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" }},
45 #======================================================================
46 # DEBUGGING AND ERROR REPORTING
47 #======================================================================
50 return unless $_[0]->{debug}; shift; # a little faster
51 my $func = (caller(1))[3];
52 warn "[$func] ", @_, "\n";
56 my($func) = (caller(1))[3];
57 Carp::carp "[$func] Warning: ", @_;
61 my($func) = (caller(1))[3];
62 Carp::croak "[$func] Fatal: ", @_;
65 sub is_literal_value ($) {
66 ref $_[0] eq 'SCALAR' ? [ ${$_[0]} ]
67 : ( ref $_[0] eq 'REF' and ref ${$_[0]} eq 'ARRAY' ) ? [ @${ $_[0] } ]
71 sub is_undef_value ($) {
75 and exists $_[0]->{-value}
76 and not defined $_[0]->{-value}
80 # FIXME XSify - this can be done so much more efficiently
81 sub is_plain_value ($) {
83 ! length ref $_[0] ? \($_[0])
85 ref $_[0] eq 'HASH' and keys %{$_[0]} == 1
87 exists $_[0]->{-value}
88 ) ? \($_[0]->{-value})
90 # reuse @_ for even moar speedz
91 defined ( $_[1] = Scalar::Util::blessed $_[0] )
93 # deliberately not using Devel::OverloadInfo - the checks we are
94 # intersted in are much more limited than the fullblown thing, and
95 # this is a very hot piece of code
97 # simply using ->can('(""') can leave behind stub methods that
98 # break actually using the overload later (see L<perldiag/Stub
99 # found while resolving method "%s" overloading "%s" in package
100 # "%s"> and the source of overload::mycan())
102 # either has stringification which DBI SHOULD prefer out of the box
103 grep { *{ (qq[${_}::(""]) }{CODE} } @{ $_[2] = mro::get_linear_isa( $_[1] ) }
105 # has nummification or boolification, AND fallback is *not* disabled
107 SQL::Abstract::_ENV_::DETECT_AUTOGENERATED_STRINGIFICATION
110 grep { *{"${_}::(0+"}{CODE} } @{$_[2]}
112 grep { *{"${_}::(bool"}{CODE} } @{$_[2]}
116 # no fallback specified at all
117 ! ( ($_[3]) = grep { *{"${_}::()"}{CODE} } @{$_[2]} )
119 # fallback explicitly undef
120 ! defined ${"$_[3]::()"}
133 #======================================================================
135 #======================================================================
139 my $class = ref($self) || $self;
140 my %opt = (ref $_[0] eq 'HASH') ? %{$_[0]} : @_;
142 # choose our case by keeping an option around
143 delete $opt{case} if $opt{case} && $opt{case} ne 'lower';
145 # default logic for interpreting arrayrefs
146 $opt{logic} = $opt{logic} ? uc $opt{logic} : 'OR';
148 # how to return bind vars
149 $opt{bindtype} ||= 'normal';
151 # default comparison is "=", but can be overridden
154 # try to recognize which are the 'equality' and 'inequality' ops
155 # (temporary quickfix (in 2007), should go through a more seasoned API)
156 $opt{equality_op} = qr/^( \Q$opt{cmp}\E | \= )$/ix;
157 $opt{inequality_op} = qr/^( != | <> )$/ix;
159 $opt{like_op} = qr/^ (is_)?r?like $/xi;
160 $opt{not_like_op} = qr/^ (is_)?not_r?like $/xi;
163 $opt{sqltrue} ||= '1=1';
164 $opt{sqlfalse} ||= '0=1';
167 $opt{special_ops} ||= [];
169 # regexes are applied in order, thus push after user-defines
170 push @{$opt{special_ops}}, @BUILTIN_SPECIAL_OPS;
172 if ($class->isa('DBIx::Class::SQLMaker')) {
173 push @{$opt{special_ops}}, our $DBIC_Compat_Op ||= {
174 regex => qr/^(?:ident|value|(?:not\s)?in)$/i, handler => sub { die "NOPE" }
176 $opt{is_dbic_sqlmaker} = 1;
180 $opt{unary_ops} ||= [];
182 # rudimentary sanity-check for user supplied bits treated as functions/operators
183 # If a purported function matches this regular expression, an exception is thrown.
184 # Literal SQL is *NOT* subject to this check, only functions (and column names
185 # when quoting is not in effect)
188 # need to guard against ()'s in column names too, but this will break tons of
189 # hacks... ideas anyone?
190 $opt{injection_guard} ||= qr/
196 $opt{expand_unary} = {};
199 -not => '_expand_not',
200 -bool => '_expand_bool',
201 -and => '_expand_op_andor',
202 -or => '_expand_op_andor',
203 -nest => '_expand_nest',
204 -bind => sub { shift; +{ @_ } },
206 -not_in => '_expand_in',
210 'between' => '_expand_between',
211 'not_between' => '_expand_between',
212 'in' => '_expand_in',
213 'not_in' => '_expand_in',
214 'nest' => '_expand_nest',
215 (map +($_ => '_expand_op_andor'), ('and', 'or')),
216 (map +($_ => '_expand_op_is'), ('is', 'is_not')),
219 # placeholder for _expand_unop system
221 my %unops = (-ident => '_expand_ident', -value => '_expand_value');
222 foreach my $name (keys %unops) {
223 $opt{expand}{$name} = $unops{$name};
224 my ($op) = $name =~ /^-(.*)$/;
225 $opt{expand_op}{$op} = sub {
226 my ($self, $op, $arg, $k) = @_;
227 return $self->_expand_expr_hashpair_cmp(
228 $k, { "-${op}" => $arg }
235 (map +("-$_", "_render_$_"), qw(op func bind ident literal list)),
240 (map +($_ => '_render_op_between'), 'between', 'not_between'),
241 (map +($_ => '_render_op_in'), 'in', 'not_in'),
242 (map +($_ => '_render_unop_postfix'),
243 'is_null', 'is_not_null', 'asc', 'desc',
245 (not => '_render_op_not'),
246 (map +($_ => '_render_op_andor'), qw(and or)),
247 ',' => '_render_op_multop',
250 return bless \%opt, $class;
253 sub sqltrue { +{ -literal => [ $_[0]->{sqltrue} ] } }
254 sub sqlfalse { +{ -literal => [ $_[0]->{sqlfalse} ] } }
256 sub _assert_pass_injection_guard {
257 if ($_[1] =~ $_[0]->{injection_guard}) {
258 my $class = ref $_[0];
259 puke "Possible SQL injection attempt '$_[1]'. If this is indeed a part of the "
260 . "desired SQL use literal SQL ( \'...' or \[ '...' ] ) or supply your own "
261 . "{injection_guard} attribute to ${class}->new()"
266 #======================================================================
268 #======================================================================
272 my $table = $self->_table(shift);
273 my $data = shift || return;
276 my $method = $self->_METHOD_FOR_refkind("_insert", $data);
277 my ($sql, @bind) = $self->$method($data);
278 $sql = join " ", $self->_sqlcase('insert into'), $table, $sql;
280 if ($options->{returning}) {
281 my ($s, @b) = $self->_insert_returning($options);
286 return wantarray ? ($sql, @bind) : $sql;
289 # So that subclasses can override INSERT ... RETURNING separately from
290 # UPDATE and DELETE (e.g. DBIx::Class::SQLMaker::Oracle does this)
291 sub _insert_returning { shift->_returning(@_) }
294 my ($self, $options) = @_;
296 my $f = $options->{returning};
298 my ($sql, @bind) = $self->render_aqt(
299 $self->_expand_maybe_list_expr($f, -ident)
302 ? $self->_sqlcase(' returning ') . $sql
303 : ($self->_sqlcase(' returning ').$sql, @bind);
306 sub _insert_HASHREF { # explicit list of fields and then values
307 my ($self, $data) = @_;
309 my @fields = sort keys %$data;
311 my ($sql, @bind) = $self->_insert_values($data);
314 $_ = $self->_quote($_) foreach @fields;
315 $sql = "( ".join(", ", @fields).") ".$sql;
317 return ($sql, @bind);
320 sub _insert_ARRAYREF { # just generate values(?,?) part (no list of fields)
321 my ($self, $data) = @_;
323 # no names (arrayref) so can't generate bindtype
324 $self->{bindtype} ne 'columns'
325 or belch "can't do 'columns' bindtype when called with arrayref";
327 my (@values, @all_bind);
328 foreach my $value (@$data) {
329 my ($values, @bind) = $self->_insert_value(undef, $value);
330 push @values, $values;
331 push @all_bind, @bind;
333 my $sql = $self->_sqlcase('values')." ( ".join(", ", @values)." )";
334 return ($sql, @all_bind);
337 sub _insert_ARRAYREFREF { # literal SQL with bind
338 my ($self, $data) = @_;
340 my ($sql, @bind) = @${$data};
341 $self->_assert_bindval_matches_bindtype(@bind);
343 return ($sql, @bind);
347 sub _insert_SCALARREF { # literal SQL without bind
348 my ($self, $data) = @_;
354 my ($self, $data) = @_;
356 my (@values, @all_bind);
357 foreach my $column (sort keys %$data) {
358 my ($values, @bind) = $self->_insert_value($column, $data->{$column});
359 push @values, $values;
360 push @all_bind, @bind;
362 my $sql = $self->_sqlcase('values')." ( ".join(", ", @values)." )";
363 return ($sql, @all_bind);
367 my ($self, $column, $v) = @_;
369 return $self->render_aqt(
370 $self->_expand_insert_value($column, $v)
374 sub _expand_insert_value {
375 my ($self, $column, $v) = @_;
377 if (ref($v) eq 'ARRAY') {
378 if ($self->{array_datatypes}) {
379 return +{ -bind => [ $column, $v ] };
381 my ($sql, @bind) = @$v;
382 $self->_assert_bindval_matches_bindtype(@bind);
383 return +{ -literal => $v };
385 if (ref($v) eq 'HASH') {
386 if (grep !/^-/, keys %$v) {
387 belch "HASH ref as bind value in insert is not supported";
388 return +{ -bind => [ $column, $v ] };
392 return +{ -bind => [ $column, undef ] };
394 local our $Cur_Col_Meta = $column;
395 return $self->expand_expr($v);
400 #======================================================================
402 #======================================================================
407 my $table = $self->_table(shift);
408 my $data = shift || return;
412 # first build the 'SET' part of the sql statement
413 puke "Unsupported data type specified to \$sql->update"
414 unless ref $data eq 'HASH';
416 my ($sql, @all_bind) = $self->_update_set_values($data);
417 $sql = $self->_sqlcase('update ') . $table . $self->_sqlcase(' set ')
421 my($where_sql, @where_bind) = $self->where($where);
423 push @all_bind, @where_bind;
426 if ($options->{returning}) {
427 my ($returning_sql, @returning_bind) = $self->_update_returning($options);
428 $sql .= $returning_sql;
429 push @all_bind, @returning_bind;
432 return wantarray ? ($sql, @all_bind) : $sql;
435 sub _update_set_values {
436 my ($self, $data) = @_;
438 return $self->render_aqt(
439 $self->_expand_update_set_values($data),
443 sub _expand_update_set_values {
444 my ($self, $data) = @_;
445 $self->_expand_maybe_list_expr( [
448 $set = { -bind => $_ } unless defined $set;
449 +{ -op => [ '=', $self->_expand_ident(-ident => $k), $set ] };
455 ? ($self->{array_datatypes}
456 ? [ $k, +{ -bind => [ $k, $v ] } ]
457 : [ $k, +{ -literal => $v } ])
459 local our $Cur_Col_Meta = $k;
