Line data Source code
1 : /*--------------------------------------------------------------------------
2 : *
3 : * spgist_name_ops.c
4 : * Test opclass for SP-GiST
5 : *
6 : * This indexes input values of type "name", but the index storage is "text",
7 : * with the same choices as made in the core SP-GiST text_ops opclass.
8 : * Much of the code is identical to src/backend/access/spgist/spgtextproc.c,
9 : * which see for a more detailed header comment.
10 : *
11 : * Unlike spgtextproc.c, we don't bother with collation-aware logic.
12 : *
13 : *
14 : * Portions Copyright (c) 1996-2024, PostgreSQL Global Development Group
15 : * Portions Copyright (c) 1994, Regents of the University of California
16 : *
17 : * IDENTIFICATION
18 : * src/test/modules/spgist_name_ops/spgist_name_ops.c
19 : *
20 : * -------------------------------------------------------------------------
21 : */
22 : #include "postgres.h"
23 :
24 : #include "access/spgist.h"
25 : #include "catalog/pg_type.h"
26 : #include "utils/datum.h"
27 : #include "varatt.h"
28 :
29 2 : PG_MODULE_MAGIC;
30 :
31 :
32 4 : PG_FUNCTION_INFO_V1(spgist_name_config);
33 : Datum
34 14 : spgist_name_config(PG_FUNCTION_ARGS)
35 : {
36 : /* spgConfigIn *cfgin = (spgConfigIn *) PG_GETARG_POINTER(0); */
37 14 : spgConfigOut *cfg = (spgConfigOut *) PG_GETARG_POINTER(1);
38 :
39 14 : cfg->prefixType = TEXTOID;
40 14 : cfg->labelType = INT2OID;
41 14 : cfg->leafType = TEXTOID;
42 14 : cfg->canReturnData = true;
43 14 : cfg->longValuesOK = true; /* suffixing will shorten long values */
44 14 : PG_RETURN_VOID();
45 : }
46 :
47 : /*
48 : * Form a text datum from the given not-necessarily-null-terminated string,
49 : * using short varlena header format if possible
50 : */
51 : static Datum
52 35924 : formTextDatum(const char *data, int datalen)
53 : {
54 : char *p;
55 :
56 35924 : p = (char *) palloc(datalen + VARHDRSZ);
57 :
58 35924 : if (datalen + VARHDRSZ_SHORT <= VARATT_SHORT_MAX)
59 : {
60 35924 : SET_VARSIZE_SHORT(p, datalen + VARHDRSZ_SHORT);
61 35924 : if (datalen)
62 35834 : memcpy(p + VARHDRSZ_SHORT, data, datalen);
63 : }
64 : else
65 : {
66 0 : SET_VARSIZE(p, datalen + VARHDRSZ);
67 0 : memcpy(p + VARHDRSZ, data, datalen);
68 : }
69 :
70 35924 : return PointerGetDatum(p);
71 : }
72 :
73 : /*
74 : * Find the length of the common prefix of a and b
75 : */
76 : static int
77 1940 : commonPrefix(const char *a, const char *b, int lena, int lenb)
78 : {
79 1940 : int i = 0;
80 :
81 4448 : while (i < lena && i < lenb && *a == *b)
82 : {
83 2508 : a++;
84 2508 : b++;
85 2508 : i++;
86 : }
87 :
88 1940 : return i;
89 : }
90 :
91 : /*
92 : * Binary search an array of int16 datums for a match to c
93 : *
94 : * On success, *i gets the match location; on failure, it gets where to insert
95 : */
96 : static bool
97 22628 : searchChar(Datum *nodeLabels, int nNodes, int16 c, int *i)
98 : {
99 22628 : int StopLow = 0,
100 22628 : StopHigh = nNodes;
101 :
102 72630 : while (StopLow < StopHigh)
103 : {
104 72398 : int StopMiddle = (StopLow + StopHigh) >> 1;
105 72398 : int16 middle = DatumGetInt16(nodeLabels[StopMiddle]);
106 :
107 72398 : if (c < middle)
108 29106 : StopHigh = StopMiddle;
