Attempt to improve the ANALYZE command so that it does a better job of

detecting lopsided indexes and makes appropriate changes to the sqlite_stat1
table.

FossilOrigin-Name: 4b70b94616ef37bac969051eee3ea6913a28f30520cdd4fc3a19e848f2cf12b7

diff --git a/manifest b/manifest
index adb892f498a5bd0d1e74250a50140f1aeb6a0a2f..c875ee5dda2a18bc1d37925d8010ec464f77d1a0 100644 (file)
--- a/manifest
+++ b/manifest
@@ -1,5 +1,5 @@
-C Minor\sdoc\stouchup\sin\sthe\sJS\sbits.
-D 2023-12-29T04:29:07.096
+C Attempt\sto\simprove\sthe\sANALYZE\scommand\sso\sthat\sit\sdoes\sa\sbetter\sjob\sof\ndetecting\slopsided\sindexes\sand\smakes\sappropriate\schanges\sto\sthe\ssqlite_stat1\ntable.
+D 2023-12-29T19:03:01.857
 F .fossil-settings/empty-dirs dbb81e8fc0401ac46a1491ab34a7f2c7c0452f2f06b54ebb845d024ca8283ef1
 F .fossil-settings/ignore-glob 35175cdfcf539b2318cb04a9901442804be81cd677d8b889fcc9149c21f239ea
 F LICENSE.md df5091916dbb40e6e9686186587125e1b2ff51f022cc334e886c19a0e9982724
@@ -670,7 +670,7 @@ F sqlite3.1 acdff36db796e2d00225b911d3047d580cd136547298435426ce9d40347973cc
 F sqlite3.pc.in 48fed132e7cb71ab676105d2a4dc77127d8c1f3a
 F sqlite_cfg.h.in baf2e409c63d4e7a765e17769b6ff17c5a82bbd9cbf1e284fd2e4cefaff3fcf2
 F src/alter.c 30c2333b8bb3af71e4eb9adeadee8aa20edb15917ed44b8422e5cd15f3dfcddc
-F src/analyze.c d4cc28738c29e009640ec20ebb6936ba6fcefff0d11aa93398d9bb9a5ead6c1f
+F src/analyze.c dd9f25aa96ab1ca1a52743a433fbc53888f41059d909a4360382835414733f82
 F src/attach.c cc9d00d30da916ff656038211410ccf04ed784b7564639b9b61d1839ed69fd39
 F src/auth.c 19b7ccacae3dfba23fc6f1d0af68134fa216e9040e53b0681b4715445ea030b4
 F src/backup.c 5c97e8023aab1ce14a42387eb3ae00ba5a0644569e3476f38661fa6f824c3523
@@ -2156,8 +2156,11 @@ F vsixtest/vsixtest.tcl 6a9a6ab600c25a91a7acc6293828957a386a8a93
 F vsixtest/vsixtest.vcxproj.data 2ed517e100c66dc455b492e1a33350c1b20fbcdc
 F vsixtest/vsixtest.vcxproj.filters 37e51ffedcdb064aad6ff33b6148725226cd608e
 F vsixtest/vsixtest_TemporaryKey.pfx e5b1b036facdb453873e7084e1cae9102ccc67a0
-P 630604a4e604bfb36c31602917bfa8d42c10c82966d0819932bf8f827b9158b8
-R 0182e48a4eb943a57927e95defc2f847
-U stephan
-Z 19ca2e7863b05c35a59e90f21d5378b3
+P 8d2120c35425081e2158d6a8a6b083c4adf8d694046b2d98f5fd235520920432
+R b58bf97977b7e8d817610de3abc3e459
+T *branch * enhanced-stat1
+T *sym-enhanced-stat1 *
+T -sym-trunk *
+U drh
+Z 87510e89218beaa47ade1953d2d2ed80
 # Remove this line to create a well-formed Fossil manifest.

diff --git a/manifest.uuid b/manifest.uuid
index c9aa0e6dea16c027671723410d7bc7255be37917..7b89cb32bf35ee8377f6ab91112c3d70c8be89f2 100644 (file)
--- a/manifest.uuid
+++ b/manifest.uuid
@@ -1 +1 @@
-8d2120c35425081e2158d6a8a6b083c4adf8d694046b2d98f5fd235520920432
\ No newline at end of file
+4b70b94616ef37bac969051eee3ea6913a28f30520cdd4fc3a19e848f2cf12b7
\ No newline at end of file

