1 package SQL::Abstract; # see doc at end of file
10 our @EXPORT_OK = qw(is_plain_value is_literal_value);
20 *SQL::Abstract::_ENV_::DETECT_AUTOGENERATED_STRINGIFICATION = $ENV{SQLA_ISVALUE_IGNORE_AUTOGENERATED_STRINGIFICATION}
26 #======================================================================
28 #======================================================================
30 our $VERSION = '1.87';
32 # This would confuse some packagers
33 $VERSION = eval $VERSION if $VERSION =~ /_/; # numify for warning-free dev releases
37 # special operators (-in, -between). May be extended/overridden by user.
38 # See section WHERE: BUILTIN SPECIAL OPERATORS below for implementation
39 my @BUILTIN_SPECIAL_OPS = (
40 {regex => qr/^ (?: not \s )? between $/ix, handler => sub { die "NOPE" }},
41 {regex => qr/^ is (?: \s+ not )? $/ix, handler => sub { die "NOPE" }},
42 {regex => qr/^ (?: not \s )? in $/ix, handler => sub { die "NOPE" }},
43 {regex => qr/^ ident $/ix, handler => sub { die "NOPE" }},
44 {regex => qr/^ value $/ix, handler => sub { die "NOPE" }},
47 #======================================================================
48 # DEBUGGING AND ERROR REPORTING
49 #======================================================================
52 return unless $_[0]->{debug}; shift; # a little faster
53 my $func = (caller(1))[3];
54 warn "[$func] ", @_, "\n";
58 my($func) = (caller(1))[3];
59 Carp::carp "[$func] Warning: ", @_;
63 my($func) = (caller(1))[3];
64 Carp::croak "[$func] Fatal: ", @_;
67 sub is_literal_value ($) {
68 ref $_[0] eq 'SCALAR' ? [ ${$_[0]} ]
69 : ( ref $_[0] eq 'REF' and ref ${$_[0]} eq 'ARRAY' ) ? [ @${ $_[0] } ]
73 sub is_undef_value ($) {
77 and exists $_[0]->{-value}
78 and not defined $_[0]->{-value}
82 # FIXME XSify - this can be done so much more efficiently
83 sub is_plain_value ($) {
85 ! length ref $_[0] ? \($_[0])
87 ref $_[0] eq 'HASH' and keys %{$_[0]} == 1
89 exists $_[0]->{-value}
90 ) ? \($_[0]->{-value})
92 # reuse @_ for even moar speedz
93 defined ( $_[1] = Scalar::Util::blessed $_[0] )
95 # deliberately not using Devel::OverloadInfo - the checks we are
96 # intersted in are much more limited than the fullblown thing, and
97 # this is a very hot piece of code
99 # simply using ->can('(""') can leave behind stub methods that
100 # break actually using the overload later (see L<perldiag/Stub
101 # found while resolving method "%s" overloading "%s" in package
102 # "%s"> and the source of overload::mycan())
104 # either has stringification which DBI SHOULD prefer out of the box
105 grep { *{ (qq[${_}::(""]) }{CODE} } @{ $_[2] = mro::get_linear_isa( $_[1] ) }
107 # has nummification or boolification, AND fallback is *not* disabled
109 SQL::Abstract::_ENV_::DETECT_AUTOGENERATED_STRINGIFICATION
112 grep { *{"${_}::(0+"}{CODE} } @{$_[2]}
114 grep { *{"${_}::(bool"}{CODE} } @{$_[2]}
118 # no fallback specified at all
119 ! ( ($_[3]) = grep { *{"${_}::()"}{CODE} } @{$_[2]} )
121 # fallback explicitly undef
122 ! defined ${"$_[3]::()"}
135 #======================================================================
137 #======================================================================
141 my $class = ref($self) || $self;
142 my %opt = (ref $_[0] eq 'HASH') ? %{$_[0]} : @_;
144 # choose our case by keeping an option around
145 delete $opt{case} if $opt{case} && $opt{case} ne 'lower';
147 # default logic for interpreting arrayrefs
148 $opt{logic} = $opt{logic} ? uc $opt{logic} : 'OR';
150 # how to return bind vars
151 $opt{bindtype} ||= 'normal';
153 # default comparison is "=", but can be overridden
156 # try to recognize which are the 'equality' and 'inequality' ops
157 # (temporary quickfix (in 2007), should go through a more seasoned API)
158 $opt{equality_op} = qr/^( \Q$opt{cmp}\E | \= )$/ix;
159 $opt{inequality_op} = qr/^( != | <> )$/ix;
161 $opt{like_op} = qr/^ (is_)?r?like $/xi;
162 $opt{not_like_op} = qr/^ (is_)?not_r?like $/xi;
165 $opt{sqltrue} ||= '1=1';
166 $opt{sqlfalse} ||= '0=1';
169 $opt{special_ops} ||= [];
171 if ($class->isa('DBIx::Class::SQLMaker')) {
172 $opt{warn_once_on_nest} = 1;
173 $opt{disable_old_special_ops} = 1;
177 $opt{unary_ops} ||= [];
179 # rudimentary sanity-check for user supplied bits treated as functions/operators
180 # If a purported function matches this regular expression, an exception is thrown.
181 # Literal SQL is *NOT* subject to this check, only functions (and column names
182 # when quoting is not in effect)
185 # need to guard against ()'s in column names too, but this will break tons of
186 # hacks... ideas anyone?
187 $opt{injection_guard} ||= qr/
193 $opt{expand_unary} = {};
196 not => '_expand_not',
197 bool => '_expand_bool',
198 and => '_expand_op_andor',
199 or => '_expand_op_andor',
200 nest => '_expand_nest',
201 bind => '_expand_bind',
203 not_in => '_expand_in',
204 row => '_expand_row',
205 between => '_expand_between',
206 not_between => '_expand_between',
208 (map +($_ => '_expand_op_is'), ('is', 'is_not')),
209 ident => '_expand_ident',
210 value => '_expand_value',
214 'between' => '_expand_between',
215 'not_between' => '_expand_between',
216 'in' => '_expand_in',
217 'not_in' => '_expand_in',
218 'nest' => '_expand_nest',
219 (map +($_ => '_expand_op_andor'), ('and', 'or')),
220 (map +($_ => '_expand_op_is'), ('is', 'is_not')),
221 'ident' => '_expand_ident',
222 'value' => '_expand_value',
226 (map +($_, "_render_$_"), qw(op func bind ident literal row)),
231 (map +($_ => '_render_op_between'), 'between', 'not_between'),
232 (map +($_ => '_render_op_in'), 'in', 'not_in'),
233 (map +($_ => '_render_unop_postfix'),
234 'is_null', 'is_not_null', 'asc', 'desc',
236 (not => '_render_op_not'),
237 (map +($_ => '_render_op_andor'), qw(and or)),
238 ',' => '_render_op_multop',
241 return bless \%opt, $class;
244 sub sqltrue { +{ -literal => [ $_[0]->{sqltrue} ] } }
245 sub sqlfalse { +{ -literal => [ $_[0]->{sqlfalse} ] } }
247 sub _assert_pass_injection_guard {
248 if ($_[1] =~ $_[0]->{injection_guard}) {
249 my $class = ref $_[0];
250 puke "Possible SQL injection attempt '$_[1]'. If this is indeed a part of the "
251 . "desired SQL use literal SQL ( \'...' or \[ '...' ] ) or supply your own "
252 . "{injection_guard} attribute to ${class}->new()"
257 #======================================================================
259 #======================================================================
263 my $table = $self->_table(shift);
264 my $data = shift || return;
267 my $method = $self->_METHOD_FOR_refkind("_insert", $data);
268 my ($sql, @bind) = $self->$method($data);
269 $sql = join " ", $self->_sqlcase('insert into'), $table, $sql;
271 if ($options->{returning}) {
272 my ($s, @b) = $self->_insert_returning($options);
277 return wantarray ? ($sql, @bind) : $sql;
280 # So that subclasses can override INSERT ... RETURNING separately from
281 # UPDATE and DELETE (e.g. DBIx::Class::SQLMaker::Oracle does this)
282 sub _insert_returning { shift->_returning(@_) }
285 my ($self, $options) = @_;
287 my $f = $options->{returning};
289 my ($sql, @bind) = $self->render_aqt(
290 $self->_expand_maybe_list_expr($f, -ident)
293 ? $self->_sqlcase(' returning ') . $sql
294 : ($self->_sqlcase(' returning ').$sql, @bind);
297 sub _insert_HASHREF { # explicit list of fields and then values
298 my ($self, $data) = @_;
300 my @fields = sort keys %$data;
302 my ($sql, @bind) = $self->_insert_values($data);
305 $_ = $self->_quote($_) foreach @fields;
306 $sql = "( ".join(", ", @fields).") ".$sql;
308 return ($sql, @bind);
311 sub _insert_ARRAYREF { # just generate values(?,?) part (no list of fields)
312 my ($self, $data) = @_;
314 # no names (arrayref) so can't generate bindtype
315 $self->{bindtype} ne 'columns'
316 or belch "can't do 'columns' bindtype when called with arrayref";
318 my (@values, @all_bind);
319 foreach my $value (@$data) {
320 my ($values, @bind) = $self->_insert_value(undef, $value);
321 push @values, $values;
322 push @all_bind, @bind;
324 my $sql = $self->_sqlcase('values')." ( ".join(", ", @values)." )";
325 return ($sql, @all_bind);
328 sub _insert_ARRAYREFREF { # literal SQL with bind
329 my ($self, $data) = @_;
331 my ($sql, @bind) = @${$data};
332 $self->_assert_bindval_matches_bindtype(@bind);
334 return ($sql, @bind);
338 sub _insert_SCALARREF { # literal SQL without bind
339 my ($self, $data) = @_;
345 my ($self, $data) = @_;
347 my (@values, @all_bind);
348 foreach my $column (sort keys %$data) {
349 my ($values, @bind) = $self->_insert_value($column, $data->{$column});
350 push @values, $values;
351 push @all_bind, @bind;
353 my $sql = $self->_sqlcase('values')." ( ".join(", ", @values)." )";
354 return ($sql, @all_bind);
358 my ($self, $column, $v) = @_;
360 return $self->render_aqt(
361 $self->_expand_insert_value($column, $v)
365 sub _expand_insert_value {
366 my ($self, $column, $v) = @_;
368 if (ref($v) eq 'ARRAY') {
369 if ($self->{array_datatypes}) {
370 return +{ -bind => [ $column, $v ] };
372 my ($sql, @bind) = @$v;
373 $self->_assert_bindval_matches_bindtype(@bind);
374 return +{ -literal => $v };
376 if (ref($v) eq 'HASH') {
377 if (grep !/^-/, keys %$v) {
378 belch "HASH ref as bind value in insert is not supported";
379 return +{ -bind => [ $column, $v ] };
383 return +{ -bind => [ $column, undef ] };
385 local our $Cur_Col_Meta = $column;
386 return $self->expand_expr($v);
391 #======================================================================
393 #======================================================================
398 my $table = $self->_table(shift);
399 my $data = shift || return;
403 # first build the 'SET' part of the sql statement
404 puke "Unsupported data type specified to \$sql->update"
405 unless ref $data eq 'HASH';
407 my ($sql, @all_bind) = $self->_update_set_values($data);
408 $sql = $self->_sqlcase('update ') . $table . $self->_sqlcase(' set ')
412 my($where_sql, @where_bind) = $self->where($where);
414 push @all_bind, @where_bind;
417 if ($options->{returning}) {
418 my ($returning_sql, @returning_bind) = $self->_update_returning($options);
419 $sql .= $returning_sql;
420 push @all_bind, @returning_bind;
423 return wantarray ? ($sql, @all_bind) : $sql;
426 sub _update_set_values {
427 my ($self, $data) = @_;
429 return $self->render_aqt(
430 $self->_expand_update_set_values($data),
434 sub _expand_update_set_values {
435 my ($self, $data) = @_;
