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 = '1.87';
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" }},
43 {regex => qr/^ ident $/ix, handler => sub { die "NOPE" }},
44 {regex => qr/^ value $/ix, handler => sub { die "NOPE" }},
47 #======================================================================
48 # DEBUGGING AND ERROR REPORTING
49 #======================================================================
52 return unless $_[0]->{debug}; shift; # a little faster
53 my $func = (caller(1))[3];
54 warn "[$func] ", @_, "\n";
58 my($func) = (caller(1))[3];
59 Carp::carp "[$func] Warning: ", @_;
63 my($func) = (caller(1))[3];
64 Carp::croak "[$func] Fatal: ", @_;
67 sub is_literal_value ($) {
68 ref $_[0] eq 'SCALAR' ? [ ${$_[0]} ]
69 : ( ref $_[0] eq 'REF' and ref ${$_[0]} eq 'ARRAY' ) ? [ @${ $_[0] } ]
73 sub is_undef_value ($) {
77 and exists $_[0]->{-value}
78 and not defined $_[0]->{-value}
82 # FIXME XSify - this can be done so much more efficiently
83 sub is_plain_value ($) {
85 ! length ref $_[0] ? \($_[0])
87 ref $_[0] eq 'HASH' and keys %{$_[0]} == 1
89 exists $_[0]->{-value}
90 ) ? \($_[0]->{-value})
92 # reuse @_ for even moar speedz
93 defined ( $_[1] = Scalar::Util::blessed $_[0] )
95 # deliberately not using Devel::OverloadInfo - the checks we are
96 # intersted in are much more limited than the fullblown thing, and
97 # this is a very hot piece of code
99 # simply using ->can('(""') can leave behind stub methods that
100 # break actually using the overload later (see L<perldiag/Stub
101 # found while resolving method "%s" overloading "%s" in package
102 # "%s"> and the source of overload::mycan())
104 # either has stringification which DBI SHOULD prefer out of the box
105 grep { *{ (qq[${_}::(""]) }{CODE} } @{ $_[2] = mro::get_linear_isa( $_[1] ) }
107 # has nummification or boolification, AND fallback is *not* disabled
109 SQL::Abstract::_ENV_::DETECT_AUTOGENERATED_STRINGIFICATION
112 grep { *{"${_}::(0+"}{CODE} } @{$_[2]}
114 grep { *{"${_}::(bool"}{CODE} } @{$_[2]}
118 # no fallback specified at all
119 ! ( ($_[3]) = grep { *{"${_}::()"}{CODE} } @{$_[2]} )
121 # fallback explicitly undef
122 ! defined ${"$_[3]::()"}
135 #======================================================================
137 #======================================================================
141 my $class = ref($self) || $self;
142 my %opt = (ref $_[0] eq 'HASH') ? %{$_[0]} : @_;
144 # choose our case by keeping an option around
145 delete $opt{case} if $opt{case} && $opt{case} ne 'lower';
147 # default logic for interpreting arrayrefs
148 $opt{logic} = $opt{logic} ? uc $opt{logic} : 'OR';
150 # how to return bind vars
151 $opt{bindtype} ||= 'normal';
153 # default comparison is "=", but can be overridden
156 # try to recognize which are the 'equality' and 'inequality' ops
157 # (temporary quickfix (in 2007), should go through a more seasoned API)
158 $opt{equality_op} = qr/^( \Q$opt{cmp}\E | \= )$/ix;
159 $opt{inequality_op} = qr/^( != | <> )$/ix;
161 $opt{like_op} = qr/^ (is_)?r?like $/xi;
162 $opt{not_like_op} = qr/^ (is_)?not_r?like $/xi;
165 $opt{sqltrue} ||= '1=1';
166 $opt{sqlfalse} ||= '0=1';
169 $opt{special_ops} ||= [];
171 # regexes are applied in order, thus push after user-defines
172 push @{$opt{special_ops}}, @BUILTIN_SPECIAL_OPS;
174 if ($class->isa('DBIx::Class::SQLMaker')) {
175 $opt{is_dbic_sqlmaker} = 1;
176 $opt{disable_old_special_ops} = 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 $opt{expand_unary} = {};
199 -not => '_expand_not',
200 -bool => '_expand_bool',
201 -and => '_expand_op_andor',
202 -or => '_expand_op_andor',
203 -nest => '_expand_nest',
204 -bind => sub { shift; +{ @_ } },
206 -not_in => '_expand_in',
208 my ($self, $node, $args) = @_;
209 +{ $node => [ map $self->expand_expr($_), @$args ] };
211 -between => '_expand_between',
212 -not_between => '_expand_between',
216 'between' => '_expand_between',
217 'not_between' => '_expand_between',
218 'in' => '_expand_in',
219 'not_in' => '_expand_in',
220 'nest' => '_expand_nest',
221 (map +($_ => '_expand_op_andor'), ('and', 'or')),
222 (map +($_ => '_expand_op_is'), ('is', 'is_not')),
225 # placeholder for _expand_unop system
227 my %unops = (-ident => '_expand_ident', -value => '_expand_value');
228 foreach my $name (keys %unops) {
229 $opt{expand}{$name} = $unops{$name};
230 my ($op) = $name =~ /^-(.*)$/;
231 $opt{expand_op}{$op} = sub {
232 my ($self, $op, $arg, $k) = @_;
233 return $self->_expand_expr_hashpair_cmp(
234 $k, { "-${op}" => $arg }
241 (map +("-$_", "_render_$_"), qw(op func bind ident literal tuple)),
246 (map +($_ => '_render_op_between'), 'between', 'not_between'),
247 (map +($_ => '_render_op_in'), 'in', 'not_in'),
248 (map +($_ => '_render_unop_postfix'),
249 'is_null', 'is_not_null', 'asc', 'desc',
251 (not => '_render_op_not'),
252 (map +($_ => '_render_op_andor'), qw(and or)),
253 ',' => '_render_op_multop',
256 return bless \%opt, $class;
259 sub sqltrue { +{ -literal => [ $_[0]->{sqltrue} ] } }
260 sub sqlfalse { +{ -literal => [ $_[0]->{sqlfalse} ] } }
262 sub _assert_pass_injection_guard {
263 if ($_[1] =~ $_[0]->{injection_guard}) {
264 my $class = ref $_[0];
265 puke "Possible SQL injection attempt '$_[1]'. If this is indeed a part of the "
266 . "desired SQL use literal SQL ( \'...' or \[ '...' ] ) or supply your own "
267 . "{injection_guard} attribute to ${class}->new()"
272 #======================================================================
274 #======================================================================
278 my $table = $self->_table(shift);
279 my $data = shift || return;
282 my $method = $self->_METHOD_FOR_refkind("_insert", $data);
283 my ($sql, @bind) = $self->$method($data);
284 $sql = join " ", $self->_sqlcase('insert into'), $table, $sql;
286 if ($options->{returning}) {
287 my ($s, @b) = $self->_insert_returning($options);
292 return wantarray ? ($sql, @bind) : $sql;
295 # So that subclasses can override INSERT ... RETURNING separately from
296 # UPDATE and DELETE (e.g. DBIx::Class::SQLMaker::Oracle does this)
297 sub _insert_returning { shift->_returning(@_) }
300 my ($self, $options) = @_;
302 my $f = $options->{returning};
304 my ($sql, @bind) = $self->render_aqt(
305 $self->_expand_maybe_list_expr($f, -ident)
308 ? $self->_sqlcase(' returning ') . $sql
309 : ($self->_sqlcase(' returning ').$sql, @bind);
312 sub _insert_HASHREF { # explicit list of fields and then values
313 my ($self, $data) = @_;
315 my @fields = sort keys %$data;
317 my ($sql, @bind) = $self->_insert_values($data);
320 $_ = $self->_quote($_) foreach @fields;
321 $sql = "( ".join(", ", @fields).") ".$sql;
323 return ($sql, @bind);
326 sub _insert_ARRAYREF { # just generate values(?,?) part (no list of fields)
327 my ($self, $data) = @_;
329 # no names (arrayref) so can't generate bindtype
330 $self->{bindtype} ne 'columns'
331 or belch "can't do 'columns' bindtype when called with arrayref";
333 my (@values, @all_bind);
334 foreach my $value (@$data) {
335 my ($values, @bind) = $self->_insert_value(undef, $value);
336 push @values, $values;
337 push @all_bind, @bind;
339 my $sql = $self->_sqlcase('values')." ( ".join(", ", @values)." )";
340 return ($sql, @all_bind);
343 sub _insert_ARRAYREFREF { # literal SQL with bind
344 my ($self, $data) = @_;
346 my ($sql, @bind) = @${$data};
347 $self->_assert_bindval_matches_bindtype(@bind);
349 return ($sql, @bind);
353 sub _insert_SCALARREF { # literal SQL without bind
354 my ($self, $data) = @_;
360 my ($self, $data) = @_;
362 my (@values, @all_bind);
363 foreach my $column (sort keys %$data) {
364 my ($values, @bind) = $self->_insert_value($column, $data->{$column});
365 push @values, $values;
366 push @all_bind, @bind;
368 my $sql = $self->_sqlcase('values')." ( ".join(", ", @values)." )";
369 return ($sql, @all_bind);
373 my ($self, $column, $v) = @_;
375 return $self->render_aqt(
376 $self->_expand_insert_value($column, $v)
380 sub _expand_insert_value {
381 my ($self, $column, $v) = @_;
383 if (ref($v) eq 'ARRAY') {
384 if ($self->{array_datatypes}) {
385 return +{ -bind => [ $column, $v ] };
387 my ($sql, @bind) = @$v;
388 $self->_assert_bindval_matches_bindtype(@bind);
389 return +{ -literal => $v };
391 if (ref($v) eq 'HASH') {
392 if (grep !/^-/, keys %$v) {
393 belch "HASH ref as bind value in insert is not supported";
394 return +{ -bind => [ $column, $v ] };
398 return +{ -bind => [ $column, undef ] };
400 local our $Cur_Col_Meta = $column;
401 return $self->expand_expr($v);
406 #======================================================================
408 #======================================================================
413 my $table = $self->_table(shift);
414 my $data = shift || return;
418 # first build the 'SET' part of the sql statement
419 puke "Unsupported data type specified to \$sql->update"
420 unless ref $data eq 'HASH';
422 my ($sql, @all_bind) = $self->_update_set_values($data);
423 $sql = $self->_sqlcase('update ') . $table . $self->_sqlcase(' set ')
427 my($where_sql, @where_bind) = $self->where($where);
429 push @all_bind, @where_bind;
432 if ($options->{returning}) {
433 my ($returning_sql, @returning_bind) = $self->_update_returning($options);
434 $sql .= $returning_sql;
435 push @all_bind, @returning_bind;
438 return wantarray ? ($sql, @all_bind) : $sql;
441 sub _update_set_values {
442 my ($self, $data) = @_;
444 return $self->render_aqt(
445 $self->_expand_update_set_values($data),
449 sub _expand_update_set_values {
450 my ($self, $data) = @_;
451 $self->_expand_maybe_list_expr( [
454 $set = { -bind => $_ } unless defined $set;
455 +{ -op => [ '=', $self->_expand_ident(-ident => $k), $set ] };
461 ? ($self->{array_datatypes}
462 ? [ $k, +{ -bind => [ $k, $v ] } ]
463 : [ $k, +{ -literal => $v } ])
465 local our $Cur_Col_Meta = $k;
466 [ $k, $self->_expand_expr($v) ]
473 # So that subclasses can override UPDATE ... RETURNING separately from
