X-Git-Url: http://git.shadowcat.co.uk/gitweb/gitweb.cgi?a=blobdiff_plain;f=lib%2FDBIx%2FClass%2FStorage%2FDBI.pm;h=84969f62eb8ce19aa51b15f0b2dc8c5972b1cac7;hb=086901c30dcd1c87b678c44bfd1d42ace0b5e18f;hp=7cb026da21a7457366b8a17be687657ac6bd5468;hpb=6f4ddea126a20d67bc599a804d097ed8d3c086aa;p=dbsrgits%2FDBIx-Class.git diff --git a/lib/DBIx/Class/Storage/DBI.pm b/lib/DBIx/Class/Storage/DBI.pm index 7cb026d..84969f6 100644 --- a/lib/DBIx/Class/Storage/DBI.pm +++ b/lib/DBIx/Class/Storage/DBI.pm @@ -1,26 +1,29 @@ package DBIx::Class::Storage::DBI; # -*- mode: cperl; cperl-indent-level: 2 -*- +use strict; +use warnings; + use base 'DBIx::Class::Storage'; +use mro 'c3'; -use strict; -use warnings; use Carp::Clan qw/^DBIx::Class/; use DBI; -use DBIx::Class::SQLAHacks; use DBIx::Class::Storage::DBI::Cursor; use DBIx::Class::Storage::Statistics; -use Scalar::Util qw/blessed weaken/; +use Scalar::Util(); +use List::Util(); __PACKAGE__->mk_group_accessors('simple' => - qw/_connect_info _dbi_connect_info _dbh _sql_maker _sql_maker_opts - _conn_pid _conn_tid transaction_depth _dbh_autocommit savepoints/ + qw/_connect_info _dbi_connect_info _dbh _sql_maker _sql_maker_opts _conn_pid + _conn_tid transaction_depth _dbh_autocommit _driver_determined savepoints/ ); # the values for these accessors are picked out (and deleted) from # the attribute hashref passed to connect_info my @storage_options = qw/ - on_connect_do on_disconnect_do disable_sth_caching unsafe auto_savepoint + on_connect_call on_disconnect_call on_connect_do on_disconnect_do + disable_sth_caching unsafe auto_savepoint /; __PACKAGE__->mk_group_accessors('simple' => @storage_options); @@ -89,8 +92,8 @@ recognized by DBIx::Class: =item * -A single code reference which returns a connected -L optionally followed by +A single code reference which returns a connected +L optionally followed by L recognized by DBIx::Class: @@ -109,7 +112,7 @@ mixed together: %extra_attributes, }]; -This is particularly useful for L based applications, allowing the +This is particularly useful for L based applications, allowing the following config (L style): @@ -128,7 +131,7 @@ Please note that the L docs recommend that you always explicitly set C to either I<0> or I<1>. L further recommends that it be set to I<1>, and that you perform transactions via our L method. L will set it -to I<1> if you do not do explicitly set it to zero. This is the default +to I<1> if you do not do explicitly set it to zero. This is the default for most DBDs. See L for details. =head3 DBIx::Class specific connection attributes @@ -153,6 +156,10 @@ the database. Its value may contain: =over +=item a scalar + +This contains one SQL statement to execute. + =item an array reference This contains SQL statements to execute in order. Each element contains @@ -173,12 +180,97 @@ immediately before disconnecting from the database. Note, this only runs if you explicitly call L on the storage object. +=item on_connect_call + +A more generalized form of L that calls the specified +C methods in your storage driver. + + on_connect_do => 'select 1' + +is equivalent to: + + on_connect_call => [ [ do_sql => 'select 1' ] ] + +Its values may contain: + +=over + +=item a scalar + +Will call the C method. + +=item a code reference + +Will execute C<< $code->($storage) >> + +=item an array reference + +Each value can be a method name or code reference. + +=item an array of arrays + +For each array, the first item is taken to be the C method name +or code reference, and the rest are parameters to it. + +=back + +Some predefined storage methods you may use: + +=over + +=item do_sql + +Executes a SQL string or a code reference that returns a SQL string. This is +what L and L use. + +It can take: + +=over + +=item a scalar + +Will execute the scalar as SQL. + +=item an arrayref + +Taken to be arguments to L, the SQL string optionally followed by the +attributes hashref and bind values. + +=item a code reference + +Will execute C<< $code->($storage) >> and execute the return array refs as +above. + +=back + +=item datetime_setup + +Execute any statements necessary to initialize the database session to return +and accept datetime/timestamp values used with +L. + +Only necessary for some databases, see your specific storage driver for +implementation details. + +=back + +=item on_disconnect_call + +Takes arguments in the same form as L and executes them +immediately before disconnecting from the database. + +Calls the C methods as opposed to the +C methods called by L. + +Note, this only runs if you explicitly call L on the +storage object. + =item disable_sth_caching If set to a true value, this option will disable the caching of statement handles via L. -=item limit_dialect +=item limit_dialect Sets the limit dialect. This is useful for JDBC-bridge among others where the remote SQL-dialect cannot be determined by the name of the @@ -186,7 +278,7 @@ driver alone. See also L. =item quote_char -Specifies what characters to use to quote table and column names. If +Specifies what characters to use to quote table and column names. If you use this you will want to specify L as well. C expects either a single character, in which case is it @@ -198,8 +290,8 @@ SQL Server you should use C<< quote_char => [qw/[ ]/] >>. =item name_sep -This only needs to be used in conjunction with C, and is used to -specify the charecter that seperates elements (schemas, tables, columns) from +This only needs to be used in conjunction with C, and is used to +specify the charecter that seperates elements (schemas, tables, columns) from each other. In most cases this is simply a C<.