109 43292 : else if (c > middle)
110 20896 : StopLow = StopMiddle + 1;
111 : else
112 : {
113 22396 : *i = StopMiddle;
114 22396 : return true;
115 : }
116 : }
117 :
118 232 : *i = StopHigh;
119 232 : return false;
120 : }
121 :
122 4 : PG_FUNCTION_INFO_V1(spgist_name_choose);
123 : Datum
124 22640 : spgist_name_choose(PG_FUNCTION_ARGS)
125 : {
126 22640 : spgChooseIn *in = (spgChooseIn *) PG_GETARG_POINTER(0);
127 22640 : spgChooseOut *out = (spgChooseOut *) PG_GETARG_POINTER(1);
128 22640 : Name inName = DatumGetName(in->datum);
129 22640 : char *inStr = NameStr(*inName);
130 22640 : int inSize = strlen(inStr);
131 22640 : char *prefixStr = NULL;
132 22640 : int prefixSize = 0;
133 22640 : int commonLen = 0;
134 22640 : int16 nodeChar = 0;
135 22640 : int i = 0;
136 :
137 : /* Check for prefix match, set nodeChar to first byte after prefix */
138 22640 : if (in->hasPrefix)
139 : {
140 1940 : text *prefixText = DatumGetTextPP(in->prefixDatum);
141 :
142 1940 : prefixStr = VARDATA_ANY(prefixText);
143 1940 : prefixSize = VARSIZE_ANY_EXHDR(prefixText);
144 :
145 1940 : commonLen = commonPrefix(inStr + in->level,
146 : prefixStr,
147 1940 : inSize - in->level,
148 : prefixSize);
149 :
150 1940 : if (commonLen == prefixSize)
151 : {
152 1928 : if (inSize - in->level > commonLen)
153 1924 : nodeChar = *(unsigned char *) (inStr + in->level + commonLen);
154 : else
155 4 : nodeChar = -1;
156 : }
157 : else
158 : {
159 : /* Must split tuple because incoming value doesn't match prefix */
160 12 : out->resultType = spgSplitTuple;
161 :
162 12 : if (commonLen == 0)
163 : {
164 4 : out->result.splitTuple.prefixHasPrefix = false;
165 : }
166 : else
167 : {
168 8 : out->result.splitTuple.prefixHasPrefix = true;
169 8 : out->result.splitTuple.prefixPrefixDatum =
170 8 : formTextDatum(prefixStr, commonLen);
171 : }
172 12 : out->result.splitTuple.prefixNNodes = 1;
173 12 : out->result.splitTuple.prefixNodeLabels =
174 12 : (Datum *) palloc(sizeof(Datum));
175 24 : out->result.splitTuple.prefixNodeLabels[0] =
176 12 : Int16GetDatum(*(unsigned char *) (prefixStr + commonLen));
177 :
178 12 : out->result.splitTuple.childNodeN = 0;
179 :
180 12 : if (prefixSize - commonLen == 1)
181 : {
182 8 : out->result.splitTuple.postfixHasPrefix = false;
183 : }
184 : else
185 : {
186 4 : out->result.splitTuple.postfixHasPrefix = true;
187 4 : out->result.splitTuple.postfixPrefixDatum =
188 4 : formTextDatum(prefixStr + commonLen + 1,
189 4 : prefixSize - commonLen - 1);
190 : }
191 :
192 12 : PG_RETURN_VOID();
193 : }
194 : }
195 20700 : else if (inSize > in->level)
196 : {
197 20660 : nodeChar = *(unsigned char *) (inStr + in->level);
198 : }
199 : else
200 : {
201 40 : nodeChar = -1;
202 : }
203 :
204 : /* Look up nodeChar in the node label array */
205 22628 : if (searchChar(in->nodeLabels, in->nNodes, nodeChar, &i))
206 : {
207 : /*
208 : * Descend to existing node. (If in->allTheSame, the core code will
209 : * ignore our nodeN specification here, but that's OK. We still have
210 : * to provide the correct levelAdd and restDatum values, and those are
211 : * the same regardless of which node gets chosen by core.)
