1 package SQL::Abstract; # see doc at end of file
10 our @EXPORT_OK = qw(is_plain_value is_literal_value);
20 *SQL::Abstract::_ENV_::DETECT_AUTOGENERATED_STRINGIFICATION = $ENV{SQLA_ISVALUE_IGNORE_AUTOGENERATED_STRINGIFICATION}
26 #======================================================================
28 #======================================================================
30 our $VERSION = '1.86';
32 # This would confuse some packagers
33 $VERSION = eval $VERSION if $VERSION =~ /_/; # numify for warning-free dev releases
37 # special operators (-in, -between). May be extended/overridden by user.
38 # See section WHERE: BUILTIN SPECIAL OPERATORS below for implementation
39 my @BUILTIN_SPECIAL_OPS = (
40 {regex => qr/^ (?: not \s )? between $/ix, handler => sub { die "NOPE" }},
41 {regex => qr/^ is (?: \s+ not )? $/ix, handler => sub { die "NOPE" }},
42 {regex => qr/^ (?: not \s )? in $/ix, handler => sub { die "NOPE" }},
45 #======================================================================
46 # DEBUGGING AND ERROR REPORTING
47 #======================================================================
50 return unless $_[0]->{debug}; shift; # a little faster
51 my $func = (caller(1))[3];
52 warn "[$func] ", @_, "\n";
56 my($func) = (caller(1))[3];
57 Carp::carp "[$func] Warning: ", @_;
61 my($func) = (caller(1))[3];
62 Carp::croak "[$func] Fatal: ", @_;
65 sub is_literal_value ($) {
66 ref $_[0] eq 'SCALAR' ? [ ${$_[0]} ]
67 : ( ref $_[0] eq 'REF' and ref ${$_[0]} eq 'ARRAY' ) ? [ @${ $_[0] } ]
71 sub is_undef_value ($) {
75 and exists $_[0]->{-value}
76 and not defined $_[0]->{-value}
80 # FIXME XSify - this can be done so much more efficiently
81 sub is_plain_value ($) {
83 ! length ref $_[0] ? \($_[0])
85 ref $_[0] eq 'HASH' and keys %{$_[0]} == 1
87 exists $_[0]->{-value}
88 ) ? \($_[0]->{-value})
90 # reuse @_ for even moar speedz
91 defined ( $_[1] = Scalar::Util::blessed $_[0] )
93 # deliberately not using Devel::OverloadInfo - the checks we are
94 # intersted in are much more limited than the fullblown thing, and
95 # this is a very hot piece of code
97 # simply using ->can('(""') can leave behind stub methods that
98 # break actually using the overload later (see L<perldiag/Stub
99 # found while resolving method "%s" overloading "%s" in package
100 # "%s"> and the source of overload::mycan())
102 # either has stringification which DBI SHOULD prefer out of the box
103 grep { *{ (qq[${_}::(""]) }{CODE} } @{ $_[2] = mro::get_linear_isa( $_[1] ) }
105 # has nummification or boolification, AND fallback is *not* disabled
107 SQL::Abstract::_ENV_::DETECT_AUTOGENERATED_STRINGIFICATION
110 grep { *{"${_}::(0+"}{CODE} } @{$_[2]}
112 grep { *{"${_}::(bool"}{CODE} } @{$_[2]}
116 # no fallback specified at all
117 ! ( ($_[3]) = grep { *{"${_}::()"}{CODE} } @{$_[2]} )
119 # fallback explicitly undef
120 ! defined ${"$_[3]::()"}
133 #======================================================================
135 #======================================================================
139 my $class = ref($self) || $self;
140 my %opt = (ref $_[0] eq 'HASH') ? %{$_[0]} : @_;
142 # choose our case by keeping an option around
143 delete $opt{case} if $opt{case} && $opt{case} ne 'lower';
145 # default logic for interpreting arrayrefs
146 $opt{logic} = $opt{logic} ? uc $opt{logic} : 'OR';
148 # how to return bind vars
149 $opt{bindtype} ||= 'normal';
151 # default comparison is "=", but can be overridden
154 # try to recognize which are the 'equality' and 'inequality' ops
155 # (temporary quickfix (in 2007), should go through a more seasoned API)
156 $opt{equality_op} = qr/^( \Q$opt{cmp}\E | \= )$/ix;
157 $opt{inequality_op} = qr/^( != | <> )$/ix;
159 $opt{like_op} = qr/^ (is_)?r?like $/xi;
160 $opt{not_like_op} = qr/^ (is_)?not_r?like $/xi;
163 $opt{sqltrue} ||= '1=1';
164 $opt{sqlfalse} ||= '0=1';
167 $opt{special_ops} ||= [];
169 # regexes are applied in order, thus push after user-defines
170 push @{$opt{special_ops}}, @BUILTIN_SPECIAL_OPS;
172 if ($class->isa('DBIx::Class::SQLMaker')) {
173 push @{$opt{special_ops}}, our $DBIC_Compat_Op ||= {
174 regex => qr/^(?:ident|value|(?:not\s)?in)$/i, handler => sub { die "NOPE" }
176 $opt{is_dbic_sqlmaker} = 1;
180 $opt{unary_ops} ||= [];
182 # rudimentary sanity-check for user supplied bits treated as functions/operators
183 # If a purported function matches this regular expression, an exception is thrown.
184 # Literal SQL is *NOT* subject to this check, only functions (and column names
185 # when quoting is not in effect)
188 # need to guard against ()'s in column names too, but this will break tons of
189 # hacks... ideas anyone?
190 $opt{injection_guard} ||= qr/
196 $opt{expand_unary} = {};
199 -not => '_expand_not',
200 -bool => '_expand_bool',
201 -and => '_expand_op_andor',
202 -or => '_expand_op_andor',
203 -nest => '_expand_nest',
204 -bind => sub { shift; +{ @_ } },
208 'between' => '_expand_between',
209 'not_between' => '_expand_between',
210 'in' => '_expand_in',
211 'not_in' => '_expand_in',
212 'nest' => '_expand_nest',
213 (map +($_ => '_expand_op_andor'), ('and', 'or')),
214 (map +($_ => '_expand_op_is'), ('is', 'is_not')),
217 # placeholder for _expand_unop system
219 my %unops = (-ident => '_expand_ident', -value => '_expand_value');
220 foreach my $name (keys %unops) {
221 $opt{expand}{$name} = $unops{$name};
222 my ($op) = $name =~ /^-(.*)$/;
223 $opt{expand_op}{$op} = sub {
224 my ($self, $op, $arg, $k) = @_;
225 return $self->_expand_expr_hashpair_cmp(
226 $k, { "-${op}" => $arg }
233 (map +("-$_", "_render_$_"), qw(op func bind ident literal list)),
238 (map +($_ => '_render_op_between'), 'between', 'not_between'),
239 (map +($_ => '_render_op_in'), 'in', 'not_in'),
240 (map +($_ => '_render_unop_postfix'),
241 'is_null', 'is_not_null', 'asc', 'desc',
243 (not => '_render_op_not'),
244 (map +($_ => '_render_op_andor'), qw(and or)),
247 return bless \%opt, $class;
250 sub sqltrue { +{ -literal => [ $_[0]->{sqltrue} ] } }
251 sub sqlfalse { +{ -literal => [ $_[0]->{sqlfalse} ] } }
253 sub _assert_pass_injection_guard {
254 if ($_[1] =~ $_[0]->{injection_guard}) {
255 my $class = ref $_[0];
256 puke "Possible SQL injection attempt '$_[1]'. If this is indeed a part of the "
257 . "desired SQL use literal SQL ( \'...' or \[ '...' ] ) or supply your own "
258 . "{injection_guard} attribute to ${class}->new()"
263 #======================================================================
265 #======================================================================
269 my $table = $self->_table(shift);
270 my $data = shift || return;
273 my $method = $self->_METHOD_FOR_refkind("_insert", $data);
274 my ($sql, @bind) = $self->$method($data);
275 $sql = join " ", $self->_sqlcase('insert into'), $table, $sql;
277 if ($options->{returning}) {
278 my ($s, @b) = $self->_insert_returning($options);
283 return wantarray ? ($sql, @bind) : $sql;
286 # So that subclasses can override INSERT ... RETURNING separately from
287 # UPDATE and DELETE (e.g. DBIx::Class::SQLMaker::Oracle does this)
288 sub _insert_returning { shift->_returning(@_) }
291 my ($self, $options) = @_;
293 my $f = $options->{returning};
295 my ($sql, @bind) = $self->render_aqt(
296 $self->_expand_maybe_list_expr($f, undef, -ident)
299 ? $self->_sqlcase(' returning ') . $sql
300 : ($self->_sqlcase(' returning ').$sql, @bind);
303 sub _insert_HASHREF { # explicit list of fields and then values
304 my ($self, $data) = @_;
306 my @fields = sort keys %$data;
308 my ($sql, @bind) = $self->_insert_values($data);
311 $_ = $self->_quote($_) foreach @fields;
312 $sql = "( ".join(", ", @fields).") ".$sql;
314 return ($sql, @bind);
317 sub _insert_ARRAYREF { # just generate values(?,?) part (no list of fields)
318 my ($self, $data) = @_;
320 # no names (arrayref) so can't generate bindtype
321 $self->{bindtype} ne 'columns'
322 or belch "can't do 'columns' bindtype when called with arrayref";
324 my (@values, @all_bind);
325 foreach my $value (@$data) {
326 my ($values, @bind) = $self->_insert_value(undef, $value);
327 push @values, $values;
328 push @all_bind, @bind;
330 my $sql = $self->_sqlcase('values')." ( ".join(", ", @values)." )";
331 return ($sql, @all_bind);
334 sub _insert_ARRAYREFREF { # literal SQL with bind
335 my ($self, $data) = @_;
337 my ($sql, @bind) = @${$data};
338 $self->_assert_bindval_matches_bindtype(@bind);
340 return ($sql, @bind);
344 sub _insert_SCALARREF { # literal SQL without bind
345 my ($self, $data) = @_;
351 my ($self, $data) = @_;
353 my (@values, @all_bind);
354 foreach my $column (sort keys %$data) {
355 my ($values, @bind) = $self->_insert_value($column, $data->{$column});
356 push @values, $values;
357 push @all_bind, @bind;
359 my $sql = $self->_sqlcase('values')." ( ".join(", ", @values)." )";
360 return ($sql, @all_bind);
364 my ($self, $column, $v) = @_;
366 return $self->render_aqt(
367 $self->_expand_insert_value($column, $v)
371 sub _expand_insert_value {
372 my ($self, $column, $v) = @_;
374 if (ref($v) eq 'ARRAY') {
375 if ($self->{array_datatypes}) {
376 return +{ -bind => [ $column, $v ] };
378 my ($sql, @bind) = @$v;
379 $self->_assert_bindval_matches_bindtype(@bind);
380 return +{ -literal => $v };
382 if (ref($v) eq 'HASH') {
383 if (grep !/^-/, keys %$v) {
384 belch "HASH ref as bind value in insert is not supported";
385 return +{ -bind => [ $column, $v ] };
389 return +{ -bind => [ $column, undef ] };
391 local our $Cur_Col_Meta = $column;
392 return $self->expand_expr($v);
397 #======================================================================
399 #======================================================================
404 my $table = $self->_table(shift);
405 my $data = shift || return;
409 # first build the 'SET' part of the sql statement
410 puke "Unsupported data type specified to \$sql->update"
411 unless ref $data eq 'HASH';
413 my ($sql, @all_bind) = $self->_update_set_values($data);
414 $sql = $self->_sqlcase('update ') . $table . $self->_sqlcase(' set ')
418 my($where_sql, @where_bind) = $self->where($where);
420 push @all_bind, @where_bind;
423 if ($options->{returning}) {
424 my ($returning_sql, @returning_bind) = $self->_update_returning($options);
425 $sql .= $returning_sql;
