X-Git-Url: http://git.shadowcat.co.uk/gitweb/gitweb.cgi?a=blobdiff_plain;f=lib%2FDBIx%2FClass%2FSQLMaker.pm;h=245847b5b4c4a3f62ab5f7ba0930bd8c25cfc45f;hb=25c1ed5f32dfcefdc12df885669e1dcf9baabbe9;hp=14fbb291ddf1c6cd90bd05ebab093075884963ed;hpb=1b5ddf236a7911b6686be06602ffd657372523c2;p=dbsrgits%2FDBIx-Class.git diff --git a/lib/DBIx/Class/SQLMaker.pm b/lib/DBIx/Class/SQLMaker.pm index 14fbb29..245847b 100644 --- a/lib/DBIx/Class/SQLMaker.pm +++ b/lib/DBIx/Class/SQLMaker.pm @@ -9,13 +9,13 @@ DBIx::Class::SQLMaker - An SQL::Abstract-based SQL maker class =head1 DESCRIPTION -This module is a subclass of L and includes a number of -DBIC-specific workarounds, not yet suitable for inclusion into the +This module is currently a subclass of L and includes a number of +DBIC-specific extensions/workarounds, not suitable for inclusion into the L core. It also provides all (and more than) the functionality of L, see L for more info. -Currently the enhancements to L are: +Currently the enhancements over L are: =over @@ -25,10 +25,102 @@ Currently the enhancements to L are: =item * C/C support (via extensions to the order_by parameter) +=item * A rudimentary multicolumn IN operator + =item * Support of C<...FOR UPDATE> type of select statement modifiers =back +=head1 ROADMAP + +Some maintainer musings on the current state of SQL generation within DBIC as +of Oct 2015 + +=head2 Folding of most (or all) of L into DBIC + +The rise of complex prefetch use, and the general streamlining of result +parsing within DBIC ended up pushing the actual SQL generation to the forefront +of many casual performance profiles. While the idea behind SQLA's API is sound, +the actual implementation is terribly inefficient (once again bumping into the +ridiculously high overhead of perl function calls). + +Given that SQLA has a B distinct life on its own, and is used within an +order of magnitude more projects compared to DBIC, it is prudent to B +disturb the current call chains within SQLA itself. Instead in the near future +an effort will be undertaken to seek a more thorough decoupling of DBIC SQL +generation from reliance on SQLA, possibly to a point where B at all. + +B library itself will continue being maintained> although +it is not likely to gain many extra features, notably dialect support, at least +not within the base C namespace. + +This work (if undertaken) will take into consideration the following +constraints: + +=over + +=item Main API compatibility + +The object returned by C<< $schema->storage->sqlmaker >> needs to be able to +satisfy most of the basic tests found in the current-at-the-time SQLA dist. +While things like L or L +or even worse L will definitely remain +unsupported, the rest of the tests should pass (within reason). + +=item Ability to plug back an SQL::Abstract (or derivative) + +During the initial work on L the test suite of DBIC turned out to +be an invaluable asset to iron out hard-to-reason-about corner cases. In +addition the test suite is much more vast and intricate than the tests of SQLA +itself. This state of affairs is way too valuable to sacrifice in order to gain +faster SQL generation. Thus a compile-time-ENV-check will be introduced along +with an extra CI configuration to ensure that DBIC is used with an off-the-CPAN +SQLA and that it continues to flawlessly run its entire test suite. While this +will undoubtedly complicate the implementation of the better performing SQL +generator, it will preserve both the usability of the test suite for external +projects and will keep L from regressions in the future. + +=back + +Aside from these constraints it is