436 $self->_expand_maybe_list_expr( [
439 $set = { -bind => $_ } unless defined $set;
440 +{ -op => [ '=', $self->_expand_ident(-ident => $k), $set ] };
446 ? ($self->{array_datatypes}
447 ? [ $k, +{ -bind => [ $k, $v ] } ]
448 : [ $k, +{ -literal => $v } ])
450 local our $Cur_Col_Meta = $k;
451 [ $k, $self->_expand_expr($v) ]
458 # So that subclasses can override UPDATE ... RETURNING separately from
460 sub _update_returning { shift->_returning(@_) }
464 #======================================================================
466 #======================================================================
471 my $table = $self->_table(shift);
472 my $fields = shift || '*';
476 my ($fields_sql, @bind) = $self->_select_fields($fields);
478 my ($where_sql, @where_bind) = $self->where($where, $order);
479 push @bind, @where_bind;
481 my $sql = join(' ', $self->_sqlcase('select'), $fields_sql,
482 $self->_sqlcase('from'), $table)
485 return wantarray ? ($sql, @bind) : $sql;
489 my ($self, $fields) = @_;
490 return $fields unless ref($fields);
491 return $self->render_aqt(
492 $self->_expand_maybe_list_expr($fields, '-ident')
496 #======================================================================
498 #======================================================================
503 my $table = $self->_table(shift);
507 my($where_sql, @bind) = $self->where($where);
508 my $sql = $self->_sqlcase('delete from ') . $table . $where_sql;
510 if ($options->{returning}) {
511 my ($returning_sql, @returning_bind) = $self->_delete_returning($options);
512 $sql .= $returning_sql;
513 push @bind, @returning_bind;
516 return wantarray ? ($sql, @bind) : $sql;
519 # So that subclasses can override DELETE ... RETURNING separately from
521 sub _delete_returning { shift->_returning(@_) }
525 #======================================================================
527 #======================================================================
531 # Finally, a separate routine just to handle WHERE clauses
533 my ($self, $where, $order) = @_;
535 local $self->{convert_where} = $self->{convert};
538 my ($sql, @bind) = defined($where)
539 ? $self->_recurse_where($where)
541 $sql = (defined $sql and length $sql) ? $self->_sqlcase(' where ') . "( $sql )" : '';
545 my ($order_sql, @order_bind) = $self->_order_by($order);
547 push @bind, @order_bind;
550 return wantarray ? ($sql, @bind) : $sql;
553 { our $Default_Scalar_To = -value }
556 my ($self, $expr, $default_scalar_to) = @_;
557 local our $Default_Scalar_To = $default_scalar_to if $default_scalar_to;
558 $self->_expand_expr($expr);
562 my ($self, $aqt) = @_;
563 my ($k, $v, @rest) = %$aqt;
565 die "Not a node type: $k" unless $k =~ s/^-//;
566 if (my $meth = $self->{render}{$k}) {
567 return $self->$meth($v);
569 die "notreached: $k";
573 my ($self, $expr) = @_;
574 $self->render_aqt($self->expand_expr($expr));
578 my ($self, $raw) = @_;
579 s/^-(?=.)//, s/\s+/_/g for my $op = lc $raw;
584 my ($self, $expr) = @_;
585 our $Expand_Depth ||= 0; local $Expand_Depth = $Expand_Depth + 1;
586 return undef unless defined($expr);
587 if (ref($expr) eq 'HASH') {
588 return undef unless my $kc = keys %$expr;
590 return $self->_expand_op_andor(and => $expr);
592 my ($key, $value) = %$expr;
593 if ($key =~ /^-/ and $key =~ s/ [_\s]? \d+ $//x ) {
594 belch 'Use of [and|or|nest]_N modifiers is deprecated and will be removed in SQLA v2.0. '
595 . "You probably wanted ...-and => [ $key => COND1, $key => COND2 ... ]";
597 return $self->_expand_hashpair($key, $value);
599 if (ref($expr) eq 'ARRAY') {
600 return $self->_expand_op_andor(lc($self->{logic}), $expr);
602 if (my $literal = is_literal_value($expr)) {
603 return +{ -literal => $literal };
605 if (!ref($expr) or Scalar::Util::blessed($expr)) {
606 return $self->_expand_scalar($expr);
611 sub _expand_hashpair {
612 my ($self, $k, $v) = @_;
613 unless (defined($k) and length($k)) {
614 if (defined($k) and my $literal = is_literal_value($v)) {
615 belch 'Hash-pairs consisting of an empty string with a literal are deprecated, and will be removed in 2.0: use -and => [ $literal ] instead';
616 return { -literal => $literal };
618 puke "Supplying an empty left hand side argument is not supported";
621 return $self->_expand_hashpair_op($k, $v);
623 return $self->_expand_hashpair_ident($k, $v);
626 sub _expand_hashpair_ident {
627 my ($self, $k, $v) = @_;
629 local our $Cur_Col_Meta = $k;
631 # hash with multiple or no elements is andor
633 if (ref($v) eq 'HASH' and keys %$v != 1) {
634 return $self->_expand_op_andor(and => $v, $k);
637 # undef needs to be re-sent with cmp to achieve IS/IS NOT NULL
639 if (is_undef_value($v)) {
640 return $self->_expand_hashpair_cmp($k => undef);
643 # scalars and objects get expanded as whatever requested or values
645 if (!ref($v) or Scalar::Util::blessed($v)) {
646 return $self->_expand_hashpair_scalar($k, $v);
649 # single key hashref is a hashtriple
651 if (ref($v) eq 'HASH') {
652 return $self->_expand_hashtriple($k, %$v);
655 # arrayref needs re-engineering over the elements
657 if (ref($v) eq 'ARRAY') {
658 return $self->sqlfalse unless @$v;
659 $self->_debug("ARRAY($k) means distribute over elements");
661 $v->[0] =~ /^-(and|or)$/i
662 ? (shift(@{$v = [ @$v ]}), $1)
663 : lc($self->{logic} || 'OR')
665 return $self->_expand_op_andor(
670 if (my $literal = is_literal_value($v)) {
672 belch 'Hash-pairs consisting of an empty string with a literal are deprecated, and will be removed in 2.0: use -and => [ $literal ] instead';
675 my ($sql, @bind) = @$literal;
676 if ($self->{bindtype} eq 'columns') {
678 $self->_assert_bindval_matches_bindtype($_);
681 return +{ -literal => [ $self->_quote($k).' '.$sql, @bind ] };
687 my ($self, $expr) = @_;
689 return $self->_expand_expr({ (our $Default_Scalar_To) => $expr });
692 sub _expand_hashpair_scalar {
693 my ($self, $k, $v) = @_;
695 return $self->_expand_hashpair_cmp(
696 $k, $self->_expand_scalar($v),
700 sub _expand_hashpair_op {
701 my ($self, $k, $v) = @_;
703 $self->_assert_pass_injection_guard($k =~ /\A-(.*)\Z/s);
705 my $op = $self->_normalize_op($k);
707 if (my $exp = $self->{expand}{$op}) {
708 return $self->$exp($op, $v);
711 # Ops prefixed with -not_ get converted
713 if (my ($rest) = $op =~/^not_(.*)$/) {
716 $self->_expand_expr({ "-${rest}", $v })
722 my $op = join(' ', split '_', $op);
724 # the old special op system requires illegality for top-level use
727 (our $Expand_Depth) == 1
729 List::Util::first { $op =~ $_->{regex} } @{$self->{special_ops}}
731 $self->{disable_old_special_ops}
732 and List::Util::first { $op =~ $_->{regex} } @BUILTIN_SPECIAL_OPS
736 puke "Illegal use of top-level '-$op'"
739 # the old unary op system means we should touch nothing and let it work
741 if (my $us = List::Util::first { $op =~ $_->{regex} } @{$self->{unary_ops}}) {
742 return { -op => [ $op, $v ] };
746 # an explicit node type is currently assumed to be expanded (this is almost
747 # certainly wrong and there should be expansion anyway)
749 if ($self->{render}{$op}) {
753 # hashref RHS values get expanded and used as op/func args
758 and (keys %$v)[0] =~ /^-/
760 my ($func) = $k =~ /^-(.*)$/;
762 if (List::Util::first { $func =~ $_->{regex} } @{$self->{special_ops}}) {
763 return +{ -op => [ $func, $self->_expand_expr($v) ] };
766 return +{ -func => [ $func, $self->_expand_expr($v) ] };
769 # scalars and literals get simply expanded
771 if (!ref($v) or is_literal_value($v)) {
772 return +{ -op => [ $op, $self->_expand_expr($v) ] };
778 sub _expand_hashpair_cmp {
779 my ($self, $k, $v) = @_;
780 $self->_expand_hashtriple($k, $self->{cmp}, $v);
783 sub _expand_hashtriple {
784 my ($self, $k, $vk, $vv) = @_;
786 my $ik = $self->_expand_ident(-ident => $k);
788 my $op = $self->_normalize_op($vk);
789 $self->_assert_pass_injection_guard($op);
791 if ($op =~ s/ _? \d+ $//x ) {
792 return $self->_expand_expr($k, { $vk, $vv });
794 if (my $x = $self->{expand_op}{$op}) {
795 local our $Cur_Col_Meta = $k;
796 return $self->$x($op, $vv, $k);
800 my $op = join(' ', split '_', $op);
802 if (my $us = List::Util::first { $op =~ $_->{regex} } @{$self->{special_ops}}) {
803 return { -op => [ $op, $ik, $vv ] };
805 if (my $us = List::Util::first { $op =~ $_->{regex} } @{$self->{unary_ops}}) {
809 { -op => [ $op, $vv ] }
813 if (ref($vv) eq 'ARRAY') {
815 my $logic = (defined($raw[0]) and $raw[0] =~ /^-(and|or)$/i)
816 ? (shift(@raw), $1) : 'or';
817 my @values = map +{ $vk => $_ }, @raw;
819 $op =~ $self->{inequality_op}
820 or $op =~ $self->{not_like_op}
822 if (lc($logic) eq 'or' and @values > 1) {
823 belch "A multi-element arrayref as an argument to the inequality op '${\uc(join ' ', split '_', $op)}' "
824 . 'is technically equivalent to an always-true 1=1 (you probably wanted '
825 . "to say ...{ \$inequality_op => [ -and => \@values ] }... instead)"
830 # try to DWIM on equality operators
831 return ($self->_dwim_op_to_is($op,
832 "Supplying an empty arrayref to '%s' is deprecated",
833 "operator '%s' applied on an empty array (field '$k')"
834 ) ? $self->sqlfalse : $self->sqltrue);
836 return $self->_expand_op_andor($logic => \@values, $k);
838 if (is_undef_value($vv)) {
839 my $is = ($self->_dwim_op_to_is($op,
840 "Supplying an undefined argument to '%s' is deprecated",
841 "unexpected operator '%s' with undef operand",
842 ) ? 'is' : 'is not');
844 return $self->_expand_hashpair($k => { $is, undef });
846 local our $Cur_Col_Meta = $k;
850 $self->_expand_expr($vv)
855 my ($self, $raw, $empty, $fail) = @_;
857 my $op = $self->_normalize_op($raw);
859 if ($op =~ /^not$/i) {
862 if ($op =~ $self->{equality_op}) {
865 if ($op =~ $self->{like_op}) {
866 belch(sprintf $empty, uc(join ' ', split '_', $op));
869 if ($op =~ $self->{inequality_op}) {
872 if ($op =~ $self->{not_like_op}) {
873 belch(sprintf $empty, uc(join ' ', split '_', $op));
876 puke(sprintf $fail, $op);
880 my ($self, undef, $body, $k) = @_;
881 return $self->_expand_hashpair_cmp(
882 $k, { -ident => $body }
884 unless (defined($body) or (ref($body) and ref($body) eq 'ARRAY')) {
885 puke "-ident requires a single plain scalar argument (a quotable identifier) or an arrayref of identifier parts";
887 my @parts = map split(/\Q${\($self->{name_sep}||'.')}\E/, $_),
888 ref($body) ? @$body : $body;
889 return { -ident => $parts[-1] } if $self->{_dequalify_idents};
