_OPR_FINFO); |
4137 | if (!OidIsValid(typentry->eq_opr_finfo.fn_oid)) |
4138 | ereport(ERROR, |
4139 | (errcode(ERRCODE_UNDEFINED_FUNCTION), |
4140 | errmsg("could not identify an equality operator for type %s" , |
4141 | format_type_be(element_type)))); |
4142 | *fn_extra = (void *) typentry; |
4143 | } |
4144 | typlen = typentry->typlen; |
4145 | typbyval = typentry->typbyval; |
4146 | typalign = typentry->typalign; |
4147 | |
4148 | /* |
4149 | * Since we probably will need to scan array2 multiple times, it's |
4150 | * worthwhile to use deconstruct_array on it. We scan array1 the hard way |
4151 | * however, since we very likely won't need to look at all of it. |
4152 | */ |
4153 | if (VARATT_IS_EXPANDED_HEADER(array2)) |
4154 | { |
4155 | /* This should be safe even if input is read-only */ |
4156 | deconstruct_expanded_array(&(array2->xpn)); |
4157 | values2 = array2->xpn.dvalues; |
4158 | nulls2 = array2->xpn.dnulls; |
4159 | nelems2 = array2->xpn.nelems; |
4160 | } |
4161 | else |
4162 | deconstruct_array((ArrayType *) array2, |
4163 | element_type, typlen, typbyval, typalign, |
4164 | &values2, &nulls2, &nelems2); |
4165 | |
4166 | /* |
4167 | * Apply the comparison operator to each pair of array elements. |
4168 | */ |
4169 | InitFunctionCallInfoData(*locfcinfo, &typentry->eq_opr_finfo, 2, |
4170 | collation, NULL, NULL); |
4171 | |
4172 | /* Loop over source data */ |
4173 | nelems1 = ArrayGetNItems(AARR_NDIM(array1), AARR_DIMS(array1)); |
4174 | array_iter_setup(&it1, array1); |
4175 | |
4176 | for (i = 0; i < nelems1; i++) |
4177 | { |
4178 | Datum elt1; |
4179 | bool isnull1; |
4180 | |
4181 | /* Get element, checking for NULL */ |
4182 | elt1 = array_iter_next(&it1, &isnull1, i, typlen, typbyval, typalign); |
4183 | |
4184 | /* |
4185 | * We assume that the comparison operator is strict, so a NULL can't |
4186 | * match anything. XXX this diverges from the "NULL=NULL" behavior of |
4187 | * array_eq, should we act like that? |
4188 | */ |
4189 | if (isnull1) |
4190 | { |
4191 | if (matchall) |
4192 | { |
4193 | result = false; |
4194 | break; |
4195 | } |
4196 | continue; |
4197 | } |
4198 | |
4199 | for (j = 0; j < nelems2; j++) |
4200 | { |
4201 | Datum elt2 = values2[j]; |
4202 | bool isnull2 = nulls2 ? nulls2[j] : false; |
4203 | bool oprresult; |
4204 | |
4205 | if (isnull2) |
4206 | continue; /* can't match */ |
4207 | |
4208 | /* |
4209 | * Apply the operator to the element pair |
4210 | */ |
4211 | locfcinfo->args[0].value = elt1; |
4212 | locfcinfo->args[0].isnull = false; |
4213 | locfcinfo->args[1].value = elt2; |
4214 | locfcinfo->args[1].isnull = false; |
4215 | locfcinfo->isnull = false; |
4216 | oprresult = DatumGetBool(FunctionCallInvoke(locfcinfo)); |
4217 | if (oprresult) |
4218 | break; |
4219 | } |
4220 | |
4221 | if (j < nelems2) |
4222 | { |
4223 | /* found a match for elt1 */ |
4224 | if (!matchall) |
4225 | { |
4226 | result = true; |
4227 | break; |
4228 | } |
4229 | } |
4230 | else |
4231 | { |
4232 | /* no match for elt1 */ |
4233 | if (matchall) |
4234 | { |
4235 | result = false; |
4236 | break; |
4237 | } |
4238 | } |
4239 | } |
4240 | |
4241 | return result; |
4242 | } |
4243 | |
4244 | Datum |
4245 | arrayoverlap(PG_FUNCTION_ARGS) |
4246 | { |
4247 | AnyArrayType *array1 = PG_GETARG_ANY_ARRAY_P(0); |
4248 | AnyArrayType *array2 = PG_GETARG_ANY_ARRAY_P(1); |
4249 | Oid collation = PG_GET_COLLATION(); |
4250 | bool result; |
4251 | |
4252 | result = array_contain_compare(array1, array2, collation, false, |
4253 | &fcinfo->flinfo->fn_extra); |
4254 | |
4255 | /* Avoid leaking memory when handed toasted input. */ |
4256 | AARR_FREE_IF_COPY(array1, 0); |
4257 | AARR_FREE_IF_COPY(array2, 1); |
4258 | |
4259 | PG_RETURN_BOOL(result); |
4260 | } |
4261 | |
4262 | Datum |
4263 | arraycontains(PG_FUNCTION_ARGS) |
4264 | { |
4265 | AnyArrayType *array1 = PG_GETARG_ANY_ARRAY_P(0); |
4266 | AnyArrayType *array2 = PG_GETARG_ANY_ARRAY_P(1); |
4267 | Oid collation = PG_GET_COLLATION(); |
4268 | bool result; |
4269 | |
4270 | result = array_contain_compare(array2, array1, collation, true, |
4271 | &fcinfo->flinfo->fn_extra); |
4272 | |
4273 | /* Avoid leaking memory when handed toasted input. */ |
4274 | AARR_FREE_IF_COPY(array1, 0); |
4275 | AARR_FREE_IF_COPY(array2, 1); |
4276 | |
4277 | PG_RETURN_BOOL(result); |
4278 | } |
4279 | |
4280 | Datum |
4281 | arraycontained(PG_FUNCTION_ARGS) |
4282 | { |
4283 | AnyArrayType *array1 = PG_GETARG_ANY_ARRAY_P(0); |
4284 | AnyArrayType *array2 = PG_GETARG_ANY_ARRAY_P(1); |
4285 | Oid collation = PG_GET_COLLATION(); |
4286 | bool result; |
4287 | |
4288 | result = array_contain_compare(array1, array2, collation, true, |
4289 | &fcinfo->flinfo->fn_extra); |
4290 | |
4291 | /* Avoid leaking memory when handed toasted input. */ |
4292 | AARR_FREE_IF_COPY(array1, 0); |
4293 | AARR_FREE_IF_COPY(array2, 1); |
4294 | |
4295 | PG_RETURN_BOOL(result); |
4296 | } |
4297 | |
4298 | |
4299 | /*----------------------------------------------------------------------------- |
4300 | * Array iteration functions |
4301 | * These functions are used to iterate efficiently through arrays |
4302 | *----------------------------------------------------------------------------- |
4303 | */ |
4304 | |
4305 | /* |
4306 | * array_create_iterator --- set up to iterate through an array |
4307 | * |
4308 | * If slice_ndim is zero, we will iterate element-by-element; the returned |
4309 | * datums are of the array's element type. |
4310 | * |
4311 | * If slice_ndim is 1..ARR_NDIM(arr), we will iterate by slices: the |
4312 | * returned datums are of the same array type as 'arr', but of size |
4313 | * equal to the rightmost N dimensions of 'arr'. |
4314 | * |
4315 | * The passed-in array must remain valid for the lifetime of the iterator. |
4316 | */ |
4317 | ArrayIterator |
4318 | array_create_iterator(ArrayType *arr, int slice_ndim, ArrayMetaState *mstate) |
4319 | { |
4320 | ArrayIterator iterator = palloc0(sizeof(ArrayIteratorData)); |
4321 | |
4322 | /* |
4323 | * Sanity-check inputs --- caller should have got this right already |
4324 | */ |
4325 | Assert(PointerIsValid(arr)); |
4326 | if (slice_ndim < 0 || slice_ndim > ARR_NDIM(arr)) |
4327 | elog(ERROR, "invalid arguments to array_create_iterator" ); |
4328 | |
4329 | /* |
4330 | * Remember basic info about the array and its element type |
4331 | */ |
4332 | iterator->arr = arr; |
4333 | iterator->nullbitmap = ARR_NULLBITMAP(arr); |
4334 | iterator->nitems = ArrayGetNItems(ARR_NDIM(arr), ARR_DIMS(arr)); |
4335 | |
4336 | if (mstate != NULL) |
4337 | { |
4338 | Assert(mstate->element_type == ARR_ELEMTYPE(arr)); |
4339 | |
4340 | iterator->typlen = mstate->typlen; |
4341 | iterator->typbyval = mstate->typbyval; |
4342 | iterator->typalign = mstate->typalign; |
4343 | } |
4344 | else |
4345 | get_typlenbyvalalign(ARR_ELEMTYPE(arr), |
4346 | &iterator->typlen, |
4347 | &iterator->typbyval, |
4348 | &iterator->typalign); |
4349 | |
4350 | /* |
4351 | * Remember the slicing parameters. |
4352 | */ |
4353 | iterator->slice_ndim = slice_ndim; |
4354 | |
4355 | if (slice_ndim > 0) |
4356 | { |
4357 | /* |
4358 | * Get pointers into the array's dims and lbound arrays to represent |
4359 | * the dims/lbound arrays of a slice. These are the same as the |
4360 | * rightmost N dimensions of the array. |
4361 | */ |
4362 | iterator->slice_dims = ARR_DIMS(arr) + ARR_NDIM(arr) - slice_ndim; |
4363 | iterator->slice_lbound = ARR_LBOUND(arr) + ARR_NDIM(arr) - slice_ndim; |
4364 | |
4365 | /* |
4366 | * Compute number of elements in a slice. |
4367 | */ |
4368 | iterator->slice_len = ArrayGetNItems(slice_ndim, |
4369 | iterator->slice_dims); |
4370 | |
4371 | /* |
4372 | * Create workspace for building sub-arrays. |
4373 | */ |
4374 | iterator->slice_values = (Datum *) |
4375 | palloc(iterator->slice_len * sizeof(Datum)); |
4376 | iterator->slice_nulls = (bool *) |
4377 | palloc(iterator->slice_len * sizeof(bool)); |
4378 | } |
4379 | |
4380 | /* |
4381 | * Initialize our data pointer and linear element number. These will |
4382 | * advance through the array during array_iterate(). |
4383 | */ |
4384 | iterator->data_ptr = ARR_DATA_PTR(arr); |
4385 | iterator->current_item = 0; |
4386 | |
4387 | return iterator; |
4388 | } |
4389 | |
4390 | /* |
4391 | * Iterate through the array referenced by 'iterator'. |
4392 | * |
4393 | * As long as there is another element (or slice), return it into |
4394 | * *value / *isnull, and return true. Return false when no more data. |
4395 | */ |
4396 | bool |
4397 | array_iterate(ArrayIterator iterator, Datum *value, bool *isnull) |
4398 | { |
4399 | /* Done if we have reached the end of the array */ |
4400 | if (iterator->current_item >= iterator->nitems) |
4401 | return false; |
4402 | |
4403 | if (iterator->slice_ndim == 0) |
4404 | { |
4405 | /* |
4406 | * Scalar case: return one element. |
4407 | */ |
4408 | if (array_get_isnull(iterator->nullbitmap, iterator->current_item++)) |
4409 | { |
4410 | *isnull = true; |
4411 | *value = (Datum) 0; |
4412 | } |
4413 | else |
4414 | { |
4415 | /* non-NULL, so fetch the individual Datum to return */ |
4416 | char *p = iterator->data_ptr; |
4417 | |
4418 | *isnull = false; |
4419 | *value = fetch_att(p, iterator->typbyval, iterator->typlen); |
4420 | |
4421 | /* Move our data pointer forward to the next element */ |
4422 | p = att_addlength_pointer(p, iterator->typlen, p); |
4423 | p = (char *) att_align_nominal(p, iterator->typalign); |
4424 | iterator->data_ptr = p; |
4425 | } |
4426 | } |
4427 | else |
4428 | { |
4429 | /* |
4430 | * Slice case: build and return an array of the requested size. |
4431 | */ |
4432 | ArrayType *result; |
4433 | Datum *values = iterator->slice_values; |
4434 | bool *nulls = iterator->slice_nulls; |
4435 | char *p = iterator->data_ptr; |
4436 | int i; |
4437 | |
4438 | for (i = 0; i < iterator->slice_len; i++) |
4439 | { |
4440 | if (array_get_isnull(iterator->nullbitmap, |
4441 | iterator->current_item++)) |
4442 | { |
4443 | nulls[i] = true; |
4444 | values[i] = (Datum) 0; |
4445 | } |
4446 | else |
4447 | { |
