SQLite

	del /Q mkkeywordhash.* keywordhash.h 2>NUL
	del /Q notasharedlib.* 2>NUL
	-rmdir /Q/S .deps 2>NUL
	-rmdir /Q/S .libs 2>NUL
	-rmdir /Q/S tsrc 2>NUL
	del /Q .target_source 2>NUL
	del /Q tclsqlite3.exe $(SQLITETCLH) $(SQLITETCLDECLSH) 2>NUL
	del /Q lsm.dll lsmtest.exe 2>NUL
	del /Q testloadext.dll 2>NUL
	del /Q testfixture.exe test.db 2>NUL
	del /Q LogEst.exe fts3view.exe rollback-test.exe showdb.exe dbdump.exe 2>NUL
	del /Q changeset.exe 2>NUL
	del /Q showjournal.exe showstat4.exe showwal.exe speedtest1.exe 2>NUL
	del /Q mptester.exe wordcount.exe rbu.exe srcck1.exe 2>NUL
	del /Q sqlite3.c sqlite3-*.c 2>NUL

     to locate the check-in desired, click on its information page link,
     then click on the "Tarball" or "ZIP Archive" links on the information
     page.

If you do want to use Fossil to check out the source tree, 
first install Fossil version 2.0 or later.
(Source tarballs and precompiled binaries available
[here](https://www.fossil-scm.org/fossil/uv/download.html).)
[here](https://www.fossil-scm.org/fossil/uv/download.html).  Fossil is
a stand-alone program.  To install, simply download or build the single 
executable file and put that file someplace on your $PATH.)
Then run commands like this:

        mkdir ~/sqlite
        cd ~/sqlite
        fossil clone https://www.sqlite.org/src sqlite.fossil
        fossil open sqlite.fossil
    

SQLite does not require [Tcl](http://www.tcl.tk/) to run, but a Tcl installation
is required by the makefiles (including those for MSVC).  SQLite contains
a lot of generated code and Tcl is used to do much of that code generation.
The makefiles also require AWK.

## Source Code Tour

Most of the core source files are in the **src/** subdirectory.  But
src/ also contains files used to build the "testfixture" test harness;
those file all begin with "test".  And src/ contains the "shell.c" file
which is the main program for the "sqlite3.exe" command-line shell and
the "tclsqlite.c" file which implements the bindings to SQLite from the
Tcl programming language.  (Historical note:  SQLite began as a Tcl
Most of the core source files are in the **src/** subdirectory.  The
**src/** folder also contains files used to build the "testfixture" test
harness. The names of the source files used by "testfixture" all begin
with "test".
The **src/** also contains the "shell.c" file
which is the main program for the "sqlite3.exe"
[command-line shell](https://sqlite.org/cli.html) and
the "tclsqlite.c" file which implements the
[TCL bindings](https://sqlite.org/tclsqlite.html) for SQLite.
(Historical note:  SQLite began as a Tcl
extension and only later escaped to the wild as an independent library.)

Test scripts and programs are found in the **test/** subdirectory.
There are other test suites for SQLite (see
[How SQLite Is Tested](http://www.sqlite.org/testing.html))
but those other test suites are
Addtional test code is found in other source repositories.
See [How SQLite Is Tested](http://www.sqlite.org/testing.html) for
additional information.
in separate source repositories.

The **ext/** subdirectory contains code for extensions.  The
Full-text search engine is in **ext/fts3**.  The R-Tree engine is in
**ext/rtree**.  The **ext/misc** subdirectory contains a number of
smaller, single-file extensions, such as a REGEXP operator.

The **tool/** subdirectory contains various scripts and programs used

The "target_source" make target will create a subdirectory "tsrc/" and
fill it with all the source files needed to build SQLite, both
manually-edited files and automatically-generated files.

The SQLite interface is defined by the **sqlite3.h** header file, which is
generated from src/sqlite.h.in, ./manifest.uuid, and ./VERSION.  The
[Tcl script](http://www.tcl.tk) at tool/mksqlite3h.tcl does the conversion.
The manifest.uuid file contains the SHA1 hash of the particular check-in
The manifest.uuid file contains the SHA3 hash of the particular check-in
and is used to generate the SQLITE\_SOURCE\_ID macro.  The VERSION file
contains the current SQLite version number.  The sqlite3.h header is really
just a copy of src/sqlite.h.in with the source-id and version number inserted
at just the right spots. Note that comment text in the sqlite3.h file is
used to generate much of the SQLite API documentation.  The Tcl scripts
used to generate that documentation are in a separate source repository.

The SQL language parser is **parse.c** which is generate from a grammar in
the src/parse.y file.  The conversion of "parse.y" into "parse.c" is done
by the [lemon](./doc/lemon.html) LALR(1) parser generator.  The source code
for lemon is at tool/lemon.c.  Lemon uses a
template for generating its parser.  A generic template is in tool/lempar.c,
for lemon is at tool/lemon.c.  Lemon uses the tool/lempar.c file as a
template for generating its parser.
but SQLite uses a slightly modified template found in src/lempar.c.

Lemon also generates the **parse.h** header file, at the same time it
generates parse.c. But the parse.h header file is
modified further (to add additional symbols) using the ./addopcodes.awk
AWK script.

The **opcodes.h** header file contains macros that define the numbers
corresponding to opcodes in the "VDBE" virtual machine.  The opcodes.h
file is generated by the scanning the src/vdbe.c source file.  The
AWK script at ./mkopcodeh.awk does this scan and generates opcodes.h.
A second AWK script, ./mkopcodec.awk, then scans opcodes.h to generate
the **opcodes.c** source file, which contains a reverse mapping from
opcode-number to opcode-name that is used for EXPLAIN output.

The **keywordhash.h** header file contains the definition of a hash table
that maps SQL language keywords (ex: "CREATE", "SELECT", "INDEX", etc.) into
the numeric codes used by the parse.c parser.  The keywordhash.h file is
generated by a C-language program at tool mkkeywordhash.c.

The **pragma.h** header file contains various definitions used to parse
and implement the PRAGMA statements.  The header is generated by a
script **tool/mkpragmatab.tcl**. If you want to add a new PRAGMA, edit
the **tool/mkpragmatab.tcl** file to insert the information needed by the
parser for your new PRAGMA, then run the script to regenerate the
**pragma.h** header file.

### The Amalgamation

All of the individual C source code and header files (both manually-edited
and automatically-generated) can be combined into a single big source file
**sqlite3.c** called "the amalgamation".  The amalgamation is the recommended
way of using SQLite in a larger application.  Combining all individual
source code files into a single big source code file allows the C compiler
to perform more cross-procedure analysis and generate better code.  SQLite
runs about 5% faster when compiled from the amalgamation versus when compiled
from individual source files.

The amalgamation is generated from the tool/mksqlite3c.tcl Tcl script.
First, all of the individual source files must be gathered into the tsrc/
subdirectory (using the equivalent of "make target_source") then the
tool/mksqlite3c.tcl script is run to copy them all together in just the
right order while resolving internal "#include" references.

The amalgamation source file is more than 100K lines long.  Some symbolic
The amalgamation source file is more than 200K lines long.  Some symbolic
debuggers (most notably MSVC) are unable to deal with files longer than 64K
lines.  To work around this, a separate Tcl script, tool/split-sqlite3c.tcl,
can be run on the amalgamation to break it up into a single small C file
called **sqlite3-all.c** that does #include on about five other files
named **sqlite3-1.c**, **sqlite3-2.c**, ..., **sqlite3-5.c**.  In this way,
all of the source code is contained within a single translation unit so
that the compiler can do extra cross-procedure optimization, but no
individual source file exceeds 32K lines in length.

## How It All Fits Together

SQLite is modular in design.
See the [architectural description](http://www.sqlite.org/arch.html)
for details. Other documents that are useful in
(helping to understand how SQLite works include the
[file format](http://www.sqlite.org/fileformat2.html) description,
the [virtual machine](http://www.sqlite.org/vdbe.html) that runs
the [virtual machine](http://www.sqlite.org/opcode.html) that runs
prepared statements, the description of
[how transactions work](http://www.sqlite.org/atomiccommit.html), and
the [overview of the query planner](http://www.sqlite.org/optoverview.html).

Unfortunately, years of effort have gone into optimizating SQLite, both
Years of effort have gone into optimizating SQLite, both
for small size and high performance.  And optimizations tend to result in
complex code.  So there is a lot of complexity in the SQLite implementation.
complex code.  So there is a lot of complexity in the current SQLite
implementation.  It will not be the easiest library in the world to hack.

Key files:

  *  **sqlite.h.in** - This file defines the public interface to the SQLite
     library.  Readers will need to be familiar with this interface before
     trying to understand how the library works internally.

  *  **sqliteInt.h** - this header file defines many of the data objects
     used internally by SQLite.

  *  **parse.y** - This file describes the LALR(1) grammer that SQLite uses
  *  **parse.y** - This file describes the LALR(1) grammar that SQLite uses
     to parse SQL statements, and the actions that are taken at each step
     in the parsing process.

  *  **vdbe.c** - This file implements the virtual machine that runs
     prepared statements.  There are various helper files whose names
     begin with "vdbe".  The VDBE has access to the vdbeInt.h header file
     which defines internal data objects.  The rest of SQLite interacts

     is the file that, when linked against sqlite3.a, generates the
     "sqlite3.exe" command-line shell.

  *  **tclsqlite.c** - This file implements the Tcl bindings for SQLite.  It
     is not part of the core SQLite library.  But as most of the tests in this
     repository are written in Tcl, the Tcl language bindings are important.

There are many other source files.  Each has a suscinct header comment that
There are many other source files.  Each has a succinct header comment that
describes its purpose and role within the larger system.


## Contacts

The main SQLite webpage is [http://www.sqlite.org/](http://www.sqlite.org/)
with geographically distributed backup servers at
with geographically distributed backups at
[http://www2.sqlite.org/](http://www2.sqlite.org) and
[http://www3.sqlite.org/](http://www3.sqlite.org).

# exception. But since bm25() can now used the cached structure record,
# it never sees the corruption introduced by funk() and so the following 
# statement no longer fails.
#
do_catchsql_test 16.2 {
  SELECT funk(), bm25(n1), funk() FROM n1 WHERE n1 MATCH 'a+b+c+d'
} {0 {{} -1e-06 {}}}
# {1 {SQL logic error or missing database}}
# {1 {SQL logic error}}

#-------------------------------------------------------------------------
#
reset_db
do_execsql_test 17.1 {
  CREATE VIRTUAL TABLE b2 USING fts5(x, detail=%DETAIL%);
  INSERT INTO b2 VALUES('a');

  SELECT rowid FROM t9('a*')
} {1}

}


finish_test

  INSERT INTO x1 VALUES($doc);
}

} ;# foreach_detail_mode...


finish_test

} {
  do_execsql_test 2.3.$tn {
    SELECT fts5_expr_tcl($expr, 'N $x')
  } [list $tclexpr]
}

finish_test

  catchsql COMMIT
}

}

finish_test

    SELECT fts5_test_phrasecount(t9) FROM t9 WHERE t9 MATCH $q LIMIT 1
  } $cnt
}

}

finish_test

do_execsql_test 5.1 {
  SELECT snippet(p1, 0, '[', ']', '...', 6) FROM p1('x');
} {{[x] a a a a a...}}

} ;# foreach_detail_mode 

finish_test

  }
}

} ;# foreach_detail_mode


finish_test

} {10000}

} ;# foreach_detail_mode

#db eval {SELECT rowid, fts5_decode(rowid, block) aS r FROM t1_data} {puts $r}

finish_test

do_execsql_test 1.2 {
  INSERT INTO t1(t1) VALUES('integrity-check');
}
}


finish_test

  }
}

do_execsql_test 2.0 { INSERT INTO t1(t1) VALUES('integrity-check') }


finish_test

  {[a b c] [c d e]}
  {[a b c d e]}
}

}

finish_test

  1 ""
  2 "fname"
  3 "fname(X'234ab')"
  4 "myfunc(-1.,'abc')"
} {
  do_test 2.2.$tn {
    catchsql { INSERT INTO ft1(ft1, rank) VALUES('rank', $defn) }
  } {1 {SQL logic error or missing database}}
  } {1 {SQL logic error}}
}

#-------------------------------------------------------------------------
# Assorted tests of the tcl interface for creating extension functions.
#

do_execsql_test 3.1 {

  SELECT *, rank FROM t3 WHERE t3 MATCH 'a' AND rank MATCH NULL
} {1 {parse error in rank function: }}

} ;# foreach_detail_mode


finish_test

do_execsql_test 3.3 {
  COMMIT;
  SELECT rowid FROM abc WHERE abc MATCH 'a';
} {1 2}

finish_test

} {
  do_auto_test 4.$tn yy $expr
}



finish_test

} {
  9 10
}



finish_test

db eval { 
  SELECT aux_function_2(f1, 2, 'A'), aux_function_2(f1, 2, 'B') 
  FROM f1 WHERE f1 MATCH 'a'
  ORDER BY rowid ASC
}

finish_test

    set doc [string repeat "$t " 150000000]
    execsql { INSERT INTO t1 VALUES($doc) }
  }
  execsql { INSERT INTO t1(t1) VALUES('integrity-check') }
} {}

finish_test

    do_execsql_test 2.[string range $v 0 0] {
      SELECT rowid FROM t1($v) ORDER BY rowid DESC
    } [lsort -integer -decr $res]
  }
}

finish_test

  do_catchsql_test 4.1 {
    SELECT * FROM t1 WHERE rowid MATCH 'a'
  } {1 {unable to use function MATCH in the requested context}}
}


finish_test

  5 "f1(x':;')"
  6 "f1(x'[]')"
  7 "f1(x'{}')"
  8 "f1('abc)"
} {
  do_catchsql_test 3.$tn {
    INSERT INTO t1(t1, rank) VALUES('rank', $val);
  } {1 {SQL logic error or missing database}}
  } {1 {SQL logic error}}
}

#-------------------------------------------------------------------------
# The parsing of SQL literals specified as part of 'rank' options.
#
do_execsql_test 4.0 {
  CREATE VIRTUAL TABLE zzz USING fts5(one);

#   9.5.* 'hashsize' options.
#
do_execsql_test 9.0 {
  CREATE VIRTUAL TABLE abc USING fts5(a, b);
} {}
do_catchsql_test 9.1.1 {
  INSERT INTO abc(abc, rank) VALUES('pgsz', -5);
} {1 {SQL logic error or missing database}}
} {1 {SQL logic error}}
do_catchsql_test 9.1.2 {
  INSERT INTO abc(abc, rank) VALUES('pgsz', 50000000);
} {1 {SQL logic error or missing database}}
} {1 {SQL logic error}}
do_catchsql_test 9.1.3 {
  INSERT INTO abc(abc, rank) VALUES('pgsz', 66.67);
} {1 {SQL logic error or missing database}}
} {1 {SQL logic error}}

do_catchsql_test 9.2.1 {
  INSERT INTO abc(abc, rank) VALUES('automerge', -5);
} {1 {SQL logic error or missing database}}
} {1 {SQL logic error}}
do_catchsql_test 9.2.2 {
  INSERT INTO abc(abc, rank) VALUES('automerge', 50000000);
} {1 {SQL logic error or missing database}}
} {1 {SQL logic error}}
do_catchsql_test 9.2.3 {
  INSERT INTO abc(abc, rank) VALUES('automerge', 66.67);
} {1 {SQL logic error or missing database}}
} {1 {SQL logic error}}
do_execsql_test 9.2.4 {
  INSERT INTO abc(abc, rank) VALUES('automerge', 1);
} {}

do_catchsql_test 9.3.1 {
  INSERT INTO abc(abc, rank) VALUES('crisismerge', -5);
} {1 {SQL logic error or missing database}}
} {1 {SQL logic error}}
do_catchsql_test 9.3.2 {
  INSERT INTO abc(abc, rank) VALUES('crisismerge', 66.67);
} {1 {SQL logic error or missing database}}
} {1 {SQL logic error}}
do_execsql_test 9.3.3 {
  INSERT INTO abc(abc, rank) VALUES('crisismerge', 1);
} {}
do_execsql_test 9.3.4 {
  INSERT INTO abc(abc, rank) VALUES('crisismerge', 50000000);
} {}

do_catchsql_test 9.4.1 {
  INSERT INTO abc(abc, rank) VALUES('nosuchoption', 1);
} {1 {SQL logic error or missing database}}
} {1 {SQL logic error}}

do_catchsql_test 9.5.1 {
  INSERT INTO abc(abc, rank) VALUES('hashsize', 'not an integer');
} {1 {SQL logic error or missing database}}
} {1 {SQL logic error}}
do_catchsql_test 9.5.2 {
  INSERT INTO abc(abc, rank) VALUES('hashsize', -500000);
} {1 {SQL logic error or missing database}}
} {1 {SQL logic error}}
do_catchsql_test 9.5.3 {
  INSERT INTO abc(abc, rank) VALUES('hashsize', 500000);
} {0 {}}

#-------------------------------------------------------------------------
# Too many prefix indexes. Maximum allowed is 31.
#

} {
  set res [list 1 {malformed detail=... directive}]
  do_catchsql_test 11.$tn "CREATE VIRTUAL TABLE f1 USING fts5(x, $opt)" $res
}

do_catchsql_test 12.1 {
  INSERT INTO t1(t1, rank) VALUES('rank', NULL);;
} {1 {SQL logic error or missing database}}
} {1 {SQL logic error}}

#-------------------------------------------------------------------------
# errors in the 'usermerge' option
#
do_execsql_test 13.0 {
  CREATE VIRTUAL TABLE tt USING fts5(ttt);
}
foreach {tn val} {
  1     -1
  2     4.2
  3     17
  4     1
} {
  set sql "INSERT INTO tt(tt, rank) VALUES('usermerge', $val)"
  do_catchsql_test 13.$tn $sql {1 {SQL logic error or missing database}}
  do_catchsql_test 13.$tn $sql {1 {SQL logic error}}
}

finish_test

  REPLACE INTO tbl VALUES(1, '4 5 6', '3 2 1');
  DELETE FROM tbl WHERE a=100;

  INSERT INTO fts_idx(fts_idx) VALUES('integrity-check');
}

finish_test

do_execsql_test 6.2 {
  DROP TABLE xx;
  SELECT name FROM sqlite_master;
} {}


finish_test

do_catchsql_test 3.1 {
  DELETE FROM t3_content WHERE rowid = 3;
  SELECT * FROM t3 WHERE t3 MATCH 'o';
} {1 {database disk image is malformed}}

finish_test

do_catchsql_test 6.2 {
  SELECT colsize(x5, 0) FROM x5 WHERE x5 MATCH 'a'
} {1 SQLITE_CORRUPT_VTAB}


sqlite3_fts5_may_be_corrupt 0
finish_test

} {}
do_catchsql_test 9.2.2 {
  SELECT * FROM t1('one AND two');
} {1 {database disk image is malformed}}

sqlite3_fts5_may_be_corrupt 0
finish_test

  for {set i 0} {$i < 5} {incr i} {
    execsql { INSERT INTO t1(t1, rank) VALUES('merge', 1) }
    execsql { INSERT INTO t1(t1) VALUES('integrity-check') }
  }
} {}

finish_test

    (SELECT sum(length(block)) from t2_data) <
    (SELECT sum(length(block)) from t3_data)
} {1}



finish_test

  } {}

  do_determin_test 1.4
}


finish_test

}

} ;# foreach_detail_mode



finish_test

do_execsql_test 1.2 {
  INSERT INTO ccc(ccc) VALUES('integrity-check');
}


finish_test

do_execsql_test 3.3 {
  SELECT rowid, bm25(e1) FROM e1 WHERE e1 MATCH '"/" OR "just"' ORDER BY rank;
} {1 -1e-06}



finish_test

    if {$ls != "2 0"} { error "fts5_level_segs says {$ls}" }
  }
}



finish_test

    );
  }
} -test {
  faultsim_test_result {0 {}}
}

finish_test

} -test {
  faultsim_test_result [list 0 {}]
}



finish_test

} -body {
  db eval { ALTER TABLE "tbl one" RENAME TO "tbl two" }
} -test {
  faultsim_test_result {0 {}}
}

finish_test

      9 x 1 {}
  ]]
}



finish_test

  faultsim_test_result {0 {}} {1 {initialization of fts5 failed: }}
  if {$testrc==0} {
    db eval { CREATE VIRTUAL TABLE temp.t1 USING fts5(x) }
  }
  db close
}
finish_test

do_faultsim_test 2.2 -faults oom-* -body {
  db eval { SELECT * FROM xy('""') }
} -test {
  faultsim_test_result {0 {}}
}

finish_test

  execsql { INSERT INTO x2(x2) VALUES('optimize') }
} -test {
  faultsim_test_result {0 {}} {1 SQLITE_NOMEM}
}


finish_test

  faultsim_test_result [list 0 {1 3}]
}


} ;# foreach_detail_mode...

finish_test

  sqlite3 db test.db
} -body {
  execsql { SELECT rowid FROM o2('a+b+c NOT xyz') }
} -test {
  faultsim_test_result {0 {1 2}}
}
finish_test

  execsql { SELECT rowid FROM t1('{a b c} : (a AND d)') }
} -test {
  faultsim_test_result {0 {2 3}}
}


finish_test

      execsql { INSERT INTO x8 VALUES( rnddoc(5) ); }
    }
  } msg] $msg
} {1 {database or disk is full}}


finish_test

reset_db
do_catchsql_test 4.1 {
  CREATE VIRTUAL TABLE f2 USING fts5(o, t);
  SELECT * FROM f2('(8 AND 9)`AND 10');
} {1 {fts5: syntax error near "`"}}

finish_test

      INSERT INTO t2 VALUES($small || ' ' || $big);
    }
  } {}

} ;# foreach_detail_mode

finish_test

      if {$res == [lsort -integer $res2]} { incr ok }
    }
    set ok
  } {1000}
}

finish_test

do_execsql_test 1.6 {
  INSERT INTO t1(rowid, str) SELECT rowid+10, x FROM x1;
  SELECT last_insert_rowid();
} {14}


finish_test

} {1 1 abc 2 {} {}}

do_execsql_test 1.2 {
  SELECT * FROM t1 LEFT JOIN vt ON (vt MATCH 'abc')
} {1 abc 2 abc}

finish_test

} {}

do_catchsql_test 14.2 {
  SELECT matchinfo(x1, 'd') FROM x1('a b c');
} {1 {unrecognized matchinfo flag: d}}

finish_test

do_execsql_test 6.3 {
  INSERT INTO g1(g1) VALUES('integrity-check');
}



finish_test

do_execsql_test 1.2 {
  INSERT INTO t1(t1) VALUES('integrity-check');
}

}

finish_test

    sql1 { INSERT INTO t1 VALUES('a b c') }
    sql3 { INSERT INTO t1(t1) VALUES('integrity-check') }
  } {}

};# do_multiclient_test
};# foreach_detail_mode
finish_test

do_near_test 1.23 "a b c d e f g h i" { NEAR(a+b+c+d i b+c, 4) } 0

do_near_test 1.24 "a b c d e f g h i" { NEAR(i a+b+c+d b+c, 5) } 1
do_near_test 1.25 "a b c d e f g h i" { NEAR(i a+b+c+d b+c, 4) } 0


finish_test

    UPDATE ttt SET x = 'A B C' WHERE rowid = 4;
    INSERT INTO ttt(rowid, x) VALUES(6, 'd e f');
  COMMIT;
} {}
do_test 4.2.2 { fts5_level_segs ttt } {3}

finish_test

  do_execsql_test 2.$tn.5 {
    INSERT INTO t1(t1) VALUES('integrity-check');
  }

  do_test 2.$tn.6 { fts5_segcount t1 } 1
}
finish_test

  FROM t3('a:f+f')
} {
  31 {h *f f*} {i j g e c} {j j f c a i j} 
  50 {*f f* c} {f f b i i} {f f a j e c i}
}

finish_test

  0 0 0 {SCAN TABLE f1 VIRTUAL TABLE INDEX 2:}
}




finish_test

    lindex [sqlite3_fts5_tokenize db porter $in] 0
  } $out
  incr i
}


finish_test

    lindex [sqlite3_fts5_tokenize db porter $in] 0
  } $out
  incr i
}


finish_test

    do_execsql_test 7.$tn {
      SELECT md5sum(id, block) FROM tt_data
    } [list $::checksum]
  }
}

finish_test

    } {}
    incr ret
  }
}


finish_test

  VTest MATCH 'wrinkle in time OR a wrinkle in time' ORDER BY rank;
} {{wrinkle in time} {Bill Smith}}




finish_test

  CREATE VIRTUAL TABLE nc USING fts5(doc, content=);
}

do_catchsql_test 2.2 {
  INSERT INTO nc(nc) VALUES('rebuild');
} {1 {'rebuild' may not be used with a contentless fts5 table}}
finish_test