212 : */
213 : int levelAdd;
214 :
215 22396 : out->resultType = spgMatchNode;
216 22396 : out->result.matchNode.nodeN = i;
217 22396 : levelAdd = commonLen;
218 22396 : if (nodeChar >= 0)
219 22352 : levelAdd++;
220 22396 : out->result.matchNode.levelAdd = levelAdd;
221 22396 : if (inSize - in->level - levelAdd > 0)
222 22306 : out->result.matchNode.restDatum =
223 22306 : formTextDatum(inStr + in->level + levelAdd,
224 22306 : inSize - in->level - levelAdd);
225 : else
226 90 : out->result.matchNode.restDatum =
227 90 : formTextDatum(NULL, 0);
228 : }
229 232 : else if (in->allTheSame)
230 : {
231 : /*
232 : * Can't use AddNode action, so split the tuple. The upper tuple has
233 : * the same prefix as before and uses a dummy node label -2 for the
234 : * lower tuple. The lower tuple has no prefix and the same node
235 : * labels as the original tuple.
236 : *
237 : * Note: it might seem tempting to shorten the upper tuple's prefix,
238 : * if it has one, then use its last byte as label for the lower tuple.
239 : * But that doesn't win since we know the incoming value matches the
240 : * whole prefix: we'd just end up splitting the lower tuple again.
241 : */
242 0 : out->resultType = spgSplitTuple;
243 0 : out->result.splitTuple.prefixHasPrefix = in->hasPrefix;
244 0 : out->result.splitTuple.prefixPrefixDatum = in->prefixDatum;
245 0 : out->result.splitTuple.prefixNNodes = 1;
246 0 : out->result.splitTuple.prefixNodeLabels = (Datum *) palloc(sizeof(Datum));
247 0 : out->result.splitTuple.prefixNodeLabels[0] = Int16GetDatum(-2);
248 0 : out->result.splitTuple.childNodeN = 0;
249 0 : out->result.splitTuple.postfixHasPrefix = false;
250 : }
251 : else
252 : {
253 : /* Add a node for the not-previously-seen nodeChar value */
254 232 : out->resultType = spgAddNode;
255 232 : out->result.addNode.nodeLabel = Int16GetDatum(nodeChar);
256 232 : out->result.addNode.nodeN = i;
257 : }
258 :
259 22628 : PG_RETURN_VOID();
260 : }
261 :
262 : /* The picksplit function is identical to the core opclass, so just use that */
263 :
264 4 : PG_FUNCTION_INFO_V1(spgist_name_inner_consistent);
265 : Datum
266 8 : spgist_name_inner_consistent(PG_FUNCTION_ARGS)
267 : {
268 8 : spgInnerConsistentIn *in = (spgInnerConsistentIn *) PG_GETARG_POINTER(0);
269 8 : spgInnerConsistentOut *out = (spgInnerConsistentOut *) PG_GETARG_POINTER(1);
270 : text *reconstructedValue;
271 : text *reconstrText;
272 : int maxReconstrLen;
273 8 : text *prefixText = NULL;
274 8 : int prefixSize = 0;
275 : int i;
276 :
277 : /*
278 : * Reconstruct values represented at this tuple, including parent data,
279 : * prefix of this tuple if any, and the node label if it's non-dummy.
280 : * in->level should be the length of the previously reconstructed value,
281 : * and the number of bytes added here is prefixSize or prefixSize + 1.
282 : *
283 : * Recall that reconstructedValues are assumed to be the same type as leaf
284 : * datums, so we must use "text" not "name" for them.
285 : *
286 : * Note: we assume that in->reconstructedValue isn't toasted and doesn't
287 : * have a short varlena header. This is okay because it must have been
288 : * created by a previous invocation of this routine, and we always emit
289 : * long-format reconstructed values.