4448 | nulls[i] = false; |
4449 | values[i] = fetch_att(p, iterator->typbyval, iterator->typlen); |
4450 | |
4451 | /* Move our data pointer forward to the next element */ |
4452 | p = att_addlength_pointer(p, iterator->typlen, p); |
4453 | p = (char *) att_align_nominal(p, iterator->typalign); |
4454 | } |
4455 | } |
4456 | |
4457 | iterator->data_ptr = p; |
4458 | |
4459 | result = construct_md_array(values, |
4460 | nulls, |
4461 | iterator->slice_ndim, |
4462 | iterator->slice_dims, |
4463 | iterator->slice_lbound, |
4464 | ARR_ELEMTYPE(iterator->arr), |
4465 | iterator->typlen, |
4466 | iterator->typbyval, |
4467 | iterator->typalign); |
4468 | |
4469 | *isnull = false; |
4470 | *value = PointerGetDatum(result); |
4471 | } |
4472 | |
4473 | return true; |
4474 | } |
4475 | |
4476 | /* |
4477 | * Release an ArrayIterator data structure |
4478 | */ |
4479 | void |
4480 | array_free_iterator(ArrayIterator iterator) |
4481 | { |
4482 | if (iterator->slice_ndim > 0) |
4483 | { |
4484 | pfree(iterator->slice_values); |
4485 | pfree(iterator->slice_nulls); |
4486 | } |
4487 | pfree(iterator); |
4488 | } |
4489 | |
4490 | |
4491 | /***************************************************************************/ |
4492 | /******************| Support Routines |*****************/ |
4493 | /***************************************************************************/ |
4494 | |
4495 | /* |
4496 | * Check whether a specific array element is NULL |
4497 | * |
4498 | * nullbitmap: pointer to array's null bitmap (NULL if none) |
4499 | * offset: 0-based linear element number of array element |
4500 | */ |
4501 | static bool |
4502 | array_get_isnull(const bits8 *nullbitmap, int offset) |
4503 | { |
4504 | if (nullbitmap == NULL) |
4505 | return false; /* assume not null */ |
4506 | if (nullbitmap[offset / 8] & (1 << (offset % 8))) |
4507 | return false; /* not null */ |
4508 | return true; |
4509 | } |
4510 | |
4511 | /* |
4512 | * Set a specific array element's null-bitmap entry |
4513 | * |
4514 | * nullbitmap: pointer to array's null bitmap (mustn't be NULL) |
4515 | * offset: 0-based linear element number of array element |
4516 | * isNull: null status to set |
4517 | */ |
4518 | static void |
4519 | array_set_isnull(bits8 *nullbitmap, int offset, bool isNull) |
4520 | { |
4521 | int bitmask; |
4522 | |
4523 | nullbitmap += offset / 8; |
4524 | bitmask = 1 << (offset % 8); |
4525 | if (isNull) |
4526 | *nullbitmap &= ~bitmask; |
4527 | else |
4528 | *nullbitmap |= bitmask; |
4529 | } |
4530 | |
4531 | /* |
4532 | * Fetch array element at pointer, converted correctly to a Datum |
4533 | * |
4534 | * Caller must have handled case of NULL element |
4535 | */ |
4536 | static Datum |
4537 | ArrayCast(char *value, bool byval, int len) |
4538 | { |
4539 | return fetch_att(value, byval, len); |
4540 | } |
4541 | |
4542 | /* |
4543 | * Copy datum to *dest and return total space used (including align padding) |
4544 | * |
4545 | * Caller must have handled case of NULL element |
4546 | */ |
4547 | static int |
4548 | ArrayCastAndSet(Datum src, |
4549 | int typlen, |
4550 | bool typbyval, |
4551 | char typalign, |
4552 | char *dest) |
4553 | { |
4554 | int inc; |
4555 | |
4556 | if (typlen > 0) |
4557 | { |
4558 | if (typbyval) |
4559 | store_att_byval(dest, src, typlen); |
4560 | else |
4561 | memmove(dest, DatumGetPointer(src), typlen); |
4562 | inc = att_align_nominal(typlen, typalign); |
4563 | } |
4564 | else |
4565 | { |
4566 | Assert(!typbyval); |
4567 | inc = att_addlength_datum(0, typlen, src); |
4568 | memmove(dest, DatumGetPointer(src), inc); |
4569 | inc = att_align_nominal(inc, typalign); |
4570 | } |
4571 | |
4572 | return inc; |
4573 | } |
4574 | |
4575 | /* |
4576 | * Advance ptr over nitems array elements |
4577 | * |
4578 | * ptr: starting location in array |
4579 | * offset: 0-based linear element number of first element (the one at *ptr) |
4580 | * nullbitmap: start of array's null bitmap, or NULL if none |
4581 | * nitems: number of array elements to advance over (>= 0) |
4582 | * typlen, typbyval, typalign: storage parameters of array element datatype |
4583 | * |
4584 | * It is caller's responsibility to ensure that nitems is within range |
4585 | */ |
4586 | static char * |
4587 | array_seek(char *ptr, int offset, bits8 *nullbitmap, int nitems, |
4588 | int typlen, bool typbyval, char typalign) |
4589 | { |
4590 | int bitmask; |
4591 | int i; |
4592 | |
4593 | /* easy if fixed-size elements and no NULLs */ |
4594 | if (typlen > 0 && !nullbitmap) |
4595 | return ptr + nitems * ((Size) att_align_nominal(typlen, typalign)); |
4596 | |
4597 | /* seems worth having separate loops for NULL and no-NULLs cases */ |
4598 | if (nullbitmap) |
4599 | { |
4600 | nullbitmap += offset / 8; |
4601 | bitmask = 1 << (offset % 8); |
4602 | |
4603 | for (i = 0; i < nitems; i++) |
4604 | { |
4605 | if (*nullbitmap & bitmask) |
4606 | { |
4607 | ptr = att_addlength_pointer(ptr, typlen, ptr); |
4608 | ptr = (char *) att_align_nominal(ptr, typalign); |
4609 | } |
4610 | bitmask <<= 1; |
4611 | if (bitmask == 0x100) |
4612 | { |
4613 | nullbitmap++; |
4614 | bitmask = 1; |
4615 | } |
4616 | } |
4617 | } |
4618 | else |
4619 | { |
4620 | for (i = 0; i < nitems; i++) |
4621 | { |
4622 | ptr = att_addlength_pointer(ptr, typlen, ptr); |
4623 | ptr = (char *) att_align_nominal(ptr, typalign); |
4624 | } |
4625 | } |
4626 | return ptr; |
4627 | } |
4628 | |
4629 | /* |
4630 | * Compute total size of the nitems array elements starting at *ptr |
4631 | * |
4632 | * Parameters same as for array_seek |
4633 | */ |
4634 | static int |
4635 | array_nelems_size(char *ptr, int offset, bits8 *nullbitmap, int nitems, |
4636 | int typlen, bool typbyval, char typalign) |
4637 | { |
4638 | return array_seek(ptr, offset, nullbitmap, nitems, |
4639 | typlen, typbyval, typalign) - ptr; |
4640 | } |
4641 | |
4642 | /* |
4643 | * Copy nitems array elements from srcptr to destptr |
4644 | * |
4645 | * destptr: starting destination location (must be enough room!) |
4646 | * nitems: number of array elements to copy (>= 0) |
4647 | * srcptr: starting location in source array |
4648 | * offset: 0-based linear element number of first element (the one at *srcptr) |
4649 | * nullbitmap: start of source array's null bitmap, or NULL if none |
4650 | * typlen, typbyval, typalign: storage parameters of array element datatype |
4651 | * |
4652 | * Returns number of bytes copied |
4653 | * |
4654 | * NB: this does not take care of setting up the destination's null bitmap! |
4655 | */ |
4656 | static int |
4657 | array_copy(char *destptr, int nitems, |
4658 | char *srcptr, int offset, bits8 *nullbitmap, |
4659 | int typlen, bool typbyval, char typalign) |
4660 | { |
4661 | int numbytes; |
4662 | |
4663 | numbytes = array_nelems_size(srcptr, offset, nullbitmap, nitems, |
4664 | typlen, typbyval, typalign); |
4665 | memcpy(destptr, srcptr, numbytes); |
4666 | return numbytes; |
4667 | } |
4668 | |
4669 | /* |
4670 | * Copy nitems null-bitmap bits from source to destination |
4671 | * |
4672 | * destbitmap: start of destination array's null bitmap (mustn't be NULL) |
4673 | * destoffset: 0-based linear element number of first dest element |
4674 | * srcbitmap: start of source array's null bitmap, or NULL if none |
4675 | * srcoffset: 0-based linear element number of first source element |
4676 | * nitems: number of bits to copy (>= 0) |
4677 | * |
4678 | * If srcbitmap is NULL then we assume the source is all-non-NULL and |
4679 | * fill 1's into the destination bitmap. Note that only the specified |
4680 | * bits in the destination map are changed, not any before or after. |
4681 | * |
4682 | * Note: this could certainly be optimized using standard bitblt methods. |
4683 | * However, it's not clear that the typical Postgres array has enough elements |
4684 | * to make it worth worrying too much. For the moment, KISS. |
4685 | */ |
4686 | void |
4687 | array_bitmap_copy(bits8 *destbitmap, int destoffset, |
4688 | const bits8 *srcbitmap, int srcoffset, |
4689 | int nitems) |
4690 | { |
4691 | int destbitmask, |
4692 | destbitval, |
4693 | srcbitmask, |
4694 | srcbitval; |
4695 | |
4696 | Assert(destbitmap); |
4697 | if (nitems <= 0) |
4698 | return; /* don't risk fetch off end of memory */ |
4699 | destbitmap += destoffset / 8; |
4700 | destbitmask = 1 << (destoffset % 8); |
4701 | destbitval = *destbitmap; |
4702 | if (srcbitmap) |
4703 | { |
4704 | srcbitmap += srcoffset / 8; |
4705 | srcbitmask = 1 << (srcoffset % 8); |
4706 | srcbitval = *srcbitmap; |
4707 | while (nitems-- > 0) |
4708 | { |
4709 | if (srcbitval & srcbitmask) |
4710 | destbitval |= destbitmask; |
4711 | else |
4712 | destbitval &= ~destbitmask; |
4713 | destbitmask <<= 1; |
4714 | if (destbitmask == 0x100) |
4715 | { |
4716 | *destbitmap++ = destbitval; |
4717 | destbitmask = 1; |
4718 | if (nitems > 0) |
4719 | destbitval = *destbitmap; |
4720 | } |
4721 | srcbitmask <<= 1; |
4722 | if (srcbitmask == 0x100) |
4723 | { |
4724 | srcbitmap++; |
4725 | srcbitmask = 1; |
4726 | if (nitems > 0) |
4727 | srcbitval = *srcbitmap; |
4728 | } |
4729 | } |
4730 | if (destbitmask != 1) |
4731 | *destbitmap = destbitval; |
4732 | } |
4733 | else |
4734 | { |
4735 | while (nitems-- > 0) |
4736 | { |
4737 | destbitval |= destbitmask; |
4738 | destbitmask <<= 1; |
4739 | if (destbitmask == 0x100) |
4740 | { |
4741 | *destbitmap++ = destbitval; |
4742 | destbitmask = 1; |
4743 | if (nitems > 0) |
4744 | destbitval = *destbitmap; |
4745 | } |
4746 | } |
4747 | if (destbitmask != 1) |
4748 | *destbitmap = destbitval; |
4749 | } |
4750 | } |
4751 | |
4752 | /* |
4753 | * Compute space needed for a slice of an array |
4754 | * |
4755 | * We assume the caller has verified that the slice coordinates are valid. |
4756 | */ |
4757 | static int |
4758 | array_slice_size(char *arraydataptr, bits8 *arraynullsptr, |
4759 | int ndim, int *dim, int *lb, |
4760 | int *st, int *endp, |
4761 | int typlen, bool typbyval, char typalign) |
4762 | { |
4763 | int src_offset, |
4764 | span[MAXDIM], |
4765 | prod[MAXDIM], |
4766 | dist[MAXDIM], |