890 unless ($self->{quote_char}) {
891 $self->_assert_pass_injection_guard($_) for @parts;
893 return +{ -ident => \@parts };
897 return $_[0]->_expand_hashpair_cmp(
898 $_[3], { -value => $_[2] },
900 +{ -bind => [ our $Cur_Col_Meta, $_[2] ] };
904 +{ -op => [ 'not', $_[0]->_expand_expr($_[2]) ] };
908 my ($self, undef, $args) = @_;
909 +{ -row => [ map $self->expand_expr($_), @$args ] };
913 my ($self, undef, $args) = @_;
914 my ($op, @opargs) = @$args;
915 +{ -op => [ $op, map $self->expand_expr($_), @opargs ] };
919 my ($self, undef, $v) = @_;
921 return $self->_expand_expr($v);
923 puke "-bool => undef not supported" unless defined($v);
924 return $self->_expand_ident(-ident => $v);
927 sub _expand_op_andor {
928 my ($self, $logop, $v, $k) = @_;
930 $v = [ map +{ $k, $_ },
932 ? (map +{ $_ => $v->{$_} }, sort keys %$v)
936 if (ref($v) eq 'HASH') {
937 return undef unless keys %$v;
940 map $self->_expand_expr({ $_ => $v->{$_} }),
944 if (ref($v) eq 'ARRAY') {
945 $logop eq 'and' or $logop eq 'or' or puke "unknown logic: $logop";
948 (ref($_) eq 'ARRAY' and @$_)
949 or (ref($_) eq 'HASH' and %$_)
955 while (my ($el) = splice @expr, 0, 1) {
956 puke "Supplying an empty left hand side argument is not supported in array-pairs"
957 unless defined($el) and length($el);
958 my $elref = ref($el);
960 local our $Expand_Depth = 0;
961 push(@res, grep defined, $self->_expand_expr({ $el, shift(@expr) }));
962 } elsif ($elref eq 'ARRAY') {
963 push(@res, grep defined, $self->_expand_expr($el)) if @$el;
964 } elsif (my $l = is_literal_value($el)) {
965 push @res, { -literal => $l };
966 } elsif ($elref eq 'HASH') {
967 local our $Expand_Depth = 0;
968 push @res, grep defined, $self->_expand_expr($el) if %$el;
974 # return $res[0] if @res == 1;
975 return { -op => [ $logop, @res ] };
981 my ($self, $op, $vv, $k) = @_;
982 ($k, $vv) = @$vv unless defined $k;
983 puke "$op can only take undef as argument"
987 and exists($vv->{-value})
988 and !defined($vv->{-value})
990 return +{ -op => [ $op.'_null', $self->expand_expr($k, -ident) ] };
993 sub _expand_between {
994 my ($self, $op, $vv, $k) = @_;
995 $k = shift @{$vv = [ @$vv ]} unless defined $k;
996 my @rhs = map $self->_expand_expr($_),
997 ref($vv) eq 'ARRAY' ? @$vv : $vv;
999 (@rhs == 1 and ref($rhs[0]) eq 'HASH' and $rhs[0]->{-literal})
1001 (@rhs == 2 and defined($rhs[0]) and defined($rhs[1]))
1003 puke "Operator '${\uc($op)}' requires either an arrayref with two defined values or expressions, or a single literal scalarref/arrayref-ref";
1007 $self->expand_expr(ref($k) ? $k : { -ident => $k }),
1013 my ($self, $op, $vv, $k) = @_;
1014 $k = shift @{$vv = [ @$vv ]} unless defined $k;
1015 if (my $literal = is_literal_value($vv)) {
1016 my ($sql, @bind) = @$literal;
1017 my $opened_sql = $self->_open_outer_paren($sql);
1019 $op, $self->expand_expr($k, -ident),
1020 [ { -literal => [ $opened_sql, @bind ] } ]
1024 'SQL::Abstract before v1.75 used to generate incorrect SQL when the '
1025 . "-${\uc($op)} operator was given an undef-containing list: !!!AUDIT YOUR CODE "
1026 . 'AND DATA!!! (the upcoming Data::Query-based version of SQL::Abstract '
1027 . 'will emit the logically correct SQL instead of raising this exception)'
1029 puke("Argument passed to the '${\uc($op)}' operator can not be undefined")
1031 my @rhs = map $self->expand_expr($_, -value),
1032 map { defined($_) ? $_: puke($undef_err) }
1033 (ref($vv) eq 'ARRAY' ? @$vv : $vv);
1034 return $self->${\($op =~ /^not/ ? 'sqltrue' : 'sqlfalse')} unless @rhs;
1038 $self->expand_expr($k, -ident),
1044 my ($self, undef, $v) = @_;
1045 # DBIx::Class requires a nest warning to be emitted once but the private
1046 # method it overrode to do so no longer exists
1047 if ($self->{warn_once_on_nest}) {
1048 unless (our $Nest_Warned) {
1050 "-nest in search conditions is deprecated, you most probably wanted:\n"
1051 .q|{..., -and => [ \%cond0, \@cond1, \'cond2', \[ 'cond3', [ col => bind ] ], etc. ], ... }|
1056 return $self->_expand_expr($v);
1060 my ($self, undef, $bind) = @_;
1061 return { -bind => $bind };
1064 sub _recurse_where {
1065 my ($self, $where, $logic) = @_;
1067 # Special case: top level simple string treated as literal
1069 my $where_exp = (ref($where)
1070 ? $self->_expand_expr($where, $logic)
1071 : { -literal => [ $where ] });
1073 # dispatch expanded expression
1075 my ($sql, @bind) = defined($where_exp) ? $self->render_aqt($where_exp) : (undef);
1076 # DBIx::Class used to call _recurse_where in scalar context
1077 # something else might too...
1079 return ($sql, @bind);
1082 belch "Calling _recurse_where in scalar context is deprecated and will go away before 2.0";
1088 my ($self, $ident) = @_;
1090 return $self->_convert($self->_quote($ident));
1094 my ($self, $values) = @_;
1095 my ($sql, @bind) = $self->_render_op([ ',', @$values ]);
1096 return "($sql)", @bind;
1100 my ($self, $rest) = @_;
1101 my ($func, @args) = @$rest;
1105 push @arg_sql, shift @x;
1107 } map [ $self->render_aqt($_) ], @args;
1108 return ($self->_sqlcase($func).'('.join(', ', @arg_sql).')', @bind);
1112 my ($self, $bind) = @_;
1113 return ($self->_convert('?'), $self->_bindtype(@$bind));
1116 sub _render_literal {
1117 my ($self, $literal) = @_;
1118 $self->_assert_bindval_matches_bindtype(@{$literal}[1..$#$literal]);
1123 my ($self, $v) = @_;
1124 my ($op, @args) = @$v;
1125 if (my $r = $self->{render_op}{$op}) {
1126 return $self->$r($op, \@args);
1131 my $op = join(' ', split '_', $op);
1133 my $us = List::Util::first { $op =~ $_->{regex} } @{$self->{special_ops}};
1134 if ($us and @args > 1) {
1135 puke "Special op '${op}' requires first value to be identifier"
1136 unless my ($ident) = map $_->{-ident}, grep ref($_) eq 'HASH', $args[0];
1137 my $k = join(($self->{name_sep}||'.'), @$ident);
1138 local our $Expand_Depth = 1;
1139 return $self->${\($us->{handler})}($k, $op, $args[1]);
1141 if (my $us = List::Util::first { $op =~ $_->{regex} } @{$self->{unary_ops}}) {
1142 return $self->${\($us->{handler})}($op, $args[0]);
1147 return $self->_render_unop_prefix($op, \@args);
1149 return $self->_render_op_multop($op, \@args);
1155 sub _render_op_between {
1156 my ($self, $op, $args) = @_;
1157 my ($left, $low, $high) = @$args;
1158 my ($rhsql, @rhbind) = do {
1160 puke "Single arg to between must be a literal"
1161 unless $low->{-literal};
1164 my ($l, $h) = map [ $self->render_aqt($_) ], $low, $high;
1165 (join(' ', $l->[0], $self->_sqlcase('and'), $h->[0]),
1166 @{$l}[1..$#$l], @{$h}[1..$#$h])
1169 my ($lhsql, @lhbind) = $self->render_aqt($left);
1173 $self->_sqlcase(join ' ', split '_', $op),
1181 my ($self, $op, $args) = @_;
1182 my ($lhs, $rhs) = @$args;
1185 my ($sql, @bind) = $self->render_aqt($_);
1186 push @in_bind, @bind;
1189 my ($lhsql, @lbind) = $self->render_aqt($lhs);
1191 $lhsql.' '.$self->_sqlcase(join ' ', split '_', $op).' ( '
1192 .join(', ', @in_sql)
1198 sub _render_op_andor {
1199 my ($self, $op, $args) = @_;
1200 my @parts = grep length($_->[0]), map [ $self->render_aqt($_) ], @$args;
1201 return '' unless @parts;
1202 return @{$parts[0]} if @parts == 1;
1203 my ($sql, @bind) = $self->_render_op_multop($op, $args);
1204 return '( '.$sql.' )', @bind;
1207 sub _render_op_multop {
1208 my ($self, $op, $args) = @_;
1209 my @parts = grep length($_->[0]), map [ $self->render_aqt($_) ], @$args;
1210 return '' unless @parts;
1211 return @{$parts[0]} if @parts == 1;
1212 my ($final_sql) = join(
1213 ($op eq ',' ? '' : ' ').$self->_sqlcase(join ' ', split '_', $op).' ',
1218 map @{$_}[1..$#$_], @parts
1221 sub _render_op_not {
1222 my ($self, $op, $v) = @_;
1223 my ($sql, @bind) = $self->_render_unop_prefix($op, $v);
1224 return "(${sql})", @bind;
1227 sub _render_unop_prefix {
1228 my ($self, $op, $v) = @_;
1229 my ($expr_sql, @bind) = $self->render_aqt($v->[0]);
1231 my $op_sql = $self->_sqlcase($op); # join ' ', split '_', $op);
1232 return ("${op_sql} ${expr_sql}", @bind);
1235 sub _render_unop_postfix {
1236 my ($self, $op, $v) = @_;
1237 my ($expr_sql, @bind) = $self->render_aqt($v->[0]);
1238 my $op_sql = $self->_sqlcase(join ' ', split '_', $op);
1239 return ($expr_sql.' '.$op_sql, @bind);
1242 # Some databases (SQLite) treat col IN (1, 2) different from
1243 # col IN ( (1, 2) ). Use this to strip all outer parens while
1244 # adding them back in the corresponding method
1245 sub _open_outer_paren {
1246 my ($self, $sql) = @_;
1248 while (my ($inner) = $sql =~ /^ \s* \( (.*) \) \s* $/xs) {
1250 # there are closing parens inside, need the heavy duty machinery
1251 # to reevaluate the extraction starting from $sql (full reevaluation)
1252 if ($inner =~ /\)/) {
1253 require Text::Balanced;
1255 my (undef, $remainder) = do {
1256 # idiotic design - writes to $@ but *DOES NOT* throw exceptions
1258 Text::Balanced::extract_bracketed($sql, '()', qr/\s*/);
1261 # the entire expression needs to be a balanced bracketed thing
1262 # (after an extract no remainder sans trailing space)
1263 last if defined $remainder and $remainder =~ /\S/;
1273 #======================================================================
1275 #======================================================================
1277 sub _expand_order_by {
1278 my ($self, $arg) = @_;
1280 return unless defined($arg) and not (ref($arg) eq 'ARRAY' and !@$arg);
1282 my $expander = sub {
1283 my ($self, $dir, $expr) = @_;
1284 my @to_expand = ref($expr) eq 'ARRAY' ? @$expr : $expr;
1285 foreach my $arg (@to_expand) {
1289 and grep /^-(asc|desc)$/, keys %$arg
1291 puke "ordering direction hash passed to order by must have exactly one key (-asc or -desc)";
1295 defined($dir) ? { -op => [ $dir =~ /^-?(.*)$/ ,=> $_ ] } : $_
1297 map $self->expand_expr($_, -ident),
1298 map ref($_) eq 'ARRAY' ? @$_ : $_, @to_expand;
1299 return undef unless @exp;
1300 return undef if @exp == 1 and not defined($exp[0]);
1301 return +{ -op => [ ',', @exp ] };
1304 local @{$self->{expand}}{qw(asc desc)} = (($expander) x 2);
1306 return $self->$expander(undef, $arg);
1310 my ($self, $arg) = @_;
1312 return '' unless defined(my $expanded = $self->_expand_order_by($arg));
1314 my ($sql, @bind) = $self->render_aqt($expanded);
1316 return '' unless length($sql);
1318 my $final_sql = $self->_sqlcase(' order by ').$sql;
1320 return wantarray ? ($final_sql, @bind) : $final_sql;
1323 # _order_by no longer needs to call this so doesn't but DBIC uses it.
