Line data Source code
1 : /*
2 : * transforming Datums to Python objects and vice versa
3 : *
4 : * src/pl/plpython/plpy_typeio.c
5 : */
6 :
7 : #include "postgres.h"
8 :
9 : #include "access/htup_details.h"
10 : #include "catalog/pg_type.h"
11 : #include "funcapi.h"
12 : #include "mb/pg_wchar.h"
13 : #include "miscadmin.h"
14 : #include "plpy_elog.h"
15 : #include "plpy_main.h"
16 : #include "plpy_typeio.h"
17 : #include "plpy_util.h"
18 : #include "utils/array.h"
19 : #include "utils/builtins.h"
20 : #include "utils/fmgroids.h"
21 : #include "utils/lsyscache.h"
22 : #include "utils/memutils.h"
23 :
24 : /* conversion from Datums to Python objects */
25 : static PyObject *PLyBool_FromBool(PLyDatumToOb *arg, Datum d);
26 : static PyObject *PLyFloat_FromFloat4(PLyDatumToOb *arg, Datum d);
27 : static PyObject *PLyFloat_FromFloat8(PLyDatumToOb *arg, Datum d);
28 : static PyObject *PLyDecimal_FromNumeric(PLyDatumToOb *arg, Datum d);
29 : static PyObject *PLyLong_FromInt16(PLyDatumToOb *arg, Datum d);
30 : static PyObject *PLyLong_FromInt32(PLyDatumToOb *arg, Datum d);
31 : static PyObject *PLyLong_FromInt64(PLyDatumToOb *arg, Datum d);
32 : static PyObject *PLyLong_FromOid(PLyDatumToOb *arg, Datum d);
33 : static PyObject *PLyBytes_FromBytea(PLyDatumToOb *arg, Datum d);
34 : static PyObject *PLyUnicode_FromScalar(PLyDatumToOb *arg, Datum d);
35 : static PyObject *PLyObject_FromTransform(PLyDatumToOb *arg, Datum d);
36 : static PyObject *PLyList_FromArray(PLyDatumToOb *arg, Datum d);
37 : static PyObject *PLyList_FromArray_recurse(PLyDatumToOb *elm, int *dims, int ndim, int dim,
38 : char **dataptr_p, bits8 **bitmap_p, int *bitmask_p);
39 : static PyObject *PLyDict_FromComposite(PLyDatumToOb *arg, Datum d);
40 : static PyObject *PLyDict_FromTuple(PLyDatumToOb *arg, HeapTuple tuple, TupleDesc desc, bool include_generated);
41 :
42 : /* conversion from Python objects to Datums */
43 : static Datum PLyObject_ToBool(PLyObToDatum *arg, PyObject *plrv,
44 : bool *isnull, bool inarray);
45 : static Datum PLyObject_ToBytea(PLyObToDatum *arg, PyObject *plrv,
46 : bool *isnull, bool inarray);
47 : static Datum PLyObject_ToComposite(PLyObToDatum *arg, PyObject *plrv,
48 : bool *isnull, bool inarray);
49 : static Datum PLyObject_ToScalar(PLyObToDatum *arg, PyObject *plrv,
50 : bool *isnull, bool inarray);
51 : static Datum PLyObject_ToDomain(PLyObToDatum *arg, PyObject *plrv,
52 : bool *isnull, bool inarray);
53 : static Datum PLyObject_ToTransform(PLyObToDatum *arg, PyObject *plrv,
54 : bool *isnull, bool inarray);
55 : static Datum PLySequence_ToArray(PLyObToDatum *arg, PyObject *plrv,
56 : bool *isnull, bool inarray);
57 : static void PLySequence_ToArray_recurse(PyObject *obj,
58 : ArrayBuildState **astatep,
59 : int *ndims, int *dims, int cur_depth,
60 : PLyObToDatum *elm, Oid elmbasetype);
61 :
62 : /* conversion from Python objects to composite Datums */
63 : static Datum PLyUnicode_ToComposite(PLyObToDatum *arg, PyObject *string, bool inarray);
64 : static Datum PLyMapping_ToComposite(PLyObToDatum *arg, TupleDesc desc, PyObject *mapping);
65 : static Datum PLySequence_ToComposite(PLyObToDatum *arg, TupleDesc desc, PyObject *sequence);
66 : static Datum PLyGenericObject_ToComposite(PLyObToDatum *arg, TupleDesc desc, PyObject *object, bool inarray);
67 :
68 :
69 : /*
70 : * Conversion functions. Remember output from Python is input to
71 : * PostgreSQL, and vice versa.
72 : */
73 :
74 : /*
75 : * Perform input conversion, given correctly-set-up state information.
76 : *
77 : * This is the outer-level entry point for any input conversion. Internally,
78 : * the conversion functions recurse directly to each other.
79 : */
80 : PyObject *
81 594 : PLy_input_convert(PLyDatumToOb *arg, Datum val)
82 : {
83 : PyObject *result;
84 594 : PLyExecutionContext *exec_ctx = PLy_current_execution_context();
85 594 : MemoryContext scratch_context = PLy_get_scratch_context(exec_ctx);
86 : MemoryContext oldcontext;
87 :
88 : /*
89 : * Do the work in the scratch context to avoid leaking memory from the
90 : * datatype output function calls. (The individual PLyDatumToObFunc
91 : * functions can't reset the scratch context, because they recurse and an
92 : * inner one might clobber data an outer one still needs. So we do it
93 : * once at the outermost recursion level.)
94 : *
95 : * We reset the scratch context before, not after, each conversion cycle.
96 : * This way we aren't on the hook to release a Python refcount on the
97 : * result object in case MemoryContextReset throws an error.
98 : */
99 594 : MemoryContextReset(scratch_context);
100 :
101 594 : oldcontext = MemoryContextSwitchTo(scratch_context);
102 :
103 594 : result = arg->func(arg, val);
104 :
105 594 : MemoryContextSwitchTo(oldcontext);
106 :
107 594 : return result;
108 : }
109 :
110 : /*
111 : * Perform output conversion, given correctly-set-up state information.
112 : *
113 : * This is the outer-level entry point for any output conversion. Internally,
114 : * the conversion functions recurse directly to each other.
115 : *
116 : * The result, as well as any cruft generated along the way, are in the
117 : * current memory context. Caller is responsible for cleanup.
118 : */
119 : Datum
120 572 : PLy_output_convert(PLyObToDatum *arg, PyObject *val, bool *isnull)
121 : {
122 : /* at outer level, we are not considering an array element */
123 572 : return arg->func(arg, val, isnull, false);
124 : }
125 :
126 : /*
127 : * Transform a tuple into a Python dict object.
128 : *
129 : * Note: the tupdesc must match the one used to set up *arg. We could
130 : * insist that this function lookup the tupdesc from what is in *arg,
131 : * but in practice all callers have the right tupdesc available.
132 : */
133 : PyObject *
134 199 : PLy_input_from_tuple(PLyDatumToOb *arg, HeapTuple tuple, TupleDesc desc, bool include_generated)
135 : {
136 : PyObject *dict;
137 199 : PLyExecutionContext *exec_ctx = PLy_current_execution_context();
138 199 : MemoryContext scratch_context = PLy_get_scratch_context(exec_ctx);
139 : MemoryContext oldcontext;
140 :
141 : /*
142 : * As in PLy_input_convert, do the work in the scratch context.
143 : */
144 199 : MemoryContextReset(scratch_context);
145 :
146 199 : oldcontext = MemoryContextSwitchTo(scratch_context);
147 :
148 199 : dict = PLyDict_FromTuple(arg, tuple, desc, include_generated);
149 :
150 199 : MemoryContextSwitchTo(oldcontext);
151 :
152 199 : return dict;
153 : }
154 :
155 : /*
156 : * Initialize, or re-initialize, per-column input info for a composite type.
