X-Git-Url: http://git.shadowcat.co.uk/gitweb/gitweb.cgi?a=blobdiff_plain;f=lib%2FDBIx%2FClass%2FStorage%2FDBI%2FSybase.pm;h=bccd0b4799796fc0e919867e5c2101440b11a1dc;hb=b561bb6fb2ddc0faada3313934fec2f14d62a625;hp=43932fac8084722eb55840b2e92a1cc890e3309f;hpb=322b7a6bb9b1c41935395b8fe809f2c36ac945dd;p=dbsrgits%2FDBIx-Class.git diff --git a/lib/DBIx/Class/Storage/DBI/Sybase.pm b/lib/DBIx/Class/Storage/DBI/Sybase.pm index 43932fa..bccd0b4 100644 --- a/lib/DBIx/Class/Storage/DBI/Sybase.pm +++ b/lib/DBIx/Class/Storage/DBI/Sybase.pm @@ -10,11 +10,23 @@ use base qw/ use mro 'c3'; use Carp::Clan qw/^DBIx::Class/; use List::Util (); +use Sub::Name (); __PACKAGE__->mk_group_accessors('simple' => - qw/_identity _blob_log_on_update _insert_dbh _identity_method/ + 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_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 +/; + =head1 NAME DBIx::Class::Storage::DBI::Sybase - Sybase support for DBIx::Class @@ -88,9 +100,6 @@ EOF $self->_rebless; } } - - $self->set_textsize; # based on LongReadLen in connect_info - } elsif (not $self->_get_dbh->{syb_dynamic_supported}) { # not necessarily FreeTDS, but no placeholders nevertheless @@ -98,15 +107,79 @@ 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); } - - $self->_set_max_connect(256); } } } +sub _init { + my $self = shift; + $self->_set_max_connect(256); + + # based on LongReadLen in connect_info + $self->set_textsize if $self->using_freetds; + +# create storage for insert/(update blob) transactions, +# unless this is that storage + return if $self->_is_extra_storage; + + my $writer_storage = (ref $self)->new; + + $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_extra_storages) { + no strict 'refs'; + no warnings 'redefine'; + + my $replaced = __PACKAGE__->can($method); + + *{$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 @@ -115,6 +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; @@ -162,6 +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, $args) = @_; @@ -236,7 +321,9 @@ sub _native_data_type { sub _fetch_identity_sql { my ($self, $source, $col) = @_; - return "SELECT MAX($col) FROM ".$source->from; + return sprintf ("SELECT MAX(%s) FROM %s", + map { $self->sql_maker->_quote ($_) } ($col, $source->from) + ); } sub _execute { @@ -262,94 +349,394 @@ sub insert { my $blob_cols = $self->_remove_blob_cols($source, $to_insert); -# insert+blob insert done atomically, on _insert_dbh - (my ($guard), local ($self->{_dbh})) = do { - $self->_insert_dbh($self->_connect(@{ $self->_dbi_connect_info })) - unless $self->_insert_dbh; + my $identity_col = List::Util::first + { $source->column_info($_)->{is_auto_increment} } + $source->columns; + + # do we need the horrific SELECT MAX(COL) hack? + my $dumb_last_insert_id = + $identity_col + && (not exists $to_insert->{$identity_col}) + && ($self->_identity_method||'') ne '@@IDENTITY'; + + my $next = $self->next::can; + + # we are already in a transaction, or there are no blobs + # and we don't need the PK - just (try to) do it + if ($self->{transaction_depth} + || (!$blob_cols && !$dumb_last_insert_id) + ) { + return $self->_insert ( + $next, $source, $to_insert, $blob_cols, $identity_col + ); + } + + # otherwise use the _writer_storage to do the insert+transaction on another + # connection + my $guard = $self->_writer_storage->txn_scope_guard; + + my $updated_cols = $self->_writer_storage->_insert ( + $next, $source, $to_insert, $blob_cols, $identity_col + ); + + $self->_identity($self->_writer_storage->_identity); + + $guard->commit; - my $new_guard = $self->txn_scope_guard; + return $updated_cols; +} + +sub _insert { + my ($self, $next, $source, $to_insert, $blob_cols, $identity_col) = @_; -# _dbh_begin_work may reconnect, if so we need to update _insert_dbh - $self->_insert_dbh($self->_dbh); + my $updated_cols = $self->$next ($source, $to_insert); - ($new_guard, $self->_insert_dbh) - } if $blob_cols; + my $final_row = { + $identity_col => $self->last_insert_id($source, $identity_col), + %$to_insert, + %$updated_cols, + }; - my $need_last_insert_id = 0; + $self->_insert_blobs ($source, $blob_cols, $final_row) if $blob_cols; - my ($identity_col) = - map $_->[0], - grep $_->[1]{is_auto_increment}, - map [ $_, $source->column_info($_) ], + return $updated_cols; +} + +sub update { + my $self = shift; + my ($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; - $need_last_insert_id = 1 - if $identity_col && (not exists $to_insert->{$identity_col}); + my $is_identity_update = $identity_col && defined $fields->{$identity_col}; + + if (not $blob_cols) { + $self->_set_identity_insert($table, 'update') if $is_identity_update; + return $self->next::method(@_); + $self->_unset_identity_insert($table, 'update') if $is_identity_update; + } + +# check that we're not updating a blob column that's also in $where + for my $blob (grep $self->_is_lob_column($source, $_), $source->columns) { + if (exists $where->{$blob} && exists $fields->{$blob}) { + croak +'Update of TEXT/IMAGE column that is also in search condition impossible'; + } + } + +# update+blob update(s) done atomically on separate connection + $self = $self->_writer_storage; - # We have to do the insert in a transaction to avoid race conditions with the - # SELECT MAX(COL) identity method used when placeholders are enabled. - my $updated_cols = do { - no warnings 'uninitialized'; - if ( - $need_last_insert_id && - $self->_identity_method ne '@@IDENTITY' && - !$self->{transaction_depth} - ) { - $self->_insert_dbh($self->_connect(@{ $self->_dbi_connect_info })) - unless $self->_insert_dbh; - local $self->{_dbh} = $self->_insert_dbh; + my $guard = $self->txn_scope_guard; - my $guard = $self->txn_scope_guard; +# 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; -# _dbh_begin_work may reconnect, if so we need to update _insert_dbh - $self->_insert_dbh($self->_dbh); + $self->next::method($source, \%blobs_to_empty, $where, @rest); - my $upd_cols = $self->next::method (@_); - $guard->commit; +# 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); - $upd_cols; + my @res; + if (%$fields) { + $self->_set_identity_insert($table, 'update') if $is_identity_update; + + if ($wantarray) { + @res = $self->next::method(@_); + } + elsif (defined $wantarray) { + $res[0] = $self->next::method(@_); } else { $self->next::method(@_); } - }; - $self->_insert_blobs($source, $blob_cols, $to_insert) if $blob_cols; + $self->_unset_identity_insert($table, 'update') if $is_identity_update; + } - $guard->commit if $guard; + $guard->commit; - return $updated_cols; + return $wantarray ? @res : $res[0]; } -sub update { - my $self = shift; - my ($source, $fields, $where) = @_; +### the insert_bulk partially stolen from DBI/MSSQL.pm - my $wantarray = wantarray; +sub _set_identity_insert { + my ($self, $table, $op) = @_; - my $blob_cols = $self->_remove_blob_cols($source, $fields); + my $sql = sprintf ( + 'SET IDENTITY_%s %s ON', + (uc($op) || 'INSERT'), + $self->sql_maker->_quote ($table), + ); -# update+blob update(s) done atomically - my $guard = $self->txn_scope_guard if $blob_cols; + $self->_query_start($sql); - my @res; - if ($wantarray) { - @res = $self->next::method(@_); + my $dbh = $self->_get_dbh; + eval { $dbh->do ($sql) }; + my $exception = $@; + + $self->_query_end($sql); + + if ($exception) { + $self->throw_exception (sprintf "Error executing '%s': %s", + $sql, + $dbh->errstr, + ); } - elsif (defined $wantarray) { - $res[0] = $self->next::method(@_); +} + +sub _unset_identity_insert { + my ($self, $table, $op) = @_; + + my $sql = sprintf ( + 'SET IDENTITY_%s %s OFF', + (uc($op) || 'INSERT'), + $self->sql_maker->_quote ($table), + ); + + $self->_query_start($sql); + + my $dbh = $self->_get_dbh; + $dbh->do ($sql); + + $self->_query_end($sql); +} + +# for tests +sub _can_insert_bulk { 1 } + +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; + + 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 +array inserts. +EOF + $self->_bulk_disabled_due_to_coderef_connect_info_warned(1); } - else { + + if (not $use_bulk_api) { + my $blob_cols = $self->_remove_blob_cols_array($source, $cols, $data); + + my $dumb_last_insert_id = + $identity_col + && (not $is_identity_insert) + && ($self->_identity_method||'') ne '@@IDENTITY'; + + ($self, my ($guard)) = do { + if ($self->{transaction_depth} == 0 && + ($blob_cols || $dumb_last_insert_id)) { + ($self->_writer_storage, $self->_writer_storage->txn_scope_guard); + } + else { + ($self, undef); + } + }; + + $self->_set_identity_insert ($source->name) if $is_identity_insert; $self->next::method(@_); + $self->_unset_identity_insert ($source->name) if $is_identity_insert; + + if ($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 + ); + } + } + + $guard->commit if $guard; + return; } - $self->_update_blobs($source, $blob_cols, $where) if $blob_cols; +# otherwise, use the bulk API - $guard->commit if $guard; +# rearrange @$data so that columns are in database order + my %orig_idx; + @orig_idx{@$cols} = 0..$#$cols; - return $wantarray ? @res : $res[0]; + my %new_idx; + @new_idx{@source_columns} = 0..$#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_attributes = $self->source_bind_attributes($source); + + foreach my $slice_idx (0..$#source_columns) { + my $col = $source_columns[$slice_idx]; + + my $attributes = $bind_attributes->{$col} + if $bind_attributes && defined $bind_attributes->{$col}; + + my @slice = map $_->[$slice_idx], @new_data; + + $sth->bind_param_array(($slice_idx + 1), \@slice, $attributes); + } + + $bulk->_query_start($sql); + +# this is stolen from DBI::insert_bulk + my $tuple_status = []; + my $rv = eval { $sth->execute_array({ArrayTupleStatus => $tuple_status}) }; + + if (my $err = $@ || $sth->errstr) { + my $i = 0; + ++$i while $i <= $#$tuple_status && !ref $tuple_status->[$i]; + + $self->throw_exception("Unexpected populate error: $err") + if ($i > $#$tuple_status); + + $self->throw_exception(sprintf "%s for populate slice:\n%s", + ($tuple_status->[$i][1] || $err), + $self->_pretty_print ({ + map { $source_columns[$_] => $new_data[$i][$_] } (0 .. $#$cols) + }), + ); + } + + $guard->commit; + $sth->finish; + + $bulk->_query_end($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) = @_; @@ -357,14 +744,47 @@ sub _remove_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} = \"''"; + 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 ($self->_is_lob_type($source->column_info($col)->{data_type})) { + 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, $where) = @_; @@ -385,14 +805,10 @@ sub _update_blobs { @row_to_update{@primary_cols} = @{$where}{@primary_cols}; @rows = \%row_to_update; } else { - my $rs = $source->resultset->search( - $where, - { - result_class => 'DBIx::Class::ResultClass::HashRefInflator', - select => \@primary_cols - } - ); - @rows = $rs->all; # statement must finish + my $cursor = $self->select ($source, \@primary_cols, $where, {}); + @rows = map { + my %row; @row{@primary_cols} = @$_; \%row + } $cursor->all; } for my $row (@rows) { @@ -404,7 +820,7 @@ sub _insert_blobs { my ($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; @@ -413,24 +829,26 @@ sub _insert_blobs { unless @primary_cols; if ((grep { defined $row{$_} } @primary_cols) != @primary_cols) { - if (@primary_cols == 1) { - my $col = $primary_cols[0]; - $row{$col} = $self->last_insert_id($source, $col); - } else { - croak "Cannot update TEXT/IMAGE column(s) without primary key values"; - } + croak "Cannot update TEXT/IMAGE column(s) without primary key values"; } for my $col (keys %$blob_cols) { my $blob = $blob_cols->{$col}; my %where = map { ($_, $row{$_}) } @primary_cols; - my $cursor = $source->resultset->search(\%where, { - select => [$col] - })->cursor; + + my $cursor = $self->select ($source, [$col], \%where, {}); $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 $sth->errstr; @@ -465,6 +883,26 @@ sub _insert_blobs { } } +sub _insert_blobs_array { + my ($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: @@ -516,10 +954,18 @@ 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 { @@ -608,6 +1054,9 @@ no concurrency issues with getting the inserted identity value using C as it's a session variable. @@ -633,7 +1082,7 @@ For example, this will not work: }); Transactions done for inserts in C mode when placeholders are in use -are not affected, as they use an extra database handle to do the insert. +are not affected, as they are done on an extra database handle. Some workarounds: @@ -688,6 +1137,27 @@ 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.