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 +($_ => '_render_unop_postfix'),
234 'is_null', 'is_not_null', 'asc', 'desc',
236 (not => '_render_unop_paren'),
237 (map +($_ => '_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 $data = shift || return;
269 my ($f_aqt, $v_aqt) = $self->_expand_insert_values($data);
271 my @parts = ([ $self->_sqlcase('insert into').' '.$table ]);
272 push @parts, [ $self->render_aqt($f_aqt) ] if $f_aqt;
273 push @parts, [ $self->render_aqt($v_aqt) ];
274 $parts[-1][0] =~ s/^/VALUES /;
276 if ($options->{returning}) {
277 push @parts, [ $self->_insert_returning($options) ];
280 my ($sql, @bind) = $self->_join_parts(' ', @parts);
282 return wantarray ? ($sql, @bind) : $sql;
285 sub _expand_insert_values {
286 my ($self, $data) = @_;
287 if (is_literal_value($data)) {
288 (undef, $self->expand_expr($data));
290 my ($fields, $values) = (
291 ref($data) eq 'HASH' ?
292 ([ sort keys %$data ], [ @{$data}{sort keys %$data} ])
296 # no names (arrayref) means can't generate bindtype
297 !($fields) && $self->{bindtype} eq 'columns'
298 && belch "can't do 'columns' bindtype when called with arrayref";
302 ? $self->expand_expr({ -row => $fields }, -ident)
307 local our $Cur_Col_Meta = $fields->[$_];
308 $self->_expand_insert_value($values->[$_])
315 # So that subclasses can override INSERT ... RETURNING separately from
316 # UPDATE and DELETE (e.g. DBIx::Class::SQLMaker::Oracle does this)
317 sub _insert_returning { shift->_returning(@_) }
320 my ($self, $options) = @_;
322 my $f = $options->{returning};
324 my ($sql, @bind) = $self->render_aqt(
325 $self->_expand_maybe_list_expr($f, -ident)
328 ? $self->_sqlcase(' returning ') . $sql
329 : ($self->_sqlcase(' returning ').$sql, @bind);
332 sub _expand_insert_value {
335 my $k = our $Cur_Col_Meta;
337 if (ref($v) eq 'ARRAY') {
338 if ($self->{array_datatypes}) {
339 return +{ -bind => [ $k, $v ] };
341 my ($sql, @bind) = @$v;
342 $self->_assert_bindval_matches_bindtype(@bind);
343 return +{ -literal => $v };
345 if (ref($v) eq 'HASH') {
346 if (grep !/^-/, keys %$v) {
347 belch "HASH ref as bind value in insert is not supported";
348 return +{ -bind => [ $k, $v ] };
352 return +{ -bind => [ $k, undef ] };
354 return $self->expand_expr($v);
359 #======================================================================
361 #======================================================================
366 my $table = $self->_table(shift);
367 my $data = shift || return;
371 # first build the 'SET' part of the sql statement
372 puke "Unsupported data type specified to \$sql->update"
373 unless ref $data eq 'HASH';
375 my ($sql, @all_bind) = $self->_update_set_values($data);
376 $sql = $self->_sqlcase('update ') . $table . $self->_sqlcase(' set ')
380 my($where_sql, @where_bind) = $self->where($where);
382 push @all_bind, @where_bind;
385 if ($options->{returning}) {
386 my ($returning_sql, @returning_bind) = $self->_update_returning($options);
387 $sql .= $returning_sql;
388 push @all_bind, @returning_bind;
391 return wantarray ? ($sql, @all_bind) : $sql;
394 sub _update_set_values {
395 my ($self, $data) = @_;
397 return $self->render_aqt(
398 $self->_expand_update_set_values($data),
402 sub _expand_update_set_values {
403 my ($self, $data) = @_;
404 $self->_expand_maybe_list_expr( [
407 $set = { -bind => $_ } unless defined $set;
408 +{ -op => [ '=', $self->_expand_ident(-ident => $k), $set ] };
414 ? ($self->{array_datatypes}
415 ? [ $k, +{ -bind => [ $k, $v ] } ]
416 : [ $k, +{ -literal => $v } ])
418 local our $Cur_Col_Meta = $k;
419 [ $k, $self->_expand_expr($v) ]
426 # So that subclasses can override UPDATE ... RETURNING separately from
428 sub _update_returning { shift->_returning(@_) }
432 #======================================================================
434 #======================================================================
439 my $table = $self->_table(shift);
440 my $fields = shift || '*';
444 my ($fields_sql, @bind) = $self->_select_fields($fields);
446 my ($where_sql, @where_bind) = $self->where($where, $order);
447 push @bind, @where_bind;
449 my $sql = join(' ', $self->_sqlcase('select'), $fields_sql,
450 $self->_sqlcase('from'), $table)
453 return wantarray ? ($sql, @bind) : $sql;
457 my ($self, $fields) = @_;
458 return $fields unless ref($fields);
459 return $self->render_aqt(
460 $self->_expand_maybe_list_expr($fields, '-ident')
464 #======================================================================
466 #======================================================================
471 my $table = $self->_table(shift);
475 my($where_sql, @bind) = $self->where($where);
476 my $sql = $self->_sqlcase('delete from ') . $table . $where_sql;
478 if ($options->{returning}) {
479 my ($returning_sql, @returning_bind) = $self->_delete_returning($options);
480 $sql .= $returning_sql;
481 push @bind, @returning_bind;
484 return wantarray ? ($sql, @bind) : $sql;
487 # So that subclasses can override DELETE ... RETURNING separately from
489 sub _delete_returning { shift->_returning(@_) }
493 #======================================================================
495 #======================================================================
499 # Finally, a separate routine just to handle WHERE clauses
501 my ($self, $where, $order) = @_;
503 local $self->{convert_where} = $self->{convert};
506 my ($sql, @bind) = defined($where)
507 ? $self->_recurse_where($where)
509 $sql = (defined $sql and length $sql) ? $self->_sqlcase(' where ') . "( $sql )" : '';
513 my ($order_sql, @order_bind) = $self->_order_by($order);
515 push @bind, @order_bind;
518 return wantarray ? ($sql, @bind) : $sql;
521 { our $Default_Scalar_To = -value }
524 my ($self, $expr, $default_scalar_to) = @_;
525 local our $Default_Scalar_To = $default_scalar_to if $default_scalar_to;
526 $self->_expand_expr($expr);
530 my ($self, $aqt) = @_;
531 my ($k, $v, @rest) = %$aqt;
533 die "Not a node type: $k" unless $k =~ s/^-//;
534 if (my $meth = $self->{render}{$k}) {
535 return $self->$meth($v);
537 die "notreached: $k";
541 my ($self, $expr, $default_scalar_to) = @_;
542 $self->render_aqt($self->expand_expr($expr, $default_scalar_to));
546 my ($self, $raw) = @_;
547 s/^-(?=.)//, s/\s+/_/g for my $op = lc $raw;
552 my ($self, $expr) = @_;
553 our $Expand_Depth ||= 0; local $Expand_Depth = $Expand_Depth + 1;
554 return undef unless defined($expr);
555 if (ref($expr) eq 'HASH') {
556 return undef unless my $kc = keys %$expr;
558 return $self->_expand_op_andor(and => $expr);
560 my ($key, $value) = %$expr;
561 if ($key =~ /^-/ and $key =~ s/ [_\s]? \d+ $//x ) {
562 belch 'Use of [and|or|nest]_N modifiers is deprecated and will be removed in SQLA v2.0. '
563 . "You probably wanted ...-and => [ $key => COND1, $key => COND2 ... ]";
565 return $self->_expand_hashpair($key, $value);
567 if (ref($expr) eq 'ARRAY') {
568 return $self->_expand_op_andor(lc($self->{logic}), $expr);
570 if (my $literal = is_literal_value($expr)) {
571 return +{ -literal => $literal };
573 if (!ref($expr) or Scalar::Util::blessed($expr)) {
574 return $self->_expand_scalar($expr);
579 sub _expand_hashpair {
580 my ($self, $k, $v) = @_;
581 unless (defined($k) and length($k)) {
582 if (defined($k) and my $literal = is_literal_value($v)) {
583 belch 'Hash-pairs consisting of an empty string with a literal are deprecated, and will be removed in 2.0: use -and => [ $literal ] instead';
584 return { -literal => $literal };
586 puke "Supplying an empty left hand side argument is not supported";
589 return $self->_expand_hashpair_op($k, $v);
591 return $self->_expand_hashpair_ident($k, $v);
594 sub _expand_hashpair_ident {
595 my ($self, $k, $v) = @_;
597 local our $Cur_Col_Meta = $k;
599 # hash with multiple or no elements is andor
601 if (ref($v) eq 'HASH' and keys %$v != 1) {
602 return $self->_expand_op_andor(and => $v, $k);
605 # undef needs to be re-sent with cmp to achieve IS/IS NOT NULL
607 if (is_undef_value($v)) {
608 return $self->_expand_hashpair_cmp($k => undef);
611 # scalars and objects get expanded as whatever requested or values
613 if (!ref($v) or Scalar::Util::blessed($v)) {
614 return $self->_expand_hashpair_scalar($k, $v);
617 # single key hashref is a hashtriple
619 if (ref($v) eq 'HASH') {
620 return $self->_expand_hashtriple($k, %$v);
623 # arrayref needs re-engineering over the elements
625 if (ref($v) eq 'ARRAY') {
626 return $self->sqlfalse unless @$v;
627 $self->_debug("ARRAY($k) means distribute over elements");
629 $v->[0] =~ /^-(and|or)$/i
630 ? (shift(@{$v = [ @$v ]}), $1)
631 : lc($self->{logic} || 'OR')
633 return $self->_expand_op_andor(
638 if (my $literal = is_literal_value($v)) {
640 belch 'Hash-pairs consisting of an empty string with a literal are deprecated, and will be removed in 2.0: use -and => [ $literal ] instead';
643 my ($sql, @bind) = @$literal;
644 if ($self->{bindtype} eq 'columns') {
646 $self->_assert_bindval_matches_bindtype($_);
649 return +{ -literal => [ $self->_quote($k).' '.$sql, @bind ] };
655 my ($self, $expr) = @_;
657 return $self->_expand_expr({ (our $Default_Scalar_To) => $expr });
