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" }},
45 #======================================================================
46 # DEBUGGING AND ERROR REPORTING
47 #======================================================================
50 return unless $_[0]->{debug}; shift; # a little faster
51 my $func = (caller(1))[3];
52 warn "[$func] ", @_, "\n";
56 my($func) = (caller(1))[3];
57 Carp::carp "[$func] Warning: ", @_;
61 my($func) = (caller(1))[3];
62 Carp::croak "[$func] Fatal: ", @_;
65 sub is_literal_value ($) {
66 ref $_[0] eq 'SCALAR' ? [ ${$_[0]} ]
67 : ( ref $_[0] eq 'REF' and ref ${$_[0]} eq 'ARRAY' ) ? [ @${ $_[0] } ]
71 sub is_undef_value ($) {
75 and exists $_[0]->{-value}
76 and not defined $_[0]->{-value}
80 # FIXME XSify - this can be done so much more efficiently
81 sub is_plain_value ($) {
83 ! length ref $_[0] ? \($_[0])
85 ref $_[0] eq 'HASH' and keys %{$_[0]} == 1
87 exists $_[0]->{-value}
88 ) ? \($_[0]->{-value})
90 # reuse @_ for even moar speedz
91 defined ( $_[1] = Scalar::Util::blessed $_[0] )
93 # deliberately not using Devel::OverloadInfo - the checks we are
94 # intersted in are much more limited than the fullblown thing, and
95 # this is a very hot piece of code
97 # simply using ->can('(""') can leave behind stub methods that
98 # break actually using the overload later (see L<perldiag/Stub
99 # found while resolving method "%s" overloading "%s" in package
100 # "%s"> and the source of overload::mycan())
102 # either has stringification which DBI SHOULD prefer out of the box
103 grep { *{ (qq[${_}::(""]) }{CODE} } @{ $_[2] = mro::get_linear_isa( $_[1] ) }
105 # has nummification or boolification, AND fallback is *not* disabled
107 SQL::Abstract::_ENV_::DETECT_AUTOGENERATED_STRINGIFICATION
110 grep { *{"${_}::(0+"}{CODE} } @{$_[2]}
112 grep { *{"${_}::(bool"}{CODE} } @{$_[2]}
116 # no fallback specified at all
117 ! ( ($_[3]) = grep { *{"${_}::()"}{CODE} } @{$_[2]} )
119 # fallback explicitly undef
120 ! defined ${"$_[3]::()"}
133 #======================================================================
135 #======================================================================
139 my $class = ref($self) || $self;
140 my %opt = (ref $_[0] eq 'HASH') ? %{$_[0]} : @_;
142 # choose our case by keeping an option around
143 delete $opt{case} if $opt{case} && $opt{case} ne 'lower';
145 # default logic for interpreting arrayrefs
146 $opt{logic} = $opt{logic} ? uc $opt{logic} : 'OR';
148 # how to return bind vars
149 $opt{bindtype} ||= 'normal';
151 # default comparison is "=", but can be overridden
154 # try to recognize which are the 'equality' and 'inequality' ops
155 # (temporary quickfix (in 2007), should go through a more seasoned API)
156 $opt{equality_op} = qr/^( \Q$opt{cmp}\E | \= )$/ix;
157 $opt{inequality_op} = qr/^( != | <> )$/ix;
159 $opt{like_op} = qr/^ (is_)?r?like $/xi;
160 $opt{not_like_op} = qr/^ (is_)?not_r?like $/xi;
163 $opt{sqltrue} ||= '1=1';
164 $opt{sqlfalse} ||= '0=1';
167 $opt{special_ops} ||= [];
169 # regexes are applied in order, thus push after user-defines
170 push @{$opt{special_ops}}, @BUILTIN_SPECIAL_OPS;
172 if ($class->isa('DBIx::Class::SQLMaker')) {
173 push @{$opt{special_ops}}, our $DBIC_Compat_Op ||= {
174 regex => qr/^(?:ident|value|(?:not\s)?in)$/i, handler => sub { die "NOPE" }
176 $opt{is_dbic_sqlmaker} = 1;
180 $opt{unary_ops} ||= [];
182 # rudimentary sanity-check for user supplied bits treated as functions/operators
183 # If a purported function matches this regular expression, an exception is thrown.
184 # Literal SQL is *NOT* subject to this check, only functions (and column names
185 # when quoting is not in effect)
188 # need to guard against ()'s in column names too, but this will break tons of
189 # hacks... ideas anyone?
190 $opt{injection_guard} ||= qr/
196 $opt{expand_unary} = {};
199 -not => '_expand_not',
200 -bool => '_expand_bool',
201 -and => '_expand_op_andor',
202 -or => '_expand_op_andor',
203 -nest => '_expand_nest',
204 -bind => sub { shift; +{ @_ } },
206 -not_in => '_expand_in',
208 my ($self, $node, $args) = @_;
209 +{ $node => [ map $self->expand_expr($_), @$args ] };
214 'between' => '_expand_between',
215 'not_between' => '_expand_between',
216 'in' => '_expand_in',
217 'not_in' => '_expand_in',
218 'nest' => '_expand_nest',
219 (map +($_ => '_expand_op_andor'), ('and', 'or')),
220 (map +($_ => '_expand_op_is'), ('is', 'is_not')),
223 # placeholder for _expand_unop system
225 my %unops = (-ident => '_expand_ident', -value => '_expand_value');
226 foreach my $name (keys %unops) {
227 $opt{expand}{$name} = $unops{$name};
228 my ($op) = $name =~ /^-(.*)$/;
229 $opt{expand_op}{$op} = sub {
230 my ($self, $op, $arg, $k) = @_;
231 return $self->_expand_expr_hashpair_cmp(
232 $k, { "-${op}" => $arg }
239 (map +("-$_", "_render_$_"), qw(op func bind ident literal list)),
244 (map +($_ => '_render_op_between'), 'between', 'not_between'),
245 (map +($_ => '_render_op_in'), 'in', 'not_in'),
246 (map +($_ => '_render_unop_postfix'),
247 'is_null', 'is_not_null', 'asc', 'desc',
249 (not => '_render_op_not'),
250 (map +($_ => '_render_op_andor'), qw(and or)),
251 ',' => sub { shift->_render_op_multop(@_, 1) },
254 return bless \%opt, $class;
257 sub sqltrue { +{ -literal => [ $_[0]->{sqltrue} ] } }
258 sub sqlfalse { +{ -literal => [ $_[0]->{sqlfalse} ] } }
260 sub _assert_pass_injection_guard {
261 if ($_[1] =~ $_[0]->{injection_guard}) {
262 my $class = ref $_[0];
263 puke "Possible SQL injection attempt '$_[1]'. If this is indeed a part of the "
264 . "desired SQL use literal SQL ( \'...' or \[ '...' ] ) or supply your own "
265 . "{injection_guard} attribute to ${class}->new()"
270 #======================================================================
272 #======================================================================
276 my $table = $self->_table(shift);
277 my $data = shift || return;
280 my $method = $self->_METHOD_FOR_refkind("_insert", $data);
281 my ($sql, @bind) = $self->$method($data);
282 $sql = join " ", $self->_sqlcase('insert into'), $table, $sql;
284 if ($options->{returning}) {
285 my ($s, @b) = $self->_insert_returning($options);
290 return wantarray ? ($sql, @bind) : $sql;
293 # So that subclasses can override INSERT ... RETURNING separately from
294 # UPDATE and DELETE (e.g. DBIx::Class::SQLMaker::Oracle does this)
295 sub _insert_returning { shift->_returning(@_) }
298 my ($self, $options) = @_;
300 my $f = $options->{returning};
302 my ($sql, @bind) = $self->render_aqt(
303 $self->_expand_maybe_list_expr($f, -ident)
306 ? $self->_sqlcase(' returning ') . $sql
307 : ($self->_sqlcase(' returning ').$sql, @bind);
310 sub _insert_HASHREF { # explicit list of fields and then values
311 my ($self, $data) = @_;
313 my @fields = sort keys %$data;
315 my ($sql, @bind) = $self->_insert_values($data);
318 $_ = $self->_quote($_) foreach @fields;
319 $sql = "( ".join(", ", @fields).") ".$sql;
321 return ($sql, @bind);
324 sub _insert_ARRAYREF { # just generate values(?,?) part (no list of fields)
325 my ($self, $data) = @_;
327 # no names (arrayref) so can't generate bindtype
328 $self->{bindtype} ne 'columns'
329 or belch "can't do 'columns' bindtype when called with arrayref";
331 my (@values, @all_bind);
332 foreach my $value (@$data) {
333 my ($values, @bind) = $self->_insert_value(undef, $value);
334 push @values, $values;
335 push @all_bind, @bind;
337 my $sql = $self->_sqlcase('values')." ( ".join(", ", @values)." )";
338 return ($sql, @all_bind);
341 sub _insert_ARRAYREFREF { # literal SQL with bind
342 my ($self, $data) = @_;
344 my ($sql, @bind) = @${$data};
345 $self->_assert_bindval_matches_bindtype(@bind);
347 return ($sql, @bind);
351 sub _insert_SCALARREF { # literal SQL without bind
352 my ($self, $data) = @_;
358 my ($self, $data) = @_;
360 my (@values, @all_bind);
361 foreach my $column (sort keys %$data) {
362 my ($values, @bind) = $self->_insert_value($column, $data->{$column});
363 push @values, $values;
364 push @all_bind, @bind;
366 my $sql = $self->_sqlcase('values')." ( ".join(", ", @values)." )";
367 return ($sql, @all_bind);
371 my ($self, $column, $v) = @_;
373 return $self->render_aqt(
374 $self->_expand_insert_value($column, $v)
378 sub _expand_insert_value {
379 my ($self, $column, $v) = @_;
381 if (ref($v) eq 'ARRAY') {
382 if ($self->{array_datatypes}) {
383 return +{ -bind => [ $column, $v ] };
385 my ($sql, @bind) = @$v;
386 $self->_assert_bindval_matches_bindtype(@bind);
387 return +{ -literal => $v };
389 if (ref($v) eq 'HASH') {
390 if (grep !/^-/, keys %$v) {
391 belch "HASH ref as bind value in insert is not supported";
392 return +{ -bind => [ $column, $v ] };
396 return +{ -bind => [ $column, undef ] };
398 local our $Cur_Col_Meta = $column;
399 return $self->expand_expr($v);
404 #======================================================================
406 #======================================================================
411 my $table = $self->_table(shift);
412 my $data = shift || return;
416 # first build the 'SET' part of the sql statement
417 puke "Unsupported data type specified to \$sql->update"
418 unless ref $data eq 'HASH';
420 my ($sql, @all_bind) = $self->_update_set_values($data);
421 $sql = $self->_sqlcase('update ') . $table . $self->_sqlcase(' set ')
425 my($where_sql, @where_bind) = $self->where($where);
427 push @all_bind, @where_bind;
430 if ($options->{returning}) {
431 my ($returning_sql, @returning_bind) = $self->_update_returning($options);
432 $sql .= $returning_sql;
433 push @all_bind, @returning_bind;
436 return wantarray ? ($sql, @all_bind) : $sql;
439 sub _update_set_values {
440 my ($self, $data) = @_;
442 return $self->render_aqt(
443 $self->_expand_update_set_values($data),
447 sub _expand_update_set_values {
448 my ($self, $data) = @_;
449 $self->_expand_maybe_list_expr( [
452 $set = { -bind => $_ } unless defined $set;
453 +{ -op => [ '=', $self->_expand_ident(-ident => $k), $set ] };
459 ? ($self->{array_datatypes}
460 ? [ $k, +{ -bind => [ $k, $v ] } ]