475 sub _update_returning { shift->_returning(@_) }
479 #======================================================================
481 #======================================================================
486 my $table = $self->_table(shift);
487 my $fields = shift || '*';
491 my ($fields_sql, @bind) = $self->_select_fields($fields);
493 my ($where_sql, @where_bind) = $self->where($where, $order);
494 push @bind, @where_bind;
496 my $sql = join(' ', $self->_sqlcase('select'), $fields_sql,
497 $self->_sqlcase('from'), $table)
500 return wantarray ? ($sql, @bind) : $sql;
504 my ($self, $fields) = @_;
505 return $fields unless ref($fields);
506 return $self->render_aqt(
507 $self->_expand_maybe_list_expr($fields, '-ident')
511 #======================================================================
513 #======================================================================
518 my $table = $self->_table(shift);
522 my($where_sql, @bind) = $self->where($where);
523 my $sql = $self->_sqlcase('delete from ') . $table . $where_sql;
525 if ($options->{returning}) {
526 my ($returning_sql, @returning_bind) = $self->_delete_returning($options);
527 $sql .= $returning_sql;
528 push @bind, @returning_bind;
531 return wantarray ? ($sql, @bind) : $sql;
534 # So that subclasses can override DELETE ... RETURNING separately from
536 sub _delete_returning { shift->_returning(@_) }
540 #======================================================================
542 #======================================================================
546 # Finally, a separate routine just to handle WHERE clauses
548 my ($self, $where, $order) = @_;
550 local $self->{convert_where} = $self->{convert};
553 my ($sql, @bind) = defined($where)
554 ? $self->_recurse_where($where)
556 $sql = (defined $sql and length $sql) ? $self->_sqlcase(' where ') . "( $sql )" : '';
560 my ($order_sql, @order_bind) = $self->_order_by($order);
562 push @bind, @order_bind;
565 return wantarray ? ($sql, @bind) : $sql;
568 { our $Default_Scalar_To = -value }
571 my ($self, $expr, $default_scalar_to) = @_;
572 local our $Default_Scalar_To = $default_scalar_to if $default_scalar_to;
573 $self->_expand_expr($expr);
577 my ($self, $aqt) = @_;
578 my ($k, $v, @rest) = %$aqt;
580 if (my $meth = $self->{render}{$k}) {
581 return $self->$meth($v);
583 die "notreached: $k";
587 my ($self, $expr) = @_;
588 $self->render_aqt($self->expand_expr($expr));
592 my ($self, $raw) = @_;
593 s/^-(?=.)//, s/\s+/_/g for my $op = lc $raw;
598 my ($self, $expr) = @_;
599 our $Expand_Depth ||= 0; local $Expand_Depth = $Expand_Depth + 1;
600 return undef unless defined($expr);
601 if (ref($expr) eq 'HASH') {
602 return undef unless my $kc = keys %$expr;
604 return $self->_expand_op_andor(-and => $expr);
606 my ($key, $value) = %$expr;
607 if ($key =~ /^-/ and $key =~ s/ [_\s]? \d+ $//x ) {
608 belch 'Use of [and|or|nest]_N modifiers is deprecated and will be removed in SQLA v2.0. '
609 . "You probably wanted ...-and => [ $key => COND1, $key => COND2 ... ]";
611 if (my $exp = $self->{expand}{$key}) {
612 return $self->$exp($key, $value);
614 return $self->_expand_expr_hashpair($key, $value);
616 if (ref($expr) eq 'ARRAY') {
617 my $logic = '-'.lc($self->{logic});
618 return $self->_expand_op_andor($logic, $expr);
620 if (my $literal = is_literal_value($expr)) {
621 return +{ -literal => $literal };
623 if (!ref($expr) or Scalar::Util::blessed($expr)) {
624 return $self->_expand_expr_scalar($expr);
629 sub _expand_expr_hashpair {
630 my ($self, $k, $v) = @_;
631 unless (defined($k) and length($k)) {
632 if (defined($k) and my $literal = is_literal_value($v)) {
633 belch 'Hash-pairs consisting of an empty string with a literal are deprecated, and will be removed in 2.0: use -and => [ $literal ] instead';
634 return { -literal => $literal };
636 puke "Supplying an empty left hand side argument is not supported";
639 return $self->_expand_expr_hashpair_op($k, $v);
641 return $self->_expand_expr_hashpair_ident($k, $v);
644 sub _expand_expr_hashpair_ident {
645 my ($self, $k, $v) = @_;
647 local our $Cur_Col_Meta = $k;
649 # hash with multiple or no elements is andor
651 if (ref($v) eq 'HASH' and keys %$v != 1) {
652 return $self->_expand_op_andor(-and => $v, $k);
655 # undef needs to be re-sent with cmp to achieve IS/IS NOT NULL
657 if (is_undef_value($v)) {
658 return $self->_expand_expr_hashpair_cmp($k => undef);
661 # scalars and objects get expanded as whatever requested or values
663 if (!ref($v) or Scalar::Util::blessed($v)) {
664 return $self->_expand_expr_hashpair_scalar($k, $v);
667 # single key hashref is a hashtriple
669 if (ref($v) eq 'HASH') {
670 return $self->_expand_expr_hashtriple($k, %$v);
673 # arrayref needs re-engineering over the elements
675 if (ref($v) eq 'ARRAY') {
676 return $self->sqlfalse unless @$v;
677 $self->_debug("ARRAY($k) means distribute over elements");
679 $v->[0] =~ /^-(and|or)$/i
680 ? shift(@{$v = [ @$v ]})
681 : '-'.lc($self->{logic} || 'OR')
683 return $self->_expand_op_andor(
688 if (my $literal = is_literal_value($v)) {
690 belch 'Hash-pairs consisting of an empty string with a literal are deprecated, and will be removed in 2.0: use -and => [ $literal ] instead';
693 my ($sql, @bind) = @$literal;
694 if ($self->{bindtype} eq 'columns') {
696 $self->_assert_bindval_matches_bindtype($_);
699 return +{ -literal => [ $self->_quote($k).' '.$sql, @bind ] };
704 sub _expand_expr_scalar {
705 my ($self, $expr) = @_;
707 return $self->_expand_expr({ (our $Default_Scalar_To) => $expr });
710 sub _expand_expr_hashpair_scalar {
711 my ($self, $k, $v) = @_;
713 return $self->_expand_expr_hashpair_cmp(
714 $k, $self->_expand_expr_scalar($v),
718 sub _expand_expr_hashpair_op {
719 my ($self, $k, $v) = @_;
721 $self->_assert_pass_injection_guard($k =~ /\A-(.*)\Z/s);
723 my $op = $self->_normalize_op($k);
725 # Ops prefixed with -not_ get converted
727 if (my ($rest) = $op =~/^not_(.*)$/) {
730 $self->_expand_expr({ "-${rest}", $v })
736 my $op = join(' ', split '_', $op);
738 # the old special op system requires illegality for top-level use
741 (our $Expand_Depth) == 1
742 and $self->{disable_old_special_ops}
743 and List::Util::first { $op =~ $_->{regex} } @{$self->{special_ops}}
745 puke "Illegal use of top-level '-$op'"
748 # the old unary op system means we should touch nothing and let it work
750 if (my $us = List::Util::first { $op =~ $_->{regex} } @{$self->{unary_ops}}) {
751 return { -op => [ $op, $v ] };
755 # an explicit node type is currently assumed to be expanded (this is almost
756 # certainly wrong and there should be expansion anyway)
758 if ($self->{render}{$k}) {
762 # hashref RHS values get expanded and used as op/func args
767 and (keys %$v)[0] =~ /^-/
769 my ($func) = $k =~ /^-(.*)$/;
771 if (List::Util::first { $func =~ $_->{regex} } @{$self->{special_ops}}) {
772 return +{ -op => [ $func, $self->_expand_expr($v) ] };
775 return +{ -func => [ $func, $self->_expand_expr($v) ] };
778 # scalars and literals get simply expanded
780 if (!ref($v) or is_literal_value($v)) {
781 return +{ -op => [ $op, $self->_expand_expr($v) ] };
787 sub _expand_expr_hashpair_cmp {
788 my ($self, $k, $v) = @_;
789 $self->_expand_expr_hashtriple($k, $self->{cmp}, $v);
792 sub _expand_expr_hashtriple {
793 my ($self, $k, $vk, $vv) = @_;
795 my $ik = $self->_expand_ident(-ident => $k);
797 my $op = $self->_normalize_op($vk);
798 $self->_assert_pass_injection_guard($op);
800 if ($op =~ s/ _? \d+ $//x ) {
801 return $self->_expand_expr($k, { $vk, $vv });
803 if (my $x = $self->{expand_op}{$op}) {
804 local our $Cur_Col_Meta = $k;
805 return $self->$x($op, $vv, $k);
809 my $op = join(' ', split '_', $op);
811 if (my $us = List::Util::first { $op =~ $_->{regex} } @{$self->{special_ops}}) {
812 return { -op => [ $op, $ik, $vv ] };
814 if (my $us = List::Util::first { $op =~ $_->{regex} } @{$self->{unary_ops}}) {
818 { -op => [ $op, $vv ] }
822 if (ref($vv) eq 'ARRAY') {
824 my $logic = (defined($raw[0]) and $raw[0] =~ /^-(and|or)$/i)
825 ? shift @raw : '-or';
826 my @values = map +{ $vk => $_ }, @raw;
828 $op =~ $self->{inequality_op}
829 or $op =~ $self->{not_like_op}
831 if (lc($logic) eq '-or' and @values > 1) {
832 belch "A multi-element arrayref as an argument to the inequality op '${\uc(join ' ', split '_', $op)}' "
833 . 'is technically equivalent to an always-true 1=1 (you probably wanted '
834 . "to say ...{ \$inequality_op => [ -and => \@values ] }... instead)"
839 # try to DWIM on equality operators
840 return ($self->_dwim_op_to_is($op,
841 "Supplying an empty arrayref to '%s' is deprecated",
842 "operator '%s' applied on an empty array (field '$k')"
843 ) ? $self->sqlfalse : $self->sqltrue);
845 return $self->_expand_op_andor($logic => \@values, $k);
847 if (is_undef_value($vv)) {
848 my $is = ($self->_dwim_op_to_is($op,
849 "Supplying an undefined argument to '%s' is deprecated",
850 "unexpected operator '%s' with undef operand",
851 ) ? 'is' : 'is not');
853 return $self->_expand_expr_hashpair($k => { $is, undef });
855 local our $Cur_Col_Meta = $k;
859 $self->_expand_expr($vv)
864 my ($self, $raw, $empty, $fail) = @_;
866 my $op = $self->_normalize_op($raw);
868 if ($op =~ /^not$/i) {
871 if ($op =~ $self->{equality_op}) {
874 if ($op =~ $self->{like_op}) {
875 belch(sprintf $empty, uc(join ' ', split '_', $op));
878 if ($op =~ $self->{inequality_op}) {
881 if ($op =~ $self->{not_like_op}) {
882 belch(sprintf $empty, uc(join ' ', split '_', $op));
885 puke(sprintf $fail, $op);
889 my ($self, $op, $body) = @_;
890 unless (defined($body) or (ref($body) and ref($body) eq 'ARRAY')) {
891 puke "$op requires a single plain scalar argument (a quotable identifier) or an arrayref of identifier parts";
893 my @parts = map split(/\Q${\($self->{name_sep}||'.')}\E/, $_),
894 ref($body) ? @$body : $body;
895 return { -ident => $parts[-1] } if $self->{_dequalify_idents};
896 unless ($self->{quote_char}) {
897 $self->_assert_pass_injection_guard($_) for @parts;
