When constructing an ephermeral table to use as the right-hand side of

an IN operator, also construct a Bloom filter to speed membership testing.

FossilOrigin-Name: baa83b460c677c210c7fa3f20314d7e05f305aed8a69026bc5fa106a3de4ea38

manifest

@@ -1,5 +1,5 @@
-C Add\sassert()\sstatements\sto\sFTS5\sto\shush-up\swarnings\sfrom\sscan-build.
+C When\sconstructing\san\sephermeral\stable\sto\suse\sas\sthe\sright-hand\sside\sof\nan\sIN\soperator,\salso\sconstruct\sa\sBloom\sfilter\sto\sspeed\smembership\stesting.
-D 2024-07-02T13:54:46.151
+D 2024-07-03T20:19:33.270
 F .fossil-settings/empty-dirs dbb81e8fc0401ac46a1491ab34a7f2c7c0452f2f06b54ebb845d024ca8283ef1
 F .fossil-settings/ignore-glob 35175cdfcf539b2318cb04a9901442804be81cd677d8b889fcc9149c21f239ea
 F LICENSE.md df5091916dbb40e6e9686186587125e1b2ff51f022cc334e886c19a0e9982724
@@ -705,7 +705,7 @@ F src/date.c 13dd752847afb32ed70510ad7345a5b9c841f51ad904dba5d010f1fa3a6a324e
 F src/dbpage.c 80e46e1df623ec40486da7a5086cb723b0275a6e2a7b01d9f9b5da0f04ba2782
 F src/dbstat.c 3b677254d512fcafd4d0b341bf267b38b235ccfddbef24f9154e19360fa22e43
 F src/delete.c cb766727c78e715f9fb7ec8a7d03658ed2a3016343ca687acfcec9083cdca500
-F src/expr.c 2b72d352a2917fe8d2cd8f84e0acd0140c6ba769b0cd7bcfd86eb9ba0c9c50e8
+F src/expr.c d6f614d0c90cf8d72bbd54dbae96040e7feb495a77991fdc1b1c67db68eeb326
 F src/fault.c 460f3e55994363812d9d60844b2a6de88826e007
 F src/fkey.c 852f93c0ef995e0c2b8983059a2b97151c194cc8259e21f5bc2b7ac508348c2a
 F src/func.c 1f61e32e7a357e615b5d2e774bee563761fce4f2fd97ecb0f72c33e62a2ada5f
@@ -755,12 +755,12 @@ F src/printf.c 8b250972305e14b365561be5117ed0fd364e4fd58968776df1ce64c6280b90f9
 F src/random.c 606b00941a1d7dd09c381d3279a058d771f406c5213c9932bbd93d5587be4b9c
 F src/resolve.c 7e8d23ce7cdbfedf351a47e759f2722e8182ca10fd7580be43f4ce1f1a228145
 F src/rowset.c 8432130e6c344b3401a8874c3cb49fefe6873fec593294de077afea2dce5ec97
-F src/select.c a1c8fadd45d0843b14793db2af49809a5327db5cca9d87d507b57aa748ee6ee2
+F src/select.c c448e2c0670ac590e5af53ef91e84a188bedb3d80d486e54948bf40d4dc4204b
 F src/shell.c.in b7d435c137eb323981adff814f172dbaabb9ba504fef17cb11d4681c1633ee13
 F src/sqlite.h.in 6c884a87bbf8828562b49272025a1e66e3801a196a58b0bdec87edcd2c9c8fc1
 F src/sqlite3.rc 5121c9e10c3964d5755191c80dd1180c122fc3a8
 F src/sqlite3ext.h 3f046c04ea3595d6bfda99b781926b17e672fd6d27da2ba6d8d8fc39981dcb54
-F src/sqliteInt.h 86cbef26e285786dfd6dc7b33cfed0210095eb3ce9b38047229b03eb795b2edd
+F src/sqliteInt.h 039b5309b49399340f260ad3d1e6b6fa13e04a7b587b9daa1e694ce0f13a04a3
 F src/sqliteLimit.h 6878ab64bdeb8c24a1d762d45635e34b96da21132179023338c93f820eee6728
 F src/status.c cb11f8589a6912af2da3bb1ec509a94dd8ef27df4d4c1a97e0bcf2309ece972b
 F src/table.c 0f141b58a16de7e2fbe81c308379e7279f4c6b50eb08efeec5892794a0ba30d1
@@ -825,7 +825,7 @@ F src/upsert.c 2e60567a0e9e8520c18671b30712a88dc73534474304af94f32bb5f3ef65ac65
 F src/utf.c f23165685a67b4caf8ec08fb274cb3f319103decfb2a980b7cfd55d18dfa855e
 F src/util.c 5d1a0134cf4240648d1c6bb5cc8efaca0ea2b5d5c840985aec7e947271f04375
 F src/vacuum.c 604fcdaebe76f3497c855afcbf91b8fa5046b32de3045bab89cc008d68e40104
-F src/vdbe.c b05777c3ff2ed7b9dfc347e7cdee18e371aa6811cef1fe83454691b0dbe2cc9f
+F src/vdbe.c f941fd8dbd1a9e073866a941e7fbb553f1de2d25b249611005d8c27d5eac8995
 F src/vdbe.h c2d78d15112c3fc5ab87f5e8e0b75d2db1c624409de2e858c3d1aafb1650bb4f
 F src/vdbeInt.h 949669dfd8a41550d27dcb905b494f2ccde9a2e6c1b0b04daa1227e2e74c2b2c
 F src/vdbeapi.c 80235ac380e9467fec1cb0883354d841f2a771976e766995f7e0c77f845406df
@@ -908,7 +908,7 @@ F test/auth2.test 9eb7fce9f34bf1f50d3f366fb3e606be5a2000a1
 F test/auth3.test 76d20a7fa136d63bcfcf8bcb65c0b1455ed71078d81f22bcd0550d3eb18594ab
