test/shared.test

# 2005 December 30
#
# 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 regression tests for SQLite library.  The
# focus of this file is testing the SELECT statement.
#
# $Id: shared.test,v 1.6 2006/01/07 13:21:04 danielk1977 Exp $

set testdir [file dirname $argv0]
source $testdir/tester.tcl
db close

ifcapable !shared_cache {
  finish_test
  return
}
set ::enable_shared_cache [sqlite3_enable_shared_cache 1]

# Test organization:
#
# shared-1.*: Simple test to verify basic sanity of table level locking when
#             two connections share a pager cache.
# shared-2.*: Test that a read transaction can co-exist with a 
#             write-transaction, including a simple test to ensure the 
#             external locking protocol is still working.
# shared-3.*: Simple test of read-uncommitted mode.
# shared-4.*: Check that the schema is locked and unlocked correctly.
# shared-5.*: Test that creating/dropping schema items works when databases
#             are attached in different orders to different handles.
# shared-6.*: Locking, UNION ALL queries and sub-queries.
#

do_test shared-1.1 {
  # Open a second database on the file test.db. It should use the same pager
  # cache and schema as the original connection. Verify that only 1 file is 
  # opened.
  sqlite3 db2 test.db
  sqlite3 db  test.db
  set ::sqlite_open_file_count
} {1}
do_test shared-1.2 {
  # Add a table and a single row of data via the first connection. 
  # Ensure that the second connection can see them.
  execsql {
    CREATE TABLE abc(a, b, c);
    INSERT INTO abc VALUES(1, 2, 3);
  } db
  execsql {
    SELECT * FROM abc;
  } db2
} {1 2 3}
do_test shared-1.3 {
  # Have the first connection begin a transaction and obtain a read-lock
  # on table abc. This should not prevent the second connection from 
  # querying abc.
  execsql {
    BEGIN;
    SELECT * FROM abc;
  }
  execsql {
    SELECT * FROM abc;
  } db2
} {1 2 3}
do_test shared-1.4 {
  # Try to insert a row into abc via connection 2. This should fail because
  # of the read-lock connection 1 is holding on table abc (obtained in the
  # previous test case).
  catchsql {
    INSERT INTO abc VALUES(4, 5, 6);
  } db2
} {1 {database table is locked: abc}}
do_test shared-1.5 {
  # Using connection 2 (the one without the open transaction), try to create
  # a new table. This should fail because of the open read transaction 
  # held by connection 1.
  catchsql {
    CREATE TABLE def(d, e, f);
  } db2
} {1 {database table is locked: sqlite_master}}
do_test shared-1.6 {
  # Upgrade connection 1's transaction to a write transaction. Create
  # a new table - def - and insert a row into it. Because the connection 1
  # transaction modifies the schema, it should not be possible for 
  # connection 2 to access the database at all until the connection 1 
  # has finished the transaction.
  execsql {
    CREATE TABLE def(d, e, f);
    INSERT INTO def VALUES('IV', 'V', 'VI');
  }
} {}
do_test shared-1.7 {
  # Read from the sqlite_master table with connection 1 (inside the 
  # transaction). Then test that we can not do this with connection 2. This
  # is because of the schema-modified lock established by connection 1 
  # in the previous test case.
  execsql {
    SELECT * FROM sqlite_master;
  }
  catchsql {
    SELECT * FROM sqlite_master;
  } db2
} {1 {database schema is locked: main}}
do_test shared-1.8 {
  # Commit the connection 1 transaction.
  execsql {
    COMMIT;
  }
} {}

do_test shared-2.1 {
  # Open connection db3 to the database. Use a different path to the same
  # file so that db3 does *not* share the same pager cache as db and db2
  # (there should be two open file handles).
  sqlite3 db3 ./test.db
  set ::sqlite_open_file_count
} {2}
do_test shared-2.2 {
  # Start read transactions on db and db2 (the shared pager cache). Ensure
  # db3 cannot write to the database.
  execsql {
    BEGIN;
    SELECT * FROM abc;
  }
  execsql {
    BEGIN;
    SELECT * FROM abc;
  } db2
  catchsql {
    INSERT INTO abc VALUES(1, 2, 3);
  } db2
} {1 {database table is locked: abc}}
do_test shared-2.3 {
  # Turn db's transaction into a write-transaction. db3 should still be
  # able to read from table def (but will not see the new row). Connection
  # db2 should not be able to read def (because of the write-lock).

