X-Git-Url: http://git.shadowcat.co.uk/gitweb/gitweb.cgi?a=blobdiff_plain;f=lib%2FDBIx%2FClass%2FStorage%2FDBIHacks.pm;h=75438d04295f7ad7b3f7a95077c919feb35402f3;hb=02562a2092543488bba4ccd98c39abca72560555;hp=4147e81f71f0c0d421cb6694a195a90b882c1df5;hpb=8d005ad9929e4bf227919cb6374e2a9e9689324f;p=dbsrgits%2FDBIx-Class.git

diff --git a/lib/DBIx/Class/Storage/DBIHacks.pm b/lib/DBIx/Class/Storage/DBIHacks.pm
index 4147e81..75438d0 100644
--- a/lib/DBIx/Class/Storage/DBIHacks.pm
+++ b/lib/DBIx/Class/Storage/DBIHacks.pm
@@ -2,9 +2,24 @@ package   #hide from PAUSE
   DBIx::Class::Storage::DBIHacks;
 
 #
-# 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 obsolete 90% of this
+# This module contains code supporting a battery of special cases and tests for
+# many corner cases pushing the envelope of what DBIC can do. When work on
+# these utilities began in mid 2009 (51a296b402c) it wasn't immediately obvious
+# that these pieces, despite their misleading on-first-sight-flakiness, will
+# become part of the generic query rewriting machinery of DBIC, allowing it to
+# both generate and process queries representing incredibly complex sets with
+# reasonable efficiency.
+#
+# Now (end of 2015), more than 6 years later the routines in this class have
+# stabilized enough, and are meticulously covered with tests, to a point where
+# an effort to formalize them into user-facing APIs might be worthwhile.
+#
+# An implementor working on publicizing and/or replacing the routines with a
+# more modern SQL generation framework should keep in mind that pretty much all
+# existing tests are constructed on the basis of real-world code used in
+# production somewhere.
+#
+# Please hack on this responsibly ;)
 #
 
 use strict;
@@ -13,9 +28,16 @@ use warnings;
 use base 'DBIx::Class::Storage';
 use mro 'c3';
 
-use List::Util 'first';
 use Scalar::Util 'blessed';
-use Sub::Name 'subname';
+use DBIx::Class::_Util qw(
+  dump_value fail_on_internal_call
+);
+use DBIx::Class::SQLMaker::Util 'extract_equality_conditions';
+use DBIx::Class::ResultSource::FromSpec::Util qw(
+  fromspec_columns_info
+  find_join_path_to_alias
+);
+use DBIx::Class::Carp;
 use namespace::clean;
 
 #
@@ -38,7 +60,11 @@ sub _prune_unused_joins {
     $self->_use_join_optimizer
   );
 
-  my $orig_aliastypes = $self->_resolve_aliastypes_from_select_args($attrs);
+  my $orig_aliastypes =
+    $attrs->{_precalculated_aliastypes}
+      ||
+    $self->_resolve_aliastypes_from_select_args($attrs)
+  ;
 
   my $new_aliastypes = { %$orig_aliastypes };
 
@@ -110,8 +136,8 @@ sub _adjust_select_args_for_complex_prefetch {
   my $outer_attrs = { %$attrs };
   delete @{$outer_attrs}{qw(from bind rows offset group_by _grouped_by_distinct having)};
 
-  my $inner_attrs = { %$attrs };
-  delete @{$inner_attrs}{qw(for collapse select as _related_results_construction)};
+  my $inner_attrs = { %$attrs, _simple_passthrough_construction => 1 };
+  delete @{$inner_attrs}{qw(for collapse select as)};
 
   # there is no point of ordering the insides if there is no limit
   delete $inner_attrs->{order_by} if (
@@ -145,7 +171,7 @@ sub _adjust_select_args_for_complex_prefetch {
     unless $root_node;
 
   # use the heavy duty resolver to take care of aliased/nonaliased naming
-  my $colinfo = $self->_resolve_column_info($inner_attrs->{from});
+  my $colinfo = fromspec_columns_info($inner_attrs->{from});
   my $selected_root_columns;
 
   for my $i (0 .. $#{$outer_attrs->{select}}) {
@@ -168,18 +194,27 @@ sub _adjust_select_args_for_complex_prefetch {
     push @{$inner_attrs->{as}}, $attrs->{as}[$i];
   }
 
-  # We will need to fetch all native columns in the inner subquery, which may
+  my $inner_aliastypes = $self->_resolve_aliastypes_from_select_args($inner_attrs);
+
+  # In the inner subq we will need to fetch *only* native columns which may
   # be a part of an *outer* join condition, or an order_by (which needs to be
   # preserved outside), or wheres. In other words everything but the inner
   # selector
   # We can not just fetch everything because a potential has_many restricting
   # join collapse *will not work* on heavy data types.
-  my $connecting_aliastypes = $self->_resolve_aliastypes_from_select_args({
-    %$inner_attrs,
-    select => [],
-  });
 
-  for (sort map { keys %{$_->{-seen_columns}||{}} } map { values %$_ } values %$connecting_aliastypes) {
+  # essentially a map of all non-selecting seen columns
+  # the sort is there for a nicer select list
+  for (
+    sort
+      map
+        { keys %{$_->{-seen_columns}||{}} }
+        map
+          { values %{$inner_aliastypes->{$_}} }
+          grep
+            { $_ ne 'selecting' }
+            keys %$inner_aliastypes
+  ) {
     my $ci = $colinfo->{$_} or next;
     if (
       $ci->{-source_alias} eq $root_alias
@@ -199,11 +234,15 @@ sub _adjust_select_args_for_complex_prefetch {
   my $inner_subq = do {
 
