take more care in mangling SELECT when applying subquery limits

[dbsrgits/DBIx-Class.git] / lib / DBIx / Class / Storage / DBI / Oracle / Generic.pm

package DBIx::Class::Storage::DBI::Oracle::Generic;

use strict;
use warnings;
use base qw/DBIx::Class::Storage::DBI/;
use mro 'c3';
use DBIx::Class::Carp;
use Scope::Guard ();
use Context::Preserve 'preserve_context';
use Try::Tiny;
use List::Util 'first';
use namespace::clean;

__PACKAGE__->sql_limit_dialect ('RowNum');
__PACKAGE__->sql_quote_char ('"');
__PACKAGE__->sql_maker_class('DBIx::Class::SQLMaker::Oracle');
__PACKAGE__->datetime_parser_type('DateTime::Format::Oracle');

sub __cache_queries_with_max_lob_parts { 2 }

=head1 NAME

DBIx::Class::Storage::DBI::Oracle::Generic - Oracle Support for DBIx::Class

=head1 SYNOPSIS

  # In your result (table) classes
  use base 'DBIx::Class::Core';
  __PACKAGE__->add_columns({ id => { sequence => 'mysequence', auto_nextval => 1 } });
  __PACKAGE__->set_primary_key('id');

  # Somewhere in your Code
  # add some data to a table with a hierarchical relationship
  $schema->resultset('Person')->create ({
        firstname => 'foo',
        lastname => 'bar',
        children => [
            {
                firstname => 'child1',
                lastname => 'bar',
                children => [
                    {
                        firstname => 'grandchild',
                        lastname => 'bar',
                    }
                ],
            },
            {
                firstname => 'child2',
                lastname => 'bar',
            },
        ],
    });

  # select from the hierarchical relationship
  my $rs = $schema->resultset('Person')->search({},
    {
      'start_with' => { 'firstname' => 'foo', 'lastname' => 'bar' },
      'connect_by' => { 'parentid' => { '-prior' => { -ident => 'personid' } },
      'order_siblings_by' => { -asc => 'name' },
    };
  );

  # this will select the whole tree starting from person "foo bar", creating
  # following query:
  # SELECT
  #     me.persionid me.firstname, me.lastname, me.parentid
  # FROM
  #     person me
  # START WITH
  #     firstname = 'foo' and lastname = 'bar'
  # CONNECT BY
  #     parentid = prior personid
  # ORDER SIBLINGS BY
  #     firstname ASC

=head1 DESCRIPTION

This class implements base Oracle support. The subclass
L<DBIx::Class::Storage::DBI::Oracle::WhereJoins> is for C<(+)> joins in Oracle
versions before 9.0.

=head1 METHODS

=cut

sub _determine_supports_insert_returning {
  my $self = shift;

# TODO find out which version supports the RETURNING syntax
# 8i has it and earlier docs are a 404 on oracle.com

  return 1
    if $self->_server_info->{normalized_dbms_version} >= 8.001;

  return 0;
}

__PACKAGE__->_use_insert_returning_bound (1);

sub deployment_statements {
  my $self = shift;;
  my ($schema, $type, $version, $dir, $sqltargs, @rest) = @_;

  $sqltargs ||= {};
  my $quote_char = $self->schema->storage->sql_maker->quote_char;
  $sqltargs->{quote_table_names} = $quote_char ? 1 : 0;
  $sqltargs->{quote_field_names} = $quote_char ? 1 : 0;

  if (
    ! exists $sqltargs->{producer_args}{oracle_version}
      and
    my $dver = $self->_server_info->{dbms_version}
  ) {
    $sqltargs->{producer_args}{oracle_version} = $dver;
  }