660 sub _expand_hashpair_scalar {
661 my ($self, $k, $v) = @_;
663 return $self->_expand_hashpair_cmp(
664 $k, $self->_expand_scalar($v),
668 sub _expand_hashpair_op {
669 my ($self, $k, $v) = @_;
671 $self->_assert_pass_injection_guard($k =~ /\A-(.*)\Z/s);
673 my $op = $self->_normalize_op($k);
675 if (my $exp = $self->{expand}{$op}) {
676 return $self->$exp($op, $v);
679 # Ops prefixed with -not_ get converted
681 if (my ($rest) = $op =~/^not_(.*)$/) {
684 $self->_expand_expr({ "-${rest}", $v })
690 my $op = join(' ', split '_', $op);
692 # the old special op system requires illegality for top-level use
695 (our $Expand_Depth) == 1
697 List::Util::first { $op =~ $_->{regex} } @{$self->{special_ops}}
699 $self->{disable_old_special_ops}
700 and List::Util::first { $op =~ $_->{regex} } @BUILTIN_SPECIAL_OPS
704 puke "Illegal use of top-level '-$op'"
707 # the old unary op system means we should touch nothing and let it work
709 if (my $us = List::Util::first { $op =~ $_->{regex} } @{$self->{unary_ops}}) {
710 return { -op => [ $op, $v ] };
714 # an explicit node type is currently assumed to be expanded (this is almost
715 # certainly wrong and there should be expansion anyway)
717 if ($self->{render}{$op}) {
721 # hashref RHS values get expanded and used as op/func args
726 and (keys %$v)[0] =~ /^-/
728 my ($func) = $k =~ /^-(.*)$/;
730 if (List::Util::first { $func =~ $_->{regex} } @{$self->{special_ops}}) {
731 return +{ -op => [ $func, $self->_expand_expr($v) ] };
734 return +{ -func => [ $func, $self->_expand_expr($v) ] };
737 # scalars and literals get simply expanded
739 if (!ref($v) or is_literal_value($v)) {
740 return +{ -op => [ $op, $self->_expand_expr($v) ] };
746 sub _expand_hashpair_cmp {
747 my ($self, $k, $v) = @_;
748 $self->_expand_hashtriple($k, $self->{cmp}, $v);
751 sub _expand_hashtriple {
752 my ($self, $k, $vk, $vv) = @_;
754 my $ik = $self->_expand_ident(-ident => $k);
756 my $op = $self->_normalize_op($vk);
757 $self->_assert_pass_injection_guard($op);
759 if ($op =~ s/ _? \d+ $//x ) {
760 return $self->_expand_expr($k, { $vk, $vv });
762 if (my $x = $self->{expand_op}{$op}) {
763 local our $Cur_Col_Meta = $k;
764 return $self->$x($op, $vv, $k);
768 my $op = join(' ', split '_', $op);
770 if (my $us = List::Util::first { $op =~ $_->{regex} } @{$self->{special_ops}}) {
771 return { -op => [ $op, $ik, $vv ] };
773 if (my $us = List::Util::first { $op =~ $_->{regex} } @{$self->{unary_ops}}) {
777 { -op => [ $op, $vv ] }
781 if (ref($vv) eq 'ARRAY') {
783 my $logic = (defined($raw[0]) and $raw[0] =~ /^-(and|or)$/i)
784 ? (shift(@raw), $1) : 'or';
785 my @values = map +{ $vk => $_ }, @raw;
787 $op =~ $self->{inequality_op}
788 or $op =~ $self->{not_like_op}
790 if (lc($logic) eq 'or' and @values > 1) {
791 belch "A multi-element arrayref as an argument to the inequality op '${\uc(join ' ', split '_', $op)}' "
792 . 'is technically equivalent to an always-true 1=1 (you probably wanted '
793 . "to say ...{ \$inequality_op => [ -and => \@values ] }... instead)"
798 # try to DWIM on equality operators
799 return ($self->_dwim_op_to_is($op,
800 "Supplying an empty arrayref to '%s' is deprecated",
801 "operator '%s' applied on an empty array (field '$k')"
802 ) ? $self->sqlfalse : $self->sqltrue);
804 return $self->_expand_op_andor($logic => \@values, $k);
806 if (is_undef_value($vv)) {
807 my $is = ($self->_dwim_op_to_is($op,
808 "Supplying an undefined argument to '%s' is deprecated",
809 "unexpected operator '%s' with undef operand",
810 ) ? 'is' : 'is not');
812 return $self->_expand_hashpair($k => { $is, undef });
814 local our $Cur_Col_Meta = $k;
818 $self->_expand_expr($vv)
823 my ($self, $raw, $empty, $fail) = @_;
825 my $op = $self->_normalize_op($raw);
827 if ($op =~ /^not$/i) {
830 if ($op =~ $self->{equality_op}) {
833 if ($op =~ $self->{like_op}) {
834 belch(sprintf $empty, uc(join ' ', split '_', $op));
837 if ($op =~ $self->{inequality_op}) {
840 if ($op =~ $self->{not_like_op}) {
841 belch(sprintf $empty, uc(join ' ', split '_', $op));
844 puke(sprintf $fail, $op);
848 my ($self, undef, $body, $k) = @_;
849 return $self->_expand_hashpair_cmp(
850 $k, { -ident => $body }
852 unless (defined($body) or (ref($body) and ref($body) eq 'ARRAY')) {
853 puke "-ident requires a single plain scalar argument (a quotable identifier) or an arrayref of identifier parts";
855 my @parts = map split(/\Q${\($self->{name_sep}||'.')}\E/, $_),
856 ref($body) ? @$body : $body;
857 return { -ident => $parts[-1] } if $self->{_dequalify_idents};
858 unless ($self->{quote_char}) {
859 $self->_assert_pass_injection_guard($_) for @parts;
861 return +{ -ident => \@parts };
865 return $_[0]->_expand_hashpair_cmp(
866 $_[3], { -value => $_[2] },
868 +{ -bind => [ our $Cur_Col_Meta, $_[2] ] };
872 +{ -op => [ 'not', $_[0]->_expand_expr($_[2]) ] };
876 my ($self, undef, $args) = @_;
877 +{ -row => [ map $self->expand_expr($_), @$args ] };
881 my ($self, undef, $args) = @_;
882 my ($op, @opargs) = @$args;
883 if (my $exp = $self->{expand_op}{$op}) {
884 return $self->$exp($op, \@opargs);
886 +{ -op => [ $op, map $self->expand_expr($_), @opargs ] };
890 my ($self, undef, $v) = @_;
892 return $self->_expand_expr($v);
894 puke "-bool => undef not supported" unless defined($v);
895 return $self->_expand_ident(-ident => $v);
898 sub _expand_op_andor {
899 my ($self, $logop, $v, $k) = @_;
901 $v = [ map +{ $k, $_ },
903 ? (map +{ $_ => $v->{$_} }, sort keys %$v)
907 if (ref($v) eq 'HASH') {
908 return undef unless keys %$v;
911 map $self->_expand_expr({ $_ => $v->{$_} }),
915 if (ref($v) eq 'ARRAY') {
916 $logop eq 'and' or $logop eq 'or' or puke "unknown logic: $logop";
919 (ref($_) eq 'ARRAY' and @$_)
920 or (ref($_) eq 'HASH' and %$_)
926 while (my ($el) = splice @expr, 0, 1) {
927 puke "Supplying an empty left hand side argument is not supported in array-pairs"
928 unless defined($el) and length($el);
929 my $elref = ref($el);
931 local our $Expand_Depth = 0;
932 push(@res, grep defined, $self->_expand_expr({ $el, shift(@expr) }));
933 } elsif ($elref eq 'ARRAY') {
934 push(@res, grep defined, $self->_expand_expr($el)) if @$el;
935 } elsif (my $l = is_literal_value($el)) {
936 push @res, { -literal => $l };
937 } elsif ($elref eq 'HASH') {
938 local our $Expand_Depth = 0;
939 push @res, grep defined, $self->_expand_expr($el) if %$el;
945 # return $res[0] if @res == 1;
946 return { -op => [ $logop, @res ] };
952 my ($self, $op, $vv, $k) = @_;
953 ($k, $vv) = @$vv unless defined $k;
954 puke "$op can only take undef as argument"
958 and exists($vv->{-value})
959 and !defined($vv->{-value})
961 return +{ -op => [ $op.'_null', $self->expand_expr($k, -ident) ] };
964 sub _expand_between {
965 my ($self, $op, $vv, $k) = @_;
966 $k = shift @{$vv = [ @$vv ]} unless defined $k;
967 my @rhs = map $self->_expand_expr($_),
968 ref($vv) eq 'ARRAY' ? @$vv : $vv;
970 (@rhs == 1 and ref($rhs[0]) eq 'HASH' and $rhs[0]->{-literal})
972 (@rhs == 2 and defined($rhs[0]) and defined($rhs[1]))
974 puke "Operator '${\uc($op)}' requires either an arrayref with two defined values or expressions, or a single literal scalarref/arrayref-ref";
978 $self->expand_expr(ref($k) ? $k : { -ident => $k }),
984 my ($self, $op, $vv, $k) = @_;
985 $k = shift @{$vv = [ @$vv ]} unless defined $k;
986 if (my $literal = is_literal_value($vv)) {
987 my ($sql, @bind) = @$literal;
988 my $opened_sql = $self->_open_outer_paren($sql);
990 $op, $self->expand_expr($k, -ident),
991 { -literal => [ $opened_sql, @bind ] }
995 'SQL::Abstract before v1.75 used to generate incorrect SQL when the '
996 . "-${\uc($op)} operator was given an undef-containing list: !!!AUDIT YOUR CODE "
997 . 'AND DATA!!! (the upcoming Data::Query-based version of SQL::Abstract '
998 . 'will emit the logically correct SQL instead of raising this exception)'
1000 puke("Argument passed to the '${\uc($op)}' operator can not be undefined")
1002 my @rhs = map $self->expand_expr($_, -value),
1003 map { defined($_) ? $_: puke($undef_err) }
1004 (ref($vv) eq 'ARRAY' ? @$vv : $vv);
1005 return $self->${\($op =~ /^not/ ? 'sqltrue' : 'sqlfalse')} unless @rhs;
1009 $self->expand_expr($k, -ident),
1015 my ($self, undef, $v) = @_;
1016 # DBIx::Class requires a nest warning to be emitted once but the private
1017 # method it overrode to do so no longer exists
1018 if ($self->{warn_once_on_nest}) {
1019 unless (our $Nest_Warned) {
1021 "-nest in search conditions is deprecated, you most probably wanted:\n"
1022 .q|{..., -and => [ \%cond0, \@cond1, \'cond2', \[ 'cond3', [ col => bind ] ], etc. ], ... }|
1027 return $self->_expand_expr($v);
1031 my ($self, undef, $bind) = @_;
1032 return { -bind => $bind };
1035 sub _recurse_where {
1036 my ($self, $where, $logic) = @_;
1038 # Special case: top level simple string treated as literal
1040 my $where_exp = (ref($where)
1041 ? $self->_expand_expr($where, $logic)
1042 : { -literal => [ $where ] });
1044 # dispatch expanded expression
1046 my ($sql, @bind) = defined($where_exp) ? $self->render_aqt($where_exp) : (undef);
1047 # DBIx::Class used to call _recurse_where in scalar context
1048 # something else might too...