461 : [ $k, +{ -literal => $v } ])
463 local our $Cur_Col_Meta = $k;
464 [ $k, $self->_expand_expr($v) ]
471 # So that subclasses can override UPDATE ... RETURNING separately from
473 sub _update_returning { shift->_returning(@_) }
477 #======================================================================
479 #======================================================================
484 my $table = $self->_table(shift);
485 my $fields = shift || '*';
489 my ($fields_sql, @bind) = $self->_select_fields($fields);
491 my ($where_sql, @where_bind) = $self->where($where, $order);
492 push @bind, @where_bind;
494 my $sql = join(' ', $self->_sqlcase('select'), $fields_sql,
495 $self->_sqlcase('from'), $table)
498 return wantarray ? ($sql, @bind) : $sql;
502 my ($self, $fields) = @_;
503 return $fields unless ref($fields);
504 return $self->render_aqt(
505 $self->_expand_maybe_list_expr($fields, '-ident')
509 #======================================================================
511 #======================================================================
516 my $table = $self->_table(shift);
520 my($where_sql, @bind) = $self->where($where);
521 my $sql = $self->_sqlcase('delete from ') . $table . $where_sql;
523 if ($options->{returning}) {
524 my ($returning_sql, @returning_bind) = $self->_delete_returning($options);
525 $sql .= $returning_sql;
526 push @bind, @returning_bind;
529 return wantarray ? ($sql, @bind) : $sql;
532 # So that subclasses can override DELETE ... RETURNING separately from
534 sub _delete_returning { shift->_returning(@_) }
538 #======================================================================
540 #======================================================================
544 # Finally, a separate routine just to handle WHERE clauses
546 my ($self, $where, $order) = @_;
548 local $self->{convert_where} = $self->{convert};
551 my ($sql, @bind) = defined($where)
552 ? $self->_recurse_where($where)
554 $sql = (defined $sql and length $sql) ? $self->_sqlcase(' where ') . "( $sql )" : '';
558 my ($order_sql, @order_bind) = $self->_order_by($order);
560 push @bind, @order_bind;
563 return wantarray ? ($sql, @bind) : $sql;
566 { our $Default_Scalar_To = -value }
569 my ($self, $expr, $default_scalar_to) = @_;
570 local our $Default_Scalar_To = $default_scalar_to if $default_scalar_to;
571 $self->_expand_expr($expr);
575 my ($self, $aqt) = @_;
576 my ($k, $v, @rest) = %$aqt;
578 if (my $meth = $self->{render}{$k}) {
579 return $self->$meth($v);
581 die "notreached: $k";
585 my ($self, $expr) = @_;
586 $self->render_aqt($self->expand_expr($expr));
590 my ($self, $raw) = @_;
591 s/^-(?=.)//, s/\s+/_/g for my $op = lc $raw;
596 my ($self, $expr) = @_;
597 our $Expand_Depth ||= 0; local $Expand_Depth = $Expand_Depth + 1;
598 return undef unless defined($expr);
599 if (ref($expr) eq 'HASH') {
600 return undef unless my $kc = keys %$expr;
602 return $self->_expand_op_andor(-and => $expr);
604 my ($key, $value) = %$expr;
605 if ($key =~ /^-/ and $key =~ s/ [_\s]? \d+ $//x ) {
606 belch 'Use of [and|or|nest]_N modifiers is deprecated and will be removed in SQLA v2.0. '
607 . "You probably wanted ...-and => [ $key => COND1, $key => COND2 ... ]";
609 if (my $exp = $self->{expand}{$key}) {
610 return $self->$exp($key, $value);
612 return $self->_expand_expr_hashpair($key, $value);
614 if (ref($expr) eq 'ARRAY') {
615 my $logic = '-'.lc($self->{logic});
616 return $self->_expand_op_andor($logic, $expr);
618 if (my $literal = is_literal_value($expr)) {
619 return +{ -literal => $literal };
621 if (!ref($expr) or Scalar::Util::blessed($expr)) {
622 return $self->_expand_expr_scalar($expr);
627 sub _expand_expr_hashpair {
628 my ($self, $k, $v) = @_;
629 unless (defined($k) and length($k)) {
630 if (defined($k) and my $literal = is_literal_value($v)) {
631 belch 'Hash-pairs consisting of an empty string with a literal are deprecated, and will be removed in 2.0: use -and => [ $literal ] instead';
632 return { -literal => $literal };
634 puke "Supplying an empty left hand side argument is not supported";
637 return $self->_expand_expr_hashpair_op($k, $v);
639 return $self->_expand_expr_hashpair_ident($k, $v);
642 sub _expand_expr_hashpair_ident {
643 my ($self, $k, $v) = @_;
645 local our $Cur_Col_Meta = $k;
647 # hash with multiple or no elements is andor
649 if (ref($v) eq 'HASH' and keys %$v != 1) {
650 return $self->_expand_op_andor(-and => $v, $k);
653 # undef needs to be re-sent with cmp to achieve IS/IS NOT NULL
655 if (is_undef_value($v)) {
656 return $self->_expand_expr_hashpair_cmp($k => undef);
659 # scalars and objects get expanded as whatever requested or values
661 if (!ref($v) or Scalar::Util::blessed($v)) {
662 return $self->_expand_expr_hashpair_scalar($k, $v);
665 # single key hashref is a hashtriple
667 if (ref($v) eq 'HASH') {
668 return $self->_expand_expr_hashtriple($k, %$v);
671 # arrayref needs re-engineering over the elements
673 if (ref($v) eq 'ARRAY') {
674 return $self->sqlfalse unless @$v;
675 $self->_debug("ARRAY($k) means distribute over elements");
677 $v->[0] =~ /^-(and|or)$/i
678 ? shift(@{$v = [ @$v ]})
679 : '-'.lc($self->{logic} || 'OR')
681 return $self->_expand_op_andor(
686 if (my $literal = is_literal_value($v)) {
688 belch 'Hash-pairs consisting of an empty string with a literal are deprecated, and will be removed in 2.0: use -and => [ $literal ] instead';
691 my ($sql, @bind) = @$literal;
692 if ($self->{bindtype} eq 'columns') {
694 $self->_assert_bindval_matches_bindtype($_);
697 return +{ -literal => [ $self->_quote($k).' '.$sql, @bind ] };
702 sub _expand_expr_scalar {
703 my ($self, $expr) = @_;
705 return $self->_expand_expr({ (our $Default_Scalar_To) => $expr });
708 sub _expand_expr_hashpair_scalar {
709 my ($self, $k, $v) = @_;
711 return $self->_expand_expr_hashpair_cmp(
712 $k, $self->_expand_expr_scalar($v),
716 sub _expand_expr_hashpair_op {
717 my ($self, $k, $v) = @_;
719 $self->_assert_pass_injection_guard($k =~ /\A-(.*)\Z/s);
721 my $op = $self->_normalize_op($k);
723 # Ops prefixed with -not_ get converted
725 if (my ($rest) = $op =~/^not_(.*)$/) {
728 $self->_expand_expr({ "-${rest}", $v })
734 my $op = join(' ', split '_', $op);
736 # the old special op system requires illegality for top-level use
739 (our $Expand_Depth) == 1
740 and List::Util::first { $op =~ $_->{regex} } @{$self->{special_ops}}
742 puke "Illegal use of top-level '-$op'"
745 # the old unary op system means we should touch nothing and let it work
747 if (my $us = List::Util::first { $op =~ $_->{regex} } @{$self->{unary_ops}}) {
748 return { -op => [ $op, $v ] };
752 # an explicit node type is currently assumed to be expanded (this is almost
753 # certainly wrong and there should be expansion anyway)
755 if ($self->{render}{$k}) {
759 # hashref RHS values get expanded and used as op/func args
764 and (keys %$v)[0] =~ /^-/
766 my ($func) = $k =~ /^-(.*)$/;
768 if (List::Util::first { $func =~ $_->{regex} } @{$self->{special_ops}}) {
769 return +{ -op => [ $func, $self->_expand_expr($v) ] };
772 return +{ -func => [ $func, $self->_expand_expr($v) ] };
775 # scalars and literals get simply expanded
777 if (!ref($v) or is_literal_value($v)) {
778 return +{ -op => [ $op, $self->_expand_expr($v) ] };
784 sub _expand_expr_hashpair_cmp {
785 my ($self, $k, $v) = @_;
786 $self->_expand_expr_hashtriple($k, $self->{cmp}, $v);
789 sub _expand_expr_hashtriple {
790 my ($self, $k, $vk, $vv) = @_;
792 my $ik = $self->_expand_ident(-ident => $k);
794 my $op = $self->_normalize_op($vk);
795 $self->_assert_pass_injection_guard($op);
797 if ($op =~ s/ _? \d+ $//x ) {
798 return $self->_expand_expr($k, { $vk, $vv });
800 if (my $x = $self->{expand_op}{$op}) {
801 local our $Cur_Col_Meta = $k;
802 return $self->$x($op, $vv, $k);
806 my $op = join(' ', split '_', $op);
808 if (my $us = List::Util::first { $op =~ $_->{regex} } @{$self->{special_ops}}) {
809 return { -op => [ $op, $ik, $vv ] };
811 if (my $us = List::Util::first { $op =~ $_->{regex} } @{$self->{unary_ops}}) {
815 { -op => [ $op, $vv ] }
819 if (ref($vv) eq 'ARRAY') {
821 my $logic = (defined($raw[0]) and $raw[0] =~ /^-(and|or)$/i)
822 ? shift @raw : '-or';
823 my @values = map +{ $vk => $_ }, @raw;
825 $op =~ $self->{inequality_op}
826 or $op =~ $self->{not_like_op}
828 if (lc($logic) eq '-or' and @values > 1) {
829 belch "A multi-element arrayref as an argument to the inequality op '${\uc(join ' ', split '_', $op)}' "
830 . 'is technically equivalent to an always-true 1=1 (you probably wanted '
831 . "to say ...{ \$inequality_op => [ -and => \@values ] }... instead)"
836 # try to DWIM on equality operators
837 return ($self->_dwim_op_to_is($op,
838 "Supplying an empty arrayref to '%s' is deprecated",
839 "operator '%s' applied on an empty array (field '$k')"
840 ) ? $self->sqlfalse : $self->sqltrue);
842 return $self->_expand_op_andor($logic => \@values, $k);
844 if (is_undef_value($vv)) {
845 my $is = ($self->_dwim_op_to_is($op,
846 "Supplying an undefined argument to '%s' is deprecated",
847 "unexpected operator '%s' with undef operand",
848 ) ? 'is' : 'is not');
850 return $self->_expand_expr_hashpair($k => { $is, undef });
852 local our $Cur_Col_Meta = $k;
856 $self->_expand_expr($vv)
861 my ($self, $raw, $empty, $fail) = @_;
863 my $op = $self->_normalize_op($raw);
865 if ($op =~ /^not$/i) {
868 if ($op =~ $self->{equality_op}) {
871 if ($op =~ $self->{like_op}) {
872 belch(sprintf $empty, uc(join ' ', split '_', $op));
875 if ($op =~ $self->{inequality_op}) {
878 if ($op =~ $self->{not_like_op}) {
879 belch(sprintf $empty, uc(join ' ', split '_', $op));
882 puke(sprintf $fail, $op);
886 my ($self, $op, $body) = @_;
887 unless (defined($body) or (ref($body) and ref($body) eq 'ARRAY')) {
888 puke "$op requires a single plain scalar argument (a quotable identifier) or an arrayref of identifier parts";
890 my @parts = map split(/\Q${\($self->{name_sep}||'.')}\E/, $_),
891 ref($body) ? @$body : $body;
892 return { -ident => $parts[-1] } if $self->{_dequalify_idents};
893 unless ($self->{quote_char}) {