  }
  set res
} {500 400 300}



finish_test

} {36}

#db eval {SELECT rowid, fts5_decode_none(rowid, block) aS r FROM x5_data} {puts $r}



finish_test

do_execsql_test 17.6 { 
  SELECT * FROM t2('x:b* OR y:a*') WHERE rowid>55
}

#db eval {SELECT rowid, fts5_decode_none(rowid, block) aS r FROM t2_data} {puts $r}
  
finish_test

} 
do_execsql_test 4.6 {
  SELECT * FROM t2('ab + xyz');
}


finish_test

do_execsql_test 7.1.2 {
  INSERT INTO t2(t2) VALUES('integrity-check');
}

} ;# foreach_detail_mode

finish_test

}

}
}

finish_test

do_catchsql_test 2.0 {
  CREATE VIRTUAL TABLE tX USING fts5tokenize(nosuchtokenizer);
} {1 {vtable constructor failed: tX}}

do_catchsql_test 2.1 {
  CREATE VIRTUAL TABLE t4 USING fts5tokenize;
  SELECT * FROM t4;
} {1 {SQL logic error or missing database}}
} {1 {SQL logic error}}


finish_test

set ::flags [list]
do_execsql_test 9.5.1 { SELECT * FROM t1('"abc xyz*"'); } {}
do_test 9.5.2 { set ::flags } {query}


finish_test

  SELECT 't1' FROM t1 WHERE t1 MATCH '\xE0\xE8\xEC';
  SELECT 't2' FROM t2 WHERE t2 MATCH '\xE0\xE8\xEC';
  SELECT 't3' FROM t3 WHERE t3 MATCH '\xE0\xE8\xEC';
" {t1 t2}


finish_test

  }
  append str {'");}
  execsql $str
} {}


finish_test

  INSERT INTO t4(t4, rowid, a, b, c) VALUES('delete', 20, 'j k l', '', 'p q r');
  DELETE FROM x4 WHERE rowid=20;
  INSERT INTO t4(t4) VALUES('integrity-check');
} {}


finish_test

} {}
do_execsql_test 2.2.integrity {
  INSERT INTO x2(x2) VALUES('integrity-check');
}

}
finish_test

  db close
  sqlite3 db test.db
  catchsql { SELECT * FROM t1 WHERE t1 MATCH 'a' }
} {1 {invalid fts5 file format (found 0, expected 4) - run 'rebuild'}}


finish_test

#
# This Makefile is designed for use with main.mk in the root directory of
# this project. After including main.mk, the users makefile should contain:
# This Makefile is designed for use with Makefile.msc in the root directory
# of this project.  The Makefile.msc should contain:
#
#    LSMDIR=$(TOP)\ext\lsm1\
#    include $(LSMDIR)\Makefile.msc
#    LSMDIR=$(TOP)\ext\lsm1
#    !INCLUDE $(LSMDIR)\Makefile.msc
#
# The most useful targets are [lsmtest.exe] and [lsm.dll].
#

LSMOBJ    = \
  lsm_ckpt.lo \
  lsm_file.lo \

             $(LSMDIR)\lsm-test\lsmtest9.c                                   \
             $(LSMDIR)\lsm-test\lsmtest_datasource.c \
             $(LSMDIR)\lsm-test\lsmtest_func.c $(LSMDIR)\lsm-test\lsmtest_io.c  \
             $(LSMDIR)\lsm-test\lsmtest_main.c $(LSMDIR)\lsm-test\lsmtest_mem.c \
             $(LSMDIR)\lsm-test\lsmtest_tdb.c $(LSMDIR)\lsm-test\lsmtest_tdb3.c \
             $(LSMDIR)\lsm-test\lsmtest_util.c $(LSMDIR)\lsm-test\lsmtest_win32.c

# all: lsm.dll
# all: lsm.dll lsmtest.exe

LSMOPTS = -DLSM_MUTEX_WIN32=1 -I$(LSMDIR)
LSMOPTS = $(NO_WARN) -DLSM_MUTEX_WIN32=1 -I$(LSMDIR)

!IF $(DEBUG)>2
LSMOPTS = $(LSMOPTS) -DLSM_DEBUG=1
!ENDIF

!IF $(MEMDEBUG)!=0
LSMOPTS = $(LSMOPTS) -DLSM_DEBUG_MEM=1
!ENDIF

lsm_ckpt.lo:	$(LSMDIR)\lsm_ckpt.c $(LSMHDR) $(SQLITE3H)
	$(LTCOMPILE) $(LSMOPTS) -c $(LSMDIR)\lsm_ckpt.c

lsm_file.lo:	$(LSMDIR)\lsm_file.c $(LSMHDR) $(SQLITE3H)
	$(LTCOMPILE) $(LSMOPTS) -c $(LSMDIR)\lsm_file.c

lsm_vtab.lo:	$(LSMDIR)\lsm_vtab.c $(LSMHDR) $(SQLITE3H)
	$(LTCOMPILE) $(LSMOPTS) -c $(LSMDIR)\lsm_vtab.c

lsm.dll:	$(LSMOBJ)
	$(LD) $(LDFLAGS) $(LTLINKOPTS) $(LTLIBPATHS) /DLL /OUT:$@ $(LSMOBJ)

lsmtest.exe: $(LSMOBJ) $(LSMTESTSRC) $(LSMTESTHDR) $(LIBOBJS1)
	$(LTLINK) $(LSMOPTS) $(LSMTESTSRC) /link $(LSMOBJ) $(LIBOBJS1)
lsmtest.exe: $(LSMOBJ) $(LSMTESTSRC) $(LSMTESTHDR) $(LIBOBJ)
	$(LTLINK) $(LSMOPTS) $(LSMTESTSRC) /link $(LSMOBJ) $(LIBOBJ)

#ifndef __WRAPPER_INT_H_
#define __WRAPPER_INT_H_

#include "lsmtest_tdb.h"
#include "sqlite3.h"
#include "lsm.h"

#include <assert.h>
#include <stdarg.h>
#include <stdlib.h>
#include <string.h>
#include <stdio.h>
#ifndef _WIN32
# include <unistd.h>
#endif
#include <sys/types.h>
#include <sys/stat.h>
#include <fcntl.h>
#include <ctype.h>
#include <stdlib.h>
#include <errno.h>

#ifdef __cplusplus
extern "C" {
#endif

#ifdef _WIN32
# include "windows.h"
# define gettimeofday win32GetTimeOfDay
# define F_OK  (0)
# define sleep(sec) Sleep(1000 * (sec))
# define usleep(usec) Sleep((usec) / 1000)
# define usleep(usec) Sleep(((usec) + 999) / 1000)
# ifdef _MSC_VER
#  include <io.h>
#  define snprintf _snprintf
#  define fsync(fd) FlushFileBuffers((HANDLE)_get_osfhandle((fd)))
#  define fdatasync(fd) FlushFileBuffers((HANDLE)_get_osfhandle((fd)))
#  define __va_copy(dst,src) ((dst) = (src))
#  define ftruncate(fd,sz) ((_chsize_s((fd), (sz))==0) ? 0 : -1)

#define MIN(x,y) ((x)<(y) ? (x) : (y))
#define MAX(x,y) ((x)>(y) ? (x) : (y))

#define unused_parameter(x) (void)(x)

#define TESTDB_DEFAULT_PAGE_SIZE   4096
#define TESTDB_DEFAULT_CACHE_SIZE  2048

#ifndef _O_BINARY
# define _O_BINARY (0)
#endif

/*
** Ideally, these should be in wrapper.c. But they are here instead so that 
** they can be used by the C++ database wrappers in wrapper2.cc.
*/
typedef struct DatabaseMethods DatabaseMethods;
struct TestDb {

  return zRet;
}

int testControlDb(TestDb **ppDb){
#ifdef HAVE_KYOTOCABINET
  return tdb_open("kyotocabinet", "tmp.db", 1, ppDb);
#else
  return tdb_open("sqlite3", ":memory:", 1, ppDb);
  return tdb_open("sqlite3", "", 1, ppDb);
#endif
}

void testDatasourceFetch(
  TestDb *pDb,                    /* Database handle */
  Datasource *pData,
  int iKey,

  int nScan,
  int *pRc
){
  if( *pRc==0 ){
    int rc;
    int iScan = 0;
    lsm_cursor *pCsr;
    int (*xAdvance)(lsm_cursor *);
    int (*xAdvance)(lsm_cursor *) = 0;
    

    rc = lsm_csr_open(pDb, &pCsr);
    testOomAssertRc(pOom, rc);

    if( rc==LSM_OK ){
      if( bReverse ){

    lsm_csr_close(pCsr);
    *pRc = rc;
  }
}

#define LSMTEST6_TESTDB "testdb.lsm" 

#ifndef _WIN32
# include <unistd.h>
#endif
#include <sys/types.h>
#include <sys/stat.h>
#include <fcntl.h>

void testDeleteLsmdb(const char *zFile){
  char *zLog = testMallocPrintf("%s-log", zFile);
  char *zShm = testMallocPrintf("%s-shm", zFile);
  unlink(zFile);
  unlink(zLog);
  unlink(zShm);
  testFree(zLog);
  testFree(zShm);
}

static void copy_file(const char *zFrom, const char *zTo){
static void copy_file(const char *zFrom, const char *zTo, int isDatabase){

  if( access(zFrom, F_OK) ){
    unlink(zTo);
  }else{
    int fd1;
    int fd2;
    off_t sz;
    off_t i;
    struct stat buf;
    u8 *aBuf;

    fd1 = open(zFrom, O_RDONLY, 0644);
    fd2 = open(zTo, O_RDWR | O_CREAT, 0644);
    fd1 = open(zFrom, O_RDONLY | _O_BINARY, 0644);
    fd2 = open(zTo, O_RDWR | O_CREAT | _O_BINARY, 0644);

    fstat(fd1, &buf);
    sz = buf.st_size;
    ftruncate(fd2, sz);

    aBuf = testMalloc(4096);
    for(i=0; i<sz; i+=4096){
      int bLockPage = isDatabase && i == 0;
      int nByte = MIN(4096, sz - i);
      int nByte = MIN((bLockPage ? 4066 : 4096), sz - i);
      memset(aBuf, 0, 4096);
      read(fd1, aBuf, nByte);
      write(fd2, aBuf, nByte);
      if( bLockPage ){
        lseek(fd1, 4096, SEEK_SET);
        lseek(fd2, 4096, SEEK_SET);
      }
    }
    testFree(aBuf);

    close(fd1);
    close(fd2);
  }
}

void testCopyLsmdb(const char *zFrom, const char *zTo){
  char *zLog1 = testMallocPrintf("%s-log", zFrom);
  char *zLog2 = testMallocPrintf("%s-log", zTo);
  char *zShm1 = testMallocPrintf("%s-shm", zFrom);
  char *zShm2 = testMallocPrintf("%s-shm", zTo);

  unlink(zShm2);
  unlink(zLog2);
  unlink(zTo);
  copy_file(zFrom, zTo);
  copy_file(zLog1, zLog2);
  copy_file(zShm1, zShm2);
  copy_file(zFrom, zTo, 1);
  copy_file(zLog1, zLog2, 0);
  copy_file(zShm1, zShm2, 0);

  testFree(zLog1); testFree(zLog2); testFree(zShm1); testFree(zShm2);
}

/*
** File zFile is the path to a database. This function makes backups
** of the database file and its log as follows:

void testSaveDb(const char *zFile, const char *zAux){
  char *zLog = testMallocPrintf("%s-%s", zFile, zAux);
  char *zFileSave = testMallocPrintf("%s-save", zFile);
  char *zLogSave = testMallocPrintf("%s-%s-save", zFile, zAux);

  unlink(zFileSave);
  unlink(zLogSave);
  copy_file(zFile, zFileSave);
  copy_file(zLog, zLogSave);
  copy_file(zFile, zFileSave, 1);
  copy_file(zLog, zLogSave, 0);

  testFree(zLog); testFree(zFileSave); testFree(zLogSave);
}

/*
** File zFile is the path to a database. This function restores
** a backup of the database made by a previous call to testSaveDb().
** Specifically, it does the equivalent of:
**
**     cp $(zFile)-save         $(zFile)
**     cp $(zFile)-save-$(zAux) $(zFile)-$(zAux)
*/
void testRestoreDb(const char *zFile, const char *zAux){
  char *zLog = testMallocPrintf("%s-%s", zFile, zAux);
  char *zFileSave = testMallocPrintf("%s-save", zFile);
  char *zLogSave = testMallocPrintf("%s-%s-save", zFile, zAux);

  copy_file(zFileSave, zFile);
  copy_file(zLogSave, zLog);
  copy_file(zFileSave, zFile, 1);
  copy_file(zLogSave, zLog, 0);

  testFree(zLog); testFree(zFileSave); testFree(zLogSave);
}


static int lsmWriteStr(lsm_db *pDb, const char *zKey, const char *zVal){
  int nKey = strlen(zKey);
  int nVal = strlen(zVal);
  return lsm_insert(pDb, (void *)zKey, nKey, (void *)zVal, nVal);
}

static void setup_delete_db(){
static void setup_delete_db(void){
  testDeleteLsmdb(LSMTEST6_TESTDB);
}

/*
** Create a small database. With the following content:
**
**    "one"   -> "one"
**    "two"   -> "four"
**    "three" -> "nine"
**    "four"  -> "sixteen"
**    "five"  -> "twentyfive"
**    "six"   -> "thirtysix"
**    "seven" -> "fourtynine"
**    "eight" -> "sixtyfour"
*/
static void setup_populate_db(){
static void setup_populate_db(void){
  const char *azStr[] = {
    "one",   "one",
    "two",   "four",
    "three", "nine",
    "four",  "sixteen",
    "five",  "twentyfive",
    "six",   "thirtysix",

** Set up a database file with the following properties:
**
**   * Page size is 1024 bytes.
**   * Block size is 64 KB.
**   * Contains 5000 key-value pairs starting at 0 from the
**     datasource returned getDatasource().
*/
static void setup_populate_db2(){
static void setup_populate_db2(void){
  Datasource *pData;
  int ii;
  int rc;
  int nBlocksize = 64*1024;
  int nPagesize = 1024;
  int nWritebuffer = 4*1024;
  lsm_db *pDb;

  }
}

#include <stdio.h>
void testReadFile(const char *zFile, int iOff, void *pOut, int nByte, int *pRc){
  if( *pRc==0 ){
    FILE *fd;
    fd = fopen(zFile, "r");
    fd = fopen(zFile, "rb");
    if( fd==0 ){
      *pRc = 1;
    }else{
      if( 0!=fseek(fd, iOff, SEEK_SET) ){
        *pRc = 1;
      }else{
        assert( nByte>=0 );
        if( nByte!=fread(pOut, 1, nByte, fd) ){
        if( (size_t)nByte!=fread(pOut, 1, nByte, fd) ){
          *pRc = 1;
        }
      }
      fclose(fd);
    }
  }
}

void testWriteFile(
  const char *zFile, 
  int iOff, 
  void *pOut, 
  int nByte, 
  int *pRc
){
  if( *pRc==0 ){
    FILE *fd;
    fd = fopen(zFile, "r+");
    fd = fopen(zFile, "r+b");
    if( fd==0 ){
      *pRc = 1;
    }else{
      if( 0!=fseek(fd, iOff, SEEK_SET) ){
        *pRc = 1;
      }else{
        assert( nByte>=0 );
        if( nByte!=fwrite(pOut, 1, nByte, fd) ){
        if( (size_t)nByte!=fwrite(pOut, 1, nByte, fd) ){
          *pRc = 1;
        }
      }
      fclose(fd);
    }
  }
}

**    1) Check that the checksum of the database is zCksum. 
**    2) Write a few keys to the database. Then delete the same keys. 
**    3) Check that the checksum is zCksum.
**    4) Flush the db to disk and run a checkpoint. 
**    5) Check once more that the checksum is still zCksum.
*/
static void doLiveRecovery(const char *zDb, const char *zCksum, int *pRc){
  if( *pRc==LSM_OK ){
  const DatasourceDefn defn = {TEST_DATASOURCE_RANDOM, 20, 25, 100, 500};
  Datasource *pData;
  const char *zCopy = "testcopy.lsm";
  char zCksum2[TEST_CKSUM_BYTES];
  TestDb *pDb = 0;
  int rc;
    const DatasourceDefn defn = {TEST_DATASOURCE_RANDOM, 20, 25, 100, 500};
    Datasource *pData;
    const char *zCopy = "testcopy.lsm";
    char zCksum2[TEST_CKSUM_BYTES];
    TestDb *pDb = 0;
    int rc;

  pData = testDatasourceNew(&defn);
    pData = testDatasourceNew(&defn);

  testCopyLsmdb(zDb, zCopy);
  rc = tdb_lsm_open("test_no_recovery=1", zCopy, 0, &pDb);
  if( rc==0 ){
    ShmHeader *pHdr;
    lsm_db *db;
    testCksumDatabase(pDb, zCksum2);
    testCompareStr(zCksum, zCksum2, &rc);
    testCopyLsmdb(zDb, zCopy);
    rc = tdb_lsm_open("test_no_recovery=1", zCopy, 0, &pDb);
    if( rc==0 ){
      ShmHeader *pHdr;
      lsm_db *db;
      testCksumDatabase(pDb, zCksum2);
      testCompareStr(zCksum, zCksum2, &rc);

    testWriteDatasourceRange(pDb, pData, 1, 10, &rc);
    testDeleteDatasourceRange(pDb, pData, 1, 10, &rc);
      testWriteDatasourceRange(pDb, pData, 1, 10, &rc);
      testDeleteDatasourceRange(pDb, pData, 1, 10, &rc);

    /* Test that the two tree-headers are now consistent. */
    pHdr = getShmHeader(zCopy);
    if( rc==0 && memcmp(&pHdr->hdr1, &pHdr->hdr2, sizeof(pHdr->hdr1)) ){
      rc = 1;
    }
    testFree(pHdr);
      /* Test that the two tree-headers are now consistent. */
      pHdr = getShmHeader(zCopy);
      if( rc==0 && memcmp(&pHdr->hdr1, &pHdr->hdr2, sizeof(pHdr->hdr1)) ){
        rc = 1;
      }
      testFree(pHdr);

    if( rc==0 ){
      int nBuf = 64;
      db = tdb_lsm(pDb);
      lsm_config(db, LSM_CONFIG_AUTOFLUSH, &nBuf);
      lsm_begin(db, 1);
      lsm_commit(db, 0);
      rc = lsm_work(db, 0, 0, 0);
    }
      if( rc==0 ){
        int nBuf = 64;
        db = tdb_lsm(pDb);
        lsm_config(db, LSM_CONFIG_AUTOFLUSH, &nBuf);
        lsm_begin(db, 1);
        lsm_commit(db, 0);
        rc = lsm_work(db, 0, 0, 0);
      }

    testCksumDatabase(pDb, zCksum2);
    testCompareStr(zCksum, zCksum2, &rc);
  }
      testCksumDatabase(pDb, zCksum2);
      testCompareStr(zCksum, zCksum2, &rc);
    }

  testDatasourceFree(pData);
  testClose(&pDb);
  testDeleteLsmdb(zCopy);
  *pRc = rc;
    testDatasourceFree(pData);
    testClose(&pDb);
    testDeleteLsmdb(zCopy);
    *pRc = rc;
  }
}

static void doWriterCrash1(int *pRc){
  const int nWrite = 2000;
  const int nStep = 10;
  const int iWriteStart = 20000;
  int rc = 0;

**     $ echo "2M@6M 1492K@4M S 4096@4K S" | lsmtest io testfile.db 4096 
**     3544K written in 127 ms
**
*/

#include "lsmtest.h"

#include <sys/types.h>
#include <sys/stat.h>
#include <fcntl.h>
#ifndef _WIN32
# include <unistd.h>
#endif
#include <ctype.h>

typedef struct IoContext IoContext;

struct IoContext {
  int fd;
  int nWrite;
};

  if( safe_isdigit(c) ){
    i64 iOff = 0;
    int nByte = 0;
    int rc = 0;
    int nPg;
    int iPg;

    nByte = getNextSize(z, &z, &rc);
    nByte = (int)getNextSize(z, &z, &rc);
    if( rc || *z!='@' ) goto bad_command;
    z++;
    iOff = getNextSize(z, &z, &rc);
    if( rc || (safe_isspace(*z)==0 && *z!='\0') ) goto bad_command;
    if( pzOut ) *pzOut = z;

    nPg = (nByte+pgsz-1) / pgsz;
    lseek(pCtx->fd, iOff, SEEK_SET);
    lseek(pCtx->fd, (off_t)iOff, SEEK_SET);
    for(iPg=0; iPg<nPg; iPg++){
      write(pCtx->fd, aData, pgsz);
    }
    pCtx->nWrite += nByte/1024;

    return 0;
  }

  if( nArg<2 ){
    testPrintUsage("FILE PGSZ ?CMD-1 ...?");
    return -1;
  }
  zFile = azArg[0];
  zPgsz = azArg[1];

  pgsz = getNextSize(zPgsz, 0, &rc);
  pgsz = (int)getNextSize(zPgsz, 0, &rc);
  if( pgsz<=0 ){
    testPrintError("Ridiculous page size: %d", pgsz);
    return -1;
  }
  aData = malloc(pgsz);
  memset(aData, 0x77, pgsz);

  ctx.fd = open(zFile, O_RDWR|O_CREAT, 0644);
  ctx.fd = open(zFile, O_RDWR|O_CREAT|_O_BINARY, 0644);
  if( ctx.fd<0 ){
    perror("open: ");
    return -1;
  }

  if( nArg==2 ){
    readStdin(&zStdin);

#include "stdarg.h"
#include "lsmtest.h"
#include "stdio.h"
#include "assert.h"
#include "string.h"
#include "stdlib.h"

#include <sqlite3.h>

#ifndef _WIN32
# include <unistd.h>
#endif
#include <sys/types.h>
#include <sys/stat.h>
#include <fcntl.h>
#include <errno.h>


void test_failed(){ 
  assert( 0 );
  return; 
}

#define testSetError(rc) testSetErrorFunc(rc, pRc, __FILE__, __LINE__)

      pLog = fopen("/tmp/speed.log", "w");
      tdb_lsm_write_hook(pDb, do_speed_write_hook2, (void *)pLog);
#endif
  
      testTimeInit();
      for(i=0; i<nRow; i+=nStep){
        int iStep;
        int nWrite1, nWrite2;
        int nWrite1 = 0, nWrite2 = 0;
        testCaseProgress(i, nRow, testCaseNDot(), &iDot);
        if( pLsm ) lsm_info(pLsm, LSM_INFO_NWRITE, &nWrite1);
        for(iStep=0; iStep<nStep; iStep++){
          u32 aKey[4];                  /* 16-byte key */
          u32 aVal[25];                 /* 100 byte value */
          testPrngArray(i+iStep, aKey, ArraySize(aKey));
          testPrngArray(i+iStep, aVal, ArraySize(aVal));

  }else{
    printf("Sync after each %d blocks.\n", nSync);
  }

  printf("Preparing file... ");
  fflush(stdout);
  unlink("writer.out");
  fd = open("writer.out", O_RDWR|O_CREAT, 0664);
  fd = open("writer.out", O_RDWR|O_CREAT|_O_BINARY, 0664);
  if( fd<0 ){
    testPrintError("open(): %d - %s\n", errno, strerror(errno));
    return -1;
  }
  testTimeInit();
  for(i=0; i<nBlock; i++){
    int iPg;

  /* Mutex to protect pFirst and aHash */
  void (*xEnterMutex)(TmGlobal*); /* Call this to enter the mutex */
  void (*xLeaveMutex)(TmGlobal*); /* Call this to leave mutex */
  void (*xDelMutex)(TmGlobal*);   /* Call this to delete mutex */
  void *pMutex;                   /* Mutex handle */

  void *xSaveMalloc;
  void *xSaveRealloc;
  void *xSaveFree;
  void *(*xSaveMalloc)(void *, size_t);
  void *(*xSaveRealloc)(void *, void *, size_t);
  void (*xSaveFree)(void *, void *);