     # must use it here regardless of user requests (vastly gentler on optimizer)
-    local $self->{_use_join_optimizer} = 1;
+    local $self->{_use_join_optimizer} = 1
+      unless $self->{_use_join_optimizer};
 
     # throw away multijoins since we def. do not care about those inside the subquery
-    ($inner_attrs->{from}, my $inner_aliastypes) = $self->_prune_unused_joins ({
-      %$inner_attrs, _force_prune_multiplying_joins => 1
+    # $inner_aliastypes *will* be redefined at this point
+    ($inner_attrs->{from}, $inner_aliastypes ) = $self->_prune_unused_joins ({
+      %$inner_attrs,
+      _force_prune_multiplying_joins => 1,
+      _precalculated_aliastypes => $inner_aliastypes,
     });
 
     # uh-oh a multiplier (which is not us) left in, this is a problem for limits
@@ -311,7 +350,7 @@ sub _adjust_select_args_for_complex_prefetch {
     ) {
       push @outer_from, $j
     }
-    elsif (first { $_->{$alias} } @outer_nonselecting_chains ) {
+    elsif (grep { $_->{$alias} } @outer_nonselecting_chains ) {
       push @outer_from, $j;
       $may_need_outer_group_by ||= $outer_aliastypes->{multiplying}{$alias} ? 1 : 0;
     }
@@ -328,27 +367,53 @@ sub _adjust_select_args_for_complex_prefetch {
     });
   }
 
-  # This is totally horrific - the {where} ends up in both the inner and outer query
-  # Unfortunately not much can be done until SQLA2 introspection arrives, and even
-  # then if where conditions apply to the *right* side of the prefetch, you may have
-  # to both filter the inner select (e.g. to apply a limit) and then have to re-filter
-  # the outer select to exclude joins you didn't want in the first place
+  # FIXME: The {where} ends up in both the inner and outer query, i.e. *twice*
+  #
+  # This is rather horrific, and while we currently *do* have enough
+  # introspection tooling available to attempt a stab at properly deciding
+  # whether or not to include the where condition on the outside, the
+  # machinery is still too slow to apply it here.
+  # Thus for the time being we do not attempt any sanitation of the where
+  # clause and just pass it through on both sides of the subquery. This *will*
+  # be addressed at a later stage, most likely after folding the SQL generator
+  # into SQLMaker proper
   #
   # OTOH it can be seen as a plus: <ash> (notes that this query would make a DBA cry ;)
+  #
   return $outer_attrs;
 }
 
+# This is probably the ickiest, yet most relied upon part of the codebase:
+# this is the place where we take arbitrary SQL input and break it into its
+# constituent parts, making sure we know which *sources* are used in what
+# *capacity* ( selecting / restricting / grouping / ordering / joining, etc )
+# Although the method is pretty horrific, the worst thing that can happen is
+# for a classification failure, which in turn will result in a vocal exception,
+# and will lead to a relatively prompt fix.
+# The code has been slowly improving and is covered with a formiddable battery
+# of tests, so can be considered "reliably stable" at this point (Oct 2015).
+#
+# A note to implementors attempting to "replace" this - keep in mind that while
+# there are multiple optimization avenues, the actual "scan literal elements"
+# part *MAY NEVER BE REMOVED*, even if it is limited only ot the (future) AST
+# nodes that are deemed opaque (i.e. contain literal expressions). The use of
+# blackbox literals is at this point firmly a user-facing API, and is one of
+# *the* reasons DBIC remains as flexible as it is. In other words, when working
+# on this keep in mind that the following is widespread and *encouraged* way
+# of using DBIC in the wild when push comes to shove:
 #
-# I KNOW THIS SUCKS! GET SQLA2 OUT THE DOOR SO THIS CAN DIE!
+# $rs->search( {}, {
+#   select => \[ $random, @stuff],
+#   from => \[ $random, @stuff ],
+#   where => \[ $random, @stuff ],
+#   group_by => \[ $random, @stuff ],
+#   order_by => \[ $random, @stuff ],
+# } )
+#
+# Various incarnations of the above are reflected in many of the tests. If one
+# gets to fail, you get to fix it. A "this is crazy, nobody does that" is not
+# acceptable going forward.
 #
-# Due to a lack of SQLA2 we fall back to crude scans of all the
-# select/where/order/group attributes, in order to determine what
-# aliases are needed to fulfill the query. This information is used
-# throughout the code to prune unnecessary JOINs from the queries
-# in an attempt to reduce the execution time.
-# Although the method is pretty horrific, the worst thing that can
-# happen is for it to fail due to some scalar SQL, which in turn will
-# result in a vocal exception.
 sub _resolve_aliastypes_from_select_args {
   my ( $self, $attrs ) = @_;
 
@@ -383,7 +448,7 @@ sub _resolve_aliastypes_from_select_args {
   }
 
   # get a column to source/alias map (including unambiguous unqualified ones)
-  my $colinfo = $self->_resolve_column_info ($attrs->{from});
+  my $colinfo = fromspec_columns_info($attrs->{from});
 
