#
# This module contains code that should never have seen the light of day,
# does not belong in the Storage, or is otherwise unfit for public
-# display. The arrival of SQLA2 should immediately oboslete 90% of this
+# display. The arrival of SQLA2 should immediately obsolete 90% of this
#
use strict;
use List::Util 'first';
use Scalar::Util 'blessed';
+use Sub::Name 'subname';
use namespace::clean;
#
my @newfrom = $from->[0]; # FROM head is always present
- my %need_joins = (map { %{$_||{}} } (values %$aliastypes) );
+ my %need_joins;
+ for (values %$aliastypes) {
+ # add all requested aliases
+ $need_joins{$_} = 1 for keys %$_;
+
+ # add all their parents (as per joinpath which is an AoH { table => alias })
+ $need_joins{$_} = 1 for map { values %$_ } map { @$_ } values %$_;
+ }
for my $j (@{$from}[1..$#$from]) {
push @newfrom, $j if (
(! $j->[0]{-alias}) # legacy crap
#
# This is the code producing joined subqueries like:
-# SELECT me.*, other.* FROM ( SELECT me.* FROM ... ) JOIN other ON ...
+# SELECT me.*, other.* FROM ( SELECT me.* FROM ... ) JOIN other ON ...
#
sub _adjust_select_args_for_complex_prefetch {
my ($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' || ref $from->[0] ne 'HASH' || ref $from->[1] ne 'ARRAY');
-
# generate inner/outer attribute lists, remove stuff that doesn't apply
my $outer_attrs = { %$attrs };
delete $outer_attrs->{$_} for qw/where bind rows offset group_by having/;
my $inner_attrs = { %$attrs };
- delete $inner_attrs->{$_} for qw/for collapse _prefetch_selector_range _collapse_order_by select as/;
+ delete $inner_attrs->{$_} for qw/for collapse _prefetch_selector_range select as/;
-
- # bring over all non-collapse-induced order_by into the inner query (if any)
- # the outer one will have to keep them all
- delete $inner_attrs->{order_by};
- if (my $ord_cnt = @{$outer_attrs->{order_by}} - @{$outer_attrs->{_collapse_order_by}} ) {
- $inner_attrs->{order_by} = [
- @{$outer_attrs->{order_by}}[ 0 .. $ord_cnt - 1]
- ];
- }
+ # if the user did not request it, there is no point using it inside
+ delete $inner_attrs->{order_by} if delete $inner_attrs->{_order_is_artificial};
# generate the inner/outer select lists
# for inside we consider only stuff *not* brought in by the prefetch
my $outer_select = [ @$select ];
my $inner_select = [];
+ my ($root_source, $root_source_offset);
+
+ for my $i (0 .. $#$from) {
+ my $node = $from->[$i];
+ my $h = (ref $node eq 'HASH') ? $node
+ : (ref $node eq 'ARRAY' and ref $node->[0] eq 'HASH') ? $node->[0]
+ : next
+ ;
+
+ if ( ($h->{-alias}||'') eq $attrs->{alias} and $root_source = $h->{-rsrc} ) {
+ $root_source_offset = $i;
+ last;
+ }
+ }
+
+ $self->throw_exception ('Complex prefetches are not supported on resultsets with a custom from attribute')
+ unless $root_source;
+
+ # use the heavy duty resolver to take care of aliased/nonaliased naming
+ my $colinfo = $self->_resolve_column_info($from);
+ my $selected_root_columns;
+
my ($p_start, $p_end) = @{$outer_attrs->{_prefetch_selector_range}};
for my $i (0 .. $p_start - 1, $p_end + 1 .. $#$outer_select) {
my $sel = $outer_select->[$i];
$sel->{-as} ||= $attrs->{as}[$i];
$outer_select->[$i] = join ('.', $attrs->{alias}, ($sel->{-as} || "inner_column_$i") );
}
+ elsif (! ref $sel and my $ci = $colinfo->{$sel}) {
+ $selected_root_columns->{$ci->{-colname}} = 1;
+ }
push @$inner_select, $sel;
push @{$inner_attrs->{as}}, $attrs->{as}[$i];
}
+ # We will need to fetch all native columns in the inner subquery, which may be a part
+ # of an *outer* join condition. We can not just fetch everything because a potential
+ # has_many restricting join collapse *will not work* on heavy data types.
+ # Time for more horrible SQL parsing, aughhhh
+
+ # MASSIVE FIXME - in fact when we are fully transitioned to DQ and the support is
+ # is sane - we will need to trim the select list to *only* fetch stuff that is
+ # necessary to build joins. In the current implementation if I am selecting a blob
+ # and the group_by kicks in - we are fucked, and all the user can do is not select
+ # that column. This is silly!
