X-Git-Url: http://git.shadowcat.co.uk/gitweb/gitweb.cgi?p=dbsrgits%2FDBIx-Class.git;a=blobdiff_plain;f=lib%2FDBIx%2FClass%2FResultSource%2FRowParser.pm;h=141037bfce2441a2c07798652dbcf9c86919325a;hp=c9d4c20c724b6d58e468caded7de8c12bf283a01;hb=fcf32d045;hpb=76031e147d6f0d80ab3ec73a12d373962ade1252 diff --git a/lib/DBIx/Class/ResultSource/RowParser.pm b/lib/DBIx/Class/ResultSource/RowParser.pm index c9d4c20..141037b 100644 --- a/lib/DBIx/Class/ResultSource/RowParser.pm +++ b/lib/DBIx/Class/ResultSource/RowParser.pm @@ -1,10 +1,11 @@ -package DBIx::Class::ResultSource::RowParser; +package # hide from the pauses + DBIx::Class::ResultSource::RowParser; use strict; use warnings; use Try::Tiny; -use List::Util 'first'; +use List::Util qw(first max); use B 'perlstring'; use namespace::clean; @@ -32,7 +33,7 @@ sub _resolve_prefetch { map { $self->_resolve_prefetch($_, $alias, $alias_map, $order, [ @$pref_path ] ), $self->related_source($_)->_resolve_prefetch( - $pre->{$_}, "${alias}.$_", $alias_map, $order, [ @$pref_path, $_] ) + $pre->{$_}, "${alias}.$_", $alias_map, $order, [ @$pref_path, $_] ) } keys %$pre; return @ret; } @@ -56,80 +57,42 @@ sub _resolve_prefetch { unless $rel_info; my $as_prefix = ($alias =~ /^.*?\.(.+)$/ ? $1.'.' : ''); - my $rel_source = $self->related_source($pre); - - if ($rel_info->{attrs}{accessor} && $rel_info->{attrs}{accessor} eq 'multi') { - $self->throw_exception( - "Can't prefetch has_many ${pre} (join cond too complex)") - unless ref($rel_info->{cond}) eq 'HASH'; - my $dots = @{[$as_prefix =~ m/\./g]} + 1; # +1 to match the ".${as_prefix}" - - #my @col = map { (/^self\.(.+)$/ ? ("${as_prefix}.$1") : ()); } - # values %{$rel_info->{cond}}; - my @key = map { (/^foreign\.(.+)$/ ? ($1) : ()); } - keys %{$rel_info->{cond}}; - - push @$order, map { "${as}.$_" } @key; - - if (my $rel_order = $rel_info->{attrs}{order_by}) { - # this is kludgy and incomplete, I am well aware - # but the parent method is going away entirely anyway - # so sod it - my $sql_maker = $self->storage->sql_maker; - my ($orig_ql, $orig_qr) = $sql_maker->_quote_chars; - my $sep = $sql_maker->name_sep; - - # install our own quoter, so we can catch unqualified stuff - local $sql_maker->{quote_char} = ["\x00", "\xFF"]; - - my $quoted_prefix = "\x00${as}\xFF"; - - for my $chunk ( $sql_maker->_order_by_chunks ($rel_order) ) { - my @bind; - ($chunk, @bind) = @$chunk if ref $chunk; - - $chunk = "${quoted_prefix}${sep}${chunk}" - unless $chunk =~ /\Q$sep/; - - $chunk =~ s/\x00/$orig_ql/g; - $chunk =~ s/\xFF/$orig_qr/g; - push @$order, \[$chunk, @bind]; - } - } - } return map { [ "${as}.$_", "${as_prefix}${pre}.$_", ] } - $rel_source->columns; + $self->related_source($pre)->columns; } } -# 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; } else { - $my_cols->{$_} = {}; # important for ||= below + $my_cols->{$_} = {}; # important for ||='s below } } my $relinfo; - # run through relationships, collect metadata, inject non-left fk-bridges from - # *INNER-JOINED* children (if any) + # run through relationships, collect metadata for my $rel (keys %$rel_cols) { my $rel_src = __get_related_source($self, $rel, $rel_cols->{$rel}); @@ -139,6 +102,7 @@ sub _resolve_collapse { $relinfo->{$rel}{is_inner} = ( $inf->{attrs}{join_type} || '' ) !~ /^left/i; $relinfo->{$rel}{rsrc} = $rel_src; + # FIME - need to use _resolve_cond here instead my $cond = $inf->{cond}; if ( @@ -146,40 +110,40 @@ sub _resolve_collapse { and keys %$cond and - ! first { $_ !~ /^foreign\./ } (keys %$cond) + ! defined first { $_ !~ /^foreign\./ } (keys %$cond) and - ! first { $_ !~ /^self\./ } (values %$cond) + ! defined 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 - ); } } } + # inject non-left fk-bridges from *INNER-JOINED* children (if any) + for my $rel (grep { $relinfo->{$_}{is_inner} } keys %$relinfo) { + my $ri = $relinfo->{$rel}; + for (keys %{$ri->{fk_map}} ) { + # need to know source from *our* pov, hence $rel.col + $my_cols->{$_} ||= { via_fk => "$rel.