157 : *
158 : * This is separate from PLy_input_setup_func() because in cases involving
159 : * anonymous record types, we need to be passed the tupdesc explicitly.
160 : * It's caller's responsibility that the tupdesc has adequate lifespan
161 : * in such cases. If the tupdesc is for a named composite or registered
162 : * record type, it does not need to be long-lived.
163 : */
164 : void
165 189 : PLy_input_setup_tuple(PLyDatumToOb *arg, TupleDesc desc, PLyProcedure *proc)
166 : {
167 : int i;
168 :
169 : /* We should be working on a previously-set-up struct */
170 : Assert(arg->func == PLyDict_FromComposite);
171 :
172 : /* Save pointer to tupdesc, but only if this is an anonymous record type */
173 189 : if (arg->typoid == RECORDOID && arg->typmod < 0)
174 91 : arg->tuple.recdesc = desc;
175 :
176 : /* (Re)allocate atts array as needed */
177 189 : if (arg->tuple.natts != desc->natts)
178 : {
179 109 : if (arg->tuple.atts)
180 1 : pfree(arg->tuple.atts);
181 109 : arg->tuple.natts = desc->natts;
182 109 : arg->tuple.atts = (PLyDatumToOb *)
183 109 : MemoryContextAllocZero(arg->mcxt,
184 109 : desc->natts * sizeof(PLyDatumToOb));
185 : }
186 :
187 : /* Fill the atts entries, except for dropped columns */
188 607 : for (i = 0; i < desc->natts; i++)
189 : {
190 418 : Form_pg_attribute attr = TupleDescAttr(desc, i);
191 418 : PLyDatumToOb *att = &arg->tuple.atts[i];
192 :
193 418 : if (attr->attisdropped)
194 3 : continue;
195 :
196 415 : if (att->typoid == attr->atttypid && att->typmod == attr->atttypmod)
197 232 : continue; /* already set up this entry */
198 :
199 183 : PLy_input_setup_func(att, arg->mcxt,
200 : attr->atttypid, attr->atttypmod,
201 : proc);
202 : }
203 189 : }
204 :
205 : /*
206 : * Initialize, or re-initialize, per-column output info for a composite type.
207 : *
208 : * This is separate from PLy_output_setup_func() because in cases involving
209 : * anonymous record types, we need to be passed the tupdesc explicitly.
210 : * It's caller's responsibility that the tupdesc has adequate lifespan
211 : * in such cases. If the tupdesc is for a named composite or registered
212 : * record type, it does not need to be long-lived.
213 : */
214 : void
215 214 : PLy_output_setup_tuple(PLyObToDatum *arg, TupleDesc desc, PLyProcedure *proc)
216 : {
217 : int i;
218 :
219 : /* We should be working on a previously-set-up struct */
220 : Assert(arg->func == PLyObject_ToComposite);
221 :
222 : /* Save pointer to tupdesc, but only if this is an anonymous record type */
223 214 : if (arg->typoid == RECORDOID && arg->typmod < 0)
224 0 : arg->tuple.recdesc = desc;
225 :
226 : /* (Re)allocate atts array as needed */
227 214 : if (arg->tuple.natts != desc->natts)
228 : {
229 74 : if (arg->tuple.atts)
230 3 : pfree(arg->tuple.atts);
231 74 : arg->tuple.natts = desc->natts;
232 74 : arg->tuple.atts = (PLyObToDatum *)
233 74 : MemoryContextAllocZero(arg->mcxt,
234 74 : desc->natts * sizeof(PLyObToDatum));
235 : }
236 :
237 : /* Fill the atts entries, except for dropped columns */
238 698 : for (i = 0; i < desc->natts; i++)
239 : {
240 484 : Form_pg_attribute attr = TupleDescAttr(desc, i);
241 484 : PLyObToDatum *att = &arg->tuple.atts[i];
242 :
243 484 : if (attr->attisdropped)
244 28 : continue;
245 :
246 456 : if (att->typoid == attr->atttypid && att->typmod == attr->atttypmod)
247 297 : continue; /* already set up this entry */
248 :
249 159 : PLy_output_setup_func(att, arg->mcxt,
250 : attr->atttypid, attr->atttypmod,
251 : proc);
252 : }
253 214 : }
254 :
255 : /*
256 : * Set up output info for a PL/Python function returning record.
257 : *
258 : * Note: the given tupdesc is not necessarily long-lived.
259 : */
260 : void
261 134 : PLy_output_setup_record(PLyObToDatum *arg, TupleDesc desc, PLyProcedure *proc)
262 : {
263 : /* Makes no sense unless RECORD */
264 : Assert(arg->typoid == RECORDOID);
265 : Assert(desc->tdtypeid == RECORDOID);
266 :
267 : /*
268 : * Bless the record type if not already done. We'd have to do this anyway
269 : * to return a tuple, so we might as well force the issue so we can use
270 : * the known-record-type code path.
271 : */
272 134 : BlessTupleDesc(desc);
273 :
274 : /*
275 : * Update arg->typmod, and clear the recdesc link if it's changed. The
276 : * next call of PLyObject_ToComposite will look up a long-lived tupdesc
277 : * for the record type.
278 : */
279 134 : arg->typmod = desc->tdtypmod;
280 134 : if (arg->tuple.recdesc &&
281 116 : arg->tuple.recdesc->tdtypmod != arg->typmod)
282 10 : arg->tuple.recdesc = NULL;
283 :
284 : /* Update derived data if necessary */
285 134 : PLy_output_setup_tuple(arg, desc, proc);
286 134 : }
287 :
288 : /*
289 : * Recursively initialize the PLyObToDatum structure(s) needed to construct
290 : * a SQL value of the specified typeOid/typmod from a Python value.
291 : * (But note that at this point we may have RECORDOID/-1, ie, an indeterminate
292 : * record type.)
293 : * proc is used to look up transform functions.
294 : */
295 : void
296 479 : PLy_output_setup_func(PLyObToDatum *arg, MemoryContext arg_mcxt,
297 : Oid typeOid, int32 typmod,
298 : PLyProcedure *proc)
299 : {
300 : TypeCacheEntry *typentry;
301 : char typtype;
302 : Oid trfuncid;
303 : Oid typinput;
304 :
305 : /* Since this is recursive, it could theoretically be driven to overflow */
306 479 : check_stack_depth();
307 :
308 479 : arg->typoid = typeOid;
309 479 : arg->typmod = typmod;
310 479 : arg->mcxt = arg_mcxt;
311 :
312 : /*
313 : * Fetch typcache entry for the target type, asking for whatever info
314 : * we'll need later. RECORD is a special case: just treat it as composite
315 : * without bothering with the typcache entry.
316 : */
317 479 : if (typeOid != RECORDOID)
318 : {
319 461 : typentry = lookup_type_cache(typeOid, TYPECACHE_DOMAIN_BASE_INFO);
320 461 : typtype = typentry->typtype;
321 461 : arg->typbyval = typentry->typbyval;
322 461 : arg->typlen = typentry->typlen;
323 461 : arg->typalign = typentry->typalign;
324 : }
325 : else
326 : {
327 18 : typentry = NULL;
328 18 : typtype = TYPTYPE_COMPOSITE;
329 : /* hard-wired knowledge about type RECORD: */
330 18 : arg->typbyval = false;
331 18 : arg->typlen = -1;
332 18 : arg->typalign = TYPALIGN_DOUBLE;
333 : }
334 :
335 : /*
336 : * Choose conversion method. Note that transform functions are checked
337 : * for composite and scalar types, but not for arrays or domains. This is
338 : * somewhat historical, but we'd have a problem allowing them on domains,
339 : * since we drill down through all levels of a domain nest without looking
340 : * at the intermediate levels at all.