4767 | indx[MAXDIM]; |
4768 | char *ptr; |
4769 | int i, |
4770 | j, |
4771 | inc; |
4772 | int count = 0; |
4773 | |
4774 | mda_get_range(ndim, span, st, endp); |
4775 | |
4776 | /* Pretty easy for fixed element length without nulls ... */ |
4777 | if (typlen > 0 && !arraynullsptr) |
4778 | return ArrayGetNItems(ndim, span) * att_align_nominal(typlen, typalign); |
4779 | |
4780 | /* Else gotta do it the hard way */ |
4781 | src_offset = ArrayGetOffset(ndim, dim, lb, st); |
4782 | ptr = array_seek(arraydataptr, 0, arraynullsptr, src_offset, |
4783 | typlen, typbyval, typalign); |
4784 | mda_get_prod(ndim, dim, prod); |
4785 | mda_get_offset_values(ndim, dist, prod, span); |
4786 | for (i = 0; i < ndim; i++) |
4787 | indx[i] = 0; |
4788 | j = ndim - 1; |
4789 | do |
4790 | { |
4791 | if (dist[j]) |
4792 | { |
4793 | ptr = array_seek(ptr, src_offset, arraynullsptr, dist[j], |
4794 | typlen, typbyval, typalign); |
4795 | src_offset += dist[j]; |
4796 | } |
4797 | if (!array_get_isnull(arraynullsptr, src_offset)) |
4798 | { |
4799 | inc = att_addlength_pointer(0, typlen, ptr); |
4800 | inc = att_align_nominal(inc, typalign); |
4801 | ptr += inc; |
4802 | count += inc; |
4803 | } |
4804 | src_offset++; |
4805 | } while ((j = mda_next_tuple(ndim, indx, span)) != -1); |
4806 | return count; |
4807 | } |
4808 | |
4809 | /* |
4810 | * Extract a slice of an array into consecutive elements in the destination |
4811 | * array. |
4812 | * |
4813 | * We assume the caller has verified that the slice coordinates are valid, |
4814 | * allocated enough storage for the result, and initialized the header |
4815 | * of the new array. |
4816 | */ |
4817 | static void |
4818 | (ArrayType *newarray, |
4819 | int ndim, |
4820 | int *dim, |
4821 | int *lb, |
4822 | char *arraydataptr, |
4823 | bits8 *arraynullsptr, |
4824 | int *st, |
4825 | int *endp, |
4826 | int typlen, |
4827 | bool typbyval, |
4828 | char typalign) |
4829 | { |
4830 | char *destdataptr = ARR_DATA_PTR(newarray); |
4831 | bits8 *destnullsptr = ARR_NULLBITMAP(newarray); |
4832 | char *srcdataptr; |
4833 | int src_offset, |
4834 | dest_offset, |
4835 | prod[MAXDIM], |
4836 | span[MAXDIM], |
4837 | dist[MAXDIM], |
4838 | indx[MAXDIM]; |
4839 | int i, |
4840 | j, |
4841 | inc; |
4842 | |
4843 | src_offset = ArrayGetOffset(ndim, dim, lb, st); |
4844 | srcdataptr = array_seek(arraydataptr, 0, arraynullsptr, src_offset, |
4845 | typlen, typbyval, typalign); |
4846 | mda_get_prod(ndim, dim, prod); |
4847 | mda_get_range(ndim, span, st, endp); |
4848 | mda_get_offset_values(ndim, dist, prod, span); |
4849 | for (i = 0; i < ndim; i++) |
4850 | indx[i] = 0; |
4851 | dest_offset = 0; |
4852 | j = ndim - 1; |
4853 | do |
4854 | { |
4855 | if (dist[j]) |
4856 | { |
4857 | /* skip unwanted elements */ |
4858 | srcdataptr = array_seek(srcdataptr, src_offset, arraynullsptr, |
4859 | dist[j], |
4860 | typlen, typbyval, typalign); |
4861 | src_offset += dist[j]; |
4862 | } |
4863 | inc = array_copy(destdataptr, 1, |
4864 | srcdataptr, src_offset, arraynullsptr, |
4865 | typlen, typbyval, typalign); |
4866 | if (destnullsptr) |
4867 | array_bitmap_copy(destnullsptr, dest_offset, |
4868 | arraynullsptr, src_offset, |
4869 | 1); |
4870 | destdataptr += inc; |
4871 | srcdataptr += inc; |
4872 | src_offset++; |
4873 | dest_offset++; |
4874 | } while ((j = mda_next_tuple(ndim, indx, span)) != -1); |
4875 | } |
4876 | |
4877 | /* |
4878 | * Insert a slice into an array. |
4879 | * |
4880 | * ndim/dim[]/lb[] are dimensions of the original array. A new array with |
4881 | * those same dimensions is to be constructed. destArray must already |
4882 | * have been allocated and its header initialized. |
4883 | * |
4884 | * st[]/endp[] identify the slice to be replaced. Elements within the slice |
4885 | * volume are taken from consecutive elements of the srcArray; elements |
4886 | * outside it are copied from origArray. |
4887 | * |
4888 | * We assume the caller has verified that the slice coordinates are valid. |
4889 | */ |
4890 | static void |
4891 | array_insert_slice(ArrayType *destArray, |
4892 | ArrayType *origArray, |
4893 | ArrayType *srcArray, |
4894 | int ndim, |
4895 | int *dim, |
4896 | int *lb, |
4897 | int *st, |
4898 | int *endp, |
4899 | int typlen, |
4900 | bool typbyval, |
4901 | char typalign) |
4902 | { |
4903 | char *destPtr = ARR_DATA_PTR(destArray); |
4904 | char *origPtr = ARR_DATA_PTR(origArray); |
4905 | char *srcPtr = ARR_DATA_PTR(srcArray); |
4906 | bits8 *destBitmap = ARR_NULLBITMAP(destArray); |
4907 | bits8 *origBitmap = ARR_NULLBITMAP(origArray); |
4908 | bits8 *srcBitmap = ARR_NULLBITMAP(srcArray); |
4909 | int orignitems = ArrayGetNItems(ARR_NDIM(origArray), |
4910 | ARR_DIMS(origArray)); |
4911 | int dest_offset, |
4912 | orig_offset, |
4913 | src_offset, |
4914 | prod[MAXDIM], |
4915 | span[MAXDIM], |
4916 | dist[MAXDIM], |
4917 | indx[MAXDIM]; |
4918 | int i, |
4919 | j, |
4920 | inc; |
4921 | |
4922 | dest_offset = ArrayGetOffset(ndim, dim, lb, st); |
4923 | /* copy items before the slice start */ |
4924 | inc = array_copy(destPtr, dest_offset, |
4925 | origPtr, 0, origBitmap, |
4926 | typlen, typbyval, typalign); |
4927 | destPtr += inc; |
4928 | origPtr += inc; |
4929 | if (destBitmap) |
4930 | array_bitmap_copy(destBitmap, 0, origBitmap, 0, dest_offset); |
4931 | orig_offset = dest_offset; |
4932 | mda_get_prod(ndim, dim, prod); |
4933 | mda_get_range(ndim, span, st, endp); |
4934 | mda_get_offset_values(ndim, dist, prod, span); |
4935 | for (i = 0; i < ndim; i++) |
4936 | indx[i] = 0; |
4937 | src_offset = 0; |
4938 | j = ndim - 1; |
4939 | do |
4940 | { |
4941 | /* Copy/advance over elements between here and next part of slice */ |
4942 | if (dist[j]) |
4943 | { |
4944 | inc = array_copy(destPtr, dist[j], |
4945 | origPtr, orig_offset, origBitmap, |
4946 | typlen, typbyval, typalign); |
4947 | destPtr += inc; |
4948 | origPtr += inc; |
4949 | if (destBitmap) |
4950 | array_bitmap_copy(destBitmap, dest_offset, |
4951 | origBitmap, orig_offset, |
4952 | dist[j]); |
4953 | dest_offset += dist[j]; |
4954 | orig_offset += dist[j]; |
4955 | } |
4956 | /* Copy new element at this slice position */ |
4957 | inc = array_copy(destPtr, 1, |
4958 | srcPtr, src_offset, srcBitmap, |
4959 | typlen, typbyval, typalign); |
4960 | if (destBitmap) |
4961 | array_bitmap_copy(destBitmap, dest_offset, |
4962 | srcBitmap, src_offset, |
4963 | 1); |
4964 | destPtr += inc; |
4965 | srcPtr += inc; |
4966 | dest_offset++; |
4967 | src_offset++; |
4968 | /* Advance over old element at this slice position */ |
4969 | origPtr = array_seek(origPtr, orig_offset, origBitmap, 1, |
4970 | typlen, typbyval, typalign); |
4971 | orig_offset++; |
4972 | } while ((j = mda_next_tuple(ndim, indx, span)) != -1); |
4973 | |
4974 | /* don't miss any data at the end */ |
4975 | array_copy(destPtr, orignitems - orig_offset, |
4976 | origPtr, orig_offset, origBitmap, |
4977 | typlen, typbyval, typalign); |
4978 | if (destBitmap) |
4979 | array_bitmap_copy(destBitmap, dest_offset, |
4980 | origBitmap, orig_offset, |
4981 | orignitems - orig_offset); |
4982 | } |
4983 | |
4984 | /* |
4985 | * initArrayResult - initialize an empty ArrayBuildState |
4986 | * |
4987 | * element_type is the array element type (must be a valid array element type) |
4988 | * rcontext is where to keep working state |
4989 | * subcontext is a flag determining whether to use a separate memory context |
4990 | * |
4991 | * Note: there are two common schemes for using accumArrayResult(). |
4992 | * In the older scheme, you start with a NULL ArrayBuildState pointer, and |
4993 | * call accumArrayResult once per element. In this scheme you end up with |
4994 | * a NULL pointer if there were no elements, which you need to special-case. |
4995 | * In the newer scheme, call initArrayResult and then call accumArrayResult |
4996 | * once per element. In this scheme you always end with a non-NULL pointer |
4997 | * that you can pass to makeArrayResult; you get an empty array if there |
4998 | * were no elements. This is preferred if an empty array is what you want. |
4999 | * |
5000 | * It's possible to choose whether to create a separate memory context for the |
5001 | * array build state, or whether to allocate it directly within rcontext. |
5002 | * |
5003 | * When there are many concurrent small states (e.g. array_agg() using hash |
5004 | * aggregation of many small groups), using a separate memory context for each |
5005 | * one may result in severe memory bloat. In such cases, use the same memory |
5006 | * context to initialize all such array build states, and pass |
5007 | * subcontext=false. |
5008 | * |
5009 | * In cases when the array build states have different lifetimes, using a |
5010 | * single memory context is impractical. Instead, pass subcontext=true so that |
5011 | * the array build states can be freed individually. |
5012 | */ |
5013 | ArrayBuildState * |
5014 | initArrayResult(Oid element_type, MemoryContext rcontext, bool subcontext) |
5015 | { |
5016 | ArrayBuildState *astate; |
5017 | MemoryContext arr_context = rcontext; |
5018 | |
5019 | /* Make a temporary context to hold all the junk */ |
5020 | if (subcontext) |
5021 | arr_context = AllocSetContextCreate(rcontext, |
5022 | "accumArrayResult" , |
5023 | ALLOCSET_DEFAULT_SIZES); |
5024 | |
5025 | astate = (ArrayBuildState *) |
5026 | MemoryContextAlloc(arr_context, sizeof(ArrayBuildState)); |
5027 | astate->mcontext = arr_context; |
5028 | astate->private_cxt = subcontext; |
5029 | astate->alen = (subcontext ? 64 : 8); /* arbitrary starting array size */ |
5030 | astate->dvalues = (Datum *) |
5031 | MemoryContextAlloc(arr_context, astate->alen * sizeof(Datum)); |
5032 | astate->dnulls = (bool *) |
5033 | MemoryContextAlloc(arr_context, astate->alen * sizeof(bool)); |
5034 | astate->nelems = 0; |
5035 | astate->element_type = element_type; |
5036 | get_typlenbyvalalign(element_type, |
5037 | &astate->typlen, |
5038 | &astate->typbyval, |
5039 | &astate->typalign); |
5040 | |
5041 | return astate; |
5042 | } |
5043 | |
5044 | /* |
