X-Git-Url: http://git.shadowcat.co.uk/gitweb/gitweb.cgi?a=blobdiff_plain;f=lib%2FDBIx%2FClass%2FStorage%2FDBI%2FSybase.pm;h=5747cb948ddcfcac8d1dfa9dfeb1051fc744b66a;hb=587daa97726631a2a5edc95ea6bfe62bf340bd8d;hp=2e01e8e5ff594f5d17f51b833d3403aa730baa38;hpb=1a58752c42ba9acf33e2943b678de4948cf3ee3f;p=dbsrgits%2FDBIx-Class.git diff --git a/lib/DBIx/Class/Storage/DBI/Sybase.pm b/lib/DBIx/Class/Storage/DBI/Sybase.pm index 2e01e8e..5747cb9 100644 --- a/lib/DBIx/Class/Storage/DBI/Sybase.pm +++ b/lib/DBIx/Class/Storage/DBI/Sybase.pm @@ -13,11 +13,16 @@ use List::Util (); use Sub::Name (); __PACKAGE__->mk_group_accessors('simple' => - qw/_identity _blob_log_on_update _writer_storage _is_writer_storage + qw/_identity _blob_log_on_update _writer_storage _is_extra_storage + _bulk_storage _is_bulk_storage _began_bulk_work + _bulk_disabled_due_to_coderef_connect_info_warned _identity_method/ ); -my @also_proxy_to_writer_storage = qw/ +my @also_proxy_to_extra_storages = qw/ + connect_call_set_auto_cast auto_cast connect_call_blob_setup + connect_call_datetime_setup + disconnect _connect_info _sql_maker _sql_maker_opts disable_sth_caching auto_savepoint unsafe cursor_class debug debugobj schema /; @@ -102,7 +107,7 @@ EOF bless $self, $no_bind_vars; $self->_rebless; } elsif (not $self->_typeless_placeholders_supported) { -# this is highly unlikely, but we check just in case + # this is highly unlikely, but we check just in case $self->auto_cast(1); } } @@ -118,28 +123,63 @@ sub _init { # create storage for insert/(update blob) transactions, # unless this is that storage - return if $self->_is_writer_storage; + return if $self->_is_extra_storage; my $writer_storage = (ref $self)->new; - $writer_storage->_is_writer_storage(1); + $writer_storage->_is_extra_storage(1); $writer_storage->connect_info($self->connect_info); + $writer_storage->auto_cast($self->auto_cast); $self->_writer_storage($writer_storage); + +# create a bulk storage unless connect_info is a coderef + return + if (Scalar::Util::reftype($self->_dbi_connect_info->[0])||'') eq 'CODE'; + + my $bulk_storage = (ref $self)->new; + + $bulk_storage->_is_extra_storage(1); + $bulk_storage->_is_bulk_storage(1); # for special ->disconnect acrobatics + $bulk_storage->connect_info($self->connect_info); + +# this is why + $bulk_storage->_dbi_connect_info->[0] .= ';bulkLogin=1'; + + $self->_bulk_storage($bulk_storage); } -for my $method (@also_proxy_to_writer_storage) { +for my $method (@also_proxy_to_extra_storages) { no strict 'refs'; + no warnings 'redefine'; my $replaced = __PACKAGE__->can($method); - *{$method} = Sub::Name::subname __PACKAGE__."::$method" => sub { + *{$method} = Sub::Name::subname $method => sub { my $self = shift; $self->_writer_storage->$replaced(@_) if $self->_writer_storage; + $self->_bulk_storage->$replaced(@_) if $self->_bulk_storage; return $self->$replaced(@_); }; } +sub disconnect { + my $self = shift; + +# Even though we call $sth->finish for uses off the bulk API, there's still an +# "active statement" warning on disconnect, which we throw away here. +# This is due to the bug described in insert_bulk. +# Currently a noop because 'prepare' is used instead of 'prepare_cached'. + local $SIG{__WARN__} = sub { + warn $_[0] unless $_[0] =~ /active statement/i; + } if $self->_is_bulk_storage; + +# so that next transaction gets a dbh + $self->_began_bulk_work(0) if $self->_is_bulk_storage; + + $self->next::method; +} + # Make sure we have CHAINED mode turned on if AutoCommit is off in non-FreeTDS # DBD::Sybase (since we don't know how DBD::Sybase was compiled.) If however # we're using FreeTDS, CHAINED mode turns on an implicit transaction which we @@ -148,6 +188,12 @@ sub _populate_dbh { my $self = shift; $self->next::method(@_); + + if ($self->_is_bulk_storage) { +# this should be cleared on every reconnect + $self->_began_bulk_work(0); + return; + } if (not $self->using_freetds) { $self->_dbh->{syb_chained_txn} = 1; @@ -195,6 +241,12 @@ sub _is_lob_type { $type && $type =~ /(?:text|image|lob|bytea|binary|memo)/i; } +sub _is_lob_column { + my ($self, $source, $column) = @_; + + return $self->_is_lob_type($source->column_info($column)->{data_type}); +} + sub _prep_for_execute { my $self = shift; my ($op, $extra_bind, $ident, $args) = @_; @@ -352,110 +404,336 @@ sub _insert { sub update { my $self = shift; - my ($source, $fields, $where) = @_; + my ($source, $fields, $where, @rest) = @_; my $wantarray = wantarray; + my $blob_cols = $self->_remove_blob_cols($source, $fields); + my $table = $source->name; + + my $identity_col = List::Util::first + { $source->column_info($_)->{is_auto_increment} } + $source->columns; + + my $is_identity_update = $identity_col && defined $fields->{$identity_col}; + if (not $blob_cols) { + $self->_set_session_identity(UPDATE => $table, 'ON') + if $is_identity_update; + return $self->next::method(@_); + + $self->_set_session_identity(UPDATE => $table, 'OFF') + if $is_identity_update; } +# If there are any blobs in $where, Sybase will return a descriptive error +# message. + # update+blob update(s) done atomically on separate connection $self = $self->_writer_storage; my $guard = $self->txn_scope_guard; - my @res; - if ($wantarray) { - @res = $self->next::method(@_); - } - elsif (defined $wantarray) { - $res[0] = $self->next::method(@_); - } - else { - $self->next::method(@_); - } +# First update the blob columns to be updated to '' (taken from $fields, where +# it is originally put by _remove_blob_cols .) + my %blobs_to_empty = map { ($_ => delete $fields->{$_}) } keys %$blob_cols; + $self->next::method($source, \%blobs_to_empty, $where, @rest); + +# Now update the blobs before the other columns in case the update of other +# columns makes the search condition invalid. $self->_update_blobs($source, $blob_cols, $where); + my @res; + if (%$fields) { + $self->_set_session_identity(UPDATE => $table, 'ON') + if $is_identity_update; + + if ($wantarray) { + @res = $self->next::method(@_); + } + elsif (defined $wantarray) { + $res[0] = $self->next::method(@_); + } + else { + $self->next::method(@_); + } + + $self->_set_session_identity(UPDATE => $table, 'OFF') + if $is_identity_update; + } + $guard->commit; return $wantarray ? @res : $res[0]; } -### the insert_bulk stuff stolen from DBI/MSSQL.pm - -sub _set_identity_insert { - my ($self, $table) = @_; +# for IDENTITY_INSERT / IDENTITY_UPDATE +sub _set_session_identity { + my ($self, $op, $table, $off_on) = @_; my $sql = sprintf ( - 'SET IDENTITY_INSERT %s ON', - $self->sql_maker->_quote ($table), + 'SET IDENTITY_%s %s %s', + uc $op, + $self->sql_maker->_quote($table), + uc $off_on, ); + $self->_query_start($sql); + my $dbh = $self->_get_dbh; - eval { $dbh->do ($sql) }; - if ($@) { + eval { + local $dbh->{RaiseError} = 1; + local $dbh->{PrintError} = 0; + $dbh->do ($sql) + }; + my $exception = $@; + + $self->_query_end($sql); + + if ($exception) { $self->throw_exception (sprintf "Error executing '%s': %s", $sql, - $dbh->errstr, + $exception, ); } } -sub _unset_identity_insert { - my ($self, $table) = @_; - - my $sql = sprintf ( - 'SET IDENTITY_INSERT %s OFF', - $self->sql_maker->_quote ($table), - ); - - my $dbh = $self->_get_dbh; - $dbh->do ($sql); -} - -# XXX this should use the DBD::Sybase bulk API, where possible sub insert_bulk { my $self = shift; my ($source, $cols, $data) = @_; + my $identity_col = List::Util::first + { $source->column_info($_)->{is_auto_increment} } + $source->columns; + my $is_identity_insert = (List::Util::first - { $source->column_info ($_)->{is_auto_increment} } - (@{$cols}) - ) - ? 1 - : 0; - - if ($is_identity_insert) { - $self->_set_identity_insert ($source->name); + { $_ eq $identity_col } + @{$cols} + ) ? 1 : 0; + + my @source_columns = $source->columns; + + my $use_bulk_api = + $self->_bulk_storage && + $self->_get_dbh->{syb_has_blk}; + + if ((not $use_bulk_api) && + (Scalar::Util::reftype($self->_dbi_connect_info->[0])||'') eq 'CODE' && + (not $self->_bulk_disabled_due_to_coderef_connect_info_warned)) { + carp <<'EOF'; +Bulk API support disabled due to use of a CODEREF connect_info. Reverting to +regular array inserts. +EOF + $self->_bulk_disabled_due_to_coderef_connect_info_warned(1); } - $self->next::method(@_); + if (not $use_bulk_api) { + my $blob_cols = $self->_remove_blob_cols_array($source, $cols, $data); + + ($self, my ($guard)) = do { + if ($self->{transaction_depth} == 0 && $blob_cols) { + ($self->_writer_storage, $self->_writer_storage->txn_scope_guard); + } + else { + ($self, undef); + } + }; + + $self->next::method(@_); + + if ($blob_cols) { + if ($is_identity_insert) { + $self->_insert_blobs_array ($source, $blob_cols, $cols, $data); + } + else { + my @cols_with_identities = (@$cols, $identity_col); + + ## calculate identities + # XXX This assumes identities always increase by 1, which may or may not + # be true. + my ($last_identity) = + $self->_dbh->selectrow_array ( + $self->_fetch_identity_sql($source, $identity_col) + ); + my @identities = (($last_identity - @$data + 1) .. $last_identity); + + my @data_with_identities = map [@$_, shift @identities], @$data; + + $self->_insert_blobs_array ( + $source, $blob_cols, \@cols_with_identities, \@data_with_identities + ); + } + } - if ($is_identity_insert) { - $self->_unset_identity_insert ($source->name); + $guard->commit if $guard; + return; } -} -### end of stolen insert_bulk section +# otherwise, use the bulk API + +# rearrange @$data so that columns are in database order + my %orig_idx; + @orig_idx{@$cols} = 0..$#$cols; + + my %new_idx; + @new_idx{@source_columns} = 0..$#source_columns; + + my @new_data; + for my $datum (@$data) { + my $new_datum = []; + for my $col (@source_columns) { +# identity data will be 'undef' if not $is_identity_insert +# columns with defaults will also be 'undef' + $new_datum->[ $new_idx{$col} ] = + exists $orig_idx{$col} ? $datum->[ $orig_idx{$col} ] : undef; + } + push @new_data, $new_datum; + } + +# bcp identity index is 1-based + my $identity_idx = exists $new_idx{$identity_col} ? + $new_idx{$identity_col} + 1 : 0; + +## Set a client-side