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)
325 return ($self->_sqlcase(' returning ').$sql, @bind);
328 sub _expand_insert_value {
331 my $k = our $Cur_Col_Meta;
333 if (ref($v) eq 'ARRAY') {
334 if ($self->{array_datatypes}) {
335 return +{ -bind => [ $k, $v ] };
337 my ($sql, @bind) = @$v;
338 $self->_assert_bindval_matches_bindtype(@bind);
339 return +{ -literal => $v };
341 if (ref($v) eq 'HASH') {
342 if (grep !/^-/, keys %$v) {
343 belch "HASH ref as bind value in insert is not supported";
344 return +{ -bind => [ $k, $v ] };
348 return +{ -bind => [ $k, undef ] };
350 return $self->expand_expr($v);
355 #======================================================================
357 #======================================================================
362 my $table = $self->_table(shift);
363 my $data = shift || return;
367 # first build the 'SET' part of the sql statement
368 puke "Unsupported data type specified to \$sql->update"
369 unless ref $data eq 'HASH';
371 my ($sql, @all_bind) = $self->_update_set_values($data);
372 $sql = $self->_sqlcase('update ') . $table . $self->_sqlcase(' set ')
376 my($where_sql, @where_bind) = $self->where($where);
378 push @all_bind, @where_bind;
381 if ($options->{returning}) {
382 my ($returning_sql, @returning_bind) = $self->_update_returning($options);
383 $sql .= $returning_sql;
384 push @all_bind, @returning_bind;
387 return wantarray ? ($sql, @all_bind) : $sql;
390 sub _update_set_values {
391 my ($self, $data) = @_;
393 return @{ $self->render_aqt(
394 $self->_expand_update_set_values(undef, $data),
398 sub _expand_update_set_values {
399 my ($self, undef, $data) = @_;
400 $self->_expand_maybe_list_expr( [
403 $set = { -bind => $_ } unless defined $set;
404 +{ -op => [ '=', $self->_expand_ident(-ident => $k), $set ] };
410 ? ($self->{array_datatypes}
411 ? [ $k, +{ -bind => [ $k, $v ] } ]
412 : [ $k, +{ -literal => $v } ])
414 local our $Cur_Col_Meta = $k;
415 [ $k, $self->_expand_expr($v) ]
422 # So that subclasses can override UPDATE ... RETURNING separately from
424 sub _update_returning { shift->_returning(@_) }
428 #======================================================================
430 #======================================================================
435 my $table = $self->_table(shift);
436 my $fields = shift || '*';
440 my ($fields_sql, @bind) = $self->_select_fields($fields);
442 my ($where_sql, @where_bind) = $self->where($where, $order);
443 push @bind, @where_bind;
445 my $sql = join(' ', $self->_sqlcase('select'), $fields_sql,
446 $self->_sqlcase('from'), $table)
449 return wantarray ? ($sql, @bind) : $sql;
453 my ($self, $fields) = @_;
454 return $fields unless ref($fields);
455 return @{ $self->render_aqt(
456 $self->_expand_maybe_list_expr($fields, '-ident')
460 #======================================================================
462 #======================================================================
467 my $table = $self->_table(shift);
471 my($where_sql, @bind) = $self->where($where);
472 my $sql = $self->_sqlcase('delete from ') . $table . $where_sql;
474 if ($options->{returning}) {
475 my ($returning_sql, @returning_bind) = $self->_delete_returning($options);
476 $sql .= $returning_sql;
477 push @bind, @returning_bind;
480 return wantarray ? ($sql, @bind) : $sql;
483 # So that subclasses can override DELETE ... RETURNING separately from
485 sub _delete_returning { shift->_returning(@_) }
489 #======================================================================
491 #======================================================================
495 # Finally, a separate routine just to handle WHERE clauses
497 my ($self, $where, $order) = @_;
499 local $self->{convert_where} = $self->{convert};
502 my ($sql, @bind) = defined($where)
503 ? $self->_recurse_where($where)
505 $sql = (defined $sql and length $sql) ? $self->_sqlcase(' where ') . "( $sql )" : '';
509 my ($order_sql, @order_bind) = $self->_order_by($order);
511 push @bind, @order_bind;
514 return wantarray ? ($sql, @bind) : $sql;
517 { our $Default_Scalar_To = -value }
520 my ($self, $expr, $default_scalar_to) = @_;
521 local our $Default_Scalar_To = $default_scalar_to if $default_scalar_to;
522 $self->_expand_expr($expr);
526 my ($self, $aqt) = @_;
527 my ($k, $v, @rest) = %$aqt;
529 die "Not a node type: $k" unless $k =~ s/^-//;
530 if (my $meth = $self->{render}{$k}) {
531 return [ $self->$meth($k, $v) ];
533 die "notreached: $k";
537 my ($self, $expr, $default_scalar_to) = @_;
538 return @{ $self->render_aqt(
539 $self->expand_expr($expr, $default_scalar_to)
544 my ($self, $raw) = @_;
545 s/^-(?=.)//, s/\s+/_/g for my $op = lc $raw;
550 my ($self, $expr) = @_;
551 our $Expand_Depth ||= 0; local $Expand_Depth = $Expand_Depth + 1;
552 return undef unless defined($expr);
553 if (ref($expr) eq 'HASH') {
554 return undef unless my $kc = keys %$expr;
556 return $self->_expand_op_andor(and => $expr);
558 my ($key, $value) = %$expr;
559 if ($key =~ /^-/ and $key =~ s/ [_\s]? \d+ $//x ) {
560 belch 'Use of [and|or|nest]_N modifiers is deprecated and will be removed in SQLA v2.0. '
561 . "You probably wanted ...-and => [ $key => COND1, $key => COND2 ... ]";
563 return $self->_expand_hashpair($key, $value);
565 if (ref($expr) eq 'ARRAY') {
566 return $self->_expand_op_andor(lc($self->{logic}), $expr);
568 if (my $literal = is_literal_value($expr)) {
569 return +{ -literal => $literal };
571 if (!ref($expr) or Scalar::Util::blessed($expr)) {
572 return $self->_expand_scalar($expr);
577 sub _expand_hashpair {
578 my ($self, $k, $v) = @_;
579 unless (defined($k) and length($k)) {
580 if (defined($k) and my $literal = is_literal_value($v)) {
581 belch 'Hash-pairs consisting of an empty string with a literal are deprecated, and will be removed in 2.0: use -and => [ $literal ] instead';
582 return { -literal => $literal };
584 puke "Supplying an empty left hand side argument is not supported";
587 return $self->_expand_hashpair_op($k, $v);
588 } elsif ($k =~ /^[^\w]/i) {
589 my ($lhs, @rhs) = @$v;
590 return $self->_expand_op(
591 -op, [ $k, $self->expand_expr($lhs, -ident), @rhs ]
594 return $self->_expand_hashpair_ident($k, $v);
597 sub _expand_hashpair_ident {
598 my ($self, $k, $v) = @_;
600 local our $Cur_Col_Meta = $k;
602 # hash with multiple or no elements is andor
604 if (ref($v) eq 'HASH' and keys %$v != 1) {
605 return $self->_expand_op_andor(and => $v, $k);
608 # undef needs to be re-sent with cmp to achieve IS/IS NOT NULL
610 if (is_undef_value($v)) {
611 return $self->_expand_hashpair_cmp($k => undef);
614 # scalars and objects get expanded as whatever requested or values
616 if (!ref($v) or Scalar::Util::blessed($v)) {
617 return $self->_expand_hashpair_scalar($k, $v);
620 # single key hashref is a hashtriple
622 if (ref($v) eq 'HASH') {
623 return $self->_expand_hashtriple($k, %$v);
626 # arrayref needs re-engineering over the elements
628 if (ref($v) eq 'ARRAY') {
629 return $self->sqlfalse unless @$v;
630 $self->_debug("ARRAY($k) means distribute over elements");
632 $v->[0] =~ /^-(and|or)$/i
633 ? (shift(@{$v = [ @$v ]}), $1)
634 : lc($self->{logic} || 'OR')
636 return $self->_expand_op_andor(
641 if (my $literal = is_literal_value($v)) {
643 belch 'Hash-pairs consisting of an empty string with a literal are deprecated, and will be removed in 2.0: use -and => [ $literal ] instead';
646 my ($sql, @bind) = @$literal;
647 if ($self->{bindtype} eq 'columns') {
649 $self->_assert_bindval_matches_bindtype($_);
652 return +{ -literal => [ $self->_quote($k).' '.$sql, @bind ] };
658 my ($self, $expr) = @_;
660 return $self->_expand_expr({ (our $Default_Scalar_To) => $expr });
663 sub _expand_hashpair_scalar {
664 my ($self, $k, $v) = @_;
666 return $self->_expand_hashpair_cmp(
667 $k, $self->_expand_scalar($v),
671 sub _expand_hashpair_op {
672 my ($self, $k, $v) = @_;
674 $self->_assert_pass_injection_guard($k =~ /\A-(.*)\Z/s);
676 my $op = $self->_normalize_op($k);
678 if (my $exp = $self->{expand}{$op}) {
679 return $self->$exp($op, $v);
682 # Ops prefixed with -not_ get converted
684 if (my ($rest) = $op =~/^not_(.*)$/) {
687 $self->_expand_expr({ "-${rest}", $v })
693 my $op = join(' ', split '_', $op);
695 # the old special op system requires illegality for top-level use
698 (our $Expand_Depth) == 1
700 List::Util::first { $op =~ $_->{regex} } @{$self->{special_ops}}