 F test/autoanalyze1.test b9cc3f32a990fa56669b668d237c6d53e983554ae80c0604992e18869a0b2dec
 F test/autoinc.test 997d6f185f138229dc4251583a1d04816423dddc2fc034871a01aeb1d728cb39
-F test/autoindex1.test d34caffb0384003ee28eae87679214c029e9be4b332d9649a79e0b94ab70502c
+F test/autoindex1.test 714cac6e60beeb5a26ed346dd46505ba60b5a5597e9122c9ed3a55f89a922aa4
 F test/autoindex2.test 12ef578928102baaa0dc23ad397601a2f4ecb0df
 F test/autoindex3.test ca502c8050166ac6107a7b4fe4e951f4d3270a23a958af02b14f1b962b83c4b6
 F test/autoindex4.test 3c2105e9172920e26f950ba3c5823e4972190e022c1e6f260ba476b0af24c593
@@ -1106,7 +1106,7 @@ F test/enc.test 9a7be5479da985381d740b15f432800f65e2c87029ee57a318f42cb2eb43763a
 F test/enc2.test 848bf05f15b011719f478dddb7b5e9aea35e39e457493cba4c4eef75d849a5ec
 F test/enc3.test 55ef64416d72975c66167310a51dc9fc544ba3ae4858b8d5ab22f4cb6500b087
 F test/enc4.test c8f1ce3618508fd0909945beb8b8831feef2c020
-F test/eqp.test 3302598f611220a6c61e29d9b7bb62fd4a43504509b978dbabdb0b3e56ae3bc0
+F test/eqp.test 815418b69f6be3a27037b1736c54699c72cc3e2e6b0bc878c01464d1dcec65fe
 F test/eqp2.test 6e8996148de88f0e7670491e92e712a2920a369b4406f21a27c3c9b6a46b68dd
 F test/errmsg.test eae9f091eb39ce7e20305de45d8e5d115b68fa856fba4ea6757b6ca3705ff7f9
 F test/eval.test 73969a2d43a511bf44080c44485a8c4d796b6a4f038d19e491867081155692c0
@@ -1518,7 +1518,7 @@ F test/printf.test 685fec5a0c5af2490ab0632775a301554361d674211d690f5bee0a97b0533
 F test/printf2.test 3f55c1871a5a65507416076f6eb97e738d5210aeda7595a74ee895f2224cce60
 F test/progress.test ebab27f670bd0d4eb9d20d49cef96e68141d92fb
 F test/ptrchng.test ef1aa72d6cf35a2bbd0869a649b744e9d84977fc
-F test/pushdown.test 3330746a897ea271b1021bc1e7e57c6f8d49bcb8ca7d58a823d01aa64a303cc7
+F test/pushdown.test 9e655df51bc6559608dcc7af89a36f727eff520b4860261ed15a5f95a5ebbcd8
 F test/queryonly.test 5f653159e0f552f0552d43259890c1089391dcca
 F test/quick.test 1681febc928d686362d50057c642f77a02c62e57
 F test/quickcheck.test a4b7e878cd97e46108291c409b0bf8214f29e18fddd68a42bc5c1375ad1fb80a
@@ -1549,7 +1549,7 @@ F test/rowid.test d27191b5ce794c05bf61081e8b2c546a1844c1641321dcaf7fb785234256cc
 F test/rowvalue.test baf4fa3ec1a8c1c920c3faa5fd25959cb454bbd99ac8960397c34549d9fc4abe
 F test/rowvalue2.test 060d238b7e5639a7c5630cb5e63e311b44efef2b
 F test/rowvalue3.test 1347e25ca11c547c5a6ff0cc5626f95aa9740e9275bfaec096029f57cb2130ce
-F test/rowvalue4.test 441e7e366ac6d939a3a95a574031c56ec2a854077a91d66eee5ff1d86cb5be58
+F test/rowvalue4.test bac9326d1e886656650f67c0ec484eb5f452244a8209c6af508e9a862ace08ed
 F test/rowvalue5.test 00740304ea6a53a8704640c7405690f0045d5d2a6b4b04dde7bccc14c3068ea7
 F test/rowvalue6.test d19b54feb604d5601f8614b15e214e0774c01087
 F test/rowvalue7.test c1cbdbf407029db01f87764097c6ac02a1c5a37efd2776eff32a9cdfdf6f2dba
@@ -2195,8 +2195,9 @@ F vsixtest/vsixtest.tcl 6a9a6ab600c25a91a7acc6293828957a386a8a93
 F vsixtest/vsixtest.vcxproj.data 2ed517e100c66dc455b492e1a33350c1b20fbcdc
 F vsixtest/vsixtest.vcxproj.filters 37e51ffedcdb064aad6ff33b6148725226cd608e
 F vsixtest/vsixtest_TemporaryKey.pfx e5b1b036facdb453873e7084e1cae9102ccc67a0
-P fc248a4a0a232a95a79e24e57faedb5d824c3bf0aa62054b72339257dc9c18b3
+P 77a76654e64c05f2c67be120f579fc60924ff51a9d0fa2cf9eb158f40171b5f5 c10a1b99d47a4c93bdb16e646f6a21add570d6dbb34fd91a32f8abacdb28712b
-R f55a5b83388b86278f89f11080405f1e
+R c969cbcf973aa6019e666c5feff1eff5
+T +closed c10a1b99d47a4c93bdb16e646f6a21add570d6dbb34fd91a32f8abacdb28712b
 U drh
-Z dda9ed678e9e469573614b965067e0d5
+Z f6a0a26b000fa84938a7a16f1457134d
 # Remove this line to create a well-formed Fossil manifest.