  /* OOM injection scheduling. If nCountdown is greater than zero when a 
  ** malloc attempt is made, it is decremented. If this means nCountdown 
  ** transitions from 1 to 0, then the allocation fails. If bPersist is true 
  ** when this happens, nCountdown is then incremented back to 1 (so that the 
  ** next attempt fails too).  
  */

static void tmFree(TmGlobal *pTm, void *p){
  if( p ){
    TmBlockHdr *pHdr;
    u8 *pUser = (u8 *)p;

    tmEnterMutex(pTm);
    pHdr = (TmBlockHdr *)&pUser[BLOCK_HDR_SIZE * -1];
    pHdr = (TmBlockHdr *)(pUser - BLOCK_HDR_SIZE);
    assert( pHdr->iForeGuard==FOREGUARD );
    assert( 0==memcmp(&pUser[pHdr->nByte], &rearguard, 4) );

    if( pHdr->pPrev ){
      assert( pHdr->pPrev->pNext==pHdr );
      pHdr->pPrev->pNext = pHdr->pNext;
    }else{

static void *tmRealloc(TmGlobal *pTm, void *p, int nByte){
  void *pNew;

  pNew = tmMalloc(pTm, nByte);
  if( pNew && p ){
    TmBlockHdr *pHdr;
    u8 *pUser = (u8 *)p;
    pHdr = (TmBlockHdr *)&pUser[BLOCK_HDR_SIZE * -1];
    pHdr = (TmBlockHdr *)(pUser - BLOCK_HDR_SIZE);
    memcpy(pNew, p, MIN(nByte, pHdr->nByte));
    tmFree(pTm, p);
  }
  return pNew;
}

static void tmMallocOom(

  pMutex->pEnv = pEnv;
  pEnv->xMutexStatic(pEnv, LSM_MUTEX_HEAP, &pMutex->pMutex);
  pGlobal->xEnterMutex = tmLsmMutexEnter;
  pGlobal->xLeaveMutex = tmLsmMutexLeave;
  pGlobal->xDelMutex = tmLsmMutexDel;
  pGlobal->pMutex = (void *)pMutex;

  pGlobal->xSaveMalloc = (void *)pEnv->xMalloc;
  pGlobal->xSaveRealloc = (void *)pEnv->xRealloc;
  pGlobal->xSaveFree = (void *)pEnv->xFree;
  pGlobal->xSaveMalloc = pEnv->xMalloc;
  pGlobal->xSaveRealloc = pEnv->xRealloc;
  pGlobal->xSaveFree = pEnv->xFree;

  /* Set up pEnv to the use the new TmGlobal */
  pEnv->pMemCtx = (void *)pGlobal;
  pEnv->xMalloc = tmLsmEnvMalloc;
  pEnv->xRealloc = tmLsmEnvRealloc;
  pEnv->xFree = tmLsmEnvFree;
}

    nLib++;
  }
  zSpec = &zLib[nLib];
  while( *zSpec==' ' ) zSpec++;
  if( *zSpec=='\0' ) zSpec = 0;

  for(i=0; i<ArraySize(aLib); i++){
    if( (int)strlen(aLib[i].zName)==nLib
    if( strlen(aLib[i].zName)==nLib && 0==memcmp(zLib, aLib[i].zName, nLib) ){
        && 0==memcmp(zLib, aLib[i].zName, nLib) ){
      rc = aLib[i].xOpen(zSpec, (zDb ? zDb : aLib[i].zDefaultDb), bClear, ppDb);
      if( rc==0 ){
        (*ppDb)->zLibrary = aLib[i].zName;
      }
      break;
    }
  }

  lsm_env *pRealEnv = tdb_lsm_env();
  LsmFile *p = (LsmFile *)pFile;
  LsmDb *pDb = p->pDb;

  if( pDb->bCrashed ) return LSM_IOERR;

  if( pDb->bPrepareCrash ){
    FileData *pData = &pDb->aFile[p->bLog];
    FileData *pData2 = &pDb->aFile[p->bLog];
    int iFirst;                 
    int iLast;
    int iSector;

    iFirst = (iOff / pDb->szSector);
    iLast =  ((iOff + nData - 1) / pDb->szSector);
    iFirst = (int)(iOff / pDb->szSector);
    iLast =  (int)((iOff + nData - 1) / pDb->szSector);

    if( pData->nSector<(iLast+1) ){
    if( pData2->nSector<(iLast+1) ){
      int nNew = ( ((iLast + 1) + 63) / 64 ) * 64;
      assert( nNew>iLast );
      pData->aSector = (FileSector *)testRealloc(
          pData->aSector, nNew*sizeof(FileSector)
      pData2->aSector = (FileSector *)testRealloc(
          pData2->aSector, nNew*sizeof(FileSector)
      );
      memset(&pData->aSector[pData->nSector], 
          0, (nNew - pData->nSector) * sizeof(FileSector)
      memset(&pData2->aSector[pData2->nSector], 
          0, (nNew - pData2->nSector) * sizeof(FileSector)
      );
      pData->nSector = nNew;
      pData2->nSector = nNew;
    }

    for(iSector=iFirst; iSector<=iLast; iSector++){
      if( pData->aSector[iSector].aOld==0 ){
      if( pData2->aSector[iSector].aOld==0 ){
        u8 *aOld = (u8 *)testMalloc(pDb->szSector);
        pRealEnv->xRead(
            p->pReal, (lsm_i64)iSector*pDb->szSector, aOld, pDb->szSector
        );
        pData->aSector[iSector].aOld = aOld;
        pData2->aSector[iSector].aOld = aOld;
      }
    }
  }

  if( pDb->xWriteHook ){
    int rc;
    int nUs;

static int waitOnWorker(LsmDb *pDb){
  int rc;
  int nLimit = -1;
  int nSleep = 0;

  rc = lsm_config(pDb->db, LSM_CONFIG_AUTOFLUSH, &nLimit);
  do {
    int nOld, nNew, rc;
    rc = lsm_info(pDb->db, LSM_INFO_TREE_SIZE, &nOld, &nNew);
    if( rc!=LSM_OK ) return rc;
    int nOld, nNew, rc2;
    rc2 = lsm_info(pDb->db, LSM_INFO_TREE_SIZE, &nOld, &nNew);
    if( rc2!=LSM_OK ) return rc2;
    if( nOld==0 || nNew<(nLimit/2) ) break;
#ifdef LSM_MUTEX_PTHREADS
    mt_signal_worker(pDb, 0);
#endif
    usleep(5000);
    nSleep += 5;
  }while( 1 );

  testPrintError("syntax error at: \"%s\"\n", z);
  return 1;
}

int tdb_lsm_config_str(TestDb *pDb, const char *zStr){
  int rc = 0;
  if( tdb_lsm(pDb) ){
#ifdef LSM_MUTEX_PTHREADS
    int i;
#endif
    LsmDb *pLsm = (LsmDb *)pDb;

    rc = test_lsm_config_str(pLsm, pLsm->db, 0, zStr, 0);
#ifdef LSM_MUTEX_PTHREADS
    for(i=0; rc==0 && i<pLsm->nWorker; i++){
      rc = test_lsm_config_str(0, pLsm->aWorker[i].pWorker, 1, zStr, 0);
    }

#define TICKS_UNIX_EPOCH      (116444736000000000LL)

int win32GetTimeOfDay(
  struct timeval *tp,
  void *tzp
){
  FILETIME fileTime;
  ULARGE_INTEGER largeInteger;
  ULONGLONG ticks;
  ULONGLONG unixTicks;

  unused_parameter(tzp);
  memset(&fileTime, 0, sizeof(FILETIME));
  GetSystemTimeAsFileTime(&fileTime);
  memset(&largeInteger, 0, sizeof(ULARGE_INTEGER));
  largeInteger.LowPart = fileTime.dwLowDateTime;
  ticks = (ULONGLONG)fileTime.dwHighDateTime << 32;
  ticks |= (ULONGLONG)fileTime.dwLowDateTime;
  largeInteger.HighPart = fileTime.dwHighDateTime;
  ticks = largeInteger.QuadPart - TICKS_UNIX_EPOCH;
  tp->tv_sec = (long)(ticks / TICKS_PER_SECOND);
  ticks -= ((ULONGLONG)tp->tv_sec * TICKS_PER_SECOND);
  tp->tv_usec = (long)(ticks / TICKS_PER_MICROSECOND);
  unixTicks = ticks - TICKS_UNIX_EPOCH;
  tp->tv_sec = (long)(unixTicks / TICKS_PER_SECOND);
  unixTicks -= ((ULONGLONG)tp->tv_sec * TICKS_PER_SECOND);
  tp->tv_usec = (long)(unixTicks / TICKS_PER_MICROSECOND);

  return 0;
}
#endif

  u32 nWrite;                     /* Total number of pages written to disk */
};
#define LSM_INITIAL_SNAPSHOT_ID 11

/*
** Functions from file "lsm_ckpt.c".
*/
int lsmCheckpointWrite(lsm_db *, int, u32 *);
int lsmCheckpointWrite(lsm_db *, u32 *);
int lsmCheckpointLevels(lsm_db *, int, void **, int *);
int lsmCheckpointLoadLevels(lsm_db *pDb, void *pVal, int nVal);

int lsmCheckpointRecover(lsm_db *);
int lsmCheckpointDeserialize(lsm_db *, int, u32 *, Snapshot **);

int lsmCheckpointLoadWorker(lsm_db *pDb);

/**************************************************************************
** Start of functions from "lsm_file.c".
*/
int lsmFsOpen(lsm_db *, const char *, int);
int lsmFsOpenLog(lsm_db *, int *);
void lsmFsCloseLog(lsm_db *);
void lsmFsClose(FileSystem *);

int lsmFsUnmap(FileSystem *);

int lsmFsConfigure(lsm_db *db);

int lsmFsBlockSize(FileSystem *);
void lsmFsSetBlockSize(FileSystem *, int);
int lsmFsMoveBlock(FileSystem *pFS, Segment *pSeg, int iTo, int iFrom);

**
** The value of the nCkpt parameter includes the two checksum values at
** the end of the checkpoint. They are not used as inputs to the checksum 
** calculation. The checksum is based on the array of (nCkpt-2) integers
** at aCkpt[].
*/
static void ckptChecksum(u32 *aCkpt, u32 nCkpt, u32 *piCksum1, u32 *piCksum2){
  int i;
  u32 i;
  u32 cksum1 = 1;
  u32 cksum2 = 2;

  if( nCkpt % 2 ){
    cksum1 += aCkpt[nCkpt-3] & 0x0000FFFF;
    cksum2 += aCkpt[nCkpt-3] & 0xFFFF0000;
  }

  Segment *pSegment               /* Populate this structure */
){
  assert( pSegment->iFirst==0 && pSegment->iLastPg==0 );
  assert( pSegment->nSize==0 && pSegment->iRoot==0 );
  pSegment->iFirst = ckptGobble64(aIn, piIn);
  pSegment->iLastPg = ckptGobble64(aIn, piIn);
  pSegment->iRoot = ckptGobble64(aIn, piIn);
  pSegment->nSize = ckptGobble64(aIn, piIn);
  pSegment->nSize = (int)ckptGobble64(aIn, piIn);
  assert( pSegment->iFirst );
}

static int ckptSetupMerge(lsm_db *pDb, u32 *aInt, int *piIn, Level *pLevel){
  Merge *pMerge;                  /* Allocated Merge object */
  int nInput;                     /* Number of input segments in merge */
  int iIn = *piIn;                /* Next value to read from aInt[] */

    if( rc==LSM_OK && bInclFreelist ){
      nFree = aCkpt[iIn++];
      if( nFree ){
        pNew->freelist.aEntry = (FreelistEntry *)lsmMallocZeroRc(
            pDb->pEnv, sizeof(FreelistEntry)*nFree, &rc
        );
        if( rc==LSM_OK ){
          int i;
          for(i=0; i<nFree; i++){
            FreelistEntry *p = &pNew->freelist.aEntry[i];
          int j;
          for(j=0; j<nFree; j++){
            FreelistEntry *p = &pNew->freelist.aEntry[j];
            p->iBlk = aCkpt[iIn++];
            p->iId = ((i64)(aCkpt[iIn])<<32) + aCkpt[iIn+1];
            iIn += 2;
          }
          pNew->freelist.nEntry = pNew->freelist.nAlloc = nFree;
        }
      }

  return ((i64)iReal*pFS->nPagesize <= pFS->nMapLimit);
}

/*
** Given that there are currently nHash slots in the hash table, return 
** the hash key for file iFile, page iPg.
*/
static int fsHashKey(int nHash, int iPg){
static int fsHashKey(int nHash, Pgno iPg){
  return (iPg % nHash);
}

/*
** This is a helper function for lsmFsOpen(). It opens a single file on
** disk (either the database or log file).
*/

/*
** Return the block number of the block that page iPg is located on. 
** Blocks are numbered starting from 1.
*/
static int fsPageToBlock(FileSystem *pFS, Pgno iPg){
  if( pFS->pCompress ){
    return (iPg / pFS->nBlocksize) + 1;
    return (int)((iPg / pFS->nBlocksize) + 1);
  }else{
    return 1 + ((iPg-1) / (pFS->nBlocksize / pFS->nPagesize));
    return (int)(1 + ((iPg-1) / (pFS->nBlocksize / pFS->nPagesize)));
  }
}

/*
** Return true if page iPg is the last page on its block.
**
** This function is only called in non-compressed database mode.

        pFix->aData += iOff;
      }
      lsmSortedRemap(pFS->pDb);
    }
    *pRc = rc;
  }
}

/*
** If it is mapped, unmap the database file.
*/
int lsmFsUnmap(FileSystem *pFS){
  int rc = LSM_OK;
  if( pFS ){
    rc = lsmEnvRemap(pFS->pEnv, pFS->fdDb, -1, &pFS->pMap, &pFS->nMap);
  }
  return rc;
}

/*
** fsync() the database file.
*/
int lsmFsSyncDb(FileSystem *pFS, int nBlock){
  return lsmEnvSync(pFS->pEnv, pFS->fdDb);
}

  i64 iEob;                       /* End of block */
  int nRead;
  int rc;

  assert( pFS->pCompress );

  iEob = fsLastPageOnPagesBlock(pFS, iOff) + 1;
  nRead = LSM_MIN(iEob - iOff, nData);
  nRead = (int)LSM_MIN(iEob - iOff, nData);

  rc = lsmEnvRead(pFS->pEnv, pFS->fdDb, iOff, aData, nRead);
  if( rc==LSM_OK && nRead!=nData ){
    int iBlk;

    rc = fsBlockNext(pFS, pSeg, fsPageToBlock(pFS, iOff), &iBlk);
    if( rc==LSM_OK ){

static int fsFreeBlock(
  FileSystem *pFS,                /* File system object */
  Snapshot *pSnapshot,            /* Worker snapshot */
  Segment *pIgnore,               /* Ignore this run when searching */
  int iBlk                        /* Block number of block to free */
){
  int rc = LSM_OK;                /* Return code */
  int iFirst;                     /* First page on block iBlk */
  int iLast;                      /* Last page on block iBlk */
  Pgno iFirst;                    /* First page on block iBlk */
  Pgno iLast;                     /* Last page on block iBlk */
  Level *pLevel;                  /* Used to iterate through levels */

  int iIn;                        /* Used to iterate through append points */
  int iOut = 0;                   /* Used to output append points */
  Pgno *aApp = pSnapshot->aiAppend;

  iFirst = fsFirstPageOnBlock(pFS, iBlk);

  int iBlk;

  assert( pRun->nSize>0 );
  assert( 0==fsSegmentRedirects(pFS, pRun) );
  assert( nPgno>0 && 0==fsPageRedirects(pFS, pRun, aPgno[0]) );

  iBlk = fsPageToBlock(pFS, pRun->iFirst);
  pRun->nSize += (pRun->iFirst - fsFirstPageOnBlock(pFS, iBlk));
  pRun->nSize += (int)(pRun->iFirst - fsFirstPageOnBlock(pFS, iBlk));

  while( rc==LSM_OK ){
    int iNext = 0;
    Pgno iFirst = firstOnBlock(pFS, iBlk, aPgno, nPgno);
    if( iFirst ){
      pRun->iFirst = iFirst;
      break;
    }
    rc = fsBlockNext(pFS, pRun, iBlk, &iNext);
    if( rc==LSM_OK ) rc = fsFreeBlock(pFS, pSnapshot, pRun, iBlk);
    pRun->nSize -= (
    pRun->nSize -= (int)(
        1 + fsLastPageOnBlock(pFS, iBlk) - fsFirstPageOnBlock(pFS, iBlk)
    );
    iBlk = iNext;
  }

  pRun->nSize -= (pRun->iFirst - fsFirstPageOnBlock(pFS, iBlk));
  pRun->nSize -= (int)(pRun->iFirst - fsFirstPageOnBlock(pFS, iBlk));
  assert( pRun->nSize>0 );
}

/*
** This function is only used in compressed database mode.
**
** Argument iPg is the page number (byte offset) of a page within segment

  Snapshot *pSnapshot,
  Level *pLvl,
  int bDefer,
  Page **ppOut
){
  int rc = LSM_OK;
  Page *pPg = 0;
  int iApp = 0;
  int iNext = 0;
  Pgno iApp = 0;
  Pgno iNext = 0;
  Segment *p = &pLvl->lhs;
  int iPrev = p->iLastPg;
  Pgno iPrev = p->iLastPg;

  *ppOut = 0;
  assert( p->pRedirect==0 );

  if( pFS->pCompress || bDefer ){
    /* In compressed database mode the page is not assigned a page number
    ** or location in the database file at this point. This will be done

      pPg->nRef = 1;
      pFS->nOut++;
    }
  }else{
    if( iPrev==0 ){
      iApp = findAppendPoint(pFS, pLvl);
    }else if( fsIsLast(pFS, iPrev) ){
      int iNext;
      rc = fsBlockNext(pFS, 0, fsPageToBlock(pFS, iPrev), &iNext);
      int iNext2;
      rc = fsBlockNext(pFS, 0, fsPageToBlock(pFS, iPrev), &iNext2);
      if( rc!=LSM_OK ) return rc;
      iApp = fsFirstPageOnBlock(pFS, iNext);
      iApp = fsFirstPageOnBlock(pFS, iNext2);
    }else{
      iApp = iPrev + 1;
    }

    /* If this is the first page allocated, or if the page allocated is the
    ** last in the block, also allocate the next block here.  */
    if( iApp==0 || fsIsLast(pFS, iApp) ){

  int nData,                      /* Size of buffer aData[] in bytes */
  int *pRc                        /* IN/OUT: Error code */
){
  Pgno iRet = 0;
  int rc = *pRc;
  assert( pFS->pCompress );
  if( rc==LSM_OK ){
    int nRem;
    int nWrite;
    int nRem = 0;
    int nWrite = 0;
    Pgno iLastOnBlock;
    Pgno iApp = pSeg->iLastPg+1;

    /* If this is the first data written into the segment, find an append-point
    ** or allocate a new block.  */
    if( iApp==1 ){
      pSeg->iFirst = iApp = findAppendPoint(pFS, 0);
      if( iApp==0 ){
        int iBlk;
        rc = lsmBlockAllocate(pFS->pDb, 0, &iBlk);
        pSeg->iFirst = iApp = fsFirstPageOnBlock(pFS, iBlk);
      }
    }
    iRet = iApp;

    /* Write as much data as is possible at iApp (usually all of it). */
    iLastOnBlock = fsLastPageOnPagesBlock(pFS, iApp);
    if( rc==LSM_OK ){
      int nSpace = iLastOnBlock - iApp + 1;
      int nSpace = (int)(iLastOnBlock - iApp + 1);
      nWrite = LSM_MIN(nData, nSpace);
      nRem = nData - nWrite;
      assert( nWrite>=0 );
      if( nWrite!=0 ){
        rc = lsmEnvWrite(pFS->pEnv, pFS->fdDb, iApp, aData, nWrite);
      }
      iApp += nWrite;

    Pgno iLast2;
    Pgno iLast = pSeg->iLastPg;     /* Current last page of segment */
    int nPad;                       /* Bytes of padding required */
    u8 aSz[3];

    iLast2 = (1 + iLast/pFS->szSector) * pFS->szSector - 1;
    assert( fsPageToBlock(pFS, iLast)==fsPageToBlock(pFS, iLast2) );
    nPad = iLast2 - iLast;
    nPad = (int)(iLast2 - iLast);

    if( iLast2>fsLastPageOnPagesBlock(pFS, iLast) ){
      nPad -= 4;
    }
    assert( nPad>=0 );

    if( nPad>=6 ){

    ** builds, this probably means the user is running an OOM injection test.
    ** Regardless, it will not be possible to run the integrity-check at this
    ** time, so assume the database is Ok and return non-zero. */
    return 1;
  }

  for(pLevel=pWorker->pLevel; pLevel; pLevel=pLevel->pNext){
    int i;
    int j;
    checkBlocks(pFS, &pLevel->lhs, (pLevel->nRight!=0), nBlock, aUsed);
    for(i=0; i<pLevel->nRight; i++){
      checkBlocks(pFS, &pLevel->aRhs[i], 0, nBlock, aUsed);
    for(j=0; j<pLevel->nRight; j++){
      checkBlocks(pFS, &pLevel->aRhs[j], 0, nBlock, aUsed);
    }
  }

  /* Mark all blocks in the free-list as used */
  ctx.aUsed = aUsed;
  ctx.nBlock = nBlock;
  rc = lsmWalkFreelist(pDb, 0, checkFreelistCb, (void *)&ctx);

    if( nPad ){
      u8 aPad[7] = {0,0,0,0,0,0,0};
      nPad = 8-nPad;
      if( nPad==1 ){
        aPad[0] = LSM_LOG_PAD1;
      }else{
        aPad[0] = LSM_LOG_PAD2;
        aPad[1] = (nPad-2);
        aPad[1] = (u8)(nPad-2);
      }
      rc = lsmStringBinAppend(&pLog->buf, aPad, nPad);
      if( rc!=LSM_OK ) return rc;
    }

    /* Append the JUMP record to the buffer. Then flush the buffer to disk
    ** and update the checksums. The next write to the log file (assuming

    while( nPad ){
      if( nPad==1 ){
        pLog->buf.z[pLog->buf.n++] = LSM_LOG_PAD1;
        nPad = 0;
      }else{
        int n = LSM_MIN(200, nPad-2);
        pLog->buf.z[pLog->buf.n++] = LSM_LOG_PAD2;
        pLog->buf.z[pLog->buf.n++] = n;
        pLog->buf.z[pLog->buf.n++] = (char)n;
        nPad -= 2;
        memset(&pLog->buf.z[pLog->buf.n], 0x2B, n);
        pLog->buf.n += n;
        nPad -= n;
      }
    }
  }

  /* Make sure there is room in the log-buffer to add the CKSUM or COMMIT
  ** record. Then add the first byte of it.  */
  rc = lsmStringExtend(&pLog->buf, 9);
  if( rc!=LSM_OK ) return rc;
  pLog->buf.z[pLog->buf.n++] = eType;
  pLog->buf.z[pLog->buf.n++] = (char)eType;
  memset(&pLog->buf.z[pLog->buf.n], 0, 8);

  rc = logCksumAndFlush(pDb);

  /* If this is a commit and synchronous=full, sync the log to disk. */
  if( rc==LSM_OK && eType==LSM_LOG_COMMIT && pDb->eSafety==LSM_SAFETY_FULL ){
    rc = lsmFsSyncLog(pDb->pFS);

  LogWriter *pLog;

  if( pDb->bUseLog==0 ) return;
  pLog = pDb->pLogWriter;

  assert( pMark->iOff<=pLog->iOff+pLog->buf.n );
  if( (pMark->iOff & 0xFFFFFFF8)>=pLog->iOff ){
    pLog->buf.n = pMark->iOff - pLog->iOff;
    pLog->buf.n = (int)(pMark->iOff - pLog->iOff);
    pLog->iCksumBuf = (pLog->buf.n & 0xFFFFFFF8);
  }else{
    pLog->buf.n = pMark->nBuf;
    memcpy(pLog->buf.z, pMark->aBuf, pMark->nBuf);
    pLog->iCksumBuf = 0;
    pLog->iOff = pMark->iOff - pMark->nBuf;
  }

#else
# define assertNotInFreelist(x,y)
#endif

/*
** Append an entry to the free-list. If (iId==-1), this is a delete.
*/
int freelistAppend(lsm_db *db, int iBlk, i64 iId){
int freelistAppend(lsm_db *db, u32 iBlk, i64 iId){
  lsm_env *pEnv = db->pEnv;
  Freelist *p;
  int i; 

  assert( iId==-1 || iId>=0 );
  p = db->bUseFreelist ? db->pFreelist : &db->pWorker->freelist;