becoming more and more practical to simply +stop using SQLA in day-to-day production deployments of DBIC. The flexibility +of the internals is simply not worth the performance cost. + +=head2 Relationship to L + +When initial work on DQ was taking place, the tools in L<::Storage::DBIHacks +|http://github.com/dbsrgits/dbix-class/blob/current/blead/lib/DBIx/Class/Storage/DBIHacks.pm> +were only beginning to take shape, and it wasn't clear how important they will +become further down the road. In fact the I was +considered an ugly stop-gap, and even a couple of highly entertaining talks +were given to that effect. As the use-cases of DBIC were progressing, and +evidence for the importance of supporting arbitrary SQL was mounting, it became +clearer that DBIC itself would not really benefit in any way from an +integration with DQ, but on the contrary is likely to lose functionality while +the corners of the brand new DQ codebase are sanded off. + +The current status of DBIC/DQ integration is that the only benefit is for DQ by +having access to the very extensive "early adopter" test suite, in the same +manner as early DBIC benefitted tremendously from usurping the Class::DBI test +suite. As far as the DBIC user-base - there are no immediate practical upsides +to DQ integration, neither in terms of API nor in performance. + +So (as described higher up) the DBIC development effort will in the foreseable +future ignore the existence of DQ, and will continue optimizing the preexisting +SQLA-based solution, potentially "organically growing" its own compatible +implementation. Also (again, as described higher up) the ability to plug a +separate SQLA-compatible class providing the necessary surface API will remain +possible, and will be protected at all costs in order to continue providing DQ +access to the test cases of DBIC. + +In the short term, after one more pass over the ResultSet internals is +undertaken I, and before the SQLA/SQLMaker integration +takes place, the preexisting DQ-based branches will be pulled/modified/rebased +to get up-to-date with the current state of the codebase, which changed very +substantially since the last migration effort, especially in the SQL +classification meta-parsing codepath. + =cut use base qw/ @@ -44,8 +136,16 @@ use namespace::clean; __PACKAGE__->mk_group_accessors (simple => qw/quote_char name_sep limit_dialect/); +sub _quoting_enabled { + ( defined $_[0]->{quote_char} and length $_[0]->{quote_char} ) ? 1 : 0 +} + # for when I need a normalized l/r pair sub _quote_chars { + + # in case we are called in the old !!$sm->_quote_chars fashion + return () if !wantarray and ( ! defined $_[0]->{quote_char} or ! length $_[0]->{quote_char} ); + map { defined $_ ? $_ : '' } ( ref $_[0]->{quote_char} ? (@{$_[0]->{quote_char}}) : ( ($_[0]->{quote_char}) x 2 ) ) @@ -110,17 +210,17 @@ sub select { my ($self, $table, $fields, $where, $rs_attrs, $limit, $offset) = @_; - $fields = $self->_recurse_fields($fields); + ($fields, @{$self->{select_bind}}) = $self->_recurse_fields($fields); if (defined $offset) { $self->throw_exception('A supplied offset must be a non-negative integer') - if ( $offset =~ /\D/ or $offset < 0 ); + if ( $offset =~ /[^0-9]/ or $offset < 0 ); } $offset ||= 0; if (defined $limit) { $self->throw_exception('A supplied limit must be a positive integer') - if ( $limit =~ /\D/ or $limit <= 0 ); + if ( $limit =~ /[^0-9]/ or $limit <= 0 ); } elsif ($offset) { $limit = $self->__max_int; @@ -138,8 +238,9 @@ sub select { if( $limiter = $self->can ('emulate_limit') ) { carp_unique( 'Support for the legacy emulate_limit() mechanism inherited from ' - . 