   # set up a botched SQLA
   my $sql_maker = $self->sql_maker;
@@ -415,7 +480,7 @@ sub _resolve_aliastypes_from_select_args {
   # generate sql chunks
   my $to_scan = {
     restricting => [
-      $sql_maker->_recurse_where ($attrs->{where}),
+      ($sql_maker->_recurse_where ($attrs->{where}))[0],
       $sql_maker->_parse_rs_attrs ({ having => $attrs->{having} }),
     ],
     grouping => [
@@ -428,111 +493,147 @@ sub _resolve_aliastypes_from_select_args {
       ),
     ],
     selecting => [
-      map { ($sql_maker->_recurse_fields($_))[0] } @{$attrs->{select}},
+      # kill all selectors which look like a proper subquery
+      # this is a sucky heuristic *BUT* - if we get it wrong the query will simply
+      # fail to run, so we are relatively safe
+      grep
+        { $_ !~ / \A \s* \( \s* SELECT \s+ .+? \s+ FROM \s+ .+? \) \s* \z /xsi }
+        map
+          {
+            length ref $_
+              ? ($sql_maker->_recurse_fields($_))[0]
+              : $sql_maker->_quote($_)
+          }
+          @{$attrs->{select}}
     ],
-    ordering => [
-      map { $_->[0] } $self->_extract_order_criteria ($attrs->{order_by}, $sql_maker),
+    ordering => [ map
+      {
+        ( my $sql = (ref $_ ? $_->[0] : $_) ) =~ s/ \s+ (?: ASC | DESC ) \s* \z //xi;
+        $sql;
+      }
+      $sql_maker->_order_by_chunks( $attrs->{order_by} ),
     ],
   };
 
-  # throw away empty chunks and all 2-value arrayrefs: the thinking is that these are
-  # bind value specs left in by the sloppy renderer above. It is ok to do this
-  # at this point, since we are going to end up rewriting this crap anyway
-  for my $v (values %$to_scan) {
-    my @nv;
-    for (@$v) {
-      next if (
-        ! defined $_
-          or
-        (
-          ref $_ eq 'ARRAY'
-            and
-          ( @$_ == 0 or @$_ == 2 )
-        )
-      );
-
-      if (ref $_) {
-        require Data::Dumper::Concise;
-        $self->throw_exception("Unexpected ref in scan-plan: " . Data::Dumper::Concise::Dumper($v) );
-      }
+  # we will be bulk-scanning anyway - pieces will not matter in that case,
+  # thus join everything up
+  # throw away empty-string chunks, and make sure no binds snuck in
+  # note that we operate over @{$to_scan->{$type}}, hence the
+  # semi-mindbending ... map ... for values ...
+  ( $_ = join ' ', map {
+
+    ( ! defined $_ )  ? ()
+  : ( length ref $_ ) ? $self->throw_exception(
+                          "Unexpected ref in scan-plan: " . dump_value $_
+                        )
+  : ( $_ =~ /^\s*$/ ) ? ()
+                      : $_
+
+  } @$_ ) for values %$to_scan;
+
+  # throw away empty to-scan's
+  (
+    length $to_scan->{$_}
+      or
+    delete $to_scan->{$_}
+  ) for keys %$to_scan;
 
-      push @nv, $_;
-    }
 
-    $v = \@nv;
-  }
 
-  # kill all selectors which look like a proper subquery
-  # this is a sucky heuristic *BUT* - if we get it wrong the query will simply
-  # fail to run, so we are relatively safe
-  $to_scan->{selecting} = [ grep {
-    $_ !~ / \A \s* \( \s* SELECT \s+ .+? \s+ FROM \s+ .+? \) \s* \z /xsi
-  } @{ $to_scan->{selecting} || [] } ];
+  # these will be used for matching in the loop below
+  my $all_aliases = join ' | ', map { quotemeta $_ } keys %$alias_list;
+  my $fq_col_re = qr/
+    $lquote ( $all_aliases ) $rquote $sep (?: $lquote ([^$rquote]+) $rquote )?
+         |
+    \b ( $all_aliases ) \. ( [^\s\)\($rquote]+ )?
+  /x;
 
-  # first see if we have any exact matches (qualified or unqualified)
-  for my $type (keys %$to_scan) {
-    for my $piece (@{$to_scan->{$type}}) {
-      if ($colinfo->{$piece} and my $alias = $colinfo->{$piece}{-source_alias}) {
-        $aliases_by_type->{$type}{$alias} ||= { -parents => $alias_list->{$alias}{-join_path}||[] };
-        $aliases_by_type->{$type}{$alias}{-seen_columns}{$colinfo->{$piece}{-fq_colname}} = $piece;
-      }
-    }
-  }
 
-  # now loop through all fully qualified columns and get the corresponding
-  # alias (should work even if they are in scalarrefs)
-  for my $alias (keys %$alias_list) {
-    my $al_re = qr/
-      $lquote $alias $rquote $sep (?: $lquote ([^$rquote]+) $rquote )?
+  my $all_unq_columns = join ' | ',
+    map
+      { quotemeta $_ }
+      grep
+        # using a regex here shows up on profiles, boggle
+        { index( $_, '.') < 0 }
+        keys %$colinfo
+  ;
+  my $unq_col_re = $all_unq_columns
+    ? qr/
+      $lquote ( $all_unq_columns ) $rquote
         |
-      \b $alias \. ([^\s\)\($rquote]+)?
-    /x;
-
-    for my $type (keys %$to_scan) {
-      for my $piece (@{$to_scan->{$type}}) {
-        if (my @matches = $piece =~ /$al_re/g) {
-          $aliases_by_type->{$type}{$alias} ||= { -parents => $alias_list->{$alias}{-join_path}||[] };
-          $aliases_by_type->{$type}{$alias}{-seen_columns}{"$alias.$_"} = "$alias.$_"
-            for grep { defined $_ } @matches;
-        }
+      (?: \A | \s ) ( $all_unq_columns ) (?: \s | \z )
+    /x
+    : undef
+  ;
+
+
+  # the actual scan, per type
+  for my $type (keys %$to_scan) {
+
+
+    # now loop through all fully qualified columns and get the corresponding
+    # alias (should work even if they are in scalarrefs)
+    #
+    # The regex captures in multiples of 4, with one of the two pairs being
+    # undef. There may be a *lot* of matches, hence the convoluted loop
+    my @matches = $to_scan->{$type} =~ /$fq_col_re/g;
+    my $i = 0;
+    while( $i < $#matches ) {
+
+      if (
+        defined $matches[$i]
+      ) {
+        $aliases_by_type->{$type}{$matches[$i]} ||= { -parents => $alias_list->{$matches[$i]}{-join_path}||[] };
+
+        $aliases_by_type->{$type}{$matches[$i]}{-seen_columns}{"$matches[$i].$matches[$i+1]"} = "$matches[$i].$matches[$i+1]"
+          if defined $matches[$i+1];
+
+        $i += 2;
       }
+
+      $i += 2;
     }
-  }
 