894 $self->_assert_pass_injection_guard($_) for @parts;
896 return +{ -ident => \@parts };
900 +{ -bind => [ our $Cur_Col_Meta, $_[2] ] };
904 +{ -op => [ 'not', $_[0]->_expand_expr($_[2]) ] };
908 my ($self, undef, $v) = @_;
910 return $self->_expand_expr($v);
912 puke "-bool => undef not supported" unless defined($v);
913 return $self->_expand_ident(-ident => $v);
916 sub _expand_op_andor {
917 my ($self, $logic, $v, $k) = @_;
919 $v = [ map +{ $k, $_ },
921 ? (map +{ $_ => $v->{$_} }, sort keys %$v)
925 my ($logop) = $logic =~ /^-?(.*)$/;
926 if (ref($v) eq 'HASH') {
927 return undef unless keys %$v;
930 map $self->_expand_expr({ $_ => $v->{$_} }),
934 if (ref($v) eq 'ARRAY') {
935 $logop eq 'and' or $logop eq 'or' or puke "unknown logic: $logop";
938 (ref($_) eq 'ARRAY' and @$_)
939 or (ref($_) eq 'HASH' and %$_)
945 while (my ($el) = splice @expr, 0, 1) {
946 puke "Supplying an empty left hand side argument is not supported in array-pairs"
947 unless defined($el) and length($el);
948 my $elref = ref($el);
950 local our $Expand_Depth = 0;
951 push(@res, grep defined, $self->_expand_expr({ $el, shift(@expr) }));
952 } elsif ($elref eq 'ARRAY') {
953 push(@res, grep defined, $self->_expand_expr($el)) if @$el;
954 } elsif (my $l = is_literal_value($el)) {
955 push @res, { -literal => $l };
956 } elsif ($elref eq 'HASH') {
957 local our $Expand_Depth = 0;
958 push @res, grep defined, $self->_expand_expr($el) if %$el;
964 # return $res[0] if @res == 1;
965 return { -op => [ $logop, @res ] };
971 my ($self, $op, $vv, $k) = @_;
972 puke "$op can only take undef as argument"
976 and exists($vv->{-value})
977 and !defined($vv->{-value})
979 return +{ -op => [ $op.'_null', $self->_expand_ident(-ident => $k) ] };
982 sub _expand_between {
983 my ($self, $op, $vv, $k) = @_;
984 local our $Cur_Col_Meta = $k;
985 my @rhs = map $self->_expand_expr($_),
986 ref($vv) eq 'ARRAY' ? @$vv : $vv;
988 (@rhs == 1 and ref($rhs[0]) eq 'HASH' and $rhs[0]->{-literal})
990 (@rhs == 2 and defined($rhs[0]) and defined($rhs[1]))
992 puke "Operator '${\uc($op)}' requires either an arrayref with two defined values or expressions, or a single literal scalarref/arrayref-ref";
996 $self->_expand_ident(-ident => $k),
1002 my ($self, $raw, $vv, $k) = @_;
1003 $k = shift @{$vv = [ @$vv ]} unless defined $k;
1004 local our $Cur_Col_Meta = $k;
1005 my $op = $self->_normalize_op($raw);
1006 if (my $literal = is_literal_value($vv)) {
1007 my ($sql, @bind) = @$literal;
1008 my $opened_sql = $self->_open_outer_paren($sql);
1010 $op, $self->expand_expr($k, -ident),
1011 [ { -literal => [ $opened_sql, @bind ] } ]
1015 'SQL::Abstract before v1.75 used to generate incorrect SQL when the '
1016 . "-${\uc($op)} operator was given an undef-containing list: !!!AUDIT YOUR CODE "
1017 . 'AND DATA!!! (the upcoming Data::Query-based version of SQL::Abstract '
1018 . 'will emit the logically correct SQL instead of raising this exception)'
1020 puke("Argument passed to the '${\uc($op)}' operator can not be undefined")
1022 my @rhs = map $self->expand_expr($_, -value),
1023 map { defined($_) ? $_: puke($undef_err) }
1024 (ref($vv) eq 'ARRAY' ? @$vv : $vv);
1025 return $self->${\($op =~ /^not/ ? 'sqltrue' : 'sqlfalse')} unless @rhs;
1029 $self->expand_expr($k, -ident),
1035 my ($self, $op, $v) = @_;
1036 # DBIx::Class requires a nest warning to be emitted once but the private
1037 # method it overrode to do so no longer exists
1038 if ($self->{is_dbic_sqlmaker}) {
1039 unless (our $Nest_Warned) {
1041 "-nest in search conditions is deprecated, you most probably wanted:\n"
1042 .q|{..., -and => [ \%cond0, \@cond1, \'cond2', \[ 'cond3', [ col => bind ] ], etc. ], ... }|
1047 return $self->_expand_expr($v);
1050 sub _recurse_where {
1051 my ($self, $where, $logic) = @_;
1053 # Special case: top level simple string treated as literal
1055 my $where_exp = (ref($where)
1056 ? $self->_expand_expr($where, $logic)
1057 : { -literal => [ $where ] });
1059 # dispatch expanded expression
1061 my ($sql, @bind) = defined($where_exp) ? $self->render_aqt($where_exp) : (undef);
1062 # DBIx::Class used to call _recurse_where in scalar context
1063 # something else might too...
1065 return ($sql, @bind);
1068 belch "Calling _recurse_where in scalar context is deprecated and will go away before 2.0";
1074 my ($self, $ident) = @_;
1076 return $self->_convert($self->_quote($ident));
1080 my ($self, $list) = @_;
1081 my ($sql, @bind) = $self->_render_op([ ',', @$list ]);
1082 return "($sql)", @bind;
1086 my ($self, $rest) = @_;
1087 my ($func, @args) = @$rest;
1091 push @arg_sql, shift @x;
1093 } map [ $self->render_aqt($_) ], @args;
1094 return ($self->_sqlcase($func).'('.join(', ', @arg_sql).')', @bind);
1098 my ($self, $bind) = @_;
1099 return ($self->_convert('?'), $self->_bindtype(@$bind));
1102 sub _render_literal {
1103 my ($self, $literal) = @_;
1104 $self->_assert_bindval_matches_bindtype(@{$literal}[1..$#$literal]);
1109 my ($self, $v) = @_;
1110 my ($op, @args) = @$v;
1111 if (my $r = $self->{render_op}{$op}) {
1112 return $self->$r($op, \@args);
1117 my $op = join(' ', split '_', $op);
1119 my $us = List::Util::first { $op =~ $_->{regex} } @{$self->{special_ops}};
1120 if ($us and @args > 1) {
1121 puke "Special op '${op}' requires first value to be identifier"
1122 unless my ($ident) = map $_->{-ident}, grep ref($_) eq 'HASH', $args[0];
1123 my $k = join(($self->{name_sep}||'.'), @$ident);
1124 local our $Expand_Depth = 1;
1125 return $self->${\($us->{handler})}($k, $op, $args[1]);
1127 if (my $us = List::Util::first { $op =~ $_->{regex} } @{$self->{unary_ops}}) {
1128 return $self->${\($us->{handler})}($op, $args[0]);
1133 return $self->_render_unop_prefix($op, \@args);
1135 return $self->_render_op_multop($op, \@args);
1141 sub _render_op_between {
1142 my ($self, $op, $args) = @_;
1143 my ($left, $low, $high) = @$args;
1144 my ($rhsql, @rhbind) = do {
1146 puke "Single arg to between must be a literal"
1147 unless $low->{-literal};
1150 my ($l, $h) = map [ $self->render_aqt($_) ], $low, $high;
1151 (join(' ', $l->[0], $self->_sqlcase('and'), $h->[0]),
1152 @{$l}[1..$#$l], @{$h}[1..$#$h])
1155 my ($lhsql, @lhbind) = $self->render_aqt($left);
1159 $self->_sqlcase(join ' ', split '_', $op),
1167 my ($self, $op, $args) = @_;
1168 my ($lhs, $rhs) = @$args;
1171 my ($sql, @bind) = $self->render_aqt($_);
1172 push @in_bind, @bind;
1175 my ($lhsql, @lbind) = $self->render_aqt($lhs);
1177 $lhsql.' '.$self->_sqlcase(join ' ', split '_', $op).' ( '
1178 .join(', ', @in_sql)
1184 sub _render_op_andor {
1185 my ($self, $op, $args) = @_;
1186 my @parts = grep length($_->[0]), map [ $self->render_aqt($_) ], @$args;
1187 return '' unless @parts;
1188 return @{$parts[0]} if @parts == 1;
1189 my ($sql, @bind) = $self->_render_op_multop($op, $args);
1190 return '( '.$sql.' )', @bind;
1193 sub _render_op_multop {
1194 my ($self, $op, $args, $strip_left) = @_;
1195 my @parts = grep length($_->[0]), map [ $self->render_aqt($_) ], @$args;
1196 return '' unless @parts;
1197 return @{$parts[0]} if @parts == 1;
1198 my ($final_sql) = join(
1199 ($strip_left ? '' : ' ').$self->_sqlcase(join ' ', split '_', $op).' ',
1204 map @{$_}[1..$#$_], @parts
1207 sub _render_op_not {
1208 my ($self, $op, $v) = @_;
1209 my ($sql, @bind) = $self->_render_unop_prefix($op, $v);
1210 return "(${sql})", @bind;
1213 sub _render_unop_prefix {
1214 my ($self, $op, $v) = @_;
1215 my ($expr_sql, @bind) = $self->render_aqt($v->[0]);
1217 my $op_sql = $self->_sqlcase($op); # join ' ', split '_', $op);
1218 return ("${op_sql} ${expr_sql}", @bind);
1221 sub _render_unop_postfix {
1222 my ($self, $op, $v) = @_;
1223 my ($expr_sql, @bind) = $self->render_aqt($v->[0]);
1224 my $op_sql = $self->_sqlcase(join ' ', split '_', $op);
1225 return ($expr_sql.' '.$op_sql, @bind);
1228 # Some databases (SQLite) treat col IN (1, 2) different from
1229 # col IN ( (1, 2) ). Use this to strip all outer parens while
1230 # adding them back in the corresponding method
1231 sub _open_outer_paren {
1232 my ($self, $sql) = @_;
1234 while (my ($inner) = $sql =~ /^ \s* \( (.*) \) \s* $/xs) {
1236 # there are closing parens inside, need the heavy duty machinery
1237 # to reevaluate the extraction starting from $sql (full reevaluation)
1238 if ($inner =~ /\)/) {
1239 require Text::Balanced;
1241 my (undef, $remainder) = do {
1242 # idiotic design - writes to $@ but *DOES NOT* throw exceptions
1244 Text::Balanced::extract_bracketed($sql, '()', qr/\s*/);
1247 # the entire expression needs to be a balanced bracketed thing
1248 # (after an extract no remainder sans trailing space)
1249 last if defined $remainder and $remainder =~ /\S/;
1259 #======================================================================
1261 #======================================================================
1263 sub _expand_order_by {
1264 my ($self, $arg) = @_;
1266 return unless defined($arg) and not (ref($arg) eq 'ARRAY' and !@$arg);
1268 my $expander = sub {
1269 my ($self, $dir, $expr) = @_;
1270 my @to_expand = ref($expr) eq 'ARRAY' ? @$expr : $expr;
1271 foreach my $arg (@to_expand) {
1275 and grep /^-(asc|desc)$/, keys %$arg
1277 puke "ordering direction hash passed to order by must have exactly one key (-asc or -desc)";
1281 defined($dir) ? { -op => [ $dir =~ /^-?(.*)$/ ,=> $_ ] } : $_
1283 map $self->expand_expr($_, -ident),
1284 map ref($_) eq 'ARRAY' ? @$_ : $_, @to_expand;
1285 return undef unless @exp;
1286 return undef if @exp == 1 and not defined($exp[0]);
1287 return +{ -op => [ ',', @exp ] };
1290 local @{$self->{expand}}{qw(-asc -desc)} = (($expander) x 2);
1292 return $self->$expander(undef, $arg);
1296 my ($self, $arg) = @_;
1298 return '' unless defined(my $expanded = $self->_expand_order_by($arg));
1300 my ($sql, @bind) = $self->render_aqt($expanded);
1302 return '' unless length($sql);
1304 my $final_sql = $self->_sqlcase(' order by ').$sql;
1306 return wantarray ? ($final_sql, @bind) : $final_sql;
1309 # _order_by no longer needs to call this so doesn't but DBIC uses it.
