1 package SQL::Abstract; # see doc at end of file
10 our @EXPORT_OK = qw(is_plain_value is_literal_value);
20 *SQL::Abstract::_ENV_::DETECT_AUTOGENERATED_STRINGIFICATION = $ENV{SQLA_ISVALUE_IGNORE_AUTOGENERATED_STRINGIFICATION}
26 #======================================================================
28 #======================================================================
30 our $VERSION = '1.86';
32 # This would confuse some packagers
33 $VERSION = eval $VERSION if $VERSION =~ /_/; # numify for warning-free dev releases
37 # special operators (-in, -between). May be extended/overridden by user.
38 # See section WHERE: BUILTIN SPECIAL OPERATORS below for implementation
39 my @BUILTIN_SPECIAL_OPS = (
40 {regex => qr/^ (?: not \s )? between $/ix, handler => sub { die "NOPE" }},
41 {regex => qr/^ is (?: \s+ not )? $/ix, handler => sub { die "NOPE" }},
42 {regex => qr/^ (?: not \s )? in $/ix, handler => sub { die "NOPE" }},
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',
211 func => '_expand_func',
215 'between' => '_expand_between',
216 'not_between' => '_expand_between',
217 'in' => '_expand_in',
218 'not_in' => '_expand_in',
219 'nest' => '_expand_nest',
220 (map +($_ => '_expand_op_andor'), ('and', 'or')),
221 (map +($_ => '_expand_op_is'), ('is', 'is_not')),
222 'ident' => '_expand_ident',
223 'value' => '_expand_value',
227 (map +($_, "_render_$_"), qw(op func bind ident literal row)),
232 (map +($_ => '_render_op_between'), 'between', 'not_between'),
233 (map +($_ => '_render_op_in'), 'in', 'not_in'),
234 (map +($_ => '_render_unop_postfix'),
235 'is_null', 'is_not_null', 'asc', 'desc',
237 (not => '_render_unop_paren'),
238 (map +($_ => '_render_op_andor'), qw(and or)),
239 ',' => '_render_op_multop',
242 return bless \%opt, $class;
245 sub sqltrue { +{ -literal => [ $_[0]->{sqltrue} ] } }
246 sub sqlfalse { +{ -literal => [ $_[0]->{sqlfalse} ] } }
248 sub _assert_pass_injection_guard {
249 if ($_[1] =~ $_[0]->{injection_guard}) {
250 my $class = ref $_[0];
251 puke "Possible SQL injection attempt '$_[1]'. If this is indeed a part of the "
252 . "desired SQL use literal SQL ( \'...' or \[ '...' ] ) or supply your own "
253 . "{injection_guard} attribute to ${class}->new()"
258 #======================================================================
260 #======================================================================
264 my $table = $self->_table(shift);
265 my $data = shift || return;
270 my ($f_aqt, $v_aqt) = $self->_expand_insert_values($data);
272 my @parts = ([ $self->_sqlcase('insert into').' '.$table ]);
273 push @parts, [ $self->render_aqt($f_aqt) ] if $f_aqt;
274 push @parts, [ $self->_sqlcase('values') ], [ $self->render_aqt($v_aqt) ];
276 if ($options->{returning}) {
277 push @parts, [ $self->_insert_returning($options) ];
280 return $self->join_query_parts(' ', @parts);
283 sub _expand_insert_values {
284 my ($self, $data) = @_;
285 if (is_literal_value($data)) {
286 (undef, $self->expand_expr($data));
288 my ($fields, $values) = (
289 ref($data) eq 'HASH' ?
290 ([ sort keys %$data ], [ @{$data}{sort keys %$data} ])
294 # no names (arrayref) means can't generate bindtype
295 !($fields) && $self->{bindtype} eq 'columns'
296 && belch "can't do 'columns' bindtype when called with arrayref";
300 ? $self->expand_expr({ -row => $fields }, -ident)
305 local our $Cur_Col_Meta = $fields->[$_];
306 $self->_expand_insert_value($values->[$_])
313 # So that subclasses can override INSERT ... RETURNING separately from
314 # UPDATE and DELETE (e.g. DBIx::Class::SQLMaker::Oracle does this)
315 sub _insert_returning { shift->_returning(@_) }
318 my ($self, $options) = @_;
320 my $f = $options->{returning};
322 my ($sql, @bind) = $self->render_aqt(
323 $self->_expand_maybe_list_expr($f, -ident)
326 ? $self->_sqlcase(' returning ') . $sql
327 : ($self->_sqlcase(' returning ').$sql, @bind);
330 sub _expand_insert_value {
333 my $k = our $Cur_Col_Meta;
335 if (ref($v) eq 'ARRAY') {
336 if ($self->{array_datatypes}) {
337 return +{ -bind => [ $k, $v ] };
339 my ($sql, @bind) = @$v;
340 $self->_assert_bindval_matches_bindtype(@bind);
341 return +{ -literal => $v };
343 if (ref($v) eq 'HASH') {
344 if (grep !/^-/, keys %$v) {
345 belch "HASH ref as bind value in insert is not supported";
346 return +{ -bind => [ $k, $v ] };
350 return +{ -bind => [ $k, undef ] };
352 return $self->expand_expr($v);
357 #======================================================================
359 #======================================================================
364 my $table = $self->_table(shift);
365 my $data = shift || return;
369 # first build the 'SET' part of the sql statement
370 puke "Unsupported data type specified to \$sql->update"
371 unless ref $data eq 'HASH';
373 my ($sql, @all_bind) = $self->_update_set_values($data);
374 $sql = $self->_sqlcase('update ') . $table . $self->_sqlcase(' set ')
378 my($where_sql, @where_bind) = $self->where($where);
380 push @all_bind, @where_bind;
383 if ($options->{returning}) {
384 my ($returning_sql, @returning_bind) = $self->_update_returning($options);
385 $sql .= $returning_sql;
386 push @all_bind, @returning_bind;
389 return wantarray ? ($sql, @all_bind) : $sql;
392 sub _update_set_values {
393 my ($self, $data) = @_;
395 return $self->render_aqt(
396 $self->_expand_update_set_values(undef, $data),
400 sub _expand_update_set_values {
401 my ($self, undef, $data) = @_;
402 $self->_expand_maybe_list_expr( [
405 $set = { -bind => $_ } unless defined $set;
406 +{ -op => [ '=', $self->_expand_ident(-ident => $k), $set ] };
412 ? ($self->{array_datatypes}
413 ? [ $k, +{ -bind => [ $k, $v ] } ]
414 : [ $k, +{ -literal => $v } ])
416 local our $Cur_Col_Meta = $k;
417 [ $k, $self->_expand_expr($v) ]
424 # So that subclasses can override UPDATE ... RETURNING separately from
426 sub _update_returning { shift->_returning(@_) }
430 #======================================================================
432 #======================================================================
437 my $table = $self->_table(shift);
438 my $fields = shift || '*';
442 my ($fields_sql, @bind) = $self->_select_fields($fields);
444 my ($where_sql, @where_bind) = $self->where($where, $order);
445 push @bind, @where_bind;
447 my $sql = join(' ', $self->_sqlcase('select'), $fields_sql,
448 $self->_sqlcase('from'), $table)
451 return wantarray ? ($sql, @bind) : $sql;
455 my ($self, $fields) = @_;
456 return $fields unless ref($fields);
457 return $self->render_aqt(
458 $self->_expand_maybe_list_expr($fields, '-ident')
462 #======================================================================
464 #======================================================================
469 my $table = $self->_table(shift);
473 my($where_sql, @bind) = $self->where($where);
474 my $sql = $self->_sqlcase('delete from ') . $table . $where_sql;
476 if ($options->{returning}) {
477 my ($returning_sql, @returning_bind) = $self->_delete_returning($options);
478 $sql .= $returning_sql;
479 push @bind, @returning_bind;
482 return wantarray ? ($sql, @bind) : $sql;
485 # So that subclasses can override DELETE ... RETURNING separately from
487 sub _delete_returning { shift->_returning(@_) }
491 #======================================================================
493 #======================================================================
497 # Finally, a separate routine just to handle WHERE clauses
499 my ($self, $where, $order) = @_;
501 local $self->{convert_where} = $self->{convert};
504 my ($sql, @bind) = defined($where)
505 ? $self->_recurse_where($where)
507 $sql = (defined $sql and length $sql) ? $self->_sqlcase(' where ') . "( $sql )" : '';
511 my ($order_sql, @order_bind) = $self->_order_by($order);
513 push @bind, @order_bind;
516 return wantarray ? ($sql, @bind) : $sql;
519 { our $Default_Scalar_To = -value }
522 my ($self, $expr, $default_scalar_to) = @_;
523 local our $Default_Scalar_To = $default_scalar_to if $default_scalar_to;
524 $self->_expand_expr($expr);
528 my ($self, $aqt) = @_;
529 my ($k, $v, @rest) = %$aqt;
531 die "Not a node type: $k" unless $k =~ s/^-//;
532 if (my $meth = $self->{render}{$k}) {
533 return $self->$meth($k, $v);
535 die "notreached: $k";
539 my ($self, $expr, $default_scalar_to) = @_;
540 my ($sql, @bind) = $self->render_aqt(
541 $self->expand_expr($expr, $default_scalar_to)
543 return (wantarray ? ($sql, @bind) : $sql);
547 my ($self, $raw) = @_;
548 s/^-(?=.)//, s/\s+/_/g for my $op = lc $raw;
553 my ($self, $expr) = @_;
554 our $Expand_Depth ||= 0; local $Expand_Depth = $Expand_Depth + 1;
555 return undef unless defined($expr);
556 if (ref($expr) eq 'HASH') {
557 return undef unless my $kc = keys %$expr;
559 return $self->_expand_op_andor(and => $expr);
561 my ($key, $value) = %$expr;
562 if ($key =~ /^-/ and $key =~ s/ [_\s]? \d+ $//x ) {
563 belch 'Use of [and|or|nest]_N modifiers is deprecated and will be removed in SQLA v2.0. '
564 . "You probably wanted ...-and => [ $key => COND1, $key => COND2 ... ]";
566 return $self->_expand_hashpair($key, $value);
568 if (ref($expr) eq 'ARRAY') {
569 return $self->_expand_op_andor(lc($self->{logic}), $expr);
571 if (my $literal = is_literal_value($expr)) {
572 return +{ -literal => $literal };
574 if (!ref($expr) or Scalar::Util::blessed($expr)) {
575 return $self->_expand_scalar($expr);
580 sub _expand_hashpair {
581 my ($self, $k, $v) = @_;
582 unless (defined($k) and length($k)) {
583 if (defined($k) and my $literal = is_literal_value($v)) {
584 belch 'Hash-pairs consisting of an empty string with a literal are deprecated, and will be removed in 2.0: use -and => [ $literal ] instead';
585 return { -literal => $literal };
587 puke "Supplying an empty left hand side argument is not supported";
590 return $self->_expand_hashpair_op($k, $v);
591 } elsif ($k =~ /^[^\w]/i) {
592 my ($lhs, @rhs) = @$v;
593 return $self->_expand_op(
594 -op, [ $k, $self->expand_expr($lhs, -ident), @rhs ]
597 return $self->_expand_hashpair_ident($k, $v);
600 sub _expand_hashpair_ident {
601 my ($self, $k, $v) = @_;
603 local our $Cur_Col_Meta = $k;
605 # hash with multiple or no elements is andor
607 if (ref($v) eq 'HASH' and keys %$v != 1) {
608 return $self->_expand_op_andor(and => $v, $k);
611 # undef needs to be re-sent with cmp to achieve IS/IS NOT NULL