460 [ $k, $self->_expand_expr($v) ]
467 # So that subclasses can override UPDATE ... RETURNING separately from
469 sub _update_returning { shift->_returning(@_) }
473 #======================================================================
475 #======================================================================
480 my $table = $self->_table(shift);
481 my $fields = shift || '*';
485 my ($fields_sql, @bind) = $self->_select_fields($fields);
487 my ($where_sql, @where_bind) = $self->where($where, $order);
488 push @bind, @where_bind;
490 my $sql = join(' ', $self->_sqlcase('select'), $fields_sql,
491 $self->_sqlcase('from'), $table)
494 return wantarray ? ($sql, @bind) : $sql;
498 my ($self, $fields) = @_;
499 return $fields unless ref($fields);
500 return $self->render_aqt(
501 $self->_expand_maybe_list_expr($fields, '-ident')
505 #======================================================================
507 #======================================================================
512 my $table = $self->_table(shift);
516 my($where_sql, @bind) = $self->where($where);
517 my $sql = $self->_sqlcase('delete from ') . $table . $where_sql;
519 if ($options->{returning}) {
520 my ($returning_sql, @returning_bind) = $self->_delete_returning($options);
521 $sql .= $returning_sql;
522 push @bind, @returning_bind;
525 return wantarray ? ($sql, @bind) : $sql;
528 # So that subclasses can override DELETE ... RETURNING separately from
530 sub _delete_returning { shift->_returning(@_) }
534 #======================================================================
536 #======================================================================
540 # Finally, a separate routine just to handle WHERE clauses
542 my ($self, $where, $order) = @_;
544 local $self->{convert_where} = $self->{convert};
547 my ($sql, @bind) = defined($where)
548 ? $self->_recurse_where($where)
550 $sql = (defined $sql and length $sql) ? $self->_sqlcase(' where ') . "( $sql )" : '';
554 my ($order_sql, @order_bind) = $self->_order_by($order);
556 push @bind, @order_bind;
559 return wantarray ? ($sql, @bind) : $sql;
562 { our $Default_Scalar_To = -value }
565 my ($self, $expr, $default_scalar_to) = @_;
566 local our $Default_Scalar_To = $default_scalar_to if $default_scalar_to;
567 $self->_expand_expr($expr);
571 my ($self, $aqt) = @_;
572 my ($k, $v, @rest) = %$aqt;
574 if (my $meth = $self->{render}{$k}) {
575 return $self->$meth($v);
577 die "notreached: $k";
581 my ($self, $expr) = @_;
582 $self->render_aqt($self->expand_expr($expr));
586 my ($self, $raw) = @_;
587 s/^-(?=.)//, s/\s+/_/g for my $op = lc $raw;
592 my ($self, $expr) = @_;
593 our $Expand_Depth ||= 0; local $Expand_Depth = $Expand_Depth + 1;
594 return undef unless defined($expr);
595 if (ref($expr) eq 'HASH') {
596 return undef unless my $kc = keys %$expr;
598 return $self->_expand_op_andor(-and => $expr);
600 my ($key, $value) = %$expr;
601 if ($key =~ /^-/ and $key =~ s/ [_\s]? \d+ $//x ) {
602 belch 'Use of [and|or|nest]_N modifiers is deprecated and will be removed in SQLA v2.0. '
603 . "You probably wanted ...-and => [ $key => COND1, $key => COND2 ... ]";
605 if (my $exp = $self->{expand}{$key}) {
606 return $self->$exp($key, $value);
608 return $self->_expand_expr_hashpair($key, $value);
610 if (ref($expr) eq 'ARRAY') {
611 my $logic = '-'.lc($self->{logic});
612 return $self->_expand_op_andor($logic, $expr);
614 if (my $literal = is_literal_value($expr)) {
615 return +{ -literal => $literal };
617 if (!ref($expr) or Scalar::Util::blessed($expr)) {
618 return $self->_expand_expr_scalar($expr);
623 sub _expand_expr_hashpair {
624 my ($self, $k, $v) = @_;
625 unless (defined($k) and length($k)) {
626 if (defined($k) and my $literal = is_literal_value($v)) {
627 belch 'Hash-pairs consisting of an empty string with a literal are deprecated, and will be removed in 2.0: use -and => [ $literal ] instead';
628 return { -literal => $literal };
630 puke "Supplying an empty left hand side argument is not supported";
633 return $self->_expand_expr_hashpair_op($k, $v);
635 return $self->_expand_expr_hashpair_ident($k, $v);
638 sub _expand_expr_hashpair_ident {
639 my ($self, $k, $v) = @_;
641 local our $Cur_Col_Meta = $k;
643 # hash with multiple or no elements is andor
645 if (ref($v) eq 'HASH' and keys %$v != 1) {
646 return $self->_expand_op_andor(-and => $v, $k);
649 # undef needs to be re-sent with cmp to achieve IS/IS NOT NULL
651 if (is_undef_value($v)) {
652 return $self->_expand_expr_hashpair_cmp($k => undef);
655 # scalars and objects get expanded as whatever requested or values
657 if (!ref($v) or Scalar::Util::blessed($v)) {
658 return $self->_expand_expr_hashpair_scalar($k, $v);
661 # single key hashref is a hashtriple
663 if (ref($v) eq 'HASH') {
664 return $self->_expand_expr_hashtriple($k, %$v);
667 # arrayref needs re-engineering over the elements
669 if (ref($v) eq 'ARRAY') {
670 return $self->sqlfalse unless @$v;
671 $self->_debug("ARRAY($k) means distribute over elements");
673 $v->[0] =~ /^-(and|or)$/i
674 ? shift(@{$v = [ @$v ]})
675 : '-'.lc($self->{logic} || 'OR')
677 return $self->_expand_op_andor(
682 if (my $literal = is_literal_value($v)) {
684 belch 'Hash-pairs consisting of an empty string with a literal are deprecated, and will be removed in 2.0: use -and => [ $literal ] instead';
687 my ($sql, @bind) = @$literal;
688 if ($self->{bindtype} eq 'columns') {
690 $self->_assert_bindval_matches_bindtype($_);
693 return +{ -literal => [ $self->_quote($k).' '.$sql, @bind ] };
698 sub _expand_expr_scalar {
699 my ($self, $expr) = @_;
701 return $self->_expand_expr({ (our $Default_Scalar_To) => $expr });
704 sub _expand_expr_hashpair_scalar {
705 my ($self, $k, $v) = @_;
707 return $self->_expand_expr_hashpair_cmp(
708 $k, $self->_expand_expr_scalar($v),
712 sub _expand_expr_hashpair_op {
713 my ($self, $k, $v) = @_;
715 $self->_assert_pass_injection_guard($k =~ /\A-(.*)\Z/s);
717 my $op = $self->_normalize_op($k);
719 # Ops prefixed with -not_ get converted
721 if (my ($rest) = $op =~/^not_(.*)$/) {
724 $self->_expand_expr({ "-${rest}", $v })
730 my $op = join(' ', split '_', $op);
732 # the old special op system requires illegality for top-level use
735 (our $Expand_Depth) == 1
736 and List::Util::first { $op =~ $_->{regex} } @{$self->{special_ops}}
738 puke "Illegal use of top-level '-$op'"
741 # the old unary op system means we should touch nothing and let it work
743 if (my $us = List::Util::first { $op =~ $_->{regex} } @{$self->{unary_ops}}) {
744 return { -op => [ $op, $v ] };
748 # an explicit node type is currently assumed to be expanded (this is almost
749 # certainly wrong and there should be expansion anyway)
751 if ($self->{render}{$k}) {
755 # hashref RHS values get expanded and used as op/func args
760 and (keys %$v)[0] =~ /^-/
762 my ($func) = $k =~ /^-(.*)$/;
764 if (List::Util::first { $func =~ $_->{regex} } @{$self->{special_ops}}) {
765 return +{ -op => [ $func, $self->_expand_expr($v) ] };
768 return +{ -func => [ $func, $self->_expand_expr($v) ] };
771 # scalars and literals get simply expanded
773 if (!ref($v) or is_literal_value($v)) {
774 return +{ -op => [ $op, $self->_expand_expr($v) ] };
780 sub _expand_expr_hashpair_cmp {
781 my ($self, $k, $v) = @_;
782 $self->_expand_expr_hashtriple($k, $self->{cmp}, $v);
785 sub _expand_expr_hashtriple {
786 my ($self, $k, $vk, $vv) = @_;
788 my $ik = $self->_expand_ident(-ident => $k);
790 my $op = $self->_normalize_op($vk);
791 $self->_assert_pass_injection_guard($op);
793 if ($op =~ s/ _? \d+ $//x ) {
794 return $self->_expand_expr($k, { $vk, $vv });
796 if (my $x = $self->{expand_op}{$op}) {
797 local our $Cur_Col_Meta = $k;
798 return $self->$x($op, $vv, $k);
802 my $op = join(' ', split '_', $op);
804 if (my $us = List::Util::first { $op =~ $_->{regex} } @{$self->{special_ops}}) {
805 return { -op => [ $op, $ik, $vv ] };
807 if (my $us = List::Util::first { $op =~ $_->{regex} } @{$self->{unary_ops}}) {
811 { -op => [ $op, $vv ] }
815 if (ref($vv) eq 'ARRAY') {
817 my $logic = (defined($raw[0]) and $raw[0] =~ /^-(and|or)$/i)
818 ? shift @raw : '-or';
819 my @values = map +{ $vk => $_ }, @raw;
821 $op =~ $self->{inequality_op}
822 or $op =~ $self->{not_like_op}
824 if (lc($logic) eq '-or' and @values > 1) {
825 belch "A multi-element arrayref as an argument to the inequality op '${\uc(join ' ', split '_', $op)}' "
826 . 'is technically equivalent to an always-true 1=1 (you probably wanted '
827 . "to say ...{ \$inequality_op => [ -and => \@values ] }... instead)"
832 # try to DWIM on equality operators
833 return ($self->_dwim_op_to_is($op,
834 "Supplying an empty arrayref to '%s' is deprecated",
835 "operator '%s' applied on an empty array (field '$k')"
836 ) ? $self->sqlfalse : $self->sqltrue);
838 return $self->_expand_op_andor($logic => \@values, $k);
840 if (is_undef_value($vv)) {
841 my $is = ($self->_dwim_op_to_is($op,
842 "Supplying an undefined argument to '%s' is deprecated",
843 "unexpected operator '%s' with undef operand",
844 ) ? 'is' : 'is not');
846 return $self->_expand_expr_hashpair($k => { $is, undef });
848 local our $Cur_Col_Meta = $k;
852 $self->_expand_expr($vv)
857 my ($self, $raw, $empty, $fail) = @_;
859 my $op = $self->_normalize_op($raw);
861 if ($op =~ /^not$/i) {
864 if ($op =~ $self->{equality_op}) {
867 if ($op =~ $self->{like_op}) {
868 belch(sprintf $empty, uc(join ' ', split '_', $op));
871 if ($op =~ $self->{inequality_op}) {
874 if ($op =~ $self->{not_like_op}) {
875 belch(sprintf $empty, uc(join ' ', split '_', $op));
878 puke(sprintf $fail, $op);
882 my ($self, $op, $body) = @_;
883 unless (defined($body) or (ref($body) and ref($body) eq 'ARRAY')) {
884 puke "$op requires a single plain scalar argument (a quotable identifier) or an arrayref of identifier parts";
886 my @parts = map split(/\Q${\($self->{name_sep}||'.')}\E/, $_),
887 ref($body) ? @$body : $body;
888 return { -ident => $parts[-1] } if $self->{_dequalify_idents};