1311 sub _order_by_chunks {
1312 my ($self, $arg) = @_;
1314 return () unless defined(my $expanded = $self->_expand_order_by($arg));
1316 return $self->_chunkify_order_by($expanded);
1319 sub _chunkify_order_by {
1320 my ($self, $expanded) = @_;
1322 return grep length, $self->render_aqt($expanded)
1323 if $expanded->{-ident} or @{$expanded->{-literal}||[]} == 1;
1326 if (ref() eq 'HASH' and $_->{-op} and $_->{-op}[0] eq ',') {
1327 my ($comma, @list) = @{$_->{-op}};
1328 return map $self->_chunkify_order_by($_), @list;
1330 return [ $self->render_aqt($_) ];
1334 #======================================================================
1335 # DATASOURCE (FOR NOW, JUST PLAIN TABLE OR LIST OF TABLES)
1336 #======================================================================
1342 $self->_expand_maybe_list_expr($from, -ident)
1347 #======================================================================
1349 #======================================================================
1351 sub _expand_maybe_list_expr {
1352 my ($self, $expr, $default) = @_;
1353 return +{ -op => [ ',',
1354 map $self->expand_expr($_, $default),
1355 ref($expr) eq 'ARRAY' ? @$expr : $expr
1359 # highly optimized, as it's called way too often
1361 # my ($self, $label) = @_;
1363 return '' unless defined $_[1];
1364 return ${$_[1]} if ref($_[1]) eq 'SCALAR';
1365 puke 'Identifier cannot be hashref' if ref($_[1]) eq 'HASH';
1367 unless ($_[0]->{quote_char}) {
1368 if (ref($_[1]) eq 'ARRAY') {
1369 return join($_[0]->{name_sep}||'.', @{$_[1]});
1371 $_[0]->_assert_pass_injection_guard($_[1]);
1376 my $qref = ref $_[0]->{quote_char};
1378 !$qref ? ($_[0]->{quote_char}, $_[0]->{quote_char})
1379 : ($qref eq 'ARRAY') ? @{$_[0]->{quote_char}}
1380 : puke "Unsupported quote_char format: $_[0]->{quote_char}";
1382 my $esc = $_[0]->{escape_char} || $r;
1384 # parts containing * are naturally unquoted
1386 $_[0]->{name_sep}||'',
1390 : do { (my $n = $_) =~ s/(\Q$esc\E|\Q$r\E)/$esc$1/g; $l . $n . $r }
1392 (ref($_[1]) eq 'ARRAY'
1396 ? split (/\Q$_[0]->{name_sep}\E/, $_[1] )
1404 # Conversion, if applicable
1406 #my ($self, $arg) = @_;
1407 if ($_[0]->{convert_where}) {
1408 return $_[0]->_sqlcase($_[0]->{convert_where}) .'(' . $_[1] . ')';
1415 #my ($self, $col, @vals) = @_;
1416 # called often - tighten code
1417 return $_[0]->{bindtype} eq 'columns'
1418 ? map {[$_[1], $_]} @_[2 .. $#_]
1423 # Dies if any element of @bind is not in [colname => value] format
1424 # if bindtype is 'columns'.
1425 sub _assert_bindval_matches_bindtype {
1426 # my ($self, @bind) = @_;
1428 if ($self->{bindtype} eq 'columns') {
1430 if (!defined $_ || ref($_) ne 'ARRAY' || @$_ != 2) {
1431 puke "bindtype 'columns' selected, you need to pass: [column_name => bind_value]"
1437 sub _join_sql_clauses {
1438 my ($self, $logic, $clauses_aref, $bind_aref) = @_;
1440 if (@$clauses_aref > 1) {
1441 my $join = " " . $self->_sqlcase($logic) . " ";
1442 my $sql = '( ' . join($join, @$clauses_aref) . ' )';
1443 return ($sql, @$bind_aref);
1445 elsif (@$clauses_aref) {
1446 return ($clauses_aref->[0], @$bind_aref); # no parentheses
1449 return (); # if no SQL, ignore @$bind_aref
1454 # Fix SQL case, if so requested
1456 # LDNOTE: if $self->{case} is true, then it contains 'lower', so we
1457 # don't touch the argument ... crooked logic, but let's not change it!
1458 return $_[0]->{case} ? $_[1] : uc($_[1]);
1462 #======================================================================
1463 # DISPATCHING FROM REFKIND
1464 #======================================================================
1467 my ($self, $data) = @_;
1469 return 'UNDEF' unless defined $data;
1471 # blessed objects are treated like scalars
1472 my $ref = (Scalar::Util::blessed $data) ? '' : ref $data;
1474 return 'SCALAR' unless $ref;
1477 while ($ref eq 'REF') {
1479 $ref = (Scalar::Util::blessed $data) ? '' : ref $data;
1483 return ($ref||'SCALAR') . ('REF' x $n_steps);
1487 my ($self, $data) = @_;
1488 my @try = ($self->_refkind($data));
1489 push @try, 'SCALAR_or_UNDEF' if $try[0] eq 'SCALAR' || $try[0] eq 'UNDEF';
1490 push @try, 'FALLBACK';
1494 sub _METHOD_FOR_refkind {
1495 my ($self, $meth_prefix, $data) = @_;
1498 for (@{$self->_try_refkind($data)}) {
1499 $method = $self->can($meth_prefix."_".$_)
1503 return $method || puke "cannot dispatch on '$meth_prefix' for ".$self->_refkind($data);
1507 sub _SWITCH_refkind {
1508 my ($self, $data, $dispatch_table) = @_;
1511 for (@{$self->_try_refkind($data)}) {
1512 $coderef = $dispatch_table->{$_}
1516 puke "no dispatch entry for ".$self->_refkind($data)
1525 #======================================================================
1526 # VALUES, GENERATE, AUTOLOAD
1527 #======================================================================
1529 # LDNOTE: original code from nwiger, didn't touch code in that section
1530 # I feel the AUTOLOAD stuff should not be the default, it should
1531 # only be activated on explicit demand by user.
1535 my $data = shift || return;
1536 puke "Argument to ", __PACKAGE__, "->values must be a \\%hash"
1537 unless ref $data eq 'HASH';
1540 foreach my $k (sort keys %$data) {
1541 my $v = $data->{$k};
1542 $self->_SWITCH_refkind($v, {
1544 if ($self->{array_datatypes}) { # array datatype
1545 push @all_bind, $self->_bindtype($k, $v);
1547 else { # literal SQL with bind
1548 my ($sql, @bind) = @$v;
1549 $self->_assert_bindval_matches_bindtype(@bind);
1550 push @all_bind, @bind;
1553 ARRAYREFREF => sub { # literal SQL with bind
1554 my ($sql, @bind) = @${$v};
1555 $self->_assert_bindval_matches_bindtype(@bind);
1556 push @all_bind, @bind;
1558 SCALARREF => sub { # literal SQL without bind
1560 SCALAR_or_UNDEF => sub {
1561 push @all_bind, $self->_bindtype($k, $v);
1572 my(@sql, @sqlq, @sqlv);
1576 if ($ref eq 'HASH') {
1577 for my $k (sort keys %$_) {
1580 my $label = $self->_quote($k);
1581 if ($r eq 'ARRAY') {
1582 # literal SQL with bind
1583 my ($sql, @bind) = @$v;
1584 $self->_assert_bindval_matches_bindtype(@bind);
1585 push @sqlq, "$label = $sql";
1587 } elsif ($r eq 'SCALAR') {
1588 # literal SQL without bind
1589 push @sqlq, "$label = $$v";
1591 push @sqlq, "$label = ?";
1592 push @sqlv, $self->_bindtype($k, $v);
1595 push @sql, $self->_sqlcase('set'), join ', ', @sqlq;
1596 } elsif ($ref eq 'ARRAY') {
1597 # unlike insert(), assume these are ONLY the column names, i.e. for SQL
1600 if ($r eq 'ARRAY') { # literal SQL with bind
1601 my ($sql, @bind) = @$v;
1602 $self->_assert_bindval_matches_bindtype(@bind);
1605 } elsif ($r eq 'SCALAR') { # literal SQL without bind
1606 # embedded literal SQL
1613 push @sql, '(' . join(', ', @sqlq) . ')';
1614 } elsif ($ref eq 'SCALAR') {
1618 # strings get case twiddled
1619 push @sql, $self->_sqlcase($_);
1623 my $sql = join ' ', @sql;
1625 # this is pretty tricky
1626 # if ask for an array, return ($stmt, @bind)
1627 # otherwise, s/?/shift @sqlv/ to put it inline
1629 return ($sql, @sqlv);
1631 1 while $sql =~ s/\?/my $d = shift(@sqlv);
1632 ref $d ? $d->[1] : $d/e;
1641 # This allows us to check for a local, then _form, attr
1643 my($name) = $AUTOLOAD =~ /.*::(.+)/;
1644 return $self->generate($name, @_);
1655 SQL::Abstract - Generate SQL from Perl data structures
1661 my $sql = SQL::Abstract->new;
1663 my($stmt, @bind) = $sql->select($source, \@fields, \%where, $order);
1665 my($stmt, @bind) = $sql->insert($table, \%fieldvals || \@values);
1667 my($stmt, @bind) = $sql->update($table, \%fieldvals, \%where);
1669 my($stmt, @bind) = $sql->delete($table, \%where);
1671 # Then, use these in your DBI statements
1672 my $sth = $dbh->prepare($stmt);
1673 $sth->execute(@bind);
1675 # Just generate the WHERE clause
1676 my($stmt, @bind) = $sql->where(\%where, $order);
1678 # Return values in the same order, for hashed queries
1679 # See PERFORMANCE section for more details
1680 my @bind = $sql->values(\%fieldvals);
1684 This module was inspired by the excellent L<DBIx::Abstract>.
1685 However, in using that module I found that what I really wanted
1686 to do was generate SQL, but still retain complete control over my
1687 statement handles and use the DBI interface. So, I set out to
1688 create an abstract SQL generation module.
1690 While based on the concepts used by L<DBIx::Abstract>, there are
1691 several important differences, especially when it comes to WHERE
1692 clauses. I have modified the concepts used to make the SQL easier
1693 to generate from Perl data structures and, IMO, more intuitive.
1694 The underlying idea is for this module to do what you mean, based
1695 on the data structures you provide it. The big advantage is that
1696 you don't have to modify your code every time your data changes,
1697 as this module figures it out.
