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 $opt{special_ops} ||= [];
171 if ($class->isa('DBIx::Class::SQLMaker')) {
172 $opt{warn_once_on_nest} = 1;
173 $opt{disable_old_special_ops} = 1;
177 $opt{unary_ops} ||= [];
179 # rudimentary sanity-check for user supplied bits treated as functions/operators
180 # If a purported function matches this regular expression, an exception is thrown.
181 # Literal SQL is *NOT* subject to this check, only functions (and column names
182 # when quoting is not in effect)
185 # need to guard against ()'s in column names too, but this will break tons of
186 # hacks... ideas anyone?
187 $opt{injection_guard} ||= qr/
193 $opt{expand_unary} = {};
196 -not => '_expand_not',
197 -bool => '_expand_bool',
198 -and => '_expand_op_andor',
199 -or => '_expand_op_andor',
200 -nest => '_expand_nest',
201 -bind => '_expand_bind',
203 -not_in => '_expand_in',
204 -row => '_expand_row',
205 -between => '_expand_between',
206 -not_between => '_expand_between',
208 (map +($_ => '_expand_op_is'), ('-is', '-is_not')),
209 -ident => '_expand_ident',
210 -value => '_expand_value',
214 'between' => '_expand_between',
215 'not_between' => '_expand_between',
216 'in' => '_expand_in',
217 'not_in' => '_expand_in',
218 'nest' => '_expand_nest',
219 (map +($_ => '_expand_op_andor'), ('and', 'or')),
220 (map +($_ => '_expand_op_is'), ('is', 'is_not')),
221 'ident' => '_expand_ident',
222 'value' => '_expand_value',
226 (map +("-$_", "_render_$_"), qw(op func bind ident literal row)),
231 (map +($_ => '_render_op_between'), 'between', 'not_between'),
232 (map +($_ => '_render_op_in'), 'in', 'not_in'),
233 (map +($_ => '_render_unop_postfix'),
234 'is_null', 'is_not_null', 'asc', 'desc',
236 (not => '_render_op_not'),
237 (map +($_ => '_render_op_andor'), qw(and or)),
238 ',' => '_render_op_multop',
241 return bless \%opt, $class;
244 sub sqltrue { +{ -literal => [ $_[0]->{sqltrue} ] } }
245 sub sqlfalse { +{ -literal => [ $_[0]->{sqlfalse} ] } }
247 sub _assert_pass_injection_guard {
248 if ($_[1] =~ $_[0]->{injection_guard}) {
249 my $class = ref $_[0];
250 puke "Possible SQL injection attempt '$_[1]'. If this is indeed a part of the "
251 . "desired SQL use literal SQL ( \'...' or \[ '...' ] ) or supply your own "
252 . "{injection_guard} attribute to ${class}->new()"
257 #======================================================================
259 #======================================================================
263 my $table = $self->_table(shift);
264 my $data = shift || return;
267 my $method = $self->_METHOD_FOR_refkind("_insert", $data);
268 my ($sql, @bind) = $self->$method($data);
269 $sql = join " ", $self->_sqlcase('insert into'), $table, $sql;
271 if ($options->{returning}) {
272 my ($s, @b) = $self->_insert_returning($options);
277 return wantarray ? ($sql, @bind) : $sql;
280 # So that subclasses can override INSERT ... RETURNING separately from
281 # UPDATE and DELETE (e.g. DBIx::Class::SQLMaker::Oracle does this)
282 sub _insert_returning { shift->_returning(@_) }
285 my ($self, $options) = @_;
287 my $f = $options->{returning};
289 my ($sql, @bind) = $self->render_aqt(
290 $self->_expand_maybe_list_expr($f, -ident)
293 ? $self->_sqlcase(' returning ') . $sql
294 : ($self->_sqlcase(' returning ').$sql, @bind);
297 sub _insert_HASHREF { # explicit list of fields and then values
298 my ($self, $data) = @_;
300 my @fields = sort keys %$data;
302 my ($sql, @bind) = $self->_insert_values($data);
305 $_ = $self->_quote($_) foreach @fields;
306 $sql = "( ".join(", ", @fields).") ".$sql;
308 return ($sql, @bind);
311 sub _insert_ARRAYREF { # just generate values(?,?) part (no list of fields)
312 my ($self, $data) = @_;
314 # no names (arrayref) so can't generate bindtype
315 $self->{bindtype} ne 'columns'
316 or belch "can't do 'columns' bindtype when called with arrayref";
318 my (@values, @all_bind);
319 foreach my $value (@$data) {
320 my ($values, @bind) = $self->_insert_value(undef, $value);
321 push @values, $values;
322 push @all_bind, @bind;
324 my $sql = $self->_sqlcase('values')." ( ".join(", ", @values)." )";
325 return ($sql, @all_bind);
328 sub _insert_ARRAYREFREF { # literal SQL with bind
329 my ($self, $data) = @_;
331 my ($sql, @bind) = @${$data};
332 $self->_assert_bindval_matches_bindtype(@bind);
334 return ($sql, @bind);
338 sub _insert_SCALARREF { # literal SQL without bind
339 my ($self, $data) = @_;
345 my ($self, $data) = @_;
347 my (@values, @all_bind);
348 foreach my $column (sort keys %$data) {
349 my ($values, @bind) = $self->_insert_value($column, $data->{$column});
350 push @values, $values;
351 push @all_bind, @bind;
353 my $sql = $self->_sqlcase('values')." ( ".join(", ", @values)." )";
354 return ($sql, @all_bind);
358 my ($self, $column, $v) = @_;
360 return $self->render_aqt(
361 $self->_expand_insert_value($column, $v)
365 sub _expand_insert_value {
366 my ($self, $column, $v) = @_;
368 if (ref($v) eq 'ARRAY') {
369 if ($self->{array_datatypes}) {
370 return +{ -bind => [ $column, $v ] };
372 my ($sql, @bind) = @$v;
373 $self->_assert_bindval_matches_bindtype(@bind);
374 return +{ -literal => $v };
376 if (ref($v) eq 'HASH') {
377 if (grep !/^-/, keys %$v) {
378 belch "HASH ref as bind value in insert is not supported";
379 return +{ -bind => [ $column, $v ] };
383 return +{ -bind => [ $column, undef ] };
385 local our $Cur_Col_Meta = $column;
386 return $self->expand_expr($v);
391 #======================================================================
393 #======================================================================
398 my $table = $self->_table(shift);
399 my $data = shift || return;
403 # first build the 'SET' part of the sql statement
404 puke "Unsupported data type specified to \$sql->update"
405 unless ref $data eq 'HASH';
407 my ($sql, @all_bind) = $self->_update_set_values($data);
408 $sql = $self->_sqlcase('update ') . $table . $self->_sqlcase(' set ')
412 my($where_sql, @where_bind) = $self->where($where);
414 push @all_bind, @where_bind;
417 if ($options->{returning}) {
418 my ($returning_sql, @returning_bind) = $self->_update_returning($options);
419 $sql .= $returning_sql;
420 push @all_bind, @returning_bind;
423 return wantarray ? ($sql, @all_bind) : $sql;
426 sub _update_set_values {
427 my ($self, $data) = @_;
429 return $self->render_aqt(
430 $self->_expand_update_set_values($data),
434 sub _expand_update_set_values {
435 my ($self, $data) = @_;
436 $self->_expand_maybe_list_expr( [
439 $set = { -bind => $_ } unless defined $set;
440 +{ -op => [ '=', $self->_expand_ident(-ident => $k), $set ] };
446 ? ($self->{array_datatypes}
447 ? [ $k, +{ -bind => [ $k, $v ] } ]
448 : [ $k, +{ -literal => $v } ])
450 local our $Cur_Col_Meta = $k;
451 [ $k, $self->_expand_expr($v) ]
458 # So that subclasses can override UPDATE ... RETURNING separately from
460 sub _update_returning { shift->_returning(@_) }
464 #======================================================================
466 #======================================================================
471 my $table = $self->_table(shift);
472 my $fields = shift || '*';
476 my ($fields_sql, @bind) = $self->_select_fields($fields);
478 my ($where_sql, @where_bind) = $self->where($where, $order);
479 push @bind, @where_bind;
481 my $sql = join(' ', $self->_sqlcase('select'), $fields_sql,
482 $self->_sqlcase('from'), $table)
485 return wantarray ? ($sql, @bind) : $sql;
489 my ($self, $fields) = @_;
490 return $fields unless ref($fields);
491 return $self->render_aqt(
492 $self->_expand_maybe_list_expr($fields, '-ident')
496 #======================================================================
498 #======================================================================
503 my $table = $self->_table(shift);
507 my($where_sql, @bind) = $self->where($where);
508 my $sql = $self->_sqlcase('delete from ') . $table . $where_sql;
510 if ($options->{returning}) {
511 my ($returning_sql, @returning_bind) = $self->_delete_returning($options);
512 $sql .= $returning_sql;
513 push @bind, @returning_bind;
516 return wantarray ? ($sql, @bind) : $sql;
519 # So that subclasses can override DELETE ... RETURNING separately from
521 sub _delete_returning { shift->_returning(@_) }
525 #======================================================================
527 #======================================================================
531 # Finally, a separate routine just to handle WHERE clauses
533 my ($self, $where, $order) = @_;
535 local $self->{convert_where} = $self->{convert};
538 my ($sql, @bind) = defined($where)
539 ? $self->_recurse_where($where)
541 $sql = (defined $sql and length $sql) ? $self->_sqlcase(' where ') . "( $sql )" : '';
545 my ($order_sql, @order_bind) = $self->_order_by($order);
547 push @bind, @order_bind;
550 return wantarray ? ($sql, @bind) : $sql;
553 { our $Default_Scalar_To = -value }
556 my ($self, $expr, $default_scalar_to) = @_;
557 local our $Default_Scalar_To = $default_scalar_to if $default_scalar_to;
558 $self->_expand_expr($expr);
562 my ($self, $aqt) = @_;
563 my ($k, $v, @rest) = %$aqt;
565 if (my $meth = $self->{render}{$k}) {