1325 sub _order_by_chunks {
1326 my ($self, $arg) = @_;
1328 return () unless defined(my $expanded = $self->_expand_order_by($arg));
1330 return $self->_chunkify_order_by($expanded);
1333 sub _chunkify_order_by {
1334 my ($self, $expanded) = @_;
1336 return grep length, $self->render_aqt($expanded)
1337 if $expanded->{-ident} or @{$expanded->{-literal}||[]} == 1;
1340 if (ref() eq 'HASH' and $_->{-op} and $_->{-op}[0] eq ',') {
1341 my ($comma, @list) = @{$_->{-op}};
1342 return map $self->_chunkify_order_by($_), @list;
1344 return [ $self->render_aqt($_) ];
1348 #======================================================================
1349 # DATASOURCE (FOR NOW, JUST PLAIN TABLE OR LIST OF TABLES)
1350 #======================================================================
1356 $self->_expand_maybe_list_expr($from, -ident)
1361 #======================================================================
1363 #======================================================================
1365 sub _expand_maybe_list_expr {
1366 my ($self, $expr, $default) = @_;
1367 return +{ -op => [ ',',
1368 map $self->expand_expr($_, $default),
1369 ref($expr) eq 'ARRAY' ? @$expr : $expr
1373 # highly optimized, as it's called way too often
1375 # my ($self, $label) = @_;
1377 return '' unless defined $_[1];
1378 return ${$_[1]} if ref($_[1]) eq 'SCALAR';
1379 puke 'Identifier cannot be hashref' if ref($_[1]) eq 'HASH';
1381 unless ($_[0]->{quote_char}) {
1382 if (ref($_[1]) eq 'ARRAY') {
1383 return join($_[0]->{name_sep}||'.', @{$_[1]});
1385 $_[0]->_assert_pass_injection_guard($_[1]);
1390 my $qref = ref $_[0]->{quote_char};
1392 !$qref ? ($_[0]->{quote_char}, $_[0]->{quote_char})
1393 : ($qref eq 'ARRAY') ? @{$_[0]->{quote_char}}
1394 : puke "Unsupported quote_char format: $_[0]->{quote_char}";
1396 my $esc = $_[0]->{escape_char} || $r;
1398 # parts containing * are naturally unquoted
1400 $_[0]->{name_sep}||'',
1404 : do { (my $n = $_) =~ s/(\Q$esc\E|\Q$r\E)/$esc$1/g; $l . $n . $r }
1406 (ref($_[1]) eq 'ARRAY'
1410 ? split (/\Q$_[0]->{name_sep}\E/, $_[1] )
1418 # Conversion, if applicable
1420 #my ($self, $arg) = @_;
1421 if ($_[0]->{convert_where}) {
1422 return $_[0]->_sqlcase($_[0]->{convert_where}) .'(' . $_[1] . ')';
1429 #my ($self, $col, @vals) = @_;
1430 # called often - tighten code
1431 return $_[0]->{bindtype} eq 'columns'
1432 ? map {[$_[1], $_]} @_[2 .. $#_]
1437 # Dies if any element of @bind is not in [colname => value] format
1438 # if bindtype is 'columns'.
1439 sub _assert_bindval_matches_bindtype {
1440 # my ($self, @bind) = @_;
1442 if ($self->{bindtype} eq 'columns') {
1444 if (!defined $_ || ref($_) ne 'ARRAY' || @$_ != 2) {
1445 puke "bindtype 'columns' selected, you need to pass: [column_name => bind_value]"
1451 sub _join_sql_clauses {
1452 my ($self, $logic, $clauses_aref, $bind_aref) = @_;
1454 if (@$clauses_aref > 1) {
1455 my $join = " " . $self->_sqlcase($logic) . " ";
1456 my $sql = '( ' . join($join, @$clauses_aref) . ' )';
1457 return ($sql, @$bind_aref);
1459 elsif (@$clauses_aref) {
1460 return ($clauses_aref->[0], @$bind_aref); # no parentheses
1463 return (); # if no SQL, ignore @$bind_aref
1468 # Fix SQL case, if so requested
1470 # LDNOTE: if $self->{case} is true, then it contains 'lower', so we
1471 # don't touch the argument ... crooked logic, but let's not change it!
1472 return $_[0]->{case} ? $_[1] : uc($_[1]);
1476 #======================================================================
1477 # DISPATCHING FROM REFKIND
1478 #======================================================================
1481 my ($self, $data) = @_;
1483 return 'UNDEF' unless defined $data;
1485 # blessed objects are treated like scalars
1486 my $ref = (Scalar::Util::blessed $data) ? '' : ref $data;
1488 return 'SCALAR' unless $ref;
1491 while ($ref eq 'REF') {
1493 $ref = (Scalar::Util::blessed $data) ? '' : ref $data;
1497 return ($ref||'SCALAR') . ('REF' x $n_steps);
1501 my ($self, $data) = @_;
1502 my @try = ($self->_refkind($data));
1503 push @try, 'SCALAR_or_UNDEF' if $try[0] eq 'SCALAR' || $try[0] eq 'UNDEF';
1504 push @try, 'FALLBACK';
1508 sub _METHOD_FOR_refkind {
1509 my ($self, $meth_prefix, $data) = @_;
1512 for (@{$self->_try_refkind($data)}) {
1513 $method = $self->can($meth_prefix."_".$_)
1517 return $method || puke "cannot dispatch on '$meth_prefix' for ".$self->_refkind($data);
1521 sub _SWITCH_refkind {
1522 my ($self, $data, $dispatch_table) = @_;
1525 for (@{$self->_try_refkind($data)}) {
1526 $coderef = $dispatch_table->{$_}
1530 puke "no dispatch entry for ".$self->_refkind($data)
1539 #======================================================================
1540 # VALUES, GENERATE, AUTOLOAD
1541 #======================================================================
1543 # LDNOTE: original code from nwiger, didn't touch code in that section
1544 # I feel the AUTOLOAD stuff should not be the default, it should
1545 # only be activated on explicit demand by user.
1549 my $data = shift || return;
1550 puke "Argument to ", __PACKAGE__, "->values must be a \\%hash"
1551 unless ref $data eq 'HASH';
1554 foreach my $k (sort keys %$data) {
1555 my $v = $data->{$k};
1556 $self->_SWITCH_refkind($v, {
1558 if ($self->{array_datatypes}) { # array datatype
1559 push @all_bind, $self->_bindtype($k, $v);
1561 else { # literal SQL with bind
1562 my ($sql, @bind) = @$v;
1563 $self->_assert_bindval_matches_bindtype(@bind);
1564 push @all_bind, @bind;
1567 ARRAYREFREF => sub { # literal SQL with bind
1568 my ($sql, @bind) = @${$v};
1569 $self->_assert_bindval_matches_bindtype(@bind);
1570 push @all_bind, @bind;
1572 SCALARREF => sub { # literal SQL without bind
1574 SCALAR_or_UNDEF => sub {
1575 push @all_bind, $self->_bindtype($k, $v);
1586 my(@sql, @sqlq, @sqlv);
1590 if ($ref eq 'HASH') {
1591 for my $k (sort keys %$_) {
1594 my $label = $self->_quote($k);
1595 if ($r eq 'ARRAY') {
1596 # literal SQL with bind
1597 my ($sql, @bind) = @$v;
1598 $self->_assert_bindval_matches_bindtype(@bind);
1599 push @sqlq, "$label = $sql";
1601 } elsif ($r eq 'SCALAR') {
1602 # literal SQL without bind
1603 push @sqlq, "$label = $$v";
1605 push @sqlq, "$label = ?";
1606 push @sqlv, $self->_bindtype($k, $v);
1609 push @sql, $self->_sqlcase('set'), join ', ', @sqlq;
1610 } elsif ($ref eq 'ARRAY') {
1611 # unlike insert(), assume these are ONLY the column names, i.e. for SQL
1614 if ($r eq 'ARRAY') { # literal SQL with bind
1615 my ($sql, @bind) = @$v;
1616 $self->_assert_bindval_matches_bindtype(@bind);
1619 } elsif ($r eq 'SCALAR') { # literal SQL without bind
1620 # embedded literal SQL
1627 push @sql, '(' . join(', ', @sqlq) . ')';
1628 } elsif ($ref eq 'SCALAR') {
1632 # strings get case twiddled
1633 push @sql, $self->_sqlcase($_);
1637 my $sql = join ' ', @sql;
1639 # this is pretty tricky
1640 # if ask for an array, return ($stmt, @bind)
1641 # otherwise, s/?/shift @sqlv/ to put it inline
1643 return ($sql, @sqlv);
1645 1 while $sql =~ s/\?/my $d = shift(@sqlv);
1646 ref $d ? $d->[1] : $d/e;
1655 # This allows us to check for a local, then _form, attr
1657 my($name) = $AUTOLOAD =~ /.*::(.+)/;
1658 return $self->generate($name, @_);
1669 SQL::Abstract - Generate SQL from Perl data structures
1675 my $sql = SQL::Abstract->new;
1677 my($stmt, @bind) = $sql->select($source, \@fields, \%where, $order);
1679 my($stmt, @bind) = $sql->insert($table, \%fieldvals || \@values);
1681 my($stmt, @bind) = $sql->update($table, \%fieldvals, \%where);
1683 my($stmt, @bind) = $sql->delete($table, \%where);
1685 # Then, use these in your DBI statements
1686 my $sth = $dbh->prepare($stmt);
1687 $sth->execute(@bind);
1689 # Just generate the WHERE clause
1690 my($stmt, @bind) = $sql->where(\%where, $order);
1692 # Return values in the same order, for hashed queries
1693 # See PERFORMANCE section for more details
1694 my @bind = $sql->values(\%fieldvals);
1698 This module was inspired by the excellent L<DBIx::Abstract>.
1699 However, in using that module I found that what I really wanted
1700 to do was generate SQL, but still retain complete control over my
1701 statement handles and use the DBI interface. So, I set out to
1702 create an abstract SQL generation module.
1704 While based on the concepts used by L<DBIx::Abstract>, there are
1705 several important differences, especially when it comes to WHERE
1706 clauses. I have modified the concepts used to make the SQL easier
1707 to generate from Perl data structures and, IMO, more intuitive.