1699 To begin with, an SQL INSERT is as easy as just specifying a hash
1700 of C<key=value> pairs:
1703 name => 'Jimbo Bobson',
1704 phone => '123-456-7890',
1705 address => '42 Sister Lane',
1706 city => 'St. Louis',
1707 state => 'Louisiana',
1710 The SQL can then be generated with this:
1712 my($stmt, @bind) = $sql->insert('people', \%data);
1714 Which would give you something like this:
1716 $stmt = "INSERT INTO people
1717 (address, city, name, phone, state)
1718 VALUES (?, ?, ?, ?, ?)";
1719 @bind = ('42 Sister Lane', 'St. Louis', 'Jimbo Bobson',
1720 '123-456-7890', 'Louisiana');
1722 These are then used directly in your DBI code:
1724 my $sth = $dbh->prepare($stmt);
1725 $sth->execute(@bind);
1727 =head2 Inserting and Updating Arrays
1729 If your database has array types (like for example Postgres),
1730 activate the special option C<< array_datatypes => 1 >>
1731 when creating the C<SQL::Abstract> object.
1732 Then you may use an arrayref to insert and update database array types:
1734 my $sql = SQL::Abstract->new(array_datatypes => 1);
1736 planets => [qw/Mercury Venus Earth Mars/]
1739 my($stmt, @bind) = $sql->insert('solar_system', \%data);
1743 $stmt = "INSERT INTO solar_system (planets) VALUES (?)"
1745 @bind = (['Mercury', 'Venus', 'Earth', 'Mars']);
1748 =head2 Inserting and Updating SQL
1750 In order to apply SQL functions to elements of your C<%data> you may
1751 specify a reference to an arrayref for the given hash value. For example,
1752 if you need to execute the Oracle C<to_date> function on a value, you can
1753 say something like this:
1757 date_entered => \[ "to_date(?,'MM/DD/YYYY')", "03/02/2003" ],
1760 The first value in the array is the actual SQL. Any other values are
1761 optional and would be included in the bind values array. This gives
1764 my($stmt, @bind) = $sql->insert('people', \%data);
1766 $stmt = "INSERT INTO people (name, date_entered)
1767 VALUES (?, to_date(?,'MM/DD/YYYY'))";
1768 @bind = ('Bill', '03/02/2003');
1770 An UPDATE is just as easy, all you change is the name of the function:
1772 my($stmt, @bind) = $sql->update('people', \%data);
1774 Notice that your C<%data> isn't touched; the module will generate
1775 the appropriately quirky SQL for you automatically. Usually you'll
1776 want to specify a WHERE clause for your UPDATE, though, which is
1777 where handling C<%where> hashes comes in handy...
1779 =head2 Complex where statements
1781 This module can generate pretty complicated WHERE statements
1782 easily. For example, simple C<key=value> pairs are taken to mean
1783 equality, and if you want to see if a field is within a set
1784 of values, you can use an arrayref. Let's say we wanted to
1785 SELECT some data based on this criteria:
1788 requestor => 'inna',
1789 worker => ['nwiger', 'rcwe', 'sfz'],
1790 status => { '!=', 'completed' }
1793 my($stmt, @bind) = $sql->select('tickets', '*', \%where);
1795 The above would give you something like this:
1797 $stmt = "SELECT * FROM tickets WHERE
1798 ( requestor = ? ) AND ( status != ? )
1799 AND ( worker = ? OR worker = ? OR worker = ? )";
1800 @bind = ('inna', 'completed', 'nwiger', 'rcwe', 'sfz');
1802 Which you could then use in DBI code like so:
1804 my $sth = $dbh->prepare($stmt);
1805 $sth->execute(@bind);
1811 The methods are simple. There's one for every major SQL operation,
1812 and a constructor you use first. The arguments are specified in a
1813 similar order for each method (table, then fields, then a where
1814 clause) to try and simplify things.
1816 =head2 new(option => 'value')
1818 The C<new()> function takes a list of options and values, and returns
1819 a new B<SQL::Abstract> object which can then be used to generate SQL
1820 through the methods below. The options accepted are:
1826 If set to 'lower', then SQL will be generated in all lowercase. By
1827 default SQL is generated in "textbook" case meaning something like:
1829 SELECT a_field FROM a_table WHERE some_field LIKE '%someval%'
1831 Any setting other than 'lower' is ignored.
1835 This determines what the default comparison operator is. By default
1836 it is C<=>, meaning that a hash like this:
1838 %where = (name => 'nwiger', email => 'nate@wiger.org');
1840 Will generate SQL like this:
1842 WHERE name = 'nwiger' AND email = 'nate@wiger.org'
1844 However, you may want loose comparisons by default, so if you set
1845 C<cmp> to C<like> you would get SQL such as:
1847 WHERE name like 'nwiger' AND email like 'nate@wiger.org'
1849 You can also override the comparison on an individual basis - see
1850 the huge section on L</"WHERE CLAUSES"> at the bottom.
1852 =item sqltrue, sqlfalse
1854 Expressions for inserting boolean values within SQL statements.
1855 By default these are C<1=1> and C<1=0>. They are used
1856 by the special operators C<-in> and C<-not_in> for generating
1857 correct SQL even when the argument is an empty array (see below).
1861 This determines the default logical operator for multiple WHERE
1862 statements in arrays or hashes. If absent, the default logic is "or"
1863 for arrays, and "and" for hashes. This means that a WHERE
1867 event_date => {'>=', '2/13/99'},
1868 event_date => {'<=', '4/24/03'},
1871 will generate SQL like this:
1873 WHERE event_date >= '2/13/99' OR event_date <= '4/24/03'
1875 This is probably not what you want given this query, though (look
1876 at the dates). To change the "OR" to an "AND", simply specify:
1878 my $sql = SQL::Abstract->new(logic => 'and');
1880 Which will change the above C<WHERE> to:
1882 WHERE event_date >= '2/13/99' AND event_date <= '4/24/03'
1884 The logic can also be changed locally by inserting
1885 a modifier in front of an arrayref:
1887 @where = (-and => [event_date => {'>=', '2/13/99'},
1888 event_date => {'<=', '4/24/03'} ]);
1890 See the L</"WHERE CLAUSES"> section for explanations.
1894 This will automatically convert comparisons using the specified SQL
1895 function for both column and value. This is mostly used with an argument
1896 of C<upper> or C<lower>, so that the SQL will have the effect of
1897 case-insensitive "searches". For example, this:
1899 $sql = SQL::Abstract->new(convert => 'upper');
1900 %where = (keywords => 'MaKe iT CAse inSeNSItive');
1902 Will turn out the following SQL:
1904 WHERE upper(keywords) like upper('MaKe iT CAse inSeNSItive')
1906 The conversion can be C<upper()>, C<lower()>, or any other SQL function
1907 that can be applied symmetrically to fields (actually B<SQL::Abstract> does
1908 not validate this option; it will just pass through what you specify verbatim).
1912 This is a kludge because many databases suck. For example, you can't
1913 just bind values using DBI's C<execute()> for Oracle C<CLOB> or C<BLOB> fields.
1914 Instead, you have to use C<bind_param()>:
1916 $sth->bind_param(1, 'reg data');
1917 $sth->bind_param(2, $lots, {ora_type => ORA_CLOB});
1919 The problem is, B<SQL::Abstract> will normally just return a C<@bind> array,
1920 which loses track of which field each slot refers to. Fear not.
1922 If you specify C<bindtype> in new, you can determine how C<@bind> is returned.
1923 Currently, you can specify either C<normal> (default) or C<columns>. If you
1924 specify C<columns>, you will get an array that looks like this:
1926 my $sql = SQL::Abstract->new(bindtype => 'columns');
1927 my($stmt, @bind) = $sql->insert(...);
1930 [ 'column1', 'value1' ],
1931 [ 'column2', 'value2' ],
1932 [ 'column3', 'value3' ],
1935 You can then iterate through this manually, using DBI's C<bind_param()>.
1937 $sth->prepare($stmt);
1940 my($col, $data) = @$_;
1941 if ($col eq 'details' || $col eq 'comments') {
1942 $sth->bind_param($i, $data, {ora_type => ORA_CLOB});
1943 } elsif ($col eq 'image') {
1944 $sth->bind_param($i, $data, {ora_type => ORA_BLOB});
1946 $sth->bind_param($i, $data);
1950 $sth->execute; # execute without @bind now
1952 Now, why would you still use B<SQL::Abstract> if you have to do this crap?
1953 Basically, the advantage is still that you don't have to care which fields
1954 are or are not included. You could wrap that above C<for> loop in a simple
1955 sub called C<bind_fields()> or something and reuse it repeatedly. You still
1956 get a layer of abstraction over manual SQL specification.
1958 Note that if you set L</bindtype> to C<columns>, the C<\[ $sql, @bind ]>
1959 construct (see L</Literal SQL with placeholders and bind values (subqueries)>)
1960 will expect the bind values in this format.
1964 This is the character that a table or column name will be quoted
1965 with. By default this is an empty string, but you could set it to
1966 the character C<`>, to generate SQL like this:
1968 SELECT `a_field` FROM `a_table` WHERE `some_field` LIKE '%someval%'
1970 Alternatively, you can supply an array ref of two items, the first being the left
1971 hand quote character, and the second the right hand quote character. For
1972 example, you could supply C<['[',']']> for SQL Server 2000 compliant quotes
1973 that generates SQL like this:
1975 SELECT [a_field] FROM [a_table] WHERE [some_field] LIKE '%someval%'
1977 Quoting is useful if you have tables or columns names that are reserved
1978 words in your database's SQL dialect.
1982 This is the character that will be used to escape L</quote_char>s appearing
1983 in an identifier before it has been quoted.
1985 The parameter default in case of a single L</quote_char> character is the quote
1988 When opening-closing-style quoting is used (L</quote_char> is an arrayref)
1989 this parameter defaults to the B<closing (right)> L</quote_char>. Occurrences
1990 of the B<opening (left)> L</quote_char> within the identifier are currently left
1991 untouched. The default for opening-closing-style quotes may change in future
1992 versions, thus you are B<strongly encouraged> to specify the escape character
1997 This is the character that separates a table and column name. It is
1998 necessary to specify this when the C<quote_char> option is selected,
1999 so that tables and column names can be individually quoted like this:
2001 SELECT `table`.`one_field` FROM `table` WHERE `table`.`other_field` = 1
2003 =item injection_guard
2005 A regular expression C<qr/.../> that is applied to any C<-function> and unquoted
2006 column name specified in a query structure. This is a safety mechanism to avoid
2007 injection attacks when mishandling user input e.g.:
2009 my %condition_as_column_value_pairs = get_values_from_user();
2010 $sqla->select( ... , \%condition_as_column_value_pairs );
2012 If the expression matches an exception is thrown. Note that literal SQL
2013 supplied via C<\'...'> or C<\['...']> is B<not> checked in any way.
2015 Defaults to checking for C<;> and the C<GO> keyword (TransactSQL)
2017 =item array_datatypes
2019 When this option is true, arrayrefs in INSERT or UPDATE are
2020 interpreted as array datatypes and are passed directly
2022 When this option is false, arrayrefs are interpreted
2023 as literal SQL, just like refs to arrayrefs
2024 (but this behavior is for backwards compatibility; when writing
2025 new queries, use the "reference to arrayref" syntax
2031 Takes a reference to a list of "special operators"
2032 to extend the syntax understood by L<SQL::Abstract>.
2033 See section L</"SPECIAL OPERATORS"> for details.
2037 Takes a reference to a list of "unary operators"
2038 to extend the syntax understood by L<SQL::Abstract>.
2039 See section L</"UNARY OPERATORS"> for details.
2045 =head2 insert($table, \@values || \%fieldvals, \%options)
2047 This is the simplest function. You simply give it a table name
2048 and either an arrayref of values or hashref of field/value pairs.
2049 It returns an SQL INSERT statement and a list of bind values.
2050 See the sections on L</"Inserting and Updating Arrays"> and
2051 L</"Inserting and Updating SQL"> for information on how to insert
2052 with those data types.