899 return +{ -ident => \@parts };
903 +{ -bind => [ our $Cur_Col_Meta, $_[2] ] };
907 +{ -op => [ 'not', $_[0]->_expand_expr($_[2]) ] };
911 my ($self, undef, $v) = @_;
913 return $self->_expand_expr($v);
915 puke "-bool => undef not supported" unless defined($v);
916 return $self->_expand_ident(-ident => $v);
919 sub _expand_op_andor {
920 my ($self, $logic, $v, $k) = @_;
922 $v = [ map +{ $k, $_ },
924 ? (map +{ $_ => $v->{$_} }, sort keys %$v)
928 my ($logop) = $logic =~ /^-?(.*)$/;
929 if (ref($v) eq 'HASH') {
930 return undef unless keys %$v;
933 map $self->_expand_expr({ $_ => $v->{$_} }),
937 if (ref($v) eq 'ARRAY') {
938 $logop eq 'and' or $logop eq 'or' or puke "unknown logic: $logop";
941 (ref($_) eq 'ARRAY' and @$_)
942 or (ref($_) eq 'HASH' and %$_)
948 while (my ($el) = splice @expr, 0, 1) {
949 puke "Supplying an empty left hand side argument is not supported in array-pairs"
950 unless defined($el) and length($el);
951 my $elref = ref($el);
953 local our $Expand_Depth = 0;
954 push(@res, grep defined, $self->_expand_expr({ $el, shift(@expr) }));
955 } elsif ($elref eq 'ARRAY') {
956 push(@res, grep defined, $self->_expand_expr($el)) if @$el;
957 } elsif (my $l = is_literal_value($el)) {
958 push @res, { -literal => $l };
959 } elsif ($elref eq 'HASH') {
960 local our $Expand_Depth = 0;
961 push @res, grep defined, $self->_expand_expr($el) if %$el;
967 # return $res[0] if @res == 1;
968 return { -op => [ $logop, @res ] };
974 my ($self, $op, $vv, $k) = @_;
975 puke "$op can only take undef as argument"
979 and exists($vv->{-value})
980 and !defined($vv->{-value})
982 return +{ -op => [ $op.'_null', $self->_expand_ident(-ident => $k) ] };
985 sub _expand_between {
986 my ($self, $op, $vv, $k) = @_;
988 $k = shift @{$vv = [ @$vv ]} unless defined $k;
989 local our $Cur_Col_Meta = $k;
990 my @rhs = map $self->_expand_expr($_),
991 ref($vv) eq 'ARRAY' ? @$vv : $vv;
993 (@rhs == 1 and ref($rhs[0]) eq 'HASH' and $rhs[0]->{-literal})
995 (@rhs == 2 and defined($rhs[0]) and defined($rhs[1]))
997 puke "Operator '${\uc($op)}' requires either an arrayref with two defined values or expressions, or a single literal scalarref/arrayref-ref";
1001 $self->_expand_ident(-ident => $k),
1007 my ($self, $raw, $vv, $k) = @_;
1008 $k = shift @{$vv = [ @$vv ]} unless defined $k;
1009 my $op = $self->_normalize_op($raw);
1010 if (my $literal = is_literal_value($vv)) {
1011 my ($sql, @bind) = @$literal;
1012 my $opened_sql = $self->_open_outer_paren($sql);
1014 $op, $self->expand_expr($k, -ident),
1015 [ { -literal => [ $opened_sql, @bind ] } ]
1019 'SQL::Abstract before v1.75 used to generate incorrect SQL when the '
1020 . "-${\uc($op)} operator was given an undef-containing list: !!!AUDIT YOUR CODE "
1021 . 'AND DATA!!! (the upcoming Data::Query-based version of SQL::Abstract '
1022 . 'will emit the logically correct SQL instead of raising this exception)'
1024 puke("Argument passed to the '${\uc($op)}' operator can not be undefined")
1026 my @rhs = map $self->expand_expr($_, -value),
1027 map { defined($_) ? $_: puke($undef_err) }
1028 (ref($vv) eq 'ARRAY' ? @$vv : $vv);
1029 return $self->${\($op =~ /^not/ ? 'sqltrue' : 'sqlfalse')} unless @rhs;
1033 $self->expand_expr($k, -ident),
1039 my ($self, $op, $v) = @_;
1040 # DBIx::Class requires a nest warning to be emitted once but the private
1041 # method it overrode to do so no longer exists
1042 if ($self->{is_dbic_sqlmaker}) {
1043 unless (our $Nest_Warned) {
1045 "-nest in search conditions is deprecated, you most probably wanted:\n"
1046 .q|{..., -and => [ \%cond0, \@cond1, \'cond2', \[ 'cond3', [ col => bind ] ], etc. ], ... }|
1051 return $self->_expand_expr($v);
1054 sub _recurse_where {
1055 my ($self, $where, $logic) = @_;
1057 # Special case: top level simple string treated as literal
1059 my $where_exp = (ref($where)
1060 ? $self->_expand_expr($where, $logic)
1061 : { -literal => [ $where ] });
1063 # dispatch expanded expression
1065 my ($sql, @bind) = defined($where_exp) ? $self->render_aqt($where_exp) : (undef);
1066 # DBIx::Class used to call _recurse_where in scalar context
1067 # something else might too...
1069 return ($sql, @bind);
1072 belch "Calling _recurse_where in scalar context is deprecated and will go away before 2.0";
1078 my ($self, $ident) = @_;
1080 return $self->_convert($self->_quote($ident));
1084 my ($self, $values) = @_;
1085 my ($sql, @bind) = $self->_render_op([ ',', @$values ]);
1086 return "($sql)", @bind;
1090 my ($self, $rest) = @_;
1091 my ($func, @args) = @$rest;
1095 push @arg_sql, shift @x;
1097 } map [ $self->render_aqt($_) ], @args;
1098 return ($self->_sqlcase($func).'('.join(', ', @arg_sql).')', @bind);
1102 my ($self, $bind) = @_;
1103 return ($self->_convert('?'), $self->_bindtype(@$bind));
1106 sub _render_literal {
1107 my ($self, $literal) = @_;
1108 $self->_assert_bindval_matches_bindtype(@{$literal}[1..$#$literal]);
1113 my ($self, $v) = @_;
1114 my ($op, @args) = @$v;
1115 if (my $r = $self->{render_op}{$op}) {
1116 return $self->$r($op, \@args);
1121 my $op = join(' ', split '_', $op);
1123 my $us = List::Util::first { $op =~ $_->{regex} } @{$self->{special_ops}};
1124 if ($us and @args > 1) {
1125 puke "Special op '${op}' requires first value to be identifier"
1126 unless my ($ident) = map $_->{-ident}, grep ref($_) eq 'HASH', $args[0];
1127 my $k = join(($self->{name_sep}||'.'), @$ident);
1128 local our $Expand_Depth = 1;
1129 return $self->${\($us->{handler})}($k, $op, $args[1]);
1131 if (my $us = List::Util::first { $op =~ $_->{regex} } @{$self->{unary_ops}}) {
1132 return $self->${\($us->{handler})}($op, $args[0]);
1137 return $self->_render_unop_prefix($op, \@args);
1139 return $self->_render_op_multop($op, \@args);
1145 sub _render_op_between {
1146 my ($self, $op, $args) = @_;
1147 my ($left, $low, $high) = @$args;
1148 my ($rhsql, @rhbind) = do {
1150 puke "Single arg to between must be a literal"
1151 unless $low->{-literal};
1154 my ($l, $h) = map [ $self->render_aqt($_) ], $low, $high;
1155 (join(' ', $l->[0], $self->_sqlcase('and'), $h->[0]),
1156 @{$l}[1..$#$l], @{$h}[1..$#$h])
1159 my ($lhsql, @lhbind) = $self->render_aqt($left);
1163 $self->_sqlcase(join ' ', split '_', $op),
1171 my ($self, $op, $args) = @_;
1172 my ($lhs, $rhs) = @$args;
1175 my ($sql, @bind) = $self->render_aqt($_);
1176 push @in_bind, @bind;
1179 my ($lhsql, @lbind) = $self->render_aqt($lhs);
1181 $lhsql.' '.$self->_sqlcase(join ' ', split '_', $op).' ( '
1182 .join(', ', @in_sql)
1188 sub _render_op_andor {
1189 my ($self, $op, $args) = @_;
1190 my @parts = grep length($_->[0]), map [ $self->render_aqt($_) ], @$args;
1191 return '' unless @parts;
1192 return @{$parts[0]} if @parts == 1;
1193 my ($sql, @bind) = $self->_render_op_multop($op, $args);
1194 return '( '.$sql.' )', @bind;
1197 sub _render_op_multop {
1198 my ($self, $op, $args) = @_;
1199 my @parts = grep length($_->[0]), map [ $self->render_aqt($_) ], @$args;
1200 return '' unless @parts;
1201 return @{$parts[0]} if @parts == 1;
1202 my ($final_sql) = join(
1203 ($op eq ',' ? '' : ' ').$self->_sqlcase(join ' ', split '_', $op).' ',
1208 map @{$_}[1..$#$_], @parts
1211 sub _render_op_not {
1212 my ($self, $op, $v) = @_;
1213 my ($sql, @bind) = $self->_render_unop_prefix($op, $v);
1214 return "(${sql})", @bind;
1217 sub _render_unop_prefix {
1218 my ($self, $op, $v) = @_;
1219 my ($expr_sql, @bind) = $self->render_aqt($v->[0]);
1221 my $op_sql = $self->_sqlcase($op); # join ' ', split '_', $op);
1222 return ("${op_sql} ${expr_sql}", @bind);
1225 sub _render_unop_postfix {
1226 my ($self, $op, $v) = @_;
1227 my ($expr_sql, @bind) = $self->render_aqt($v->[0]);
1228 my $op_sql = $self->_sqlcase(join ' ', split '_', $op);
1229 return ($expr_sql.' '.$op_sql, @bind);
1232 # Some databases (SQLite) treat col IN (1, 2) different from
1233 # col IN ( (1, 2) ). Use this to strip all outer parens while
1234 # adding them back in the corresponding method
1235 sub _open_outer_paren {
1236 my ($self, $sql) = @_;
1238 while (my ($inner) = $sql =~ /^ \s* \( (.*) \) \s* $/xs) {
1240 # there are closing parens inside, need the heavy duty machinery
1241 # to reevaluate the extraction starting from $sql (full reevaluation)
1242 if ($inner =~ /\)/) {
1243 require Text::Balanced;
1245 my (undef, $remainder) = do {
1246 # idiotic design - writes to $@ but *DOES NOT* throw exceptions
1248 Text::Balanced::extract_bracketed($sql, '()', qr/\s*/);
1251 # the entire expression needs to be a balanced bracketed thing
1252 # (after an extract no remainder sans trailing space)
1253 last if defined $remainder and $remainder =~ /\S/;
1263 #======================================================================
1265 #======================================================================
1267 sub _expand_order_by {
1268 my ($self, $arg) = @_;
1270 return unless defined($arg) and not (ref($arg) eq 'ARRAY' and !@$arg);
1272 my $expander = sub {
1273 my ($self, $dir, $expr) = @_;
1274 my @to_expand = ref($expr) eq 'ARRAY' ? @$expr : $expr;
1275 foreach my $arg (@to_expand) {
1279 and grep /^-(asc|desc)$/, keys %$arg
1281 puke "ordering direction hash passed to order by must have exactly one key (-asc or -desc)";
1285 defined($dir) ? { -op => [ $dir =~ /^-?(.*)$/ ,=> $_ ] } : $_
1287 map $self->expand_expr($_, -ident),
1288 map ref($_) eq 'ARRAY' ? @$_ : $_, @to_expand;
1289 return undef unless @exp;
1290 return undef if @exp == 1 and not defined($exp[0]);
1291 return +{ -op => [ ',', @exp ] };
1294 local @{$self->{expand}}{qw(-asc -desc)} = (($expander) x 2);
1296 return $self->$expander(undef, $arg);
1300 my ($self, $arg) = @_;
1302 return '' unless defined(my $expanded = $self->_expand_order_by($arg));
1304 my ($sql, @bind) = $self->render_aqt($expanded);
1306 return '' unless length($sql);
1308 my $final_sql = $self->_sqlcase(' order by ').$sql;
1310 return wantarray ? ($final_sql, @bind) : $final_sql;
1313 # _order_by no longer needs to call this so doesn't but DBIC uses it.