  $self->next::method($schema, $type, $version, $dir, $sqltargs, @rest);
}

sub _dbh_last_insert_id {
  my ($self, $dbh, $source, @columns) = @_;
  my @ids = ();
  foreach my $col (@columns) {
    my $seq = ($source->column_info($col)->{sequence} ||= $self->get_autoinc_seq($source,$col));
    my $id = $self->_sequence_fetch( 'CURRVAL', $seq );
    push @ids, $id;
  }
  return @ids;
}

sub _dbh_get_autoinc_seq {
  my ($self, $dbh, $source, $col) = @_;

  my $sql_maker = $self->sql_maker;
  my ($ql, $qr) = map { $_ ? (quotemeta $_) : '' } $sql_maker->_quote_chars;

  my $source_name;
  if ( ref $source->name eq 'SCALAR' ) {
    $source_name = ${$source->name};

    # the ALL_TRIGGERS match further on is case sensitive - thus uppercase
    # stuff unless it is already quoted
    $source_name = uc ($source_name) if $source_name !~ /\"/;
  }
  else {
    $source_name = $source->name;
    $source_name = uc($source_name) unless $ql;
  }

  # trigger_body is a LONG
  local $dbh->{LongReadLen} = 64 * 1024 if ($dbh->{LongReadLen} < 64 * 1024);

  # disable default bindtype
  local $sql_maker->{bindtype} = 'normal';

  # look up the correct sequence automatically
  my ( $schema, $table ) = $source_name =~ /( (?:${ql})? \w+ (?:${qr})? ) \. ( (?:${ql})? \w+ (?:${qr})? )/x;

  # if no explicit schema was requested - use the default schema (which in the case of Oracle is the db user)
  $schema ||= uc( ($self->_dbi_connect_info||[])->[1] || '');

  my ($sql, @bind) = $sql_maker->select (
    'ALL_TRIGGERS',
    [qw/TRIGGER_BODY TABLE_OWNER TRIGGER_NAME/],
    {
      $schema ? (OWNER => $schema) : (),
      TABLE_NAME => $table || $source_name,
      TRIGGERING_EVENT => { -like => '%INSERT%' },  # this will also catch insert_or_update
      TRIGGER_TYPE => { -like => '%BEFORE%' },      # we care only about 'before' triggers
      STATUS => 'ENABLED',
     },
  );

  # to find all the triggers that mention the column in question a simple
  # regex grep since the trigger_body above is a LONG and hence not searchable
  my @triggers = ( map
    { my %inf; @inf{qw/body schema name/} = @$_; \%inf }
    ( grep
      { $_->[0] =~ /\:new\.${ql}${col}${qr} | \:new\.$col/xi }
      @{ $dbh->selectall_arrayref( $sql, {}, @bind ) }
    )
  );

  # extract all sequence names mentioned in each trigger
  for (@triggers) {
    $_->{sequences} = [ $_->{body} =~ / ( "? [\.\w\"\-]+ "? ) \. nextval /xig ];
  }

  my $chosen_trigger;

  # if only one trigger matched things are easy
  if (@triggers == 1) {

    if ( @{$triggers[0]{sequences}} == 1 ) {
      $chosen_trigger = $triggers[0];
    }
    else {
      $self->throw_exception( sprintf (
        "Unable to introspect trigger '%s' for column %s.%s (references multiple sequences). "
      . "You need to specify the correct 'sequence' explicitly in '%s's column_info.",
        $triggers[0]{name},
        $source_name,
        $col,
        $col,
      ) );
    }
  }
  # got more than one matching trigger - see if we can narrow it down
  elsif (@triggers > 1) {

    my @candidates = grep
      { $_->{body} =~ / into \s+ \:new\.$col /xi }
      @triggers
    ;

    if (@candidates == 1 && @{$candidates[0]{sequences}} == 1) {
      $chosen_trigger = $candidates[0];
    }
    else {
      $self->throw_exception( sprintf (
        "Unable to reliably select a BEFORE INSERT trigger for column %s.%s (possibilities: %s). "
      . "You need to specify the correct 'sequence' explicitly in '%s's column_info.",
        $source_name,
        $col,
        ( join ', ', map { "'$_->{name}'" } @triggers ),
        $col,
      ) );
    }
  }

  if ($chosen_trigger) {
    my $seq_name = $chosen_trigger->{sequences}[0];