2054 The optional C<\%options> hash reference may contain additional
2055 options to generate the insert SQL. Currently supported options
2062 Takes either a scalar of raw SQL fields, or an array reference of
2063 field names, and adds on an SQL C<RETURNING> statement at the end.
2064 This allows you to return data generated by the insert statement
2065 (such as row IDs) without performing another C<SELECT> statement.
2066 Note, however, this is not part of the SQL standard and may not
2067 be supported by all database engines.
2071 =head2 update($table, \%fieldvals, \%where, \%options)
2073 This takes a table, hashref of field/value pairs, and an optional
2074 hashref L<WHERE clause|/WHERE CLAUSES>. It returns an SQL UPDATE function and a list
2076 See the sections on L</"Inserting and Updating Arrays"> and
2077 L</"Inserting and Updating SQL"> for information on how to insert
2078 with those data types.
2080 The optional C<\%options> hash reference may contain additional
2081 options to generate the update SQL. Currently supported options
2088 See the C<returning> option to
2089 L<insert|/insert($table, \@values || \%fieldvals, \%options)>.
2093 =head2 select($source, $fields, $where, $order)
2095 This returns a SQL SELECT statement and associated list of bind values, as
2096 specified by the arguments:
2102 Specification of the 'FROM' part of the statement.
2103 The argument can be either a plain scalar (interpreted as a table
2104 name, will be quoted), or an arrayref (interpreted as a list
2105 of table names, joined by commas, quoted), or a scalarref
2106 (literal SQL, not quoted).
2110 Specification of the list of fields to retrieve from
2112 The argument can be either an arrayref (interpreted as a list
2113 of field names, will be joined by commas and quoted), or a
2114 plain scalar (literal SQL, not quoted).
2115 Please observe that this API is not as flexible as that of
2116 the first argument C<$source>, for backwards compatibility reasons.
2120 Optional argument to specify the WHERE part of the query.
2121 The argument is most often a hashref, but can also be
2122 an arrayref or plain scalar --
2123 see section L<WHERE clause|/"WHERE CLAUSES"> for details.
2127 Optional argument to specify the ORDER BY part of the query.
2128 The argument can be a scalar, a hashref or an arrayref
2129 -- see section L<ORDER BY clause|/"ORDER BY CLAUSES">
2135 =head2 delete($table, \%where, \%options)
2137 This takes a table name and optional hashref L<WHERE clause|/WHERE CLAUSES>.
2138 It returns an SQL DELETE statement and list of bind values.
2140 The optional C<\%options> hash reference may contain additional
2141 options to generate the delete SQL. Currently supported options
2148 See the C<returning> option to
2149 L<insert|/insert($table, \@values || \%fieldvals, \%options)>.
2153 =head2 where(\%where, $order)
2155 This is used to generate just the WHERE clause. For example,
2156 if you have an arbitrary data structure and know what the
2157 rest of your SQL is going to look like, but want an easy way
2158 to produce a WHERE clause, use this. It returns an SQL WHERE
2159 clause and list of bind values.
2162 =head2 values(\%data)
2164 This just returns the values from the hash C<%data>, in the same
2165 order that would be returned from any of the other above queries.
2166 Using this allows you to markedly speed up your queries if you
2167 are affecting lots of rows. See below under the L</"PERFORMANCE"> section.
2169 =head2 generate($any, 'number', $of, \@data, $struct, \%types)
2171 Warning: This is an experimental method and subject to change.
2173 This returns arbitrarily generated SQL. It's a really basic shortcut.
2174 It will return two different things, depending on return context:
2176 my($stmt, @bind) = $sql->generate('create table', \$table, \@fields);
2177 my $stmt_and_val = $sql->generate('create table', \$table, \@fields);
2179 These would return the following:
2181 # First calling form
2182 $stmt = "CREATE TABLE test (?, ?)";
2183 @bind = (field1, field2);
2185 # Second calling form
2186 $stmt_and_val = "CREATE TABLE test (field1, field2)";
2188 Depending on what you're trying to do, it's up to you to choose the correct
2189 format. In this example, the second form is what you would want.
2193 $sql->generate('alter session', { nls_date_format => 'MM/YY' });
2197 ALTER SESSION SET nls_date_format = 'MM/YY'
2199 You get the idea. Strings get their case twiddled, but everything
2200 else remains verbatim.
2202 =head1 EXPORTABLE FUNCTIONS
2204 =head2 is_plain_value
2206 Determines if the supplied argument is a plain value as understood by this
2211 =item * The value is C<undef>
2213 =item * The value is a non-reference
2215 =item * The value is an object with stringification overloading
2217 =item * The value is of the form C<< { -value => $anything } >>
2221 On failure returns C<undef>, on success returns a B<scalar> reference
2222 to the original supplied argument.
2228 The stringification overloading detection is rather advanced: it takes
2229 into consideration not only the presence of a C<""> overload, but if that
2230 fails also checks for enabled
2231 L<autogenerated versions of C<"">|overload/Magic Autogeneration>, based
2232 on either C<0+> or C<bool>.
2234 Unfortunately testing in the field indicates that this
2235 detection B<< may tickle a latent bug in perl versions before 5.018 >>,
2236 but only when very large numbers of stringifying objects are involved.
2237 At the time of writing ( Sep 2014 ) there is no clear explanation of
2238 the direct cause, nor is there a manageably small test case that reliably
2239 reproduces the problem.
2241 If you encounter any of the following exceptions in B<random places within
2242 your application stack> - this module may be to blame:
2244 Operation "ne": no method found,
2245 left argument in overloaded package <something>,
2246 right argument in overloaded package <something>
2250 Stub found while resolving method "???" overloading """" in package <something>
2252 If you fall victim to the above - please attempt to reduce the problem
2253 to something that could be sent to the L<SQL::Abstract developers
2254 |DBIx::Class/GETTING HELP/SUPPORT>
2255 (either publicly or privately). As a workaround in the meantime you can
2256 set C<$ENV{SQLA_ISVALUE_IGNORE_AUTOGENERATED_STRINGIFICATION}> to a true
2257 value, which will most likely eliminate your problem (at the expense of
2258 not being able to properly detect exotic forms of stringification).
2260 This notice and environment variable will be removed in a future version,
2261 as soon as the underlying problem is found and a reliable workaround is
2266 =head2 is_literal_value
2268 Determines if the supplied argument is a literal value as understood by this
2273 =item * C<\$sql_string>
2275 =item * C<\[ $sql_string, @bind_values ]>
2279 On failure returns C<undef>, on success returns an B<array> reference
2280 containing the unpacked version of the supplied literal SQL and bind values.
2282 =head1 WHERE CLAUSES
2286 This module uses a variation on the idea from L<DBIx::Abstract>. It
2287 is B<NOT>, repeat I<not> 100% compatible. B<The main logic of this
2288 module is that things in arrays are OR'ed, and things in hashes
2291 The easiest way to explain is to show lots of examples. After
2292 each C<%where> hash shown, it is assumed you used:
2294 my($stmt, @bind) = $sql->where(\%where);
2296 However, note that the C<%where> hash can be used directly in any
2297 of the other functions as well, as described above.
2299 =head2 Key-value pairs
2301 So, let's get started. To begin, a simple hash:
2305 status => 'completed'
2308 Is converted to SQL C<key = val> statements:
2310 $stmt = "WHERE user = ? AND status = ?";
2311 @bind = ('nwiger', 'completed');
2313 One common thing I end up doing is having a list of values that
2314 a field can be in. To do this, simply specify a list inside of
2319 status => ['assigned', 'in-progress', 'pending'];
2322 This simple code will create the following:
2324 $stmt = "WHERE user = ? AND ( status = ? OR status = ? OR status = ? )";
2325 @bind = ('nwiger', 'assigned', 'in-progress', 'pending');
2327 A field associated to an empty arrayref will be considered a
2328 logical false and will generate 0=1.
2330 =head2 Tests for NULL values
2332 If the value part is C<undef> then this is converted to SQL <IS NULL>
2341 $stmt = "WHERE user = ? AND status IS NULL";
2344 To test if a column IS NOT NULL:
2348 status => { '!=', undef },
2351 =head2 Specific comparison operators
2353 If you want to specify a different type of operator for your comparison,
2354 you can use a hashref for a given column:
2358 status => { '!=', 'completed' }
2361 Which would generate:
2363 $stmt = "WHERE user = ? AND status != ?";
2364 @bind = ('nwiger', 'completed');
2366 To test against multiple values, just enclose the values in an arrayref:
2368 status => { '=', ['assigned', 'in-progress', 'pending'] };
2370 Which would give you:
2372 "WHERE status = ? OR status = ? OR status = ?"
2375 The hashref can also contain multiple pairs, in which case it is expanded
2376 into an C<AND> of its elements:
2380 status => { '!=', 'completed', -not_like => 'pending%' }
2383 # Or more dynamically, like from a form
2384 $where{user} = 'nwiger';
2385 $where{status}{'!='} = 'completed';
2386 $where{status}{'-not_like'} = 'pending%';
2388 # Both generate this
2389 $stmt = "WHERE user = ? AND status != ? AND status NOT LIKE ?";
2390 @bind = ('nwiger', 'completed', 'pending%');
2393 To get an OR instead, you can combine it with the arrayref idea:
2397 priority => [ { '=', 2 }, { '>', 5 } ]
2400 Which would generate:
2402 $stmt = "WHERE ( priority = ? OR priority > ? ) AND user = ?";
2403 @bind = ('2', '5', 'nwiger');
2405 If you want to include literal SQL (with or without bind values), just use a
2406 scalar reference or reference to an arrayref as the value:
2409 date_entered => { '>' => \["to_date(?, 'MM/DD/YYYY')", "11/26/2008"] },
2410 date_expires => { '<' => \"now()" }
2413 Which would generate:
2415 $stmt = "WHERE date_entered > to_date(?, 'MM/DD/YYYY') AND date_expires < now()";
2416 @bind = ('11/26/2008');
2419 =head2 Logic and nesting operators
2421 In the example above,
2422 there is a subtle trap if you want to say something like
2423 this (notice the C<AND>):
2425 WHERE priority != ? AND priority != ?
2427 Because, in Perl you I<can't> do this:
2429 priority => { '!=' => 2, '!=' => 1 }
2431 As the second C<!=> key will obliterate the first. The solution
2432 is to use the special C<-modifier> form inside an arrayref:
2434 priority => [ -and => {'!=', 2},
2438 Normally, these would be joined by C<OR>, but the modifier tells it
2439 to use C<AND> instead. (Hint: You can use this in conjunction with the
2440 C<logic> option to C<new()> in order to change the way your queries
2441 work by default.) B<Important:> Note that the C<-modifier> goes
2442 B<INSIDE> the arrayref, as an extra first element. This will
2443 B<NOT> do what you think it might:
2445 priority => -and => [{'!=', 2}, {'!=', 1}] # WRONG!
2447 Here is a quick list of equivalencies, since there is some overlap:
2450 status => {'!=', 'completed', 'not like', 'pending%' }
2451 status => [ -and => {'!=', 'completed'}, {'not like', 'pending%'}]
2454 status => {'=', ['assigned', 'in-progress']}
2455 status => [ -or => {'=', 'assigned'}, {'=', 'in-progress'}]
2456 status => [ {'=', 'assigned'}, {'=', 'in-progress'} ]
2460 =head2 Special operators: IN, BETWEEN, etc.