1050 return ($sql, @bind);
1053 belch "Calling _recurse_where in scalar context is deprecated and will go away before 2.0";
1059 my ($self, $ident) = @_;
1061 return $self->_convert($self->_quote($ident));
1065 my ($self, $values) = @_;
1066 my ($sql, @bind) = $self->_render_op([ ',', @$values ]);
1067 return "($sql)", @bind;
1071 my ($self, $rest) = @_;
1072 my ($func, @args) = @$rest;
1076 push @arg_sql, shift @x;
1078 } map [ $self->render_aqt($_) ], @args;
1079 return ($self->_sqlcase($func).'('.join(', ', @arg_sql).')', @bind);
1083 my ($self, $bind) = @_;
1084 return ($self->_convert('?'), $self->_bindtype(@$bind));
1087 sub _render_literal {
1088 my ($self, $literal) = @_;
1089 $self->_assert_bindval_matches_bindtype(@{$literal}[1..$#$literal]);
1094 my ($self, $v) = @_;
1095 my ($op, @args) = @$v;
1096 if (my $r = $self->{render_op}{$op}) {
1097 return $self->$r($op, \@args);
1102 my $op = join(' ', split '_', $op);
1104 my $ss = List::Util::first { $op =~ $_->{regex} } @{$self->{special_ops}};
1105 if ($ss and @args > 1) {
1106 puke "Special op '${op}' requires first value to be identifier"
1107 unless my ($ident) = map $_->{-ident}, grep ref($_) eq 'HASH', $args[0];
1108 my $k = join(($self->{name_sep}||'.'), @$ident);
1109 local our $Expand_Depth = 1;
1110 return $self->${\($ss->{handler})}($k, $op, $args[1]);
1112 if (my $us = List::Util::first { $op =~ $_->{regex} } @{$self->{unary_ops}}) {
1113 return $self->${\($us->{handler})}($op, $args[0]);
1116 return $self->_render_unop_paren($op, \@args);
1120 return $self->_render_unop_prefix($op, \@args);
1122 return $self->_render_op_multop($op, \@args);
1128 sub _render_op_between {
1129 my ($self, $op, $args) = @_;
1130 my ($left, $low, $high) = @$args;
1131 my ($rhsql, @rhbind) = do {
1133 puke "Single arg to between must be a literal"
1134 unless $low->{-literal};
1137 my ($l, $h) = map [ $self->render_aqt($_) ], $low, $high;
1138 (join(' ', $l->[0], $self->_sqlcase('and'), $h->[0]),
1139 @{$l}[1..$#$l], @{$h}[1..$#$h])
1142 my ($lhsql, @lhbind) = $self->render_aqt($left);
1146 $self->_sqlcase(join ' ', split '_', $op),
1154 my ($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(join ' ', split '_', $op).' ( '
1165 .join(', ', @in_sql)
1171 sub _render_op_andor {
1172 my ($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) = $self->_join_parts(' '.$self->_sqlcase($op).' ', @parts);
1177 return '( '.$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 $join = ($op eq ','
1187 : ' '.$self->_sqlcase(join ' ', split '_', $op).' '
1189 return $self->_join_parts($join, @parts);
1193 my ($self, $join, @parts) = @_;
1195 join($join, map $_->[0], @parts),
1196 map @{$_}[1..$#$_], @parts
1200 sub _render_unop_paren {
1201 my ($self, $op, $v) = @_;
1202 my ($sql, @bind) = $self->_render_unop_prefix($op, $v);
1203 return "(${sql})", @bind;
1206 sub _render_unop_prefix {
1207 my ($self, $op, $v) = @_;
1208 my ($expr_sql, @bind) = $self->render_aqt($v->[0]);
1210 my $op_sql = $self->_sqlcase($op); # join ' ', split '_', $op);
1211 return ("${op_sql} ${expr_sql}", @bind);
1214 sub _render_unop_postfix {
1215 my ($self, $op, $v) = @_;
1216 my ($expr_sql, @bind) = $self->render_aqt($v->[0]);
1217 my $op_sql = $self->_sqlcase(join ' ', split '_', $op);
1218 return ($expr_sql.' '.$op_sql, @bind);
1221 # Some databases (SQLite) treat col IN (1, 2) different from
1222 # col IN ( (1, 2) ). Use this to strip all outer parens while
1223 # adding them back in the corresponding method
1224 sub _open_outer_paren {
1225 my ($self, $sql) = @_;
1227 while (my ($inner) = $sql =~ /^ \s* \( (.*) \) \s* $/xs) {
1229 # there are closing parens inside, need the heavy duty machinery
1230 # to reevaluate the extraction starting from $sql (full reevaluation)
1231 if ($inner =~ /\)/) {
1232 require Text::Balanced;
1234 my (undef, $remainder) = do {
1235 # idiotic design - writes to $@ but *DOES NOT* throw exceptions
1237 Text::Balanced::extract_bracketed($sql, '()', qr/\s*/);
1240 # the entire expression needs to be a balanced bracketed thing
1241 # (after an extract no remainder sans trailing space)
1242 last if defined $remainder and $remainder =~ /\S/;
1252 #======================================================================
1254 #======================================================================
1256 sub _expand_order_by {
1257 my ($self, $arg) = @_;
1259 return unless defined($arg) and not (ref($arg) eq 'ARRAY' and !@$arg);
1261 my $expander = sub {
1262 my ($self, $dir, $expr) = @_;
1263 my @to_expand = ref($expr) eq 'ARRAY' ? @$expr : $expr;
1264 foreach my $arg (@to_expand) {
1268 and grep /^-(asc|desc)$/, keys %$arg
1270 puke "ordering direction hash passed to order by must have exactly one key (-asc or -desc)";
1274 defined($dir) ? { -op => [ $dir =~ /^-?(.*)$/ ,=> $_ ] } : $_
1276 map $self->expand_expr($_, -ident),
1277 map ref($_) eq 'ARRAY' ? @$_ : $_, @to_expand;
1278 return undef unless @exp;
1279 return undef if @exp == 1 and not defined($exp[0]);
1280 return +{ -op => [ ',', @exp ] };
1283 local @{$self->{expand}}{qw(asc desc)} = (($expander) x 2);
1285 return $self->$expander(undef, $arg);
1289 my ($self, $arg) = @_;
1291 return '' unless defined(my $expanded = $self->_expand_order_by($arg));
1293 my ($sql, @bind) = $self->render_aqt($expanded);
1295 return '' unless length($sql);
1297 my $final_sql = $self->_sqlcase(' order by ').$sql;
1299 return wantarray ? ($final_sql, @bind) : $final_sql;
1302 # _order_by no longer needs to call this so doesn't but DBIC uses it.
1304 sub _order_by_chunks {
1305 my ($self, $arg) = @_;
1307 return () unless defined(my $expanded = $self->_expand_order_by($arg));
1309 return $self->_chunkify_order_by($expanded);
1312 sub _chunkify_order_by {
1313 my ($self, $expanded) = @_;
1315 return grep length, $self->render_aqt($expanded)
1316 if $expanded->{-ident} or @{$expanded->{-literal}||[]} == 1;
1319 if (ref() eq 'HASH' and $_->{-op} and $_->{-op}[0] eq ',') {
1320 my ($comma, @list) = @{$_->{-op}};
1321 return map $self->_chunkify_order_by($_), @list;
1323 return [ $self->render_aqt($_) ];
1327 #======================================================================
1328 # DATASOURCE (FOR NOW, JUST PLAIN TABLE OR LIST OF TABLES)
1329 #======================================================================
1335 $self->_expand_maybe_list_expr($from, -ident)
1340 #======================================================================
1342 #======================================================================
1344 sub _expand_maybe_list_expr {
1345 my ($self, $expr, $default) = @_;
1346 return +{ -op => [ ',',
1347 map $self->expand_expr($_, $default),
1348 ref($expr) eq 'ARRAY' ? @$expr : $expr
1352 # highly optimized, as it's called way too often
1354 # my ($self, $label) = @_;
1356 return '' unless defined $_[1];
1357 return ${$_[1]} if ref($_[1]) eq 'SCALAR';
1358 puke 'Identifier cannot be hashref' if ref($_[1]) eq 'HASH';
1360 unless ($_[0]->{quote_char}) {
1361 if (ref($_[1]) eq 'ARRAY') {
1362 return join($_[0]->{name_sep}||'.', @{$_[1]});
1364 $_[0]->_assert_pass_injection_guard($_[1]);
1369 my $qref = ref $_[0]->{quote_char};
1371 !$qref ? ($_[0]->{quote_char}, $_[0]->{quote_char})
1372 : ($qref eq 'ARRAY') ? @{$_[0]->{quote_char}}
1373 : puke "Unsupported quote_char format: $_[0]->{quote_char}";
1375 my $esc = $_[0]->{escape_char} || $r;
1377 # parts containing * are naturally unquoted
1379 $_[0]->{name_sep}||'',
1383 : do { (my $n = $_) =~ s/(\Q$esc\E|\Q$r\E)/$esc$1/g; $l . $n . $r }
1385 (ref($_[1]) eq 'ARRAY'
1389 ? split (/\Q$_[0]->{name_sep}\E/, $_[1] )
1397 # Conversion, if applicable
1399 #my ($self, $arg) = @_;
1400 if ($_[0]->{convert_where}) {
1401 return $_[0]->_sqlcase($_[0]->{convert_where}) .'(' . $_[1] . ')';
1408 #my ($self, $col, @vals) = @_;
1409 # called often - tighten code
1410 return $_[0]->{bindtype} eq 'columns'
1411 ? map {[$_[1], $_]} @_[2 .. $#_]
1416 # Dies if any element of @bind is not in [colname => value] format
1417 # if bindtype is 'columns'.
1418 sub _assert_bindval_matches_bindtype {
1419 # my ($self, @bind) = @_;
1421 if ($self->{bindtype} eq 'columns') {
1423 if (!defined $_ || ref($_) ne 'ARRAY' || @$_ != 2) {
1424 puke "bindtype 'columns' selected, you need to pass: [column_name => bind_value]"
1430 sub _join_sql_clauses {
1431 my ($self, $logic, $clauses_aref, $bind_aref) = @_;
1433 if (@$clauses_aref > 1) {
1434 my $join = " " . $self->_sqlcase($logic) . " ";
1435 my $sql = '( ' . join($join, @$clauses_aref) . ' )';
1436 return ($sql, @$bind_aref);
1438 elsif (@$clauses_aref) {
1439 return ($clauses_aref->[0], @$bind_aref); # no parentheses
1442 return (); # if no SQL, ignore @$bind_aref
1447 # Fix SQL case, if so requested
1449 # LDNOTE: if $self->{case} is true, then it contains 'lower', so we
1450 # don't touch the argument ... crooked logic, but let's not change it!