341 : */
342 479 : if (typtype == TYPTYPE_DOMAIN)
343 : {
344 : /* Domain */
345 14 : arg->func = PLyObject_ToDomain;
346 14 : arg->domain.domain_info = NULL;
347 : /* Recursively set up conversion info for the element type */
348 14 : arg->domain.base = (PLyObToDatum *)
349 14 : MemoryContextAllocZero(arg_mcxt, sizeof(PLyObToDatum));
350 14 : PLy_output_setup_func(arg->domain.base, arg_mcxt,
351 : typentry->domainBaseType,
352 : typentry->domainBaseTypmod,
353 : proc);
354 : }
355 465 : else if (typentry &&
356 447 : IsTrueArrayType(typentry))
357 : {
358 : /* Standard array */
359 39 : arg->func = PLySequence_ToArray;
360 : /* Get base type OID to insert into constructed array */
361 : /* (note this might not be the same as the immediate child type) */
362 39 : arg->array.elmbasetype = getBaseType(typentry->typelem);
363 : /* Recursively set up conversion info for the element type */
364 39 : arg->array.elm = (PLyObToDatum *)
365 39 : MemoryContextAllocZero(arg_mcxt, sizeof(PLyObToDatum));
366 39 : PLy_output_setup_func(arg->array.elm, arg_mcxt,
367 : typentry->typelem, typmod,
368 : proc);
369 : }
370 426 : else if ((trfuncid = get_transform_tosql(typeOid,
371 : proc->langid,
372 : proc->trftypes)))
373 : {
374 12 : arg->func = PLyObject_ToTransform;
375 12 : fmgr_info_cxt(trfuncid, &arg->transform.typtransform, arg_mcxt);
376 : }
377 414 : else if (typtype == TYPTYPE_COMPOSITE)
378 : {
379 : /* Named composite type, or RECORD */
380 73 : arg->func = PLyObject_ToComposite;
381 : /* We'll set up the per-field data later */
382 73 : arg->tuple.recdesc = NULL;
383 73 : arg->tuple.typentry = typentry;
384 73 : arg->tuple.tupdescid = INVALID_TUPLEDESC_IDENTIFIER;
385 73 : arg->tuple.atts = NULL;
386 73 : arg->tuple.natts = 0;
387 : /* Mark this invalid till needed, too */
388 73 : arg->tuple.recinfunc.fn_oid = InvalidOid;
389 : }
390 : else
391 : {
392 : /* Scalar type, but we have a couple of special cases */
393 341 : switch (typeOid)
394 : {
395 14 : case BOOLOID:
396 14 : arg->func = PLyObject_ToBool;
397 14 : break;
398 6 : case BYTEAOID:
399 6 : arg->func = PLyObject_ToBytea;
400 6 : break;
401 321 : default:
402 321 : arg->func = PLyObject_ToScalar;
403 321 : getTypeInputInfo(typeOid, &typinput, &arg->scalar.typioparam);
404 321 : fmgr_info_cxt(typinput, &arg->scalar.typfunc, arg_mcxt);
405 321 : break;
406 : }
407 : }
408 479 : }
409 :
410 : /*
411 : * Recursively initialize the PLyDatumToOb structure(s) needed to construct
412 : * a Python value from a SQL value of the specified typeOid/typmod.
413 : * (But note that at this point we may have RECORDOID/-1, ie, an indeterminate
414 : * record type.)
415 : * proc is used to look up transform functions.
416 : */
417 : void
418 467 : PLy_input_setup_func(PLyDatumToOb *arg, MemoryContext arg_mcxt,
419 : Oid typeOid, int32 typmod,
420 : PLyProcedure *proc)
421 : {
422 : TypeCacheEntry *typentry;
423 : char typtype;
424 : Oid trfuncid;
425 : Oid typoutput;
426 : bool typisvarlena;
427 :
428 : /* Since this is recursive, it could theoretically be driven to overflow */
429 467 : check_stack_depth();
430 :
431 467 : arg->typoid = typeOid;
432 467 : arg->typmod = typmod;
433 467 : arg->mcxt = arg_mcxt;
434 :
435 : /*
436 : * Fetch typcache entry for the target type, asking for whatever info
437 : * we'll need later. RECORD is a special case: just treat it as composite
438 : * without bothering with the typcache entry.
439 : */
440 467 : if (typeOid != RECORDOID)
441 : {
442 389 : typentry = lookup_type_cache(typeOid, TYPECACHE_DOMAIN_BASE_INFO);
443 389 : typtype = typentry->typtype;
444 389 : arg->typbyval = typentry->typbyval;
445 389 : arg->typlen = typentry->typlen;
446 389 : arg->typalign = typentry->typalign;
447 : }
448 : else
449 : {
450 78 : typentry = NULL;
451 78 : typtype = TYPTYPE_COMPOSITE;
452 : /* hard-wired knowledge about type RECORD: */
453 78 : arg->typbyval = false;
454 78 : arg->typlen = -1;
455 78 : arg->typalign = TYPALIGN_DOUBLE;
456 : }
457 :
458 : /*
459 : * Choose conversion method. Note that transform functions are checked
460 : * for composite and scalar types, but not for arrays or domains. This is
461 : * somewhat historical, but we'd have a problem allowing them on domains,
462 : * since we drill down through all levels of a domain nest without looking
463 : * at the intermediate levels at all.
464 : */
465 467 : if (typtype == TYPTYPE_DOMAIN)
466 : {
467 : /* Domain --- we don't care, just recurse down to the base type */
468 9 : PLy_input_setup_func(arg, arg_mcxt,
469 : typentry->domainBaseType,
470 : typentry->domainBaseTypmod,
471 : proc);
472 : }
473 458 : else if (typentry &&
474 380 : IsTrueArrayType(typentry))
475 : {
476 : /* Standard array */
477 13 : arg->func = PLyList_FromArray;
478 : /* Recursively set up conversion info for the element type */
479 13 : arg->array.elm = (PLyDatumToOb *)
480 13 : MemoryContextAllocZero(arg_mcxt, sizeof(PLyDatumToOb));
481 13 : PLy_input_setup_func(arg->array.elm, arg_mcxt,
482 : typentry->typelem, typmod,
483 : proc);
484 : }
485 445 : else if ((trfuncid = get_transform_fromsql(typeOid,
486 : proc->langid,
487 : proc->trftypes)))
488 : {
489 16 : arg->func = PLyObject_FromTransform;
490 16 : fmgr_info_cxt(trfuncid, &arg->transform.typtransform, arg_mcxt);
491 : }
492 429 : else if (typtype == TYPTYPE_COMPOSITE)
493 : {
494 : /* Named composite type, or RECORD */
495 119 : arg->func = PLyDict_FromComposite;
496 : /* We'll set up the per-field data later */
497 119 : arg->tuple.recdesc = NULL;
498 119 : arg->tuple.typentry = typentry;
499 119 : arg->tuple.tupdescid = INVALID_TUPLEDESC_IDENTIFIER;
500 119 : arg->tuple.atts = NULL;
501 119 : arg->tuple.natts = 0;
502 : }
503 : else
504 : {
505 : /* Scalar type, but we have a couple of special cases */
506 310 : switch (typeOid)
507 : {
508 14 : case BOOLOID:
509 14 : arg->func = PLyBool_FromBool;
510 14 : break;
511 1 : case FLOAT4OID:
512 1 : arg->func = PLyFloat_FromFloat4;
513 1 : break;
514 1 : case FLOAT8OID:
515 1 : arg->func = PLyFloat_FromFloat8;
516 1 : break;
517 1 : case NUMERICOID:
518 1 : arg->func = PLyDecimal_FromNumeric;
519 1 : break;
520 4 : case INT2OID:
521 4 : arg->func = PLyLong_FromInt16;
522 4 : break;
523 147 : case INT4OID:
524 147 : arg->func = PLyLong_FromInt32;
525 147 : break;
526 5 : case INT8OID:
527 5 : arg->func = PLyLong_FromInt64;
528 5 : break;
529 1 : case OIDOID:
530 1 : arg->func = PLyLong_FromOid;
531 1 : break;
532 7 : case BYTEAOID:
533 7 : arg->func = PLyBytes_FromBytea;
534 7 : break;
535 129 : default:
536 129 : arg->func = PLyUnicode_FromScalar;
537 129 : getTypeOutputInfo(typeOid, &typoutput, &typisvarlena);
538 129 : fmgr_info_cxt(typoutput, &arg->scalar.typfunc, arg_mcxt);
539 129 : break;
540 : }
541 : }
542 467 : }
543 :
544 :
545 : /*
546 : * Special-purpose input converters.