889 unless ($self->{quote_char}) {
890 $self->_assert_pass_injection_guard($_) for @parts;
892 return +{ -ident => \@parts };
896 +{ -bind => [ our $Cur_Col_Meta, $_[2] ] };
900 +{ -op => [ 'not', $_[0]->_expand_expr($_[2]) ] };
904 my ($self, undef, $v) = @_;
906 return $self->_expand_expr($v);
908 puke "-bool => undef not supported" unless defined($v);
909 return $self->_expand_ident(-ident => $v);
912 sub _expand_op_andor {
913 my ($self, $logic, $v, $k) = @_;
915 $v = [ map +{ $k, $_ },
917 ? (map +{ $_ => $v->{$_} }, sort keys %$v)
921 my ($logop) = $logic =~ /^-?(.*)$/;
922 if (ref($v) eq 'HASH') {
923 return undef unless keys %$v;
926 map $self->_expand_expr({ $_ => $v->{$_} }),
930 if (ref($v) eq 'ARRAY') {
931 $logop eq 'and' or $logop eq 'or' or puke "unknown logic: $logop";
934 (ref($_) eq 'ARRAY' and @$_)
935 or (ref($_) eq 'HASH' and %$_)
941 while (my ($el) = splice @expr, 0, 1) {
942 puke "Supplying an empty left hand side argument is not supported in array-pairs"
943 unless defined($el) and length($el);
944 my $elref = ref($el);
946 local our $Expand_Depth = 0;
947 push(@res, grep defined, $self->_expand_expr({ $el, shift(@expr) }));
948 } elsif ($elref eq 'ARRAY') {
949 push(@res, grep defined, $self->_expand_expr($el)) if @$el;
950 } elsif (my $l = is_literal_value($el)) {
951 push @res, { -literal => $l };
952 } elsif ($elref eq 'HASH') {
953 local our $Expand_Depth = 0;
954 push @res, grep defined, $self->_expand_expr($el) if %$el;
960 # return $res[0] if @res == 1;
961 return { -op => [ $logop, @res ] };
967 my ($self, $op, $vv, $k) = @_;
968 puke "$op can only take undef as argument"
972 and exists($vv->{-value})
973 and !defined($vv->{-value})
975 return +{ -op => [ $op.'_null', $self->_expand_ident(-ident => $k) ] };
978 sub _expand_between {
979 my ($self, $op, $vv, $k) = @_;
980 local our $Cur_Col_Meta = $k;
981 my @rhs = map $self->_expand_expr($_),
982 ref($vv) eq 'ARRAY' ? @$vv : $vv;
984 (@rhs == 1 and ref($rhs[0]) eq 'HASH' and $rhs[0]->{-literal})
986 (@rhs == 2 and defined($rhs[0]) and defined($rhs[1]))
988 puke "Operator '${\uc($op)}' requires either an arrayref with two defined values or expressions, or a single literal scalarref/arrayref-ref";
992 $self->_expand_ident(-ident => $k),
998 my ($self, $raw, $vv, $k) = @_;
999 $k = shift @{$vv = [ @$vv ]} unless defined $k;
1000 local our $Cur_Col_Meta = $k;
1001 my $op = $self->_normalize_op($raw);
1002 if (my $literal = is_literal_value($vv)) {
1003 my ($sql, @bind) = @$literal;
1004 my $opened_sql = $self->_open_outer_paren($sql);
1006 $op, $self->_expand_ident(-ident => $k),
1007 [ { -literal => [ $opened_sql, @bind ] } ]
1011 'SQL::Abstract before v1.75 used to generate incorrect SQL when the '
1012 . "-${\uc($op)} operator was given an undef-containing list: !!!AUDIT YOUR CODE "
1013 . 'AND DATA!!! (the upcoming Data::Query-based version of SQL::Abstract '
1014 . 'will emit the logically correct SQL instead of raising this exception)'
1016 puke("Argument passed to the '${\uc($op)}' operator can not be undefined")
1018 my @rhs = map $self->expand_expr($_, -value),
1019 map { defined($_) ? $_: puke($undef_err) }
1020 (ref($vv) eq 'ARRAY' ? @$vv : $vv);
1021 return $self->${\($op =~ /^not/ ? 'sqltrue' : 'sqlfalse')} unless @rhs;
1025 $self->_expand_ident(-ident => $k),
1031 my ($self, $op, $v) = @_;
1032 # DBIx::Class requires a nest warning to be emitted once but the private
1033 # method it overrode to do so no longer exists
1034 if ($self->{is_dbic_sqlmaker}) {
1035 unless (our $Nest_Warned) {
1037 "-nest in search conditions is deprecated, you most probably wanted:\n"
1038 .q|{..., -and => [ \%cond0, \@cond1, \'cond2', \[ 'cond3', [ col => bind ] ], etc. ], ... }|
1043 return $self->_expand_expr($v);
1046 sub _recurse_where {
1047 my ($self, $where, $logic) = @_;
1049 # Special case: top level simple string treated as literal
1051 my $where_exp = (ref($where)
1052 ? $self->_expand_expr($where, $logic)
1053 : { -literal => [ $where ] });
1055 # dispatch expanded expression
1057 my ($sql, @bind) = defined($where_exp) ? $self->render_aqt($where_exp) : (undef);
1058 # DBIx::Class used to call _recurse_where in scalar context
1059 # something else might too...
1061 return ($sql, @bind);
1064 belch "Calling _recurse_where in scalar context is deprecated and will go away before 2.0";
1070 my ($self, $ident) = @_;
1072 return $self->_convert($self->_quote($ident));
1076 my ($self, $list) = @_;
1077 my @parts = grep length($_->[0]), map [ $self->render_aqt($_) ], @$list;
1078 return join(', ', map $_->[0], @parts), map @{$_}[1..$#$_], @parts;
1082 my ($self, $rest) = @_;
1083 my ($func, @args) = @$rest;
1087 push @arg_sql, shift @x;
1089 } map [ $self->render_aqt($_) ], @args;
1090 return ($self->_sqlcase($func).'('.join(', ', @arg_sql).')', @bind);
1094 my ($self, $bind) = @_;
1095 return ($self->_convert('?'), $self->_bindtype(@$bind));
1098 sub _render_literal {
1099 my ($self, $literal) = @_;
1100 $self->_assert_bindval_matches_bindtype(@{$literal}[1..$#$literal]);
1105 my ($self, $v) = @_;
1106 my ($op, @args) = @$v;
1107 if (my $r = $self->{render_op}{$op}) {
1108 return $self->$r($op, \@args);
1113 my $op = join(' ', split '_', $op);
1115 my $us = List::Util::first { $op =~ $_->{regex} } @{$self->{special_ops}};
1116 if ($us and @args > 1) {
1117 puke "Special op '${op}' requires first value to be identifier"
1118 unless my ($ident) = map $_->{-ident}, grep ref($_) eq 'HASH', $args[0];
1119 my $k = join(($self->{name_sep}||'.'), @$ident);
1120 local our $Expand_Depth = 1;
1121 return $self->${\($us->{handler})}($k, $op, $args[1]);
1123 if (my $us = List::Util::first { $op =~ $_->{regex} } @{$self->{unary_ops}}) {
1124 return $self->${\($us->{handler})}($op, $args[0]);
1129 return $self->_render_unop_prefix($op, \@args);
1131 return $self->_render_op_multop($op, \@args);
1137 sub _render_op_between {
1138 my ($self, $op, $args) = @_;
1139 my ($left, $low, $high) = @$args;
1140 my ($rhsql, @rhbind) = do {
1142 puke "Single arg to between must be a literal"
1143 unless $low->{-literal};
1146 my ($l, $h) = map [ $self->render_aqt($_) ], $low, $high;
1147 (join(' ', $l->[0], $self->_sqlcase('and'), $h->[0]),
1148 @{$l}[1..$#$l], @{$h}[1..$#$h])
1151 my ($lhsql, @lhbind) = $self->render_aqt($left);
1155 $self->_sqlcase(join ' ', split '_', $op),
1163 my ($self, $op, $args) = @_;
1164 my ($lhs, $rhs) = @$args;
1167 my ($sql, @bind) = $self->render_aqt($_);
1168 push @in_bind, @bind;
1171 my ($lhsql, @lbind) = $self->render_aqt($lhs);
1173 $lhsql.' '.$self->_sqlcase(join ' ', split '_', $op).' ( '
1174 .join(', ', @in_sql)
1180 sub _render_op_andor {
1181 my ($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 ($sql, @bind) = $self->_render_op_multop($op, $args);
1186 return '( '.$sql.' )', @bind;
1189 sub _render_op_multop {
1190 my ($self, $op, $args) = @_;
1191 my @parts = grep length($_->[0]), map [ $self->render_aqt($_) ], @$args;
1192 return '' unless @parts;
1193 return @{$parts[0]} if @parts == 1;
1194 my ($final_sql) = join(
1195 ' '.$self->_sqlcase(join ' ', split '_', $op).' ',
1200 map @{$_}[1..$#$_], @parts
1203 sub _render_op_not {
1204 my ($self, $op, $v) = @_;
1205 my ($sql, @bind) = $self->_render_unop_prefix($op, $v);
1206 return "(${sql})", @bind;
1209 sub _render_unop_prefix {
1210 my ($self, $op, $v) = @_;
1211 my ($expr_sql, @bind) = $self->render_aqt($v->[0]);
1213 my $op_sql = $self->_sqlcase($op); # join ' ', split '_', $op);
1214 return ("${op_sql} ${expr_sql}", @bind);
1217 sub _render_unop_postfix {
1218 my ($self, $op, $v) = @_;
1219 my ($expr_sql, @bind) = $self->render_aqt($v->[0]);
1220 my $op_sql = $self->_sqlcase(join ' ', split '_', $op);
1221 return ($expr_sql.' '.$op_sql, @bind);
1224 # Some databases (SQLite) treat col IN (1, 2) different from
1225 # col IN ( (1, 2) ). Use this to strip all outer parens while
1226 # adding them back in the corresponding method
1227 sub _open_outer_paren {
1228 my ($self, $sql) = @_;
1230 while (my ($inner) = $sql =~ /^ \s* \( (.*) \) \s* $/xs) {
1232 # there are closing parens inside, need the heavy duty machinery
1233 # to reevaluate the extraction starting from $sql (full reevaluation)
1234 if ($inner =~ /\)/) {
1235 require Text::Balanced;
1237 my (undef, $remainder) = do {
1238 # idiotic design - writes to $@ but *DOES NOT* throw exceptions
1240 Text::Balanced::extract_bracketed($sql, '()', qr/\s*/);
1243 # the entire expression needs to be a balanced bracketed thing
1244 # (after an extract no remainder sans trailing space)
1245 last if defined $remainder and $remainder =~ /\S/;
1255 #======================================================================
1257 #======================================================================
1259 sub _expand_order_by {
1260 my ($self, $arg) = @_;
1262 return unless defined($arg) and not (ref($arg) eq 'ARRAY' and !@$arg);
1264 my $expander = sub {
1265 my ($self, $dir, $expr) = @_;
1266 my @to_expand = ref($expr) eq 'ARRAY' ? @$expr : $expr;
1267 foreach my $arg (@to_expand) {
1271 and grep /^-(asc|desc)$/, keys %$arg
1273 puke "ordering direction hash passed to order by must have exactly one key (-asc or -desc)";
1277 defined($dir) ? { -op => [ $dir =~ /^-?(.*)$/ ,=> $_ ] } : $_
1279 map $self->expand_expr($_, -ident),
1280 map ref($_) eq 'ARRAY' ? @$_ : $_, @to_expand;
1281 return undef unless @exp;
1282 return undef if @exp == 1 and not defined($exp[0]);
1283 return +{ -op => [ ',', @exp ] };
1286 local @{$self->{expand}}{qw(-asc -desc)} = (($expander) x 2);
1288 return $self->$expander(undef, $arg);
1292 my ($self, $arg) = @_;
1294 return '' unless defined(my $expanded = $self->_expand_order_by($arg));
1296 my ($sql, @bind) = $self->render_aqt($expanded);
1298 return '' unless length($sql);
1300 my $final_sql = $self->_sqlcase(' order by ').$sql;
1302 return wantarray ? ($final_sql, @bind) : $final_sql;
1305 # _order_by no longer needs to call this so doesn't but DBIC uses it.