5045 | * accumArrayResult - accumulate one (more) Datum for an array result |
5046 | * |
5047 | * astate is working state (can be NULL on first call) |
5048 | * dvalue/disnull represent the new Datum to append to the array |
5049 | * element_type is the Datum's type (must be a valid array element type) |
5050 | * rcontext is where to keep working state |
5051 | */ |
5052 | ArrayBuildState * |
5053 | accumArrayResult(ArrayBuildState *astate, |
5054 | Datum dvalue, bool disnull, |
5055 | Oid element_type, |
5056 | MemoryContext rcontext) |
5057 | { |
5058 | MemoryContext oldcontext; |
5059 | |
5060 | if (astate == NULL) |
5061 | { |
5062 | /* First time through --- initialize */ |
5063 | astate = initArrayResult(element_type, rcontext, true); |
5064 | } |
5065 | else |
5066 | { |
5067 | Assert(astate->element_type == element_type); |
5068 | } |
5069 | |
5070 | oldcontext = MemoryContextSwitchTo(astate->mcontext); |
5071 | |
5072 | /* enlarge dvalues[]/dnulls[] if needed */ |
5073 | if (astate->nelems >= astate->alen) |
5074 | { |
5075 | astate->alen *= 2; |
5076 | astate->dvalues = (Datum *) |
5077 | repalloc(astate->dvalues, astate->alen * sizeof(Datum)); |
5078 | astate->dnulls = (bool *) |
5079 | repalloc(astate->dnulls, astate->alen * sizeof(bool)); |
5080 | } |
5081 | |
5082 | /* |
5083 | * Ensure pass-by-ref stuff is copied into mcontext; and detoast it too if |
5084 | * it's varlena. (You might think that detoasting is not needed here |
5085 | * because construct_md_array can detoast the array elements later. |
5086 | * However, we must not let construct_md_array modify the ArrayBuildState |
5087 | * because that would mean array_agg_finalfn damages its input, which is |
5088 | * verboten. Also, this way frequently saves one copying step.) |
5089 | */ |
5090 | if (!disnull && !astate->typbyval) |
5091 | { |
5092 | if (astate->typlen == -1) |
5093 | dvalue = PointerGetDatum(PG_DETOAST_DATUM_COPY(dvalue)); |
5094 | else |
5095 | dvalue = datumCopy(dvalue, astate->typbyval, astate->typlen); |
5096 | } |
5097 | |
5098 | astate->dvalues[astate->nelems] = dvalue; |
5099 | astate->dnulls[astate->nelems] = disnull; |
5100 | astate->nelems++; |
5101 | |
5102 | MemoryContextSwitchTo(oldcontext); |
5103 | |
5104 | return astate; |
5105 | } |
5106 | |
5107 | /* |
5108 | * makeArrayResult - produce 1-D final result of accumArrayResult |
5109 | * |
5110 | * Note: only releases astate if it was initialized within a separate memory |
5111 | * context (i.e. using subcontext=true when calling initArrayResult). |
5112 | * |
5113 | * astate is working state (must not be NULL) |
5114 | * rcontext is where to construct result |
5115 | */ |
5116 | Datum |
5117 | makeArrayResult(ArrayBuildState *astate, |
5118 | MemoryContext rcontext) |
5119 | { |
5120 | int ndims; |
5121 | int dims[1]; |
5122 | int lbs[1]; |
5123 | |
5124 | /* If no elements were presented, we want to create an empty array */ |
5125 | ndims = (astate->nelems > 0) ? 1 : 0; |
5126 | dims[0] = astate->nelems; |
5127 | lbs[0] = 1; |
5128 | |
5129 | return makeMdArrayResult(astate, ndims, dims, lbs, rcontext, |
5130 | astate->private_cxt); |
5131 | } |
5132 | |
5133 | /* |
5134 | * makeMdArrayResult - produce multi-D final result of accumArrayResult |
5135 | * |
5136 | * beware: no check that specified dimensions match the number of values |
5137 | * accumulated. |
5138 | * |
5139 | * Note: if the astate was not initialized within a separate memory context |
5140 | * (that is, initArrayResult was called with subcontext=false), then using |
5141 | * release=true is illegal. Instead, release astate along with the rest of its |
5142 | * context when appropriate. |
5143 | * |
5144 | * astate is working state (must not be NULL) |
5145 | * rcontext is where to construct result |
5146 | * release is true if okay to release working state |
5147 | */ |
5148 | Datum |
5149 | makeMdArrayResult(ArrayBuildState *astate, |
5150 | int ndims, |
5151 | int *dims, |
5152 | int *lbs, |
5153 | MemoryContext rcontext, |
5154 | bool release) |
5155 | { |
5156 | ArrayType *result; |
5157 | MemoryContext oldcontext; |
5158 | |
5159 | /* Build the final array result in rcontext */ |
5160 | oldcontext = MemoryContextSwitchTo(rcontext); |
5161 | |
5162 | result = construct_md_array(astate->dvalues, |
5163 | astate->dnulls, |
5164 | ndims, |
5165 | dims, |
5166 | lbs, |
5167 | astate->element_type, |
5168 | astate->typlen, |
5169 | astate->typbyval, |
5170 | astate->typalign); |
5171 | |
5172 | MemoryContextSwitchTo(oldcontext); |
5173 | |
5174 | /* Clean up all the junk */ |
5175 | if (release) |
5176 | { |
5177 | Assert(astate->private_cxt); |
5178 | MemoryContextDelete(astate->mcontext); |
5179 | } |
5180 | |
5181 | return PointerGetDatum(result); |
5182 | } |
5183 | |
5184 | /* |
5185 | * The following three functions provide essentially the same API as |
5186 | * initArrayResult/accumArrayResult/makeArrayResult, but instead of accepting |
5187 | * inputs that are array elements, they accept inputs that are arrays and |
5188 | * produce an output array having N+1 dimensions. The inputs must all have |
5189 | * identical dimensionality as well as element type. |
5190 | */ |
5191 | |
5192 | /* |
5193 | * initArrayResultArr - initialize an empty ArrayBuildStateArr |
5194 | * |
5195 | * array_type is the array type (must be a valid varlena array type) |
5196 | * element_type is the type of the array's elements (lookup if InvalidOid) |
5197 | * rcontext is where to keep working state |
5198 | * subcontext is a flag determining whether to use a separate memory context |
5199 | */ |
5200 | ArrayBuildStateArr * |
5201 | initArrayResultArr(Oid array_type, Oid element_type, MemoryContext rcontext, |
5202 | bool subcontext) |
5203 | { |
5204 | ArrayBuildStateArr *astate; |
5205 | MemoryContext arr_context = rcontext; /* by default use the parent ctx */ |
5206 | |
5207 | /* Lookup element type, unless element_type already provided */ |
5208 | if (!OidIsValid(element_type)) |
5209 | { |
5210 | element_type = get_element_type(array_type); |
5211 | |
5212 | if (!OidIsValid(element_type)) |
5213 | ereport(ERROR, |
5214 | (errcode(ERRCODE_DATATYPE_MISMATCH), |
5215 | errmsg("data type %s is not an array type" , |
5216 | format_type_be(array_type)))); |
5217 | } |
5218 | |
5219 | /* Make a temporary context to hold all the junk */ |
5220 | if (subcontext) |
5221 | arr_context = AllocSetContextCreate(rcontext, |
5222 | "accumArrayResultArr" , |
5223 | ALLOCSET_DEFAULT_SIZES); |
5224 | |
5225 | /* Note we initialize all fields to zero */ |
5226 | astate = (ArrayBuildStateArr *) |
5227 | MemoryContextAllocZero(arr_context, sizeof(ArrayBuildStateArr)); |
5228 | astate->mcontext = arr_context; |
5229 | astate->private_cxt = subcontext; |
5230 | |
5231 | /* Save relevant datatype information */ |
5232 | astate->array_type = array_type; |
5233 | astate->element_type = element_type; |
5234 | |
5235 | return astate; |
5236 | } |
5237 | |
5238 | /* |
5239 | * accumArrayResultArr - accumulate one (more) sub-array for an array result |
5240 | * |
5241 | * astate is working state (can be NULL on first call) |
5242 | * dvalue/disnull represent the new sub-array to append to the array |
5243 | * array_type is the array type (must be a valid varlena array type) |
5244 | * rcontext is where to keep working state |
5245 | */ |
5246 | ArrayBuildStateArr * |
5247 | accumArrayResultArr(ArrayBuildStateArr *astate, |
5248 | Datum dvalue, bool disnull, |
5249 | Oid array_type, |
5250 | MemoryContext rcontext) |
5251 | { |
5252 | ArrayType *arg; |
5253 | MemoryContext oldcontext; |
5254 | int *dims, |
5255 | *lbs, |
5256 | ndims, |
5257 | nitems, |
5258 | ndatabytes; |
5259 | char *data; |
5260 | int i; |
5261 | |
5262 | /* |
5263 | * We disallow accumulating null subarrays. Another plausible definition |
5264 | * is to ignore them, but callers that want that can just skip calling |
5265 | * this function. |
5266 | */ |
5267 | if (disnull) |
5268 | ereport(ERROR, |
5269 | (errcode(ERRCODE_NULL_VALUE_NOT_ALLOWED), |
5270 | errmsg("cannot accumulate null arrays" ))); |
5271 | |
5272 | /* Detoast input array in caller's context */ |
5273 | arg = DatumGetArrayTypeP(dvalue); |
5274 | |
5275 | if (astate == NULL) |
5276 | astate = initArrayResultArr(array_type, InvalidOid, rcontext, true); |
5277 | else |
5278 | Assert(astate->array_type == array_type); |
5279 | |
5280 | oldcontext = MemoryContextSwitchTo(astate->mcontext); |
5281 | |
5282 | /* Collect this input's dimensions */ |
5283 | ndims = ARR_NDIM(arg); |
5284 | dims = ARR_DIMS(arg); |
5285 | lbs = ARR_LBOUND(arg); |
5286 | data = ARR_DATA_PTR(arg); |
5287 | nitems = ArrayGetNItems(ndims, dims); |
5288 | ndatabytes = ARR_SIZE(arg) - ARR_DATA_OFFSET(arg); |
5289 | |
5290 | if (astate->ndims == 0) |
5291 | { |
5292 | /* First input; check/save the dimensionality info */ |
5293 | |
5294 | /* Should we allow empty inputs and just produce an empty output? */ |
5295 | if (ndims == 0) |
5296 | ereport(ERROR, |
5297 | (errcode(ERRCODE_ARRAY_SUBSCRIPT_ERROR), |
5298 | errmsg("cannot accumulate empty arrays" ))); |
5299 | if (ndims + 1 > MAXDIM) |
5300 | ereport(ERROR, |
5301 | (errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED), |
5302 | errmsg("number of array dimensions (%d) exceeds the maximum allowed (%d)" , |
5303 | ndims + 1, MAXDIM))); |
5304 | |
5305 | /* |
5306 | * The output array will have n+1 dimensions, with the ones after the |
5307 | * first matching the input's dimensions. |
5308 | */ |
5309 | astate->ndims = ndims + 1; |
5310 | astate->dims[0] = 0; |
5311 | memcpy(&astate->dims[1], dims, ndims * sizeof(int)); |
5312 | astate->lbs[0] = 1; |
5313 | memcpy(&astate->lbs[1], lbs, ndims * sizeof(int)); |
5314 | |
5315 | /* Allocate at least enough data space for this item */ |
5316 | astate->abytes = 1024; |
5317 | while (astate->abytes <= ndatabytes) |
5318 | astate->abytes *= 2; |
5319 | astate->data = (char *) palloc(astate->abytes); |
5320 | } |
5321 | else |
5322 | { |
5323 | /* Second or later input: must match first input's dimensionality */ |
5324 | if (astate->ndims != ndims + 1) |