1315 sub _order_by_chunks {
1316 my ($self, $arg) = @_;
1318 return () unless defined(my $expanded = $self->_expand_order_by($arg));
1320 return $self->_chunkify_order_by($expanded);
1323 sub _chunkify_order_by {
1324 my ($self, $expanded) = @_;
1326 return grep length, $self->render_aqt($expanded)
1327 if $expanded->{-ident} or @{$expanded->{-literal}||[]} == 1;
1330 if (ref() eq 'HASH' and $_->{-op} and $_->{-op}[0] eq ',') {
1331 my ($comma, @list) = @{$_->{-op}};
1332 return map $self->_chunkify_order_by($_), @list;
1334 return [ $self->render_aqt($_) ];
1338 #======================================================================
1339 # DATASOURCE (FOR NOW, JUST PLAIN TABLE OR LIST OF TABLES)
1340 #======================================================================
1346 $self->_expand_maybe_list_expr($from, -ident)
1351 #======================================================================
1353 #======================================================================
1355 sub _expand_maybe_list_expr {
1356 my ($self, $expr, $default) = @_;
1357 return +{ -op => [ ',',
1358 map $self->expand_expr($_, $default),
1359 ref($expr) eq 'ARRAY' ? @$expr : $expr
1363 # highly optimized, as it's called way too often
1365 # my ($self, $label) = @_;
1367 return '' unless defined $_[1];
1368 return ${$_[1]} if ref($_[1]) eq 'SCALAR';
1369 puke 'Identifier cannot be hashref' if ref($_[1]) eq 'HASH';
1371 unless ($_[0]->{quote_char}) {
1372 if (ref($_[1]) eq 'ARRAY') {
1373 return join($_[0]->{name_sep}||'.', @{$_[1]});
1375 $_[0]->_assert_pass_injection_guard($_[1]);
1380 my $qref = ref $_[0]->{quote_char};
1382 !$qref ? ($_[0]->{quote_char}, $_[0]->{quote_char})
1383 : ($qref eq 'ARRAY') ? @{$_[0]->{quote_char}}
1384 : puke "Unsupported quote_char format: $_[0]->{quote_char}";
1386 my $esc = $_[0]->{escape_char} || $r;
1388 # parts containing * are naturally unquoted
1390 $_[0]->{name_sep}||'',
1394 : do { (my $n = $_) =~ s/(\Q$esc\E|\Q$r\E)/$esc$1/g; $l . $n . $r }
1396 (ref($_[1]) eq 'ARRAY'
1400 ? split (/\Q$_[0]->{name_sep}\E/, $_[1] )
1408 # Conversion, if applicable
1410 #my ($self, $arg) = @_;
1411 if ($_[0]->{convert_where}) {
1412 return $_[0]->_sqlcase($_[0]->{convert_where}) .'(' . $_[1] . ')';
1419 #my ($self, $col, @vals) = @_;
1420 # called often - tighten code
1421 return $_[0]->{bindtype} eq 'columns'
1422 ? map {[$_[1], $_]} @_[2 .. $#_]
1427 # Dies if any element of @bind is not in [colname => value] format
1428 # if bindtype is 'columns'.
1429 sub _assert_bindval_matches_bindtype {
1430 # my ($self, @bind) = @_;
1432 if ($self->{bindtype} eq 'columns') {
1434 if (!defined $_ || ref($_) ne 'ARRAY' || @$_ != 2) {
1435 puke "bindtype 'columns' selected, you need to pass: [column_name => bind_value]"
1441 sub _join_sql_clauses {
1442 my ($self, $logic, $clauses_aref, $bind_aref) = @_;
1444 if (@$clauses_aref > 1) {
1445 my $join = " " . $self->_sqlcase($logic) . " ";
1446 my $sql = '( ' . join($join, @$clauses_aref) . ' )';
1447 return ($sql, @$bind_aref);
1449 elsif (@$clauses_aref) {
1450 return ($clauses_aref->[0], @$bind_aref); # no parentheses
1453 return (); # if no SQL, ignore @$bind_aref
1458 # Fix SQL case, if so requested
1460 # LDNOTE: if $self->{case} is true, then it contains 'lower', so we
1461 # don't touch the argument ... crooked logic, but let's not change it!
1462 return $_[0]->{case} ? $_[1] : uc($_[1]);
1466 #======================================================================
1467 # DISPATCHING FROM REFKIND
1468 #======================================================================
1471 my ($self, $data) = @_;
1473 return 'UNDEF' unless defined $data;
1475 # blessed objects are treated like scalars
1476 my $ref = (Scalar::Util::blessed $data) ? '' : ref $data;
1478 return 'SCALAR' unless $ref;
1481 while ($ref eq 'REF') {
1483 $ref = (Scalar::Util::blessed $data) ? '' : ref $data;
1487 return ($ref||'SCALAR') . ('REF' x $n_steps);
1491 my ($self, $data) = @_;
1492 my @try = ($self->_refkind($data));
1493 push @try, 'SCALAR_or_UNDEF' if $try[0] eq 'SCALAR' || $try[0] eq 'UNDEF';
1494 push @try, 'FALLBACK';
1498 sub _METHOD_FOR_refkind {
1499 my ($self, $meth_prefix, $data) = @_;
1502 for (@{$self->_try_refkind($data)}) {
1503 $method = $self->can($meth_prefix."_".$_)
1507 return $method || puke "cannot dispatch on '$meth_prefix' for ".$self->_refkind($data);
1511 sub _SWITCH_refkind {
1512 my ($self, $data, $dispatch_table) = @_;
1515 for (@{$self->_try_refkind($data)}) {
1516 $coderef = $dispatch_table->{$_}
1520 puke "no dispatch entry for ".$self->_refkind($data)
1529 #======================================================================
1530 # VALUES, GENERATE, AUTOLOAD
1531 #======================================================================
1533 # LDNOTE: original code from nwiger, didn't touch code in that section
1534 # I feel the AUTOLOAD stuff should not be the default, it should
1535 # only be activated on explicit demand by user.
1539 my $data = shift || return;
1540 puke "Argument to ", __PACKAGE__, "->values must be a \\%hash"
1541 unless ref $data eq 'HASH';
1544 foreach my $k (sort keys %$data) {
1545 my $v = $data->{$k};
1546 $self->_SWITCH_refkind($v, {
1548 if ($self->{array_datatypes}) { # array datatype
1549 push @all_bind, $self->_bindtype($k, $v);
1551 else { # literal SQL with bind
1552 my ($sql, @bind) = @$v;
1553 $self->_assert_bindval_matches_bindtype(@bind);
1554 push @all_bind, @bind;
1557 ARRAYREFREF => sub { # literal SQL with bind
1558 my ($sql, @bind) = @${$v};
1559 $self->_assert_bindval_matches_bindtype(@bind);
1560 push @all_bind, @bind;
1562 SCALARREF => sub { # literal SQL without bind
1564 SCALAR_or_UNDEF => sub {
1565 push @all_bind, $self->_bindtype($k, $v);
1576 my(@sql, @sqlq, @sqlv);
1580 if ($ref eq 'HASH') {
1581 for my $k (sort keys %$_) {
1584 my $label = $self->_quote($k);
1585 if ($r eq 'ARRAY') {
1586 # literal SQL with bind
1587 my ($sql, @bind) = @$v;
1588 $self->_assert_bindval_matches_bindtype(@bind);
1589 push @sqlq, "$label = $sql";
1591 } elsif ($r eq 'SCALAR') {
1592 # literal SQL without bind
1593 push @sqlq, "$label = $$v";
1595 push @sqlq, "$label = ?";
1596 push @sqlv, $self->_bindtype($k, $v);
1599 push @sql, $self->_sqlcase('set'), join ', ', @sqlq;
1600 } elsif ($ref eq 'ARRAY') {
1601 # unlike insert(), assume these are ONLY the column names, i.e. for SQL
1604 if ($r eq 'ARRAY') { # literal SQL with bind
1605 my ($sql, @bind) = @$v;
1606 $self->_assert_bindval_matches_bindtype(@bind);
1609 } elsif ($r eq 'SCALAR') { # literal SQL without bind
1610 # embedded literal SQL
1617 push @sql, '(' . join(', ', @sqlq) . ')';
1618 } elsif ($ref eq 'SCALAR') {
1622 # strings get case twiddled
1623 push @sql, $self->_sqlcase($_);
1627 my $sql = join ' ', @sql;
1629 # this is pretty tricky
1630 # if ask for an array, return ($stmt, @bind)
1631 # otherwise, s/?/shift @sqlv/ to put it inline
1633 return ($sql, @sqlv);
1635 1 while $sql =~ s/\?/my $d = shift(@sqlv);
1636 ref $d ? $d->[1] : $d/e;
1645 # This allows us to check for a local, then _form, attr
1647 my($name) = $AUTOLOAD =~ /.*::(.+)/;
1648 return $self->generate($name, @_);
1659 SQL::Abstract - Generate SQL from Perl data structures
1665 my $sql = SQL::Abstract->new;
1667 my($stmt, @bind) = $sql->select($source, \@fields, \%where, $order);
1669 my($stmt, @bind) = $sql->insert($table, \%fieldvals || \@values);
1671 my($stmt, @bind) = $sql->update($table, \%fieldvals, \%where);
1673 my($stmt, @bind) = $sql->delete($table, \%where);
1675 # Then, use these in your DBI statements
1676 my $sth = $dbh->prepare($stmt);
1677 $sth->execute(@bind);
1679 # Just generate the WHERE clause
1680 my($stmt, @bind) = $sql->where(\%where, $order);
1682 # Return values in the same order, for hashed queries
1683 # See PERFORMANCE section for more details
1684 my @bind = $sql->values(\%fieldvals);
1688 This module was inspired by the excellent L<DBIx::Abstract>.
1689 However, in using that module I found that what I really wanted
1690 to do was generate SQL, but still retain complete control over my
1691 statement handles and use the DBI interface. So, I set out to
1692 create an abstract SQL generation module.
1694 While based on the concepts used by L<DBIx::Abstract>, there are
1695 several important differences, especially when it comes to WHERE
1696 clauses. I have modified the concepts used to make the SQL easier
1697 to generate from Perl data structures and, IMO, more intuitive.
1698 The underlying idea is for this module to do what you mean, based
1699 on the data structures you provide it. The big advantage is that
1700 you don't have to modify your code every time your data changes,
1701 as this module figures it out.