1451 return $_[0]->{case} ? $_[1] : uc($_[1]);
1455 #======================================================================
1456 # DISPATCHING FROM REFKIND
1457 #======================================================================
1460 my ($self, $data) = @_;
1462 return 'UNDEF' unless defined $data;
1464 # blessed objects are treated like scalars
1465 my $ref = (Scalar::Util::blessed $data) ? '' : ref $data;
1467 return 'SCALAR' unless $ref;
1470 while ($ref eq 'REF') {
1472 $ref = (Scalar::Util::blessed $data) ? '' : ref $data;
1476 return ($ref||'SCALAR') . ('REF' x $n_steps);
1480 my ($self, $data) = @_;
1481 my @try = ($self->_refkind($data));
1482 push @try, 'SCALAR_or_UNDEF' if $try[0] eq 'SCALAR' || $try[0] eq 'UNDEF';
1483 push @try, 'FALLBACK';
1487 sub _METHOD_FOR_refkind {
1488 my ($self, $meth_prefix, $data) = @_;
1491 for (@{$self->_try_refkind($data)}) {
1492 $method = $self->can($meth_prefix."_".$_)
1496 return $method || puke "cannot dispatch on '$meth_prefix' for ".$self->_refkind($data);
1500 sub _SWITCH_refkind {
1501 my ($self, $data, $dispatch_table) = @_;
1504 for (@{$self->_try_refkind($data)}) {
1505 $coderef = $dispatch_table->{$_}
1509 puke "no dispatch entry for ".$self->_refkind($data)
1518 #======================================================================
1519 # VALUES, GENERATE, AUTOLOAD
1520 #======================================================================
1522 # LDNOTE: original code from nwiger, didn't touch code in that section
1523 # I feel the AUTOLOAD stuff should not be the default, it should
1524 # only be activated on explicit demand by user.
1528 my $data = shift || return;
1529 puke "Argument to ", __PACKAGE__, "->values must be a \\%hash"
1530 unless ref $data eq 'HASH';
1533 foreach my $k (sort keys %$data) {
1534 my $v = $data->{$k};
1535 $self->_SWITCH_refkind($v, {
1537 if ($self->{array_datatypes}) { # array datatype
1538 push @all_bind, $self->_bindtype($k, $v);
1540 else { # literal SQL with bind
1541 my ($sql, @bind) = @$v;
1542 $self->_assert_bindval_matches_bindtype(@bind);
1543 push @all_bind, @bind;
1546 ARRAYREFREF => sub { # literal SQL with bind
1547 my ($sql, @bind) = @${$v};
1548 $self->_assert_bindval_matches_bindtype(@bind);
1549 push @all_bind, @bind;
1551 SCALARREF => sub { # literal SQL without bind
1553 SCALAR_or_UNDEF => sub {
1554 push @all_bind, $self->_bindtype($k, $v);
1565 my(@sql, @sqlq, @sqlv);
1569 if ($ref eq 'HASH') {
1570 for my $k (sort keys %$_) {
1573 my $label = $self->_quote($k);
1574 if ($r eq 'ARRAY') {
1575 # literal SQL with bind
1576 my ($sql, @bind) = @$v;
1577 $self->_assert_bindval_matches_bindtype(@bind);
1578 push @sqlq, "$label = $sql";
1580 } elsif ($r eq 'SCALAR') {
1581 # literal SQL without bind
1582 push @sqlq, "$label = $$v";
1584 push @sqlq, "$label = ?";
1585 push @sqlv, $self->_bindtype($k, $v);
1588 push @sql, $self->_sqlcase('set'), join ', ', @sqlq;
1589 } elsif ($ref eq 'ARRAY') {
1590 # unlike insert(), assume these are ONLY the column names, i.e. for SQL
1593 if ($r eq 'ARRAY') { # literal SQL with bind
1594 my ($sql, @bind) = @$v;
1595 $self->_assert_bindval_matches_bindtype(@bind);
1598 } elsif ($r eq 'SCALAR') { # literal SQL without bind
1599 # embedded literal SQL
1606 push @sql, '(' . join(', ', @sqlq) . ')';
1607 } elsif ($ref eq 'SCALAR') {
1611 # strings get case twiddled
1612 push @sql, $self->_sqlcase($_);
1616 my $sql = join ' ', @sql;
1618 # this is pretty tricky
1619 # if ask for an array, return ($stmt, @bind)
1620 # otherwise, s/?/shift @sqlv/ to put it inline
1622 return ($sql, @sqlv);
1624 1 while $sql =~ s/\?/my $d = shift(@sqlv);
1625 ref $d ? $d->[1] : $d/e;
1634 # This allows us to check for a local, then _form, attr
1636 my($name) = $AUTOLOAD =~ /.*::(.+)/;
1637 return $self->generate($name, @_);
1648 SQL::Abstract - Generate SQL from Perl data structures
1654 my $sql = SQL::Abstract->new;
1656 my($stmt, @bind) = $sql->select($source, \@fields, \%where, $order);
1658 my($stmt, @bind) = $sql->insert($table, \%fieldvals || \@values);
1660 my($stmt, @bind) = $sql->update($table, \%fieldvals, \%where);
1662 my($stmt, @bind) = $sql->delete($table, \%where);
1664 # Then, use these in your DBI statements
1665 my $sth = $dbh->prepare($stmt);
1666 $sth->execute(@bind);
1668 # Just generate the WHERE clause
1669 my($stmt, @bind) = $sql->where(\%where, $order);
1671 # Return values in the same order, for hashed queries
1672 # See PERFORMANCE section for more details
1673 my @bind = $sql->values(\%fieldvals);
1677 This module was inspired by the excellent L<DBIx::Abstract>.
1678 However, in using that module I found that what I really wanted
1679 to do was generate SQL, but still retain complete control over my
1680 statement handles and use the DBI interface. So, I set out to
1681 create an abstract SQL generation module.
1683 While based on the concepts used by L<DBIx::Abstract>, there are
1684 several important differences, especially when it comes to WHERE
1685 clauses. I have modified the concepts used to make the SQL easier
1686 to generate from Perl data structures and, IMO, more intuitive.
1687 The underlying idea is for this module to do what you mean, based
1688 on the data structures you provide it. The big advantage is that
1689 you don't have to modify your code every time your data changes,
1690 as this module figures it out.
1692 To begin with, an SQL INSERT is as easy as just specifying a hash
1693 of C<key=value> pairs:
1696 name => 'Jimbo Bobson',
1697 phone => '123-456-7890',
1698 address => '42 Sister Lane',
1699 city => 'St. Louis',
1700 state => 'Louisiana',
1703 The SQL can then be generated with this:
1705 my($stmt, @bind) = $sql->insert('people', \%data);
1707 Which would give you something like this:
1709 $stmt = "INSERT INTO people
1710 (address, city, name, phone, state)
1711 VALUES (?, ?, ?, ?, ?)";
1712 @bind = ('42 Sister Lane', 'St. Louis', 'Jimbo Bobson',
1713 '123-456-7890', 'Louisiana');
1715 These are then used directly in your DBI code:
1717 my $sth = $dbh->prepare($stmt);
1718 $sth->execute(@bind);
1720 =head2 Inserting and Updating Arrays
1722 If your database has array types (like for example Postgres),
1723 activate the special option C<< array_datatypes => 1 >>
1724 when creating the C<SQL::Abstract> object.
1725 Then you may use an arrayref to insert and update database array types:
1727 my $sql = SQL::Abstract->new(array_datatypes => 1);
1729 planets => [qw/Mercury Venus Earth Mars/]
1732 my($stmt, @bind) = $sql->insert('solar_system', \%data);
1736 $stmt = "INSERT INTO solar_system (planets) VALUES (?)"
1738 @bind = (['Mercury', 'Venus', 'Earth', 'Mars']);
1741 =head2 Inserting and Updating SQL
1743 In order to apply SQL functions to elements of your C<%data> you may
1744 specify a reference to an arrayref for the given hash value. For example,
1745 if you need to execute the Oracle C<to_date> function on a value, you can
1746 say something like this:
1750 date_entered => \[ "to_date(?,'MM/DD/YYYY')", "03/02/2003" ],
1753 The first value in the array is the actual SQL. Any other values are
1754 optional and would be included in the bind values array. This gives
1757 my($stmt, @bind) = $sql->insert('people', \%data);
1759 $stmt = "INSERT INTO people (name, date_entered)
1760 VALUES (?, to_date(?,'MM/DD/YYYY'))";
1761 @bind = ('Bill', '03/02/2003');
1763 An UPDATE is just as easy, all you change is the name of the function:
1765 my($stmt, @bind) = $sql->update('people', \%data);
1767 Notice that your C<%data> isn't touched; the module will generate
1768 the appropriately quirky SQL for you automatically. Usually you'll
1769 want to specify a WHERE clause for your UPDATE, though, which is
1770 where handling C<%where> hashes comes in handy...
1772 =head2 Complex where statements
1774 This module can generate pretty complicated WHERE statements
1775 easily. For example, simple C<key=value> pairs are taken to mean
1776 equality, and if you want to see if a field is within a set
1777 of values, you can use an arrayref. Let's say we wanted to
1778 SELECT some data based on this criteria:
1781 requestor => 'inna',
1782 worker => ['nwiger', 'rcwe', 'sfz'],
1783 status => { '!=', 'completed' }
1786 my($stmt, @bind) = $sql->select('tickets', '*', \%where);
1788 The above would give you something like this:
1790 $stmt = "SELECT * FROM tickets WHERE
1791 ( requestor = ? ) AND ( status != ? )
1792 AND ( worker = ? OR worker = ? OR worker = ? )";
1793 @bind = ('inna', 'completed', 'nwiger', 'rcwe', 'sfz');
1795 Which you could then use in DBI code like so:
1797 my $sth = $dbh->prepare($stmt);
1798 $sth->execute(@bind);
1804 The methods are simple. There's one for every major SQL operation,
1805 and a constructor you use first. The arguments are specified in a
1806 similar order for each method (table, then fields, then a where
1807 clause) to try and simplify things.
