X-Git-Url: http://git.shadowcat.co.uk/gitweb/gitweb.cgi?a=blobdiff_plain;f=lib%2FDBIx%2FClass%2FResultSource%2FRowParser.pm;h=c74ab9e952e31562dfac9e2260a813af1ad72bf9;hb=574dd7c9e960fd2fa6ad083454776adf3cd967e9;hp=d71ca309ea94467f83e39a3e495d74a956358219;hpb=4e9fc3f33df616fb7340d05e304ff985b9cce9cb;p=dbsrgits%2FDBIx-Class.git diff --git a/lib/DBIx/Class/ResultSource/RowParser.pm b/lib/DBIx/Class/ResultSource/RowParser.pm index d71ca30..c74ab9e 100644 --- a/lib/DBIx/Class/ResultSource/RowParser.pm +++ b/lib/DBIx/Class/ResultSource/RowParser.pm @@ -63,22 +63,26 @@ sub _resolve_prefetch { } } -# Takes a selection list and generates a collapse-map representing +# Takes an arrayref 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. +# data of individual to-be-row-objects. See t/resultset/rowparser_internals.t +# for extensive RV examples sub _resolve_collapse { - my ($self, $as, $as_fq_idx, $rel_chain, $parent_info, $node_idx_ref) = @_; + my ($self, $args, $common_args) = @_; # for comprehensible error messages put ourselves at the head of the relationship chain - $rel_chain ||= [ $self->source_name ]; + $args->{_rel_chain} ||= [ $self->source_name ]; - # record top-level fully-qualified column index - $as_fq_idx ||= { %$as }; + # record top-level fully-qualified column index, start nodecount + $common_args ||= { + _as_fq_idx => { %{$args->{as}} }, + _node_idx => 1, # this is *deliberately* not 0-based + }; my ($my_cols, $rel_cols); - for (keys %$as) { + for (keys %{$args->{as}}) { if ($_ =~ /^ ([^\.]+) \. (.+) /x) { $rel_cols->{$1}{$2} = 1; } @@ -131,14 +135,14 @@ sub _resolve_collapse { # 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}) { + unless ($args->{_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} || {}} + values %{$args->{_parent_info}{rel_condition} || {}} }; - $my_cols->{$_} = { via_collapse => $parent_info->{collapse_on} } + $my_cols->{$_} = { via_collapse => $args->{_parent_info}{collapse_on_idcols} } for keys %{$assumed_from_parent->{columns}}; } @@ -150,8 +154,18 @@ sub _resolve_collapse { my $collapse_map; - # try to resolve based on our columns (plus already inserted FK bridges) + # first try to reuse the parent's collapser (i.e. reuse collapser over 1:1) + # (makes for a leaner coderef later) + unless ($collapse_map->{-idcols_current_node}) { + $collapse_map->{-idcols_current_node} = $args->{_parent_info}{collapse_on_idcols} + if $args->{_parent_info}{collapser_reusable}; + } + + + # Still dont know how to collapse - try to resolve based on our columns (plus already inserted FK bridges) if ( + ! $collapse_map->{-idcols_current_node} + and $my_cols and my $idset = $self->_identifying_column_set ({map { $_ => $my_cols->{$_}{colinfo} } keys %$my_cols}) @@ -160,38 +174,38 @@ sub _resolve_collapse { # and fix stuff up if this is the case my @reduced_set = grep { ! $assumed_from_parent->{columns}{$_} } @$idset; - $collapse_map->{-node_id} = __unique_numlist( - (@reduced_set != @$idset) ? @{$parent_info->{collapse_on}} : (), + $collapse_map->{-idcols_current_node} = [ __unique_numlist( + @{ $args->{_parent_info}{collapse_on_idcols}||[] }, + (map { my $fqc = join ('.', - @{$rel_chain}[1 .. $#$rel_chain], + @{$args->{_rel_chain}}[1 .. $#{$args->{_rel_chain}}], ( $my_cols->{$_}{via_fk} || $_ ), ); - $as_fq_idx->{$fqc}; + $common_args->{_as_fq_idx}->{$fqc}; } @reduced_set ), - ); + )]; } # 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 ($collapse_map->{-node_id}) { + unless ($collapse_map->{-idcols_current_node}) { 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}; + if ( my $rel_collapse = $relinfo->{$rel}{rsrc}->_resolve_collapse ({ + as => $rel_cols->{$rel}, + _rel_chain => [ @{$args->{_rel_chain}}, $rel ], + _parent_info => { underdefined => 1 }, + }, $common_args)) { + push @candidates, $rel_collapse->{-idcols_current_node}; } } @@ -199,30 +213,23 @@ sub _resolve_collapse { # FIXME - maybe need to implement a data type order as well (i.e. prefer several ints # to a single varchar) if (@candidates) { - ($collapse_map->{-node_id}) = sort { scalar @$a <=> scalar @$b } (@candidates); + ($collapse_map->{-idcols_current_node}) = sort { scalar @$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} ? $collapse_map : undef + if ($args->{_parent_info}{underdefined}) { + return $collapse_map->{-idcols_current_node} ? $collapse_map : undef } # nothing down the chain resolved - can't calculate a collapse-map - elsif (! $collapse_map->{-node_id}) { + elsif (! $collapse_map->{-idcols_current_node}) { $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 ' -> ', @$rel_chain ) + @{$args->{_rel_chain}} > 1 + ? sprintf (' (last member of the %s chain)', join ' -> ', @{$args->{_rel_chain}} ) : '' , ); @@ -231,21 +238,19 @@ sub _resolve_collapse { # 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->{-is_optional} = 1 if $args->{_parent_info}{is_optional}; + $collapse_map->{-node_index} = $common_args->{_node_idx}++; - $collapse_map->{$rel} = $relinfo->{$rel}{rsrc}->_resolve_collapse ( - { map { $_ => 1 } ( keys %{$rel_cols->{$rel}} ) }, - $as_fq_idx, - - [ @$rel_chain, $rel], + my @id_sets; + for my $rel (sort keys %$relinfo) { - { - collapse_on => [ @{$collapse_map->{-node_id}} ], + $collapse_map->{$rel} = $relinfo->{$rel}{rsrc}->_resolve_collapse ({ + as => { map { $_ => 1 } ( keys %{$rel_cols->{$rel}} ) }, + _rel_chain => [ @{$args->{_rel_chain}}, $rel], + _parent_info => { + # shallow copy + collapse_on_idcols => [ @{$collapse_map->{-idcols_current_node}} ], rel_condition => $relinfo->{$rel}{fk_map}, @@ -255,69 +260,50 @@ sub _resolve_collapse { # (regardless of left or not) collapser_reusable => $relinfo->{$rel}{is_single}, }, - - $node_idx_ref, - ); + }, $common_args ); $collapse_map->{$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} }; + push @id_sets, ( map { @$_ } ( + $collapse_map->{$rel}{-idcols_current_node}, + $collapse_map->{$rel}{-idcols_extra_from_children} || (), + )); } - $collapse_map->{-branch_id} = __unique_numlist( @id_sets, @{$collapse_map->{-node_id}} ); + if (@id_sets) { + my $cur_nodeid_hash = { map { $_ => 1 } @{$collapse_map->{-idcols_current_node}} }; + $collapse_map->{-idcols_extra_from_children} = [ grep + { ! $cur_nodeid_hash->{$_} } + __unique_numlist( @id_sets ) + ]; + } return $collapse_map; } # 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. +# attributes from the same $rs (the selector requirement is a temporary +# workaround... I hope), and returns a coderef capable of: +# my $me_pref_clps = $coderef->([$rs->cursor->next/all]) +# Where the $me_pref_clps arrayref is the future argument to inflate_result() # # For an example of this coderef in action (and to see its guts) look at -# t/prefetch/_internals.t +# t/resultset/rowparser_internals.t # -# This is a huge performance win, as we call the same code for -# every row returned from the db, thus avoiding repeated method -# lookups when traversing relationships +# This is a huge performance win, as we call the same code for # every row +# returned from the db, thus avoiding repeated method lookups when traversing +# relationships # # Also since the coderef