547 : */
548 :
549 : static PyObject *
550 121 : PLyBool_FromBool(PLyDatumToOb *arg, Datum d)
551 : {
552 121 : if (DatumGetBool(d))
553 26 : Py_RETURN_TRUE;
554 95 : Py_RETURN_FALSE;
555 : }
556 :
557 : static PyObject *
558 3 : PLyFloat_FromFloat4(PLyDatumToOb *arg, Datum d)
559 : {
560 3 : return PyFloat_FromDouble(DatumGetFloat4(d));
561 : }
562 :
563 : static PyObject *
564 4 : PLyFloat_FromFloat8(PLyDatumToOb *arg, Datum d)
565 : {
566 4 : return PyFloat_FromDouble(DatumGetFloat8(d));
567 : }
568 :
569 : static PyObject *
570 7 : PLyDecimal_FromNumeric(PLyDatumToOb *arg, Datum d)
571 : {
572 : static PyObject *decimal_constructor;
573 : char *str;
574 : PyObject *pyvalue;
575 :
576 : /* Try to import cdecimal. If it doesn't exist, fall back to decimal. */
577 7 : if (!decimal_constructor)
578 : {
579 : PyObject *decimal_module;
580 :
581 1 : decimal_module = PyImport_ImportModule("cdecimal");
582 1 : if (!decimal_module)
583 : {
584 1 : PyErr_Clear();
585 1 : decimal_module = PyImport_ImportModule("decimal");
586 : }
587 1 : if (!decimal_module)
588 0 : PLy_elog(ERROR, "could not import a module for Decimal constructor");
589 :
590 1 : decimal_constructor = PyObject_GetAttrString(decimal_module, "Decimal");
591 1 : if (!decimal_constructor)
592 0 : PLy_elog(ERROR, "no Decimal attribute in module");
593 : }
594 :
595 7 : str = DatumGetCString(DirectFunctionCall1(numeric_out, d));
596 7 : pyvalue = PyObject_CallFunction(decimal_constructor, "s", str);
597 7 : if (!pyvalue)
598 0 : PLy_elog(ERROR, "conversion from numeric to Decimal failed");
599 :
600 7 : return pyvalue;
601 : }
602 :
603 : static PyObject *
604 7 : PLyLong_FromInt16(PLyDatumToOb *arg, Datum d)
605 : {
606 7 : return PyLong_FromLong(DatumGetInt16(d));
607 : }
608 :
609 : static PyObject *
610 409 : PLyLong_FromInt32(PLyDatumToOb *arg, Datum d)
611 : {
612 409 : return PyLong_FromLong(DatumGetInt32(d));
613 : }
614 :
615 : static PyObject *
616 15 : PLyLong_FromInt64(PLyDatumToOb *arg, Datum d)
617 : {
618 15 : return PyLong_FromLongLong(DatumGetInt64(d));
619 : }
620 :
621 : static PyObject *
622 2 : PLyLong_FromOid(PLyDatumToOb *arg, Datum d)
623 : {
624 2 : return PyLong_FromUnsignedLong(DatumGetObjectId(d));
625 : }
626 :
627 : static PyObject *
628 11 : PLyBytes_FromBytea(PLyDatumToOb *arg, Datum d)
629 : {
630 11 : text *txt = DatumGetByteaPP(d);
631 11 : char *str = VARDATA_ANY(txt);
632 11 : size_t size = VARSIZE_ANY_EXHDR(txt);
633 :
634 11 : return PyBytes_FromStringAndSize(str, size);
635 : }
636 :
637 :
638 : /*
639 : * Generic input conversion using a SQL type's output function.
640 : */
641 : static PyObject *
642 487 : PLyUnicode_FromScalar(PLyDatumToOb *arg, Datum d)
643 : {
644 487 : char *x = OutputFunctionCall(&arg->scalar.typfunc, d);
645 487 : PyObject *r = PLyUnicode_FromString(x);
646 :
647 487 : pfree(x);
648 487 : return r;
649 : }
650 :
651 : /*
652 : * Convert using a from-SQL transform function.
653 : */
654 : static PyObject *
655 35 : PLyObject_FromTransform(PLyDatumToOb *arg, Datum d)
656 : {
657 : Datum t;
658 :
659 35 : t = FunctionCall1(&arg->transform.typtransform, d);
660 35 : return (PyObject *) DatumGetPointer(t);
661 : }
662 :
663 : /*
664 : * Convert a SQL array to a Python list.
665 : */
666 : static PyObject *
667 21 : PLyList_FromArray(PLyDatumToOb *arg, Datum d)
668 : {
669 21 : ArrayType *array = DatumGetArrayTypeP(d);
670 21 : PLyDatumToOb *elm = arg->array.elm;
671 : int ndim;
672 : int *dims;
673 : char *dataptr;
674 : bits8 *bitmap;
675 : int bitmask;
676 :
677 21 : if (ARR_NDIM(array) == 0)
678 1 : return PyList_New(0);
679 :
680 : /* Array dimensions and left bounds */
681 20 : ndim = ARR_NDIM(array);
682 20 : dims = ARR_DIMS(array);
683 : Assert(ndim <= MAXDIM);
684 :
685 : /*
686 : * We iterate the SQL array in the physical order it's stored in the
687 : * datum. For example, for a 3-dimensional array the order of iteration
688 : * would be the following: [0,0,0] elements through [0,0,k], then [0,1,0]
689 : * through [0,1,k] till [0,m,k], then [1,0,0] through [1,0,k] till
690 : * [1,m,k], and so on.
691 : *
692 : * In Python, there are no multi-dimensional lists as such, but they are
693 : * represented as a list of lists. So a 3-d array of [n,m,k] elements is a
694 : * list of n m-element arrays, each element of which is k-element array.
695 : * PLyList_FromArray_recurse() builds the Python list for a single
696 : * dimension, and recurses for the next inner dimension.
697 : */
698 20 : dataptr = ARR_DATA_PTR(array);
699 20 : bitmap = ARR_NULLBITMAP(array);
700 20 : bitmask = 1;
701 :
702 20 : return PLyList_FromArray_recurse(elm, dims, ndim, 0,
703 : &dataptr, &bitmap, &bitmask);
704 : }
705 :
706 : static PyObject *
707 48 : PLyList_FromArray_recurse(PLyDatumToOb *elm, int *dims, int ndim, int dim,
708 : char **dataptr_p, bits8 **bitmap_p, int *bitmask_p)
709 : {
710 : int i;
711 : PyObject *list;
712 :
713 48 : list = PyList_New(dims[dim]);
714 48 : if (!list)
715 0 : return NULL;
716 :
717 48 : if (dim < ndim - 1)
718 : {
719 : /* Outer dimension. Recurse for each inner slice. */
720 42 : for (i = 0; i < dims[dim]; i++)
721 : {
722 : PyObject *sublist;
723 :
724 28 : sublist = PLyList_FromArray_recurse(elm, dims, ndim, dim + 1,
725 : dataptr_p, bitmap_p, bitmask_p);
726 28 : PyList_SetItem(list, i, sublist);
727 : }
728 : }
729 : else
730 : {
731 : /*
732 : * Innermost dimension. Fill the list with the values from the array
733 : * for this slice.