1307 sub _order_by_chunks {
1308 my ($self, $arg) = @_;
1310 return () unless defined(my $expanded = $self->_expand_order_by($arg));
1312 return $self->_chunkify_order_by($expanded);
1315 sub _chunkify_order_by {
1316 my ($self, $expanded) = @_;
1318 return grep length, $self->render_aqt($expanded)
1319 if $expanded->{-ident} or @{$expanded->{-literal}||[]} == 1;
1322 if (ref() eq 'HASH' and $_->{-op} and $_->{-op}[0] eq ',') {
1323 my ($comma, @list) = @{$_->{-op}};
1324 return map $self->_chunkify_order_by($_), @list;
1326 return [ $self->render_aqt($_) ];
1330 #======================================================================
1331 # DATASOURCE (FOR NOW, JUST PLAIN TABLE OR LIST OF TABLES)
1332 #======================================================================
1338 $self->_expand_maybe_list_expr($from, -ident)
1343 #======================================================================
1345 #======================================================================
1347 sub _expand_maybe_list_expr {
1348 my ($self, $expr, $default) = @_;
1350 if (ref($expr) eq 'ARRAY') {
1352 map $self->expand_expr($_, $default), @$expr
1359 return $self->expand_expr($e, $default);
1362 # highly optimized, as it's called way too often
1364 # my ($self, $label) = @_;
1366 return '' unless defined $_[1];
1367 return ${$_[1]} if ref($_[1]) eq 'SCALAR';
1368 puke 'Identifier cannot be hashref' if ref($_[1]) eq 'HASH';
1370 unless ($_[0]->{quote_char}) {
1371 if (ref($_[1]) eq 'ARRAY') {
1372 return join($_[0]->{name_sep}||'.', @{$_[1]});
1374 $_[0]->_assert_pass_injection_guard($_[1]);
1379 my $qref = ref $_[0]->{quote_char};
1381 !$qref ? ($_[0]->{quote_char}, $_[0]->{quote_char})
1382 : ($qref eq 'ARRAY') ? @{$_[0]->{quote_char}}
1383 : puke "Unsupported quote_char format: $_[0]->{quote_char}";
1385 my $esc = $_[0]->{escape_char} || $r;
1387 # parts containing * are naturally unquoted
1389 $_[0]->{name_sep}||'',
1393 : do { (my $n = $_) =~ s/(\Q$esc\E|\Q$r\E)/$esc$1/g; $l . $n . $r }
1395 (ref($_[1]) eq 'ARRAY'
1399 ? split (/\Q$_[0]->{name_sep}\E/, $_[1] )
1407 # Conversion, if applicable
1409 #my ($self, $arg) = @_;
1410 if ($_[0]->{convert_where}) {
1411 return $_[0]->_sqlcase($_[0]->{convert_where}) .'(' . $_[1] . ')';
1418 #my ($self, $col, @vals) = @_;
1419 # called often - tighten code
1420 return $_[0]->{bindtype} eq 'columns'
1421 ? map {[$_[1], $_]} @_[2 .. $#_]
1426 # Dies if any element of @bind is not in [colname => value] format
1427 # if bindtype is 'columns'.
1428 sub _assert_bindval_matches_bindtype {
1429 # my ($self, @bind) = @_;
1431 if ($self->{bindtype} eq 'columns') {
1433 if (!defined $_ || ref($_) ne 'ARRAY' || @$_ != 2) {
1434 puke "bindtype 'columns' selected, you need to pass: [column_name => bind_value]"
1440 sub _join_sql_clauses {
1441 my ($self, $logic, $clauses_aref, $bind_aref) = @_;
1443 if (@$clauses_aref > 1) {
1444 my $join = " " . $self->_sqlcase($logic) . " ";
1445 my $sql = '( ' . join($join, @$clauses_aref) . ' )';
1446 return ($sql, @$bind_aref);
1448 elsif (@$clauses_aref) {
1449 return ($clauses_aref->[0], @$bind_aref); # no parentheses
1452 return (); # if no SQL, ignore @$bind_aref
1457 # Fix SQL case, if so requested
1459 # LDNOTE: if $self->{case} is true, then it contains 'lower', so we
1460 # don't touch the argument ... crooked logic, but let's not change it!
1461 return $_[0]->{case} ? $_[1] : uc($_[1]);
1465 #======================================================================
1466 # DISPATCHING FROM REFKIND
1467 #======================================================================
1470 my ($self, $data) = @_;
1472 return 'UNDEF' unless defined $data;
1474 # blessed objects are treated like scalars
1475 my $ref = (Scalar::Util::blessed $data) ? '' : ref $data;
1477 return 'SCALAR' unless $ref;
1480 while ($ref eq 'REF') {
1482 $ref = (Scalar::Util::blessed $data) ? '' : ref $data;
1486 return ($ref||'SCALAR') . ('REF' x $n_steps);
1490 my ($self, $data) = @_;
1491 my @try = ($self->_refkind($data));
1492 push @try, 'SCALAR_or_UNDEF' if $try[0] eq 'SCALAR' || $try[0] eq 'UNDEF';
1493 push @try, 'FALLBACK';
1497 sub _METHOD_FOR_refkind {
1498 my ($self, $meth_prefix, $data) = @_;
1501 for (@{$self->_try_refkind($data)}) {
1502 $method = $self->can($meth_prefix."_".$_)
1506 return $method || puke "cannot dispatch on '$meth_prefix' for ".$self->_refkind($data);
1510 sub _SWITCH_refkind {
1511 my ($self, $data, $dispatch_table) = @_;
1514 for (@{$self->_try_refkind($data)}) {
1515 $coderef = $dispatch_table->{$_}
1519 puke "no dispatch entry for ".$self->_refkind($data)
1528 #======================================================================
1529 # VALUES, GENERATE, AUTOLOAD
1530 #======================================================================
1532 # LDNOTE: original code from nwiger, didn't touch code in that section
1533 # I feel the AUTOLOAD stuff should not be the default, it should
1534 # only be activated on explicit demand by user.
1538 my $data = shift || return;
1539 puke "Argument to ", __PACKAGE__, "->values must be a \\%hash"
1540 unless ref $data eq 'HASH';
1543 foreach my $k (sort keys %$data) {
1544 my $v = $data->{$k};
1545 $self->_SWITCH_refkind($v, {
1547 if ($self->{array_datatypes}) { # array datatype
1548 push @all_bind, $self->_bindtype($k, $v);
1550 else { # literal SQL with bind
1551 my ($sql, @bind) = @$v;
1552 $self->_assert_bindval_matches_bindtype(@bind);
1553 push @all_bind, @bind;
1556 ARRAYREFREF => sub { # literal SQL with bind
1557 my ($sql, @bind) = @${$v};
1558 $self->_assert_bindval_matches_bindtype(@bind);
1559 push @all_bind, @bind;
1561 SCALARREF => sub { # literal SQL without bind
1563 SCALAR_or_UNDEF => sub {
1564 push @all_bind, $self->_bindtype($k, $v);
1575 my(@sql, @sqlq, @sqlv);
1579 if ($ref eq 'HASH') {
1580 for my $k (sort keys %$_) {
1583 my $label = $self->_quote($k);
1584 if ($r eq 'ARRAY') {
1585 # literal SQL with bind
1586 my ($sql, @bind) = @$v;
1587 $self->_assert_bindval_matches_bindtype(@bind);
1588 push @sqlq, "$label = $sql";
1590 } elsif ($r eq 'SCALAR') {
1591 # literal SQL without bind
1592 push @sqlq, "$label = $$v";
1594 push @sqlq, "$label = ?";
1595 push @sqlv, $self->_bindtype($k, $v);
1598 push @sql, $self->_sqlcase('set'), join ', ', @sqlq;
1599 } elsif ($ref eq 'ARRAY') {
1600 # unlike insert(), assume these are ONLY the column names, i.e. for SQL
1603 if ($r eq 'ARRAY') { # literal SQL with bind
1604 my ($sql, @bind) = @$v;
1605 $self->_assert_bindval_matches_bindtype(@bind);
1608 } elsif ($r eq 'SCALAR') { # literal SQL without bind
1609 # embedded literal SQL
1616 push @sql, '(' . join(', ', @sqlq) . ')';
1617 } elsif ($ref eq 'SCALAR') {
1621 # strings get case twiddled
1622 push @sql, $self->_sqlcase($_);
1626 my $sql = join ' ', @sql;
1628 # this is pretty tricky
1629 # if ask for an array, return ($stmt, @bind)
1630 # otherwise, s/?/shift @sqlv/ to put it inline
1632 return ($sql, @sqlv);
1634 1 while $sql =~ s/\?/my $d = shift(@sqlv);
1635 ref $d ? $d->[1] : $d/e;
1644 # This allows us to check for a local, then _form, attr
1646 my($name) = $AUTOLOAD =~ /.*::(.+)/;
1647 return $self->generate($name, @_);
1658 SQL::Abstract - Generate SQL from Perl data structures
1664 my $sql = SQL::Abstract->new;
1666 my($stmt, @bind) = $sql->select($source, \@fields, \%where, $order);
1668 my($stmt, @bind) = $sql->insert($table, \%fieldvals || \@values);
1670 my($stmt, @bind) = $sql->update($table, \%fieldvals, \%where);
1672 my($stmt, @bind) = $sql->delete($table, \%where);
1674 # Then, use these in your DBI statements
1675 my $sth = $dbh->prepare($stmt);
1676 $sth->execute(@bind);
1678 # Just generate the WHERE clause
1679 my($stmt, @bind) = $sql->where(\%where, $order);
1681 # Return values in the same order, for hashed queries
1682 # See PERFORMANCE section for more details
1683 my @bind = $sql->values(\%fieldvals);
1687 This module was inspired by the excellent L<DBIx::Abstract>.
1688 However, in using that module I found that what I really wanted
1689 to do was generate SQL, but still retain complete control over my
1690 statement handles and use the DBI interface. So, I set out to
1691 create an abstract SQL generation module.
1693 While based on the concepts used by L<DBIx::Abstract>, there are
1694 several important differences, especially when it comes to WHERE
1695 clauses. I have modified the concepts used to make the SQL easier
1696 to generate from Perl data structures and, IMO, more intuitive.
1697 The underlying idea is for this module to do what you mean, based
1698 on the data structures you provide it. The big advantage is that
1699 you don't have to modify your code every time your data changes,
1700 as this module figures it out.
1702 To begin with, an SQL INSERT is as easy as just specifying a hash
1703 of C<key=value> pairs:
1706 name => 'Jimbo Bobson',
1707 phone => '123-456-7890',
1708 address => '42 Sister Lane',
1709 city => 'St. Louis',
1710 state => 'Louisiana',
1713 The SQL can then be generated with this:
1715 my($stmt, @bind) = $sql->insert('people', \%data);
1717 Which would give you something like this:
1719 $stmt = "INSERT INTO people
1720 (address, city, name, phone, state)
1721 VALUES (?, ?, ?, ?, ?)";
1722 @bind = ('42 Sister Lane', 'St. Louis', 'Jimbo Bobson',
1723 '123-456-7890', 'Louisiana');
1725 These are then used directly in your DBI code:
1727 my $sth = $dbh->prepare($stmt);
1728 $sth->execute(@bind);
1730 =head2 Inserting and Updating Arrays
1732 If your database has array types (like for example Postgres),
1733 activate the special option C<< array_datatypes => 1 >>
1734 when creating the C<SQL::Abstract> object.