>. The consequences of not supplying this value is that L @@ -355,6 +447,34 @@ sub connect_info { This method is deprecated in favour of setting via L. +=cut + +=head2 on_disconnect_do + +This method is deprecated in favour of setting via L. + +=cut + +sub _parse_connect_do { + my ($self, $type) = @_; + + my $val = $self->$type; + return () if not defined $val; + + my @res; + + if (not ref($val)) { + push @res, [ 'do_sql', $val ]; + } elsif (ref($val) eq 'CODE') { + push @res, $val; + } elsif (ref($val) eq 'ARRAY') { + push @res, map { [ 'do_sql', $_ ] } @$val; + } else { + $self->throw_exception("Invalid type for $type: ".ref($val)); + } + + return \@res; +} =head2 dbh_do @@ -502,8 +622,12 @@ sub disconnect { my ($self) = @_; if( $self->connected ) { - my $connection_do = $self->on_disconnect_do; - $self->_do_connection_actions($connection_do) if ref($connection_do); + my @actions; + + push @actions, ( $self->on_disconnect_call || () ); + push @actions, $self->_parse_connect_do ('on_disconnect_do'); + + $self->_do_connection_actions(disconnect_call_ => $_) for @actions; $self->_dbh->rollback unless $self->_dbh_autocommit; $self->_dbh->disconnect; @@ -547,12 +671,20 @@ sub connected { $self->_verify_pid; return 0 if !$self->_dbh; } - return ($dbh->FETCH('Active') && $dbh->ping); + return ($dbh->FETCH('Active') && $self->_ping); } return 0; } +sub _ping { + my $self = shift; + + my $dbh = $self->_dbh or return 0; + + return $dbh->ping; +} + # handle pid changes correctly # NOTE: assumes $self->_dbh is a valid $dbh sub _verify_pid { @@ -590,7 +722,7 @@ sub dbh { sub _sql_maker_args { my ($self) = @_; - + return ( bindtype=>'columns', array_datatypes => 1, limit_dialect => $self->dbh, %{$self->_sql_maker_opts} ); } @@ -598,6 +730,7 @@ sub sql_maker { my ($self) = @_; unless ($self->_sql_maker) { my $sql_maker_class = $self->sql_maker_class; + $self->ensure_class_loaded ($sql_maker_class); $self->_sql_maker($sql_maker_class->new( $self->_sql_maker_args )); } return $self->_sql_maker; @@ -610,39 +743,86 @@ sub _populate_dbh { my @info = @{$self->_dbi_connect_info || []}; $self->_dbh($self->_connect(@info)); + $self->_conn_pid($$); + $self->_conn_tid(threads->tid) if $INC{'threads.pm'}; + + $self->_determine_driver; + # Always set the transaction depth on connect, since # there is no transaction in progress by definition $self->{transaction_depth} = $self->_dbh_autocommit ? 0 : 1; - if(ref $self eq 'DBIx::Class::Storage::DBI') { - my $driver = $self->_dbh->{Driver}->{Name}; - if ($self->load_optional_class("DBIx::Class::Storage::DBI::${driver}")) { - bless $self, "DBIx::Class::Storage::DBI::${driver}"; - $self->_rebless(); - } - } + my @actions; - $self->_conn_pid($$); - $self->_conn_tid(threads->tid) if $INC{'threads.pm'}; + push @actions, ( $self->on_connect_call || () ); + push @actions, $self->_parse_connect_do ('on_connect_do'); - my $connection_do = $self->on_connect_do; - $self->_do_connection_actions($connection_do) if ref($connection_do); + $self->_do_connection_actions(connect_call_ => $_) for @actions; } -sub _do_connection_actions { - my $self = shift; - my $connection_do = shift; +sub _determine_driver { + my ($self) = @_; + + if (not $self->_driver_determined) { + if (ref($self) eq __PACKAGE__) { + my $driver; + + if ($self->_dbh) { # we are connected + $driver = $self->_dbh->{Driver}{Name}; + } else { + # try to use dsn to not require being connected, the driver may still + # force a connection in _rebless to determine version + ($driver) = $self->_dbi_connect_info->[0] =~ /dbi:([^:]+):/i; + } - if (ref $connection_do eq 'ARRAY') { - $self->_do_query($_) foreach @$connection_do; + my $storage_class = "DBIx::Class::Storage::DBI::${driver}"; + if ($self->load_optional_class($storage_class)) { + mro::set_mro($storage_class, 'c3'); + bless $self, $storage_class; + $self->_rebless(); + } + } + + $self->_driver_determined(1); } - elsif (ref $connection_do eq 'CODE') { - $connection_do->($self); +} + +sub _do_connection_actions { + my $self = shift; + my $method_prefix = shift; + my $call = shift; + + if (not ref($call)) { + my $method = $method_prefix . $call; + $self->$method(@_); + } elsif (ref($call) eq 'CODE') { + $self->$call(@_); + } elsif (ref($call) eq 'ARRAY') { + if (ref($call->[0]) ne 'ARRAY') { + $self->_do_connection_actions($method_prefix, $_) for @$call; + } else { + $self->_do_connection_actions($method_prefix, @$_) for @$call; + } + } else { + $self->throw_exception (sprintf ("Don't know how to process conection actions of type '%s'", ref($call)) ); } return $self; } +sub connect_call_do_sql { + my $self = shift; + $self->_do_query(@_); +} + +sub disconnect_call_do_sql { + my $self = shift; + $self->_do_query(@_); +} + +# override in db-specific backend when necessary +sub connect_call_datetime_setup { 1 } + sub _do_query { my ($self, $action) = @_; @@ -691,7 +871,7 @@ sub _connect { if($dbh && !