426 push @all_bind, @returning_bind;
429 return wantarray ? ($sql, @all_bind) : $sql;
432 sub _update_set_values {
433 my ($self, $data) = @_;
435 return $self->render_aqt(
436 $self->_expand_update_set_values($data),
440 sub _expand_update_set_values {
441 my ($self, $data) = @_;
442 $self->_expand_maybe_list_expr( [
445 $set = { -bind => $_ } unless defined $set;
446 +{ -op => [ '=', $self->_expand_ident(-ident => $k), $set ] };
452 ? ($self->{array_datatypes}
453 ? [ $k, +{ -bind => [ $k, $v ] } ]
454 : [ $k, +{ -literal => $v } ])
456 local our $Cur_Col_Meta = $k;
457 [ $k, $self->_expand_expr($v) ]
464 # So that subclasses can override UPDATE ... RETURNING separately from
466 sub _update_returning { shift->_returning(@_) }
470 #======================================================================
472 #======================================================================
477 my $table = $self->_table(shift);
478 my $fields = shift || '*';
482 my ($fields_sql, @bind) = $self->_select_fields($fields);
484 my ($where_sql, @where_bind) = $self->where($where, $order);
485 push @bind, @where_bind;
487 my $sql = join(' ', $self->_sqlcase('select'), $fields_sql,
488 $self->_sqlcase('from'), $table)
491 return wantarray ? ($sql, @bind) : $sql;
495 my ($self, $fields) = @_;
496 return $fields unless ref($fields);
497 return $self->render_aqt(
498 $self->_expand_maybe_list_expr($fields, undef, '-ident')
502 #======================================================================
504 #======================================================================
509 my $table = $self->_table(shift);
513 my($where_sql, @bind) = $self->where($where);
514 my $sql = $self->_sqlcase('delete from ') . $table . $where_sql;
516 if ($options->{returning}) {
517 my ($returning_sql, @returning_bind) = $self->_delete_returning($options);
518 $sql .= $returning_sql;
519 push @bind, @returning_bind;
522 return wantarray ? ($sql, @bind) : $sql;
525 # So that subclasses can override DELETE ... RETURNING separately from
527 sub _delete_returning { shift->_returning(@_) }
531 #======================================================================
533 #======================================================================
537 # Finally, a separate routine just to handle WHERE clauses
539 my ($self, $where, $order) = @_;
541 local $self->{convert_where} = $self->{convert};
544 my ($sql, @bind) = defined($where)
545 ? $self->_recurse_where($where)
547 $sql = (defined $sql and length $sql) ? $self->_sqlcase(' where ') . "( $sql )" : '';
551 my ($order_sql, @order_bind) = $self->_order_by($order);
553 push @bind, @order_bind;
556 return wantarray ? ($sql, @bind) : $sql;
559 { our $Default_Scalar_To = -value }
562 my ($self, $expr, $default_scalar_to) = @_;
563 local our $Default_Scalar_To = $default_scalar_to if $default_scalar_to;
564 $self->_expand_expr($expr);
568 my ($self, $aqt) = @_;
569 my ($k, $v, @rest) = %$aqt;
571 if (my $meth = $self->{render}{$k}) {
572 return $self->$meth($v);
574 die "notreached: $k";
578 my ($self, $expr) = @_;
579 $self->render_aqt($self->expand_expr($expr));
583 my ($self, $raw) = @_;
584 s/^-(?=.)//, s/\s+/_/g for my $op = lc $raw;
589 my ($self, $expr) = @_;
590 our $Expand_Depth ||= 0; local $Expand_Depth = $Expand_Depth + 1;
591 return undef unless defined($expr);
592 if (ref($expr) eq 'HASH') {
593 return undef unless my $kc = keys %$expr;
595 return $self->_expand_op_andor(-and => $expr);
597 my ($key, $value) = %$expr;
598 if ($key =~ /^-/ and $key =~ s/ [_\s]? \d+ $//x ) {
599 belch 'Use of [and|or|nest]_N modifiers is deprecated and will be removed in SQLA v2.0. '
600 . "You probably wanted ...-and => [ $key => COND1, $key => COND2 ... ]";
602 if (my $exp = $self->{expand}{$key}) {
603 return $self->$exp($key, $value);
605 return $self->_expand_expr_hashpair($key, $value);
607 if (ref($expr) eq 'ARRAY') {
608 my $logic = '-'.lc($self->{logic});
609 return $self->_expand_op_andor($logic, $expr);
611 if (my $literal = is_literal_value($expr)) {
612 return +{ -literal => $literal };
614 if (!ref($expr) or Scalar::Util::blessed($expr)) {
615 return $self->_expand_expr_scalar($expr);
620 sub _expand_expr_hashpair {
621 my ($self, $k, $v) = @_;
622 unless (defined($k) and length($k)) {
623 if (defined($k) and my $literal = is_literal_value($v)) {
624 belch 'Hash-pairs consisting of an empty string with a literal are deprecated, and will be removed in 2.0: use -and => [ $literal ] instead';
625 return { -literal => $literal };
627 puke "Supplying an empty left hand side argument is not supported";
630 return $self->_expand_expr_hashpair_op($k, $v);
632 return $self->_expand_expr_hashpair_ident($k, $v);
635 sub _expand_expr_hashpair_ident {
636 my ($self, $k, $v) = @_;
638 local our $Cur_Col_Meta = $k;
640 # hash with multiple or no elements is andor
642 if (ref($v) eq 'HASH' and keys %$v != 1) {
643 return $self->_expand_op_andor(-and => $v, $k);
646 # undef needs to be re-sent with cmp to achieve IS/IS NOT NULL
648 if (is_undef_value($v)) {
649 return $self->_expand_expr_hashpair_cmp($k => undef);
652 # scalars and objects get expanded as whatever requested or values
654 if (!ref($v) or Scalar::Util::blessed($v)) {
655 return $self->_expand_expr_hashpair_scalar($k, $v);
658 # single key hashref is a hashtriple
660 if (ref($v) eq 'HASH') {
661 return $self->_expand_expr_hashtriple($k, %$v);
664 # arrayref needs re-engineering over the elements
666 if (ref($v) eq 'ARRAY') {
667 return $self->sqlfalse unless @$v;
668 $self->_debug("ARRAY($k) means distribute over elements");
670 $v->[0] =~ /^-(and|or)$/i
671 ? shift(@{$v = [ @$v ]})
672 : '-'.lc($self->{logic} || 'OR')
674 return $self->_expand_op_andor(
679 if (my $literal = is_literal_value($v)) {
681 belch 'Hash-pairs consisting of an empty string with a literal are deprecated, and will be removed in 2.0: use -and => [ $literal ] instead';
684 my ($sql, @bind) = @$literal;
685 if ($self->{bindtype} eq 'columns') {
687 $self->_assert_bindval_matches_bindtype($_);
690 return +{ -literal => [ $self->_quote($k).' '.$sql, @bind ] };
695 sub _expand_expr_scalar {
696 my ($self, $expr) = @_;
698 return $self->_expand_expr({ (our $Default_Scalar_To) => $expr });
701 sub _expand_expr_hashpair_scalar {
702 my ($self, $k, $v) = @_;
704 return $self->_expand_expr_hashpair_cmp(
705 $k, $self->_expand_expr_scalar($v),
709 sub _expand_expr_hashpair_op {
710 my ($self, $k, $v) = @_;
712 $self->_assert_pass_injection_guard($k =~ /\A-(.*)\Z/s);
714 my $op = $self->_normalize_op($k);
716 # Ops prefixed with -not_ get converted
718 if (my ($rest) = $op =~/^not_(.*)$/) {
721 $self->_expand_expr({ "-${rest}", $v })
727 my $op = join(' ', split '_', $op);
729 # the old special op system requires illegality for top-level use
732 (our $Expand_Depth) == 1
733 and List::Util::first { $op =~ $_->{regex} } @{$self->{special_ops}}
735 puke "Illegal use of top-level '-$op'"
738 # the old unary op system means we should touch nothing and let it work
740 if (my $us = List::Util::first { $op =~ $_->{regex} } @{$self->{unary_ops}}) {
741 return { -op => [ $op, $v ] };
745 # an explicit node type is currently assumed to be expanded (this is almost
746 # certainly wrong and there should be expansion anyway)
748 if ($self->{render}{$k}) {
752 # hashref RHS values get expanded and used as op/func args
757 and (keys %$v)[0] =~ /^-/
759 my ($func) = $k =~ /^-(.*)$/;
761 if (List::Util::first { $func =~ $_->{regex} } @{$self->{special_ops}}) {
762 return +{ -op => [ $func, $self->_expand_expr($v) ] };
765 return +{ -func => [ $func, $self->_expand_expr($v) ] };
768 # scalars and literals get simply expanded
770 if (!ref($v) or is_literal_value($v)) {
771 return +{ -op => [ $op, $self->_expand_expr($v) ] };
777 sub _expand_expr_hashpair_cmp {
778 my ($self, $k, $v) = @_;
779 $self->_expand_expr_hashtriple($k, $self->{cmp}, $v);
782 sub _expand_expr_hashtriple {
783 my ($self, $k, $vk, $vv) = @_;
785 my $ik = $self->_expand_ident(-ident => $k);
787 my $op = $self->_normalize_op($vk);
788 $self->_assert_pass_injection_guard($op);
790 if ($op =~ s/ _? \d+ $//x ) {
791 return $self->_expand_expr($k, { $vk, $vv });
793 if (my $x = $self->{expand_op}{$op}) {
794 local our $Cur_Col_Meta = $k;
795 return $self->$x($op, $vv, $k);
799 my $op = join(' ', split '_', $op);
801 if (my $us = List::Util::first { $op =~ $_->{regex} } @{$self->{special_ops}}) {
802 return { -op => [ $op, $ik, $vv ] };
804 if (my $us = List::Util::first { $op =~ $_->{regex} } @{$self->{unary_ops}}) {
808 { -op => [ $op, $vv ] }
812 if (ref($vv) eq 'ARRAY') {
814 my $logic = (defined($raw[0]) and $raw[0] =~ /^-(and|or)$/i)
815 ? shift @raw : '-or';
816 my @values = map +{ $vk => $_ }, @raw;
818 $op =~ $self->{inequality_op}
819 or $op =~ $self->{not_like_op}
821 if (lc($logic) eq '-or' and @values > 1) {
822 belch "A multi-element arrayref as an argument to the inequality op '${\uc(join ' ', split '_', $op)}' "
823 . 'is technically equivalent to an always-true 1=1 (you probably wanted '
824 . "to say ...{ \$inequality_op => [ -and => \@values ] }... instead)"
829 # try to DWIM on equality operators
830 return ($self->_dwim_op_to_is($op,
831 "Supplying an empty arrayref to '%s' is deprecated",
832 "operator '%s' applied on an empty array (field '$k')"
833 ) ? $self->sqlfalse : $self->sqltrue);
835 return $self->_expand_op_andor($logic => \@values, $k);
837 if (is_undef_value($vv)) {
838 my $is = ($self->_dwim_op_to_is($op,
839 "Supplying an undefined argument to '%s' is deprecated",
840 "unexpected operator '%s' with undef operand",
841 ) ? 'is' : 'is not');
843 return $self->_expand_expr_hashpair($k => { $is, undef });
845 local our $Cur_Col_Meta = $k;
849 $self->_expand_expr($vv)
854 my ($self, $raw, $empty, $fail) = @_;
856 my $op = $self->_normalize_op($raw);
858 if ($op =~ /^not$/i) {
861 if ($op =~ $self->{equality_op}) {
864 if ($op =~ $self->{like_op}) {
865 belch(sprintf $empty, uc(join ' ', split '_', $op));
868 if ($op =~ $self->{inequality_op}) {
871 if ($op =~ $self->{not_like_op}) {
872 belch(sprintf $empty, uc(join ' ', split '_', $op));
875 puke(sprintf $fail, $op);
879 my ($self, $op, $body) = @_;
880 unless (defined($body) or (ref($body) and ref($body) eq 'ARRAY')) {