5325 | ereport(ERROR, |
5326 | (errcode(ERRCODE_ARRAY_SUBSCRIPT_ERROR), |
5327 | errmsg("cannot accumulate arrays of different dimensionality" ))); |
5328 | for (i = 0; i < ndims; i++) |
5329 | { |
5330 | if (astate->dims[i + 1] != dims[i] || astate->lbs[i + 1] != lbs[i]) |
5331 | ereport(ERROR, |
5332 | (errcode(ERRCODE_ARRAY_SUBSCRIPT_ERROR), |
5333 | errmsg("cannot accumulate arrays of different dimensionality" ))); |
5334 | } |
5335 | |
5336 | /* Enlarge data space if needed */ |
5337 | if (astate->nbytes + ndatabytes > astate->abytes) |
5338 | { |
5339 | astate->abytes = Max(astate->abytes * 2, |
5340 | astate->nbytes + ndatabytes); |
5341 | astate->data = (char *) repalloc(astate->data, astate->abytes); |
5342 | } |
5343 | } |
5344 | |
5345 | /* |
5346 | * Copy the data portion of the sub-array. Note we assume that the |
5347 | * advertised data length of the sub-array is properly aligned. We do not |
5348 | * have to worry about detoasting elements since whatever's in the |
5349 | * sub-array should be OK already. |
5350 | */ |
5351 | memcpy(astate->data + astate->nbytes, data, ndatabytes); |
5352 | astate->nbytes += ndatabytes; |
5353 | |
5354 | /* Deal with null bitmap if needed */ |
5355 | if (astate->nullbitmap || ARR_HASNULL(arg)) |
5356 | { |
5357 | int newnitems = astate->nitems + nitems; |
5358 | |
5359 | if (astate->nullbitmap == NULL) |
5360 | { |
5361 | /* |
5362 | * First input with nulls; we must retrospectively handle any |
5363 | * previous inputs by marking all their items non-null. |
5364 | */ |
5365 | astate->aitems = 256; |
5366 | while (astate->aitems <= newnitems) |
5367 | astate->aitems *= 2; |
5368 | astate->nullbitmap = (bits8 *) palloc((astate->aitems + 7) / 8); |
5369 | array_bitmap_copy(astate->nullbitmap, 0, |
5370 | NULL, 0, |
5371 | astate->nitems); |
5372 | } |
5373 | else if (newnitems > astate->aitems) |
5374 | { |
5375 | astate->aitems = Max(astate->aitems * 2, newnitems); |
5376 | astate->nullbitmap = (bits8 *) |
5377 | repalloc(astate->nullbitmap, (astate->aitems + 7) / 8); |
5378 | } |
5379 | array_bitmap_copy(astate->nullbitmap, astate->nitems, |
5380 | ARR_NULLBITMAP(arg), 0, |
5381 | nitems); |
5382 | } |
5383 | |
5384 | astate->nitems += nitems; |
5385 | astate->dims[0] += 1; |
5386 | |
5387 | MemoryContextSwitchTo(oldcontext); |
5388 | |
5389 | /* Release detoasted copy if any */ |
5390 | if ((Pointer) arg != DatumGetPointer(dvalue)) |
5391 | pfree(arg); |
5392 | |
5393 | return astate; |
5394 | } |
5395 | |
5396 | /* |
5397 | * makeArrayResultArr - produce N+1-D final result of accumArrayResultArr |
5398 | * |
5399 | * astate is working state (must not be NULL) |
5400 | * rcontext is where to construct result |
5401 | * release is true if okay to release working state |
5402 | */ |
5403 | Datum |
5404 | makeArrayResultArr(ArrayBuildStateArr *astate, |
5405 | MemoryContext rcontext, |
5406 | bool release) |
5407 | { |
5408 | ArrayType *result; |
5409 | MemoryContext oldcontext; |
5410 | |
5411 | /* Build the final array result in rcontext */ |
5412 | oldcontext = MemoryContextSwitchTo(rcontext); |
5413 | |
5414 | if (astate->ndims == 0) |
5415 | { |
5416 | /* No inputs, return empty array */ |
5417 | result = construct_empty_array(astate->element_type); |
5418 | } |
5419 | else |
5420 | { |
5421 | int dataoffset, |
5422 | nbytes; |
5423 | |
5424 | /* Compute required space */ |
5425 | nbytes = astate->nbytes; |
5426 | if (astate->nullbitmap != NULL) |
5427 | { |
5428 | dataoffset = ARR_OVERHEAD_WITHNULLS(astate->ndims, astate->nitems); |
5429 | nbytes += dataoffset; |
5430 | } |
5431 | else |
5432 | { |
5433 | dataoffset = 0; |
5434 | nbytes += ARR_OVERHEAD_NONULLS(astate->ndims); |
5435 | } |
5436 | |
5437 | result = (ArrayType *) palloc0(nbytes); |
5438 | SET_VARSIZE(result, nbytes); |
5439 | result->ndim = astate->ndims; |
5440 | result->dataoffset = dataoffset; |
5441 | result->elemtype = astate->element_type; |
5442 | |
5443 | memcpy(ARR_DIMS(result), astate->dims, astate->ndims * sizeof(int)); |
5444 | memcpy(ARR_LBOUND(result), astate->lbs, astate->ndims * sizeof(int)); |
5445 | memcpy(ARR_DATA_PTR(result), astate->data, astate->nbytes); |
5446 | |
5447 | if (astate->nullbitmap != NULL) |
5448 | array_bitmap_copy(ARR_NULLBITMAP(result), 0, |
5449 | astate->nullbitmap, 0, |
5450 | astate->nitems); |
5451 | } |
5452 | |
5453 | MemoryContextSwitchTo(oldcontext); |
5454 | |
5455 | /* Clean up all the junk */ |
5456 | if (release) |
5457 | { |
5458 | Assert(astate->private_cxt); |
5459 | MemoryContextDelete(astate->mcontext); |
5460 | } |
5461 | |
5462 | return PointerGetDatum(result); |
5463 | } |
5464 | |
5465 | /* |
5466 | * The following three functions provide essentially the same API as |
5467 | * initArrayResult/accumArrayResult/makeArrayResult, but can accept either |
5468 | * scalar or array inputs, invoking the appropriate set of functions above. |
5469 | */ |
5470 | |
5471 | /* |
5472 | * initArrayResultAny - initialize an empty ArrayBuildStateAny |
5473 | * |
5474 | * input_type is the input datatype (either element or array type) |
5475 | * rcontext is where to keep working state |
5476 | * subcontext is a flag determining whether to use a separate memory context |
5477 | */ |
5478 | ArrayBuildStateAny * |
5479 | initArrayResultAny(Oid input_type, MemoryContext rcontext, bool subcontext) |
5480 | { |
5481 | ArrayBuildStateAny *astate; |
5482 | Oid element_type = get_element_type(input_type); |
5483 | |
5484 | if (OidIsValid(element_type)) |
5485 | { |
5486 | /* Array case */ |
5487 | ArrayBuildStateArr *arraystate; |
5488 | |
5489 | arraystate = initArrayResultArr(input_type, InvalidOid, rcontext, subcontext); |
5490 | astate = (ArrayBuildStateAny *) |
5491 | MemoryContextAlloc(arraystate->mcontext, |
5492 | sizeof(ArrayBuildStateAny)); |
5493 | astate->scalarstate = NULL; |
5494 | astate->arraystate = arraystate; |
5495 | } |
5496 | else |
5497 | { |
5498 | /* Scalar case */ |
5499 | ArrayBuildState *scalarstate; |
5500 | |
5501 | /* Let's just check that we have a type that can be put into arrays */ |
5502 | Assert(OidIsValid(get_array_type(input_type))); |
5503 | |
5504 | scalarstate = initArrayResult(input_type, rcontext, subcontext); |
5505 | astate = (ArrayBuildStateAny *) |
5506 | MemoryContextAlloc(scalarstate->mcontext, |
5507 | sizeof(ArrayBuildStateAny)); |
5508 | astate->scalarstate = scalarstate; |
5509 | astate->arraystate = NULL; |
5510 | } |
5511 | |
5512 | return astate; |
5513 | } |
5514 | |
5515 | /* |
5516 | * accumArrayResultAny - accumulate one (more) input for an array result |
5517 | * |
5518 | * astate is working state (can be NULL on first call) |
5519 | * dvalue/disnull represent the new input to append to the array |
5520 | * input_type is the input datatype (either element or array type) |
5521 | * rcontext is where to keep working state |
5522 | */ |
5523 | ArrayBuildStateAny * |
5524 | accumArrayResultAny(ArrayBuildStateAny *astate, |
5525 | Datum dvalue, bool disnull, |
5526 | Oid input_type, |
5527 | MemoryContext rcontext) |
5528 | { |
5529 | if (astate == NULL) |
5530 | astate = initArrayResultAny(input_type, rcontext, true); |
5531 | |
5532 | if (astate->scalarstate) |
5533 | (void) accumArrayResult(astate->scalarstate, |
5534 | dvalue, disnull, |
5535 | input_type, rcontext); |
5536 | else |
5537 | (void) accumArrayResultArr(astate->arraystate, |
5538 | dvalue, disnull, |
5539 | input_type, rcontext); |
5540 | |
5541 | return astate; |
5542 | } |
5543 | |
5544 | /* |
5545 | * makeArrayResultAny - produce final result of accumArrayResultAny |
5546 | * |
5547 | * astate is working state (must not be NULL) |
5548 | * rcontext is where to construct result |
5549 | * release is true if okay to release working state |
5550 | */ |
5551 | Datum |
5552 | makeArrayResultAny(ArrayBuildStateAny *astate, |
5553 | MemoryContext rcontext, bool release) |
5554 | { |
5555 | Datum result; |
5556 | |
5557 | if (astate->scalarstate) |
5558 | { |
5559 | /* Must use makeMdArrayResult to support "release" parameter */ |
5560 | int ndims; |
5561 | int dims[1]; |
5562 | int lbs[1]; |
5563 | |
5564 | /* If no elements were presented, we want to create an empty array */ |
5565 | ndims = (astate->scalarstate->nelems > 0) ? 1 : 0; |
5566 | dims[0] = astate->scalarstate->nelems; |
5567 | lbs[0] = 1; |
5568 | |
5569 | result = makeMdArrayResult(astate->scalarstate, ndims, dims, lbs, |
5570 | rcontext, release); |
5571 | } |
5572 | else |
5573 | { |
5574 | result = makeArrayResultArr(astate->arraystate, |
5575 | rcontext, release); |
5576 | } |
5577 | return result; |
5578 | } |
5579 | |
5580 | |
5581 | Datum |
5582 | array_larger(PG_FUNCTION_ARGS) |
5583 | { |
5584 | if (array_cmp(fcinfo) > 0) |
5585 | PG_RETURN_DATUM(PG_GETARG_DATUM(0)); |
5586 | else |
5587 | PG_RETURN_DATUM(PG_GETARG_DATUM(1)); |
5588 | } |
5589 | |
5590 | Datum |
5591 | array_smaller(PG_FUNCTION_ARGS) |
5592 | { |
5593 | if (array_cmp(fcinfo) < 0) |
5594 | PG_RETURN_DATUM(PG_GETARG_DATUM(0)); |
5595 | else |
5596 | PG_RETURN_DATUM(PG_GETARG_DATUM(1)); |
5597 | } |
5598 | |
5599 | |
5600 | typedef struct generate_subscripts_fctx |
5601 | { |
5602 | int32 lower; |
5603 | int32 upper; |
5604 | bool reverse; |
5605 | } generate_subscripts_fctx; |
5606 | |
5607 | /* |
5608 | * generate_subscripts(array anyarray, dim int [, reverse bool]) |
5609 | * Returns all subscripts of the array for any dimension |
5610 | */ |
5611 | Datum |
5612 | generate_subscripts(PG_FUNCTION_ARGS) |
5613 | { |
5614 | FuncCallContext *funcctx; |
5615 | MemoryContext oldcontext; |
5616 | generate_subscripts_fctx *fctx; |
5617 | |
5618 | /* stuff done only on the first call of the function */ |
5619 | if (SRF_IS_FIRSTCALL()) |
5620 | { |
5621 | AnyArrayType *v = PG_GETARG_ANY_ARRAY_P(0); |
5622 | int reqdim = PG_GETARG_INT32(1); |
5623 | int *lb, |
5624 | *dimv; |
5625 | |
5626 | /* create a function context for cross-call persistence */ |
5627 | funcctx = SRF_FIRSTCALL_INIT(); |
5628 | |
5629 | /* Sanity check: does it look like an array at all? */ |
5630 | if (AARR_NDIM(v) <= 0 || AARR_NDIM(v) > MAXDIM) |