    ** contains data. */
    ctx.nBlock = pDb->pWorker->nBlock;
    ctx.iInUse = -1;
    rc = lsmWalkFreelist(pDb, 1, dbTruncateCb, (void *)&ctx);

    /* If the last block that contains data is not already the last block in
    ** the database file, truncate the database file so that it is. */
    if( rc==LSM_OK && ctx.nBlock!=pDb->pWorker->nBlock ){
    if( rc==LSM_OK ){
      rc = lsmFsTruncateDb(
          pDb->pFS, (i64)ctx.nBlock*lsmFsBlockSize(pDb->pFS)
      );
    }
  }

  lsmFreeSnapshot(pDb->pEnv, pDb->pWorker);

  if( pDb->bReadonly ){
    lsmShmLock(pDb, LSM_LOCK_DMS3, LSM_LOCK_UNLOCK, 0);
  }else{
    /* Block for an exclusive lock on DMS1. This lock serializes all calls
    ** to doDbConnect() and doDbDisconnect() across all processes.  */
    rc = lsmShmLock(pDb, LSM_LOCK_DMS1, LSM_LOCK_EXCL, 1);
    if( rc==LSM_OK ){

      lsmShmLock(pDb, LSM_LOCK_DMS2, LSM_LOCK_UNLOCK, 0);

      /* Try an exclusive lock on DMS2. If successful, this is the last
      ** connection to the database. In this case flush the contents of the
      ** in-memory tree to disk and write a checkpoint.  */
      rc = lsmShmTestLock(pDb, LSM_LOCK_DMS2, 1, LSM_LOCK_EXCL);
      if( rc==LSM_OK ){
        rc = lsmShmTestLock(pDb, LSM_LOCK_CHECKPOINTER, 1, LSM_LOCK_EXCL);

            bReadonly = 1;
            rc = LSM_OK;
          }
        }

        /* Write a checkpoint to disk. */
        if( rc==LSM_OK ){
          rc = lsmCheckpointWrite(pDb, (bReadonly==0), 0);
          rc = lsmCheckpointWrite(pDb, 0);
        }

        /* If the checkpoint was written successfully, delete the log file
        ** and, if possible, truncate the database file.  */
        if( rc==LSM_OK ){
          int bRotrans = 0;
          Database *p = pDb->pDatabase;

          /* The log file may only be deleted if there are no clients 
          ** read-only clients running rotrans transactions.  */
          rc = lsmDetectRoTrans(pDb, &bRotrans);
          if( rc==LSM_OK && bRotrans==0 ){
            lsmFsCloseAndDeleteLog(pDb->pFS);
          }

          /* The database may only be truncated if there exist no read-only
          ** clients - either connected or running rotrans transactions. */
          if( bReadonly==0 && bRotrans==0 ){
            lsmFsUnmap(pDb->pFS);
            dbTruncateFile(pDb);
            if( p->pFile && p->bMultiProc ){
              lsmEnvShmUnmap(pDb->pEnv, p->pFile, 1);
            }
          }
        }
      }
    }

    if( pDb->iRwclient>=0 ){
      lsmShmLock(pDb, LSM_LOCK_RWCLIENT(pDb->iRwclient), LSM_LOCK_UNLOCK, 0);
      pDb->iRwclient = -1;
    }

    lsmShmLock(pDb, LSM_LOCK_DMS2, LSM_LOCK_UNLOCK, 0);
    lsmShmLock(pDb, LSM_LOCK_DMS1, LSM_LOCK_UNLOCK, 0);
  }
  pDb->pShmhdr = 0;
}

static int doDbConnect(lsm_db *pDb){
  const int nUsMax = 100000;      /* Max value for nUs */

      ** attempt to take the exclusive lock on DMS2.  */
      if( rc==LSM_OK ){
        int bReadonly = (pDb->bReadonly && pDb->bMultiProc);
        rc = dbOpenSharedFd(pDb->pEnv, p, bReadonly);
      }

      if( rc==LSM_OK && p->bMultiProc==0 ){
        /* Hold an exclusive lock DMS1 while grabbing DMS2. This ensures
        ** that any ongoing call to doDbDisconnect() (even one in another
        ** process) is finished before proceeding.  */
        assert( p->bReadonly==0 );
        rc = lsmEnvLock(pDb->pEnv, p->pFile, LSM_LOCK_DMS1, LSM_LOCK_EXCL);
        if( rc==LSM_OK ){
        rc = lsmEnvLock(pDb->pEnv, p->pFile, LSM_LOCK_DMS2, LSM_LOCK_EXCL);
          rc = lsmEnvLock(pDb->pEnv, p->pFile, LSM_LOCK_DMS2, LSM_LOCK_EXCL);
          lsmEnvLock(pDb->pEnv, p->pFile, LSM_LOCK_DMS1, LSM_LOCK_UNLOCK);
        }
      }

      if( rc==LSM_OK ){
        p->pDbNext = gShared.pDatabase;
        gShared.pDatabase = p;
      }else{
        freeDatabase(pEnv, p);

  if( p ){
    lsm_db **ppDb;

    if( pDb->pShmhdr ){
      doDbDisconnect(pDb);
    }

    lsmFsUnmap(pDb->pFS);
    lsmMutexEnter(pDb->pEnv, p->pClientMutex);
    for(ppDb=&p->pConn; *ppDb!=pDb; ppDb=&((*ppDb)->pNext));
    *ppDb = pDb->pNext;
    dbDeferClose(pDb);
    lsmMutexLeave(pDb->pEnv, p->pClientMutex);

    enterGlobalMutex(pDb->pEnv);

*/
static int walkFreelistCb(void *pCtx, int iBlk, i64 iSnapshot){
  WalkFreelistCtx *p = (WalkFreelistCtx *)pCtx;
  const int iDir = (p->bReverse ? -1 : 1);
  Freelist *pFree = p->pFreelist;

  assert( p->bDone==0 );
  assert( iBlk>=0 );
  if( pFree ){
    while( (p->iFree < pFree->nEntry) && p->iFree>=0 ){
      FreelistEntry *pEntry = &pFree->aEntry[p->iFree];
      if( (p->bReverse==0 && pEntry->iBlk>iBlk)
       || (p->bReverse!=0 && pEntry->iBlk<iBlk)
      if( (p->bReverse==0 && pEntry->iBlk>(u32)iBlk)
       || (p->bReverse!=0 && pEntry->iBlk<(u32)iBlk)
      ){
        break;
      }else{
        p->iFree += iDir;
        if( pEntry->iId>=0 
            && p->xUsr(p->pUsrctx, pEntry->iBlk, pEntry->iId) 
          ){
          p->bDone = 1;
          return 1;
        }
        if( pEntry->iBlk==iBlk ) return 0;
        if( pEntry->iBlk==(u32)iBlk ) return 0;
      }
    }
  }

  if( p->xUsr(p->pUsrctx, iBlk, iSnapshot) ){
    p->bDone = 1;
    return 1;

** database itself.
**
** The WORKER lock must not be held when this is called. This is because
** this function may indirectly call fsync(). And the WORKER lock should
** not be held that long (in case it is required by a client flushing an
** in-memory tree to disk).
*/
int lsmCheckpointWrite(lsm_db *pDb, int bTruncate, u32 *pnWrite){
int lsmCheckpointWrite(lsm_db *pDb, u32 *pnWrite){
  int rc;                         /* Return Code */
  u32 nWrite = 0;

  assert( pDb->pWorker==0 );
  assert( 1 || pDb->pClient==0 );
  assert( lsmShmAssertLock(pDb, LSM_LOCK_WORKER, LSM_LOCK_UNLOCK) );

    /* Check if this checkpoint has already been written to the database
    ** file. If so, set variable bDone to true.  */
    if( pShm->iMetaPage ){
      MetaPage *pPg;              /* Meta page */
      u8 *aData;                  /* Meta-page data buffer */
      int nData;                  /* Size of aData[] in bytes */
      i64 iCkpt;                  /* Id of checkpoint just loaded */
      i64 iDisk;                  /* Id of checkpoint already stored in db */
      i64 iDisk = 0;              /* Id of checkpoint already stored in db */
      iCkpt = lsmCheckpointId(pDb->aSnapshot, 0);
      rc = lsmFsMetaPageGet(pDb->pFS, 0, pShm->iMetaPage, &pPg);
      if( rc==LSM_OK ){
        aData = lsmFsMetaPageData(pPg, &nData);
        iDisk = lsmCheckpointId((u32 *)aData, 1);
        nWrite = lsmCheckpointNWrite((u32 *)aData, 1);
        lsmFsMetaPageRelease(pPg);

      }
#ifdef LSM_LOG_WORK
      lsmLogMessage(pDb, 0, "finish checkpoint %d", 
          (int)lsmCheckpointId(pDb->aSnapshot, 0)
      );
#endif
    }

    if( rc==LSM_OK && bTruncate && nBlock>0 ){
      rc = lsmFsTruncateDb(pDb->pFS, (i64)nBlock*lsmFsBlockSize(pDb->pFS));
    }
  }

  lsmShmLock(pDb, LSM_LOCK_CHECKPOINTER, LSM_LOCK_UNLOCK, 0);
  if( pnWrite && rc==LSM_OK ) *pnWrite = nWrite;
  return rc;
}

  for(i=iLock; i<(iLock+nLock); i++){
    mask |= ((u64)1 << (iLock-1));
    if( eOp==LSM_LOCK_EXCL ) mask |= ((u64)1 << (iLock+32-1));
  }

  lsmMutexEnter(db->pEnv, p->pClientMutex);
  for(pIter=p->pConn; pIter; pIter=pIter->pNext){
    if( pIter!=db && (pIter->mLock & mask) ) break;
    if( pIter!=db && (pIter->mLock & mask) ){
      assert( pIter!=db );
      break;
    }
  }

  if( pIter ){
    rc = LSM_BUSY;
  }else if( p->bMultiProc ){
    rc = lsmEnvTestLock(db->pEnv, p->pFile, iLock, nLock, eOp);
  }

** follows:
**
**   (eOp==LSM_LOCK_UNLOCK) -> true if db has no lock on iLock
**   (eOp==LSM_LOCK_SHARED) -> true if db has at least a SHARED lock on iLock.
**   (eOp==LSM_LOCK_EXCL)   -> true if db has an EXCLUSIVE lock on iLock.
*/
int lsmShmAssertLock(lsm_db *db, int iLock, int eOp){
  int ret;
  int ret = 0;
  int eHave;

  assert( iLock>=1 && iLock<=LSM_LOCK_READER(LSM_LOCK_NREADER-1) );
  assert( iLock<=16 );
  assert( eOp==LSM_LOCK_UNLOCK || eOp==LSM_LOCK_SHARED || eOp==LSM_LOCK_EXCL );

  eHave = shmLockType(db, iLock);

int lsm_checkpoint(lsm_db *pDb, int *pnKB){
  int rc;                         /* Return code */
  u32 nWrite = 0;                 /* Number of pages checkpointed */

  /* Attempt the checkpoint. If successful, nWrite is set to the number of
  ** pages written between this and the previous checkpoint.  */
  rc = lsmCheckpointWrite(pDb, 0, &nWrite);
  rc = lsmCheckpointWrite(pDb, &nWrite);

  /* If required, calculate the output variable (KB of data checkpointed). 
  ** Set it to zero if an error occured.  */
  if( pnKB ){
    int nKB = 0;
    if( rc==LSM_OK && nWrite ){
      nKB = (((i64)nWrite * lsmFsPageSize(pDb->pFS)) + 1023) / 1024;
    }
    *pnKB = nKB;
  }

  return rc;
}

}

static int btreeCursorPtr(u8 *aData, int nData, int iCell){
  int nCell;

  nCell = pageGetNRec(aData, nData);
  if( iCell>=nCell ){
    return pageGetPtr(aData, nData);
    return (int)pageGetPtr(aData, nData);
  }
  return pageGetRecordPtr(aData, nData, iCell);
  return (int)pageGetRecordPtr(aData, nData, iCell);
}

static int btreeCursorNext(BtreeCursor *pCsr){
  int rc = LSM_OK;

  BtreePg *pPg = &pCsr->aPg[pCsr->iPg];
  int nCell; 

}

static int btreeCursorFirst(BtreeCursor *pCsr){
  int rc;

  Page *pPg = 0;
  FileSystem *pFS = pCsr->pFS;
  int iPg = pCsr->pSeg->iRoot;
  int iPg = (int)pCsr->pSeg->iRoot;

  do {
    rc = lsmFsDbPageGet(pFS, pCsr->pSeg, iPg, &pPg);
    assert( (rc==LSM_OK)==(pPg!=0) );
    if( rc==LSM_OK ){
      u8 *aData;
      int nData;

        }
      }

      if( rc==LSM_OK ){
        assert( pCsr->aPg[pCsr->nDepth].iCell==0 );
        pCsr->aPg[pCsr->nDepth].pPage = pPg;
        pCsr->nDepth++;
        iPg = pageGetRecordPtr(aData, nData, 0);
        iPg = (int)pageGetRecordPtr(aData, nData, 0);
      }
    }
  }while( rc==LSM_OK );
  lsmFsPageRelease(pPg);
  pCsr->iPg = pCsr->nDepth-1;

  if( rc==LSM_OK && pCsr->nDepth ){

    /* Populate any other aPg[] array entries */
    if( rc==LSM_OK && nDepth>1 ){
      Blob blob = {0,0,0};
      void *pSeek;
      int nSeek;
      int iTopicSeek;
      int iPg = 0;
      int iLoad = pSeg->iRoot;
      int iLoad = (int)pSeg->iRoot;
      Page *pPg = pCsr->aPg[nDepth-1].pPage;
 
      if( pageObjGetNRec(pPg)==0 ){
        /* This can happen when pPg is the right-most leaf in the b-tree.
        ** In this case, set the iTopicSeek/pSeek/nSeek key to a value
        ** greater than any real key.  */
        assert( iCell==-1 );
        iTopicSeek = 1000;
        pSeek = 0;
        nSeek = 0;
      }else{
        Pgno dummy;
        rc = pageGetBtreeKey(pSeg, pPg,
            0, &dummy, &iTopicSeek, &pSeek, &nSeek, &pCsr->blob
        );
      }

      do {
        Page *pPg;
        rc = lsmFsDbPageGet(pCsr->pFS, pSeg, iLoad, &pPg);
        assert( rc==LSM_OK || pPg==0 );
        Page *pPg2;
        rc = lsmFsDbPageGet(pCsr->pFS, pSeg, iLoad, &pPg2);
        assert( rc==LSM_OK || pPg2==0 );
        if( rc==LSM_OK ){
          u8 *aData;                  /* Buffer containing page data */
          int nData;                  /* Size of aData[] in bytes */
          int iMin;
          int iMax;
          int iCell;
          int iCell2;

          aData = fsPageData(pPg, &nData);
          aData = fsPageData(pPg2, &nData);
          assert( (pageGetFlags(aData, nData) & SEGMENT_BTREE_FLAG) );

          iLoad = pageGetPtr(aData, nData);
          iCell = pageGetNRec(aData, nData); 
          iMax = iCell-1;
          iLoad = (int)pageGetPtr(aData, nData);
          iCell2 = pageGetNRec(aData, nData); 
          iMax = iCell2-1;
          iMin = 0;

          while( iMax>=iMin ){
            int iTry = (iMin+iMax)/2;
            void *pKey; int nKey;         /* Key for cell iTry */
            int iTopic;                   /* Topic for key pKeyT/nKeyT */
            Pgno iPtr;                    /* Pointer for cell iTry */
            int res;                      /* (pSeek - pKeyT) */

            rc = pageGetBtreeKey(
                pSeg, pPg, iTry, &iPtr, &iTopic, &pKey, &nKey, &blob
                pSeg, pPg2, iTry, &iPtr, &iTopic, &pKey, &nKey, &blob
            );
            if( rc!=LSM_OK ) break;

            res = sortedKeyCompare(
                xCmp, iTopicSeek, pSeek, nSeek, iTopic, pKey, nKey
            );
            assert( res!=0 );

            if( res<0 ){
              iLoad = iPtr;
              iCell = iTry;
              iLoad = (int)iPtr;
              iCell2 = iTry;
              iMax = iTry-1;
            }else{
              iMin = iTry+1;
            }
          }

          pCsr->aPg[iPg].pPage = pPg;
          pCsr->aPg[iPg].iCell = iCell;
          pCsr->aPg[iPg].pPage = pPg2;
          pCsr->aPg[iPg].iCell = iCell2;
          iPg++;
          assert( iPg!=nDepth-1 
               || lsmFsRedirectPage(pCsr->pFS, pSeg->pRedirect, iLoad)==iLeaf
          );
        }
      }while( rc==LSM_OK && iPg<(nDepth-1) );
      sortedBlobFree(&blob);

static void segmentPtrSetPage(SegmentPtr *pPtr, Page *pNext){
  lsmFsPageRelease(pPtr->pPg);
  if( pNext ){
    int nData;
    u8 *aData = fsPageData(pNext, &nData);
    pPtr->nCell = pageGetNRec(aData, nData);
    pPtr->flags = pageGetFlags(aData, nData);
    pPtr->flags = (u16)pageGetFlags(aData, nData);
    pPtr->iPtr = pageGetPtr(aData, nData);
  }
  pPtr->pPg = pNext;
}

/*
** Load a new page into the SegmentPtr object pPtr.

  int iTopic,                     /* Key topic to seek to */
  void *pKey, int nKey,           /* Key to seek to */
  int eSeek,                      /* Search bias - see above */
  int *piPtr,                     /* OUT: FC pointer */
  int *pbStop
){
  int (*xCmp)(void *, int, void *, int) = pCsr->pDb->xCmp;
  int res;                        /* Result of comparison operation */
  int res = 0;                        /* Result of comparison operation */
  int rc = LSM_OK;
  int iMin;
  int iMax;
  Pgno iPtrOut = 0;

  /* If the current page contains an oversized entry, then there are no
  ** pointers to one or more of the subsequent pages in the sorted run.

    ){
      assert( eSeek!=LSM_SEEK_EQ );
      rc = segmentPtrAdvance(pCsr, pPtr, eSeek==LSM_SEEK_LE);
    }
  }

  assert( rc!=LSM_OK || assertSeekResult(pCsr,pPtr,iTopic,pKey,nKey,eSeek) );
  *piPtr = iPtrOut;
  *piPtr = (int)iPtrOut;
  return rc;
}

static int seekInBtree(
  MultiCursor *pCsr,              /* Multi-cursor object */
  Segment *pSeg,                  /* Seek within this segment */
  int iTopic,
  void *pKey, int nKey,           /* Key to seek to */
  Pgno *aPg,                      /* OUT: Page numbers */
  Page **ppPg                     /* OUT: Leaf (sorted-run) page reference */
){
  int i = 0;
  int rc;
  int iPg;
  Page *pPg = 0;
  Blob blob = {0, 0, 0};

  iPg = pSeg->iRoot;
  iPg = (int)pSeg->iRoot;
  do {
    Pgno *piFirst = 0;
    if( aPg ){
      aPg[i++] = iPg;
      piFirst = &aPg[i];
    }

      int nRec;
      int flags;

      aData = fsPageData(pPg, &nData);
      flags = pageGetFlags(aData, nData);
      if( (flags & SEGMENT_BTREE_FLAG)==0 ) break;

      iPg = pageGetPtr(aData, nData);
      iPg = (int)pageGetPtr(aData, nData);
      nRec = pageGetNRec(aData, nData);

      iMin = 0;
      iMax = nRec-1;
      while( iMax>=iMin ){
        int iTry = (iMin+iMax)/2;
        void *pKeyT; int nKeyT;       /* Key for cell iTry */

          i++;
        }

        res = sortedKeyCompare(
            pCsr->pDb->xCmp, iTopic, pKey, nKey, iTopicT, pKeyT, nKeyT
        );
        if( res<0 ){
          iPg = iPtr;
          iPg = (int)iPtr;
          iMax = iTry-1;
        }else{
          iMin = iTry+1;
        }
      }
      lsmFsPageRelease(pPg);
      pPg = 0;

  if( pPtr->pSeg->iRoot ){
    Page *pPg;
    assert( pPtr->pSeg->iRoot!=0 );
    rc = seekInBtree(pCsr, pPtr->pSeg, iTopic, pKey, nKey, 0, &pPg);
    if( rc==LSM_OK ) segmentPtrSetPage(pPtr, pPg);
  }else{
    if( iPtr==0 ){
      iPtr = pPtr->pSeg->iFirst;
      iPtr = (int)pPtr->pSeg->iFirst;
    }
    if( rc==LSM_OK ){
      rc = segmentPtrLoadPage(pCsr->pDb->pFS, pPtr, iPtr);
    }
  }

  if( rc==LSM_OK ){

  }

  /* If (res<0), then key pKey/nKey is smaller than the split-key (or this
  ** is not a composite level and there is no split-key). Search the 
  ** left-hand-side of the level in this case.  */
  if( res<0 ){
    int iPtr = 0;
    if( nRhs==0 ) iPtr = *piPgno;
    if( nRhs==0 ) iPtr = (int)*piPgno;

    rc = seekInSegment(
        pCsr, &aPtr[0], iTopic, pKey, nKey, iPtr, eSeek, &iOut, &bStop
    );
    if( rc==LSM_OK && nRhs>0 && eSeek==LSM_SEEK_GE && aPtr[0].pPg==0 ){
      res = 0;
    }
  }
  
  if( res>=0 ){
    int bHit = 0;                 /* True if at least one rhs is not EOF */
    int iPtr = *piPgno;
    int iPtr = (int)*piPgno;
    int i;
    for(i=1; rc==LSM_OK && i<=nRhs && bStop==0; i++){
      SegmentPtr *pPtr = &aPtr[i];
      iOut = 0;
      rc = seekInSegment(
          pCsr, pPtr, iTopic, pKey, nKey, iPtr, eSeek, &iOut, &bStop
      );

    case CURSOR_DATA_SYSTEM: {
      Snapshot *pWorker = pCsr->pDb->pWorker;
      if( pWorker && (pCsr->flags & CURSOR_FLUSH_FREELIST) ){
        int nEntry = pWorker->freelist.nEntry;
        if( pCsr->iFree < (nEntry*2) ){
          FreelistEntry *aEntry = pWorker->freelist.aEntry;
          int i = nEntry - 1 - (pCsr->iFree / 2);
          u32 iKey = 0;
          u32 iKey2 = 0;

          if( (pCsr->iFree % 2) ){
            eType = LSM_END_DELETE|LSM_SYSTEMKEY;
            iKey = aEntry[i].iBlk-1;
            iKey2 = aEntry[i].iBlk-1;
          }else if( aEntry[i].iId>=0 ){
            eType = LSM_INSERT|LSM_SYSTEMKEY;
            iKey = aEntry[i].iBlk;
            iKey2 = aEntry[i].iBlk;

            /* If the in-memory entry immediately before this one was a
             ** DELETE, and the block number is one greater than the current
             ** block number, mark this entry as an "end-delete-range". */
            if( i<(nEntry-1) && aEntry[i+1].iBlk==iKey+1 && aEntry[i+1].iId<0 ){
            if( i<(nEntry-1) && aEntry[i+1].iBlk==iKey2+1 && aEntry[i+1].iId<0 ){
              eType |= LSM_END_DELETE;
            }

          }else{
            eType = LSM_START_DELETE|LSM_SYSTEMKEY;
            iKey = aEntry[i].iBlk + 1;
            iKey2 = aEntry[i].iBlk + 1;
          }

