1.0.6:

* fix a problem when compiling for Oracle 8.0.
  (reported by Axel Reinhold as a ruby-oci8 2.0 issue)

* [dbi] fix to pass a newly added sanity check in dbi 0.4.1.
  (reported by Dirk Herzhauser as a ruby-oci8 2.0 issue)

* [dbi] fix dbh#columns for Oracle 8.1 or lower.
  The data dictionary all_constraints doesn't have index_name
  column on Oracle 8i. Rewrite DBI::DBD::OCI8::Database#columns
  by using all_cons_columns.
  (backport from ruby-oci8 trunk)

1.0.5:

No changes except who try to install on Mac OS X ppc.

- fix big/little endian checking problem on Mac OS X ppc.
  (contributed by unknown. See: Bug ID 24284 on rubyforge.)

1.0.4:

1. [dbi] support ruby-dbi 0.4
   (Thanks to James Cao and Peter Zingg.)

   Ruby-dbi 0.4 is supported by DBD::OCI8 after 6 months from the
   release of 0.4.

   If you are using ruby-dbi 0.1.1, you need to copy dbd/OCI8.rb
   to DBD/OCI8/OCI8.rb. The file itself works fine with ruby-dbi
   0.1.1, but the dbi cannot find the file by default.

2. Rowid column data are fetched as a String by default.

3. The return type of OCI8::Cursor#rowid is changed from OCIRowid
   to String.

   Note: It needs one network roundtrip to convert OCIRowid to
   String on server side when using Oracle 8.x client.
   On Oracle 9.0 or upper it is converted in the client side.

4. An exception is raised if a connection is used in a process which is
   different with a process to establish the connection.

   If you encounter an exception "The connection cannot be reused in the
   forked process," see the following link.

     http://groups.google.com/group/phusion-passenger/browse_thread/thread/b00d4770bc1d00b8

5. fix for 11.1.0.7.0 instant rpm package.
   (reported by Luis Parravicini.)

   The directory tree is a bit different with that of 11.1.0.6.0
   or earlier.

6. suppress compilation warnings.
   (contributed by Daniel Berger)

1.0.3:

1. add workaround code for a losing character problem when reading CLOB.
   (reported by Efren Yevale and Thomas Witt and fixed with Thomas Witt's help.)

   The problem is happened at the following condition.

   (a) Oracle client version is 10.2.0.4 or 11.1.0.6.
       It doesn't happend when using 10.2.0.3 client or lower.
       It doesn't depend on Oracle server version.

   (b) The character set is a variable-length one.
       e.g. AL32UTF8

2. fix a problem when reading BLOB/CLOB over 1GB.
   (reported by Jonathan Hadders.)

3. [rails] fix a problem that OraNumber#to_json returns "{}" when using Rails.
   (reported by Alex Moore)

   Rails add to_json method. But it doesn't know what OraNumber is.
   OraNumber#to_json is added for Rails.

4. [dbi] fix a problem when using Oracle 8i and ruby-dbi.
   (reported by Glauco Magnelli.)

1.0.2:

1. add a gemspec file.
   (Thanks to Jesse Hu)

   To make a pure ruby gems package:
     gem build ruby-oci8.gemspec

   To make a binary gems package:
     gem build ruby-oci8.gemspec -- current


2. [BUG FIX] fix a segv bug in gc mark.
   (Thanks to authorNari)

1.0.1:

1. [SPEC CHANGE] OCI8#exec and OCI8::Cursor#exec's return value is changed
   for statement except select, insert, update and delete statement.

   It had been true. But now it is the number of rows processed for all
   non-select statements.

   For example:

     # 1.0.0 or before
     conn = OCI8.new('user/pass')
     conn.exec('rename FOO_TABLE to BAR_TABLE') # => true

     # 1.0.1
     conn = OCI8.new('user/pass')
     conn.exec('rename FOO_TABLE to BAR_TABLE') # => 0

2. fix a bug that OraDate.new made a core dump on x86_64 linux.
  (reported by Alun Eyre)

3. fix OCI8#non_blocking = false problem.
  Once the connection became non-bocking mode, it could
  not be reset to blocking mode. Now it can be reset.
  (reported by Cagdas Gerede)

4. support cursors in a result set.
   (contributed by Randy Gordon)

   For example:
      conn = OCI8.new('ruby/oci8')
      sql = 'SELECT CURSOR(SELECT * FROM foo where foo.c1 = bar.c1 ) FROM bar'
      conn.exec(sql) do |row|
        cursor_in_result_set = row[0]
        cursor_in_result_set.fetch do |row|
          puts row.join(',')
        end
      end

5. fix oraconf.rb for official x86_64 linux rpms.
   (contributed by Pat)

1.0.0:

1. fix a BUG when binding a Bignum. This bug was added in
   ruby-oci8 1.0.0-rc2. (reported by Daniel Berger)

2. add OCI8#describe_table(table_name), OCI8::Cursor#column_metadata,
   OCI8::Metadata::Table, OCI8::Metadata::View and OCI8::Metadata::Column.

   table information:
     con = OCI8.new('username/password')
     table_info = con.describe_table('table_name') #=> OCI8::Metadata::Table
     table_info.columns #=> an array of OCI8::Metadata::Column.

   query information:
     con = OCI8.new('username/password')
     csr = con.exec('select * from table_name')
     csr.column_metadata #=> an array of OCI8::Metadata::Column.

