X-Git-Url: http://git.shadowcat.co.uk/gitweb/gitweb.cgi?a=blobdiff_plain;f=lib%2FDBIx%2FClass%2FResultSource.pm;h=b4dc288dedd8922f6b95217e5d546f07f03238a0;hb=908aa1bb761ec1da5c061fe9f687598e3f1934bc;hp=d596d2acb6dbe0b00a44cbeed8e09da788ed9cbb;hpb=1605376709663b035385b41828ce13ae3ed45a4d;p=dbsrgits%2FDBIx-Class.git diff --git a/lib/DBIx/Class/ResultSource.pm b/lib/DBIx/Class/ResultSource.pm index d596d2a..b4dc288 100644 --- a/lib/DBIx/Class/ResultSource.pm +++ b/lib/DBIx/Class/ResultSource.pm @@ -3,26 +3,34 @@ package DBIx::Class::ResultSource; use strict; use warnings; +use base 'DBIx::Class'; + use DBIx::Class::ResultSet; use DBIx::Class::ResultSourceHandle; use DBIx::Class::Exception; -use Carp::Clan qw/^DBIx::Class/; +use DBIx::Class::Carp; +use DBIx::Class::GlobalDestruction; use Try::Tiny; use List::Util 'first'; -use Scalar::Util qw/weaken isweak/; -use Storable qw/nfreeze thaw/; +use Scalar::Util qw/blessed weaken isweak/; +use B 'perlstring'; + use namespace::clean; -use base qw/DBIx::Class/; +__PACKAGE__->mk_group_accessors(simple => qw/ + source_name name source_info + _ordered_columns _columns _primaries _unique_constraints + _relationships resultset_attributes + column_info_from_storage +/); -__PACKAGE__->mk_group_accessors('simple' => qw/_ordered_columns - _columns _primaries _unique_constraints name resultset_attributes - from _relationships column_info_from_storage source_info - source_name sqlt_deploy_callback/); +__PACKAGE__->mk_group_accessors(component_class => qw/ + resultset_class + result_class +/); -__PACKAGE__->mk_group_accessors('component_class' => qw/resultset_class - result_class/); +__PACKAGE__->mk_classdata( sqlt_deploy_callback => 'default_sqlt_deploy_hook' ); =head1 NAME @@ -32,18 +40,18 @@ DBIx::Class::ResultSource - Result source object # Create a table based result source, in a result class. - package MyDB::Schema::Result::Artist; + package MyApp::Schema::Result::Artist; use base qw/DBIx::Class::Core/; __PACKAGE__->table('artist'); __PACKAGE__->add_columns(qw/ artistid name /); __PACKAGE__->set_primary_key('artistid'); - __PACKAGE__->has_many(cds => 'MyDB::Schema::Result::CD'); + __PACKAGE__->has_many(cds => 'MyApp::Schema::Result::CD'); 1; # Create a query (view) based result source, in a result class - package MyDB::Schema::Result::Year2000CDs; + package MyApp::Schema::Result::Year2000CDs; use base qw/DBIx::Class::Core/; __PACKAGE__->load_components('InflateColumn::DateTime'); @@ -116,7 +124,6 @@ sub new { $new->{_relationships} = { %{$new->{_relationships}||{}} }; $new->{name} ||= "!!NAME NOT SET!!"; $new->{_columns_info_loaded} ||= 0; - $new->{sqlt_deploy_callback} ||= "default_sqlt_deploy_hook"; return $new; } @@ -254,8 +261,20 @@ generate a new key value. If not specified, L will attempt to retrieve the name of the sequence from the database automatically. +=item retrieve_on_insert + + { retrieve_on_insert => 1 } + +For every column where this is set to true, DBIC will retrieve the RDBMS-side +value upon a new row insertion (normally only the autoincrement PK is +retrieved on insert). C is used automatically if +supported by the underlying storage, otherwise an extra SELECT statement is +executed to retrieve the missing data. + =item auto_nextval + { auto_nextval => 1 } + Set this to a true value for a column whose value is retrieved automatically from a sequence or function (if supported by your Storage driver.) For a sequence, if you do not use a trigger to get the nextval, you have to set the @@ -428,7 +447,7 @@ sub columns { my $columns_info = $source->columns_info; Like L but returns information for the requested columns. If -the optional column-list arrayref is ommitted it returns info on all columns +the optional column-list arrayref is omitted it returns info on all columns currently defined on the ResultSource via L. =cut @@ -634,7 +653,7 @@ sub sequence { my ($self,$seq) = @_; my @pks = $self->primary_columns - or next; + or return; $_->{sequence} = $seq for values %{ $self->columns_info (\@pks) }; @@ -875,12 +894,21 @@ sub unique_constraint_columns { =over -=item Arguments: $callback +=item Arguments: $callback_name | \&callback_code + +=item Return value: $callback_name | \&callback_code =back __PACKAGE__->sqlt_deploy_callback('mycallbackmethod'); + or + + __PACKAGE__->sqlt_deploy_callback(sub { + my ($source_instance, $sqlt_table) = @_; + ... + } ); + An accessor to set a callback to be called during deployment of the schema via L or L. @@ -888,7 +916,7 @@ L. The callback can be set as either a code reference or the name of a method in the current result class. -If not set, the L is called. +Defaults to L. Your callback will be passed the $source object representing the ResultSource instance being deployed, and the @@ -908,19 +936,13 @@ and call L. =head2 