          /* If the in-memory entry immediately after this one is a
          ** DELETE, and the block number is one less than the current
          ** key, mark this entry as an "start-delete-range".  */
          if( i>0 && aEntry[i-1].iBlk==iKey-1 && aEntry[i-1].iId<0 ){
          if( i>0 && aEntry[i-1].iBlk==iKey2-1 && aEntry[i-1].iId<0 ){
            eType |= LSM_START_DELETE;
          }

          pKey = pCsr->pSystemVal;
          nKey = 4;
          lsmPutU32(pKey, ~iKey);
          lsmPutU32(pKey, ~iKey2);
        }
      }
      break;
    }

    default: {
      int iPtr = iKey - CURSOR_DATA_SEGMENT;

      rc = lsmMCursorLast(pCsr);
    }else{
      rc = lsmMCursorSeek(pCsr, 1, "", 0, LSM_SEEK_GE);
    }

    while( rc==LSM_OK && lsmMCursorValid(pCsr) && rtIsSystem(pCsr->eType) ){
      void *pKey; int nKey;
      void *pVal; int nVal;
      void *pVal = 0; int nVal = 0;

      rc = lsmMCursorKey(pCsr, &pKey, &nKey);
      if( rc==LSM_OK ) rc = lsmMCursorValue(pCsr, &pVal, &nVal);
      if( rc==LSM_OK && (nKey!=4 || nVal!=8) ) rc = LSM_CORRUPT_BKPT;

      if( rc==LSM_OK ){
        int iBlk;

  p = &pMW->hier;

  if( p->apHier==0 && pSeg->iRoot!=0 ){
    FileSystem *pFS = pMW->pDb->pFS;
    lsm_env *pEnv = pMW->pDb->pEnv;
    Page **apHier = 0;
    int nHier = 0;
    int iPg = pSeg->iRoot;
    int iPg = (int)pSeg->iRoot;

    do {
      Page *pPg = 0;
      u8 *aData;
      int nData;
      int flags;

          break;
        }
        apHier = apNew;
        memmove(&apHier[1], &apHier[0], sizeof(Page *) * nHier);
        nHier++;

        apHier[0] = pPg;
        iPg = pageGetPtr(aData, nData);
        iPg = (int)pageGetPtr(aData, nData);
      }else{
        lsmFsPageRelease(pPg);
        break;
      }
    }while( 1 );

    if( rc==LSM_OK ){

      /* If the key will fit on this page, break out of the loop here.
      ** The new entry will be written to page apHier[iLevel]. */
      pOld = p->apHier[iLevel];
      assert( lsmFsPageWritable(pOld) );
      aData = fsPageData(pOld, &nData);
      if( eType==0 ){
        nByte = 2 + 1 + lsmVarintLen32(iPtr) + lsmVarintLen32(iKeyPg);
        nByte = 2 + 1 + lsmVarintLen32((int)iPtr) + lsmVarintLen32((int)iKeyPg);
      }else{
        nByte = 2 + 1 + lsmVarintLen32(iPtr) + lsmVarintLen32(nKey) + nKey;
        nByte = 2 + 1 + lsmVarintLen32((int)iPtr) + lsmVarintLen32(nKey) + nKey;
      }
      nRec = pageGetNRec(aData, nData);
      nFree = SEGMENT_EOF(nData, nRec) - mergeWorkerPageOffset(aData, nData);
      if( nByte<=nFree ) break;

      /* Otherwise, this page is full. Set the right-hand-child pointer
      ** to iPtr and release it.  */

    }
  }

  /* Write the key into page apHier[iLevel]. */
  aData = fsPageData(p->apHier[iLevel], &nData);
  iOff = mergeWorkerPageOffset(aData, nData);
  nRec = pageGetNRec(aData, nData);
  lsmPutU16(&aData[SEGMENT_CELLPTR_OFFSET(nData, nRec)], iOff);
  lsmPutU16(&aData[SEGMENT_NRECORD_OFFSET(nData)], nRec+1);
  lsmPutU16(&aData[SEGMENT_CELLPTR_OFFSET(nData, nRec)], (u16)iOff);
  lsmPutU16(&aData[SEGMENT_NRECORD_OFFSET(nData)], (u16)(nRec+1));
  if( eType==0 ){
    aData[iOff++] = 0x00;
    iOff += lsmVarintPut32(&aData[iOff], iPtr);
    iOff += lsmVarintPut32(&aData[iOff], iKeyPg);
    iOff += lsmVarintPut32(&aData[iOff], (int)iPtr);
    iOff += lsmVarintPut32(&aData[iOff], (int)iKeyPg);
  }else{
    aData[iOff++] = eType;
    iOff += lsmVarintPut32(&aData[iOff], iPtr);
    iOff += lsmVarintPut32(&aData[iOff], (int)iPtr);
    iOff += lsmVarintPut32(&aData[iOff], nKey);
    memcpy(&aData[iOff], pKey, nKey);
  }

  return rc;
}

  int iFPtr = 0;                  /* Pointer value read from footer of pPg */
  MultiCursor *pCsr = pMW->pCsr;

  assert( pMW->pPage==0 );

  if( pCsr->pBtCsr ){
    rc = LSM_OK;
    iFPtr = pMW->pLevel->pNext->lhs.iFirst;
    iFPtr = (int)pMW->pLevel->pNext->lhs.iFirst;
  }else if( pCsr->nPtr>0 ){
    Segment *pSeg;
    pSeg = pCsr->aPtr[pCsr->nPtr-1].pSeg;
    rc = lsmFsDbPageGet(pMW->pDb->pFS, pSeg, pSeg->iFirst, &pPg);
    if( rc==LSM_OK ){
      u8 *aData;                    /* Buffer for page pPg */
      int nData;                    /* Size of aData[] in bytes */
      aData = fsPageData(pPg, &nData);
      iFPtr = pageGetPtr(aData, nData);
      iFPtr = (int)pageGetPtr(aData, nData);
      lsmFsPageRelease(pPg);
    }
  }

  if( rc==LSM_OK ){
    rc = mergeWorkerNextPage(pMW, iFPtr);
    if( pCsr->pPrevMergePtr ) *pCsr->pPrevMergePtr = iFPtr;

  int iPtr                        /* Absolute value of page pointer, or 0 */
){
  int rc = LSM_OK;                /* Return code */
  Merge *pMerge;                  /* Persistent part of level merge state */
  int nHdr;                       /* Space required for this record header */
  Page *pPg;                      /* Page to write to */
  u8 *aData;                      /* Data buffer for page pWriter->pPage */
  int nData;                      /* Size of buffer aData[] in bytes */
  int nRec;                       /* Number of records on page pPg */
  int iFPtr;                      /* Value of pointer in footer of pPg */
  int nData = 0;                  /* Size of buffer aData[] in bytes */
  int nRec = 0;                   /* Number of records on page pPg */
  int iFPtr = 0;                  /* Value of pointer in footer of pPg */
  int iRPtr = 0;                  /* Value of pointer written into record */
  int iOff;                       /* Current write offset within page pPg */
  int iOff = 0;                   /* Current write offset within page pPg */
  Segment *pSeg;                  /* Segment being written */
  int flags = 0;                  /* If != 0, flags value for page footer */
  int bFirst = 0;                 /* True for first key of output run */

  pMerge = pMW->pLevel->pMerge;    
  pSeg = &pMW->pLevel->lhs;

  if( pSeg->iFirst==0 && pMW->pPage==0 ){
    rc = mergeWorkerFirstPage(pMW);
    bFirst = 1;
  }
  pPg = pMW->pPage;
  if( pPg ){
    aData = fsPageData(pPg, &nData);
    nRec = pageGetNRec(aData, nData);
    iFPtr = pageGetPtr(aData, nData);
    iFPtr = (int)pageGetPtr(aData, nData);
    iRPtr = iPtr - iFPtr;
  }
     
  /* Figure out how much space is required by the new record. The space
  ** required is divided into two sections: the header and the body. The
  ** header consists of the intial varint fields. The body are the blobs 
  ** of data that correspond to the key and value data. The entire header 

    nHdr = 1 + lsmVarintLen32(iRPtr) + lsmVarintLen32(nKey);
    if( rtIsWrite(eType) ) nHdr += lsmVarintLen32(nVal);

    /* If the entire header will not fit on page pPg, or if page pPg is 
    ** marked read-only, advance to the next page of the output run. */
    iOff = pMerge->iOutputOff;
    if( iOff<0 || pPg==0 || iOff+nHdr > SEGMENT_EOF(nData, nRec+1) ){
      iFPtr = *pMW->pCsr->pPrevMergePtr;
      iFPtr = (int)*pMW->pCsr->pPrevMergePtr;
      iRPtr = iPtr - iFPtr;
      iOff = 0;
      nRec = 0;
      rc = mergeWorkerNextPage(pMW, iFPtr);
      pPg = pMW->pPage;
    }
  }

  }

  /* Update the output segment */
  if( rc==LSM_OK ){
    aData = fsPageData(pPg, &nData);

    /* Update the page footer. */
    lsmPutU16(&aData[SEGMENT_NRECORD_OFFSET(nData)], nRec+1);
    lsmPutU16(&aData[SEGMENT_CELLPTR_OFFSET(nData, nRec)], iOff);
    if( flags ) lsmPutU16(&aData[SEGMENT_FLAGS_OFFSET(nData)], flags);
    lsmPutU16(&aData[SEGMENT_NRECORD_OFFSET(nData)], (u16)(nRec+1));
    lsmPutU16(&aData[SEGMENT_CELLPTR_OFFSET(nData, nRec)], (u16)iOff);
    if( flags ) lsmPutU16(&aData[SEGMENT_FLAGS_OFFSET(nData)], (u16)flags);

    /* Write the entry header into the current page. */
    aData[iOff++] = eType;                                               /* 1 */
    aData[iOff++] = (u8)eType;                                           /* 1 */
    iOff += lsmVarintPut32(&aData[iOff], iRPtr);                         /* 2 */
    iOff += lsmVarintPut32(&aData[iOff], nKey);                          /* 3 */
    if( rtIsWrite(eType) ) iOff += lsmVarintPut32(&aData[iOff], nVal);   /* 4 */
    pMerge->iOutputOff = iOff;

    /* Write the key and data into the segment. */
    assert( iFPtr==pageGetPtr(aData, nData) );

      rc = multiCursorGetVal(pCsr, iVal, &pVal, &nVal);
      if( pVal && rc==LSM_OK ){
        assert( nVal>=0 );
        rc = sortedBlobSet(pDb->pEnv, &pCsr->val, pVal, nVal);
        pVal = pCsr->val.pData;
      }
      if( rc==LSM_OK ){
        rc = mergeWorkerWrite(pMW, eType, pKey, nKey, pVal, nVal, iPtr);
        rc = mergeWorkerWrite(pMW, eType, pKey, nKey, pVal, nVal, (int)iPtr);
      }
    }
  }

  /* Advance the cursor to the next input record (assuming one exists). */
  assert( lsmMCursorValid(pMW->pCsr) );
  if( rc==LSM_OK ) rc = lsmMCursorNext(pMW->pCsr);

      int i;
      for(i=0; rc==LSM_OK && i<pCsr->nPtr; i++){
        MergeInput *pInput = &pMerge->aInput[i];
        if( pInput->iPg ){
          SegmentPtr *pPtr;
          assert( pCsr->aPtr[i].pPg==0 );
          pPtr = &pCsr->aPtr[i];
          rc = segmentPtrLoadPage(pDb->pFS, pPtr, pInput->iPg);
          rc = segmentPtrLoadPage(pDb->pFS, pPtr, (int)pInput->iPg);
          if( rc==LSM_OK && pPtr->nCell>0 ){
            rc = segmentPtrLoadCell(pPtr, pInput->iCell);
          }
        }
      }

      if( rc==LSM_OK && pCsr->pBtCsr ){

    u32 nUnsync;
    int nPgsz;

    lsmCheckpointSynced(pDb, 0, 0, &nSync);
    nUnsync = lsmCheckpointNWrite(pDb->pShmhdr->aSnap1, 0);
    nPgsz = lsmCheckpointPgsz(pDb->pShmhdr->aSnap1);

    nMax = LSM_MIN(nMax, (pDb->nAutockpt/nPgsz) - (int)(nUnsync-nSync));
    nMax = (int)LSM_MIN(nMax, (pDb->nAutockpt/nPgsz) - (int)(nUnsync-nSync));
    if( nMax<nRem ){
      bCkpt = 1;
      nRem = LSM_MAX(nMax, 0);
    }
  }

  /* If there exists in-memory data ready to be flushed to disk, attempt

  int flags;
  u8 *aData;
  int nData;

  aData = fsPageData(pPg, &nData);

  nRec = pageGetNRec(aData, nData);
  iPtr = pageGetPtr(aData, nData);
  iPtr = (int)pageGetPtr(aData, nData);
  flags = pageGetFlags(aData, nData);

  lsmStringInit(&s, pDb->pEnv);
  lsmStringAppendf(&s,"nCell=%d iPtr=%d flags=%d {", nRec, iPtr, flags);
  if( flags&SEGMENT_BTREE_FLAG ) iPtr = 0;

  for(i=0; i<nRec; i++){
    Page *pRef = 0;               /* Pointer to page iRef */
    int iChar;
    u8 *aKey; int nKey = 0;       /* Key */
    u8 *aVal; int nVal = 0;       /* Value */
    u8 *aVal = 0; int nVal = 0;   /* Value */
    int iTopic;
    u8 *aCell;
    int iPgPtr;
    int eType;

    aCell = pageGetCell(aData, nData, i);
    eType = *aCell++;

  int *piPgPtr,
  u8 **paKey, int *pnKey,
  u8 **paVal, int *pnVal,
  Blob *pBlob
){
  u8 *aData; int nData;           /* Page data */
  u8 *aKey; int nKey = 0;         /* Key */
  u8 *aVal; int nVal = 0;         /* Value */
  u8 *aVal = 0; int nVal = 0;     /* Value */
  int eType;
  int iPgPtr;
  Page *pRef = 0;                 /* Pointer to page iRef */
  u8 *aCell;

  aData = fsPageData(pPg, &nData);

  if( rc==LSM_OK ){
    Blob blob = {0, 0, 0, 0};
    int nKeyWidth = 0;
    LsmString str;
    int nRec;
    int iPtr;
    int flags;
    int flags2;
    int iCell;
    u8 *aData; int nData;         /* Page data and size thereof */

    aData = fsPageData(pPg, &nData);
    nRec = pageGetNRec(aData, nData);
    iPtr = pageGetPtr(aData, nData);
    flags = pageGetFlags(aData, nData);
    iPtr = (int)pageGetPtr(aData, nData);
    flags2 = pageGetFlags(aData, nData);

    lsmStringInit(&str, pDb->pEnv);
    lsmStringAppendf(&str, "Page : %lld  (%d bytes)\n", iPg, nData);
    lsmStringAppendf(&str, "nRec : %d\n", nRec);
    lsmStringAppendf(&str, "iPtr : %d\n", iPtr);
    lsmStringAppendf(&str, "flags: %04x\n", flags);
    lsmStringAppendf(&str, "flags: %04x\n", flags2);
    lsmStringAppendf(&str, "\n");

    for(iCell=0; iCell<nRec; iCell++){
      int nKey;
      infoCellDump(
          pDb, pSeg, bIndirect, pPg, iCell, 0, 0, 0, &nKey, 0, 0, &blob
      );

      int eType;
      Pgno iAbsPtr;
      char zFlags[8];

      infoCellDump(pDb, pSeg, bIndirect, pPg, iCell, &eType, &iPgPtr,
          &aKey, &nKey, &aVal, &nVal, &blob
      );
      iAbsPtr = iPgPtr + ((flags & SEGMENT_BTREE_FLAG) ? 0 : iPtr);
      iAbsPtr = iPgPtr + ((flags2 & SEGMENT_BTREE_FLAG) ? 0 : iPtr);

      lsmFlagsToString(eType, zFlags);
      lsmStringAppendf(&str, "%s %d (%s) ", 
          zFlags, iAbsPtr, (rtTopic(eType) ? "sys" : "usr")
      );
      infoAppendBlob(&str, bHex, aKey, nKey); 
      if( nVal>0 && bValues ){

/*
** The pointer passed as the first argument points to an interior node,
** not a leaf. This function returns the offset of the iCell'th child
** sub-tree of the node.
*/
static u32 getChildPtr(TreeNode *p, int iVersion, int iCell){
  assert( iVersion>=0 );
  assert( iCell>=0 && iCell<=array_size(p->aiChildPtr) );
  if( p->iV2 && p->iV2<=iVersion && iCell==p->iV2Child ) return p->iV2Ptr;
  if( p->iV2 && p->iV2<=(u32)iVersion && iCell==p->iV2Child ) return p->iV2Ptr;
  return p->aiChildPtr[iCell];
}

/*
** Given an offset within the *-shm file, return the associated chunk number.
*/
static int treeOffsetToChunk(u32 iOff){
  assert( LSM_SHM_CHUNK_SIZE==(1<<15) );
  return (int)(iOff>>15);
}

#define treeShmptrUnsafe(pDb, iPtr) \
(&((u8*)((pDb)->apShm[(iPtr)>>15]))[(iPtr) & (LSM_SHM_CHUNK_SIZE-1)])

/*
** Return a pointer to the mapped memory location associated with *-shm 
** file offset iPtr.
*/
static void *treeShmptr(lsm_db *pDb, u32 iPtr){

  assert( (iPtr>>15)<pDb->nShm );
  assert( (iPtr>>15)<(u32)pDb->nShm );
  assert( pDb->apShm[iPtr>>15] );

  return iPtr ? treeShmptrUnsafe(pDb, iPtr) : 0;
}

static ShmChunk * treeShmChunk(lsm_db *pDb, int iChunk){
  return (ShmChunk *)(pDb->apShm[iChunk]);

    printf("% 6d %.*sleaf%.*s: %s\n", 
        iNode, nPath, zPath, 20-nPath-4, zSpace, s.z
    );
    lsmStringClear(&s);
  }else{
    for(i=0; i<4 && nHeight>0; i++){
      u32 iPtr = getChildPtr(pNode, pDb->treehdr.root.iTransId, i);
      zPath[nPath] = i+'0';
      zPath[nPath] = (char)(i+'0');
      zPath[nPath+1] = '/';

      if( iPtr ){
        dump_node_contents(pDb, iPtr, zPath, nPath+2, nHeight-1);
      }
      if( i!=3 && pNode->aiKeyPtr[i] ){
        TreeKey *pKey = treeShmkey(pDb, pNode->aiKeyPtr[i], TKV_LOADKEY,&b,&rc);

      iWrite = (iWrite + 3) & ~0x0003;
      assert( (iWrite % 4)==0 );
    }

    assert( iWrite );
    iChunk = treeOffsetToChunk(iWrite-1);
    iEof = (iChunk+1) * CHUNK_SIZE;
    assert( iEof>=iWrite && (iEof-iWrite)<CHUNK_SIZE );
    assert( iEof>=iWrite && (iEof-iWrite)<(u32)CHUNK_SIZE );
    if( (iWrite+nByte)>iEof ){
      ShmChunk *pHdr;           /* Header of chunk just finished (iChunk) */
      ShmChunk *pFirst;         /* Header of chunk treehdr.iFirst */
      ShmChunk *pNext;          /* Header of new chunk */
      int iNext = 0;            /* Next chunk */
      int rc = LSM_OK;

    while( nRem>0 ){
      u8 *aAlloc;
      int nAlloc;
      u32 iWrite;

      iWrite = (pDb->treehdr.iWrite & (LSM_SHM_CHUNK_SIZE-1));
      iWrite = LSM_MAX(iWrite, LSM_SHM_CHUNK_HDR);
      nAlloc = LSM_MIN((LSM_SHM_CHUNK_SIZE-iWrite), nRem);
      nAlloc = LSM_MIN((LSM_SHM_CHUNK_SIZE-iWrite), (u32)nRem);

      aAlloc = treeShmptr(pDb, treeShmalloc(pDb, 0, nAlloc, pRc));
      if( aAlloc==0 ) break;
      memcpy(aAlloc, &a[n-nRem], nAlloc);
      nRem -= nAlloc;
    }
    a = pVal;

      }
    }else{
      p = 0;
    }
    nVisit++;
  }

  if( rc==LSM_OK && nVisit!=db->treehdr.nChunk-1 ){
  if( rc==LSM_OK && (u32)nVisit!=db->treehdr.nChunk-1 ){
    rc = LSM_CORRUPT_BKPT;
  }
  return rc;
}

/*
** Iterate through the current in-memory tree. If there are any v2-pointers

  int i;
  ShmChunk *p;
  ShmChunk *pMin = 0;
  u32 iMin = 0;

  /* Iterate through all shm chunks. Find the smallest shm-id present in
  ** the shared-memory region. */
  for(i=1; rc==LSM_OK && i<db->treehdr.nChunk; i++){
  for(i=1; rc==LSM_OK && (u32)i<db->treehdr.nChunk; i++){
    p = treeShmChunkRc(db, i, &rc);
    if( p && (pMin==0 || shm_sequence_ge(pMin->iShmid, p->iShmid)) ){
      pMin = p;
      iMin = i;
    }
  }

  /* Fix the shm-id values on any chunks with a shm-id greater than or 
  ** equal to treehdr.iNextShmid. Then do a merge-sort of all chunks to 
  ** fix the ShmChunk.iNext pointers.
  */
  if( rc==LSM_OK ){
    int nSort;
    int nByte;
    u32 iPrevShmid;
    ShmChunkLoc *aSort;

    /* Allocate space for a merge sort. */
    nSort = 1;
    while( nSort < (db->treehdr.nChunk-1) ) nSort = nSort * 2;
    while( (u32)nSort < (db->treehdr.nChunk-1) ) nSort = nSort * 2;
    nByte = sizeof(ShmChunkLoc) * nSort * 2;
    aSort = lsmMallocZeroRc(db->pEnv, nByte, &rc);
    iPrevShmid = pMin->iShmid;

    /* Fix all shm-ids, if required. */
    if( rc==LSM_OK ){
      iPrevShmid = pMin->iShmid-1;
      for(i=1; i<db->treehdr.nChunk; i++){
      for(i=1; (u32)i<db->treehdr.nChunk; i++){
        p = treeShmChunk(db, i);
        aSort[i-1].pShm = p;
        aSort[i-1].iLoc = i;
        if( i!=db->treehdr.iFirst ){
        if( (u32)i!=db->treehdr.iFirst ){
          if( shm_sequence_ge(p->iShmid, db->treehdr.iNextShmid) ){
            p->iShmid = iPrevShmid--;
          }
        }
      }
      if( iMin!=db->treehdr.iFirst ){
        p = treeShmChunk(db, db->treehdr.iFirst);

    TreeRoot *p = &db->treehdr.root;
    TreeNode *pNew;
    u32 iNew;
    TreeNode *pNode = pCsr->apTreeNode[pCsr->iNode];
    int iCell = pCsr->aiCell[pCsr->iNode];

    /* Create a copy of this node */
    if( (pCsr->iNode>0 && pCsr->iNode==(p->nHeight-1)) ){
    if( (pCsr->iNode>0 && (u32)pCsr->iNode==(p->nHeight-1)) ){
      pNew = copyTreeLeaf(db, (TreeLeaf *)pNode, &iNew, pRc);
    }else{
      pNew = copyTreeNode(db, pNode, &iNew, pRc);
    }

    if( pNew ){
      /* Modify the value in the new version */

}

static int treeNextIsEndDelete(lsm_db *db, TreeCursor *pCsr){
  int iNode = pCsr->iNode;
  int iCell = pCsr->aiCell[iNode]+1;

  /* Cursor currently points to a leaf node. */
  assert( pCsr->iNode==(db->treehdr.root.nHeight-1) );
  assert( (u32)pCsr->iNode==(db->treehdr.root.nHeight-1) );

  while( iNode>=0 ){
    TreeNode *pNode = pCsr->apTreeNode[iNode];
    if( iCell<3 && pNode->aiKeyPtr[iCell] ){
      int rc = LSM_OK;
      TreeKey *pKey = treeShmptr(db, pNode->aiKeyPtr[iCell]);
      assert( rc==LSM_OK );
      return ((pKey->flags & LSM_END_DELETE) ? 1 : 0);
    }
    iNode--;
    iCell = pCsr->aiCell[iNode];
  }

  return 0;
}

static int treePrevIsStartDelete(lsm_db *db, TreeCursor *pCsr){
  int iNode = pCsr->iNode;