$self->unsafe) { my $weak_self = $self; - weaken($weak_self); + Scalar::Util::weaken($weak_self); $dbh->{HandleError} = sub { if ($weak_self) { $weak_self->throw_exception("DBI Exception: $_[0]"); @@ -727,11 +907,11 @@ sub svp_begin { $self->throw_exception ("Your Storage implementation doesn't support savepoints") unless $self->can('_svp_begin'); - + push @{ $self->{savepoints} }, $name; $self->debugobj->svp_begin($name) if $self->debug; - + return $self->_svp_begin($name); } @@ -791,7 +971,7 @@ sub svp_rollback { } $self->debugobj->svp_rollback($name) if $self->debug; - + return $self->_svp_rollback($name); } @@ -872,7 +1052,7 @@ sub txn_rollback { sub _prep_for_execute { my ($self, $op, $extra_bind, $ident, $args) = @_; - if( blessed($ident) && $ident->isa("DBIx::Class::ResultSource") ) { + if( Scalar::Util::blessed($ident) && $ident->isa("DBIx::Class::ResultSource") ) { $ident = $ident->from(); } @@ -884,6 +1064,7 @@ sub _prep_for_execute { return ($sql, \@bind); } + sub _fix_bind_params { my ($self, @bind) = @_; @@ -905,7 +1086,7 @@ sub _query_start { if ( $self->debug ) { @bind = $self->_fix_bind_params(@bind); - + $self->debugobj->query_start( $sql, @bind ); } } @@ -928,7 +1109,7 @@ sub _dbh_execute { my $sth = $self->sth($sql,$op); - my $placeholder_index = 1; + my $placeholder_index = 1; foreach my $bound (@$bind) { my $attributes = {}; @@ -964,13 +1145,18 @@ sub _execute { sub insert { my ($self, $source, $to_insert) = @_; - - my $ident = $source->from; + +# redispatch to insert method of storage we reblessed into, if necessary + if (not $self->_driver_determined) { + $self->_determine_driver; + goto $self->can('insert'); + } + + my $ident = $source->from; my $bind_attributes = $self->source_bind_attributes($source); my $updated_cols = {}; - $self->ensure_connected; foreach my $col ( $source->columns ) { if ( !defined $to_insert->{$col} ) { my $col_info = $source->column_info($col); @@ -987,7 +1173,7 @@ sub insert { } ## Still not quite perfect, and EXPERIMENTAL -## Currently it is assumed that all values passed will be "normal", i.e. not +## Currently it is assumed that all values passed will be "normal", i.e. not ## scalar refs, or at least, all the same type as the first set, the statement is ## only prepped once. sub insert_bulk { @@ -996,7 +1182,7 @@ sub insert_bulk { my $table = $source->from; @colvalues{@$cols} = (0..$#$cols); my ($sql, @bind) = $self->sql_maker->insert($table, \%colvalues); - + $self->_query_start( $sql, @bind ); my $sth = $self->sth($sql); @@ -1009,7 +1195,7 @@ sub insert_bulk { my $bind_attributes = $self->source_bind_attributes($source); ## Bind the values and execute - my $placeholder_index = 1; + my $placeholder_index = 1; foreach my $bound (@bind) { @@ -1026,7 +1212,27 @@ sub insert_bulk { $sth->bind_param_array( $placeholder_index, [@data], $attributes ); $placeholder_index++; } - my $rv = $sth->execute_array({ArrayTupleStatus => $tuple_status}); + my $rv = eval { $sth->execute_array({ArrayTupleStatus => $tuple_status}) }; + if (my $err = $@) { + my $i = 0; + ++$i while $i <= $#$tuple_status && !ref $tuple_status->[$i]; + + $self->throw_exception($sth->errstr || "Unexpected populate error: $err") + if ($i > $#$tuple_status); + + require Data::Dumper; + local $Data::Dumper::Terse = 1; + local $Data::Dumper::Indent = 1; + local $Data::Dumper::Useqq = 1; + local $Data::Dumper::Quotekeys = 0; + + $self->throw_exception(sprintf "%s for populate slice:\n%s", + $tuple_status->[$i][1], + Data::Dumper::Dumper( + { map { $cols->[$_] => $data->[$i][$_] } (0 .. $#$cols) } + ), + ); + } $self->throw_exception($sth->errstr) if !