566 return $self->$meth($v);
568 die "notreached: $k";
572 my ($self, $expr) = @_;
573 $self->render_aqt($self->expand_expr($expr));
577 my ($self, $raw) = @_;
578 s/^-(?=.)//, s/\s+/_/g for my $op = lc $raw;
583 my ($self, $expr) = @_;
584 our $Expand_Depth ||= 0; local $Expand_Depth = $Expand_Depth + 1;
585 return undef unless defined($expr);
586 if (ref($expr) eq 'HASH') {
587 return undef unless my $kc = keys %$expr;
589 return $self->_expand_op_andor(-and => $expr);
591 my ($key, $value) = %$expr;
592 if ($key =~ /^-/ and $key =~ s/ [_\s]? \d+ $//x ) {
593 belch 'Use of [and|or|nest]_N modifiers is deprecated and will be removed in SQLA v2.0. '
594 . "You probably wanted ...-and => [ $key => COND1, $key => COND2 ... ]";
596 if (my $exp = $self->{expand}{$key}) {
597 return $self->$exp($key, $value);
599 return $self->_expand_expr_hashpair($key, $value);
601 if (ref($expr) eq 'ARRAY') {
602 my $logic = '-'.lc($self->{logic});
603 return $self->_expand_op_andor($logic, $expr);
605 if (my $literal = is_literal_value($expr)) {
606 return +{ -literal => $literal };
608 if (!ref($expr) or Scalar::Util::blessed($expr)) {
609 return $self->_expand_expr_scalar($expr);
614 sub _expand_expr_hashpair {
615 my ($self, $k, $v) = @_;
616 unless (defined($k) and length($k)) {
617 if (defined($k) and my $literal = is_literal_value($v)) {
618 belch 'Hash-pairs consisting of an empty string with a literal are deprecated, and will be removed in 2.0: use -and => [ $literal ] instead';
619 return { -literal => $literal };
621 puke "Supplying an empty left hand side argument is not supported";
624 return $self->_expand_expr_hashpair_op($k, $v);
626 return $self->_expand_expr_hashpair_ident($k, $v);
629 sub _expand_expr_hashpair_ident {
630 my ($self, $k, $v) = @_;
632 local our $Cur_Col_Meta = $k;
634 # hash with multiple or no elements is andor
636 if (ref($v) eq 'HASH' and keys %$v != 1) {
637 return $self->_expand_op_andor(-and => $v, $k);
640 # undef needs to be re-sent with cmp to achieve IS/IS NOT NULL
642 if (is_undef_value($v)) {
643 return $self->_expand_expr_hashpair_cmp($k => undef);
646 # scalars and objects get expanded as whatever requested or values
648 if (!ref($v) or Scalar::Util::blessed($v)) {
649 return $self->_expand_expr_hashpair_scalar($k, $v);
652 # single key hashref is a hashtriple
654 if (ref($v) eq 'HASH') {
655 return $self->_expand_expr_hashtriple($k, %$v);
658 # arrayref needs re-engineering over the elements
660 if (ref($v) eq 'ARRAY') {
661 return $self->sqlfalse unless @$v;
662 $self->_debug("ARRAY($k) means distribute over elements");
664 $v->[0] =~ /^-(and|or)$/i
665 ? shift(@{$v = [ @$v ]})
666 : '-'.lc($self->{logic} || 'OR')
668 return $self->_expand_op_andor(
673 if (my $literal = is_literal_value($v)) {
675 belch 'Hash-pairs consisting of an empty string with a literal are deprecated, and will be removed in 2.0: use -and => [ $literal ] instead';
678 my ($sql, @bind) = @$literal;
679 if ($self->{bindtype} eq 'columns') {
681 $self->_assert_bindval_matches_bindtype($_);
684 return +{ -literal => [ $self->_quote($k).' '.$sql, @bind ] };
689 sub _expand_expr_scalar {
690 my ($self, $expr) = @_;
692 return $self->_expand_expr({ (our $Default_Scalar_To) => $expr });
695 sub _expand_expr_hashpair_scalar {
696 my ($self, $k, $v) = @_;
698 return $self->_expand_expr_hashpair_cmp(
699 $k, $self->_expand_expr_scalar($v),
703 sub _expand_expr_hashpair_op {
704 my ($self, $k, $v) = @_;
706 $self->_assert_pass_injection_guard($k =~ /\A-(.*)\Z/s);
708 my $op = $self->_normalize_op($k);
710 # Ops prefixed with -not_ get converted
712 if (my ($rest) = $op =~/^not_(.*)$/) {
715 $self->_expand_expr({ "-${rest}", $v })
721 my $op = join(' ', split '_', $op);
723 # the old special op system requires illegality for top-level use
726 (our $Expand_Depth) == 1
728 List::Util::first { $op =~ $_->{regex} } @{$self->{special_ops}}
730 $self->{disable_old_special_ops}
731 and List::Util::first { $op =~ $_->{regex} } @BUILTIN_SPECIAL_OPS
735 puke "Illegal use of top-level '-$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, $k) = @_;
880 return $self->_expand_expr_hashpair_cmp(
881 $k, { -ident => $body }
883 unless (defined($body) or (ref($body) and ref($body) eq 'ARRAY')) {
884 puke "$op requires a single plain scalar argument (a quotable identifier) or an arrayref of identifier parts";
886 my @parts = map split(/\Q${\($self->{name_sep}||'.')}\E/, $_),
887 ref($body) ? @$body : $body;
888 return { -ident => $parts[-1] } if $self->{_dequalify_idents};
889 unless ($self->{quote_char}) {
890 $self->_assert_pass_injection_guard($_) for @parts;
892 return +{ -ident => \@parts };
896 return $_[0]->_expand_expr_hashpair_cmp(
897 $_[3], { -value => $_[2] },
899 +{ -bind => [ our $Cur_Col_Meta, $_[2] ] };
903 +{ -op => [ 'not', $_[0]->_expand_expr($_[2]) ] };
907 my ($self, $node, $args) = @_;
908 +{ $node => [ map $self->expand_expr($_), @$args ] };
912 my ($self, $node, $args) = @_;
913 my ($op, @opargs) = @$args;
914 +{ $node => [ $op, map $self->expand_expr($_), @opargs ] };
918 my ($self, undef, $v) = @_;
920 return $self->_expand_expr($v);
922 puke "-bool => undef not supported" unless defined($v);
923 return $self->_expand_ident(-ident => $v);
926 sub _expand_op_andor {
927 my ($self, $logic, $v, $k) = @_;
929 $v = [ map +{ $k, $_ },
931 ? (map +{ $_ => $v->{$_} }, sort keys %$v)
935 my ($logop) = $logic =~ /^-?(.*)$/;
936 if (ref($v) eq 'HASH') {
937 return undef unless keys %$v;
940 map $self->_expand_expr({ $_ => $v->{$_} }),
944 if (ref($v) eq 'ARRAY') {
945 $logop eq 'and' or $logop eq 'or' or puke "unknown logic: $logop";
948 (ref($_) eq 'ARRAY' and @$_)
949 or (ref($_) eq 'HASH' and %$_)
955 while (my ($el) = splice @expr, 0, 1) {
956 puke "Supplying an empty left hand side argument is not supported in array-pairs"
957 unless defined($el) and length($el);
958 my $elref = ref($el);
960 local our $Expand_Depth = 0;
961 push(@res, grep defined, $self->_expand_expr({ $el, shift(@expr) }));
962 } elsif ($elref eq 'ARRAY') {
963 push(@res, grep defined, $self->_expand_expr($el)) if @$el;
964 } elsif (my $l = is_literal_value($el)) {
965 push @res, { -literal => $l };
966 } elsif ($elref eq 'HASH') {
967 local our $Expand_Depth = 0;
968 push @res, grep defined, $self->_expand_expr($el) if %$el;
974 # return $res[0] if @res == 1;
975 return { -op => [ $logop, @res ] };
981 my ($self, $op, $vv, $k) = @_;
983 ($k, $vv) = @$vv unless defined $k;
984 puke "$op can only take undef as argument"
988 and exists($vv->{-value})
989 and !defined($vv->{-value})
991 return +{ -op => [ $op.'_null', $self->expand_expr($k, -ident) ] };
994 sub _expand_between {
995 my ($self, $op, $vv, $k) = @_;
997 $k = shift @{$vv = [ @$vv ]} unless defined $k;
998 my @rhs = map $self->_expand_expr($_),
999 ref($vv) eq 'ARRAY' ? @$vv : $vv;
1001 (@rhs == 1 and ref($rhs[0]) eq 'HASH' and $rhs[0]->{-literal})
1003 (@rhs == 2 and defined($rhs[0]) and defined($rhs[1]))
1005 puke "Operator '${\uc($op)}' requires either an arrayref with two defined values or expressions, or a single literal scalarref/arrayref-ref";
1009 $self->expand_expr(ref($k) ? $k : { -ident => $k }),
1015 my ($self, $raw, $vv, $k) = @_;
1016 $k = shift @{$vv = [ @$vv ]} unless defined $k;
1017 my $op = $self->_normalize_op($raw);
1018 if (my $literal = is_literal_value($vv)) {
1019 my ($sql, @bind) = @$literal;
1020 my $opened_sql = $self->_open_outer_paren($sql);
1022 $op, $self->expand_expr($k, -ident),
1023 [ { -literal => [ $opened_sql, @bind ] } ]
1027 'SQL::Abstract before v1.75 used to generate incorrect SQL when the '
1028 . "-${\uc($op)} operator was given an undef-containing list: !!!AUDIT YOUR CODE "
1029 . 'AND DATA!!! (the upcoming Data::Query-based version of SQL::Abstract '
1030 . 'will emit the logically correct SQL instead of raising this exception)'
1032 puke("Argument passed to the '${\uc($op)}' operator can not be undefined")
1034 my @rhs = map $self->expand_expr($_, -value),
1035 map { defined($_) ? $_: puke($undef_err) }
1036 (ref($vv) eq 'ARRAY' ? @$vv : $vv);
1037 return $self->${\($op =~ /^not/ ? 'sqltrue' : 'sqlfalse')} unless @rhs;
1041 $self->expand_expr($k, -ident),
1047 my ($self, $op, $v) = @_;
1048 # DBIx::Class requires a nest warning to be emitted once but the private
1049 # method it overrode to do so no longer exists
1050 if ($self->{warn_once_on_nest}) {
1051 unless (our $Nest_Warned) {
1053 "-nest in search conditions is deprecated, you most probably wanted:\n"
1054 .q|{..., -and => [ \%cond0, \@cond1, \'cond2', \[ 'cond3', [ col => bind ] ], etc. ], ... }|
1059 return $self->_expand_expr($v);
1063 my ($self, $op, $bind) = @_;
1064 return { $op => $bind };
1067 sub _recurse_where {
1068 my ($self, $where, $logic) = @_;
1070 # Special case: top level simple string treated as literal
1072 my $where_exp = (ref($where)
1073 ? $self->_expand_expr($where, $logic)
1074 : { -literal => [ $where ] });
1076 # dispatch expanded expression
1078 my ($sql, @bind) = defined($where_exp) ? $self->render_aqt($where_exp) : (undef);
1079 # DBIx::Class used to call _recurse_where in scalar context
1080 # something else might too...