  /* Cursor currently points to a leaf node. */
  assert( pCsr->iNode==(db->treehdr.root.nHeight-1) );
  assert( (u32)pCsr->iNode==(db->treehdr.root.nHeight-1) );

  while( iNode>=0 ){
    TreeNode *pNode = pCsr->apTreeNode[iNode];
    int iCell = pCsr->aiCell[iNode]-1;
    if( iCell>=0 && pNode->aiKeyPtr[iCell] ){
      int rc = LSM_OK;
      TreeKey *pKey = treeShmptr(db, pNode->aiKeyPtr[iCell]);

  int nVal                        /* Bytes in value data (or -ve for delete) */
){
  int rc = LSM_OK;                /* Return Code */
  TreeKey *pTreeKey;              /* New key-value being inserted */
  u32 iTreeKey;
  TreeRoot *p = &pDb->treehdr.root;
  TreeCursor csr;                 /* Cursor to seek to pKey/nKey */
  int res;                        /* Result of seek operation on csr */
  int res = 0;                    /* Result of seek operation on csr */

  assert( nVal>=0 || pVal==0 );
  assert_tree_looks_ok(LSM_OK, pTree);
  assert( flags==LSM_INSERT       || flags==LSM_POINT_DELETE 
       || flags==LSM_START_DELETE || flags==LSM_END_DELETE 
  );
  assert( (flags & LSM_CONTIGUOUS)==0 );

  int iSlot = pCsr->aiCell[pCsr->iNode];
  int bLeaf;
  int rc = LSM_OK;

  assert( pNode->aiKeyPtr[1] );
  assert( pNode->aiKeyPtr[iSlot] );
  assert( iSlot==0 || iSlot==1 || iSlot==2 );
  assert( (pCsr->iNode==(db->treehdr.root.nHeight-1))==(iNewptr==0) );
  assert( ((u32)pCsr->iNode==(db->treehdr.root.nHeight-1))==(iNewptr==0) );

  bLeaf = (pCsr->iNode==(p->nHeight-1) && p->nHeight>1);
  bLeaf = ((u32)pCsr->iNode==(p->nHeight-1) && p->nHeight>1);
  
  if( pNode->aiKeyPtr[0] || pNode->aiKeyPtr[2] ){
    /* There are currently at least 2 keys on this node. So just create
    ** a new copy of the node with one of the keys removed. If the node
    ** happens to be the root node of the tree, allocate an entire 
    ** TreeNode structure instead of just a TreeLeaf.  */
    TreeNode *pNew;

          }
        }
      }
      if( iDir==-1 ){
        iPSlot--;
        pNew1->aiKeyPtr[iKOut++] = pParent->aiKeyPtr[iPSlot];
        if( bLeaf==0 ) pNew1->aiChildPtr[iPOut++] = iNewptr;
        pCsr->aiCell[pCsr->iNode] = iPSlot;
        pCsr->aiCell[pCsr->iNode] = (u8)iPSlot;
      }

      rc = treeDeleteEntry(db, pCsr, iNew1);
    }
  }

  return rc;

    bDone = 1;
    if( lsmTreeCursorValid(&csr) ){
      lsmTreeCursorKey(&csr, 0, &pDel, &nDel);
      if( treeKeycmp(pDel, nDel, pKey2, nKey2)<0 ) bDone = 0;
    }

    if( bDone==0 ){
      if( csr.iNode==(p->nHeight-1) ){
      if( (u32)csr.iNode==(p->nHeight-1) ){
        /* The element to delete already lies on a leaf node */
        rc = treeDeleteEntry(db, &csr, 0);
      }else{
        /* 1. Overwrite the current key with a copy of the next key in the 
        **    tree (key N).
        **
        ** 2. Seek to key N (cursor will stop at the internal node copy of
        **    N). Move to the next key (original copy of N). Delete
        **    this entry. 
        */
        u32 iKey;
        TreeKey *pKey;
        int iNode = csr.iNode;
        lsmTreeCursorNext(&csr);
        assert( csr.iNode==(p->nHeight-1) );
        assert( (u32)csr.iNode==(p->nHeight-1) );

        iKey = csr.apTreeNode[csr.iNode]->aiKeyPtr[csr.aiCell[csr.iNode]];
        lsmTreeCursorPrev(&csr);

        treeOverwriteKey(db, &csr, iKey, &rc);
        pKey = treeShmkey(db, iKey, TKV_LOADKEY, &blob, &rc);
        if( pKey ){

        pTreeKey = (TreeKey*)treeShmptrUnsafe(pDb, iTreeKey);
        if( !(pTreeKey->flags & LSM_CONTIGUOUS) ){
          pTreeKey = treeShmkey(pDb, iTreeKey, TKV_LOADKEY, &b, &rc);
          if( rc ) break;
        }
        res = treeKeycmp((void *)&pTreeKey[1], pTreeKey->nKey, pKey, nKey);
        if( res==0 ){
          pCsr->aiCell[iNode] = iTest;
          pCsr->aiCell[iNode] = (u8)iTest;
          break;
        }
      }else{
        iTest = 1;
      }

      if( iNode<(pRoot->nHeight-1) ){
      if( (u32)iNode<(pRoot->nHeight-1) ){
        iNodePtr = getChildPtr(pNode, pRoot->iTransId, iTest + (res<0));
      }else{
        iNodePtr = 0;
      }
      pCsr->aiCell[iNode] = iTest + (iNodePtr && (res<0));
      pCsr->aiCell[iNode] = (u8)(iTest + (iNodePtr && (res<0)));
    }

    *pRes = res;
    pCsr->iNode = iNode;
    tblobFree(pDb, &b);
  }

      u32 iNodePtr;
      pCsr->iNode++;
      iNodePtr = getChildPtr(pNode, pRoot->iTransId, iCell);
      pNode = (TreeNode *)treeShmptr(pDb, iNodePtr);
      if( rc!=LSM_OK ) break;
      pCsr->apTreeNode[pCsr->iNode] = pNode;
      iCell = 1 + (pNode->aiKeyPtr[2]!=0) + (pCsr->iNode < iLeaf);
      pCsr->aiCell[pCsr->iNode] = iCell;
      pCsr->aiCell[pCsr->iNode] = (u8)iCell;
    }while( pCsr->iNode < iLeaf );
  }

  /* Otherwise, the next key is found by following pointer up the tree until
  ** there is a key immediately to the left of the pointer followed to reach
  ** the sub-tree containing the current key. */
  else{
    do {
      iCell = pCsr->aiCell[pCsr->iNode]-1;
      if( iCell>=0 && pCsr->apTreeNode[pCsr->iNode]->aiKeyPtr[iCell] ) break;
    }while( (--pCsr->iNode)>=0 );
    pCsr->aiCell[pCsr->iNode] = iCell;
    pCsr->aiCell[pCsr->iNode] = (u8)iCell;
  }

#ifndef NDEBUG
  if( pCsr->iNode>=0 ){
    TreeKey *pK2 = csrGetKey(pCsr, &pCsr->blob, &rc);
    assert( rc || treeKeycmp(TKV_KEY(pK2),pK2->nKey,TKV_KEY(pK1),pK1->nKey)<0 );
  }

      iCell = ((pNode->aiKeyPtr[2]==0) ? 2 : 3);
    }else{
      iCell = ((pNode->aiKeyPtr[0]==0) ? 1 : 0);
    }
    pCsr->iNode++;
    pCsr->apTreeNode[pCsr->iNode] = pNode;

    if( pCsr->iNode<pRoot->nHeight-1 ){
    if( (u32)pCsr->iNode<pRoot->nHeight-1 ){
      iNodePtr = getChildPtr(pNode, pRoot->iTransId, iCell);
    }else{
      iNodePtr = 0;
    }
    pCsr->aiCell[pCsr->iNode] = iCell - (iNodePtr==0 && bLast);
    pCsr->aiCell[pCsr->iNode] = (u8)(iCell - (iNodePtr==0 && bLast));
  }

  return rc;
}

int lsmTreeCursorFlags(TreeCursor *pCsr){
  int flags = 0;

  }
  if( z[0]==250 ){
    *piVal = (z[1]<<16) + (z[2]<<8) + z[3];
    return 4;
  }

  ret = lsmSqlite4GetVarint64(z, &i);
  *piVal = i;
  *piVal = (int)i;
  return ret;
}

int lsmVarintLen32(int n){
  u8 aData[9];
  return lsmVarintPut32(aData, n);
}

/*
** Advance a lsm1_cursor to its next row of output.
*/
static int lsm1Next(sqlite3_vtab_cursor *cur){
  lsm1_cursor *pCur = (lsm1_cursor*)cur;
  int rc;
  int rc = LSM_OK;
  if( pCur->bUnique ){
    pCur->atEof = 1;
  }else{
    if( pCur->isDesc ){
      rc = lsm_csr_prev(pCur->pLsmCur);
    }else{
      rc = lsm_csr_next(pCur->pLsmCur);

        pVal = (const void*)sqlite3_value_text(pValue);
        if( pVal==0 ) return SQLITE_NOMEM;
      }
      if( nVal+1>nSpace ){
        pSpace = sqlite3_malloc( nVal+1 );
        if( pSpace==0 ) return SQLITE_NOMEM;
      }
      pSpace[0] = eType;
      pSpace[0] = (unsigned char)eType;
      memcpy(&pSpace[1], pVal, nVal);
      *ppKey = pSpace;
      *pnKey = nVal+1;
      break;
    }
    case SQLITE_INTEGER: {
      sqlite3_int64 iVal = sqlite3_value_int64(pValue);
      sqlite3_uint64 uVal;
      if( iVal<0 ){
        if( iVal==0xffffffffffffffffLL ) return SQLITE_ERROR;
        uVal = *(sqlite3_uint64*)&iVal;
        eType = LSM1_TYPE_NEGATIVE;
      }else{
        uVal = iVal;
        eType = LSM1_TYPE_POSITIVE;
      }
      pSpace[0] = eType;
      pSpace[0] = (unsigned char)eType;
      *ppKey = pSpace;
      *pnKey = 1 + lsm1PutVarint64(&pSpace[1], uVal);
    }
  }
  return SQLITE_OK;
}

  sqlite3_value **argv,
  sqlite_int64 *pRowid
){
  lsm1_vtab *p = (lsm1_vtab*)pVTab;
  const void *pKey;
  int nKey;
  int eType;
  int rc;
  int rc = LSM_OK;
  sqlite3_value *pValue;
  const unsigned char *pVal;
  unsigned char *pData;
  int nVal;
  unsigned char pSpace[100];

  if( argc==1 ){

          pVal = (unsigned char*)sqlite3_value_blob(pValue);
        }
        nVal = sqlite3_value_bytes(pValue);
        pData = sqlite3_malloc( nVal+1 );
        if( pData==0 ){
          rc = SQLITE_NOMEM;
        }else{
          pData[0] = eType;
          pData[0] = (unsigned char)eType;
          memcpy(&pData[1], pVal, nVal);
          rc = lsm_insert(p->pDb, pKey, nKey, pData, nVal+1);
          sqlite3_free(pData);
        }
        break;
      }
      case SQLITE_INTEGER:

          assert( sizeof(r)==sizeof(x) );
          memcpy(&x, &r, sizeof(r));
        }
        for(i=8; x>0 && i>=1; i--){
          aVal[i] = x & 0xff;
          x >>= 8;
        }
        aVal[i] = eType;
        aVal[i] = (unsigned char)eType;
        rc = lsm_insert(p->pDb, pKey, nKey, &aVal[i], 9-i);
        break;
      }
    }
  }
  if( pKey!=(const void*)pSpace ) sqlite3_free((void*)pKey);
  return rc==LSM_OK ? SQLITE_OK : SQLITE_ERROR;

  LPVOID pMap;                    /* Pointer to mapping of file fd */
  size_t nMap;                    /* Size of mapping at pMap in bytes */
  int nShm;                       /* Number of entries in ahShm[]/apShm[] */
  LPHANDLE ahShm;                 /* Array of handles for shared mappings */
  LPVOID *apShm;                  /* Array of 32K shared memory segments */
};

static char *win32ShmFile(Win32File *p){
static char *win32ShmFile(Win32File *pWin32File){
  char *zShm;
  int nName = strlen(p->zName);
  zShm = (char *)lsmMallocZero(p->pEnv, nName+4+1);
  int nName = strlen(pWin32File->zName);
  zShm = (char *)lsmMallocZero(pWin32File->pEnv, nName+4+1);
  if( zShm ){
    memcpy(zShm, p->zName, nName);
    memcpy(zShm, pWin32File->zName, nName);
    memcpy(&zShm[nName], "-shm", 5);
  }
  return zShm;
}

static int win32Sleep(int us){
  Sleep((us + 999) / 1000);

  int rc = LSM_OK;
  Win32File *pWin32File;

  pWin32File = lsmMallocZero(pEnv, sizeof(Win32File));
  if( pWin32File==0 ){
    rc = LSM_NOMEM_BKPT;
  }else{
    HANDLE hFile;
    HANDLE hFile = NULL;

    rc = win32Open(pEnv, zFile, flags, &hFile);
    if( rc==LSM_OK ){
      memset(&pWin32File->sysInfo, 0, sizeof(SYSTEM_INFO));
      GetSystemInfo(&pWin32File->sysInfo);
      pWin32File->pEnv = pEnv;
      pWin32File->zName = zFile;

  return rc;
}

static int lsmWin32OsSectorSize(lsm_file *pFile){
  return 512;
}

static void win32Unmap(Win32File *pWin32File){
  if( pWin32File->pMap!=NULL ){
    UnmapViewOfFile(pWin32File->pMap);
    pWin32File->pMap = NULL;
    pWin32File->nMap = 0;
  }
  if( pWin32File->hMap!=NULL ){
    CloseHandle(pWin32File->hMap);
    pWin32File->hMap = NULL;
  }
}

static int lsmWin32OsRemap(
  lsm_file *pFile,
  lsm_i64 iMin,
  void **ppOut,
  lsm_i64 *pnOut
){
  Win32File *pWin32File = (Win32File *)pFile;

  /* If the file is between 0 and 2MB in size, extend it in chunks of 256K.
  ** Thereafter, in chunks of 1MB at a time.  */
  const int aIncrSz[] = {256*1024, 1024*1024};
  int nIncrSz = aIncrSz[iMin>(2*1024*1024)];

  if( pWin32File->pMap!=NULL ){
    UnmapViewOfFile(pWin32File->pMap);
    *ppOut = pWin32File->pMap = NULL;
    *pnOut = pWin32File->nMap = 0;
  }
  *ppOut = NULL;
  *pnOut = 0;

  if( pWin32File->hMap!=NULL ){
    CloseHandle(pWin32File->hMap);
    pWin32File->hMap = NULL;
  win32Unmap(pWin32File);
  }
  if( iMin>=0 ){
    LARGE_INTEGER fileSize;
    DWORD dwSizeHigh;
    DWORD dwSizeLow;
    HANDLE hMap;
    LPVOID pMap;
    memset(&fileSize, 0, sizeof(LARGE_INTEGER));

}

#if !defined(win32IsLockBusy)
#define win32IsLockBusy(a) (((a)==ERROR_LOCK_VIOLATION) || \
                            ((a)==ERROR_IO_PENDING))
#endif

static int lsmWin32OsLock(lsm_file *pFile, int iLock, int eType){
  Win32File *pWin32File = (Win32File *)pFile;
static int win32LockFile(
  Win32File *pWin32File,
  int iLock,
  int nLock,
  int eType
){
  OVERLAPPED ovlp;

  assert( LSM_LOCK_UNLOCK==0 );
  assert( LSM_LOCK_SHARED==1 );
  assert( LSM_LOCK_EXCL==2 );
  assert( eType>=LSM_LOCK_UNLOCK && eType<=LSM_LOCK_EXCL );
  assert( nLock>=0 );
  assert( iLock>0 && iLock<=32 );

  memset(&ovlp, 0, sizeof(OVERLAPPED));
  ovlp.Offset = (4096-iLock);
  ovlp.Offset = (4096-iLock-nLock+1);
  if( eType>LSM_LOCK_UNLOCK ){
    DWORD flags = LOCKFILE_FAIL_IMMEDIATELY;
    if( eType>=LSM_LOCK_EXCL ) flags |= LOCKFILE_EXCLUSIVE_LOCK;
    if( !LockFileEx(pWin32File->hFile, flags, 0, 1, 0, &ovlp) ){
    if( !LockFileEx(pWin32File->hFile, flags, 0, (DWORD)nLock, 0, &ovlp) ){
      if( win32IsLockBusy(GetLastError()) ){
        return LSM_BUSY;
      }else{
        return LSM_IOERR_BKPT;
      }
    }
  }else{
    if( !UnlockFileEx(pWin32File->hFile, 0, 1, 0, &ovlp) ){
    if( !UnlockFileEx(pWin32File->hFile, 0, (DWORD)nLock, 0, &ovlp) ){
      return LSM_IOERR_BKPT;
    }
  }
  return LSM_OK;
}

static int lsmWin32OsTestLock(lsm_file *pFile, int iLock, int nLock, int eType){
static int lsmWin32OsLock(lsm_file *pFile, int iLock, int eType){
  Win32File *pWin32File = (Win32File *)pFile;
  DWORD flags = LOCKFILE_FAIL_IMMEDIATELY;
  return win32LockFile(pWin32File, iLock, 1, eType);
  OVERLAPPED ovlp;

}
  assert( LSM_LOCK_UNLOCK==0 );
  assert( LSM_LOCK_SHARED==1 );
  assert( LSM_LOCK_EXCL==2 );
  assert( eType==LSM_LOCK_SHARED || eType==LSM_LOCK_EXCL );
  assert( nLock>=0 );
  assert( iLock>0 && iLock<=32 );

  if( eType>=LSM_LOCK_EXCL ) flags |= LOCKFILE_EXCLUSIVE_LOCK;
  memset(&ovlp, 0, sizeof(OVERLAPPED));
  ovlp.Offset = (4096-iLock-nLock+1);
  if( !LockFileEx(pWin32File->hFile, flags, 0, (DWORD)nLock, 0, &ovlp) ){
    if( win32IsLockBusy(GetLastError()) ){
      return LSM_BUSY;
static int lsmWin32OsTestLock(lsm_file *pFile, int iLock, int nLock, int eType){
  int rc;
  Win32File *pWin32File = (Win32File *)pFile;
  rc = win32LockFile(pWin32File, iLock, nLock, eType);
  if( rc!=LSM_OK ) return rc;
    }else{
      return LSM_IOERR_BKPT;
    }
  }
  UnlockFileEx(pWin32File->hFile, 0, (DWORD)nLock, 0, &ovlp);
  win32LockFile(pWin32File, iLock, nLock, LSM_LOCK_UNLOCK);
  return LSM_OK;
}

static int lsmWin32OsShmMap(lsm_file *pFile, int iChunk, int sz, void **ppShm){
  int rc;
  Win32File *pWin32File = (Win32File *)pFile;
  int iOffset = iChunk * sz;
  int iOffsetShift = iOffset % pWin32File->sysInfo.dwAllocationGranularity;
  int nNew = iChunk + 1;
  lsm_i64 nReq = nNew * sz;

  *ppShm = NULL;
  assert( sz>=0 );
  assert( sz==LSM_SHM_CHUNK_SIZE );
  if( iChunk>=pWin32File->nShm ){
    int i;
    LPHANDLE ahNew;
    LPVOID *apNew;
    LARGE_INTEGER fileSize;

    /* If the shared-memory file has not been opened, open it now. */
    if( pWin32File->hShmFile==NULL ){
      char *zShm = win32ShmFile(pWin32File);

    if( fileSize.QuadPart<nReq ){
      rc = win32Truncate(pWin32File->hShmFile, nReq);
      if( rc!=LSM_OK ){
        return rc;
      }
    }

    ahNew = (LPHANDLE)lsmRealloc(pWin32File->pEnv, pWin32File->ahShm,
    ahNew = (LPHANDLE)lsmMallocZero(pWin32File->pEnv, sizeof(HANDLE) * nNew);
                                 sizeof(LPHANDLE) * nNew);
    if( !ahNew ) return LSM_NOMEM_BKPT;
    apNew = (LPVOID *)lsmRealloc(pWin32File->pEnv, pWin32File->apShm,
                                 sizeof(LPVOID) * nNew);
    apNew = (LPVOID *)lsmMallocZero(pWin32File->pEnv, sizeof(LPVOID) * nNew);
    if( !apNew ){
      lsmFree(pWin32File->pEnv, ahNew);
      return LSM_NOMEM_BKPT;
    }
    for(i=pWin32File->nShm; i<nNew; i++){
    memcpy(ahNew, pWin32File->ahShm, sizeof(HANDLE) * pWin32File->nShm);
      ahNew[i] = NULL;
      apNew[i] = NULL;
    memcpy(apNew, pWin32File->apShm, sizeof(LPVOID) * pWin32File->nShm);
    }
    lsmFree(pWin32File->pEnv, pWin32File->ahShm);
    pWin32File->ahShm = ahNew;
    lsmFree(pWin32File->pEnv, pWin32File->apShm);
    pWin32File->apShm = apNew;
    pWin32File->nShm = nNew;
  }

  if( pWin32File->ahShm[iChunk]==NULL ){
    HANDLE hMap;
    assert( nReq<=0xFFFFFFFF );

#define MX_CLOSE_ATTEMPT 3
static int lsmWin32OsClose(lsm_file *pFile){
  int rc;
  int nRetry = 0;
  Win32File *pWin32File = (Win32File *)pFile;
  lsmWin32OsShmUnmap(pFile, 0);
  if( pWin32File->pMap!=NULL ){
    UnmapViewOfFile(pWin32File->pMap);
    pWin32File->pMap = NULL;
    pWin32File->nMap = 0;
  win32Unmap(pWin32File);
  }
  if( pWin32File->hMap!=NULL ){
    CloseHandle(pWin32File->hMap);
    pWin32File->hMap = NULL;
  }
  do{
    if( pWin32File->hFile==NULL ){
      rc = LSM_IOERR_BKPT;
      break;
    }
    rc = CloseHandle(pWin32File->hFile);
    if( rc ){

}
static int lsmWin32OsMutexNotHeld(lsm_mutex *p){
  Win32Mutex *pMutex = (Win32Mutex *)p;
  return pMutex ? pMutex->owner!=GetCurrentThreadId() : 1;
}
#endif
/*
** End of pthreads mutex implementation.
** End of Win32 mutex implementation.
*************************************************************************/
#else
/*************************************************************************
** Noop mutex implementation
*/
typedef struct NoopMutex NoopMutex;
struct NoopMutex {

/*
** 2017-07-10
**
** The author disclaims copyright to this source code.  In place of
** a legal notice, here is a blessing:
**
**    May you do good and not evil.
**    May you find forgiveness for yourself and forgive others.
**    May you share freely, never taking more than you give.
**
*************************************************************************
**
** This file implements an eponymous virtual table that returns suggested
** completions for a partial SQL input.
**
** Suggested usage:
**
**     SELECT DISTINCT candidate COLLATE nocase
**       FROM completion($prefix,$wholeline)
**      ORDER BY 1;
**
** The two query parameters are optional.  $prefix is the text of the
** current word being typed and that is to be completed.  $wholeline is
** the complete input line, used for context.
**
** The raw completion() table might return the same candidate multiple
** times, for example if the same column name is used to two or more
** tables.  And the candidates are returned in an arbitrary order.  Hence,
** the DISTINCT and ORDER BY are recommended.
**
** This virtual table operates at the speed of human typing, and so there
** is no attempt to make it fast.  Even a slow implementation will be much
** faster than any human can type.
**
*/
#include "sqlite3ext.h"
SQLITE_EXTENSION_INIT1
#include <assert.h>
#include <string.h>
#include <ctype.h>

#ifndef SQLITE_OMIT_VIRTUALTABLE

/* completion_vtab is a subclass of sqlite3_vtab which will
** serve as the underlying representation of a completion virtual table
*/
typedef struct completion_vtab completion_vtab;
struct completion_vtab {
  sqlite3_vtab base;  /* Base class - must be first */
  sqlite3 *db;        /* Database connection for this completion vtab */
};

/* completion_cursor is a subclass of sqlite3_vtab_cursor which will
** serve as the underlying representation of a cursor that scans
** over rows of the result
*/
typedef struct completion_cursor completion_cursor;
struct completion_cursor {
  sqlite3_vtab_cursor base;  /* Base class - must be first */
  sqlite3 *db;               /* Database connection for this cursor */
  int nPrefix, nLine;        /* Number of bytes in zPrefix and zLine */
  char *zPrefix;             /* The prefix for the word we want to complete */
  char *zLine;               /* The whole that we want to complete */
  const char *zCurrentRow;   /* Current output row */
  sqlite3_stmt *pStmt;       /* Current statement */
  sqlite3_int64 iRowid;      /* The rowid */
  int ePhase;                /* Current phase */
  int j;                     /* inter-phase counter */
};

/* Values for ePhase:
*/
#define COMPLETION_FIRST_PHASE   1
#define COMPLETION_KEYWORDS      1
#define COMPLETION_PRAGMAS       2
#define COMPLETION_FUNCTIONS     3
#define COMPLETION_COLLATIONS    4
#define COMPLETION_INDEXES       5
#define COMPLETION_TRIGGERS      6
#define COMPLETION_DATABASES     7
#define COMPLETION_TABLES        8
#define COMPLETION_COLUMNS       9
#define COMPLETION_MODULES       10
#define COMPLETION_EOF           11