881 puke "$op requires a single plain scalar argument (a quotable identifier) or an arrayref of identifier parts";
883 my @parts = map split(/\Q${\($self->{name_sep}||'.')}\E/, $_),
884 ref($body) ? @$body : $body;
885 return { -ident => $parts[-1] } if $self->{_dequalify_idents};
886 unless ($self->{quote_char}) {
887 $self->_assert_pass_injection_guard($_) for @parts;
889 return +{ -ident => \@parts };
893 +{ -bind => [ our $Cur_Col_Meta, $_[2] ] };
897 +{ -op => [ 'not', $_[0]->_expand_expr($_[2]) ] };
901 my ($self, undef, $v) = @_;
903 return $self->_expand_expr($v);
905 puke "-bool => undef not supported" unless defined($v);
906 return $self->_expand_ident(-ident => $v);
909 sub _expand_op_andor {
910 my ($self, $logic, $v, $k) = @_;
912 $v = [ map +{ $k, $_ },
914 ? (map +{ $_ => $v->{$_} }, sort keys %$v)
918 my ($logop) = $logic =~ /^-?(.*)$/;
919 if (ref($v) eq 'HASH') {
920 return undef unless keys %$v;
923 map $self->_expand_expr({ $_ => $v->{$_} }),
927 if (ref($v) eq 'ARRAY') {
928 $logop eq 'and' or $logop eq 'or' or puke "unknown logic: $logop";
931 (ref($_) eq 'ARRAY' and @$_)
932 or (ref($_) eq 'HASH' and %$_)
938 while (my ($el) = splice @expr, 0, 1) {
939 puke "Supplying an empty left hand side argument is not supported in array-pairs"
940 unless defined($el) and length($el);
941 my $elref = ref($el);
943 local our $Expand_Depth = 0;
944 push(@res, grep defined, $self->_expand_expr({ $el, shift(@expr) }));
945 } elsif ($elref eq 'ARRAY') {
946 push(@res, grep defined, $self->_expand_expr($el)) if @$el;
947 } elsif (my $l = is_literal_value($el)) {
948 push @res, { -literal => $l };
949 } elsif ($elref eq 'HASH') {
950 local our $Expand_Depth = 0;
951 push @res, grep defined, $self->_expand_expr($el) if %$el;
957 # return $res[0] if @res == 1;
958 return { -op => [ $logop, @res ] };
964 my ($self, $op, $vv, $k) = @_;
965 puke "$op can only take undef as argument"
969 and exists($vv->{-value})
970 and !defined($vv->{-value})
972 return +{ -op => [ $op.'_null', $self->_expand_ident(-ident => $k) ] };
975 sub _expand_between {
976 my ($self, $op, $vv, $k) = @_;
977 local our $Cur_Col_Meta = $k;
978 my @rhs = map $self->_expand_expr($_),
979 ref($vv) eq 'ARRAY' ? @$vv : $vv;
981 (@rhs == 1 and ref($rhs[0]) eq 'HASH' and $rhs[0]->{-literal})
983 (@rhs == 2 and defined($rhs[0]) and defined($rhs[1]))
985 puke "Operator '${\uc($op)}' requires either an arrayref with two defined values or expressions, or a single literal scalarref/arrayref-ref";
989 $self->_expand_ident(-ident => $k),
995 my ($self, $op, $vv, $k) = @_;
996 if (my $literal = is_literal_value($vv)) {
997 my ($sql, @bind) = @$literal;
998 my $opened_sql = $self->_open_outer_paren($sql);
1000 $op, $self->_expand_ident(-ident => $k),
1001 [ { -literal => [ $opened_sql, @bind ] } ]
1005 'SQL::Abstract before v1.75 used to generate incorrect SQL when the '
1006 . "-${\uc($op)} operator was given an undef-containing list: !!!AUDIT YOUR CODE "
1007 . 'AND DATA!!! (the upcoming Data::Query-based version of SQL::Abstract '
1008 . 'will emit the logically correct SQL instead of raising this exception)'
1010 puke("Argument passed to the '${\uc($op)}' operator can not be undefined")
1012 my @rhs = map $self->_expand_expr($_),
1013 map { ref($_) ? $_ : { -bind => [ $k, $_ ] } }
1014 map { defined($_) ? $_: puke($undef_err) }
1015 (ref($vv) eq 'ARRAY' ? @$vv : $vv);
1016 return $self->${\($op =~ /^not/ ? 'sqltrue' : 'sqlfalse')} unless @rhs;
1020 $self->_expand_ident(-ident => $k),
1026 my ($self, $op, $v) = @_;
1027 # DBIx::Class requires a nest warning to be emitted once but the private
1028 # method it overrode to do so no longer exists
1029 if ($self->{is_dbic_sqlmaker}) {
1030 unless (our $Nest_Warned) {
1032 "-nest in search conditions is deprecated, you most probably wanted:\n"
1033 .q|{..., -and => [ \%cond0, \@cond1, \'cond2', \[ 'cond3', [ col => bind ] ], etc. ], ... }|
1038 return $self->_expand_expr($v);
1041 sub _recurse_where {
1042 my ($self, $where, $logic) = @_;
1044 # Special case: top level simple string treated as literal
1046 my $where_exp = (ref($where)
1047 ? $self->_expand_expr($where, $logic)
1048 : { -literal => [ $where ] });
1050 # dispatch expanded expression
1052 my ($sql, @bind) = defined($where_exp) ? $self->render_aqt($where_exp) : (undef);
1053 # DBIx::Class used to call _recurse_where in scalar context
1054 # something else might too...
1056 return ($sql, @bind);
1059 belch "Calling _recurse_where in scalar context is deprecated and will go away before 2.0";
1065 my ($self, $ident) = @_;
1067 return $self->_convert($self->_quote($ident));
1071 my ($self, $list) = @_;
1072 my @parts = grep length($_->[0]), map [ $self->render_aqt($_) ], @$list;
1073 return join(', ', map $_->[0], @parts), map @{$_}[1..$#$_], @parts;
1077 my ($self, $rest) = @_;
1078 my ($func, @args) = @$rest;
1082 push @arg_sql, shift @x;
1084 } map [ $self->render_aqt($_) ], @args;
1085 return ($self->_sqlcase($func).'('.join(', ', @arg_sql).')', @bind);
1089 my ($self, $bind) = @_;
1090 return ($self->_convert('?'), $self->_bindtype(@$bind));
1093 sub _render_literal {
1094 my ($self, $literal) = @_;
1095 $self->_assert_bindval_matches_bindtype(@{$literal}[1..$#$literal]);
1100 my ($self, $v) = @_;
1101 my ($op, @args) = @$v;
1102 if (my $r = $self->{render_op}{$op}) {
1103 return $self->$r($op, \@args);
1108 my $op = join(' ', split '_', $op);
1110 my $us = List::Util::first { $op =~ $_->{regex} } @{$self->{special_ops}};
1111 if ($us and @args > 1) {
1112 puke "Special op '${op}' requires first value to be identifier"
1113 unless my ($ident) = map $_->{-ident}, grep ref($_) eq 'HASH', $args[0];
1114 my $k = join(($self->{name_sep}||'.'), @$ident);
1115 local our $Expand_Depth = 1;
1116 return $self->${\($us->{handler})}($k, $op, $args[1]);
1118 if (my $us = List::Util::first { $op =~ $_->{regex} } @{$self->{unary_ops}}) {
1119 return $self->${\($us->{handler})}($op, $args[0]);
1124 return $self->_render_unop_prefix($op, \@args);
1126 return $self->_render_op_multop($op, \@args);
1132 sub _render_op_between {
1133 my ($self, $op, $args) = @_;
1134 my ($left, $low, $high) = @$args;
1135 my ($rhsql, @rhbind) = do {
1137 puke "Single arg to between must be a literal"
1138 unless $low->{-literal};
1141 my ($l, $h) = map [ $self->render_aqt($_) ], $low, $high;
1142 (join(' ', $l->[0], $self->_sqlcase('and'), $h->[0]),
1143 @{$l}[1..$#$l], @{$h}[1..$#$h])
1146 my ($lhsql, @lhbind) = $self->render_aqt($left);
1150 $self->_sqlcase(join ' ', split '_', $op),
1158 my ($self, $op, $args) = @_;
1159 my ($lhs, $rhs) = @$args;
1162 my ($sql, @bind) = $self->render_aqt($_);
1163 push @in_bind, @bind;
1166 my ($lhsql, @lbind) = $self->render_aqt($lhs);
1168 $lhsql.' '.$self->_sqlcase(join ' ', split '_', $op).' ( '
1169 .join(', ', @in_sql)
1175 sub _render_op_andor {
1176 my ($self, $op, $args) = @_;
1177 my @parts = grep length($_->[0]), map [ $self->render_aqt($_) ], @$args;
1178 return '' unless @parts;
1179 return @{$parts[0]} if @parts == 1;
1180 my ($sql, @bind) = $self->_render_op_multop($op, $args);
1181 return '( '.$sql.' )', @bind;
1184 sub _render_op_multop {
1185 my ($self, $op, $args) = @_;
1186 my @parts = grep length($_->[0]), map [ $self->render_aqt($_) ], @$args;
1187 return '' unless @parts;
1188 return @{$parts[0]} if @parts == 1;
1189 my ($final_sql) = join(
1190 ' '.$self->_sqlcase(join ' ', split '_', $op).' ',
1195 map @{$_}[1..$#$_], @parts
1198 sub _render_op_not {
1199 my ($self, $op, $v) = @_;
1200 my ($sql, @bind) = $self->_render_unop_prefix($op, $v);
1201 return "(${sql})", @bind;
1204 sub _render_unop_prefix {
1205 my ($self, $op, $v) = @_;
1206 my ($expr_sql, @bind) = $self->render_aqt($v->[0]);
1208 my $op_sql = $self->_sqlcase($op); # join ' ', split '_', $op);
1209 return ("${op_sql} ${expr_sql}", @bind);
1212 sub _render_unop_postfix {
1213 my ($self, $op, $v) = @_;
1214 my ($expr_sql, @bind) = $self->render_aqt($v->[0]);
1215 my $op_sql = $self->_sqlcase(join ' ', split '_', $op);
1216 return ($expr_sql.' '.$op_sql, @bind);
1219 # Some databases (SQLite) treat col IN (1, 2) different from
1220 # col IN ( (1, 2) ). Use this to strip all outer parens while
1221 # adding them back in the corresponding method
1222 sub _open_outer_paren {
1223 my ($self, $sql) = @_;
1225 while (my ($inner) = $sql =~ /^ \s* \( (.*) \) \s* $/xs) {
1227 # there are closing parens inside, need the heavy duty machinery
1228 # to reevaluate the extraction starting from $sql (full reevaluation)
1229 if ($inner =~ /\)/) {
1230 require Text::Balanced;
1232 my (undef, $remainder) = do {
1233 # idiotic design - writes to $@ but *DOES NOT* throw exceptions
1235 Text::Balanced::extract_bracketed($sql, '()', qr/\s*/);
1238 # the entire expression needs to be a balanced bracketed thing
1239 # (after an extract no remainder sans trailing space)
1240 last if defined $remainder and $remainder =~ /\S/;
1250 #======================================================================
1252 #======================================================================
1254 sub _expand_order_by {
1255 my ($self, $arg) = @_;
1257 return unless defined($arg) and not (ref($arg) eq 'ARRAY' and !@$arg);
1259 my $expander = sub {
1260 my ($self, $dir, $expr) = @_;
1261 my @to_expand = ref($expr) eq 'ARRAY' ? @$expr : $expr;
1262 foreach my $arg (@to_expand) {
1266 and grep /^-(asc|desc)$/, keys %$arg
1268 puke "ordering direction hash passed to order by must have exactly one key (-asc or -desc)";
1272 defined($dir) ? { -op => [ $dir =~ /^-?(.*)$/ ,=> $_ ] } : $_
1274 map $self->expand_expr($_, -ident),
1275 map ref($_) eq 'ARRAY' ? @$_ : $_, @to_expand;
1276 return (@exp > 1 ? { -list => \@exp } : $exp[0]);
1279 local @{$self->{expand}}{qw(-asc -desc)} = (($expander) x 2);
1281 return $self->$expander(undef, $arg);
1285 my ($self, $arg) = @_;
1287 return '' unless defined(my $expanded = $self->_expand_order_by($arg));
1289 my ($sql, @bind) = $self->render_aqt($expanded);
1291 return '' unless length($sql);
1293 my $final_sql = $self->_sqlcase(' order by ').$sql;
1295 return wantarray ? ($final_sql, @bind) : $final_sql;
1298 # _order_by no longer needs to call this so doesn't but DBIC uses it.