+
+ my $retardo_sqla_cache = {};
+ for my $cond ( map { $_->[1] } @{$from}[$root_source_offset + 1 .. $#$from] ) {
+ for my $col (@{$self->_extract_condition_columns($cond, $retardo_sqla_cache)}) {
+ my $ci = $colinfo->{$col};
+ if (
+ $ci
+ and
+ $ci->{-source_alias} eq $attrs->{alias}
+ and
+ ! $selected_root_columns->{$ci->{-colname}}++
+ ) {
+ # adding it to both to keep limits not supporting dark selectors happy
+ push @$inner_select, $ci->{-fq_colname};
+ push @{$inner_attrs->{as}}, $ci->{-fq_colname};
+ }
+ }
+ }
+
# construct the inner $from and lock it in a subquery
# we need to prune first, because this will determine if we need a group_by below
# the fake group_by is so that the pruner throws away all non-selecting, non-restricting
# multijoins (since we def. do not care about those inside the subquery)
- my $subq_joinspec = do {
+ my $inner_subq = do {
# must use it here regardless of user requests
local $self->{_use_join_optimizer} = 1;
my $inner_aliastypes =
$self->_resolve_aliastypes_from_select_args( $inner_from, $inner_select, $where, $inner_attrs );
- # if a multi-type non-selecting (only restricting) join was needed in the subquery
- # add a group_by to simulate the collapse in the subq
+ # we need to simulate collapse in the subq if a multiplying join is pulled
+ # by being a non-selecting restrictor
if (
! $inner_attrs->{group_by}
and
local $self->{_use_join_optimizer} = 0;
# generate the subquery
- my $subq = $self->_select_args_to_query (
+ $self->_select_args_to_query (
$inner_from,
$inner_select,
$where,
$inner_attrs,
);
-
- +{
- -alias => $attrs->{alias},
- -rsrc => $inner_from->[0]{-rsrc},
- $attrs->{alias} => $subq,
- };
};
# Generate the outer from - this is relatively easy (really just replace
# - it is part of the restrictions, in which case we need to collapse the outer
# result by tackling yet another group_by to the outside of the query
+ # work on a shallow copy
$from = [ @$from ];
- # so first generate the outer_from, up to the substitution point
my @outer_from;
- while (my $j = shift @$from) {
- $j = [ $j ] unless ref $j eq 'ARRAY'; # promote the head-from to an AoH
- if ($j->[0]{-alias} eq $attrs->{alias}) { # time to swap
- push @outer_from, [
- $subq_joinspec,
- @{$j}[1 .. $#$j],
- ];
- last; # we'll take care of what's left in $from below
- }
- else {
- push @outer_from, $j;
- }
+ # we may not be the head
+ if ($root_source_offset) {
+ # first generate the outer_from, up to the substitution point
+ @outer_from = splice @$from, 0, $root_source_offset;
+
+ my $root_node = shift @$from;
+
+ push @outer_from, [
+ {
+ -alias => $attrs->{alias},
+ -rsrc => $root_node->[0]{-rsrc},
+ $attrs->{alias} => $inner_subq,
+ },
+ @{$root_node}[1 .. $#$root_node],
+ ];
+ }
+ else {
+ my $root_node = shift @$from;
+
+ @outer_from = {
+ -alias => $attrs->{alias},
+ -rsrc => $root_node->{-rsrc},
+ $attrs->{alias} => $inner_subq,
+ };
}
# scan the *remaining* from spec against different attributes, and see which joins are needed
my $outer_aliastypes =
$self->_resolve_aliastypes_from_select_args( $from, $outer_select, $where, $outer_attrs );
+ # unroll parents
+ my ($outer_select_chain, $outer_restrict_chain) = map { +{
+ map { $_ => 1 } map { values %$_} map { @$_ } values %{ $outer_aliastypes->{$_} || {} }
+ } } qw/selecting restricting/;
+
# see what's left - throw away if not selecting/restricting
- # also throw in a group_by if restricting to guard against
- # cross-join explosions
- #
+ # also throw in a group_by if a non-selecting multiplier,
+ # to guard against cross-join explosions
my $need_outer_group_by;
while (my $j = shift @$from) {
my $alias = $j->[0]{-alias};
- if ($outer_aliastypes->{selecting}{$alias}) {
- push @outer_from, $j;
+ if (
+ $outer_select_chain->{$alias}
+ ) {
+ push @outer_from, $j
}
- elsif ($outer_aliastypes->{restricting}{$alias}) {
+ elsif ($outer_restrict_chain->{$alias}) {
