Many hyperlinks are disabled.
Use anonymous login
to enable hyperlinks.
Overview
Comment: | Fix a couple of out-of-date comments in where.c. |
---|---|
Downloads: | Tarball | ZIP archive |
Timelines: | family | ancestors | descendants | both | experimental-costs |
Files: | files | file ages | folders |
SHA1: |
eefeda32d54efbbdf7d20b719299eda4 |
User & Date: | dan 2014-04-30 14:47:01.628 |
Context
2014-04-30
| ||
14:53 | Update a couple of test cases to account for the fact that this branch prefers an index scan and partial sort over a full-table scan and full external sort. (check-in: 9b975bf33c user: dan tags: experimental-costs) | |
14:47 | Fix a couple of out-of-date comments in where.c. (check-in: eefeda32d5 user: dan tags: experimental-costs) | |
14:22 | Improved rendering of LogEst values corresponding to real values near 0.0 in the tool/logest.c utility program. (check-in: 32910c8c59 user: drh tags: experimental-costs) | |
Changes
Changes to src/where.c.
︙ | ︙ | |||
4018 4019 4020 4021 4022 4023 4024 | if( j<0 ){ pLoop->nOut += (pTerm->truthProb<=0 ? pTerm->truthProb : -1); } } } /* | | | > > > > > | 4018 4019 4020 4021 4022 4023 4024 4025 4026 4027 4028 4029 4030 4031 4032 4033 4034 4035 4036 4037 4038 | if( j<0 ){ pLoop->nOut += (pTerm->truthProb<=0 ? pTerm->truthProb : -1); } } } /* ** We have so far matched pBuilder->pNew->u.btree.nEq terms of the ** index pIndex. Try to match one more. ** ** When this function is called, pBuilder->pNew->nOut contains the ** number of rows expected to be visited by filtering using the nEq ** terms only. If it is modified, this value is restored before this ** function returns. ** ** If pProbe->tnum==0, that means pIndex is a fake index used for the ** INTEGER PRIMARY KEY. */ static int whereLoopAddBtreeIndex( WhereLoopBuilder *pBuilder, /* The WhereLoop factory */ struct SrcList_item *pSrc, /* FROM clause term being analyzed */ |
︙ | ︙ | |||
4081 4082 4083 4084 4085 4086 4087 | saved_prereq = pNew->prereq; saved_nOut = pNew->nOut; pNew->rSetup = 0; rLogSize = estLog(pProbe->aiRowLogEst[0]); /* Consider using a skip-scan if there are no WHERE clause constraints ** available for the left-most terms of the index, and if the average | | < < | > > > > > > | 4086 4087 4088 4089 4090 4091 4092 4093 4094 4095 4096 4097 4098 4099 4100 4101 4102 4103 4104 4105 4106 4107 | saved_prereq = pNew->prereq; saved_nOut = pNew->nOut; pNew->rSetup = 0; rLogSize = estLog(pProbe->aiRowLogEst[0]); /* Consider using a skip-scan if there are no WHERE clause constraints ** available for the left-most terms of the index, and if the average ** number of repeats in the left-most terms is at least 18. ** ** The magic number 18 is selected on the basis that scanning 17 rows ** is almost always quicker than an index seek (even though if the index ** contains fewer than 2^17 rows we assume otherwise in other parts of ** the code). And, even if it is not, it should not be too much slower. ** On the other hand, the extra seeks could end up being significantly ** more expensive. */ assert( 42==sqlite3LogEst(18) ); if( pTerm==0 && saved_nEq==saved_nSkip && saved_nEq+1<pProbe->nKeyCol && pProbe->aiRowLogEst[saved_nEq+1]>=42 /* TUNING: Minimum for skip-scan */ && (rc = whereLoopResize(db, pNew, pNew->nLTerm+1))==SQLITE_OK ){ |
︙ | ︙ | |||
5222 5223 5224 5225 5226 5227 5228 | nOut = pFrom->nRow + pWLoop->nOut; maskNew = pFrom->maskLoop | pWLoop->maskSelf; if( isOrdered<0 ){ isOrdered = wherePathSatisfiesOrderBy(pWInfo, pWInfo->pOrderBy, pFrom, pWInfo->wctrlFlags, iLoop, pWLoop, &revMask); if( isOrdered>=0 && isOrdered<nOrderBy ){ | | > > > > | | | | < < | | > > | | | 5231 5232 5233 5234 5235 5236 5237 5238 5239 5240 5241 5242 5243 5244 5245 5246 5247 5248 5249 5250 5251 5252 5253 5254 5255 5256 5257 5258 5259 5260 5261 5262 5263 5264 5265 | nOut = pFrom->nRow + pWLoop->nOut; maskNew = pFrom->maskLoop | pWLoop->maskSelf; if( isOrdered<0 ){ isOrdered = wherePathSatisfiesOrderBy(pWInfo, pWInfo->pOrderBy, pFrom, pWInfo->wctrlFlags, iLoop, pWLoop, &revMask); if( isOrdered>=0 && isOrdered<nOrderBy ){ /* TUNING: Estimated cost of a full external sort, where N is ** the number of rows to sort is: ** ** cost = (3.0 * N * log(N)). ** ** Or, if the order-by clause has X terms but only the last Y ** terms are out of order, then block-sorting will reduce the ** sorting cost to: ** ** cost = (3.0 * N * log(N)) * (Y/X) ** ** The (Y/X) term is implemented using stack variable rScale ** below. */ LogEst rScale, rSortCost; assert( nOrderBy>0 && 66==sqlite3LogEst(100) ); rScale = sqlite3LogEst((nOrderBy-isOrdered)*100/nOrderBy) - 66; rSortCost = nRowEst + estLog(nRowEst) + rScale + 16; /* TUNING: The cost of implementing DISTINCT using a B-TREE is ** similar but with a larger constant of proportionality. ** Multiply by an additional factor of 3.0. */ if( pWInfo->wctrlFlags & WHERE_WANT_DISTINCT ){ rSortCost += 16; } WHERETRACE(0x002, ("---- sort cost=%-3d (%d/%d) increases cost %3d to %-3d\n", rSortCost, (nOrderBy-isOrdered), nOrderBy, rCost, sqlite3LogEstAdd(rCost,rSortCost))); |
︙ | ︙ |