    $seq_name = "$chosen_trigger->{schema}.$seq_name"
      unless $seq_name =~ /\./;

    return \$seq_name if $seq_name =~ /\"/; # may already be quoted in-trigger
    return $seq_name;
  }

  $self->throw_exception( sprintf (
    "No suitable BEFORE INSERT triggers found for column %s.%s. "
  . "You need to specify the correct 'sequence' explicitly in '%s's column_info.",
    $source_name,
    $col,
    $col,
  ));
}

sub _sequence_fetch {
  my ( $self, $type, $seq ) = @_;

  # use the maker to leverage quoting settings
  my $sql_maker = $self->sql_maker;
  my ($id) = $self->_get_dbh->selectrow_array ($sql_maker->select('DUAL', [ ref $seq ? \"$$seq.$type" : "$seq.$type" ] ) );
  return $id;
}

sub _ping {
  my $self = shift;

  my $dbh = $self->_dbh or return 0;

  local $dbh->{RaiseError} = 1;
  local $dbh->{PrintError} = 0;

  return try {
    $dbh->do('select 1 from dual');
    1;
  } catch {
    0;
  };
}

sub _dbh_execute {
  my ($self, $dbh, $sql, @args) = @_;

  my (@res, $tried);
  my $want = wantarray;
  my $next = $self->next::can;
  do {
    try {
      my $exec = sub {
        # Turn off sth caching for multi-part LOBs. See _prep_for_execute above.
        local $self->{disable_sth_caching} = 1
          if first {
            ($_->[0]{_ora_lob_autosplit_part}||0)
              > (__cache_queries_with_max_lob_parts-1)
          } @{ $args[0] };

        $self->$next($dbh, $sql, @args)
      };

      if (!defined $want) {
        $exec->();
      }
      elsif (! $want) {
        $res[0] = $exec->();
      }
      else {
        @res = $exec->();
      }

      $tried++;
    }
    catch {
      if (! $tried and $_ =~ /ORA-01003/) {
        # ORA-01003: no statement parsed (someone changed the table somehow,
        # invalidating your cursor.)
        delete $dbh->{CachedKids}{$sql};
      }
      else {
        $self->throw_exception($_);
      }
    };
  } while (! $tried++);

  return wantarray ? @res : $res[0];
}

sub _dbh_execute_array {
  #my ($self, $sth, $tuple_status, @extra) = @_;

  # DBD::Oracle warns loudly on partial execute_array failures
  local $_[1]->{PrintWarn} = 0;

  shift->next::method(@_);
}

=head2 get_autoinc_seq

Returns the sequence name for an autoincrement column

=cut

sub get_autoinc_seq {
  my ($self, $source, $col) = @_;

  $self->dbh_do('_dbh_get_autoinc_seq', $source, $col);
}

=head2 datetime_parser_type

This sets the proper DateTime::Format module for use with
L<DBIx::Class::InflateColumn::DateTime>.

=head2 connect_call_datetime_setup

Used as:

    on_connect_call => 'datetime_setup'

In L<connect_info|DBIx::Class::Storage::DBI/connect_info> to set the session nls
date, and timestamp values for use with L<DBIx::Class::InflateColumn::DateTime>
and the necessary environment variables for L<DateTime::Format::Oracle>, which
is used by it.

Maximum allowable precision is used, unless the environment variables have
already been set.