1300 sub _order_by_chunks {
1301 my ($self, $arg) = @_;
1303 return () unless defined(my $expanded = $self->_expand_order_by($arg));
1305 return $self->_chunkify_order_by($expanded);
1308 sub _chunkify_order_by {
1309 my ($self, $expanded) = @_;
1311 return grep length, $self->render_aqt($expanded)
1312 if $expanded->{-ident} or @{$expanded->{-literal}||[]} == 1;
1315 if (ref() eq 'HASH' and my $l = $_->{-list}) {
1316 return map $self->_chunkify_order_by($_), @$l;
1318 return [ $self->render_aqt($_) ];
1322 #======================================================================
1323 # DATASOURCE (FOR NOW, JUST PLAIN TABLE OR LIST OF TABLES)
1324 #======================================================================
1330 $self->_expand_maybe_list_expr($from, undef, -ident)
1335 #======================================================================
1337 #======================================================================
1339 sub _expand_maybe_list_expr {
1340 my ($self, $expr, $logic, $default) = @_;
1342 if (ref($expr) eq 'ARRAY') {
1344 map $self->expand_expr($_, $default), @$expr
1351 return $self->expand_expr($e, $default);
1354 # highly optimized, as it's called way too often
1356 # my ($self, $label) = @_;
1358 return '' unless defined $_[1];
1359 return ${$_[1]} if ref($_[1]) eq 'SCALAR';
1360 puke 'Identifier cannot be hashref' if ref($_[1]) eq 'HASH';
1362 unless ($_[0]->{quote_char}) {
1363 if (ref($_[1]) eq 'ARRAY') {
1364 return join($_[0]->{name_sep}||'.', @{$_[1]});
1366 $_[0]->_assert_pass_injection_guard($_[1]);
1371 my $qref = ref $_[0]->{quote_char};
1373 !$qref ? ($_[0]->{quote_char}, $_[0]->{quote_char})
1374 : ($qref eq 'ARRAY') ? @{$_[0]->{quote_char}}
1375 : puke "Unsupported quote_char format: $_[0]->{quote_char}";
1377 my $esc = $_[0]->{escape_char} || $r;
1379 # parts containing * are naturally unquoted
1381 $_[0]->{name_sep}||'',
1385 : do { (my $n = $_) =~ s/(\Q$esc\E|\Q$r\E)/$esc$1/g; $l . $n . $r }
1387 (ref($_[1]) eq 'ARRAY'
1391 ? split (/\Q$_[0]->{name_sep}\E/, $_[1] )
1399 # Conversion, if applicable
1401 #my ($self, $arg) = @_;
1402 if ($_[0]->{convert_where}) {
1403 return $_[0]->_sqlcase($_[0]->{convert_where}) .'(' . $_[1] . ')';
1410 #my ($self, $col, @vals) = @_;
1411 # called often - tighten code
1412 return $_[0]->{bindtype} eq 'columns'
1413 ? map {[$_[1], $_]} @_[2 .. $#_]
1418 # Dies if any element of @bind is not in [colname => value] format
1419 # if bindtype is 'columns'.
1420 sub _assert_bindval_matches_bindtype {
1421 # my ($self, @bind) = @_;
1423 if ($self->{bindtype} eq 'columns') {
1425 if (!defined $_ || ref($_) ne 'ARRAY' || @$_ != 2) {
1426 puke "bindtype 'columns' selected, you need to pass: [column_name => bind_value]"
1432 sub _join_sql_clauses {
1433 my ($self, $logic, $clauses_aref, $bind_aref) = @_;
1435 if (@$clauses_aref > 1) {
1436 my $join = " " . $self->_sqlcase($logic) . " ";
1437 my $sql = '( ' . join($join, @$clauses_aref) . ' )';
1438 return ($sql, @$bind_aref);
1440 elsif (@$clauses_aref) {
1441 return ($clauses_aref->[0], @$bind_aref); # no parentheses
1444 return (); # if no SQL, ignore @$bind_aref
1449 # Fix SQL case, if so requested
1451 # LDNOTE: if $self->{case} is true, then it contains 'lower', so we
1452 # don't touch the argument ... crooked logic, but let's not change it!
1453 return $_[0]->{case} ? $_[1] : uc($_[1]);
1457 #======================================================================
1458 # DISPATCHING FROM REFKIND
1459 #======================================================================
1462 my ($self, $data) = @_;
1464 return 'UNDEF' unless defined $data;
1466 # blessed objects are treated like scalars
1467 my $ref = (Scalar::Util::blessed $data) ? '' : ref $data;
1469 return 'SCALAR' unless $ref;
1472 while ($ref eq 'REF') {
1474 $ref = (Scalar::Util::blessed $data) ? '' : ref $data;
1478 return ($ref||'SCALAR') . ('REF' x $n_steps);
1482 my ($self, $data) = @_;
1483 my @try = ($self->_refkind($data));
1484 push @try, 'SCALAR_or_UNDEF' if $try[0] eq 'SCALAR' || $try[0] eq 'UNDEF';
1485 push @try, 'FALLBACK';
1489 sub _METHOD_FOR_refkind {
1490 my ($self, $meth_prefix, $data) = @_;
1493 for (@{$self->_try_refkind($data)}) {
1494 $method = $self->can($meth_prefix."_".$_)
1498 return $method || puke "cannot dispatch on '$meth_prefix' for ".$self->_refkind($data);
1502 sub _SWITCH_refkind {
1503 my ($self, $data, $dispatch_table) = @_;
1506 for (@{$self->_try_refkind($data)}) {
1507 $coderef = $dispatch_table->{$_}
1511 puke "no dispatch entry for ".$self->_refkind($data)
1520 #======================================================================
1521 # VALUES, GENERATE, AUTOLOAD
1522 #======================================================================
1524 # LDNOTE: original code from nwiger, didn't touch code in that section
1525 # I feel the AUTOLOAD stuff should not be the default, it should
1526 # only be activated on explicit demand by user.
1530 my $data = shift || return;
1531 puke "Argument to ", __PACKAGE__, "->values must be a \\%hash"
1532 unless ref $data eq 'HASH';
1535 foreach my $k (sort keys %$data) {
1536 my $v = $data->{$k};
1537 $self->_SWITCH_refkind($v, {
1539 if ($self->{array_datatypes}) { # array datatype
1540 push @all_bind, $self->_bindtype($k, $v);
1542 else { # literal SQL with bind
1543 my ($sql, @bind) = @$v;
1544 $self->_assert_bindval_matches_bindtype(@bind);
1545 push @all_bind, @bind;
1548 ARRAYREFREF => sub { # literal SQL with bind
1549 my ($sql, @bind) = @${$v};
1550 $self->_assert_bindval_matches_bindtype(@bind);
1551 push @all_bind, @bind;
1553 SCALARREF => sub { # literal SQL without bind
1555 SCALAR_or_UNDEF => sub {
1556 push @all_bind, $self->_bindtype($k, $v);
1567 my(@sql, @sqlq, @sqlv);
1571 if ($ref eq 'HASH') {
1572 for my $k (sort keys %$_) {
1575 my $label = $self->_quote($k);
1576 if ($r eq 'ARRAY') {
1577 # literal SQL with bind
1578 my ($sql, @bind) = @$v;
1579 $self->_assert_bindval_matches_bindtype(@bind);
1580 push @sqlq, "$label = $sql";
1582 } elsif ($r eq 'SCALAR') {
1583 # literal SQL without bind
1584 push @sqlq, "$label = $$v";
1586 push @sqlq, "$label = ?";
1587 push @sqlv, $self->_bindtype($k, $v);
1590 push @sql, $self->_sqlcase('set'), join ', ', @sqlq;
1591 } elsif ($ref eq 'ARRAY') {
1592 # unlike insert(), assume these are ONLY the column names, i.e. for SQL
1595 if ($r eq 'ARRAY') { # literal SQL with bind
1596 my ($sql, @bind) = @$v;
1597 $self->_assert_bindval_matches_bindtype(@bind);
1600 } elsif ($r eq 'SCALAR') { # literal SQL without bind
1601 # embedded literal SQL
1608 push @sql, '(' . join(', ', @sqlq) . ')';
1609 } elsif ($ref eq 'SCALAR') {
1613 # strings get case twiddled
1614 push @sql, $self->_sqlcase($_);
1618 my $sql = join ' ', @sql;
1620 # this is pretty tricky
1621 # if ask for an array, return ($stmt, @bind)
1622 # otherwise, s/?/shift @sqlv/ to put it inline
1624 return ($sql, @sqlv);
1626 1 while $sql =~ s/\?/my $d = shift(@sqlv);
1627 ref $d ? $d->[1] : $d/e;
1636 # This allows us to check for a local, then _form, attr
1638 my($name) = $AUTOLOAD =~ /.*::(.+)/;
1639 return $self->generate($name, @_);
1650 SQL::Abstract - Generate SQL from Perl data structures
1656 my $sql = SQL::Abstract->new;
1658 my($stmt, @bind) = $sql->select($source, \@fields, \%where, $order);
1660 my($stmt, @bind) = $sql->insert($table, \%fieldvals || \@values);
1662 my($stmt, @bind) = $sql->update($table, \%fieldvals, \%where);
1664 my($stmt, @bind) = $sql->delete($table, \%where);
1666 # Then, use these in your DBI statements
1667 my $sth = $dbh->prepare($stmt);
1668 $sth->execute(@bind);
1670 # Just generate the WHERE clause
1671 my($stmt, @bind) = $sql->where(\%where, $order);
1673 # Return values in the same order, for hashed queries
1674 # See PERFORMANCE section for more details
1675 my @bind = $sql->values(\%fieldvals);
1679 This module was inspired by the excellent L<DBIx::Abstract>.
1680 However, in using that module I found that what I really wanted
1681 to do was generate SQL, but still retain complete control over my
1682 statement handles and use the DBI interface. So, I set out to
1683 create an abstract SQL generation module.
1685 While based on the concepts used by L<DBIx::Abstract>, there are
1686 several important differences, especially when it comes to WHERE
1687 clauses. I have modified the concepts used to make the SQL easier
1688 to generate from Perl data structures and, IMO, more intuitive.
1689 The underlying idea is for this module to do what you mean, based
1690 on the data structures you provide it. The big advantage is that
1691 you don't have to modify your code every time your data changes,
1692 as this module figures it out.