734 : */
735 34 : char *dataptr = *dataptr_p;
736 34 : bits8 *bitmap = *bitmap_p;
737 34 : int bitmask = *bitmask_p;
738 34 : uint8 typalignby = typalign_to_alignby(elm->typalign);
739 :
740 120 : for (i = 0; i < dims[dim]; i++)
741 : {
742 : /* checking for NULL */
743 86 : if (bitmap && (*bitmap & bitmask) == 0)
744 : {
745 : Py_INCREF(Py_None);
746 14 : PyList_SetItem(list, i, Py_None);
747 : }
748 : else
749 : {
750 : Datum itemvalue;
751 :
752 72 : itemvalue = fetch_att(dataptr, elm->typbyval, elm->typlen);
753 72 : PyList_SetItem(list, i, elm->func(elm, itemvalue));
754 72 : dataptr = att_addlength_pointer(dataptr, elm->typlen, dataptr);
755 72 : dataptr = (char *) att_nominal_alignby(dataptr, typalignby);
756 : }
757 :
758 : /* advance bitmap pointer if any */
759 86 : if (bitmap)
760 : {
761 52 : bitmask <<= 1;
762 52 : if (bitmask == 0x100 /* (1<<8) */ )
763 : {
764 4 : bitmap++;
765 4 : bitmask = 1;
766 : }
767 : }
768 : }
769 :
770 34 : *dataptr_p = dataptr;
771 34 : *bitmap_p = bitmap;
772 34 : *bitmask_p = bitmask;
773 : }
774 :
775 48 : return list;
776 : }
777 :
778 : /*
779 : * Convert a composite SQL value to a Python dict.
780 : */
781 : static PyObject *
782 49 : PLyDict_FromComposite(PLyDatumToOb *arg, Datum d)
783 : {
784 : PyObject *dict;
785 : HeapTupleHeader td;
786 : Oid tupType;
787 : int32 tupTypmod;
788 : TupleDesc tupdesc;
789 : HeapTupleData tmptup;
790 :
791 49 : td = DatumGetHeapTupleHeader(d);
792 : /* Extract rowtype info and find a tupdesc */
793 49 : tupType = HeapTupleHeaderGetTypeId(td);
794 49 : tupTypmod = HeapTupleHeaderGetTypMod(td);
795 49 : tupdesc = lookup_rowtype_tupdesc(tupType, tupTypmod);
796 :
797 : /* Set up I/O funcs if not done yet */
798 49 : PLy_input_setup_tuple(arg, tupdesc,
799 49 : PLy_current_execution_context()->curr_proc);
800 :
801 : /* Build a temporary HeapTuple control structure */
802 49 : tmptup.t_len = HeapTupleHeaderGetDatumLength(td);
803 49 : tmptup.t_data = td;
804 :
805 49 : dict = PLyDict_FromTuple(arg, &tmptup, tupdesc, true);
806 :
807 49 : ReleaseTupleDesc(tupdesc);
808 :
809 49 : return dict;
810 : }
811 :
812 : /*
813 : * Transform a tuple into a Python dict object.
814 : */
815 : static PyObject *
816 248 : PLyDict_FromTuple(PLyDatumToOb *arg, HeapTuple tuple, TupleDesc desc, bool include_generated)
817 : {
818 : PyObject *volatile dict;
819 :
820 : /* Simple sanity check that desc matches */
821 : Assert(desc->natts == arg->tuple.natts);
822 :
823 248 : dict = PyDict_New();
824 248 : if (dict == NULL)
825 0 : return NULL;
826 :
827 248 : PG_TRY();
828 : {
829 : int i;
830 :
831 781 : for (i = 0; i < arg->tuple.natts; i++)
832 : {
833 533 : PLyDatumToOb *att = &arg->tuple.atts[i];
834 533 : Form_pg_attribute attr = TupleDescAttr(desc, i);
835 : char *key;
836 : Datum vattr;
837 : bool is_null;
838 : PyObject *value;
839 :
840 533 : if (attr->attisdropped)
841 15 : continue;
842 :
843 530 : if (attr->attgenerated)
844 : {
845 : /* don't include unless requested */
846 18 : if (!include_generated)
847 6 : continue;
848 : /* never include virtual columns */
849 12 : if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL)
850 6 : continue;
851 : }
852 :
853 518 : key = NameStr(attr->attname);
854 518 : vattr = heap_getattr(tuple, (i + 1), desc, &is_null);
855 :
856 518 : if (is_null)
857 13 : PyDict_SetItemString(dict, key, Py_None);
858 : else
859 : {
860 505 : value = att->func(att, vattr);
861 505 : PyDict_SetItemString(dict, key, value);
862 : Py_DECREF(value);
863 : }
864 : }
865 : }
866 0 : PG_CATCH();
867 : {
868 0 : Py_DECREF(dict);
869 0 : PG_RE_THROW();
870 : }
871 248 : PG_END_TRY();
872 :
873 248 : return dict;
874 : }
875 :
876 : /*
877 : * Convert a Python object to a PostgreSQL bool datum. This can't go
878 : * through the generic conversion function, because Python attaches a
879 : * Boolean value to everything, more things than the PostgreSQL bool
880 : * type can parse.
881 : */
882 : static Datum
883 23 : PLyObject_ToBool(PLyObToDatum *arg, PyObject *plrv,
884 : bool *isnull, bool inarray)
885 : {
886 23 : if (plrv == Py_None)
887 : {
888 1 : *isnull = true;
889 1 : return (Datum) 0;
890 : }
891 22 : *isnull = false;
892 22 : return BoolGetDatum(PyObject_IsTrue(plrv));
893 : }
894 :
895 : /*
896 : * Convert a Python object to a PostgreSQL bytea datum. This doesn't
897 : * go through the generic conversion function to circumvent problems
898 : * with embedded nulls. And it's faster this way.
899 : */
900 : static Datum
901 11 : PLyObject_ToBytea(PLyObToDatum *arg, PyObject *plrv,
902 : bool *isnull, bool inarray)
903 : {
904 11 : PyObject *volatile plrv_so = NULL;
905 11 : Datum rv = (Datum) 0;
906 :
907 11 : if (plrv == Py_None)
908 : {
909 3 : *isnull = true;
910 3 : return (Datum) 0;
911 : }
912 8 : *isnull = false;
913 :
914 8 : plrv_so = PyObject_Bytes(plrv);
915 8 : if (!plrv_so)
916 0 : PLy_elog(ERROR, "could not create bytes representation of Python object");
917 :
918 8 : PG_TRY();
919 : {
920 8 : char *plrv_sc = PyBytes_AsString(plrv_so);
921 8 : size_t len = PyBytes_Size(plrv_so);
922 8 : size_t size = len + VARHDRSZ;
923 8 : bytea *result = palloc(size);
924 :
925 8 : SET_VARSIZE(result, size);
926 8 : memcpy(VARDATA(result), plrv_sc, len);
927 8 : rv = PointerGetDatum(result);
928 : }
929 0 : PG_FINALLY();
930 : {
931 8 : Py_XDECREF(plrv_so);
932 : }
933 8 : PG_END_TRY();
934 :
935 8 : return rv;
936 : }
937 :
938 :
939 : /*
940 : * Convert a Python object to a composite type. First look up the type's
941 : * description, then route the Python object through the conversion function
942 : * for obtaining PostgreSQL tuples.
943 : */
944 : static Datum
945 291 : PLyObject_ToComposite(PLyObToDatum *arg, PyObject *plrv,
946 : bool *isnull, bool inarray)
947 : {
948 : Datum rv;
949 : TupleDesc desc;
950 :
951 291 : if (plrv == Py_None)
952 : {
953 21 : *isnull = true;
954 21 : return (Datum) 0;
955 : }
956 270 : *isnull = false;
957 :
958 : /*
959 : * The string conversion case doesn't require a tupdesc, nor per-field
960 : * conversion data, so just go for it if that's the case to use.
