SQLite

      ** are no outstanding references to any page other than page 1. And
      ** page 1 is never written to the log until the transaction is
      ** committed. As a result, the call to xUndo may not fail.
      */
      assert( walFramePgno(pWal, iFrame)!=1 );
      rc = xUndo(pUndoCtx, walFramePgno(pWal, iFrame));
    }
    if( iMax!=pWal->hdr.mxFrame ) walCleanupHash(pWal);
  }
  assert( rc==SQLITE_OK );
  return rc;
}

/* 
** Argument aWalData must point to an array of WAL_SAVEPOINT_NDATA u32 

# 2012 October 15
#
# 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 test case tests that a problem causing a failing assert() has
# been fixed. The problem occurred if a writer process with a subset
# of the *shm file mapped rolled back a transaction begun after the
# entire WAL file was checkpointed into the db file (i.e. a transaction
# that would have restarted the WAL file from the beginning).
#

set testdir [file dirname $argv0]
source $testdir/tester.tcl
set testprefix wal9

sqlite3 db2 test.db

do_execsql_test 1.0 {
  PRAGMA page_size = 1024;
  PRAGMA journal_mode = WAL;
  PRAGMA wal_autocheckpoint = 0;
  CREATE TABLE t(x);
} {wal 0}

do_test 1.1 { 
  execsql "SELECT * FROM t" db2
} {}

do_execsql_test 1.2 {
  BEGIN;
    INSERT INTO t VALUES(randomblob(100));
    INSERT INTO t SELECT randomblob(100) FROM t;
    INSERT INTO t SELECT randomblob(100) FROM t;
    INSERT INTO t SELECT randomblob(100) FROM t;
    INSERT INTO t SELECT randomblob(100) FROM t;
    INSERT INTO t SELECT randomblob(100) FROM t;
    INSERT INTO t SELECT randomblob(100) FROM t;
    INSERT INTO t SELECT randomblob(100) FROM t;

    INSERT INTO t SELECT randomblob(100) FROM t;
    INSERT INTO t SELECT randomblob(100) FROM t;
    INSERT INTO t SELECT randomblob(100) FROM t;
    INSERT INTO t SELECT randomblob(100) FROM t;
    INSERT INTO t SELECT randomblob(100) FROM t;
    INSERT INTO t SELECT randomblob(100) FROM t;
    INSERT INTO t SELECT randomblob(100) FROM t;
    INSERT INTO t SELECT randomblob(100) FROM t;

    INSERT INTO t SELECT randomblob(100) FROM t;
    INSERT INTO t SELECT randomblob(100) FROM t;
  COMMIT;
} {}

# Check file sizes are as expected. The real requirement here is that 
# the *shm file is now more than one chunk (>32KiB).
do_test 1.3 { file size test.db     } {1024}
do_test 1.4 { file size test.db-wal } {15421352}
do_test 1.5 { file size test.db-shm } {131072}

do_execsql_test 1.6 { PRAGMA wal_checkpoint } {0 14715 14715}

# At this point connection [db2] has mapped the first 32KB of the *shm file
# only. Because the entire WAL file has been checkpointed, it is not 
# necessary to map any more of the *-shm file to read or write the database
# (since all data will be read directly from the db file). 
#
# However, at one point if a transaction that had not yet written to the 
# WAL file was rolled back an assert() attempting to verify that the entire 
# *-shm file was mapped would fail. If NDEBUG was defined (and the assert() 
# disabled) this bug caused SQLite to ignore the return code of a mmap() 
# call.
#
do_test 1.7 {
  execsql { 
    BEGIN;
      INSERT INTO t VALUES('hello');
    ROLLBACK;
  } db2
} {}
db2 close

finish_test