1694 To begin with, an SQL INSERT is as easy as just specifying a hash
1695 of C<key=value> pairs:
1698 name => 'Jimbo Bobson',
1699 phone => '123-456-7890',
1700 address => '42 Sister Lane',
1701 city => 'St. Louis',
1702 state => 'Louisiana',
1705 The SQL can then be generated with this:
1707 my($stmt, @bind) = $sql->insert('people', \%data);
1709 Which would give you something like this:
1711 $stmt = "INSERT INTO people
1712 (address, city, name, phone, state)
1713 VALUES (?, ?, ?, ?, ?)";
1714 @bind = ('42 Sister Lane', 'St. Louis', 'Jimbo Bobson',
1715 '123-456-7890', 'Louisiana');
1717 These are then used directly in your DBI code:
1719 my $sth = $dbh->prepare($stmt);
1720 $sth->execute(@bind);
1722 =head2 Inserting and Updating Arrays
1724 If your database has array types (like for example Postgres),
1725 activate the special option C<< array_datatypes => 1 >>
1726 when creating the C<SQL::Abstract> object.
1727 Then you may use an arrayref to insert and update database array types:
1729 my $sql = SQL::Abstract->new(array_datatypes => 1);
1731 planets => [qw/Mercury Venus Earth Mars/]
1734 my($stmt, @bind) = $sql->insert('solar_system', \%data);
1738 $stmt = "INSERT INTO solar_system (planets) VALUES (?)"
1740 @bind = (['Mercury', 'Venus', 'Earth', 'Mars']);
1743 =head2 Inserting and Updating SQL
1745 In order to apply SQL functions to elements of your C<%data> you may
1746 specify a reference to an arrayref for the given hash value. For example,
1747 if you need to execute the Oracle C<to_date> function on a value, you can
1748 say something like this:
1752 date_entered => \[ "to_date(?,'MM/DD/YYYY')", "03/02/2003" ],
1755 The first value in the array is the actual SQL. Any other values are
1756 optional and would be included in the bind values array. This gives
1759 my($stmt, @bind) = $sql->insert('people', \%data);
1761 $stmt = "INSERT INTO people (name, date_entered)
1762 VALUES (?, to_date(?,'MM/DD/YYYY'))";
1763 @bind = ('Bill', '03/02/2003');
1765 An UPDATE is just as easy, all you change is the name of the function:
1767 my($stmt, @bind) = $sql->update('people', \%data);
1769 Notice that your C<%data> isn't touched; the module will generate
1770 the appropriately quirky SQL for you automatically. Usually you'll
1771 want to specify a WHERE clause for your UPDATE, though, which is
1772 where handling C<%where> hashes comes in handy...
1774 =head2 Complex where statements
1776 This module can generate pretty complicated WHERE statements
1777 easily. For example, simple C<key=value> pairs are taken to mean
1778 equality, and if you want to see if a field is within a set
1779 of values, you can use an arrayref. Let's say we wanted to
1780 SELECT some data based on this criteria:
1783 requestor => 'inna',
1784 worker => ['nwiger', 'rcwe', 'sfz'],
1785 status => { '!=', 'completed' }
1788 my($stmt, @bind) = $sql->select('tickets', '*', \%where);
1790 The above would give you something like this:
1792 $stmt = "SELECT * FROM tickets WHERE
1793 ( requestor = ? ) AND ( status != ? )
1794 AND ( worker = ? OR worker = ? OR worker = ? )";
1795 @bind = ('inna', 'completed', 'nwiger', 'rcwe', 'sfz');
1797 Which you could then use in DBI code like so:
1799 my $sth = $dbh->prepare($stmt);
1800 $sth->execute(@bind);
1806 The methods are simple. There's one for every major SQL operation,
1807 and a constructor you use first. The arguments are specified in a
1808 similar order for each method (table, then fields, then a where
1809 clause) to try and simplify things.
1811 =head2 new(option => 'value')
1813 The C<new()> function takes a list of options and values, and returns
1814 a new B<SQL::Abstract> object which can then be used to generate SQL
1815 through the methods below. The options accepted are:
1821 If set to 'lower', then SQL will be generated in all lowercase. By
1822 default SQL is generated in "textbook" case meaning something like:
1824 SELECT a_field FROM a_table WHERE some_field LIKE '%someval%'
1826 Any setting other than 'lower' is ignored.
1830 This determines what the default comparison operator is. By default
1831 it is C<=>, meaning that a hash like this:
1833 %where = (name => 'nwiger', email => 'nate@wiger.org');
1835 Will generate SQL like this:
1837 WHERE name = 'nwiger' AND email = 'nate@wiger.org'
1839 However, you may want loose comparisons by default, so if you set
1840 C<cmp> to C<like> you would get SQL such as:
1842 WHERE name like 'nwiger' AND email like 'nate@wiger.org'
1844 You can also override the comparison on an individual basis - see
1845 the huge section on L</"WHERE CLAUSES"> at the bottom.
1847 =item sqltrue, sqlfalse
1849 Expressions for inserting boolean values within SQL statements.
1850 By default these are C<1=1> and C<1=0>. They are used
1851 by the special operators C<-in> and C<-not_in> for generating
1852 correct SQL even when the argument is an empty array (see below).
1856 This determines the default logical operator for multiple WHERE
1857 statements in arrays or hashes. If absent, the default logic is "or"
1858 for arrays, and "and" for hashes. This means that a WHERE
1862 event_date => {'>=', '2/13/99'},
1863 event_date => {'<=', '4/24/03'},
1866 will generate SQL like this:
1868 WHERE event_date >= '2/13/99' OR event_date <= '4/24/03'
1870 This is probably not what you want given this query, though (look
1871 at the dates). To change the "OR" to an "AND", simply specify:
1873 my $sql = SQL::Abstract->new(logic => 'and');
1875 Which will change the above C<WHERE> to:
1877 WHERE event_date >= '2/13/99' AND event_date <= '4/24/03'
1879 The logic can also be changed locally by inserting
1880 a modifier in front of an arrayref:
1882 @where = (-and => [event_date => {'>=', '2/13/99'},
1883 event_date => {'<=', '4/24/03'} ]);
1885 See the L</"WHERE CLAUSES"> section for explanations.
1889 This will automatically convert comparisons using the specified SQL
1890 function for both column and value. This is mostly used with an argument
1891 of C<upper> or C<lower>, so that the SQL will have the effect of
1892 case-insensitive "searches". For example, this:
1894 $sql = SQL::Abstract->new(convert => 'upper');
1895 %where = (keywords => 'MaKe iT CAse inSeNSItive');
1897 Will turn out the following SQL:
1899 WHERE upper(keywords) like upper('MaKe iT CAse inSeNSItive')
1901 The conversion can be C<upper()>, C<lower()>, or any other SQL function
1902 that can be applied symmetrically to fields (actually B<SQL::Abstract> does
1903 not validate this option; it will just pass through what you specify verbatim).
1907 This is a kludge because many databases suck. For example, you can't
1908 just bind values using DBI's C<execute()> for Oracle C<CLOB> or C<BLOB> fields.
1909 Instead, you have to use C<bind_param()>:
1911 $sth->bind_param(1, 'reg data');
1912 $sth->bind_param(2, $lots, {ora_type => ORA_CLOB});
1914 The problem is, B<SQL::Abstract> will normally just return a C<@bind> array,
1915 which loses track of which field each slot refers to. Fear not.
1917 If you specify C<bindtype> in new, you can determine how C<@bind> is returned.
1918 Currently, you can specify either C<normal> (default) or C<columns>. If you
1919 specify C<columns>, you will get an array that looks like this:
1921 my $sql = SQL::Abstract->new(bindtype => 'columns');
1922 my($stmt, @bind) = $sql->insert(...);
1925 [ 'column1', 'value1' ],
1926 [ 'column2', 'value2' ],
1927 [ 'column3', 'value3' ],
1930 You can then iterate through this manually, using DBI's C<bind_param()>.
1932 $sth->prepare($stmt);
1935 my($col, $data) = @$_;
1936 if ($col eq 'details' || $col eq 'comments') {
1937 $sth->bind_param($i, $data, {ora_type => ORA_CLOB});
1938 } elsif ($col eq 'image') {
1939 $sth->bind_param($i, $data, {ora_type => ORA_BLOB});
1941 $sth->bind_param($i, $data);
1945 $sth->execute; # execute without @bind now
1947 Now, why would you still use B<SQL::Abstract> if you have to do this crap?
1948 Basically, the advantage is still that you don't have to care which fields
1949 are or are not included. You could wrap that above C<for> loop in a simple
1950 sub called C<bind_fields()> or something and reuse it repeatedly. You still
1951 get a layer of abstraction over manual SQL specification.
1953 Note that if you set L</bindtype> to C<columns>, the C<\[ $sql, @bind ]>
1954 construct (see L</Literal SQL with placeholders and bind values (subqueries)>)
1955 will expect the bind values in this format.
1959 This is the character that a table or column name will be quoted
1960 with. By default this is an empty string, but you could set it to
1961 the character C<`>, to generate SQL like this:
1963 SELECT `a_field` FROM `a_table` WHERE `some_field` LIKE '%someval%'
1965 Alternatively, you can supply an array ref of two items, the first being the left
1966 hand quote character, and the second the right hand quote character. For
1967 example, you could supply C<['[',']']> for SQL Server 2000 compliant quotes
1968 that generates SQL like this:
1970 SELECT [a_field] FROM [a_table] WHERE [some_field] LIKE '%someval%'
1972 Quoting is useful if you have tables or columns names that are reserved
1973 words in your database's SQL dialect.
1977 This is the character that will be used to escape L</quote_char>s appearing
1978 in an identifier before it has been quoted.
1980 The parameter default in case of a single L</quote_char> character is the quote
1983 When opening-closing-style quoting is used (L</quote_char> is an arrayref)
1984 this parameter defaults to the B<closing (right)> L</quote_char>. Occurrences
1985 of the B<opening (left)> L</quote_char> within the identifier are currently left
1986 untouched. The default for opening-closing-style quotes may change in future
1987 versions, thus you are B<strongly encouraged> to specify the escape character
1992 This is the character that separates a table and column name. It is
1993 necessary to specify this when the C<quote_char> option is selected,
1994 so that tables and column names can be individually quoted like this:
1996 SELECT `table`.`one_field` FROM `table` WHERE `table`.`other_field` = 1
1998 =item injection_guard
2000 A regular expression C<qr/.../> that is applied to any C<-function> and unquoted
2001 column name specified in a query structure. This is a safety mechanism to avoid
2002 injection attacks when mishandling user input e.g.:
2004 my %condition_as_column_value_pairs = get_values_from_user();
2005 $sqla->select( ... , \%condition_as_column_value_pairs );
2007 If the expression matches an exception is thrown. Note that literal SQL
2008 supplied via C<\'...'> or C<\['...']> is B<not> checked in any way.
2010 Defaults to checking for C<;> and the C<GO> keyword (TransactSQL)
2012 =item array_datatypes
2014 When this option is true, arrayrefs in INSERT or UPDATE are
2015 interpreted as array datatypes and are passed directly
2017 When this option is false, arrayrefs are interpreted
2018 as literal SQL, just like refs to arrayrefs
2019 (but this behavior is for backwards compatibility; when writing
2020 new queries, use the "reference to arrayref" syntax
2026 Takes a reference to a list of "special operators"
2027 to extend the syntax understood by L<SQL::Abstract>.
2028 See section L</"SPECIAL OPERATORS"> for details.
2032 Takes a reference to a list of "unary operators"
2033 to extend the syntax understood by L<SQL::Abstract>.
2034 See section L</"UNARY OPERATORS"> for details.
2040 =head2 insert($table, \@values || \%fieldvals, \%options)
2042 This is the simplest function. You simply give it a table name
2043 and either an arrayref of values or hashref of field/value pairs.