push @outer_from, $j;
- $need_outer_group_by ||= ! $j->[0]{-is_single};
+ $need_outer_group_by ||= $outer_aliastypes->{multiplying}{$alias} ? 1 : 0;
}
}
- # demote the outer_from head
- $outer_from[0] = $outer_from[0][0];
-
if ($need_outer_group_by and ! $outer_attrs->{group_by}) {
my $unprocessed_order_chunks;
or next;
$alias_list->{$al} = $j;
- $aliases_by_type->{multiplying}{$al} = 1
- if ref($_) eq 'ARRAY' and ! $j->{-is_single}; # not array == {from} head == can't be multiplying
+ $aliases_by_type->{multiplying}{$al} ||= $j->{-join_path}||[] if (
+ # not array == {from} head == can't be multiplying
+ ( ref($_) eq 'ARRAY' and ! $j->{-is_single} )
+ or
+ # a parent of ours is already a multiplier
+ ( grep { $aliases_by_type->{multiplying}{$_} } @{ $j->{-join_path}||[] } )
+ );
}
# get a column to source/alias map (including unqualified ones)
for my $type (keys %$to_scan) {
for my $piece (@{$to_scan->{$type}}) {
- $aliases_by_type->{$type}{$alias} = 1 if ($piece =~ $al_re);
+ $aliases_by_type->{$type}{$alias} ||= $alias_list->{$alias}{-join_path}||[]
+ if ($piece =~ $al_re);
}
}
}
for my $type (keys %$to_scan) {
for my $piece (@{$to_scan->{$type}}) {
- $aliases_by_type->{$type}{$colinfo->{$col}{-source_alias}} = 1 if ($piece =~ $col_re);
+ if ($piece =~ $col_re) {
+ my $alias = $colinfo->{$col}{-source_alias};
+ $aliases_by_type->{$type}{$alias} ||= $alias_list->{$alias}{-join_path}||[];
+ }
}
}
}
# Add any non-left joins to the restriction list (such joins are indeed restrictions)
for my $j (values %$alias_list) {
my $alias = $j->{-alias} or next;
- $aliases_by_type->{restricting}{$alias} = 1 if (
+ $aliases_by_type->{restricting}{$alias} ||= $j->{-join_path}||[] if (
(not $j->{-join_type})
or
($j->{-join_type} !~ /^left (?: \s+ outer)? $/xi)
);
}
- # mark all restricting/selecting join parents as such
- # (e.g. join => { cds => 'tracks' } - tracks will need to bring cds too )
- for my $type (qw/restricting selecting/) {
- for my $alias (keys %{$aliases_by_type->{$type}||{}}) {
- $aliases_by_type->{$type}{$_} = 1
- for (map { values %$_ } @{ $alias_list->{$alias}{-join_path} || [] });
- }
- }
-
return $aliases_by_type;
}
or next;
$return{$col} = {
- %{ ( $colinfos->{$source_alias} ||= $rsrc->columns_info )->{$colname} },
+ %{
+ ( $colinfos->{$source_alias} ||= $rsrc->columns_info )->{$colname}
+ ||
+ $self->throw_exception(
+ "No such column '$colname' on source " . $rsrc->source_name
+ );
+ },
-result_source => $rsrc,
-source_alias => $source_alias,
+ -fq_colname => $col eq $colname ? "$source_alias.$col" : $col,
+ -colname => $colname,
};
+
+ $return{"$source_alias.$colname"} = $return{$col} if $col eq $colname;
}
return \%return;
# So it looks like we will have to switch some stuff around.
# local() is useless here as we will be leaving the scope
# anyway, and deep cloning is just too fucking expensive
- # So replace the first hashref in the node arrayref manually
+ # So replace the first hashref in the node arrayref manually
my @new_from = ($from->[0]);
my $sw_idx = { map { (values %$_), 1 } @$switch_branch }; #there's one k/v per join-path
return \@new_from;
}
-# Most databases do not allow aliasing of tables in UPDATE/DELETE. Thus
-# a condition containing 'me' or other table prefixes will not work
-# at all. What this code tries to do (badly) is introspect the condition
-# and remove all column qualifiers. If it bails out early (returns undef)
-# the calling code should try another approach (e.g. a subquery)
-
-sub _strip_cond_qualifiers_from_array {
- my ($self, $where) = @_;
- my @cond;
- for (my $i = 0; $i < @$where; $i++) {
- my $entry = $where->[$i];
- my $hash;
- my $ref = ref $entry;
- if ($ref eq 'HASH' or $ref eq 'ARRAY') {
- $hash = $self->_strip_cond_qualifiers($entry);
- }
- elsif (! $ref) {
- $entry =~ /([^.]+)$/;
- $hash->{$1} = $where->[++$i];
- }
- push @cond, $hash;
- }