-  # now loop through unqualified column names, and try to locate them within
-  # the chunks
-  for my $col (keys %$colinfo) {
-    next if $col =~ / \. /x;   # if column is qualified it was caught by the above
-
-    my $col_re = qr/ $lquote ($col) $rquote /x;
-
-    for my $type (keys %$to_scan) {
-      for my $piece (@{$to_scan->{$type}}) {
-        if ( my @matches = $piece =~ /$col_re/g) {
-          my $alias = $colinfo->{$col}{-source_alias};
-          $aliases_by_type->{$type}{$alias} ||= { -parents => $alias_list->{$alias}{-join_path}||[] };
-          $aliases_by_type->{$type}{$alias}{-seen_columns}{"$alias.$_"} = $_
-            for grep { defined $_ } @matches;
-        }
-      }
+
+    # now loop through unqualified column names, and try to locate them within
+    # the chunks, if there are any unqualified columns in the 1st place
+    next unless $unq_col_re;
+
+    # The regex captures in multiples of 2, one of the two being undef
+    for ( $to_scan->{$type} =~ /$unq_col_re/g ) {
+      defined $_ or next;
+      my $alias = $colinfo->{$_}{-source_alias} or next;
+      $aliases_by_type->{$type}{$alias} ||= { -parents => $alias_list->{$alias}{-join_path}||[] };
+      $aliases_by_type->{$type}{$alias}{-seen_columns}{"$alias.$_"} = $_
     }
   }
 
+
   # 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} ||= { -parents => $j->{-join_path}||[] } if (
-      (not $j->{-join_type})
+  (
+    $_->{-alias}
+      and
+    ! $aliases_by_type->{restricting}{ $_->{-alias} }
+      and
+    (
+      not $_->{-join_type}
         or
-      ($j->{-join_type} !~ /^left (?: \s+ outer)? $/xi)
-    );
-  }
+      $_->{-join_type} !~ /^left (?: \s+ outer)? $/xi
+    )
+      and
+    $aliases_by_type->{restricting}{ $_->{-alias} } = { -parents => $_->{-join_path}||[] }
+  ) for values %$alias_list;
 
-  for (keys %$aliases_by_type) {
-    delete $aliases_by_type->{$_} unless keys %{$aliases_by_type->{$_}};
-  }
 
-  return $aliases_by_type;
+  # final cleanup
+  (
+    keys %{$aliases_by_type->{$_}}
+      or
+    delete $aliases_by_type->{$_}
+  ) for keys %$aliases_by_type;
+
+
+  $aliases_by_type;
 }
 
 # This is the engine behind { distinct => 1 } and the general
@@ -540,7 +641,7 @@ sub _resolve_aliastypes_from_select_args {
 sub _group_over_selection {
   my ($self, $attrs) = @_;
 
-  my $colinfos = $self->_resolve_column_info ($attrs->{from});
+  my $colinfos = fromspec_columns_info($attrs->{from});
 
   my (@group_by, %group_index);
 
@@ -628,12 +729,9 @@ sub _group_over_selection {
       # of the external order and convert them to MIN(X) for ASC or MAX(X)
       # for DESC, and group_by the root columns. The end result should be
       # exactly what we expect
+      #
 
-      # FIXME - this code is a joke, will need to be completely rewritten in
-      # the DQ branch. But I need to push a POC here, otherwise the
-      # pesky tests won't pass
-      # wrap any part of the order_by that "responds" to an ordering alias
-      # into a MIN/MAX
+      # both populated on the first loop over $o_idx
       $sql_maker ||= $self->sql_maker;
       $order_chunks ||= [
         map { ref $_ eq 'ARRAY' ? $_ : [ $_ ] } $sql_maker->_order_by_chunks($attrs->{order_by})
@@ -641,9 +739,11 @@ sub _group_over_selection {
 
       my ($chunk, $is_desc) = $sql_maker->_split_order_chunk($order_chunks->[$o_idx][0]);
 
+      # we reached that far - wrap any part of the order_by that "responded"
+      # to an ordering alias into a MIN/MAX
       $new_order_by[$o_idx] = \[
         sprintf( '%s( %s )%s',
-          ($is_desc ? 'MAX' : 'MIN'),
+          $self->_minmax_operator_for_datatype($chunk_ci->{data_type}, $is_desc),
           $chunk,
           ($is_desc ? ' DESC' : ''),
         ),
@@ -653,9 +753,10 @@ sub _group_over_selection {
   }
 