1809 =head2 new(option => 'value')
1811 The C<new()> function takes a list of options and values, and returns
1812 a new B<SQL::Abstract> object which can then be used to generate SQL
1813 through the methods below. The options accepted are:
1819 If set to 'lower', then SQL will be generated in all lowercase. By
1820 default SQL is generated in "textbook" case meaning something like:
1822 SELECT a_field FROM a_table WHERE some_field LIKE '%someval%'
1824 Any setting other than 'lower' is ignored.
1828 This determines what the default comparison operator is. By default
1829 it is C<=>, meaning that a hash like this:
1831 %where = (name => 'nwiger', email => 'nate@wiger.org');
1833 Will generate SQL like this:
1835 WHERE name = 'nwiger' AND email = 'nate@wiger.org'
1837 However, you may want loose comparisons by default, so if you set
1838 C<cmp> to C<like> you would get SQL such as:
1840 WHERE name like 'nwiger' AND email like 'nate@wiger.org'
1842 You can also override the comparison on an individual basis - see
1843 the huge section on L</"WHERE CLAUSES"> at the bottom.
1845 =item sqltrue, sqlfalse
1847 Expressions for inserting boolean values within SQL statements.
1848 By default these are C<1=1> and C<1=0>. They are used
1849 by the special operators C<-in> and C<-not_in> for generating
1850 correct SQL even when the argument is an empty array (see below).
1854 This determines the default logical operator for multiple WHERE
1855 statements in arrays or hashes. If absent, the default logic is "or"
1856 for arrays, and "and" for hashes. This means that a WHERE
1860 event_date => {'>=', '2/13/99'},
1861 event_date => {'<=', '4/24/03'},
1864 will generate SQL like this:
1866 WHERE event_date >= '2/13/99' OR event_date <= '4/24/03'
1868 This is probably not what you want given this query, though (look
1869 at the dates). To change the "OR" to an "AND", simply specify:
1871 my $sql = SQL::Abstract->new(logic => 'and');
1873 Which will change the above C<WHERE> to:
1875 WHERE event_date >= '2/13/99' AND event_date <= '4/24/03'
1877 The logic can also be changed locally by inserting
1878 a modifier in front of an arrayref:
1880 @where = (-and => [event_date => {'>=', '2/13/99'},
1881 event_date => {'<=', '4/24/03'} ]);
1883 See the L</"WHERE CLAUSES"> section for explanations.
1887 This will automatically convert comparisons using the specified SQL
1888 function for both column and value. This is mostly used with an argument
1889 of C<upper> or C<lower>, so that the SQL will have the effect of
1890 case-insensitive "searches". For example, this:
1892 $sql = SQL::Abstract->new(convert => 'upper');
1893 %where = (keywords => 'MaKe iT CAse inSeNSItive');
1895 Will turn out the following SQL:
1897 WHERE upper(keywords) like upper('MaKe iT CAse inSeNSItive')
1899 The conversion can be C<upper()>, C<lower()>, or any other SQL function
1900 that can be applied symmetrically to fields (actually B<SQL::Abstract> does
1901 not validate this option; it will just pass through what you specify verbatim).
1905 This is a kludge because many databases suck. For example, you can't
1906 just bind values using DBI's C<execute()> for Oracle C<CLOB> or C<BLOB> fields.
1907 Instead, you have to use C<bind_param()>:
1909 $sth->bind_param(1, 'reg data');
1910 $sth->bind_param(2, $lots, {ora_type => ORA_CLOB});
1912 The problem is, B<SQL::Abstract> will normally just return a C<@bind> array,
1913 which loses track of which field each slot refers to. Fear not.
1915 If you specify C<bindtype> in new, you can determine how C<@bind> is returned.
1916 Currently, you can specify either C<normal> (default) or C<columns>. If you
1917 specify C<columns>, you will get an array that looks like this:
1919 my $sql = SQL::Abstract->new(bindtype => 'columns');
1920 my($stmt, @bind) = $sql->insert(...);
1923 [ 'column1', 'value1' ],
1924 [ 'column2', 'value2' ],
1925 [ 'column3', 'value3' ],
1928 You can then iterate through this manually, using DBI's C<bind_param()>.
1930 $sth->prepare($stmt);
1933 my($col, $data) = @$_;
1934 if ($col eq 'details' || $col eq 'comments') {
1935 $sth->bind_param($i, $data, {ora_type => ORA_CLOB});
1936 } elsif ($col eq 'image') {
1937 $sth->bind_param($i, $data, {ora_type => ORA_BLOB});
1939 $sth->bind_param($i, $data);
1943 $sth->execute; # execute without @bind now
1945 Now, why would you still use B<SQL::Abstract> if you have to do this crap?
1946 Basically, the advantage is still that you don't have to care which fields
1947 are or are not included. You could wrap that above C<for> loop in a simple
1948 sub called C<bind_fields()> or something and reuse it repeatedly. You still
1949 get a layer of abstraction over manual SQL specification.
1951 Note that if you set L</bindtype> to C<columns>, the C<\[ $sql, @bind ]>
1952 construct (see L</Literal SQL with placeholders and bind values (subqueries)>)
1953 will expect the bind values in this format.
1957 This is the character that a table or column name will be quoted
1958 with. By default this is an empty string, but you could set it to
1959 the character C<`>, to generate SQL like this:
1961 SELECT `a_field` FROM `a_table` WHERE `some_field` LIKE '%someval%'
1963 Alternatively, you can supply an array ref of two items, the first being the left
1964 hand quote character, and the second the right hand quote character. For
1965 example, you could supply C<['[',']']> for SQL Server 2000 compliant quotes
1966 that generates SQL like this:
1968 SELECT [a_field] FROM [a_table] WHERE [some_field] LIKE '%someval%'
1970 Quoting is useful if you have tables or columns names that are reserved
1971 words in your database's SQL dialect.
1975 This is the character that will be used to escape L</quote_char>s appearing
1976 in an identifier before it has been quoted.
1978 The parameter default in case of a single L</quote_char> character is the quote
1981 When opening-closing-style quoting is used (L</quote_char> is an arrayref)
1982 this parameter defaults to the B<closing (right)> L</quote_char>. Occurrences
1983 of the B<opening (left)> L</quote_char> within the identifier are currently left
1984 untouched. The default for opening-closing-style quotes may change in future
1985 versions, thus you are B<strongly encouraged> to specify the escape character
1990 This is the character that separates a table and column name. It is
1991 necessary to specify this when the C<quote_char> option is selected,
1992 so that tables and column names can be individually quoted like this:
1994 SELECT `table`.`one_field` FROM `table` WHERE `table`.`other_field` = 1
1996 =item injection_guard
1998 A regular expression C<qr/.../> that is applied to any C<-function> and unquoted
1999 column name specified in a query structure. This is a safety mechanism to avoid
2000 injection attacks when mishandling user input e.g.:
2002 my %condition_as_column_value_pairs = get_values_from_user();
2003 $sqla->select( ... , \%condition_as_column_value_pairs );
2005 If the expression matches an exception is thrown. Note that literal SQL
2006 supplied via C<\'...'> or C<\['...']> is B<not> checked in any way.
2008 Defaults to checking for C<;> and the C<GO> keyword (TransactSQL)
2010 =item array_datatypes
2012 When this option is true, arrayrefs in INSERT or UPDATE are
2013 interpreted as array datatypes and are passed directly
2015 When this option is false, arrayrefs are interpreted
2016 as literal SQL, just like refs to arrayrefs
2017 (but this behavior is for backwards compatibility; when writing
2018 new queries, use the "reference to arrayref" syntax
2024 Takes a reference to a list of "special operators"
2025 to extend the syntax understood by L<SQL::Abstract>.
2026 See section L</"SPECIAL OPERATORS"> for details.
2030 Takes a reference to a list of "unary operators"
2031 to extend the syntax understood by L<SQL::Abstract>.
2032 See section L</"UNARY OPERATORS"> for details.
2038 =head2 insert($table, \@values || \%fieldvals, \%options)
2040 This is the simplest function. You simply give it a table name
2041 and either an arrayref of values or hashref of field/value pairs.
2042 It returns an SQL INSERT statement and a list of bind values.
2043 See the sections on L</"Inserting and Updating Arrays"> and
2044 L</"Inserting and Updating SQL"> for information on how to insert
2045 with those data types.
2047 The optional C<\%options> hash reference may contain additional
2048 options to generate the insert SQL. Currently supported options
2055 Takes either a scalar of raw SQL fields, or an array reference of
2056 field names, and adds on an SQL C<RETURNING> statement at the end.
2057 This allows you to return data generated by the insert statement
2058 (such as row IDs) without performing another C<SELECT> statement.
2059 Note, however, this is not part of the SQL standard and may not
2060 be supported by all database engines.
2064 =head2 update($table, \%fieldvals, \%where, \%options)
2066 This takes a table, hashref of field/value pairs, and an optional
2067 hashref L<WHERE clause|/WHERE CLAUSES>. It returns an SQL UPDATE function and a list
2069 See the sections on L</"Inserting and Updating Arrays"> and
2070 L</"Inserting and Updating SQL"> for information on how to insert
2071 with those data types.
2073 The optional C<\%options> hash reference may contain additional
2074 options to generate the update SQL. Currently supported options
2081 See the C<returning> option to
2082 L<insert|/insert($table, \@values || \%fieldvals, \%options)>.
2086 =head2 select($source, $fields, $where, $order)
2088 This returns a SQL SELECT statement and associated list of bind values, as
2089 specified by the arguments:
2095 Specification of the 'FROM' part of the statement.
2096 The argument can be either a plain scalar (interpreted as a table
2097 name, will be quoted), or an arrayref (interpreted as a list
2098 of table names, joined by commas, quoted), or a scalarref
2099 (literal SQL, not quoted).
2103 Specification of the list of fields to retrieve from
2105 The argument can be either an arrayref (interpreted as a list
2106 of field names, will be joined by commas and quoted), or a
2107 plain scalar (literal SQL, not quoted).
2108 Please observe that this API is not as flexible as that of
2109 the first argument C<$source>, for backwards compatibility reasons.
2113 Optional argument to specify the WHERE part of the query.
2114 The argument is most often a hashref, but can also be
2115 an arrayref or plain scalar --
2116 see section L<WHERE clause|/"WHERE CLAUSES"> for details.
2120 Optional argument to specify the ORDER BY part of the query.
2121 The argument can be a scalar, a hashref or an arrayref
2122 -- see section L<ORDER BY clause|/"ORDER BY CLAUSES">
2128 =head2 delete($table, \%where, \%options)
2130 This takes a table name and optional hashref L<WHERE clause|/WHERE CLAUSES>.
2131 It returns an SQL DELETE statement and list of bind values.
2133 The optional C<\%options> hash reference may contain additional
2134 options to generate the delete SQL. Currently supported options
2141 See the C<returning> option to
2142 L<insert|/insert($table, \@values || \%fieldvals, \%options)>.
2146 =head2 where(\%where, $order)
2148 This is used to generate just the WHERE clause. For example,
2149 if you have an arbitrary data structure and know what the
2150 rest of your SQL is going to look like, but want an easy way
2151 to produce a WHERE clause, use this. It returns an SQL WHERE
2152 clause and list of bind values.