961 : */
962 270 : if (PyUnicode_Check(plrv))
963 18 : return PLyUnicode_ToComposite(arg, plrv, inarray);
964 :
965 : /*
966 : * If we're dealing with a named composite type, we must look up the
967 : * tupdesc every time, to protect against possible changes to the type.
968 : * RECORD types can't change between calls; but we must still be willing
969 : * to set up the info the first time, if nobody did yet.
970 : */
971 252 : if (arg->typoid != RECORDOID)
972 : {
973 125 : desc = lookup_rowtype_tupdesc(arg->typoid, arg->typmod);
974 : /* We should have the descriptor of the type's typcache entry */
975 : Assert(desc == arg->tuple.typentry->tupDesc);
976 : /* Detect change of descriptor, update cache if needed */
977 125 : if (arg->tuple.tupdescid != arg->tuple.typentry->tupDesc_identifier)
978 : {
979 31 : PLy_output_setup_tuple(arg, desc,
980 31 : PLy_current_execution_context()->curr_proc);
981 31 : arg->tuple.tupdescid = arg->tuple.typentry->tupDesc_identifier;
982 : }
983 : }
984 : else
985 : {
986 127 : desc = arg->tuple.recdesc;
987 127 : if (desc == NULL)
988 : {
989 28 : desc = lookup_rowtype_tupdesc(arg->typoid, arg->typmod);
990 28 : arg->tuple.recdesc = desc;
991 : }
992 : else
993 : {
994 : /* Pin descriptor to match unpin below */
995 99 : PinTupleDesc(desc);
996 : }
997 : }
998 :
999 : /* Simple sanity check on our caching */
1000 : Assert(desc->natts == arg->tuple.natts);
1001 :
1002 : /*
1003 : * Convert, using the appropriate method depending on the type of the
1004 : * supplied Python object.
1005 : */
1006 252 : if (PySequence_Check(plrv))
1007 : /* composite type as sequence (tuple, list etc) */
1008 132 : rv = PLySequence_ToComposite(arg, desc, plrv);
1009 120 : else if (PyMapping_Check(plrv))
1010 : /* composite type as mapping (currently only dict) */
1011 95 : rv = PLyMapping_ToComposite(arg, desc, plrv);
1012 : else
1013 : /* returned as smth, must provide method __getattr__(name) */
1014 25 : rv = PLyGenericObject_ToComposite(arg, desc, plrv, inarray);
1015 :
1016 243 : ReleaseTupleDesc(desc);
1017 :
1018 243 : return rv;
1019 : }
1020 :
1021 :
1022 : /*
1023 : * Convert Python object to C string in server encoding.
1024 : *
1025 : * Note: this is exported for use by add-on transform modules.
1026 : */
1027 : char *
1028 1584 : PLyObject_AsString(PyObject *plrv)
1029 : {
1030 : PyObject *plrv_bo;
1031 : char *plrv_sc;
1032 : size_t plen;
1033 : size_t slen;
1034 :
1035 1584 : if (PyUnicode_Check(plrv))
1036 341 : plrv_bo = PLyUnicode_Bytes(plrv);
1037 1243 : else if (PyFloat_Check(plrv))
1038 : {
1039 : /* use repr() for floats, str() is lossy */
1040 7 : PyObject *s = PyObject_Repr(plrv);
1041 :
1042 7 : plrv_bo = PLyUnicode_Bytes(s);
1043 7 : Py_XDECREF(s);
1044 : }
1045 : else
1046 : {
1047 1236 : PyObject *s = PyObject_Str(plrv);
1048 :
1049 1236 : plrv_bo = PLyUnicode_Bytes(s);
1050 1236 : Py_XDECREF(s);
1051 : }
1052 1584 : if (!plrv_bo)
1053 0 : PLy_elog(ERROR, "could not create string representation of Python object");
1054 :
1055 1584 : plrv_sc = pstrdup(PyBytes_AsString(plrv_bo));
1056 1584 : plen = PyBytes_Size(plrv_bo);
1057 1584 : slen = strlen(plrv_sc);
1058 :
1059 1584 : Py_XDECREF(plrv_bo);
1060 :
1061 1584 : if (slen < plen)
1062 0 : ereport(ERROR,
1063 : (errcode(ERRCODE_DATATYPE_MISMATCH),
1064 : errmsg("could not convert Python object into cstring: Python string representation appears to contain null bytes")));
1065 1584 : else if (slen > plen)
1066 0 : elog(ERROR, "could not convert Python object into cstring: Python string longer than reported length");
1067 1584 : pg_verifymbstr(plrv_sc, slen, false);
1068 :
1069 1584 : return plrv_sc;
1070 : }
1071 :
1072 :
1073 : /*
1074 : * Generic output conversion function: convert PyObject to cstring and
1075 : * cstring into PostgreSQL type.
1076 : */
1077 : static Datum
1078 1603 : PLyObject_ToScalar(PLyObToDatum *arg, PyObject *plrv,
1079 : bool *isnull, bool inarray)
1080 : {
1081 : char *str;
1082 :
1083 1603 : if (plrv == Py_None)
1084 : {
1085 98 : *isnull = true;
1086 98 : return (Datum) 0;
1087 : }
1088 1505 : *isnull = false;
1089 :
1090 1505 : str = PLyObject_AsString(plrv);
1091 :
1092 1505 : return InputFunctionCall(&arg->scalar.typfunc,
1093 : str,
1094 : arg->scalar.typioparam,
1095 : arg->typmod);
1096 : }
1097 :
1098 :
1099 : /*
1100 : * Convert to a domain type.
1101 : */
1102 : static Datum
1103 29 : PLyObject_ToDomain(PLyObToDatum *arg, PyObject *plrv,
1104 : bool *isnull, bool inarray)
1105 : {
1106 : Datum result;
1107 29 : PLyObToDatum *base = arg->domain.base;
1108 :
1109 29 : result = base->func(base, plrv, isnull, inarray);
1110 27 : domain_check(result, *isnull, arg->typoid,
1111 : &arg->domain.domain_info, arg->mcxt);
1112 16 : return result;
1113 : }
1114 :
1115 :
1116 : /*
1117 : * Convert using a to-SQL transform function.
1118 : */
1119 : static Datum
1120 31 : PLyObject_ToTransform(PLyObToDatum *arg, PyObject *plrv,
1121 : bool *isnull, bool inarray)
1122 : {
1123 31 : if (plrv == Py_None)
1124 : {
1125 1 : *isnull = true;
1126 1 : return (Datum) 0;
1127 : }
1128 30 : *isnull = false;
1129 30 : return FunctionCall1(&arg->transform.typtransform, PointerGetDatum(plrv));
1130 : }
1131 :
1132 :
1133 : /*
1134 : * Convert Python sequence (or list of lists) to SQL array.
1135 : */
1136 : static Datum
1137 58 : PLySequence_ToArray(PLyObToDatum *arg, PyObject *plrv,
1138 : bool *isnull, bool inarray)
1139 : {
1140 58 : ArrayBuildState *astate = NULL;
1141 58 : int ndims = 1;
1142 : int dims[MAXDIM];
1143 : int lbs[MAXDIM];
1144 :
1145 58 : if (plrv == Py_None)
1146 : {
1147 2 : *isnull = true;
1148 2 : return (Datum) 0;
1149 : }
1150 56 : *isnull = false;
1151 :
1152 : /*
1153 : * For historical reasons, we allow any sequence (not only a list) at the
1154 : * top level when converting a Python object to a SQL array. However, a
1155 : * multi-dimensional array is recognized only when the object contains
1156 : * true lists.
1157 : */
1158 56 : if (!PySequence_Check(plrv))
1159 3 : ereport(ERROR,
1160 : (errcode(ERRCODE_DATATYPE_MISMATCH),
1161 : errmsg("return value of function with array return type is not a Python sequence")));
1162 :
1163 : /* Initialize dimensionality info with first-level dimension */
1164 53 : memset(dims, 0, sizeof(dims));
1165 53 : dims[0] = PySequence_Length(plrv);
1166 :
1167 : /*
1168 : * Traverse the Python lists, in depth-first order, and collect all the
1169 : * elements at the bottom level into an ArrayBuildState.