1708 The underlying idea is for this module to do what you mean, based
1709 on the data structures you provide it. The big advantage is that
1710 you don't have to modify your code every time your data changes,
1711 as this module figures it out.
1713 To begin with, an SQL INSERT is as easy as just specifying a hash
1714 of C<key=value> pairs:
1717 name => 'Jimbo Bobson',
1718 phone => '123-456-7890',
1719 address => '42 Sister Lane',
1720 city => 'St. Louis',
1721 state => 'Louisiana',
1724 The SQL can then be generated with this:
1726 my($stmt, @bind) = $sql->insert('people', \%data);
1728 Which would give you something like this:
1730 $stmt = "INSERT INTO people
1731 (address, city, name, phone, state)
1732 VALUES (?, ?, ?, ?, ?)";
1733 @bind = ('42 Sister Lane', 'St. Louis', 'Jimbo Bobson',
1734 '123-456-7890', 'Louisiana');
1736 These are then used directly in your DBI code:
1738 my $sth = $dbh->prepare($stmt);
1739 $sth->execute(@bind);
1741 =head2 Inserting and Updating Arrays
1743 If your database has array types (like for example Postgres),
1744 activate the special option C<< array_datatypes => 1 >>
1745 when creating the C<SQL::Abstract> object.
1746 Then you may use an arrayref to insert and update database array types:
1748 my $sql = SQL::Abstract->new(array_datatypes => 1);
1750 planets => [qw/Mercury Venus Earth Mars/]
1753 my($stmt, @bind) = $sql->insert('solar_system', \%data);
1757 $stmt = "INSERT INTO solar_system (planets) VALUES (?)"
1759 @bind = (['Mercury', 'Venus', 'Earth', 'Mars']);
1762 =head2 Inserting and Updating SQL
1764 In order to apply SQL functions to elements of your C<%data> you may
1765 specify a reference to an arrayref for the given hash value. For example,
1766 if you need to execute the Oracle C<to_date> function on a value, you can
1767 say something like this:
1771 date_entered => \[ "to_date(?,'MM/DD/YYYY')", "03/02/2003" ],
1774 The first value in the array is the actual SQL. Any other values are
1775 optional and would be included in the bind values array. This gives
1778 my($stmt, @bind) = $sql->insert('people', \%data);
1780 $stmt = "INSERT INTO people (name, date_entered)
1781 VALUES (?, to_date(?,'MM/DD/YYYY'))";
1782 @bind = ('Bill', '03/02/2003');
1784 An UPDATE is just as easy, all you change is the name of the function:
1786 my($stmt, @bind) = $sql->update('people', \%data);
1788 Notice that your C<%data> isn't touched; the module will generate
1789 the appropriately quirky SQL for you automatically. Usually you'll
1790 want to specify a WHERE clause for your UPDATE, though, which is
1791 where handling C<%where> hashes comes in handy...
1793 =head2 Complex where statements
1795 This module can generate pretty complicated WHERE statements
1796 easily. For example, simple C<key=value> pairs are taken to mean
1797 equality, and if you want to see if a field is within a set
1798 of values, you can use an arrayref. Let's say we wanted to
1799 SELECT some data based on this criteria:
1802 requestor => 'inna',
1803 worker => ['nwiger', 'rcwe', 'sfz'],
1804 status => { '!=', 'completed' }
1807 my($stmt, @bind) = $sql->select('tickets', '*', \%where);
1809 The above would give you something like this:
1811 $stmt = "SELECT * FROM tickets WHERE
1812 ( requestor = ? ) AND ( status != ? )
1813 AND ( worker = ? OR worker = ? OR worker = ? )";
1814 @bind = ('inna', 'completed', 'nwiger', 'rcwe', 'sfz');
1816 Which you could then use in DBI code like so:
1818 my $sth = $dbh->prepare($stmt);
1819 $sth->execute(@bind);
1825 The methods are simple. There's one for every major SQL operation,
1826 and a constructor you use first. The arguments are specified in a
1827 similar order for each method (table, then fields, then a where
1828 clause) to try and simplify things.
1830 =head2 new(option => 'value')
1832 The C<new()> function takes a list of options and values, and returns
1833 a new B<SQL::Abstract> object which can then be used to generate SQL
1834 through the methods below. The options accepted are:
1840 If set to 'lower', then SQL will be generated in all lowercase. By
1841 default SQL is generated in "textbook" case meaning something like:
1843 SELECT a_field FROM a_table WHERE some_field LIKE '%someval%'
1845 Any setting other than 'lower' is ignored.
1849 This determines what the default comparison operator is. By default
1850 it is C<=>, meaning that a hash like this:
1852 %where = (name => 'nwiger', email => 'nate@wiger.org');
1854 Will generate SQL like this:
1856 WHERE name = 'nwiger' AND email = 'nate@wiger.org'
1858 However, you may want loose comparisons by default, so if you set
1859 C<cmp> to C<like> you would get SQL such as:
1861 WHERE name like 'nwiger' AND email like 'nate@wiger.org'
1863 You can also override the comparison on an individual basis - see
1864 the huge section on L</"WHERE CLAUSES"> at the bottom.
1866 =item sqltrue, sqlfalse
1868 Expressions for inserting boolean values within SQL statements.
1869 By default these are C<1=1> and C<1=0>. They are used
1870 by the special operators C<-in> and C<-not_in> for generating
1871 correct SQL even when the argument is an empty array (see below).
1875 This determines the default logical operator for multiple WHERE
1876 statements in arrays or hashes. If absent, the default logic is "or"
1877 for arrays, and "and" for hashes. This means that a WHERE
1881 event_date => {'>=', '2/13/99'},
1882 event_date => {'<=', '4/24/03'},
1885 will generate SQL like this:
1887 WHERE event_date >= '2/13/99' OR event_date <= '4/24/03'
1889 This is probably not what you want given this query, though (look
1890 at the dates). To change the "OR" to an "AND", simply specify:
1892 my $sql = SQL::Abstract->new(logic => 'and');
1894 Which will change the above C<WHERE> to:
1896 WHERE event_date >= '2/13/99' AND event_date <= '4/24/03'
1898 The logic can also be changed locally by inserting
1899 a modifier in front of an arrayref:
1901 @where = (-and => [event_date => {'>=', '2/13/99'},
1902 event_date => {'<=', '4/24/03'} ]);
1904 See the L</"WHERE CLAUSES"> section for explanations.
1908 This will automatically convert comparisons using the specified SQL
1909 function for both column and value. This is mostly used with an argument
1910 of C<upper> or C<lower>, so that the SQL will have the effect of
1911 case-insensitive "searches". For example, this:
1913 $sql = SQL::Abstract->new(convert => 'upper');
1914 %where = (keywords => 'MaKe iT CAse inSeNSItive');
1916 Will turn out the following SQL:
1918 WHERE upper(keywords) like upper('MaKe iT CAse inSeNSItive')
1920 The conversion can be C<upper()>, C<lower()>, or any other SQL function
1921 that can be applied symmetrically to fields (actually B<SQL::Abstract> does
1922 not validate this option; it will just pass through what you specify verbatim).
1926 This is a kludge because many databases suck. For example, you can't
1927 just bind values using DBI's C<execute()> for Oracle C<CLOB> or C<BLOB> fields.
1928 Instead, you have to use C<bind_param()>:
1930 $sth->bind_param(1, 'reg data');
1931 $sth->bind_param(2, $lots, {ora_type => ORA_CLOB});
1933 The problem is, B<SQL::Abstract> will normally just return a C<@bind> array,
1934 which loses track of which field each slot refers to. Fear not.
1936 If you specify C<bindtype> in new, you can determine how C<@bind> is returned.
1937 Currently, you can specify either C<normal> (default) or C<columns>. If you
1938 specify C<columns>, you will get an array that looks like this:
1940 my $sql = SQL::Abstract->new(bindtype => 'columns');
1941 my($stmt, @bind) = $sql->insert(...);
1944 [ 'column1', 'value1' ],
1945 [ 'column2', 'value2' ],
1946 [ 'column3', 'value3' ],
1949 You can then iterate through this manually, using DBI's C<bind_param()>.
1951 $sth->prepare($stmt);
1954 my($col, $data) = @$_;
1955 if ($col eq 'details' || $col eq 'comments') {
1956 $sth->bind_param($i, $data, {ora_type => ORA_CLOB});
1957 } elsif ($col eq 'image') {
1958 $sth->bind_param($i, $data, {ora_type => ORA_BLOB});
1960 $sth->bind_param($i, $data);
1964 $sth->execute; # execute without @bind now
1966 Now, why would you still use B<SQL::Abstract> if you have to do this crap?
1967 Basically, the advantage is still that you don't have to care which fields
1968 are or are not included. You could wrap that above C<for> loop in a simple
1969 sub called C<bind_fields()> or something and reuse it repeatedly. You still
1970 get a layer of abstraction over manual SQL specification.
1972 Note that if you set L</bindtype> to C<columns>, the C<\[ $sql, @bind ]>
1973 construct (see L</Literal SQL with placeholders and bind values (subqueries)>)
1974 will expect the bind values in this format.
1978 This is the character that a table or column name will be quoted
1979 with. By default this is an empty string, but you could set it to
1980 the character C<`>, to generate SQL like this:
1982 SELECT `a_field` FROM `a_table` WHERE `some_field` LIKE '%someval%'
1984 Alternatively, you can supply an array ref of two items, the first being the left
1985 hand quote character, and the second the right hand quote character. For
1986 example, you could supply C<['[',']']> for SQL Server 2000 compliant quotes
1987 that generates SQL like this:
1989 SELECT [a_field] FROM [a_table] WHERE [some_field] LIKE '%someval%'
1991 Quoting is useful if you have tables or columns names that are reserved
1992 words in your database's SQL dialect.
1996 This is the character that will be used to escape L</quote_char>s appearing
1997 in an identifier before it has been quoted.
1999 The parameter default in case of a single L</quote_char> character is the quote
2002 When opening-closing-style quoting is used (L</quote_char> is an arrayref)
2003 this parameter defaults to the B<closing (right)> L</quote_char>. Occurrences
2004 of the B<opening (left)> L</quote_char> within the identifier are currently left
2005 untouched. The default for opening-closing-style quotes may change in future
2006 versions, thus you are B<strongly encouraged> to specify the escape character
2011 This is the character that separates a table and column name. It is
2012 necessary to specify this when the C<quote_char> option is selected,
2013 so that tables and column names can be individually quoted like this:
2015 SELECT `table`.`one_field` FROM `table` WHERE `table`.`other_field` = 1
2017 =item injection_guard
2019 A regular expression C<qr/.../> that is applied to any C<-function> and unquoted
2020 column name specified in a query structure. This is a safety mechanism to avoid
2021 injection attacks when mishandling user input e.g.:
2023 my %condition_as_column_value_pairs = get_values_from_user();
2024 $sqla->select( ... , \%condition_as_column_value_pairs );
2026 If the expression matches an exception is thrown. Note that literal SQL
2027 supplied via C<\'...'> or C<\['...']> is B<not> checked in any way.
2029 Defaults to checking for C<;> and the C<GO> keyword (TransactSQL)
2031 =item array_datatypes
2033 When this option is true, arrayrefs in INSERT or UPDATE are
2034 interpreted as array datatypes and are passed directly
2036 When this option is false, arrayrefs are interpreted
2037 as literal SQL, just like refs to arrayrefs
2038 (but this behavior is for backwards compatibility; when writing
2039 new queries, use the "reference to arrayref" syntax
2045 Takes a reference to a list of "special operators"
2046 to extend the syntax understood by L<SQL::Abstract>.
2047 See section L</"SPECIAL OPERATORS"> for details.
2051 Takes a reference to a list of "unary operators"
2052 to extend the syntax understood by L<SQL::Abstract>.