613 if (is_undef_value($v)) {
614 return $self->_expand_hashpair_cmp($k => undef);
617 # scalars and objects get expanded as whatever requested or values
619 if (!ref($v) or Scalar::Util::blessed($v)) {
620 return $self->_expand_hashpair_scalar($k, $v);
623 # single key hashref is a hashtriple
625 if (ref($v) eq 'HASH') {
626 return $self->_expand_hashtriple($k, %$v);
629 # arrayref needs re-engineering over the elements
631 if (ref($v) eq 'ARRAY') {
632 return $self->sqlfalse unless @$v;
633 $self->_debug("ARRAY($k) means distribute over elements");
635 $v->[0] =~ /^-(and|or)$/i
636 ? (shift(@{$v = [ @$v ]}), $1)
637 : lc($self->{logic} || 'OR')
639 return $self->_expand_op_andor(
644 if (my $literal = is_literal_value($v)) {
646 belch 'Hash-pairs consisting of an empty string with a literal are deprecated, and will be removed in 2.0: use -and => [ $literal ] instead';
649 my ($sql, @bind) = @$literal;
650 if ($self->{bindtype} eq 'columns') {
652 $self->_assert_bindval_matches_bindtype($_);
655 return +{ -literal => [ $self->_quote($k).' '.$sql, @bind ] };
661 my ($self, $expr) = @_;
663 return $self->_expand_expr({ (our $Default_Scalar_To) => $expr });
666 sub _expand_hashpair_scalar {
667 my ($self, $k, $v) = @_;
669 return $self->_expand_hashpair_cmp(
670 $k, $self->_expand_scalar($v),
674 sub _expand_hashpair_op {
675 my ($self, $k, $v) = @_;
677 $self->_assert_pass_injection_guard($k =~ /\A-(.*)\Z/s);
679 my $op = $self->_normalize_op($k);
681 if (my $exp = $self->{expand}{$op}) {
682 return $self->$exp($op, $v);
685 # Ops prefixed with -not_ get converted
687 if (my ($rest) = $op =~/^not_(.*)$/) {
690 $self->_expand_expr({ "-${rest}", $v })
696 my $op = join(' ', split '_', $op);
698 # the old special op system requires illegality for top-level use
701 (our $Expand_Depth) == 1
703 List::Util::first { $op =~ $_->{regex} } @{$self->{special_ops}}
705 $self->{disable_old_special_ops}
706 and List::Util::first { $op =~ $_->{regex} } @BUILTIN_SPECIAL_OPS
710 puke "Illegal use of top-level '-$op'"
713 # the old unary op system means we should touch nothing and let it work
715 if (my $us = List::Util::first { $op =~ $_->{regex} } @{$self->{unary_ops}}) {
716 return { -op => [ $op, $v ] };
720 # an explicit node type is currently assumed to be expanded (this is almost
721 # certainly wrong and there should be expansion anyway)
723 if ($self->{render}{$op}) {
727 my $type = $self->{unknown_unop_always_func} ? -func : -op;
734 and (keys %$v)[0] =~ /^-/
737 (List::Util::first { $op =~ $_->{regex} } @{$self->{special_ops}})
746 ($type eq -func and ref($v) eq 'ARRAY')
747 ? map $self->_expand_expr($_), @$v
748 : $self->_expand_expr($v)
752 sub _expand_hashpair_cmp {
753 my ($self, $k, $v) = @_;
754 $self->_expand_hashtriple($k, $self->{cmp}, $v);
757 sub _expand_hashtriple {
758 my ($self, $k, $vk, $vv) = @_;
760 my $ik = $self->_expand_ident(-ident => $k);
762 my $op = $self->_normalize_op($vk);
763 $self->_assert_pass_injection_guard($op);
765 if ($op =~ s/ _? \d+ $//x ) {
766 return $self->_expand_expr($k, { $vk, $vv });
768 if (my $x = $self->{expand_op}{$op}) {
769 local our $Cur_Col_Meta = $k;
770 return $self->$x($op, $vv, $k);
774 my $op = join(' ', split '_', $op);
776 if (my $us = List::Util::first { $op =~ $_->{regex} } @{$self->{special_ops}}) {
777 return { -op => [ $op, $ik, $vv ] };
779 if (my $us = List::Util::first { $op =~ $_->{regex} } @{$self->{unary_ops}}) {
783 { -op => [ $op, $vv ] }
787 if (ref($vv) eq 'ARRAY') {
789 my $logic = (defined($raw[0]) and $raw[0] =~ /^-(and|or)$/i)
790 ? (shift(@raw), $1) : 'or';
791 my @values = map +{ $vk => $_ }, @raw;
793 $op =~ $self->{inequality_op}
794 or $op =~ $self->{not_like_op}
796 if (lc($logic) eq 'or' and @values > 1) {
797 belch "A multi-element arrayref as an argument to the inequality op '${\uc(join ' ', split '_', $op)}' "
798 . 'is technically equivalent to an always-true 1=1 (you probably wanted '
799 . "to say ...{ \$inequality_op => [ -and => \@values ] }... instead)"
804 # try to DWIM on equality operators
805 return ($self->_dwim_op_to_is($op,
806 "Supplying an empty arrayref to '%s' is deprecated",
807 "operator '%s' applied on an empty array (field '$k')"
808 ) ? $self->sqlfalse : $self->sqltrue);
810 return $self->_expand_op_andor($logic => \@values, $k);
812 if (is_undef_value($vv)) {
813 my $is = ($self->_dwim_op_to_is($op,
814 "Supplying an undefined argument to '%s' is deprecated",
815 "unexpected operator '%s' with undef operand",
816 ) ? 'is' : 'is not');
818 return $self->_expand_hashpair($k => { $is, undef });
820 local our $Cur_Col_Meta = $k;
824 $self->_expand_expr($vv)
829 my ($self, $raw, $empty, $fail) = @_;
831 my $op = $self->_normalize_op($raw);
833 if ($op =~ /^not$/i) {
836 if ($op =~ $self->{equality_op}) {
839 if ($op =~ $self->{like_op}) {
840 belch(sprintf $empty, uc(join ' ', split '_', $op));
843 if ($op =~ $self->{inequality_op}) {
846 if ($op =~ $self->{not_like_op}) {
847 belch(sprintf $empty, uc(join ' ', split '_', $op));
850 puke(sprintf $fail, $op);
854 my ($self, undef, $args) = @_;
855 my ($func, @args) = @$args;
856 return { -func => [ $func, map $self->expand_expr($_), @args ] };
860 my ($self, undef, $body, $k) = @_;
861 return $self->_expand_hashpair_cmp(
862 $k, { -ident => $body }
864 unless (defined($body) or (ref($body) and ref($body) eq 'ARRAY')) {
865 puke "-ident requires a single plain scalar argument (a quotable identifier) or an arrayref of identifier parts";
867 my @parts = map split(/\Q${\($self->{name_sep}||'.')}\E/, $_),
868 ref($body) ? @$body : $body;
869 return { -ident => $parts[-1] } if $self->{_dequalify_idents};
870 unless ($self->{quote_char}) {
871 $self->_assert_pass_injection_guard($_) for @parts;
873 return +{ -ident => \@parts };
877 return $_[0]->_expand_hashpair_cmp(
878 $_[3], { -value => $_[2] },
880 +{ -bind => [ our $Cur_Col_Meta, $_[2] ] };
884 +{ -op => [ 'not', $_[0]->_expand_expr($_[2]) ] };
888 my ($self, undef, $args) = @_;
889 +{ -row => [ map $self->expand_expr($_), @$args ] };
893 my ($self, undef, $args) = @_;
894 my ($op, @opargs) = @$args;
895 if (my $exp = $self->{expand_op}{$op}) {
896 return $self->$exp($op, \@opargs);
898 +{ -op => [ $op, map $self->expand_expr($_), @opargs ] };
902 my ($self, undef, $v) = @_;
904 return $self->_expand_expr($v);
906 puke "-bool => undef not supported" unless defined($v);
907 return $self->_expand_ident(-ident => $v);
910 sub _expand_op_andor {
911 my ($self, $logop, $v, $k) = @_;
913 $v = [ map +{ $k, $_ },
915 ? (map +{ $_ => $v->{$_} }, sort keys %$v)
919 if (ref($v) eq 'HASH') {
920 return undef unless keys %$v;
923 map $self->_expand_expr({ $_ => $v->{$_} }),
927 if (ref($v) eq 'ARRAY') {
928 $logop eq 'and' or $logop eq 'or' or puke "unknown logic: $logop";
931 (ref($_) eq 'ARRAY' and @$_)
932 or (ref($_) eq 'HASH' and %$_)
938 while (my ($el) = splice @expr, 0, 1) {
939 puke "Supplying an empty left hand side argument is not supported in array-pairs"
940 unless defined($el) and length($el);
941 my $elref = ref($el);
943 local our $Expand_Depth = 0;
944 push(@res, grep defined, $self->_expand_expr({ $el, shift(@expr) }));
945 } elsif ($elref eq 'ARRAY') {
946 push(@res, grep defined, $self->_expand_expr($el)) if @$el;
947 } elsif (my $l = is_literal_value($el)) {
948 push @res, { -literal => $l };
949 } elsif ($elref eq 'HASH') {
950 local our $Expand_Depth = 0;
951 push @res, grep defined, $self->_expand_expr($el) if %$el;
957 # return $res[0] if @res == 1;
958 return { -op => [ $logop, @res ] };
964 my ($self, $op, $vv, $k) = @_;
965 ($k, $vv) = @$vv unless defined $k;
966 puke "$op can only take undef as argument"
970 and exists($vv->{-value})
971 and !defined($vv->{-value})
973 return +{ -op => [ $op.'_null', $self->expand_expr($k, -ident) ] };
976 sub _expand_between {
977 my ($self, $op, $vv, $k) = @_;
978 $k = shift @{$vv = [ @$vv ]} unless defined $k;
979 my @rhs = map $self->_expand_expr($_),
980 ref($vv) eq 'ARRAY' ? @$vv : $vv;
982 (@rhs == 1 and ref($rhs[0]) eq 'HASH' and $rhs[0]->{-literal})
984 (@rhs == 2 and defined($rhs[0]) and defined($rhs[1]))
986 puke "Operator '${\uc($op)}' requires either an arrayref with two defined values or expressions, or a single literal scalarref/arrayref-ref";
990 $self->expand_expr(ref($k) ? $k : { -ident => $k }),
996 my ($self, $op, $vv, $k) = @_;
997 $k = shift @{$vv = [ @$vv ]} unless defined $k;
998 if (my $literal = is_literal_value($vv)) {
999 my ($sql, @bind) = @$literal;
1000 my $opened_sql = $self->_open_outer_paren($sql);
1002 $op, $self->expand_expr($k, -ident),
1003 { -literal => [ $opened_sql, @bind ] }
1007 'SQL::Abstract before v1.75 used to generate incorrect SQL when the '
1008 . "-${\uc($op)} operator was given an undef-containing list: !!!AUDIT YOUR CODE "
1009 . 'AND DATA!!! (the upcoming Data::Query-based version of SQL::Abstract '
1010 . 'will emit the logically correct SQL instead of raising this exception)'
1012 puke("Argument passed to the '${\uc($op)}' operator can not be undefined")
1014 my @rhs = map $self->expand_expr($_, -value),
1015 map { defined($_) ? $_: puke($undef_err) }
1016 (ref($vv) eq 'ARRAY' ? @$vv : $vv);
1017 return $self->${\($op =~ /^not/ ? 'sqltrue' : 'sqlfalse')} unless @rhs;
1021 $self->expand_expr($k, -ident),
1027 my ($self, undef, $v) = @_;
1028 # DBIx::Class requires a nest warning to be emitted once but the private
1029 # method it overrode to do so no longer exists
1030 if ($self->{warn_once_on_nest}) {
1031 unless (our $Nest_Warned) {
1033 "-nest in search conditions is deprecated, you most probably wanted:\n"
1034 .q|{..., -and => [ \%cond0, \@cond1, \'cond2', \[ 'cond3', [ col => bind ] ], etc. ], ... }|
1039 return $self->_expand_expr($v);
1043 my ($self, undef, $bind) = @_;
1044 return { -bind => $bind };
1047 sub _recurse_where {
1048 my ($self, $where, $logic) = @_;
1050 # Special case: top level simple string treated as literal
1052 my $where_exp = (ref($where)
1053 ? $self->_expand_expr($where, $logic)
1054 : { -literal => [ $where ] });
1056 # dispatch expanded expression
1058 my ($sql, @bind) = defined($where_exp) ? $self->render_aqt($where_exp) : (undef);
1059 # DBIx::Class used to call _recurse_where in scalar context
1060 # something else might too...