702 $self->{disable_old_special_ops}
703 and List::Util::first { $op =~ $_->{regex} } @BUILTIN_SPECIAL_OPS
707 puke "Illegal use of top-level '-$op'"
710 # the old unary op system means we should touch nothing and let it work
712 if (my $us = List::Util::first { $op =~ $_->{regex} } @{$self->{unary_ops}}) {
713 return { -op => [ $op, $v ] };
717 # an explicit node type is currently assumed to be expanded (this is almost
718 # certainly wrong and there should be expansion anyway)
720 if ($self->{render}{$op}) {
724 my $type = $self->{unknown_unop_always_func} ? -func : -op;
731 and (keys %$v)[0] =~ /^-/
734 (List::Util::first { $op =~ $_->{regex} } @{$self->{special_ops}})
743 ($type eq -func and ref($v) eq 'ARRAY')
744 ? map $self->_expand_expr($_), @$v
745 : $self->_expand_expr($v)
749 sub _expand_hashpair_cmp {
750 my ($self, $k, $v) = @_;
751 $self->_expand_hashtriple($k, $self->{cmp}, $v);
754 sub _expand_hashtriple {
755 my ($self, $k, $vk, $vv) = @_;
757 my $ik = $self->_expand_ident(-ident => $k);
759 my $op = $self->_normalize_op($vk);
760 $self->_assert_pass_injection_guard($op);
762 if ($op =~ s/ _? \d+ $//x ) {
763 return $self->_expand_expr($k, { $vk, $vv });
765 if (my $x = $self->{expand_op}{$op}) {
766 local our $Cur_Col_Meta = $k;
767 return $self->$x($op, $vv, $k);
771 my $op = join(' ', split '_', $op);
773 if (my $us = List::Util::first { $op =~ $_->{regex} } @{$self->{special_ops}}) {
774 return { -op => [ $op, $ik, $vv ] };
776 if (my $us = List::Util::first { $op =~ $_->{regex} } @{$self->{unary_ops}}) {
780 { -op => [ $op, $vv ] }
784 if (ref($vv) eq 'ARRAY') {
786 my $logic = (defined($raw[0]) and $raw[0] =~ /^-(and|or)$/i)
787 ? (shift(@raw), $1) : 'or';
788 my @values = map +{ $vk => $_ }, @raw;
790 $op =~ $self->{inequality_op}
791 or $op =~ $self->{not_like_op}
793 if (lc($logic) eq 'or' and @values > 1) {
794 belch "A multi-element arrayref as an argument to the inequality op '${\uc(join ' ', split '_', $op)}' "
795 . 'is technically equivalent to an always-true 1=1 (you probably wanted '
796 . "to say ...{ \$inequality_op => [ -and => \@values ] }... instead)"
801 # try to DWIM on equality operators
802 return ($self->_dwim_op_to_is($op,
803 "Supplying an empty arrayref to '%s' is deprecated",
804 "operator '%s' applied on an empty array (field '$k')"
805 ) ? $self->sqlfalse : $self->sqltrue);
807 return $self->_expand_op_andor($logic => \@values, $k);
809 if (is_undef_value($vv)) {
810 my $is = ($self->_dwim_op_to_is($op,
811 "Supplying an undefined argument to '%s' is deprecated",
812 "unexpected operator '%s' with undef operand",
813 ) ? 'is' : 'is not');
815 return $self->_expand_hashpair($k => { $is, undef });
817 local our $Cur_Col_Meta = $k;
821 $self->_expand_expr($vv)
826 my ($self, $raw, $empty, $fail) = @_;
828 my $op = $self->_normalize_op($raw);
830 if ($op =~ /^not$/i) {
833 if ($op =~ $self->{equality_op}) {
836 if ($op =~ $self->{like_op}) {
837 belch(sprintf $empty, uc(join ' ', split '_', $op));
840 if ($op =~ $self->{inequality_op}) {
843 if ($op =~ $self->{not_like_op}) {
844 belch(sprintf $empty, uc(join ' ', split '_', $op));
847 puke(sprintf $fail, $op);
851 my ($self, undef, $args) = @_;
852 my ($func, @args) = @$args;
853 return { -func => [ $func, map $self->expand_expr($_), @args ] };
857 my ($self, undef, $body, $k) = @_;
858 return $self->_expand_hashpair_cmp(
859 $k, { -ident => $body }
861 unless (defined($body) or (ref($body) and ref($body) eq 'ARRAY')) {
862 puke "-ident requires a single plain scalar argument (a quotable identifier) or an arrayref of identifier parts";
864 my @parts = map split(/\Q${\($self->{name_sep}||'.')}\E/, $_),
865 ref($body) ? @$body : $body;
866 return { -ident => $parts[-1] } if $self->{_dequalify_idents};
867 unless ($self->{quote_char}) {
868 $self->_assert_pass_injection_guard($_) for @parts;
870 return +{ -ident => \@parts };
874 return $_[0]->_expand_hashpair_cmp(
875 $_[3], { -value => $_[2] },
877 +{ -bind => [ our $Cur_Col_Meta, $_[2] ] };
881 +{ -op => [ 'not', $_[0]->_expand_expr($_[2]) ] };
885 my ($self, undef, $args) = @_;
886 +{ -row => [ map $self->expand_expr($_), @$args ] };
890 my ($self, undef, $args) = @_;
891 my ($op, @opargs) = @$args;
892 if (my $exp = $self->{expand_op}{$op}) {
893 return $self->$exp($op, \@opargs);
895 +{ -op => [ $op, map $self->expand_expr($_), @opargs ] };
899 my ($self, undef, $v) = @_;
901 return $self->_expand_expr($v);
903 puke "-bool => undef not supported" unless defined($v);
904 return $self->_expand_ident(-ident => $v);
907 sub _expand_op_andor {
908 my ($self, $logop, $v, $k) = @_;
910 $v = [ map +{ $k, $_ },
912 ? (map +{ $_ => $v->{$_} }, sort keys %$v)
916 if (ref($v) eq 'HASH') {
917 return undef unless keys %$v;
920 map $self->_expand_expr({ $_ => $v->{$_} }),
924 if (ref($v) eq 'ARRAY') {
925 $logop eq 'and' or $logop eq 'or' or puke "unknown logic: $logop";
928 (ref($_) eq 'ARRAY' and @$_)
929 or (ref($_) eq 'HASH' and %$_)
935 while (my ($el) = splice @expr, 0, 1) {
936 puke "Supplying an empty left hand side argument is not supported in array-pairs"
937 unless defined($el) and length($el);
938 my $elref = ref($el);
940 local our $Expand_Depth = 0;
941 push(@res, grep defined, $self->_expand_expr({ $el, shift(@expr) }));
942 } elsif ($elref eq 'ARRAY') {
943 push(@res, grep defined, $self->_expand_expr($el)) if @$el;
944 } elsif (my $l = is_literal_value($el)) {
945 push @res, { -literal => $l };
946 } elsif ($elref eq 'HASH') {
947 local our $Expand_Depth = 0;
948 push @res, grep defined, $self->_expand_expr($el) if %$el;
954 # return $res[0] if @res == 1;
955 return { -op => [ $logop, @res ] };
961 my ($self, $op, $vv, $k) = @_;
962 ($k, $vv) = @$vv unless defined $k;
963 puke "$op can only take undef as argument"
967 and exists($vv->{-value})
968 and !defined($vv->{-value})
970 return +{ -op => [ $op.'_null', $self->expand_expr($k, -ident) ] };
973 sub _expand_between {
974 my ($self, $op, $vv, $k) = @_;
975 $k = shift @{$vv = [ @$vv ]} unless defined $k;
976 my @rhs = map $self->_expand_expr($_),
977 ref($vv) eq 'ARRAY' ? @$vv : $vv;
979 (@rhs == 1 and ref($rhs[0]) eq 'HASH' and $rhs[0]->{-literal})
981 (@rhs == 2 and defined($rhs[0]) and defined($rhs[1]))
983 puke "Operator '${\uc($op)}' requires either an arrayref with two defined values or expressions, or a single literal scalarref/arrayref-ref";
987 $self->expand_expr(ref($k) ? $k : { -ident => $k }),
993 my ($self, $op, $vv, $k) = @_;
994 $k = shift @{$vv = [ @$vv ]} unless defined $k;
995 if (my $literal = is_literal_value($vv)) {
996 my ($sql, @bind) = @$literal;
997 my $opened_sql = $self->_open_outer_paren($sql);
999 $op, $self->expand_expr($k, -ident),
1000 { -literal => [ $opened_sql, @bind ] }
1004 'SQL::Abstract before v1.75 used to generate incorrect SQL when the '
1005 . "-${\uc($op)} operator was given an undef-containing list: !!!AUDIT YOUR CODE "
1006 . 'AND DATA!!! (the upcoming Data::Query-based version of SQL::Abstract '
1007 . 'will emit the logically correct SQL instead of raising this exception)'
1009 puke("Argument passed to the '${\uc($op)}' operator can not be undefined")
1011 my @rhs = map $self->expand_expr($_, -value),
1012 map { defined($_) ? $_: puke($undef_err) }
1013 (ref($vv) eq 'ARRAY' ? @$vv : $vv);
1014 return $self->${\($op =~ /^not/ ? 'sqltrue' : 'sqlfalse')} unless @rhs;
1018 $self->expand_expr($k, -ident),
1024 my ($self, undef, $v) = @_;
1025 # DBIx::Class requires a nest warning to be emitted once but the private
1026 # method it overrode to do so no longer exists
1027 if ($self->{warn_once_on_nest}) {
1028 unless (our $Nest_Warned) {
1030 "-nest in search conditions is deprecated, you most probably wanted:\n"
1031 .q|{..., -and => [ \%cond0, \@cond1, \'cond2', \[ 'cond3', [ col => bind ] ], etc. ], ... }|
1036 return $self->_expand_expr($v);
1040 my ($self, undef, $bind) = @_;
1041 return { -bind => $bind };
1044 sub _recurse_where {
1045 my ($self, $where, $logic) = @_;
1047 # Special case: top level simple string treated as literal
1049 my $where_exp = (ref($where)
1050 ? $self->_expand_expr($where, $logic)
1051 : { -literal => [ $where ] });
1053 # dispatch expanded expression
1055 my ($sql, @bind) = defined($where_exp) ? @{ $self->render_aqt($where_exp) } : (undef);
1056 # DBIx::Class used to call _recurse_where in scalar context
1057 # something else might too...