manifest.uuid

@@ -1 +1 @@
-77a76654e64c05f2c67be120f579fc60924ff51a9d0fa2cf9eb158f40171b5f5
+baa83b460c677c210c7fa3f20314d7e05f305aed8a69026bc5fa106a3de4ea38

src/expr.c

@@ -3536,19 +3536,34 @@ void sqlite3CodeRhsOfIN(
       SelectDest dest;
       int i;
       int rc;
+      int addrBloom = 0;
       sqlite3SelectDestInit(&dest, SRT_Set, iTab);
       dest.zAffSdst = exprINAffinity(pParse, pExpr);
       pSelect->iLimit = 0;
+      if( addrOnce && OptimizationEnabled(pParse->db, SQLITE_BloomFilter) ){
+        int regBloom = ++pParse->nMem;
+        addrBloom = sqlite3VdbeAddOp2(v, OP_Blob, 10000, regBloom);
+        VdbeComment((v, "Bloom filter"));
+        dest.iSDParm2 = regBloom;
+      }
       testcase( pSelect->selFlags & SF_Distinct );
       testcase( pKeyInfo==0 ); /* Caused by OOM in sqlite3KeyInfoAlloc() */
       pCopy = sqlite3SelectDup(pParse->db, pSelect, 0);
       rc = pParse->db->mallocFailed ? 1 :sqlite3Select(pParse, pCopy, &dest);
       sqlite3SelectDelete(pParse->db, pCopy);
       sqlite3DbFree(pParse->db, dest.zAffSdst);
+      if( addrBloom ){
+        sqlite3VdbeGetOp(v, addrOnce)->p3 = dest.iSDParm2;
+        if( dest.iSDParm2==0 ){
+          sqlite3VdbeChangeToNoop(v, addrBloom);
+        }else{
+          sqlite3VdbeGetOp(v, addrOnce)->p3 = dest.iSDParm2;
+        }
+      }
       if( rc ){
         sqlite3KeyInfoUnref(pKeyInfo);
         return;
-      }     
+      }
       assert( pKeyInfo!=0 ); /* OOM will cause exit after sqlite3Select() */
       assert( pEList!=0 );
       assert( pEList->nExpr>0 );
@@ -3987,6 +4002,15 @@ static void sqlite3ExprCodeIN(
     sqlite3VdbeAddOp4(v, OP_Affinity, rLhs, nVector, 0, zAff, nVector);
     if( destIfFalse==destIfNull ){
       /* Combine Step 3 and Step 5 into a single opcode */
+      if( ExprHasProperty(pExpr, EP_Subrtn) ){
+        const VdbeOp *pOp = sqlite3VdbeGetOp(v, pExpr->y.sub.iAddr);
+        assert( pOp->opcode==OP_Once || pParse->nErr );
+        if( pOp->opcode==OP_Once && pOp->p3>0 ){
+          assert( OptimizationEnabled(pParse->db, SQLITE_BloomFilter) );
+          sqlite3VdbeAddOp4Int(v, OP_Filter, pOp->p3, destIfFalse,
+                               rLhs, nVector); VdbeCoverage(v);
+        }
+      }
       sqlite3VdbeAddOp4Int(v, OP_NotFound, iTab, destIfFalse,
                            rLhs, nVector); VdbeCoverage(v);
       goto sqlite3ExprCodeIN_finished;