1703 To begin with, an SQL INSERT is as easy as just specifying a hash
1704 of C<key=value> pairs:
1707 name => 'Jimbo Bobson',
1708 phone => '123-456-7890',
1709 address => '42 Sister Lane',
1710 city => 'St. Louis',
1711 state => 'Louisiana',
1714 The SQL can then be generated with this:
1716 my($stmt, @bind) = $sql->insert('people', \%data);
1718 Which would give you something like this:
1720 $stmt = "INSERT INTO people
1721 (address, city, name, phone, state)
1722 VALUES (?, ?, ?, ?, ?)";
1723 @bind = ('42 Sister Lane', 'St. Louis', 'Jimbo Bobson',
1724 '123-456-7890', 'Louisiana');
1726 These are then used directly in your DBI code:
1728 my $sth = $dbh->prepare($stmt);
1729 $sth->execute(@bind);
1731 =head2 Inserting and Updating Arrays
1733 If your database has array types (like for example Postgres),
1734 activate the special option C<< array_datatypes => 1 >>
1735 when creating the C<SQL::Abstract> object.
1736 Then you may use an arrayref to insert and update database array types:
1738 my $sql = SQL::Abstract->new(array_datatypes => 1);
1740 planets => [qw/Mercury Venus Earth Mars/]
1743 my($stmt, @bind) = $sql->insert('solar_system', \%data);
1747 $stmt = "INSERT INTO solar_system (planets) VALUES (?)"
1749 @bind = (['Mercury', 'Venus', 'Earth', 'Mars']);
1752 =head2 Inserting and Updating SQL
1754 In order to apply SQL functions to elements of your C<%data> you may
1755 specify a reference to an arrayref for the given hash value. For example,
1756 if you need to execute the Oracle C<to_date> function on a value, you can
1757 say something like this:
1761 date_entered => \[ "to_date(?,'MM/DD/YYYY')", "03/02/2003" ],
1764 The first value in the array is the actual SQL. Any other values are
1765 optional and would be included in the bind values array. This gives
1768 my($stmt, @bind) = $sql->insert('people', \%data);
1770 $stmt = "INSERT INTO people (name, date_entered)
1771 VALUES (?, to_date(?,'MM/DD/YYYY'))";
1772 @bind = ('Bill', '03/02/2003');
1774 An UPDATE is just as easy, all you change is the name of the function:
1776 my($stmt, @bind) = $sql->update('people', \%data);
1778 Notice that your C<%data> isn't touched; the module will generate
1779 the appropriately quirky SQL for you automatically. Usually you'll
1780 want to specify a WHERE clause for your UPDATE, though, which is
1781 where handling C<%where> hashes comes in handy...
1783 =head2 Complex where statements
1785 This module can generate pretty complicated WHERE statements
1786 easily. For example, simple C<key=value> pairs are taken to mean
1787 equality, and if you want to see if a field is within a set
1788 of values, you can use an arrayref. Let's say we wanted to
1789 SELECT some data based on this criteria:
1792 requestor => 'inna',
1793 worker => ['nwiger', 'rcwe', 'sfz'],
1794 status => { '!=', 'completed' }
1797 my($stmt, @bind) = $sql->select('tickets', '*', \%where);
1799 The above would give you something like this:
1801 $stmt = "SELECT * FROM tickets WHERE
1802 ( requestor = ? ) AND ( status != ? )
1803 AND ( worker = ? OR worker = ? OR worker = ? )";
1804 @bind = ('inna', 'completed', 'nwiger', 'rcwe', 'sfz');
1806 Which you could then use in DBI code like so:
1808 my $sth = $dbh->prepare($stmt);
1809 $sth->execute(@bind);
1815 The methods are simple. There's one for every major SQL operation,
1816 and a constructor you use first. The arguments are specified in a
1817 similar order for each method (table, then fields, then a where
1818 clause) to try and simplify things.
1820 =head2 new(option => 'value')
1822 The C<new()> function takes a list of options and values, and returns
1823 a new B<SQL::Abstract> object which can then be used to generate SQL
1824 through the methods below. The options accepted are:
1830 If set to 'lower', then SQL will be generated in all lowercase. By
1831 default SQL is generated in "textbook" case meaning something like:
1833 SELECT a_field FROM a_table WHERE some_field LIKE '%someval%'
1835 Any setting other than 'lower' is ignored.
1839 This determines what the default comparison operator is. By default
1840 it is C<=>, meaning that a hash like this:
1842 %where = (name => 'nwiger', email => 'nate@wiger.org');
1844 Will generate SQL like this:
1846 WHERE name = 'nwiger' AND email = 'nate@wiger.org'
1848 However, you may want loose comparisons by default, so if you set
1849 C<cmp> to C<like> you would get SQL such as:
1851 WHERE name like 'nwiger' AND email like 'nate@wiger.org'
1853 You can also override the comparison on an individual basis - see
1854 the huge section on L</"WHERE CLAUSES"> at the bottom.
1856 =item sqltrue, sqlfalse
1858 Expressions for inserting boolean values within SQL statements.
1859 By default these are C<1=1> and C<1=0>. They are used
1860 by the special operators C<-in> and C<-not_in> for generating
1861 correct SQL even when the argument is an empty array (see below).
1865 This determines the default logical operator for multiple WHERE
1866 statements in arrays or hashes. If absent, the default logic is "or"
1867 for arrays, and "and" for hashes. This means that a WHERE
1871 event_date => {'>=', '2/13/99'},
1872 event_date => {'<=', '4/24/03'},
1875 will generate SQL like this:
1877 WHERE event_date >= '2/13/99' OR event_date <= '4/24/03'
1879 This is probably not what you want given this query, though (look
1880 at the dates). To change the "OR" to an "AND", simply specify:
1882 my $sql = SQL::Abstract->new(logic => 'and');
1884 Which will change the above C<WHERE> to:
1886 WHERE event_date >= '2/13/99' AND event_date <= '4/24/03'
1888 The logic can also be changed locally by inserting
1889 a modifier in front of an arrayref:
1891 @where = (-and => [event_date => {'>=', '2/13/99'},
1892 event_date => {'<=', '4/24/03'} ]);
1894 See the L</"WHERE CLAUSES"> section for explanations.
1898 This will automatically convert comparisons using the specified SQL
1899 function for both column and value. This is mostly used with an argument
1900 of C<upper> or C<lower>, so that the SQL will have the effect of
1901 case-insensitive "searches". For example, this:
1903 $sql = SQL::Abstract->new(convert => 'upper');
1904 %where = (keywords => 'MaKe iT CAse inSeNSItive');
1906 Will turn out the following SQL:
1908 WHERE upper(keywords) like upper('MaKe iT CAse inSeNSItive')
1910 The conversion can be C<upper()>, C<lower()>, or any other SQL function
1911 that can be applied symmetrically to fields (actually B<SQL::Abstract> does
1912 not validate this option; it will just pass through what you specify verbatim).
1916 This is a kludge because many databases suck. For example, you can't
1917 just bind values using DBI's C<execute()> for Oracle C<CLOB> or C<BLOB> fields.
1918 Instead, you have to use C<bind_param()>:
1920 $sth->bind_param(1, 'reg data');
1921 $sth->bind_param(2, $lots, {ora_type => ORA_CLOB});
1923 The problem is, B<SQL::Abstract> will normally just return a C<@bind> array,
1924 which loses track of which field each slot refers to. Fear not.
1926 If you specify C<bindtype> in new, you can determine how C<@bind> is returned.
1927 Currently, you can specify either C<normal> (default) or C<columns>. If you
1928 specify C<columns>, you will get an array that looks like this:
1930 my $sql = SQL::Abstract->new(bindtype => 'columns');
1931 my($stmt, @bind) = $sql->insert(...);
1934 [ 'column1', 'value1' ],
1935 [ 'column2', 'value2' ],
1936 [ 'column3', 'value3' ],
1939 You can then iterate through this manually, using DBI's C<bind_param()>.
1941 $sth->prepare($stmt);
1944 my($col, $data) = @$_;
1945 if ($col eq 'details' || $col eq 'comments') {
1946 $sth->bind_param($i, $data, {ora_type => ORA_CLOB});
1947 } elsif ($col eq 'image') {
1948 $sth->bind_param($i, $data, {ora_type => ORA_BLOB});
1950 $sth->bind_param($i, $data);
1954 $sth->execute; # execute without @bind now
1956 Now, why would you still use B<SQL::Abstract> if you have to do this crap?
1957 Basically, the advantage is still that you don't have to care which fields
1958 are or are not included. You could wrap that above C<for> loop in a simple
1959 sub called C<bind_fields()> or something and reuse it repeatedly. You still
1960 get a layer of abstraction over manual SQL specification.
1962 Note that if you set L</bindtype> to C<columns>, the C<\[ $sql, @bind ]>
1963 construct (see L</Literal SQL with placeholders and bind values (subqueries)>)
1964 will expect the bind values in this format.
1968 This is the character that a table or column name will be quoted
1969 with. By default this is an empty string, but you could set it to
1970 the character C<`>, to generate SQL like this:
1972 SELECT `a_field` FROM `a_table` WHERE `some_field` LIKE '%someval%'
1974 Alternatively, you can supply an array ref of two items, the first being the left
1975 hand quote character, and the second the right hand quote character. For
1976 example, you could supply C<['[',']']> for SQL Server 2000 compliant quotes
1977 that generates SQL like this:
1979 SELECT [a_field] FROM [a_table] WHERE [some_field] LIKE '%someval%'
1981 Quoting is useful if you have tables or columns names that are reserved
1982 words in your database's SQL dialect.
1986 This is the character that will be used to escape L</quote_char>s appearing
1987 in an identifier before it has been quoted.
1989 The parameter default in case of a single L</quote_char> character is the quote
1992 When opening-closing-style quoting is used (L</quote_char> is an arrayref)
1993 this parameter defaults to the B<closing (right)> L</quote_char>. Occurrences
1994 of the B<opening (left)> L</quote_char> within the identifier are currently left
1995 untouched. The default for opening-closing-style quotes may change in future
1996 versions, thus you are B<strongly encouraged> to specify the escape character
2001 This is the character that separates a table and column name. It is
2002 necessary to specify this when the C<quote_char> option is selected,
2003 so that tables and column names can be individually quoted like this:
2005 SELECT `table`.`one_field` FROM `table` WHERE `table`.`other_field` = 1
2007 =item injection_guard
2009 A regular expression C<qr/.../> that is applied to any C<-function> and unquoted
2010 column name specified in a query structure. This is a safety mechanism to avoid
2011 injection attacks when mishandling user input e.g.:
2013 my %condition_as_column_value_pairs = get_values_from_user();
2014 $sqla->select( ... , \%condition_as_column_value_pairs );
2016 If the expression matches an exception is thrown. Note that literal SQL
2017 supplied via C<\'...'> or C<\['...']> is B<not> checked in any way.
2019 Defaults to checking for C<;> and the C<GO> keyword (TransactSQL)
2021 =item array_datatypes
2023 When this option is true, arrayrefs in INSERT or UPDATE are
2024 interpreted as array datatypes and are passed directly
2026 When this option is false, arrayrefs are interpreted
2027 as literal SQL, just like refs to arrayrefs
2028 (but this behavior is for backwards compatibility; when writing
2029 new queries, use the "reference to arrayref" syntax
2035 Takes a reference to a list of "special operators"
2036 to extend the syntax understood by L<SQL::Abstract>.
2037 See section L</"SPECIAL OPERATORS"> for details.
2041 Takes a reference to a list of "unary operators"
2042 to extend the syntax understood by L<SQL::Abstract>.
2043 See section L</"UNARY OPERATORS"> for details.
2049 =head2 insert($table, \@values || \%fieldvals, \%options)
2051 This is the simplest function. You simply give it a table name
2052 and either an arrayref of values or hashref of field/value pairs.
2053 It returns an SQL INSERT statement and a list of bind values.
2054 See the sections on L</"Inserting and Updating Arrays"> and
2055 L</"Inserting and Updating SQL"> for information on how to insert
2056 with those data types.