2155 =head2 values(\%data)
2157 This just returns the values from the hash C<%data>, in the same
2158 order that would be returned from any of the other above queries.
2159 Using this allows you to markedly speed up your queries if you
2160 are affecting lots of rows. See below under the L</"PERFORMANCE"> section.
2162 =head2 generate($any, 'number', $of, \@data, $struct, \%types)
2164 Warning: This is an experimental method and subject to change.
2166 This returns arbitrarily generated SQL. It's a really basic shortcut.
2167 It will return two different things, depending on return context:
2169 my($stmt, @bind) = $sql->generate('create table', \$table, \@fields);
2170 my $stmt_and_val = $sql->generate('create table', \$table, \@fields);
2172 These would return the following:
2174 # First calling form
2175 $stmt = "CREATE TABLE test (?, ?)";
2176 @bind = (field1, field2);
2178 # Second calling form
2179 $stmt_and_val = "CREATE TABLE test (field1, field2)";
2181 Depending on what you're trying to do, it's up to you to choose the correct
2182 format. In this example, the second form is what you would want.
2186 $sql->generate('alter session', { nls_date_format => 'MM/YY' });
2190 ALTER SESSION SET nls_date_format = 'MM/YY'
2192 You get the idea. Strings get their case twiddled, but everything
2193 else remains verbatim.
2195 =head1 EXPORTABLE FUNCTIONS
2197 =head2 is_plain_value
2199 Determines if the supplied argument is a plain value as understood by this
2204 =item * The value is C<undef>
2206 =item * The value is a non-reference
2208 =item * The value is an object with stringification overloading
2210 =item * The value is of the form C<< { -value => $anything } >>
2214 On failure returns C<undef>, on success returns a B<scalar> reference
2215 to the original supplied argument.
2221 The stringification overloading detection is rather advanced: it takes
2222 into consideration not only the presence of a C<""> overload, but if that
2223 fails also checks for enabled
2224 L<autogenerated versions of C<"">|overload/Magic Autogeneration>, based
2225 on either C<0+> or C<bool>.
2227 Unfortunately testing in the field indicates that this
2228 detection B<< may tickle a latent bug in perl versions before 5.018 >>,
2229 but only when very large numbers of stringifying objects are involved.
2230 At the time of writing ( Sep 2014 ) there is no clear explanation of
2231 the direct cause, nor is there a manageably small test case that reliably
2232 reproduces the problem.
2234 If you encounter any of the following exceptions in B<random places within
2235 your application stack> - this module may be to blame:
2237 Operation "ne": no method found,
2238 left argument in overloaded package <something>,
2239 right argument in overloaded package <something>
2243 Stub found while resolving method "???" overloading """" in package <something>
2245 If you fall victim to the above - please attempt to reduce the problem
2246 to something that could be sent to the L<SQL::Abstract developers
2247 |DBIx::Class/GETTING HELP/SUPPORT>
2248 (either publicly or privately). As a workaround in the meantime you can
2249 set C<$ENV{SQLA_ISVALUE_IGNORE_AUTOGENERATED_STRINGIFICATION}> to a true
2250 value, which will most likely eliminate your problem (at the expense of
2251 not being able to properly detect exotic forms of stringification).
2253 This notice and environment variable will be removed in a future version,
2254 as soon as the underlying problem is found and a reliable workaround is
2259 =head2 is_literal_value
2261 Determines if the supplied argument is a literal value as understood by this
2266 =item * C<\$sql_string>
2268 =item * C<\[ $sql_string, @bind_values ]>
2272 On failure returns C<undef>, on success returns an B<array> reference
2273 containing the unpacked version of the supplied literal SQL and bind values.
2275 =head1 WHERE CLAUSES
2279 This module uses a variation on the idea from L<DBIx::Abstract>. It
2280 is B<NOT>, repeat I<not> 100% compatible. B<The main logic of this
2281 module is that things in arrays are OR'ed, and things in hashes
2284 The easiest way to explain is to show lots of examples. After
2285 each C<%where> hash shown, it is assumed you used:
2287 my($stmt, @bind) = $sql->where(\%where);
2289 However, note that the C<%where> hash can be used directly in any
2290 of the other functions as well, as described above.
2292 =head2 Key-value pairs
2294 So, let's get started. To begin, a simple hash:
2298 status => 'completed'
2301 Is converted to SQL C<key = val> statements:
2303 $stmt = "WHERE user = ? AND status = ?";
2304 @bind = ('nwiger', 'completed');
2306 One common thing I end up doing is having a list of values that
2307 a field can be in. To do this, simply specify a list inside of
2312 status => ['assigned', 'in-progress', 'pending'];
2315 This simple code will create the following:
2317 $stmt = "WHERE user = ? AND ( status = ? OR status = ? OR status = ? )";
2318 @bind = ('nwiger', 'assigned', 'in-progress', 'pending');
2320 A field associated to an empty arrayref will be considered a
2321 logical false and will generate 0=1.
2323 =head2 Tests for NULL values
2325 If the value part is C<undef> then this is converted to SQL <IS NULL>
2334 $stmt = "WHERE user = ? AND status IS NULL";
2337 To test if a column IS NOT NULL:
2341 status => { '!=', undef },
2344 =head2 Specific comparison operators
2346 If you want to specify a different type of operator for your comparison,
2347 you can use a hashref for a given column:
2351 status => { '!=', 'completed' }
2354 Which would generate:
2356 $stmt = "WHERE user = ? AND status != ?";
2357 @bind = ('nwiger', 'completed');
2359 To test against multiple values, just enclose the values in an arrayref:
2361 status => { '=', ['assigned', 'in-progress', 'pending'] };
2363 Which would give you:
2365 "WHERE status = ? OR status = ? OR status = ?"
2368 The hashref can also contain multiple pairs, in which case it is expanded
2369 into an C<AND> of its elements:
2373 status => { '!=', 'completed', -not_like => 'pending%' }
2376 # Or more dynamically, like from a form
2377 $where{user} = 'nwiger';
2378 $where{status}{'!='} = 'completed';
2379 $where{status}{'-not_like'} = 'pending%';
2381 # Both generate this
2382 $stmt = "WHERE user = ? AND status != ? AND status NOT LIKE ?";
2383 @bind = ('nwiger', 'completed', 'pending%');
2386 To get an OR instead, you can combine it with the arrayref idea:
2390 priority => [ { '=', 2 }, { '>', 5 } ]
2393 Which would generate:
2395 $stmt = "WHERE ( priority = ? OR priority > ? ) AND user = ?";
2396 @bind = ('2', '5', 'nwiger');
2398 If you want to include literal SQL (with or without bind values), just use a
2399 scalar reference or reference to an arrayref as the value:
2402 date_entered => { '>' => \["to_date(?, 'MM/DD/YYYY')", "11/26/2008"] },
2403 date_expires => { '<' => \"now()" }
2406 Which would generate:
2408 $stmt = "WHERE date_entered > to_date(?, 'MM/DD/YYYY') AND date_expires < now()";
2409 @bind = ('11/26/2008');
2412 =head2 Logic and nesting operators
2414 In the example above,
2415 there is a subtle trap if you want to say something like
2416 this (notice the C<AND>):
2418 WHERE priority != ? AND priority != ?
2420 Because, in Perl you I<can't> do this:
2422 priority => { '!=' => 2, '!=' => 1 }
2424 As the second C<!=> key will obliterate the first. The solution
2425 is to use the special C<-modifier> form inside an arrayref:
2427 priority => [ -and => {'!=', 2},
2431 Normally, these would be joined by C<OR>, but the modifier tells it
2432 to use C<AND> instead. (Hint: You can use this in conjunction with the
2433 C<logic> option to C<new()> in order to change the way your queries
2434 work by default.) B<Important:> Note that the C<-modifier> goes
2435 B<INSIDE> the arrayref, as an extra first element. This will
2436 B<NOT> do what you think it might:
2438 priority => -and => [{'!=', 2}, {'!=', 1}] # WRONG!
2440 Here is a quick list of equivalencies, since there is some overlap:
2443 status => {'!=', 'completed', 'not like', 'pending%' }
2444 status => [ -and => {'!=', 'completed'}, {'not like', 'pending%'}]
2447 status => {'=', ['assigned', 'in-progress']}
2448 status => [ -or => {'=', 'assigned'}, {'=', 'in-progress'}]
2449 status => [ {'=', 'assigned'}, {'=', 'in-progress'} ]
2453 =head2 Special operators: IN, BETWEEN, etc.
2455 You can also use the hashref format to compare a list of fields using the
2456 C<IN> comparison operator, by specifying the list as an arrayref:
2459 status => 'completed',
2460 reportid => { -in => [567, 2335, 2] }
2463 Which would generate:
2465 $stmt = "WHERE status = ? AND reportid IN (?,?,?)";
2466 @bind = ('completed', '567', '2335', '2');
2468 The reverse operator C<-not_in> generates SQL C<NOT IN> and is used in
2471 If the argument to C<-in> is an empty array, 'sqlfalse' is generated
2472 (by default: C<1=0>). Similarly, C<< -not_in => [] >> generates
2473 'sqltrue' (by default: C<1=1>).
2475 In addition to the array you can supply a chunk of literal sql or
2476 literal sql with bind:
2479 customer => { -in => \[
2480 'SELECT cust_id FROM cust WHERE balance > ?',
2483 status => { -in => \'SELECT status_codes FROM states' },
2489 customer IN ( SELECT cust_id FROM cust WHERE balance > ? )
2490 AND status IN ( SELECT status_codes FROM states )
2494 Finally, if the argument to C<-in> is not a reference, it will be
2495 treated as a single-element array.
2497 Another pair of operators is C<-between> and C<-not_between>,
2498 used with an arrayref of two values:
2502 completion_date => {
2503 -not_between => ['2002-10-01', '2003-02-06']
2509 WHERE user = ? AND completion_date NOT BETWEEN ( ? AND ? )
2511 Just like with C<-in> all plausible combinations of literal SQL
2515 start0 => { -between => [ 1, 2 ] },
2516 start1 => { -between => \["? AND ?", 1, 2] },
2517 start2 => { -between => \"lower(x) AND upper(y)" },
2518 start3 => { -between => [
2520 \["upper(?)", 'stuff' ],
2527 ( start0 BETWEEN ? AND ? )
2528 AND ( start1 BETWEEN ? AND ? )
2529 AND ( start2 BETWEEN lower(x) AND upper(y) )
2530 AND ( start3 BETWEEN lower(x) AND upper(?) )
2532 @bind = (1, 2, 1, 2, 'stuff');
2535 These are the two builtin "special operators"; but the
2536 list can be expanded: see section L</"SPECIAL OPERATORS"> below.