   $self->throw_exception ( sprintf
-    'A required group_by clause could not be constructed automatically due to a complex '
-  . 'order_by criteria (%s). Either order_by columns only (no functions) or construct a suitable '
-  . 'group_by by hand',
+    'Unable to programatically derive a required group_by from the supplied '
+  . 'order_by criteria. To proceed either add an explicit group_by, or '
+  . 'simplify your order_by to only include plain columns '
+  . '(supplied order_by: %s)',
     join ', ', map { "'$_'" } @$leftovers,
   ) if $leftovers;
 
@@ -670,168 +771,10 @@ sub _group_over_selection {
   );
 }
 
-sub _resolve_ident_sources {
-  my ($self, $ident) = @_;
-
-  my $alias2source = {};
-
-  # the reason this is so contrived is that $ident may be a {from}
-  # structure, specifying multiple tables to join
-  if ( blessed $ident && $ident->isa("DBIx::Class::ResultSource") ) {
-    # this is compat mode for insert/update/delete which do not deal with aliases
-    $alias2source->{me} = $ident;
-  }
-  elsif (ref $ident eq 'ARRAY') {
-
-    for (@$ident) {
-      my $tabinfo;
-      if (ref $_ eq 'HASH') {
-        $tabinfo = $_;
-      }
-      if (ref $_ eq 'ARRAY' and ref $_->[0] eq 'HASH') {
-        $tabinfo = $_->[0];
-      }
-
-      $alias2source->{$tabinfo->{-alias}} = $tabinfo->{-rsrc}
-        if ($tabinfo->{-rsrc});
-    }
-  }
-
-  return $alias2source;
-}
-
-# Takes $ident, \@column_names
-#
-# returns { $column_name => \%column_info, ... }
-# also note: this adds -result_source => $rsrc to the column info
-#
-# If no columns_names are supplied returns info about *all* columns
-# for all sources
-sub _resolve_column_info {
-  my ($self, $ident, $colnames) = @_;
-
-  return {} if $colnames and ! @$colnames;
-
-  my $alias2src = $self->_resolve_ident_sources($ident);
-
-  my (%seen_cols, @auto_colnames);
-
-  # compile a global list of column names, to be able to properly
-  # disambiguate unqualified column names (if at all possible)
-  for my $alias (keys %$alias2src) {
-    my $rsrc = $alias2src->{$alias};
-    for my $colname ($rsrc->columns) {
-      push @{$seen_cols{$colname}}, $alias;
-      push @auto_colnames, "$alias.$colname" unless $colnames;
-    }
-  }
-
-  $colnames ||= [
-    @auto_colnames,
-    grep { @{$seen_cols{$_}} == 1 } (keys %seen_cols),
-  ];
-
-  my (%return, $colinfos);
-  foreach my $col (@$colnames) {
-    my ($source_alias, $colname) = $col =~ m/^ (?: ([^\.]+) \. )? (.+) $/x;
-
-    # if the column was seen exactly once - we know which rsrc it came from
-    $source_alias ||= $seen_cols{$colname}[0]
-      if ($seen_cols{$colname} and @{$seen_cols{$colname}} == 1);
-
-    next unless $source_alias;
-
-    my $rsrc = $alias2src->{$source_alias}
-      or next;
-
-    $return{$col} = {
-      %{
-          ( $colinfos->{$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;
-}
-
-# The DBIC relationship chaining implementation is pretty simple - every
-# new related_relationship is pushed onto the {from} stack, and the {select}
-# window simply slides further in. This means that when we count somewhere
-# in the middle, we got to make sure that everything in the join chain is an
-# actual inner join, otherwise the count will come back with unpredictable
-# results (a resultset may be generated with _some_ rows regardless of if
-# the relation which the $rs currently selects has rows or not). E.g.
-# $artist_rs->cds->count - normally generates:
-# SELECT COUNT( * ) FROM artist me LEFT JOIN cd cds ON cds.artist = me.artistid
-# which actually returns the number of artists * (number of cds || 1)
-#
-# So what we do here is crawl {from}, determine if the current alias is at
-# the top of the stack, and if not - make sure the chain is inner-joined down
-# to the root.
-#
-sub _inner_join_to_node {
-  my ($self, $from, $alias) = @_;
-
-  # subqueries and other oddness are naturally not supported
-  return $from if (
-    ref $from ne 'ARRAY'
-      ||
-    @$from <= 1
-      ||
-    ref $from->[0] ne 'HASH'
-      ||
-    ! $from->[0]{-alias}
-      ||
-    $from->[0]{-alias} eq $alias  # this last bit means $alias is the head of $from - nothing to do
-  );
-
-  # find the current $alias in the $from structure
-  my $switch_branch;
-  JOINSCAN:
-  for my $j (@{$from}[1 .. $#$from]) {
-    if ($j->[0]{-alias} eq $alias) {
-      $switch_branch = $j->[0]{-join_path};
-      last JOINSCAN;
-    }
-  }
+sub _minmax_operator_for_datatype {
+  #my ($self, $datatype, $want_max) = @_;
 