1059 return ($sql, @bind);
1062 belch "Calling _recurse_where in scalar context is deprecated and will go away before 2.0";
1068 my ($self, undef, $ident) = @_;
1070 return $self->_convert($self->_quote($ident));
1074 my ($self, undef, $values) = @_;
1075 my ($sql, @bind) = $self->_render_op(undef, [ ',', @$values ]);
1076 return $self->join_query_parts('', [ '(' ], [ $sql, @bind ], [ ')' ]);
1080 my ($self, undef, $rest) = @_;
1081 my ($func, @args) = @$rest;
1082 return $self->join_query_parts('',
1083 [ $self->_sqlcase($func) ],
1085 [ $self->join_query_parts(', ', @args) ],
1091 my ($self, undef, $bind) = @_;
1092 return ($self->_convert('?'), $self->_bindtype(@$bind));
1095 sub _render_literal {
1096 my ($self, undef, $literal) = @_;
1097 $self->_assert_bindval_matches_bindtype(@{$literal}[1..$#$literal]);
1102 my ($self, undef, $v) = @_;
1103 my ($op, @args) = @$v;
1104 if (my $r = $self->{render_op}{$op}) {
1105 return $self->$r($op, \@args);
1110 my $op = join(' ', split '_', $op);
1112 my $ss = List::Util::first { $op =~ $_->{regex} } @{$self->{special_ops}};
1113 if ($ss and @args > 1) {
1114 puke "Special op '${op}' requires first value to be identifier"
1115 unless my ($ident) = map $_->{-ident}, grep ref($_) eq 'HASH', $args[0];
1116 my $k = join(($self->{name_sep}||'.'), @$ident);
1117 local our $Expand_Depth = 1;
1118 return $self->${\($ss->{handler})}($k, $op, $args[1]);
1120 if (my $us = List::Util::first { $op =~ $_->{regex} } @{$self->{unary_ops}}) {
1121 return $self->${\($us->{handler})}($op, $args[0]);
1124 return $self->_render_unop_paren($op, \@args);
1128 return $self->_render_unop_prefix($op, \@args);
1130 return $self->_render_op_multop($op, \@args);
1136 sub _render_op_between {
1137 my ($self, $op, $args) = @_;
1138 my ($left, $low, $high) = @$args;
1141 puke "Single arg to between must be a literal"
1142 unless $low->{-literal};
1145 +($low, $self->format_keyword('and'), $high);
1148 return $self->join_query_parts(' ',
1149 '(', $left, $self->format_keyword($op), @rh, ')',
1154 my ($self, $op, $args) = @_;
1155 my ($lhs, @rhs) = @$args;
1157 return $self->join_query_parts(' ',
1159 $self->format_keyword($op),
1161 [ $self->join_query_parts(', ', @rhs) ],
1166 sub _render_op_andor {
1167 my ($self, $op, $args) = @_;
1168 return '' unless @$args;
1169 return $self->join_query_parts('', $args->[0]) if @$args == 1;
1170 return $self->join_query_parts(
1171 ' ' => '(',[ $self->_render_op_multop($op, $args) ], ')'
1175 sub _render_op_multop {
1176 my ($self, $op, $args) = @_;
1178 return '' unless @parts;
1179 return @{$self->render_aqt($parts[0])} if @parts == 1;
1180 my $join = ($op eq ','
1182 : ' '.$self->format_keyword($op).' '
1184 return $self->join_query_parts($join, @parts);
1187 sub join_query_parts {
1188 my ($self, $join, @parts) = @_;
1191 ? $self->render_aqt($_)
1192 : ref($_) eq 'ARRAY' ? $_ : [ $_ ]),
1195 join($join, map $_->[0], @final),
1196 (map @{$_}[1..$#$_], @final),
1200 sub _render_unop_paren {
1201 my ($self, $op, $v) = @_;
1202 return $self->join_query_parts('',
1203 '(', [ $self->_render_unop_prefix($op, $v) ], ')'
1207 sub _render_unop_prefix {
1208 my ($self, $op, $v) = @_;
1209 return $self->join_query_parts(' ',
1210 $self->_sqlcase($op), $v->[0]
1214 sub _render_unop_postfix {
1215 my ($self, $op, $v) = @_;
1216 return $self->join_query_parts(' ',
1217 $v->[0], $self->format_keyword($op),
1221 # Some databases (SQLite) treat col IN (1, 2) different from
1222 # col IN ( (1, 2) ). Use this to strip all outer parens while
1223 # adding them back in the corresponding method
1224 sub _open_outer_paren {
1225 my ($self, $sql) = @_;
1227 while (my ($inner) = $sql =~ /^ \s* \( (.*) \) \s* $/xs) {
1229 # there are closing parens inside, need the heavy duty machinery
1230 # to reevaluate the extraction starting from $sql (full reevaluation)
1231 if ($inner =~ /\)/) {
1232 require Text::Balanced;
1234 my (undef, $remainder) = do {
1235 # idiotic design - writes to $@ but *DOES NOT* throw exceptions
1237 Text::Balanced::extract_bracketed($sql, '()', qr/\s*/);
1240 # the entire expression needs to be a balanced bracketed thing
1241 # (after an extract no remainder sans trailing space)
1242 last if defined $remainder and $remainder =~ /\S/;
1252 #======================================================================
1254 #======================================================================
1256 sub _expand_order_by {
1257 my ($self, $arg) = @_;
1259 return unless defined($arg) and not (ref($arg) eq 'ARRAY' and !@$arg);
1261 return $self->_expand_maybe_list_expr($arg)
1262 if ref($arg) eq 'HASH' and ($arg->{-op}||[''])->[0] eq ',';
1264 my $expander = sub {
1265 my ($self, $dir, $expr) = @_;
1266 my @to_expand = ref($expr) eq 'ARRAY' ? @$expr : $expr;
1267 foreach my $arg (@to_expand) {
1271 and grep /^-(asc|desc)$/, keys %$arg
1273 puke "ordering direction hash passed to order by must have exactly one key (-asc or -desc)";
1277 defined($dir) ? { -op => [ $dir =~ /^-?(.*)$/ ,=> $_ ] } : $_
1279 map $self->expand_expr($_, -ident),
1280 map ref($_) eq 'ARRAY' ? @$_ : $_, @to_expand;
1281 return undef unless @exp;
1282 return undef if @exp == 1 and not defined($exp[0]);
1283 return +{ -op => [ ',', @exp ] };
1286 local @{$self->{expand}}{qw(asc desc)} = (($expander) x 2);
1288 return $self->$expander(undef, $arg);
1292 my ($self, $arg) = @_;
1294 return '' unless defined(my $expanded = $self->_expand_order_by($arg));
1296 my ($sql, @bind) = @{ $self->render_aqt($expanded) };
1298 return '' unless length($sql);
1300 my $final_sql = $self->_sqlcase(' order by ').$sql;
1302 return ($final_sql, @bind);
1305 # _order_by no longer needs to call this so doesn't but DBIC uses it.
1307 sub _order_by_chunks {
1308 my ($self, $arg) = @_;
1310 return () unless defined(my $expanded = $self->_expand_order_by($arg));
1312 return $self->_chunkify_order_by($expanded);
1315 sub _chunkify_order_by {
1316 my ($self, $expanded) = @_;
1318 return grep length, @{ $self->render_aqt($expanded) }
1319 if $expanded->{-ident} or @{$expanded->{-literal}||[]} == 1;
1322 if (ref() eq 'HASH' and $_->{-op} and $_->{-op}[0] eq ',') {
1323 my ($comma, @list) = @{$_->{-op}};
1324 return map $self->_chunkify_order_by($_), @list;
1326 return $self->render_aqt($_);
1330 #======================================================================
1331 # DATASOURCE (FOR NOW, JUST PLAIN TABLE OR LIST OF TABLES)
1332 #======================================================================
1338 $self->_expand_maybe_list_expr($from, -ident)
1343 #======================================================================
1345 #======================================================================
1347 sub _expand_maybe_list_expr {
1348 my ($self, $expr, $default) = @_;
1350 ',', map $self->expand_expr($_, $default),
1351 @{$expr->{-op}}[1..$#{$expr->{-op}}]
1352 ] } if ref($expr) eq 'HASH' and ($expr->{-op}||[''])->[0] eq ',';
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]);
1461 sub format_keyword { $_[0]->_sqlcase(join ' ', split '_', $_[1]) }
1463 #======================================================================
1464 # DISPATCHING FROM REFKIND
1465 #======================================================================
1468 my ($self, $data) = @_;
1470 return 'UNDEF' unless defined $data;
1472 # blessed objects are treated like scalars
1473 my $ref = (Scalar::Util::blessed $data) ? '' : ref $data;
1475 return 'SCALAR' unless $ref;
1478 while ($ref eq 'REF') {
1480 $ref = (Scalar::Util::blessed $data) ? '' : ref $data;
1484 return ($ref||'SCALAR') . ('REF' x $n_steps);
1488 my ($self, $data) = @_;
1489 my @try = ($self->_refkind($data));
1490 push @try, 'SCALAR_or_UNDEF' if $try[0] eq 'SCALAR' || $try[0] eq 'UNDEF';
1491 push @try, 'FALLBACK';
1495 sub _METHOD_FOR_refkind {
1496 my ($self, $meth_prefix, $data) = @_;
1499 for (@{$self->_try_refkind($data)}) {
1500 $method = $self->can($meth_prefix."_".$_)
1504 return $method || puke "cannot dispatch on '$meth_prefix' for ".$self->_refkind($data);
1508 sub _SWITCH_refkind {
1509 my ($self, $data, $dispatch_table) = @_;
1512 for (@{$self->_try_refkind($data)}) {
1513 $coderef = $dispatch_table->{$_}
1517 puke "no dispatch entry for ".$self->_refkind($data)
1526 #======================================================================
1527 # VALUES, GENERATE, AUTOLOAD
1528 #======================================================================
1530 # LDNOTE: original code from nwiger, didn't touch code in that section
1531 # I feel the AUTOLOAD stuff should not be the default, it should
1532 # only be activated on explicit demand by user.