diff --git a/src/analyze.c b/src/analyze.c
index a7a8b6d66541553530ef1b84a93b6c197028251a..2c1fbdaac634baebdc657f8dd39913fc0cf3b1b2 100644 (file)
--- a/src/analyze.c
+++ b/src/analyze.c
@@ -266,6 +266,7 @@ typedef struct StatSample StatSample;
 struct StatSample {
   tRowcnt *anEq;                  /* sqlite_stat4.nEq */
   tRowcnt *anDLt;                 /* sqlite_stat4.nDLt */
+  tRowcnt *amxEq;                 /* Maximum length run of equal values */
 #ifdef SQLITE_ENABLE_STAT4
   tRowcnt *anLt;                  /* sqlite_stat4.nLt */
   union {
@@ -425,6 +426,7 @@ static void statInit(
   /* Allocate the space required for the StatAccum object */
   n = sizeof(*p) 
     + sizeof(tRowcnt)*nColUp                  /* StatAccum.anEq */
+    + sizeof(tRowcnt)*nColUp                  /* StatAccum.amxEq */
     + sizeof(tRowcnt)*nColUp;                 /* StatAccum.anDLt */
 #ifdef SQLITE_ENABLE_STAT4
   if( mxSample ){
@@ -447,7 +449,8 @@ static void statInit(
   p->nKeyCol = nKeyCol;
   p->nSkipAhead = 0;
   p->current.anDLt = (tRowcnt*)&p[1];
-  p->current.anEq = &p->current.anDLt[nColUp];
+  p->current.amxEq = &p->current.anDLt[nColUp];
+  p->current.anEq = &p->current.amxEq[nColUp];
 
 #ifdef SQLITE_ENABLE_STAT4
   p->mxSample = p->nLimit==0 ? mxSample : 0;
@@ -716,7 +719,10 @@ static void statPush(
 
   if( p->nRow==0 ){
     /* This is the first call to this function. Do initialization. */
-    for(i=0; i<p->nCol; i++) p->current.anEq[i] = 1;
+    for(i=0; i<p->nCol; i++){
+      p->current.anEq[i] = 1;
+      p->current.amxEq[i] = 1;
+    }
   }else{
     /* Second and subsequent calls get processed here */
 #ifdef SQLITE_ENABLE_STAT4
@@ -733,6 +739,9 @@ static void statPush(
 #ifdef SQLITE_ENABLE_STAT4
       if( p->mxSample ) p->current.anLt[i] += p->current.anEq[i];
 #endif
+      if( p->current.amxEq[i]<p->current.anEq[i] ){
+        p->current.amxEq[i] = p->current.anEq[i];
+      }
       p->current.anEq[i] = 1;
     }
   }
@@ -841,20 +850,18 @@ static void statGet(
     ** a key with the corresponding number of fields. In other words,
     ** if the index is on columns (a,b) and the sqlite_stat1 value is 
     ** "100 10 2", then SQLite estimates that:
-    **
-    **   * the index contains 100 rows,
-    **   * "WHERE a=?" matches 10 rows, and
-    **   * "WHERE a=? AND b=?" matches 2 rows.
+    **   |   | |
+    **   |   | `--  "WHERE a=? AND b=?" matches approximately 2 rows
+    **   |   `----  "WHERE a=?" matches approximately 10 rows
+    **   `--------  There are approximately 100 rows in the index total
     **
     ** If D is the count of distinct values and K is the total number of 
     ** rows, then each estimate is usually computed as:
     **
     **        I = (K+D-1)/D
     **
-    ** In other words, I is K/D rounded up to the next whole integer.
-    ** However, if I is between 1.0 and 1.1 (in other words if I is
-    ** close to 1.0 but just a little larger) then do not round up but
-    ** instead keep the I value at 1.0.
+    ** Adjustments to the I value are made in some cases.  See comments
+    ** in-line below.
     */
     sqlite3_str sStat;   /* Text of the constructed "stat" line */
     int i;               /* Loop counter */
@@ -863,9 +870,23 @@ static void statGet(
     sqlite3_str_appendf(&sStat, "%llu", 
         p->nSkipAhead ? (u64)p->nEst : (u64)p->nRow);
     for(i=0; i<p->nKeyCol; i++){
-      u64 nDistinct = p->current.anDLt[i] + 1;
-      u64 iVal = (p->nRow + nDistinct - 1) / nDistinct;
-      if( iVal==2 && p->nRow*10 <= nDistinct*11 ) iVal = 1;
+      u64 nDistinct, iVal, mx;
+      nDistinct = p->current.anDLt[i] + 1;
+      iVal = (p->nRow + nDistinct - 1) / nDistinct;
+      mx = p->current.amxEq[i];
+      if( nDistinct==1 && p->nLimit>0 ){
+        /* If we never saw more than a single value in a PRAGMA analysis_limit
+        ** search, then set the estimated number of matching rows to the
+        ** estimated number of rows in the index. */
+        iVal = p->nEst;
+      }else if( iVal<mx/8 ){
+        /* Never let the estimated number of matching rows be less than
+        ** 1/8th the greatest number of identical rows */
+        iVal = mx/8;
+      }else if( iVal==2 && p->nRow*10 <= nDistinct*11 ){
+        /* If the value is less than or equal to 1.1, round it down to 1.0 */
+        iVal = 1;
+      }
       sqlite3_str_appendf(&sStat, " %llu", iVal);
       assert( p->current.anEq[i] );
     }