1062 return ($sql, @bind);
1065 belch "Calling _recurse_where in scalar context is deprecated and will go away before 2.0";
1071 my ($self, undef, $ident) = @_;
1073 return $self->_convert($self->_quote($ident));
1077 my ($self, undef, $values) = @_;
1078 my ($sql, @bind) = $self->_render_op(undef, [ ',', @$values ]);
1079 return "($sql)", @bind;
1083 my ($self, undef, $rest) = @_;
1084 my ($func, @args) = @$rest;
1085 return $self->join_query_parts('',
1086 [ $self->_sqlcase($func) ],
1088 [ $self->join_query_parts(', ', @args) ],
1094 my ($self, undef, $bind) = @_;
1095 return ($self->_convert('?'), $self->_bindtype(@$bind));
1098 sub _render_literal {
1099 my ($self, undef, $literal) = @_;
1100 $self->_assert_bindval_matches_bindtype(@{$literal}[1..$#$literal]);
1105 my ($self, undef, $v) = @_;
1106 my ($op, @args) = @$v;
1107 if (my $r = $self->{render_op}{$op}) {
1108 return $self->$r($op, \@args);
1113 my $op = join(' ', split '_', $op);
1115 my $ss = List::Util::first { $op =~ $_->{regex} } @{$self->{special_ops}};
1116 if ($ss and @args > 1) {
1117 puke "Special op '${op}' requires first value to be identifier"
1118 unless my ($ident) = map $_->{-ident}, grep ref($_) eq 'HASH', $args[0];
1119 my $k = join(($self->{name_sep}||'.'), @$ident);
1120 local our $Expand_Depth = 1;
1121 return $self->${\($ss->{handler})}($k, $op, $args[1]);
1123 if (my $us = List::Util::first { $op =~ $_->{regex} } @{$self->{unary_ops}}) {
1124 return $self->${\($us->{handler})}($op, $args[0]);
1127 return $self->_render_unop_paren($op, \@args);
1131 return $self->_render_unop_prefix($op, \@args);
1133 return $self->_render_op_multop($op, \@args);
1139 sub _render_op_between {
1140 my ($self, $op, $args) = @_;
1141 my ($left, $low, $high) = @$args;
1142 my ($rhsql, @rhbind) = do {
1144 puke "Single arg to between must be a literal"
1145 unless $low->{-literal};
1148 my ($l, $h) = map [ $self->render_aqt($_) ], $low, $high;
1149 (join(' ', $l->[0], $self->_sqlcase('and'), $h->[0]),
1150 @{$l}[1..$#$l], @{$h}[1..$#$h])
1153 my ($lhsql, @lhbind) = $self->render_aqt($left);
1157 $self->_sqlcase(join ' ', split '_', $op),
1165 my ($self, $op, $args) = @_;
1166 my ($lhs, @rhs) = @$args;
1169 my ($sql, @bind) = $self->render_aqt($_);
1170 push @in_bind, @bind;
1173 my ($lhsql, @lbind) = $self->render_aqt($lhs);
1175 $lhsql.' '.$self->_sqlcase(join ' ', split '_', $op).' ( '
1176 .join(', ', @in_sql)
1182 sub _render_op_andor {
1183 my ($self, $op, $args) = @_;
1184 my @parts = map [ $self->render_aqt($_) ], @$args;
1185 return '' unless @parts;
1186 return @{$parts[0]} if @parts == 1;
1187 my ($sql, @bind) = $self->join_query_parts(' '.$self->_sqlcase($op).' ', @parts);
1188 return '( '.$sql.' )', @bind;
1191 sub _render_op_multop {
1192 my ($self, $op, $args) = @_;
1193 my @parts = map [ $self->render_aqt($_) ], @$args;
1194 return '' unless @parts;
1195 return @{$parts[0]} if @parts == 1;
1196 my $join = ($op eq ','
1198 : ' '.$self->_sqlcase(join ' ', split '_', $op).' '
1200 return $self->join_query_parts($join, @parts);
1203 sub join_query_parts {
1204 my ($self, $join, @parts) = @_;
1205 my @final = map +(ref($_) eq 'HASH' ? [ $self->render_aqt($_) ] : $_),
1208 join($join, map $_->[0], @final),
1209 (wantarray ? (map @{$_}[1..$#$_], @final) : ()),
1213 sub _render_unop_paren {
1214 my ($self, $op, $v) = @_;
1215 my ($sql, @bind) = $self->_render_unop_prefix($op, $v);
1216 return "(${sql})", @bind;
1219 sub _render_unop_prefix {
1220 my ($self, $op, $v) = @_;
1221 my ($expr_sql, @bind) = $self->render_aqt($v->[0]);
1223 my $op_sql = $self->_sqlcase($op); # join ' ', split '_', $op);
1224 return ("${op_sql} ${expr_sql}", @bind);
1227 sub _render_unop_postfix {
1228 my ($self, $op, $v) = @_;
1229 my ($expr_sql, @bind) = $self->render_aqt($v->[0]);
1230 my $op_sql = $self->_sqlcase(join ' ', split '_', $op);
1231 return ($expr_sql.' '.$op_sql, @bind);
1234 # Some databases (SQLite) treat col IN (1, 2) different from
1235 # col IN ( (1, 2) ). Use this to strip all outer parens while
1236 # adding them back in the corresponding method
1237 sub _open_outer_paren {
1238 my ($self, $sql) = @_;
1240 while (my ($inner) = $sql =~ /^ \s* \( (.*) \) \s* $/xs) {
1242 # there are closing parens inside, need the heavy duty machinery
1243 # to reevaluate the extraction starting from $sql (full reevaluation)
1244 if ($inner =~ /\)/) {
1245 require Text::Balanced;
1247 my (undef, $remainder) = do {
1248 # idiotic design - writes to $@ but *DOES NOT* throw exceptions
1250 Text::Balanced::extract_bracketed($sql, '()', qr/\s*/);
1253 # the entire expression needs to be a balanced bracketed thing
1254 # (after an extract no remainder sans trailing space)
1255 last if defined $remainder and $remainder =~ /\S/;
1265 #======================================================================
1267 #======================================================================
1269 sub _expand_order_by {
1270 my ($self, $arg) = @_;
1272 return unless defined($arg) and not (ref($arg) eq 'ARRAY' and !@$arg);
1274 return $self->_expand_maybe_list_expr($arg)
1275 if ref($arg) eq 'HASH' and ($arg->{-op}||[''])->[0] eq ',';
1277 my $expander = sub {
1278 my ($self, $dir, $expr) = @_;
1279 my @to_expand = ref($expr) eq 'ARRAY' ? @$expr : $expr;
1280 foreach my $arg (@to_expand) {
1284 and grep /^-(asc|desc)$/, keys %$arg
1286 puke "ordering direction hash passed to order by must have exactly one key (-asc or -desc)";
1290 defined($dir) ? { -op => [ $dir =~ /^-?(.*)$/ ,=> $_ ] } : $_
1292 map $self->expand_expr($_, -ident),
1293 map ref($_) eq 'ARRAY' ? @$_ : $_, @to_expand;
1294 return undef unless @exp;
1295 return undef if @exp == 1 and not defined($exp[0]);
1296 return +{ -op => [ ',', @exp ] };
1299 local @{$self->{expand}}{qw(asc desc)} = (($expander) x 2);
1301 return $self->$expander(undef, $arg);
1305 my ($self, $arg) = @_;
1307 return '' unless defined(my $expanded = $self->_expand_order_by($arg));
1309 my ($sql, @bind) = $self->render_aqt($expanded);
1311 return '' unless length($sql);
1313 my $final_sql = $self->_sqlcase(' order by ').$sql;
1315 return wantarray ? ($final_sql, @bind) : $final_sql;
1318 # _order_by no longer needs to call this so doesn't but DBIC uses it.