2462 You can also use the hashref format to compare a list of fields using the
2463 C<IN> comparison operator, by specifying the list as an arrayref:
2466 status => 'completed',
2467 reportid => { -in => [567, 2335, 2] }
2470 Which would generate:
2472 $stmt = "WHERE status = ? AND reportid IN (?,?,?)";
2473 @bind = ('completed', '567', '2335', '2');
2475 The reverse operator C<-not_in> generates SQL C<NOT IN> and is used in
2478 If the argument to C<-in> is an empty array, 'sqlfalse' is generated
2479 (by default: C<1=0>). Similarly, C<< -not_in => [] >> generates
2480 'sqltrue' (by default: C<1=1>).
2482 In addition to the array you can supply a chunk of literal sql or
2483 literal sql with bind:
2486 customer => { -in => \[
2487 'SELECT cust_id FROM cust WHERE balance > ?',
2490 status => { -in => \'SELECT status_codes FROM states' },
2496 customer IN ( SELECT cust_id FROM cust WHERE balance > ? )
2497 AND status IN ( SELECT status_codes FROM states )
2501 Finally, if the argument to C<-in> is not a reference, it will be
2502 treated as a single-element array.
2504 Another pair of operators is C<-between> and C<-not_between>,
2505 used with an arrayref of two values:
2509 completion_date => {
2510 -not_between => ['2002-10-01', '2003-02-06']
2516 WHERE user = ? AND completion_date NOT BETWEEN ( ? AND ? )
2518 Just like with C<-in> all plausible combinations of literal SQL
2522 start0 => { -between => [ 1, 2 ] },
2523 start1 => { -between => \["? AND ?", 1, 2] },
2524 start2 => { -between => \"lower(x) AND upper(y)" },
2525 start3 => { -between => [
2527 \["upper(?)", 'stuff' ],
2534 ( start0 BETWEEN ? AND ? )
2535 AND ( start1 BETWEEN ? AND ? )
2536 AND ( start2 BETWEEN lower(x) AND upper(y) )
2537 AND ( start3 BETWEEN lower(x) AND upper(?) )
2539 @bind = (1, 2, 1, 2, 'stuff');
2542 These are the two builtin "special operators"; but the
2543 list can be expanded: see section L</"SPECIAL OPERATORS"> below.
2545 =head2 Unary operators: bool
2547 If you wish to test against boolean columns or functions within your
2548 database you can use the C<-bool> and C<-not_bool> operators. For
2549 example to test the column C<is_user> being true and the column
2550 C<is_enabled> being false you would use:-
2554 -not_bool => 'is_enabled',
2559 WHERE is_user AND NOT is_enabled
2561 If a more complex combination is required, testing more conditions,
2562 then you should use the and/or operators:-
2567 -not_bool => { two=> { -rlike => 'bar' } },
2568 -not_bool => { three => [ { '=', 2 }, { '>', 5 } ] },
2579 (NOT ( three = ? OR three > ? ))
2582 =head2 Nested conditions, -and/-or prefixes
2584 So far, we've seen how multiple conditions are joined with a top-level
2585 C<AND>. We can change this by putting the different conditions we want in
2586 hashes and then putting those hashes in an array. For example:
2591 status => { -like => ['pending%', 'dispatched'] },
2595 status => 'unassigned',
2599 This data structure would create the following:
2601 $stmt = "WHERE ( user = ? AND ( status LIKE ? OR status LIKE ? ) )
2602 OR ( user = ? AND status = ? ) )";
2603 @bind = ('nwiger', 'pending', 'dispatched', 'robot', 'unassigned');
2606 Clauses in hashrefs or arrayrefs can be prefixed with an C<-and> or C<-or>
2607 to change the logic inside:
2613 -and => [ workhrs => {'>', 20}, geo => 'ASIA' ],
2614 -or => { workhrs => {'<', 50}, geo => 'EURO' },
2621 $stmt = "WHERE ( user = ?
2622 AND ( ( workhrs > ? AND geo = ? )
2623 OR ( workhrs < ? OR geo = ? ) ) )";
2624 @bind = ('nwiger', '20', 'ASIA', '50', 'EURO');
2626 =head3 Algebraic inconsistency, for historical reasons
2628 C<Important note>: when connecting several conditions, the C<-and->|C<-or>
2629 operator goes C<outside> of the nested structure; whereas when connecting
2630 several constraints on one column, the C<-and> operator goes
2631 C<inside> the arrayref. Here is an example combining both features:
2634 -and => [a => 1, b => 2],
2635 -or => [c => 3, d => 4],
2636 e => [-and => {-like => 'foo%'}, {-like => '%bar'} ]
2641 WHERE ( ( ( a = ? AND b = ? )
2642 OR ( c = ? OR d = ? )
2643 OR ( e LIKE ? AND e LIKE ? ) ) )
2645 This difference in syntax is unfortunate but must be preserved for
2646 historical reasons. So be careful: the two examples below would
2647 seem algebraically equivalent, but they are not
2650 { -like => 'foo%' },
2651 { -like => '%bar' },
2653 # yields: WHERE ( ( col LIKE ? AND col LIKE ? ) )
2656 { col => { -like => 'foo%' } },
2657 { col => { -like => '%bar' } },
2659 # yields: WHERE ( ( col LIKE ? OR col LIKE ? ) )
2662 =head2 Literal SQL and value type operators
2664 The basic premise of SQL::Abstract is that in WHERE specifications the "left
2665 side" is a column name and the "right side" is a value (normally rendered as
2666 a placeholder). This holds true for both hashrefs and arrayref pairs as you
2667 see in the L</WHERE CLAUSES> examples above. Sometimes it is necessary to
2668 alter this behavior. There are several ways of doing so.
2672 This is a virtual operator that signals the string to its right side is an
2673 identifier (a column name) and not a value. For example to compare two
2674 columns you would write:
2677 priority => { '<', 2 },
2678 requestor => { -ident => 'submitter' },
2683 $stmt = "WHERE priority < ? AND requestor = submitter";
2686 If you are maintaining legacy code you may see a different construct as
2687 described in L</Deprecated usage of Literal SQL>, please use C<-ident> in new
2692 This is a virtual operator that signals that the construct to its right side
2693 is a value to be passed to DBI. This is for example necessary when you want
2694 to write a where clause against an array (for RDBMS that support such
2695 datatypes). For example:
2698 array => { -value => [1, 2, 3] }
2703 $stmt = 'WHERE array = ?';
2704 @bind = ([1, 2, 3]);
2706 Note that if you were to simply say:
2712 the result would probably not be what you wanted:
2714 $stmt = 'WHERE array = ? OR array = ? OR array = ?';
2719 Finally, sometimes only literal SQL will do. To include a random snippet
2720 of SQL verbatim, you specify it as a scalar reference. Consider this only
2721 as a last resort. Usually there is a better way. For example:
2724 priority => { '<', 2 },
2725 requestor => { -in => \'(SELECT name FROM hitmen)' },
2730 $stmt = "WHERE priority < ? AND requestor IN (SELECT name FROM hitmen)"
2733 Note that in this example, you only get one bind parameter back, since
2734 the verbatim SQL is passed as part of the statement.
2738 Never use untrusted input as a literal SQL argument - this is a massive
2739 security risk (there is no way to check literal snippets for SQL
2740 injections and other nastyness). If you need to deal with untrusted input
2741 use literal SQL with placeholders as described next.
2743 =head3 Literal SQL with placeholders and bind values (subqueries)
2745 If the literal SQL to be inserted has placeholders and bind values,
2746 use a reference to an arrayref (yes this is a double reference --
2747 not so common, but perfectly legal Perl). For example, to find a date
2748 in Postgres you can use something like this:
2751 date_column => \[ "= date '2008-09-30' - ?::integer", 10 ]
2756 $stmt = "WHERE ( date_column = date '2008-09-30' - ?::integer )"
2759 Note that you must pass the bind values in the same format as they are returned
2760 by L<where|/where(\%where, $order)>. This means that if you set L</bindtype>
2761 to C<columns>, you must provide the bind values in the
2762 C<< [ column_meta => value ] >> format, where C<column_meta> is an opaque
2763 scalar value; most commonly the column name, but you can use any scalar value
2764 (including references and blessed references), L<SQL::Abstract> will simply
2765 pass it through intact. So if C<bindtype> is set to C<columns> the above
2766 example will look like:
2769 date_column => \[ "= date '2008-09-30' - ?::integer", [ {} => 10 ] ]
2772 Literal SQL is especially useful for nesting parenthesized clauses in the
2773 main SQL query. Here is a first example:
2775 my ($sub_stmt, @sub_bind) = ("SELECT c1 FROM t1 WHERE c2 < ? AND c3 LIKE ?",
2779 bar => \["IN ($sub_stmt)" => @sub_bind],
2784 $stmt = "WHERE (foo = ? AND bar IN (SELECT c1 FROM t1
2785 WHERE c2 < ? AND c3 LIKE ?))";
2786 @bind = (1234, 100, "foo%");
2788 Other subquery operators, like for example C<"E<gt> ALL"> or C<"NOT IN">,
2789 are expressed in the same way. Of course the C<$sub_stmt> and
2790 its associated bind values can be generated through a former call
2793 my ($sub_stmt, @sub_bind)
2794 = $sql->select("t1", "c1", {c2 => {"<" => 100},
2795 c3 => {-like => "foo%"}});
2798 bar => \["> ALL ($sub_stmt)" => @sub_bind],
2801 In the examples above, the subquery was used as an operator on a column;
2802 but the same principle also applies for a clause within the main C<%where>
2803 hash, like an EXISTS subquery:
2805 my ($sub_stmt, @sub_bind)
2806 = $sql->select("t1", "*", {c1 => 1, c2 => \"> t0.c0"});
2807 my %where = ( -and => [
2809 \["EXISTS ($sub_stmt)" => @sub_bind],
2814 $stmt = "WHERE (foo = ? AND EXISTS (SELECT * FROM t1
2815 WHERE c1 = ? AND c2 > t0.c0))";
2819 Observe that the condition on C<c2> in the subquery refers to
2820 column C<t0.c0> of the main query: this is I<not> a bind
2821 value, so we have to express it through a scalar ref.
2822 Writing C<< c2 => {">" => "t0.c0"} >> would have generated
2823 C<< c2 > ? >> with bind value C<"t0.c0"> ... not exactly
2824 what we wanted here.
2826 Finally, here is an example where a subquery is used
2827 for expressing unary negation:
2829 my ($sub_stmt, @sub_bind)
2830 = $sql->where({age => [{"<" => 10}, {">" => 20}]});
2831 $sub_stmt =~ s/^ where //i; # don't want "WHERE" in the subclause
2833 lname => {like => '%son%'},
2834 \["NOT ($sub_stmt)" => @sub_bind],
2839 $stmt = "lname LIKE ? AND NOT ( age < ? OR age > ? )"
2840 @bind = ('%son%', 10, 20)
2842 =head3 Deprecated usage of Literal SQL
2844 Below are some examples of archaic use of literal SQL. It is shown only as
2845 reference for those who deal with legacy code. Each example has a much
2846 better, cleaner and safer alternative that users should opt for in new code.
2852 my %where = ( requestor => \'IS NOT NULL' )
2854 $stmt = "WHERE requestor IS NOT NULL"
2856 This used to be the way of generating NULL comparisons, before the handling
2857 of C<undef> got formalized. For new code please use the superior syntax as
2858 described in L</Tests for NULL values>.
2862 my %where = ( requestor => \'= submitter' )
2864 $stmt = "WHERE requestor = submitter"
2866 This used to be the only way to compare columns. Use the superior L</-ident>
2867 method for all new code. For example an identifier declared in such a way
2868 will be properly quoted if L</quote_char> is properly set, while the legacy
2869 form will remain as supplied.