conversion error handler, straight from DBD::Sybase docs. +# This ignores any data conversion errors detected by the client side libs, as +# they are usually harmless. + my $orig_cslib_cb = DBD::Sybase::set_cslib_cb( + Sub::Name::subname insert_bulk => sub { + my ($layer, $origin, $severity, $errno, $errmsg, $osmsg, $blkmsg) = @_; + + return 1 if $errno == 36; + + carp + "Layer: $layer, Origin: $origin, Severity: $severity, Error: $errno" . + ($errmsg ? "\n$errmsg" : '') . + ($osmsg ? "\n$osmsg" : '') . + ($blkmsg ? "\n$blkmsg" : ''); + + return 0; + }); + + eval { + my $bulk = $self->_bulk_storage; + + my $guard = $bulk->txn_scope_guard; + +## XXX get this to work instead of our own $sth +## will require SQLA or *Hacks changes for ordered columns +# $bulk->next::method($source, \@source_columns, \@new_data, { +# syb_bcp_attribs => { +# identity_flag => $is_identity_insert, +# identity_column => $identity_idx, +# } +# }); + my $sql = 'INSERT INTO ' . + $bulk->sql_maker->_quote($source->name) . ' (' . +# colname list is ignored for BCP, but does no harm + (join ', ', map $bulk->sql_maker->_quote($_), @source_columns) . ') '. + ' VALUES ('. (join ', ', ('?') x @source_columns) . ')'; + +## XXX there's a bug in the DBD::Sybase bulk support that makes $sth->finish for +## a prepare_cached statement ineffective. Replace with ->sth when fixed, or +## better yet the version above. Should be fixed in DBD::Sybase . + my $sth = $bulk->_get_dbh->prepare($sql, +# 'insert', # op + { + syb_bcp_attribs => { + identity_flag => $is_identity_insert, + identity_column => $identity_idx, + } + } + ); + + my @bind = do { + my $idx = 0; + map [ $_, $idx++ ], @source_columns; + }; + $self->_execute_array( + $source, $sth, \@bind, \@source_columns, \@new_data, $guard + ); + + $bulk->_query_end($sql); + }; + my $exception = $@; + if ($exception =~ /-Y option/) { + carp <<"EOF"; + +Sybase bulk API operation failed due to character set incompatibility, reverting +to regular array inserts: + +*** Try unsetting the LANG environment variable. + +$@ +EOF + $self->_bulk_storage(undef); + DBD::Sybase::set_cslib_cb($orig_cslib_cb); + unshift @_, $self; + goto \&insert_bulk; + } + elsif ($exception) { + DBD::Sybase::set_cslib_cb($orig_cslib_cb); +# rollback makes the bulkLogin connection unusable + $self->_bulk_storage->disconnect; + $self->throw_exception($exception); + } + + DBD::Sybase::set_cslib_cb($orig_cslib_cb); +} + +# Make sure blobs are not bound as placeholders, and return any non-empty ones +# as a hash. sub _remove_blob_cols { my ($self, $source, $fields) = @_; my %blob_cols; for my $col (keys %$fields) { - if ($self->_is_lob_type($source->column_info($col)->{data_type})) { - $blob_cols{$col} = delete $fields->{$col}; - $fields->{$col} = \"''"; + if ($self->_is_lob_column($source, $col)) { + my $blob_val = delete $fields->{$col}; + if (not defined $blob_val) { + $fields->{$col} = \'NULL'; + } + else { + $fields->{$col} = \"''"; + $blob_cols{$col} = $blob_val unless $blob_val eq ''; + } } } return keys %blob_cols ? \%blob_cols : undef; } +# same for insert_bulk +sub _remove_blob_cols_array { + my ($self, $source, $cols, $data) = @_; + + my @blob_cols; + + for my $i (0..$#$cols) { + my $col = $cols->[$i]; + + if ($self->_is_lob_column($source, $col)) { + for my $j (0..