1536 my $data = shift || return;
1537 puke "Argument to ", __PACKAGE__, "->values must be a \\%hash"
1538 unless ref $data eq 'HASH';
1541 foreach my $k (sort keys %$data) {
1542 my $v = $data->{$k};
1543 $self->_SWITCH_refkind($v, {
1545 if ($self->{array_datatypes}) { # array datatype
1546 push @all_bind, $self->_bindtype($k, $v);
1548 else { # literal SQL with bind
1549 my ($sql, @bind) = @$v;
1550 $self->_assert_bindval_matches_bindtype(@bind);
1551 push @all_bind, @bind;
1554 ARRAYREFREF => sub { # literal SQL with bind
1555 my ($sql, @bind) = @${$v};
1556 $self->_assert_bindval_matches_bindtype(@bind);
1557 push @all_bind, @bind;
1559 SCALARREF => sub { # literal SQL without bind
1561 SCALAR_or_UNDEF => sub {
1562 push @all_bind, $self->_bindtype($k, $v);
1573 my(@sql, @sqlq, @sqlv);
1577 if ($ref eq 'HASH') {
1578 for my $k (sort keys %$_) {
1581 my $label = $self->_quote($k);
1582 if ($r eq 'ARRAY') {
1583 # literal SQL with bind
1584 my ($sql, @bind) = @$v;
1585 $self->_assert_bindval_matches_bindtype(@bind);
1586 push @sqlq, "$label = $sql";
1588 } elsif ($r eq 'SCALAR') {
1589 # literal SQL without bind
1590 push @sqlq, "$label = $$v";
1592 push @sqlq, "$label = ?";
1593 push @sqlv, $self->_bindtype($k, $v);
1596 push @sql, $self->_sqlcase('set'), join ', ', @sqlq;
1597 } elsif ($ref eq 'ARRAY') {
1598 # unlike insert(), assume these are ONLY the column names, i.e. for SQL
1601 if ($r eq 'ARRAY') { # literal SQL with bind
1602 my ($sql, @bind) = @$v;
1603 $self->_assert_bindval_matches_bindtype(@bind);
1606 } elsif ($r eq 'SCALAR') { # literal SQL without bind
1607 # embedded literal SQL
1614 push @sql, '(' . join(', ', @sqlq) . ')';
1615 } elsif ($ref eq 'SCALAR') {
1619 # strings get case twiddled
1620 push @sql, $self->_sqlcase($_);
1624 my $sql = join ' ', @sql;
1626 # this is pretty tricky
1627 # if ask for an array, return ($stmt, @bind)
1628 # otherwise, s/?/shift @sqlv/ to put it inline
1630 return ($sql, @sqlv);
1632 1 while $sql =~ s/\?/my $d = shift(@sqlv);
1633 ref $d ? $d->[1] : $d/e;
1642 # This allows us to check for a local, then _form, attr
1644 my($name) = $AUTOLOAD =~ /.*::(.+)/;
1645 puke "AUTOLOAD invoked for method name ${name} and allow_autoload option not set" unless $self->{allow_autoload};
1646 return $self->generate($name, @_);
1657 SQL::Abstract - Generate SQL from Perl data structures
1663 my $sql = SQL::Abstract->new;
1665 my($stmt, @bind) = $sql->select($source, \@fields, \%where, $order);
1667 my($stmt, @bind) = $sql->insert($table, \%fieldvals || \@values);
1669 my($stmt, @bind) = $sql->update($table, \%fieldvals, \%where);
1671 my($stmt, @bind) = $sql->delete($table, \%where);
1673 # Then, use these in your DBI statements
1674 my $sth = $dbh->prepare($stmt);
1675 $sth->execute(@bind);
1677 # Just generate the WHERE clause
1678 my($stmt, @bind) = $sql->where(\%where, $order);
1680 # Return values in the same order, for hashed queries
1681 # See PERFORMANCE section for more details
1682 my @bind = $sql->values(\%fieldvals);
1686 This module was inspired by the excellent L<DBIx::Abstract>.
1687 However, in using that module I found that what I really wanted
1688 to do was generate SQL, but still retain complete control over my
1689 statement handles and use the DBI interface. So, I set out to
1690 create an abstract SQL generation module.
1692 While based on the concepts used by L<DBIx::Abstract>, there are
1693 several important differences, especially when it comes to WHERE
1694 clauses. I have modified the concepts used to make the SQL easier
1695 to generate from Perl data structures and, IMO, more intuitive.
1696 The underlying idea is for this module to do what you mean, based
1697 on the data structures you provide it. The big advantage is that
1698 you don't have to modify your code every time your data changes,
1699 as this module figures it out.
1701 To begin with, an SQL INSERT is as easy as just specifying a hash
1702 of C<key=value> pairs:
1705 name => 'Jimbo Bobson',
1706 phone => '123-456-7890',
1707 address => '42 Sister Lane',
1708 city => 'St. Louis',
1709 state => 'Louisiana',
1712 The SQL can then be generated with this:
1714 my($stmt, @bind) = $sql->insert('people', \%data);
1716 Which would give you something like this:
1718 $stmt = "INSERT INTO people
1719 (address, city, name, phone, state)
1720 VALUES (?, ?, ?, ?, ?)";
1721 @bind = ('42 Sister Lane', 'St. Louis', 'Jimbo Bobson',
1722 '123-456-7890', 'Louisiana');
1724 These are then used directly in your DBI code:
1726 my $sth = $dbh->prepare($stmt);
1727 $sth->execute(@bind);
1729 =head2 Inserting and Updating Arrays
1731 If your database has array types (like for example Postgres),
1732 activate the special option C<< array_datatypes => 1 >>
1733 when creating the C<SQL::Abstract> object.
1734 Then you may use an arrayref to insert and update database array types:
1736 my $sql = SQL::Abstract->new(array_datatypes => 1);
1738 planets => [qw/Mercury Venus Earth Mars/]
1741 my($stmt, @bind) = $sql->insert('solar_system', \%data);
1745 $stmt = "INSERT INTO solar_system (planets) VALUES (?)"
1747 @bind = (['Mercury', 'Venus', 'Earth', 'Mars']);
1750 =head2 Inserting and Updating SQL
1752 In order to apply SQL functions to elements of your C<%data> you may
1753 specify a reference to an arrayref for the given hash value. For example,
1754 if you need to execute the Oracle C<to_date> function on a value, you can
1755 say something like this:
1759 date_entered => \[ "to_date(?,'MM/DD/YYYY')", "03/02/2003" ],
1762 The first value in the array is the actual SQL. Any other values are
1763 optional and would be included in the bind values array. This gives
1766 my($stmt, @bind) = $sql->insert('people', \%data);
1768 $stmt = "INSERT INTO people (name, date_entered)
1769 VALUES (?, to_date(?,'MM/DD/YYYY'))";
1770 @bind = ('Bill', '03/02/2003');
1772 An UPDATE is just as easy, all you change is the name of the function:
1774 my($stmt, @bind) = $sql->update('people', \%data);
1776 Notice that your C<%data> isn't touched; the module will generate
1777 the appropriately quirky SQL for you automatically. Usually you'll
1778 want to specify a WHERE clause for your UPDATE, though, which is
1779 where handling C<%where> hashes comes in handy...
1781 =head2 Complex where statements
1783 This module can generate pretty complicated WHERE statements
1784 easily. For example, simple C<key=value> pairs are taken to mean
1785 equality, and if you want to see if a field is within a set
1786 of values, you can use an arrayref. Let's say we wanted to
1787 SELECT some data based on this criteria:
1790 requestor => 'inna',
1791 worker => ['nwiger', 'rcwe', 'sfz'],
1792 status => { '!=', 'completed' }
1795 my($stmt, @bind) = $sql->select('tickets', '*', \%where);
1797 The above would give you something like this:
1799 $stmt = "SELECT * FROM tickets WHERE
1800 ( requestor = ? ) AND ( status != ? )
1801 AND ( worker = ? OR worker = ? OR worker = ? )";
1802 @bind = ('inna', 'completed', 'nwiger', 'rcwe', 'sfz');
1804 Which you could then use in DBI code like so:
1806 my $sth = $dbh->prepare($stmt);
1807 $sth->execute(@bind);
1813 The methods are simple. There's one for every major SQL operation,
1814 and a constructor you use first. The arguments are specified in a
1815 similar order for each method (table, then fields, then a where
1816 clause) to try and simplify things.
1818 =head2 new(option => 'value')
1820 The C<new()> function takes a list of options and values, and returns
1821 a new B<SQL::Abstract> object which can then be used to generate SQL
1822 through the methods below. The options accepted are:
1828 If set to 'lower', then SQL will be generated in all lowercase. By
1829 default SQL is generated in "textbook" case meaning something like:
1831 SELECT a_field FROM a_table WHERE some_field LIKE '%someval%'
1833 Any setting other than 'lower' is ignored.
1837 This determines what the default comparison operator is. By default
1838 it is C<=>, meaning that a hash like this:
1840 %where = (name => 'nwiger', email => 'nate@wiger.org');
1842 Will generate SQL like this:
1844 WHERE name = 'nwiger' AND email = 'nate@wiger.org'
1846 However, you may want loose comparisons by default, so if you set
1847 C<cmp> to C<like> you would get SQL such as:
1849 WHERE name like 'nwiger' AND email like 'nate@wiger.org'
1851 You can also override the comparison on an individual basis - see
1852 the huge section on L</"WHERE CLAUSES"> at the bottom.
1854 =item sqltrue, sqlfalse
1856 Expressions for inserting boolean values within SQL statements.
1857 By default these are C<1=1> and C<1=0>. They are used
1858 by the special operators C<-in> and C<-not_in> for generating
1859 correct SQL even when the argument is an empty array (see below).
1863 This determines the default logical operator for multiple WHERE
1864 statements in arrays or hashes. If absent, the default logic is "or"
1865 for arrays, and "and" for hashes. This means that a WHERE
1869 event_date => {'>=', '2/13/99'},
1870 event_date => {'<=', '4/24/03'},
1873 will generate SQL like this:
1875 WHERE event_date >= '2/13/99' OR event_date <= '4/24/03'
1877 This is probably not what you want given this query, though (look
1878 at the dates). To change the "OR" to an "AND", simply specify:
1880 my $sql = SQL::Abstract->new(logic => 'and');
1882 Which will change the above C<WHERE> to:
1884 WHERE event_date >= '2/13/99' AND event_date <= '4/24/03'
1886 The logic can also be changed locally by inserting
1887 a modifier in front of an arrayref:
1889 @where = (-and => [event_date => {'>=', '2/13/99'},
1890 event_date => {'<=', '4/24/03'} ]);
1892 See the L</"WHERE CLAUSES"> section for explanations.
1896 This will automatically convert comparisons using the specified SQL
1897 function for both column and value. This is mostly used with an argument
1898 of C<upper> or C<lower>, so that the SQL will have the effect of
1899 case-insensitive "searches". For example, this:
1901 $sql = SQL::Abstract->new(convert => 'upper');
1902 %where = (keywords => 'MaKe iT CAse inSeNSItive');
1904 Will turn out the following SQL:
1906 WHERE upper(keywords) like upper('MaKe iT CAse inSeNSItive')
1908 The conversion can be C<upper()>, C<lower()>, or any other SQL function
1909 that can be applied symmetrically to fields (actually B<SQL::Abstract> does
1910 not validate this option; it will just pass through what you specify verbatim).