/*
** The completionConnect() method is invoked to create a new
** completion_vtab that describes the completion virtual table.
**
** Think of this routine as the constructor for completion_vtab objects.
**
** All this routine needs to do is:
**
**    (1) Allocate the completion_vtab object and initialize all fields.
**
**    (2) Tell SQLite (via the sqlite3_declare_vtab() interface) what the
**        result set of queries against completion will look like.
*/
static int completionConnect(
  sqlite3 *db,
  void *pAux,
  int argc, const char *const*argv,
  sqlite3_vtab **ppVtab,
  char **pzErr
){
  completion_vtab *pNew;
  int rc;

  (void)(pAux);    /* Unused parameter */
  (void)(argc);    /* Unused parameter */
  (void)(argv);    /* Unused parameter */
  (void)(pzErr);   /* Unused parameter */

/* Column numbers */
#define COMPLETION_COLUMN_CANDIDATE 0  /* Suggested completion of the input */
#define COMPLETION_COLUMN_PREFIX    1  /* Prefix of the word to be completed */
#define COMPLETION_COLUMN_WHOLELINE 2  /* Entire line seen so far */
#define COMPLETION_COLUMN_PHASE     3  /* ePhase - used for debugging only */

  rc = sqlite3_declare_vtab(db,
      "CREATE TABLE x("
      "  candidate TEXT,"
      "  prefix TEXT HIDDEN,"
      "  wholeline TEXT HIDDEN,"
      "  phase INT HIDDEN"        /* Used for debugging only */
      ")");
  if( rc==SQLITE_OK ){
    pNew = sqlite3_malloc( sizeof(*pNew) );
    *ppVtab = (sqlite3_vtab*)pNew;
    if( pNew==0 ) return SQLITE_NOMEM;
    memset(pNew, 0, sizeof(*pNew));
    pNew->db = db;
  }
  return rc;
}

/*
** This method is the destructor for completion_cursor objects.
*/
static int completionDisconnect(sqlite3_vtab *pVtab){
  sqlite3_free(pVtab);
  return SQLITE_OK;
}

/*
** Constructor for a new completion_cursor object.
*/
static int completionOpen(sqlite3_vtab *p, sqlite3_vtab_cursor **ppCursor){
  completion_cursor *pCur;
  pCur = sqlite3_malloc( sizeof(*pCur) );
  if( pCur==0 ) return SQLITE_NOMEM;
  memset(pCur, 0, sizeof(*pCur));
  pCur->db = ((completion_vtab*)p)->db;
  *ppCursor = &pCur->base;
  return SQLITE_OK;
}

/*
** Reset the completion_cursor.
*/
static void completionCursorReset(completion_cursor *pCur){
  sqlite3_free(pCur->zPrefix);   pCur->zPrefix = 0;  pCur->nPrefix = 0;
  sqlite3_free(pCur->zLine);     pCur->zLine = 0;    pCur->nLine = 0;
  sqlite3_finalize(pCur->pStmt); pCur->pStmt = 0;
  pCur->j = 0;
}

/*
** Destructor for a completion_cursor.
*/
static int completionClose(sqlite3_vtab_cursor *cur){
  completionCursorReset((completion_cursor*)cur);
  sqlite3_free(cur);
  return SQLITE_OK;
}

/*
** All SQL keywords understood by SQLite
*/
static const char *completionKwrds[] = {
  "ABORT", "ACTION", "ADD", "AFTER", "ALL", "ALTER", "ANALYZE", "AND", "AS",
  "ASC", "ATTACH", "AUTOINCREMENT", "BEFORE", "BEGIN", "BETWEEN", "BY",
  "CASCADE", "CASE", "CAST", "CHECK", "COLLATE", "COLUMN", "COMMIT",
  "CONFLICT", "CONSTRAINT", "CREATE", "CROSS", "CURRENT_DATE",
  "CURRENT_TIME", "CURRENT_TIMESTAMP", "DATABASE", "DEFAULT", "DEFERRABLE",
  "DEFERRED", "DELETE", "DESC", "DETACH", "DISTINCT", "DROP", "EACH",
  "ELSE", "END", "ESCAPE", "EXCEPT", "EXCLUSIVE", "EXISTS", "EXPLAIN",
  "FAIL", "FOR", "FOREIGN", "FROM", "FULL", "GLOB", "GROUP", "HAVING", "IF",
  "IGNORE", "IMMEDIATE", "IN", "INDEX", "INDEXED", "INITIALLY", "INNER",
  "INSERT", "INSTEAD", "INTERSECT", "INTO", "IS", "ISNULL", "JOIN", "KEY",
  "LEFT", "LIKE", "LIMIT", "MATCH", "NATURAL", "NO", "NOT", "NOTNULL",
  "NULL", "OF", "OFFSET", "ON", "OR", "ORDER", "OUTER", "PLAN", "PRAGMA",
  "PRIMARY", "QUERY", "RAISE", "RECURSIVE", "REFERENCES", "REGEXP",
  "REINDEX", "RELEASE", "RENAME", "REPLACE", "RESTRICT", "RIGHT",
  "ROLLBACK", "ROW", "SAVEPOINT", "SELECT", "SET", "TABLE", "TEMP",
  "TEMPORARY", "THEN", "TO", "TRANSACTION", "TRIGGER", "UNION", "UNIQUE",
  "UPDATE", "USING", "VACUUM", "VALUES", "VIEW", "VIRTUAL", "WHEN", "WHERE",
  "WITH", "WITHOUT",
};
#define completionKwCount \
   (int)(sizeof(completionKwrds)/sizeof(completionKwrds[0]))

/*
** Advance a completion_cursor to its next row of output.
**
** The ->ePhase, ->j, and ->pStmt fields of the completion_cursor object
** record the current state of the scan.  This routine sets ->zCurrentRow
** to the current row of output and then returns.  If no more rows remain,
** then ->ePhase is set to COMPLETION_EOF which will signal the virtual
** table that has reached the end of its scan.
**
** The current implementation just lists potential identifiers and
** keywords and filters them by zPrefix.  Future enhancements should
** take zLine into account to try to restrict the set of identifiers and
** keywords based on what would be legal at the current point of input.
*/
static int completionNext(sqlite3_vtab_cursor *cur){
  completion_cursor *pCur = (completion_cursor*)cur;
  int eNextPhase = 0;  /* Next phase to try if current phase reaches end */
  int iCol = -1;       /* If >=0, step pCur->pStmt and use the i-th column */
  pCur->iRowid++;
  while( pCur->ePhase!=COMPLETION_EOF ){
    switch( pCur->ePhase ){
      case COMPLETION_KEYWORDS: {
        if( pCur->j >= completionKwCount ){
          pCur->zCurrentRow = 0;
          pCur->ePhase = COMPLETION_DATABASES;
        }else{
          pCur->zCurrentRow = completionKwrds[pCur->j++];
        }
        iCol = -1;
        break;
      }
      case COMPLETION_DATABASES: {
        if( pCur->pStmt==0 ){
          sqlite3_prepare_v2(pCur->db, "PRAGMA database_list", -1,
                             &pCur->pStmt, 0);
        }
        iCol = 1;
        eNextPhase = COMPLETION_TABLES;
        break;
      }
      case COMPLETION_TABLES: {
        if( pCur->pStmt==0 ){
          sqlite3_stmt *pS2;
          char *zSql = 0;
          const char *zSep = "";
          sqlite3_prepare_v2(pCur->db, "PRAGMA database_list", -1, &pS2, 0);
          while( sqlite3_step(pS2)==SQLITE_ROW ){
            const char *zDb = (const char*)sqlite3_column_text(pS2, 1);
            zSql = sqlite3_mprintf(
               "%z%s"
               "SELECT name FROM \"%w\".sqlite_master"
               " WHERE type='table'",
               zSql, zSep, zDb
            );
            if( zSql==0 ) return SQLITE_NOMEM;
            zSep = " UNION ";
          }
          sqlite3_finalize(pS2);
          sqlite3_prepare_v2(pCur->db, zSql, -1, &pCur->pStmt, 0);
          sqlite3_free(zSql);
        }
        iCol = 0;
        eNextPhase = COMPLETION_COLUMNS;
        break;
      }
      case COMPLETION_COLUMNS: {
        if( pCur->pStmt==0 ){
          sqlite3_stmt *pS2;
          char *zSql = 0;
          const char *zSep = "";
          sqlite3_prepare_v2(pCur->db, "PRAGMA database_list", -1, &pS2, 0);
          while( sqlite3_step(pS2)==SQLITE_ROW ){
            const char *zDb = (const char*)sqlite3_column_text(pS2, 1);
            zSql = sqlite3_mprintf(
               "%z%s"
               "SELECT pti.name FROM \"%w\".sqlite_master AS sm"
                       " JOIN pragma_table_info(sm.name,%Q) AS pti"
               " WHERE sm.type='table'",
               zSql, zSep, zDb, zDb
            );
            if( zSql==0 ) return SQLITE_NOMEM;
            zSep = " UNION ";
          }
          sqlite3_finalize(pS2);
          sqlite3_prepare_v2(pCur->db, zSql, -1, &pCur->pStmt, 0);
          sqlite3_free(zSql);
        }
        iCol = 0;
        eNextPhase = COMPLETION_EOF;
        break;
      }
    }
    if( iCol<0 ){
      /* This case is when the phase presets zCurrentRow */
      if( pCur->zCurrentRow==0 ) continue;
    }else{
      if( sqlite3_step(pCur->pStmt)==SQLITE_ROW ){
        /* Extract the next row of content */
        pCur->zCurrentRow = (const char*)sqlite3_column_text(pCur->pStmt, iCol);
      }else{
        /* When all rows are finished, advance to the next phase */
        sqlite3_finalize(pCur->pStmt);
        pCur->pStmt = 0;
        pCur->ePhase = eNextPhase;
        continue;
      }
    }
    if( pCur->nPrefix==0 ) break;
    if( sqlite3_strnicmp(pCur->zPrefix, pCur->zCurrentRow, pCur->nPrefix)==0 ){
      break;
    }
  }

  return SQLITE_OK;
}

/*
** Return values of columns for the row at which the completion_cursor
** is currently pointing.
*/
static int completionColumn(
  sqlite3_vtab_cursor *cur,   /* The cursor */
  sqlite3_context *ctx,       /* First argument to sqlite3_result_...() */
  int i                       /* Which column to return */
){
  completion_cursor *pCur = (completion_cursor*)cur;
  switch( i ){
    case COMPLETION_COLUMN_CANDIDATE: {
      sqlite3_result_text(ctx, pCur->zCurrentRow, -1, SQLITE_TRANSIENT);
      break;
    }
    case COMPLETION_COLUMN_PREFIX: {
      sqlite3_result_text(ctx, pCur->zPrefix, -1, SQLITE_TRANSIENT);
      break;
    }
    case COMPLETION_COLUMN_WHOLELINE: {
      sqlite3_result_text(ctx, pCur->zLine, -1, SQLITE_TRANSIENT);
      break;
    }
    case COMPLETION_COLUMN_PHASE: {
      sqlite3_result_int(ctx, pCur->ePhase);
      break;
    }
  }
  return SQLITE_OK;
}

/*
** Return the rowid for the current row.  In this implementation, the
** rowid is the same as the output value.
*/
static int completionRowid(sqlite3_vtab_cursor *cur, sqlite_int64 *pRowid){
  completion_cursor *pCur = (completion_cursor*)cur;
  *pRowid = pCur->iRowid;
  return SQLITE_OK;
}

/*
** Return TRUE if the cursor has been moved off of the last
** row of output.
*/
static int completionEof(sqlite3_vtab_cursor *cur){
  completion_cursor *pCur = (completion_cursor*)cur;
  return pCur->ePhase >= COMPLETION_EOF;
}

/*
** This method is called to "rewind" the completion_cursor object back
** to the first row of output.  This method is always called at least
** once prior to any call to completionColumn() or completionRowid() or 
** completionEof().
*/
static int completionFilter(
  sqlite3_vtab_cursor *pVtabCursor, 
  int idxNum, const char *idxStr,
  int argc, sqlite3_value **argv
){
  completion_cursor *pCur = (completion_cursor *)pVtabCursor;
  int iArg = 0;
  (void)(idxStr);   /* Unused parameter */
  (void)(argc);     /* Unused parameter */
  completionCursorReset(pCur);
  if( idxNum & 1 ){
    pCur->nPrefix = sqlite3_value_bytes(argv[iArg]);
    if( pCur->nPrefix>0 ){
      pCur->zPrefix = sqlite3_mprintf("%s", sqlite3_value_text(argv[iArg]));
      if( pCur->zPrefix==0 ) return SQLITE_NOMEM;
    }
    iArg++;
  }
  if( idxNum & 2 ){
    pCur->nLine = sqlite3_value_bytes(argv[iArg]);
    if( pCur->nLine>0 ){
      pCur->zLine = sqlite3_mprintf("%s", sqlite3_value_text(argv[iArg]));
      if( pCur->zLine==0 ) return SQLITE_NOMEM;
    }
    iArg++;
  }
  if( pCur->zLine!=0 && pCur->zPrefix==0 ){
    int i = pCur->nLine;
    while( i>0 && (isalnum(pCur->zLine[i-1]) || pCur->zLine[i-1]=='_') ){
      i--;
    }
    pCur->nPrefix = pCur->nLine - i;
    if( pCur->nPrefix>0 ){
      pCur->zPrefix = sqlite3_mprintf("%.*s", pCur->nPrefix, pCur->zLine + i);
      if( pCur->zPrefix==0 ) return SQLITE_NOMEM;
    }
  }
  pCur->iRowid = 0;
  pCur->ePhase = COMPLETION_FIRST_PHASE;
  return completionNext(pVtabCursor);
}

/*
** SQLite will invoke this method one or more times while planning a query
** that uses the completion virtual table.  This routine needs to create
** a query plan for each invocation and compute an estimated cost for that
** plan.
**
** There are two hidden parameters that act as arguments to the table-valued
** function:  "prefix" and "wholeline".  Bit 0 of idxNum is set if "prefix"
** is available and bit 1 is set if "wholeline" is available.
*/
static int completionBestIndex(
  sqlite3_vtab *tab,
  sqlite3_index_info *pIdxInfo
){
  int i;                 /* Loop over constraints */
  int idxNum = 0;        /* The query plan bitmask */
  int prefixIdx = -1;    /* Index of the start= constraint, or -1 if none */
  int wholelineIdx = -1; /* Index of the stop= constraint, or -1 if none */
  int nArg = 0;          /* Number of arguments that completeFilter() expects */
  const struct sqlite3_index_constraint *pConstraint;

  (void)(tab);    /* Unused parameter */
  pConstraint = pIdxInfo->aConstraint;
  for(i=0; i<pIdxInfo->nConstraint; i++, pConstraint++){
    if( pConstraint->usable==0 ) continue;
    if( pConstraint->op!=SQLITE_INDEX_CONSTRAINT_EQ ) continue;
    switch( pConstraint->iColumn ){
      case COMPLETION_COLUMN_PREFIX:
        prefixIdx = i;
        idxNum |= 1;
        break;
      case COMPLETION_COLUMN_WHOLELINE:
        wholelineIdx = i;
        idxNum |= 2;
        break;
    }
  }
  if( prefixIdx>=0 ){
    pIdxInfo->aConstraintUsage[prefixIdx].argvIndex = ++nArg;
    pIdxInfo->aConstraintUsage[prefixIdx].omit = 1;
  }
  if( wholelineIdx>=0 ){
    pIdxInfo->aConstraintUsage[wholelineIdx].argvIndex = ++nArg;
    pIdxInfo->aConstraintUsage[wholelineIdx].omit = 1;
  }
  pIdxInfo->idxNum = idxNum;
  pIdxInfo->estimatedCost = (double)5000 - 1000*nArg;
  pIdxInfo->estimatedRows = 500 - 100*nArg;
  return SQLITE_OK;
}

/*
** This following structure defines all the methods for the 
** completion virtual table.
*/
static sqlite3_module completionModule = {
  0,                         /* iVersion */
  0,                         /* xCreate */
  completionConnect,         /* xConnect */
  completionBestIndex,       /* xBestIndex */
  completionDisconnect,      /* xDisconnect */
  0,                         /* xDestroy */
  completionOpen,            /* xOpen - open a cursor */
  completionClose,           /* xClose - close a cursor */
  completionFilter,          /* xFilter - configure scan constraints */
  completionNext,            /* xNext - advance a cursor */
  completionEof,             /* xEof - check for end of scan */
  completionColumn,          /* xColumn - read data */
  completionRowid,           /* xRowid - read data */
  0,                         /* xUpdate */
  0,                         /* xBegin */
  0,                         /* xSync */
  0,                         /* xCommit */
  0,                         /* xRollback */
  0,                         /* xFindMethod */
  0,                         /* xRename */
  0,                         /* xSavepoint */
  0,                         /* xRelease */
  0                          /* xRollbackTo */
};

#endif /* SQLITE_OMIT_VIRTUALTABLE */

int sqlite3CompletionVtabInit(sqlite3 *db){
  int rc = SQLITE_OK;
#ifndef SQLITE_OMIT_VIRTUALTABLE
  rc = sqlite3_create_module(db, "completion", &completionModule, 0);
#endif
  return rc;
}

#ifdef _WIN32
__declspec(dllexport)
#endif
int sqlite3_completion_init(
  sqlite3 *db, 
  char **pzErrMsg, 
  const sqlite3_api_routines *pApi
){
  int rc = SQLITE_OK;
  SQLITE_EXTENSION_INIT2(pApi);
  (void)(pzErrMsg);  /* Unused parameter */
#ifndef SQLITE_OMIT_VIRTUALTABLE
  rc = sqlite3CompletionVtabInit(db);
#endif
  return rc;
}

  sqlite3_value **argv
){
  const char *zName;
  FILE *in;
  long nIn;
  void *pBuf;

  (void)(argc);  /* Unused parameter */
  zName = (const char*)sqlite3_value_text(argv[0]);
  if( zName==0 ) return;
  in = fopen(zName, "rb");
  if( in==0 ) return;
  fseek(in, 0, SEEK_END);
  nIn = ftell(in);
  rewind(in);

  sqlite3_value **argv
){
  FILE *out;
  const char *z;
  sqlite3_int64 rc;
  const char *zFile;

  (void)(argc);  /* Unused parameter */
  zFile = (const char*)sqlite3_value_text(argv[0]);
  if( zFile==0 ) return;
  out = fopen(zFile, "wb");
  if( out==0 ) return;
  z = (const char*)sqlite3_value_blob(argv[1]);
  if( z==0 ){
    rc = 0;

  sqlite3 *db;               /* Database connection for this cursor */
  sqlite3_stmt *pStmt;       /* Statement cursor is currently pointing at */
  sqlite3_int64 iRowid;      /* The rowid */
};

/*
** The stmtConnect() method is invoked to create a new
** stmt_vtab that describes the generate_stmt virtual table.
** stmt_vtab that describes the stmt virtual table.
**
** Think of this routine as the constructor for stmt_vtab objects.
**
** All this routine needs to do is:
**
**    (1) Allocate the stmt_vtab object and initialize all fields.
**
**    (2) Tell SQLite (via the sqlite3_declare_vtab() interface) what the
**        result set of queries against generate_stmt will look like.
**        result set of queries against stmt will look like.
*/
static int stmtConnect(
  sqlite3 *db,
  void *pAux,
  int argc, const char *const*argv,
  sqlite3_vtab **ppVtab,
  char **pzErr

  pCur->pStmt = 0;
  pCur->iRowid = 0;
  return stmtNext(pVtabCursor);
}

/*
** SQLite will invoke this method one or more times while planning a query
** that uses the generate_stmt virtual table.  This routine needs to create
** that uses the stmt virtual table.  This routine needs to create
** a query plan for each invocation and compute an estimated cost for that
** plan.
*/
static int stmtBestIndex(
  sqlite3_vtab *tab,
  sqlite3_index_info *pIdxInfo
){
  pIdxInfo->estimatedCost = (double)500;
  pIdxInfo->estimatedRows = 500;
  return SQLITE_OK;
}

/*
** This following structure defines all the methods for the 
** generate_stmt virtual table.
** stmt virtual table.
*/
static sqlite3_module stmtModule = {
  0,                         /* iVersion */
  0,                         /* xCreate */
  stmtConnect,               /* xConnect */
  stmtBestIndex,             /* xBestIndex */
  stmtDisconnect,            /* xDisconnect */

  0,                         /* xUpdate */
  0,                         /* xBegin */
  0,                         /* xSync */
  0,                         /* xCommit */
  0,                         /* xRollback */
  0,                         /* xFindMethod */
  0,                         /* xRename */
  0,                         /* xSavepoint */
  0,                         /* xRelease */
  0,                         /* xRollbackTo */
};

#endif /* SQLITE_OMIT_VIRTUALTABLE */

int sqlite3StmtVtabInit(sqlite3 *db){
  int rc = SQLITE_OK;
#ifndef SQLITE_OMIT_VIRTUALTABLE

# An error midway through a rename operation.
do_test rtree-7.2.1 {
  execsql { 
    CREATE TABLE t4_node(a);
  }
  catchsql { ALTER TABLE "abc 123" RENAME TO t4 }
} {1 {SQL logic error or missing database}}
} {1 {SQL logic error}}
do_test rtree-7.2.2 {
  execsql_intout { SELECT * FROM "abc 123" }
} {1 2 3 4 5 6 7}
do_test rtree-7.2.3 {
  execsql { 
    DROP TABLE t4_node;
    CREATE TABLE t4_rowid(a);
  }
  catchsql { ALTER TABLE "abc 123" RENAME TO t4 }
} {1 {SQL logic error or missing database}}
} {1 {SQL logic error}}
do_test rtree-7.2.4 {
  db close
  sqlite3 db test.db
  execsql_intout { SELECT * FROM "abc 123" }
} {1 2 3 4 5 6 7}
do_test rtree-7.2.5 {
  execsql { DROP TABLE t4_rowid }

#-------------------------------------------------------------------------
# Test that trying to use the MATCH operator with the r-tree module does
# not confuse it. 
#
populate_t1 10
do_catchsql_test rtree8-3.1 { 
  SELECT * FROM t1 WHERE x1 MATCH '1234'
} {1 {SQL logic error or missing database}}
} {1 {SQL logic error}}

#-------------------------------------------------------------------------
# Test a couple of invalid arguments to rtreedepth().
#
do_catchsql_test rtree8-4.1 {
  SELECT rtreedepth('hello world')
} {1 {Invalid argument to rtreedepth()}}

do_execsql_test rtree9-3.3 {
  SELECT id FROM rt32 WHERE id MATCH cube(5.5, 5.5, 5.5, 1, 1, 1) ORDER BY id;
} {555 556 565 566 655 656 665 666}


do_catchsql_test rtree9-4.1 {
  SELECT id FROM rt32 WHERE id MATCH cube(5.5, 5.5, 1, 1, 1) ORDER BY id;
} {1 {SQL logic error or missing database}}
} {1 {SQL logic error}}
for {set x 2} {$x<200} {incr x 2} {
  do_catchsql_test rtree9-4.2.[expr $x/2] {
    SELECT id FROM rt WHERE id MATCH randomblob($x)
  } {1 {SQL logic error or missing database}}
  } {1 {SQL logic error}}
}
do_catchsql_test rtree9-4.3 {
  SELECT id FROM rt WHERE id MATCH CAST( 
    (cube(5.5, 5.5, 5.5, 1, 1, 1) || X'1234567812345678') AS blob 
  )
} {1 {SQL logic error or missing database}}
} {1 {SQL logic error}}


#-------------------------------------------------------------------------
# Test the example 2d "circle" geometry callback.
#
register_circle_geom db

  do_test 1.$tn.4 {
    list [catch { sql2 { SELECT * FROM rt } } msg] $msg
  } {1 {database is locked}}
}

finish_test

  assert( CORRUPT_DB || iEnd <= pPage->pBt->usableSize );
  assert( sqlite3_mutex_held(pPage->pBt->mutex) );
  assert( iSize>=4 );   /* Minimum cell size is 4 */
  assert( iStart<=iLast );

  /* Overwrite deleted information with zeros when the secure_delete
  ** option is enabled */
  if( pPage->pBt->btsFlags & BTS_SECURE_DELETE ){
  if( pPage->pBt->btsFlags & BTS_FAST_SECURE ){
    memset(&data[iStart], 0, iSize);
  }