1320 sub _order_by_chunks {
1321 my ($self, $arg) = @_;
1323 return () unless defined(my $expanded = $self->_expand_order_by($arg));
1325 return $self->_chunkify_order_by($expanded);
1328 sub _chunkify_order_by {
1329 my ($self, $expanded) = @_;
1331 return grep length, $self->render_aqt($expanded)
1332 if $expanded->{-ident} or @{$expanded->{-literal}||[]} == 1;
1335 if (ref() eq 'HASH' and $_->{-op} and $_->{-op}[0] eq ',') {
1336 my ($comma, @list) = @{$_->{-op}};
1337 return map $self->_chunkify_order_by($_), @list;
1339 return [ $self->render_aqt($_) ];
1343 #======================================================================
1344 # DATASOURCE (FOR NOW, JUST PLAIN TABLE OR LIST OF TABLES)
1345 #======================================================================
1351 $self->_expand_maybe_list_expr($from, -ident)
1356 #======================================================================
1358 #======================================================================
1360 sub _expand_maybe_list_expr {
1361 my ($self, $expr, $default) = @_;
1363 ',', map $self->expand_expr($_, $default),
1364 @{$expr->{-op}}[1..$#{$expr->{-op}}]
1365 ] } if ref($expr) eq 'HASH' and ($expr->{-op}||[''])->[0] eq ',';
1366 return +{ -op => [ ',',
1367 map $self->expand_expr($_, $default),
1368 ref($expr) eq 'ARRAY' ? @$expr : $expr
1372 # highly optimized, as it's called way too often
1374 # my ($self, $label) = @_;
1376 return '' unless defined $_[1];
1377 return ${$_[1]} if ref($_[1]) eq 'SCALAR';
1378 puke 'Identifier cannot be hashref' if ref($_[1]) eq 'HASH';
1380 unless ($_[0]->{quote_char}) {
1381 if (ref($_[1]) eq 'ARRAY') {
1382 return join($_[0]->{name_sep}||'.', @{$_[1]});
1384 $_[0]->_assert_pass_injection_guard($_[1]);
1389 my $qref = ref $_[0]->{quote_char};
1391 !$qref ? ($_[0]->{quote_char}, $_[0]->{quote_char})
1392 : ($qref eq 'ARRAY') ? @{$_[0]->{quote_char}}
1393 : puke "Unsupported quote_char format: $_[0]->{quote_char}";
1395 my $esc = $_[0]->{escape_char} || $r;
1397 # parts containing * are naturally unquoted
1399 $_[0]->{name_sep}||'',
1403 : do { (my $n = $_) =~ s/(\Q$esc\E|\Q$r\E)/$esc$1/g; $l . $n . $r }
1405 (ref($_[1]) eq 'ARRAY'
1409 ? split (/\Q$_[0]->{name_sep}\E/, $_[1] )
1417 # Conversion, if applicable
1419 #my ($self, $arg) = @_;
1420 if ($_[0]->{convert_where}) {
1421 return $_[0]->_sqlcase($_[0]->{convert_where}) .'(' . $_[1] . ')';
1428 #my ($self, $col, @vals) = @_;
1429 # called often - tighten code
1430 return $_[0]->{bindtype} eq 'columns'
1431 ? map {[$_[1], $_]} @_[2 .. $#_]
1436 # Dies if any element of @bind is not in [colname => value] format
1437 # if bindtype is 'columns'.
1438 sub _assert_bindval_matches_bindtype {
1439 # my ($self, @bind) = @_;
1441 if ($self->{bindtype} eq 'columns') {
1443 if (!defined $_ || ref($_) ne 'ARRAY' || @$_ != 2) {
1444 puke "bindtype 'columns' selected, you need to pass: [column_name => bind_value]"
1450 sub _join_sql_clauses {
1451 my ($self, $logic, $clauses_aref, $bind_aref) = @_;
1453 if (@$clauses_aref > 1) {
1454 my $join = " " . $self->_sqlcase($logic) . " ";
1455 my $sql = '( ' . join($join, @$clauses_aref) . ' )';
1456 return ($sql, @$bind_aref);
1458 elsif (@$clauses_aref) {
1459 return ($clauses_aref->[0], @$bind_aref); # no parentheses
1462 return (); # if no SQL, ignore @$bind_aref
1467 # Fix SQL case, if so requested
1469 # LDNOTE: if $self->{case} is true, then it contains 'lower', so we
1470 # don't touch the argument ... crooked logic, but let's not change it!
1471 return $_[0]->{case} ? $_[1] : uc($_[1]);
1475 #======================================================================
1476 # DISPATCHING FROM REFKIND
1477 #======================================================================
1480 my ($self, $data) = @_;
1482 return 'UNDEF' unless defined $data;
1484 # blessed objects are treated like scalars
1485 my $ref = (Scalar::Util::blessed $data) ? '' : ref $data;
1487 return 'SCALAR' unless $ref;
1490 while ($ref eq 'REF') {
1492 $ref = (Scalar::Util::blessed $data) ? '' : ref $data;
1496 return ($ref||'SCALAR') . ('REF' x $n_steps);
1500 my ($self, $data) = @_;
1501 my @try = ($self->_refkind($data));
1502 push @try, 'SCALAR_or_UNDEF' if $try[0] eq 'SCALAR' || $try[0] eq 'UNDEF';
1503 push @try, 'FALLBACK';
1507 sub _METHOD_FOR_refkind {
1508 my ($self, $meth_prefix, $data) = @_;
1511 for (@{$self->_try_refkind($data)}) {
1512 $method = $self->can($meth_prefix."_".$_)
1516 return $method || puke "cannot dispatch on '$meth_prefix' for ".$self->_refkind($data);
1520 sub _SWITCH_refkind {
1521 my ($self, $data, $dispatch_table) = @_;
1524 for (@{$self->_try_refkind($data)}) {
1525 $coderef = $dispatch_table->{$_}
1529 puke "no dispatch entry for ".$self->_refkind($data)
1538 #======================================================================
1539 # VALUES, GENERATE, AUTOLOAD
1540 #======================================================================
1542 # LDNOTE: original code from nwiger, didn't touch code in that section
1543 # I feel the AUTOLOAD stuff should not be the default, it should
1544 # only be activated on explicit demand by user.
1548 my $data = shift || return;
1549 puke "Argument to ", __PACKAGE__, "->values must be a \\%hash"
1550 unless ref $data eq 'HASH';
1553 foreach my $k (sort keys %$data) {
1554 my $v = $data->{$k};
1555 $self->_SWITCH_refkind($v, {
1557 if ($self->{array_datatypes}) { # array datatype
1558 push @all_bind, $self->_bindtype($k, $v);
1560 else { # literal SQL with bind
1561 my ($sql, @bind) = @$v;
1562 $self->_assert_bindval_matches_bindtype(@bind);
1563 push @all_bind, @bind;
1566 ARRAYREFREF => sub { # literal SQL with bind
1567 my ($sql, @bind) = @${$v};
1568 $self->_assert_bindval_matches_bindtype(@bind);
1569 push @all_bind, @bind;
1571 SCALARREF => sub { # literal SQL without bind
1573 SCALAR_or_UNDEF => sub {
1574 push @all_bind, $self->_bindtype($k, $v);
1585 my(@sql, @sqlq, @sqlv);
1589 if ($ref eq 'HASH') {
1590 for my $k (sort keys %$_) {
1593 my $label = $self->_quote($k);
1594 if ($r eq 'ARRAY') {
1595 # literal SQL with bind
1596 my ($sql, @bind) = @$v;
1597 $self->_assert_bindval_matches_bindtype(@bind);
1598 push @sqlq, "$label = $sql";
1600 } elsif ($r eq 'SCALAR') {
1601 # literal SQL without bind
1602 push @sqlq, "$label = $$v";
1604 push @sqlq, "$label = ?";
1605 push @sqlv, $self->_bindtype($k, $v);
1608 push @sql, $self->_sqlcase('set'), join ', ', @sqlq;
1609 } elsif ($ref eq 'ARRAY') {
1610 # unlike insert(), assume these are ONLY the column names, i.e. for SQL
1613 if ($r eq 'ARRAY') { # literal SQL with bind
1614 my ($sql, @bind) = @$v;
1615 $self->_assert_bindval_matches_bindtype(@bind);
1618 } elsif ($r eq 'SCALAR') { # literal SQL without bind
1619 # embedded literal SQL
1626 push @sql, '(' . join(', ', @sqlq) . ')';
1627 } elsif ($ref eq 'SCALAR') {
1631 # strings get case twiddled
1632 push @sql, $self->_sqlcase($_);
1636 my $sql = join ' ', @sql;
1638 # this is pretty tricky
1639 # if ask for an array, return ($stmt, @bind)
1640 # otherwise, s/?/shift @sqlv/ to put it inline
1642 return ($sql, @sqlv);
1644 1 while $sql =~ s/\?/my $d = shift(@sqlv);
1645 ref $d ? $d->[1] : $d/e;
1654 # This allows us to check for a local, then _form, attr
1656 my($name) = $AUTOLOAD =~ /.*::(.+)/;
1657 return $self->generate($name, @_);
1668 SQL::Abstract - Generate SQL from Perl data structures
1674 my $sql = SQL::Abstract->new;
1676 my($stmt, @bind) = $sql->select($source, \@fields, \%where, $order);
1678 my($stmt, @bind) = $sql->insert($table, \%fieldvals || \@values);
1680 my($stmt, @bind) = $sql->update($table, \%fieldvals, \%where);
1682 my($stmt, @bind) = $sql->delete($table, \%where);
1684 # Then, use these in your DBI statements
1685 my $sth = $dbh->prepare($stmt);
1686 $sth->execute(@bind);
1688 # Just generate the WHERE clause
1689 my($stmt, @bind) = $sql->where(\%where, $order);
1691 # Return values in the same order, for hashed queries
1692 # See PERFORMANCE section for more details
1693 my @bind = $sql->values(\%fieldvals);
1697 This module was inspired by the excellent L<DBIx::Abstract>.
1698 However, in using that module I found that what I really wanted
1699 to do was generate SQL, but still retain complete control over my
1700 statement handles and use the DBI interface. So, I set out to
1701 create an abstract SQL generation module.
1703 While based on the concepts used by L<DBIx::Abstract>, there are
1704 several important differences, especially when it comes to WHERE
1705 clauses. I have modified the concepts used to make the SQL easier
1706 to generate from Perl data structures and, IMO, more intuitive.
1707 The underlying idea is for this module to do what you mean, based
1708 on the data structures you provide it. The big advantage is that
1709 you don't have to modify your code every time your data changes,
1710 as this module figures it out.
1712 To begin with, an SQL INSERT is as easy as just specifying a hash
1713 of C<key=value> pairs:
1716 name => 'Jimbo Bobson',
1717 phone => '123-456-7890',
1718 address => '42 Sister Lane',
1719 city => 'St. Louis',
1720 state => 'Louisiana',
1723 The SQL can then be generated with this:
1725 my($stmt, @bind) = $sql->insert('people', \%data);
1727 Which would give you something like this:
1729 $stmt = "INSERT INTO people
1730 (address, city, name, phone, state)
1731 VALUES (?, ?, ?, ?, ?)";
1732 @bind = ('42 Sister Lane', 'St. Louis', 'Jimbo Bobson',
1733 '123-456-7890', 'Louisiana');
1735 These are then used directly in your DBI code:
1737 my $sth = $dbh->prepare($stmt);
1738 $sth->execute(@bind);
1740 =head2 Inserting and Updating Arrays
1742 If your database has array types (like for example Postgres),
1743 activate the special option C<< array_datatypes => 1 >>
1744 when creating the C<SQL::Abstract> object.
1745 Then you may use an arrayref to insert and update database array types:
1747 my $sql = SQL::Abstract->new(array_datatypes => 1);
1749 planets => [qw/Mercury Venus Earth Mars/]
1752 my($stmt, @bind) = $sql->insert('solar_system', \%data);
1756 $stmt = "INSERT INTO solar_system (planets) VALUES (?)"
1758 @bind = (['Mercury', 'Venus', 'Earth', 'Mars']);
1761 =head2 Inserting and Updating SQL
1763 In order to apply SQL functions to elements of your C<%data> you may
1764 specify a reference to an arrayref for the given hash value. For example,
1765 if you need to execute the Oracle C<to_date> function on a value, you can
1766 say something like this:
1770 date_entered => \[ "to_date(?,'MM/DD/YYYY')", "03/02/2003" ],
1773 The first value in the array is the actual SQL. Any other values are
1774 optional and would be included in the bind values array. This gives
1777 my($stmt, @bind) = $sql->insert('people', \%data);
1779 $stmt = "INSERT INTO people (name, date_entered)
1780 VALUES (?, to_date(?,'MM/DD/YYYY'))";
1781 @bind = ('Bill', '03/02/2003');
1783 An UPDATE is just as easy, all you change is the name of the function:
1785 my($stmt, @bind) = $sql->update('people', \%data);
1787 Notice that your C<%data> isn't touched; the module will generate
1788 the appropriately quirky SQL for you automatically. Usually you'll
1789 want to specify a WHERE clause for your UPDATE, though, which is
1790 where handling C<%where> hashes comes in handy...