290 : */
291 8 : reconstructedValue = (text *) DatumGetPointer(in->reconstructedValue);
292 : Assert(reconstructedValue == NULL ? in->level == 0 :
293 : VARSIZE_ANY_EXHDR(reconstructedValue) == in->level);
294 :
295 8 : maxReconstrLen = in->level + 1;
296 8 : if (in->hasPrefix)
297 : {
298 0 : prefixText = DatumGetTextPP(in->prefixDatum);
299 0 : prefixSize = VARSIZE_ANY_EXHDR(prefixText);
300 0 : maxReconstrLen += prefixSize;
301 : }
302 :
303 8 : reconstrText = palloc(VARHDRSZ + maxReconstrLen);
304 8 : SET_VARSIZE(reconstrText, VARHDRSZ + maxReconstrLen);
305 :
306 8 : if (in->level)
307 4 : memcpy(VARDATA(reconstrText),
308 4 : VARDATA(reconstructedValue),
309 4 : in->level);
310 8 : if (prefixSize)
311 0 : memcpy(((char *) VARDATA(reconstrText)) + in->level,
312 0 : VARDATA_ANY(prefixText),
313 : prefixSize);
314 : /* last byte of reconstrText will be filled in below */
315 :
316 : /*
317 : * Scan the child nodes. For each one, complete the reconstructed value
318 : * and see if it's consistent with the query. If so, emit an entry into
319 : * the output arrays.
320 : */
321 8 : out->nodeNumbers = (int *) palloc(sizeof(int) * in->nNodes);
322 8 : out->levelAdds = (int *) palloc(sizeof(int) * in->nNodes);
323 8 : out->reconstructedValues = (Datum *) palloc(sizeof(Datum) * in->nNodes);
324 8 : out->nNodes = 0;
325 :
326 140 : for (i = 0; i < in->nNodes; i++)
327 : {
328 132 : int16 nodeChar = DatumGetInt16(in->nodeLabels[i]);
329 : int thisLen;
330 132 : bool res = true;
331 : int j;
332 :
333 : /* If nodeChar is a dummy value, don't include it in data */
334 132 : if (nodeChar <= 0)
335 0 : thisLen = maxReconstrLen - 1;
336 : else
337 : {
338 132 : ((unsigned char *) VARDATA(reconstrText))[maxReconstrLen - 1] = nodeChar;
339 132 : thisLen = maxReconstrLen;
340 : }
341 :
342 256 : for (j = 0; j < in->nkeys; j++)
343 : {
344 248 : StrategyNumber strategy = in->scankeys[j].sk_strategy;
345 : Name inName;
346 : char *inStr;
347 : int inSize;
348 : int r;
349 :
350 248 : inName = DatumGetName(in->scankeys[j].sk_argument);
351 248 : inStr = NameStr(*inName);
352 248 : inSize = strlen(inStr);
353 :
354 248 : r = memcmp(VARDATA(reconstrText), inStr,
355 248 : Min(inSize, thisLen));
356 :
357 248 : switch (strategy)
358 : {
359 116 : case BTLessStrategyNumber:
360 : case BTLessEqualStrategyNumber:
361 116 : if (r > 0)
362 108 : res = false;
363 116 : break;
364 0 : case BTEqualStrategyNumber:
365 0 : if (r != 0 || inSize < thisLen)
366 0 : res = false;
367 0 : break;
368 132 : case BTGreaterEqualStrategyNumber:
369 : case BTGreaterStrategyNumber:
370 132 : if (r < 0)
371 16 : res = false;
372 132 : break;
373 0 : default:
374 0 : elog(ERROR, "unrecognized strategy number: %d",
375 : in->scankeys[j].sk_strategy);
376 : break;
377 : }
378 :
379 248 : if (!res)
380 124 : break; /* no need to consider remaining conditions */
381 : }
382 :
383 132 : if (res)
384 : {
385 8 : out->nodeNumbers[out->nNodes] = i;
386 8 : out->levelAdds[out->nNodes] = thisLen - in->level;
387 8 : SET_VARSIZE(reconstrText, VARHDRSZ + thisLen);
388 16 : out->reconstructedValues[out->nNodes] =
389 8 : datumCopy(PointerGetDatum(reconstrText), false, -1);
390 8 : out->nNodes++;
391 : }
392 : }
393 :
394 8 : PG_RETURN_VOID();
395 : }