$ri->{fk_map}{$_}" } + if defined $rel_cols->{$rel}{$ri->{fk_map}{$_}} # in fact selected + } + } + # 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}}; + unless ($args->{_parent_info}{underdefined}) { + for my $col ( values %{$args->{_parent_info}{rel_condition} || {}} ) { + next if exists $my_cols->{$col}; + $my_cols->{$col} = { via_collapse => $args->{_parent_info}{collapse_on_idcols} }; + $assumed_from_parent->{columns}{$col}++; + } } # get colinfo for everything @@ -190,48 +154,58 @@ 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 $uset = $self->_unique_column_set ($my_cols) + my $idset = $self->_identifying_column_set ({map { $_ => $my_cols->{$_}{colinfo} } keys %$my_cols}) ) { # see if the resulting collapser relies on any implied columns, # and fix stuff up if this is the case + my @reduced_set = grep { ! $assumed_from_parent->{columns}{$_} } @$idset; + + $collapse_map->{-idcols_current_node} = [ __unique_numlist( + @{ $args->{_parent_info}{collapse_on_idcols}||[] }, - 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], + @{$args->{_rel_chain}}[1 .. $#{$args->{_rel_chain}}], ( $my_cols->{$_}{via_fk} || $_ ), ); - $as_fq_idx->{$fqc}; + $common_args->{_as_fq_idx}->{$fqc}; } - keys %$uset + @reduced_set ), - ); + )]; } # Stil don't know how to collapse - keep descending down 1:1 chains - if # a related non-LEFT 1:1 is resolvable - its condition will collapse us # too - unless ($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, $rel_collapse->{-idcols_current_node}; } } @@ -239,30 +213,99 @@ sub _resolve_collapse { # FIXME - maybe need to implement a data type order as well (i.e. prefer several ints # to a single varchar) if (@candidates) { - ($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}; + # Stil don't know how to collapse, and we are the root node. Last ditch + # effort in case we are *NOT* premultiplied. + # Run through *each multi* all the way down, left or not, and all + # *left* singles (a single may become a multi underneath) . When everything + # gets back see if all the rels link to us definitively. If this is the + # case we are good - either one of them will define us, or if all are NULLs + # we know we are "unique" due to the "non-premultiplied" check + if ( + ! $collapse_map->{-idcols_current_node} + and + ! $args->{premultiplied} + and + $common_args->{_node_idx} == 1 + ) { + my (@collapse_sets, $uncollapsible_chain); + + for my $rel (keys %$relinfo) { + + # we already looked at these higher up + next if ($relinfo->{$rel}{is_single} && $relinfo->{$rel}{is_inner}); + + if (my $clps = $relinfo->{$rel}{rsrc}->_resolve_collapse ({ + as => $rel_cols->{$rel}, + _rel_chain => [ @{$args->{_rel_chain}}, $rel ], + _parent_info => { underdefined => 1 }, + }, $common_args) ) { + + # for singles use the idcols wholesale (either there or not) + if ($relinfo->{$rel}{is_single}) { + push @collapse_sets, $clps->{-idcols_current_node}; + } + elsif (! $relinfo->{$rel}{fk_map}) { + $uncollapsible_chain = 1; + last; + } + else { + my $defined_cols_parent_side; + + for my $fq_col ( grep { /^$rel\.[^\.]+$/ } keys %{$args->{as}} ) { + my ($col) = $fq_col =~ /([^\.]+)$/; + + $defined_cols_parent_side->{$_} = $args->{as}{$fq_col} for grep + { $relinfo->{$rel}{fk_map}{$_} eq $col } + keys %{$relinfo->{$rel}{fk_map}} + ; + } + + if (my $set = $self->_identifying_column_set([ keys %$defined_cols_parent_side ]) ) { + push @collapse_sets, [ sort map { $defined_cols_parent_side->{$_} } @$set ]; + } + else { + $uncollapsible_chain = 1; + last; + } + } + } + else { + $uncollapsible_chain = 1; + last; + } + } + + unless ($uncollapsible_chain) { + # if we got here - we are good to go, but the construction is tricky + # since our children will want to include our collapse criteria - we + # don't give them anything (safe, since they are