1914 This is a kludge because many databases suck. For example, you can't
1915 just bind values using DBI's C<execute()> for Oracle C<CLOB> or C<BLOB> fields.
1916 Instead, you have to use C<bind_param()>:
1918 $sth->bind_param(1, 'reg data');
1919 $sth->bind_param(2, $lots, {ora_type => ORA_CLOB});
1921 The problem is, B<SQL::Abstract> will normally just return a C<@bind> array,
1922 which loses track of which field each slot refers to. Fear not.
1924 If you specify C<bindtype> in new, you can determine how C<@bind> is returned.
1925 Currently, you can specify either C<normal> (default) or C<columns>. If you
1926 specify C<columns>, you will get an array that looks like this:
1928 my $sql = SQL::Abstract->new(bindtype => 'columns');
1929 my($stmt, @bind) = $sql->insert(...);
1932 [ 'column1', 'value1' ],
1933 [ 'column2', 'value2' ],
1934 [ 'column3', 'value3' ],
1937 You can then iterate through this manually, using DBI's C<bind_param()>.
1939 $sth->prepare($stmt);
1942 my($col, $data) = @$_;
1943 if ($col eq 'details' || $col eq 'comments') {
1944 $sth->bind_param($i, $data, {ora_type => ORA_CLOB});
1945 } elsif ($col eq 'image') {
1946 $sth->bind_param($i, $data, {ora_type => ORA_BLOB});
1948 $sth->bind_param($i, $data);
1952 $sth->execute; # execute without @bind now
1954 Now, why would you still use B<SQL::Abstract> if you have to do this crap?
1955 Basically, the advantage is still that you don't have to care which fields
1956 are or are not included. You could wrap that above C<for> loop in a simple
1957 sub called C<bind_fields()> or something and reuse it repeatedly. You still
1958 get a layer of abstraction over manual SQL specification.
1960 Note that if you set L</bindtype> to C<columns>, the C<\[ $sql, @bind ]>
1961 construct (see L</Literal SQL with placeholders and bind values (subqueries)>)
1962 will expect the bind values in this format.
1966 This is the character that a table or column name will be quoted
1967 with. By default this is an empty string, but you could set it to
1968 the character C<`>, to generate SQL like this:
1970 SELECT `a_field` FROM `a_table` WHERE `some_field` LIKE '%someval%'
1972 Alternatively, you can supply an array ref of two items, the first being the left
1973 hand quote character, and the second the right hand quote character. For
1974 example, you could supply C<['[',']']> for SQL Server 2000 compliant quotes
1975 that generates SQL like this:
1977 SELECT [a_field] FROM [a_table] WHERE [some_field] LIKE '%someval%'
1979 Quoting is useful if you have tables or columns names that are reserved
1980 words in your database's SQL dialect.
1984 This is the character that will be used to escape L</quote_char>s appearing
1985 in an identifier before it has been quoted.
1987 The parameter default in case of a single L</quote_char> character is the quote
1990 When opening-closing-style quoting is used (L</quote_char> is an arrayref)
1991 this parameter defaults to the B<closing (right)> L</quote_char>. Occurrences
1992 of the B<opening (left)> L</quote_char> within the identifier are currently left
1993 untouched. The default for opening-closing-style quotes may change in future
1994 versions, thus you are B<strongly encouraged> to specify the escape character
1999 This is the character that separates a table and column name. It is
2000 necessary to specify this when the C<quote_char> option is selected,
2001 so that tables and column names can be individually quoted like this:
2003 SELECT `table`.`one_field` FROM `table` WHERE `table`.`other_field` = 1
2005 =item injection_guard
2007 A regular expression C<qr/.../> that is applied to any C<-function> and unquoted
2008 column name specified in a query structure. This is a safety mechanism to avoid
2009 injection attacks when mishandling user input e.g.:
2011 my %condition_as_column_value_pairs = get_values_from_user();
2012 $sqla->select( ... , \%condition_as_column_value_pairs );
2014 If the expression matches an exception is thrown. Note that literal SQL
2015 supplied via C<\'...'> or C<\['...']> is B<not> checked in any way.
2017 Defaults to checking for C<;> and the C<GO> keyword (TransactSQL)
2019 =item array_datatypes
2021 When this option is true, arrayrefs in INSERT or UPDATE are
2022 interpreted as array datatypes and are passed directly
2024 When this option is false, arrayrefs are interpreted
2025 as literal SQL, just like refs to arrayrefs
2026 (but this behavior is for backwards compatibility; when writing
2027 new queries, use the "reference to arrayref" syntax
2033 Takes a reference to a list of "special operators"
2034 to extend the syntax understood by L<SQL::Abstract>.
2035 See section L</"SPECIAL OPERATORS"> for details.
2039 Takes a reference to a list of "unary operators"
2040 to extend the syntax understood by L<SQL::Abstract>.
2041 See section L</"UNARY OPERATORS"> for details.
2047 =head2 insert($table, \@values || \%fieldvals, \%options)
2049 This is the simplest function. You simply give it a table name
2050 and either an arrayref of values or hashref of field/value pairs.
2051 It returns an SQL INSERT statement and a list of bind values.
2052 See the sections on L</"Inserting and Updating Arrays"> and
2053 L</"Inserting and Updating SQL"> for information on how to insert
2054 with those data types.
2056 The optional C<\%options> hash reference may contain additional
2057 options to generate the insert SQL. Currently supported options
2064 Takes either a scalar of raw SQL fields, or an array reference of
2065 field names, and adds on an SQL C<RETURNING> statement at the end.
2066 This allows you to return data generated by the insert statement
2067 (such as row IDs) without performing another C<SELECT> statement.
2068 Note, however, this is not part of the SQL standard and may not
2069 be supported by all database engines.
2073 =head2 update($table, \%fieldvals, \%where, \%options)
2075 This takes a table, hashref of field/value pairs, and an optional
2076 hashref L<WHERE clause|/WHERE CLAUSES>. It returns an SQL UPDATE function and a list
2078 See the sections on L</"Inserting and Updating Arrays"> and
2079 L</"Inserting and Updating SQL"> for information on how to insert
2080 with those data types.
2082 The optional C<\%options> hash reference may contain additional
2083 options to generate the update SQL. Currently supported options
2090 See the C<returning> option to
2091 L<insert|/insert($table, \@values || \%fieldvals, \%options)>.
2095 =head2 select($source, $fields, $where, $order)
2097 This returns a SQL SELECT statement and associated list of bind values, as
2098 specified by the arguments:
2104 Specification of the 'FROM' part of the statement.
2105 The argument can be either a plain scalar (interpreted as a table
2106 name, will be quoted), or an arrayref (interpreted as a list
2107 of table names, joined by commas, quoted), or a scalarref
2108 (literal SQL, not quoted).
2112 Specification of the list of fields to retrieve from
2114 The argument can be either an arrayref (interpreted as a list
2115 of field names, will be joined by commas and quoted), or a
2116 plain scalar (literal SQL, not quoted).
2117 Please observe that this API is not as flexible as that of
2118 the first argument C<$source>, for backwards compatibility reasons.
2122 Optional argument to specify the WHERE part of the query.
2123 The argument is most often a hashref, but can also be
2124 an arrayref or plain scalar --
2125 see section L<WHERE clause|/"WHERE CLAUSES"> for details.
2129 Optional argument to specify the ORDER BY part of the query.
2130 The argument can be a scalar, a hashref or an arrayref
2131 -- see section L<ORDER BY clause|/"ORDER BY CLAUSES">
2137 =head2 delete($table, \%where, \%options)
2139 This takes a table name and optional hashref L<WHERE clause|/WHERE CLAUSES>.
2140 It returns an SQL DELETE statement and list of bind values.
2142 The optional C<\%options> hash reference may contain additional
2143 options to generate the delete SQL. Currently supported options
2150 See the C<returning> option to
2151 L<insert|/insert($table, \@values || \%fieldvals, \%options)>.
2155 =head2 where(\%where, $order)
2157 This is used to generate just the WHERE clause. For example,
2158 if you have an arbitrary data structure and know what the
2159 rest of your SQL is going to look like, but want an easy way
2160 to produce a WHERE clause, use this. It returns an SQL WHERE
2161 clause and list of bind values.
2164 =head2 values(\%data)
2166 This just returns the values from the hash C<%data>, in the same
2167 order that would be returned from any of the other above queries.
2168 Using this allows you to markedly speed up your queries if you
2169 are affecting lots of rows. See below under the L</"PERFORMANCE"> section.
2171 =head2 generate($any, 'number', $of, \@data, $struct, \%types)
2173 Warning: This is an experimental method and subject to change.
2175 This returns arbitrarily generated SQL. It's a really basic shortcut.
2176 It will return two different things, depending on return context:
2178 my($stmt, @bind) = $sql->generate('create table', \$table, \@fields);
2179 my $stmt_and_val = $sql->generate('create table', \$table, \@fields);
2181 These would return the following:
2183 # First calling form
2184 $stmt = "CREATE TABLE test (?, ?)";
2185 @bind = (field1, field2);
2187 # Second calling form
2188 $stmt_and_val = "CREATE TABLE test (field1, field2)";
2190 Depending on what you're trying to do, it's up to you to choose the correct
2191 format. In this example, the second form is what you would want.
2195 $sql->generate('alter session', { nls_date_format => 'MM/YY' });
2199 ALTER SESSION SET nls_date_format = 'MM/YY'
2201 You get the idea. Strings get their case twiddled, but everything
2202 else remains verbatim.