-  # something else went quite wrong
-  return $from unless $switch_branch;
-
-  # So it looks like we will have to switch some stuff around.
-  # local() is useless here as we will be leaving the scope
-  # anyway, and deep cloning is just too fucking expensive
-  # So replace the first hashref in the node arrayref manually
-  my @new_from = ($from->[0]);
-  my $sw_idx = { map { (values %$_), 1 } @$switch_branch }; #there's one k/v per join-path
-
-  for my $j (@{$from}[1 .. $#$from]) {
-    my $jalias = $j->[0]{-alias};
-
-    if ($sw_idx->{$jalias}) {
-      my %attrs = %{$j->[0]};
-      delete $attrs{-join_type};
-      push @new_from, [
-        \%attrs,
-        @{$j}[ 1 .. $#$j ],
-      ];
-    }
-    else {
-      push @new_from, $j;
-    }
-  }
-
-  return \@new_from;
+  $_[2] ? 'MAX' : 'MIN';
 }
 
 sub _extract_order_criteria {
@@ -884,14 +827,14 @@ sub _order_by_is_stable {
 
   my @cols = (
     ( map { $_->[0] } $self->_extract_order_criteria($order_by) ),
-    ( $where ? @{ $self->_extract_fixed_condition_columns($where) || [] } : () ),
-  ) or return undef;
+    ( $where ? keys %{ extract_equality_conditions( $where ) } : () ),
+  ) or return 0;
 
-  my $colinfo = $self->_resolve_column_info($ident, \@cols);
+  my $colinfo = fromspec_columns_info($ident, \@cols);
 
   return keys %$colinfo
     ? $self->_columns_comprise_identifying_set( $colinfo,  \@cols )
-    : undef
+    : 0
   ;
 }
 
@@ -907,323 +850,114 @@ sub _columns_comprise_identifying_set {
     return 1 if $src->_identifying_column_set($_);
   }
 
-  return undef;
+  return 0;
 }
 
-# this is almost identical to the above, except it accepts only
+# this is almost similar to _order_by_is_stable, except it takes
 # a single rsrc, and will succeed only if the first portion of the order
 # by is stable.
 # returns that portion as a colinfo hashref on success
-sub _main_source_order_by_portion_is_stable {
-  my ($self, $main_rsrc, $order_by, $where) = @_;
+sub _extract_colinfo_of_stable_main_source_order_by_portion {
+  my ($self, $attrs) = @_;
 
-  die "Huh... I expect a blessed result_source..."
-    if ref($main_rsrc) eq 'ARRAY';
+  my $nodes = find_join_path_to_alias($attrs->{from}, $attrs->{alias});
+
+  return unless defined $nodes;
 
   my @ord_cols = map
     { $_->[0] }
-    ( $self->_extract_order_criteria($order_by) )
+    ( $self->_extract_order_criteria($attrs->{order_by}) )
   ;
   return unless @ord_cols;
 
-  my $colinfos = $self->_resolve_column_info($main_rsrc);
+  my $valid_aliases = { map { $_ => 1 } (
+    $attrs->{from}[0]{-alias},
+    map { values %$_ } @$nodes,
+  ) };
 
-  for (0 .. $#ord_cols) {
-    if (
-      ! $colinfos->{$ord_cols[$_]}
-        or
-      $colinfos->{$ord_cols[$_]}{-result_source} != $main_rsrc
-    ) {
-      $#ord_cols =  $_ - 1;
-      last;
-    }
-  }
+  my $colinfos = fromspec_columns_info($attrs->{from});
 
-  # we just truncated it above
-  return unless @ord_cols;
+  my ($colinfos_to_return, $seen_main_src_cols);
 
-  my $order_portion_ci = { map {
-    $colinfos->{$_}{-colname} => $colinfos->{$_},
-    $colinfos->{$_}{-fq_colname} => $colinfos->{$_},
-  } @ord_cols };
+  for my $col (@ord_cols) {
+    # if order criteria is unresolvable - there is nothing we can do
+    my $colinfo = $colinfos->{$col} or last;
 
-  # since all we check here are the start of the order_by belonging to the
-  # top level $rsrc, a present identifying set will mean that the resultset
-  # is ordered by its leftmost table in a stable manner
-  #
-  # RV of _identifying_column_set contains unqualified names only
-  my $unqualified_idset = $main_rsrc->_identifying_column_set({
-    ( $where ? %{
-      $self->_resolve_column_info(
-        $main_rsrc, $self->_extract_fixed_condition_columns($where)||[]
-      )
-    } : () ),
-    %$order_portion_ci
-  }) or return;
-
-  my $ret_info;
-  my %unqualified_idcols_from_order = map {
-    $order_portion_ci->{$_} ? ( $_ => $order_portion_ci->{$_} ) : ()
-  } @$unqualified_idset;
-
-  # extra optimization - cut the order_by at the end of the identifying set
-  # (just in case the user was stupid and overlooked the obvious)
-  for my $i (0 .. $#ord_cols) {
-    my $col = $ord_cols[$i];
-    my $unqualified_colname = $order_portion_ci->{$col}{-colname};
-    $ret_info->{$col} = { %{$order_portion_ci->{$col}}, -idx_in_order_subset => $i };
-    delete $unqualified_idcols_from_order{$ret_info->{$col}{-colname}};
-
-    # we didn't reach the end of the identifying portion yet
-    return $ret_info unless keys %unqualified_idcols_from_order;
-  }
-
-  die 'How did we get here...';
-}
+    # if we reached the end of the allowed aliases - also nothing we can do
+    last unless $valid_aliases->{$colinfo->{-source_alias}};
 
-# Attempts to flatten a passed in SQLA condition as much as possible towards
-# a plain hashref, *without* altering its semantics. Required by
-# create/populate being able to extract definitive conditions from preexisting
-# resultset {where} stacks
-#
-# FIXME - while relatively robust, this is still imperfect, one of the first
-# things to tackle with DQ
-sub _collapse_cond {
-  my ($self, $where, $where_is_anded_array) = @_;
+    $colinfos_to_return->{$col} = $colinfo;
 