all collapsible on their own) + # in addition we record the individual collapse posibilities + # of all left children node collapsers, and merge them in the rowparser + # coderef later + $collapse_map->{-idcols_current_node} = []; + $collapse_map->{-root_node_idcol_variants} = [ sort { + (scalar @$a) <=> (scalar @$b) or max(@$a) <=> max(@$b) + } @collapse_sets ]; + } } # stop descending into children if we were called by a parent for first-pass # and don't despair if nothing was found (there may be other parallel branches # to dive into) - if ($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}} ) : '' , ); @@ -271,21 +314,19 @@ 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) { - - $collapse_map->{$rel} = $relinfo->{$rel}{rsrc}->_resolve_collapse ( - { map { $_ => 1 } ( keys %{$rel_cols->{$rel}} ) }, + $collapse_map->{-is_optional} = 1 if $args->{_parent_info}{is_optional}; + $collapse_map->{-node_index} = $common_args->{_node_idx}++; - $as_fq_idx, - [ @$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}, @@ -293,91 +334,52 @@ sub _resolve_collapse { # 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}, + collapser_reusable => @{$collapse_map->{-idcols_current_node}} && $relinfo->{$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}} ); - - 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 }; + 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 undef; + 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) = @_; @@ -387,46 +389,89 @@ sub _mk_row_parser { ( 0 .. $#{$args->{inflate_map}} ) }; - my ($parser_src); - if ($args->{collapse}) { - # FIXME - deal with unorderedness - # unordered => $unordered + 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 + )); - my $collapse_map = $self->_resolve_collapse ( + # change the quoted placeholders to unquoted alias-references + $parser_src =~ s/ \' \xFF__VALPOS__(\d+)__\xFF \' /"\$_->[$1]"/gex; + } + + # the collapsing parser is more complicated - it needs to keep a lot of state + # + else { + my $collapse_map = $self->_resolve_collapse ({ + premultiplied => $args->{premultiplied}, # 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 @all_idcols = sort { $a <=> $b } map { @$_ } ( + $collapse_map->{-idcols_current_node}, + $collapse_map->{-idcols_extra_from_children} || (), ); - my $unrolled_top_branch_id_indexes = join (', ', @{$collapse_map->{-branch_id}}); + my ($top_node_id_path, $top_node_id_cacher, @path_variants); + if (scalar @{$collapse_map->{-idcols_current_node}}) { + $top_node_id_path = join ('', map + { "{'\xFF__IDVALPOS__${_}__\xFF'}" } + @{$collapse_map->{-idcols_current_node}} + ); + } + elsif( my @variants = @{$collapse_map->{-root_node_idcol_variants}} ) { + my @path_parts; - my ($sequenced_top_branch_id, $sequenced_top_node_id) = map - { join ('', map { "{'\xFF__IDVALPOS__${_}__\xFF'}" } @$_ ) } - $collapse_map->{-branch_id}, $collapse_map->{-node_id} - ; + for (@variants) { - 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, - $rolled_out_assemblers, - ); + push @path_variants, sprintf "(join qq(\xFF), '', %s, '')", + ( join ', ', map { "'\xFF__VALPOS__${_}__\xFF'" } @$_ ) + ; - $parser_src = sprintf (<<'EOS', @sprintf_args); + push @path_parts, sprintf "( %s && %s)", + ( join ' && ', map { "( defined '\xFF__VALPOS__${_}__\xFF' )" } @$_ ), + $path_variants[-1]; + ; + } -### BEGIN STRING EVAL - my ($rows_pos, $result_pos, $cur_row, @cur_row_id_values, $is_new_res, @collapse_idx) = (0,0); + $top_node_id_cacher = sprintf '$cur_row_ids[%d] = (%s);', + $all_idcols[-1] + 1, + "\n" . join( "\n or\n", @path_parts, qq{"\0\$rows_pos\0"} ); + $top_node_id_path = sprintf '{$cur_row_ids[%d]}', $all_idcols[-1] + 1; + } + else { + $self->throw_exception('Unexpected collapse map contents'); + } + + my $rel_assemblers = __visit_infmap_collapse ( + $inflate_index, { %$collapse_map, -custom_node_id => $top_node_id_path }, + ); + + $parser_src = sprintf (<<'EOS', join(', ', @all_idcols), $top_node_id_path, $top_node_id_cacher||'', $rel_assemblers); +### BEGIN LITERAL STRING EVAL + my ($rows_pos, $result_pos, $cur_row, @cur_row_ids, @collapse_idx, $is_new_res) = (0,0); # this loop is a bit arcane - the rationale is that the passed in # $_[0] will either have only one row (->next) or will have all @@ -435,55 +480,45 @@ sub _mk_row_parser { # array, since the collapsed prefetch is smaller by definition. # At the end we cut the leftovers away and move on. while ($cur_row = - ( ( $rows_pos >= 0 and $_[0][$rows_pos++] ) or do { $rows_pos = -1; 0 } ) + ( ( $rows_pos >= 0 and $_[0][$rows_pos++] ) or do { $rows_pos = -1; undef } ) || ($_[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 (%1$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 $collapse_idx[0]%2$s++; + # due to left joins some of the ids may be NULL/undef, and + # won't play well when used as hash lookups + # we also need to differentiate NULLs on per-row/per-col basis + #(otherwise folding of optional 1:1s will be greatly confused + $cur_row_ids[$_] = defined $cur_row->[$_] ? $cur_row->[$_] : "\0NULL\xFF$rows_pos\xFF$_\0" + for (%1$s); - $is_new_res = ! $collapse_idx[1]%3$s; + # maybe(!) cache the top node id calculation + %3$s - # lazify - # fire on ordered only -# if ($is_new_res = ! $collapse_idx[1]{$cur_row_id_values[2]}) { -# } + $is_new_res = ! $collapse_idx[1]%2$s and ( + $_[1] and $result_pos and (unshift @{$_[2]}, $cur_row) and last + ); %4$s - $_[0][$result_pos++] = $collapse_idx[1]%3$s + $_[0][$result_pos++] = $collapse_idx[1]%2$s if $is_new_res; } splice @{$_[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_id_values[$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 =~ s/ \' \xFF__IDVALPOS__(\d+)__\xFF \' /"\$cur_row_ids[$1]"/gex; } $parser_src; } +# the simple non-collapsing nested structure recursor sub __visit_infmap_simple { my ($val_idx, $args) = @_; @@ -500,18 +535,18 @@ 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->{$rel}} @@ -535,6 +570,7 @@ sub __visit_infmap_simple { ); } +# the collapsing nested structure recursor sub __visit_infmap_collapse { my ($val_idx, $collapse_map, $parent_info) = @_; @@ -550,14 +586,14 @@ sub __visit_infmap_collapse { } } - my $sequenced_node_id = join ('', map + my $sequenced_node_id = $collapse_map->{-custom_node_id} || join ('', map { "{'\xFF__IDVALPOS__${_}__\xFF'}" } - @{$collapse_map->{-node_id}} + @{$collapse_map->{-idcols_current_node}} ); my $me_struct = keys %$my_cols ? __visit_dump([{ map { $_ => "\xFF__VALPOS__$my_cols->{$_}__\xFF" } (keys %$my_cols) }]) - : 'undef' + : undef ; my $node_idx_ref = sprintf '$collapse_idx[%d]%s', $collapse_map->{-node_index}, $sequenced_node_id; @@ -571,44 +607,47 @@ sub __visit_infmap_collapse { push @src, sprintf( '%s ||= %s;', $node_idx_ref, $me_struct, - ); + ) if $me_struct; } elsif ($collapse_map->{-is_single}) { - push @src, sprintf ( '%s = %s ||= %s;', + push @src, sprintf ( '%s ||= %s%s;', $parent_idx_ref, $node_idx_ref, - $me_struct, + $me_struct ? " ||= $me_struct" : '', ); } else { - push @src, sprintf('push @{%s}, %s = %s if !%s;', + push @src, sprintf('push @{%s}, %s%s unless %s;', $parent_idx_ref, $node_idx_ref, - $me_struct, + $me_struct ? " ||= $me_struct" : '', $node_idx_ref, ); } + # 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) ); +# 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->{$rel}{-is_optional} and my @null_checks = map - # { "(! defined '\xFF__VALPOS__${_}__\xFF')" } + # { "(! 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 ), @@ -622,7 +661,7 @@ 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 { $_ => 1 } @_ }} } # This error must be thrown from two distinct codepaths, joining them is @@ -651,7 +690,7 @@ sub __visit_dump { ($dumper_obj ||= do { require Data::Dumper; Data::Dumper->new([]) - ->Useperl (1) + ->Useperl (0) ->Purity (1) ->Pad ('') ->Useqq (0)