5631 | SRF_RETURN_DONE(funcctx); |
5632 | |
5633 | /* Sanity check: was the requested dim valid */ |
5634 | if (reqdim <= 0 || reqdim > AARR_NDIM(v)) |
5635 | SRF_RETURN_DONE(funcctx); |
5636 | |
5637 | /* |
5638 | * switch to memory context appropriate for multiple function calls |
5639 | */ |
5640 | oldcontext = MemoryContextSwitchTo(funcctx->multi_call_memory_ctx); |
5641 | fctx = (generate_subscripts_fctx *) palloc(sizeof(generate_subscripts_fctx)); |
5642 | |
5643 | lb = AARR_LBOUND(v); |
5644 | dimv = AARR_DIMS(v); |
5645 | |
5646 | fctx->lower = lb[reqdim - 1]; |
5647 | fctx->upper = dimv[reqdim - 1] + lb[reqdim - 1] - 1; |
5648 | fctx->reverse = (PG_NARGS() < 3) ? false : PG_GETARG_BOOL(2); |
5649 | |
5650 | funcctx->user_fctx = fctx; |
5651 | |
5652 | MemoryContextSwitchTo(oldcontext); |
5653 | } |
5654 | |
5655 | funcctx = SRF_PERCALL_SETUP(); |
5656 | |
5657 | fctx = funcctx->user_fctx; |
5658 | |
5659 | if (fctx->lower <= fctx->upper) |
5660 | { |
5661 | if (!fctx->reverse) |
5662 | SRF_RETURN_NEXT(funcctx, Int32GetDatum(fctx->lower++)); |
5663 | else |
5664 | SRF_RETURN_NEXT(funcctx, Int32GetDatum(fctx->upper--)); |
5665 | } |
5666 | else |
5667 | /* done when there are no more elements left */ |
5668 | SRF_RETURN_DONE(funcctx); |
5669 | } |
5670 | |
5671 | /* |
5672 | * generate_subscripts_nodir |
5673 | * Implements the 2-argument version of generate_subscripts |
5674 | */ |
5675 | Datum |
5676 | generate_subscripts_nodir(PG_FUNCTION_ARGS) |
5677 | { |
5678 | /* just call the other one -- it can handle both cases */ |
5679 | return generate_subscripts(fcinfo); |
5680 | } |
5681 | |
5682 | /* |
5683 | * array_fill_with_lower_bounds |
5684 | * Create and fill array with defined lower bounds. |
5685 | */ |
5686 | Datum |
5687 | array_fill_with_lower_bounds(PG_FUNCTION_ARGS) |
5688 | { |
5689 | ArrayType *dims; |
5690 | ArrayType *lbs; |
5691 | ArrayType *result; |
5692 | Oid elmtype; |
5693 | Datum value; |
5694 | bool isnull; |
5695 | |
5696 | if (PG_ARGISNULL(1) || PG_ARGISNULL(2)) |
5697 | ereport(ERROR, |
5698 | (errcode(ERRCODE_NULL_VALUE_NOT_ALLOWED), |
5699 | errmsg("dimension array or low bound array cannot be null" ))); |
5700 | |
5701 | dims = PG_GETARG_ARRAYTYPE_P(1); |
5702 | lbs = PG_GETARG_ARRAYTYPE_P(2); |
5703 | |
5704 | if (!PG_ARGISNULL(0)) |
5705 | { |
5706 | value = PG_GETARG_DATUM(0); |
5707 | isnull = false; |
5708 | } |
5709 | else |
5710 | { |
5711 | value = 0; |
5712 | isnull = true; |
5713 | } |
5714 | |
5715 | elmtype = get_fn_expr_argtype(fcinfo->flinfo, 0); |
5716 | if (!OidIsValid(elmtype)) |
5717 | elog(ERROR, "could not determine data type of input" ); |
5718 | |
5719 | result = array_fill_internal(dims, lbs, value, isnull, elmtype, fcinfo); |
5720 | PG_RETURN_ARRAYTYPE_P(result); |
5721 | } |
5722 | |
5723 | /* |
5724 | * array_fill |
5725 | * Create and fill array with default lower bounds. |
5726 | */ |
5727 | Datum |
5728 | array_fill(PG_FUNCTION_ARGS) |
5729 | { |
5730 | ArrayType *dims; |
5731 | ArrayType *result; |
5732 | Oid elmtype; |
5733 | Datum value; |
5734 | bool isnull; |
5735 | |
5736 | if (PG_ARGISNULL(1)) |
5737 | ereport(ERROR, |
5738 | (errcode(ERRCODE_NULL_VALUE_NOT_ALLOWED), |
5739 | errmsg("dimension array or low bound array cannot be null" ))); |
5740 | |
5741 | dims = PG_GETARG_ARRAYTYPE_P(1); |
5742 | |
5743 | if (!PG_ARGISNULL(0)) |
5744 | { |
5745 | value = PG_GETARG_DATUM(0); |
5746 | isnull = false; |
5747 | } |
5748 | else |
5749 | { |
5750 | value = 0; |
5751 | isnull = true; |
5752 | } |
5753 | |
5754 | elmtype = get_fn_expr_argtype(fcinfo->flinfo, 0); |
5755 | if (!OidIsValid(elmtype)) |
5756 | elog(ERROR, "could not determine data type of input" ); |
5757 | |
5758 | result = array_fill_internal(dims, NULL, value, isnull, elmtype, fcinfo); |
5759 | PG_RETURN_ARRAYTYPE_P(result); |
5760 | } |
5761 | |
5762 | static ArrayType * |
5763 | create_array_envelope(int ndims, int *dimv, int *lbsv, int nbytes, |
5764 | Oid elmtype, int dataoffset) |
5765 | { |
5766 | ArrayType *result; |
5767 | |
5768 | result = (ArrayType *) palloc0(nbytes); |
5769 | SET_VARSIZE(result, nbytes); |
5770 | result->ndim = ndims; |
5771 | result->dataoffset = dataoffset; |
5772 | result->elemtype = elmtype; |
5773 | memcpy(ARR_DIMS(result), dimv, ndims * sizeof(int)); |
5774 | memcpy(ARR_LBOUND(result), lbsv, ndims * sizeof(int)); |
5775 | |
5776 | return result; |
5777 | } |
5778 | |
5779 | static ArrayType * |
5780 | array_fill_internal(ArrayType *dims, ArrayType *lbs, |
5781 | Datum value, bool isnull, Oid elmtype, |
5782 | FunctionCallInfo fcinfo) |
5783 | { |
5784 | ArrayType *result; |
5785 | int *dimv; |
5786 | int *lbsv; |
5787 | int ndims; |
5788 | int nitems; |
5789 | int deflbs[MAXDIM]; |
5790 | int16 elmlen; |
5791 | bool elmbyval; |
5792 | char elmalign; |
5793 | ArrayMetaState *; |
5794 | |
5795 | /* |
5796 | * Params checks |
5797 | */ |
5798 | if (ARR_NDIM(dims) > 1) |
5799 | ereport(ERROR, |
5800 | (errcode(ERRCODE_ARRAY_SUBSCRIPT_ERROR), |
5801 | errmsg("wrong number of array subscripts" ), |
5802 | errdetail("Dimension array must be one dimensional." ))); |
5803 | |
5804 | if (array_contains_nulls(dims)) |
5805 | ereport(ERROR, |
5806 | (errcode(ERRCODE_NULL_VALUE_NOT_ALLOWED), |
5807 | errmsg("dimension values cannot be null" ))); |
5808 | |
5809 | dimv = (int *) ARR_DATA_PTR(dims); |
5810 | ndims = (ARR_NDIM(dims) > 0) ? ARR_DIMS(dims)[0] : 0; |
5811 | |
5812 | if (ndims < 0) /* we do allow zero-dimension arrays */ |
5813 | ereport(ERROR, |
5814 | (errcode(ERRCODE_INVALID_PARAMETER_VALUE), |
5815 | errmsg("invalid number of dimensions: %d" , ndims))); |
5816 | if (ndims > MAXDIM) |
5817 | ereport(ERROR, |
5818 | (errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED), |
5819 | errmsg("number of array dimensions (%d) exceeds the maximum allowed (%d)" , |
5820 | ndims, MAXDIM))); |
5821 | |
5822 | if (lbs != NULL) |
5823 | { |
5824 | if (ARR_NDIM(lbs) > 1) |
5825 | ereport(ERROR, |
5826 | (errcode(ERRCODE_ARRAY_SUBSCRIPT_ERROR), |
5827 | errmsg("wrong number of array subscripts" ), |
5828 | errdetail("Dimension array must be one dimensional." ))); |
5829 | |
5830 | if (array_contains_nulls(lbs)) |
5831 | ereport(ERROR, |
5832 | (errcode(ERRCODE_NULL_VALUE_NOT_ALLOWED), |
5833 | errmsg("dimension values cannot be null" ))); |
5834 | |
5835 | if (ndims != ((ARR_NDIM(lbs) > 0) ? ARR_DIMS(lbs)[0] : 0)) |
5836 | ereport(ERROR, |
5837 | (errcode(ERRCODE_ARRAY_SUBSCRIPT_ERROR), |
5838 | errmsg("wrong number of array subscripts" ), |
5839 | errdetail("Low bound array has different size than dimensions array." ))); |
5840 | |
5841 | lbsv = (int *) ARR_DATA_PTR(lbs); |
5842 | } |
5843 | else |
5844 | { |
5845 | int i; |
5846 | |
5847 | for (i = 0; i < MAXDIM; i++) |
5848 | deflbs[i] = 1; |
5849 | |
5850 | lbsv = deflbs; |
5851 | } |
5852 | |
5853 | nitems = ArrayGetNItems(ndims, dimv); |
5854 | |
5855 | /* fast track for empty array */ |
5856 | if (nitems <= 0) |
5857 | return construct_empty_array(elmtype); |
5858 | |
5859 | /* |
5860 | * We arrange to look up info about element type only once per series of |
5861 | * calls, assuming the element type doesn't change underneath us. |
5862 | */ |
5863 | my_extra = (ArrayMetaState *) fcinfo->flinfo->fn_extra; |
5864 | if (my_extra == NULL) |
5865 | { |
5866 | fcinfo->flinfo->fn_extra = MemoryContextAlloc(fcinfo->flinfo->fn_mcxt, |
5867 | sizeof(ArrayMetaState)); |
5868 | my_extra = (ArrayMetaState *) fcinfo->flinfo->fn_extra; |
5869 | my_extra->element_type = InvalidOid; |
5870 | } |
5871 | |
5872 | if (my_extra->element_type != elmtype) |
5873 | { |
5874 | /* Get info about element type */ |
5875 | get_typlenbyvalalign(elmtype, |
5876 | &my_extra->typlen, |
5877 | &my_extra->typbyval, |
5878 | &my_extra->typalign); |
5879 | my_extra->element_type = elmtype; |
5880 | } |
5881 | |
5882 | elmlen = my_extra->typlen; |
5883 | elmbyval = my_extra->typbyval; |
5884 | elmalign = my_extra->typalign; |
5885 | |
5886 | /* compute required space */ |
5887 | if (!isnull) |
5888 | { |
5889 | int i; |
5890 | char *p; |
5891 | int nbytes; |
5892 | int totbytes; |
5893 | |
5894 | /* make sure data is not toasted */ |
5895 | if (elmlen == -1) |
5896 | value = PointerGetDatum(PG_DETOAST_DATUM(value)); |
5897 | |
5898 | nbytes = att_addlength_datum(0, elmlen, value); |
5899 | nbytes = att_align_nominal(nbytes, elmalign); |
5900 | Assert(nbytes > 0); |
5901 | |
5902 | totbytes = nbytes * nitems; |
5903 | |
5904 | /* check for overflow of multiplication or total request */ |
5905 | if (totbytes / nbytes != nitems || |
5906 | !AllocSizeIsValid(totbytes)) |
5907 | ereport(ERROR, |
5908 | (errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED), |
5909 | errmsg("array size exceeds the maximum allowed (%d)" , |
5910 | (int) MaxAllocSize))); |
5911 | |
5912 | /* |
5913 | * This addition can't overflow, but it might cause us to go past |
5914 | * MaxAllocSize. We leave it to palloc to complain in that case. |
5915 | */ |
5916 | totbytes += ARR_OVERHEAD_NONULLS(ndims); |
5917 | |
5918 | result = create_array_envelope(ndims, dimv, lbsv, totbytes, |
5919 | elmtype, 0); |
5920 | |
5921 | p = ARR_DATA_PTR(result); |
5922 | for (i = 0; i < nitems; i++) |
5923 | p += ArrayCastAndSet(value, elmlen, elmbyval, elmalign, p); |
5924 | } |
5925 | else |
5926 | { |
5927 | int nbytes; |
5928 | int dataoffset; |
5929 | |
5930 | dataoffset = ARR_OVERHEAD_WITHNULLS(ndims, nitems); |
5931 | nbytes = dataoffset; |
5932 | |
5933 | result = create_array_envelope(ndims, dimv, lbsv, nbytes, |
5934 | elmtype, dataoffset); |
5935 | |
5936 | /* create_array_envelope already zeroed the bitmap, so we're done */ |
5937 | } |
5938 | |
5939 | return result; |
5940 | } |
5941 | |
5942 | |
5943 | /* |
5944 | * UNNEST |
5945 | */ |
5946 | Datum |
5947 | array_unnest(PG_FUNCTION_ARGS) |
5948 | { |
5949 | typedef struct |
5950 | { |
5951 | array_iter iter; |
5952 | int nextelem; |
5953 | int numelems; |
5954 | int16 elmlen; |
5955 | bool elmbyval; |
5956 | char elmalign; |
5957 | } array_unnest_fctx; |
5958 | |
5959 | FuncCallContext *funcctx; |