1735 Then you may use an arrayref to insert and update database array types:
1737 my $sql = SQL::Abstract->new(array_datatypes => 1);
1739 planets => [qw/Mercury Venus Earth Mars/]
1742 my($stmt, @bind) = $sql->insert('solar_system', \%data);
1746 $stmt = "INSERT INTO solar_system (planets) VALUES (?)"
1748 @bind = (['Mercury', 'Venus', 'Earth', 'Mars']);
1751 =head2 Inserting and Updating SQL
1753 In order to apply SQL functions to elements of your C<%data> you may
1754 specify a reference to an arrayref for the given hash value. For example,
1755 if you need to execute the Oracle C<to_date> function on a value, you can
1756 say something like this:
1760 date_entered => \[ "to_date(?,'MM/DD/YYYY')", "03/02/2003" ],
1763 The first value in the array is the actual SQL. Any other values are
1764 optional and would be included in the bind values array. This gives
1767 my($stmt, @bind) = $sql->insert('people', \%data);
1769 $stmt = "INSERT INTO people (name, date_entered)
1770 VALUES (?, to_date(?,'MM/DD/YYYY'))";
1771 @bind = ('Bill', '03/02/2003');
1773 An UPDATE is just as easy, all you change is the name of the function:
1775 my($stmt, @bind) = $sql->update('people', \%data);
1777 Notice that your C<%data> isn't touched; the module will generate
1778 the appropriately quirky SQL for you automatically. Usually you'll
1779 want to specify a WHERE clause for your UPDATE, though, which is
1780 where handling C<%where> hashes comes in handy...
1782 =head2 Complex where statements
1784 This module can generate pretty complicated WHERE statements
1785 easily. For example, simple C<key=value> pairs are taken to mean
1786 equality, and if you want to see if a field is within a set
1787 of values, you can use an arrayref. Let's say we wanted to
1788 SELECT some data based on this criteria:
1791 requestor => 'inna',
1792 worker => ['nwiger', 'rcwe', 'sfz'],
1793 status => { '!=', 'completed' }
1796 my($stmt, @bind) = $sql->select('tickets', '*', \%where);
1798 The above would give you something like this:
1800 $stmt = "SELECT * FROM tickets WHERE
1801 ( requestor = ? ) AND ( status != ? )
1802 AND ( worker = ? OR worker = ? OR worker = ? )";
1803 @bind = ('inna', 'completed', 'nwiger', 'rcwe', 'sfz');
1805 Which you could then use in DBI code like so:
1807 my $sth = $dbh->prepare($stmt);
1808 $sth->execute(@bind);
1814 The methods are simple. There's one for every major SQL operation,
1815 and a constructor you use first. The arguments are specified in a
1816 similar order for each method (table, then fields, then a where
1817 clause) to try and simplify things.
1819 =head2 new(option => 'value')
1821 The C<new()> function takes a list of options and values, and returns
1822 a new B<SQL::Abstract> object which can then be used to generate SQL
1823 through the methods below. The options accepted are:
1829 If set to 'lower', then SQL will be generated in all lowercase. By
1830 default SQL is generated in "textbook" case meaning something like:
1832 SELECT a_field FROM a_table WHERE some_field LIKE '%someval%'
1834 Any setting other than 'lower' is ignored.
1838 This determines what the default comparison operator is. By default
1839 it is C<=>, meaning that a hash like this:
1841 %where = (name => 'nwiger', email => 'nate@wiger.org');
1843 Will generate SQL like this:
1845 WHERE name = 'nwiger' AND email = 'nate@wiger.org'
1847 However, you may want loose comparisons by default, so if you set
1848 C<cmp> to C<like> you would get SQL such as:
1850 WHERE name like 'nwiger' AND email like 'nate@wiger.org'
1852 You can also override the comparison on an individual basis - see
1853 the huge section on L</"WHERE CLAUSES"> at the bottom.
1855 =item sqltrue, sqlfalse
1857 Expressions for inserting boolean values within SQL statements.
1858 By default these are C<1=1> and C<1=0>. They are used
1859 by the special operators C<-in> and C<-not_in> for generating
1860 correct SQL even when the argument is an empty array (see below).
1864 This determines the default logical operator for multiple WHERE
1865 statements in arrays or hashes. If absent, the default logic is "or"
1866 for arrays, and "and" for hashes. This means that a WHERE
1870 event_date => {'>=', '2/13/99'},
1871 event_date => {'<=', '4/24/03'},
1874 will generate SQL like this:
1876 WHERE event_date >= '2/13/99' OR event_date <= '4/24/03'
1878 This is probably not what you want given this query, though (look
1879 at the dates). To change the "OR" to an "AND", simply specify:
1881 my $sql = SQL::Abstract->new(logic => 'and');
1883 Which will change the above C<WHERE> to:
1885 WHERE event_date >= '2/13/99' AND event_date <= '4/24/03'
1887 The logic can also be changed locally by inserting
1888 a modifier in front of an arrayref:
1890 @where = (-and => [event_date => {'>=', '2/13/99'},
1891 event_date => {'<=', '4/24/03'} ]);
1893 See the L</"WHERE CLAUSES"> section for explanations.
1897 This will automatically convert comparisons using the specified SQL
1898 function for both column and value. This is mostly used with an argument
1899 of C<upper> or C<lower>, so that the SQL will have the effect of
1900 case-insensitive "searches". For example, this:
1902 $sql = SQL::Abstract->new(convert => 'upper');
1903 %where = (keywords => 'MaKe iT CAse inSeNSItive');
1905 Will turn out the following SQL:
1907 WHERE upper(keywords) like upper('MaKe iT CAse inSeNSItive')
1909 The conversion can be C<upper()>, C<lower()>, or any other SQL function
1910 that can be applied symmetrically to fields (actually B<SQL::Abstract> does
1911 not validate this option; it will just pass through what you specify verbatim).
1915 This is a kludge because many databases suck. For example, you can't
1916 just bind values using DBI's C<execute()> for Oracle C<CLOB> or C<BLOB> fields.
1917 Instead, you have to use C<bind_param()>:
1919 $sth->bind_param(1, 'reg data');
1920 $sth->bind_param(2, $lots, {ora_type => ORA_CLOB});
1922 The problem is, B<SQL::Abstract> will normally just return a C<@bind> array,
1923 which loses track of which field each slot refers to. Fear not.
1925 If you specify C<bindtype> in new, you can determine how C<@bind> is returned.
1926 Currently, you can specify either C<normal> (default) or C<columns>. If you
1927 specify C<columns>, you will get an array that looks like this:
1929 my $sql = SQL::Abstract->new(bindtype => 'columns');
1930 my($stmt, @bind) = $sql->insert(...);
1933 [ 'column1', 'value1' ],
1934 [ 'column2', 'value2' ],
1935 [ 'column3', 'value3' ],
1938 You can then iterate through this manually, using DBI's C<bind_param()>.
1940 $sth->prepare($stmt);
1943 my($col, $data) = @$_;
1944 if ($col eq 'details' || $col eq 'comments') {
1945 $sth->bind_param($i, $data, {ora_type => ORA_CLOB});
1946 } elsif ($col eq 'image') {
1947 $sth->bind_param($i, $data, {ora_type => ORA_BLOB});
1949 $sth->bind_param($i, $data);
1953 $sth->execute; # execute without @bind now
1955 Now, why would you still use B<SQL::Abstract> if you have to do this crap?
1956 Basically, the advantage is still that you don't have to care which fields
1957 are or are not included. You could wrap that above C<for> loop in a simple
1958 sub called C<bind_fields()> or something and reuse it repeatedly. You still
1959 get a layer of abstraction over manual SQL specification.
1961 Note that if you set L</bindtype> to C<columns>, the C<\[ $sql, @bind ]>
1962 construct (see L</Literal SQL with placeholders and bind values (subqueries)>)
1963 will expect the bind values in this format.
1967 This is the character that a table or column name will be quoted
1968 with. By default this is an empty string, but you could set it to
1969 the character C<`>, to generate SQL like this:
1971 SELECT `a_field` FROM `a_table` WHERE `some_field` LIKE '%someval%'
1973 Alternatively, you can supply an array ref of two items, the first being the left
1974 hand quote character, and the second the right hand quote character. For
1975 example, you could supply C<['[',']']> for SQL Server 2000 compliant quotes
1976 that generates SQL like this:
1978 SELECT [a_field] FROM [a_table] WHERE [some_field] LIKE '%someval%'
1980 Quoting is useful if you have tables or columns names that are reserved
1981 words in your database's SQL dialect.
1985 This is the character that will be used to escape L</quote_char>s appearing
1986 in an identifier before it has been quoted.
1988 The parameter default in case of a single L</quote_char> character is the quote
1991 When opening-closing-style quoting is used (L</quote_char> is an arrayref)
1992 this parameter defaults to the B<closing (right)> L</quote_char>. Occurrences
1993 of the B<opening (left)> L</quote_char> within the identifier are currently left
1994 untouched. The default for opening-closing-style quotes may change in future
1995 versions, thus you are B<strongly encouraged> to specify the escape character
2000 This is the character that separates a table and column name. It is
2001 necessary to specify this when the C<quote_char> option is selected,
2002 so that tables and column names can be individually quoted like this:
2004 SELECT `table`.`one_field` FROM `table` WHERE `table`.`other_field` = 1
2006 =item injection_guard
2008 A regular expression C<qr/.../> that is applied to any C<-function> and unquoted
2009 column name specified in a query structure. This is a safety mechanism to avoid
2010 injection attacks when mishandling user input e.g.:
2012 my %condition_as_column_value_pairs = get_values_from_user();
2013 $sqla->select( ... , \%condition_as_column_value_pairs );
2015 If the expression matches an exception is thrown. Note that literal SQL
2016 supplied via C<\'...'> or C<\['...']> is B<not> checked in any way.
2018 Defaults to checking for C<;> and the C<GO> keyword (TransactSQL)
2020 =item array_datatypes
2022 When this option is true, arrayrefs in INSERT or UPDATE are
2023 interpreted as array datatypes and are passed directly
2025 When this option is false, arrayrefs are interpreted
2026 as literal SQL, just like refs to arrayrefs
2027 (but this behavior is for backwards compatibility; when writing
2028 new queries, use the "reference to arrayref" syntax
2034 Takes a reference to a list of "special operators"
2035 to extend the syntax understood by L<SQL::Abstract>.
2036 See section L</"SPECIAL OPERATORS"> for details.
2040 Takes a reference to a list of "unary operators"
2041 to extend the syntax understood by L<SQL::Abstract>.
2042 See section L</"UNARY OPERATORS"> for details.
2048 =head2 insert($table, \@values || \%fieldvals, \%options)
2050 This is the simplest function. You simply give it a table name
2051 and either an arrayref of values or hashref of field/value pairs.
2052 It returns an SQL INSERT statement and a list of bind values.
2053 See the sections on L</"Inserting and Updating Arrays"> and
2054 L</"Inserting and Updating SQL"> for information on how to insert
2055 with those data types.