2204 =head1 EXPORTABLE FUNCTIONS
2206 =head2 is_plain_value
2208 Determines if the supplied argument is a plain value as understood by this
2213 =item * The value is C<undef>
2215 =item * The value is a non-reference
2217 =item * The value is an object with stringification overloading
2219 =item * The value is of the form C<< { -value => $anything } >>
2223 On failure returns C<undef>, on success returns a B<scalar> reference
2224 to the original supplied argument.
2230 The stringification overloading detection is rather advanced: it takes
2231 into consideration not only the presence of a C<""> overload, but if that
2232 fails also checks for enabled
2233 L<autogenerated versions of C<"">|overload/Magic Autogeneration>, based
2234 on either C<0+> or C<bool>.
2236 Unfortunately testing in the field indicates that this
2237 detection B<< may tickle a latent bug in perl versions before 5.018 >>,
2238 but only when very large numbers of stringifying objects are involved.
2239 At the time of writing ( Sep 2014 ) there is no clear explanation of
2240 the direct cause, nor is there a manageably small test case that reliably
2241 reproduces the problem.
2243 If you encounter any of the following exceptions in B<random places within
2244 your application stack> - this module may be to blame:
2246 Operation "ne": no method found,
2247 left argument in overloaded package <something>,
2248 right argument in overloaded package <something>
2252 Stub found while resolving method "???" overloading """" in package <something>
2254 If you fall victim to the above - please attempt to reduce the problem
2255 to something that could be sent to the L<SQL::Abstract developers
2256 |DBIx::Class/GETTING HELP/SUPPORT>
2257 (either publicly or privately). As a workaround in the meantime you can
2258 set C<$ENV{SQLA_ISVALUE_IGNORE_AUTOGENERATED_STRINGIFICATION}> to a true
2259 value, which will most likely eliminate your problem (at the expense of
2260 not being able to properly detect exotic forms of stringification).
2262 This notice and environment variable will be removed in a future version,
2263 as soon as the underlying problem is found and a reliable workaround is
2268 =head2 is_literal_value
2270 Determines if the supplied argument is a literal value as understood by this
2275 =item * C<\$sql_string>
2277 =item * C<\[ $sql_string, @bind_values ]>
2281 On failure returns C<undef>, on success returns an B<array> reference
2282 containing the unpacked version of the supplied literal SQL and bind values.
2284 =head1 WHERE CLAUSES
2288 This module uses a variation on the idea from L<DBIx::Abstract>. It
2289 is B<NOT>, repeat I<not> 100% compatible. B<The main logic of this
2290 module is that things in arrays are OR'ed, and things in hashes
2293 The easiest way to explain is to show lots of examples. After
2294 each C<%where> hash shown, it is assumed you used:
2296 my($stmt, @bind) = $sql->where(\%where);
2298 However, note that the C<%where> hash can be used directly in any
2299 of the other functions as well, as described above.
2301 =head2 Key-value pairs
2303 So, let's get started. To begin, a simple hash:
2307 status => 'completed'
2310 Is converted to SQL C<key = val> statements:
2312 $stmt = "WHERE user = ? AND status = ?";
2313 @bind = ('nwiger', 'completed');
2315 One common thing I end up doing is having a list of values that
2316 a field can be in. To do this, simply specify a list inside of
2321 status => ['assigned', 'in-progress', 'pending'];
2324 This simple code will create the following:
2326 $stmt = "WHERE user = ? AND ( status = ? OR status = ? OR status = ? )";
2327 @bind = ('nwiger', 'assigned', 'in-progress', 'pending');
2329 A field associated to an empty arrayref will be considered a
2330 logical false and will generate 0=1.
2332 =head2 Tests for NULL values
2334 If the value part is C<undef> then this is converted to SQL <IS NULL>
2343 $stmt = "WHERE user = ? AND status IS NULL";
2346 To test if a column IS NOT NULL:
2350 status => { '!=', undef },
2353 =head2 Specific comparison operators
2355 If you want to specify a different type of operator for your comparison,
2356 you can use a hashref for a given column:
2360 status => { '!=', 'completed' }
2363 Which would generate:
2365 $stmt = "WHERE user = ? AND status != ?";
2366 @bind = ('nwiger', 'completed');
2368 To test against multiple values, just enclose the values in an arrayref:
2370 status => { '=', ['assigned', 'in-progress', 'pending'] };
2372 Which would give you:
2374 "WHERE status = ? OR status = ? OR status = ?"
2377 The hashref can also contain multiple pairs, in which case it is expanded
2378 into an C<AND> of its elements:
2382 status => { '!=', 'completed', -not_like => 'pending%' }
2385 # Or more dynamically, like from a form
2386 $where{user} = 'nwiger';
2387 $where{status}{'!='} = 'completed';
2388 $where{status}{'-not_like'} = 'pending%';
2390 # Both generate this
2391 $stmt = "WHERE user = ? AND status != ? AND status NOT LIKE ?";
2392 @bind = ('nwiger', 'completed', 'pending%');
2395 To get an OR instead, you can combine it with the arrayref idea:
2399 priority => [ { '=', 2 }, { '>', 5 } ]
2402 Which would generate:
2404 $stmt = "WHERE ( priority = ? OR priority > ? ) AND user = ?";
2405 @bind = ('2', '5', 'nwiger');
2407 If you want to include literal SQL (with or without bind values), just use a
2408 scalar reference or reference to an arrayref as the value:
2411 date_entered => { '>' => \["to_date(?, 'MM/DD/YYYY')", "11/26/2008"] },
2412 date_expires => { '<' => \"now()" }
2415 Which would generate:
2417 $stmt = "WHERE date_entered > to_date(?, 'MM/DD/YYYY') AND date_expires < now()";
2418 @bind = ('11/26/2008');
2421 =head2 Logic and nesting operators
2423 In the example above,
2424 there is a subtle trap if you want to say something like
2425 this (notice the C<AND>):
2427 WHERE priority != ? AND priority != ?
2429 Because, in Perl you I<can't> do this:
2431 priority => { '!=' => 2, '!=' => 1 }
2433 As the second C<!=> key will obliterate the first. The solution
2434 is to use the special C<-modifier> form inside an arrayref:
2436 priority => [ -and => {'!=', 2},
2440 Normally, these would be joined by C<OR>, but the modifier tells it
2441 to use C<AND> instead. (Hint: You can use this in conjunction with the
2442 C<logic> option to C<new()> in order to change the way your queries
2443 work by default.) B<Important:> Note that the C<-modifier> goes
2444 B<INSIDE> the arrayref, as an extra first element. This will
2445 B<NOT> do what you think it might:
2447 priority => -and => [{'!=', 2}, {'!=', 1}] # WRONG!
2449 Here is a quick list of equivalencies, since there is some overlap:
2452 status => {'!=', 'completed', 'not like', 'pending%' }
2453 status => [ -and => {'!=', 'completed'}, {'not like', 'pending%'}]
2456 status => {'=', ['assigned', 'in-progress']}
2457 status => [ -or => {'=', 'assigned'}, {'=', 'in-progress'}]
2458 status => [ {'=', 'assigned'}, {'=', 'in-progress'} ]
2462 =head2 Special operators: IN, BETWEEN, etc.
2464 You can also use the hashref format to compare a list of fields using the
2465 C<IN> comparison operator, by specifying the list as an arrayref:
2468 status => 'completed',
2469 reportid => { -in => [567, 2335, 2] }
2472 Which would generate:
2474 $stmt = "WHERE status = ? AND reportid IN (?,?,?)";
2475 @bind = ('completed', '567', '2335', '2');
2477 The reverse operator C<-not_in> generates SQL C<NOT IN> and is used in
2480 If the argument to C<-in> is an empty array, 'sqlfalse' is generated
2481 (by default: C<1=0>). Similarly, C<< -not_in => [] >> generates
2482 'sqltrue' (by default: C<1=1>).
2484 In addition to the array you can supply a chunk of literal sql or
2485 literal sql with bind:
2488 customer => { -in => \[
2489 'SELECT cust_id FROM cust WHERE balance > ?',
2492 status => { -in => \'SELECT status_codes FROM states' },
2498 customer IN ( SELECT cust_id FROM cust WHERE balance > ? )
2499 AND status IN ( SELECT status_codes FROM states )
2503 Finally, if the argument to C<-in> is not a reference, it will be
2504 treated as a single-element array.
2506 Another pair of operators is C<-between> and C<-not_between>,
2507 used with an arrayref of two values:
2511 completion_date => {
2512 -not_between => ['2002-10-01', '2003-02-06']
2518 WHERE user = ? AND completion_date NOT BETWEEN ( ? AND ? )
2520 Just like with C<-in> all plausible combinations of literal SQL
2524 start0 => { -between => [ 1, 2 ] },
2525 start1 => { -between => \["? AND ?", 1, 2] },
2526 start2 => { -between => \"lower(x) AND upper(y)" },
2527 start3 => { -between => [
2529 \["upper(?)", 'stuff' ],
2536 ( start0 BETWEEN ? AND ? )
2537 AND ( start1 BETWEEN ? AND ? )
2538 AND ( start2 BETWEEN lower(x) AND upper(y) )
2539 AND ( start3 BETWEEN lower(x) AND upper(?) )
2541 @bind = (1, 2, 1, 2, 'stuff');
2544 These are the two builtin "special operators"; but the
2545 list can be expanded: see section L</"SPECIAL OPERATORS"> below.
2547 =head2 Unary operators: bool
2549 If you wish to test against boolean columns or functions within your
2550 database you can use the C<-bool> and C<-not_bool> operators. For
2551 example to test the column C<is_user> being true and the column
2552 C<is_enabled> being false you would use:-
2556 -not_bool => 'is_enabled',
2561 WHERE is_user AND NOT is_enabled
2563 If a more complex combination is required, testing more conditions,
2564 then you should use the and/or operators:-
2569 -not_bool => { two=> { -rlike => 'bar' } },
2570 -not_bool => { three => [ { '=', 2 }, { '>', 5 } ] },
2581 (NOT ( three = ? OR three > ? ))
2584 =head2 Nested conditions, -and/-or prefixes
2586 So far, we've seen how multiple conditions are joined with a top-level
2587 C<AND>. We can change this by putting the different conditions we want in
2588 hashes and then putting those hashes in an array. For example:
2593 status => { -like => ['pending%', 'dispatched'] },
2597 status => 'unassigned',
2601 This data structure would create the following:
2603 $stmt = "WHERE ( user = ? AND ( status LIKE ? OR status LIKE ? ) )
2604 OR ( user = ? AND status = ? ) )";
2605 @bind = ('nwiger', 'pending', 'dispatched', 'robot', 'unassigned');
2608 Clauses in hashrefs or arrayrefs can be prefixed with an C<-and> or C<-or>
2609 to change the logic inside:
2615 -and => [ workhrs => {'>', 20}, geo => 'ASIA' ],
2616 -or => { workhrs => {'<', 50}, geo => 'EURO' },
2623 $stmt = "WHERE ( user = ?