1082 return ($sql, @bind);
1085 belch "Calling _recurse_where in scalar context is deprecated and will go away before 2.0";
1091 my ($self, $ident) = @_;
1093 return $self->_convert($self->_quote($ident));
1097 my ($self, $values) = @_;
1098 my ($sql, @bind) = $self->_render_op([ ',', @$values ]);
1099 return "($sql)", @bind;
1103 my ($self, $rest) = @_;
1104 my ($func, @args) = @$rest;
1108 push @arg_sql, shift @x;
1110 } map [ $self->render_aqt($_) ], @args;
1111 return ($self->_sqlcase($func).'('.join(', ', @arg_sql).')', @bind);
1115 my ($self, $bind) = @_;
1116 return ($self->_convert('?'), $self->_bindtype(@$bind));
1119 sub _render_literal {
1120 my ($self, $literal) = @_;
1121 $self->_assert_bindval_matches_bindtype(@{$literal}[1..$#$literal]);
1126 my ($self, $v) = @_;
1127 my ($op, @args) = @$v;
1128 if (my $r = $self->{render_op}{$op}) {
1129 return $self->$r($op, \@args);
1134 my $op = join(' ', split '_', $op);
1136 my $us = List::Util::first { $op =~ $_->{regex} } @{$self->{special_ops}};
1137 if ($us and @args > 1) {
1138 puke "Special op '${op}' requires first value to be identifier"
1139 unless my ($ident) = map $_->{-ident}, grep ref($_) eq 'HASH', $args[0];
1140 my $k = join(($self->{name_sep}||'.'), @$ident);
1141 local our $Expand_Depth = 1;
1142 return $self->${\($us->{handler})}($k, $op, $args[1]);
1144 if (my $us = List::Util::first { $op =~ $_->{regex} } @{$self->{unary_ops}}) {
1145 return $self->${\($us->{handler})}($op, $args[0]);
1150 return $self->_render_unop_prefix($op, \@args);
1152 return $self->_render_op_multop($op, \@args);
1158 sub _render_op_between {
1159 my ($self, $op, $args) = @_;
1160 my ($left, $low, $high) = @$args;
1161 my ($rhsql, @rhbind) = do {
1163 puke "Single arg to between must be a literal"
1164 unless $low->{-literal};
1167 my ($l, $h) = map [ $self->render_aqt($_) ], $low, $high;
1168 (join(' ', $l->[0], $self->_sqlcase('and'), $h->[0]),
1169 @{$l}[1..$#$l], @{$h}[1..$#$h])
1172 my ($lhsql, @lhbind) = $self->render_aqt($left);
1176 $self->_sqlcase(join ' ', split '_', $op),
1184 my ($self, $op, $args) = @_;
1185 my ($lhs, $rhs) = @$args;
1188 my ($sql, @bind) = $self->render_aqt($_);
1189 push @in_bind, @bind;
1192 my ($lhsql, @lbind) = $self->render_aqt($lhs);
1194 $lhsql.' '.$self->_sqlcase(join ' ', split '_', $op).' ( '
1195 .join(', ', @in_sql)
1201 sub _render_op_andor {
1202 my ($self, $op, $args) = @_;
1203 my @parts = grep length($_->[0]), map [ $self->render_aqt($_) ], @$args;
1204 return '' unless @parts;
1205 return @{$parts[0]} if @parts == 1;
1206 my ($sql, @bind) = $self->_render_op_multop($op, $args);
1207 return '( '.$sql.' )', @bind;
1210 sub _render_op_multop {
1211 my ($self, $op, $args) = @_;
1212 my @parts = grep length($_->[0]), map [ $self->render_aqt($_) ], @$args;
1213 return '' unless @parts;
1214 return @{$parts[0]} if @parts == 1;
1215 my ($final_sql) = join(
1216 ($op eq ',' ? '' : ' ').$self->_sqlcase(join ' ', split '_', $op).' ',
1221 map @{$_}[1..$#$_], @parts
1224 sub _render_op_not {
1225 my ($self, $op, $v) = @_;
1226 my ($sql, @bind) = $self->_render_unop_prefix($op, $v);
1227 return "(${sql})", @bind;
1230 sub _render_unop_prefix {
1231 my ($self, $op, $v) = @_;
1232 my ($expr_sql, @bind) = $self->render_aqt($v->[0]);
1234 my $op_sql = $self->_sqlcase($op); # join ' ', split '_', $op);
1235 return ("${op_sql} ${expr_sql}", @bind);
1238 sub _render_unop_postfix {
1239 my ($self, $op, $v) = @_;
1240 my ($expr_sql, @bind) = $self->render_aqt($v->[0]);
1241 my $op_sql = $self->_sqlcase(join ' ', split '_', $op);
1242 return ($expr_sql.' '.$op_sql, @bind);
1245 # Some databases (SQLite) treat col IN (1, 2) different from
1246 # col IN ( (1, 2) ). Use this to strip all outer parens while
1247 # adding them back in the corresponding method
1248 sub _open_outer_paren {
1249 my ($self, $sql) = @_;
1251 while (my ($inner) = $sql =~ /^ \s* \( (.*) \) \s* $/xs) {
1253 # there are closing parens inside, need the heavy duty machinery
1254 # to reevaluate the extraction starting from $sql (full reevaluation)
1255 if ($inner =~ /\)/) {
1256 require Text::Balanced;
1258 my (undef, $remainder) = do {
1259 # idiotic design - writes to $@ but *DOES NOT* throw exceptions
1261 Text::Balanced::extract_bracketed($sql, '()', qr/\s*/);
1264 # the entire expression needs to be a balanced bracketed thing
1265 # (after an extract no remainder sans trailing space)
1266 last if defined $remainder and $remainder =~ /\S/;
1276 #======================================================================
1278 #======================================================================
1280 sub _expand_order_by {
1281 my ($self, $arg) = @_;
1283 return unless defined($arg) and not (ref($arg) eq 'ARRAY' and !@$arg);
1285 my $expander = sub {
1286 my ($self, $dir, $expr) = @_;
1287 my @to_expand = ref($expr) eq 'ARRAY' ? @$expr : $expr;
1288 foreach my $arg (@to_expand) {
1292 and grep /^-(asc|desc)$/, keys %$arg
1294 puke "ordering direction hash passed to order by must have exactly one key (-asc or -desc)";
1298 defined($dir) ? { -op => [ $dir =~ /^-?(.*)$/ ,=> $_ ] } : $_
1300 map $self->expand_expr($_, -ident),
1301 map ref($_) eq 'ARRAY' ? @$_ : $_, @to_expand;
1302 return undef unless @exp;
1303 return undef if @exp == 1 and not defined($exp[0]);
1304 return +{ -op => [ ',', @exp ] };
1307 local @{$self->{expand}}{qw(-asc -desc)} = (($expander) x 2);
1309 return $self->$expander(undef, $arg);
1313 my ($self, $arg) = @_;
1315 return '' unless defined(my $expanded = $self->_expand_order_by($arg));
1317 my ($sql, @bind) = $self->render_aqt($expanded);
1319 return '' unless length($sql);
1321 my $final_sql = $self->_sqlcase(' order by ').$sql;
1323 return wantarray ? ($final_sql, @bind) : $final_sql;
1326 # _order_by no longer needs to call this so doesn't but DBIC uses it.
1328 sub _order_by_chunks {
1329 my ($self, $arg) = @_;
1331 return () unless defined(my $expanded = $self->_expand_order_by($arg));
1333 return $self->_chunkify_order_by($expanded);
1336 sub _chunkify_order_by {
1337 my ($self, $expanded) = @_;
1339 return grep length, $self->render_aqt($expanded)
1340 if $expanded->{-ident} or @{$expanded->{-literal}||[]} == 1;
1343 if (ref() eq 'HASH' and $_->{-op} and $_->{-op}[0] eq ',') {
1344 my ($comma, @list) = @{$_->{-op}};
1345 return map $self->_chunkify_order_by($_), @list;
1347 return [ $self->render_aqt($_) ];
1351 #======================================================================
1352 # DATASOURCE (FOR NOW, JUST PLAIN TABLE OR LIST OF TABLES)
1353 #======================================================================
1359 $self->_expand_maybe_list_expr($from, -ident)
1364 #======================================================================
1366 #======================================================================
1368 sub _expand_maybe_list_expr {
1369 my ($self, $expr, $default) = @_;
1370 return +{ -op => [ ',',
1371 map $self->expand_expr($_, $default),
1372 ref($expr) eq 'ARRAY' ? @$expr : $expr
1376 # highly optimized, as it's called way too often
1378 # my ($self, $label) = @_;
1380 return '' unless defined $_[1];
1381 return ${$_[1]} if ref($_[1]) eq 'SCALAR';
1382 puke 'Identifier cannot be hashref' if ref($_[1]) eq 'HASH';
1384 unless ($_[0]->{quote_char}) {
1385 if (ref($_[1]) eq 'ARRAY') {
1386 return join($_[0]->{name_sep}||'.', @{$_[1]});
1388 $_[0]->_assert_pass_injection_guard($_[1]);
1393 my $qref = ref $_[0]->{quote_char};
1395 !$qref ? ($_[0]->{quote_char}, $_[0]->{quote_char})
1396 : ($qref eq 'ARRAY') ? @{$_[0]->{quote_char}}
1397 : puke "Unsupported quote_char format: $_[0]->{quote_char}";
1399 my $esc = $_[0]->{escape_char} || $r;
1401 # parts containing * are naturally unquoted
1403 $_[0]->{name_sep}||'',
1407 : do { (my $n = $_) =~ s/(\Q$esc\E|\Q$r\E)/$esc$1/g; $l . $n . $r }
1409 (ref($_[1]) eq 'ARRAY'
1413 ? split (/\Q$_[0]->{name_sep}\E/, $_[1] )
1421 # Conversion, if applicable
1423 #my ($self, $arg) = @_;
1424 if ($_[0]->{convert_where}) {
1425 return $_[0]->_sqlcase($_[0]->{convert_where}) .'(' . $_[1] . ')';
1432 #my ($self, $col, @vals) = @_;
1433 # called often - tighten code
1434 return $_[0]->{bindtype} eq 'columns'
1435 ? map {[$_[1], $_]} @_[2 .. $#_]
1440 # Dies if any element of @bind is not in [colname => value] format
1441 # if bindtype is 'columns'.
1442 sub _assert_bindval_matches_bindtype {
1443 # my ($self, @bind) = @_;
1445 if ($self->{bindtype} eq 'columns') {
1447 if (!defined $_ || ref($_) ne 'ARRAY' || @$_ != 2) {
1448 puke "bindtype 'columns' selected, you need to pass: [column_name => bind_value]"
1454 sub _join_sql_clauses {
1455 my ($self, $logic, $clauses_aref, $bind_aref) = @_;
1457 if (@$clauses_aref > 1) {
1458 my $join = " " . $self->_sqlcase($logic) . " ";
1459 my $sql = '( ' . join($join, @$clauses_aref) . ' )';
1460 return ($sql, @$bind_aref);
1462 elsif (@$clauses_aref) {
1463 return ($clauses_aref->[0], @$bind_aref); # no parentheses
1466 return (); # if no SQL, ignore @$bind_aref
1471 # Fix SQL case, if so requested
1473 # LDNOTE: if $self->{case} is true, then it contains 'lower', so we
1474 # don't touch the argument ... crooked logic, but let's not change it!
