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/^ (?: not \s )? in $/ix, handler => sub { die "NOPE" }},
42 {regex => qr/^ is (?: \s+ not )? $/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' ) ? [ @${ $_[0] } ]
71 # FIXME XSify - this can be done so much more efficiently
72 sub is_plain_value ($) {
74 ! length ref $_[0] ? \($_[0])
76 ref $_[0] eq 'HASH' and keys %{$_[0]} == 1
78 exists $_[0]->{-value}
79 ) ? \($_[0]->{-value})
81 # reuse @_ for even moar speedz
82 defined ( $_[1] = Scalar::Util::blessed $_[0] )
84 # deliberately not using Devel::OverloadInfo - the checks we are
85 # intersted in are much more limited than the fullblown thing, and
86 # this is a very hot piece of code
88 # simply using ->can('(""') can leave behind stub methods that
89 # break actually using the overload later (see L<perldiag/Stub
90 # found while resolving method "%s" overloading "%s" in package
91 # "%s"> and the source of overload::mycan())
93 # either has stringification which DBI SHOULD prefer out of the box
94 grep { *{ (qq[${_}::(""]) }{CODE} } @{ $_[2] = mro::get_linear_isa( $_[1] ) }
96 # has nummification or boolification, AND fallback is *not* disabled
98 SQL::Abstract::_ENV_::DETECT_AUTOGENERATED_STRINGIFICATION
101 grep { *{"${_}::(0+"}{CODE} } @{$_[2]}
103 grep { *{"${_}::(bool"}{CODE} } @{$_[2]}
107 # no fallback specified at all
108 ! ( ($_[3]) = grep { *{"${_}::()"}{CODE} } @{$_[2]} )
110 # fallback explicitly undef
111 ! defined ${"$_[3]::()"}
124 #======================================================================
126 #======================================================================
130 my $class = ref($self) || $self;
131 my %opt = (ref $_[0] eq 'HASH') ? %{$_[0]} : @_;
133 # choose our case by keeping an option around
134 delete $opt{case} if $opt{case} && $opt{case} ne 'lower';
136 # default logic for interpreting arrayrefs
137 $opt{logic} = $opt{logic} ? uc $opt{logic} : 'OR';
139 # how to return bind vars
140 $opt{bindtype} ||= 'normal';
142 # default comparison is "=", but can be overridden
145 # try to recognize which are the 'equality' and 'inequality' ops
146 # (temporary quickfix (in 2007), should go through a more seasoned API)
147 $opt{equality_op} = qr/^( \Q$opt{cmp}\E | \= )$/ix;
148 $opt{inequality_op} = qr/^( != | <> )$/ix;
150 $opt{like_op} = qr/^ (is\s+)? r?like $/xi;
151 $opt{not_like_op} = qr/^ (is\s+)? not \s+ r?like $/xi;
154 $opt{sqltrue} ||= '1=1';
155 $opt{sqlfalse} ||= '0=1';
158 $opt{special_ops} ||= [];
160 # regexes are applied in order, thus push after user-defines
161 push @{$opt{special_ops}}, @BUILTIN_SPECIAL_OPS;
163 if ($class->isa('DBIx::Class::SQLMaker')) {
164 push @{$opt{special_ops}}, our $DBIC_Compat_Op ||= {
165 regex => qr/^(?:ident|value)$/i, handler => sub { die "NOPE" }
167 $opt{is_dbic_sqlmaker} = 1;
171 $opt{unary_ops} ||= [];
173 # rudimentary sanity-check for user supplied bits treated as functions/operators
174 # If a purported function matches this regular expression, an exception is thrown.
175 # Literal SQL is *NOT* subject to this check, only functions (and column names
176 # when quoting is not in effect)
179 # need to guard against ()'s in column names too, but this will break tons of
180 # hacks... ideas anyone?