5960 | array_unnest_fctx *fctx; |
5961 | MemoryContext oldcontext; |
5962 | |
5963 | /* stuff done only on the first call of the function */ |
5964 | if (SRF_IS_FIRSTCALL()) |
5965 | { |
5966 | AnyArrayType *arr; |
5967 | |
5968 | /* create a function context for cross-call persistence */ |
5969 | funcctx = SRF_FIRSTCALL_INIT(); |
5970 | |
5971 | /* |
5972 | * switch to memory context appropriate for multiple function calls |
5973 | */ |
5974 | oldcontext = MemoryContextSwitchTo(funcctx->multi_call_memory_ctx); |
5975 | |
5976 | /* |
5977 | * Get the array value and detoast if needed. We can't do this |
5978 | * earlier because if we have to detoast, we want the detoasted copy |
5979 | * to be in multi_call_memory_ctx, so it will go away when we're done |
5980 | * and not before. (If no detoast happens, we assume the originally |
5981 | * passed array will stick around till then.) |
5982 | */ |
5983 | arr = PG_GETARG_ANY_ARRAY_P(0); |
5984 | |
5985 | /* allocate memory for user context */ |
5986 | fctx = (array_unnest_fctx *) palloc(sizeof(array_unnest_fctx)); |
5987 | |
5988 | /* initialize state */ |
5989 | array_iter_setup(&fctx->iter, arr); |
5990 | fctx->nextelem = 0; |
5991 | fctx->numelems = ArrayGetNItems(AARR_NDIM(arr), AARR_DIMS(arr)); |
5992 | |
5993 | if (VARATT_IS_EXPANDED_HEADER(arr)) |
5994 | { |
5995 | /* we can just grab the type data from expanded array */ |
5996 | fctx->elmlen = arr->xpn.typlen; |
5997 | fctx->elmbyval = arr->xpn.typbyval; |
5998 | fctx->elmalign = arr->xpn.typalign; |
5999 | } |
6000 | else |
6001 | get_typlenbyvalalign(AARR_ELEMTYPE(arr), |
6002 | &fctx->elmlen, |
6003 | &fctx->elmbyval, |
6004 | &fctx->elmalign); |
6005 | |
6006 | funcctx->user_fctx = fctx; |
6007 | MemoryContextSwitchTo(oldcontext); |
6008 | } |
6009 | |
6010 | /* stuff done on every call of the function */ |
6011 | funcctx = SRF_PERCALL_SETUP(); |
6012 | fctx = funcctx->user_fctx; |
6013 | |
6014 | if (fctx->nextelem < fctx->numelems) |
6015 | { |
6016 | int offset = fctx->nextelem++; |
6017 | Datum elem; |
6018 | |
6019 | elem = array_iter_next(&fctx->iter, &fcinfo->isnull, offset, |
6020 | fctx->elmlen, fctx->elmbyval, fctx->elmalign); |
6021 | |
6022 | SRF_RETURN_NEXT(funcctx, elem); |
6023 | } |
6024 | else |
6025 | { |
6026 | /* do when there is no more left */ |
6027 | SRF_RETURN_DONE(funcctx); |
6028 | } |
6029 | } |
6030 | |
6031 | /* |
6032 | * Planner support function for array_unnest(anyarray) |
6033 | */ |
6034 | Datum |
6035 | array_unnest_support(PG_FUNCTION_ARGS) |
6036 | { |
6037 | Node *rawreq = (Node *) PG_GETARG_POINTER(0); |
6038 | Node *ret = NULL; |
6039 | |
6040 | if (IsA(rawreq, SupportRequestRows)) |
6041 | { |
6042 | /* Try to estimate the number of rows returned */ |
6043 | SupportRequestRows *req = (SupportRequestRows *) rawreq; |
6044 | |
6045 | if (is_funcclause(req->node)) /* be paranoid */ |
6046 | { |
6047 | List *args = ((FuncExpr *) req->node)->args; |
6048 | Node *arg1; |
6049 | |
6050 | /* We can use estimated argument values here */ |
6051 | arg1 = estimate_expression_value(req->root, linitial(args)); |
6052 | |
6053 | req->rows = estimate_array_length(arg1); |
6054 | ret = (Node *) req; |
6055 | } |
6056 | } |
6057 | |
6058 | PG_RETURN_POINTER(ret); |
6059 | } |
6060 | |
6061 | |
6062 | /* |
6063 | * array_replace/array_remove support |
6064 | * |
6065 | * Find all array entries matching (not distinct from) search/search_isnull, |
6066 | * and delete them if remove is true, else replace them with |
6067 | * replace/replace_isnull. Comparisons are done using the specified |
6068 | * collation. fcinfo is passed only for caching purposes. |
6069 | */ |
6070 | static ArrayType * |
6071 | array_replace_internal(ArrayType *array, |
6072 | Datum search, bool search_isnull, |
6073 | Datum replace, bool replace_isnull, |
6074 | bool remove, Oid collation, |
6075 | FunctionCallInfo fcinfo) |
6076 | { |
6077 | LOCAL_FCINFO(locfcinfo, 2); |
6078 | ArrayType *result; |
6079 | Oid element_type; |
6080 | Datum *values; |
6081 | bool *nulls; |
6082 | int *dim; |
6083 | int ndim; |
6084 | int nitems, |
6085 | nresult; |
6086 | int i; |
6087 | int32 nbytes = 0; |
6088 | int32 dataoffset; |
6089 | bool hasnulls; |
6090 | int typlen; |
6091 | bool typbyval; |
6092 | char typalign; |
6093 | char *arraydataptr; |
6094 | bits8 *bitmap; |
6095 | int bitmask; |
6096 | bool changed = false; |
6097 | TypeCacheEntry *typentry; |
6098 | |
6099 | element_type = ARR_ELEMTYPE(array); |
6100 | ndim = ARR_NDIM(array); |
6101 | dim = ARR_DIMS(array); |
6102 | nitems = ArrayGetNItems(ndim, dim); |
6103 | |
6104 | /* Return input array unmodified if it is empty */ |
6105 | if (nitems <= 0) |
6106 | return array; |
6107 | |
6108 | /* |
6109 | * We can't remove elements from multi-dimensional arrays, since the |
6110 | * result might not be rectangular. |
6111 | */ |
6112 | if (remove && ndim > 1) |
6113 | ereport(ERROR, |
6114 | (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), |
6115 | errmsg("removing elements from multidimensional arrays is not supported" ))); |
6116 | |
6117 | /* |
6118 | * We arrange to look up the equality function only once per series of |
6119 | * calls, assuming the element type doesn't change underneath us. |
6120 | */ |
6121 | typentry = (TypeCacheEntry *) fcinfo->flinfo->fn_extra; |
6122 | if (typentry == NULL || |
6123 | typentry->type_id != element_type) |
6124 | { |
6125 | typentry = lookup_type_cache(element_type, |
6126 | TYPECACHE_EQ_OPR_FINFO); |
6127 | if (!OidIsValid(typentry->eq_opr_finfo.fn_oid)) |
6128 | ereport(ERROR, |
6129 | (errcode(ERRCODE_UNDEFINED_FUNCTION), |
6130 | errmsg("could not identify an equality operator for type %s" , |
6131 | format_type_be(element_type)))); |
6132 | fcinfo->flinfo->fn_extra = (void *) typentry; |
6133 | } |
6134 | typlen = typentry->typlen; |
6135 | typbyval = typentry->typbyval; |
6136 | typalign = typentry->typalign; |
6137 | |
6138 | /* |
6139 | * Detoast values if they are toasted. The replacement value must be |
6140 | * detoasted for insertion into the result array, while detoasting the |
6141 | * search value only once saves cycles. |
6142 | */ |
6143 | if (typlen == -1) |
6144 | { |
6145 | if (!search_isnull) |
6146 | search = PointerGetDatum(PG_DETOAST_DATUM(search)); |
6147 | if (!replace_isnull) |
6148 | replace = PointerGetDatum(PG_DETOAST_DATUM(replace)); |
6149 | } |
6150 | |
6151 | /* Prepare to apply the comparison operator */ |
6152 | InitFunctionCallInfoData(*locfcinfo, &typentry->eq_opr_finfo, 2, |
6153 | collation, NULL, NULL); |
6154 | |
6155 | /* Allocate temporary arrays for new values */ |
6156 | values = (Datum *) palloc(nitems * sizeof(Datum)); |
6157 | nulls = (bool *) palloc(nitems * sizeof(bool)); |
6158 | |
6159 | /* Loop over source data */ |
6160 | arraydataptr = ARR_DATA_PTR(array); |
6161 | bitmap = ARR_NULLBITMAP(array); |
6162 | bitmask = 1; |
6163 | hasnulls = false; |
6164 | nresult = 0; |
6165 | |
6166 | for (i = 0; i < nitems; i++) |
6167 | { |
6168 | Datum elt; |
6169 | bool isNull; |
6170 | bool oprresult; |
6171 | bool skip = false; |
6172 | |
6173 | /* Get source element, checking for NULL */ |
6174 | if (bitmap && (*bitmap & bitmask) == 0) |
6175 | { |
6176 | isNull = true; |
6177 | /* If searching for NULL, we have a match */ |
6178 | if (search_isnull) |
6179 | { |
6180 | if (remove) |
6181 | { |
6182 | skip = true; |
6183 | changed = true; |
6184 | } |
6185 | else if (!replace_isnull) |
6186 | { |
6187 | values[nresult] = replace; |
6188 | isNull = false; |
6189 | changed = true; |
6190 | } |
6191 | } |
6192 | } |
6193 | else |
6194 | { |
6195 | isNull = false; |
6196 | elt = fetch_att(arraydataptr, typbyval, typlen); |
6197 | arraydataptr = att_addlength_datum(arraydataptr, typlen, elt); |
6198 | arraydataptr = (char *) att_align_nominal(arraydataptr, typalign); |
6199 | |
6200 | if (search_isnull) |
6201 | { |
6202 | /* no match possible, keep element */ |
6203 | values[nresult] = elt; |
6204 | } |
6205 | else |
6206 | { |
6207 | /* |
6208 | * Apply the operator to the element pair |
6209 | */ |
6210 | locfcinfo->args[0].value = elt; |
6211 | locfcinfo->args[0].isnull = false; |
6212 | locfcinfo->args[1].value = search; |
6213 | locfcinfo->args[1].isnull = false; |
6214 | locfcinfo->isnull = false; |
6215 | oprresult = DatumGetBool(FunctionCallInvoke(locfcinfo)); |
6216 | if (!oprresult) |
6217 | { |
6218 | /* no match, keep element */ |
6219 | values[nresult] = elt; |
6220 | } |
6221 | else |
6222 | { |
6223 | /* match, so replace or delete */ |
6224 | changed = true; |
6225 | if (remove) |
6226 | skip = true; |
6227 | else |
6228 | { |
6229 | values[nresult] = replace; |
6230 | isNull = replace_isnull; |
6231 | } |
6232 | } |
6233 | } |
6234 | } |
6235 | |
6236 | if (!skip) |
6237 | { |
6238 | nulls[nresult] = isNull; |
6239 | if (isNull) |
6240 | hasnulls = true; |
6241 | else |
6242 | { |
6243 | /* Update total result size */ |
6244 | nbytes = att_addlength_datum(nbytes, typlen, values[nresult]); |
6245 | nbytes = att_align_nominal(nbytes, typalign); |
6246 | /* check for overflow of total request */ |
6247 | if (!AllocSizeIsValid(nbytes)) |
6248 | ereport(ERROR, |
6249 | (errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED), |
6250 | errmsg("array size exceeds the maximum allowed (%d)" , |
6251 | (int) MaxAllocSize))); |
6252 | } |
6253 | nresult++; |
6254 | } |
6255 | |
6256 | /* advance bitmap pointer if any */ |
6257 | if (bitmap) |
6258 | { |
6259 | bitmask <<= 1; |
6260 | if (bitmask == 0x100) |
6261 | { |
6262 | bitmap++; |
6263 | bitmask = 1; |
6264 | } |
6265 | } |
6266 | } |
6267 | |
6268 | /* |
6269 | * If not changed just return the original array |
6270 | */ |
6271 | if (!changed) |
6272 | { |
6273 | pfree(values); |
6274 | pfree(nulls); |
6275 | return array; |
6276 | } |
6277 | |
6278 | /* If all elements were removed return an empty array */ |
6279 | if (nresult == 0) |
6280 | { |
6281 | pfree(values); |
6282 | pfree(nulls); |
6283 | return construct_empty_array(element_type); |
6284 | } |
6285 | |
6286 | /* Allocate and initialize the result array */ |