1475 return $_[0]->{case} ? $_[1] : uc($_[1]);
1479 #======================================================================
1480 # DISPATCHING FROM REFKIND
1481 #======================================================================
1484 my ($self, $data) = @_;
1486 return 'UNDEF' unless defined $data;
1488 # blessed objects are treated like scalars
1489 my $ref = (Scalar::Util::blessed $data) ? '' : ref $data;
1491 return 'SCALAR' unless $ref;
1494 while ($ref eq 'REF') {
1496 $ref = (Scalar::Util::blessed $data) ? '' : ref $data;
1500 return ($ref||'SCALAR') . ('REF' x $n_steps);
1504 my ($self, $data) = @_;
1505 my @try = ($self->_refkind($data));
1506 push @try, 'SCALAR_or_UNDEF' if $try[0] eq 'SCALAR' || $try[0] eq 'UNDEF';
1507 push @try, 'FALLBACK';
1511 sub _METHOD_FOR_refkind {
1512 my ($self, $meth_prefix, $data) = @_;
1515 for (@{$self->_try_refkind($data)}) {
1516 $method = $self->can($meth_prefix."_".$_)
1520 return $method || puke "cannot dispatch on '$meth_prefix' for ".$self->_refkind($data);
1524 sub _SWITCH_refkind {
1525 my ($self, $data, $dispatch_table) = @_;
1528 for (@{$self->_try_refkind($data)}) {
1529 $coderef = $dispatch_table->{$_}
1533 puke "no dispatch entry for ".$self->_refkind($data)
1542 #======================================================================
1543 # VALUES, GENERATE, AUTOLOAD
1544 #======================================================================
1546 # LDNOTE: original code from nwiger, didn't touch code in that section
1547 # I feel the AUTOLOAD stuff should not be the default, it should
1548 # only be activated on explicit demand by user.
1552 my $data = shift || return;
1553 puke "Argument to ", __PACKAGE__, "->values must be a \\%hash"
1554 unless ref $data eq 'HASH';
1557 foreach my $k (sort keys %$data) {
1558 my $v = $data->{$k};
1559 $self->_SWITCH_refkind($v, {
1561 if ($self->{array_datatypes}) { # array datatype
1562 push @all_bind, $self->_bindtype($k, $v);
1564 else { # literal SQL with bind
1565 my ($sql, @bind) = @$v;
1566 $self->_assert_bindval_matches_bindtype(@bind);
1567 push @all_bind, @bind;
1570 ARRAYREFREF => sub { # literal SQL with bind
1571 my ($sql, @bind) = @${$v};
1572 $self->_assert_bindval_matches_bindtype(@bind);
1573 push @all_bind, @bind;
1575 SCALARREF => sub { # literal SQL without bind
1577 SCALAR_or_UNDEF => sub {
1578 push @all_bind, $self->_bindtype($k, $v);
1589 my(@sql, @sqlq, @sqlv);
1593 if ($ref eq 'HASH') {
1594 for my $k (sort keys %$_) {
1597 my $label = $self->_quote($k);
1598 if ($r eq 'ARRAY') {
1599 # literal SQL with bind
1600 my ($sql, @bind) = @$v;
1601 $self->_assert_bindval_matches_bindtype(@bind);
1602 push @sqlq, "$label = $sql";
1604 } elsif ($r eq 'SCALAR') {
1605 # literal SQL without bind
1606 push @sqlq, "$label = $$v";
1608 push @sqlq, "$label = ?";
1609 push @sqlv, $self->_bindtype($k, $v);
1612 push @sql, $self->_sqlcase('set'), join ', ', @sqlq;
1613 } elsif ($ref eq 'ARRAY') {
1614 # unlike insert(), assume these are ONLY the column names, i.e. for SQL
1617 if ($r eq 'ARRAY') { # literal SQL with bind
1618 my ($sql, @bind) = @$v;
1619 $self->_assert_bindval_matches_bindtype(@bind);
1622 } elsif ($r eq 'SCALAR') { # literal SQL without bind
1623 # embedded literal SQL
1630 push @sql, '(' . join(', ', @sqlq) . ')';
1631 } elsif ($ref eq 'SCALAR') {
1635 # strings get case twiddled
1636 push @sql, $self->_sqlcase($_);
1640 my $sql = join ' ', @sql;
1642 # this is pretty tricky
1643 # if ask for an array, return ($stmt, @bind)
1644 # otherwise, s/?/shift @sqlv/ to put it inline
1646 return ($sql, @sqlv);
1648 1 while $sql =~ s/\?/my $d = shift(@sqlv);
1649 ref $d ? $d->[1] : $d/e;
1658 # This allows us to check for a local, then _form, attr
1660 my($name) = $AUTOLOAD =~ /.*::(.+)/;
1661 return $self->generate($name, @_);
1672 SQL::Abstract - Generate SQL from Perl data structures
1678 my $sql = SQL::Abstract->new;
1680 my($stmt, @bind) = $sql->select($source, \@fields, \%where, $order);
1682 my($stmt, @bind) = $sql->insert($table, \%fieldvals || \@values);
1684 my($stmt, @bind) = $sql->update($table, \%fieldvals, \%where);
1686 my($stmt, @bind) = $sql->delete($table, \%where);
1688 # Then, use these in your DBI statements
1689 my $sth = $dbh->prepare($stmt);
1690 $sth->execute(@bind);
1692 # Just generate the WHERE clause
1693 my($stmt, @bind) = $sql->where(\%where, $order);
1695 # Return values in the same order, for hashed queries
1696 # See PERFORMANCE section for more details
1697 my @bind = $sql->values(\%fieldvals);
1701 This module was inspired by the excellent L<DBIx::Abstract>.
1702 However, in using that module I found that what I really wanted
1703 to do was generate SQL, but still retain complete control over my
1704 statement handles and use the DBI interface. So, I set out to
1705 create an abstract SQL generation module.
1707 While based on the concepts used by L<DBIx::Abstract>, there are
1708 several important differences, especially when it comes to WHERE
1709 clauses. I have modified the concepts used to make the SQL easier
1710 to generate from Perl data structures and, IMO, more intuitive.
1711 The underlying idea is for this module to do what you mean, based
1712 on the data structures you provide it. The big advantage is that
1713 you don't have to modify your code every time your data changes,
1714 as this module figures it out.
1716 To begin with, an SQL INSERT is as easy as just specifying a hash
1717 of C<key=value> pairs:
1720 name => 'Jimbo Bobson',
1721 phone => '123-456-7890',
1722 address => '42 Sister Lane',
1723 city => 'St. Louis',
1724 state => 'Louisiana',
1727 The SQL can then be generated with this:
1729 my($stmt, @bind) = $sql->insert('people', \%data);
1731 Which would give you something like this:
1733 $stmt = "INSERT INTO people
1734 (address, city, name, phone, state)
1735 VALUES (?, ?, ?, ?, ?)";
1736 @bind = ('42 Sister Lane', 'St. Louis', 'Jimbo Bobson',
1737 '123-456-7890', 'Louisiana');
1739 These are then used directly in your DBI code:
1741 my $sth = $dbh->prepare($stmt);
1742 $sth->execute(@bind);
1744 =head2 Inserting and Updating Arrays
1746 If your database has array types (like for example Postgres),
1747 activate the special option C<< array_datatypes => 1 >>
1748 when creating the C<SQL::Abstract> object.
1749 Then you may use an arrayref to insert and update database array types:
1751 my $sql = SQL::Abstract->new(array_datatypes => 1);
1753 planets => [qw/Mercury Venus Earth Mars/]
1756 my($stmt, @bind) = $sql->insert('solar_system', \%data);
1760 $stmt = "INSERT INTO solar_system (planets) VALUES (?)"
1762 @bind = (['Mercury', 'Venus', 'Earth', 'Mars']);
1765 =head2 Inserting and Updating SQL
1767 In order to apply SQL functions to elements of your C<%data> you may
1768 specify a reference to an arrayref for the given hash value. For example,
1769 if you need to execute the Oracle C<to_date> function on a value, you can
1770 say something like this:
1774 date_entered => \[ "to_date(?,'MM/DD/YYYY')", "03/02/2003" ],
1777 The first value in the array is the actual SQL. Any other values are
1778 optional and would be included in the bind values array. This gives
1781 my($stmt, @bind) = $sql->insert('people', \%data);
1783 $stmt = "INSERT INTO people (name, date_entered)
1784 VALUES (?, to_date(?,'MM/DD/YYYY'))";
1785 @bind = ('Bill', '03/02/2003');
1787 An UPDATE is just as easy, all you change is the name of the function:
1789 my($stmt, @bind) = $sql->update('people', \%data);
1791 Notice that your C<%data> isn't touched; the module will generate
1792 the appropriately quirky SQL for you automatically. Usually you'll
1793 want to specify a WHERE clause for your UPDATE, though, which is
1794 where handling C<%where> hashes comes in handy...
1796 =head2 Complex where statements
1798 This module can generate pretty complicated WHERE statements
1799 easily. For example, simple C<key=value> pairs are taken to mean
1800 equality, and if you want to see if a field is within a set
1801 of values, you can use an arrayref. Let's say we wanted to
1802 SELECT some data based on this criteria:
1805 requestor => 'inna',
1806 worker => ['nwiger', 'rcwe', 'sfz'],
1807 status => { '!=', 'completed' }
1810 my($stmt, @bind) = $sql->select('tickets', '*', \%where);
1812 The above would give you something like this:
1814 $stmt = "SELECT * FROM tickets WHERE
1815 ( requestor = ? ) AND ( status != ? )
1816 AND ( worker = ? OR worker = ? OR worker = ? )";
1817 @bind = ('inna', 'completed', 'nwiger', 'rcwe', 'sfz');
1819 Which you could then use in DBI code like so:
1821 my $sth = $dbh->prepare($stmt);
1822 $sth->execute(@bind);
1828 The methods are simple. There's one for every major SQL operation,
1829 and a constructor you use first. The arguments are specified in a
1830 similar order for each method (table, then fields, then a where
1831 clause) to try and simplify things.