1792 =head2 Complex where statements
1794 This module can generate pretty complicated WHERE statements
1795 easily. For example, simple C<key=value> pairs are taken to mean
1796 equality, and if you want to see if a field is within a set
1797 of values, you can use an arrayref. Let's say we wanted to
1798 SELECT some data based on this criteria:
1801 requestor => 'inna',
1802 worker => ['nwiger', 'rcwe', 'sfz'],
1803 status => { '!=', 'completed' }
1806 my($stmt, @bind) = $sql->select('tickets', '*', \%where);
1808 The above would give you something like this:
1810 $stmt = "SELECT * FROM tickets WHERE
1811 ( requestor = ? ) AND ( status != ? )
1812 AND ( worker = ? OR worker = ? OR worker = ? )";
1813 @bind = ('inna', 'completed', 'nwiger', 'rcwe', 'sfz');
1815 Which you could then use in DBI code like so:
1817 my $sth = $dbh->prepare($stmt);
1818 $sth->execute(@bind);
1824 The methods are simple. There's one for every major SQL operation,
1825 and a constructor you use first. The arguments are specified in a
1826 similar order for each method (table, then fields, then a where
1827 clause) to try and simplify things.
1829 =head2 new(option => 'value')
1831 The C<new()> function takes a list of options and values, and returns
1832 a new B<SQL::Abstract> object which can then be used to generate SQL
1833 through the methods below. The options accepted are:
1839 If set to 'lower', then SQL will be generated in all lowercase. By
1840 default SQL is generated in "textbook" case meaning something like:
1842 SELECT a_field FROM a_table WHERE some_field LIKE '%someval%'
1844 Any setting other than 'lower' is ignored.
1848 This determines what the default comparison operator is. By default
1849 it is C<=>, meaning that a hash like this:
1851 %where = (name => 'nwiger', email => 'nate@wiger.org');
1853 Will generate SQL like this:
1855 WHERE name = 'nwiger' AND email = 'nate@wiger.org'
1857 However, you may want loose comparisons by default, so if you set
1858 C<cmp> to C<like> you would get SQL such as:
1860 WHERE name like 'nwiger' AND email like 'nate@wiger.org'
1862 You can also override the comparison on an individual basis - see
1863 the huge section on L</"WHERE CLAUSES"> at the bottom.
1865 =item sqltrue, sqlfalse
1867 Expressions for inserting boolean values within SQL statements.
1868 By default these are C<1=1> and C<1=0>. They are used
1869 by the special operators C<-in> and C<-not_in> for generating
1870 correct SQL even when the argument is an empty array (see below).
1874 This determines the default logical operator for multiple WHERE
1875 statements in arrays or hashes. If absent, the default logic is "or"
1876 for arrays, and "and" for hashes. This means that a WHERE
1880 event_date => {'>=', '2/13/99'},
1881 event_date => {'<=', '4/24/03'},
1884 will generate SQL like this:
1886 WHERE event_date >= '2/13/99' OR event_date <= '4/24/03'
1888 This is probably not what you want given this query, though (look
1889 at the dates). To change the "OR" to an "AND", simply specify:
1891 my $sql = SQL::Abstract->new(logic => 'and');
1893 Which will change the above C<WHERE> to:
1895 WHERE event_date >= '2/13/99' AND event_date <= '4/24/03'
1897 The logic can also be changed locally by inserting
1898 a modifier in front of an arrayref:
1900 @where = (-and => [event_date => {'>=', '2/13/99'},
1901 event_date => {'<=', '4/24/03'} ]);
1903 See the L</"WHERE CLAUSES"> section for explanations.
1907 This will automatically convert comparisons using the specified SQL
1908 function for both column and value. This is mostly used with an argument
1909 of C<upper> or C<lower>, so that the SQL will have the effect of
1910 case-insensitive "searches". For example, this:
1912 $sql = SQL::Abstract->new(convert => 'upper');
1913 %where = (keywords => 'MaKe iT CAse inSeNSItive');
1915 Will turn out the following SQL:
1917 WHERE upper(keywords) like upper('MaKe iT CAse inSeNSItive')
1919 The conversion can be C<upper()>, C<lower()>, or any other SQL function
1920 that can be applied symmetrically to fields (actually B<SQL::Abstract> does
1921 not validate this option; it will just pass through what you specify verbatim).
1925 This is a kludge because many databases suck. For example, you can't
1926 just bind values using DBI's C<execute()> for Oracle C<CLOB> or C<BLOB> fields.
1927 Instead, you have to use C<bind_param()>:
1929 $sth->bind_param(1, 'reg data');
1930 $sth->bind_param(2, $lots, {ora_type => ORA_CLOB});
1932 The problem is, B<SQL::Abstract> will normally just return a C<@bind> array,
1933 which loses track of which field each slot refers to. Fear not.
1935 If you specify C<bindtype> in new, you can determine how C<@bind> is returned.
1936 Currently, you can specify either C<normal> (default) or C<columns>. If you
1937 specify C<columns>, you will get an array that looks like this:
1939 my $sql = SQL::Abstract->new(bindtype => 'columns');
1940 my($stmt, @bind) = $sql->insert(...);
1943 [ 'column1', 'value1' ],
1944 [ 'column2', 'value2' ],
1945 [ 'column3', 'value3' ],
1948 You can then iterate through this manually, using DBI's C<bind_param()>.
1950 $sth->prepare($stmt);
1953 my($col, $data) = @$_;
1954 if ($col eq 'details' || $col eq 'comments') {
1955 $sth->bind_param($i, $data, {ora_type => ORA_CLOB});
1956 } elsif ($col eq 'image') {
1957 $sth->bind_param($i, $data, {ora_type => ORA_BLOB});
1959 $sth->bind_param($i, $data);
1963 $sth->execute; # execute without @bind now
1965 Now, why would you still use B<SQL::Abstract> if you have to do this crap?
1966 Basically, the advantage is still that you don't have to care which fields
1967 are or are not included. You could wrap that above C<for> loop in a simple
1968 sub called C<bind_fields()> or something and reuse it repeatedly. You still
1969 get a layer of abstraction over manual SQL specification.
1971 Note that if you set L</bindtype> to C<columns>, the C<\[ $sql, @bind ]>
1972 construct (see L</Literal SQL with placeholders and bind values (subqueries)>)
1973 will expect the bind values in this format.
1977 This is the character that a table or column name will be quoted
1978 with. By default this is an empty string, but you could set it to
1979 the character C<`>, to generate SQL like this:
1981 SELECT `a_field` FROM `a_table` WHERE `some_field` LIKE '%someval%'
1983 Alternatively, you can supply an array ref of two items, the first being the left
1984 hand quote character, and the second the right hand quote character. For
1985 example, you could supply C<['[',']']> for SQL Server 2000 compliant quotes
1986 that generates SQL like this:
1988 SELECT [a_field] FROM [a_table] WHERE [some_field] LIKE '%someval%'
1990 Quoting is useful if you have tables or columns names that are reserved
1991 words in your database's SQL dialect.
1995 This is the character that will be used to escape L</quote_char>s appearing
1996 in an identifier before it has been quoted.
1998 The parameter default in case of a single L</quote_char> character is the quote
2001 When opening-closing-style quoting is used (L</quote_char> is an arrayref)
2002 this parameter defaults to the B<closing (right)> L</quote_char>. Occurrences
2003 of the B<opening (left)> L</quote_char> within the identifier are currently left
2004 untouched. The default for opening-closing-style quotes may change in future
2005 versions, thus you are B<strongly encouraged> to specify the escape character
2010 This is the character that separates a table and column name. It is
2011 necessary to specify this when the C<quote_char> option is selected,
2012 so that tables and column names can be individually quoted like this:
2014 SELECT `table`.`one_field` FROM `table` WHERE `table`.`other_field` = 1
2016 =item injection_guard
2018 A regular expression C<qr/.../> that is applied to any C<-function> and unquoted
2019 column name specified in a query structure. This is a safety mechanism to avoid
2020 injection attacks when mishandling user input e.g.:
2022 my %condition_as_column_value_pairs = get_values_from_user();
2023 $sqla->select( ... , \%condition_as_column_value_pairs );
2025 If the expression matches an exception is thrown. Note that literal SQL
2026 supplied via C<\'...'> or C<\['...']> is B<not> checked in any way.
2028 Defaults to checking for C<;> and the C<GO> keyword (TransactSQL)
2030 =item array_datatypes
2032 When this option is true, arrayrefs in INSERT or UPDATE are
2033 interpreted as array datatypes and are passed directly
2035 When this option is false, arrayrefs are interpreted
2036 as literal SQL, just like refs to arrayrefs
2037 (but this behavior is for backwards compatibility; when writing
2038 new queries, use the "reference to arrayref" syntax
2044 Takes a reference to a list of "special operators"
2045 to extend the syntax understood by L<SQL::Abstract>.
2046 See section L</"SPECIAL OPERATORS"> for details.
2050 Takes a reference to a list of "unary operators"
2051 to extend the syntax understood by L<SQL::Abstract>.
2052 See section L</"UNARY OPERATORS"> for details.
2058 =head2 insert($table, \@values || \%fieldvals, \%options)
2060 This is the simplest function. You simply give it a table name
2061 and either an arrayref of values or hashref of field/value pairs.
2062 It returns an SQL INSERT statement and a list of bind values.
2063 See the sections on L</"Inserting and Updating Arrays"> and
2064 L</"Inserting and Updating SQL"> for information on how to insert
2065 with those data types.
2067 The optional C<\%options> hash reference may contain additional
2068 options to generate the insert SQL. Currently supported options
2075 Takes either a scalar of raw SQL fields, or an array reference of
2076 field names, and adds on an SQL C<RETURNING> statement at the end.
2077 This allows you to return data generated by the insert statement
2078 (such as row IDs) without performing another C<SELECT> statement.
2079 Note, however, this is not part of the SQL standard and may not
2080 be supported by all database engines.
2084 =head2 update($table, \%fieldvals, \%where, \%options)
2086 This takes a table, hashref of field/value pairs, and an optional
2087 hashref L<WHERE clause|/WHERE CLAUSES>. It returns an SQL UPDATE function and a list
2089 See the sections on L</"Inserting and Updating Arrays"> and
2090 L</"Inserting and Updating SQL"> for information on how to insert
2091 with those data types.
2093 The optional C<\%options> hash reference may contain additional
2094 options to generate the update SQL. Currently supported options
2101 See the C<returning> option to
2102 L<insert|/insert($table, \@values || \%fieldvals, \%options)>.
2106 =head2 select($source, $fields, $where, $order)
2108 This returns a SQL SELECT statement and associated list of bind values, as
2109 specified by the arguments:
2115 Specification of the 'FROM' part of the statement.
2116 The argument can be either a plain scalar (interpreted as a table
2117 name, will be quoted), or an arrayref (interpreted as a list
2118 of table names, joined by commas, quoted), or a scalarref
2119 (literal SQL, not quoted).