# Todo: The failed "INSERT INTO abc ..." statement in the above test
# has started a write-transaction on db2 (should this be so?). This 
# would prevent connection db from starting a write-transaction. So roll the
# db2 transaction back and replace it with a new read transaction.
  execsql {
    ROLLBACK;
    BEGIN;
    SELECT * FROM abc;
  } db2

  execsql {
    INSERT INTO def VALUES('VII', 'VIII', 'IX');
  }
  concat [
    catchsql { SELECT * FROM def; } db3
  ] [
    catchsql { SELECT * FROM def; } db2
  ]
} {0 {IV V VI} 1 {database table is locked: def}}
do_test shared-2.4 {
  # Commit the open transaction on db. db2 still holds a read-transaction.
  # This should prevent db3 from writing to the database, but not from 
  # reading.
  execsql {
    COMMIT;
  }
  concat [
    catchsql { SELECT * FROM def; } db3
  ] [
    catchsql { INSERT INTO def VALUES('X', 'XI', 'XII'); } db3
  ]
} {0 {IV V VI VII VIII IX} 1 {database is locked}}

catchsql COMMIT db2

do_test shared-3.1.1 {
  # This test case starts a linear scan of table 'seq' using a 
  # read-uncommitted connection. In the middle of the scan, rows are added
  # to the end of the seq table (ahead of the current cursor position).
  # The uncommitted rows should be included in the results of the scan.
  execsql "
    CREATE TABLE seq(i, x);
    INSERT INTO seq VALUES(1, '[string repeat X 500]');
    INSERT INTO seq VALUES(2, '[string repeat X 500]');
  "
  execsql {SELECT * FROM sqlite_master} db2
  execsql {PRAGMA read_uncommitted = 1} db2

  set ret [list]
  db2 eval {SELECT i FROM seq} {
    if {$i < 4} {
      execsql {
        INSERT INTO seq SELECT i + (SELECT max(i) FROM seq), x FROM seq;
      }
    }
    lappend ret $i
  }
  set ret
} {1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16}
do_test shared-3.1.2 {
  # Another linear scan through table seq using a read-uncommitted connection.
  # This time, delete each row as it is read. Should not affect the results of
  # the scan, but the table should be empty after the scan is concluded 
  # (test 3.1.3 verifies this).
  set ret [list]
  db2 eval {SELECT i FROM seq} {
    db eval {DELETE FROM seq WHERE i = $i}
    lappend ret $i
  }
  set ret
} {1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16}
do_test shared-3.1.3 {
  execsql {
    SELECT * FROM seq;
  }
} {}

catch {db close}
catch {db2 close}
catch {db3 close}

#--------------------------------------------------------------------------
# Tests shared-4.* test that the schema locking rules are applied 
# correctly. i.e.:
#
# 1. All transactions require a read-lock on the schemas of databases they
#    access.
# 2. Transactions that modify a database schema require a write-lock on that
#    schema.
# 3. It is not possible to compile a statement while another handle has a 
#    write-lock on the schema.
#

# Open two database handles db and db2. Each has a single attach database
# (as well as main):
#
#     db.main   ->   ./test.db
#     db.test2  ->   ./test2.db
#     db2.main  ->   ./test2.db
#     db2.test  ->   ./test.db
#
file delete -force test.db
file delete -force test2.db
file delete -force test2.db-journal
sqlite3 db  test.db
sqlite3 db2 test2.db
do_test shared-4.1.1 {
  set sqlite_open_file_count
} {2}
do_test shared-4.1.2 {
  execsql {ATTACH 'test2.db' AS test2}
  set sqlite_open_file_count
} {2}
do_test shared-4.1.3 {
  execsql {ATTACH 'test.db' AS test} db2
  set sqlite_open_file_count
} {2}

# Sanity check: Create a table in ./test.db via handle db, and test that handle
# db2 can "see" the new table immediately. A handle using a seperate pager
# cache would have to reload the database schema before this were possible.
#
do_test shared-4.2.1 {
  execsql {
    CREATE TABLE abc(a, b, c);
    CREATE TABLE def(d, e, f);
    INSERT INTO abc VALUES('i', 'ii', 'iii');
    INSERT INTO def VALUES('I', 'II', 'III');
  }
} {}
do_test shared-4.2.2 {
  execsql {
    SELECT * FROM test.abc;
  } db2
} {i ii iii}

# Open a read-transaction and read from table abc via handle 2. Check that
# handle 1 can read table abc. Check that handle 1 cannot modify table abc
# or the database schema. Then check that handle 1 can modify table def.
#
do_test shared-4.3.1 {
  execsql {
    BEGIN;
    SELECT * FROM test.abc;
  } db2
} {i ii iii}
do_test shared-4.3.2 {
  catchsql {
    INSERT INTO abc VALUES('iv', 'v', 'vi');
  }
} {1 {database table is locked: abc}}
do_test shared-4.3.3 {
  catchsql {
    CREATE TABLE ghi(g, h, i);
  }
} {1 {database table is locked: sqlite_master}}
do_test shared-4.3.3 {
  catchsql {
    INSERT INTO def VALUES('IV', 'V', 'VI');
  }
} {0 {}}
do_test shared-4.3.4 {
  # Cleanup: commit the transaction opened by db2.
  execsql {
    COMMIT
  } db2
} {}