is completely stateless (the returned structure is # always fresh on every new invocation) this is a very good opportunity for # memoization if further speed improvements are needed # -# The way we construct this coderef is somewhat fugly, although I am not -# sure if the string eval is *that* bad of an idea. The alternative is to -# have a *very* large number of anon coderefs calling each other in a twisty -# maze, whereas the current result is a nice, smooth, single-pass function. +# The way we construct this coderef is somewhat fugly, although the result is +# really worth it. The final coderef does not perform any kind of recursion - +# the entire nested structure constructor is rolled out into a single scope. +# # In any case - the output of this thing is meticulously micro-tested, so -# any sort of rewrite should be relatively easy +# any sort of adjustment/rewrite should be relatively easy (fsvo relatively) # sub _mk_row_parser { my ($self, $args) = @_; @@ -327,36 +313,61 @@ sub _mk_row_parser { ( 0 .. $#{$args->{inflate_map}} ) }; - my ($parser_src); - if ($args->{collapse}) { + my $parser_src; + + # the non-collapsing assembler is easy + # FIXME SUBOPTIMAL there could be a yet faster way to do things here, but + # need to try an actual implementation and benchmark it: + # + # First setup the nested data structure you want for each row + # Then call bind_col() to alias the row fields into the right place in + # the data structure, then to fetch the data do: + # push @rows, dclone($row_data_struct) while ($sth->fetchrow); + # + if (!$args->{collapse}) { + $parser_src = sprintf('$_ = %s for @{$_[0]}', __visit_infmap_simple( + $inflate_index, + { rsrc => $self }, # need the $rsrc to sanity-check inflation map once + )); + + # change the quoted placeholders to unquoted alias-references + $parser_src =~ s/ \' \xFF__VALPOS__(\d+)__\xFF \' /"\$_->[$1]"/gex; + } - my $collapse_map = $self->_resolve_collapse ( + # the collapsing parser is more complicated - it needs to keep a lot of state + # + else { + 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 + # (it is now trivial as the attrs specify where things go out of sync + # needs MOAR tests) + as => { map { ref $args->{selection}[$inflate_index->{$_}] ? () : ( $_ => $inflate_index->{$_} ) } keys %$inflate_index } - ); + }); - my $top_branch_idx_list = join (', ', @{$collapse_map->{-branch_id}}); + my $all_idcols_as_list = join ', ', sort map { @$_ } ( + $collapse_map->{-idcols_current_node}, + $collapse_map->{-idcols_extra_from_children} || (), + ); my $top_node_id_path = join ('', map { "{'\xFF__IDVALPOS__${_}__\xFF'}" } - @{$collapse_map->{-node_id}} + @{$collapse_map->{-idcols_current_node}} ); my $rel_assemblers = __visit_infmap_collapse ( $inflate_index, $collapse_map ); - $parser_src = sprintf (<<'EOS', $top_branch_idx_list, $top_node_id_path, $rel_assemblers); -### BEGIN STRING EVAL + $parser_src = sprintf (<<'EOS', $all_idcols_as_list, $top_node_id_path, $rel_assemblers); +### BEGIN LITERAL STRING EVAL my ($rows_pos, $result_pos, $cur_row, @cur_row_ids, @collapse_idx, $is_new_res) = (0,0); @@ -372,8 +383,10 @@ sub _mk_row_parser { ($_[1] and $_[1]->()) ) { + # due to left joins some of the ids may be NULL/undef, and + # won't play well when used as hash lookups $cur_row_ids[$_] = defined $cur_row->[$_] ? $cur_row->[$_] : "\xFF\xFFN\xFFU\xFFL\xFFL\xFF\xFF" - for (%1$s); # the top branch_id includes all id values + for (%1$s); $is_new_res = ! $collapse_idx[1]%2$s and ( $_[1] and $result_pos and (unshift @{$_[2]}, $cur_row) and