2057 The optional C<\%options> hash reference may contain additional
2058 options to generate the insert SQL. Currently supported options
2065 Takes either a scalar of raw SQL fields, or an array reference of
2066 field names, and adds on an SQL C<RETURNING> statement at the end.
2067 This allows you to return data generated by the insert statement
2068 (such as row IDs) without performing another C<SELECT> statement.
2069 Note, however, this is not part of the SQL standard and may not
2070 be supported by all database engines.
2074 =head2 update($table, \%fieldvals, \%where, \%options)
2076 This takes a table, hashref of field/value pairs, and an optional
2077 hashref L<WHERE clause|/WHERE CLAUSES>. It returns an SQL UPDATE function and a list
2079 See the sections on L</"Inserting and Updating Arrays"> and
2080 L</"Inserting and Updating SQL"> for information on how to insert
2081 with those data types.
2083 The optional C<\%options> hash reference may contain additional
2084 options to generate the update SQL. Currently supported options
2091 See the C<returning> option to
2092 L<insert|/insert($table, \@values || \%fieldvals, \%options)>.
2096 =head2 select($source, $fields, $where, $order)
2098 This returns a SQL SELECT statement and associated list of bind values, as
2099 specified by the arguments:
2105 Specification of the 'FROM' part of the statement.
2106 The argument can be either a plain scalar (interpreted as a table
2107 name, will be quoted), or an arrayref (interpreted as a list
2108 of table names, joined by commas, quoted), or a scalarref
2109 (literal SQL, not quoted).
2113 Specification of the list of fields to retrieve from
2115 The argument can be either an arrayref (interpreted as a list
2116 of field names, will be joined by commas and quoted), or a
2117 plain scalar (literal SQL, not quoted).
2118 Please observe that this API is not as flexible as that of
2119 the first argument C<$source>, for backwards compatibility reasons.
2123 Optional argument to specify the WHERE part of the query.
2124 The argument is most often a hashref, but can also be
2125 an arrayref or plain scalar --
2126 see section L<WHERE clause|/"WHERE CLAUSES"> for details.
2130 Optional argument to specify the ORDER BY part of the query.
2131 The argument can be a scalar, a hashref or an arrayref
2132 -- see section L<ORDER BY clause|/"ORDER BY CLAUSES">
2138 =head2 delete($table, \%where, \%options)
2140 This takes a table name and optional hashref L<WHERE clause|/WHERE CLAUSES>.
2141 It returns an SQL DELETE statement and list of bind values.
2143 The optional C<\%options> hash reference may contain additional
2144 options to generate the delete SQL. Currently supported options
2151 See the C<returning> option to
2152 L<insert|/insert($table, \@values || \%fieldvals, \%options)>.
2156 =head2 where(\%where, $order)
2158 This is used to generate just the WHERE clause. For example,
2159 if you have an arbitrary data structure and know what the
2160 rest of your SQL is going to look like, but want an easy way
2161 to produce a WHERE clause, use this. It returns an SQL WHERE
2162 clause and list of bind values.
2165 =head2 values(\%data)
2167 This just returns the values from the hash C<%data>, in the same
2168 order that would be returned from any of the other above queries.
2169 Using this allows you to markedly speed up your queries if you
2170 are affecting lots of rows. See below under the L</"PERFORMANCE"> section.
2172 =head2 generate($any, 'number', $of, \@data, $struct, \%types)
2174 Warning: This is an experimental method and subject to change.
2176 This returns arbitrarily generated SQL. It's a really basic shortcut.
2177 It will return two different things, depending on return context:
2179 my($stmt, @bind) = $sql->generate('create table', \$table, \@fields);
2180 my $stmt_and_val = $sql->generate('create table', \$table, \@fields);
2182 These would return the following:
2184 # First calling form
2185 $stmt = "CREATE TABLE test (?, ?)";
2186 @bind = (field1, field2);
2188 # Second calling form
2189 $stmt_and_val = "CREATE TABLE test (field1, field2)";
2191 Depending on what you're trying to do, it's up to you to choose the correct
2192 format. In this example, the second form is what you would want.
2196 $sql->generate('alter session', { nls_date_format => 'MM/YY' });
2200 ALTER SESSION SET nls_date_format = 'MM/YY'
2202 You get the idea. Strings get their case twiddled, but everything
2203 else remains verbatim.
2205 =head1 EXPORTABLE FUNCTIONS
2207 =head2 is_plain_value
2209 Determines if the supplied argument is a plain value as understood by this
2214 =item * The value is C<undef>
2216 =item * The value is a non-reference
2218 =item * The value is an object with stringification overloading
2220 =item * The value is of the form C<< { -value => $anything } >>
2224 On failure returns C<undef>, on success returns a B<scalar> reference
2225 to the original supplied argument.
2231 The stringification overloading detection is rather advanced: it takes
2232 into consideration not only the presence of a C<""> overload, but if that
2233 fails also checks for enabled
2234 L<autogenerated versions of C<"">|overload/Magic Autogeneration>, based
2235 on either C<0+> or C<bool>.
2237 Unfortunately testing in the field indicates that this
2238 detection B<< may tickle a latent bug in perl versions before 5.018 >>,
2239 but only when very large numbers of stringifying objects are involved.
2240 At the time of writing ( Sep 2014 ) there is no clear explanation of
2241 the direct cause, nor is there a manageably small test case that reliably
2242 reproduces the problem.
2244 If you encounter any of the following exceptions in B<random places within
2245 your application stack> - this module may be to blame:
2247 Operation "ne": no method found,
2248 left argument in overloaded package <something>,
2249 right argument in overloaded package <something>
2253 Stub found while resolving method "???" overloading """" in package <something>
2255 If you fall victim to the above - please attempt to reduce the problem
2256 to something that could be sent to the L<SQL::Abstract developers
2257 |DBIx::Class/GETTING HELP/SUPPORT>
2258 (either publicly or privately). As a workaround in the meantime you can
2259 set C<$ENV{SQLA_ISVALUE_IGNORE_AUTOGENERATED_STRINGIFICATION}> to a true
2260 value, which will most likely eliminate your problem (at the expense of
2261 not being able to properly detect exotic forms of stringification).
2263 This notice and environment variable will be removed in a future version,
2264 as soon as the underlying problem is found and a reliable workaround is
2269 =head2 is_literal_value
2271 Determines if the supplied argument is a literal value as understood by this
2276 =item * C<\$sql_string>
2278 =item * C<\[ $sql_string, @bind_values ]>
2282 On failure returns C<undef>, on success returns an B<array> reference
2283 containing the unpacked version of the supplied literal SQL and bind values.
2285 =head1 WHERE CLAUSES
2289 This module uses a variation on the idea from L<DBIx::Abstract>. It
2290 is B<NOT>, repeat I<not> 100% compatible. B<The main logic of this
2291 module is that things in arrays are OR'ed, and things in hashes
2294 The easiest way to explain is to show lots of examples. After
2295 each C<%where> hash shown, it is assumed you used:
2297 my($stmt, @bind) = $sql->where(\%where);
2299 However, note that the C<%where> hash can be used directly in any
2300 of the other functions as well, as described above.
2302 =head2 Key-value pairs
2304 So, let's get started. To begin, a simple hash:
2308 status => 'completed'
2311 Is converted to SQL C<key = val> statements:
2313 $stmt = "WHERE user = ? AND status = ?";
2314 @bind = ('nwiger', 'completed');
2316 One common thing I end up doing is having a list of values that
2317 a field can be in. To do this, simply specify a list inside of
2322 status => ['assigned', 'in-progress', 'pending'];
2325 This simple code will create the following:
2327 $stmt = "WHERE user = ? AND ( status = ? OR status = ? OR status = ? )";
2328 @bind = ('nwiger', 'assigned', 'in-progress', 'pending');
2330 A field associated to an empty arrayref will be considered a
2331 logical false and will generate 0=1.
2333 =head2 Tests for NULL values
2335 If the value part is C<undef> then this is converted to SQL <IS NULL>
2344 $stmt = "WHERE user = ? AND status IS NULL";
2347 To test if a column IS NOT NULL:
2351 status => { '!=', undef },
2354 =head2 Specific comparison operators
2356 If you want to specify a different type of operator for your comparison,
2357 you can use a hashref for a given column:
2361 status => { '!=', 'completed' }
2364 Which would generate:
2366 $stmt = "WHERE user = ? AND status != ?";
2367 @bind = ('nwiger', 'completed');
2369 To test against multiple values, just enclose the values in an arrayref:
2371 status => { '=', ['assigned', 'in-progress', 'pending'] };
2373 Which would give you:
2375 "WHERE status = ? OR status = ? OR status = ?"
2378 The hashref can also contain multiple pairs, in which case it is expanded
2379 into an C<AND> of its elements:
2383 status => { '!=', 'completed', -not_like => 'pending%' }
2386 # Or more dynamically, like from a form
2387 $where{user} = 'nwiger';
2388 $where{status}{'!='} = 'completed';
2389 $where{status}{'-not_like'} = 'pending%';
2391 # Both generate this
2392 $stmt = "WHERE user = ? AND status != ? AND status NOT LIKE ?";
2393 @bind = ('nwiger', 'completed', 'pending%');
2396 To get an OR instead, you can combine it with the arrayref idea:
2400 priority => [ { '=', 2 }, { '>', 5 } ]
2403 Which would generate:
2405 $stmt = "WHERE ( priority = ? OR priority > ? ) AND user = ?";
2406 @bind = ('2', '5', 'nwiger');
2408 If you want to include literal SQL (with or without bind values), just use a
2409 scalar reference or reference to an arrayref as the value:
2412 date_entered => { '>' => \["to_date(?, 'MM/DD/YYYY')", "11/26/2008"] },
2413 date_expires => { '<' => \"now()" }
2416 Which would generate:
2418 $stmt = "WHERE date_entered > to_date(?, 'MM/DD/YYYY') AND date_expires < now()";
2419 @bind = ('11/26/2008');
2422 =head2 Logic and nesting operators
2424 In the example above,
2425 there is a subtle trap if you want to say something like
2426 this (notice the C<AND>):
2428 WHERE priority != ? AND priority != ?
2430 Because, in Perl you I<can't> do this:
2432 priority => { '!=' => 2, '!=' => 1 }
2434 As the second C<!=> key will obliterate the first. The solution
2435 is to use the special C<-modifier> form inside an arrayref:
2437 priority => [ -and => {'!=', 2},
2441 Normally, these would be joined by C<OR>, but the modifier tells it
2442 to use C<AND> instead. (Hint: You can use this in conjunction with the
2443 C<logic> option to C<new()> in order to change the way your queries
2444 work by default.) B<Important:> Note that the C<-modifier> goes
2445 B<INSIDE> the arrayref, as an extra first element. This will
2446 B<NOT> do what you think it might:
2448 priority => -and => [{'!=', 2}, {'!=', 1}] # WRONG!
2450 Here is a quick list of equivalencies, since there is some overlap:
2453 status => {'!=', 'completed', 'not like', 'pending%' }
2454 status => [ -and => {'!=', 'completed'}, {'not like', 'pending%'}]
2457 status => {'=', ['assigned', 'in-progress']}
2458 status => [ -or => {'=', 'assigned'}, {'=', 'in-progress'}]
2459 status => [ {'=', 'assigned'}, {'=', 'in-progress'} ]
2463 =head2 Special operators: IN, BETWEEN, etc.
2465 You can also use the hashref format to compare a list of fields using the
2466 C<IN> comparison operator, by specifying the list as an arrayref:
2469 status => 'completed',
2470 reportid => { -in => [567, 2335, 2] }
2473 Which would generate:
2475 $stmt = "WHERE status = ? AND reportid IN (?,?,?)";
2476 @bind = ('completed', '567', '2335', '2');
2478 The reverse operator C<-not_in> generates SQL C<NOT IN> and is used in
2481 If the argument to C<-in> is an empty array, 'sqlfalse' is generated
2482 (by default: C<1=0>). Similarly, C<< -not_in => [] >> generates
2483 'sqltrue' (by default: C<1=1>).