2624 AND ( ( workhrs > ? AND geo = ? )
2625 OR ( workhrs < ? OR geo = ? ) ) )";
2626 @bind = ('nwiger', '20', 'ASIA', '50', 'EURO');
2628 =head3 Algebraic inconsistency, for historical reasons
2630 C<Important note>: when connecting several conditions, the C<-and->|C<-or>
2631 operator goes C<outside> of the nested structure; whereas when connecting
2632 several constraints on one column, the C<-and> operator goes
2633 C<inside> the arrayref. Here is an example combining both features:
2636 -and => [a => 1, b => 2],
2637 -or => [c => 3, d => 4],
2638 e => [-and => {-like => 'foo%'}, {-like => '%bar'} ]
2643 WHERE ( ( ( a = ? AND b = ? )
2644 OR ( c = ? OR d = ? )
2645 OR ( e LIKE ? AND e LIKE ? ) ) )
2647 This difference in syntax is unfortunate but must be preserved for
2648 historical reasons. So be careful: the two examples below would
2649 seem algebraically equivalent, but they are not
2652 { -like => 'foo%' },
2653 { -like => '%bar' },
2655 # yields: WHERE ( ( col LIKE ? AND col LIKE ? ) )
2658 { col => { -like => 'foo%' } },
2659 { col => { -like => '%bar' } },
2661 # yields: WHERE ( ( col LIKE ? OR col LIKE ? ) )
2664 =head2 Literal SQL and value type operators
2666 The basic premise of SQL::Abstract is that in WHERE specifications the "left
2667 side" is a column name and the "right side" is a value (normally rendered as
2668 a placeholder). This holds true for both hashrefs and arrayref pairs as you
2669 see in the L</WHERE CLAUSES> examples above. Sometimes it is necessary to
2670 alter this behavior. There are several ways of doing so.
2674 This is a virtual operator that signals the string to its right side is an
2675 identifier (a column name) and not a value. For example to compare two
2676 columns you would write:
2679 priority => { '<', 2 },
2680 requestor => { -ident => 'submitter' },
2685 $stmt = "WHERE priority < ? AND requestor = submitter";
2688 If you are maintaining legacy code you may see a different construct as
2689 described in L</Deprecated usage of Literal SQL>, please use C<-ident> in new
2694 This is a virtual operator that signals that the construct to its right side
2695 is a value to be passed to DBI. This is for example necessary when you want
2696 to write a where clause against an array (for RDBMS that support such
2697 datatypes). For example:
2700 array => { -value => [1, 2, 3] }
2705 $stmt = 'WHERE array = ?';
2706 @bind = ([1, 2, 3]);
2708 Note that if you were to simply say:
2714 the result would probably not be what you wanted:
2716 $stmt = 'WHERE array = ? OR array = ? OR array = ?';
2721 Finally, sometimes only literal SQL will do. To include a random snippet
2722 of SQL verbatim, you specify it as a scalar reference. Consider this only
2723 as a last resort. Usually there is a better way. For example:
2726 priority => { '<', 2 },
2727 requestor => { -in => \'(SELECT name FROM hitmen)' },
2732 $stmt = "WHERE priority < ? AND requestor IN (SELECT name FROM hitmen)"
2735 Note that in this example, you only get one bind parameter back, since
2736 the verbatim SQL is passed as part of the statement.
2740 Never use untrusted input as a literal SQL argument - this is a massive
2741 security risk (there is no way to check literal snippets for SQL
2742 injections and other nastyness). If you need to deal with untrusted input
2743 use literal SQL with placeholders as described next.
2745 =head3 Literal SQL with placeholders and bind values (subqueries)
2747 If the literal SQL to be inserted has placeholders and bind values,
2748 use a reference to an arrayref (yes this is a double reference --
2749 not so common, but perfectly legal Perl). For example, to find a date
2750 in Postgres you can use something like this:
2753 date_column => \[ "= date '2008-09-30' - ?::integer", 10 ]
2758 $stmt = "WHERE ( date_column = date '2008-09-30' - ?::integer )"
2761 Note that you must pass the bind values in the same format as they are returned
2762 by L<where|/where(\%where, $order)>. This means that if you set L</bindtype>
2763 to C<columns>, you must provide the bind values in the
2764 C<< [ column_meta => value ] >> format, where C<column_meta> is an opaque
2765 scalar value; most commonly the column name, but you can use any scalar value
2766 (including references and blessed references), L<SQL::Abstract> will simply
2767 pass it through intact. So if C<bindtype> is set to C<columns> the above
2768 example will look like:
2771 date_column => \[ "= date '2008-09-30' - ?::integer", [ {} => 10 ] ]
2774 Literal SQL is especially useful for nesting parenthesized clauses in the
2775 main SQL query. Here is a first example:
2777 my ($sub_stmt, @sub_bind) = ("SELECT c1 FROM t1 WHERE c2 < ? AND c3 LIKE ?",
2781 bar => \["IN ($sub_stmt)" => @sub_bind],
2786 $stmt = "WHERE (foo = ? AND bar IN (SELECT c1 FROM t1
2787 WHERE c2 < ? AND c3 LIKE ?))";
2788 @bind = (1234, 100, "foo%");
2790 Other subquery operators, like for example C<"E<gt> ALL"> or C<"NOT IN">,
2791 are expressed in the same way. Of course the C<$sub_stmt> and
2792 its associated bind values can be generated through a former call
2795 my ($sub_stmt, @sub_bind)
2796 = $sql->select("t1", "c1", {c2 => {"<" => 100},
2797 c3 => {-like => "foo%"}});
2800 bar => \["> ALL ($sub_stmt)" => @sub_bind],
2803 In the examples above, the subquery was used as an operator on a column;
2804 but the same principle also applies for a clause within the main C<%where>
2805 hash, like an EXISTS subquery:
2807 my ($sub_stmt, @sub_bind)
2808 = $sql->select("t1", "*", {c1 => 1, c2 => \"> t0.c0"});
2809 my %where = ( -and => [
2811 \["EXISTS ($sub_stmt)" => @sub_bind],
2816 $stmt = "WHERE (foo = ? AND EXISTS (SELECT * FROM t1
2817 WHERE c1 = ? AND c2 > t0.c0))";
2821 Observe that the condition on C<c2> in the subquery refers to
2822 column C<t0.c0> of the main query: this is I<not> a bind
2823 value, so we have to express it through a scalar ref.
2824 Writing C<< c2 => {">" => "t0.c0"} >> would have generated
2825 C<< c2 > ? >> with bind value C<"t0.c0"> ... not exactly
2826 what we wanted here.
2828 Finally, here is an example where a subquery is used
2829 for expressing unary negation:
2831 my ($sub_stmt, @sub_bind)
2832 = $sql->where({age => [{"<" => 10}, {">" => 20}]});
2833 $sub_stmt =~ s/^ where //i; # don't want "WHERE" in the subclause
2835 lname => {like => '%son%'},
2836 \["NOT ($sub_stmt)" => @sub_bind],
2841 $stmt = "lname LIKE ? AND NOT ( age < ? OR age > ? )"
2842 @bind = ('%son%', 10, 20)
2844 =head3 Deprecated usage of Literal SQL
2846 Below are some examples of archaic use of literal SQL. It is shown only as
2847 reference for those who deal with legacy code. Each example has a much
2848 better, cleaner and safer alternative that users should opt for in new code.
2854 my %where = ( requestor => \'IS NOT NULL' )
2856 $stmt = "WHERE requestor IS NOT NULL"
2858 This used to be the way of generating NULL comparisons, before the handling
2859 of C<undef> got formalized. For new code please use the superior syntax as
2860 described in L</Tests for NULL values>.
2864 my %where = ( requestor => \'= submitter' )
2866 $stmt = "WHERE requestor = submitter"
2868 This used to be the only way to compare columns. Use the superior L</-ident>
2869 method for all new code. For example an identifier declared in such a way
2870 will be properly quoted if L</quote_char> is properly set, while the legacy
2871 form will remain as supplied.
2875 my %where = ( is_ready => \"", completed => { '>', '2012-12-21' } )
2877 $stmt = "WHERE completed > ? AND is_ready"
2878 @bind = ('2012-12-21')
2880 Using an empty string literal used to be the only way to express a boolean.
2881 For all new code please use the much more readable
2882 L<-bool|/Unary operators: bool> operator.
2888 These pages could go on for a while, since the nesting of the data
2889 structures this module can handle are pretty much unlimited (the
2890 module implements the C<WHERE> expansion as a recursive function
2891 internally). Your best bet is to "play around" with the module a
2892 little to see how the data structures behave, and choose the best
2893 format for your data based on that.
2895 And of course, all the values above will probably be replaced with
2896 variables gotten from forms or the command line. After all, if you
2897 knew everything ahead of time, you wouldn't have to worry about
2898 dynamically-generating SQL and could just hardwire it into your
2901 =head1 ORDER BY CLAUSES
2903 Some functions take an order by clause. This can either be a scalar (just a
2904 column name), a hashref of C<< { -desc => 'col' } >> or C<< { -asc => 'col' }
2905 >>, a scalarref, an arrayref-ref, or an arrayref of any of the previous
2908 Given | Will Generate
2909 ---------------------------------------------------------------
2911 'colA' | ORDER BY colA
2913 [qw/colA colB/] | ORDER BY colA, colB
2915 {-asc => 'colA'} | ORDER BY colA ASC
2917 {-desc => 'colB'} | ORDER BY colB DESC
2919 ['colA', {-asc => 'colB'}] | ORDER BY colA, colB ASC
2921 { -asc => [qw/colA colB/] } | ORDER BY colA ASC, colB ASC
2923 \'colA DESC' | ORDER BY colA DESC
2925 \[ 'FUNC(colA, ?)', $x ] | ORDER BY FUNC(colA, ?)