src/select.c

@@ -1377,12 +1377,18 @@ static void selectInnerLoop(
         ** case the order does matter */
         pushOntoSorter(
             pParse, pSort, p, regResult, regOrig, nResultCol, nPrefixReg);
+        pDest->iSDParm2 = 0; /* Signal that any Bloom filter is unpopulated */
       }else{
         int r1 = sqlite3GetTempReg(pParse);
         assert( sqlite3Strlen30(pDest->zAffSdst)==nResultCol );
         sqlite3VdbeAddOp4(v, OP_MakeRecord, regResult, nResultCol,
             r1, pDest->zAffSdst, nResultCol);
         sqlite3VdbeAddOp4Int(v, OP_IdxInsert, iParm, r1, regResult, nResultCol);
+        if( pDest->iSDParm2 ){
+          sqlite3VdbeAddOp4Int(v, OP_FilterAdd, pDest->iSDParm2, 0,
+                               regResult, nResultCol);
+          ExplainQueryPlan((pParse, 0, "CREATE BLOOM FILTER"));
+        }
         sqlite3ReleaseTempReg(pParse, r1);
       }
       break;
@@ -3316,6 +3322,11 @@ static int generateOutputSubroutine(
           r1, pDest->zAffSdst, pIn->nSdst);
       sqlite3VdbeAddOp4Int(v, OP_IdxInsert, pDest->iSDParm, r1,
                            pIn->iSdst, pIn->nSdst);
+      if( pDest->iSDParm2>0 ){
+        sqlite3VdbeAddOp4Int(v, OP_FilterAdd, pDest->iSDParm2, 0,
+                             pIn->iSdst, pIn->nSdst);
+        ExplainQueryPlan((pParse, 0, "CREATE BLOOM FILTER"));
+      }
       sqlite3ReleaseTempReg(pParse, r1);
       break;
     }

src/sqliteInt.h

@@ -3622,7 +3622,11 @@ struct Select {
 **     SRT_Set         The result must be a single column.  Store each
 **                     row of result as the key in table pDest->iSDParm.
 **                     Apply the affinity pDest->affSdst before storing
-**                     results.  Used to implement "IN (SELECT ...)".
+**                     results.  if pDest->iSDParm2 is positive, then it is
+**                     a regsiter holding a Bloom filter for the IN operator
+**                     that should be populated in addition to the 
+**                     pDest->iSDParm table.  This SRT is used to
+**                     implement "IN (SELECT ...)".
 **
 **     SRT_EphemTab    Create an temporary table pDest->iSDParm and store
 **                     the result there. The cursor is left open after

src/vdbe.c

@@ -5308,6 +5308,7 @@ case OP_Found: {        /* jump, in3, ncycle */
     r.pKeyInfo = pC->pKeyInfo;
     r.default_rc = 0;
 #ifdef SQLITE_DEBUG
+    (void)sqlite3FaultSim(50);  /* For use by --counter in TH3 */
     for(ii=0; ii<r.nField; ii++){
       assert( memIsValid(&r.aMem[ii]) );
       assert( (r.aMem[ii].flags & MEM_Zero)==0 || r.aMem[ii].n==0 );