'SQL::Abstract::Limit has been deprecated, and will be removed when ' - . 'DBIC transitions to Data::Query. If your code uses this type of ' + . 'SQL::Abstract::Limit has been deprecated, and will be removed at ' + . 'some future point, as it gets in the way of architectural and/or ' + . 'performance advances within DBIC. If your code uses this type of ' . 'limit specification please file an RT and provide the source of ' . 'your emulate_limit() implementation, so an acceptable upgrade-path ' . 'can be devised' @@ -203,9 +304,9 @@ sub insert { # optimized due to hotttnesss # my ($self, $table, $data, $options) = @_; - # SQLA will emit INSERT INTO $table ( ) VALUES ( ) + # FIXME SQLA will emit INSERT INTO $table ( ) VALUES ( ) # which is sadly understood only by MySQL. Change default behavior here, - # until SQLA2 comes with proper dialect support + # until we fold the extra pieces into SQLMaker properly if (! $_[2] or (ref $_[2] eq 'HASH' and !keys %{$_[2]} ) ) { my @bind; my $sql = sprintf( @@ -231,42 +332,47 @@ sub _recurse_fields { return $$fields if $ref eq 'SCALAR'; if ($ref eq 'ARRAY') { - return join(', ', map { $self->_recurse_fields($_) } @$fields); + my (@select, @bind); + for my $field (@$fields) { + my ($select, @new_bind) = $self->_recurse_fields($field); + push @select, $select; + push @bind, @new_bind; + } + return (join(', ', @select), @bind); } elsif ($ref eq 'HASH') { my %hash = %$fields; # shallow copy my $as = delete $hash{-as}; # if supplied - my ($func, $args, @toomany) = %hash; + my ($func, $rhs, @toomany) = %hash; # there should be only one pair if (@toomany) { $self->throw_exception( "Malformed select argument - too many keys in hash: " . join (',', keys %$fields ) ); } - if (lc ($func) eq 'distinct' && ref $args eq 'ARRAY' && @$args > 1) { + if (lc ($func) eq 'distinct' && ref $rhs eq 'ARRAY' && @$rhs > 1) { $self->throw_exception ( 'The select => { distinct => ... } syntax is not supported for multiple columns.' - .' Instead please use { group_by => [ qw/' . (join ' ', @$args) . '/ ] }' - .' or { select => [ qw/' . (join ' ', @$args) . '/ ], distinct => 1 }' + .' Instead please use { group_by => [ qw/' . (join ' ', @$rhs) . '/ ] }' + .' or { select => [ qw/' . (join ' ', @$rhs) . '/ ], distinct => 1 }' ); } + my ($rhs_sql, @rhs_bind) = $self->_recurse_fields($rhs); my $select = sprintf ('%s( %s )%s', $self->_sqlcase($func), - $self->_recurse_fields($args), + $rhs_sql, $as ? sprintf (' %s %s', $self->_sqlcase('as'), $self->_quote ($as) ) : '' ); - return $select; + return ($select, @rhs_bind); } - # Is the second check absolutely necessary? elsif ( $ref eq 'REF' and ref($$fields) eq 'ARRAY' ) { - push @{$self->{select_bind}}, @{$$fields}[1..$#$$fields]; - return $$fields->[0]; + return @{$$fields}; } else { $self->throw_exception( $ref . qq{ unexpected in _recurse_fields()} ); @@ -278,31 +384,39 @@ sub _recurse_fields { # What we have been doing forever is hijacking the $order arg of # SQLA::select to pass in arbitrary pieces of data (first the group_by, # then pretty much the entire resultset attr-hash, as more and more -# things in the SQLA space need to have mopre info about the $rs they +# things in the SQLA space need to have more info about the $rs they # create SQL for. The alternative would be to keep expanding the # signature of _select with more and more positional parameters, which -# is just gross. All hail SQLA2! +# is just gross. +# +# FIXME - this will have to transition out to a subclass when the effort +# of folding the SQLA machinery into SQLMaker takes place sub _parse_rs_attrs { my ($self, $arg) = @_; my $sql = ''; + my @sqlbind; - if ($arg->{group_by}) { - # horible horrible, waiting for refactor - local $self->{select_bind}; - if (my $g = $self->_recurse_fields($arg->{group_by}) ) { - $sql .