1833 =head2 new(option => 'value')
1835 The C<new()> function takes a list of options and values, and returns
1836 a new B<SQL::Abstract> object which can then be used to generate SQL
1837 through the methods below. The options accepted are:
1843 If set to 'lower', then SQL will be generated in all lowercase. By
1844 default SQL is generated in "textbook" case meaning something like:
1846 SELECT a_field FROM a_table WHERE some_field LIKE '%someval%'
1848 Any setting other than 'lower' is ignored.
1852 This determines what the default comparison operator is. By default
1853 it is C<=>, meaning that a hash like this:
1855 %where = (name => 'nwiger', email => 'nate@wiger.org');
1857 Will generate SQL like this:
1859 WHERE name = 'nwiger' AND email = 'nate@wiger.org'
1861 However, you may want loose comparisons by default, so if you set
1862 C<cmp> to C<like> you would get SQL such as:
1864 WHERE name like 'nwiger' AND email like 'nate@wiger.org'
1866 You can also override the comparison on an individual basis - see
1867 the huge section on L</"WHERE CLAUSES"> at the bottom.
1869 =item sqltrue, sqlfalse
1871 Expressions for inserting boolean values within SQL statements.
1872 By default these are C<1=1> and C<1=0>. They are used
1873 by the special operators C<-in> and C<-not_in> for generating
1874 correct SQL even when the argument is an empty array (see below).
1878 This determines the default logical operator for multiple WHERE
1879 statements in arrays or hashes. If absent, the default logic is "or"
1880 for arrays, and "and" for hashes. This means that a WHERE
1884 event_date => {'>=', '2/13/99'},
1885 event_date => {'<=', '4/24/03'},
1888 will generate SQL like this:
1890 WHERE event_date >= '2/13/99' OR event_date <= '4/24/03'
1892 This is probably not what you want given this query, though (look
1893 at the dates). To change the "OR" to an "AND", simply specify:
1895 my $sql = SQL::Abstract->new(logic => 'and');
1897 Which will change the above C<WHERE> to:
1899 WHERE event_date >= '2/13/99' AND event_date <= '4/24/03'
1901 The logic can also be changed locally by inserting
1902 a modifier in front of an arrayref:
1904 @where = (-and => [event_date => {'>=', '2/13/99'},
1905 event_date => {'<=', '4/24/03'} ]);
1907 See the L</"WHERE CLAUSES"> section for explanations.
1911 This will automatically convert comparisons using the specified SQL
1912 function for both column and value. This is mostly used with an argument
1913 of C<upper> or C<lower>, so that the SQL will have the effect of
1914 case-insensitive "searches". For example, this:
1916 $sql = SQL::Abstract->new(convert => 'upper');
1917 %where = (keywords => 'MaKe iT CAse inSeNSItive');
1919 Will turn out the following SQL:
1921 WHERE upper(keywords) like upper('MaKe iT CAse inSeNSItive')
1923 The conversion can be C<upper()>, C<lower()>, or any other SQL function
1924 that can be applied symmetrically to fields (actually B<SQL::Abstract> does
1925 not validate this option; it will just pass through what you specify verbatim).
1929 This is a kludge because many databases suck. For example, you can't
1930 just bind values using DBI's C<execute()> for Oracle C<CLOB> or C<BLOB> fields.
1931 Instead, you have to use C<bind_param()>:
1933 $sth->bind_param(1, 'reg data');
1934 $sth->bind_param(2, $lots, {ora_type => ORA_CLOB});
1936 The problem is, B<SQL::Abstract> will normally just return a C<@bind> array,
1937 which loses track of which field each slot refers to. Fear not.
1939 If you specify C<bindtype> in new, you can determine how C<@bind> is returned.
1940 Currently, you can specify either C<normal> (default) or C<columns>. If you
1941 specify C<columns>, you will get an array that looks like this:
1943 my $sql = SQL::Abstract->new(bindtype => 'columns');
1944 my($stmt, @bind) = $sql->insert(...);
1947 [ 'column1', 'value1' ],
1948 [ 'column2', 'value2' ],
1949 [ 'column3', 'value3' ],
1952 You can then iterate through this manually, using DBI's C<bind_param()>.
1954 $sth->prepare($stmt);
1957 my($col, $data) = @$_;
1958 if ($col eq 'details' || $col eq 'comments') {
1959 $sth->bind_param($i, $data, {ora_type => ORA_CLOB});
1960 } elsif ($col eq 'image') {
1961 $sth->bind_param($i, $data, {ora_type => ORA_BLOB});
1963 $sth->bind_param($i, $data);
1967 $sth->execute; # execute without @bind now
1969 Now, why would you still use B<SQL::Abstract> if you have to do this crap?
1970 Basically, the advantage is still that you don't have to care which fields
1971 are or are not included. You could wrap that above C<for> loop in a simple
1972 sub called C<bind_fields()> or something and reuse it repeatedly. You still
1973 get a layer of abstraction over manual SQL specification.
1975 Note that if you set L</bindtype> to C<columns>, the C<\[ $sql, @bind ]>
1976 construct (see L</Literal SQL with placeholders and bind values (subqueries)>)
1977 will expect the bind values in this format.
1981 This is the character that a table or column name will be quoted
1982 with. By default this is an empty string, but you could set it to
1983 the character C<`>, to generate SQL like this:
1985 SELECT `a_field` FROM `a_table` WHERE `some_field` LIKE '%someval%'
1987 Alternatively, you can supply an array ref of two items, the first being the left
1988 hand quote character, and the second the right hand quote character. For
1989 example, you could supply C<['[',']']> for SQL Server 2000 compliant quotes
1990 that generates SQL like this:
1992 SELECT [a_field] FROM [a_table] WHERE [some_field] LIKE '%someval%'
1994 Quoting is useful if you have tables or columns names that are reserved
1995 words in your database's SQL dialect.
1999 This is the character that will be used to escape L</quote_char>s appearing
2000 in an identifier before it has been quoted.
2002 The parameter default in case of a single L</quote_char> character is the quote
2005 When opening-closing-style quoting is used (L</quote_char> is an arrayref)
2006 this parameter defaults to the B<closing (right)> L</quote_char>. Occurrences
2007 of the B<opening (left)> L</quote_char> within the identifier are currently left
2008 untouched. The default for opening-closing-style quotes may change in future
2009 versions, thus you are B<strongly encouraged> to specify the escape character
2014 This is the character that separates a table and column name. It is
2015 necessary to specify this when the C<quote_char> option is selected,
2016 so that tables and column names can be individually quoted like this:
2018 SELECT `table`.`one_field` FROM `table` WHERE `table`.`other_field` = 1
2020 =item injection_guard
2022 A regular expression C<qr/.../> that is applied to any C<-function> and unquoted
2023 column name specified in a query structure. This is a safety mechanism to avoid
2024 injection attacks when mishandling user input e.g.:
2026 my %condition_as_column_value_pairs = get_values_from_user();
2027 $sqla->select( ... , \%condition_as_column_value_pairs );
2029 If the expression matches an exception is thrown. Note that literal SQL
2030 supplied via C<\'...'> or C<\['...']> is B<not> checked in any way.
2032 Defaults to checking for C<;> and the C<GO> keyword (TransactSQL)
2034 =item array_datatypes
2036 When this option is true, arrayrefs in INSERT or UPDATE are
2037 interpreted as array datatypes and are passed directly
2039 When this option is false, arrayrefs are interpreted
2040 as literal SQL, just like refs to arrayrefs
2041 (but this behavior is for backwards compatibility; when writing
2042 new queries, use the "reference to arrayref" syntax
2048 Takes a reference to a list of "special operators"
2049 to extend the syntax understood by L<SQL::Abstract>.
2050 See section L</"SPECIAL OPERATORS"> for details.
2054 Takes a reference to a list of "unary operators"
2055 to extend the syntax understood by L<SQL::Abstract>.
2056 See section L</"UNARY OPERATORS"> for details.
2062 =head2 insert($table, \@values || \%fieldvals, \%options)
2064 This is the simplest function. You simply give it a table name
2065 and either an arrayref of values or hashref of field/value pairs.
2066 It returns an SQL INSERT statement and a list of bind values.
2067 See the sections on L</"Inserting and Updating Arrays"> and
2068 L</"Inserting and Updating SQL"> for information on how to insert
2069 with those data types.
2071 The optional C<\%options> hash reference may contain additional
2072 options to generate the insert SQL. Currently supported options
2079 Takes either a scalar of raw SQL fields, or an array reference of
2080 field names, and adds on an SQL C<RETURNING> statement at the end.
2081 This allows you to return data generated by the insert statement
2082 (such as row IDs) without performing another C<SELECT> statement.
2083 Note, however, this is not part of the SQL standard and may not
2084 be supported by all database engines.
2088 =head2 update($table, \%fieldvals, \%where, \%options)
2090 This takes a table, hashref of field/value pairs, and an optional
2091 hashref L<WHERE clause|/WHERE CLAUSES>. It returns an SQL UPDATE function and a list
2093 See the sections on L</"Inserting and Updating Arrays"> and
2094 L</"Inserting and Updating SQL"> for information on how to insert
2095 with those data types.
2097 The optional C<\%options> hash reference may contain additional
2098 options to generate the update SQL. Currently supported options
2105 See the C<returning> option to
2106 L<insert|/insert($table, \@values || \%fieldvals, \%options)>.
2110 =head2 select($source, $fields, $where, $order)
2112 This returns a SQL SELECT statement and associated list of bind values, as
2113 specified by the arguments:
2119 Specification of the 'FROM' part of the statement.
2120 The argument can be either a plain scalar (interpreted as a table
2121 name, will be quoted), or an arrayref (interpreted as a list
2122 of table names, joined by commas, quoted), or a scalarref
2123 (literal SQL, not quoted).
2127 Specification of the list of fields to retrieve from
2129 The argument can be either an arrayref (interpreted as a list
2130 of field names, will be joined by commas and quoted), or a
2131 plain scalar (literal SQL, not quoted).
2132 Please observe that this API is not as flexible as that of
2133 the first argument C<$source>, for backwards compatibility reasons.
2137 Optional argument to specify the WHERE part of the query.
2138 The argument is most often a hashref, but can also be
2139 an arrayref or plain scalar --
2140 see section L<WHERE clause|/"WHERE CLAUSES"> for details.
2144 Optional argument to specify the ORDER BY part of the query.
2145 The argument can be a scalar, a hashref or an arrayref
2146 -- see section L<ORDER BY clause|/"ORDER BY CLAUSES">
2152 =head2 delete($table, \%where, \%options)
2154 This takes a table name and optional hashref L<WHERE clause|/WHERE CLAUSES>.
2155 It returns an SQL DELETE statement and list of bind values.
2157 The optional C<\%options> hash reference may contain additional
2158 options to generate the delete SQL. Currently supported options
2165 See the C<returning> option to
2166 L<insert|/insert($table, \@values || \%fieldvals, \%options)>.
2170 =head2 where(\%where, $order)
2172 This is used to generate just the WHERE clause. For example,
2173 if you have an arbitrary data structure and know what the
2174 rest of your SQL is going to look like, but want an easy way
2175 to produce a WHERE clause, use this. It returns an SQL WHERE
2176 clause and list of bind values.