These are the defaults used:

  $ENV{NLS_DATE_FORMAT}         ||= 'YYYY-MM-DD HH24:MI:SS';
  $ENV{NLS_TIMESTAMP_FORMAT}    ||= 'YYYY-MM-DD HH24:MI:SS.FF';
  $ENV{NLS_TIMESTAMP_TZ_FORMAT} ||= 'YYYY-MM-DD HH24:MI:SS.FF TZHTZM';

To get more than second precision with L<DBIx::Class::InflateColumn::DateTime>
for your timestamps, use something like this:

  use Time::HiRes 'time';
  my $ts = DateTime->from_epoch(epoch => time);

=cut

sub connect_call_datetime_setup {
  my $self = shift;

  my $date_format = $ENV{NLS_DATE_FORMAT} ||= 'YYYY-MM-DD HH24:MI:SS';
  my $timestamp_format = $ENV{NLS_TIMESTAMP_FORMAT} ||=
    'YYYY-MM-DD HH24:MI:SS.FF';
  my $timestamp_tz_format = $ENV{NLS_TIMESTAMP_TZ_FORMAT} ||=
    'YYYY-MM-DD HH24:MI:SS.FF TZHTZM';

  $self->_do_query(
    "alter session set nls_date_format = '$date_format'"
  );
  $self->_do_query(
    "alter session set nls_timestamp_format = '$timestamp_format'"
  );
  $self->_do_query(
    "alter session set nls_timestamp_tz_format='$timestamp_tz_format'"
  );
}

### Note originally by Ron "Quinn" Straight <quinnfazigu@gmail.org>
### http://git.shadowcat.co.uk/gitweb/gitweb.cgi?p=dbsrgits/DBIx-Class.git;a=commitdiff;h=5db2758de644d53e07cd3e05f0e9037bf40116fc
#
# Handle LOB types in Oracle.  Under a certain size (4k?), you can get away
# with the driver assuming your input is the deprecated LONG type if you
# encode it as a hex string.  That ain't gonna fly at larger values, where
# you'll discover you have to do what this does.
#
# This method had to be overridden because we need to set ora_field to the
# actual column, and that isn't passed to the call (provided by Storage) to
# bind_attribute_by_data_type.
#
# According to L<DBD::Oracle>, the ora_field isn't always necessary, but
# adding it doesn't hurt, and will save your bacon if you're modifying a
# table with more than one LOB column.
#
sub _dbi_attrs_for_bind {
  my ($self, $ident, $bind) = @_;

  my $attrs = $self->next::method($ident, $bind);

  for my $i (0 .. $#$attrs) {
    if (keys %{$attrs->[$i]||{}} and my $col = $bind->[$i][0]{dbic_colname}) {
      $attrs->[$i]{ora_field} = $col;
    }
  }

  $attrs;
}

my $dbd_loaded;
sub bind_attribute_by_data_type {
  my ($self, $dt) = @_;

  $dbd_loaded ||= do {
    require DBD::Oracle;
    if ($DBD::Oracle::VERSION eq '1.23') {
      $self->throw_exception(
        "BLOB/CLOB support in DBD::Oracle == 1.23 is broken, use an earlier or later ".
        "version.\n\nSee: https://rt.cpan.org/Public/Bug/Display.html?id=46016\n"
      );
    }
    1;
  };

  if ($self->_is_lob_type($dt)) {
    return {
      ora_type => $self->_is_text_lob_type($dt)
        ? DBD::Oracle::ORA_CLOB()
        : DBD::Oracle::ORA_BLOB()
    };
  }
}