-  if (! $where) {
-    return;
+    $seen_main_src_cols->{$colinfo->{-colname}} = 1
+      if $colinfo->{-source_alias} eq $attrs->{alias};
   }
-  elsif ($where_is_anded_array or ref $where eq 'HASH') {
-
-    my @pairs;
-
-    my @pieces = $where_is_anded_array ? @$where : $where;
-    while (@pieces) {
-      my $chunk = shift @pieces;
-
-      if (ref $chunk eq 'HASH') {
-        push @pairs, map { [ $_ => $chunk->{$_} ] } sort keys %$chunk;
-      }
-      elsif (ref $chunk eq 'ARRAY') {
-        push @pairs, [ -or => $chunk ]
-          if @$chunk;
-      }
-      elsif ( ! ref $chunk) {
-        push @pairs, [ $chunk, shift @pieces ];
-      }
-      else {
-        push @pairs, [ '', $chunk ];
-      }
-    }
-
-    return unless @pairs;
 
-    my @conds = $self->_collapse_cond_unroll_pairs(\@pairs)
-      or return;
+  # FIXME: the condition may be singling out things on its own, so we
+  # conceivably could come back with "stable-ordered by nothing"
+  # not confident enough in the parser yet, so punt for the time being
+  return unless $seen_main_src_cols;
 
-    # Consolidate various @conds back into something more compact
-    my $fin;
-
-    for my $c (@conds) {
-      if (ref $c ne 'HASH') {
-        push @{$fin->{-and}}, $c;
-      }
-      else {
-        for my $col (sort keys %$c) {
-          if (exists $fin->{$col}) {
-            my ($l, $r) = ($fin->{$col}, $c->{$col});
-
-            (ref $_ ne 'ARRAY' or !@$_) and $_ = [ -and => $_ ] for ($l, $r);
-
-            if (@$l and @$r and $l->[0] eq $r->[0] and $l->[0] eq '-and') {
-              $fin->{$col} = [ -and => map { @$_[1..$#$_] } ($l, $r) ];
-            }
-            else {
-              $fin->{$col} = [ -and => $fin->{$col}, $c->{$col} ];
-            }
-          }
-          else {
-            $fin->{$col} = $c->{$col};
-          }
-        }
-      }
-    }
-
-    if ( ref $fin->{-and} eq 'ARRAY' and @{$fin->{-and}} == 1 ) {
-      my $piece = (delete $fin->{-and})->[0];
-      if (ref $piece eq 'ARRAY') {
-        $fin->{-or} = $fin->{-or} ? [ $piece, $fin->{-or} ] : $piece;
-      }
-      elsif (! exists $fin->{''}) {
-        $fin->{''} = $piece;
+  my $main_src_fixed_cols_from_cond = [ $attrs->{where}
+    ? (
+      map
+      {
+        ( $colinfos->{$_} and $colinfos->{$_}{-source_alias} eq $attrs->{alias} )
+          ? $colinfos->{$_}{-colname}
+          : ()
       }
-    }
-
-    return $fin;
-  }
-  elsif (ref $where eq 'ARRAY') {
-    my @w = @$where;
-
-    while ( @w and (
-      (ref $w[0] eq 'ARRAY' and ! @{$w[0]} )
-        or
-      (ref $w[0] eq 'HASH' and ! keys %{$w[0]})
-    )) { shift @w };
+      keys %{ extract_equality_conditions( $attrs->{where} ) }
+    )
+    : ()
+  ];
 
-    return unless @w;
+  return $attrs->{result_source}->_identifying_column_set([
+    keys %$seen_main_src_cols,
+    @$main_src_fixed_cols_from_cond,
+  ]) ? $colinfos_to_return : ();
+}
 
-    if ( @w == 1 ) {
-      return ( ref $w[0] )
-        ? $self->_collapse_cond($w[0])
-        : { $w[0] => undef }
-      ;
-    }
-    elsif ( ref $w[0] ) {
-      return \@w;
-    }
-    elsif ( @w == 2 ) {
-      if ( ( $w[0]||'' ) =~ /^\-and$/i ) {
-        return (ref $w[1] eq 'HASH' or ref $w[1] eq 'ARRAY')
-          ? $self->_collapse_cond($w[1], (ref $w[1] eq 'ARRAY') )
-          : $self->throw_exception("Unsupported top-level op/arg pair: [ $w[0] => $w[1] ]")
-        ;
-      }
-      else {
-        return $self->_collapse_cond({ @w });
-      }
-    }
-  }
-  else {
-    # not a hash not an array
-    return { '' => $where };
-  }
+sub _resolve_column_info :DBIC_method_is_indirect_sugar {
+  DBIx::Class::_ENV_::ASSERT_NO_INTERNAL_INDIRECT_CALLS and fail_on_internal_call;
+  carp_unique("_resolve_column_info() is deprecated, ask on IRC for a better alternative");
 
-  # catchall, some of the things above fall through
-  return $where;
+  fromspec_columns_info( @_[1,2] );
 }
 
-sub _collapse_cond_unroll_pairs {
-  my ($self, $pairs) = @_;
-
-  my @conds;
+sub _find_join_path_to_node :DBIC_method_is_indirect_sugar {
+  DBIx::Class::_ENV_::ASSERT_NO_INTERNAL_INDIRECT_CALLS and fail_on_internal_call;
+  carp_unique("_find_join_path_to_node() is deprecated, ask on IRC for a better alternative");
 