2179 =head2 values(\%data)
2181 This just returns the values from the hash C<%data>, in the same
2182 order that would be returned from any of the other above queries.
2183 Using this allows you to markedly speed up your queries if you
2184 are affecting lots of rows. See below under the L</"PERFORMANCE"> section.
2186 =head2 generate($any, 'number', $of, \@data, $struct, \%types)
2188 Warning: This is an experimental method and subject to change.
2190 This returns arbitrarily generated SQL. It's a really basic shortcut.
2191 It will return two different things, depending on return context:
2193 my($stmt, @bind) = $sql->generate('create table', \$table, \@fields);
2194 my $stmt_and_val = $sql->generate('create table', \$table, \@fields);
2196 These would return the following:
2198 # First calling form
2199 $stmt = "CREATE TABLE test (?, ?)";
2200 @bind = (field1, field2);
2202 # Second calling form
2203 $stmt_and_val = "CREATE TABLE test (field1, field2)";
2205 Depending on what you're trying to do, it's up to you to choose the correct
2206 format. In this example, the second form is what you would want.
2210 $sql->generate('alter session', { nls_date_format => 'MM/YY' });
2214 ALTER SESSION SET nls_date_format = 'MM/YY'
2216 You get the idea. Strings get their case twiddled, but everything
2217 else remains verbatim.
2219 =head1 EXPORTABLE FUNCTIONS
2221 =head2 is_plain_value
2223 Determines if the supplied argument is a plain value as understood by this
2228 =item * The value is C<undef>
2230 =item * The value is a non-reference
2232 =item * The value is an object with stringification overloading
2234 =item * The value is of the form C<< { -value => $anything } >>
2238 On failure returns C<undef>, on success returns a B<scalar> reference
2239 to the original supplied argument.
2245 The stringification overloading detection is rather advanced: it takes
2246 into consideration not only the presence of a C<""> overload, but if that
2247 fails also checks for enabled
2248 L<autogenerated versions of C<"">|overload/Magic Autogeneration>, based
2249 on either C<0+> or C<bool>.
2251 Unfortunately testing in the field indicates that this
2252 detection B<< may tickle a latent bug in perl versions before 5.018 >>,
2253 but only when very large numbers of stringifying objects are involved.
2254 At the time of writing ( Sep 2014 ) there is no clear explanation of
2255 the direct cause, nor is there a manageably small test case that reliably
2256 reproduces the problem.
2258 If you encounter any of the following exceptions in B<random places within
2259 your application stack> - this module may be to blame:
2261 Operation "ne": no method found,
2262 left argument in overloaded package <something>,
2263 right argument in overloaded package <something>
2267 Stub found while resolving method "???" overloading """" in package <something>
2269 If you fall victim to the above - please attempt to reduce the problem
2270 to something that could be sent to the L<SQL::Abstract developers
2271 |DBIx::Class/GETTING HELP/SUPPORT>
2272 (either publicly or privately). As a workaround in the meantime you can
2273 set C<$ENV{SQLA_ISVALUE_IGNORE_AUTOGENERATED_STRINGIFICATION}> to a true
2274 value, which will most likely eliminate your problem (at the expense of
2275 not being able to properly detect exotic forms of stringification).
2277 This notice and environment variable will be removed in a future version,
2278 as soon as the underlying problem is found and a reliable workaround is
2283 =head2 is_literal_value
2285 Determines if the supplied argument is a literal value as understood by this
2290 =item * C<\$sql_string>
2292 =item * C<\[ $sql_string, @bind_values ]>
2296 On failure returns C<undef>, on success returns an B<array> reference
2297 containing the unpacked version of the supplied literal SQL and bind values.
2299 =head1 WHERE CLAUSES
2303 This module uses a variation on the idea from L<DBIx::Abstract>. It
2304 is B<NOT>, repeat I<not> 100% compatible. B<The main logic of this
2305 module is that things in arrays are OR'ed, and things in hashes
2308 The easiest way to explain is to show lots of examples. After
2309 each C<%where> hash shown, it is assumed you used:
2311 my($stmt, @bind) = $sql->where(\%where);
2313 However, note that the C<%where> hash can be used directly in any
2314 of the other functions as well, as described above.
2316 =head2 Key-value pairs
2318 So, let's get started. To begin, a simple hash:
2322 status => 'completed'
2325 Is converted to SQL C<key = val> statements:
2327 $stmt = "WHERE user = ? AND status = ?";
2328 @bind = ('nwiger', 'completed');
2330 One common thing I end up doing is having a list of values that
2331 a field can be in. To do this, simply specify a list inside of
2336 status => ['assigned', 'in-progress', 'pending'];
2339 This simple code will create the following:
2341 $stmt = "WHERE user = ? AND ( status = ? OR status = ? OR status = ? )";
2342 @bind = ('nwiger', 'assigned', 'in-progress', 'pending');
2344 A field associated to an empty arrayref will be considered a
2345 logical false and will generate 0=1.
2347 =head2 Tests for NULL values
2349 If the value part is C<undef> then this is converted to SQL <IS NULL>
2358 $stmt = "WHERE user = ? AND status IS NULL";
2361 To test if a column IS NOT NULL:
2365 status => { '!=', undef },
2368 =head2 Specific comparison operators
2370 If you want to specify a different type of operator for your comparison,
2371 you can use a hashref for a given column:
2375 status => { '!=', 'completed' }
2378 Which would generate:
2380 $stmt = "WHERE user = ? AND status != ?";
2381 @bind = ('nwiger', 'completed');
2383 To test against multiple values, just enclose the values in an arrayref:
2385 status => { '=', ['assigned', 'in-progress', 'pending'] };
2387 Which would give you:
2389 "WHERE status = ? OR status = ? OR status = ?"
2392 The hashref can also contain multiple pairs, in which case it is expanded
2393 into an C<AND> of its elements:
2397 status => { '!=', 'completed', -not_like => 'pending%' }
2400 # Or more dynamically, like from a form
2401 $where{user} = 'nwiger';
2402 $where{status}{'!='} = 'completed';
2403 $where{status}{'-not_like'} = 'pending%';
2405 # Both generate this
2406 $stmt = "WHERE user = ? AND status != ? AND status NOT LIKE ?";
2407 @bind = ('nwiger', 'completed', 'pending%');
2410 To get an OR instead, you can combine it with the arrayref idea:
2414 priority => [ { '=', 2 }, { '>', 5 } ]
2417 Which would generate:
2419 $stmt = "WHERE ( priority = ? OR priority > ? ) AND user = ?";
2420 @bind = ('2', '5', 'nwiger');
2422 If you want to include literal SQL (with or without bind values), just use a
2423 scalar reference or reference to an arrayref as the value:
2426 date_entered => { '>' => \["to_date(?, 'MM/DD/YYYY')", "11/26/2008"] },
2427 date_expires => { '<' => \"now()" }
2430 Which would generate:
2432 $stmt = "WHERE date_entered > to_date(?, 'MM/DD/YYYY') AND date_expires < now()";
2433 @bind = ('11/26/2008');
2436 =head2 Logic and nesting operators
2438 In the example above,
2439 there is a subtle trap if you want to say something like
2440 this (notice the C<AND>):
2442 WHERE priority != ? AND priority != ?
2444 Because, in Perl you I<can't> do this:
2446 priority => { '!=' => 2, '!=' => 1 }
2448 As the second C<!=> key will obliterate the first. The solution
2449 is to use the special C<-modifier> form inside an arrayref:
2451 priority => [ -and => {'!=', 2},
2455 Normally, these would be joined by C<OR>, but the modifier tells it
2456 to use C<AND> instead. (Hint: You can use this in conjunction with the
2457 C<logic> option to C<new()> in order to change the way your queries
2458 work by default.) B<Important:> Note that the C<-modifier> goes
2459 B<INSIDE> the arrayref, as an extra first element. This will
2460 B<NOT> do what you think it might:
2462 priority => -and => [{'!=', 2}, {'!=', 1}] # WRONG!
2464 Here is a quick list of equivalencies, since there is some overlap:
2467 status => {'!=', 'completed', 'not like', 'pending%' }
2468 status => [ -and => {'!=', 'completed'}, {'not like', 'pending%'}]
2471 status => {'=', ['assigned', 'in-progress']}
2472 status => [ -or => {'=', 'assigned'}, {'=', 'in-progress'}]
2473 status => [ {'=', 'assigned'}, {'=', 'in-progress'} ]
2477 =head2 Special operators: IN, BETWEEN, etc.
2479 You can also use the hashref format to compare a list of fields using the
2480 C<IN> comparison operator, by specifying the list as an arrayref:
2483 status => 'completed',
2484 reportid => { -in => [567, 2335, 2] }
2487 Which would generate:
2489 $stmt = "WHERE status = ? AND reportid IN (?,?,?)";
2490 @bind = ('completed', '567', '2335', '2');
2492 The reverse operator C<-not_in> generates SQL C<NOT IN> and is used in
2495 If the argument to C<-in> is an empty array, 'sqlfalse' is generated
2496 (by default: C<1=0>). Similarly, C<< -not_in => [] >> generates
2497 'sqltrue' (by default: C<1=1>).
2499 In addition to the array you can supply a chunk of literal sql or
2500 literal sql with bind:
2503 customer => { -in => \[
2504 'SELECT cust_id FROM cust WHERE balance > ?',
2507 status => { -in => \'SELECT status_codes FROM states' },
2513 customer IN ( SELECT cust_id FROM cust WHERE balance > ? )
2514 AND status IN ( SELECT status_codes FROM states )
2518 Finally, if the argument to C<-in> is not a reference, it will be
2519 treated as a single-element array.
2521 Another pair of operators is C<-between> and C<-not_between>,
2522 used with an arrayref of two values:
2526 completion_date => {
2527 -not_between => ['2002-10-01', '2003-02-06']
2533 WHERE user = ? AND completion_date NOT BETWEEN ( ? AND ? )
2535 Just like with C<-in> all plausible combinations of literal SQL
2539 start0 => { -between => [ 1, 2 ] },
2540 start1 => { -between => \["? AND ?", 1, 2] },
2541 start2 => { -between => \"lower(x) AND upper(y)" },
2542 start3 => { -between => [
2544 \["upper(?)", 'stuff' ],
2551 ( start0 BETWEEN ? AND ? )
2552 AND ( start1 BETWEEN ? AND ? )
2553 AND ( start2 BETWEEN lower(x) AND upper(y) )
2554 AND ( start3 BETWEEN lower(x) AND upper(?) )
2556 @bind = (1, 2, 1, 2, 'stuff');
2559 These are the two builtin "special operators"; but the
2560 list can be expanded: see section L</"SPECIAL OPERATORS"> below.