2058 The optional C<\%options> hash reference may contain additional
2059 options to generate the insert SQL. Currently supported options
2066 Takes either a scalar of raw SQL fields, or an array reference of
2067 field names, and adds on an SQL C<RETURNING> statement at the end.
2068 This allows you to return data generated by the insert statement
2069 (such as row IDs) without performing another C<SELECT> statement.
2070 Note, however, this is not part of the SQL standard and may not
2071 be supported by all database engines.
2075 =head2 update($table, \%fieldvals, \%where, \%options)
2077 This takes a table, hashref of field/value pairs, and an optional
2078 hashref L<WHERE clause|/WHERE CLAUSES>. It returns an SQL UPDATE function and a list
2080 See the sections on L</"Inserting and Updating Arrays"> and
2081 L</"Inserting and Updating SQL"> for information on how to insert
2082 with those data types.
2084 The optional C<\%options> hash reference may contain additional
2085 options to generate the update SQL. Currently supported options
2092 See the C<returning> option to
2093 L<insert|/insert($table, \@values || \%fieldvals, \%options)>.
2097 =head2 select($source, $fields, $where, $order)
2099 This returns a SQL SELECT statement and associated list of bind values, as
2100 specified by the arguments:
2106 Specification of the 'FROM' part of the statement.
2107 The argument can be either a plain scalar (interpreted as a table
2108 name, will be quoted), or an arrayref (interpreted as a list
2109 of table names, joined by commas, quoted), or a scalarref
2110 (literal SQL, not quoted).
2114 Specification of the list of fields to retrieve from
2116 The argument can be either an arrayref (interpreted as a list
2117 of field names, will be joined by commas and quoted), or a
2118 plain scalar (literal SQL, not quoted).
2119 Please observe that this API is not as flexible as that of
2120 the first argument C<$source>, for backwards compatibility reasons.
2124 Optional argument to specify the WHERE part of the query.
2125 The argument is most often a hashref, but can also be
2126 an arrayref or plain scalar --
2127 see section L<WHERE clause|/"WHERE CLAUSES"> for details.
2131 Optional argument to specify the ORDER BY part of the query.
2132 The argument can be a scalar, a hashref or an arrayref
2133 -- see section L<ORDER BY clause|/"ORDER BY CLAUSES">
2139 =head2 delete($table, \%where, \%options)
2141 This takes a table name and optional hashref L<WHERE clause|/WHERE CLAUSES>.
2142 It returns an SQL DELETE statement and list of bind values.
2144 The optional C<\%options> hash reference may contain additional
2145 options to generate the delete SQL. Currently supported options
2152 See the C<returning> option to
2153 L<insert|/insert($table, \@values || \%fieldvals, \%options)>.
2157 =head2 where(\%where, $order)
2159 This is used to generate just the WHERE clause. For example,
2160 if you have an arbitrary data structure and know what the
2161 rest of your SQL is going to look like, but want an easy way
2162 to produce a WHERE clause, use this. It returns an SQL WHERE
2163 clause and list of bind values.
2166 =head2 values(\%data)
2168 This just returns the values from the hash C<%data>, in the same
2169 order that would be returned from any of the other above queries.
2170 Using this allows you to markedly speed up your queries if you
2171 are affecting lots of rows. See below under the L</"PERFORMANCE"> section.
2173 =head2 generate($any, 'number', $of, \@data, $struct, \%types)
2175 Warning: This is an experimental method and subject to change.
2177 This returns arbitrarily generated SQL. It's a really basic shortcut.
2178 It will return two different things, depending on return context:
2180 my($stmt, @bind) = $sql->generate('create table', \$table, \@fields);
2181 my $stmt_and_val = $sql->generate('create table', \$table, \@fields);
2183 These would return the following:
2185 # First calling form
2186 $stmt = "CREATE TABLE test (?, ?)";
2187 @bind = (field1, field2);
2189 # Second calling form
2190 $stmt_and_val = "CREATE TABLE test (field1, field2)";
2192 Depending on what you're trying to do, it's up to you to choose the correct
2193 format. In this example, the second form is what you would want.
2197 $sql->generate('alter session', { nls_date_format => 'MM/YY' });
2201 ALTER SESSION SET nls_date_format = 'MM/YY'
2203 You get the idea. Strings get their case twiddled, but everything
2204 else remains verbatim.
2206 =head1 EXPORTABLE FUNCTIONS
2208 =head2 is_plain_value
2210 Determines if the supplied argument is a plain value as understood by this
2215 =item * The value is C<undef>
2217 =item * The value is a non-reference
2219 =item * The value is an object with stringification overloading
2221 =item * The value is of the form C<< { -value => $anything } >>
2225 On failure returns C<undef>, on success returns a B<scalar> reference
2226 to the original supplied argument.
2232 The stringification overloading detection is rather advanced: it takes
2233 into consideration not only the presence of a C<""> overload, but if that
2234 fails also checks for enabled
2235 L<autogenerated versions of C<"">|overload/Magic Autogeneration>, based
2236 on either C<0+> or C<bool>.
2238 Unfortunately testing in the field indicates that this
2239 detection B<< may tickle a latent bug in perl versions before 5.018 >>,
2240 but only when very large numbers of stringifying objects are involved.
2241 At the time of writing ( Sep 2014 ) there is no clear explanation of
2242 the direct cause, nor is there a manageably small test case that reliably
2243 reproduces the problem.
2245 If you encounter any of the following exceptions in B<random places within
2246 your application stack> - this module may be to blame:
2248 Operation "ne": no method found,
2249 left argument in overloaded package <something>,
2250 right argument in overloaded package <something>
2254 Stub found while resolving method "???" overloading """" in package <something>
2256 If you fall victim to the above - please attempt to reduce the problem
2257 to something that could be sent to the L<SQL::Abstract developers
2258 |DBIx::Class/GETTING HELP/SUPPORT>
2259 (either publicly or privately). As a workaround in the meantime you can
2260 set C<$ENV{SQLA_ISVALUE_IGNORE_AUTOGENERATED_STRINGIFICATION}> to a true
2261 value, which will most likely eliminate your problem (at the expense of
2262 not being able to properly detect exotic forms of stringification).
2264 This notice and environment variable will be removed in a future version,
2265 as soon as the underlying problem is found and a reliable workaround is
2270 =head2 is_literal_value
2272 Determines if the supplied argument is a literal value as understood by this
2277 =item * C<\$sql_string>
2279 =item * C<\[ $sql_string, @bind_values ]>
2283 On failure returns C<undef>, on success returns an B<array> reference
2284 containing the unpacked version of the supplied literal SQL and bind values.
2286 =head1 WHERE CLAUSES
2290 This module uses a variation on the idea from L<DBIx::Abstract>. It
2291 is B<NOT>, repeat I<not> 100% compatible. B<The main logic of this
2292 module is that things in arrays are OR'ed, and things in hashes
2295 The easiest way to explain is to show lots of examples. After
2296 each C<%where> hash shown, it is assumed you used:
2298 my($stmt, @bind) = $sql->where(\%where);
2300 However, note that the C<%where> hash can be used directly in any
2301 of the other functions as well, as described above.
2303 =head2 Key-value pairs
2305 So, let's get started. To begin, a simple hash:
2309 status => 'completed'
2312 Is converted to SQL C<key = val> statements:
2314 $stmt = "WHERE user = ? AND status = ?";
2315 @bind = ('nwiger', 'completed');
2317 One common thing I end up doing is having a list of values that
2318 a field can be in. To do this, simply specify a list inside of
2323 status => ['assigned', 'in-progress', 'pending'];
2326 This simple code will create the following:
2328 $stmt = "WHERE user = ? AND ( status = ? OR status = ? OR status = ? )";
2329 @bind = ('nwiger', 'assigned', 'in-progress', 'pending');
2331 A field associated to an empty arrayref will be considered a
2332 logical false and will generate 0=1.
2334 =head2 Tests for NULL values
2336 If the value part is C<undef> then this is converted to SQL <IS NULL>
2345 $stmt = "WHERE user = ? AND status IS NULL";
2348 To test if a column IS NOT NULL:
2352 status => { '!=', undef },
2355 =head2 Specific comparison operators
2357 If you want to specify a different type of operator for your comparison,
2358 you can use a hashref for a given column:
2362 status => { '!=', 'completed' }
2365 Which would generate:
2367 $stmt = "WHERE user = ? AND status != ?";
2368 @bind = ('nwiger', 'completed');
2370 To test against multiple values, just enclose the values in an arrayref:
2372 status => { '=', ['assigned', 'in-progress', 'pending'] };
2374 Which would give you:
2376 "WHERE status = ? OR status = ? OR status = ?"
2379 The hashref can also contain multiple pairs, in which case it is expanded
2380 into an C<AND> of its elements:
2384 status => { '!=', 'completed', -not_like => 'pending%' }
2387 # Or more dynamically, like from a form
2388 $where{user} = 'nwiger';
2389 $where{status}{'!='} = 'completed';
2390 $where{status}{'-not_like'} = 'pending%';
2392 # Both generate this
2393 $stmt = "WHERE user = ? AND status != ? AND status NOT LIKE ?";
2394 @bind = ('nwiger', 'completed', 'pending%');
2397 To get an OR instead, you can combine it with the arrayref idea:
2401 priority => [ { '=', 2 }, { '>', 5 } ]
2404 Which would generate:
2406 $stmt = "WHERE ( priority = ? OR priority > ? ) AND user = ?";
2407 @bind = ('2', '5', 'nwiger');
2409 If you want to include literal SQL (with or without bind values), just use a
2410 scalar reference or reference to an arrayref as the value:
2413 date_entered => { '>' => \["to_date(?, 'MM/DD/YYYY')", "11/26/2008"] },
2414 date_expires => { '<' => \"now()" }
2417 Which would generate:
2419 $stmt = "WHERE date_entered > to_date(?, 'MM/DD/YYYY') AND date_expires < now()";
2420 @bind = ('11/26/2008');
2423 =head2 Logic and nesting operators
2425 In the example above,
2426 there is a subtle trap if you want to say something like
2427 this (notice the C<AND>):
2429 WHERE priority != ? AND priority != ?
2431 Because, in Perl you I<can't> do this:
2433 priority => { '!=' => 2, '!=' => 1 }
2435 As the second C<!=> key will obliterate the first. The solution
2436 is to use the special C<-modifier> form inside an arrayref:
2438 priority => [ -and => {'!=', 2},
2442 Normally, these would be joined by C<OR>, but the modifier tells it
2443 to use C<AND> instead. (Hint: You can use this in conjunction with the
2444 C<logic> option to C<new()> in order to change the way your queries
2445 work by default.) B<Important:> Note that the C<-modifier> goes
2446 B<INSIDE> the arrayref, as an extra first element. This will
2447 B<NOT> do what you think it might:
2449 priority => -and => [{'!=', 2}, {'!=', 1}] # WRONG!
2451 Here is a quick list of equivalencies, since there is some overlap:
2454 status => {'!=', 'completed', 'not like', 'pending%' }
2455 status => [ -and => {'!=', 'completed'}, {'not like', 'pending%'}]
2458 status => {'=', ['assigned', 'in-progress']}
2459 status => [ -or => {'=', 'assigned'}, {'=', 'in-progress'}]
2460 status => [ {'=', 'assigned'}, {'=', 'in-progress'} ]
2464 =head2 Special operators: IN, BETWEEN, etc.
2466 You can also use the hashref format to compare a list of fields using the
2467 C<IN> comparison operator, by specifying the list as an arrayref:
2470 status => 'completed',
2471 reportid => { -in => [567, 2335, 2] }
2474 Which would generate:
2476 $stmt = "WHERE status = ? AND reportid IN (?,?,?)";
2477 @bind = ('completed', '567', '2335', '2');
2479 The reverse operator C<-not_in> generates SQL C<NOT IN> and is used in
2482 If the argument to C<-in> is an empty array, 'sqlfalse' is generated
2483 (by default: C<1=0>). Similarly, C<< -not_in => [] >> generates
2484 'sqltrue' (by default: C<1=1>).