default_sqlt_deploy_hook -=over - -=item Arguments: $source, $sqlt_table - -=item Return value: undefined - -=back - -This is the sensible default for L. - -If a method named C exists in your Result class, it -will be called and passed the current C<$source> and the -C<$sqlt_table> being deployed. +This is the default deploy hook implementation which checks if your +current Result class has a C method, and if present +invokes it B. This is to preserve the +semantics of C which was originally designed to expect +the Result class name and the +L<$sqlt_table instance|SQL::Translator::Schema::Table> of the table being +deployed. =cut @@ -1022,6 +1044,20 @@ sub resultset { ); } +=head2 name + +=over 4 + +=item Arguments: None + +=item Result value: $name + +=back + +Returns the name of the result source, which will typically be the table +name. This may be a scalar reference if the result source has a non-standard +name. + =head2 source_name =over 4 @@ -1060,6 +1096,10 @@ Returns an expression of the source to be supplied to storage to specify retrieval from this source. In the case of a database, the required FROM clause contents. +=cut + +sub from { die 'Virtual method!' } + =head2 schema =over 4 @@ -1327,56 +1367,74 @@ L. sub reverse_relationship_info { my ($self, $rel) = @_; - my $rel_info = $self->relationship_info($rel); + + my $rel_info = $self->relationship_info($rel) + or $self->throw_exception("No such relationship '$rel'"); + my $ret = {}; return $ret unless ((ref $rel_info->{cond}) eq 'HASH'); - my @cond = keys(%{$rel_info->{cond}}); - my @refkeys = map {/^\w+\.(\w+)$/} @cond; - my @keys = map {$rel_info->{cond}->{$_} =~ /^\w+\.(\w+)$/} @cond; + my $stripped_cond = $self->__strip_relcond ($rel_info->{cond}); - # Get the related result source for this relationship - my $othertable = $self->related_source($rel); + my $rsrc_schema_moniker = $self->source_name + if try { $self->schema }; + + # this may be a partial schema or something else equally esoteric + my $other_rsrc = try { $self->related_source($rel) } + or return $ret; # Get all the relationships for that source that related to this source # whose foreign column set are our self columns on $rel and whose self - # columns are our foreign columns on $rel. - my @otherrels = $othertable->relationships(); - my $otherrelationship; - foreach my $otherrel (@otherrels) { - # this may be a partial schema with the related source not being - # available at all - my $back = try { $othertable->related_source($otherrel) } or next; - - # did we get back to ourselves? - next unless $back->source_name eq $self->source_name; - - my $otherrel_info = $othertable->relationship_info($otherrel); - my @othertestconds; - - if (ref $otherrel_info->{cond} eq 'HASH') { - @othertestconds = ($otherrel_info->{cond}); - } - elsif (ref $otherrel_info->{cond} eq 'ARRAY') { - @othertestconds = @{$otherrel_info->{cond}}; + # columns are our foreign columns on $rel + foreach my $other_rel ($other_rsrc->relationships) { + + # only consider stuff that points back to us + # "us" here is tricky - if we are in a schema registration, we want + # to use the source_names, otherwise we will use the actual classes + + # the schema may be partial + my $roundtrip_rsrc = try { $other_rsrc->related_source($other_rel) } + or next; + + if ($rsrc_schema_moniker and try { $roundtrip_rsrc->schema } ) { + next unless $rsrc_schema_moniker eq $roundtrip_rsrc->source_name; } else { - next; + next unless $self->result_class eq $roundtrip_rsrc->result_class; } - foreach my $othercond (@othertestconds) { - my @other_cond = keys(%$othercond); - my @other_refkeys = map {/^\w+\.(\w+)$/} @other_cond; - my @other_keys = map {$othercond->{$_} =~ /^\w+\.(\w+)$/} @other_cond; - next if (!$self->_compare_relationship_keys(\@refkeys, \@other_keys) || - !$self->_compare_relationship_keys(\@other_refkeys, \@keys)); - $ret->{$otherrel} = $otherrel_info; - } + my $other_rel_info = $other_rsrc->relationship_info($other_rel); + + # this can happen when we have a self-referential class + next if $other_rel_info eq $rel_info; + + next unless ref $other_rel_info->{cond} eq 'HASH'; + my $other_stripped_cond = $self->__strip_relcond($other_rel_info->{cond}); + + $ret->{$other_rel} = $other_rel_info if ( + $self->_compare_relationship_keys ( + [ keys %$stripped_cond ], [ values %$other_stripped_cond ] + ) + and + $self->_compare_relationship_keys ( + [ values %$stripped_cond ], [ keys %$other_stripped_cond ] + ) + ); } + return $ret; } +# all this does is removes the foreign/self prefix from a condition +sub __strip_relcond { + +{ + map + { map { /^ (?:foreign|self) \. (\w+) $/x } ($_, $_[1]{$_}) } + keys %{$_[1]} + } +} + sub compare_relationship_keys { carp 'compare_relationship_keys is a private method, stop calling it'; my $self = shift; @@ -1385,36 +1443,38 @@ sub compare_relationship_keys { # Returns true if both sets of keynames are the same, false otherwise. sub _compare_relationship_keys { - my ($self, $keys1, $keys2) = @_; - - # Make sure every keys1 is in keys2 - my $found; - foreach my $key (@$keys1) { - $found = 0; - foreach my $prim (@$keys2) { - if ($prim eq $key) { - $found = 1; - last; - } - } - last unless $found; - } +# my ($self, $keys1, $keys2) = @_; + return + join ("\x00", sort @{$_[1]}) + eq + join ("\x00", sort @{$_[2]}) + ; +} - # Make sure every key2 is in key1 - if ($found) { - foreach my $prim (@$keys2) { - $found = 0; - foreach my $key (@$keys1) { - if ($prim eq $key) { - $found = 1; - last; - } - } - last unless $found; +# optionally takes either an arrayref of column names, or a hashref of already +# retrieved colinfos +# returns an arrayref of column names of the shortest unique constraint +# (matching some of the input if any), giving preference to the PK +sub _identifying_column_set { + my ($self, $cols) = @_; + + my %unique = $self->unique_constraints; + my $colinfos = ref $cols eq 'HASH' ? $cols : $self->columns_info($cols||()); + + # always prefer the PK first, and then shortest constraints first + USET: + for my $set (delete $unique{primary}, sort { @$a <=> @$b } (values %unique) ) { + next unless $set && @$set; + + for (@$set) { + next USET unless ($colinfos->{$_} && !$colinfos->{$_}{is_nullable} ); } + + # copy so we can mangle it at will + return [ @$set ]; } - return $found; + return undef; } # Returns the {from} structure used to express JOIN conditions @@ -1430,7 +1490,7 @@ sub _resolve_join { $jpath = [@$jpath]; # copy - if (not defined $join) { + if (not defined $join or not length $join) { return (); } elsif (ref $join eq 'ARRAY') { @@ -1486,14 +1546,14 @@ sub _resolve_join { , -join_path => [@$jpath, { $join => $as } ], -is_single => ( - $rel_info->{attrs}{accessor} - && + (! $rel_info->{attrs}{accessor}) + or first { $rel_info->{attrs}{accessor} eq $_ } (qw/single filter/) ), -alias => $as, -relation_chain_depth => $seen->{-relation_chain_depth} || 0, }, - $self->_resolve_condition($rel_info->{cond}, $as, $alias, $join) + scalar $self->_resolve_condition($rel_info->{cond}, $as, $alias, $join) ]; } } @@ -1546,26 +1606,89 @@ sub resolve_condition { $self->_resolve_condition (@_); } -# Resolves the passed condition to a concrete query fragment. If given an alias, -# returns a join condition; if given an object, inverts that object to produce -# a related conditional from that object. our $UNRESOLVABLE_CONDITION = \ '1 = 0'; +# Resolves the passed condition to a concrete query fragment and a flag +# indicating whether this is a cross-table condition. Also an optional +# list of non-triviail values (notmally conditions) returned as a part +# of a joinfree condition hash sub _resolve_condition { my ($self, $cond, $as, $for, $relname) = @_; - if (ref $cond eq 'CODE') { - my $obj_rel = !!ref $for; + my $obj_rel = !!blessed $for; - return $cond->({ + if (ref $cond eq 'CODE') { + my $relalias = $obj_rel ? 'me' : $as; + + my ($crosstable_cond, $joinfree_cond) = $cond->({ self_alias => $obj_rel ? $as : $for, - foreign_alias => $obj_rel ? 'me' : $as, + foreign_alias => $relalias, self_resultsource => $self, foreign_relname => $relname || ($obj_rel ? $as : $for), self_rowobj => $obj_rel ? $for : undef }); - } elsif (ref $cond eq 'HASH') { + my $cond_cols; + if ($joinfree_cond) { + + # FIXME sanity check until things stabilize, remove at some point + $self->throw_exception ( + "A join-free condition returned for relationship '$relname' without a row-object to chain from" + ) unless $obj_rel; + + # FIXME another sanity check + if ( + ref $joinfree_cond ne 'HASH' + or + first { $_ !~ /^\Q$relalias.\E.+/ } keys %$joinfree_cond + ) { + $self->throw_exception ( + "The join-free condition returned for relationship '$relname' must be a hash " + .'reference with all keys being valid columns on the related result source' + ); + } + + # normalize + for (values %$joinfree_cond) { + $_ = $_->{'='} if ( + ref $_ eq 'HASH' + and + keys %$_ == 1 + and + exists $_->{'='} + ); + } + + # see which parts of the joinfree cond are conditionals + my $relcol_list = { map { $_ => 1 } $self->related_source($relname)->columns }; + + for my $c (keys %$joinfree_cond) { + my ($colname) = $c =~ /^ (?: \Q$relalias.\E )? (.+)/x; + + unless ($relcol_list->{$colname}) { + push @$cond_cols, $colname; + next; + } + + if ( + ref $joinfree_cond->{$c} + and + ref $joinfree_cond->{$c} ne 'SCALAR' + and + ref $joinfree_cond->{$c} ne 'REF' + ) { + push @$cond_cols, $colname; + next; + } + } + + return wantarray ? ($joinfree_cond, 0, $cond_cols) : $joinfree_cond; + } + else { + return wantarray ? ($crosstable_cond, 1) : $crosstable_cond; + } + } + elsif (ref $cond eq 'HASH') { my %ret; foreach my $k (keys %{$cond}) { my $v = $cond->{$k}; @@ -1605,38 +1728,48 @@ sub _resolve_condition { $ret{"${as}.