-  while (@$pairs) {
-    my ($lhs, $rhs) = @{ shift @$pairs };
+  find_join_path_to_alias( @_[1,2] );
+}
 
-    if ($lhs eq '') {
-      push @conds, $self->_collapse_cond($rhs);
-    }
-    elsif ( $lhs =~ /^\-and$/i ) {
-      push @conds, $self->_collapse_cond($rhs, (ref $rhs eq 'ARRAY'));
-    }
-    elsif ( $lhs =~ /^\-or$/i ) {
-      push @conds, $self->_collapse_cond(
-        (ref $rhs eq 'HASH') ? [ map { $_ => $rhs->{$_} } sort keys %$rhs ] : $rhs
-      );
-    }
-    else {
-      if (ref $rhs eq 'HASH' and ! keys %$rhs) {
-        # FIXME - SQLA seems to be doing... nothing...?
-      }
-      elsif (ref $rhs eq 'HASH' and keys %$rhs == 1 and exists $rhs->{'='}) {
-        for my $p ($self->_collapse_cond_unroll_pairs([ [ $lhs => $rhs->{'='} ] ])) {
-
-          # extra sanity check
-          if (keys %$p > 1) {
-            require Data::Dumper::Concise;
-            local $Data::Dumper::Deepcopy = 1;
-            $self->throw_exception(
-              "Internal error: unexpected collapse unroll:"
-            . Data::Dumper::Concise::Dumper { in => { $lhs => $rhs }, out => $p }
-            );
-          }
+sub _collapse_cond :DBIC_method_is_indirect_sugar {
+  DBIx::Class::_ENV_::ASSERT_NO_INTERNAL_INDIRECT_CALLS and fail_on_internal_call;
+  carp_unique("_collapse_cond() is deprecated, ask on IRC for a better alternative");
 
-          my ($l, $r) = %$p;
+  shift;
+  DBIx::Class::SQLMaker::Util::normalize_sqla_condition(@_);
+}
 
-          push @conds, ( ! ref $r or overload::Method($r, '""' ) )
-            ? { $l => $r }
-            : { $l => { '=' => $r } }
-          ;
-        }
-      }
-      elsif (ref $rhs eq 'ARRAY') {
-        # some of these conditionals encounter multi-values - roll them out using
-        # an unshift, which will cause extra looping in the while{} above
-        if (! @$rhs ) {
-          push @conds, { $lhs => [] };
-        }
-        elsif ( ($rhs->[0]||'') =~ /^\-(?:and|or)$/i ) {
-          $self->throw_exception("Value modifier not followed by any values: $lhs => [ $rhs->[0] ] ")
-            if  @$rhs == 1;
-
-          if( $rhs->[0] =~ /^\-and$/i ) {
-            unshift @$pairs, map { [ $lhs => $_ ] } @{$rhs}[1..$#$rhs];
-          }
-          # if not an AND then it's an OR
-          elsif(@$rhs == 2) {
-            unshift @$pairs, [ $lhs => $rhs->[1] ];
-          }
-          else {
-            push @conds, { $lhs => $rhs };
-          }
-        }
-        elsif (@$rhs == 1) {
-          unshift @$pairs, [ $lhs => $rhs->[0] ];
-        }
-        else {
-          push @conds, { $lhs => $rhs };
-        }
-      }
-      else {
-        push @conds, { $lhs => $rhs };
-      }
-    }
-  }
+sub _extract_fixed_condition_columns :DBIC_method_is_indirect_sugar {
+  DBIx::Class::_ENV_::ASSERT_NO_INTERNAL_INDIRECT_CALLS and fail_on_internal_call;
+  carp_unique("_extract_fixed_condition_columns() is deprecated, ask on IRC for a better alternative");
 
-  return @conds;
+  shift;
+  extract_equality_conditions(@_);
 }
 
+sub _resolve_ident_sources :DBIC_method_is_indirect_sugar {
+  DBIx::Class::Exception->throw(
+    '_resolve_ident_sources() has been removed with no replacement, '
+  . 'ask for advice on IRC if this affected you'
+  );
+}
 
-# returns an arrayref of column names which *definitely* have some
-# sort of non-nullable *single* 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
-sub _extract_fixed_condition_columns {
-  my $self = shift;
-  my $where_hash = $self->_collapse_cond(shift);
-
-  my $res;
-  for my $c (keys %$where_hash) {
-    if (defined (my $v = $where_hash->{$c}) ) {
-      if (
-        ! ref $v
-          or
-        (ref $v eq 'HASH' and keys %$v == 1 and defined $v->{'='} and (
-          ! ref $v->{'='}
-            or
-          ref $v->{'='} eq 'SCALAR'
-            or
-          ( ref $v->{'='} eq 'REF' and ref ${$v->{'='}} eq 'ARRAY' )
-            or
-          overload::Method($v->{'='}, '""')
-        ))
-      ) {
-        $res->{$c} = 1;
-      }
-      elsif (ref $v eq 'ARRAY' and ($v->[0]||'') eq '-and') {
-        $res->{$_} = 1 for map { @{ $self->_extract_fixed_condition_columns({ $c => $_ }) } } @{$v}[1..$#$v];
-      }
-    }
-  }
-
-  return [ sort keys %$res ];
+sub _inner_join_to_node :DBIC_method_is_indirect_sugar {
+  DBIx::Class::Exception->throw(
+    '_inner_join_to_node() has been removed with no replacement, '
+  . 'ask for advice on IRC if this affected you'
+  );
 }
 
 1;