5. add OCI8::BindType::BinaryDouble if oci8lib.so is compiled for
   Oracle 10g client. You can fetch exactly same values in Oracle.

   You could fetch the values in previous version by adding the
   following line.

     OCI8::BindType::Mapping[101] = OCI8::BindType::Float
     (This is default in ruby-oci8 1.0.0-rc2.)

   But a fetched value may not be exactly same with the value in
   Oracle. That's because OCI8::BindType::Float's network
   representation is Oracle NUMBER data type. A BINARY_DOUBLE
   value in Oracle become NUMBER on the network and is converted
   to Float on the client side. It makes a bit difference by
   round-off error.

   In addition, Oracle NUMBER doesn't have NaN, +Infinity and -Infinity.
   You cannot fetch the values by older versions.

   Note: When you bind a Float value, it is bound by
   OCI8::BindType::Float by default. You have two ways to bind
   it by OCI8::BindType::BinaryDouble.

   a. bind explicitly.

      con = OCI8.new('ruby/oci8')
      csr = con.parse('insert into table_name values (:1)')
      csr.bind_param(1, 10.5, OCI8::SQLT_IBDOUBLE)
      csr.exec

   b. change the default behaviour if and only if you are sure that
      the Oracle client version and Oracle server version are both
      10g or upper.

      # use OCI8::BindType::BinaryDouble for ruby's Float values.
      OCI8::BindType::Mapping[Float] = OCI8::BindType::BinaryDouble
      con = OCI8.new('ruby/oci8')
      csr = con.exec('insert into table_name values (:1)', 10.5)

4. OCI8::Cursor#fetch_hash accepts a block as a iterator.

5. [dbi] support external OS authentication.
   (contributed by Dan Fitch)

     dbh = DBI.connect('dbi:OCI8', nil, nil)

6. [dbi] add DatabaseHandle#columns and improve StatementHandle#column_info
   (suggested by Venkat Pa)

7. [bug] fix a problem when binding ''(empty string) via dbi.

1.0.0-rc3:

1. support external OS authentication.
   (suggested by Dan Fitch)

      conn = OCI8.new('/')
   or
      conn = OCI8.new(nil, nil)

2. incompatible change on connect string.

   You need to pass only one argument to OCI8.new to parse the
   first argument as a SQL*Plus-like connect string.

   For example, OCI8.new('scott/tiger', nil) was parsed as the
   username is 'scott' and the password is 'tiger' in 1.0.0-rc2.
   But from now on, it is treated as the username is 'scott/tiger'
   and the password is nil because it has two arguments. You need
   to use OCI8.new('scott/tiger') instead.

3. fix NCLOB support on Oracle 9i.
   (reported by Matthew Noble)

   1.0.0-rc2's NCLOB support worked on Oracle 10g only.

4. fix a problem to compile with a third-party x86_64 rpm package.
   (contributed by Jeff Abbott)

   The x86_64 source rpm packages are found on the following URLs.
     http://freshrpms.net/docs/oracle/
     http://remi.collet.free.fr/index.php?2007/01/22/279-oracle-instant-client-10203-on-x86_64

5. '--with-instant-client' option become obsolete and is ignored.

   Use LD_LIBRARY_PATH(Linux), PATH(Windows), etc. to indicate
   the instant client's location.

6. fix some other minor problems.

1.0.0-rc2:

1. support SQL*Plus-like connect strings.
   (contributed by Oracle Corporation)

     require 'oci8'
     conn = OCI8.new('username/password@tns_name')
     # instead of OCI8.new('username', 'password', 'tns_name')

2. Add a rpm spec file.
   (contributed by Brian Candler)

     rpmbuild -tb ruby-oci8-x.y.z.tar.gz

     /usr/src/redhat/RPMS/i386/ruby-oci8-x.y.z-1.i386.rpm

3. support BINARY_FLOAT and BINADY_DOUBLE columns.
   (This problem is reported by Duncan McCaffery.)