$rv; $self->_query_end( $sql, @bind ); @@ -1037,7 +1243,7 @@ sub update { my $self = shift @_; my $source = shift @_; my $bind_attributes = $self->source_bind_attributes($source); - + return $self->_execute('update' => [], $source, $bind_attributes, @_); } @@ -1045,56 +1251,528 @@ sub update { sub delete { my $self = shift @_; my $source = shift @_; - - my $bind_attrs = {}; ## If ever it's needed... - + + my $bind_attrs = $self->source_bind_attributes($source); + return $self->_execute('delete' => [], $source, $bind_attrs, @_); } +# We were sent here because the $rs contains a complex search +# which will require a subquery to select the correct rows +# (i.e. joined or limited resultsets) +# +# Genarating a single PK column subquery is trivial and supported +# by all RDBMS. However if we have a multicolumn PK, things get ugly. +# Look at _multipk_update_delete() +sub _subq_update_delete { + my $self = shift; + my ($rs, $op, $values) = @_; + + my $rsrc = $rs->result_source; + + # we already check this, but double check naively just in case. Should be removed soon + my $sel = $rs->_resolved_attrs->{select}; + $sel = [ $sel ] unless ref $sel eq 'ARRAY'; + my @pcols = $rsrc->primary_columns; + if (@$sel != @pcols) { + $self->throw_exception ( + 'Subquery update/delete can not be called on resultsets selecting a' + .' number of columns different than the number of primary keys' + ); + } + + if (@pcols == 1) { + return $self->$op ( + $rsrc, + $op eq 'update' ? $values : (), + { $pcols[0] => { -in => $rs->as_query } }, + ); + } + + else { + return $self->_multipk_update_delete (@_); + } +} + +# ANSI SQL does not provide a reliable way to perform a multicol-PK +# resultset update/delete involving subqueries. So by default resort +# to simple (and inefficient) delete_all style per-row opearations, +# while allowing specific storages to override this with a faster +# implementation. +# +sub _multipk_update_delete { + return shift->_per_row_update_delete (@_); +} + +# This is the default loop used to delete/update rows for multi PK +# resultsets, and used by mysql exclusively (because it can't do anything +# else). +# +# We do not use $row->$op style queries, because resultset update/delete +# is not expected to cascade (this is what delete_all/update_all is for). +# +# There should be no race conditions as the entire operation is rolled +# in a transaction. +# +sub _per_row_update_delete { + my $self = shift; + my ($rs, $op, $values) = @_; + + my $rsrc = $rs->result_source; + my @pcols = $rsrc->primary_columns; + + my $guard = $self->txn_scope_guard; + + # emulate the return value of $sth->execute for non-selects + my $row_cnt = '0E0'; + + my $subrs_cur = $rs->cursor; + while (my @pks = $subrs_cur->next) { + + my $cond; + for my $i (0.. $#pcols) { + $cond->{$pcols[$i]} = $pks[$i]; + } + + $self->$op ( + $rsrc, + $op eq 'update' ? $values : (), + $cond, + ); + + $row_cnt++; + } + + $guard->commit; + + return $row_cnt; +} + sub _select { my $self = shift; + + # localization is neccessary as + # 1) there is no infrastructure to pass this around before SQLA2 + # 2) _select_args sets it and _prep_for_execute consumes it my $sql_maker = $self->sql_maker; - local $sql_maker->{for}; + local $sql_maker->{_dbic_rs_attrs}; + return $self->_execute($self->_select_args(@_)); } +sub _select_args_to_query { + my $self = shift; + + # localization is neccessary as + # 1) there is no infrastructure to pass this around before SQLA2 + # 2) _select_args sets it and _prep_for_execute consumes it + my $sql_maker = $self->sql_maker; + local $sql_maker->{_dbic_rs_attrs}; + + # my ($op, $bind, $ident, $bind_attrs, $select, $cond, $order, $rows, $offset) + # = $self->_select_args($ident, $select, $cond, $attrs); + my ($op, $bind, $ident, $bind_attrs, @args) = + $self->_select_args(@_); + + # my ($sql, $prepared_bind) = $self->_prep_for_execute($op, $bind, $ident, [ $select, $cond, $order, $rows, $offset ]); + my ($sql, $prepared_bind) = $self->_prep_for_execute($op, $bind, $ident, \@args); + $prepared_bind ||= []; + + return wantarray + ? ($sql, $prepared_bind, $bind_attrs) + : \[ "($sql)", @$prepared_bind ] + ; +} + sub _select_args { - my ($self, $ident, $select, $condition, $attrs) = @_; - my $order = $attrs->{order_by}; + my ($self, $ident, $select, $where, $attrs) = @_; + + my ($alias2source, $rs_alias) = $self->_resolve_ident_sources ($ident); - my $for = delete $attrs->{for}; my $sql_maker = $self->sql_maker; - $sql_maker->{for} = $for; + $sql_maker->{_dbic_rs_attrs} = { + %$attrs, + select => $select, + from => $ident, + where => $where, + $rs_alias + ? ( _source_handle => $alias2source->{$rs_alias}->handle ) + : () + , + }; - if (exists $attrs->{group_by} || $attrs->{having}) { - $order = { - group_by => $attrs->{group_by}, - having => $attrs->{having}, - ($order ? (order_by => $order) : ()) - }; + # calculate bind_attrs before possible $ident mangling + my $bind_attrs = {}; + for my $alias (keys %$alias2source) { + my $bindtypes = $self->source_bind_attributes ($alias2source->{$alias}) || {}; + for my $col (keys %$bindtypes) { + + my $fqcn = join ('.', $alias, $col); + $bind_attrs->{$fqcn} = $bindtypes->{$col} if $bindtypes->{$col}; + + # Unqialified column names are nice, but at the same time can be + # rather ambiguous. What we do here is basically go along with + # the loop, adding an unqualified column slot to $bind_attrs, + # alongside the fully qualified name. As soon as we encounter + # another column by that name (which would imply another table) + # we