# Open a write-transaction using handle 1 and modify the database schema.
# Then try to execute a compiled statement to read from the same 
# database via handle 2 (fails to get the lock on sqlite_master). Also
# try to compile a read of the same database using handle 2 (also fails).
# Finally, compile a read of the other database using handle 2. This
# should also fail.
#
do_test shared-4.4.1.2 {
  # Sanity check 1: Check that the schema is what we think it is when viewed
  # via handle 1.
  execsql {
    CREATE TABLE test2.ghi(g, h, i);
    SELECT 'test.db:'||name FROM sqlite_master 
    UNION ALL
    SELECT 'test2.db:'||name FROM test2.sqlite_master;
  }
} {test.db:abc test.db:def test2.db:ghi}
do_test shared-4.4.1.2 {
  # Sanity check 2: Check that the schema is what we think it is when viewed
  # via handle 2.
  execsql {
    SELECT 'test2.db:'||name FROM sqlite_master 
    UNION ALL
    SELECT 'test.db:'||name FROM test.sqlite_master;
  } db2
} {test2.db:ghi test.db:abc test.db:def}

do_test shared-4.4.2 {
  set ::DB2 [sqlite3_connection_pointer db2]
  set sql {SELECT * FROM abc}
  set ::STMT1 [sqlite3_prepare $::DB2 $sql -1 DUMMY]
  execsql {
    BEGIN;
    CREATE TABLE jkl(j, k, l);
  }
  sqlite3_step $::STMT1
} {SQLITE_ERROR}
do_test shared-4.4.3 {
  sqlite3_finalize $::STMT1
} {SQLITE_LOCKED}
do_test shared-4.4.4 {
  set rc [catch {
    set ::STMT1 [sqlite3_prepare $::DB2 $sql -1 DUMMY]
  } msg]
  list $rc $msg
} {1 {(6) database schema is locked: test}}
do_test shared-4.4.5 {
  set rc [catch {
    set ::STMT1 [sqlite3_prepare $::DB2 "SELECT * FROM ghi" -1 DUMMY]
  } msg]
  list $rc $msg
} {1 {(6) database schema is locked: test}}


catch {db2 close}
catch {db close}

#--------------------------------------------------------------------------
# Tests shared-5.* 
#
foreach db [list test.db test1.db test2.db test3.db] {
  file delete -force $db ${db}-journal
}
do_test shared-5.1.1 {
  sqlite3 db1 test.db
  sqlite3 db2 test.db
  execsql {
    ATTACH 'test1.db' AS test1;
    ATTACH 'test2.db' AS test2;
    ATTACH 'test3.db' AS test3;
  } db1
  execsql {
    ATTACH 'test3.db' AS test3;
    ATTACH 'test2.db' AS test2;
    ATTACH 'test1.db' AS test1;
  } db2
} {}
do_test shared-5.1.2 {
  execsql {
    CREATE TABLE test1.t1(a, b);
    CREATE INDEX test1.i1 ON t1(a, b);
    CREATE VIEW test1.v1 AS SELECT * FROM t1;
    CREATE TRIGGER test1.trig1 AFTER INSERT ON t1 BEGIN
      INSERT INTO t1 VALUES(new.a, new.b);
    END;
  } db1
  execsql {
    DROP INDEX i1;
    DROP VIEW v1;
    DROP TRIGGER trig1;
    DROP TABLE t1;
  } db2
} {}
do_test shared-5.1.2 {
  execsql {
    SELECT * FROM sqlite_master UNION ALL SELECT * FROM test1.sqlite_master
  } db1
} {}

#--------------------------------------------------------------------------
# Tests shared-6.* test that a query obtains all the read-locks it needs
# before starting execution of the query. This means that there is no chance
# some rows of data will be returned before a lock fails and SQLITE_LOCK
# is returned.
#
do_test shared-6.1.1 {
  execsql {
    CREATE TABLE t1(a, b);
    CREATE TABLE t2(a, b);
    INSERT INTO t1 VALUES(1, 2);
    INSERT INTO t2 VALUES(3, 4);
  } db1
  execsql {
    SELECT * FROM t1 UNION ALL SELECT * FROM t2;
  } db2
} {1 2 3 4}
do_test shared-6.1.2 {
  # Establish a write lock on table t2 via connection db2. Then make a 
  # UNION all query using connection db1 that first accesses t1, followed 
  # by t2. If the locks are grabbed at the start of the statement (as 
  # they should be), no rows are returned. If (as was previously the case)
  # they are grabbed as the tables are accessed, the t1 rows will be 
  # returned before the query fails.
  #
  execsql {
    BEGIN;
    INSERT INTO t2 VALUES(5, 6);
  } db2
  set ret [list]
  catch {
    db1 eval {SELECT * FROM t1 UNION ALL SELECT * FROM t2} {
      lappend ret $a $b
    }
  }
  set ret
} {}
do_test shared-6.1.3 {
  execsql {
    COMMIT;
    BEGIN;
    INSERT INTO t1 VALUES(7, 8);
  } db2
  set ret [list]
  catch {
    db1 eval {
      SELECT (CASE WHEN a>4 THEN (SELECT a FROM t1) ELSE 0 END) AS d FROM t2;
    } {
      lappend ret $d
    }
  }
  set ret
} {}

catch {db1 close}
catch {db2 close}

finish_test
sqlite3_enable_shared_cache $::enable_shared_cache