2123 Specification of the list of fields to retrieve from
2125 The argument can be either an arrayref (interpreted as a list
2126 of field names, will be joined by commas and quoted), or a
2127 plain scalar (literal SQL, not quoted).
2128 Please observe that this API is not as flexible as that of
2129 the first argument C<$source>, for backwards compatibility reasons.
2133 Optional argument to specify the WHERE part of the query.
2134 The argument is most often a hashref, but can also be
2135 an arrayref or plain scalar --
2136 see section L<WHERE clause|/"WHERE CLAUSES"> for details.
2140 Optional argument to specify the ORDER BY part of the query.
2141 The argument can be a scalar, a hashref or an arrayref
2142 -- see section L<ORDER BY clause|/"ORDER BY CLAUSES">
2148 =head2 delete($table, \%where, \%options)
2150 This takes a table name and optional hashref L<WHERE clause|/WHERE CLAUSES>.
2151 It returns an SQL DELETE statement and list of bind values.
2153 The optional C<\%options> hash reference may contain additional
2154 options to generate the delete SQL. Currently supported options
2161 See the C<returning> option to
2162 L<insert|/insert($table, \@values || \%fieldvals, \%options)>.
2166 =head2 where(\%where, $order)
2168 This is used to generate just the WHERE clause. For example,
2169 if you have an arbitrary data structure and know what the
2170 rest of your SQL is going to look like, but want an easy way
2171 to produce a WHERE clause, use this. It returns an SQL WHERE
2172 clause and list of bind values.
2175 =head2 values(\%data)
2177 This just returns the values from the hash C<%data>, in the same
2178 order that would be returned from any of the other above queries.
2179 Using this allows you to markedly speed up your queries if you
2180 are affecting lots of rows. See below under the L</"PERFORMANCE"> section.
2182 =head2 generate($any, 'number', $of, \@data, $struct, \%types)
2184 Warning: This is an experimental method and subject to change.
2186 This returns arbitrarily generated SQL. It's a really basic shortcut.
2187 It will return two different things, depending on return context:
2189 my($stmt, @bind) = $sql->generate('create table', \$table, \@fields);
2190 my $stmt_and_val = $sql->generate('create table', \$table, \@fields);
2192 These would return the following:
2194 # First calling form
2195 $stmt = "CREATE TABLE test (?, ?)";
2196 @bind = (field1, field2);
2198 # Second calling form
2199 $stmt_and_val = "CREATE TABLE test (field1, field2)";
2201 Depending on what you're trying to do, it's up to you to choose the correct
2202 format. In this example, the second form is what you would want.
2206 $sql->generate('alter session', { nls_date_format => 'MM/YY' });
2210 ALTER SESSION SET nls_date_format = 'MM/YY'
2212 You get the idea. Strings get their case twiddled, but everything
2213 else remains verbatim.
2215 =head1 EXPORTABLE FUNCTIONS
2217 =head2 is_plain_value
2219 Determines if the supplied argument is a plain value as understood by this
2224 =item * The value is C<undef>
2226 =item * The value is a non-reference
2228 =item * The value is an object with stringification overloading
2230 =item * The value is of the form C<< { -value => $anything } >>
2234 On failure returns C<undef>, on success returns a B<scalar> reference
2235 to the original supplied argument.
2241 The stringification overloading detection is rather advanced: it takes
2242 into consideration not only the presence of a C<""> overload, but if that
2243 fails also checks for enabled
2244 L<autogenerated versions of C<"">|overload/Magic Autogeneration>, based
2245 on either C<0+> or C<bool>.
2247 Unfortunately testing in the field indicates that this
2248 detection B<< may tickle a latent bug in perl versions before 5.018 >>,
2249 but only when very large numbers of stringifying objects are involved.
2250 At the time of writing ( Sep 2014 ) there is no clear explanation of
2251 the direct cause, nor is there a manageably small test case that reliably
2252 reproduces the problem.
2254 If you encounter any of the following exceptions in B<random places within
2255 your application stack> - this module may be to blame:
2257 Operation "ne": no method found,
2258 left argument in overloaded package <something>,
2259 right argument in overloaded package <something>
2263 Stub found while resolving method "???" overloading """" in package <something>
2265 If you fall victim to the above - please attempt to reduce the problem
2266 to something that could be sent to the L<SQL::Abstract developers
2267 |DBIx::Class/GETTING HELP/SUPPORT>
2268 (either publicly or privately). As a workaround in the meantime you can
2269 set C<$ENV{SQLA_ISVALUE_IGNORE_AUTOGENERATED_STRINGIFICATION}> to a true
2270 value, which will most likely eliminate your problem (at the expense of
2271 not being able to properly detect exotic forms of stringification).
2273 This notice and environment variable will be removed in a future version,
2274 as soon as the underlying problem is found and a reliable workaround is
2279 =head2 is_literal_value
2281 Determines if the supplied argument is a literal value as understood by this
2286 =item * C<\$sql_string>
2288 =item * C<\[ $sql_string, @bind_values ]>
2292 On failure returns C<undef>, on success returns an B<array> reference
2293 containing the unpacked version of the supplied literal SQL and bind values.
2295 =head1 WHERE CLAUSES
2299 This module uses a variation on the idea from L<DBIx::Abstract>. It
2300 is B<NOT>, repeat I<not> 100% compatible. B<The main logic of this
2301 module is that things in arrays are OR'ed, and things in hashes
2304 The easiest way to explain is to show lots of examples. After
2305 each C<%where> hash shown, it is assumed you used:
2307 my($stmt, @bind) = $sql->where(\%where);
2309 However, note that the C<%where> hash can be used directly in any
2310 of the other functions as well, as described above.
2312 =head2 Key-value pairs
2314 So, let's get started. To begin, a simple hash:
2318 status => 'completed'
2321 Is converted to SQL C<key = val> statements:
2323 $stmt = "WHERE user = ? AND status = ?";
2324 @bind = ('nwiger', 'completed');
2326 One common thing I end up doing is having a list of values that
2327 a field can be in. To do this, simply specify a list inside of
2332 status => ['assigned', 'in-progress', 'pending'];
2335 This simple code will create the following:
2337 $stmt = "WHERE user = ? AND ( status = ? OR status = ? OR status = ? )";
2338 @bind = ('nwiger', 'assigned', 'in-progress', 'pending');
2340 A field associated to an empty arrayref will be considered a
2341 logical false and will generate 0=1.
2343 =head2 Tests for NULL values
2345 If the value part is C<undef> then this is converted to SQL <IS NULL>
2354 $stmt = "WHERE user = ? AND status IS NULL";
2357 To test if a column IS NOT NULL:
2361 status => { '!=', undef },
2364 =head2 Specific comparison operators
2366 If you want to specify a different type of operator for your comparison,
2367 you can use a hashref for a given column:
2371 status => { '!=', 'completed' }
2374 Which would generate:
2376 $stmt = "WHERE user = ? AND status != ?";
2377 @bind = ('nwiger', 'completed');
2379 To test against multiple values, just enclose the values in an arrayref:
2381 status => { '=', ['assigned', 'in-progress', 'pending'] };
2383 Which would give you:
2385 "WHERE status = ? OR status = ? OR status = ?"
2388 The hashref can also contain multiple pairs, in which case it is expanded
2389 into an C<AND> of its elements:
2393 status => { '!=', 'completed', -not_like => 'pending%' }
2396 # Or more dynamically, like from a form
2397 $where{user} = 'nwiger';
2398 $where{status}{'!='} = 'completed';
2399 $where{status}{'-not_like'} = 'pending%';
2401 # Both generate this
2402 $stmt = "WHERE user = ? AND status != ? AND status NOT LIKE ?";
2403 @bind = ('nwiger', 'completed', 'pending%');
2406 To get an OR instead, you can combine it with the arrayref idea:
2410 priority => [ { '=', 2 }, { '>', 5 } ]
2413 Which would generate:
2415 $stmt = "WHERE ( priority = ? OR priority > ? ) AND user = ?";
2416 @bind = ('2', '5', 'nwiger');
2418 If you want to include literal SQL (with or without bind values), just use a
2419 scalar reference or reference to an arrayref as the value:
2422 date_entered => { '>' => \["to_date(?, 'MM/DD/YYYY')", "11/26/2008"] },
2423 date_expires => { '<' => \"now()" }
2426 Which would generate:
2428 $stmt = "WHERE date_entered > to_date(?, 'MM/DD/YYYY') AND date_expires < now()";
2429 @bind = ('11/26/2008');
2432 =head2 Logic and nesting operators
2434 In the example above,
2435 there is a subtle trap if you want to say something like
2436 this (notice the C<AND>):
2438 WHERE priority != ? AND priority != ?
2440 Because, in Perl you I<can't> do this:
2442 priority => { '!=' => 2, '!=' => 1 }
2444 As the second C<!=> key will obliterate the first. The solution
2445 is to use the special C<-modifier> form inside an arrayref:
2447 priority => [ -and => {'!=', 2},
2451 Normally, these would be joined by C<OR>, but the modifier tells it
2452 to use C<AND> instead. (Hint: You can use this in conjunction with the
2453 C<logic> option to C<new()> in order to change the way your queries
2454 work by default.) B<Important:> Note that the C<-modifier> goes
2455 B<INSIDE> the arrayref, as an extra first element. This will
2456 B<NOT> do what you think it might:
2458 priority => -and => [{'!=', 2}, {'!=', 1}] # WRONG!
2460 Here is a quick list of equivalencies, since there is some overlap:
2463 status => {'!=', 'completed', 'not like', 'pending%' }
2464 status => [ -and => {'!=', 'completed'}, {'not like', 'pending%'}]
2467 status => {'=', ['assigned', 'in-progress']}
2468 status => [ -or => {'=', 'assigned'}, {'=', 'in-progress'}]
2469 status => [ {'=', 'assigned'}, {'=', 'in-progress'} ]
2473 =head2 Special operators: IN, BETWEEN, etc.
2475 You can also use the hashref format to compare a list of fields using the
2476 C<IN> comparison operator, by specifying the list as an arrayref:
2479 status => 'completed',
2480 reportid => { -in => [567, 2335, 2] }
2483 Which would generate:
2485 $stmt = "WHERE status = ? AND reportid IN (?,?,?)";
2486 @bind = ('completed', '567', '2335', '2');
2488 The reverse operator C<-not_in> generates SQL C<NOT IN> and is used in
2491 If the argument to C<-in> is an empty array, 'sqlfalse' is generated
2492 (by default: C<1=0>). Similarly, C<< -not_in => [] >> generates
2493 'sqltrue' (by default: C<1=1>).
2495 In addition to the array you can supply a chunk of literal sql or
2496 literal sql with bind:
2499 customer => { -in => \[
2500 'SELECT cust_id FROM cust WHERE balance > ?',
2503 status => { -in => \'SELECT status_codes FROM states' },
2509 customer IN ( SELECT cust_id FROM cust WHERE balance > ? )
2510 AND status IN ( SELECT status_codes FROM states )
2514 Finally, if the argument to C<-in> is not a reference, it will be
2515 treated as a single-element array.