unset the unqualified slot and never add any info to it + # to avoid erroneous type binding. If this happens the users + # only choice will be to fully qualify his column name + + if (exists $bind_attrs->{$col}) { + $bind_attrs->{$col} = {}; + } + else { + $bind_attrs->{$col} = $bind_attrs->{$fqcn}; + } + } } - my $bind_attrs = {}; ## Future support - my @args = ('select', $attrs->{bind}, $ident, $bind_attrs, $select, $condition, $order); - if ($attrs->{software_limit} || - $self->sql_maker->_default_limit_syntax eq "GenericSubQ") { - $attrs->{software_limit} = 1; - } else { + + # adjust limits + if ( + $attrs->{software_limit} + || + $sql_maker->_default_limit_syntax eq "GenericSubQ" + ) { + $attrs->{software_limit} = 1; + } + else { $self->throw_exception("rows attribute must be positive if present") if (defined($attrs->{rows}) && !($attrs->{rows} > 0)); # MySQL actually recommends this approach. I cringe. $attrs->{rows} = 2**48 if not defined $attrs->{rows} and defined $attrs->{offset}; - push @args, $attrs->{rows}, $attrs->{offset}; } - return @args; + + my @limit; + + # see if we need to tear the prefetch apart (either limited has_many or grouped prefetch) + # otherwise delegate the limiting to the storage, unless software limit was requested + if ( + ( $attrs->{rows} && keys %{$attrs->{collapse}} ) + || + ( $attrs->{group_by} && @{$attrs->{group_by}} && + $attrs->{_prefetch_select} && @{$attrs->{_prefetch_select}} ) + ) { + ($ident, $select, $where, $attrs) + = $self->_adjust_select_args_for_complex_prefetch ($ident, $select, $where, $attrs); + } + elsif (! $attrs->{software_limit} ) { + push @limit, $attrs->{rows}, $attrs->{offset}; + } + +### + # This would be the point to deflate anything found in $where + # (and leave $attrs->{bind} intact). Problem is - inflators historically + # expect a row object. And all we have is a resultsource (it is trivial + # to extract deflator coderefs via $alias2source above). + # + # I don't see a way forward other than changing the way deflators are + # invoked, and that's just bad... +### + + my $order = { map + { $attrs->{$_} ? ( $_ => $attrs->{$_} ) : () } + (qw/order_by group_by having/ ) + }; + + return ('select', $attrs->{bind}, $ident, $bind_attrs, $select, $where, $order, @limit); +} + +# +# This is the code producing joined subqueries like: +# SELECT me.*, other.* FROM ( SELECT me.* FROM ... ) JOIN other ON ... +# +sub _adjust_select_args_for_complex_prefetch { + my ($self, $from, $select, $where, $attrs) = @_; + + $self->throw_exception ('Complex prefetches are not supported on resultsets with a custom from attribute') + if (ref $from ne 'ARRAY'); + + # copies for mangling + $from = [ @$from ]; + $select = [ @$select ]; + $attrs = { %$attrs }; + + # separate attributes + my $sub_attrs = { %$attrs }; + delete $attrs->{$_} for qw/where bind rows offset group_by having/; + delete $sub_attrs->{$_} for qw/for collapse _prefetch_select _collapse_order_by select as/; + + my $select_root_alias = $attrs->{alias}; + my $sql_maker = $self->sql_maker; + + # create subquery select list - consider only stuff *not* brought in by the prefetch + my $sub_select = []; + my $sub_group_by; + for my $i (0 .. @{$attrs->{select}} - @{$attrs->{_prefetch_select}} - 1) { + my $sel = $attrs->{select}[$i]; + + # alias any functions to the dbic-side 'as' label + # adjust the outer select accordingly + if (ref $sel eq 'HASH' && !$sel->{-select}) { + $sel = { -select => $sel, -as => $attrs->{as}[$i] }; + $select->[$i] = join ('.', $attrs->{alias}, ($attrs->{as}[$i] || "select_$i") ); + } + + push @$sub_select, $sel; + } + + # bring over all non-collapse-induced order_by into the inner query (if any) + # the outer one will have to keep them all + delete $sub_attrs->{order_by}; + if (my $ord_cnt = @{$attrs->{order_by}} - @{$attrs->{_collapse_order_by}} ) { + $sub_attrs->{order_by} = [ + @{$attrs->{order_by}}[ 0 .. $ord_cnt - 1] + ]; + } + + # mangle {from}, keep in mind that $from is "headless" from here on + my $join_root = shift @$from; + + my %inner_joins; + my %join_info = map { $_->[0]{-alias} => $_->[0] } (@$from); + + # in complex search_related chains $select_root_alias may *not* be + # 'me' so always include it in the inner join + $inner_joins{$select_root_alias} = 1 if ($join_root->{-alias} ne $select_root_alias); + + + # decide which parts of the join will remain on the inside + # + # this is not a very viable optimisation, but it was written + # before I realised this, so might as well remain. We can throw + # away _any_ branches of the join tree that are: + # 1) not mentioned in the condition/order + # 2) left-join leaves (or left-join leaf chains) + # Most of the join conditions will not satisfy this, but for real + # complex queries some might, and we might make some RDBMS happy. + # + # + # since we do not have introspectable SQLA, we fall back to ugly + # scanning of raw SQL for WHERE, and for