181 $opt{injection_guard} ||= qr/
187 $opt{expand_unary} = {};
190 -not => '_expand_not',
191 -bool => '_expand_bool',
192 -and => '_expand_op_andor',
193 -or => '_expand_op_andor',
194 -nest => '_expand_nest',
198 'between' => '_expand_between',
199 'not between' => '_expand_between',
200 'in' => '_expand_in',
201 'not in' => '_expand_in',
202 'nest' => '_expand_nest',
203 (map +($_ => '_expand_op_andor'),
207 # placeholder for _expand_unop system
209 my %unops = (-ident => '_expand_ident', -value => '_expand_value');
210 foreach my $name (keys %unops) {
211 $opt{expand}{$name} = $unops{$name};
212 my ($op) = $name =~ /^-(.*)$/;
213 $opt{expand_op}{$op} = sub {
214 my ($self, $op, $arg, $k) = @_;
217 $self->_expand_ident(-ident => $k),
218 $self->_expand_expr({ '-'.$op => $arg }),
225 (map +("-$_", "_render_$_"), qw(op func bind ident literal list)),
229 $opt{render_op} = our $RENDER_OP;
231 return bless \%opt, $class;
234 sub sqltrue { +{ -literal => [ $_[0]->{sqltrue} ] } }
235 sub sqlfalse { +{ -literal => [ $_[0]->{sqlfalse} ] } }
237 sub _assert_pass_injection_guard {
238 if ($_[1] =~ $_[0]->{injection_guard}) {
239 my $class = ref $_[0];
240 puke "Possible SQL injection attempt '$_[1]'. If this is indeed a part of the "
241 . "desired SQL use literal SQL ( \'...' or \[ '...' ] ) or supply your own "
242 . "{injection_guard} attribute to ${class}->new()"
247 #======================================================================
249 #======================================================================
253 my $table = $self->_table(shift);
254 my $data = shift || return;
257 my $method = $self->_METHOD_FOR_refkind("_insert", $data);
258 my ($sql, @bind) = $self->$method($data);
259 $sql = join " ", $self->_sqlcase('insert into'), $table, $sql;
261 if ($options->{returning}) {
262 my ($s, @b) = $self->_insert_returning($options);
267 return wantarray ? ($sql, @bind) : $sql;
270 # So that subclasses can override INSERT ... RETURNING separately from
271 # UPDATE and DELETE (e.g. DBIx::Class::SQLMaker::Oracle does this)
272 sub _insert_returning { shift->_returning(@_) }
275 my ($self, $options) = @_;
277 my $f = $options->{returning};
279 my ($sql, @bind) = $self->render_aqt(
280 $self->_expand_maybe_list_expr($f, undef, -ident)
283 ? $self->_sqlcase(' returning ') . $sql
284 : ($self->_sqlcase(' returning ').$sql, @bind);
287 sub _insert_HASHREF { # explicit list of fields and then values
288 my ($self, $data) = @_;
290 my @fields = sort keys %$data;
292 my ($sql, @bind) = $self->_insert_values($data);
295 $_ = $self->_quote($_) foreach @fields;
296 $sql = "( ".join(", ", @fields).") ".$sql;
298 return ($sql, @bind);
301 sub _insert_ARRAYREF { # just generate values(?,?) part (no list of fields)
302 my ($self, $data) = @_;
304 # no names (arrayref) so can't generate bindtype
305 $self->{bindtype} ne 'columns'
306 or belch "can't do 'columns' bindtype when called with arrayref";
308 my (@values, @all_bind);
309 foreach my $value (@$data) {
310 my ($values, @bind) = $self->_insert_value(undef, $value);
311 push @values, $values;
312 push @all_bind, @bind;
314 my $sql = $self->_sqlcase('values')." ( ".join(", ", @values)." )";
315 return ($sql, @all_bind);
318 sub _insert_ARRAYREFREF { # literal SQL with bind
319 my ($self, $data) = @_;
321 my ($sql, @bind) = @${$data};
322 $self->_assert_bindval_matches_bindtype(@bind);
324 return ($sql, @bind);
328 sub _insert_SCALARREF { # literal SQL without bind
329 my ($self, $data) = @_;
335 my ($self, $data) = @_;
337 my (@values, @all_bind);
338 foreach my $column (sort keys %$data) {
339 my ($values, @bind) = $self->_insert_value($column, $data->{$column});
340 push @values, $values;
341 push @all_bind, @bind;
343 my $sql = $self->_sqlcase('values')." ( ".join(", ", @values)." )";
344 return ($sql, @all_bind);
348 my ($self, $column, $v) = @_;
350 return $self->render_aqt(
351 $self->_expand_insert_value($column, $v)
355 sub _expand_insert_value {
356 my ($self, $column, $v) = @_;
358 if (ref($v) eq 'ARRAY') {
359 if ($self->{array_datatypes}) {
360 return +{ -bind => [ $column, $v ] };
362 my ($sql, @bind) = @$v;
363 $self->_assert_bindval_matches_bindtype(@bind);
364 return +{ -literal => $v };
366 if (ref($v) eq 'HASH') {
367 if (grep !/^-/, keys %$v) {
368 belch "HASH ref as bind value in insert is not supported";