$#$data) { + my $blob_val = delete $data->[$j][$i]; + if (not defined $blob_val) { + $data->[$j][$i] = \'NULL'; + } + else { + $data->[$j][$i] = \"''"; + $blob_cols[$j][$i] = $blob_val + unless $blob_val eq ''; + } + } + } + } + + return @blob_cols ? \@blob_cols : undef; +} + sub _update_blobs { my ($self, $source, $blob_cols, $where) = @_; @@ -491,7 +769,7 @@ sub _insert_blobs { my ($self, $source, $blob_cols, $row) = @_; my $dbh = $self->_get_dbh; - my $table = $source->from; + my $table = $source->name; my %row = %$row; my (@primary_cols) = $source->primary_columns; @@ -511,6 +789,14 @@ sub _insert_blobs { $cursor->next; my $sth = $cursor->sth; + if (not $sth) { + + $self->throw_exception( + "Could not find row in table '$table' for blob update:\n" + . $self->_pretty_print (\%where) + ); + } + eval { do { $sth->func('CS_GET', 1, 'ct_data_info') or die $sth->errstr; @@ -545,6 +831,26 @@ sub _insert_blobs { } } +sub _insert_blobs_array { + my ($self, $source, $blob_cols, $cols, $data) = @_; + + for my $i (0..$#$data) { + my $datum = $data->[$i]; + + my %row; + @row{ @$cols } = @$datum; + + my %blob_vals; + for my $j (0..$#$cols) { + if (exists $blob_cols->[$i][$j]) { + $blob_vals{ $cols->[$j] } = $blob_cols->[$i][$j]; + } + } + + $self->_insert_blobs ($source, \%blob_vals, \%row); + } +} + =head2 connect_call_datetime_setup Used as: @@ -596,10 +902,18 @@ sub datetime_parser_type { "DateTime::Format::Sybase" } sub _dbh_begin_work { my $self = shift; + +# bulkLogin=1 connections are always in a transaction, and can only call BEGIN +# TRAN once. However, we need to make sure there's a $dbh. + return if $self->_is_bulk_storage && $self->_dbh && $self->_began_bulk_work; + $self->next::method(@_); + if ($self->using_freetds) { $self->_get_dbh->do('BEGIN TRAN'); } + + $self->_began_bulk_work(1) if $self->_is_bulk_storage; } sub _dbh_commit { @@ -698,10 +1012,10 @@ session variable. =head1 TRANSACTIONS Due to limitations of the TDS protocol, L, or both; you cannot -begin a transaction while there are active cursors. An active cursor is, for -example, a L that has been executed using -C or C but has not been exhausted or -L. +begin a transaction while there are active cursors; nor can you use multiple +active cursors within a transaction. An active cursor is, for example, a +L that has been executed using C or +C but has not been exhausted or L. For example, this will not work: @@ -715,6 +1029,11 @@ For example, this will not work: } }); +This won't either: + + my $first_row = $large_rs->first; + $schema->txn_do(sub { ... }); + Transactions done for inserts in C mode when placeholders are in use are not affected, as they are done on an extra database handle. @@ -771,6 +1090,27 @@ C command on connection. See L for a L setting you need to work with C columns. +=head1 BULK API + +The experimental L Bulk API support is used for +L in B context, in a transaction +on a separate connection. + +To use this feature effectively, use a large number of rows for each +L call, eg.: + + while (my $rows = $data_source->get_100_rows()) { + $rs->populate($rows); + } + +B the L +calls in your C classes B list columns in database order for this +to work. Also, you may have to unset the C environment variable before +loading your app, if it doesn't match the character set of your database. + +When inserting IMAGE columns using this method, you'll need to use +L as well. + =head1 AUTHOR See L.