2873 my %where = ( is_ready => \"", completed => { '>', '2012-12-21' } )
2875 $stmt = "WHERE completed > ? AND is_ready"
2876 @bind = ('2012-12-21')
2878 Using an empty string literal used to be the only way to express a boolean.
2879 For all new code please use the much more readable
2880 L<-bool|/Unary operators: bool> operator.
2886 These pages could go on for a while, since the nesting of the data
2887 structures this module can handle are pretty much unlimited (the
2888 module implements the C<WHERE> expansion as a recursive function
2889 internally). Your best bet is to "play around" with the module a
2890 little to see how the data structures behave, and choose the best
2891 format for your data based on that.
2893 And of course, all the values above will probably be replaced with
2894 variables gotten from forms or the command line. After all, if you
2895 knew everything ahead of time, you wouldn't have to worry about
2896 dynamically-generating SQL and could just hardwire it into your
2899 =head1 ORDER BY CLAUSES
2901 Some functions take an order by clause. This can either be a scalar (just a
2902 column name), a hashref of C<< { -desc => 'col' } >> or C<< { -asc => 'col' }
2903 >>, a scalarref, an arrayref-ref, or an arrayref of any of the previous
2906 Given | Will Generate
2907 ---------------------------------------------------------------
2909 'colA' | ORDER BY colA
2911 [qw/colA colB/] | ORDER BY colA, colB
2913 {-asc => 'colA'} | ORDER BY colA ASC
2915 {-desc => 'colB'} | ORDER BY colB DESC
2917 ['colA', {-asc => 'colB'}] | ORDER BY colA, colB ASC
2919 { -asc => [qw/colA colB/] } | ORDER BY colA ASC, colB ASC
2921 \'colA DESC' | ORDER BY colA DESC
2923 \[ 'FUNC(colA, ?)', $x ] | ORDER BY FUNC(colA, ?)
2924 | /* ...with $x bound to ? */
2927 { -asc => 'colA' }, | colA ASC,
2928 { -desc => [qw/colB/] }, | colB DESC,
2929 { -asc => [qw/colC colD/] },| colC ASC, colD ASC,
2930 \'colE DESC', | colE DESC,
2931 \[ 'FUNC(colF, ?)', $x ], | FUNC(colF, ?)
2932 ] | /* ...with $x bound to ? */
2933 ===============================================================
2937 =head1 SPECIAL OPERATORS
2939 my $sqlmaker = SQL::Abstract->new(special_ops => [
2943 my ($self, $field, $op, $arg) = @_;
2949 handler => 'method_name',
2953 A "special operator" is a SQL syntactic clause that can be
2954 applied to a field, instead of a usual binary operator.
2957 WHERE field IN (?, ?, ?)
2958 WHERE field BETWEEN ? AND ?
2959 WHERE MATCH(field) AGAINST (?, ?)
2961 Special operators IN and BETWEEN are fairly standard and therefore
2962 are builtin within C<SQL::Abstract> (as the overridable methods
2963 C<_where_field_IN> and C<_where_field_BETWEEN>). For other operators,
2964 like the MATCH .. AGAINST example above which is specific to MySQL,
2965 you can write your own operator handlers - supply a C<special_ops>
2966 argument to the C<new> method. That argument takes an arrayref of
2967 operator definitions; each operator definition is a hashref with two
2974 the regular expression to match the operator
2978 Either a coderef or a plain scalar method name. In both cases
2979 the expected return is C<< ($sql, @bind) >>.
2981 When supplied with a method name, it is simply called on the
2982 L<SQL::Abstract> object as:
2984 $self->$method_name($field, $op, $arg)
2988 $field is the LHS of the operator
2989 $op is the part that matched the handler regex
2992 When supplied with a coderef, it is called as:
2994 $coderef->($self, $field, $op, $arg)
2999 For example, here is an implementation
3000 of the MATCH .. AGAINST syntax for MySQL
3002 my $sqlmaker = SQL::Abstract->new(special_ops => [
3004 # special op for MySql MATCH (field) AGAINST(word1, word2, ...)
3005 {regex => qr/^match$/i,
3007 my ($self, $field, $op, $arg) = @_;
3008 $arg = [$arg] if not ref $arg;
3009 my $label = $self->_quote($field);
3010 my ($placeholder) = $self->_convert('?');
3011 my $placeholders = join ", ", (($placeholder) x @$arg);
3012 my $sql = $self->_sqlcase('match') . " ($label) "
3013 . $self->_sqlcase('against') . " ($placeholders) ";
3014 my @bind = $self->_bindtype($field, @$arg);
3015 return ($sql, @bind);
3022 =head1 UNARY OPERATORS
3024 my $sqlmaker = SQL::Abstract->new(unary_ops => [
3028 my ($self, $op, $arg) = @_;
3034 handler => 'method_name',
3038 A "unary operator" is a SQL syntactic clause that can be
3039 applied to a field - the operator goes before the field
3041 You can write your own operator handlers - supply a C<unary_ops>
3042 argument to the C<new> method. That argument takes an arrayref of
3043 operator definitions; each operator definition is a hashref with two
3050 the regular expression to match the operator
3054 Either a coderef or a plain scalar method name. In both cases
3055 the expected return is C<< $sql >>.
3057 When supplied with a method name, it is simply called on the
3058 L<SQL::Abstract> object as:
3060 $self->$method_name($op, $arg)
3064 $op is the part that matched the handler regex
3065 $arg is the RHS or argument of the operator
3067 When supplied with a coderef, it is called as:
3069 $coderef->($self, $op, $arg)
3077 Thanks to some benchmarking by Mark Stosberg, it turns out that
3078 this module is many orders of magnitude faster than using C<DBIx::Abstract>.
3079 I must admit this wasn't an intentional design issue, but it's a
3080 byproduct of the fact that you get to control your C<DBI> handles
3083 To maximize performance, use a code snippet like the following:
3085 # prepare a statement handle using the first row
3086 # and then reuse it for the rest of the rows
3088 for my $href (@array_of_hashrefs) {
3089 $stmt ||= $sql->insert('table', $href);
3090 $sth ||= $dbh->prepare($stmt);
3091 $sth->execute($sql->values($href));
3094 The reason this works is because the keys in your C<$href> are sorted
3095 internally by B<SQL::Abstract>. Thus, as long as your data retains
3096 the same structure, you only have to generate the SQL the first time
3097 around. On subsequent queries, simply use the C<values> function provided
3098 by this module to return your values in the correct order.
3100 However this depends on the values having the same type - if, for
3101 example, the values of a where clause may either have values
3102 (resulting in sql of the form C<column = ?> with a single bind
3103 value), or alternatively the values might be C<undef> (resulting in
3104 sql of the form C<column IS NULL> with no bind value) then the
3105 caching technique suggested will not work.
3109 If you use my C<CGI::FormBuilder> module at all, you'll hopefully
3110 really like this part (I do, at least). Building up a complex query
3111 can be as simple as the following:
3118 use CGI::FormBuilder;
3121 my $form = CGI::FormBuilder->new(...);
3122 my $sql = SQL::Abstract->new;
3124 if ($form->submitted) {
3125 my $field = $form->field;
3126 my $id = delete $field->{id};
3127 my($stmt, @bind) = $sql->update('table', $field, {id => $id});
3130 Of course, you would still have to connect using C<DBI> to run the
3131 query, but the point is that if you make your form look like your
3132 table, the actual query script can be extremely simplistic.
3134 If you're B<REALLY> lazy (I am), check out C<HTML::QuickTable> for
3135 a fast interface to returning and formatting data. I frequently
3136 use these three modules together to write complex database query
3137 apps in under 50 lines.
3139 =head1 HOW TO CONTRIBUTE
3141 Contributions are always welcome, in all usable forms (we especially
3142 welcome documentation improvements). The delivery methods include git-
3143 or unified-diff formatted patches, GitHub pull requests, or plain bug
3144 reports either via RT or the Mailing list. Contributors are generally
3145 granted full access to the official repository after their first several
3146 patches pass successful review.
3148 This project is maintained in a git repository. The code and related tools are
3149 accessible at the following locations:
3153 =item * Official repo: L<git://git.shadowcat.co.uk/dbsrgits/SQL-Abstract.git>
3155 =item * Official gitweb: L<http://git.shadowcat.co.uk/gitweb/gitweb.cgi?p=dbsrgits/SQL-Abstract.git>
3157 =item * GitHub mirror: L<https://github.com/dbsrgits/sql-abstract>
3159 =item * Authorized committers: L<ssh://dbsrgits@git.shadowcat.co.uk/SQL-Abstract.git>
3165 Version 1.50 was a major internal refactoring of C<SQL::Abstract>.
3166 Great care has been taken to preserve the I<published> behavior
3167 documented in previous versions in the 1.* family; however,
3168 some features that were previously undocumented, or behaved
3169 differently from the documentation, had to be changed in order
3170 to clarify the semantics. Hence, client code that was relying
3171 on some dark areas of C<SQL::Abstract> v1.*
3172 B<might behave differently> in v1.50.
3174 The main changes are:
3180 support for literal SQL through the C<< \ [ $sql, @bind ] >> syntax.
3184 support for the { operator => \"..." } construct (to embed literal SQL)
3188 support for the { operator => \["...", @bind] } construct (to embed literal SQL with bind values)
3192 optional support for L<array datatypes|/"Inserting and Updating Arrays">
3196 defensive programming: check arguments
3200 fixed bug with global logic, which was previously implemented
3201 through global variables yielding side-effects. Prior versions would
3202 interpret C<< [ {cond1, cond2}, [cond3, cond4] ] >>
3203 as C<< "(cond1 AND cond2) OR (cond3 AND cond4)" >>.
3204 Now this is interpreted
3205 as C<< "(cond1 AND cond2) OR (cond3 OR cond4)" >>.
3210 fixed semantics of _bindtype on array args
3214 dropped the C<_anoncopy> of the %where tree. No longer necessary,
3215 we just avoid shifting arrays within that tree.
3219 dropped the C<_modlogic> function
3223 =head1 ACKNOWLEDGEMENTS
3225 There are a number of individuals that have really helped out with
3226 this module. Unfortunately, most of them submitted bugs via CPAN
3227 so I have no idea who they are! But the people I do know are:
3229 Ash Berlin (order_by hash term support)
3230 Matt Trout (DBIx::Class support)
3231 Mark Stosberg (benchmarking)
3232 Chas Owens (initial "IN" operator support)
3233 Philip Collins (per-field SQL functions)
3234 Eric Kolve (hashref "AND" support)
3235 Mike Fragassi (enhancements to "BETWEEN" and "LIKE")
3236 Dan Kubb (support for "quote_char" and "name_sep")
3237 Guillermo Roditi (patch to cleanup "IN" and "BETWEEN", fix and tests for _order_by)
3238 Laurent Dami (internal refactoring, extensible list of special operators, literal SQL)
3239 Norbert Buchmuller (support for literal SQL in hashpair, misc. fixes & tests)
3240 Peter Rabbitson (rewrite of SQLA::Test, misc. fixes & tests)
3241 Oliver Charles (support for "RETURNING" after "INSERT")
3247 L<DBIx::Class>, L<DBIx::Abstract>, L<CGI::FormBuilder>, L<HTML::QuickTable>.
3251 Copyright (c) 2001-2007 Nathan Wiger <nwiger@cpan.org>. All Rights Reserved.
3253 This module is actively maintained by Matt Trout <mst@shadowcatsystems.co.uk>
3255 For support, your best bet is to try the C<DBIx::Class> users mailing list.
3256 While not an official support venue, C<DBIx::Class> makes heavy use of
3257 C<SQL::Abstract>, and as such list members there are very familiar with
3258 how to create queries.
3262 This module is free software; you may copy this under the same
3263 terms as perl itself (either the GNU General Public License or
3264 the Artistic License)