2562 =head2 Unary operators: bool
2564 If you wish to test against boolean columns or functions within your
2565 database you can use the C<-bool> and C<-not_bool> operators. For
2566 example to test the column C<is_user> being true and the column
2567 C<is_enabled> being false you would use:-
2571 -not_bool => 'is_enabled',
2576 WHERE is_user AND NOT is_enabled
2578 If a more complex combination is required, testing more conditions,
2579 then you should use the and/or operators:-
2584 -not_bool => { two=> { -rlike => 'bar' } },
2585 -not_bool => { three => [ { '=', 2 }, { '>', 5 } ] },
2596 (NOT ( three = ? OR three > ? ))
2599 =head2 Nested conditions, -and/-or prefixes
2601 So far, we've seen how multiple conditions are joined with a top-level
2602 C<AND>. We can change this by putting the different conditions we want in
2603 hashes and then putting those hashes in an array. For example:
2608 status => { -like => ['pending%', 'dispatched'] },
2612 status => 'unassigned',
2616 This data structure would create the following:
2618 $stmt = "WHERE ( user = ? AND ( status LIKE ? OR status LIKE ? ) )
2619 OR ( user = ? AND status = ? ) )";
2620 @bind = ('nwiger', 'pending', 'dispatched', 'robot', 'unassigned');
2623 Clauses in hashrefs or arrayrefs can be prefixed with an C<-and> or C<-or>
2624 to change the logic inside:
2630 -and => [ workhrs => {'>', 20}, geo => 'ASIA' ],
2631 -or => { workhrs => {'<', 50}, geo => 'EURO' },
2638 $stmt = "WHERE ( user = ?
2639 AND ( ( workhrs > ? AND geo = ? )
2640 OR ( workhrs < ? OR geo = ? ) ) )";
2641 @bind = ('nwiger', '20', 'ASIA', '50', 'EURO');
2643 =head3 Algebraic inconsistency, for historical reasons
2645 C<Important note>: when connecting several conditions, the C<-and->|C<-or>
2646 operator goes C<outside> of the nested structure; whereas when connecting
2647 several constraints on one column, the C<-and> operator goes
2648 C<inside> the arrayref. Here is an example combining both features:
2651 -and => [a => 1, b => 2],
2652 -or => [c => 3, d => 4],
2653 e => [-and => {-like => 'foo%'}, {-like => '%bar'} ]
2658 WHERE ( ( ( a = ? AND b = ? )
2659 OR ( c = ? OR d = ? )
2660 OR ( e LIKE ? AND e LIKE ? ) ) )
2662 This difference in syntax is unfortunate but must be preserved for
2663 historical reasons. So be careful: the two examples below would
2664 seem algebraically equivalent, but they are not
2667 { -like => 'foo%' },
2668 { -like => '%bar' },
2670 # yields: WHERE ( ( col LIKE ? AND col LIKE ? ) )
2673 { col => { -like => 'foo%' } },
2674 { col => { -like => '%bar' } },
2676 # yields: WHERE ( ( col LIKE ? OR col LIKE ? ) )
2679 =head2 Literal SQL and value type operators
2681 The basic premise of SQL::Abstract is that in WHERE specifications the "left
2682 side" is a column name and the "right side" is a value (normally rendered as
2683 a placeholder). This holds true for both hashrefs and arrayref pairs as you
2684 see in the L</WHERE CLAUSES> examples above. Sometimes it is necessary to
2685 alter this behavior. There are several ways of doing so.
2689 This is a virtual operator that signals the string to its right side is an
2690 identifier (a column name) and not a value. For example to compare two
2691 columns you would write:
2694 priority => { '<', 2 },
2695 requestor => { -ident => 'submitter' },
2700 $stmt = "WHERE priority < ? AND requestor = submitter";
2703 If you are maintaining legacy code you may see a different construct as
2704 described in L</Deprecated usage of Literal SQL>, please use C<-ident> in new
2709 This is a virtual operator that signals that the construct to its right side
2710 is a value to be passed to DBI. This is for example necessary when you want
2711 to write a where clause against an array (for RDBMS that support such
2712 datatypes). For example:
2715 array => { -value => [1, 2, 3] }
2720 $stmt = 'WHERE array = ?';
2721 @bind = ([1, 2, 3]);
2723 Note that if you were to simply say:
2729 the result would probably not be what you wanted:
2731 $stmt = 'WHERE array = ? OR array = ? OR array = ?';
2736 Finally, sometimes only literal SQL will do. To include a random snippet
2737 of SQL verbatim, you specify it as a scalar reference. Consider this only
2738 as a last resort. Usually there is a better way. For example:
2741 priority => { '<', 2 },
2742 requestor => { -in => \'(SELECT name FROM hitmen)' },
2747 $stmt = "WHERE priority < ? AND requestor IN (SELECT name FROM hitmen)"
2750 Note that in this example, you only get one bind parameter back, since
2751 the verbatim SQL is passed as part of the statement.
2755 Never use untrusted input as a literal SQL argument - this is a massive
2756 security risk (there is no way to check literal snippets for SQL
2757 injections and other nastyness). If you need to deal with untrusted input
2758 use literal SQL with placeholders as described next.
2760 =head3 Literal SQL with placeholders and bind values (subqueries)
2762 If the literal SQL to be inserted has placeholders and bind values,
2763 use a reference to an arrayref (yes this is a double reference --
2764 not so common, but perfectly legal Perl). For example, to find a date
2765 in Postgres you can use something like this:
2768 date_column => \[ "= date '2008-09-30' - ?::integer", 10 ]
2773 $stmt = "WHERE ( date_column = date '2008-09-30' - ?::integer )"
2776 Note that you must pass the bind values in the same format as they are returned
2777 by L<where|/where(\%where, $order)>. This means that if you set L</bindtype>
2778 to C<columns>, you must provide the bind values in the
2779 C<< [ column_meta => value ] >> format, where C<column_meta> is an opaque
2780 scalar value; most commonly the column name, but you can use any scalar value
2781 (including references and blessed references), L<SQL::Abstract> will simply
2782 pass it through intact. So if C<bindtype> is set to C<columns> the above
2783 example will look like:
2786 date_column => \[ "= date '2008-09-30' - ?::integer", [ {} => 10 ] ]
2789 Literal SQL is especially useful for nesting parenthesized clauses in the
2790 main SQL query. Here is a first example:
2792 my ($sub_stmt, @sub_bind) = ("SELECT c1 FROM t1 WHERE c2 < ? AND c3 LIKE ?",
2796 bar => \["IN ($sub_stmt)" => @sub_bind],
2801 $stmt = "WHERE (foo = ? AND bar IN (SELECT c1 FROM t1
2802 WHERE c2 < ? AND c3 LIKE ?))";
2803 @bind = (1234, 100, "foo%");
2805 Other subquery operators, like for example C<"E<gt> ALL"> or C<"NOT IN">,
2806 are expressed in the same way. Of course the C<$sub_stmt> and
2807 its associated bind values can be generated through a former call
2810 my ($sub_stmt, @sub_bind)
2811 = $sql->select("t1", "c1", {c2 => {"<" => 100},
2812 c3 => {-like => "foo%"}});
2815 bar => \["> ALL ($sub_stmt)" => @sub_bind],
2818 In the examples above, the subquery was used as an operator on a column;
2819 but the same principle also applies for a clause within the main C<%where>
2820 hash, like an EXISTS subquery:
2822 my ($sub_stmt, @sub_bind)
2823 = $sql->select("t1", "*", {c1 => 1, c2 => \"> t0.c0"});
2824 my %where = ( -and => [
2826 \["EXISTS ($sub_stmt)" => @sub_bind],
2831 $stmt = "WHERE (foo = ? AND EXISTS (SELECT * FROM t1
2832 WHERE c1 = ? AND c2 > t0.c0))";
2836 Observe that the condition on C<c2> in the subquery refers to
2837 column C<t0.c0> of the main query: this is I<not> a bind
2838 value, so we have to express it through a scalar ref.
2839 Writing C<< c2 => {">" => "t0.c0"} >> would have generated
2840 C<< c2 > ? >> with bind value C<"t0.c0"> ... not exactly
2841 what we wanted here.
2843 Finally, here is an example where a subquery is used
2844 for expressing unary negation:
2846 my ($sub_stmt, @sub_bind)
2847 = $sql->where({age => [{"<" => 10}, {">" => 20}]});
2848 $sub_stmt =~ s/^ where //i; # don't want "WHERE" in the subclause
2850 lname => {like => '%son%'},
2851 \["NOT ($sub_stmt)" => @sub_bind],
2856 $stmt = "lname LIKE ? AND NOT ( age < ? OR age > ? )"
2857 @bind = ('%son%', 10, 20)
2859 =head3 Deprecated usage of Literal SQL
2861 Below are some examples of archaic use of literal SQL. It is shown only as
2862 reference for those who deal with legacy code. Each example has a much
2863 better, cleaner and safer alternative that users should opt for in new code.
2869 my %where = ( requestor => \'IS NOT NULL' )
2871 $stmt = "WHERE requestor IS NOT NULL"
2873 This used to be the way of generating NULL comparisons, before the handling
2874 of C<undef> got formalized. For new code please use the superior syntax as
2875 described in L</Tests for NULL values>.
2879 my %where = ( requestor => \'= submitter' )
2881 $stmt = "WHERE requestor = submitter"
2883 This used to be the only way to compare columns. Use the superior L</-ident>
2884 method for all new code. For example an identifier declared in such a way
2885 will be properly quoted if L</quote_char> is properly set, while the legacy
2886 form will remain as supplied.
2890 my %where = ( is_ready => \"", completed => { '>', '2012-12-21' } )
2892 $stmt = "WHERE completed > ? AND is_ready"
2893 @bind = ('2012-12-21')
2895 Using an empty string literal used to be the only way to express a boolean.
2896 For all new code please use the much more readable
2897 L<-bool|/Unary operators: bool> operator.
2903 These pages could go on for a while, since the nesting of the data
2904 structures this module can handle are pretty much unlimited (the
2905 module implements the C<WHERE> expansion as a recursive function
2906 internally). Your best bet is to "play around" with the module a
2907 little to see how the data structures behave, and choose the best
2908 format for your data based on that.