2517 Another pair of operators is C<-between> and C<-not_between>,
2518 used with an arrayref of two values:
2522 completion_date => {
2523 -not_between => ['2002-10-01', '2003-02-06']
2529 WHERE user = ? AND completion_date NOT BETWEEN ( ? AND ? )
2531 Just like with C<-in> all plausible combinations of literal SQL
2535 start0 => { -between => [ 1, 2 ] },
2536 start1 => { -between => \["? AND ?", 1, 2] },
2537 start2 => { -between => \"lower(x) AND upper(y)" },
2538 start3 => { -between => [
2540 \["upper(?)", 'stuff' ],
2547 ( start0 BETWEEN ? AND ? )
2548 AND ( start1 BETWEEN ? AND ? )
2549 AND ( start2 BETWEEN lower(x) AND upper(y) )
2550 AND ( start3 BETWEEN lower(x) AND upper(?) )
2552 @bind = (1, 2, 1, 2, 'stuff');
2555 These are the two builtin "special operators"; but the
2556 list can be expanded: see section L</"SPECIAL OPERATORS"> below.
2558 =head2 Unary operators: bool
2560 If you wish to test against boolean columns or functions within your
2561 database you can use the C<-bool> and C<-not_bool> operators. For
2562 example to test the column C<is_user> being true and the column
2563 C<is_enabled> being false you would use:-
2567 -not_bool => 'is_enabled',
2572 WHERE is_user AND NOT is_enabled
2574 If a more complex combination is required, testing more conditions,
2575 then you should use the and/or operators:-
2580 -not_bool => { two=> { -rlike => 'bar' } },
2581 -not_bool => { three => [ { '=', 2 }, { '>', 5 } ] },
2592 (NOT ( three = ? OR three > ? ))
2595 =head2 Nested conditions, -and/-or prefixes
2597 So far, we've seen how multiple conditions are joined with a top-level
2598 C<AND>. We can change this by putting the different conditions we want in
2599 hashes and then putting those hashes in an array. For example:
2604 status => { -like => ['pending%', 'dispatched'] },
2608 status => 'unassigned',
2612 This data structure would create the following:
2614 $stmt = "WHERE ( user = ? AND ( status LIKE ? OR status LIKE ? ) )
2615 OR ( user = ? AND status = ? ) )";
2616 @bind = ('nwiger', 'pending', 'dispatched', 'robot', 'unassigned');
2619 Clauses in hashrefs or arrayrefs can be prefixed with an C<-and> or C<-or>
2620 to change the logic inside:
2626 -and => [ workhrs => {'>', 20}, geo => 'ASIA' ],
2627 -or => { workhrs => {'<', 50}, geo => 'EURO' },
2634 $stmt = "WHERE ( user = ?
2635 AND ( ( workhrs > ? AND geo = ? )
2636 OR ( workhrs < ? OR geo = ? ) ) )";
2637 @bind = ('nwiger', '20', 'ASIA', '50', 'EURO');
2639 =head3 Algebraic inconsistency, for historical reasons
2641 C<Important note>: when connecting several conditions, the C<-and->|C<-or>
2642 operator goes C<outside> of the nested structure; whereas when connecting
2643 several constraints on one column, the C<-and> operator goes
2644 C<inside> the arrayref. Here is an example combining both features:
2647 -and => [a => 1, b => 2],
2648 -or => [c => 3, d => 4],
2649 e => [-and => {-like => 'foo%'}, {-like => '%bar'} ]
2654 WHERE ( ( ( a = ? AND b = ? )
2655 OR ( c = ? OR d = ? )
2656 OR ( e LIKE ? AND e LIKE ? ) ) )
2658 This difference in syntax is unfortunate but must be preserved for
2659 historical reasons. So be careful: the two examples below would
2660 seem algebraically equivalent, but they are not
2663 { -like => 'foo%' },
2664 { -like => '%bar' },
2666 # yields: WHERE ( ( col LIKE ? AND col LIKE ? ) )
2669 { col => { -like => 'foo%' } },
2670 { col => { -like => '%bar' } },
2672 # yields: WHERE ( ( col LIKE ? OR col LIKE ? ) )
2675 =head2 Literal SQL and value type operators
2677 The basic premise of SQL::Abstract is that in WHERE specifications the "left
2678 side" is a column name and the "right side" is a value (normally rendered as
2679 a placeholder). This holds true for both hashrefs and arrayref pairs as you
2680 see in the L</WHERE CLAUSES> examples above. Sometimes it is necessary to
2681 alter this behavior. There are several ways of doing so.
2685 This is a virtual operator that signals the string to its right side is an
2686 identifier (a column name) and not a value. For example to compare two
2687 columns you would write:
2690 priority => { '<', 2 },
2691 requestor => { -ident => 'submitter' },
2696 $stmt = "WHERE priority < ? AND requestor = submitter";
2699 If you are maintaining legacy code you may see a different construct as
2700 described in L</Deprecated usage of Literal SQL>, please use C<-ident> in new
2705 This is a virtual operator that signals that the construct to its right side
2706 is a value to be passed to DBI. This is for example necessary when you want
2707 to write a where clause against an array (for RDBMS that support such
2708 datatypes). For example:
2711 array => { -value => [1, 2, 3] }
2716 $stmt = 'WHERE array = ?';
2717 @bind = ([1, 2, 3]);
2719 Note that if you were to simply say:
2725 the result would probably not be what you wanted:
2727 $stmt = 'WHERE array = ? OR array = ? OR array = ?';
2732 Finally, sometimes only literal SQL will do. To include a random snippet
2733 of SQL verbatim, you specify it as a scalar reference. Consider this only
2734 as a last resort. Usually there is a better way. For example:
2737 priority => { '<', 2 },
2738 requestor => { -in => \'(SELECT name FROM hitmen)' },
2743 $stmt = "WHERE priority < ? AND requestor IN (SELECT name FROM hitmen)"
2746 Note that in this example, you only get one bind parameter back, since
2747 the verbatim SQL is passed as part of the statement.
2751 Never use untrusted input as a literal SQL argument - this is a massive
2752 security risk (there is no way to check literal snippets for SQL
2753 injections and other nastyness). If you need to deal with untrusted input
2754 use literal SQL with placeholders as described next.
2756 =head3 Literal SQL with placeholders and bind values (subqueries)
2758 If the literal SQL to be inserted has placeholders and bind values,
2759 use a reference to an arrayref (yes this is a double reference --
2760 not so common, but perfectly legal Perl). For example, to find a date
2761 in Postgres you can use something like this:
2764 date_column => \[ "= date '2008-09-30' - ?::integer", 10 ]
2769 $stmt = "WHERE ( date_column = date '2008-09-30' - ?::integer )"
2772 Note that you must pass the bind values in the same format as they are returned
2773 by L<where|/where(\%where, $order)>. This means that if you set L</bindtype>
2774 to C<columns>, you must provide the bind values in the
2775 C<< [ column_meta => value ] >> format, where C<column_meta> is an opaque
2776 scalar value; most commonly the column name, but you can use any scalar value
2777 (including references and blessed references), L<SQL::Abstract> will simply
2778 pass it through intact. So if C<bindtype> is set to C<columns> the above
2779 example will look like:
2782 date_column => \[ "= date '2008-09-30' - ?::integer", [ {} => 10 ] ]
2785 Literal SQL is especially useful for nesting parenthesized clauses in the
2786 main SQL query. Here is a first example:
2788 my ($sub_stmt, @sub_bind) = ("SELECT c1 FROM t1 WHERE c2 < ? AND c3 LIKE ?",
2792 bar => \["IN ($sub_stmt)" => @sub_bind],
2797 $stmt = "WHERE (foo = ? AND bar IN (SELECT c1 FROM t1
2798 WHERE c2 < ? AND c3 LIKE ?))";
2799 @bind = (1234, 100, "foo%");
2801 Other subquery operators, like for example C<"E<gt> ALL"> or C<"NOT IN">,
2802 are expressed in the same way. Of course the C<$sub_stmt> and
2803 its associated bind values can be generated through a former call
2806 my ($sub_stmt, @sub_bind)
2807 = $sql->select("t1", "c1", {c2 => {"<" => 100},
2808 c3 => {-like => "foo%"}});
2811 bar => \["> ALL ($sub_stmt)" => @sub_bind],
2814 In the examples above, the subquery was used as an operator on a column;
2815 but the same principle also applies for a clause within the main C<%where>
2816 hash, like an EXISTS subquery:
2818 my ($sub_stmt, @sub_bind)
2819 = $sql->select("t1", "*", {c1 => 1, c2 => \"> t0.c0"});
2820 my %where = ( -and => [
2822 \["EXISTS ($sub_stmt)" => @sub_bind],
2827 $stmt = "WHERE (foo = ? AND EXISTS (SELECT * FROM t1
2828 WHERE c1 = ? AND c2 > t0.c0))";
2832 Observe that the condition on C<c2> in the subquery refers to
2833 column C<t0.c0> of the main query: this is I<not> a bind
2834 value, so we have to express it through a scalar ref.
2835 Writing C<< c2 => {">" => "t0.c0"} >> would have generated
2836 C<< c2 > ? >> with bind value C<"t0.c0"> ... not exactly
2837 what we wanted here.
2839 Finally, here is an example where a subquery is used
2840 for expressing unary negation:
2842 my ($sub_stmt, @sub_bind)
2843 = $sql->where({age => [{"<" => 10}, {">" => 20}]});
2844 $sub_stmt =~ s/^ where //i; # don't want "WHERE" in the subclause
2846 lname => {like => '%son%'},
2847 \["NOT ($sub_stmt)" => @sub_bind],
2852 $stmt = "lname LIKE ? AND NOT ( age < ? OR age > ? )"
2853 @bind = ('%son%', 10, 20)
2855 =head3 Deprecated usage of Literal SQL
2857 Below are some examples of archaic use of literal SQL. It is shown only as
2858 reference for those who deal with legacy code. Each example has a much
2859 better, cleaner and safer alternative that users should opt for in new code.
2865 my %where = ( requestor => \'IS NOT NULL' )
2867 $stmt = "WHERE requestor IS NOT NULL"
2869 This used to be the way of generating NULL comparisons, before the handling
2870 of C<undef> got formalized. For new code please use the superior syntax as
2871 described in L</Tests for NULL values>.
2875 my %where = ( requestor => \'= submitter' )
2877 $stmt = "WHERE requestor = submitter"
2879 This used to be the only way to compare columns. Use the superior L</-ident>
2880 method for all new code. For example an identifier declared in such a way
2881 will be properly quoted if L</quote_char> is properly set, while the legacy
2882 form will remain as supplied.
2886 my %where = ( is_ready => \"", completed => { '>', '2012-12-21' } )
2888 $stmt = "WHERE completed > ? AND is_ready"
2889 @bind = ('2012-12-21')
2891 Using an empty string literal used to be the only way to express a boolean.
2892 For all new code please use the much more readable
2893 L<-bool|/Unary operators: bool> operator.
2899 These pages could go on for a while, since the nesting of the data
2900 structures this module can handle are pretty much unlimited (the
2901 module implements the C<WHERE> expansion as a recursive function
2902 internally). Your best bet is to "play around" with the module a
2903 little to see how the data structures behave, and choose the best
2904 format for your data based on that.