2044 It returns an SQL INSERT statement and a list of bind values.
2045 See the sections on L</"Inserting and Updating Arrays"> and
2046 L</"Inserting and Updating SQL"> for information on how to insert
2047 with those data types.
2049 The optional C<\%options> hash reference may contain additional
2050 options to generate the insert SQL. Currently supported options
2057 Takes either a scalar of raw SQL fields, or an array reference of
2058 field names, and adds on an SQL C<RETURNING> statement at the end.
2059 This allows you to return data generated by the insert statement
2060 (such as row IDs) without performing another C<SELECT> statement.
2061 Note, however, this is not part of the SQL standard and may not
2062 be supported by all database engines.
2066 =head2 update($table, \%fieldvals, \%where, \%options)
2068 This takes a table, hashref of field/value pairs, and an optional
2069 hashref L<WHERE clause|/WHERE CLAUSES>. It returns an SQL UPDATE function and a list
2071 See the sections on L</"Inserting and Updating Arrays"> and
2072 L</"Inserting and Updating SQL"> for information on how to insert
2073 with those data types.
2075 The optional C<\%options> hash reference may contain additional
2076 options to generate the update SQL. Currently supported options
2083 See the C<returning> option to
2084 L<insert|/insert($table, \@values || \%fieldvals, \%options)>.
2088 =head2 select($source, $fields, $where, $order)
2090 This returns a SQL SELECT statement and associated list of bind values, as
2091 specified by the arguments:
2097 Specification of the 'FROM' part of the statement.
2098 The argument can be either a plain scalar (interpreted as a table
2099 name, will be quoted), or an arrayref (interpreted as a list
2100 of table names, joined by commas, quoted), or a scalarref
2101 (literal SQL, not quoted).
2105 Specification of the list of fields to retrieve from
2107 The argument can be either an arrayref (interpreted as a list
2108 of field names, will be joined by commas and quoted), or a
2109 plain scalar (literal SQL, not quoted).
2110 Please observe that this API is not as flexible as that of
2111 the first argument C<$source>, for backwards compatibility reasons.
2115 Optional argument to specify the WHERE part of the query.
2116 The argument is most often a hashref, but can also be
2117 an arrayref or plain scalar --
2118 see section L<WHERE clause|/"WHERE CLAUSES"> for details.
2122 Optional argument to specify the ORDER BY part of the query.
2123 The argument can be a scalar, a hashref or an arrayref
2124 -- see section L<ORDER BY clause|/"ORDER BY CLAUSES">
2130 =head2 delete($table, \%where, \%options)
2132 This takes a table name and optional hashref L<WHERE clause|/WHERE CLAUSES>.
2133 It returns an SQL DELETE statement and list of bind values.
2135 The optional C<\%options> hash reference may contain additional
2136 options to generate the delete SQL. Currently supported options
2143 See the C<returning> option to
2144 L<insert|/insert($table, \@values || \%fieldvals, \%options)>.
2148 =head2 where(\%where, $order)
2150 This is used to generate just the WHERE clause. For example,
2151 if you have an arbitrary data structure and know what the
2152 rest of your SQL is going to look like, but want an easy way
2153 to produce a WHERE clause, use this. It returns an SQL WHERE
2154 clause and list of bind values.
2157 =head2 values(\%data)
2159 This just returns the values from the hash C<%data>, in the same
2160 order that would be returned from any of the other above queries.
2161 Using this allows you to markedly speed up your queries if you
2162 are affecting lots of rows. See below under the L</"PERFORMANCE"> section.
2164 =head2 generate($any, 'number', $of, \@data, $struct, \%types)
2166 Warning: This is an experimental method and subject to change.
2168 This returns arbitrarily generated SQL. It's a really basic shortcut.
2169 It will return two different things, depending on return context:
2171 my($stmt, @bind) = $sql->generate('create table', \$table, \@fields);
2172 my $stmt_and_val = $sql->generate('create table', \$table, \@fields);
2174 These would return the following:
2176 # First calling form
2177 $stmt = "CREATE TABLE test (?, ?)";
2178 @bind = (field1, field2);
2180 # Second calling form
2181 $stmt_and_val = "CREATE TABLE test (field1, field2)";
2183 Depending on what you're trying to do, it's up to you to choose the correct
2184 format. In this example, the second form is what you would want.
2188 $sql->generate('alter session', { nls_date_format => 'MM/YY' });
2192 ALTER SESSION SET nls_date_format = 'MM/YY'
2194 You get the idea. Strings get their case twiddled, but everything
2195 else remains verbatim.
2197 =head1 EXPORTABLE FUNCTIONS
2199 =head2 is_plain_value
2201 Determines if the supplied argument is a plain value as understood by this
2206 =item * The value is C<undef>
2208 =item * The value is a non-reference
2210 =item * The value is an object with stringification overloading
2212 =item * The value is of the form C<< { -value => $anything } >>
2216 On failure returns C<undef>, on success returns a B<scalar> reference
2217 to the original supplied argument.
2223 The stringification overloading detection is rather advanced: it takes
2224 into consideration not only the presence of a C<""> overload, but if that
2225 fails also checks for enabled
2226 L<autogenerated versions of C<"">|overload/Magic Autogeneration>, based
2227 on either C<0+> or C<bool>.
2229 Unfortunately testing in the field indicates that this
2230 detection B<< may tickle a latent bug in perl versions before 5.018 >>,
2231 but only when very large numbers of stringifying objects are involved.
2232 At the time of writing ( Sep 2014 ) there is no clear explanation of
2233 the direct cause, nor is there a manageably small test case that reliably
2234 reproduces the problem.
2236 If you encounter any of the following exceptions in B<random places within
2237 your application stack> - this module may be to blame:
2239 Operation "ne": no method found,
2240 left argument in overloaded package <something>,
2241 right argument in overloaded package <something>
2245 Stub found while resolving method "???" overloading """" in package <something>
2247 If you fall victim to the above - please attempt to reduce the problem
2248 to something that could be sent to the L<SQL::Abstract developers
2249 |DBIx::Class/GETTING HELP/SUPPORT>
2250 (either publicly or privately). As a workaround in the meantime you can
2251 set C<$ENV{SQLA_ISVALUE_IGNORE_AUTOGENERATED_STRINGIFICATION}> to a true
2252 value, which will most likely eliminate your problem (at the expense of
2253 not being able to properly detect exotic forms of stringification).
2255 This notice and environment variable will be removed in a future version,
2256 as soon as the underlying problem is found and a reliable workaround is
2261 =head2 is_literal_value
2263 Determines if the supplied argument is a literal value as understood by this
2268 =item * C<\$sql_string>
2270 =item * C<\[ $sql_string, @bind_values ]>
2274 On failure returns C<undef>, on success returns an B<array> reference
2275 containing the unpacked version of the supplied literal SQL and bind values.
2277 =head1 WHERE CLAUSES
2281 This module uses a variation on the idea from L<DBIx::Abstract>. It
2282 is B<NOT>, repeat I<not> 100% compatible. B<The main logic of this
2283 module is that things in arrays are OR'ed, and things in hashes
2286 The easiest way to explain is to show lots of examples. After
2287 each C<%where> hash shown, it is assumed you used:
2289 my($stmt, @bind) = $sql->where(\%where);
2291 However, note that the C<%where> hash can be used directly in any
2292 of the other functions as well, as described above.
2294 =head2 Key-value pairs
2296 So, let's get started. To begin, a simple hash:
2300 status => 'completed'
2303 Is converted to SQL C<key = val> statements:
2305 $stmt = "WHERE user = ? AND status = ?";
2306 @bind = ('nwiger', 'completed');
2308 One common thing I end up doing is having a list of values that
2309 a field can be in. To do this, simply specify a list inside of
2314 status => ['assigned', 'in-progress', 'pending'];
2317 This simple code will create the following:
2319 $stmt = "WHERE user = ? AND ( status = ? OR status = ? OR status = ? )";
2320 @bind = ('nwiger', 'assigned', 'in-progress', 'pending');
2322 A field associated to an empty arrayref will be considered a
2323 logical false and will generate 0=1.
2325 =head2 Tests for NULL values
2327 If the value part is C<undef> then this is converted to SQL <IS NULL>
2336 $stmt = "WHERE user = ? AND status IS NULL";
2339 To test if a column IS NOT NULL:
2343 status => { '!=', undef },
2346 =head2 Specific comparison operators
2348 If you want to specify a different type of operator for your comparison,
2349 you can use a hashref for a given column:
2353 status => { '!=', 'completed' }
2356 Which would generate:
2358 $stmt = "WHERE user = ? AND status != ?";
2359 @bind = ('nwiger', 'completed');
2361 To test against multiple values, just enclose the values in an arrayref:
2363 status => { '=', ['assigned', 'in-progress', 'pending'] };
2365 Which would give you:
2367 "WHERE status = ? OR status = ? OR status = ?"
2370 The hashref can also contain multiple pairs, in which case it is expanded
2371 into an C<AND> of its elements:
2375 status => { '!=', 'completed', -not_like => 'pending%' }
2378 # Or more dynamically, like from a form
2379 $where{user} = 'nwiger';
2380 $where{status}{'!='} = 'completed';
2381 $where{status}{'-not_like'} = 'pending%';
2383 # Both generate this
2384 $stmt = "WHERE user = ? AND status != ? AND status NOT LIKE ?";
2385 @bind = ('nwiger', 'completed', 'pending%');
2388 To get an OR instead, you can combine it with the arrayref idea:
2392 priority => [ { '=', 2 }, { '>', 5 } ]
2395 Which would generate:
2397 $stmt = "WHERE ( priority = ? OR priority > ? ) AND user = ?";
2398 @bind = ('2', '5', 'nwiger');
2400 If you want to include literal SQL (with or without bind values), just use a
2401 scalar reference or reference to an arrayref as the value:
2404 date_entered => { '>' => \["to_date(?, 'MM/DD/YYYY')", "11/26/2008"] },
2405 date_expires => { '<' => \"now()" }
2408 Which would generate:
2410 $stmt = "WHERE date_entered > to_date(?, 'MM/DD/YYYY') AND date_expires < now()";
2411 @bind = ('11/26/2008');
2414 =head2 Logic and nesting operators
2416 In the example above,
2417 there is a subtle trap if you want to say something like
2418 this (notice the C<AND>):
2420 WHERE priority != ? AND priority != ?
2422 Because, in Perl you I<can't> do this:
2424 priority => { '!=' => 2, '!=' => 1 }
2426 As the second C<!=> key will obliterate the first. The solution
2427 is to use the special C<-modifier> form inside an arrayref:
2429 priority => [ -and => {'!=', 2},
2433 Normally, these would be joined by C<OR>, but the modifier tells it
2434 to use C<AND> instead. (Hint: You can use this in conjunction with the
2435 C<logic> option to C<new()> in order to change the way your queries
2436 work by default.) B<Important:> Note that the C<-modifier> goes
2437 B<INSIDE> the arrayref, as an extra first element. This will
2438 B<NOT> do what you think it might:
2440 priority => -and => [{'!=', 2}, {'!=', 1}] # WRONG!
2442 Here is a quick list of equivalencies, since there is some overlap:
2445 status => {'!=', 'completed', 'not like', 'pending%' }
2446 status => [ -and => {'!=', 'completed'}, {'not like', 'pending%'}]
2449 status => {'=', ['assigned', 'in-progress']}
2450 status => [ -or => {'=', 'assigned'}, {'=', 'in-progress'}]
2451 status => [ {'=', 'assigned'}, {'=', 'in-progress'} ]
2455 =head2 Special operators: IN, BETWEEN, etc.
