X-Git-Url: http://git.shadowcat.co.uk/gitweb/gitweb.cgi?a=blobdiff_plain;f=lib%2FDBIx%2FClass%2FStorage%2FDBI.pm;h=5a97bb3acef3eb04f357843e73eb86f810b1edd1;hb=a34702e855e9de44e748595a721b171fbb79a836;hp=a5025fcc30d81b7c5b182335a420e1c86a81860b;hpb=3299c939663473c56afb15a4f6024642f30a91c4;p=dbsrgits%2FDBIx-Class.git diff --git a/lib/DBIx/Class/Storage/DBI.pm b/lib/DBIx/Class/Storage/DBI.pm index a5025fc..5a97bb3 100644 --- a/lib/DBIx/Class/Storage/DBI.pm +++ b/lib/DBIx/Class/Storage/DBI.pm @@ -1,10 +1,12 @@ package DBIx::Class::Storage::DBI; # -*- mode: cperl; cperl-indent-level: 2 -*- -use base 'DBIx::Class::Storage'; - use strict; use warnings; + +use base 'DBIx::Class::Storage'; +use mro 'c3'; + use Carp::Clan qw/^DBIx::Class/; use DBI; use DBIx::Class::Storage::DBI::Cursor; @@ -13,14 +15,15 @@ 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 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); @@ -177,6 +180,91 @@ 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 @@ -359,6 +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 @@ -506,8 +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; @@ -551,12 +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 { @@ -624,8 +752,12 @@ sub _populate_dbh { # there is no transaction in progress by definition $self->{transaction_depth} = $self->_dbh_autocommit ? 0 : 1; - my $connection_do = $self->on_connect_do; - $self->_do_connection_actions($connection_do) if $connection_do; + my @actions; + + push @actions, ( $self->on_connect_call || () ); + push @actions, $self->_parse_connect_do ('on_connect_do'); + + $self->_do_connection_actions(connect_call_ => $_) for @actions; } sub _determine_driver { @@ -642,33 +774,51 @@ sub _determine_driver { ($driver) = $self->_dbi_connect_info->[0] =~ /dbi:([^:]+):/i; } - if ($self->load_optional_class("DBIx::Class::Storage::DBI::${driver}")) { - bless $self, "DBIx::Class::Storage::DBI::${driver}"; + 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(); } } } sub _do_connection_actions { - my $self = shift; - my $connection_do = shift; - - if (!ref $connection_do) { - $self->_do_query($connection_do); - } - elsif (ref $connection_do eq 'ARRAY') { - $self->_do_query($_) foreach @$connection_do; - } - elsif (ref $connection_do eq 'CODE') { - $connection_do->($self); - } - else { - $self->throw_exception (sprintf ("Don't know how to process conection actions of type '%s'", ref $connection_do) ); + 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) = @_; @@ -1193,10 +1343,10 @@ sub _select { my $self = shift; # localization is neccessary as - # 1) there is no infrastructure to pass this around (easy to do, but will wait) + # 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(@_)); } @@ -1205,10 +1355,10 @@ sub _select_args_to_query { my $self = shift; # localization is neccessary as - # 1) there is no infrastructure to pass this around (easy to do, but will wait) + # 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 $sql_maker->{_dbic_rs_attrs}; # my ($op, $bind, $ident, $bind_attrs, $select, $cond, $order, $rows, $offset) # = $self->_select_args($ident, $select, $cond, $attrs); @@ -1226,21 +1376,24 @@ sub _select_args_to_query { } sub _select_args { - my ($self, $ident, $select, $condition, $attrs) = @_; + my ($self, $ident, $select, $where, $attrs) = @_; - my $sql_maker = $self->sql_maker; - $sql_maker->{for} = delete $attrs->{for}; + my ($alias2source, $rs_alias) = $self->_resolve_ident_sources ($ident); - my $order = { map - { $attrs->{$_} ? ( $_ => $attrs->{$_} ) : () } - (qw/order_by group_by having _virtual_order_by/ ) + my $sql_maker = $self->sql_maker; + $sql_maker->{_dbic_rs_attrs} = { + %$attrs, + select => $select, + from => $ident, + where => $where, + $rs_alias + ? ( _source_handle => $alias2source->{$rs_alias}->handle ) + : () + , }; - + # calculate bind_attrs before possible $ident mangling my $bind_attrs = {}; - - my $alias2source = $self->_resolve_ident_sources ($ident); - for my $alias (keys %$alias2source) { my $bindtypes = $self->source_bind_attributes ($alias2source->{$alias}) || {}; for my $col (keys %$bindtypes) { @@ -1248,44 +1401,260 @@ sub _select_args { my $fqcn = join ('.', $alias, $col); $bind_attrs->{$fqcn} = $bindtypes->{$col} if $bindtypes->{$col}; - # so that unqualified searches can be bound too - $bind_attrs->{$col} = $bind_attrs->{$fqcn} if $alias eq 'me'; + # 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}; + } } } - # This would be the point to deflate anything found in $condition + # 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}; + } + + 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 @args = ('select', $attrs->{bind}, $ident, $bind_attrs, $select, $condition, $order); - if ($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)); + my $order = { map + { $attrs->{$_} ? ( $_ => $attrs->{$_} ) : () } + (qw/order_by group_by having/ ) + }; - # 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 ('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 $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 = []; + 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; } - return @args; + + # 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} + my $join_root = shift @$from; + my @outer_from = @$from; + + my %inner_joins; + my %join_info = map { $_->[0]{-alias} => $_->[0] } (@$from); + + # in complex search_related chains $alias may *not* be 'me' + # so always include it in the inner join, and also shift away + # from the outer stack, so that the two datasets actually do + # meet + if ($join_root->{-alias} ne $alias) { + $inner_joins{$alias} = 1; + + while (@outer_from && $outer_from[0][0]{-alias} ne $alias) { + shift @outer_from; + } + if (! @outer_from) { + $self->throw_exception ("Unable to find '$alias' in the {from} stack, something is wrong"); + } + + shift @outer_from; # the new subquery will represent this alias, so get rid of it + } + + + # 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 + 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 + ); + + # put it in the new {from} + unshift @outer_from, { + -alias => $alias, + -source_handle => $join_root->{-source_handle}, + $alias => $subq, + }; + + # 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') { @@ -1293,6 +1662,7 @@ sub _resolve_ident_sources { my $tabinfo; if (ref $_ eq 'HASH') { $tabinfo = $_; + $rs_alias = $tabinfo->{-alias}; } if (ref $_ eq 'ARRAY' and ref $_->[0] eq 'HASH') { $tabinfo = $_->[0]; @@ -1303,7 +1673,7 @@ sub _resolve_ident_sources { } } - return $alias2source; + return ($alias2source, $rs_alias); } # Takes $ident, \@column_names @@ -1315,81 +1685,50 @@ sub _resolve_ident_sources { # my $col_sources = $self->_resolve_column_info($ident, [map $_->[0], @{$bind}]); sub _resolve_column_info { my ($self, $ident, $colnames) = @_; - my $alias2src = $self->_resolve_ident_sources($ident); - my $name_sep = $self->_sql_maker_opts->{name_sep} || '.'; - my %return; - foreach my $col (@{$colnames}) { - $col =~ m/^([^\Q${name_sep}\E]*)\Q${name_sep}\E/; - my $alias = $1 || 'me'; - my $rsrc = $alias2src->{$alias}; - $return{$col} = $rsrc && { %{$rsrc->column_info($col)}, -result_source => $rsrc }; - } - return \%return; -} + my ($alias2src, $root_alias) = $self->_resolve_ident_sources($ident); -sub count { - my ($self, $source, $attrs) = @_; + my $sep = $self->_sql_maker_opts->{name_sep} || '.'; + $sep = "\Q$sep\E"; - my $tmp_attrs = { %$attrs }; - - # take off any pagers, record_filter is cdbi, and no point of ordering a count - delete $tmp_attrs->{$_} for (qw/select as rows offset page order_by record_filter/); - - # overwrite the selector - $tmp_attrs->{select} = { count => '*' }; - - my $tmp_rs = $source->resultset_class->new($source, $tmp_attrs); - my ($count) = $tmp_rs->cursor->next; - - # if the offset/rows attributes are still present, we did not use - # a subquery, so we need to make the calculations in software - $count -= $attrs->{offset} if $attrs->{offset}; - $count = $attrs->{rows} if $attrs->{rows} and $attrs->{rows} < $count; - $count = 0 if ($count < 0); - - return $count; -} + my (%return, %converted); + foreach my $col (@$colnames) { + my ($alias, $colname) = $col =~ m/^ (?: ([^$sep]+) $sep)? (.+) $/x; -sub count_grouped { - my ($self, $source, $attrs) = @_; + # 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; - # copy for the subquery, we need to do some adjustments to it too - my $sub_attrs = { %$attrs }; - - # these can not go in the subquery, and there is no point of ordering it - delete $sub_attrs->{$_} for qw/prefetch collapse select as order_by/; - - # if we prefetch, we group_by primary keys only as this is what we would get out of the rs via ->next/->all - # simply deleting group_by suffices, as the code below will re-fill it - # Note: we check $attrs, as $sub_attrs has collapse deleted - if (ref $attrs->{collapse} and keys %{$attrs->{collapse}} ) { - delete $sub_attrs->{group_by}; + my $rsrc = $alias2src->{$alias}; + $return{$col} = $rsrc && { %{$rsrc->column_info($colname)}, -result_source => $rsrc }; } + return \%return; +} - $sub_attrs->{group_by} ||= [ map { "$attrs->{alias}.$_" } ($source->primary_columns) ]; - $sub_attrs->{select} = $self->_grouped_count_select ($source, $sub_attrs); - - $attrs->{from} = [{ - count_subq => $source->resultset_class->new ($source, $sub_attrs )->as_query - }]; - - # the subquery replaces this - delete $attrs->{$_} for qw/where bind prefetch collapse group_by having having_bind rows offset page pager/; - - return $self->count ($source, $attrs); +# Returns a counting SELECT for a simple count +# query. Abstracted so that a storage could override +# this to { count => 'firstcol' } or whatever makes +# sense as a performance optimization +sub _count_select { + #my ($self, $source, $rs_attrs) = @_; + return { count => '*' }; } +# Returns a SELECT which will end up in the subselect +# There may or may not be a group_by, as the subquery +# might have been called to accomodate a limit # -# Returns a SELECT to go with a supplied GROUP BY -# (caled by count_grouped so a group_by is present) -# Most databases expect them to match, but some -# choke in various ways. +# Most databases would be happy with whatever ends up +# here, but some choke in various ways. # -sub _grouped_count_select { - my ($self, $source, $rs_args) = @_; - return $rs_args->{group_by}; +sub _subq_count_select { + my ($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) = @_;