2538 =head2 Unary operators: bool
2540 If you wish to test against boolean columns or functions within your
2541 database you can use the C<-bool> and C<-not_bool> operators. For
2542 example to test the column C<is_user> being true and the column
2543 C<is_enabled> being false you would use:-
2547 -not_bool => 'is_enabled',
2552 WHERE is_user AND NOT is_enabled
2554 If a more complex combination is required, testing more conditions,
2555 then you should use the and/or operators:-
2560 -not_bool => { two=> { -rlike => 'bar' } },
2561 -not_bool => { three => [ { '=', 2 }, { '>', 5 } ] },
2572 (NOT ( three = ? OR three > ? ))
2575 =head2 Nested conditions, -and/-or prefixes
2577 So far, we've seen how multiple conditions are joined with a top-level
2578 C<AND>. We can change this by putting the different conditions we want in
2579 hashes and then putting those hashes in an array. For example:
2584 status => { -like => ['pending%', 'dispatched'] },
2588 status => 'unassigned',
2592 This data structure would create the following:
2594 $stmt = "WHERE ( user = ? AND ( status LIKE ? OR status LIKE ? ) )
2595 OR ( user = ? AND status = ? ) )";
2596 @bind = ('nwiger', 'pending', 'dispatched', 'robot', 'unassigned');
2599 Clauses in hashrefs or arrayrefs can be prefixed with an C<-and> or C<-or>
2600 to change the logic inside:
2606 -and => [ workhrs => {'>', 20}, geo => 'ASIA' ],
2607 -or => { workhrs => {'<', 50}, geo => 'EURO' },
2614 $stmt = "WHERE ( user = ?
2615 AND ( ( workhrs > ? AND geo = ? )
2616 OR ( workhrs < ? OR geo = ? ) ) )";
2617 @bind = ('nwiger', '20', 'ASIA', '50', 'EURO');
2619 =head3 Algebraic inconsistency, for historical reasons
2621 C<Important note>: when connecting several conditions, the C<-and->|C<-or>
2622 operator goes C<outside> of the nested structure; whereas when connecting
2623 several constraints on one column, the C<-and> operator goes
2624 C<inside> the arrayref. Here is an example combining both features:
2627 -and => [a => 1, b => 2],
2628 -or => [c => 3, d => 4],
2629 e => [-and => {-like => 'foo%'}, {-like => '%bar'} ]
2634 WHERE ( ( ( a = ? AND b = ? )
2635 OR ( c = ? OR d = ? )
2636 OR ( e LIKE ? AND e LIKE ? ) ) )
2638 This difference in syntax is unfortunate but must be preserved for
2639 historical reasons. So be careful: the two examples below would
2640 seem algebraically equivalent, but they are not
2643 { -like => 'foo%' },
2644 { -like => '%bar' },
2646 # yields: WHERE ( ( col LIKE ? AND col LIKE ? ) )
2649 { col => { -like => 'foo%' } },
2650 { col => { -like => '%bar' } },
2652 # yields: WHERE ( ( col LIKE ? OR col LIKE ? ) )
2655 =head2 Literal SQL and value type operators
2657 The basic premise of SQL::Abstract is that in WHERE specifications the "left
2658 side" is a column name and the "right side" is a value (normally rendered as
2659 a placeholder). This holds true for both hashrefs and arrayref pairs as you
2660 see in the L</WHERE CLAUSES> examples above. Sometimes it is necessary to
2661 alter this behavior. There are several ways of doing so.
2665 This is a virtual operator that signals the string to its right side is an
2666 identifier (a column name) and not a value. For example to compare two
2667 columns you would write:
2670 priority => { '<', 2 },
2671 requestor => { -ident => 'submitter' },
2676 $stmt = "WHERE priority < ? AND requestor = submitter";
2679 If you are maintaining legacy code you may see a different construct as
2680 described in L</Deprecated usage of Literal SQL>, please use C<-ident> in new
2685 This is a virtual operator that signals that the construct to its right side
2686 is a value to be passed to DBI. This is for example necessary when you want
2687 to write a where clause against an array (for RDBMS that support such
2688 datatypes). For example:
2691 array => { -value => [1, 2, 3] }
2696 $stmt = 'WHERE array = ?';
2697 @bind = ([1, 2, 3]);
2699 Note that if you were to simply say:
2705 the result would probably not be what you wanted:
2707 $stmt = 'WHERE array = ? OR array = ? OR array = ?';
2712 Finally, sometimes only literal SQL will do. To include a random snippet
2713 of SQL verbatim, you specify it as a scalar reference. Consider this only
2714 as a last resort. Usually there is a better way. For example:
2717 priority => { '<', 2 },
2718 requestor => { -in => \'(SELECT name FROM hitmen)' },
2723 $stmt = "WHERE priority < ? AND requestor IN (SELECT name FROM hitmen)"
2726 Note that in this example, you only get one bind parameter back, since
2727 the verbatim SQL is passed as part of the statement.
2731 Never use untrusted input as a literal SQL argument - this is a massive
2732 security risk (there is no way to check literal snippets for SQL
2733 injections and other nastyness). If you need to deal with untrusted input
2734 use literal SQL with placeholders as described next.
2736 =head3 Literal SQL with placeholders and bind values (subqueries)
2738 If the literal SQL to be inserted has placeholders and bind values,
2739 use a reference to an arrayref (yes this is a double reference --
2740 not so common, but perfectly legal Perl). For example, to find a date
2741 in Postgres you can use something like this:
2744 date_column => \[ "= date '2008-09-30' - ?::integer", 10 ]
2749 $stmt = "WHERE ( date_column = date '2008-09-30' - ?::integer )"
2752 Note that you must pass the bind values in the same format as they are returned
2753 by L<where|/where(\%where, $order)>. This means that if you set L</bindtype>
2754 to C<columns>, you must provide the bind values in the
2755 C<< [ column_meta => value ] >> format, where C<column_meta> is an opaque
2756 scalar value; most commonly the column name, but you can use any scalar value
2757 (including references and blessed references), L<SQL::Abstract> will simply
2758 pass it through intact. So if C<bindtype> is set to C<columns> the above
2759 example will look like:
2762 date_column => \[ "= date '2008-09-30' - ?::integer", [ {} => 10 ] ]
2765 Literal SQL is especially useful for nesting parenthesized clauses in the
2766 main SQL query. Here is a first example:
2768 my ($sub_stmt, @sub_bind) = ("SELECT c1 FROM t1 WHERE c2 < ? AND c3 LIKE ?",
2772 bar => \["IN ($sub_stmt)" => @sub_bind],
2777 $stmt = "WHERE (foo = ? AND bar IN (SELECT c1 FROM t1
2778 WHERE c2 < ? AND c3 LIKE ?))";
2779 @bind = (1234, 100, "foo%");
2781 Other subquery operators, like for example C<"E<gt> ALL"> or C<"NOT IN">,
2782 are expressed in the same way. Of course the C<$sub_stmt> and
2783 its associated bind values can be generated through a former call
2786 my ($sub_stmt, @sub_bind)
2787 = $sql->select("t1", "c1", {c2 => {"<" => 100},
2788 c3 => {-like => "foo%"}});
2791 bar => \["> ALL ($sub_stmt)" => @sub_bind],
2794 In the examples above, the subquery was used as an operator on a column;
2795 but the same principle also applies for a clause within the main C<%where>
2796 hash, like an EXISTS subquery:
2798 my ($sub_stmt, @sub_bind)
2799 = $sql->select("t1", "*", {c1 => 1, c2 => \"> t0.c0"});
2800 my %where = ( -and => [
2802 \["EXISTS ($sub_stmt)" => @sub_bind],
2807 $stmt = "WHERE (foo = ? AND EXISTS (SELECT * FROM t1
2808 WHERE c1 = ? AND c2 > t0.c0))";
2812 Observe that the condition on C<c2> in the subquery refers to
2813 column C<t0.c0> of the main query: this is I<not> a bind
2814 value, so we have to express it through a scalar ref.
2815 Writing C<< c2 => {">" => "t0.c0"} >> would have generated
2816 C<< c2 > ? >> with bind value C<"t0.c0"> ... not exactly
2817 what we wanted here.
2819 Finally, here is an example where a subquery is used
2820 for expressing unary negation:
2822 my ($sub_stmt, @sub_bind)
2823 = $sql->where({age => [{"<" => 10}, {">" => 20}]});
2824 $sub_stmt =~ s/^ where //i; # don't want "WHERE" in the subclause
2826 lname => {like => '%son%'},
2827 \["NOT ($sub_stmt)" => @sub_bind],
2832 $stmt = "lname LIKE ? AND NOT ( age < ? OR age > ? )"
2833 @bind = ('%son%', 10, 20)
2835 =head3 Deprecated usage of Literal SQL
2837 Below are some examples of archaic use of literal SQL. It is shown only as
2838 reference for those who deal with legacy code. Each example has a much
2839 better, cleaner and safer alternative that users should opt for in new code.
2845 my %where = ( requestor => \'IS NOT NULL' )
2847 $stmt = "WHERE requestor IS NOT NULL"
2849 This used to be the way of generating NULL comparisons, before the handling
2850 of C<undef> got formalized. For new code please use the superior syntax as
2851 described in L</Tests for NULL values>.
2855 my %where = ( requestor => \'= submitter' )
2857 $stmt = "WHERE requestor = submitter"
2859 This used to be the only way to compare columns. Use the superior L</-ident>
2860 method for all new code. For example an identifier declared in such a way
2861 will be properly quoted if L</quote_char> is properly set, while the legacy
2862 form will remain as supplied.
2866 my %where = ( is_ready => \"", completed => { '>', '2012-12-21' } )
2868 $stmt = "WHERE completed > ? AND is_ready"
2869 @bind = ('2012-12-21')
2871 Using an empty string literal used to be the only way to express a boolean.
2872 For all new code please use the much more readable
2873 L<-bool|/Unary operators: bool> operator.
2879 These pages could go on for a while, since the nesting of the data
2880 structures this module can handle are pretty much unlimited (the
2881 module implements the C<WHERE> expansion as a recursive function
2882 internally). Your best bet is to "play around" with the module a
2883 little to see how the data structures behave, and choose the best
2884 format for your data based on that.