# Handle blob columns in WHERE.
#
# For equality comparisons:
#
# We split data intended for comparing to a LOB into 2000 character chunks and
# compare them using dbms_lob.substr on the LOB column.
#
# We turn off DBD::Oracle LOB binds for these partial LOB comparisons by passing
# dbd_attrs => undef, because these are regular varchar2 comparisons and
# otherwise the query will fail.
#
# Since the most common comparison size is likely to be under 4000 characters
# (TEXT comparisons previously deployed to other RDBMSes) we disable
# prepare_cached for queries with more than two part comparisons to a LOB
# column. This is done in _dbh_execute (above) which was previously overridden
# to gracefully recover from an Oracle error. This is to be careful to not
# exhaust your application's open cursor limit.
#
# See:
# http://itcareershift.com/blog1/2011/02/21/oracle-max-number-of-open-cursors-complete-reference-for-the-new-oracle-dba/
# on the open_cursor limit.
#
# For everything else:
#
# We assume that everything that is not a LOB comparison, will most likely be a
# LIKE query or some sort of function invocation. This may prove to be a naive
# assumption in the future, but for now it should cover the two most likely
# things users would want to do with a BLOB or CLOB, an equality test or a LIKE
# query (on a CLOB.)
#
# For these expressions, the bind must NOT have the attributes of a LOB bind for
# DBD::Oracle, otherwise the query will fail. This is done by passing
# dbd_attrs => undef.

sub _prep_for_execute {
  my $self = shift;
  my ($op) = @_;

  my ($sql, $bind) = $self->next::method(@_);

  return ($sql, $bind) if $op eq 'insert';

  my $blob_bind_index;
  for (0 .. $#$bind) {
    $blob_bind_index->{$_} = 1 if $self->_is_lob_type(
      $bind->[$_][0]{sqlt_datatype}
    );
  }

  return ($sql, $bind) unless $blob_bind_index;

  my (@sql_parts, $new_sql, @new_binds);

  if ($op eq 'select' || $op eq 'delete') {
    @sql_parts = split /\?/, $sql;
  }
  elsif ($op eq 'update') {
    $self->throw_exception('Update with complex WHERE clauses currently not supported')
      if $sql =~ /\bWHERE\b .+ \bWHERE\b/xs;

    my ($set_part, $where_part) = $sql =~ /^ (.+?) ( \bWHERE\b .+) /xs;

    my $set_bind_count = $set_part =~ y/?//;
    @new_binds = splice @$bind, 0, $set_bind_count;

    @sql_parts = split /\?/, $where_part;
    $new_sql  = $set_part;
  }
  else {
    $self->throw_exception("Unsupported \$op: $op");
  }

  my $col_equality_re = qr/ (?<=\s) ([\w."]+) (\s*=\s*) $/x;

  for my $b_idx (0 .. $#$bind) {
    my $bound = $bind->[$b_idx];

    if ($blob_bind_index->{$b_idx}) {
      if (my ($col, $eq) = $sql_parts[0] =~ $col_equality_re) {
        my $data = $bound->[1];

        $data = "$data" if ref $data;

        my @parts = unpack '(a2000)*', $data;

        my @sql_frag;

        for my $idx (0..$#parts) {
          push @sql_frag, sprintf (
            'UTL_RAW.CAST_TO_VARCHAR2(RAWTOHEX(DBMS_LOB.SUBSTR(%s, 2000, %d))) = ?',
            $col, ($idx*2000 + 1),
          );
        }

        my $sql_frag = '( ' . (join ' AND ', @sql_frag) . ' )';

        $sql_parts[0] =~ s/$col_equality_re/$sql_frag/;

        $new_sql .= shift @sql_parts;

        for my $idx (0..$#parts) {
          push @new_binds, [
            {
              %{ $bound->[0] },
              _ora_lob_autosplit_part => $idx,
              dbd_attrs => undef,
            },
            $parts[$idx]
          ];
        }
      }
      else {
        $new_sql .= shift(@sql_parts) . '?';

        push @new_binds, [
          {
            %{ $bound->[0] },
            dbd_attrs => undef,
          },
          $bound->[1],
        ];
      }
    }
    else {
      $new_sql .= shift(@sql_parts) . '?';
      push @new_binds, $bound;
    }
  }

  if (@sql_parts > 1) {
    carp "There are more placeholders than binds, this should not happen!";
    @sql_parts = join ('?', @sql_parts);
  }

  $new_sql .= $sql_parts[0];

  return ($new_sql, \@new_binds);
}

# Savepoints stuff.

sub _exec_svp_begin {
  my ($self, $name) = @_;
  $self->_dbh->do("SAVEPOINT $name");
}

# Oracle automatically releases a savepoint when you start another one with the
# same name.
sub _exec_svp_release { 1 }

sub _exec_svp_rollback {
  my ($self, $name) = @_;
  $self->_dbh->do("ROLLBACK TO SAVEPOINT $name")
}

=head2 relname_to_table_alias

L<DBIx::Class> uses L<DBIx::Class::Relationship> names as table aliases in
queries.