1170 : */
1171 53 : PLySequence_ToArray_recurse(plrv, &astate,
1172 : &ndims, dims, 1,
1173 : arg->array.elm,
1174 : arg->array.elmbasetype);
1175 :
1176 : /* ensure we get zero-D array for no inputs, as per PG convention */
1177 39 : if (astate == NULL)
1178 2 : return PointerGetDatum(construct_empty_array(arg->array.elmbasetype));
1179 :
1180 98 : for (int i = 0; i < ndims; i++)
1181 61 : lbs[i] = 1;
1182 :
1183 37 : return makeMdArrayResult(astate, ndims, dims, lbs,
1184 : CurrentMemoryContext, true);
1185 : }
1186 :
1187 : /*
1188 : * Helper function for PLySequence_ToArray. Traverse a Python list of lists in
1189 : * depth-first order, storing the elements in *astatep.
1190 : *
1191 : * The ArrayBuildState is created only when we first find a scalar element;
1192 : * if we didn't do it like that, we'd need some other convention for knowing
1193 : * whether we'd already found any scalars (and thus the number of dimensions
1194 : * is frozen).
1195 : */
1196 : static void
1197 255 : PLySequence_ToArray_recurse(PyObject *obj, ArrayBuildState **astatep,
1198 : int *ndims, int *dims, int cur_depth,
1199 : PLyObToDatum *elm, Oid elmbasetype)
1200 : {
1201 : int i;
1202 255 : int len = PySequence_Length(obj);
1203 :
1204 : /* We should not get here with a non-sequence object */
1205 255 : if (len < 0)
1206 0 : PLy_elog(ERROR, "could not determine sequence length for function return value");
1207 :
1208 1358 : for (i = 0; i < len; i++)
1209 : {
1210 : /* fetch the array element */
1211 1126 : PyObject *subobj = PySequence_GetItem(obj, i);
1212 :
1213 : /* need PG_TRY to ensure we release the subobj's refcount */
1214 1126 : PG_TRY();
1215 : {
1216 : /* multi-dimensional array? */
1217 1126 : if (PyList_Check(subobj))
1218 : {
1219 : /* set size when at first element in this level, else compare */
1220 208 : if (i == 0 && *ndims == cur_depth)
1221 : {
1222 : /* array after some scalars at same level? */
1223 40 : if (*astatep != NULL)
1224 1 : ereport(ERROR,
1225 : (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
1226 : errmsg("multidimensional arrays must have array expressions with matching dimensions")));
1227 : /* too many dimensions? */
1228 39 : if (cur_depth >= MAXDIM)
1229 1 : ereport(ERROR,
1230 : (errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
1231 : errmsg("number of array dimensions exceeds the maximum allowed (%d)",
1232 : MAXDIM)));
1233 : /* OK, add a dimension */
1234 38 : dims[*ndims] = PySequence_Length(subobj);
1235 38 : (*ndims)++;
1236 : }
1237 168 : else if (cur_depth >= *ndims ||
1238 166 : PySequence_Length(subobj) != dims[cur_depth])
1239 4 : ereport(ERROR,
1240 : (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
1241 : errmsg("multidimensional arrays must have array expressions with matching dimensions")));
1242 :
1243 : /* recurse to fetch elements of this sub-array */
1244 202 : PLySequence_ToArray_recurse(subobj, astatep,
1245 : ndims, dims, cur_depth + 1,
1246 : elm, elmbasetype);
1247 : }
1248 : else
1249 : {
1250 : Datum dat;
1251 : bool isnull;
1252 :
1253 : /* scalar after some sub-arrays at same level? */
1254 918 : if (*ndims != cur_depth)
1255 2 : ereport(ERROR,
1256 : (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
1257 : errmsg("multidimensional arrays must have array expressions with matching dimensions")));
1258 :
1259 : /* convert non-list object to Datum */
1260 916 : dat = elm->func(elm, subobj, &isnull, true);
1261 :
1262 : /* create ArrayBuildState if we didn't already */
1263 910 : if (*astatep == NULL)
1264 43 : *astatep = initArrayResult(elmbasetype,
1265 : CurrentMemoryContext, true);
1266 :
1267 : /* ... and save the element value in it */
1268 910 : (void) accumArrayResult(*astatep, dat, isnull,
1269 : elmbasetype, CurrentMemoryContext);
1270 : }
1271 : }
1272 23 : PG_FINALLY();
1273 : {
1274 1126 : Py_XDECREF(subobj);
1275 : }
1276 1126 : PG_END_TRY();
1277 : }
1278 232 : }
1279 :
1280 :
1281 : /*
1282 : * Convert a Python string to composite, using record_in.
1283 : */
1284 : static Datum
1285 18 : PLyUnicode_ToComposite(PLyObToDatum *arg, PyObject *string, bool inarray)
1286 : {
1287 : char *str;
1288 :
1289 : /*
1290 : * Set up call data for record_in, if we didn't already. (We can't just
1291 : * use DirectFunctionCall, because record_in needs a fn_extra field.)
1292 : */
1293 18 : if (!OidIsValid(arg->tuple.recinfunc.fn_oid))
1294 5 : fmgr_info_cxt(F_RECORD_IN, &arg->tuple.recinfunc, arg->mcxt);
1295 :
1296 18 : str = PLyObject_AsString(string);
1297 :
1298 : /*
1299 : * If we are parsing a composite type within an array, and the string
1300 : * isn't a valid record literal, there's a high chance that the function
1301 : * did something like:
1302 : *
1303 : * CREATE FUNCTION .. RETURNS comptype[] AS $$ return [['foo', 'bar']] $$
1304 : * LANGUAGE plpython;
1305 : *
1306 : * Before PostgreSQL 10, that was interpreted as a single-dimensional
1307 : * array, containing record ('foo', 'bar'). PostgreSQL 10 added support
1308 : * for multi-dimensional arrays, and it is now interpreted as a
1309 : * two-dimensional array, containing two records, 'foo', and 'bar'.
1310 : * record_in() will throw an error, because "foo" is not a valid record
1311 : * literal.
1312 : *
1313 : * To make that less confusing to users who are upgrading from older
1314 : * versions, try to give a hint in the typical instances of that. If we
1315 : * are parsing an array of composite types, and we see a string literal
1316 : * that is not a valid record literal, give a hint. We only want to give
1317 : * the hint in the narrow case of a malformed string literal, not any
1318 : * error from record_in(), so check for that case here specifically.
1319 : *
1320 : * This check better match the one in record_in(), so that we don't forbid
1321 : * literals that are actually valid!