2053 See section L</"UNARY OPERATORS"> for details.
2059 =head2 insert($table, \@values || \%fieldvals, \%options)
2061 This is the simplest function. You simply give it a table name
2062 and either an arrayref of values or hashref of field/value pairs.
2063 It returns an SQL INSERT statement and a list of bind values.
2064 See the sections on L</"Inserting and Updating Arrays"> and
2065 L</"Inserting and Updating SQL"> for information on how to insert
2066 with those data types.
2068 The optional C<\%options> hash reference may contain additional
2069 options to generate the insert SQL. Currently supported options
2076 Takes either a scalar of raw SQL fields, or an array reference of
2077 field names, and adds on an SQL C<RETURNING> statement at the end.
2078 This allows you to return data generated by the insert statement
2079 (such as row IDs) without performing another C<SELECT> statement.
2080 Note, however, this is not part of the SQL standard and may not
2081 be supported by all database engines.
2085 =head2 update($table, \%fieldvals, \%where, \%options)
2087 This takes a table, hashref of field/value pairs, and an optional
2088 hashref L<WHERE clause|/WHERE CLAUSES>. It returns an SQL UPDATE function and a list
2090 See the sections on L</"Inserting and Updating Arrays"> and
2091 L</"Inserting and Updating SQL"> for information on how to insert
2092 with those data types.
2094 The optional C<\%options> hash reference may contain additional
2095 options to generate the update SQL. Currently supported options
2102 See the C<returning> option to
2103 L<insert|/insert($table, \@values || \%fieldvals, \%options)>.
2107 =head2 select($source, $fields, $where, $order)
2109 This returns a SQL SELECT statement and associated list of bind values, as
2110 specified by the arguments:
2116 Specification of the 'FROM' part of the statement.
2117 The argument can be either a plain scalar (interpreted as a table
2118 name, will be quoted), or an arrayref (interpreted as a list
2119 of table names, joined by commas, quoted), or a scalarref
2120 (literal SQL, not quoted).
2124 Specification of the list of fields to retrieve from
2126 The argument can be either an arrayref (interpreted as a list
2127 of field names, will be joined by commas and quoted), or a
2128 plain scalar (literal SQL, not quoted).
2129 Please observe that this API is not as flexible as that of
2130 the first argument C<$source>, for backwards compatibility reasons.
2134 Optional argument to specify the WHERE part of the query.
2135 The argument is most often a hashref, but can also be
2136 an arrayref or plain scalar --
2137 see section L<WHERE clause|/"WHERE CLAUSES"> for details.
2141 Optional argument to specify the ORDER BY part of the query.
2142 The argument can be a scalar, a hashref or an arrayref
2143 -- see section L<ORDER BY clause|/"ORDER BY CLAUSES">
2149 =head2 delete($table, \%where, \%options)
2151 This takes a table name and optional hashref L<WHERE clause|/WHERE CLAUSES>.
2152 It returns an SQL DELETE statement and list of bind values.
2154 The optional C<\%options> hash reference may contain additional
2155 options to generate the delete SQL. Currently supported options
2162 See the C<returning> option to
2163 L<insert|/insert($table, \@values || \%fieldvals, \%options)>.
2167 =head2 where(\%where, $order)
2169 This is used to generate just the WHERE clause. For example,
2170 if you have an arbitrary data structure and know what the
2171 rest of your SQL is going to look like, but want an easy way
2172 to produce a WHERE clause, use this. It returns an SQL WHERE
2173 clause and list of bind values.
2176 =head2 values(\%data)
2178 This just returns the values from the hash C<%data>, in the same
2179 order that would be returned from any of the other above queries.
2180 Using this allows you to markedly speed up your queries if you
2181 are affecting lots of rows. See below under the L</"PERFORMANCE"> section.
2183 =head2 generate($any, 'number', $of, \@data, $struct, \%types)
2185 Warning: This is an experimental method and subject to change.
2187 This returns arbitrarily generated SQL. It's a really basic shortcut.
2188 It will return two different things, depending on return context:
2190 my($stmt, @bind) = $sql->generate('create table', \$table, \@fields);
2191 my $stmt_and_val = $sql->generate('create table', \$table, \@fields);
2193 These would return the following:
2195 # First calling form
2196 $stmt = "CREATE TABLE test (?, ?)";
2197 @bind = (field1, field2);
2199 # Second calling form
2200 $stmt_and_val = "CREATE TABLE test (field1, field2)";
2202 Depending on what you're trying to do, it's up to you to choose the correct
2203 format. In this example, the second form is what you would want.
2207 $sql->generate('alter session', { nls_date_format => 'MM/YY' });
2211 ALTER SESSION SET nls_date_format = 'MM/YY'
2213 You get the idea. Strings get their case twiddled, but everything
2214 else remains verbatim.
2216 =head1 EXPORTABLE FUNCTIONS
2218 =head2 is_plain_value
2220 Determines if the supplied argument is a plain value as understood by this
2225 =item * The value is C<undef>
2227 =item * The value is a non-reference
2229 =item * The value is an object with stringification overloading
2231 =item * The value is of the form C<< { -value => $anything } >>
2235 On failure returns C<undef>, on success returns a B<scalar> reference
2236 to the original supplied argument.
2242 The stringification overloading detection is rather advanced: it takes
2243 into consideration not only the presence of a C<""> overload, but if that
2244 fails also checks for enabled
2245 L<autogenerated versions of C<"">|overload/Magic Autogeneration>, based
2246 on either C<0+> or C<bool>.
2248 Unfortunately testing in the field indicates that this
2249 detection B<< may tickle a latent bug in perl versions before 5.018 >>,
2250 but only when very large numbers of stringifying objects are involved.
2251 At the time of writing ( Sep 2014 ) there is no clear explanation of
2252 the direct cause, nor is there a manageably small test case that reliably
2253 reproduces the problem.
2255 If you encounter any of the following exceptions in B<random places within
2256 your application stack> - this module may be to blame:
2258 Operation "ne": no method found,
2259 left argument in overloaded package <something>,
2260 right argument in overloaded package <something>
2264 Stub found while resolving method "???" overloading """" in package <something>
2266 If you fall victim to the above - please attempt to reduce the problem
2267 to something that could be sent to the L<SQL::Abstract developers
2268 |DBIx::Class/GETTING HELP/SUPPORT>
2269 (either publicly or privately). As a workaround in the meantime you can
2270 set C<$ENV{SQLA_ISVALUE_IGNORE_AUTOGENERATED_STRINGIFICATION}> to a true
2271 value, which will most likely eliminate your problem (at the expense of
2272 not being able to properly detect exotic forms of stringification).
2274 This notice and environment variable will be removed in a future version,
2275 as soon as the underlying problem is found and a reliable workaround is
2280 =head2 is_literal_value
2282 Determines if the supplied argument is a literal value as understood by this
2287 =item * C<\$sql_string>
2289 =item * C<\[ $sql_string, @bind_values ]>
2293 On failure returns C<undef>, on success returns an B<array> reference
2294 containing the unpacked version of the supplied literal SQL and bind values.
2296 =head1 WHERE CLAUSES
2300 This module uses a variation on the idea from L<DBIx::Abstract>. It
2301 is B<NOT>, repeat I<not> 100% compatible. B<The main logic of this
2302 module is that things in arrays are OR'ed, and things in hashes
2305 The easiest way to explain is to show lots of examples. After
2306 each C<%where> hash shown, it is assumed you used:
2308 my($stmt, @bind) = $sql->where(\%where);
2310 However, note that the C<%where> hash can be used directly in any
2311 of the other functions as well, as described above.
2313 =head2 Key-value pairs
2315 So, let's get started. To begin, a simple hash:
2319 status => 'completed'
2322 Is converted to SQL C<key = val> statements:
2324 $stmt = "WHERE user = ? AND status = ?";
2325 @bind = ('nwiger', 'completed');
2327 One common thing I end up doing is having a list of values that
2328 a field can be in. To do this, simply specify a list inside of
2333 status => ['assigned', 'in-progress', 'pending'];
2336 This simple code will create the following:
2338 $stmt = "WHERE user = ? AND ( status = ? OR status = ? OR status = ? )";
2339 @bind = ('nwiger', 'assigned', 'in-progress', 'pending');
2341 A field associated to an empty arrayref will be considered a
2342 logical false and will generate 0=1.
2344 =head2 Tests for NULL values
2346 If the value part is C<undef> then this is converted to SQL <IS NULL>
2355 $stmt = "WHERE user = ? AND status IS NULL";
2358 To test if a column IS NOT NULL:
2362 status => { '!=', undef },
2365 =head2 Specific comparison operators
2367 If you want to specify a different type of operator for your comparison,
2368 you can use a hashref for a given column:
2372 status => { '!=', 'completed' }
2375 Which would generate:
2377 $stmt = "WHERE user = ? AND status != ?";
2378 @bind = ('nwiger', 'completed');
2380 To test against multiple values, just enclose the values in an arrayref:
2382 status => { '=', ['assigned', 'in-progress', 'pending'] };
2384 Which would give you:
2386 "WHERE status = ? OR status = ? OR status = ?"
2389 The hashref can also contain multiple pairs, in which case it is expanded
2390 into an C<AND> of its elements:
2394 status => { '!=', 'completed', -not_like => 'pending%' }
2397 # Or more dynamically, like from a form
2398 $where{user} = 'nwiger';
2399 $where{status}{'!='} = 'completed';
2400 $where{status}{'-not_like'} = 'pending%';
2402 # Both generate this
2403 $stmt = "WHERE user = ? AND status != ? AND status NOT LIKE ?";
2404 @bind = ('nwiger', 'completed', 'pending%');
2407 To get an OR instead, you can combine it with the arrayref idea:
2411 priority => [ { '=', 2 }, { '>', 5 } ]
2414 Which would generate:
2416 $stmt = "WHERE ( priority = ? OR priority > ? ) AND user = ?";
2417 @bind = ('2', '5', 'nwiger');
2419 If you want to include literal SQL (with or without bind values), just use a
2420 scalar reference or reference to an arrayref as the value:
2423 date_entered => { '>' => \["to_date(?, 'MM/DD/YYYY')", "11/26/2008"] },
2424 date_expires => { '<' => \"now()" }
2427 Which would generate:
2429 $stmt = "WHERE date_entered > to_date(?, 'MM/DD/YYYY') AND date_expires < now()";
2430 @bind = ('11/26/2008');
2433 =head2 Logic and nesting operators
2435 In the example above,
2436 there is a subtle trap if you want to say something like
2437 this (notice the C<AND>):
2439 WHERE priority != ? AND priority != ?
2441 Because, in Perl you I<can't> do this:
2443 priority => { '!=' => 2, '!=' => 1 }
2445 As the second C<!=> key will obliterate the first. The solution
2446 is to use the special C<-modifier> form inside an arrayref:
2448 priority => [ -and => {'!=', 2},
2452 Normally, these would be joined by C<OR>, but the modifier tells it
2453 to use C<AND> instead. (Hint: You can use this in conjunction with the
2454 C<logic> option to C<new()> in order to change the way your queries
2455 work by default.) B<Important:> Note that the C<-modifier> goes
2456 B<INSIDE> the arrayref, as an extra first element. This will
2457 B<NOT> do what you think it might:
2459 priority => -and => [{'!=', 2}, {'!=', 1}] # WRONG!
2461 Here is a quick list of equivalencies, since there is some overlap:
2464 status => {'!=', 'completed', 'not like', 'pending%' }
2465 status => [ -and => {'!=', 'completed'}, {'not like', 'pending%'}]
2468 status => {'=', ['assigned', 'in-progress']}
2469 status => [ -or => {'=', 'assigned'}, {'=', 'in-progress'}]
2470 status => [ {'=', 'assigned'}, {'=', 'in-progress'} ]
2474 =head2 Special operators: IN, BETWEEN, etc.
2476 You can also use the hashref format to compare a list of fields using the
2477 C<IN> comparison operator, by specifying the list as an arrayref:
2480 status => 'completed',
2481 reportid => { -in => [567, 2335, 2] }
2484 Which would generate:
2486 $stmt = "WHERE status = ? AND reportid IN (?,?,?)";
2487 @bind = ('completed', '567', '2335', '2');
2489 The reverse operator C<-not_in> generates SQL C<NOT IN> and is used in
2492 If the argument to C<-in> is an empty array, 'sqlfalse' is generated
2493 (by default: C<1=0>). Similarly, C<< -not_in => [] >> generates
2494 'sqltrue' (by default: C<1=1>).
2496 In addition to the array you can supply a chunk of literal sql or
2497 literal sql with bind:
2500 customer => { -in => \[
2501 'SELECT cust_id FROM cust WHERE balance > ?',
2504 status => { -in => \'SELECT status_codes FROM states' },
2510 customer IN ( SELECT cust_id FROM cust WHERE balance > ? )
2511 AND status IN ( SELECT status_codes FROM states )
2515 Finally, if the argument to C<-in> is not a reference, it will be
2516 treated as a single-element array.
2518 Another pair of operators is C<-between> and C<-not_between>,
2519 used with an arrayref of two values:
2523 completion_date => {
2524 -not_between => ['2002-10-01', '2003-02-06']
2530 WHERE user = ? AND completion_date NOT BETWEEN ( ? AND ? )
2532 Just like with C<-in> all plausible combinations of literal SQL
2536 start0 => { -between => [ 1, 2 ] },
2537 start1 => { -between => \["? AND ?", 1, 2] },
2538 start2 => { -between => \"lower(x) AND upper(y)" },
2539 start3 => { -between => [
2541 \["upper(?)", 'stuff' ],
2548 ( start0 BETWEEN ? AND ? )
2549 AND ( start1 BETWEEN ? AND ? )
2550 AND ( start2 BETWEEN lower(x) AND upper(y) )
2551 AND ( start3 BETWEEN lower(x) AND upper(?) )
2553 @bind = (1, 2, 1, 2, 'stuff');
2556 These are the two builtin "special operators"; but the
2557 list can be expanded: see section L</"SPECIAL OPERATORS"> below.
2559 =head2 Unary operators: bool
2561 If you wish to test against boolean columns or functions within your
2562 database you can use the C<-bool> and C<-not_bool> operators. For
2563 example to test the column C<is_user> being true and the column
2564 C<is_enabled> being false you would use:-
2568 -not_bool => 'is_enabled',
2573 WHERE is_user AND NOT is_enabled
2575 If a more complex combination is required, testing more conditions,
2576 then you should use the and/or operators:-
2581 -not_bool => { two=> { -rlike => 'bar' } },
2582 -not_bool => { three => [ { '=', 2 }, { '>', 5 } ] },
2593 (NOT ( three = ? OR three > ? ))
2596 =head2 Nested conditions, -and/-or prefixes
2598 So far, we've seen how multiple conditions are joined with a top-level
2599 C<AND>. We can change this by putting the different conditions we want in
2600 hashes and then putting those hashes in an array. For example:
2605 status => { -like => ['pending%', 'dispatched'] },
2609 status => 'unassigned',
2613 This data structure would create the following:
2615 $stmt = "WHERE ( user = ? AND ( status LIKE ? OR status LIKE ? ) )
2616 OR ( user = ? AND status = ? ) )";
2617 @bind = ('nwiger', 'pending', 'dispatched', 'robot', 'unassigned');
2620 Clauses in hashrefs or arrayrefs can be prefixed with an C<-and> or C<-or>
2621 to change the logic inside:
2627 -and => [ workhrs => {'>', 20}, geo => 'ASIA' ],
2628 -or => { workhrs => {'<', 50}, geo => 'EURO' },
2635 $stmt = "WHERE ( user = ?
2636 AND ( ( workhrs > ? AND geo = ? )
2637 OR ( workhrs < ? OR geo = ? ) ) )";
2638 @bind = ('nwiger', '20', 'ASIA', '50', 'EURO');
2640 =head3 Algebraic inconsistency, for historical reasons
2642 C<Important note>: when connecting several conditions, the C<-and->|C<-or>
2643 operator goes C<outside> of the nested structure; whereas when connecting
2644 several constraints on one column, the C<-and> operator goes
2645 C<inside> the arrayref. Here is an example combining both features:
2648 -and => [a => 1, b => 2],
2649 -or => [c => 3, d => 4],
2650 e => [-and => {-like => 'foo%'}, {-like => '%bar'} ]
2655 WHERE ( ( ( a = ? AND b = ? )
2656 OR ( c = ? OR d = ? )
2657 OR ( e LIKE ? AND e LIKE ? ) ) )
2659 This difference in syntax is unfortunate but must be preserved for
2660 historical reasons. So be careful: the two examples below would
2661 seem algebraically equivalent, but they are not
2664 { -like => 'foo%' },
2665 { -like => '%bar' },
2667 # yields: WHERE ( ( col LIKE ? AND col LIKE ? ) )
2670 { col => { -like => 'foo%' } },
2671 { col => { -like => '%bar' } },
2673 # yields: WHERE ( ( col LIKE ? OR col LIKE ? ) )
2676 =head2 Literal SQL and value type operators
2678 The basic premise of SQL::Abstract is that in WHERE specifications the "left
2679 side" is a column name and the "right side" is a value (normally rendered as
2680 a placeholder). This holds true for both hashrefs and arrayref pairs as you
2681 see in the L</WHERE CLAUSES> examples above. Sometimes it is necessary to
2682 alter this behavior. There are several ways of doing so.
2686 This is a virtual operator that signals the string to its right side is an
2687 identifier (a column name) and not a value. For example to compare two
2688 columns you would write:
2691 priority => { '<', 2 },
2692 requestor => { -ident => 'submitter' },
2697 $stmt = "WHERE priority < ? AND requestor = submitter";
2700 If you are maintaining legacy code you may see a different construct as
2701 described in L</Deprecated usage of Literal SQL>, please use C<-ident> in new
2706 This is a virtual operator that signals that the construct to its right side
2707 is a value to be passed to DBI. This is for example necessary when you want
2708 to write a where clause against an array (for RDBMS that support such
2709 datatypes). For example:
2712 array => { -value => [1, 2, 3] }
2717 $stmt = 'WHERE array = ?';
2718 @bind = ([1, 2, 3]);
2720 Note that if you were to simply say:
2726 the result would probably not be what you wanted:
2728 $stmt = 'WHERE array = ? OR array = ? OR array = ?';
2733 Finally, sometimes only literal SQL will do. To include a random snippet
2734 of SQL verbatim, you specify it as a scalar reference. Consider this only
2735 as a last resort. Usually there is a better way. For example:
2738 priority => { '<', 2 },
2739 requestor => { -in => \'(SELECT name FROM hitmen)' },
2744 $stmt = "WHERE priority < ? AND requestor IN (SELECT name FROM hitmen)"
2747 Note that in this example, you only get one bind parameter back, since
2748 the verbatim SQL is passed as part of the statement.
2752 Never use untrusted input as a literal SQL argument - this is a massive
2753 security risk (there is no way to check literal snippets for SQL
2754 injections and other nastyness). If you need to deal with untrusted input
2755 use literal SQL with placeholders as described next.
2757 =head3 Literal SQL with placeholders and bind values (subqueries)
2759 If the literal SQL to be inserted has placeholders and bind values,
2760 use a reference to an arrayref (yes this is a double reference --
2761 not so common, but perfectly legal Perl). For example, to find a date
2762 in Postgres you can use something like this:
2765 date_column => \[ "= date '2008-09-30' - ?::integer", 10 ]
2770 $stmt = "WHERE ( date_column = date '2008-09-30' - ?::integer )"
2773 Note that you must pass the bind values in the same format as they are returned
2774 by L<where|/where(\%where, $order)>. This means that if you set L</bindtype>
2775 to C<columns>, you must provide the bind values in the
2776 C<< [ column_meta => value ] >> format, where C<column_meta> is an opaque
2777 scalar value; most commonly the column name, but you can use any scalar value
2778 (including references and blessed references), L<SQL::Abstract> will simply
2779 pass it through intact. So if C<bindtype> is set to C<columns> the above
2780 example will look like:
2783 date_column => \[ "= date '2008-09-30' - ?::integer", [ {} => 10 ] ]
2786 Literal SQL is especially useful for nesting parenthesized clauses in the
2787 main SQL query. Here is a first example:
2789 my ($sub_stmt, @sub_bind) = ("SELECT c1 FROM t1 WHERE c2 < ? AND c3 LIKE ?",
2793 bar => \["IN ($sub_stmt)" => @sub_bind],
2798 $stmt = "WHERE (foo = ? AND bar IN (SELECT c1 FROM t1
2799 WHERE c2 < ? AND c3 LIKE ?))";
2800 @bind = (1234, 100, "foo%");
2802 Other subquery operators, like for example C<"E<gt> ALL"> or C<"NOT IN">,
2803 are expressed in the same way. Of course the C<$sub_stmt> and
2804 its associated bind values can be generated through a former call
2807 my ($sub_stmt, @sub_bind)
2808 = $sql->select("t1", "c1", {c2 => {"<" => 100},
2809 c3 => {-like => "foo%"}});
2812 bar => \["> ALL ($sub_stmt)" => @sub_bind],
2815 In the examples above, the subquery was used as an operator on a column;
2816 but the same principle also applies for a clause within the main C<%where>
2817 hash, like an EXISTS subquery:
2819 my ($sub_stmt, @sub_bind)
2820 = $sql->select("t1", "*", {c1 => 1, c2 => \"> t0.c0"});
2821 my %where = ( -and => [
2823 \["EXISTS ($sub_stmt)" => @sub_bind],
2828 $stmt = "WHERE (foo = ? AND EXISTS (SELECT * FROM t1
2829 WHERE c1 = ? AND c2 > t0.c0))";
2833 Observe that the condition on C<c2> in the subquery refers to
2834 column C<t0.c0> of the main query: this is I<not> a bind
2835 value, so we have to express it through a scalar ref.
2836 Writing C<< c2 => {">" => "t0.c0"} >> would have generated
2837 C<< c2 > ? >> with bind value C<"t0.c0"> ... not exactly
2838 what we wanted here.
2840 Finally, here is an example where a subquery is used
2841 for expressing unary negation:
2843 my ($sub_stmt, @sub_bind)
2844 = $sql->where({age => [{"<" => 10}, {">" => 20}]});
2845 $sub_stmt =~ s/^ where //i; # don't want "WHERE" in the subclause
2847 lname => {like => '%son%'},
2848 \["NOT ($sub_stmt)" => @sub_bind],
2853 $stmt = "lname LIKE ? AND NOT ( age < ? OR age > ? )"
2854 @bind = ('%son%', 10, 20)
2856 =head3 Deprecated usage of Literal SQL
2858 Below are some examples of archaic use of literal SQL. It is shown only as
2859 reference for those who deal with legacy code. Each example has a much
2860 better, cleaner and safer alternative that users should opt for in new code.
2866 my %where = ( requestor => \'IS NOT NULL' )
2868 $stmt = "WHERE requestor IS NOT NULL"
2870 This used to be the way of generating NULL comparisons, before the handling
2871 of C<undef> got formalized. For new code please use the superior syntax as
2872 described in L</Tests for NULL values>.
2876 my %where = ( requestor => \'= submitter' )
2878 $stmt = "WHERE requestor = submitter"
2880 This used to be the only way to compare columns. Use the superior L</-ident>
2881 method for all new code. For example an identifier declared in such a way
2882 will be properly quoted if L</quote_char> is properly set, while the legacy
2883 form will remain as supplied.