2457 You can also use the hashref format to compare a list of fields using the
2458 C<IN> comparison operator, by specifying the list as an arrayref:
2461 status => 'completed',
2462 reportid => { -in => [567, 2335, 2] }
2465 Which would generate:
2467 $stmt = "WHERE status = ? AND reportid IN (?,?,?)";
2468 @bind = ('completed', '567', '2335', '2');
2470 The reverse operator C<-not_in> generates SQL C<NOT IN> and is used in
2473 If the argument to C<-in> is an empty array, 'sqlfalse' is generated
2474 (by default: C<1=0>). Similarly, C<< -not_in => [] >> generates
2475 'sqltrue' (by default: C<1=1>).
2477 In addition to the array you can supply a chunk of literal sql or
2478 literal sql with bind:
2481 customer => { -in => \[
2482 'SELECT cust_id FROM cust WHERE balance > ?',
2485 status => { -in => \'SELECT status_codes FROM states' },
2491 customer IN ( SELECT cust_id FROM cust WHERE balance > ? )
2492 AND status IN ( SELECT status_codes FROM states )
2496 Finally, if the argument to C<-in> is not a reference, it will be
2497 treated as a single-element array.
2499 Another pair of operators is C<-between> and C<-not_between>,
2500 used with an arrayref of two values:
2504 completion_date => {
2505 -not_between => ['2002-10-01', '2003-02-06']
2511 WHERE user = ? AND completion_date NOT BETWEEN ( ? AND ? )
2513 Just like with C<-in> all plausible combinations of literal SQL
2517 start0 => { -between => [ 1, 2 ] },
2518 start1 => { -between => \["? AND ?", 1, 2] },
2519 start2 => { -between => \"lower(x) AND upper(y)" },
2520 start3 => { -between => [
2522 \["upper(?)", 'stuff' ],
2529 ( start0 BETWEEN ? AND ? )
2530 AND ( start1 BETWEEN ? AND ? )
2531 AND ( start2 BETWEEN lower(x) AND upper(y) )
2532 AND ( start3 BETWEEN lower(x) AND upper(?) )
2534 @bind = (1, 2, 1, 2, 'stuff');
2537 These are the two builtin "special operators"; but the
2538 list can be expanded: see section L</"SPECIAL OPERATORS"> below.
2540 =head2 Unary operators: bool
2542 If you wish to test against boolean columns or functions within your
2543 database you can use the C<-bool> and C<-not_bool> operators. For
2544 example to test the column C<is_user> being true and the column
2545 C<is_enabled> being false you would use:-
2549 -not_bool => 'is_enabled',
2554 WHERE is_user AND NOT is_enabled
2556 If a more complex combination is required, testing more conditions,
2557 then you should use the and/or operators:-
2562 -not_bool => { two=> { -rlike => 'bar' } },
2563 -not_bool => { three => [ { '=', 2 }, { '>', 5 } ] },
2574 (NOT ( three = ? OR three > ? ))
2577 =head2 Nested conditions, -and/-or prefixes
2579 So far, we've seen how multiple conditions are joined with a top-level
2580 C<AND>. We can change this by putting the different conditions we want in
2581 hashes and then putting those hashes in an array. For example:
2586 status => { -like => ['pending%', 'dispatched'] },
2590 status => 'unassigned',
2594 This data structure would create the following:
2596 $stmt = "WHERE ( user = ? AND ( status LIKE ? OR status LIKE ? ) )
2597 OR ( user = ? AND status = ? ) )";
2598 @bind = ('nwiger', 'pending', 'dispatched', 'robot', 'unassigned');
2601 Clauses in hashrefs or arrayrefs can be prefixed with an C<-and> or C<-or>
2602 to change the logic inside:
2608 -and => [ workhrs => {'>', 20}, geo => 'ASIA' ],
2609 -or => { workhrs => {'<', 50}, geo => 'EURO' },
2616 $stmt = "WHERE ( user = ?
2617 AND ( ( workhrs > ? AND geo = ? )
2618 OR ( workhrs < ? OR geo = ? ) ) )";
2619 @bind = ('nwiger', '20', 'ASIA', '50', 'EURO');
2621 =head3 Algebraic inconsistency, for historical reasons
2623 C<Important note>: when connecting several conditions, the C<-and->|C<-or>
2624 operator goes C<outside> of the nested structure; whereas when connecting
2625 several constraints on one column, the C<-and> operator goes
2626 C<inside> the arrayref. Here is an example combining both features:
2629 -and => [a => 1, b => 2],
2630 -or => [c => 3, d => 4],
2631 e => [-and => {-like => 'foo%'}, {-like => '%bar'} ]
2636 WHERE ( ( ( a = ? AND b = ? )
2637 OR ( c = ? OR d = ? )
2638 OR ( e LIKE ? AND e LIKE ? ) ) )
2640 This difference in syntax is unfortunate but must be preserved for
2641 historical reasons. So be careful: the two examples below would
2642 seem algebraically equivalent, but they are not
2645 { -like => 'foo%' },
2646 { -like => '%bar' },
2648 # yields: WHERE ( ( col LIKE ? AND col LIKE ? ) )
2651 { col => { -like => 'foo%' } },
2652 { col => { -like => '%bar' } },
2654 # yields: WHERE ( ( col LIKE ? OR col LIKE ? ) )
2657 =head2 Literal SQL and value type operators
2659 The basic premise of SQL::Abstract is that in WHERE specifications the "left
2660 side" is a column name and the "right side" is a value (normally rendered as
2661 a placeholder). This holds true for both hashrefs and arrayref pairs as you
2662 see in the L</WHERE CLAUSES> examples above. Sometimes it is necessary to
2663 alter this behavior. There are several ways of doing so.
2667 This is a virtual operator that signals the string to its right side is an
2668 identifier (a column name) and not a value. For example to compare two
2669 columns you would write:
2672 priority => { '<', 2 },
2673 requestor => { -ident => 'submitter' },
2678 $stmt = "WHERE priority < ? AND requestor = submitter";
2681 If you are maintaining legacy code you may see a different construct as
2682 described in L</Deprecated usage of Literal SQL>, please use C<-ident> in new
2687 This is a virtual operator that signals that the construct to its right side
2688 is a value to be passed to DBI. This is for example necessary when you want
2689 to write a where clause against an array (for RDBMS that support such
2690 datatypes). For example:
2693 array => { -value => [1, 2, 3] }
2698 $stmt = 'WHERE array = ?';
2699 @bind = ([1, 2, 3]);
2701 Note that if you were to simply say:
2707 the result would probably not be what you wanted:
2709 $stmt = 'WHERE array = ? OR array = ? OR array = ?';
2714 Finally, sometimes only literal SQL will do. To include a random snippet
2715 of SQL verbatim, you specify it as a scalar reference. Consider this only
2716 as a last resort. Usually there is a better way. For example:
2719 priority => { '<', 2 },
2720 requestor => { -in => \'(SELECT name FROM hitmen)' },
2725 $stmt = "WHERE priority < ? AND requestor IN (SELECT name FROM hitmen)"
2728 Note that in this example, you only get one bind parameter back, since
2729 the verbatim SQL is passed as part of the statement.
2733 Never use untrusted input as a literal SQL argument - this is a massive
2734 security risk (there is no way to check literal snippets for SQL
2735 injections and other nastyness). If you need to deal with untrusted input
2736 use literal SQL with placeholders as described next.
2738 =head3 Literal SQL with placeholders and bind values (subqueries)
2740 If the literal SQL to be inserted has placeholders and bind values,
2741 use a reference to an arrayref (yes this is a double reference --
2742 not so common, but perfectly legal Perl). For example, to find a date
2743 in Postgres you can use something like this:
2746 date_column => \[ "= date '2008-09-30' - ?::integer", 10 ]
2751 $stmt = "WHERE ( date_column = date '2008-09-30' - ?::integer )"
2754 Note that you must pass the bind values in the same format as they are returned
2755 by L<where|/where(\%where, $order)>. This means that if you set L</bindtype>
2756 to C<columns>, you must provide the bind values in the
2757 C<< [ column_meta => value ] >> format, where C<column_meta> is an opaque
2758 scalar value; most commonly the column name, but you can use any scalar value
2759 (including references and blessed references), L<SQL::Abstract> will simply
2760 pass it through intact. So if C<bindtype> is set to C<columns> the above
2761 example will look like:
2764 date_column => \[ "= date '2008-09-30' - ?::integer", [ {} => 10 ] ]
2767 Literal SQL is especially useful for nesting parenthesized clauses in the
2768 main SQL query. Here is a first example:
2770 my ($sub_stmt, @sub_bind) = ("SELECT c1 FROM t1 WHERE c2 < ? AND c3 LIKE ?",
2774 bar => \["IN ($sub_stmt)" => @sub_bind],
2779 $stmt = "WHERE (foo = ? AND bar IN (SELECT c1 FROM t1
2780 WHERE c2 < ? AND c3 LIKE ?))";
2781 @bind = (1234, 100, "foo%");
2783 Other subquery operators, like for example C<"E<gt> ALL"> or C<"NOT IN">,
2784 are expressed in the same way. Of course the C<$sub_stmt> and
2785 its associated bind values can be generated through a former call
2788 my ($sub_stmt, @sub_bind)
2789 = $sql->select("t1", "c1", {c2 => {"<" => 100},
2790 c3 => {-like => "foo%"}});
2793 bar => \["> ALL ($sub_stmt)" => @sub_bind],
2796 In the examples above, the subquery was used as an operator on a column;
2797 but the same principle also applies for a clause within the main C<%where>
2798 hash, like an EXISTS subquery:
2800 my ($sub_stmt, @sub_bind)
2801 = $sql->select("t1", "*", {c1 => 1, c2 => \"> t0.c0"});
2802 my %where = ( -and => [
2804 \["EXISTS ($sub_stmt)" => @sub_bind],
2809 $stmt = "WHERE (foo = ? AND EXISTS (SELECT * FROM t1
2810 WHERE c1 = ? AND c2 > t0.c0))";
2814 Observe that the condition on C<c2> in the subquery refers to
2815 column C<t0.c0> of the main query: this is I<not> a bind
2816 value, so we have to express it through a scalar ref.
2817 Writing C<< c2 => {">" => "t0.c0"} >> would have generated
2818 C<< c2 > ? >> with bind value C<"t0.c0"> ... not exactly
2819 what we wanted here.
2821 Finally, here is an example where a subquery is used
2822 for expressing unary negation:
2824 my ($sub_stmt, @sub_bind)
2825 = $sql->where({age => [{"<" => 10}, {">" => 20}]});
2826 $sub_stmt =~ s/^ where //i; # don't want "WHERE" in the subclause
2828 lname => {like => '%son%'},
2829 \["NOT ($sub_stmt)" => @sub_bind],
2834 $stmt = "lname LIKE ? AND NOT ( age < ? OR age > ? )"
2835 @bind = ('%son%', 10, 20)
2837 =head3 Deprecated usage of Literal SQL
2839 Below are some examples of archaic use of literal SQL. It is shown only as
2840 reference for those who deal with legacy code. Each example has a much
2841 better, cleaner and safer alternative that users should opt for in new code.
2847 my %where = ( requestor => \'IS NOT NULL' )
2849 $stmt = "WHERE requestor IS NOT NULL"
2851 This used to be the way of generating NULL comparisons, before the handling
2852 of C<undef> got formalized. For new code please use the superior syntax as
2853 described in L</Tests for NULL values>.
2857 my %where = ( requestor => \'= submitter' )
2859 $stmt = "WHERE requestor = submitter"
2861 This used to be the only way to compare columns. Use the superior L</-ident>
2862 method for all new code. For example an identifier declared in such a way
2863 will be properly quoted if L</quote_char> is properly set, while the legacy
2864 form will remain as supplied.