2486 In addition to the array you can supply a chunk of literal sql or
2487 literal sql with bind:
2490 customer => { -in => \[
2491 'SELECT cust_id FROM cust WHERE balance > ?',
2494 status => { -in => \'SELECT status_codes FROM states' },
2500 customer IN ( SELECT cust_id FROM cust WHERE balance > ? )
2501 AND status IN ( SELECT status_codes FROM states )
2505 Finally, if the argument to C<-in> is not a reference, it will be
2506 treated as a single-element array.
2508 Another pair of operators is C<-between> and C<-not_between>,
2509 used with an arrayref of two values:
2513 completion_date => {
2514 -not_between => ['2002-10-01', '2003-02-06']
2520 WHERE user = ? AND completion_date NOT BETWEEN ( ? AND ? )
2522 Just like with C<-in> all plausible combinations of literal SQL
2526 start0 => { -between => [ 1, 2 ] },
2527 start1 => { -between => \["? AND ?", 1, 2] },
2528 start2 => { -between => \"lower(x) AND upper(y)" },
2529 start3 => { -between => [
2531 \["upper(?)", 'stuff' ],
2538 ( start0 BETWEEN ? AND ? )
2539 AND ( start1 BETWEEN ? AND ? )
2540 AND ( start2 BETWEEN lower(x) AND upper(y) )
2541 AND ( start3 BETWEEN lower(x) AND upper(?) )
2543 @bind = (1, 2, 1, 2, 'stuff');
2546 These are the two builtin "special operators"; but the
2547 list can be expanded: see section L</"SPECIAL OPERATORS"> below.
2549 =head2 Unary operators: bool
2551 If you wish to test against boolean columns or functions within your
2552 database you can use the C<-bool> and C<-not_bool> operators. For
2553 example to test the column C<is_user> being true and the column
2554 C<is_enabled> being false you would use:-
2558 -not_bool => 'is_enabled',
2563 WHERE is_user AND NOT is_enabled
2565 If a more complex combination is required, testing more conditions,
2566 then you should use the and/or operators:-
2571 -not_bool => { two=> { -rlike => 'bar' } },
2572 -not_bool => { three => [ { '=', 2 }, { '>', 5 } ] },
2583 (NOT ( three = ? OR three > ? ))
2586 =head2 Nested conditions, -and/-or prefixes
2588 So far, we've seen how multiple conditions are joined with a top-level
2589 C<AND>. We can change this by putting the different conditions we want in
2590 hashes and then putting those hashes in an array. For example:
2595 status => { -like => ['pending%', 'dispatched'] },
2599 status => 'unassigned',
2603 This data structure would create the following:
2605 $stmt = "WHERE ( user = ? AND ( status LIKE ? OR status LIKE ? ) )
2606 OR ( user = ? AND status = ? ) )";
2607 @bind = ('nwiger', 'pending', 'dispatched', 'robot', 'unassigned');
2610 Clauses in hashrefs or arrayrefs can be prefixed with an C<-and> or C<-or>
2611 to change the logic inside:
2617 -and => [ workhrs => {'>', 20}, geo => 'ASIA' ],
2618 -or => { workhrs => {'<', 50}, geo => 'EURO' },
2625 $stmt = "WHERE ( user = ?
2626 AND ( ( workhrs > ? AND geo = ? )
2627 OR ( workhrs < ? OR geo = ? ) ) )";
2628 @bind = ('nwiger', '20', 'ASIA', '50', 'EURO');
2630 =head3 Algebraic inconsistency, for historical reasons
2632 C<Important note>: when connecting several conditions, the C<-and->|C<-or>
2633 operator goes C<outside> of the nested structure; whereas when connecting
2634 several constraints on one column, the C<-and> operator goes
2635 C<inside> the arrayref. Here is an example combining both features:
2638 -and => [a => 1, b => 2],
2639 -or => [c => 3, d => 4],
2640 e => [-and => {-like => 'foo%'}, {-like => '%bar'} ]
2645 WHERE ( ( ( a = ? AND b = ? )
2646 OR ( c = ? OR d = ? )
2647 OR ( e LIKE ? AND e LIKE ? ) ) )
2649 This difference in syntax is unfortunate but must be preserved for
2650 historical reasons. So be careful: the two examples below would
2651 seem algebraically equivalent, but they are not
2654 { -like => 'foo%' },
2655 { -like => '%bar' },
2657 # yields: WHERE ( ( col LIKE ? AND col LIKE ? ) )
2660 { col => { -like => 'foo%' } },
2661 { col => { -like => '%bar' } },
2663 # yields: WHERE ( ( col LIKE ? OR col LIKE ? ) )
2666 =head2 Literal SQL and value type operators
2668 The basic premise of SQL::Abstract is that in WHERE specifications the "left
2669 side" is a column name and the "right side" is a value (normally rendered as
2670 a placeholder). This holds true for both hashrefs and arrayref pairs as you
2671 see in the L</WHERE CLAUSES> examples above. Sometimes it is necessary to
2672 alter this behavior. There are several ways of doing so.
2676 This is a virtual operator that signals the string to its right side is an
2677 identifier (a column name) and not a value. For example to compare two
2678 columns you would write:
2681 priority => { '<', 2 },
2682 requestor => { -ident => 'submitter' },
2687 $stmt = "WHERE priority < ? AND requestor = submitter";
2690 If you are maintaining legacy code you may see a different construct as
2691 described in L</Deprecated usage of Literal SQL>, please use C<-ident> in new
2696 This is a virtual operator that signals that the construct to its right side
2697 is a value to be passed to DBI. This is for example necessary when you want
2698 to write a where clause against an array (for RDBMS that support such
2699 datatypes). For example:
2702 array => { -value => [1, 2, 3] }
2707 $stmt = 'WHERE array = ?';
2708 @bind = ([1, 2, 3]);
2710 Note that if you were to simply say:
2716 the result would probably not be what you wanted:
2718 $stmt = 'WHERE array = ? OR array = ? OR array = ?';
2723 Finally, sometimes only literal SQL will do. To include a random snippet
2724 of SQL verbatim, you specify it as a scalar reference. Consider this only
2725 as a last resort. Usually there is a better way. For example:
2728 priority => { '<', 2 },
2729 requestor => { -in => \'(SELECT name FROM hitmen)' },
2734 $stmt = "WHERE priority < ? AND requestor IN (SELECT name FROM hitmen)"
2737 Note that in this example, you only get one bind parameter back, since
2738 the verbatim SQL is passed as part of the statement.
2742 Never use untrusted input as a literal SQL argument - this is a massive
2743 security risk (there is no way to check literal snippets for SQL
2744 injections and other nastyness). If you need to deal with untrusted input
2745 use literal SQL with placeholders as described next.
2747 =head3 Literal SQL with placeholders and bind values (subqueries)
2749 If the literal SQL to be inserted has placeholders and bind values,
2750 use a reference to an arrayref (yes this is a double reference --
2751 not so common, but perfectly legal Perl). For example, to find a date
2752 in Postgres you can use something like this:
2755 date_column => \[ "= date '2008-09-30' - ?::integer", 10 ]
2760 $stmt = "WHERE ( date_column = date '2008-09-30' - ?::integer )"
2763 Note that you must pass the bind values in the same format as they are returned
2764 by L<where|/where(\%where, $order)>. This means that if you set L</bindtype>
2765 to C<columns>, you must provide the bind values in the
2766 C<< [ column_meta => value ] >> format, where C<column_meta> is an opaque
2767 scalar value; most commonly the column name, but you can use any scalar value
2768 (including references and blessed references), L<SQL::Abstract> will simply
2769 pass it through intact. So if C<bindtype> is set to C<columns> the above
2770 example will look like:
2773 date_column => \[ "= date '2008-09-30' - ?::integer", [ {} => 10 ] ]
2776 Literal SQL is especially useful for nesting parenthesized clauses in the
2777 main SQL query. Here is a first example:
2779 my ($sub_stmt, @sub_bind) = ("SELECT c1 FROM t1 WHERE c2 < ? AND c3 LIKE ?",
2783 bar => \["IN ($sub_stmt)" => @sub_bind],
2788 $stmt = "WHERE (foo = ? AND bar IN (SELECT c1 FROM t1
2789 WHERE c2 < ? AND c3 LIKE ?))";
2790 @bind = (1234, 100, "foo%");
2792 Other subquery operators, like for example C<"E<gt> ALL"> or C<"NOT IN">,
2793 are expressed in the same way. Of course the C<$sub_stmt> and
2794 its associated bind values can be generated through a former call
2797 my ($sub_stmt, @sub_bind)
2798 = $sql->select("t1", "c1", {c2 => {"<" => 100},
2799 c3 => {-like => "foo%"}});
2802 bar => \["> ALL ($sub_stmt)" => @sub_bind],
2805 In the examples above, the subquery was used as an operator on a column;
2806 but the same principle also applies for a clause within the main C<%where>
2807 hash, like an EXISTS subquery:
2809 my ($sub_stmt, @sub_bind)
2810 = $sql->select("t1", "*", {c1 => 1, c2 => \"> t0.c0"});
2811 my %where = ( -and => [
2813 \["EXISTS ($sub_stmt)" => @sub_bind],
2818 $stmt = "WHERE (foo = ? AND EXISTS (SELECT * FROM t1
2819 WHERE c1 = ? AND c2 > t0.c0))";
2823 Observe that the condition on C<c2> in the subquery refers to
2824 column C<t0.c0> of the main query: this is I<not> a bind
2825 value, so we have to express it through a scalar ref.
2826 Writing C<< c2 => {">" => "t0.c0"} >> would have generated
2827 C<< c2 > ? >> with bind value C<"t0.c0"> ... not exactly
2828 what we wanted here.
2830 Finally, here is an example where a subquery is used
2831 for expressing unary negation:
2833 my ($sub_stmt, @sub_bind)
2834 = $sql->where({age => [{"<" => 10}, {">" => 20}]});
2835 $sub_stmt =~ s/^ where //i; # don't want "WHERE" in the subclause
2837 lname => {like => '%son%'},
2838 \["NOT ($sub_stmt)" => @sub_bind],
2843 $stmt = "lname LIKE ? AND NOT ( age < ? OR age > ? )"
2844 @bind = ('%son%', 10, 20)
2846 =head3 Deprecated usage of Literal SQL
2848 Below are some examples of archaic use of literal SQL. It is shown only as
2849 reference for those who deal with legacy code. Each example has a much
2850 better, cleaner and safer alternative that users should opt for in new code.
2856 my %where = ( requestor => \'IS NOT NULL' )
2858 $stmt = "WHERE requestor IS NOT NULL"
2860 This used to be the way of generating NULL comparisons, before the handling
2861 of C<undef> got formalized. For new code please use the superior syntax as
2862 described in L</Tests for NULL values>.
2866 my %where = ( requestor => \'= submitter' )
2868 $stmt = "WHERE requestor = submitter"
2870 This used to be the only way to compare columns. Use the superior L</-ident>
2871 method for all new code. For example an identifier declared in such a way
2872 will be properly quoted if L</quote_char> is properly set, while the legacy
2873 form will remain as supplied.