4. support NCLOB columns.
   (This problem is reported by Dmitry Zadvornykh.)

     cursor = conn.exec('select nclob_column from nclob_table')
     cursor.fetch # => [ instance of OCI8::NCLOB ]

5. add new CLOB/NCLOB/BLOB insert method.
   (This is available on Oracle 8.1.5 or later.)

     cursor = conn.parse("INSERT INTO clob_table VALUES (:1)")
     cursor.bind_param(1, OCI8::CLOB.new(conn, lob_data))
     cursor.exec

     cursor = conn.parse("INSERT INTO nclob_table VALUES (:1)")
     cursor.exec(OCI8::NCLOB.new(conn, lob_data))

     conn.exec("INSERT INTO blob_table VALUES (:1)",
               OCI8::BLOB.new(conn, lob_data))

   This is added to insert CLOB as a SDO_GEOMETRY's parameter
   to solve a problem reported by Pete Schwamb.

6. add an error location in C code to OCIException's backtrace.
   This is for debug information.

1.0.0-rc1:

1. support BFILE columns.

2. check whether Full Client or Instant Client by checking load library
   path on configuration to compile.

3. raise an OCIError with "OCI Library Initialization Error" message if
   no message can not be retrieved when requiring oci8lib.so.

4. raise "OCI Library Initialization Error" when first connect instead
   of requiring oci8lib.so. This is a workaround to prevent Rails from
   being stopped on invalid Oracle environment.

0.1.16:

1. Add LONG and LONG RAW support longer than 65535 bytes.

   Use OCI8#long_read_len and OCI8#long_read_len= to set max long length.
   The default value is 65535.

2. OraDate and OraNumber can be serialized by Marshal now.

3. OraDate and OraNumber can be serialized by YAML now.

4. Fix bug on OraDate#dup, OraDate#clone, OraNumber#dup and OraNumber#clone.
   (reported by Daniel Berger.)

   These methods had not duplicated their contents, which caused segmentation
   fault. Now they duplicate their contents also by adding initialize_copy
   method.

5. Add a workaround code on cygwin environment variables problem.

   Cygwin manages environment variables by itself. They don't synchroize
   with Win32's ones. When setting Oracle's environment variables, they
   are set to cygwin's environment. But Oracle client library read
   Win32's. Now Oracle-related variables are forcedly set to Win32.

6. Accept all Fixnum values when binding as a Fixnum on 64-bit ruby.

   A Fixnum is a 31-bit integer on 32-bit ruby, 63-bit integer on 64-bit
   ruby. But it had been bound to 32-bit integer on ruby-oci8. Now it is
   bound to integer whose bit width is same with ruby.

0.1.15:

1. no new features. :-<

2. suppress memory usage when binding a placeholder which is already bound.

     require 'oci8'
     conn = OCI8.new('ruby', 'oci8')
     cursor = conn.parse('select * from foo where key = :key')
     0.upto(10000) do |key|
       # bind values had not freed until cursor.close.
       # Now they are freed when GC is run.
       cursor.bind_param('key', key)
       cursor.exec
       .... fetch ....
     end

3. fix potentially segmentation fault while exiting process.
   When exiting process, ruby's objects are GCed with unexpectable
   order. If an OCIEnv instance was GCed in prior to OCIRowid objects,
   it caused segmentation fault. 

4. add a workaround code when setting $0.
   The following code made enviroment variables empty strings on
   Unix when using ruby 1.8.4, ruby-oci8 0.1.14 and Oracle 10.1.0.
   (Oracle 10.2.0 and ruby 1.8.2 doesn't have this problem.)

     require 'oci8'
     ENV['HOME'] # --> valid value
     $0 = 'foo'
     ENV['HOME'] # --> ''

5. fix registry search path on compilation.
   (OracleXE Windows version)

0.1.14:

1. add OCI8#prefetch_rows=.
    
   This specifies the number of rows in a single round trip.
   See: http://www.easysoft.com/products/data_access/odbc_oracle_driver/performance_tuning_guide.html#multiple_rows 
   In addition, setting zero may become a workaround
   of some Oracle internal problems.
   The default value is one.

2. fix bug when binding a Bignum.
3. (low-level API) add attrSet(OCI_ATTR_DESC_PUBLIC, -1).
4. (low-level API) add a constant OCI_PTYPE_UNK.
5. (internal) use OCIEnv.logon when no priviledge login.
6. (internal) delete code to register malloc/realloc/free callbacks to OCI.

0.1.13:

1. fix a memory leak.
   (fixed by Nathaniel Talbott.)

   It is recommended to update to 0.1.13 if a program executes many
   queries. That's bacause 0.1.12 cause 70-byte memory leak per a
   query.

2. ruby's DateTime support is added.
   (suggested by Daniel Berger.)

   To fetch Oracle's DATE datatype as ruby's DateTime by default, add
   the following code to your program.

     OCI8::BindType::Mapping[OCI8::SQLT_DAT] = OCI8::BindType::DateTime
     OCI8::BindType::Mapping[OCI8::SQLT_TIMESTAMP] = OCI8::BindType::DateTime