2485 In addition to the array you can supply a chunk of literal sql or
2486 literal sql with bind:
2489 customer => { -in => \[
2490 'SELECT cust_id FROM cust WHERE balance > ?',
2493 status => { -in => \'SELECT status_codes FROM states' },
2499 customer IN ( SELECT cust_id FROM cust WHERE balance > ? )
2500 AND status IN ( SELECT status_codes FROM states )
2504 Finally, if the argument to C<-in> is not a reference, it will be
2505 treated as a single-element array.
2507 Another pair of operators is C<-between> and C<-not_between>,
2508 used with an arrayref of two values:
2512 completion_date => {
2513 -not_between => ['2002-10-01', '2003-02-06']
2519 WHERE user = ? AND completion_date NOT BETWEEN ( ? AND ? )
2521 Just like with C<-in> all plausible combinations of literal SQL
2525 start0 => { -between => [ 1, 2 ] },
2526 start1 => { -between => \["? AND ?", 1, 2] },
2527 start2 => { -between => \"lower(x) AND upper(y)" },
2528 start3 => { -between => [
2530 \["upper(?)", 'stuff' ],
2537 ( start0 BETWEEN ? AND ? )
2538 AND ( start1 BETWEEN ? AND ? )
2539 AND ( start2 BETWEEN lower(x) AND upper(y) )
2540 AND ( start3 BETWEEN lower(x) AND upper(?) )
2542 @bind = (1, 2, 1, 2, 'stuff');
2545 These are the two builtin "special operators"; but the
2546 list can be expanded: see section L</"SPECIAL OPERATORS"> below.
2548 =head2 Unary operators: bool
2550 If you wish to test against boolean columns or functions within your
2551 database you can use the C<-bool> and C<-not_bool> operators. For
2552 example to test the column C<is_user> being true and the column
2553 C<is_enabled> being false you would use:-
2557 -not_bool => 'is_enabled',
2562 WHERE is_user AND NOT is_enabled
2564 If a more complex combination is required, testing more conditions,
2565 then you should use the and/or operators:-
2570 -not_bool => { two=> { -rlike => 'bar' } },
2571 -not_bool => { three => [ { '=', 2 }, { '>', 5 } ] },
2582 (NOT ( three = ? OR three > ? ))
2585 =head2 Nested conditions, -and/-or prefixes
2587 So far, we've seen how multiple conditions are joined with a top-level
2588 C<AND>. We can change this by putting the different conditions we want in
2589 hashes and then putting those hashes in an array. For example:
2594 status => { -like => ['pending%', 'dispatched'] },
2598 status => 'unassigned',
2602 This data structure would create the following:
2604 $stmt = "WHERE ( user = ? AND ( status LIKE ? OR status LIKE ? ) )
2605 OR ( user = ? AND status = ? ) )";
2606 @bind = ('nwiger', 'pending', 'dispatched', 'robot', 'unassigned');
2609 Clauses in hashrefs or arrayrefs can be prefixed with an C<-and> or C<-or>
2610 to change the logic inside:
2616 -and => [ workhrs => {'>', 20}, geo => 'ASIA' ],
2617 -or => { workhrs => {'<', 50}, geo => 'EURO' },
2624 $stmt = "WHERE ( user = ?
2625 AND ( ( workhrs > ? AND geo = ? )
2626 OR ( workhrs < ? OR geo = ? ) ) )";
2627 @bind = ('nwiger', '20', 'ASIA', '50', 'EURO');
2629 =head3 Algebraic inconsistency, for historical reasons
2631 C<Important note>: when connecting several conditions, the C<-and->|C<-or>
2632 operator goes C<outside> of the nested structure; whereas when connecting
2633 several constraints on one column, the C<-and> operator goes
2634 C<inside> the arrayref. Here is an example combining both features:
2637 -and => [a => 1, b => 2],
2638 -or => [c => 3, d => 4],
2639 e => [-and => {-like => 'foo%'}, {-like => '%bar'} ]
2644 WHERE ( ( ( a = ? AND b = ? )
2645 OR ( c = ? OR d = ? )
2646 OR ( e LIKE ? AND e LIKE ? ) ) )
2648 This difference in syntax is unfortunate but must be preserved for
2649 historical reasons. So be careful: the two examples below would
2650 seem algebraically equivalent, but they are not
2653 { -like => 'foo%' },
2654 { -like => '%bar' },
2656 # yields: WHERE ( ( col LIKE ? AND col LIKE ? ) )
2659 { col => { -like => 'foo%' } },
2660 { col => { -like => '%bar' } },
2662 # yields: WHERE ( ( col LIKE ? OR col LIKE ? ) )
2665 =head2 Literal SQL and value type operators
2667 The basic premise of SQL::Abstract is that in WHERE specifications the "left
2668 side" is a column name and the "right side" is a value (normally rendered as
2669 a placeholder). This holds true for both hashrefs and arrayref pairs as you
2670 see in the L</WHERE CLAUSES> examples above. Sometimes it is necessary to
2671 alter this behavior. There are several ways of doing so.
2675 This is a virtual operator that signals the string to its right side is an
2676 identifier (a column name) and not a value. For example to compare two
2677 columns you would write:
2680 priority => { '<', 2 },
2681 requestor => { -ident => 'submitter' },
2686 $stmt = "WHERE priority < ? AND requestor = submitter";
2689 If you are maintaining legacy code you may see a different construct as
2690 described in L</Deprecated usage of Literal SQL>, please use C<-ident> in new
2695 This is a virtual operator that signals that the construct to its right side
2696 is a value to be passed to DBI. This is for example necessary when you want
2697 to write a where clause against an array (for RDBMS that support such
2698 datatypes). For example:
2701 array => { -value => [1, 2, 3] }
2706 $stmt = 'WHERE array = ?';
2707 @bind = ([1, 2, 3]);
2709 Note that if you were to simply say:
2715 the result would probably not be what you wanted:
2717 $stmt = 'WHERE array = ? OR array = ? OR array = ?';
2722 Finally, sometimes only literal SQL will do. To include a random snippet
2723 of SQL verbatim, you specify it as a scalar reference. Consider this only
2724 as a last resort. Usually there is a better way. For example:
2727 priority => { '<', 2 },
2728 requestor => { -in => \'(SELECT name FROM hitmen)' },
2733 $stmt = "WHERE priority < ? AND requestor IN (SELECT name FROM hitmen)"
2736 Note that in this example, you only get one bind parameter back, since
2737 the verbatim SQL is passed as part of the statement.
2741 Never use untrusted input as a literal SQL argument - this is a massive
2742 security risk (there is no way to check literal snippets for SQL
2743 injections and other nastyness). If you need to deal with untrusted input
2744 use literal SQL with placeholders as described next.
2746 =head3 Literal SQL with placeholders and bind values (subqueries)
2748 If the literal SQL to be inserted has placeholders and bind values,
2749 use a reference to an arrayref (yes this is a double reference --
2750 not so common, but perfectly legal Perl). For example, to find a date
2751 in Postgres you can use something like this:
2754 date_column => \[ "= date '2008-09-30' - ?::integer", 10 ]
2759 $stmt = "WHERE ( date_column = date '2008-09-30' - ?::integer )"
2762 Note that you must pass the bind values in the same format as they are returned
2763 by L<where|/where(\%where, $order)>. This means that if you set L</bindtype>
2764 to C<columns>, you must provide the bind values in the
2765 C<< [ column_meta => value ] >> format, where C<column_meta> is an opaque
2766 scalar value; most commonly the column name, but you can use any scalar value
2767 (including references and blessed references), L<SQL::Abstract> will simply
2768 pass it through intact. So if C<bindtype> is set to C<columns> the above
2769 example will look like:
2772 date_column => \[ "= date '2008-09-30' - ?::integer", [ {} => 10 ] ]
2775 Literal SQL is especially useful for nesting parenthesized clauses in the
2776 main SQL query. Here is a first example:
2778 my ($sub_stmt, @sub_bind) = ("SELECT c1 FROM t1 WHERE c2 < ? AND c3 LIKE ?",
2782 bar => \["IN ($sub_stmt)" => @sub_bind],
2787 $stmt = "WHERE (foo = ? AND bar IN (SELECT c1 FROM t1
2788 WHERE c2 < ? AND c3 LIKE ?))";
2789 @bind = (1234, 100, "foo%");
2791 Other subquery operators, like for example C<"E<gt> ALL"> or C<"NOT IN">,
2792 are expressed in the same way. Of course the C<$sub_stmt> and
2793 its associated bind values can be generated through a former call
2796 my ($sub_stmt, @sub_bind)
2797 = $sql->select("t1", "c1", {c2 => {"<" => 100},
2798 c3 => {-like => "foo%"}});
2801 bar => \["> ALL ($sub_stmt)" => @sub_bind],
2804 In the examples above, the subquery was used as an operator on a column;
2805 but the same principle also applies for a clause within the main C<%where>
2806 hash, like an EXISTS subquery:
2808 my ($sub_stmt, @sub_bind)
2809 = $sql->select("t1", "*", {c1 => 1, c2 => \"> t0.c0"});
2810 my %where = ( -and => [
2812 \["EXISTS ($sub_stmt)" => @sub_bind],
2817 $stmt = "WHERE (foo = ? AND EXISTS (SELECT * FROM t1
2818 WHERE c1 = ? AND c2 > t0.c0))";
2822 Observe that the condition on C<c2> in the subquery refers to
2823 column C<t0.c0> of the main query: this is I<not> a bind
2824 value, so we have to express it through a scalar ref.
2825 Writing C<< c2 => {">" => "t0.c0"} >> would have generated
2826 C<< c2 > ? >> with bind value C<"t0.c0"> ... not exactly
2827 what we wanted here.
2829 Finally, here is an example where a subquery is used
2830 for expressing unary negation:
2832 my ($sub_stmt, @sub_bind)
2833 = $sql->where({age => [{"<" => 10}, {">" => 20}]});
2834 $sub_stmt =~ s/^ where //i; # don't want "WHERE" in the subclause
2836 lname => {like => '%son%'},
2837 \["NOT ($sub_stmt)" => @sub_bind],
2842 $stmt = "lname LIKE ? AND NOT ( age < ? OR age > ? )"
2843 @bind = ('%son%', 10, 20)
2845 =head3 Deprecated usage of Literal SQL
2847 Below are some examples of archaic use of literal SQL. It is shown only as
2848 reference for those who deal with legacy code. Each example has a much
2849 better, cleaner and safer alternative that users should opt for in new code.
2855 my %where = ( requestor => \'IS NOT NULL' )
2857 $stmt = "WHERE requestor IS NOT NULL"
2859 This used to be the way of generating NULL comparisons, before the handling
2860 of C<undef> got formalized. For new code please use the superior syntax as
2861 described in L</Tests for NULL values>.
2865 my %where = ( requestor => \'= submitter' )
2867 $stmt = "WHERE requestor = submitter"
2869 This used to be the only way to compare columns. Use the superior L</-ident>
2870 method for all new code. For example an identifier declared in such a way
2871 will be properly quoted if L</quote_char> is properly set, while the legacy
2872 form will remain as supplied.