2868 my %where = ( is_ready => \"", completed => { '>', '2012-12-21' } )
2870 $stmt = "WHERE completed > ? AND is_ready"
2871 @bind = ('2012-12-21')
2873 Using an empty string literal used to be the only way to express a boolean.
2874 For all new code please use the much more readable
2875 L<-bool|/Unary operators: bool> operator.
2881 These pages could go on for a while, since the nesting of the data
2882 structures this module can handle are pretty much unlimited (the
2883 module implements the C<WHERE> expansion as a recursive function
2884 internally). Your best bet is to "play around" with the module a
2885 little to see how the data structures behave, and choose the best
2886 format for your data based on that.
2888 And of course, all the values above will probably be replaced with
2889 variables gotten from forms or the command line. After all, if you
2890 knew everything ahead of time, you wouldn't have to worry about
2891 dynamically-generating SQL and could just hardwire it into your
2894 =head1 ORDER BY CLAUSES
2896 Some functions take an order by clause. This can either be a scalar (just a
2897 column name), a hashref of C<< { -desc => 'col' } >> or C<< { -asc => 'col' }
2898 >>, a scalarref, an arrayref-ref, or an arrayref of any of the previous
2901 Given | Will Generate
2902 ---------------------------------------------------------------
2904 'colA' | ORDER BY colA
2906 [qw/colA colB/] | ORDER BY colA, colB
2908 {-asc => 'colA'} | ORDER BY colA ASC
2910 {-desc => 'colB'} | ORDER BY colB DESC
2912 ['colA', {-asc => 'colB'}] | ORDER BY colA, colB ASC
2914 { -asc => [qw/colA colB/] } | ORDER BY colA ASC, colB ASC
2916 \'colA DESC' | ORDER BY colA DESC
2918 \[ 'FUNC(colA, ?)', $x ] | ORDER BY FUNC(colA, ?)
2919 | /* ...with $x bound to ? */
2922 { -asc => 'colA' }, | colA ASC,
2923 { -desc => [qw/colB/] }, | colB DESC,
2924 { -asc => [qw/colC colD/] },| colC ASC, colD ASC,
2925 \'colE DESC', | colE DESC,
2926 \[ 'FUNC(colF, ?)', $x ], | FUNC(colF, ?)
2927 ] | /* ...with $x bound to ? */
2928 ===============================================================
2932 =head1 SPECIAL OPERATORS
2934 my $sqlmaker = SQL::Abstract->new(special_ops => [
2938 my ($self, $field, $op, $arg) = @_;
2944 handler => 'method_name',
2948 A "special operator" is a SQL syntactic clause that can be
2949 applied to a field, instead of a usual binary operator.
2952 WHERE field IN (?, ?, ?)
2953 WHERE field BETWEEN ? AND ?
2954 WHERE MATCH(field) AGAINST (?, ?)
2956 Special operators IN and BETWEEN are fairly standard and therefore
2957 are builtin within C<SQL::Abstract> (as the overridable methods
2958 C<_where_field_IN> and C<_where_field_BETWEEN>). For other operators,
2959 like the MATCH .. AGAINST example above which is specific to MySQL,
2960 you can write your own operator handlers - supply a C<special_ops>
2961 argument to the C<new> method. That argument takes an arrayref of
2962 operator definitions; each operator definition is a hashref with two
2969 the regular expression to match the operator
2973 Either a coderef or a plain scalar method name. In both cases
2974 the expected return is C<< ($sql, @bind) >>.
2976 When supplied with a method name, it is simply called on the
2977 L<SQL::Abstract> object as:
2979 $self->$method_name($field, $op, $arg)
2983 $field is the LHS of the operator
2984 $op is the part that matched the handler regex
2987 When supplied with a coderef, it is called as:
2989 $coderef->($self, $field, $op, $arg)
2994 For example, here is an implementation
2995 of the MATCH .. AGAINST syntax for MySQL
2997 my $sqlmaker = SQL::Abstract->new(special_ops => [
2999 # special op for MySql MATCH (field) AGAINST(word1, word2, ...)
3000 {regex => qr/^match$/i,
3002 my ($self, $field, $op, $arg) = @_;
3003 $arg = [$arg] if not ref $arg;
3004 my $label = $self->_quote($field);
3005 my ($placeholder) = $self->_convert('?');
3006 my $placeholders = join ", ", (($placeholder) x @$arg);
3007 my $sql = $self->_sqlcase('match') . " ($label) "
3008 . $self->_sqlcase('against') . " ($placeholders) ";
3009 my @bind = $self->_bindtype($field, @$arg);
3010 return ($sql, @bind);
3017 =head1 UNARY OPERATORS
3019 my $sqlmaker = SQL::Abstract->new(unary_ops => [
3023 my ($self, $op, $arg) = @_;
3029 handler => 'method_name',
3033 A "unary operator" is a SQL syntactic clause that can be
3034 applied to a field - the operator goes before the field
3036 You can write your own operator handlers - supply a C<unary_ops>
3037 argument to the C<new> method. That argument takes an arrayref of
3038 operator definitions; each operator definition is a hashref with two
3045 the regular expression to match the operator
3049 Either a coderef or a plain scalar method name. In both cases
3050 the expected return is C<< $sql >>.
3052 When supplied with a method name, it is simply called on the
3053 L<SQL::Abstract> object as:
3055 $self->$method_name($op, $arg)
3059 $op is the part that matched the handler regex
3060 $arg is the RHS or argument of the operator
3062 When supplied with a coderef, it is called as:
3064 $coderef->($self, $op, $arg)
3072 Thanks to some benchmarking by Mark Stosberg, it turns out that
3073 this module is many orders of magnitude faster than using C<DBIx::Abstract>.
3074 I must admit this wasn't an intentional design issue, but it's a
3075 byproduct of the fact that you get to control your C<DBI> handles
3078 To maximize performance, use a code snippet like the following:
3080 # prepare a statement handle using the first row
3081 # and then reuse it for the rest of the rows
3083 for my $href (@array_of_hashrefs) {
3084 $stmt ||= $sql->insert('table', $href);
3085 $sth ||= $dbh->prepare($stmt);
3086 $sth->execute($sql->values($href));
3089 The reason this works is because the keys in your C<$href> are sorted
3090 internally by B<SQL::Abstract>. Thus, as long as your data retains
3091 the same structure, you only have to generate the SQL the first time
3092 around. On subsequent queries, simply use the C<values> function provided
3093 by this module to return your values in the correct order.
3095 However this depends on the values having the same type - if, for
3096 example, the values of a where clause may either have values
3097 (resulting in sql of the form C<column = ?> with a single bind
3098 value), or alternatively the values might be C<undef> (resulting in
3099 sql of the form C<column IS NULL> with no bind value) then the
3100 caching technique suggested will not work.
3104 If you use my C<CGI::FormBuilder> module at all, you'll hopefully
3105 really like this part (I do, at least). Building up a complex query
3106 can be as simple as the following:
3113 use CGI::FormBuilder;
3116 my $form = CGI::FormBuilder->new(...);
3117 my $sql = SQL::Abstract->new;
3119 if ($form->submitted) {
3120 my $field = $form->field;
3121 my $id = delete $field->{id};
3122 my($stmt, @bind) = $sql->update('table', $field, {id => $id});
3125 Of course, you would still have to connect using C<DBI> to run the
3126 query, but the point is that if you make your form look like your
3127 table, the actual query script can be extremely simplistic.
3129 If you're B<REALLY> lazy (I am), check out C<HTML::QuickTable> for
3130 a fast interface to returning and formatting data. I frequently
3131 use these three modules together to write complex database query
3132 apps in under 50 lines.
3134 =head1 HOW TO CONTRIBUTE
3136 Contributions are always welcome, in all usable forms (we especially
3137 welcome documentation improvements). The delivery methods include git-
3138 or unified-diff formatted patches, GitHub pull requests, or plain bug
3139 reports either via RT or the Mailing list. Contributors are generally
3140 granted full access to the official repository after their first several
3141 patches pass successful review.
3143 This project is maintained in a git repository. The code and related tools are
3144 accessible at the following locations:
3148 =item * Official repo: L<git://git.shadowcat.co.uk/dbsrgits/SQL-Abstract.git>
3150 =item * Official gitweb: L<http://git.shadowcat.co.uk/gitweb/gitweb.cgi?p=dbsrgits/SQL-Abstract.git>
3152 =item * GitHub mirror: L<https://github.com/dbsrgits/sql-abstract>
3154 =item * Authorized committers: L<ssh://dbsrgits@git.shadowcat.co.uk/SQL-Abstract.git>
3160 Version 1.50 was a major internal refactoring of C<SQL::Abstract>.
3161 Great care has been taken to preserve the I<published> behavior
3162 documented in previous versions in the 1.* family; however,
3163 some features that were previously undocumented, or behaved
3164 differently from the documentation, had to be changed in order
3165 to clarify the semantics. Hence, client code that was relying
3166 on some dark areas of C<SQL::Abstract> v1.*
3167 B<might behave differently> in v1.50.
3169 The main changes are:
3175 support for literal SQL through the C<< \ [ $sql, @bind ] >> syntax.
3179 support for the { operator => \"..." } construct (to embed literal SQL)
3183 support for the { operator => \["...", @bind] } construct (to embed literal SQL with bind values)
3187 optional support for L<array datatypes|/"Inserting and Updating Arrays">
3191 defensive programming: check arguments
3195 fixed bug with global logic, which was previously implemented
3196 through global variables yielding side-effects. Prior versions would
3197 interpret C<< [ {cond1, cond2}, [cond3, cond4] ] >>
3198 as C<< "(cond1 AND cond2) OR (cond3 AND cond4)" >>.
3199 Now this is interpreted
3200 as C<< "(cond1 AND cond2) OR (cond3 OR cond4)" >>.
3205 fixed semantics of _bindtype on array args
3209 dropped the C<_anoncopy> of the %where tree. No longer necessary,
3210 we just avoid shifting arrays within that tree.
3214 dropped the C<_modlogic> function
3218 =head1 ACKNOWLEDGEMENTS
3220 There are a number of individuals that have really helped out with
3221 this module. Unfortunately, most of them submitted bugs via CPAN
3222 so I have no idea who they are! But the people I do know are:
3224 Ash Berlin (order_by hash term support)
3225 Matt Trout (DBIx::Class support)
3226 Mark Stosberg (benchmarking)
3227 Chas Owens (initial "IN" operator support)
3228 Philip Collins (per-field SQL functions)
3229 Eric Kolve (hashref "AND" support)
3230 Mike Fragassi (enhancements to "BETWEEN" and "LIKE")
3231 Dan Kubb (support for "quote_char" and "name_sep")
3232 Guillermo Roditi (patch to cleanup "IN" and "BETWEEN", fix and tests for _order_by)
3233 Laurent Dami (internal refactoring, extensible list of special operators, literal SQL)
3234 Norbert Buchmuller (support for literal SQL in hashpair, misc. fixes & tests)
3235 Peter Rabbitson (rewrite of SQLA::Test, misc. fixes & tests)
3236 Oliver Charles (support for "RETURNING" after "INSERT")
3242 L<DBIx::Class>, L<DBIx::Abstract>, L<CGI::FormBuilder>, L<HTML::QuickTable>.
3246 Copyright (c) 2001-2007 Nathan Wiger <nwiger@cpan.org>. All Rights Reserved.
3248 This module is actively maintained by Matt Trout <mst@shadowcatsystems.co.uk>
3250 For support, your best bet is to try the C<DBIx::Class> users mailing list.
3251 While not an official support venue, C<DBIx::Class> makes heavy use of
3252 C<SQL::Abstract>, and as such list members there are very familiar with
3253 how to create queries.
3257 This module is free software; you may copy this under the same
3258 terms as perl itself (either the GNU General Public License or
3259 the Artistic License)