- return \@cond;
-}
-
-sub _strip_cond_qualifiers {
- my ($self, $where) = @_;
-
- my $cond = {};
-
- # No-op. No condition, we're updating/deleting everything
- return $cond unless $where;
-
- if (ref $where eq 'ARRAY') {
- $cond = $self->_strip_cond_qualifiers_from_array($where);
- }
- elsif (ref $where eq 'HASH') {
- if ( (keys %$where) == 1 && ( (keys %{$where})[0] eq '-and' )) {
- $cond->{-and} =
- $self->_strip_cond_qualifiers_from_array($where->{-and});
- }
- else {
- foreach my $key (keys %$where) {
- if ($key eq '-or' && ref $where->{$key} eq 'ARRAY') {
- $cond->{$key} = $self->_strip_cond_qualifiers($where->{$key});
+# yet another atrocity: attempt to extract all columns from a
+# where condition by hooking _quote
+sub _extract_condition_columns {
+ my ($self, $cond, $sql_maker_cache) = @_;
+
+ return [] unless $cond;
+
+ my $sm = $sql_maker_cache->{condparser} ||= $self->{_sql_ident_capturer} ||= do {
+ # FIXME - replace with a Moo trait
+ my $orig_sm_class = ref $self->sql_maker;
+ my $smic_class = "${orig_sm_class}::_IdentCapture_";
+
+ unless ($smic_class->isa('SQL::Abstract')) {
+
+ no strict 'refs';
+ *{"${smic_class}::_quote"} = subname "${smic_class}::_quote" => sub {
+ my ($self, $ident) = @_;
+ if (ref $ident eq 'SCALAR') {
+ $ident = $$ident;
+ my $storage_quotes = $self->sql_quote_char || '"';
+ my ($ql, $qr) = map
+ { quotemeta $_ }
+ (ref $storage_quotes eq 'ARRAY' ? @$storage_quotes : ($storage_quotes) x 2 )
+ ;
+
+ while ($ident =~ /
+ $ql (\w+) $qr
+ |
+ ([\w\.]+)
+ /xg) {
+ $self->{_captured_idents}{$1||$2}++;
+ }
}
else {
- $key =~ /([^.]+)$/;
- $cond->{$1} = $where->{$key};
+ $self->{_captured_idents}{$ident}++;
}
- }
+ return $ident;
+ };
+
+ *{"${smic_class}::_get_captured_idents"} = subname "${smic_class}::_get_captures" => sub {
+ (delete shift->{_captured_idents}) || {};
+ };
+
+ $self->inject_base ($smic_class, $orig_sm_class);
+
}
- }
- else {
- return undef;
- }
- return $cond;
+ $smic_class->new();
+ };
+
+ $sm->_recurse_where($cond);
+
+ return [ sort keys %{$sm->_get_captured_idents} ];
}
sub _extract_order_criteria {
}
}
+sub _order_by_is_stable {
+ my ($self, $ident, $order_by, $where) = @_;
+
+ my $colinfo = $self->_resolve_column_info($ident, [
+ (map { $_->[0] } $self->_extract_order_criteria($order_by)),
+ $where ? @{$self->_extract_fixed_condition_columns($where)} :(),
+ ]);
+
+ return undef unless keys %$colinfo;
+
+ my $cols_per_src;
+ $cols_per_src->{$_->{-source_alias}}{$_->{-colname}} = $_ for values %$colinfo;
+
+ for (values %$cols_per_src) {
+ my $src = (values %$_)[0]->{-result_source};
+ return 1 if $src->_identifying_column_set($_);
+ }
+
+ return undef;
+}
+
+# returns an arrayref of column names which *definitely* have som
+# sort of non-nullable equality requested in the given condition
+# specification. This is used to figure out if a resultset is
+# constrained to a column which is part of a unique constraint,
+# which in turn allows us to better predict how ordering will behave
+# etc.
+#
+# this is a rudimentary, incomplete, and error-prone extractor
+# however this is OK - it is conservative, and if we can not find
+# something that is in fact there - the stack will recover gracefully
+# Also - DQ and the mst it rode in on will save us all RSN!!!
+sub _extract_fixed_condition_columns {
+ my ($self, $where, $nested) = @_;
+
+ return unless ref $where eq 'HASH';
+
+ my @cols;
+ for my $lhs (keys %$where) {
+ if ($lhs =~ /^\-and$/i) {
+ push @cols, ref $where->{$lhs} eq 'ARRAY'
+ ? ( map { $self->_extract_fixed_condition_columns($_, 1) } @{$where->{$lhs}} )
+ : $self->_extract_fixed_condition_columns($where->{$lhs}, 1)
+ ;
+ }
+ elsif ($lhs !~ /^\-/) {
+ my $val = $where->{$lhs};
+
+ push @cols, $lhs if (defined $val and (
+ ! ref $val
+ or
+ (ref $val eq 'HASH' and keys %$val == 1 and defined $val->{'='})
+ ));
+ }
+ }
+ return $nested ? @cols : \@cols;
+}
+
1;