  /* The list of freeblocks must be in ascending order.  Find the 
  ** spot on the list where iStart should be inserted.
  */
  hdr = pPage->hdrOffset;

  u16 first;

  assert( sqlite3PagerPagenumber(pPage->pDbPage)==pPage->pgno );
  assert( sqlite3PagerGetExtra(pPage->pDbPage) == (void*)pPage );
  assert( sqlite3PagerGetData(pPage->pDbPage) == data );
  assert( sqlite3PagerIswriteable(pPage->pDbPage) );
  assert( sqlite3_mutex_held(pBt->mutex) );
  if( pBt->btsFlags & BTS_SECURE_DELETE ){
  if( pBt->btsFlags & BTS_FAST_SECURE ){
    memset(&data[hdr], 0, pBt->usableSize - hdr);
  }
  data[hdr] = (char)flags;
  first = hdr + ((flags&PTF_LEAF)==0 ? 12 : 8);
  memset(&data[hdr+1], 0, 4);
  data[hdr+7] = 0;
  put2byte(&data[hdr+5], pBt->usableSize);

    pBt->db = db;
    sqlite3PagerSetBusyhandler(pBt->pPager, btreeInvokeBusyHandler, pBt);
    p->pBt = pBt;
  
    pBt->pCursor = 0;
    pBt->pPage1 = 0;
    if( sqlite3PagerIsreadonly(pBt->pPager) ) pBt->btsFlags |= BTS_READ_ONLY;
#ifdef SQLITE_SECURE_DELETE
#if defined(SQLITE_SECURE_DELETE)
    pBt->btsFlags |= BTS_SECURE_DELETE;
#elif defined(SQLITE_FAST_SECURE_DELETE)
    pBt->btsFlags |= BTS_OVERWRITE;
#endif
    /* EVIDENCE-OF: R-51873-39618 The page size for a database file is
    ** determined by the 2-byte integer located at an offset of 16 bytes from
    ** the beginning of the database file. */
    pBt->pageSize = (zDbHeader[16]<<8) | (zDbHeader[17]<<16);
    if( pBt->pageSize<512 || pBt->pageSize>SQLITE_MAX_PAGE_SIZE
         || ((pBt->pageSize-1)&pBt->pageSize)!=0 ){

  sqlite3BtreeEnter(p);
  n = sqlite3PagerMaxPageCount(p->pBt->pPager, mxPage);
  sqlite3BtreeLeave(p);
  return n;
}

/*
** Set the BTS_SECURE_DELETE flag if newFlag is 0 or 1.  If newFlag is -1,
** then make no changes.  Always return the value of the BTS_SECURE_DELETE
** setting after the change.
** Change the values for the BTS_SECURE_DELETE and BTS_OVERWRITE flags:
**
**    newFlag==0       Both BTS_SECURE_DELETE and BTS_OVERWRITE are cleared
**    newFlag==1       BTS_SECURE_DELETE set and BTS_OVERWRITE is cleared
**    newFlag==2       BTS_SECURE_DELETE cleared and BTS_OVERWRITE is set
**    newFlag==(-1)    No changes
**
** This routine acts as a query if newFlag is less than zero
**
** With BTS_OVERWRITE set, deleted content is overwritten by zeros, but
** freelist leaf pages are not written back to the database.  Thus in-page
** deleted content is cleared, but freelist deleted content is not.
**
** With BTS_SECURE_DELETE, operation is like BTS_OVERWRITE with the addition
** that freelist leaf pages are written back into the database, increasing
** the amount of disk I/O.
*/
int sqlite3BtreeSecureDelete(Btree *p, int newFlag){
  int b;
  if( p==0 ) return 0;
  sqlite3BtreeEnter(p);
  assert( BTS_OVERWRITE==BTS_SECURE_DELETE*2 );
  assert( BTS_FAST_SECURE==(BTS_OVERWRITE|BTS_SECURE_DELETE) );
  if( newFlag>=0 ){
    p->pBt->btsFlags &= ~BTS_SECURE_DELETE;
    if( newFlag ) p->pBt->btsFlags |= BTS_SECURE_DELETE;
  } 
  b = (p->pBt->btsFlags & BTS_SECURE_DELETE)!=0;
    p->pBt->btsFlags &= ~BTS_FAST_SECURE;
    p->pBt->btsFlags |= BTS_SECURE_DELETE*newFlag;
  }
  b = (p->pBt->btsFlags & BTS_FAST_SECURE)/BTS_SECURE_DELETE;
  sqlite3BtreeLeave(p);
  return b;
}

/*
** Change the 'auto-vacuum' property of the database. If the 'autoVacuum'
** parameter is non-zero, then auto-vacuum mode is enabled. If zero, it

**
** The main entry point is sqlite3BtreeNext().  That routine is optimized
** for the common case of merely incrementing the cell counter BtCursor.aiIdx
** to the next cell on the current page.  The (slower) btreeNext() helper
** routine is called when it is necessary to move to a different page or
** to restore the cursor.
**
** If bit 0x01 of the flags argument is 1, then the cursor corresponds to
** an SQL index and this routine could have been skipped if the SQL index
** had been a unique index.  The flags argument is a hint to the implement.
** SQLite btree implementation does not use this hint, but COMDB2 does.
** If bit 0x01 of the F argument in sqlite3BtreeNext(C,F) is 1, then the
** cursor corresponds to an SQL index and this routine could have been
** skipped if the SQL index had been a unique index.  The F argument
** is a hint to the implement.  SQLite btree implementation does not use
** this hint, but COMDB2 does.
*/
static SQLITE_NOINLINE int btreeNext(BtCursor *pCur, int flags){
static SQLITE_NOINLINE int btreeNext(BtCursor *pCur){
  int rc;
  int idx;
  MemPage *pPage;

  assert( cursorOwnsBtShared(pCur) );
  assert( pCur->skipNext==0 || pCur->eState!=CURSOR_VALID );
  assert( flags==0 );
  if( pCur->eState!=CURSOR_VALID ){
    assert( (pCur->curFlags & BTCF_ValidOvfl)==0 );
    rc = restoreCursorPosition(pCur);
    if( rc!=SQLITE_OK ){
      return rc;
    }
    if( CURSOR_INVALID==pCur->eState ){

        pCur->eState = CURSOR_INVALID;
        return SQLITE_DONE;
      }
      moveToParent(pCur);
      pPage = pCur->apPage[pCur->iPage];
    }while( pCur->ix>=pPage->nCell );
    if( pPage->intKey ){
      return sqlite3BtreeNext(pCur, flags);
      return sqlite3BtreeNext(pCur, 0);
    }else{
      return SQLITE_OK;
    }
  }
  if( pPage->leaf ){
    return SQLITE_OK;
  }else{
    return moveToLeftmost(pCur);
  }
}
int sqlite3BtreeNext(BtCursor *pCur, int flags){
  MemPage *pPage;
  UNUSED_PARAMETER( flags );  /* Used in COMDB2 but not native SQLite */
  assert( cursorOwnsBtShared(pCur) );
  assert( flags==0 || flags==1 );
  assert( pCur->skipNext==0 || pCur->eState!=CURSOR_VALID );
  pCur->info.nSize = 0;
  pCur->curFlags &= ~(BTCF_ValidNKey|BTCF_ValidOvfl);
  if( pCur->eState!=CURSOR_VALID ) return btreeNext(pCur, 0);
  if( pCur->eState!=CURSOR_VALID ) return btreeNext(pCur);
  pPage = pCur->apPage[pCur->iPage];
  if( (++pCur->ix)>=pPage->nCell ){
    pCur->ix--;
    return btreeNext(pCur, 0);
    return btreeNext(pCur);
  }
  if( pPage->leaf ){
    return SQLITE_OK;
  }else{
    return moveToLeftmost(pCur);
  }
}

**
** The main entry point is sqlite3BtreePrevious().  That routine is optimized
** for the common case of merely decrementing the cell counter BtCursor.aiIdx
** to the previous cell on the current page.  The (slower) btreePrevious()
** helper routine is called when it is necessary to move to a different page
** or to restore the cursor.
**
**
** If bit 0x01 of the flags argument is 1, then the cursor corresponds to
** an SQL index and this routine could have been skipped if the SQL index
** had been a unique index.  The flags argument is a hint to the implement.
** SQLite btree implementation does not use this hint, but COMDB2 does.
** If bit 0x01 of the F argument to sqlite3BtreePrevious(C,F) is 1, then
** the cursor corresponds to an SQL index and this routine could have been
** skipped if the SQL index had been a unique index.  The F argument is a
** hint to the implement.  The native SQLite btree implementation does not
** use this hint, but COMDB2 does.
*/
static SQLITE_NOINLINE int btreePrevious(BtCursor *pCur, int flags){
static SQLITE_NOINLINE int btreePrevious(BtCursor *pCur){
  int rc;
  MemPage *pPage;

  assert( cursorOwnsBtShared(pCur) );
  assert( flags==0 );
  assert( pCur->skipNext==0 || pCur->eState!=CURSOR_VALID );
  assert( (pCur->curFlags & (BTCF_AtLast|BTCF_ValidOvfl|BTCF_ValidNKey))==0 );
  assert( pCur->info.nSize==0 );
  if( pCur->eState!=CURSOR_VALID ){
    rc = restoreCursorPosition(pCur);
    if( rc!=SQLITE_OK ){
      return rc;

    }
    assert( pCur->info.nSize==0 );
    assert( (pCur->curFlags & (BTCF_ValidOvfl))==0 );

    pCur->ix--;
    pPage = pCur->apPage[pCur->iPage];
    if( pPage->intKey && !pPage->leaf ){
      rc = sqlite3BtreePrevious(pCur, flags);
      rc = sqlite3BtreePrevious(pCur, 0);
    }else{
      rc = SQLITE_OK;
    }
  }
  return rc;
}
int sqlite3BtreePrevious(BtCursor *pCur, int flags){
  assert( cursorOwnsBtShared(pCur) );
  assert( flags==0 || flags==1 );
  assert( pCur->skipNext==0 || pCur->eState!=CURSOR_VALID );
  UNUSED_PARAMETER( flags );  /* Used in COMDB2 but not native SQLite */
  pCur->curFlags &= ~(BTCF_AtLast|BTCF_ValidOvfl|BTCF_ValidNKey);
  pCur->info.nSize = 0;
  if( pCur->eState!=CURSOR_VALID
   || pCur->ix==0
   || pCur->apPage[pCur->iPage]->leaf==0
  ){
    return btreePrevious(pCur, 0);
    return btreePrevious(pCur);
  }
  pCur->ix--;
  return SQLITE_OK;
}

/*
** Allocate a new page from the database file.

      ** later on.  
      **
      ** But not if we are in secure-delete mode. In secure-delete mode,
      ** the dropCell() routine will overwrite the entire cell with zeroes.
      ** In this case, temporarily copy the cell into the aOvflSpace[]
      ** buffer. It will be copied out again as soon as the aSpace[] buffer
      ** is allocated.  */
      if( pBt->btsFlags & BTS_SECURE_DELETE ){
      if( pBt->btsFlags & BTS_FAST_SECURE ){
        int iOff;

        iOff = SQLITE_PTR_TO_INT(apDiv[i]) - SQLITE_PTR_TO_INT(pParent->aData);
        if( (iOff+szNew[i])>(int)pBt->usableSize ){
          rc = SQLITE_CORRUPT_BKPT;
          memset(apOld, 0, (i+1)*sizeof(MemPage*));
          goto balance_cleanup;

/*
** Allowed values for BtShared.btsFlags
*/
#define BTS_READ_ONLY        0x0001   /* Underlying file is readonly */
#define BTS_PAGESIZE_FIXED   0x0002   /* Page size can no longer be changed */
#define BTS_SECURE_DELETE    0x0004   /* PRAGMA secure_delete is enabled */
#define BTS_OVERWRITE        0x0008   /* Overwrite deleted content with zeros */
#define BTS_FAST_SECURE      0x000c   /* Combination of the previous two */
#define BTS_INITIALLY_EMPTY  0x0008   /* Database was empty at trans start */
#define BTS_NO_WAL           0x0010   /* Do not open write-ahead-log files */
#define BTS_EXCLUSIVE        0x0020   /* pWriter has an exclusive lock */
#define BTS_PENDING          0x0040   /* Waiting for read-locks to clear */
#define BTS_INITIALLY_EMPTY  0x0010   /* Database was empty at trans start */
#define BTS_NO_WAL           0x0020   /* Do not open write-ahead-log files */
#define BTS_EXCLUSIVE        0x0040   /* pWriter has an exclusive lock */
#define BTS_PENDING          0x0080   /* Waiting for read-locks to clear */

/*
** An instance of the following structure is used to hold information
** about a cell.  The parseCellPtr() function fills in this structure
** based on information extract from the raw disk page.
*/
struct CellInfo {

    0,                            /* xUpdate */
    0,                            /* xBegin */
    0,                            /* xSync */
    0,                            /* xCommit */
    0,                            /* xRollback */
    0,                            /* xFindMethod */
    0,                            /* xRename */
    0,                            /* xSavepoint */
    0,                            /* xRelease */
    0,                            /* xRollbackTo */
  };
  return sqlite3_create_module(db, "dbstat", &dbstat_module, 0);
}
#elif defined(SQLITE_ENABLE_DBSTAT_VTAB)
int sqlite3DbstatRegister(sqlite3 *db){ return SQLITE_OK; }
#endif /* SQLITE_ENABLE_DBSTAT_VTAB */

# define sqlite3_complete16             0
# define sqlite3_create_collation16     0
# define sqlite3_create_function16      0
# define sqlite3_errmsg16               0
# define sqlite3_open16                 0
# define sqlite3_prepare16              0
# define sqlite3_prepare16_v2           0
# define sqlite3_prepare16_v3           0
# define sqlite3_result_error16         0
# define sqlite3_result_text16          0
# define sqlite3_result_text16be        0
# define sqlite3_result_text16le        0
# define sqlite3_value_text16           0
# define sqlite3_value_text16be         0
# define sqlite3_value_text16le         0

  sqlite3_db_cacheflush,
  /* Version 3.12.0 and later */
  sqlite3_system_errno,
  /* Version 3.14.0 and later */
  sqlite3_trace_v2,
  sqlite3_expanded_sql,
  /* Version 3.18.0 and later */
  sqlite3_set_last_insert_rowid
  sqlite3_set_last_insert_rowid,
  /* Version 3.20.0 and later */
  sqlite3_prepare_v3,
  sqlite3_prepare16_v3
};

/*
** Attempt to load an SQLite extension library contained in the file
** zFile.  The entry point is zProc.  zProc may be 0 in which case a
** default entry point name (sqlite3_extension_init) is used.  Use
** of the default name is recommended.

/*
** Return a static string that describes the kind of error specified in the
** argument.
*/
const char *sqlite3ErrStr(int rc){
  static const char* const aMsg[] = {
    /* SQLITE_OK          */ "not an error",
    /* SQLITE_ERROR       */ "SQL logic error or missing database",
    /* SQLITE_ERROR       */ "SQL logic error",
    /* SQLITE_INTERNAL    */ 0,
    /* SQLITE_PERM        */ "access permission denied",
    /* SQLITE_ABORT       */ "callback requested query abort",
    /* SQLITE_ABORT       */ "query aborted",
    /* SQLITE_BUSY        */ "database is locked",
    /* SQLITE_LOCKED      */ "database table is locked",
    /* SQLITE_NOMEM       */ "out of memory",
    /* SQLITE_READONLY    */ "attempt to write a readonly database",
    /* SQLITE_INTERRUPT   */ "interrupted",
    /* SQLITE_IOERR       */ "disk I/O error",
    /* SQLITE_CORRUPT     */ "database disk image is malformed",
    /* SQLITE_NOTFOUND    */ "unknown operation",
    /* SQLITE_FULL        */ "database or disk is full",
    /* SQLITE_CANTOPEN    */ "unable to open database file",
    /* SQLITE_PROTOCOL    */ "locking protocol",
    /* SQLITE_EMPTY       */ "table contains no data",
    /* SQLITE_EMPTY       */ 0,
    /* SQLITE_SCHEMA      */ "database schema has changed",
    /* SQLITE_TOOBIG      */ "string or blob too big",
    /* SQLITE_CONSTRAINT  */ "constraint failed",
    /* SQLITE_MISMATCH    */ "datatype mismatch",
    /* SQLITE_MISUSE      */ "library routine called out of sequence",
    /* SQLITE_MISUSE      */ "bad parameter or other API misuse",
#ifdef SQLITE_DISABLE_LFS
    /* SQLITE_NOLFS       */ "large file support is disabled",
#else
    /* SQLITE_NOLFS       */ 0,
#endif
    /* SQLITE_AUTH        */ "authorization denied",
    /* SQLITE_FORMAT      */ "auxiliary database format error",
    /* SQLITE_RANGE       */ "bind or column index out of range",
    /* SQLITE_NOTADB      */ "file is encrypted or is not a database",
    /* SQLITE_FORMAT      */ 0,
    /* SQLITE_RANGE       */ "column index out of range",
    /* SQLITE_NOTADB      */ "file is not a database",
  };
  const char *zErr = "unknown error";
  switch( rc ){
    case SQLITE_ABORT_ROLLBACK: {
      zErr = "abort due to ROLLBACK";
      break;
    }

** error.
*/
const void *sqlite3_errmsg16(sqlite3 *db){
  static const u16 outOfMem[] = {
    'o', 'u', 't', ' ', 'o', 'f', ' ', 'm', 'e', 'm', 'o', 'r', 'y', 0
  };
  static const u16 misuse[] = {
    'l', 'i', 'b', 'r', 'a', 'r', 'y', ' ', 
    'b', 'a', 'd', ' ', 'p', 'a', 'r', 'a', 'm', 'e', 't', 'e', 'r', ' ',
    'r', 'o', 'u', 't', 'i', 'n', 'e', ' ', 
    'c', 'a', 'l', 'l', 'e', 'd', ' ', 
    'o', 'u', 't', ' ', 
    'o', 'f', ' ', 
    'o', 'r', ' ', 'o', 't', 'h', 'e', 'r', ' ', 'A', 'P', 'I', ' ',
    'm', 'i', 's', 'u', 's', 'e', 0
    's', 'e', 'q', 'u', 'e', 'n', 'c', 'e', 0
  };

  const void *z;
  if( !db ){
    return (void *)outOfMem;
  }
  if( !sqlite3SafetyCheckSickOrOk(db) ){

#endif
  *ppDb = 0;
#ifndef SQLITE_OMIT_AUTOINIT
  rc = sqlite3_initialize();
  if( rc ) return rc;
#endif

  /* Only allow sensible combinations of bits in the flags argument.  
  ** Throw an error if any non-sense combination is used.  If we
  ** do not block illegal combinations here, it could trigger
  ** assert() statements in deeper layers.  Sensible combinations
  ** are:
  **
  **  1:  SQLITE_OPEN_READONLY
  **  2:  SQLITE_OPEN_READWRITE
  **  6:  SQLITE_OPEN_READWRITE | SQLITE_OPEN_CREATE
  */
  assert( SQLITE_OPEN_READONLY  == 0x01 );
  assert( SQLITE_OPEN_READWRITE == 0x02 );
  assert( SQLITE_OPEN_CREATE    == 0x04 );
  testcase( (1<<(flags&7))==0x02 ); /* READONLY */
  testcase( (1<<(flags&7))==0x04 ); /* READWRITE */
  testcase( (1<<(flags&7))==0x40 ); /* READWRITE | CREATE */
  if( ((1<<(flags&7)) & 0x46)==0 ){
    return SQLITE_MISUSE_BKPT;  /* IMP: R-65497-44594 */
  }

  if( sqlite3GlobalConfig.bCoreMutex==0 ){
    isThreadsafe = 0;
  }else if( flags & SQLITE_OPEN_NOMUTEX ){
    isThreadsafe = 0;
  }else if( flags & SQLITE_OPEN_FULLMUTEX ){
    isThreadsafe = 1;
  }else{

  }
  /* EVIDENCE-OF: R-08308-17224 The default collating function for all
  ** strings is BINARY. 
  */
  db->pDfltColl = sqlite3FindCollSeq(db, SQLITE_UTF8, sqlite3StrBINARY, 0);
  assert( db->pDfltColl!=0 );

  /* Parse the filename/URI argument. */
  /* Parse the filename/URI argument
  **
  ** Only allow sensible combinations of bits in the flags argument.  
  ** Throw an error if any non-sense combination is used.  If we
  ** do not block illegal combinations here, it could trigger
  ** assert() statements in deeper layers.  Sensible combinations
  ** are:
  **
  **  1:  SQLITE_OPEN_READONLY
  **  2:  SQLITE_OPEN_READWRITE
  **  6:  SQLITE_OPEN_READWRITE | SQLITE_OPEN_CREATE
  */
  db->openFlags = flags;
  assert( SQLITE_OPEN_READONLY  == 0x01 );
  assert( SQLITE_OPEN_READWRITE == 0x02 );
  assert( SQLITE_OPEN_CREATE    == 0x04 );
  testcase( (1<<(flags&7))==0x02 ); /* READONLY */
  testcase( (1<<(flags&7))==0x04 ); /* READWRITE */
  testcase( (1<<(flags&7))==0x40 ); /* READWRITE | CREATE */
  if( ((1<<(flags&7)) & 0x46)==0 ){
    rc = SQLITE_MISUSE_BKPT;  /* IMP: R-65497-44594 */
  }else{
  rc = sqlite3ParseUri(zVfs, zFilename, &flags, &db->pVfs, &zOpen, &zErrMsg);
    rc = sqlite3ParseUri(zVfs, zFilename, &flags, &db->pVfs, &zOpen, &zErrMsg);
  }
  if( rc!=SQLITE_OK ){
    if( rc==SQLITE_NOMEM ) sqlite3OomFault(db);
    sqlite3ErrorWithMsg(db, rc, zErrMsg ? "%s" : 0, zErrMsg);
    sqlite3_free(zErrMsg);
    goto opendb_out;
  }

    pOut->zEnd = &t.z[t.n];
  }
}

expr(A) ::= term(A).
expr(A) ::= LP(B) expr(X) RP(E).
            {spanSet(&A,&B,&E); /*A-overwrites-B*/  A.pExpr = X.pExpr;}
term(A) ::= NULL(X).        {spanExpr(&A,pParse,@X,X);/*A-overwrites-X*/}
expr(A) ::= id(X).          {spanExpr(&A,pParse,TK_ID,X); /*A-overwrites-X*/}
expr(A) ::= JOIN_KW(X).     {spanExpr(&A,pParse,TK_ID,X); /*A-overwrites-X*/}
expr(A) ::= nm(X) DOT nm(Y). {
  Expr *temp1 = sqlite3ExprAlloc(pParse->db, TK_ID, &X, 1);
  Expr *temp2 = sqlite3ExprAlloc(pParse->db, TK_ID, &Y, 1);
  spanSet(&A,&X,&Y); /*A-overwrites-X*/
  A.pExpr = sqlite3PExpr(pParse, TK_DOT, temp1, temp2);
}
expr(A) ::= nm(X) DOT nm(Y) DOT nm(Z). {
  Expr *temp1 = sqlite3ExprAlloc(pParse->db, TK_ID, &X, 1);
  Expr *temp2 = sqlite3ExprAlloc(pParse->db, TK_ID, &Y, 1);
  Expr *temp3 = sqlite3ExprAlloc(pParse->db, TK_ID, &Z, 1);
  Expr *temp4 = sqlite3PExpr(pParse, TK_DOT, temp2, temp3);
  spanSet(&A,&X,&Z); /*A-overwrites-X*/
  A.pExpr = sqlite3PExpr(pParse, TK_DOT, temp1, temp4);
}
term(A) ::= FLOAT|BLOB(X). {spanExpr(&A,pParse,@X,X);/*A-overwrites-X*/}
term(A) ::= STRING(X).     {spanExpr(&A,pParse,@X,X);/*A-overwrites-X*/}
term(A) ::= NULL|FLOAT|BLOB(X). {spanExpr(&A,pParse,@X,X); /*A-overwrites-X*/}
term(A) ::= STRING(X).          {spanExpr(&A,pParse,@X,X); /*A-overwrites-X*/}
term(A) ::= INTEGER(X). {
  A.pExpr = sqlite3ExprAlloc(pParse->db, TK_INTEGER, &X, 1);
  A.zStart = X.z;
  A.zEnd = X.z + X.n;
  if( A.pExpr ) A.pExpr->flags |= EP_Leaf|EP_Resolved;
}
expr(A) ::= VARIABLE(X).     {
  if( !(X.z[0]=='#' && sqlite3Isdigit(X.z[1])) ){
    u32 n = X.n;
    spanExpr(&A, pParse, TK_VARIABLE, X);
    sqlite3ExprAssignVarNumber(pParse, A.pExpr, n);
  }else{