1322 : */
1323 18 : if (inarray)
1324 : {
1325 1 : char *ptr = str;
1326 :
1327 : /* Allow leading whitespace */
1328 1 : while (*ptr && isspace((unsigned char) *ptr))
1329 0 : ptr++;
1330 1 : if (*ptr++ != '(')
1331 1 : ereport(ERROR,
1332 : (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
1333 : errmsg("malformed record literal: \"%s\"", str),
1334 : errdetail("Missing left parenthesis."),
1335 : errhint("To return a composite type in an array, return the composite type as a Python tuple, e.g., \"[('foo',)]\".")));
1336 : }
1337 :
1338 17 : return InputFunctionCall(&arg->tuple.recinfunc,
1339 : str,
1340 : arg->typoid,
1341 : arg->typmod);
1342 : }
1343 :
1344 :
1345 : static Datum
1346 95 : PLyMapping_ToComposite(PLyObToDatum *arg, TupleDesc desc, PyObject *mapping)
1347 : {
1348 : Datum result;
1349 : HeapTuple tuple;
1350 : Datum *values;
1351 : bool *nulls;
1352 : volatile int i;
1353 :
1354 : Assert(PyMapping_Check(mapping));
1355 :
1356 : /* Build tuple */
1357 95 : values = palloc_array(Datum, desc->natts);
1358 95 : nulls = palloc_array(bool, desc->natts);
1359 365 : for (i = 0; i < desc->natts; ++i)
1360 : {
1361 : char *key;
1362 : PyObject *volatile value;
1363 : PLyObToDatum *att;
1364 273 : Form_pg_attribute attr = TupleDescAttr(desc, i);
1365 :
1366 273 : if (attr->attisdropped)
1367 : {
1368 47 : values[i] = (Datum) 0;
1369 47 : nulls[i] = true;
1370 47 : continue;
1371 : }
1372 :
1373 226 : key = NameStr(attr->attname);
1374 226 : value = NULL;
1375 226 : att = &arg->tuple.atts[i];
1376 226 : PG_TRY();
1377 : {
1378 226 : value = PyMapping_GetItemString(mapping, key);
1379 226 : if (!value)
1380 2 : ereport(ERROR,
1381 : (errcode(ERRCODE_UNDEFINED_COLUMN),
1382 : errmsg("key \"%s\" not found in mapping", key),
1383 : errhint("To return null in a column, "
1384 : "add the value None to the mapping with the key named after the column.")));
1385 :
1386 224 : values[i] = att->func(att, value, &nulls[i], false);
1387 :
1388 223 : Py_XDECREF(value);
1389 223 : value = NULL;
1390 : }
1391 3 : PG_CATCH();
1392 : {
1393 3 : Py_XDECREF(value);
1394 3 : PG_RE_THROW();
1395 : }
1396 223 : PG_END_TRY();
1397 : }
1398 :
1399 92 : tuple = heap_form_tuple(desc, values, nulls);
1400 92 : result = heap_copy_tuple_as_datum(tuple, desc);
1401 92 : heap_freetuple(tuple);
1402 :
1403 92 : pfree(values);
1404 92 : pfree(nulls);
1405 :
1406 92 : return result;
1407 : }
1408 :
1409 :
1410 : static Datum
1411 132 : PLySequence_ToComposite(PLyObToDatum *arg, TupleDesc desc, PyObject *sequence)
1412 : {
1413 : Datum result;
1414 : HeapTuple tuple;
1415 : Datum *values;
1416 : bool *nulls;
1417 : volatile int idx;
1418 : volatile int i;
1419 :
1420 : Assert(PySequence_Check(sequence));
1421 :
1422 : /*
1423 : * Check that sequence length is exactly same as PG tuple's. We actually
1424 : * can ignore exceeding items or assume missing ones as null but to avoid
1425 : * plpython developer's errors we are strict here
1426 : */
1427 132 : idx = 0;
1428 443 : for (i = 0; i < desc->natts; i++)
1429 : {
1430 311 : if (!TupleDescCompactAttr(desc, i)->attisdropped)
1431 260 : idx++;
1432 : }
1433 132 : if (PySequence_Length(sequence) != idx)
1434 3 : ereport(ERROR,
1435 : (errcode(ERRCODE_DATATYPE_MISMATCH),
1436 : errmsg("length of returned sequence did not match number of columns in row")));
1437 :
1438 : /* Build tuple */
1439 129 : values = palloc_array(Datum, desc->natts);
1440 129 : nulls = palloc_array(bool, desc->natts);
1441 129 : idx = 0;
1442 430 : for (i = 0; i < desc->natts; ++i)
1443 : {
1444 : PyObject *volatile value;
1445 : PLyObToDatum *att;
1446 :
1447 303 : if (TupleDescCompactAttr(desc, i)->attisdropped)
1448 : {
1449 47 : values[i] = (Datum) 0;
1450 47 : nulls[i] = true;
1451 47 : continue;
1452 : }
1453 :
1454 256 : value = NULL;
1455 256 : att = &arg->tuple.atts[i];
1456 256 : PG_TRY();
1457 : {
1458 256 : value = PySequence_GetItem(sequence, idx);
1459 : Assert(value);
1460 :
1461 256 : values[i] = att->func(att, value, &nulls[i], false);
1462 :
1463 254 : Py_XDECREF(value);
1464 254 : value = NULL;
1465 : }
1466 2 : PG_CATCH();
1467 : {
1468 2 : Py_XDECREF(value);
1469 2 : PG_RE_THROW();
1470 : }
1471 254 : PG_END_TRY();
1472 :
1473 254 : idx++;
1474 : }
1475 :
1476 127 : tuple = heap_form_tuple(desc, values, nulls);
1477 127 : result = heap_copy_tuple_as_datum(tuple, desc);
1478 127 : heap_freetuple(tuple);
1479 :
1480 127 : pfree(values);
1481 127 : pfree(nulls);
1482 :
1483 127 : return result;
1484 : }
1485 :
1486 :
1487 : static Datum
1488 25 : PLyGenericObject_ToComposite(PLyObToDatum *arg, TupleDesc desc, PyObject *object, bool inarray)
1489 : {
1490 : Datum result;
1491 : HeapTuple tuple;
1492 : Datum *values;
1493 : bool *nulls;
1494 : volatile int i;
1495 :
1496 : /* Build tuple */
1497 25 : values = palloc_array(Datum, desc->natts);
1498 25 : nulls = palloc_array(bool, desc->natts);
1499 98 : for (i = 0; i < desc->natts; ++i)
1500 : {
1501 : char *key;
1502 : PyObject *volatile value;
1503 : PLyObToDatum *att;
1504 74 : Form_pg_attribute attr = TupleDescAttr(desc, i);
1505 :
1506 74 : if (attr->attisdropped)
1507 : {
1508 24 : values[i] = (Datum) 0;
1509 24 : nulls[i] = true;
1510 24 : continue;
1511 : }
1512 :
1513 50 : key = NameStr(attr->attname);
1514 50 : value = NULL;
1515 50 : att = &arg->tuple.atts[i];
1516 50 : PG_TRY();
1517 : {
1518 50 : value = PyObject_GetAttrString(object, key);
1519 50 : if (!value)
1520 : {
1521 : /*
1522 : * No attribute for this column in the object.
1523 : *
1524 : * If we are parsing a composite type in an array, a likely
1525 : * cause is that the function contained something like "[[123,
1526 : * 'foo']]". Before PostgreSQL 10, that was interpreted as an
1527 : * array, with a composite type (123, 'foo') in it. But now
1528 : * it's interpreted as a two-dimensional array, and we try to
1529 : * interpret "123" as the composite type. See also similar
1530 : * heuristic in PLyObject_ToScalar().
1531 : */
1532 1 : ereport(ERROR,
1533 : (errcode(ERRCODE_UNDEFINED_COLUMN),
1534 : errmsg("attribute \"%s\" does not exist in Python object", key),
1535 : inarray ?
1536 : errhint("To return a composite type in an array, return the composite type as a Python tuple, e.g., \"[('foo',)]\".") :
1537 : errhint("To return null in a column, let the returned object have an attribute named after column with value None.")));
1538 : }
1539 :
1540 49 : values[i] = att->func(att, value, &nulls[i], false);
1541 :
1542 49 : Py_XDECREF(value);
1543 49 : value = NULL;
1544 : }
1545 1 : PG_CATCH();
1546 : {
1547 1 : Py_XDECREF(value);
1548 1 : PG_RE_THROW();
1549 : }
1550 49 : PG_END_TRY();
1551 : }
1552 :
1553 24 : tuple = heap_form_tuple(desc, values, nulls);
1554 24 : result = heap_copy_tuple_as_datum(tuple, desc);
1555 24 : heap_freetuple(tuple);
1556 :
1557 24 : pfree(values);
1558 24 : pfree(nulls);
1559 :
1560 24 : return result;
1561 : }
|