6287 | if (hasnulls) |
6288 | { |
6289 | dataoffset = ARR_OVERHEAD_WITHNULLS(ndim, nresult); |
6290 | nbytes += dataoffset; |
6291 | } |
6292 | else |
6293 | { |
6294 | dataoffset = 0; /* marker for no null bitmap */ |
6295 | nbytes += ARR_OVERHEAD_NONULLS(ndim); |
6296 | } |
6297 | result = (ArrayType *) palloc0(nbytes); |
6298 | SET_VARSIZE(result, nbytes); |
6299 | result->ndim = ndim; |
6300 | result->dataoffset = dataoffset; |
6301 | result->elemtype = element_type; |
6302 | memcpy(ARR_DIMS(result), ARR_DIMS(array), ndim * sizeof(int)); |
6303 | memcpy(ARR_LBOUND(result), ARR_LBOUND(array), ndim * sizeof(int)); |
6304 | |
6305 | if (remove) |
6306 | { |
6307 | /* Adjust the result length */ |
6308 | ARR_DIMS(result)[0] = nresult; |
6309 | } |
6310 | |
6311 | /* Insert data into result array */ |
6312 | CopyArrayEls(result, |
6313 | values, nulls, nresult, |
6314 | typlen, typbyval, typalign, |
6315 | false); |
6316 | |
6317 | pfree(values); |
6318 | pfree(nulls); |
6319 | |
6320 | return result; |
6321 | } |
6322 | |
6323 | /* |
6324 | * Remove any occurrences of an element from an array |
6325 | * |
6326 | * If used on a multi-dimensional array this will raise an error. |
6327 | */ |
6328 | Datum |
6329 | array_remove(PG_FUNCTION_ARGS) |
6330 | { |
6331 | ArrayType *array; |
6332 | Datum search = PG_GETARG_DATUM(1); |
6333 | bool search_isnull = PG_ARGISNULL(1); |
6334 | |
6335 | if (PG_ARGISNULL(0)) |
6336 | PG_RETURN_NULL(); |
6337 | array = PG_GETARG_ARRAYTYPE_P(0); |
6338 | |
6339 | array = array_replace_internal(array, |
6340 | search, search_isnull, |
6341 | (Datum) 0, true, |
6342 | true, PG_GET_COLLATION(), |
6343 | fcinfo); |
6344 | PG_RETURN_ARRAYTYPE_P(array); |
6345 | } |
6346 | |
6347 | /* |
6348 | * Replace any occurrences of an element in an array |
6349 | */ |
6350 | Datum |
6351 | array_replace(PG_FUNCTION_ARGS) |
6352 | { |
6353 | ArrayType *array; |
6354 | Datum search = PG_GETARG_DATUM(1); |
6355 | bool search_isnull = PG_ARGISNULL(1); |
6356 | Datum replace = PG_GETARG_DATUM(2); |
6357 | bool replace_isnull = PG_ARGISNULL(2); |
6358 | |
6359 | if (PG_ARGISNULL(0)) |
6360 | PG_RETURN_NULL(); |
6361 | array = PG_GETARG_ARRAYTYPE_P(0); |
6362 | |
6363 | array = array_replace_internal(array, |
6364 | search, search_isnull, |
6365 | replace, replace_isnull, |
6366 | false, PG_GET_COLLATION(), |
6367 | fcinfo); |
6368 | PG_RETURN_ARRAYTYPE_P(array); |
6369 | } |
6370 | |
6371 | /* |
6372 | * Implements width_bucket(anyelement, anyarray). |
6373 | * |
6374 | * 'thresholds' is an array containing lower bound values for each bucket; |
6375 | * these must be sorted from smallest to largest, or bogus results will be |
6376 | * produced. If N thresholds are supplied, the output is from 0 to N: |
6377 | * 0 is for inputs < first threshold, N is for inputs >= last threshold. |
6378 | */ |
6379 | Datum |
6380 | width_bucket_array(PG_FUNCTION_ARGS) |
6381 | { |
6382 | Datum operand = PG_GETARG_DATUM(0); |
6383 | ArrayType *thresholds = PG_GETARG_ARRAYTYPE_P(1); |
6384 | Oid collation = PG_GET_COLLATION(); |
6385 | Oid element_type = ARR_ELEMTYPE(thresholds); |
6386 | int result; |
6387 | |
6388 | /* Check input */ |
6389 | if (ARR_NDIM(thresholds) > 1) |
6390 | ereport(ERROR, |
6391 | (errcode(ERRCODE_ARRAY_SUBSCRIPT_ERROR), |
6392 | errmsg("thresholds must be one-dimensional array" ))); |
6393 | |
6394 | if (array_contains_nulls(thresholds)) |
6395 | ereport(ERROR, |
6396 | (errcode(ERRCODE_NULL_VALUE_NOT_ALLOWED), |
6397 | errmsg("thresholds array must not contain NULLs" ))); |
6398 | |
6399 | /* We have a dedicated implementation for float8 data */ |
6400 | if (element_type == FLOAT8OID) |
6401 | result = width_bucket_array_float8(operand, thresholds); |
6402 | else |
6403 | { |
6404 | TypeCacheEntry *typentry; |
6405 | |
6406 | /* Cache information about the input type */ |
6407 | typentry = (TypeCacheEntry *) fcinfo->flinfo->fn_extra; |
6408 | if (typentry == NULL || |
6409 | typentry->type_id != element_type) |
6410 | { |
6411 | typentry = lookup_type_cache(element_type, |
6412 | TYPECACHE_CMP_PROC_FINFO); |
6413 | if (!OidIsValid(typentry->cmp_proc_finfo.fn_oid)) |
6414 | ereport(ERROR, |
6415 | (errcode(ERRCODE_UNDEFINED_FUNCTION), |
6416 | errmsg("could not identify a comparison function for type %s" , |
6417 | format_type_be(element_type)))); |
6418 | fcinfo->flinfo->fn_extra = (void *) typentry; |
6419 | } |
6420 | |
6421 | /* |
6422 | * We have separate implementation paths for fixed- and variable-width |
6423 | * types, since indexing the array is a lot cheaper in the first case. |
6424 | */ |
6425 | if (typentry->typlen > 0) |
6426 | result = width_bucket_array_fixed(operand, thresholds, |
6427 | collation, typentry); |
6428 | else |
6429 | result = width_bucket_array_variable(operand, thresholds, |
6430 | collation, typentry); |
6431 | } |
6432 | |
6433 | /* Avoid leaking memory when handed toasted input. */ |
6434 | PG_FREE_IF_COPY(thresholds, 1); |
6435 | |
6436 | PG_RETURN_INT32(result); |
6437 | } |
6438 | |
6439 | /* |
6440 | * width_bucket_array for float8 data. |
6441 | */ |
6442 | static int |
6443 | width_bucket_array_float8(Datum operand, ArrayType *thresholds) |
6444 | { |
6445 | float8 op = DatumGetFloat8(operand); |
6446 | float8 *thresholds_data; |
6447 | int left; |
6448 | int right; |
6449 | |
6450 | /* |
6451 | * Since we know the array contains no NULLs, we can just index it |
6452 | * directly. |
6453 | */ |
6454 | thresholds_data = (float8 *) ARR_DATA_PTR(thresholds); |
6455 | |
6456 | left = 0; |
6457 | right = ArrayGetNItems(ARR_NDIM(thresholds), ARR_DIMS(thresholds)); |
6458 | |
6459 | /* |
6460 | * If the probe value is a NaN, it's greater than or equal to all possible |
6461 | * threshold values (including other NaNs), so we need not search. Note |
6462 | * that this would give the same result as searching even if the array |
6463 | * contains multiple NaNs (as long as they're correctly sorted), since the |
6464 | * loop logic will find the rightmost of multiple equal threshold values. |
6465 | */ |
6466 | if (isnan(op)) |
6467 | return right; |
6468 | |
6469 | /* Find the bucket */ |
6470 | while (left < right) |
6471 | { |
6472 | int mid = (left + right) / 2; |
6473 | |
6474 | if (isnan(thresholds_data[mid]) || op < thresholds_data[mid]) |
6475 | right = mid; |
6476 | else |
6477 | left = mid + 1; |
6478 | } |
6479 | |
6480 | return left; |
6481 | } |
6482 | |
6483 | /* |
6484 | * width_bucket_array for generic fixed-width data types. |
6485 | */ |
6486 | static int |
6487 | width_bucket_array_fixed(Datum operand, |
6488 | ArrayType *thresholds, |
6489 | Oid collation, |
6490 | TypeCacheEntry *typentry) |
6491 | { |
6492 | LOCAL_FCINFO(locfcinfo, 2); |
6493 | char *thresholds_data; |
6494 | int typlen = typentry->typlen; |
6495 | bool typbyval = typentry->typbyval; |
6496 | int left; |
6497 | int right; |
6498 | |
6499 | /* |
6500 | * Since we know the array contains no NULLs, we can just index it |
6501 | * directly. |
6502 | */ |
6503 | thresholds_data = (char *) ARR_DATA_PTR(thresholds); |
6504 | |
6505 | InitFunctionCallInfoData(*locfcinfo, &typentry->cmp_proc_finfo, 2, |
6506 | collation, NULL, NULL); |
6507 | |
6508 | /* Find the bucket */ |
6509 | left = 0; |
6510 | right = ArrayGetNItems(ARR_NDIM(thresholds), ARR_DIMS(thresholds)); |
6511 | while (left < right) |
6512 | { |
6513 | int mid = (left + right) / 2; |
6514 | char *ptr; |
6515 | int32 cmpresult; |
6516 | |
6517 | ptr = thresholds_data + mid * typlen; |
6518 | |
6519 | locfcinfo->args[0].value = operand; |
6520 | locfcinfo->args[0].isnull = false; |
6521 | locfcinfo->args[1].value = fetch_att(ptr, typbyval, typlen); |
6522 | locfcinfo->args[1].isnull = false; |
6523 | locfcinfo->isnull = false; |
6524 | |
6525 | cmpresult = DatumGetInt32(FunctionCallInvoke(locfcinfo)); |
6526 | |
6527 | if (cmpresult < 0) |
6528 | right = mid; |
6529 | else |
6530 | left = mid + 1; |
6531 | } |
6532 | |
6533 | return left; |
6534 | } |
6535 | |
6536 | /* |
6537 | * width_bucket_array for generic variable-width data types. |
6538 | */ |
6539 | static int |
6540 | width_bucket_array_variable(Datum operand, |
6541 | ArrayType *thresholds, |
6542 | Oid collation, |
6543 | TypeCacheEntry *typentry) |
6544 | { |
6545 | LOCAL_FCINFO(locfcinfo, 2); |
6546 | char *thresholds_data; |
6547 | int typlen = typentry->typlen; |
6548 | bool typbyval = typentry->typbyval; |
6549 | char typalign = typentry->typalign; |
6550 | int left; |
6551 | int right; |
6552 | |
6553 | thresholds_data = (char *) ARR_DATA_PTR(thresholds); |
6554 | |
6555 | InitFunctionCallInfoData(*locfcinfo, &typentry->cmp_proc_finfo, 2, |
6556 | collation, NULL, NULL); |
6557 | |
6558 | /* Find the bucket */ |
6559 | left = 0; |
6560 | right = ArrayGetNItems(ARR_NDIM(thresholds), ARR_DIMS(thresholds)); |
6561 | while (left < right) |
6562 | { |
6563 | int mid = (left + right) / 2; |
6564 | char *ptr; |
6565 | int i; |
6566 | int32 cmpresult; |
6567 | |
6568 | /* Locate mid'th array element by advancing from left element */ |
6569 | ptr = thresholds_data; |
6570 | for (i = left; i < mid; i++) |
6571 | { |
6572 | ptr = att_addlength_pointer(ptr, typlen, ptr); |
6573 | ptr = (char *) att_align_nominal(ptr, typalign); |
6574 | } |
6575 | |
6576 | locfcinfo->args[0].value = operand; |
6577 | locfcinfo->args[0].isnull = false; |
6578 | locfcinfo->args[1].value = fetch_att(ptr, typbyval, typlen); |
6579 | locfcinfo->args[1].isnull = false; |
6580 | |
6581 | cmpresult = DatumGetInt32(FunctionCallInvoke(locfcinfo)); |
6582 | |
6583 | if (cmpresult < 0) |
6584 | right = mid; |
6585 | else |
6586 | { |
6587 | left = mid + 1; |
6588 | |
6589 | /* |
6590 | * Move the thresholds pointer to match new "left" index, so we |
6591 | * don't have to seek over those elements again. This trick |
6592 | * ensures we do only O(N) array indexing work, not O(N^2). |
6593 | */ |
6594 | ptr = att_addlength_pointer(ptr, typlen, ptr); |
6595 | thresholds_data = (char *) att_align_nominal(ptr, typalign); |
6596 | } |
6597 | } |
6598 | |
6599 | return left; |
6600 | } |
6601 | |