2910 And of course, all the values above will probably be replaced with
2911 variables gotten from forms or the command line. After all, if you
2912 knew everything ahead of time, you wouldn't have to worry about
2913 dynamically-generating SQL and could just hardwire it into your
2916 =head1 ORDER BY CLAUSES
2918 Some functions take an order by clause. This can either be a scalar (just a
2919 column name), a hashref of C<< { -desc => 'col' } >> or C<< { -asc => 'col' }
2920 >>, a scalarref, an arrayref-ref, or an arrayref of any of the previous
2923 Given | Will Generate
2924 ---------------------------------------------------------------
2926 'colA' | ORDER BY colA
2928 [qw/colA colB/] | ORDER BY colA, colB
2930 {-asc => 'colA'} | ORDER BY colA ASC
2932 {-desc => 'colB'} | ORDER BY colB DESC
2934 ['colA', {-asc => 'colB'}] | ORDER BY colA, colB ASC
2936 { -asc => [qw/colA colB/] } | ORDER BY colA ASC, colB ASC
2938 \'colA DESC' | ORDER BY colA DESC
2940 \[ 'FUNC(colA, ?)', $x ] | ORDER BY FUNC(colA, ?)
2941 | /* ...with $x bound to ? */
2944 { -asc => 'colA' }, | colA ASC,
2945 { -desc => [qw/colB/] }, | colB DESC,
2946 { -asc => [qw/colC colD/] },| colC ASC, colD ASC,
2947 \'colE DESC', | colE DESC,
2948 \[ 'FUNC(colF, ?)', $x ], | FUNC(colF, ?)
2949 ] | /* ...with $x bound to ? */
2950 ===============================================================
2954 =head1 SPECIAL OPERATORS
2956 my $sqlmaker = SQL::Abstract->new(special_ops => [
2960 my ($self, $field, $op, $arg) = @_;
2966 handler => 'method_name',
2970 A "special operator" is a SQL syntactic clause that can be
2971 applied to a field, instead of a usual binary operator.
2974 WHERE field IN (?, ?, ?)
2975 WHERE field BETWEEN ? AND ?
2976 WHERE MATCH(field) AGAINST (?, ?)
2978 Special operators IN and BETWEEN are fairly standard and therefore
2979 are builtin within C<SQL::Abstract> (as the overridable methods
2980 C<_where_field_IN> and C<_where_field_BETWEEN>). For other operators,
2981 like the MATCH .. AGAINST example above which is specific to MySQL,
2982 you can write your own operator handlers - supply a C<special_ops>
2983 argument to the C<new> method. That argument takes an arrayref of
2984 operator definitions; each operator definition is a hashref with two
2991 the regular expression to match the operator
2995 Either a coderef or a plain scalar method name. In both cases
2996 the expected return is C<< ($sql, @bind) >>.
2998 When supplied with a method name, it is simply called on the
2999 L<SQL::Abstract> object as:
3001 $self->$method_name($field, $op, $arg)
3005 $field is the LHS of the operator
3006 $op is the part that matched the handler regex
3009 When supplied with a coderef, it is called as:
3011 $coderef->($self, $field, $op, $arg)
3016 For example, here is an implementation
3017 of the MATCH .. AGAINST syntax for MySQL
3019 my $sqlmaker = SQL::Abstract->new(special_ops => [
3021 # special op for MySql MATCH (field) AGAINST(word1, word2, ...)
3022 {regex => qr/^match$/i,
3024 my ($self, $field, $op, $arg) = @_;
3025 $arg = [$arg] if not ref $arg;
3026 my $label = $self->_quote($field);
3027 my ($placeholder) = $self->_convert('?');
3028 my $placeholders = join ", ", (($placeholder) x @$arg);
3029 my $sql = $self->_sqlcase('match') . " ($label) "
3030 . $self->_sqlcase('against') . " ($placeholders) ";
3031 my @bind = $self->_bindtype($field, @$arg);
3032 return ($sql, @bind);
3039 =head1 UNARY OPERATORS
3041 my $sqlmaker = SQL::Abstract->new(unary_ops => [
3045 my ($self, $op, $arg) = @_;
3051 handler => 'method_name',
3055 A "unary operator" is a SQL syntactic clause that can be
3056 applied to a field - the operator goes before the field
3058 You can write your own operator handlers - supply a C<unary_ops>
3059 argument to the C<new> method. That argument takes an arrayref of
3060 operator definitions; each operator definition is a hashref with two
3067 the regular expression to match the operator
3071 Either a coderef or a plain scalar method name. In both cases
3072 the expected return is C<< $sql >>.
3074 When supplied with a method name, it is simply called on the
3075 L<SQL::Abstract> object as:
3077 $self->$method_name($op, $arg)
3081 $op is the part that matched the handler regex
3082 $arg is the RHS or argument of the operator
3084 When supplied with a coderef, it is called as:
3086 $coderef->($self, $op, $arg)
3094 Thanks to some benchmarking by Mark Stosberg, it turns out that
3095 this module is many orders of magnitude faster than using C<DBIx::Abstract>.
3096 I must admit this wasn't an intentional design issue, but it's a
3097 byproduct of the fact that you get to control your C<DBI> handles
3100 To maximize performance, use a code snippet like the following:
3102 # prepare a statement handle using the first row
3103 # and then reuse it for the rest of the rows
3105 for my $href (@array_of_hashrefs) {
3106 $stmt ||= $sql->insert('table', $href);
3107 $sth ||= $dbh->prepare($stmt);
3108 $sth->execute($sql->values($href));
3111 The reason this works is because the keys in your C<$href> are sorted
3112 internally by B<SQL::Abstract>. Thus, as long as your data retains
3113 the same structure, you only have to generate the SQL the first time
3114 around. On subsequent queries, simply use the C<values> function provided
3115 by this module to return your values in the correct order.
3117 However this depends on the values having the same type - if, for
3118 example, the values of a where clause may either have values
3119 (resulting in sql of the form C<column = ?> with a single bind
3120 value), or alternatively the values might be C<undef> (resulting in
3121 sql of the form C<column IS NULL> with no bind value) then the
3122 caching technique suggested will not work.
3126 If you use my C<CGI::FormBuilder> module at all, you'll hopefully
3127 really like this part (I do, at least). Building up a complex query
3128 can be as simple as the following:
3135 use CGI::FormBuilder;
3138 my $form = CGI::FormBuilder->new(...);
3139 my $sql = SQL::Abstract->new;
3141 if ($form->submitted) {
3142 my $field = $form->field;
3143 my $id = delete $field->{id};
3144 my($stmt, @bind) = $sql->update('table', $field, {id => $id});
3147 Of course, you would still have to connect using C<DBI> to run the
3148 query, but the point is that if you make your form look like your
3149 table, the actual query script can be extremely simplistic.
3151 If you're B<REALLY> lazy (I am), check out C<HTML::QuickTable> for
3152 a fast interface to returning and formatting data. I frequently
3153 use these three modules together to write complex database query
3154 apps in under 50 lines.
3156 =head1 HOW TO CONTRIBUTE
3158 Contributions are always welcome, in all usable forms (we especially
3159 welcome documentation improvements). The delivery methods include git-
3160 or unified-diff formatted patches, GitHub pull requests, or plain bug
3161 reports either via RT or the Mailing list. Contributors are generally
3162 granted full access to the official repository after their first several
3163 patches pass successful review.
3165 This project is maintained in a git repository. The code and related tools are
3166 accessible at the following locations:
3170 =item * Official repo: L<git://git.shadowcat.co.uk/dbsrgits/SQL-Abstract.git>
3172 =item * Official gitweb: L<http://git.shadowcat.co.uk/gitweb/gitweb.cgi?p=dbsrgits/SQL-Abstract.git>
3174 =item * GitHub mirror: L<https://github.com/dbsrgits/sql-abstract>
3176 =item * Authorized committers: L<ssh://dbsrgits@git.shadowcat.co.uk/SQL-Abstract.git>
3182 Version 1.50 was a major internal refactoring of C<SQL::Abstract>.
3183 Great care has been taken to preserve the I<published> behavior
3184 documented in previous versions in the 1.* family; however,
3185 some features that were previously undocumented, or behaved
3186 differently from the documentation, had to be changed in order
3187 to clarify the semantics. Hence, client code that was relying
3188 on some dark areas of C<SQL::Abstract> v1.*
3189 B<might behave differently> in v1.50.
3191 The main changes are:
3197 support for literal SQL through the C<< \ [ $sql, @bind ] >> syntax.
3201 support for the { operator => \"..." } construct (to embed literal SQL)
3205 support for the { operator => \["...", @bind] } construct (to embed literal SQL with bind values)
3209 optional support for L<array datatypes|/"Inserting and Updating Arrays">
3213 defensive programming: check arguments
3217 fixed bug with global logic, which was previously implemented
3218 through global variables yielding side-effects. Prior versions would
3219 interpret C<< [ {cond1, cond2}, [cond3, cond4] ] >>
3220 as C<< "(cond1 AND cond2) OR (cond3 AND cond4)" >>.
3221 Now this is interpreted
3222 as C<< "(cond1 AND cond2) OR (cond3 OR cond4)" >>.
3227 fixed semantics of _bindtype on array args
3231 dropped the C<_anoncopy> of the %where tree. No longer necessary,
3232 we just avoid shifting arrays within that tree.
3236 dropped the C<_modlogic> function
3240 =head1 ACKNOWLEDGEMENTS
3242 There are a number of individuals that have really helped out with
3243 this module. Unfortunately, most of them submitted bugs via CPAN
3244 so I have no idea who they are! But the people I do know are:
3246 Ash Berlin (order_by hash term support)
3247 Matt Trout (DBIx::Class support)
3248 Mark Stosberg (benchmarking)
3249 Chas Owens (initial "IN" operator support)
3250 Philip Collins (per-field SQL functions)
3251 Eric Kolve (hashref "AND" support)
3252 Mike Fragassi (enhancements to "BETWEEN" and "LIKE")
3253 Dan Kubb (support for "quote_char" and "name_sep")
3254 Guillermo Roditi (patch to cleanup "IN" and "BETWEEN", fix and tests for _order_by)
3255 Laurent Dami (internal refactoring, extensible list of special operators, literal SQL)
3256 Norbert Buchmuller (support for literal SQL in hashpair, misc. fixes & tests)
3257 Peter Rabbitson (rewrite of SQLA::Test, misc. fixes & tests)
3258 Oliver Charles (support for "RETURNING" after "INSERT")
3264 L<DBIx::Class>, L<DBIx::Abstract>, L<CGI::FormBuilder>, L<HTML::QuickTable>.
3268 Copyright (c) 2001-2007 Nathan Wiger <nwiger@cpan.org>. All Rights Reserved.
3270 This module is actively maintained by Matt Trout <mst@shadowcatsystems.co.uk>
3272 For support, your best bet is to try the C<DBIx::Class> users mailing list.
3273 While not an official support venue, C<DBIx::Class> makes heavy use of
3274 C<SQL::Abstract>, and as such list members there are very familiar with
3275 how to create queries.
3279 This module is free software; you may copy this under the same
3280 terms as perl itself (either the GNU General Public License or
3281 the Artistic License)