2906 And of course, all the values above will probably be replaced with
2907 variables gotten from forms or the command line. After all, if you
2908 knew everything ahead of time, you wouldn't have to worry about
2909 dynamically-generating SQL and could just hardwire it into your
2912 =head1 ORDER BY CLAUSES
2914 Some functions take an order by clause. This can either be a scalar (just a
2915 column name), a hashref of C<< { -desc => 'col' } >> or C<< { -asc => 'col' }
2916 >>, a scalarref, an arrayref-ref, or an arrayref of any of the previous
2919 Given | Will Generate
2920 ---------------------------------------------------------------
2922 'colA' | ORDER BY colA
2924 [qw/colA colB/] | ORDER BY colA, colB
2926 {-asc => 'colA'} | ORDER BY colA ASC
2928 {-desc => 'colB'} | ORDER BY colB DESC
2930 ['colA', {-asc => 'colB'}] | ORDER BY colA, colB ASC
2932 { -asc => [qw/colA colB/] } | ORDER BY colA ASC, colB ASC
2934 \'colA DESC' | ORDER BY colA DESC
2936 \[ 'FUNC(colA, ?)', $x ] | ORDER BY FUNC(colA, ?)
2937 | /* ...with $x bound to ? */
2940 { -asc => 'colA' }, | colA ASC,
2941 { -desc => [qw/colB/] }, | colB DESC,
2942 { -asc => [qw/colC colD/] },| colC ASC, colD ASC,
2943 \'colE DESC', | colE DESC,
2944 \[ 'FUNC(colF, ?)', $x ], | FUNC(colF, ?)
2945 ] | /* ...with $x bound to ? */
2946 ===============================================================
2950 =head1 SPECIAL OPERATORS
2952 my $sqlmaker = SQL::Abstract->new(special_ops => [
2956 my ($self, $field, $op, $arg) = @_;
2962 handler => 'method_name',
2966 A "special operator" is a SQL syntactic clause that can be
2967 applied to a field, instead of a usual binary operator.
2970 WHERE field IN (?, ?, ?)
2971 WHERE field BETWEEN ? AND ?
2972 WHERE MATCH(field) AGAINST (?, ?)
2974 Special operators IN and BETWEEN are fairly standard and therefore
2975 are builtin within C<SQL::Abstract> (as the overridable methods
2976 C<_where_field_IN> and C<_where_field_BETWEEN>). For other operators,
2977 like the MATCH .. AGAINST example above which is specific to MySQL,
2978 you can write your own operator handlers - supply a C<special_ops>
2979 argument to the C<new> method. That argument takes an arrayref of
2980 operator definitions; each operator definition is a hashref with two
2987 the regular expression to match the operator
2991 Either a coderef or a plain scalar method name. In both cases
2992 the expected return is C<< ($sql, @bind) >>.
2994 When supplied with a method name, it is simply called on the
2995 L<SQL::Abstract> object as:
2997 $self->$method_name($field, $op, $arg)
3001 $field is the LHS of the operator
3002 $op is the part that matched the handler regex
3005 When supplied with a coderef, it is called as:
3007 $coderef->($self, $field, $op, $arg)
3012 For example, here is an implementation
3013 of the MATCH .. AGAINST syntax for MySQL
3015 my $sqlmaker = SQL::Abstract->new(special_ops => [
3017 # special op for MySql MATCH (field) AGAINST(word1, word2, ...)
3018 {regex => qr/^match$/i,
3020 my ($self, $field, $op, $arg) = @_;
3021 $arg = [$arg] if not ref $arg;
3022 my $label = $self->_quote($field);
3023 my ($placeholder) = $self->_convert('?');
3024 my $placeholders = join ", ", (($placeholder) x @$arg);
3025 my $sql = $self->_sqlcase('match') . " ($label) "
3026 . $self->_sqlcase('against') . " ($placeholders) ";
3027 my @bind = $self->_bindtype($field, @$arg);
3028 return ($sql, @bind);
3035 =head1 UNARY OPERATORS
3037 my $sqlmaker = SQL::Abstract->new(unary_ops => [
3041 my ($self, $op, $arg) = @_;
3047 handler => 'method_name',
3051 A "unary operator" is a SQL syntactic clause that can be
3052 applied to a field - the operator goes before the field
3054 You can write your own operator handlers - supply a C<unary_ops>
3055 argument to the C<new> method. That argument takes an arrayref of
3056 operator definitions; each operator definition is a hashref with two
3063 the regular expression to match the operator
3067 Either a coderef or a plain scalar method name. In both cases
3068 the expected return is C<< $sql >>.
3070 When supplied with a method name, it is simply called on the
3071 L<SQL::Abstract> object as:
3073 $self->$method_name($op, $arg)
3077 $op is the part that matched the handler regex
3078 $arg is the RHS or argument of the operator
3080 When supplied with a coderef, it is called as:
3082 $coderef->($self, $op, $arg)
3090 Thanks to some benchmarking by Mark Stosberg, it turns out that
3091 this module is many orders of magnitude faster than using C<DBIx::Abstract>.
3092 I must admit this wasn't an intentional design issue, but it's a
3093 byproduct of the fact that you get to control your C<DBI> handles
3096 To maximize performance, use a code snippet like the following:
3098 # prepare a statement handle using the first row
3099 # and then reuse it for the rest of the rows
3101 for my $href (@array_of_hashrefs) {
3102 $stmt ||= $sql->insert('table', $href);
3103 $sth ||= $dbh->prepare($stmt);
3104 $sth->execute($sql->values($href));
3107 The reason this works is because the keys in your C<$href> are sorted
3108 internally by B<SQL::Abstract>. Thus, as long as your data retains
3109 the same structure, you only have to generate the SQL the first time
3110 around. On subsequent queries, simply use the C<values> function provided
3111 by this module to return your values in the correct order.
3113 However this depends on the values having the same type - if, for
3114 example, the values of a where clause may either have values
3115 (resulting in sql of the form C<column = ?> with a single bind
3116 value), or alternatively the values might be C<undef> (resulting in
3117 sql of the form C<column IS NULL> with no bind value) then the
3118 caching technique suggested will not work.
3122 If you use my C<CGI::FormBuilder> module at all, you'll hopefully
3123 really like this part (I do, at least). Building up a complex query
3124 can be as simple as the following:
3131 use CGI::FormBuilder;
3134 my $form = CGI::FormBuilder->new(...);
3135 my $sql = SQL::Abstract->new;
3137 if ($form->submitted) {
3138 my $field = $form->field;
3139 my $id = delete $field->{id};
3140 my($stmt, @bind) = $sql->update('table', $field, {id => $id});
3143 Of course, you would still have to connect using C<DBI> to run the
3144 query, but the point is that if you make your form look like your
3145 table, the actual query script can be extremely simplistic.
3147 If you're B<REALLY> lazy (I am), check out C<HTML::QuickTable> for
3148 a fast interface to returning and formatting data. I frequently
3149 use these three modules together to write complex database query
3150 apps in under 50 lines.
3152 =head1 HOW TO CONTRIBUTE
3154 Contributions are always welcome, in all usable forms (we especially
3155 welcome documentation improvements). The delivery methods include git-
3156 or unified-diff formatted patches, GitHub pull requests, or plain bug
3157 reports either via RT or the Mailing list. Contributors are generally
3158 granted full access to the official repository after their first several
3159 patches pass successful review.
3161 This project is maintained in a git repository. The code and related tools are
3162 accessible at the following locations:
3166 =item * Official repo: L<git://git.shadowcat.co.uk/dbsrgits/SQL-Abstract.git>
3168 =item * Official gitweb: L<http://git.shadowcat.co.uk/gitweb/gitweb.cgi?p=dbsrgits/SQL-Abstract.git>
3170 =item * GitHub mirror: L<https://github.com/dbsrgits/sql-abstract>
3172 =item * Authorized committers: L<ssh://dbsrgits@git.shadowcat.co.uk/SQL-Abstract.git>
3178 Version 1.50 was a major internal refactoring of C<SQL::Abstract>.
3179 Great care has been taken to preserve the I<published> behavior
3180 documented in previous versions in the 1.* family; however,
3181 some features that were previously undocumented, or behaved
3182 differently from the documentation, had to be changed in order
3183 to clarify the semantics. Hence, client code that was relying
3184 on some dark areas of C<SQL::Abstract> v1.*
3185 B<might behave differently> in v1.50.
3187 The main changes are:
3193 support for literal SQL through the C<< \ [ $sql, @bind ] >> syntax.
3197 support for the { operator => \"..." } construct (to embed literal SQL)
3201 support for the { operator => \["...", @bind] } construct (to embed literal SQL with bind values)
3205 optional support for L<array datatypes|/"Inserting and Updating Arrays">
3209 defensive programming: check arguments
3213 fixed bug with global logic, which was previously implemented
3214 through global variables yielding side-effects. Prior versions would
3215 interpret C<< [ {cond1, cond2}, [cond3, cond4] ] >>
3216 as C<< "(cond1 AND cond2) OR (cond3 AND cond4)" >>.
3217 Now this is interpreted
3218 as C<< "(cond1 AND cond2) OR (cond3 OR cond4)" >>.
3223 fixed semantics of _bindtype on array args
3227 dropped the C<_anoncopy> of the %where tree. No longer necessary,
3228 we just avoid shifting arrays within that tree.
3232 dropped the C<_modlogic> function
3236 =head1 ACKNOWLEDGEMENTS
3238 There are a number of individuals that have really helped out with
3239 this module. Unfortunately, most of them submitted bugs via CPAN
3240 so I have no idea who they are! But the people I do know are:
3242 Ash Berlin (order_by hash term support)
3243 Matt Trout (DBIx::Class support)
3244 Mark Stosberg (benchmarking)
3245 Chas Owens (initial "IN" operator support)
3246 Philip Collins (per-field SQL functions)
3247 Eric Kolve (hashref "AND" support)
3248 Mike Fragassi (enhancements to "BETWEEN" and "LIKE")
3249 Dan Kubb (support for "quote_char" and "name_sep")
3250 Guillermo Roditi (patch to cleanup "IN" and "BETWEEN", fix and tests for _order_by)
3251 Laurent Dami (internal refactoring, extensible list of special operators, literal SQL)
3252 Norbert Buchmuller (support for literal SQL in hashpair, misc. fixes & tests)
3253 Peter Rabbitson (rewrite of SQLA::Test, misc. fixes & tests)
3254 Oliver Charles (support for "RETURNING" after "INSERT")
3260 L<DBIx::Class>, L<DBIx::Abstract>, L<CGI::FormBuilder>, L<HTML::QuickTable>.
3264 Copyright (c) 2001-2007 Nathan Wiger <nwiger@cpan.org>. All Rights Reserved.
3266 This module is actively maintained by Matt Trout <mst@shadowcatsystems.co.uk>
3268 For support, your best bet is to try the C<DBIx::Class> users mailing list.
3269 While not an official support venue, C<DBIx::Class> makes heavy use of
3270 C<SQL::Abstract>, and as such list members there are very familiar with
3271 how to create queries.
3275 This module is free software; you may copy this under the same
3276 terms as perl itself (either the GNU General Public License or
3277 the Artistic License)