= $self->_sqlcase(' group by ') . $g; - push @{$self->{group_bind} ||= []}, @{$self->{select_bind}||[]}; - } + if ( + $arg->{group_by} + and + @sqlbind = $self->_recurse_fields($arg->{group_by}) + ) { + $sql .= $self->_sqlcase(' group by ') . shift @sqlbind; + push @{$self->{group_bind}}, @sqlbind; } - if (defined $arg->{having}) { - my ($frag, @bind) = $self->_recurse_where($arg->{having}); - push(@{$self->{having_bind}}, @bind); - $sql .= $self->_sqlcase(' having ') . $frag; + if ( + $arg->{having} + and + @sqlbind = $self->_recurse_where($arg->{having}) + ) { + $sql .= $self->_sqlcase(' having ') . shift @sqlbind; + push(@{$self->{having_bind}}, @sqlbind); } - if (defined $arg->{order_by}) { + if ($arg->{order_by}) { + # unlike the 2 above, _order_by injects into @{...bind...} for us $sql .= $self->_order_by ($arg->{order_by}); } @@ -313,14 +427,18 @@ sub _order_by { my ($self, $arg) = @_; # check that we are not called in legacy mode (order_by as 4th argument) - if (ref $arg eq 'HASH' and not grep { $_ =~ /^-(?:desc|asc)/i } keys %$arg ) { - return $self->_parse_rs_attrs ($arg); - } - else { - my ($sql, @bind) = $self->next::method($arg); - push @{$self->{order_bind}}, @bind; - return $sql; - } + ( + ref $arg eq 'HASH' + and + not grep { $_ =~ /^-(?:desc|asc)/i } keys %$arg + ) + ? $self->_parse_rs_attrs ($arg) + : do { + my ($sql, @bind) = $self->next::method($arg); + push @{$self->{order_bind}}, @bind; + $sql; # RV + } + ; } sub _split_order_chunk { @@ -441,8 +559,6 @@ sub _join_condition { # Backcompat for the old days when a plain hashref # { 't1.col1' => 't2.col2' } meant ON t1.col1 = t2.col2 - # Once things settle we should start warning here so that - # folks unroll their hacks if ( ref $cond eq 'HASH' and @@ -452,6 +568,12 @@ sub _join_condition { and ! ref ( (values %$cond)[0] ) ) { + carp_unique( + "ResultSet {from} structures with conditions not conforming to the " + . "SQL::Abstract syntax are deprecated: you either need to stop abusing " + . "{from} altogether, or express the condition properly using the " + . "{ -ident => ... } operator" + ); $cond = { keys %$cond => { -ident => values %$cond } } } elsif ( ref $cond eq 'ARRAY' ) { @@ -469,9 +591,14 @@ sub _join_condition { return $self->_recurse_where($cond); } -# This is hideously ugly, but SQLA does not understand multicol IN expressions -# FIXME TEMPORARY - DQ should have native syntax for this -# moved here to raise API questions +# !!! EXPERIMENTAL API !!! WILL CHANGE !!! +# +# This is rather odd, but vanilla SQLA does not have support for multicolumn IN +# expressions +# Currently has only one callsite in ResultSet, body moved into this subclass +# of SQLA to raise API questions like: +# - how do we convey a list of idents...? +# - can binds reside on lhs? # # !!! EXPERIMENTAL API !!! WILL CHANGE !!! sub _where_op_multicolumn_in { @@ -512,20 +639,23 @@ sub _where_op_multicolumn_in { for ($lhs, $rhs) { $$_->[0] = "( $$_->[0] )" - unless $$_->[0] =~ /^ \s* \( .* \) \s* ^/xs; + unless $$_->[0] =~ /^ \s* \( .* \) \s* $/xs; } \[ join( ' IN ', shift @$$lhs, shift @$$rhs ), @$$lhs, @$$rhs ]; } -1; - -=head1 AUTHORS +=head1 FURTHER QUESTIONS? -See L. +Check the list of L. -=head1 LICENSE +=head1 COPYRIGHT AND LICENSE -You may distribute this code under the same terms as Perl itself. +This module is free software L +by the L. You can +redistribute it and/or modify it under the same terms as the +L. =cut + +1;