2877 my %where = ( is_ready => \"", completed => { '>', '2012-12-21' } )
2879 $stmt = "WHERE completed > ? AND is_ready"
2880 @bind = ('2012-12-21')
2882 Using an empty string literal used to be the only way to express a boolean.
2883 For all new code please use the much more readable
2884 L<-bool|/Unary operators: bool> operator.
2890 These pages could go on for a while, since the nesting of the data
2891 structures this module can handle are pretty much unlimited (the
2892 module implements the C<WHERE> expansion as a recursive function
2893 internally). Your best bet is to "play around" with the module a
2894 little to see how the data structures behave, and choose the best
2895 format for your data based on that.
2897 And of course, all the values above will probably be replaced with
2898 variables gotten from forms or the command line. After all, if you
2899 knew everything ahead of time, you wouldn't have to worry about
2900 dynamically-generating SQL and could just hardwire it into your
2903 =head1 ORDER BY CLAUSES
2905 Some functions take an order by clause. This can either be a scalar (just a
2906 column name), a hashref of C<< { -desc => 'col' } >> or C<< { -asc => 'col' }
2907 >>, a scalarref, an arrayref-ref, or an arrayref of any of the previous
2910 Given | Will Generate
2911 ---------------------------------------------------------------
2913 'colA' | ORDER BY colA
2915 [qw/colA colB/] | ORDER BY colA, colB
2917 {-asc => 'colA'} | ORDER BY colA ASC
2919 {-desc => 'colB'} | ORDER BY colB DESC
2921 ['colA', {-asc => 'colB'}] | ORDER BY colA, colB ASC
2923 { -asc => [qw/colA colB/] } | ORDER BY colA ASC, colB ASC
2925 \'colA DESC' | ORDER BY colA DESC
2927 \[ 'FUNC(colA, ?)', $x ] | ORDER BY FUNC(colA, ?)
2928 | /* ...with $x bound to ? */
2931 { -asc => 'colA' }, | colA ASC,
2932 { -desc => [qw/colB/] }, | colB DESC,
2933 { -asc => [qw/colC colD/] },| colC ASC, colD ASC,
2934 \'colE DESC', | colE DESC,
2935 \[ 'FUNC(colF, ?)', $x ], | FUNC(colF, ?)
2936 ] | /* ...with $x bound to ? */
2937 ===============================================================
2941 =head1 SPECIAL OPERATORS
2943 my $sqlmaker = SQL::Abstract->new(special_ops => [
2947 my ($self, $field, $op, $arg) = @_;
2953 handler => 'method_name',
2957 A "special operator" is a SQL syntactic clause that can be
2958 applied to a field, instead of a usual binary operator.
2961 WHERE field IN (?, ?, ?)
2962 WHERE field BETWEEN ? AND ?
2963 WHERE MATCH(field) AGAINST (?, ?)
2965 Special operators IN and BETWEEN are fairly standard and therefore
2966 are builtin within C<SQL::Abstract> (as the overridable methods
2967 C<_where_field_IN> and C<_where_field_BETWEEN>). For other operators,
2968 like the MATCH .. AGAINST example above which is specific to MySQL,
2969 you can write your own operator handlers - supply a C<special_ops>
2970 argument to the C<new> method. That argument takes an arrayref of
2971 operator definitions; each operator definition is a hashref with two
2978 the regular expression to match the operator
2982 Either a coderef or a plain scalar method name. In both cases
2983 the expected return is C<< ($sql, @bind) >>.
2985 When supplied with a method name, it is simply called on the
2986 L<SQL::Abstract> object as:
2988 $self->$method_name($field, $op, $arg)
2992 $field is the LHS of the operator
2993 $op is the part that matched the handler regex
2996 When supplied with a coderef, it is called as:
2998 $coderef->($self, $field, $op, $arg)
3003 For example, here is an implementation
3004 of the MATCH .. AGAINST syntax for MySQL
3006 my $sqlmaker = SQL::Abstract->new(special_ops => [
3008 # special op for MySql MATCH (field) AGAINST(word1, word2, ...)
3009 {regex => qr/^match$/i,
3011 my ($self, $field, $op, $arg) = @_;
3012 $arg = [$arg] if not ref $arg;
3013 my $label = $self->_quote($field);
3014 my ($placeholder) = $self->_convert('?');
3015 my $placeholders = join ", ", (($placeholder) x @$arg);
3016 my $sql = $self->_sqlcase('match') . " ($label) "
3017 . $self->_sqlcase('against') . " ($placeholders) ";
3018 my @bind = $self->_bindtype($field, @$arg);
3019 return ($sql, @bind);
3026 =head1 UNARY OPERATORS
3028 my $sqlmaker = SQL::Abstract->new(unary_ops => [
3032 my ($self, $op, $arg) = @_;
3038 handler => 'method_name',
3042 A "unary operator" is a SQL syntactic clause that can be
3043 applied to a field - the operator goes before the field
3045 You can write your own operator handlers - supply a C<unary_ops>
3046 argument to the C<new> method. That argument takes an arrayref of
3047 operator definitions; each operator definition is a hashref with two
3054 the regular expression to match the operator
3058 Either a coderef or a plain scalar method name. In both cases
3059 the expected return is C<< $sql >>.
3061 When supplied with a method name, it is simply called on the
3062 L<SQL::Abstract> object as:
3064 $self->$method_name($op, $arg)
3068 $op is the part that matched the handler regex
3069 $arg is the RHS or argument of the operator
3071 When supplied with a coderef, it is called as:
3073 $coderef->($self, $op, $arg)
3081 Thanks to some benchmarking by Mark Stosberg, it turns out that
3082 this module is many orders of magnitude faster than using C<DBIx::Abstract>.
3083 I must admit this wasn't an intentional design issue, but it's a
3084 byproduct of the fact that you get to control your C<DBI> handles
3087 To maximize performance, use a code snippet like the following:
3089 # prepare a statement handle using the first row
3090 # and then reuse it for the rest of the rows
3092 for my $href (@array_of_hashrefs) {
3093 $stmt ||= $sql->insert('table', $href);
3094 $sth ||= $dbh->prepare($stmt);
3095 $sth->execute($sql->values($href));
3098 The reason this works is because the keys in your C<$href> are sorted
3099 internally by B<SQL::Abstract>. Thus, as long as your data retains
3100 the same structure, you only have to generate the SQL the first time
3101 around. On subsequent queries, simply use the C<values> function provided
3102 by this module to return your values in the correct order.
3104 However this depends on the values having the same type - if, for
3105 example, the values of a where clause may either have values
3106 (resulting in sql of the form C<column = ?> with a single bind
3107 value), or alternatively the values might be C<undef> (resulting in
3108 sql of the form C<column IS NULL> with no bind value) then the
3109 caching technique suggested will not work.
3113 If you use my C<CGI::FormBuilder> module at all, you'll hopefully
3114 really like this part (I do, at least). Building up a complex query
3115 can be as simple as the following:
3122 use CGI::FormBuilder;
3125 my $form = CGI::FormBuilder->new(...);
3126 my $sql = SQL::Abstract->new;
3128 if ($form->submitted) {
3129 my $field = $form->field;
3130 my $id = delete $field->{id};
3131 my($stmt, @bind) = $sql->update('table', $field, {id => $id});
3134 Of course, you would still have to connect using C<DBI> to run the
3135 query, but the point is that if you make your form look like your
3136 table, the actual query script can be extremely simplistic.
3138 If you're B<REALLY> lazy (I am), check out C<HTML::QuickTable> for
3139 a fast interface to returning and formatting data. I frequently
3140 use these three modules together to write complex database query
3141 apps in under 50 lines.
3143 =head1 HOW TO CONTRIBUTE
3145 Contributions are always welcome, in all usable forms (we especially
3146 welcome documentation improvements). The delivery methods include git-
3147 or unified-diff formatted patches, GitHub pull requests, or plain bug
3148 reports either via RT or the Mailing list. Contributors are generally
3149 granted full access to the official repository after their first several
3150 patches pass successful review.
3152 This project is maintained in a git repository. The code and related tools are
3153 accessible at the following locations:
3157 =item * Official repo: L<git://git.shadowcat.co.uk/dbsrgits/SQL-Abstract.git>
3159 =item * Official gitweb: L<http://git.shadowcat.co.uk/gitweb/gitweb.cgi?p=dbsrgits/SQL-Abstract.git>
3161 =item * GitHub mirror: L<https://github.com/dbsrgits/sql-abstract>
3163 =item * Authorized committers: L<ssh://dbsrgits@git.shadowcat.co.uk/SQL-Abstract.git>
3169 Version 1.50 was a major internal refactoring of C<SQL::Abstract>.
3170 Great care has been taken to preserve the I<published> behavior
3171 documented in previous versions in the 1.* family; however,
3172 some features that were previously undocumented, or behaved
3173 differently from the documentation, had to be changed in order
3174 to clarify the semantics. Hence, client code that was relying
3175 on some dark areas of C<SQL::Abstract> v1.*
3176 B<might behave differently> in v1.50.
3178 The main changes are:
3184 support for literal SQL through the C<< \ [ $sql, @bind ] >> syntax.
3188 support for the { operator => \"..." } construct (to embed literal SQL)
3192 support for the { operator => \["...", @bind] } construct (to embed literal SQL with bind values)
3196 optional support for L<array datatypes|/"Inserting and Updating Arrays">
3200 defensive programming: check arguments
3204 fixed bug with global logic, which was previously implemented
3205 through global variables yielding side-effects. Prior versions would
3206 interpret C<< [ {cond1, cond2}, [cond3, cond4] ] >>
3207 as C<< "(cond1 AND cond2) OR (cond3 AND cond4)" >>.
3208 Now this is interpreted
3209 as C<< "(cond1 AND cond2) OR (cond3 OR cond4)" >>.
3214 fixed semantics of _bindtype on array args
3218 dropped the C<_anoncopy> of the %where tree. No longer necessary,
3219 we just avoid shifting arrays within that tree.
3223 dropped the C<_modlogic> function
3227 =head1 ACKNOWLEDGEMENTS
3229 There are a number of individuals that have really helped out with
3230 this module. Unfortunately, most of them submitted bugs via CPAN
3231 so I have no idea who they are! But the people I do know are:
3233 Ash Berlin (order_by hash term support)
3234 Matt Trout (DBIx::Class support)
3235 Mark Stosberg (benchmarking)
3236 Chas Owens (initial "IN" operator support)
3237 Philip Collins (per-field SQL functions)
3238 Eric Kolve (hashref "AND" support)
3239 Mike Fragassi (enhancements to "BETWEEN" and "LIKE")
3240 Dan Kubb (support for "quote_char" and "name_sep")
3241 Guillermo Roditi (patch to cleanup "IN" and "BETWEEN", fix and tests for _order_by)
3242 Laurent Dami (internal refactoring, extensible list of special operators, literal SQL)
3243 Norbert Buchmuller (support for literal SQL in hashpair, misc. fixes & tests)
3244 Peter Rabbitson (rewrite of SQLA::Test, misc. fixes & tests)
3245 Oliver Charles (support for "RETURNING" after "INSERT")
3251 L<DBIx::Class>, L<DBIx::Abstract>, L<CGI::FormBuilder>, L<HTML::QuickTable>.
3255 Copyright (c) 2001-2007 Nathan Wiger <nwiger@cpan.org>. All Rights Reserved.
3257 This module is actively maintained by Matt Trout <mst@shadowcatsystems.co.uk>
3259 For support, your best bet is to try the C<DBIx::Class> users mailing list.
3260 While not an official support venue, C<DBIx::Class> makes heavy use of
3261 C<SQL::Abstract>, and as such list members there are very familiar with
3262 how to create queries.
3266 This module is free software; you may copy this under the same
3267 terms as perl itself (either the GNU General Public License or
3268 the Artistic License)