396 :
397 4 : PG_FUNCTION_INFO_V1(spgist_name_leaf_consistent);
398 : Datum
399 248 : spgist_name_leaf_consistent(PG_FUNCTION_ARGS)
400 : {
401 248 : spgLeafConsistentIn *in = (spgLeafConsistentIn *) PG_GETARG_POINTER(0);
402 248 : spgLeafConsistentOut *out = (spgLeafConsistentOut *) PG_GETARG_POINTER(1);
403 248 : int level = in->level;
404 : text *leafValue,
405 248 : *reconstrValue = NULL;
406 : char *fullValue;
407 : int fullLen;
408 : bool res;
409 : int j;
410 :
411 : /* all tests are exact */
412 248 : out->recheck = false;
413 :
414 248 : leafValue = DatumGetTextPP(in->leafDatum);
415 :
416 : /* As above, in->reconstructedValue isn't toasted or short. */
417 248 : if (DatumGetPointer(in->reconstructedValue))
418 248 : reconstrValue = (text *) DatumGetPointer(in->reconstructedValue);
419 :
420 : Assert(reconstrValue == NULL ? level == 0 :
421 : VARSIZE_ANY_EXHDR(reconstrValue) == level);
422 :
423 : /* Reconstruct the Name represented by this leaf tuple */
424 248 : fullValue = palloc0(NAMEDATALEN);
425 248 : fullLen = level + VARSIZE_ANY_EXHDR(leafValue);
426 : Assert(fullLen < NAMEDATALEN);
427 248 : if (VARSIZE_ANY_EXHDR(leafValue) == 0 && level > 0)
428 : {
429 0 : memcpy(fullValue, VARDATA(reconstrValue),
430 0 : VARSIZE_ANY_EXHDR(reconstrValue));
431 : }
432 : else
433 : {
434 248 : if (level)
435 248 : memcpy(fullValue, VARDATA(reconstrValue), level);
436 248 : if (VARSIZE_ANY_EXHDR(leafValue) > 0)
437 248 : memcpy(fullValue + level, VARDATA_ANY(leafValue),
438 248 : VARSIZE_ANY_EXHDR(leafValue));
439 : }
440 248 : out->leafValue = PointerGetDatum(fullValue);
441 :
442 : /* Perform the required comparison(s) */
443 248 : res = true;
444 516 : for (j = 0; j < in->nkeys; j++)
445 : {
446 464 : StrategyNumber strategy = in->scankeys[j].sk_strategy;
447 464 : Name queryName = DatumGetName(in->scankeys[j].sk_argument);
448 464 : char *queryStr = NameStr(*queryName);
449 464 : int queryLen = strlen(queryStr);
450 : int r;
451 :
452 : /* Non-collation-aware comparison */
453 464 : r = memcmp(fullValue, queryStr, Min(queryLen, fullLen));
454 :
455 464 : if (r == 0)
456 : {
457 52 : if (queryLen > fullLen)
458 0 : r = -1;
459 52 : else if (queryLen < fullLen)
460 52 : r = 1;
461 : }
462 :
463 464 : switch (strategy)
464 : {
465 216 : case BTLessStrategyNumber:
466 216 : res = (r < 0);
467 216 : break;
468 0 : case BTLessEqualStrategyNumber:
469 0 : res = (r <= 0);
470 0 : break;
471 0 : case BTEqualStrategyNumber:
472 0 : res = (r == 0);
473 0 : break;
474 0 : case BTGreaterEqualStrategyNumber:
475 0 : res = (r >= 0);
476 0 : break;
477 248 : case BTGreaterStrategyNumber:
478 248 : res = (r > 0);
479 248 : break;
480 0 : default:
481 0 : elog(ERROR, "unrecognized strategy number: %d",
482 : in->scankeys[j].sk_strategy);
483 : res = false;
484 : break;
485 : }
486 :
487 464 : if (!res)
488 196 : break; /* no need to consider remaining conditions */
489 : }
490 :
491 248 : PG_RETURN_BOOL(res);
492 : }
493 :
494 4 : PG_FUNCTION_INFO_V1(spgist_name_compress);
495 : Datum
496 13516 : spgist_name_compress(PG_FUNCTION_ARGS)
497 : {
498 13516 : Name inName = PG_GETARG_NAME(0);
499 13516 : char *inStr = NameStr(*inName);
500 :
501 13516 : PG_RETURN_DATUM(formTextDatum(inStr, strlen(inStr)));
502 : }
|