last @@ -386,27 +399,19 @@ sub _mk_row_parser { } splice @{$_[0]}, $result_pos; # truncate the passed in array for cases of collapsing ->all() -### END STRING EVAL +### END LITERAL STRING EVAL EOS + # !!! note - different var than the one above # change the quoted placeholders to unquoted alias-references $parser_src =~ s/ \' \xFF__VALPOS__(\d+)__\xFF \' /"\$cur_row->[$1]"/gex; $parser_src =~ s/ \' \xFF__IDVALPOS__(\d+)__\xFF \' /"\$cur_row_ids[$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 - # !!! note - different var than the one above - $parser_src =~ s/ \' \xFF__VALPOS__(\d+)__\xFF \' /"\$_->[$1]"/gex; - } - $parser_src; } +# the simple non-collapsing nested structure recursor sub __visit_infmap_simple { my ($val_idx, $args) = @_; @@ -423,18 +428,18 @@ sub __visit_infmap_simple { my @relperl; for my $rel (sort keys %$rel_cols) { - my $rel_rsrc = __get_related_source($args->{rsrc}, $rel, $rel_cols->{$rel}); - + # DISABLEPRUNE #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->{$rel}, { - non_top => 1, + rsrc => __get_related_source($args->{rsrc}, $rel, $rel_cols->{$rel}), + # DISABLEPRUNE + #non_top => 1, #is_optional => $optional, - rsrc => $rel_rsrc, }); - # FIXME SUBOPTIMAL - disabled to satisfy t/resultset/inflate_result_api.t + # FIXME SUBOPTIMAL DISABLEPRUNE - disabled to satisfy t/resultset/inflate_result_api.t #if ($optional and my @branch_null_checks = map # { "(! defined '\xFF__VALPOS__${_}__\xFF')" } # sort { $a <=> $b } values %{$rel_cols->{$rel}} @@ -458,6 +463,7 @@ sub __visit_infmap_simple { ); } +# the collapsing nested structure recursor sub __visit_infmap_collapse { my ($val_idx, $collapse_map, $parent_info) = @_; @@ -475,7 +481,7 @@ sub __visit_infmap_collapse { my $sequenced_node_id = join ('', map { "{'\xFF__IDVALPOS__${_}__\xFF'}" } - @{$collapse_map->{-node_id}} + @{$collapse_map->{-idcols_current_node}} ); my $me_struct = keys %$my_cols @@ -512,27 +518,30 @@ sub __visit_infmap_collapse { ); } + # DISABLEPRUNE #my $known_defined = { %{ $parent_info->{known_defined} || {} } }; - #$known_defined->{$_}++ for @{$collapse_map->{-node_id}}; + #$known_defined->{$_}++ for @{$collapse_map->{-idcols_current_node}}; for my $rel (sort keys %$rel_cols) { - push @src, sprintf( '%s[1]{%s} ||= [];', $node_idx_ref, perlstring($rel) ) - unless $collapse_map->{$rel}{-is_single}; +# push @src, sprintf( +# '%s[1]{%s} ||= [];', $node_idx_ref, perlstring($rel) +# ) unless $collapse_map->{$rel}{-is_single}; 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, + # DISABLEPRUNE #known_defined => $known_defined, }); - # FIXME SUBOPTIMAL - disabled to satisfy t/resultset/inflate_result_api.t + # FIXME SUBOPTIMAL DISABLEPRUNE - disabled to satisfy t/resultset/inflate_result_api.t #if ($collapse_map->{$rel}{-is_optional} and my @null_checks = map # { "(! defined '\xFF__IDVALPOS__${_}__\xFF')" } # sort { $a <=> $b } grep # { ! $known_defined->{$_} } - # @{$collapse_map->{$rel}{-node_id}} + # @{$collapse_map->{$rel}{-idcols_current_node}} #) { # $src[-1] = sprintf( '(%s) or %s', # join (' || ', @null_checks ), @@ -546,7 +555,7 @@ sub __visit_infmap_collapse { # adding a dep on MoreUtils *just* for this is retarded sub __unique_numlist { - [ sort { $a <=> $b } keys %{ {map { $_ => 1 } @_ }} ] + sort { $a <=> $b } keys %{ {map { $_ => 1 } @_ }} } # This error must be thrown from two distinct codepaths, joining them is @@ -575,7 +584,7 @@ sub __visit_dump { ($dumper_obj ||= do { require Data::Dumper; Data::Dumper->new([]) - ->Useperl (1) + ->Useperl (0) ->Purity (1) ->Pad ('') ->Useqq (0)