test/autoindex1.test

@@ -185,7 +185,8 @@ do_eqp_test autoindex1-500.1 {
   QUERY PLAN
   |--SEARCH t501 USING INTEGER PRIMARY KEY (rowid=?)
   `--LIST SUBQUERY xxxxxx
-     `--SCAN t502
+     |--SCAN t502
+     `--CREATE BLOOM FILTER
 }
 do_eqp_test autoindex1-501 {
   SELECT b FROM t501

test/eqp.test

@@ -311,7 +311,8 @@ det 3.3.1 {
   QUERY PLAN
   |--SCAN t1
   `--LIST SUBQUERY xxxxxx
-     `--SCAN t2
+     |--SCAN t2
+     `--CREATE BLOOM FILTER
 }
 det 3.3.2 {
   SELECT * FROM t1 WHERE y IN (SELECT y FROM t2 WHERE t1.x!=t2.x)

test/pushdown.test

@@ -275,14 +275,17 @@ do_eqp_test 6.1 {
   |     |  `--LIST SUBQUERY xxxxxx
   |     |     |--MATERIALIZE k
   |     |     |  `--SCAN 3 CONSTANT ROWS
-  |     |     `--SCAN k
+  |     |     |--SCAN k
+  |     |     `--CREATE BLOOM FILTER
   |     `--UNION ALL
   |        |--SEARCH t02 USING INDEX t02x (w=? AND x=? AND y>? AND y<?)
   |        `--LIST SUBQUERY xxxxxx
-  |           `--SCAN k
+  |           |--SCAN k
+  |           `--CREATE BLOOM FILTER
   |--SEARCH t0
   `--LIST SUBQUERY xxxxxx
-     `--SCAN k
+     |--SCAN k
+     `--CREATE BLOOM FILTER
 }
 # ^^^^--- The key feature above is that the SEARCH for each subquery
 # uses all three fields of the index w, x, and y.  Prior to the push-down
@@ -300,18 +303,21 @@ do_eqp_test 6.2 {
   |     |  `--LIST SUBQUERY xxxxxx
   |     |     |--CO-ROUTINE v1
   |     |     |  `--SCAN 3 CONSTANT ROWS
-  |     |     `--SCAN v1
+  |     |     |--SCAN v1
+  |     |     `--CREATE BLOOM FILTER
   |     `--UNION ALL
   |        |--SEARCH t02 USING INDEX t02x (w=? AND x=? AND y>? AND y<?)
   |        `--LIST SUBQUERY xxxxxx
   |           |--CO-ROUTINE v1
   |           |  `--SCAN 3 CONSTANT ROWS
-  |           `--SCAN v1
+  |           |--SCAN v1
+  |           `--CREATE BLOOM FILTER
   |--SEARCH t0
   `--LIST SUBQUERY xxxxxx
      |--CO-ROUTINE v1
      |  `--SCAN 3 CONSTANT ROWS
-     `--SCAN v1
+     |--SCAN v1
+     `--CREATE BLOOM FILTER
 }
 do_eqp_test 6.3 {
   SELECT max(z) FROM t0 WHERE w=123 AND x IN k1 AND y BETWEEN 44 AND 55;
@@ -322,14 +328,17 @@ do_eqp_test 6.3 {
   |     |--LEFT-MOST SUBQUERY
   |     |  |--SEARCH t01 USING INDEX t01x (w=? AND x=? AND y>? AND y<?)
   |     |  `--LIST SUBQUERY xxxxxx
-  |     |     `--SCAN k1
+  |     |     |--SCAN k1
+  |     |     `--CREATE BLOOM FILTER
   |     `--UNION ALL
   |        |--SEARCH t02 USING INDEX t02x (w=? AND x=? AND y>? AND y<?)
   |        `--LIST SUBQUERY xxxxxx
-  |           `--SCAN k1
+  |           |--SCAN k1
+  |           `--CREATE BLOOM FILTER
   |--SEARCH t0
   `--LIST SUBQUERY xxxxxx
-     `--SCAN k1
+     |--SCAN k1
+     `--CREATE BLOOM FILTER
 }
 
 finish_test

test/rowvalue4.test

@@ -236,9 +236,11 @@ do_eqp_test 5.1 {
   QUERY PLAN
   |--SEARCH d2 USING INDEX d2ab (a=? AND b=?)
   |--LIST SUBQUERY xxxxxx
-  |  `--SCAN d1
+  |  |--SCAN d1
+  |  `--CREATE BLOOM FILTER
   `--LIST SUBQUERY xxxxxx
-     `--SCAN d1
+     |--SCAN d1
+     `--CREATE BLOOM FILTER
 }
 
 do_execsql_test 6.0 {