${k}"} = { -ident => "${for}.${v}" }; } } - return \%ret; - } elsif (ref $cond eq 'ARRAY') { - return [ map { $self->_resolve_condition($_, $as, $for, $relname) } @$cond ]; - } else { - $self->throw_exception ("Can't handle condition $cond yet :("); + + return wantarray + ? ( \%ret, ($obj_rel || !defined $as || ref $as) ? 0 : 1 ) + : \%ret + ; + } + elsif (ref $cond eq 'ARRAY') { + my (@ret, $crosstable); + for (@$cond) { + my ($cond, $crosstab) = $self->_resolve_condition($_, $as, $for, $relname); + push @ret, $cond; + $crosstable ||= $crosstab; + } + return wantarray ? (\@ret, $crosstable) : \@ret; + } + else { + $self->throw_exception ("Can't handle condition $cond for relationship '$relname' yet :("); } } - # Accepts one or more relationships for the current source and returns an # array of column names for each of those relationships. Column names are # prefixed relative to the current source, in accordance with where they appear # in the supplied relationships. - sub _resolve_prefetch { - my ($self, $pre, $alias, $alias_map, $order, $collapse, $pref_path) = @_; + my ($self, $pre, $alias, $alias_map, $order, $pref_path) = @_; $pref_path ||= []; - if (not defined $pre) { + if (not defined $pre or not length $pre) { return (); } elsif( ref $pre eq 'ARRAY' ) { return - map { $self->_resolve_prefetch( $_, $alias, $alias_map, $order, $collapse, [ @$pref_path ] ) } + map { $self->_resolve_prefetch( $_, $alias, $alias_map, $order, [ @$pref_path ] ) } @$pre; } elsif( ref $pre eq 'HASH' ) { my @ret = map { - $self->_resolve_prefetch($_, $alias, $alias_map, $order, $collapse, [ @$pref_path ] ), + $self->_resolve_prefetch($_, $alias, $alias_map, $order, [ @$pref_path ] ), $self->related_source($_)->_resolve_prefetch( - $pre->{$_}, "${alias}.$_", $alias_map, $order, $collapse, [ @$pref_path, $_] ) + $pre->{$_}, "${alias}.$_", $alias_map, $order, [ @$pref_path, $_] ) } keys %$pre; return @ret; } @@ -1658,6 +1791,7 @@ sub _resolve_prefetch { my $rel_info = $self->relationship_info( $pre ); $self->throw_exception( $self->source_name . " has no such relationship '$pre'" ) unless $rel_info; + my $as_prefix = ($alias =~ /^.*?\.(.+)$/ ? $1.'.' : ''); my $rel_source = $self->related_source($pre); @@ -1667,27 +1801,11 @@ sub _resolve_prefetch { unless ref($rel_info->{cond}) eq 'HASH'; my $dots = @{[$as_prefix =~ m/\./g]} + 1; # +1 to match the ".${as_prefix}" - if (my ($fail) = grep { @{[$_ =~ m/\./g]} == $dots } - keys %{$collapse}) { - my ($last) = ($fail =~ /([^\.]+)$/); - carp ( - "Prefetching multiple has_many rels ${last} and ${pre} " - .(length($as_prefix) - ? "at the same level (${as_prefix}) " - : "at top level " - ) - . 'will explode the number of row objects retrievable via ->next or ->all. ' - . 'Use at your own risk.' - ); - } - #my @col = map { (/^self\.(.+)$/ ? ("${as_prefix}.$1") : ()); } # values %{$rel_info->{cond}}; - $collapse->{".${as_prefix}${pre}"} = [ $rel_source->_pri_cols ]; - # action at a distance. prepending the '.' allows simpler code - # in ResultSet->_collapse_result my @key = map { (/^foreign\.(.+)$/ ? ($1) : ()); } keys %{$rel_info->{cond}}; + push @$order, map { "${as}.$_" } @key; if (my $rel_order = $rel_info->{attrs}{order_by}) { @@ -1722,6 +1840,567 @@ sub _resolve_prefetch { } } +# adding a dep on MoreUtils *just* for this is retarded +my $unique_numlist = sub { [ sort { $a <=> $b } keys %{ {map { $_ => 1 } @_ }} ] }; + +# This error must be thrown from two distinct codepaths, joining them is +# rather hard. Go for this hack instead. +my $get_related_source = sub { + my ($rsrc, $rel, $relcols) = @_; + try { + $rsrc->related_source ($rel) + } catch { + $rsrc->throw_exception(sprintf( + "Can't inflate prefetch into non-existent relationship '%s' from '%s', " + . "check the inflation specification (columns/as) ending in '...%s.%s'.", + $rel, + $rsrc->source_name, + $rel, + (sort { length($a) <=> length ($b) } keys %$relcols)[0], + ))}; +}; + +# Takes a selection list and generates a collapse-map representing +# row-object fold-points. Every relationship is assigned a set of unique, +# non-nullable columns (which may *not even be* from the same resultset) +# and the collapser will use this information to correctly distinguish +# data of individual to-be-row-objects. +sub _resolve_collapse { + my ($self, $as, $as_fq_idx, $rel_chain, $parent_info, $node_idx_ref) = @_; + + # for comprehensible error messages put ourselves at the head of the relationship chain + $rel_chain ||= [ $self->source_name ]; + + # record top-level fully-qualified column index + $as_fq_idx ||= { %$as }; + + my ($my_cols, $rel_cols); + for (keys %$as) { + if ($_ =~ /^ ([^\.]+) \. (.+) /x) { + $rel_cols->{$1}{$2} = 1; + } + else { + $my_cols->{$_} = {}; # important for ||= below + } + } + + my $relinfo; + # run through relationships, collect metadata, inject non-left fk-bridges from + # *INNER-JOINED* children (if any) + for my $rel (keys %$rel_cols) { + my $rel_src = $get_related_source->($self, $rel, $rel_cols->{$rel}); + + my $inf = $self->relationship_info ($rel); + + $relinfo->{$rel}{is_single} = $inf->{attrs}{accessor} && $inf->{attrs}{accessor} ne 'multi'; + $relinfo->{$rel}{is_inner} = ( $inf->{attrs}{join_type} || '' ) !~ /^left/i; + $relinfo->{$rel}{rsrc} = $rel_src; + + my $cond = $inf->{cond}; + + if ( + ref $cond eq 'HASH' + and + keys %$cond + and + ! first { $_ !~ /^foreign\./ } (keys %$cond) + and + ! first { $_ !~ /^self\./ } (values %$cond) + ) { + for my $f (keys %$cond) { + my $s = $cond->{$f}; + $_ =~ s/^ (?: foreign | self ) \.//x for ($f, $s); + $relinfo->{$rel}{fk_map}{$s} = $f; + + # need to know source from *our* pov, hnce $rel. + $my_cols->{$s} ||= { via_fk => "$rel.$f" } if ( + defined $rel_cols->{$rel}{$f} # in fact selected + and + (! $node_idx_ref or $relinfo->{$rel}{is_inner}) # either top-level or an inner join + ); + } + } + } + + # if the parent is already defined, assume all of its related FKs are selected + # (even if they in fact are NOT in the select list). Keep a record of what we + # assumed, and if any such phantom-column becomes part of our own collapser, + # throw everything assumed-from-parent away and replace with the collapser of + # the parent (whatever it may be) + my $assumed_from_parent; + unless ($parent_info->{underdefined}) { + $assumed_from_parent->{columns} = { map + # only add to the list if we do not already select said columns + { ! exists $my_cols->{$_} ? ( $_ => 1 ) : () } + values %{$parent_info->{rel_condition} || {}} + }; + + $my_cols->{$_} = { via_collapse => $parent_info->{collapse_on} } + for keys %{$assumed_from_parent->{columns}}; + } + + # get colinfo for everything + if ($my_cols) { + my $ci = $self->columns_info; + $my_cols->{$_}{colinfo} = $ci->{$_} for keys %$my_cols; + } + + my $collapse_map; + + # try to resolve based on our columns (plus already inserted FK bridges) + if ( + $my_cols + and + my $uset = $self->_unique_column_set ($my_cols) + ) { + # see if the resulting collapser relies on any implied columns, + # and fix stuff up if this is the case + + my $parent_collapser_used = defined delete @{$uset}{keys %{$assumed_from_parent->{columns}}}; + $collapse_map->{-node_id} = $unique_numlist->( + $parent_collapser_used ? @{$parent_info->{collapse_on}} : (), + (map + { + my $fqc = join ('.', + @{$rel_chain}[1 .. $#$rel_chain], + ( $my_cols->{$_}{via_fk} || $_ ), + ); + + $as_fq_idx->{$fqc}; + } + keys %$uset + ), + ); + } + + # Stil don't know how to collapse - keep descending down 1:1 chains - if + # a related non-LEFT 1:1 is resolvable - its condition will collapse us + # too + unless ($collapse_map->{-node_id}) { + my @candidates; + + for my $rel (keys %$relinfo) { + next unless ($relinfo->{$rel}{is_single} && $relinfo->{$rel}{is_inner}); + + if ( my $rel_collapse = $relinfo->{$rel}{rsrc}->_resolve_collapse ( + $rel_cols->{$rel}, + $as_fq_idx, + [ @$rel_chain, $rel ], + { underdefined => 1 } + )) { + push @candidates, $rel_collapse->{-node_id}; + } + } + + # get the set with least amount of columns + # FIXME - maybe need to implement a data type order as well (i.e. prefer several ints + # to a single varchar) + if (@candidates) { + ($collapse_map->{-node_id}) = sort { scalar @$a <=> scalar @$b } (@candidates); + } + } + + # Still dont know how to collapse - see if the parent passed us anything + # (i.e. reuse collapser over 1:1) + unless ($collapse_map->{-node_id}) { + $collapse_map->{-node_id} = $parent_info->{collapse_on} + if $parent_info->{collapser_reusable}; + } + + # stop descending into children if we were called by a parent for first-pass + # and don't despair if nothing was found (there may be other parallel branches + # to dive into) + if ($parent_info->{underdefined}) { + return $collapse_map->{-node_id} ? $collapse_map : undef + } + # nothing down the chain resolved - can't calculate a collapse-map + elsif (! $collapse_map->{-node_id}) { + $self->throw_exception ( sprintf + "Unable to calculate a definitive collapse column set for %s%s: fetch more unique non-nullable columns", + $self->source_name, + @$rel_chain > 1 + ? sprintf (' (last member of the %s chain)', join ' -> ', @$rel_chain ) + : '' + , + ); + } + + # If we got that far - we are collapsable - GREAT! Now go down all children + # a second time, and fill in the rest + + $collapse_map->{-is_optional} = 1 if $parent_info->{is_optional}; + $collapse_map->{-node_index} = ${ $node_idx_ref ||= \do { my $x = 1 } }++; # this is *deliberately* not 0-based + + my (@id_sets, $multis_in_chain); + for my $rel (sort keys %$relinfo) { + + $collapse_map->{$rel} = $relinfo->{$rel}{rsrc}->_resolve_collapse ( + { map { $_ => 1 } ( keys %{$rel_cols->{$rel}} ) }, + + $as_fq_idx, + + [ @$rel_chain, $rel], + + { + collapse_on => [ @{$collapse_map->{-node_id}} ], + + rel_condition => $relinfo->{$rel}{fk_map}, + + is_optional => $collapse_map->{-is_optional}, + + # if this is a 1:1 our own collapser can be used as a collapse-map + # (regardless of left or not) + collapser_reusable => $relinfo->{$rel}{is_single}, + }, + + $node_idx_ref, + ); + + $collapse_map->{$rel}{-is_single} = 1 if $relinfo->{$rel}{is_single}; + $collapse_map->{$rel}{-is_optional} ||= 1 unless $relinfo->{$rel}{is_inner}; + push @id_sets, @{ $collapse_map->{$rel}{-branch_id} }; + } + + $collapse_map->{-branch_id} = $unique_numlist->( @id_sets, @{$collapse_map->{-node_id}} ); + + return $collapse_map; +} + +sub _unique_column_set { + my ($self, $cols) = @_; + + my %unique = $self->unique_constraints; + + # always prefer the PK first, and then shortest constraints first + USET: + for my $set (delete $unique{primary}, sort { @$a <=> @$b } (values %unique) ) { + next unless $set && @$set; + + for (@$set) { + next USET unless ($cols->{$_} && $cols->{$_}{colinfo} && !$cols->{$_}{colinfo}{is_nullable} ); + } + + return { map { $_ => 1 } @$set }; + } + + return undef; +} + +# Takes an arrayref of {as} dbic column aliases and the collapse and select +# attributes from the same $rs (the slector requirement is a temporary +# workaround), and returns a coderef capable of: +# my $me_pref_clps = $coderef->([$rs->cursor->next]) +# Where the $me_pref_clps arrayref is the future argument to +# ::ResultSet::_collapse_result. +# +# $me_pref_clps->[0] is always returned (even if as an empty hash with no +# rowdata), however branches of related data in $me_pref_clps->[1] may be +# pruned short of what was originally requested based on {as}, depending +# on: +# +# * If collapse is requested, a definitive collapse map is calculated for +# every relationship "fold-point", consisting of a set of values (which +# may not even be contained in the future 'me' of said relationship +# (for example a cd.artist_id defines the related inner-joined artist)). +# Thus a definedness check is carried on all collapse-condition values +# and if at least one is undef it is assumed that we are dealing with a +# NULLed right-side of a left-join, so we don't return a related data +# container at all, which implies no related objects +# +# * If we are not collapsing, there is no constraint on having a selector +# uniquely identifying all possible objects, and the user might have very +# well requested a column that just *happens* to be all NULLs. What we do +# in this case is fallback to the old behavior (which is a potential FIXME) +# by always returning a data container, but only filling it with columns +# IFF at least one of them is defined. This way we do not get an object +# with a bunch of has_column_loaded to undef, but at the same time do not +# further relationships based off this "null" object (e.g. in case the user +# deliberately skipped link-table values). I am pretty sure there are some +# tests that codify this behavior, need to find the exact testname. +# +# For an example of this coderef in action (and to see its guts) look at +# t/prefetch/_internals.t +# +# This is a huge performance win, as we call the same code for +# every row returned from the db, thus avoiding repeated method +# lookups when traversing relationships +# +# Also since the coderef is completely stateless (the returned structure is +# always fresh on every new invocation) this is a very good opportunity for +# memoization if further speed improvements are needed +# +# The way we construct this coderef is somewhat fugly, although I am not +# sure if the string eval is *that* bad of an idea. The alternative is to +# have a *very* large number of anon coderefs calling each other in a twisty +# maze, whereas the current result is a nice, smooth, single-pass function. +# In any case - the output of this thing is meticulously micro-tested, so +# any sort of rewrite should be relatively easy +# +sub _mk_row_parser { + my ($self, $args) = @_; + + my $inflate_index = { map + { $args->{inflate_map}[$_] => $_ } + ( 0 .. $#{$args->{inflate_map}} ) + }; + + my ($parser_src); + if ($args->{collapse}) { + # FIXME - deal with unorderedness + # unordered => $unordered + + my $collapse_map = $self->_resolve_collapse ( + # FIXME + # only consider real columns (not functions) during collapse resolution + # this check shouldn't really be here, as fucktards are not supposed to + # alias random crap to existing column names anyway, but still - just in + # case + # FIXME !!!! - this does not yet deal with unbalanced selectors correctly + # (it is now trivial as the attrs specify where things go out of sync) + { map + { ref $args->{selection}[$inflate_index->{$_}] ? () : ( $_ => $inflate_index->{$_} ) } + keys %$inflate_index + } + ); + + my $unrolled_top_branch_id_indexes = join (', ', @{$collapse_map->{-branch_id}}); + + my ($sequenced_top_branch_id, $sequenced_top_node_id) = map + { join ('', map { "{'\xFF__IDVALPOS__${_}__\xFF'}" } @$_ ) } + $collapse_map->{-branch_id}, $collapse_map->{-node_id} + ; + + my $rolled_out_assemblers = __visit_infmap_collapse ( + $inflate_index, $collapse_map + ); + + my @sprintf_args = ( + $unrolled_top_branch_id_indexes, + $sequenced_top_branch_id, + $sequenced_top_node_id, + $rolled_out_assemblers, + $sequenced_top_node_id, + ); + $parser_src = sprintf (<<'EOS', @sprintf_args); + +### BEGIN STRING EVAL + my ($rows_pos, $result_pos, $cur_row, @cur_row_id_values, $is_new_res, @collapse_idx) = (0,0); + + # this loop is a bit arcane - the rationale is that the passed in + # $_[0] will either have only one row (->next) or will have all + # rows already pulled in (->all and/or unordered). Given that the + # result can be rather large - we reuse the same already allocated + # array, since the collapsed prefetch is smaller by definition. + # At the end we cut the leftovers away and move on. + while ($cur_row = + ($rows_pos >= 0 and $_[0][$rows_pos++] or do { $rows_pos = -1; 0 } ) + || + ($_[1] and $_[1]->()) + ) { + + # FIXME + # optimize this away when we know we have no undefs in the collapse map + $cur_row_id_values[$_] = defined $cur_row->[$_] ? $cur_row->[$_] : "\xFF\xFFN\xFFU\xFFL\xFFL\xFF\xFF" + for (%s); # the top branch_id includes all id values + + # check top branch for doubling via a has_many non-selecting join or something + # 0 is reserved for this (node indexes start from 1) + next if $collapse_idx[0]%s++; + + $is_new_res = ! $collapse_idx[1]%s; + + # lazify + # fire on ordered only +# if ($is_new_res = ! $collapse_idx[1]{$cur_row_id_values[2]}) { +# } + + %s + + $_[0][$result_pos++] = $collapse_idx[1]%s + if $is_new_res; + } + + splice @{$_[0]}, $result_pos; # truncate the passed in array for cases of collapsing ->all() + +### END STRING EVAL +EOS + + # change the quoted placeholders to unquoted alias-references + $parser_src =~ s/ \' \xFF__VALPOS__(\d+)__\xFF \' /sprintf ('$cur_row->[%d]', $1)/gex; + $parser_src =~ s/ \' \xFF__IDVALPOS__(\d+)__\xFF \' /sprintf ('$cur_row_id_values[%d]', $1)/gex; + } + + else { + $parser_src = sprintf( + '$_ = %s for @{$_[0]}', + __visit_infmap_simple($inflate_index, { rsrc => $self }), # need the $rsrc to determine left-ness + ); + + # change the quoted placeholders to unquoted alias-references + $parser_src =~ s/ \' \xFF__VALPOS__(\d+)__\xFF \' /sprintf ('$_->[%d]', $1)/gex; + } + + eval "sub { no strict; no warnings; $parser_src }" or die "$@\n\n$parser_src"; +} + +{ + # keep our own DD object around so we don't have to fitz with quoting + my $dumper_obj; + my $visit_dump = sub { + # we actually will be producing functional perl code here, + # thus no second-guessing of what these globals might have + # been set to. DO NOT CHANGE! + ($dumper_obj ||= do { + require Data::Dumper; + Data::Dumper->new([]) + ->Purity (1) + ->Pad ('') + ->Useqq (0) + ->Terse (1) + ->Quotekeys (1) + ->Deepcopy (1) + ->Deparse (0) + ->Maxdepth (0) + ->Indent (0) + })->Values ([shift])->Dump, + }; + + sub __visit_infmap_simple { + my ($val_idx, $args) = @_; + + my $my_cols = {}; + my $rel_cols; + for (keys %$val_idx) { + if ($_ =~ /^ ([^\.]+) \. (.+) /x) { + $rel_cols->{$1}{$2} = $val_idx->{$_}; + } + else { + $my_cols->{$_} = $val_idx->{$_}; + } + } + my @relperl; + for my $rel (sort keys %$rel_cols) { + + my $rel_rsrc = $get_related_source->($args->{rsrc}, $rel, $rel_cols->{$rel}); + + #my $optional = $args->{is_optional}; + #$optional ||= ($args->{rsrc}->relationship_info($rel)->{attrs}{join_type} || '') =~ /^left/i; + + push @relperl, join ' => ', perlstring($rel), __visit_infmap_simple($rel_cols->{$rel}, { + non_top => 1, + #is_optional => $optional, + rsrc => $rel_rsrc, + }); + + # FIXME SUBOPTIMAL - disabled to satisfy t/resultset/inflate_result_api.t + #if ($optional and my @branch_null_checks = map + # { "(! defined '\xFF__VALPOS__${_}__\xFF')" } + # sort { $a <=> $b } values %{$rel_cols->{$rel}} + #) { + # $relperl[-1] = sprintf ( '(%s) ? ( %s => [] ) : ( %s )', + # join (' && ', @branch_null_checks ), + # perlstring($rel), + # $relperl[-1], + # ); + #} + } + + my $me_struct = keys %$my_cols + ? $visit_dump->({ map { $_ => "\xFF__VALPOS__$my_cols->{$_}__\xFF" } (keys %$my_cols) }) + : 'undef' + ; + + return sprintf '[%s]', join (',', + $me_struct, + @relperl ? sprintf ('{ %s }', join (',', @relperl)) : (), + ); + } + + sub __visit_infmap_collapse { + my ($val_idx, $collapse_map, $parent_info) = @_; + + my $my_cols = {}; + my $rel_cols; + for (keys %$val_idx) { + if ($_ =~ /^ ([^\.]+) \. (.+) /x) { + $rel_cols->{$1}{$2} = $val_idx->{$_}; + } + else { + $my_cols->{$_} = $val_idx->{$_}; + } + } + + my $sequenced_node_id = join ('', map + { "{'\xFF__IDVALPOS__${_}__\xFF'}" } + @{$collapse_map->{-node_id}} + ); + + my $me_struct = keys %$my_cols + ? $visit_dump->([{ map { $_ => "\xFF__VALPOS__$my_cols->{$_}__\xFF" } (keys %$my_cols) }]) + : 'undef' + ; + my $node_idx_ref = sprintf '$collapse_idx[%d]%s', $collapse_map->{-node_index}, $sequenced_node_id; + + my $parent_idx_ref = sprintf( '$collapse_idx[%d]%s[1]{%s}', + @{$parent_info}{qw/node_idx sequenced_node_id/}, + perlstring($parent_info->{relname}), + ) if $parent_info; + + my @src; + if ($collapse_map->{-node_index} == 1) { + push @src, sprintf( '%s ||= %s;', + $node_idx_ref, + $me_struct, + ); + } + elsif ($collapse_map->{-is_single}) { + push @src, sprintf ( '%s = %s ||= %s;', + $parent_idx_ref, + $node_idx_ref, + $me_struct, + ); + } + else { + push @src, sprintf('push @{%s}, %s = %s if !%s;', + $parent_idx_ref, + $node_idx_ref, + $me_struct, + $node_idx_ref, + ); + } + + #my $known_defined = { %{ $parent_info->{known_defined} || {} } }; + #$known_defined->{$_}++ for @{$collapse_map->{-node_id}}; + + for my $rel (sort keys %$rel_cols) { + + push @src, sprintf( '%s[1]{%s} ||= [];', $node_idx_ref, perlstring($rel) ); + + push @src, __visit_infmap_collapse($rel_cols->{$rel}, $collapse_map->{$rel}, { + node_idx => $collapse_map->{-node_index}, + sequenced_node_id => $sequenced_node_id, + relname => $rel, + #known_defined => $known_defined, + }); + + # FIXME SUBOPTIMAL - disabled to satisfy t/resultset/inflate_result_api.t + #if ($collapse_map->{$rel}{-is_optional} and my @null_checks = map + # { "(! defined '\xFF__VALPOS__${_}__\xFF')" } + # sort { $a <=> $b } grep + # { ! $known_defined->{$_} } + # @{$collapse_map->{$rel}{-node_id}} + #) { + # $src[-1] = sprintf( '(%s) or %s', + # join (' || ', @null_checks ), + # $src[-1], + # ); + #} + } + + join "\n", @src; + } +} + =head2 related_source =over 4 @@ -1741,7 +2420,18 @@ sub related_source { if( !$self->has_relationship( $rel ) ) { $self->throw_exception("No such relationship '$rel' on " . $self->source_name); } - return $self->schema->source($self->relationship_info($rel)->{source}); + + # if we are not registered with a schema - just use the prototype + # however if we do have a schema - ask for the source by name (and + # throw in the process if all fails) + if (my $schema = try { $self->schema }) { + $schema->source($self->relationship_info($rel)->{source}); + } + else { + my $class = $self->relationship_info($rel)->{class}; + $self->ensure_class_loaded($class); + $class->result_source_instance; + } } =head2 related_class @@ -1795,13 +2485,9 @@ sub handle { }); } -{ - my $global_phase_destroy; - - END { $global_phase_destroy++ } - - sub DESTROY { - return if $global_phase_destroy; +my $global_phase_destroy; +sub DESTROY { + return if $global_phase_destroy ||= in_global_destruction; ###### # !!! ACHTUNG !!!! @@ -1813,31 +2499,44 @@ sub handle { # we are trying to save to reattach back to the source we are destroying. # The relevant code checking refcounts is in ::Schema::DESTROY() - # if we are not a schema instance holder - we don't matter - return if( - ! ref $_[0]->{schema} - or - isweak $_[0]->{schema} - ); + # if we are not a schema instance holder - we don't matter + return if( + ! ref $_[0]->{schema} + or + isweak $_[0]->{schema} + ); - # weaken our schema hold forcing the schema to find somewhere else to live + # weaken our schema hold forcing the schema to find somewhere else to live + # during global destruction (if we have not yet bailed out) this will throw + # which will serve as a signal to not try doing anything else + # however beware - on older perls the exception seems randomly untrappable + # due to some weird race condition during thread joining :((( + local $@; + eval { weaken $_[0]->{schema}; - # if schema is still there reintroduce ourselves with strong refs back + # if schema is still there reintroduce ourselves with strong refs back to us if ($_[0]->{schema}) { my $srcregs = $_[0]->{schema}->source_registrations; for (keys %$srcregs) { + next unless $srcregs->{$_}; $srcregs->{$_} = $_[0] if $srcregs->{$_} == $_[0]; } } - } + + 1; + } or do { + $global_phase_destroy = 1; + }; + + return; } -sub STORABLE_freeze { nfreeze($_[0]->handle) } +sub STORABLE_freeze { Storable::nfreeze($_[0]->handle) } sub STORABLE_thaw { my ($self, $cloning, $ice) = @_; - %$self = %{ (thaw $ice)->resolve }; + %$self = %{ (Storable::thaw($ice))->resolve }; } =head2 throw_exception