Unfortunately, Oracle doesn't support identifiers over 30 chars in length, so
the L<DBIx::Class::Relationship> name is shortened and appended with half of an
MD5 hash.

See L<DBIx::Class::Storage/"relname_to_table_alias">.

=cut

sub relname_to_table_alias {
  my $self = shift;
  my ($relname, $join_count) = @_;

  my $alias = $self->next::method(@_);

  # we need to shorten here in addition to the shortening in SQLA itself,
  # since the final relnames are a crucial for the join optimizer
  return $self->sql_maker->_shorten_identifier($alias);
}

=head2 with_deferred_fk_checks

Runs a coderef between:

  alter session set constraints = deferred
  ...
  alter session set constraints = immediate

to defer foreign key checks.

Constraints must be declared C<DEFERRABLE> for this to work.

=cut

sub with_deferred_fk_checks {
  my ($self, $sub) = @_;

  my $txn_scope_guard = $self->txn_scope_guard;

  $self->_do_query('alter session set constraints = deferred');

  my $sg = Scope::Guard->new(sub {
    $self->_do_query('alter session set constraints = immediate');
  });

  return
    preserve_context { $sub->() } after => sub { $txn_scope_guard->commit };
}

=head1 ATTRIBUTES

Following additional attributes can be used in resultsets.

=head2 connect_by or connect_by_nocycle

=over 4

=item Value: \%connect_by

=back

A hashref of conditions used to specify the relationship between parent rows
and child rows of the hierarchy.


  connect_by => { parentid => 'prior personid' }

  # adds a connect by statement to the query:
  # SELECT
  #     me.persionid me.firstname, me.lastname, me.parentid
  # FROM
  #     person me
  # CONNECT BY
  #     parentid = prior persionid
  

  connect_by_nocycle => { parentid => 'prior personid' }

  # adds a connect by statement to the query:
  # SELECT
  #     me.persionid me.firstname, me.lastname, me.parentid
  # FROM
  #     person me
  # CONNECT BY NOCYCLE
  #     parentid = prior persionid


=head2 start_with

=over 4

=item Value: \%condition

=back

A hashref of conditions which specify the root row(s) of the hierarchy.

It uses the same syntax as L<DBIx::Class::ResultSet/search>

  start_with => { firstname => 'Foo', lastname => 'Bar' }

  # SELECT
  #     me.persionid me.firstname, me.lastname, me.parentid
  # FROM
  #     person me
  # START WITH
  #     firstname = 'foo' and lastname = 'bar'
  # CONNECT BY
  #     parentid = prior persionid

=head2 order_siblings_by

=over 4

=item Value: ($order_siblings_by | \@order_siblings_by)

=back

Which column(s) to order the siblings by.

It uses the same syntax as L<DBIx::Class::ResultSet/order_by>

  'order_siblings_by' => 'firstname ASC'

  # SELECT
  #     me.persionid me.firstname, me.lastname, me.parentid
  # FROM
  #     person me
  # CONNECT BY
  #     parentid = prior persionid
  # ORDER SIBLINGS BY
  #     firstname ASC

=head1 AUTHOR

See L<DBIx::Class/AUTHOR> and L<DBIx::Class/CONTRIBUTORS>.

=head1 LICENSE

You may distribute this code under the same terms as Perl itself.

=cut

1;
# vim:sts=2 sw=2: