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" }},
44 #======================================================================
45 # DEBUGGING AND ERROR REPORTING
46 #======================================================================
49 return unless $_[0]->{debug}; shift; # a little faster
50 my $func = (caller(1))[3];
51 warn "[$func] ", @_, "\n";
55 my($func) = (caller(1))[3];
56 Carp::carp "[$func] Warning: ", @_;
60 my($func) = (caller(1))[3];
61 Carp::croak "[$func] Fatal: ", @_;
64 sub is_literal_value ($) {
65 ref $_[0] eq 'SCALAR' ? [ ${$_[0]} ]
66 : ( ref $_[0] eq 'REF' and ref ${$_[0]} eq 'ARRAY' ) ? [ @${ $_[0] } ]
70 sub is_undef_value ($) {
74 and exists $_[0]->{-value}
75 and not defined $_[0]->{-value}
79 # FIXME XSify - this can be done so much more efficiently
80 sub is_plain_value ($) {
82 ! length ref $_[0] ? \($_[0])
84 ref $_[0] eq 'HASH' and keys %{$_[0]} == 1
86 exists $_[0]->{-value}
87 ) ? \($_[0]->{-value})
89 # reuse @_ for even moar speedz
90 defined ( $_[1] = Scalar::Util::blessed $_[0] )
92 # deliberately not using Devel::OverloadInfo - the checks we are
93 # intersted in are much more limited than the fullblown thing, and
94 # this is a very hot piece of code
96 # simply using ->can('(""') can leave behind stub methods that
97 # break actually using the overload later (see L<perldiag/Stub
98 # found while resolving method "%s" overloading "%s" in package
99 # "%s"> and the source of overload::mycan())
101 # either has stringification which DBI SHOULD prefer out of the box
102 grep { *{ (qq[${_}::(""]) }{CODE} } @{ $_[2] = mro::get_linear_isa( $_[1] ) }
104 # has nummification or boolification, AND fallback is *not* disabled
106 SQL::Abstract::_ENV_::DETECT_AUTOGENERATED_STRINGIFICATION
109 grep { *{"${_}::(0+"}{CODE} } @{$_[2]}
111 grep { *{"${_}::(bool"}{CODE} } @{$_[2]}
115 # no fallback specified at all
116 ! ( ($_[3]) = grep { *{"${_}::()"}{CODE} } @{$_[2]} )
118 # fallback explicitly undef
119 ! defined ${"$_[3]::()"}
132 #======================================================================
134 #======================================================================
138 my $class = ref($self) || $self;
139 my %opt = (ref $_[0] eq 'HASH') ? %{$_[0]} : @_;
141 # choose our case by keeping an option around
142 delete $opt{case} if $opt{case} && $opt{case} ne 'lower';
144 # default logic for interpreting arrayrefs
145 $opt{logic} = $opt{logic} ? uc $opt{logic} : 'OR';
147 # how to return bind vars
148 $opt{bindtype} ||= 'normal';
150 # default comparison is "=", but can be overridden
153 # try to recognize which are the 'equality' and 'inequality' ops
154 # (temporary quickfix (in 2007), should go through a more seasoned API)
155 $opt{equality_op} = qr/^( \Q$opt{cmp}\E | \= )$/ix;
156 $opt{inequality_op} = qr/^( != | <> )$/ix;
158 $opt{like_op} = qr/^ (is\s+)? r?like $/xi;
159 $opt{not_like_op} = qr/^ (is\s+)? not \s+ r?like $/xi;
162 $opt{sqltrue} ||= '1=1';
163 $opt{sqlfalse} ||= '0=1';
166 $opt{special_ops} ||= [];
168 # regexes are applied in order, thus push after user-defines
169 push @{$opt{special_ops}}, @BUILTIN_SPECIAL_OPS;
171 if ($class->isa('DBIx::Class::SQLMaker')) {
172 push @{$opt{special_ops}}, our $DBIC_Compat_Op ||= {
173 regex => qr/^(?:ident|value|(?:not\s)?in)$/i, handler => sub { die "NOPE" }
175 $opt{is_dbic_sqlmaker} = 1;
179 $opt{unary_ops} ||= [];
181 # rudimentary sanity-check for user supplied bits treated as functions/operators
182 # If a purported function matches this regular expression, an exception is thrown.
183 # Literal SQL is *NOT* subject to this check, only functions (and column names
184 # when quoting is not in effect)
187 # need to guard against ()'s in column names too, but this will break tons of
188 # hacks... ideas anyone?
189 $opt{injection_guard} ||= qr/
195 $opt{expand_unary} = {};
198 -not => '_expand_not',
199 -bool => '_expand_bool',
200 -and => '_expand_op_andor',
201 -or => '_expand_op_andor',
202 -nest => '_expand_nest',
203 -bind => sub { shift; +{ @_ } },
207 'between' => '_expand_between',
208 'not between' => '_expand_between',
209 'in' => '_expand_in',
210 'not in' => '_expand_in',
211 'nest' => '_expand_nest',
212 (map +($_ => '_expand_op_andor'), ('and', 'or')),
213 (map +($_ => '_expand_op_is'), ('is', 'is not')),
216 # placeholder for _expand_unop system
218 my %unops = (-ident => '_expand_ident', -value => '_expand_value');
219 foreach my $name (keys %unops) {
220 $opt{expand}{$name} = $unops{$name};
221 my ($op) = $name =~ /^-(.*)$/;
222 $opt{expand_op}{$op} = sub {
223 my ($self, $op, $arg, $k) = @_;
224 return $self->_expand_expr_hashpair_cmp(
225 $k, { "-${op}" => $arg }
232 (map +("-$_", "_render_$_"), qw(op func bind ident literal list)),
237 (map +($_ => '_render_op_between'), 'between', 'not between'),
238 (map +($_ => '_render_op_in'), 'in', 'not in'),
239 (map +($_ => '_render_unop_postfix'),
240 'is null', 'is not null', 'asc', 'desc',
242 (not => '_render_op_not'),
243 (map +($_ => '_render_op_andor'), qw(and or)),
246 return bless \%opt, $class;
249 sub sqltrue { +{ -literal => [ $_[0]->{sqltrue} ] } }
250 sub sqlfalse { +{ -literal => [ $_[0]->{sqlfalse} ] } }
252 sub _assert_pass_injection_guard {
253 if ($_[1] =~ $_[0]->{injection_guard}) {
254 my $class = ref $_[0];
255 puke "Possible SQL injection attempt '$_[1]'. If this is indeed a part of the "
256 . "desired SQL use literal SQL ( \'...' or \[ '...' ] ) or supply your own "
257 . "{injection_guard} attribute to ${class}->new()"
262 #======================================================================
264 #======================================================================
268 my $table = $self->_table(shift);
269 my $data = shift || return;
272 my $method = $self->_METHOD_FOR_refkind("_insert", $data);
273 my ($sql, @bind) = $self->$method($data);
274 $sql = join " ", $self->_sqlcase('insert into'), $table, $sql;
276 if ($options->{returning}) {
277 my ($s, @b) = $self->_insert_returning($options);
282 return wantarray ? ($sql, @bind) : $sql;
285 # So that subclasses can override INSERT ... RETURNING separately from
286 # UPDATE and DELETE (e.g. DBIx::Class::SQLMaker::Oracle does this)
287 sub _insert_returning { shift->_returning(@_) }
290 my ($self, $options) = @_;
292 my $f = $options->{returning};
294 my ($sql, @bind) = $self->render_aqt(
295 $self->_expand_maybe_list_expr($f, undef, -ident)
298 ? $self->_sqlcase(' returning ') . $sql
299 : ($self->_sqlcase(' returning ').$sql, @bind);
302 sub _insert_HASHREF { # explicit list of fields and then values
303 my ($self, $data) = @_;
305 my @fields = sort keys %$data;
307 my ($sql, @bind) = $self->_insert_values($data);
310 $_ = $self->_quote($_) foreach @fields;
311 $sql = "( ".join(", ", @fields).") ".$sql;
313 return ($sql, @bind);
316 sub _insert_ARRAYREF { # just generate values(?,?) part (no list of fields)
317 my ($self, $data) = @_;
319 # no names (arrayref) so can't generate bindtype
320 $self->{bindtype} ne 'columns'
321 or belch "can't do 'columns' bindtype when called with arrayref";
323 my (@values, @all_bind);
324 foreach my $value (@$data) {
325 my ($values, @bind) = $self->_insert_value(undef, $value);
326 push @values, $values;
327 push @all_bind, @bind;
329 my $sql = $self->_sqlcase('values')." ( ".join(", ", @values)." )";
330 return ($sql, @all_bind);
333 sub _insert_ARRAYREFREF { # literal SQL with bind
334 my ($self, $data) = @_;
336 my ($sql, @bind) = @${$data};
337 $self->_assert_bindval_matches_bindtype(@bind);
339 return ($sql, @bind);
343 sub _insert_SCALARREF { # literal SQL without bind
344 my ($self, $data) = @_;
350 my ($self, $data) = @_;
352 my (@values, @all_bind);
353 foreach my $column (sort keys %$data) {
354 my ($values, @bind) = $self->_insert_value($column, $data->{$column});
355 push @values, $values;
356 push @all_bind, @bind;
358 my $sql = $self->_sqlcase('values')." ( ".join(", ", @values)." )";
359 return ($sql, @all_bind);
363 my ($self, $column, $v) = @_;
365 return $self->render_aqt(
366 $self->_expand_insert_value($column, $v)
370 sub _expand_insert_value {
371 my ($self, $column, $v) = @_;
373 if (ref($v) eq 'ARRAY') {
374 if ($self->{array_datatypes}) {
375 return +{ -bind => [ $column, $v ] };
377 my ($sql, @bind) = @$v;
378 $self->_assert_bindval_matches_bindtype(@bind);
379 return +{ -literal => $v };
381 if (ref($v) eq 'HASH') {
382 if (grep !/^-/, keys %$v) {
383 belch "HASH ref as bind value in insert is not supported";
384 return +{ -bind => [ $column, $v ] };
388 return +{ -bind => [ $column, undef ] };
390 local our $Cur_Col_Meta = $column;
391 return $self->expand_expr($v);
396 #======================================================================
398 #======================================================================
403 my $table = $self->_table(shift);
404 my $data = shift || return;
408 # first build the 'SET' part of the sql statement
409 puke "Unsupported data type specified to \$sql->update"
410 unless ref $data eq 'HASH';
412 my ($sql, @all_bind) = $self->_update_set_values($data);
413 $sql = $self->_sqlcase('update ') . $table . $self->_sqlcase(' set ')
417 my($where_sql, @where_bind) = $self->where($where);
419 push @all_bind, @where_bind;
422 if ($options->{returning}) {
423 my ($returning_sql, @returning_bind) = $self->_update_returning($options);
424 $sql .= $returning_sql;
425 push @all_bind, @returning_bind;
428 return wantarray ? ($sql, @all_bind) : $sql;
431 sub _update_set_values {
432 my ($self, $data) = @_;
434 return $self->render_aqt(
435 $self->_expand_update_set_values($data),
439 sub _expand_update_set_values {
440 my ($self, $data) = @_;
441 $self->_expand_maybe_list_expr( [
444 $set = { -bind => $_ } unless defined $set;
445 +{ -op => [ '=', $self->_expand_ident(-ident => $k), $set ] };
451 ? ($self->{array_datatypes}
452 ? [ $k, +{ -bind => [ $k, $v ] } ]
453 : [ $k, +{ -literal => $v } ])
455 local our $Cur_Col_Meta = $k;
456 [ $k, $self->_expand_expr($v) ]
463 # So that subclasses can override UPDATE ... RETURNING separately from
465 sub _update_returning { shift->_returning(@_) }
469 #======================================================================
471 #======================================================================
476 my $table = $self->_table(shift);
477 my $fields = shift || '*';
481 my ($fields_sql, @bind) = $self->_select_fields($fields);
483 my ($where_sql, @where_bind) = $self->where($where, $order);
484 push @bind, @where_bind;
486 my $sql = join(' ', $self->_sqlcase('select'), $fields_sql,
487 $self->_sqlcase('from'), $table)
490 return wantarray ? ($sql, @bind) : $sql;
494 my ($self, $fields) = @_;
495 return $fields unless ref($fields);
496 return $self->render_aqt(
497 $self->_expand_maybe_list_expr($fields, undef, '-ident')
501 #======================================================================
503 #======================================================================
508 my $table = $self->_table(shift);
512 my($where_sql, @bind) = $self->where($where);
513 my $sql = $self->_sqlcase('delete from ') . $table . $where_sql;
515 if ($options->{returning}) {
516 my ($returning_sql, @returning_bind) = $self->_delete_returning($options);
517 $sql .= $returning_sql;
518 push @bind, @returning_bind;
521 return wantarray ? ($sql, @bind) : $sql;
524 # So that subclasses can override DELETE ... RETURNING separately from
526 sub _delete_returning { shift->_returning(@_) }
530 #======================================================================
532 #======================================================================
536 # Finally, a separate routine just to handle WHERE clauses
538 my ($self, $where, $order) = @_;
540 local $self->{convert_where} = $self->{convert};
543 my ($sql, @bind) = defined($where)
544 ? $self->_recurse_where($where)
546 $sql = (defined $sql and length $sql) ? $self->_sqlcase(' where ') . "( $sql )" : '';
550 my ($order_sql, @order_bind) = $self->_order_by($order);
552 push @bind, @order_bind;
555 return wantarray ? ($sql, @bind) : $sql;
558 { our $Default_Scalar_To = -value }
561 my ($self, $expr, $default_scalar_to) = @_;
562 local our $Default_Scalar_To = $default_scalar_to if $default_scalar_to;
563 $self->_expand_expr($expr);
567 my ($self, $aqt) = @_;
568 my ($k, $v, @rest) = %$aqt;
570 if (my $meth = $self->{render}{$k}) {
571 return $self->$meth($v);
573 die "notreached: $k";
577 my ($self, $expr) = @_;
578 $self->render_aqt($self->expand_expr($expr));
582 my ($self, $raw) = @_;
583 s/^-(?=[a-z])//, s/\s+/_/g for my $op = lc $raw;
588 my ($self, $expr) = @_;
589 our $Expand_Depth ||= 0; local $Expand_Depth = $Expand_Depth + 1;
590 return undef unless defined($expr);
591 if (ref($expr) eq 'HASH') {
592 return undef unless my $kc = keys %$expr;
594 return $self->_expand_op_andor(-and => $expr);
596 my ($key, $value) = %$expr;
597 if ($key =~ /^-/ and $key =~ s/ [_\s]? \d+ $//x ) {
598 belch 'Use of [and|or|nest]_N modifiers is deprecated and will be removed in SQLA v2.0. '
599 . "You probably wanted ...-and => [ $key => COND1, $key => COND2 ... ]";
601 if (my $exp = $self->{expand}{$key}) {
602 return $self->$exp($key, $value);
604 return $self->_expand_expr_hashpair($key, $value);
606 if (ref($expr) eq 'ARRAY') {
607 my $logic = '-'.lc($self->{logic});
608 return $self->_expand_op_andor($logic, $expr);
610 if (my $literal = is_literal_value($expr)) {
611 return +{ -literal => $literal };
613 if (!ref($expr) or Scalar::Util::blessed($expr)) {
614 return $self->_expand_expr_scalar($expr);
619 sub _expand_expr_hashpair {
620 my ($self, $k, $v) = @_;
621 unless (defined($k) and length($k)) {
622 if (defined($k) and my $literal = is_literal_value($v)) {
623 belch 'Hash-pairs consisting of an empty string with a literal are deprecated, and will be removed in 2.0: use -and => [ $literal ] instead';
624 return { -literal => $literal };
626 puke "Supplying an empty left hand side argument is not supported";
629 return $self->_expand_expr_hashpair_op($k, $v);
631 return $self->_expand_expr_hashpair_ident($k, $v);
634 sub _expand_expr_hashpair_ident {
635 my ($self, $k, $v) = @_;
637 local our $Cur_Col_Meta = $k;
639 # hash with multiple or no elements is andor
641 if (ref($v) eq 'HASH' and keys %$v != 1) {
642 return $self->_expand_op_andor(-and => $v, $k);
645 # undef needs to be re-sent with cmp to achieve IS/IS NOT NULL
647 if (is_undef_value($v)) {
648 return $self->_expand_expr_hashpair_cmp($k => undef);
651 # scalars and objects get expanded as whatever requested or values
653 if (!ref($v) or Scalar::Util::blessed($v)) {
654 return $self->_expand_expr_hashpair_scalar($k, $v);
657 # single key hashref is a hashtriple
659 if (ref($v) eq 'HASH') {
660 return $self->_expand_expr_hashtriple($k, %$v);
663 # arrayref needs re-engineering over the elements
665 if (ref($v) eq 'ARRAY') {
666 return $self->sqlfalse unless @$v;
667 $self->_debug("ARRAY($k) means distribute over elements");
669 $v->[0] =~ /^-(and|or)$/i
670 ? shift(@{$v = [ @$v ]})
671 : '-'.lc($self->{logic} || 'OR')
673 return $self->_expand_op_andor(
678 if (my $literal = is_literal_value($v)) {
680 belch 'Hash-pairs consisting of an empty string with a literal are deprecated, and will be removed in 2.0: use -and => [ $literal ] instead';
683 my ($sql, @bind) = @$literal;
684 if ($self->{bindtype} eq 'columns') {
686 $self->_assert_bindval_matches_bindtype($_);
689 return +{ -literal => [ $self->_quote($k).' '.$sql, @bind ] };
694 sub _expand_expr_scalar {
695 my ($self, $expr) = @_;
697 return $self->_expand_expr({ (our $Default_Scalar_To) => $expr });
700 sub _expand_expr_hashpair_scalar {
701 my ($self, $k, $v) = @_;
703 return $self->_expand_expr_hashpair_cmp(
704 $k, $self->_expand_expr_scalar($v),
708 sub _expand_expr_hashpair_op {
709 my ($self, $k, $v) = @_;
711 my $op = $self->_normalize_op($k);
713 $self->_assert_pass_injection_guard($op);
715 # Ops prefixed with -not_ get converted
717 if (my ($rest) = $op =~/^not_(.*)$/) {
720 $self->_expand_expr({ "-${rest}", $v })
727 my $op = join(' ', split '_', $op);
729 # the old special op system requires illegality for top-level use
732 (our $Expand_Depth) == 1
733 and List::Util::first { $op =~ $_->{regex} } @{$self->{special_ops}}
735 puke "Illegal use of top-level '-$op'"
738 # the old unary op system means we should touch nothing and let it work
740 if (my $us = List::Util::first { $op =~ $_->{regex} } @{$self->{unary_ops}}) {
741 return { -op => [ $op, $v ] };
745 # an explicit node type is currently assumed to be expanded (this is almost
746 # certainly wrong and there should be expansion anyway)
748 if ($self->{render}{$k}) {
752 # hashref RHS values get expanded and used as op/func args
757 and (keys %$v)[0] =~ /^-/
759 my ($func) = $k =~ /^-(.*)$/;
761 if (List::Util::first { $func =~ $_->{regex} } @{$self->{special_ops}}) {
762 return +{ -op => [ $func, $self->_expand_expr($v) ] };
765 return +{ -func => [ $func, $self->_expand_expr($v) ] };
768 # scalars and literals get simply expanded
770 if (!ref($v) or is_literal_value($v)) {
771 return +{ -op => [ $op, $self->_expand_expr($v) ] };
777 sub _expand_expr_hashpair_cmp {
778 my ($self, $k, $v) = @_;
779 $self->_expand_expr_hashtriple($k, $self->{cmp}, $v);
782 sub _expand_expr_hashtriple {
783 my ($self, $k, $vk, $vv) = @_;
785 my $ik = $self->_expand_ident(-ident => $k);
787 my $op = $self->_normalize_op($vk);
788 $self->_assert_pass_injection_guard($op);
790 if ($op =~ s/ _? \d+ $//x ) {
791 return $self->_expand_expr($k, { $vk, $vv });
793 if (my $x = $self->{expand_op}{$op}) {
794 local our $Cur_Col_Meta = $k;
795 return $self->$x($op, $vv, $k);
799 my $op = join(' ', split '_', $op);
801 if (my $us = List::Util::first { $op =~ $_->{regex} } @{$self->{special_ops}}) {
802 return { -op => [ $op, $ik, $vv ] };
804 if (my $us = List::Util::first { $op =~ $_->{regex} } @{$self->{unary_ops}}) {
808 { -op => [ $op, $vv ] }
812 if (ref($vv) eq 'ARRAY') {
814 my $logic = (defined($raw[0]) and $raw[0] =~ /^-(and|or)$/i)
815 ? shift @raw : '-or';
816 my @values = map +{ $vk => $_ }, @raw;
818 $op =~ $self->{inequality_op}
819 or $op =~ $self->{not_like_op}
821 if (lc($logic) eq '-or' and @values > 1) {
822 belch "A multi-element arrayref as an argument to the inequality op '${\uc($op)}' "
823 . 'is technically equivalent to an always-true 1=1 (you probably wanted '
824 . "to say ...{ \$inequality_op => [ -and => \@values ] }... instead)"
829 # try to DWIM on equality operators
830 return ($self->_dwim_op_to_is($op,
831 "Supplying an empty arrayref to '%s' is deprecated",
832 "operator '%s' applied on an empty array (field '$k')"
833 ) ? $self->sqlfalse : $self->sqltrue);
835 return $self->_expand_op_andor($logic => \@values, $k);
837 if (is_undef_value($vv)) {
838 my $is = ($self->_dwim_op_to_is($op,
839 "Supplying an undefined argument to '%s' is deprecated",
840 "unexpected operator '%s' with undef operand",
841 ) ? 'is' : 'is not');
843 return $self->_expand_expr_hashpair($k => { $is, undef });
845 local our $Cur_Col_Meta = $k;
849 $self->_expand_expr($vv)
854 my ($self, $raw, $empty, $fail) = @_;
856 my $op = $self->_normalize_op($raw);
858 if ($op =~ /^not$/i) {
861 if ($op =~ $self->{equality_op}) {
864 if ($op =~ $self->{like_op}) {
865 belch(sprintf $empty, uc($op));
868 if ($op =~ $self->{inequality_op}) {
871 if ($op =~ $self->{not_like_op}) {
872 belch(sprintf $empty, uc($op));
875 puke(sprintf $fail, $op);
879 my ($self, $op, $body) = @_;
880 unless (defined($body) or (ref($body) and ref($body) eq 'ARRAY')) {
881 puke "$op requires a single plain scalar argument (a quotable identifier) or an arrayref of identifier parts";
883 my @parts = map split(/\Q${\($self->{name_sep}||'.')}\E/, $_),
884 ref($body) ? @$body : $body;
885 return { -ident => $parts[-1] } if $self->{_dequalify_idents};
886 unless ($self->{quote_char}) {
887 $self->_assert_pass_injection_guard($_) for @parts;
889 return +{ -ident => \@parts };
893 +{ -bind => [ our $Cur_Col_Meta, $_[2] ] };
897 +{ -op => [ 'not', $_[0]->_expand_expr($_[2]) ] };
901 my ($self, undef, $v) = @_;
903 return $self->_expand_expr($v);
905 puke "-bool => undef not supported" unless defined($v);
906 return $self->_expand_ident(-ident => $v);
909 sub _expand_op_andor {
910 my ($self, $logic, $v, $k) = @_;
912 $v = [ map +{ $k, $_ },
914 ? (map +{ $_ => $v->{$_} }, sort keys %$v)
918 my ($logop) = $logic =~ /^-?(.*)$/;
919 if (ref($v) eq 'HASH') {
920 return undef unless keys %$v;
923 map $self->_expand_expr({ $_ => $v->{$_} }),
927 if (ref($v) eq 'ARRAY') {
928 $logop eq 'and' or $logop eq 'or' or puke "unknown logic: $logop";
931 (ref($_) eq 'ARRAY' and @$_)
932 or (ref($_) eq 'HASH' and %$_)
938 while (my ($el) = splice @expr, 0, 1) {
939 puke "Supplying an empty left hand side argument is not supported in array-pairs"
940 unless defined($el) and length($el);
941 my $elref = ref($el);
943 local our $Expand_Depth = 0;
944 push(@res, grep defined, $self->_expand_expr({ $el, shift(@expr) }));
945 } elsif ($elref eq 'ARRAY') {
946 push(@res, grep defined, $self->_expand_expr($el)) if @$el;
947 } elsif (my $l = is_literal_value($el)) {
948 push @res, { -literal => $l };
949 } elsif ($elref eq 'HASH') {
950 local our $Expand_Depth = 0;
951 push @res, grep defined, $self->_expand_expr($el) if %$el;
957 # return $res[0] if @res == 1;
958 return { -op => [ $logop, @res ] };
964 my ($self, $op, $vv, $k) = @_;
965 puke "$op can only take undef as argument"
969 and exists($vv->{-value})
970 and !defined($vv->{-value})
972 return +{ -op => [ $op.' null', $self->_expand_ident(-ident => $k) ] };
975 sub _expand_between {
976 my ($self, $op, $vv, $k) = @_;
977 local our $Cur_Col_Meta = $k;
978 my @rhs = map $self->_expand_expr($_),
979 ref($vv) eq 'ARRAY' ? @$vv : $vv;
981 (@rhs == 1 and ref($rhs[0]) eq 'HASH' and $rhs[0]->{-literal})
983 (@rhs == 2 and defined($rhs[0]) and defined($rhs[1]))
985 puke "Operator '${\uc($op)}' requires either an arrayref with two defined values or expressions, or a single literal scalarref/arrayref-ref";
989 $self->_expand_ident(-ident => $k),
995 my ($self, $op, $vv, $k) = @_;
996 if (my $literal = is_literal_value($vv)) {
997 my ($sql, @bind) = @$literal;
998 my $opened_sql = $self->_open_outer_paren($sql);
1000 $op, $self->_expand_ident(-ident => $k),
1001 [ { -literal => [ $opened_sql, @bind ] } ]
1005 'SQL::Abstract before v1.75 used to generate incorrect SQL when the '
1006 . "-${\uc($op)} operator was given an undef-containing list: !!!AUDIT YOUR CODE "
1007 . 'AND DATA!!! (the upcoming Data::Query-based version of SQL::Abstract '
1008 . 'will emit the logically correct SQL instead of raising this exception)'
1010 puke("Argument passed to the '${\uc($op)}' operator can not be undefined")
1012 my @rhs = map $self->_expand_expr($_),
1013 map { ref($_) ? $_ : { -bind => [ $k, $_ ] } }
1014 map { defined($_) ? $_: puke($undef_err) }
1015 (ref($vv) eq 'ARRAY' ? @$vv : $vv);
1016 return $self->${\($op =~ /^not/ ? 'sqltrue' : 'sqlfalse')} unless @rhs;
1020 $self->_expand_ident(-ident => $k),
1026 my ($self, $op, $v) = @_;
1027 # DBIx::Class requires a nest warning to be emitted once but the private
1028 # method it overrode to do so no longer exists
1029 if ($self->{is_dbic_sqlmaker}) {
1030 unless (our $Nest_Warned) {
1032 "-nest in search conditions is deprecated, you most probably wanted:\n"
1033 .q|{..., -and => [ \%cond0, \@cond1, \'cond2', \[ 'cond3', [ col => bind ] ], etc. ], ... }|
1038 return $self->_expand_expr($v);
1041 sub _recurse_where {
1042 my ($self, $where, $logic) = @_;
1044 # Special case: top level simple string treated as literal
1046 my $where_exp = (ref($where)
1047 ? $self->_expand_expr($where, $logic)
1048 : { -literal => [ $where ] });
1050 # dispatch expanded expression
1052 my ($sql, @bind) = defined($where_exp) ? $self->render_aqt($where_exp) : (undef);
1053 # DBIx::Class used to call _recurse_where in scalar context
1054 # something else might too...
1056 return ($sql, @bind);
1059 belch "Calling _recurse_where in scalar context is deprecated and will go away before 2.0";
1065 my ($self, $ident) = @_;
1067 return $self->_convert($self->_quote($ident));
1071 my ($self, $list) = @_;
1072 my @parts = grep length($_->[0]), map [ $self->render_aqt($_) ], @$list;
1073 return join(', ', map $_->[0], @parts), map @{$_}[1..$#$_], @parts;
1077 my ($self, $rest) = @_;
1078 my ($func, @args) = @$rest;
1082 push @arg_sql, shift @x;
1084 } map [ $self->render_aqt($_) ], @args;
1085 return ($self->_sqlcase($func).'('.join(', ', @arg_sql).')', @bind);
1089 my ($self, $bind) = @_;
1090 return ($self->_convert('?'), $self->_bindtype(@$bind));
1093 sub _render_literal {
1094 my ($self, $literal) = @_;
1095 $self->_assert_bindval_matches_bindtype(@{$literal}[1..$#$literal]);
1100 my ($self, $v) = @_;
1101 my ($op, @args) = @$v;
1102 if (my $r = $self->{render_op}{$op}) {
1103 return $self->$r($op, \@args);
1108 my $op = join(' ', split '_', $op);
1110 my $us = List::Util::first { $op =~ $_->{regex} } @{$self->{special_ops}};
1111 if ($us and @args > 1) {
1112 puke "Special op '${op}' requires first value to be identifier"
1113 unless my ($ident) = map $_->{-ident}, grep ref($_) eq 'HASH', $args[0];
1114 my $k = join(($self->{name_sep}||'.'), @$ident);
1115 local our $Expand_Depth = 1;
1116 return $self->${\($us->{handler})}($k, $op, $args[1]);
1118 if (my $us = List::Util::first { $op =~ $_->{regex} } @{$self->{unary_ops}}) {
1119 return $self->${\($us->{handler})}($op, $args[0]);
1124 return $self->_render_unop_prefix($op, \@args);
1126 return $self->_render_op_multop($op, \@args);
1132 sub _render_op_between {
1133 my ($self, $op, $args) = @_;
1134 my ($left, $low, $high) = @$args;
1135 my ($rhsql, @rhbind) = do {
1137 puke "Single arg to between must be a literal"
1138 unless $low->{-literal};
1141 my ($l, $h) = map [ $self->render_aqt($_) ], $low, $high;
1142 (join(' ', $l->[0], $self->_sqlcase('and'), $h->[0]),
1143 @{$l}[1..$#$l], @{$h}[1..$#$h])
1146 my ($lhsql, @lhbind) = $self->render_aqt($left);
1148 join(' ', '(', $lhsql, $self->_sqlcase($op), $rhsql, ')'),
1154 my ($self, $op, $args) = @_;
1155 my ($lhs, $rhs) = @$args;
1158 my ($sql, @bind) = $self->render_aqt($_);
1159 push @in_bind, @bind;
1162 my ($lhsql, @lbind) = $self->render_aqt($lhs);
1164 $lhsql.' '.$self->_sqlcase($op).' ( '
1165 .join(', ', @in_sql)
1171 sub _render_op_andor {
1172 my ($self, $op, $args) = @_;
1173 my @parts = grep length($_->[0]), map [ $self->render_aqt($_) ], @$args;
1174 return '' unless @parts;
1175 return @{$parts[0]} if @parts == 1;
1176 my ($sql, @bind) = $self->_render_op_multop($op, $args);
1177 return '( '.$sql.' )', @bind;
1180 sub _render_op_multop {
1181 my ($self, $op, $args) = @_;
1182 my @parts = grep length($_->[0]), map [ $self->render_aqt($_) ], @$args;
1183 return '' unless @parts;
1184 return @{$parts[0]} if @parts == 1;
1185 my ($final_sql) = join(
1186 ' '.$self->_sqlcase($op).' ',
1191 map @{$_}[1..$#$_], @parts
1194 sub _render_op_not {
1195 my ($self, $op, $v) = @_;
1196 my ($sql, @bind) = $self->_render_unop_prefix($op, $v);
1197 return "(${sql})", @bind;
1200 sub _render_unop_prefix {
1201 my ($self, $op, $v) = @_;
1202 my ($expr_sql, @bind) = $self->render_aqt($v->[0]);
1204 my $op_sql = $self->_sqlcase(join ' ', split '_', $op);
1205 return ("${op_sql} ${expr_sql}", @bind);
1208 sub _render_unop_postfix {
1209 my ($self, $op, $v) = @_;
1210 my ($expr_sql, @bind) = $self->render_aqt($v->[0]);
1211 my $op_sql = $self->_sqlcase(join ' ', split '_', $op);
1212 return ($expr_sql.' '.$op_sql, @bind);
1215 # Some databases (SQLite) treat col IN (1, 2) different from
1216 # col IN ( (1, 2) ). Use this to strip all outer parens while
1217 # adding them back in the corresponding method
1218 sub _open_outer_paren {
1219 my ($self, $sql) = @_;
1221 while (my ($inner) = $sql =~ /^ \s* \( (.*) \) \s* $/xs) {
1223 # there are closing parens inside, need the heavy duty machinery
1224 # to reevaluate the extraction starting from $sql (full reevaluation)
1225 if ($inner =~ /\)/) {
1226 require Text::Balanced;
1228 my (undef, $remainder) = do {
1229 # idiotic design - writes to $@ but *DOES NOT* throw exceptions
1231 Text::Balanced::extract_bracketed($sql, '()', qr/\s*/);
1234 # the entire expression needs to be a balanced bracketed thing
1235 # (after an extract no remainder sans trailing space)
1236 last if defined $remainder and $remainder =~ /\S/;
1246 #======================================================================
1248 #======================================================================
1250 sub _expand_order_by {
1251 my ($self, $arg) = @_;
1253 return unless defined($arg) and not (ref($arg) eq 'ARRAY' and !@$arg);
1255 my $expander = sub {
1256 my ($self, $dir, $expr) = @_;
1257 my @to_expand = ref($expr) eq 'ARRAY' ? @$expr : $expr;
1258 foreach my $arg (@to_expand) {
1262 and grep /^-(asc|desc)$/, keys %$arg
1264 puke "ordering direction hash passed to order by must have exactly one key (-asc or -desc)";
1268 defined($dir) ? { -op => [ $dir =~ /^-?(.*)$/ ,=> $_ ] } : $_
1270 map $self->expand_expr($_, -ident),
1271 map ref($_) eq 'ARRAY' ? @$_ : $_, @to_expand;
1272 return (@exp > 1 ? { -list => \@exp } : $exp[0]);
1275 local @{$self->{expand}}{qw(-asc -desc)} = (($expander) x 2);
1277 return $self->$expander(undef, $arg);
1281 my ($self, $arg) = @_;
1283 return '' unless defined(my $expanded = $self->_expand_order_by($arg));
1285 my ($sql, @bind) = $self->render_aqt($expanded);
1287 return '' unless length($sql);
1289 my $final_sql = $self->_sqlcase(' order by ').$sql;
1291 return wantarray ? ($final_sql, @bind) : $final_sql;
1294 # _order_by no longer needs to call this so doesn't but DBIC uses it.
1296 sub _order_by_chunks {
1297 my ($self, $arg) = @_;
1299 return () unless defined(my $expanded = $self->_expand_order_by($arg));
1301 return $self->_chunkify_order_by($expanded);
1304 sub _chunkify_order_by {
1305 my ($self, $expanded) = @_;
1307 return grep length, $self->render_aqt($expanded)
1308 if $expanded->{-ident} or @{$expanded->{-literal}||[]} == 1;
1311 if (ref() eq 'HASH' and my $l = $_->{-list}) {
1312 return map $self->_chunkify_order_by($_), @$l;
1314 return [ $self->render_aqt($_) ];
1318 #======================================================================
1319 # DATASOURCE (FOR NOW, JUST PLAIN TABLE OR LIST OF TABLES)
1320 #======================================================================
1326 $self->_expand_maybe_list_expr($from, undef, -ident)
1331 #======================================================================
1333 #======================================================================
1335 sub _expand_maybe_list_expr {
1336 my ($self, $expr, $logic, $default) = @_;
1338 if (ref($expr) eq 'ARRAY') {
1340 map $self->expand_expr($_, $default), @$expr
1347 return $self->expand_expr($e, $default);
1350 # highly optimized, as it's called way too often
1352 # my ($self, $label) = @_;
1354 return '' unless defined $_[1];
1355 return ${$_[1]} if ref($_[1]) eq 'SCALAR';
1356 puke 'Identifier cannot be hashref' if ref($_[1]) eq 'HASH';
1358 unless ($_[0]->{quote_char}) {
1359 if (ref($_[1]) eq 'ARRAY') {
1360 return join($_[0]->{name_sep}||'.', @{$_[1]});
1362 $_[0]->_assert_pass_injection_guard($_[1]);
1367 my $qref = ref $_[0]->{quote_char};
1369 !$qref ? ($_[0]->{quote_char}, $_[0]->{quote_char})
1370 : ($qref eq 'ARRAY') ? @{$_[0]->{quote_char}}
1371 : puke "Unsupported quote_char format: $_[0]->{quote_char}";
1373 my $esc = $_[0]->{escape_char} || $r;
1375 # parts containing * are naturally unquoted
1377 $_[0]->{name_sep}||'',
1381 : do { (my $n = $_) =~ s/(\Q$esc\E|\Q$r\E)/$esc$1/g; $l . $n . $r }
1383 (ref($_[1]) eq 'ARRAY'
1387 ? split (/\Q$_[0]->{name_sep}\E/, $_[1] )
1395 # Conversion, if applicable
1397 #my ($self, $arg) = @_;
1398 if ($_[0]->{convert_where}) {
1399 return $_[0]->_sqlcase($_[0]->{convert_where}) .'(' . $_[1] . ')';
1406 #my ($self, $col, @vals) = @_;
1407 # called often - tighten code
1408 return $_[0]->{bindtype} eq 'columns'
1409 ? map {[$_[1], $_]} @_[2 .. $#_]
1414 # Dies if any element of @bind is not in [colname => value] format
1415 # if bindtype is 'columns'.
1416 sub _assert_bindval_matches_bindtype {
1417 # my ($self, @bind) = @_;
1419 if ($self->{bindtype} eq 'columns') {
1421 if (!defined $_ || ref($_) ne 'ARRAY' || @$_ != 2) {
1422 puke "bindtype 'columns' selected, you need to pass: [column_name => bind_value]"
1428 sub _join_sql_clauses {
1429 my ($self, $logic, $clauses_aref, $bind_aref) = @_;
1431 if (@$clauses_aref > 1) {
1432 my $join = " " . $self->_sqlcase($logic) . " ";
1433 my $sql = '( ' . join($join, @$clauses_aref) . ' )';
1434 return ($sql, @$bind_aref);
1436 elsif (@$clauses_aref) {
1437 return ($clauses_aref->[0], @$bind_aref); # no parentheses
1440 return (); # if no SQL, ignore @$bind_aref
1445 # Fix SQL case, if so requested
1447 # LDNOTE: if $self->{case} is true, then it contains 'lower', so we
1448 # don't touch the argument ... crooked logic, but let's not change it!
1449 return $_[0]->{case} ? $_[1] : uc($_[1]);
1453 #======================================================================
1454 # DISPATCHING FROM REFKIND
1455 #======================================================================
1458 my ($self, $data) = @_;
1460 return 'UNDEF' unless defined $data;
1462 # blessed objects are treated like scalars
1463 my $ref = (Scalar::Util::blessed $data) ? '' : ref $data;
1465 return 'SCALAR' unless $ref;
1468 while ($ref eq 'REF') {
1470 $ref = (Scalar::Util::blessed $data) ? '' : ref $data;
1474 return ($ref||'SCALAR') . ('REF' x $n_steps);
1478 my ($self, $data) = @_;
1479 my @try = ($self->_refkind($data));
1480 push @try, 'SCALAR_or_UNDEF' if $try[0] eq 'SCALAR' || $try[0] eq 'UNDEF';
1481 push @try, 'FALLBACK';
1485 sub _METHOD_FOR_refkind {
1486 my ($self, $meth_prefix, $data) = @_;
1489 for (@{$self->_try_refkind($data)}) {
1490 $method = $self->can($meth_prefix."_".$_)
1494 return $method || puke "cannot dispatch on '$meth_prefix' for ".$self->_refkind($data);
1498 sub _SWITCH_refkind {
1499 my ($self, $data, $dispatch_table) = @_;
1502 for (@{$self->_try_refkind($data)}) {
1503 $coderef = $dispatch_table->{$_}
1507 puke "no dispatch entry for ".$self->_refkind($data)
1516 #======================================================================
1517 # VALUES, GENERATE, AUTOLOAD
1518 #======================================================================
1520 # LDNOTE: original code from nwiger, didn't touch code in that section
1521 # I feel the AUTOLOAD stuff should not be the default, it should
1522 # only be activated on explicit demand by user.
1526 my $data = shift || return;
1527 puke "Argument to ", __PACKAGE__, "->values must be a \\%hash"
1528 unless ref $data eq 'HASH';
1531 foreach my $k (sort keys %$data) {
1532 my $v = $data->{$k};
1533 $self->_SWITCH_refkind($v, {
1535 if ($self->{array_datatypes}) { # array datatype
1536 push @all_bind, $self->_bindtype($k, $v);
1538 else { # literal SQL with bind
1539 my ($sql, @bind) = @$v;
1540 $self->_assert_bindval_matches_bindtype(@bind);
1541 push @all_bind, @bind;
1544 ARRAYREFREF => sub { # literal SQL with bind
1545 my ($sql, @bind) = @${$v};
1546 $self->_assert_bindval_matches_bindtype(@bind);
1547 push @all_bind, @bind;
1549 SCALARREF => sub { # literal SQL without bind
1551 SCALAR_or_UNDEF => sub {
1552 push @all_bind, $self->_bindtype($k, $v);
1563 my(@sql, @sqlq, @sqlv);
1567 if ($ref eq 'HASH') {
1568 for my $k (sort keys %$_) {
1571 my $label = $self->_quote($k);
1572 if ($r eq 'ARRAY') {
1573 # literal SQL with bind
1574 my ($sql, @bind) = @$v;
1575 $self->_assert_bindval_matches_bindtype(@bind);
1576 push @sqlq, "$label = $sql";
1578 } elsif ($r eq 'SCALAR') {
1579 # literal SQL without bind
1580 push @sqlq, "$label = $$v";
1582 push @sqlq, "$label = ?";
1583 push @sqlv, $self->_bindtype($k, $v);
1586 push @sql, $self->_sqlcase('set'), join ', ', @sqlq;
1587 } elsif ($ref eq 'ARRAY') {
1588 # unlike insert(), assume these are ONLY the column names, i.e. for SQL
1591 if ($r eq 'ARRAY') { # literal SQL with bind
1592 my ($sql, @bind) = @$v;
1593 $self->_assert_bindval_matches_bindtype(@bind);
1596 } elsif ($r eq 'SCALAR') { # literal SQL without bind
1597 # embedded literal SQL
1604 push @sql, '(' . join(', ', @sqlq) . ')';
1605 } elsif ($ref eq 'SCALAR') {
1609 # strings get case twiddled
1610 push @sql, $self->_sqlcase($_);
1614 my $sql = join ' ', @sql;
1616 # this is pretty tricky
1617 # if ask for an array, return ($stmt, @bind)
1618 # otherwise, s/?/shift @sqlv/ to put it inline
1620 return ($sql, @sqlv);
1622 1 while $sql =~ s/\?/my $d = shift(@sqlv);
1623 ref $d ? $d->[1] : $d/e;
1632 # This allows us to check for a local, then _form, attr
1634 my($name) = $AUTOLOAD =~ /.*::(.+)/;
1635 return $self->generate($name, @_);
1646 SQL::Abstract - Generate SQL from Perl data structures
1652 my $sql = SQL::Abstract->new;
1654 my($stmt, @bind) = $sql->select($source, \@fields, \%where, $order);
1656 my($stmt, @bind) = $sql->insert($table, \%fieldvals || \@values);
1658 my($stmt, @bind) = $sql->update($table, \%fieldvals, \%where);
1660 my($stmt, @bind) = $sql->delete($table, \%where);
1662 # Then, use these in your DBI statements
1663 my $sth = $dbh->prepare($stmt);
1664 $sth->execute(@bind);
1666 # Just generate the WHERE clause
1667 my($stmt, @bind) = $sql->where(\%where, $order);
1669 # Return values in the same order, for hashed queries
1670 # See PERFORMANCE section for more details
1671 my @bind = $sql->values(\%fieldvals);
1675 This module was inspired by the excellent L<DBIx::Abstract>.
1676 However, in using that module I found that what I really wanted
1677 to do was generate SQL, but still retain complete control over my
1678 statement handles and use the DBI interface. So, I set out to
1679 create an abstract SQL generation module.
1681 While based on the concepts used by L<DBIx::Abstract>, there are
1682 several important differences, especially when it comes to WHERE
1683 clauses. I have modified the concepts used to make the SQL easier
1684 to generate from Perl data structures and, IMO, more intuitive.
1685 The underlying idea is for this module to do what you mean, based
1686 on the data structures you provide it. The big advantage is that
1687 you don't have to modify your code every time your data changes,
1688 as this module figures it out.
1690 To begin with, an SQL INSERT is as easy as just specifying a hash
1691 of C<key=value> pairs:
1694 name => 'Jimbo Bobson',
1695 phone => '123-456-7890',
1696 address => '42 Sister Lane',
1697 city => 'St. Louis',
1698 state => 'Louisiana',
1701 The SQL can then be generated with this:
1703 my($stmt, @bind) = $sql->insert('people', \%data);
1705 Which would give you something like this:
1707 $stmt = "INSERT INTO people
1708 (address, city, name, phone, state)
1709 VALUES (?, ?, ?, ?, ?)";
1710 @bind = ('42 Sister Lane', 'St. Louis', 'Jimbo Bobson',
1711 '123-456-7890', 'Louisiana');
1713 These are then used directly in your DBI code:
1715 my $sth = $dbh->prepare($stmt);
1716 $sth->execute(@bind);
1718 =head2 Inserting and Updating Arrays
1720 If your database has array types (like for example Postgres),
1721 activate the special option C<< array_datatypes => 1 >>
1722 when creating the C<SQL::Abstract> object.
1723 Then you may use an arrayref to insert and update database array types:
1725 my $sql = SQL::Abstract->new(array_datatypes => 1);
1727 planets => [qw/Mercury Venus Earth Mars/]
1730 my($stmt, @bind) = $sql->insert('solar_system', \%data);
1734 $stmt = "INSERT INTO solar_system (planets) VALUES (?)"
1736 @bind = (['Mercury', 'Venus', 'Earth', 'Mars']);
1739 =head2 Inserting and Updating SQL
1741 In order to apply SQL functions to elements of your C<%data> you may
1742 specify a reference to an arrayref for the given hash value. For example,
1743 if you need to execute the Oracle C<to_date> function on a value, you can
1744 say something like this:
1748 date_entered => \[ "to_date(?,'MM/DD/YYYY')", "03/02/2003" ],
1751 The first value in the array is the actual SQL. Any other values are
1752 optional and would be included in the bind values array. This gives
1755 my($stmt, @bind) = $sql->insert('people', \%data);
1757 $stmt = "INSERT INTO people (name, date_entered)
1758 VALUES (?, to_date(?,'MM/DD/YYYY'))";
1759 @bind = ('Bill', '03/02/2003');
1761 An UPDATE is just as easy, all you change is the name of the function:
1763 my($stmt, @bind) = $sql->update('people', \%data);
1765 Notice that your C<%data> isn't touched; the module will generate
1766 the appropriately quirky SQL for you automatically. Usually you'll
1767 want to specify a WHERE clause for your UPDATE, though, which is
1768 where handling C<%where> hashes comes in handy...
1770 =head2 Complex where statements
1772 This module can generate pretty complicated WHERE statements
1773 easily. For example, simple C<key=value> pairs are taken to mean
1774 equality, and if you want to see if a field is within a set
1775 of values, you can use an arrayref. Let's say we wanted to
1776 SELECT some data based on this criteria:
1779 requestor => 'inna',
1780 worker => ['nwiger', 'rcwe', 'sfz'],
1781 status => { '!=', 'completed' }
1784 my($stmt, @bind) = $sql->select('tickets', '*', \%where);
1786 The above would give you something like this:
1788 $stmt = "SELECT * FROM tickets WHERE
1789 ( requestor = ? ) AND ( status != ? )
1790 AND ( worker = ? OR worker = ? OR worker = ? )";
1791 @bind = ('inna', 'completed', 'nwiger', 'rcwe', 'sfz');
1793 Which you could then use in DBI code like so:
1795 my $sth = $dbh->prepare($stmt);
1796 $sth->execute(@bind);
1802 The methods are simple. There's one for every major SQL operation,
1803 and a constructor you use first. The arguments are specified in a
1804 similar order for each method (table, then fields, then a where
1805 clause) to try and simplify things.
1807 =head2 new(option => 'value')
1809 The C<new()> function takes a list of options and values, and returns
1810 a new B<SQL::Abstract> object which can then be used to generate SQL
1811 through the methods below. The options accepted are:
1817 If set to 'lower', then SQL will be generated in all lowercase. By
1818 default SQL is generated in "textbook" case meaning something like:
1820 SELECT a_field FROM a_table WHERE some_field LIKE '%someval%'
1822 Any setting other than 'lower' is ignored.
1826 This determines what the default comparison operator is. By default
1827 it is C<=>, meaning that a hash like this:
1829 %where = (name => 'nwiger', email => 'nate@wiger.org');
1831 Will generate SQL like this:
1833 WHERE name = 'nwiger' AND email = 'nate@wiger.org'
1835 However, you may want loose comparisons by default, so if you set
1836 C<cmp> to C<like> you would get SQL such as:
1838 WHERE name like 'nwiger' AND email like 'nate@wiger.org'
1840 You can also override the comparison on an individual basis - see
1841 the huge section on L</"WHERE CLAUSES"> at the bottom.
1843 =item sqltrue, sqlfalse
1845 Expressions for inserting boolean values within SQL statements.
1846 By default these are C<1=1> and C<1=0>. They are used
1847 by the special operators C<-in> and C<-not_in> for generating
1848 correct SQL even when the argument is an empty array (see below).
1852 This determines the default logical operator for multiple WHERE
1853 statements in arrays or hashes. If absent, the default logic is "or"
1854 for arrays, and "and" for hashes. This means that a WHERE
1858 event_date => {'>=', '2/13/99'},
1859 event_date => {'<=', '4/24/03'},
1862 will generate SQL like this:
1864 WHERE event_date >= '2/13/99' OR event_date <= '4/24/03'
1866 This is probably not what you want given this query, though (look
1867 at the dates). To change the "OR" to an "AND", simply specify:
1869 my $sql = SQL::Abstract->new(logic => 'and');
1871 Which will change the above C<WHERE> to:
1873 WHERE event_date >= '2/13/99' AND event_date <= '4/24/03'
1875 The logic can also be changed locally by inserting
1876 a modifier in front of an arrayref:
1878 @where = (-and => [event_date => {'>=', '2/13/99'},
1879 event_date => {'<=', '4/24/03'} ]);
1881 See the L</"WHERE CLAUSES"> section for explanations.
1885 This will automatically convert comparisons using the specified SQL
1886 function for both column and value. This is mostly used with an argument
1887 of C<upper> or C<lower>, so that the SQL will have the effect of
1888 case-insensitive "searches". For example, this:
1890 $sql = SQL::Abstract->new(convert => 'upper');
1891 %where = (keywords => 'MaKe iT CAse inSeNSItive');
1893 Will turn out the following SQL:
1895 WHERE upper(keywords) like upper('MaKe iT CAse inSeNSItive')
1897 The conversion can be C<upper()>, C<lower()>, or any other SQL function
1898 that can be applied symmetrically to fields (actually B<SQL::Abstract> does
1899 not validate this option; it will just pass through what you specify verbatim).
1903 This is a kludge because many databases suck. For example, you can't
1904 just bind values using DBI's C<execute()> for Oracle C<CLOB> or C<BLOB> fields.
1905 Instead, you have to use C<bind_param()>:
1907 $sth->bind_param(1, 'reg data');
1908 $sth->bind_param(2, $lots, {ora_type => ORA_CLOB});
1910 The problem is, B<SQL::Abstract> will normally just return a C<@bind> array,
1911 which loses track of which field each slot refers to. Fear not.
1913 If you specify C<bindtype> in new, you can determine how C<@bind> is returned.
1914 Currently, you can specify either C<normal> (default) or C<columns>. If you
1915 specify C<columns>, you will get an array that looks like this:
1917 my $sql = SQL::Abstract->new(bindtype => 'columns');
1918 my($stmt, @bind) = $sql->insert(...);
1921 [ 'column1', 'value1' ],
1922 [ 'column2', 'value2' ],
1923 [ 'column3', 'value3' ],
1926 You can then iterate through this manually, using DBI's C<bind_param()>.
1928 $sth->prepare($stmt);
1931 my($col, $data) = @$_;
1932 if ($col eq 'details' || $col eq 'comments') {
1933 $sth->bind_param($i, $data, {ora_type => ORA_CLOB});
1934 } elsif ($col eq 'image') {
1935 $sth->bind_param($i, $data, {ora_type => ORA_BLOB});
1937 $sth->bind_param($i, $data);
1941 $sth->execute; # execute without @bind now
1943 Now, why would you still use B<SQL::Abstract> if you have to do this crap?
1944 Basically, the advantage is still that you don't have to care which fields
1945 are or are not included. You could wrap that above C<for> loop in a simple
1946 sub called C<bind_fields()> or something and reuse it repeatedly. You still
1947 get a layer of abstraction over manual SQL specification.
1949 Note that if you set L</bindtype> to C<columns>, the C<\[ $sql, @bind ]>
1950 construct (see L</Literal SQL with placeholders and bind values (subqueries)>)
1951 will expect the bind values in this format.
1955 This is the character that a table or column name will be quoted
1956 with. By default this is an empty string, but you could set it to
1957 the character C<`>, to generate SQL like this:
1959 SELECT `a_field` FROM `a_table` WHERE `some_field` LIKE '%someval%'
1961 Alternatively, you can supply an array ref of two items, the first being the left
1962 hand quote character, and the second the right hand quote character. For
1963 example, you could supply C<['[',']']> for SQL Server 2000 compliant quotes
1964 that generates SQL like this:
1966 SELECT [a_field] FROM [a_table] WHERE [some_field] LIKE '%someval%'
1968 Quoting is useful if you have tables or columns names that are reserved
1969 words in your database's SQL dialect.
1973 This is the character that will be used to escape L</quote_char>s appearing
1974 in an identifier before it has been quoted.
1976 The parameter default in case of a single L</quote_char> character is the quote
1979 When opening-closing-style quoting is used (L</quote_char> is an arrayref)
1980 this parameter defaults to the B<closing (right)> L</quote_char>. Occurrences
1981 of the B<opening (left)> L</quote_char> within the identifier are currently left
1982 untouched. The default for opening-closing-style quotes may change in future
1983 versions, thus you are B<strongly encouraged> to specify the escape character
1988 This is the character that separates a table and column name. It is
1989 necessary to specify this when the C<quote_char> option is selected,
1990 so that tables and column names can be individually quoted like this:
1992 SELECT `table`.`one_field` FROM `table` WHERE `table`.`other_field` = 1
1994 =item injection_guard
1996 A regular expression C<qr/.../> that is applied to any C<-function> and unquoted
1997 column name specified in a query structure. This is a safety mechanism to avoid
1998 injection attacks when mishandling user input e.g.:
2000 my %condition_as_column_value_pairs = get_values_from_user();
2001 $sqla->select( ... , \%condition_as_column_value_pairs );
2003 If the expression matches an exception is thrown. Note that literal SQL
2004 supplied via C<\'...'> or C<\['...']> is B<not> checked in any way.
2006 Defaults to checking for C<;> and the C<GO> keyword (TransactSQL)
2008 =item array_datatypes
2010 When this option is true, arrayrefs in INSERT or UPDATE are
2011 interpreted as array datatypes and are passed directly
2013 When this option is false, arrayrefs are interpreted
2014 as literal SQL, just like refs to arrayrefs
2015 (but this behavior is for backwards compatibility; when writing
2016 new queries, use the "reference to arrayref" syntax
2022 Takes a reference to a list of "special operators"
2023 to extend the syntax understood by L<SQL::Abstract>.
2024 See section L</"SPECIAL OPERATORS"> for details.
2028 Takes a reference to a list of "unary operators"
2029 to extend the syntax understood by L<SQL::Abstract>.
2030 See section L</"UNARY OPERATORS"> for details.
2036 =head2 insert($table, \@values || \%fieldvals, \%options)
2038 This is the simplest function. You simply give it a table name
2039 and either an arrayref of values or hashref of field/value pairs.
2040 It returns an SQL INSERT statement and a list of bind values.
2041 See the sections on L</"Inserting and Updating Arrays"> and
2042 L</"Inserting and Updating SQL"> for information on how to insert
2043 with those data types.
2045 The optional C<\%options> hash reference may contain additional
2046 options to generate the insert SQL. Currently supported options
2053 Takes either a scalar of raw SQL fields, or an array reference of
2054 field names, and adds on an SQL C<RETURNING> statement at the end.
2055 This allows you to return data generated by the insert statement
2056 (such as row IDs) without performing another C<SELECT> statement.
2057 Note, however, this is not part of the SQL standard and may not
2058 be supported by all database engines.
2062 =head2 update($table, \%fieldvals, \%where, \%options)
2064 This takes a table, hashref of field/value pairs, and an optional
2065 hashref L<WHERE clause|/WHERE CLAUSES>. It returns an SQL UPDATE function and a list
2067 See the sections on L</"Inserting and Updating Arrays"> and
2068 L</"Inserting and Updating SQL"> for information on how to insert
2069 with those data types.
2071 The optional C<\%options> hash reference may contain additional
2072 options to generate the update SQL. Currently supported options
2079 See the C<returning> option to
2080 L<insert|/insert($table, \@values || \%fieldvals, \%options)>.
2084 =head2 select($source, $fields, $where, $order)
2086 This returns a SQL SELECT statement and associated list of bind values, as
2087 specified by the arguments:
2093 Specification of the 'FROM' part of the statement.
2094 The argument can be either a plain scalar (interpreted as a table
2095 name, will be quoted), or an arrayref (interpreted as a list
2096 of table names, joined by commas, quoted), or a scalarref
2097 (literal SQL, not quoted).
2101 Specification of the list of fields to retrieve from
2103 The argument can be either an arrayref (interpreted as a list
2104 of field names, will be joined by commas and quoted), or a
2105 plain scalar (literal SQL, not quoted).
2106 Please observe that this API is not as flexible as that of
2107 the first argument C<$source>, for backwards compatibility reasons.
2111 Optional argument to specify the WHERE part of the query.
2112 The argument is most often a hashref, but can also be
2113 an arrayref or plain scalar --
2114 see section L<WHERE clause|/"WHERE CLAUSES"> for details.
2118 Optional argument to specify the ORDER BY part of the query.
2119 The argument can be a scalar, a hashref or an arrayref
2120 -- see section L<ORDER BY clause|/"ORDER BY CLAUSES">
2126 =head2 delete($table, \%where, \%options)
2128 This takes a table name and optional hashref L<WHERE clause|/WHERE CLAUSES>.
2129 It returns an SQL DELETE statement and list of bind values.
2131 The optional C<\%options> hash reference may contain additional
2132 options to generate the delete SQL. Currently supported options
2139 See the C<returning> option to
2140 L<insert|/insert($table, \@values || \%fieldvals, \%options)>.
2144 =head2 where(\%where, $order)
2146 This is used to generate just the WHERE clause. For example,
2147 if you have an arbitrary data structure and know what the
2148 rest of your SQL is going to look like, but want an easy way
2149 to produce a WHERE clause, use this. It returns an SQL WHERE
2150 clause and list of bind values.
2153 =head2 values(\%data)
2155 This just returns the values from the hash C<%data>, in the same
2156 order that would be returned from any of the other above queries.
2157 Using this allows you to markedly speed up your queries if you
2158 are affecting lots of rows. See below under the L</"PERFORMANCE"> section.
2160 =head2 generate($any, 'number', $of, \@data, $struct, \%types)
2162 Warning: This is an experimental method and subject to change.
2164 This returns arbitrarily generated SQL. It's a really basic shortcut.
2165 It will return two different things, depending on return context:
2167 my($stmt, @bind) = $sql->generate('create table', \$table, \@fields);
2168 my $stmt_and_val = $sql->generate('create table', \$table, \@fields);
2170 These would return the following:
2172 # First calling form
2173 $stmt = "CREATE TABLE test (?, ?)";
2174 @bind = (field1, field2);
2176 # Second calling form
2177 $stmt_and_val = "CREATE TABLE test (field1, field2)";
2179 Depending on what you're trying to do, it's up to you to choose the correct
2180 format. In this example, the second form is what you would want.
2184 $sql->generate('alter session', { nls_date_format => 'MM/YY' });
2188 ALTER SESSION SET nls_date_format = 'MM/YY'
2190 You get the idea. Strings get their case twiddled, but everything
2191 else remains verbatim.
2193 =head1 EXPORTABLE FUNCTIONS
2195 =head2 is_plain_value
2197 Determines if the supplied argument is a plain value as understood by this
2202 =item * The value is C<undef>
2204 =item * The value is a non-reference
2206 =item * The value is an object with stringification overloading
2208 =item * The value is of the form C<< { -value => $anything } >>
2212 On failure returns C<undef>, on success returns a B<scalar> reference
2213 to the original supplied argument.
2219 The stringification overloading detection is rather advanced: it takes
2220 into consideration not only the presence of a C<""> overload, but if that
2221 fails also checks for enabled
2222 L<autogenerated versions of C<"">|overload/Magic Autogeneration>, based
2223 on either C<0+> or C<bool>.
2225 Unfortunately testing in the field indicates that this
2226 detection B<< may tickle a latent bug in perl versions before 5.018 >>,
2227 but only when very large numbers of stringifying objects are involved.
2228 At the time of writing ( Sep 2014 ) there is no clear explanation of
2229 the direct cause, nor is there a manageably small test case that reliably
2230 reproduces the problem.
2232 If you encounter any of the following exceptions in B<random places within
2233 your application stack> - this module may be to blame:
2235 Operation "ne": no method found,
2236 left argument in overloaded package <something>,
2237 right argument in overloaded package <something>
2241 Stub found while resolving method "???" overloading """" in package <something>
2243 If you fall victim to the above - please attempt to reduce the problem
2244 to something that could be sent to the L<SQL::Abstract developers
2245 |DBIx::Class/GETTING HELP/SUPPORT>
2246 (either publicly or privately). As a workaround in the meantime you can
2247 set C<$ENV{SQLA_ISVALUE_IGNORE_AUTOGENERATED_STRINGIFICATION}> to a true
2248 value, which will most likely eliminate your problem (at the expense of
2249 not being able to properly detect exotic forms of stringification).
2251 This notice and environment variable will be removed in a future version,
2252 as soon as the underlying problem is found and a reliable workaround is
2257 =head2 is_literal_value
2259 Determines if the supplied argument is a literal value as understood by this
2264 =item * C<\$sql_string>
2266 =item * C<\[ $sql_string, @bind_values ]>
2270 On failure returns C<undef>, on success returns an B<array> reference
2271 containing the unpacked version of the supplied literal SQL and bind values.
2273 =head1 WHERE CLAUSES
2277 This module uses a variation on the idea from L<DBIx::Abstract>. It
2278 is B<NOT>, repeat I<not> 100% compatible. B<The main logic of this
2279 module is that things in arrays are OR'ed, and things in hashes
2282 The easiest way to explain is to show lots of examples. After
2283 each C<%where> hash shown, it is assumed you used:
2285 my($stmt, @bind) = $sql->where(\%where);
2287 However, note that the C<%where> hash can be used directly in any
2288 of the other functions as well, as described above.
2290 =head2 Key-value pairs
2292 So, let's get started. To begin, a simple hash:
2296 status => 'completed'
2299 Is converted to SQL C<key = val> statements:
2301 $stmt = "WHERE user = ? AND status = ?";
2302 @bind = ('nwiger', 'completed');
2304 One common thing I end up doing is having a list of values that
2305 a field can be in. To do this, simply specify a list inside of
2310 status => ['assigned', 'in-progress', 'pending'];
2313 This simple code will create the following:
2315 $stmt = "WHERE user = ? AND ( status = ? OR status = ? OR status = ? )";
2316 @bind = ('nwiger', 'assigned', 'in-progress', 'pending');
2318 A field associated to an empty arrayref will be considered a
2319 logical false and will generate 0=1.
2321 =head2 Tests for NULL values
2323 If the value part is C<undef> then this is converted to SQL <IS NULL>
2332 $stmt = "WHERE user = ? AND status IS NULL";
2335 To test if a column IS NOT NULL:
2339 status => { '!=', undef },
2342 =head2 Specific comparison operators
2344 If you want to specify a different type of operator for your comparison,
2345 you can use a hashref for a given column:
2349 status => { '!=', 'completed' }
2352 Which would generate:
2354 $stmt = "WHERE user = ? AND status != ?";
2355 @bind = ('nwiger', 'completed');
2357 To test against multiple values, just enclose the values in an arrayref:
2359 status => { '=', ['assigned', 'in-progress', 'pending'] };
2361 Which would give you:
2363 "WHERE status = ? OR status = ? OR status = ?"
2366 The hashref can also contain multiple pairs, in which case it is expanded
2367 into an C<AND> of its elements:
2371 status => { '!=', 'completed', -not_like => 'pending%' }
2374 # Or more dynamically, like from a form
2375 $where{user} = 'nwiger';
2376 $where{status}{'!='} = 'completed';
2377 $where{status}{'-not_like'} = 'pending%';
2379 # Both generate this
2380 $stmt = "WHERE user = ? AND status != ? AND status NOT LIKE ?";
2381 @bind = ('nwiger', 'completed', 'pending%');
2384 To get an OR instead, you can combine it with the arrayref idea:
2388 priority => [ { '=', 2 }, { '>', 5 } ]
2391 Which would generate:
2393 $stmt = "WHERE ( priority = ? OR priority > ? ) AND user = ?";
2394 @bind = ('2', '5', 'nwiger');
2396 If you want to include literal SQL (with or without bind values), just use a
2397 scalar reference or reference to an arrayref as the value:
2400 date_entered => { '>' => \["to_date(?, 'MM/DD/YYYY')", "11/26/2008"] },
2401 date_expires => { '<' => \"now()" }
2404 Which would generate:
2406 $stmt = "WHERE date_entered > to_date(?, 'MM/DD/YYYY') AND date_expires < now()";
2407 @bind = ('11/26/2008');
2410 =head2 Logic and nesting operators
2412 In the example above,
2413 there is a subtle trap if you want to say something like
2414 this (notice the C<AND>):
2416 WHERE priority != ? AND priority != ?
2418 Because, in Perl you I<can't> do this:
2420 priority => { '!=' => 2, '!=' => 1 }
2422 As the second C<!=> key will obliterate the first. The solution
2423 is to use the special C<-modifier> form inside an arrayref:
2425 priority => [ -and => {'!=', 2},
2429 Normally, these would be joined by C<OR>, but the modifier tells it
2430 to use C<AND> instead. (Hint: You can use this in conjunction with the
2431 C<logic> option to C<new()> in order to change the way your queries
2432 work by default.) B<Important:> Note that the C<-modifier> goes
2433 B<INSIDE> the arrayref, as an extra first element. This will
2434 B<NOT> do what you think it might:
2436 priority => -and => [{'!=', 2}, {'!=', 1}] # WRONG!
2438 Here is a quick list of equivalencies, since there is some overlap:
2441 status => {'!=', 'completed', 'not like', 'pending%' }
2442 status => [ -and => {'!=', 'completed'}, {'not like', 'pending%'}]
2445 status => {'=', ['assigned', 'in-progress']}
2446 status => [ -or => {'=', 'assigned'}, {'=', 'in-progress'}]
2447 status => [ {'=', 'assigned'}, {'=', 'in-progress'} ]
2451 =head2 Special operators: IN, BETWEEN, etc.
2453 You can also use the hashref format to compare a list of fields using the
2454 C<IN> comparison operator, by specifying the list as an arrayref:
2457 status => 'completed',
2458 reportid => { -in => [567, 2335, 2] }
2461 Which would generate:
2463 $stmt = "WHERE status = ? AND reportid IN (?,?,?)";
2464 @bind = ('completed', '567', '2335', '2');
2466 The reverse operator C<-not_in> generates SQL C<NOT IN> and is used in
2469 If the argument to C<-in> is an empty array, 'sqlfalse' is generated
2470 (by default: C<1=0>). Similarly, C<< -not_in => [] >> generates
2471 'sqltrue' (by default: C<1=1>).
2473 In addition to the array you can supply a chunk of literal sql or
2474 literal sql with bind:
2477 customer => { -in => \[
2478 'SELECT cust_id FROM cust WHERE balance > ?',
2481 status => { -in => \'SELECT status_codes FROM states' },
2487 customer IN ( SELECT cust_id FROM cust WHERE balance > ? )
2488 AND status IN ( SELECT status_codes FROM states )
2492 Finally, if the argument to C<-in> is not a reference, it will be
2493 treated as a single-element array.
2495 Another pair of operators is C<-between> and C<-not_between>,
2496 used with an arrayref of two values:
2500 completion_date => {
2501 -not_between => ['2002-10-01', '2003-02-06']
2507 WHERE user = ? AND completion_date NOT BETWEEN ( ? AND ? )
2509 Just like with C<-in> all plausible combinations of literal SQL
2513 start0 => { -between => [ 1, 2 ] },
2514 start1 => { -between => \["? AND ?", 1, 2] },
2515 start2 => { -between => \"lower(x) AND upper(y)" },
2516 start3 => { -between => [
2518 \["upper(?)", 'stuff' ],
2525 ( start0 BETWEEN ? AND ? )
2526 AND ( start1 BETWEEN ? AND ? )
2527 AND ( start2 BETWEEN lower(x) AND upper(y) )
2528 AND ( start3 BETWEEN lower(x) AND upper(?) )
2530 @bind = (1, 2, 1, 2, 'stuff');
2533 These are the two builtin "special operators"; but the
2534 list can be expanded: see section L</"SPECIAL OPERATORS"> below.
2536 =head2 Unary operators: bool
2538 If you wish to test against boolean columns or functions within your
2539 database you can use the C<-bool> and C<-not_bool> operators. For
2540 example to test the column C<is_user> being true and the column
2541 C<is_enabled> being false you would use:-
2545 -not_bool => 'is_enabled',
2550 WHERE is_user AND NOT is_enabled
2552 If a more complex combination is required, testing more conditions,
2553 then you should use the and/or operators:-
2558 -not_bool => { two=> { -rlike => 'bar' } },
2559 -not_bool => { three => [ { '=', 2 }, { '>', 5 } ] },
2570 (NOT ( three = ? OR three > ? ))
2573 =head2 Nested conditions, -and/-or prefixes
2575 So far, we've seen how multiple conditions are joined with a top-level
2576 C<AND>. We can change this by putting the different conditions we want in
2577 hashes and then putting those hashes in an array. For example:
2582 status => { -like => ['pending%', 'dispatched'] },
2586 status => 'unassigned',
2590 This data structure would create the following:
2592 $stmt = "WHERE ( user = ? AND ( status LIKE ? OR status LIKE ? ) )
2593 OR ( user = ? AND status = ? ) )";
2594 @bind = ('nwiger', 'pending', 'dispatched', 'robot', 'unassigned');
2597 Clauses in hashrefs or arrayrefs can be prefixed with an C<-and> or C<-or>
2598 to change the logic inside:
2604 -and => [ workhrs => {'>', 20}, geo => 'ASIA' ],
2605 -or => { workhrs => {'<', 50}, geo => 'EURO' },
2612 $stmt = "WHERE ( user = ?
2613 AND ( ( workhrs > ? AND geo = ? )
2614 OR ( workhrs < ? OR geo = ? ) ) )";
2615 @bind = ('nwiger', '20', 'ASIA', '50', 'EURO');
2617 =head3 Algebraic inconsistency, for historical reasons
2619 C<Important note>: when connecting several conditions, the C<-and->|C<-or>
2620 operator goes C<outside> of the nested structure; whereas when connecting
2621 several constraints on one column, the C<-and> operator goes
2622 C<inside> the arrayref. Here is an example combining both features:
2625 -and => [a => 1, b => 2],
2626 -or => [c => 3, d => 4],
2627 e => [-and => {-like => 'foo%'}, {-like => '%bar'} ]
2632 WHERE ( ( ( a = ? AND b = ? )
2633 OR ( c = ? OR d = ? )
2634 OR ( e LIKE ? AND e LIKE ? ) ) )
2636 This difference in syntax is unfortunate but must be preserved for
2637 historical reasons. So be careful: the two examples below would
2638 seem algebraically equivalent, but they are not
2641 { -like => 'foo%' },
2642 { -like => '%bar' },
2644 # yields: WHERE ( ( col LIKE ? AND col LIKE ? ) )
2647 { col => { -like => 'foo%' } },
2648 { col => { -like => '%bar' } },
2650 # yields: WHERE ( ( col LIKE ? OR col LIKE ? ) )
2653 =head2 Literal SQL and value type operators
2655 The basic premise of SQL::Abstract is that in WHERE specifications the "left
2656 side" is a column name and the "right side" is a value (normally rendered as
2657 a placeholder). This holds true for both hashrefs and arrayref pairs as you
2658 see in the L</WHERE CLAUSES> examples above. Sometimes it is necessary to
2659 alter this behavior. There are several ways of doing so.
2663 This is a virtual operator that signals the string to its right side is an
2664 identifier (a column name) and not a value. For example to compare two
2665 columns you would write:
2668 priority => { '<', 2 },
2669 requestor => { -ident => 'submitter' },
2674 $stmt = "WHERE priority < ? AND requestor = submitter";
2677 If you are maintaining legacy code you may see a different construct as
2678 described in L</Deprecated usage of Literal SQL>, please use C<-ident> in new
2683 This is a virtual operator that signals that the construct to its right side
2684 is a value to be passed to DBI. This is for example necessary when you want
2685 to write a where clause against an array (for RDBMS that support such
2686 datatypes). For example:
2689 array => { -value => [1, 2, 3] }
2694 $stmt = 'WHERE array = ?';
2695 @bind = ([1, 2, 3]);
2697 Note that if you were to simply say:
2703 the result would probably not be what you wanted:
2705 $stmt = 'WHERE array = ? OR array = ? OR array = ?';
2710 Finally, sometimes only literal SQL will do. To include a random snippet
2711 of SQL verbatim, you specify it as a scalar reference. Consider this only
2712 as a last resort. Usually there is a better way. For example:
2715 priority => { '<', 2 },
2716 requestor => { -in => \'(SELECT name FROM hitmen)' },
2721 $stmt = "WHERE priority < ? AND requestor IN (SELECT name FROM hitmen)"
2724 Note that in this example, you only get one bind parameter back, since
2725 the verbatim SQL is passed as part of the statement.
2729 Never use untrusted input as a literal SQL argument - this is a massive
2730 security risk (there is no way to check literal snippets for SQL
2731 injections and other nastyness). If you need to deal with untrusted input
2732 use literal SQL with placeholders as described next.
2734 =head3 Literal SQL with placeholders and bind values (subqueries)
2736 If the literal SQL to be inserted has placeholders and bind values,
2737 use a reference to an arrayref (yes this is a double reference --
2738 not so common, but perfectly legal Perl). For example, to find a date
2739 in Postgres you can use something like this:
2742 date_column => \[ "= date '2008-09-30' - ?::integer", 10 ]
2747 $stmt = "WHERE ( date_column = date '2008-09-30' - ?::integer )"
2750 Note that you must pass the bind values in the same format as they are returned
2751 by L<where|/where(\%where, $order)>. This means that if you set L</bindtype>
2752 to C<columns>, you must provide the bind values in the
2753 C<< [ column_meta => value ] >> format, where C<column_meta> is an opaque
2754 scalar value; most commonly the column name, but you can use any scalar value
2755 (including references and blessed references), L<SQL::Abstract> will simply
2756 pass it through intact. So if C<bindtype> is set to C<columns> the above
2757 example will look like:
2760 date_column => \[ "= date '2008-09-30' - ?::integer", [ {} => 10 ] ]
2763 Literal SQL is especially useful for nesting parenthesized clauses in the
2764 main SQL query. Here is a first example:
2766 my ($sub_stmt, @sub_bind) = ("SELECT c1 FROM t1 WHERE c2 < ? AND c3 LIKE ?",
2770 bar => \["IN ($sub_stmt)" => @sub_bind],
2775 $stmt = "WHERE (foo = ? AND bar IN (SELECT c1 FROM t1
2776 WHERE c2 < ? AND c3 LIKE ?))";
2777 @bind = (1234, 100, "foo%");
2779 Other subquery operators, like for example C<"E<gt> ALL"> or C<"NOT IN">,
2780 are expressed in the same way. Of course the C<$sub_stmt> and
2781 its associated bind values can be generated through a former call
2784 my ($sub_stmt, @sub_bind)
2785 = $sql->select("t1", "c1", {c2 => {"<" => 100},
2786 c3 => {-like => "foo%"}});
2789 bar => \["> ALL ($sub_stmt)" => @sub_bind],
2792 In the examples above, the subquery was used as an operator on a column;
2793 but the same principle also applies for a clause within the main C<%where>
2794 hash, like an EXISTS subquery:
2796 my ($sub_stmt, @sub_bind)
2797 = $sql->select("t1", "*", {c1 => 1, c2 => \"> t0.c0"});
2798 my %where = ( -and => [
2800 \["EXISTS ($sub_stmt)" => @sub_bind],
2805 $stmt = "WHERE (foo = ? AND EXISTS (SELECT * FROM t1
2806 WHERE c1 = ? AND c2 > t0.c0))";
2810 Observe that the condition on C<c2> in the subquery refers to
2811 column C<t0.c0> of the main query: this is I<not> a bind
2812 value, so we have to express it through a scalar ref.
2813 Writing C<< c2 => {">" => "t0.c0"} >> would have generated
2814 C<< c2 > ? >> with bind value C<"t0.c0"> ... not exactly
2815 what we wanted here.
2817 Finally, here is an example where a subquery is used
2818 for expressing unary negation:
2820 my ($sub_stmt, @sub_bind)
2821 = $sql->where({age => [{"<" => 10}, {">" => 20}]});
2822 $sub_stmt =~ s/^ where //i; # don't want "WHERE" in the subclause
2824 lname => {like => '%son%'},
2825 \["NOT ($sub_stmt)" => @sub_bind],
2830 $stmt = "lname LIKE ? AND NOT ( age < ? OR age > ? )"
2831 @bind = ('%son%', 10, 20)
2833 =head3 Deprecated usage of Literal SQL
2835 Below are some examples of archaic use of literal SQL. It is shown only as
2836 reference for those who deal with legacy code. Each example has a much
2837 better, cleaner and safer alternative that users should opt for in new code.
2843 my %where = ( requestor => \'IS NOT NULL' )
2845 $stmt = "WHERE requestor IS NOT NULL"
2847 This used to be the way of generating NULL comparisons, before the handling
2848 of C<undef> got formalized. For new code please use the superior syntax as
2849 described in L</Tests for NULL values>.
2853 my %where = ( requestor => \'= submitter' )
2855 $stmt = "WHERE requestor = submitter"
2857 This used to be the only way to compare columns. Use the superior L</-ident>
2858 method for all new code. For example an identifier declared in such a way
2859 will be properly quoted if L</quote_char> is properly set, while the legacy
2860 form will remain as supplied.
2864 my %where = ( is_ready => \"", completed => { '>', '2012-12-21' } )
2866 $stmt = "WHERE completed > ? AND is_ready"
2867 @bind = ('2012-12-21')
2869 Using an empty string literal used to be the only way to express a boolean.
2870 For all new code please use the much more readable
2871 L<-bool|/Unary operators: bool> operator.
2877 These pages could go on for a while, since the nesting of the data
2878 structures this module can handle are pretty much unlimited (the
2879 module implements the C<WHERE> expansion as a recursive function
2880 internally). Your best bet is to "play around" with the module a
2881 little to see how the data structures behave, and choose the best
2882 format for your data based on that.
2884 And of course, all the values above will probably be replaced with
2885 variables gotten from forms or the command line. After all, if you
2886 knew everything ahead of time, you wouldn't have to worry about
2887 dynamically-generating SQL and could just hardwire it into your
2890 =head1 ORDER BY CLAUSES
2892 Some functions take an order by clause. This can either be a scalar (just a
2893 column name), a hashref of C<< { -desc => 'col' } >> or C<< { -asc => 'col' }
2894 >>, a scalarref, an arrayref-ref, or an arrayref of any of the previous
2897 Given | Will Generate
2898 ---------------------------------------------------------------
2900 'colA' | ORDER BY colA
2902 [qw/colA colB/] | ORDER BY colA, colB
2904 {-asc => 'colA'} | ORDER BY colA ASC
2906 {-desc => 'colB'} | ORDER BY colB DESC
2908 ['colA', {-asc => 'colB'}] | ORDER BY colA, colB ASC
2910 { -asc => [qw/colA colB/] } | ORDER BY colA ASC, colB ASC
2912 \'colA DESC' | ORDER BY colA DESC
2914 \[ 'FUNC(colA, ?)', $x ] | ORDER BY FUNC(colA, ?)
2915 | /* ...with $x bound to ? */
2918 { -asc => 'colA' }, | colA ASC,
2919 { -desc => [qw/colB/] }, | colB DESC,
2920 { -asc => [qw/colC colD/] },| colC ASC, colD ASC,
2921 \'colE DESC', | colE DESC,
2922 \[ 'FUNC(colF, ?)', $x ], | FUNC(colF, ?)
2923 ] | /* ...with $x bound to ? */
2924 ===============================================================
2928 =head1 SPECIAL OPERATORS
2930 my $sqlmaker = SQL::Abstract->new(special_ops => [
2934 my ($self, $field, $op, $arg) = @_;
2940 handler => 'method_name',
2944 A "special operator" is a SQL syntactic clause that can be
2945 applied to a field, instead of a usual binary operator.
2948 WHERE field IN (?, ?, ?)
2949 WHERE field BETWEEN ? AND ?
2950 WHERE MATCH(field) AGAINST (?, ?)
2952 Special operators IN and BETWEEN are fairly standard and therefore
2953 are builtin within C<SQL::Abstract> (as the overridable methods
2954 C<_where_field_IN> and C<_where_field_BETWEEN>). For other operators,
2955 like the MATCH .. AGAINST example above which is specific to MySQL,
2956 you can write your own operator handlers - supply a C<special_ops>
2957 argument to the C<new> method. That argument takes an arrayref of
2958 operator definitions; each operator definition is a hashref with two
2965 the regular expression to match the operator
2969 Either a coderef or a plain scalar method name. In both cases
2970 the expected return is C<< ($sql, @bind) >>.
2972 When supplied with a method name, it is simply called on the
2973 L<SQL::Abstract> object as:
2975 $self->$method_name($field, $op, $arg)
2979 $field is the LHS of the operator
2980 $op is the part that matched the handler regex
2983 When supplied with a coderef, it is called as:
2985 $coderef->($self, $field, $op, $arg)
2990 For example, here is an implementation
2991 of the MATCH .. AGAINST syntax for MySQL
2993 my $sqlmaker = SQL::Abstract->new(special_ops => [
2995 # special op for MySql MATCH (field) AGAINST(word1, word2, ...)
2996 {regex => qr/^match$/i,
2998 my ($self, $field, $op, $arg) = @_;
2999 $arg = [$arg] if not ref $arg;
3000 my $label = $self->_quote($field);
3001 my ($placeholder) = $self->_convert('?');
3002 my $placeholders = join ", ", (($placeholder) x @$arg);
3003 my $sql = $self->_sqlcase('match') . " ($label) "
3004 . $self->_sqlcase('against') . " ($placeholders) ";
3005 my @bind = $self->_bindtype($field, @$arg);
3006 return ($sql, @bind);
3013 =head1 UNARY OPERATORS
3015 my $sqlmaker = SQL::Abstract->new(unary_ops => [
3019 my ($self, $op, $arg) = @_;
3025 handler => 'method_name',
3029 A "unary operator" is a SQL syntactic clause that can be
3030 applied to a field - the operator goes before the field
3032 You can write your own operator handlers - supply a C<unary_ops>
3033 argument to the C<new> method. That argument takes an arrayref of
3034 operator definitions; each operator definition is a hashref with two
3041 the regular expression to match the operator
3045 Either a coderef or a plain scalar method name. In both cases
3046 the expected return is C<< $sql >>.
3048 When supplied with a method name, it is simply called on the
3049 L<SQL::Abstract> object as:
3051 $self->$method_name($op, $arg)
3055 $op is the part that matched the handler regex
3056 $arg is the RHS or argument of the operator
3058 When supplied with a coderef, it is called as:
3060 $coderef->($self, $op, $arg)
3068 Thanks to some benchmarking by Mark Stosberg, it turns out that
3069 this module is many orders of magnitude faster than using C<DBIx::Abstract>.
3070 I must admit this wasn't an intentional design issue, but it's a
3071 byproduct of the fact that you get to control your C<DBI> handles
3074 To maximize performance, use a code snippet like the following:
3076 # prepare a statement handle using the first row
3077 # and then reuse it for the rest of the rows
3079 for my $href (@array_of_hashrefs) {
3080 $stmt ||= $sql->insert('table', $href);
3081 $sth ||= $dbh->prepare($stmt);
3082 $sth->execute($sql->values($href));
3085 The reason this works is because the keys in your C<$href> are sorted
3086 internally by B<SQL::Abstract>. Thus, as long as your data retains
3087 the same structure, you only have to generate the SQL the first time
3088 around. On subsequent queries, simply use the C<values> function provided
3089 by this module to return your values in the correct order.
3091 However this depends on the values having the same type - if, for
3092 example, the values of a where clause may either have values
3093 (resulting in sql of the form C<column = ?> with a single bind
3094 value), or alternatively the values might be C<undef> (resulting in
3095 sql of the form C<column IS NULL> with no bind value) then the
3096 caching technique suggested will not work.
3100 If you use my C<CGI::FormBuilder> module at all, you'll hopefully
3101 really like this part (I do, at least). Building up a complex query
3102 can be as simple as the following:
3109 use CGI::FormBuilder;
3112 my $form = CGI::FormBuilder->new(...);
3113 my $sql = SQL::Abstract->new;
3115 if ($form->submitted) {
3116 my $field = $form->field;
3117 my $id = delete $field->{id};
3118 my($stmt, @bind) = $sql->update('table', $field, {id => $id});
3121 Of course, you would still have to connect using C<DBI> to run the
3122 query, but the point is that if you make your form look like your
3123 table, the actual query script can be extremely simplistic.
3125 If you're B<REALLY> lazy (I am), check out C<HTML::QuickTable> for
3126 a fast interface to returning and formatting data. I frequently
3127 use these three modules together to write complex database query
3128 apps in under 50 lines.
3130 =head1 HOW TO CONTRIBUTE
3132 Contributions are always welcome, in all usable forms (we especially
3133 welcome documentation improvements). The delivery methods include git-
3134 or unified-diff formatted patches, GitHub pull requests, or plain bug
3135 reports either via RT or the Mailing list. Contributors are generally
3136 granted full access to the official repository after their first several
3137 patches pass successful review.
3139 This project is maintained in a git repository. The code and related tools are
3140 accessible at the following locations:
3144 =item * Official repo: L<git://git.shadowcat.co.uk/dbsrgits/SQL-Abstract.git>
3146 =item * Official gitweb: L<http://git.shadowcat.co.uk/gitweb/gitweb.cgi?p=dbsrgits/SQL-Abstract.git>
3148 =item * GitHub mirror: L<https://github.com/dbsrgits/sql-abstract>
3150 =item * Authorized committers: L<ssh://dbsrgits@git.shadowcat.co.uk/SQL-Abstract.git>
3156 Version 1.50 was a major internal refactoring of C<SQL::Abstract>.
3157 Great care has been taken to preserve the I<published> behavior
3158 documented in previous versions in the 1.* family; however,
3159 some features that were previously undocumented, or behaved
3160 differently from the documentation, had to be changed in order
3161 to clarify the semantics. Hence, client code that was relying
3162 on some dark areas of C<SQL::Abstract> v1.*
3163 B<might behave differently> in v1.50.
3165 The main changes are:
3171 support for literal SQL through the C<< \ [ $sql, @bind ] >> syntax.
3175 support for the { operator => \"..." } construct (to embed literal SQL)
3179 support for the { operator => \["...", @bind] } construct (to embed literal SQL with bind values)
3183 optional support for L<array datatypes|/"Inserting and Updating Arrays">
3187 defensive programming: check arguments
3191 fixed bug with global logic, which was previously implemented
3192 through global variables yielding side-effects. Prior versions would
3193 interpret C<< [ {cond1, cond2}, [cond3, cond4] ] >>
3194 as C<< "(cond1 AND cond2) OR (cond3 AND cond4)" >>.
3195 Now this is interpreted
3196 as C<< "(cond1 AND cond2) OR (cond3 OR cond4)" >>.
3201 fixed semantics of _bindtype on array args
3205 dropped the C<_anoncopy> of the %where tree. No longer necessary,
3206 we just avoid shifting arrays within that tree.
3210 dropped the C<_modlogic> function
3214 =head1 ACKNOWLEDGEMENTS
3216 There are a number of individuals that have really helped out with
3217 this module. Unfortunately, most of them submitted bugs via CPAN
3218 so I have no idea who they are! But the people I do know are:
3220 Ash Berlin (order_by hash term support)
3221 Matt Trout (DBIx::Class support)
3222 Mark Stosberg (benchmarking)
3223 Chas Owens (initial "IN" operator support)
3224 Philip Collins (per-field SQL functions)
3225 Eric Kolve (hashref "AND" support)
3226 Mike Fragassi (enhancements to "BETWEEN" and "LIKE")
3227 Dan Kubb (support for "quote_char" and "name_sep")
3228 Guillermo Roditi (patch to cleanup "IN" and "BETWEEN", fix and tests for _order_by)
3229 Laurent Dami (internal refactoring, extensible list of special operators, literal SQL)
3230 Norbert Buchmuller (support for literal SQL in hashpair, misc. fixes & tests)
3231 Peter Rabbitson (rewrite of SQLA::Test, misc. fixes & tests)
3232 Oliver Charles (support for "RETURNING" after "INSERT")
3238 L<DBIx::Class>, L<DBIx::Abstract>, L<CGI::FormBuilder>, L<HTML::QuickTable>.
3242 Copyright (c) 2001-2007 Nathan Wiger <nwiger@cpan.org>. All Rights Reserved.
3244 This module is actively maintained by Matt Trout <mst@shadowcatsystems.co.uk>
3246 For support, your best bet is to try the C<DBIx::Class> users mailing list.
3247 While not an official support venue, C<DBIx::Class> makes heavy use of
3248 C<SQL::Abstract>, and as such list members there are very familiar with
3249 how to create queries.
3253 This module is free software; you may copy this under the same
3254 terms as perl itself (either the GNU General Public License or
3255 the Artistic License)