2926 | /* ...with $x bound to ? */
2929 { -asc => 'colA' }, | colA ASC,
2930 { -desc => [qw/colB/] }, | colB DESC,
2931 { -asc => [qw/colC colD/] },| colC ASC, colD ASC,
2932 \'colE DESC', | colE DESC,
2933 \[ 'FUNC(colF, ?)', $x ], | FUNC(colF, ?)
2934 ] | /* ...with $x bound to ? */
2935 ===============================================================
2939 =head1 SPECIAL OPERATORS
2941 my $sqlmaker = SQL::Abstract->new(special_ops => [
2945 my ($self, $field, $op, $arg) = @_;
2951 handler => 'method_name',
2955 A "special operator" is a SQL syntactic clause that can be
2956 applied to a field, instead of a usual binary operator.
2959 WHERE field IN (?, ?, ?)
2960 WHERE field BETWEEN ? AND ?
2961 WHERE MATCH(field) AGAINST (?, ?)
2963 Special operators IN and BETWEEN are fairly standard and therefore
2964 are builtin within C<SQL::Abstract> (as the overridable methods
2965 C<_where_field_IN> and C<_where_field_BETWEEN>). For other operators,
2966 like the MATCH .. AGAINST example above which is specific to MySQL,
2967 you can write your own operator handlers - supply a C<special_ops>
2968 argument to the C<new> method. That argument takes an arrayref of
2969 operator definitions; each operator definition is a hashref with two
2976 the regular expression to match the operator
2980 Either a coderef or a plain scalar method name. In both cases
2981 the expected return is C<< ($sql, @bind) >>.
2983 When supplied with a method name, it is simply called on the
2984 L<SQL::Abstract> object as:
2986 $self->$method_name($field, $op, $arg)
2990 $field is the LHS of the operator
2991 $op is the part that matched the handler regex
2994 When supplied with a coderef, it is called as:
2996 $coderef->($self, $field, $op, $arg)
3001 For example, here is an implementation
3002 of the MATCH .. AGAINST syntax for MySQL
3004 my $sqlmaker = SQL::Abstract->new(special_ops => [
3006 # special op for MySql MATCH (field) AGAINST(word1, word2, ...)
3007 {regex => qr/^match$/i,
3009 my ($self, $field, $op, $arg) = @_;
3010 $arg = [$arg] if not ref $arg;
3011 my $label = $self->_quote($field);
3012 my ($placeholder) = $self->_convert('?');
3013 my $placeholders = join ", ", (($placeholder) x @$arg);
3014 my $sql = $self->_sqlcase('match') . " ($label) "
3015 . $self->_sqlcase('against') . " ($placeholders) ";
3016 my @bind = $self->_bindtype($field, @$arg);
3017 return ($sql, @bind);
3024 =head1 UNARY OPERATORS
3026 my $sqlmaker = SQL::Abstract->new(unary_ops => [
3030 my ($self, $op, $arg) = @_;
3036 handler => 'method_name',
3040 A "unary operator" is a SQL syntactic clause that can be
3041 applied to a field - the operator goes before the field
3043 You can write your own operator handlers - supply a C<unary_ops>
3044 argument to the C<new> method. That argument takes an arrayref of
3045 operator definitions; each operator definition is a hashref with two
3052 the regular expression to match the operator
3056 Either a coderef or a plain scalar method name. In both cases
3057 the expected return is C<< $sql >>.
3059 When supplied with a method name, it is simply called on the
3060 L<SQL::Abstract> object as:
3062 $self->$method_name($op, $arg)
3066 $op is the part that matched the handler regex
3067 $arg is the RHS or argument of the operator
3069 When supplied with a coderef, it is called as:
3071 $coderef->($self, $op, $arg)
3079 Thanks to some benchmarking by Mark Stosberg, it turns out that
3080 this module is many orders of magnitude faster than using C<DBIx::Abstract>.
3081 I must admit this wasn't an intentional design issue, but it's a
3082 byproduct of the fact that you get to control your C<DBI> handles
3085 To maximize performance, use a code snippet like the following:
3087 # prepare a statement handle using the first row
3088 # and then reuse it for the rest of the rows
3090 for my $href (@array_of_hashrefs) {
3091 $stmt ||= $sql->insert('table', $href);
3092 $sth ||= $dbh->prepare($stmt);
3093 $sth->execute($sql->values($href));
3096 The reason this works is because the keys in your C<$href> are sorted
3097 internally by B<SQL::Abstract>. Thus, as long as your data retains
3098 the same structure, you only have to generate the SQL the first time
3099 around. On subsequent queries, simply use the C<values> function provided
3100 by this module to return your values in the correct order.
3102 However this depends on the values having the same type - if, for
3103 example, the values of a where clause may either have values
3104 (resulting in sql of the form C<column = ?> with a single bind
3105 value), or alternatively the values might be C<undef> (resulting in
3106 sql of the form C<column IS NULL> with no bind value) then the
3107 caching technique suggested will not work.
3111 If you use my C<CGI::FormBuilder> module at all, you'll hopefully
3112 really like this part (I do, at least). Building up a complex query
3113 can be as simple as the following:
3120 use CGI::FormBuilder;
3123 my $form = CGI::FormBuilder->new(...);
3124 my $sql = SQL::Abstract->new;
3126 if ($form->submitted) {
3127 my $field = $form->field;
3128 my $id = delete $field->{id};
3129 my($stmt, @bind) = $sql->update('table', $field, {id => $id});
3132 Of course, you would still have to connect using C<DBI> to run the
3133 query, but the point is that if you make your form look like your
3134 table, the actual query script can be extremely simplistic.
3136 If you're B<REALLY> lazy (I am), check out C<HTML::QuickTable> for
3137 a fast interface to returning and formatting data. I frequently
3138 use these three modules together to write complex database query
3139 apps in under 50 lines.
3141 =head1 HOW TO CONTRIBUTE
3143 Contributions are always welcome, in all usable forms (we especially
3144 welcome documentation improvements). The delivery methods include git-
3145 or unified-diff formatted patches, GitHub pull requests, or plain bug
3146 reports either via RT or the Mailing list. Contributors are generally
3147 granted full access to the official repository after their first several
3148 patches pass successful review.
3150 This project is maintained in a git repository. The code and related tools are
3151 accessible at the following locations:
3155 =item * Official repo: L<git://git.shadowcat.co.uk/dbsrgits/SQL-Abstract.git>
3157 =item * Official gitweb: L<http://git.shadowcat.co.uk/gitweb/gitweb.cgi?p=dbsrgits/SQL-Abstract.git>
3159 =item * GitHub mirror: L<https://github.com/dbsrgits/sql-abstract>
3161 =item * Authorized committers: L<ssh://dbsrgits@git.shadowcat.co.uk/SQL-Abstract.git>
3167 Version 1.50 was a major internal refactoring of C<SQL::Abstract>.
3168 Great care has been taken to preserve the I<published> behavior
3169 documented in previous versions in the 1.* family; however,
3170 some features that were previously undocumented, or behaved
3171 differently from the documentation, had to be changed in order
3172 to clarify the semantics. Hence, client code that was relying
3173 on some dark areas of C<SQL::Abstract> v1.*
3174 B<might behave differently> in v1.50.
3176 The main changes are:
3182 support for literal SQL through the C<< \ [ $sql, @bind ] >> syntax.
3186 support for the { operator => \"..." } construct (to embed literal SQL)
3190 support for the { operator => \["...", @bind] } construct (to embed literal SQL with bind values)
3194 optional support for L<array datatypes|/"Inserting and Updating Arrays">
3198 defensive programming: check arguments
3202 fixed bug with global logic, which was previously implemented
3203 through global variables yielding side-effects. Prior versions would
3204 interpret C<< [ {cond1, cond2}, [cond3, cond4] ] >>
3205 as C<< "(cond1 AND cond2) OR (cond3 AND cond4)" >>.
3206 Now this is interpreted
3207 as C<< "(cond1 AND cond2) OR (cond3 OR cond4)" >>.
3212 fixed semantics of _bindtype on array args
3216 dropped the C<_anoncopy> of the %where tree. No longer necessary,
3217 we just avoid shifting arrays within that tree.
3221 dropped the C<_modlogic> function
3225 =head1 ACKNOWLEDGEMENTS
3227 There are a number of individuals that have really helped out with
3228 this module. Unfortunately, most of them submitted bugs via CPAN
3229 so I have no idea who they are! But the people I do know are:
3231 Ash Berlin (order_by hash term support)
3232 Matt Trout (DBIx::Class support)
3233 Mark Stosberg (benchmarking)
3234 Chas Owens (initial "IN" operator support)
3235 Philip Collins (per-field SQL functions)
3236 Eric Kolve (hashref "AND" support)
3237 Mike Fragassi (enhancements to "BETWEEN" and "LIKE")
3238 Dan Kubb (support for "quote_char" and "name_sep")
3239 Guillermo Roditi (patch to cleanup "IN" and "BETWEEN", fix and tests for _order_by)
3240 Laurent Dami (internal refactoring, extensible list of special operators, literal SQL)
3241 Norbert Buchmuller (support for literal SQL in hashpair, misc. fixes & tests)
3242 Peter Rabbitson (rewrite of SQLA::Test, misc. fixes & tests)
3243 Oliver Charles (support for "RETURNING" after "INSERT")
3249 L<DBIx::Class>, L<DBIx::Abstract>, L<CGI::FormBuilder>, L<HTML::QuickTable>.
3253 Copyright (c) 2001-2007 Nathan Wiger <nwiger@cpan.org>. All Rights Reserved.
3255 This module is actively maintained by Matt Trout <mst@shadowcatsystems.co.uk>
3257 For support, your best bet is to try the C<DBIx::Class> users mailing list.
3258 While not an official support venue, C<DBIx::Class> makes heavy use of
3259 C<SQL::Abstract>, and as such list members there are very familiar with
3260 how to create queries.
3264 This module is free software; you may copy this under the same
3265 terms as perl itself (either the GNU General Public License or
3266 the Artistic License)