2887 my %where = ( is_ready => \"", completed => { '>', '2012-12-21' } )
2889 $stmt = "WHERE completed > ? AND is_ready"
2890 @bind = ('2012-12-21')
2892 Using an empty string literal used to be the only way to express a boolean.
2893 For all new code please use the much more readable
2894 L<-bool|/Unary operators: bool> operator.
2900 These pages could go on for a while, since the nesting of the data
2901 structures this module can handle are pretty much unlimited (the
2902 module implements the C<WHERE> expansion as a recursive function
2903 internally). Your best bet is to "play around" with the module a
2904 little to see how the data structures behave, and choose the best
2905 format for your data based on that.
2907 And of course, all the values above will probably be replaced with
2908 variables gotten from forms or the command line. After all, if you
2909 knew everything ahead of time, you wouldn't have to worry about
2910 dynamically-generating SQL and could just hardwire it into your
2913 =head1 ORDER BY CLAUSES
2915 Some functions take an order by clause. This can either be a scalar (just a
2916 column name), a hashref of C<< { -desc => 'col' } >> or C<< { -asc => 'col' }
2917 >>, a scalarref, an arrayref-ref, or an arrayref of any of the previous
2920 Given | Will Generate
2921 ---------------------------------------------------------------
2923 'colA' | ORDER BY colA
2925 [qw/colA colB/] | ORDER BY colA, colB
2927 {-asc => 'colA'} | ORDER BY colA ASC
2929 {-desc => 'colB'} | ORDER BY colB DESC
2931 ['colA', {-asc => 'colB'}] | ORDER BY colA, colB ASC
2933 { -asc => [qw/colA colB/] } | ORDER BY colA ASC, colB ASC
2935 \'colA DESC' | ORDER BY colA DESC
2937 \[ 'FUNC(colA, ?)', $x ] | ORDER BY FUNC(colA, ?)
2938 | /* ...with $x bound to ? */
2941 { -asc => 'colA' }, | colA ASC,
2942 { -desc => [qw/colB/] }, | colB DESC,
2943 { -asc => [qw/colC colD/] },| colC ASC, colD ASC,
2944 \'colE DESC', | colE DESC,
2945 \[ 'FUNC(colF, ?)', $x ], | FUNC(colF, ?)
2946 ] | /* ...with $x bound to ? */
2947 ===============================================================
2951 =head1 SPECIAL OPERATORS
2953 my $sqlmaker = SQL::Abstract->new(special_ops => [
2957 my ($self, $field, $op, $arg) = @_;
2963 handler => 'method_name',
2967 A "special operator" is a SQL syntactic clause that can be
2968 applied to a field, instead of a usual binary operator.
2971 WHERE field IN (?, ?, ?)
2972 WHERE field BETWEEN ? AND ?
2973 WHERE MATCH(field) AGAINST (?, ?)
2975 Special operators IN and BETWEEN are fairly standard and therefore
2976 are builtin within C<SQL::Abstract> (as the overridable methods
2977 C<_where_field_IN> and C<_where_field_BETWEEN>). For other operators,
2978 like the MATCH .. AGAINST example above which is specific to MySQL,
2979 you can write your own operator handlers - supply a C<special_ops>
2980 argument to the C<new> method. That argument takes an arrayref of
2981 operator definitions; each operator definition is a hashref with two
2988 the regular expression to match the operator
2992 Either a coderef or a plain scalar method name. In both cases
2993 the expected return is C<< ($sql, @bind) >>.
2995 When supplied with a method name, it is simply called on the
2996 L<SQL::Abstract> object as:
2998 $self->$method_name($field, $op, $arg)
3002 $field is the LHS of the operator
3003 $op is the part that matched the handler regex
3006 When supplied with a coderef, it is called as:
3008 $coderef->($self, $field, $op, $arg)
3013 For example, here is an implementation
3014 of the MATCH .. AGAINST syntax for MySQL
3016 my $sqlmaker = SQL::Abstract->new(special_ops => [
3018 # special op for MySql MATCH (field) AGAINST(word1, word2, ...)
3019 {regex => qr/^match$/i,
3021 my ($self, $field, $op, $arg) = @_;
3022 $arg = [$arg] if not ref $arg;
3023 my $label = $self->_quote($field);
3024 my ($placeholder) = $self->_convert('?');
3025 my $placeholders = join ", ", (($placeholder) x @$arg);
3026 my $sql = $self->_sqlcase('match') . " ($label) "
3027 . $self->_sqlcase('against') . " ($placeholders) ";
3028 my @bind = $self->_bindtype($field, @$arg);
3029 return ($sql, @bind);
3036 =head1 UNARY OPERATORS
3038 my $sqlmaker = SQL::Abstract->new(unary_ops => [
3042 my ($self, $op, $arg) = @_;
3048 handler => 'method_name',
3052 A "unary operator" is a SQL syntactic clause that can be
3053 applied to a field - the operator goes before the field
3055 You can write your own operator handlers - supply a C<unary_ops>
3056 argument to the C<new> method. That argument takes an arrayref of
3057 operator definitions; each operator definition is a hashref with two
3064 the regular expression to match the operator
3068 Either a coderef or a plain scalar method name. In both cases
3069 the expected return is C<< $sql >>.
3071 When supplied with a method name, it is simply called on the
3072 L<SQL::Abstract> object as:
3074 $self->$method_name($op, $arg)
3078 $op is the part that matched the handler regex
3079 $arg is the RHS or argument of the operator
3081 When supplied with a coderef, it is called as:
3083 $coderef->($self, $op, $arg)
3091 Thanks to some benchmarking by Mark Stosberg, it turns out that
3092 this module is many orders of magnitude faster than using C<DBIx::Abstract>.
3093 I must admit this wasn't an intentional design issue, but it's a
3094 byproduct of the fact that you get to control your C<DBI> handles
3097 To maximize performance, use a code snippet like the following:
3099 # prepare a statement handle using the first row
3100 # and then reuse it for the rest of the rows
3102 for my $href (@array_of_hashrefs) {
3103 $stmt ||= $sql->insert('table', $href);
3104 $sth ||= $dbh->prepare($stmt);
3105 $sth->execute($sql->values($href));
3108 The reason this works is because the keys in your C<$href> are sorted
3109 internally by B<SQL::Abstract>. Thus, as long as your data retains
3110 the same structure, you only have to generate the SQL the first time
3111 around. On subsequent queries, simply use the C<values> function provided
3112 by this module to return your values in the correct order.
3114 However this depends on the values having the same type - if, for
3115 example, the values of a where clause may either have values
3116 (resulting in sql of the form C<column = ?> with a single bind
3117 value), or alternatively the values might be C<undef> (resulting in
3118 sql of the form C<column IS NULL> with no bind value) then the
3119 caching technique suggested will not work.
3123 If you use my C<CGI::FormBuilder> module at all, you'll hopefully
3124 really like this part (I do, at least). Building up a complex query
3125 can be as simple as the following:
3132 use CGI::FormBuilder;
3135 my $form = CGI::FormBuilder->new(...);
3136 my $sql = SQL::Abstract->new;
3138 if ($form->submitted) {
3139 my $field = $form->field;
3140 my $id = delete $field->{id};
3141 my($stmt, @bind) = $sql->update('table', $field, {id => $id});
3144 Of course, you would still have to connect using C<DBI> to run the
3145 query, but the point is that if you make your form look like your
3146 table, the actual query script can be extremely simplistic.
3148 If you're B<REALLY> lazy (I am), check out C<HTML::QuickTable> for
3149 a fast interface to returning and formatting data. I frequently
3150 use these three modules together to write complex database query
3151 apps in under 50 lines.
3153 =head1 HOW TO CONTRIBUTE
3155 Contributions are always welcome, in all usable forms (we especially
3156 welcome documentation improvements). The delivery methods include git-
3157 or unified-diff formatted patches, GitHub pull requests, or plain bug
3158 reports either via RT or the Mailing list. Contributors are generally
3159 granted full access to the official repository after their first several
3160 patches pass successful review.
3162 This project is maintained in a git repository. The code and related tools are
3163 accessible at the following locations:
3167 =item * Official repo: L<git://git.shadowcat.co.uk/dbsrgits/SQL-Abstract.git>
3169 =item * Official gitweb: L<http://git.shadowcat.co.uk/gitweb/gitweb.cgi?p=dbsrgits/SQL-Abstract.git>
3171 =item * GitHub mirror: L<https://github.com/dbsrgits/sql-abstract>
3173 =item * Authorized committers: L<ssh://dbsrgits@git.shadowcat.co.uk/SQL-Abstract.git>
3179 Version 1.50 was a major internal refactoring of C<SQL::Abstract>.
3180 Great care has been taken to preserve the I<published> behavior
3181 documented in previous versions in the 1.* family; however,
3182 some features that were previously undocumented, or behaved
3183 differently from the documentation, had to be changed in order
3184 to clarify the semantics. Hence, client code that was relying
3185 on some dark areas of C<SQL::Abstract> v1.*
3186 B<might behave differently> in v1.50.
3188 The main changes are:
3194 support for literal SQL through the C<< \ [ $sql, @bind ] >> syntax.
3198 support for the { operator => \"..." } construct (to embed literal SQL)
3202 support for the { operator => \["...", @bind] } construct (to embed literal SQL with bind values)
3206 optional support for L<array datatypes|/"Inserting and Updating Arrays">
3210 defensive programming: check arguments
3214 fixed bug with global logic, which was previously implemented
3215 through global variables yielding side-effects. Prior versions would
3216 interpret C<< [ {cond1, cond2}, [cond3, cond4] ] >>
3217 as C<< "(cond1 AND cond2) OR (cond3 AND cond4)" >>.
3218 Now this is interpreted
3219 as C<< "(cond1 AND cond2) OR (cond3 OR cond4)" >>.
3224 fixed semantics of _bindtype on array args
3228 dropped the C<_anoncopy> of the %where tree. No longer necessary,
3229 we just avoid shifting arrays within that tree.
3233 dropped the C<_modlogic> function
3237 =head1 ACKNOWLEDGEMENTS
3239 There are a number of individuals that have really helped out with
3240 this module. Unfortunately, most of them submitted bugs via CPAN
3241 so I have no idea who they are! But the people I do know are:
3243 Ash Berlin (order_by hash term support)
3244 Matt Trout (DBIx::Class support)
3245 Mark Stosberg (benchmarking)
3246 Chas Owens (initial "IN" operator support)
3247 Philip Collins (per-field SQL functions)
3248 Eric Kolve (hashref "AND" support)
3249 Mike Fragassi (enhancements to "BETWEEN" and "LIKE")
3250 Dan Kubb (support for "quote_char" and "name_sep")
3251 Guillermo Roditi (patch to cleanup "IN" and "BETWEEN", fix and tests for _order_by)
3252 Laurent Dami (internal refactoring, extensible list of special operators, literal SQL)
3253 Norbert Buchmuller (support for literal SQL in hashpair, misc. fixes & tests)
3254 Peter Rabbitson (rewrite of SQLA::Test, misc. fixes & tests)
3255 Oliver Charles (support for "RETURNING" after "INSERT")
3261 L<DBIx::Class>, L<DBIx::Abstract>, L<CGI::FormBuilder>, L<HTML::QuickTable>.
3265 Copyright (c) 2001-2007 Nathan Wiger <nwiger@cpan.org>. All Rights Reserved.
3267 This module is actively maintained by Matt Trout <mst@shadowcatsystems.co.uk>
3269 For support, your best bet is to try the C<DBIx::Class> users mailing list.
3270 While not an official support venue, C<DBIx::Class> makes heavy use of
3271 C<SQL::Abstract>, and as such list members there are very familiar with
3272 how to create queries.
3276 This module is free software; you may copy this under the same
3277 terms as perl itself (either the GNU General Public License or
3278 the Artistic License)