pieces of ORDER BY + # in order to determine what goes into %inner_joins + # It may not be very efficient, but it's a reasonable stop-gap + { + # produce stuff unquoted, so it can be scanned + local $sql_maker->{quote_char}; + my $sep = $self->_sql_maker_opts->{name_sep} || '.'; + $sep = "\Q$sep\E"; + + my @order_by = (map + { ref $_ ? $_->[0] : $_ } + $sql_maker->_order_by_chunks ($sub_attrs->{order_by}) + ); + + my $where_sql = $sql_maker->where ($where); + my $select_sql = $sql_maker->_recurse_fields ($sub_select); + + # sort needed joins + for my $alias (keys %join_info) { + + # any table alias found on a column name in where or order_by + # gets included in %inner_joins + # Also any parent joins that are needed to reach this particular alias + for my $piece ($select_sql, $where_sql, @order_by ) { + if ($piece =~ /\b $alias $sep/x) { + $inner_joins{$alias} = 1; + } + } + } + } + + # scan for non-leaf/non-left joins and mark as needed + # also mark all ancestor joins that are needed to reach this particular alias + # (e.g. join => { cds => 'tracks' } - tracks will bring cds too ) + # + # traverse by the size of the -join_path i.e. reverse depth first + for my $alias (sort { @{$join_info{$b}{-join_path}} <=> @{$join_info{$a}{-join_path}} } (keys %join_info) ) { + + my $j = $join_info{$alias}; + $inner_joins{$alias} = 1 if (! $j->{-join_type} || ($j->{-join_type} !~ /^left$/i) ); + + if ($inner_joins{$alias}) { + $inner_joins{$_} = 1 for (@{$j->{-join_path}}); + } + } + + # construct the inner $from for the subquery + my $inner_from = [ $join_root ]; + for my $j (@$from) { + push @$inner_from, $j if $inner_joins{$j->[0]{-alias}}; + } + + # if a multi-type join was needed in the subquery ("multi" is indicated by + # presence in {collapse}) - add a group_by to simulate the collapse in the subq + unless ($sub_attrs->{group_by}) { + for my $alias (keys %inner_joins) { + + # the dot comes from some weirdness in collapse + # remove after the rewrite + if ($attrs->{collapse}{".$alias"}) { + $sub_attrs->{group_by} ||= $sub_select; + last; + } + } + } + + # generate the subquery + my $subq = $self->_select_args_to_query ( + $inner_from, + $sub_select, + $where, + $sub_attrs + ); + my $subq_joinspec = { + -alias => $select_root_alias, + -source_handle => $join_root->{-source_handle}, + $select_root_alias => $subq, + }; + + # Generate a new from (really just replace the join slot with the subquery) + # Before we would start the outer chain from the subquery itself (i.e. + # SELECT ... FROM (SELECT ... ) alias JOIN ..., but this turned out to be + # a bad idea for search_related, as the root of the chain was effectively + # lost (i.e. $artist_rs->search_related ('cds'... ) would result in alias + # of 'cds', which would prevent from doing things like order_by artist.*) + # See t/prefetch/via_search_related.t for a better idea + my @outer_from; + if ($join_root->{-alias} eq $select_root_alias) { # just swap the root part and we're done + @outer_from = ( + $subq_joinspec, + @$from, + ) + } + else { # this is trickier + @outer_from = ($join_root); + + for my $j (@$from) { + if ($j->[0]{-alias} eq $select_root_alias) { + push @outer_from, [ + $subq_joinspec, + @{$j}[1 .. $#$j], + ]; + } + else { + push @outer_from, $j; + } + } + } + + # This is totally horrific - the $where ends up in both the inner and outer query + # Unfortunately not much can be done until SQLA2 introspection arrives, and even + # then if where conditions apply to the *right* side of the prefetch, you may have + # to both filter the inner select (e.g. to apply a limit) and then have to re-filter + # the outer select to exclude joins you didin't want in the first place + # + # OTOH it can be seen as a plus: (notes that this query would make a DBA cry ;) + return (\@outer_from, $select, $where, $attrs); +} + +sub _resolve_ident_sources { + my ($self, $ident) = @_; + + my $alias2source = {}; + my $rs_alias; + + # the reason this is so contrived is that $ident may be a {from} + # structure, specifying multiple tables to join + if ( Scalar::Util::blessed($ident) && $ident->isa("DBIx::Class::ResultSource") ) { + # this is compat mode for insert/update/delete which do not deal with aliases + $alias2source->{me} = $ident; + $rs_alias = 'me'; + } + elsif (ref $ident eq 'ARRAY') { + + for (@$ident) { + my $tabinfo; + if (ref $_ eq 'HASH') { + $tabinfo = $_; + $rs_alias = $tabinfo->{-alias}; + } + if (ref $_ eq 'ARRAY' and ref $_->[0] eq 'HASH') { + $tabinfo = $_->[0]; + } + + $alias2source->{$tabinfo->{-alias}} = $tabinfo->{-source_handle}->resolve + if ($tabinfo->{-source_handle}); + } + } + + return ($alias2source, $rs_alias); } +# Takes $ident, \@column_names +# +# returns { $column_name => \%column_info, ... } +# also note: this adds -result_source => $rsrc to the column info +# +# usage: +# my $col_sources = $self->_resolve_column_info($ident, [map $_->[0], @{$bind}]); +sub _resolve_column_info { + my ($self, $ident, $colnames) = @_; + my ($alias2src, $root_alias) = $self->_resolve_ident_sources($ident); + + my $sep = $self->_sql_maker_opts->{name_sep} || '.'; + $sep = "\Q$sep\E"; + + my (%return, %converted); + + if (not $colnames) { + $colnames = [ map { + my $alias = $_; + my $source = $alias2src->{$alias}; + map "${alias}${sep}$_", $source->columns + } keys %$alias2src ]; + +# also add unqualified columns for 'me' table + push @$colnames, $alias2src->{$root_alias}->columns; + } + + foreach my $col (@$colnames) { + my ($alias, $colname) = $col =~ m/^ (?: ([^$sep]+) $sep)? (.