2876 my %where = ( is_ready => \"", completed => { '>', '2012-12-21' } )
2878 $stmt = "WHERE completed > ? AND is_ready"
2879 @bind = ('2012-12-21')
2881 Using an empty string literal used to be the only way to express a boolean.
2882 For all new code please use the much more readable
2883 L<-bool|/Unary operators: bool> operator.
2889 These pages could go on for a while, since the nesting of the data
2890 structures this module can handle are pretty much unlimited (the
2891 module implements the C<WHERE> expansion as a recursive function
2892 internally). Your best bet is to "play around" with the module a
2893 little to see how the data structures behave, and choose the best
2894 format for your data based on that.
2896 And of course, all the values above will probably be replaced with
2897 variables gotten from forms or the command line. After all, if you
2898 knew everything ahead of time, you wouldn't have to worry about
2899 dynamically-generating SQL and could just hardwire it into your
2902 =head1 ORDER BY CLAUSES
2904 Some functions take an order by clause. This can either be a scalar (just a
2905 column name), a hashref of C<< { -desc => 'col' } >> or C<< { -asc => 'col' }
2906 >>, a scalarref, an arrayref-ref, or an arrayref of any of the previous
2909 Given | Will Generate
2910 ---------------------------------------------------------------
2912 'colA' | ORDER BY colA
2914 [qw/colA colB/] | ORDER BY colA, colB
2916 {-asc => 'colA'} | ORDER BY colA ASC
2918 {-desc => 'colB'} | ORDER BY colB DESC
2920 ['colA', {-asc => 'colB'}] | ORDER BY colA, colB ASC
2922 { -asc => [qw/colA colB/] } | ORDER BY colA ASC, colB ASC
2924 \'colA DESC' | ORDER BY colA DESC
2926 \[ 'FUNC(colA, ?)', $x ] | ORDER BY FUNC(colA, ?)
2927 | /* ...with $x bound to ? */
2930 { -asc => 'colA' }, | colA ASC,
2931 { -desc => [qw/colB/] }, | colB DESC,
2932 { -asc => [qw/colC colD/] },| colC ASC, colD ASC,
2933 \'colE DESC', | colE DESC,
2934 \[ 'FUNC(colF, ?)', $x ], | FUNC(colF, ?)
2935 ] | /* ...with $x bound to ? */
2936 ===============================================================
2940 =head1 SPECIAL OPERATORS
2942 my $sqlmaker = SQL::Abstract->new(special_ops => [
2946 my ($self, $field, $op, $arg) = @_;
2952 handler => 'method_name',
2956 A "special operator" is a SQL syntactic clause that can be
2957 applied to a field, instead of a usual binary operator.
2960 WHERE field IN (?, ?, ?)
2961 WHERE field BETWEEN ? AND ?
2962 WHERE MATCH(field) AGAINST (?, ?)
2964 Special operators IN and BETWEEN are fairly standard and therefore
2965 are builtin within C<SQL::Abstract> (as the overridable methods
2966 C<_where_field_IN> and C<_where_field_BETWEEN>). For other operators,
2967 like the MATCH .. AGAINST example above which is specific to MySQL,
2968 you can write your own operator handlers - supply a C<special_ops>
2969 argument to the C<new> method. That argument takes an arrayref of
2970 operator definitions; each operator definition is a hashref with two
2977 the regular expression to match the operator
2981 Either a coderef or a plain scalar method name. In both cases
2982 the expected return is C<< ($sql, @bind) >>.
2984 When supplied with a method name, it is simply called on the
2985 L<SQL::Abstract> object as:
2987 $self->$method_name($field, $op, $arg)
2991 $field is the LHS of the operator
2992 $op is the part that matched the handler regex
2995 When supplied with a coderef, it is called as:
2997 $coderef->($self, $field, $op, $arg)
3002 For example, here is an implementation
3003 of the MATCH .. AGAINST syntax for MySQL
3005 my $sqlmaker = SQL::Abstract->new(special_ops => [
3007 # special op for MySql MATCH (field) AGAINST(word1, word2, ...)
3008 {regex => qr/^match$/i,
3010 my ($self, $field, $op, $arg) = @_;
3011 $arg = [$arg] if not ref $arg;
3012 my $label = $self->_quote($field);
3013 my ($placeholder) = $self->_convert('?');
3014 my $placeholders = join ", ", (($placeholder) x @$arg);
3015 my $sql = $self->_sqlcase('match') . " ($label) "
3016 . $self->_sqlcase('against') . " ($placeholders) ";
3017 my @bind = $self->_bindtype($field, @$arg);
3018 return ($sql, @bind);
3025 =head1 UNARY OPERATORS
3027 my $sqlmaker = SQL::Abstract->new(unary_ops => [
3031 my ($self, $op, $arg) = @_;
3037 handler => 'method_name',
3041 A "unary operator" is a SQL syntactic clause that can be
3042 applied to a field - the operator goes before the field
3044 You can write your own operator handlers - supply a C<unary_ops>
3045 argument to the C<new> method. That argument takes an arrayref of
3046 operator definitions; each operator definition is a hashref with two
3053 the regular expression to match the operator
3057 Either a coderef or a plain scalar method name. In both cases
3058 the expected return is C<< $sql >>.
3060 When supplied with a method name, it is simply called on the
3061 L<SQL::Abstract> object as:
3063 $self->$method_name($op, $arg)
3067 $op is the part that matched the handler regex
3068 $arg is the RHS or argument of the operator
3070 When supplied with a coderef, it is called as:
3072 $coderef->($self, $op, $arg)
3080 Thanks to some benchmarking by Mark Stosberg, it turns out that
3081 this module is many orders of magnitude faster than using C<DBIx::Abstract>.
3082 I must admit this wasn't an intentional design issue, but it's a
3083 byproduct of the fact that you get to control your C<DBI> handles
3086 To maximize performance, use a code snippet like the following:
3088 # prepare a statement handle using the first row
3089 # and then reuse it for the rest of the rows
3091 for my $href (@array_of_hashrefs) {
3092 $stmt ||= $sql->insert('table', $href);
3093 $sth ||= $dbh->prepare($stmt);
3094 $sth->execute($sql->values($href));
3097 The reason this works is because the keys in your C<$href> are sorted
3098 internally by B<SQL::Abstract>. Thus, as long as your data retains
3099 the same structure, you only have to generate the SQL the first time
3100 around. On subsequent queries, simply use the C<values> function provided
3101 by this module to return your values in the correct order.
3103 However this depends on the values having the same type - if, for
3104 example, the values of a where clause may either have values
3105 (resulting in sql of the form C<column = ?> with a single bind
3106 value), or alternatively the values might be C<undef> (resulting in
3107 sql of the form C<column IS NULL> with no bind value) then the
3108 caching technique suggested will not work.
3112 If you use my C<CGI::FormBuilder> module at all, you'll hopefully
3113 really like this part (I do, at least). Building up a complex query
3114 can be as simple as the following:
3121 use CGI::FormBuilder;
3124 my $form = CGI::FormBuilder->new(...);
3125 my $sql = SQL::Abstract->new;
3127 if ($form->submitted) {
3128 my $field = $form->field;
3129 my $id = delete $field->{id};
3130 my($stmt, @bind) = $sql->update('table', $field, {id => $id});
3133 Of course, you would still have to connect using C<DBI> to run the
3134 query, but the point is that if you make your form look like your
3135 table, the actual query script can be extremely simplistic.
3137 If you're B<REALLY> lazy (I am), check out C<HTML::QuickTable> for
3138 a fast interface to returning and formatting data. I frequently
3139 use these three modules together to write complex database query
3140 apps in under 50 lines.
3142 =head1 HOW TO CONTRIBUTE
3144 Contributions are always welcome, in all usable forms (we especially
3145 welcome documentation improvements). The delivery methods include git-
3146 or unified-diff formatted patches, GitHub pull requests, or plain bug
3147 reports either via RT or the Mailing list. Contributors are generally
3148 granted full access to the official repository after their first several
3149 patches pass successful review.
3151 This project is maintained in a git repository. The code and related tools are
3152 accessible at the following locations:
3156 =item * Official repo: L<git://git.shadowcat.co.uk/dbsrgits/SQL-Abstract.git>
3158 =item * Official gitweb: L<http://git.shadowcat.co.uk/gitweb/gitweb.cgi?p=dbsrgits/SQL-Abstract.git>
3160 =item * GitHub mirror: L<https://github.com/dbsrgits/sql-abstract>
3162 =item * Authorized committers: L<ssh://dbsrgits@git.shadowcat.co.uk/SQL-Abstract.git>
3168 Version 1.50 was a major internal refactoring of C<SQL::Abstract>.
3169 Great care has been taken to preserve the I<published> behavior
3170 documented in previous versions in the 1.* family; however,
3171 some features that were previously undocumented, or behaved
3172 differently from the documentation, had to be changed in order
3173 to clarify the semantics. Hence, client code that was relying
3174 on some dark areas of C<SQL::Abstract> v1.*
3175 B<might behave differently> in v1.50.
3177 The main changes are:
3183 support for literal SQL through the C<< \ [ $sql, @bind ] >> syntax.
3187 support for the { operator => \"..." } construct (to embed literal SQL)
3191 support for the { operator => \["...", @bind] } construct (to embed literal SQL with bind values)
3195 optional support for L<array datatypes|/"Inserting and Updating Arrays">
3199 defensive programming: check arguments
3203 fixed bug with global logic, which was previously implemented
3204 through global variables yielding side-effects. Prior versions would
3205 interpret C<< [ {cond1, cond2}, [cond3, cond4] ] >>
3206 as C<< "(cond1 AND cond2) OR (cond3 AND cond4)" >>.
3207 Now this is interpreted
3208 as C<< "(cond1 AND cond2) OR (cond3 OR cond4)" >>.
3213 fixed semantics of _bindtype on array args
3217 dropped the C<_anoncopy> of the %where tree. No longer necessary,
3218 we just avoid shifting arrays within that tree.
3222 dropped the C<_modlogic> function
3226 =head1 ACKNOWLEDGEMENTS
3228 There are a number of individuals that have really helped out with
3229 this module. Unfortunately, most of them submitted bugs via CPAN
3230 so I have no idea who they are! But the people I do know are:
3232 Ash Berlin (order_by hash term support)
3233 Matt Trout (DBIx::Class support)
3234 Mark Stosberg (benchmarking)
3235 Chas Owens (initial "IN" operator support)
3236 Philip Collins (per-field SQL functions)
3237 Eric Kolve (hashref "AND" support)
3238 Mike Fragassi (enhancements to "BETWEEN" and "LIKE")
3239 Dan Kubb (support for "quote_char" and "name_sep")
3240 Guillermo Roditi (patch to cleanup "IN" and "BETWEEN", fix and tests for _order_by)
3241 Laurent Dami (internal refactoring, extensible list of special operators, literal SQL)
3242 Norbert Buchmuller (support for literal SQL in hashpair, misc. fixes & tests)
3243 Peter Rabbitson (rewrite of SQLA::Test, misc. fixes & tests)
3244 Oliver Charles (support for "RETURNING" after "INSERT")
3250 L<DBIx::Class>, L<DBIx::Abstract>, L<CGI::FormBuilder>, L<HTML::QuickTable>.
3254 Copyright (c) 2001-2007 Nathan Wiger <nwiger@cpan.org>. All Rights Reserved.
3256 This module is actively maintained by Matt Trout <mst@shadowcatsystems.co.uk>
3258 For support, your best bet is to try the C<DBIx::Class> users mailing list.
3259 While not an official support venue, C<DBIx::Class> makes heavy use of
3260 C<SQL::Abstract>, and as such list members there are very familiar with
3261 how to create queries.
3265 This module is free software; you may copy this under the same
3266 terms as perl itself (either the GNU General Public License or
3267 the Artistic License)