2886 And of course, all the values above will probably be replaced with
2887 variables gotten from forms or the command line. After all, if you
2888 knew everything ahead of time, you wouldn't have to worry about
2889 dynamically-generating SQL and could just hardwire it into your
2892 =head1 ORDER BY CLAUSES
2894 Some functions take an order by clause. This can either be a scalar (just a
2895 column name), a hashref of C<< { -desc => 'col' } >> or C<< { -asc => 'col' }
2896 >>, a scalarref, an arrayref-ref, or an arrayref of any of the previous
2899 Given | Will Generate
2900 ---------------------------------------------------------------
2902 'colA' | ORDER BY colA
2904 [qw/colA colB/] | ORDER BY colA, colB
2906 {-asc => 'colA'} | ORDER BY colA ASC
2908 {-desc => 'colB'} | ORDER BY colB DESC
2910 ['colA', {-asc => 'colB'}] | ORDER BY colA, colB ASC
2912 { -asc => [qw/colA colB/] } | ORDER BY colA ASC, colB ASC
2914 \'colA DESC' | ORDER BY colA DESC
2916 \[ 'FUNC(colA, ?)', $x ] | ORDER BY FUNC(colA, ?)
2917 | /* ...with $x bound to ? */
2920 { -asc => 'colA' }, | colA ASC,
2921 { -desc => [qw/colB/] }, | colB DESC,
2922 { -asc => [qw/colC colD/] },| colC ASC, colD ASC,
2923 \'colE DESC', | colE DESC,
2924 \[ 'FUNC(colF, ?)', $x ], | FUNC(colF, ?)
2925 ] | /* ...with $x bound to ? */
2926 ===============================================================
2930 =head1 SPECIAL OPERATORS
2932 my $sqlmaker = SQL::Abstract->new(special_ops => [
2936 my ($self, $field, $op, $arg) = @_;
2942 handler => 'method_name',
2946 A "special operator" is a SQL syntactic clause that can be
2947 applied to a field, instead of a usual binary operator.
2950 WHERE field IN (?, ?, ?)
2951 WHERE field BETWEEN ? AND ?
2952 WHERE MATCH(field) AGAINST (?, ?)
2954 Special operators IN and BETWEEN are fairly standard and therefore
2955 are builtin within C<SQL::Abstract> (as the overridable methods
2956 C<_where_field_IN> and C<_where_field_BETWEEN>). For other operators,
2957 like the MATCH .. AGAINST example above which is specific to MySQL,
2958 you can write your own operator handlers - supply a C<special_ops>
2959 argument to the C<new> method. That argument takes an arrayref of
2960 operator definitions; each operator definition is a hashref with two
2967 the regular expression to match the operator
2971 Either a coderef or a plain scalar method name. In both cases
2972 the expected return is C<< ($sql, @bind) >>.
2974 When supplied with a method name, it is simply called on the
2975 L<SQL::Abstract> object as:
2977 $self->$method_name($field, $op, $arg)
2981 $field is the LHS of the operator
2982 $op is the part that matched the handler regex
2985 When supplied with a coderef, it is called as:
2987 $coderef->($self, $field, $op, $arg)
2992 For example, here is an implementation
2993 of the MATCH .. AGAINST syntax for MySQL
2995 my $sqlmaker = SQL::Abstract->new(special_ops => [
2997 # special op for MySql MATCH (field) AGAINST(word1, word2, ...)
2998 {regex => qr/^match$/i,
3000 my ($self, $field, $op, $arg) = @_;
3001 $arg = [$arg] if not ref $arg;
3002 my $label = $self->_quote($field);
3003 my ($placeholder) = $self->_convert('?');
3004 my $placeholders = join ", ", (($placeholder) x @$arg);
3005 my $sql = $self->_sqlcase('match') . " ($label) "
3006 . $self->_sqlcase('against') . " ($placeholders) ";
3007 my @bind = $self->_bindtype($field, @$arg);
3008 return ($sql, @bind);
3015 =head1 UNARY OPERATORS
3017 my $sqlmaker = SQL::Abstract->new(unary_ops => [
3021 my ($self, $op, $arg) = @_;
3027 handler => 'method_name',
3031 A "unary operator" is a SQL syntactic clause that can be
3032 applied to a field - the operator goes before the field
3034 You can write your own operator handlers - supply a C<unary_ops>
3035 argument to the C<new> method. That argument takes an arrayref of
3036 operator definitions; each operator definition is a hashref with two
3043 the regular expression to match the operator
3047 Either a coderef or a plain scalar method name. In both cases
3048 the expected return is C<< $sql >>.
3050 When supplied with a method name, it is simply called on the
3051 L<SQL::Abstract> object as:
3053 $self->$method_name($op, $arg)
3057 $op is the part that matched the handler regex
3058 $arg is the RHS or argument of the operator
3060 When supplied with a coderef, it is called as:
3062 $coderef->($self, $op, $arg)
3070 Thanks to some benchmarking by Mark Stosberg, it turns out that
3071 this module is many orders of magnitude faster than using C<DBIx::Abstract>.
3072 I must admit this wasn't an intentional design issue, but it's a
3073 byproduct of the fact that you get to control your C<DBI> handles
3076 To maximize performance, use a code snippet like the following:
3078 # prepare a statement handle using the first row
3079 # and then reuse it for the rest of the rows
3081 for my $href (@array_of_hashrefs) {
3082 $stmt ||= $sql->insert('table', $href);
3083 $sth ||= $dbh->prepare($stmt);
3084 $sth->execute($sql->values($href));
3087 The reason this works is because the keys in your C<$href> are sorted
3088 internally by B<SQL::Abstract>. Thus, as long as your data retains
3089 the same structure, you only have to generate the SQL the first time
3090 around. On subsequent queries, simply use the C<values> function provided
3091 by this module to return your values in the correct order.
3093 However this depends on the values having the same type - if, for
3094 example, the values of a where clause may either have values
3095 (resulting in sql of the form C<column = ?> with a single bind
3096 value), or alternatively the values might be C<undef> (resulting in
3097 sql of the form C<column IS NULL> with no bind value) then the
3098 caching technique suggested will not work.
3102 If you use my C<CGI::FormBuilder> module at all, you'll hopefully
3103 really like this part (I do, at least). Building up a complex query
3104 can be as simple as the following:
3111 use CGI::FormBuilder;
3114 my $form = CGI::FormBuilder->new(...);
3115 my $sql = SQL::Abstract->new;
3117 if ($form->submitted) {
3118 my $field = $form->field;
3119 my $id = delete $field->{id};
3120 my($stmt, @bind) = $sql->update('table', $field, {id => $id});
3123 Of course, you would still have to connect using C<DBI> to run the
3124 query, but the point is that if you make your form look like your
3125 table, the actual query script can be extremely simplistic.
3127 If you're B<REALLY> lazy (I am), check out C<HTML::QuickTable> for
3128 a fast interface to returning and formatting data. I frequently
3129 use these three modules together to write complex database query
3130 apps in under 50 lines.
3132 =head1 HOW TO CONTRIBUTE
3134 Contributions are always welcome, in all usable forms (we especially
3135 welcome documentation improvements). The delivery methods include git-
3136 or unified-diff formatted patches, GitHub pull requests, or plain bug
3137 reports either via RT or the Mailing list. Contributors are generally
3138 granted full access to the official repository after their first several
3139 patches pass successful review.
3141 This project is maintained in a git repository. The code and related tools are
3142 accessible at the following locations:
3146 =item * Official repo: L<git://git.shadowcat.co.uk/dbsrgits/SQL-Abstract.git>
3148 =item * Official gitweb: L<http://git.shadowcat.co.uk/gitweb/gitweb.cgi?p=dbsrgits/SQL-Abstract.git>
3150 =item * GitHub mirror: L<https://github.com/dbsrgits/sql-abstract>
3152 =item * Authorized committers: L<ssh://dbsrgits@git.shadowcat.co.uk/SQL-Abstract.git>
3158 Version 1.50 was a major internal refactoring of C<SQL::Abstract>.
3159 Great care has been taken to preserve the I<published> behavior
3160 documented in previous versions in the 1.* family; however,
3161 some features that were previously undocumented, or behaved
3162 differently from the documentation, had to be changed in order
3163 to clarify the semantics. Hence, client code that was relying
3164 on some dark areas of C<SQL::Abstract> v1.*
3165 B<might behave differently> in v1.50.
3167 The main changes are:
3173 support for literal SQL through the C<< \ [ $sql, @bind ] >> syntax.
3177 support for the { operator => \"..." } construct (to embed literal SQL)
3181 support for the { operator => \["...", @bind] } construct (to embed literal SQL with bind values)
3185 optional support for L<array datatypes|/"Inserting and Updating Arrays">
3189 defensive programming: check arguments
3193 fixed bug with global logic, which was previously implemented
3194 through global variables yielding side-effects. Prior versions would
3195 interpret C<< [ {cond1, cond2}, [cond3, cond4] ] >>
3196 as C<< "(cond1 AND cond2) OR (cond3 AND cond4)" >>.
3197 Now this is interpreted
3198 as C<< "(cond1 AND cond2) OR (cond3 OR cond4)" >>.
3203 fixed semantics of _bindtype on array args
3207 dropped the C<_anoncopy> of the %where tree. No longer necessary,
3208 we just avoid shifting arrays within that tree.
3212 dropped the C<_modlogic> function
3216 =head1 ACKNOWLEDGEMENTS
3218 There are a number of individuals that have really helped out with
3219 this module. Unfortunately, most of them submitted bugs via CPAN
3220 so I have no idea who they are! But the people I do know are:
3222 Ash Berlin (order_by hash term support)
3223 Matt Trout (DBIx::Class support)
3224 Mark Stosberg (benchmarking)
3225 Chas Owens (initial "IN" operator support)
3226 Philip Collins (per-field SQL functions)
3227 Eric Kolve (hashref "AND" support)
3228 Mike Fragassi (enhancements to "BETWEEN" and "LIKE")
3229 Dan Kubb (support for "quote_char" and "name_sep")
3230 Guillermo Roditi (patch to cleanup "IN" and "BETWEEN", fix and tests for _order_by)
3231 Laurent Dami (internal refactoring, extensible list of special operators, literal SQL)
3232 Norbert Buchmuller (support for literal SQL in hashpair, misc. fixes & tests)
3233 Peter Rabbitson (rewrite of SQLA::Test, misc. fixes & tests)
3234 Oliver Charles (support for "RETURNING" after "INSERT")
3240 L<DBIx::Class>, L<DBIx::Abstract>, L<CGI::FormBuilder>, L<HTML::QuickTable>.
3244 Copyright (c) 2001-2007 Nathan Wiger <nwiger@cpan.org>. All Rights Reserved.
3246 This module is actively maintained by Matt Trout <mst@shadowcatsystems.co.uk>
3248 For support, your best bet is to try the C<DBIx::Class> users mailing list.
3249 While not an official support venue, C<DBIx::Class> makes heavy use of
3250 C<SQL::Abstract>, and as such list members there are very familiar with
3251 how to create queries.
3255 This module is free software; you may copy this under the same
3256 terms as perl itself (either the GNU General Public License or
3257 the Artistic License)