+) $/x; + + # deal with unqualified cols - we assume the main alias for all + # unqualified ones, ugly but can't think of anything better right now + $alias ||= $root_alias; + + my $rsrc = $alias2src->{$alias}; + $return{$col} = $rsrc && { %{$rsrc->column_info($colname)}, -result_source => $rsrc }; + } + return \%return; +} + +# Returns a counting SELECT for a simple count +# query. Abstracted so that a storage could override +# this to { count => 'firstcol' } or whatever makes +# sense as a performance optimization +sub _count_select { + #my ($self, $source, $rs_attrs) = @_; + return { count => '*' }; +} + +# Returns a SELECT which will end up in the subselect +# There may or may not be a group_by, as the subquery +# might have been called to accomodate a limit +# +# Most databases would be happy with whatever ends up +# here, but some choke in various ways. +# +sub _subq_count_select { + my ($self, $source, $rs_attrs) = @_; + return $rs_attrs->{group_by} if $rs_attrs->{group_by}; + + my @pcols = map { join '.', $rs_attrs->{alias}, $_ } ($source->primary_columns); + return @pcols ? \@pcols : [ 1 ]; +} + + sub source_bind_attributes { my ($self, $source) = @_; - + my $bind_attributes; foreach my $column ($source->columns) { - + my $data_type = $source->column_info($column)->{data_type} || ''; $bind_attributes->{$column} = $self->bind_attribute_by_data_type($data_type) if $data_type; @@ -1276,7 +1954,28 @@ sub bind_attribute_by_data_type { return; } -=head2 create_ddl_dir +=head2 is_datatype_numeric + +Given a datatype from column_info, returns a boolean value indicating if +the current RDBMS considers it a numeric value. This controls how +L decides whether to mark the column as +dirty - when the datatype is deemed numeric a C<< != >> comparison will +be performed instead of the usual C. + +=cut + +sub is_datatype_numeric { + my ($self, $dt) = @_; + + return 0 unless $dt; + + return $dt =~ /^ (?: + numeric | int(?:eger)? | (?:tiny|small|medium|big)int | dec(?:imal)? | real | float | double (?: \s+ precision)? | (?:big)?serial + ) $/ix; +} + + +=head2 create_ddl_dir (EXPERIMENTAL) =over 4 @@ -1285,24 +1984,61 @@ sub bind_attribute_by_data_type { =back Creates a SQL file based on the Schema, for each of the specified -database types, in the given directory. +database engines in C<\@databases> in the given directory. +(note: specify L names, not L driver names). + +Given a previous version number, this will also create a file containing +the ALTER TABLE statements to transform the previous schema into the +current one. Note that these statements may contain C or +C statements that can potentially destroy data. + +The file names are created using the C method below, please +override this method in your schema if you would like a different file +name format. For the ALTER file, the same format is used, replacing +$version in the name with "$preversion-$version". + +See L for a list of values for C<\%sqlt_args>. +The most common value for this would be C<< { add_drop_table => 1 } >> +to have the SQL produced include a C statement for each table +created. For quoting purposes supply C and +C. + +If no arguments are passed, then the following default values are assumed: + +=over 4 + +=item databases - ['MySQL', 'SQLite', 'PostgreSQL'] + +=item version - $schema->schema_version + +=item directory - './' + +=item preversion - + +=back By default, C<\%sqlt_args> will have { add_drop_table => 1, ignore_constraint_names => 1, ignore_index_names => 1 } -merged with the hash passed in. To disable any of those features, pass in a +merged with the hash passed in. To disable any of those features, pass in a hashref like the following { ignore_constraint_names => 0, # ... other options } + +Note that this feature is currently EXPERIMENTAL and may not work correctly +across all databases, or fully handle complex relationships. + +WARNING: Please check all SQL files created, before applying them. + =cut sub create_ddl_dir { my ($self, $schema, $databases, $version, $dir, $preversion, $sqltargs) = @_; if(!$dir || !-d $dir) { - warn "No directory given, using ./\n"; + carp "No directory given, using ./\n"; $dir = "./"; } $databases ||= ['MySQL', 'SQLite', 'PostgreSQL']; @@ -1312,7 +2048,7 @@ sub create_ddl_dir { $version ||= $schema_version; $sqltargs = { - add_drop_table => 1, + add_drop_table => 1, ignore_constraint_names => 1, ignore_index_names => 1, %{$sqltargs || {}} @@ -1325,7 +2061,8 @@ sub create_ddl_dir { my $sqlt = SQL::Translator->new( $sqltargs ); $sqlt->parser('SQL::Translator::Parser::DBIx::Class'); - my $sqlt_schema = $sqlt->translate({ data => $schema }) or die $sqlt->error; + my $sqlt_schema = $sqlt->translate({ data => $schema }) + or $self->throw_exception ($sqlt->error); foreach my $db (@$databases) { $sqlt->reset(); @@ -1336,13 +2073,13 @@ sub create_ddl_dir { my $filename = $schema->ddl_filename($db, $version, $dir); if (-e $filename && ($version eq $schema_version )) { # if we are dumping the current version, overwrite the DDL - warn "Overwriting existing DDL file - $filename"; + carp "Overwriting existing DDL file - $filename"; unlink($filename); } my $output = $sqlt->translate; if(!$output) { - warn("Failed to translate to $db, skipping. (" . $sqlt->error . ")"); + carp("Failed to translate to $db, skipping. (" . $sqlt->error . ")"); next; } if(!open($file, ">$filename")) { @@ -1351,57 +2088,68 @@ sub create_ddl_dir { } print $file $output; close($file); - + next unless ($preversion); require SQL::Translator::Diff; my $prefilename = $schema->ddl_filename($db, $preversion, $dir); if(!-e $prefilename) { - warn("No previous schema file found ($prefilename)"); + carp("No previous schema file found ($prefilename)"); next; } my $difffile = $schema->ddl_filename($db, $version, $dir, $preversion); if(-e $difffile) { - warn("Overwriting existing diff file - $difffile"); + carp("Overwriting existing diff file - $difffile"); unlink($difffile); } - + my $source_schema; { my $t = SQL::Translator->new($sqltargs); $t->debug( 0 ); $t->trace( 0 ); - $t->parser( $db ) or die $t->error; - my $out = $t->translate( $prefilename ) or die $t->error; + + $t->parser( $db ) + or $self->throw_exception ($t->error); + + my $out = $t->translate( $prefilename ) + or $self->throw_exception ($t->error); + $source_schema = $t->schema; - unless ( $source_schema->name ) { - $source_schema->name( $prefilename ); - } + + $source_schema->name( $prefilename ) + unless ( $source_schema->name ); } - # The "new" style of producers have sane normalization and can support + # The "new" style of producers have sane normalization and can support # diffing a SQL file against a DBIC->SQLT schema. Old style ones don't # And we have to diff parsed SQL against parsed SQL. my $dest_schema = $sqlt_schema; - + unless ( "SQL::Translator::Producer::$db"->can('preprocess_schema') ) { my $t = SQL::Translator->new($sqltargs); $t->debug( 0 ); $t->trace( 0 ); - $t->parser( $db ) or die $t->error; - my $out = $t->translate( $filename ) or die $t->error; + + $t->parser( $db ) + or $self->throw_exception ($t->error); + + my $out = $t->translate( $filename ) + or $self->throw_exception ($t->error); + $dest_schema = $t->schema; + $dest_schema->name( $filename ) unless $dest_schema->name; } - + my $diff = SQL::Translator::Diff::schema_diff($source_schema, $db, $dest_schema, $db, $sqltargs ); - if(!open $file, ">$difffile") { + if(!open $file, ">$difffile") { $self->throw_exception("Can't write to $difffile ($!)"); next; } @@ -1419,8 +2167,9 @@ sub create_ddl_dir { =back Returns the statements used by L and L. -The database driver name is given by C<$type>, though the value from -L is used if it is not specified. + +The L (not L) database driver name can be explicitly +provided in C<$type>, otherwise the result of L is used as default. C<$directory> is used to return statements from files in a previously created L directory and is optional. The filenames are constructed @@ -1444,7 +2193,7 @@ sub deployment_statements { if(-f $filename) { my $file; - open($file, "<$filename") + open($file, "<$filename") or $self->throw_exception("Can't open $filename ($!)"); my @rows = <$file>; close($file); @@ -1459,7 +2208,7 @@ sub deployment_statements { eval qq{use SQL::Translator::Producer::${type}}; $self->throw_exception($@) if $@; - # sources needs to be a parser arg, but for simplicty allow at top level + # sources needs to be a parser arg, but for simplicty allow at top level # coming in $sqltargs->{parser_args}{sources} = delete $sqltargs->{sources} if exists $sqltargs->{sources}; @@ -1484,11 +2233,11 @@ sub deploy { $self->dbh->do($line); # shouldn't be using ->dbh ? }; if ($@) { - warn qq{$@ (running "${line}")}; + carp qq{$@ (running "${line}")}; } $self->_query_end($line); }; - my @statements = $self->deployment_statements($schema, $type, undef, $dir, { no_comments => 1, %{ $sqltargs || {} } } ); + my @statements = $self->deployment_statements($schema, $type, undef, $dir, { %{ $sqltargs || {} }, no_comments => 1 } ); if (@statements > 1) { foreach my $statement (@statements) { $deploy->( $statement ); @@ -1564,7 +2313,7 @@ returned by databases that don't support replication. sub is_replicating { return; - + } =head2 lag_behind_master