369 return +{ -bind => [ $column, $v ] };
373 return +{ -bind => [ $column, undef ] };
375 local our $Cur_Col_Meta = $column;
376 return $self->expand_expr($v);
381 #======================================================================
383 #======================================================================
388 my $table = $self->_table(shift);
389 my $data = shift || return;
393 # first build the 'SET' part of the sql statement
394 puke "Unsupported data type specified to \$sql->update"
395 unless ref $data eq 'HASH';
397 my ($sql, @all_bind) = $self->_update_set_values($data);
398 $sql = $self->_sqlcase('update ') . $table . $self->_sqlcase(' set ')
402 my($where_sql, @where_bind) = $self->where($where);
404 push @all_bind, @where_bind;
407 if ($options->{returning}) {
408 my ($returning_sql, @returning_bind) = $self->_update_returning($options);
409 $sql .= $returning_sql;
410 push @all_bind, @returning_bind;
413 return wantarray ? ($sql, @all_bind) : $sql;
416 sub _update_set_values {
417 my ($self, $data) = @_;
419 return $self->render_aqt(
420 $self->_expand_update_set_values($data),
424 sub _expand_update_set_values {
425 my ($self, $data) = @_;
426 $self->_expand_maybe_list_expr( [
429 $set = { -bind => $_ } unless defined $set;
430 +{ -op => [ '=', $self->_expand_ident(-ident => $k), $set ] };
436 ? ($self->{array_datatypes}
437 ? [ $k, +{ -bind => [ $k, $v ] } ]
438 : [ $k, +{ -literal => $v } ])
440 local our $Cur_Col_Meta = $k;
441 [ $k, $self->_expand_expr($v) ]
448 # So that subclasses can override UPDATE ... RETURNING separately from
450 sub _update_returning { shift->_returning(@_) }
454 #======================================================================
456 #======================================================================
461 my $table = $self->_table(shift);
462 my $fields = shift || '*';
466 my ($fields_sql, @bind) = $self->_select_fields($fields);
468 my ($where_sql, @where_bind) = $self->where($where, $order);
469 push @bind, @where_bind;
471 my $sql = join(' ', $self->_sqlcase('select'), $fields_sql,
472 $self->_sqlcase('from'), $table)
475 return wantarray ? ($sql, @bind) : $sql;
479 my ($self, $fields) = @_;
480 return $fields unless ref($fields);
481 return $self->render_aqt(
482 $self->_expand_maybe_list_expr($fields, undef, '-ident')
486 #======================================================================
488 #======================================================================
493 my $table = $self->_table(shift);
497 my($where_sql, @bind) = $self->where($where);
498 my $sql = $self->_sqlcase('delete from ') . $table . $where_sql;
500 if ($options->{returning}) {
501 my ($returning_sql, @returning_bind) = $self->_delete_returning($options);
502 $sql .= $returning_sql;
503 push @bind, @returning_bind;
506 return wantarray ? ($sql, @bind) : $sql;
509 # So that subclasses can override DELETE ... RETURNING separately from
511 sub _delete_returning { shift->_returning(@_) }
515 #======================================================================
517 #======================================================================
521 # Finally, a separate routine just to handle WHERE clauses
523 my ($self, $where, $order) = @_;
525 local $self->{convert_where} = $self->{convert};
528 my ($sql, @bind) = defined($where)
529 ? $self->_recurse_where($where)
531 $sql = (defined $sql and length $sql) ? $self->_sqlcase(' where ') . "( $sql )" : '';
535 my ($order_sql, @order_bind) = $self->_order_by($order);
537 push @bind, @order_bind;
540 return wantarray ? ($sql, @bind) : $sql;
544 my ($self, $expr, $default_scalar_to) = @_;
545 local our $Default_Scalar_To = $default_scalar_to if $default_scalar_to;
546 $self->_expand_expr($expr);
550 my ($self, $aqt) = @_;
551 my ($k, $v, @rest) = %$aqt;
553 if (my $meth = $self->{render}{$k}) {
554 return $self->$meth($v);
556 die "notreached: $k";
560 my ($self, $expr) = @_;
561 $self->render_aqt($self->expand_expr($expr));
565 my ($self, $expr) = @_;
566 our $Expand_Depth ||= 0; local $Expand_Depth = $Expand_Depth + 1;
567 return undef unless defined($expr);
568 if (ref($expr) eq 'HASH') {
569 return undef unless my $kc = keys %$expr;
571 return $self->_expand_op_andor(-and => $expr);
573 my ($key, $value) = %$expr;
574 if ($key =~ /^-/ and $key =~ s/ [_\s]? \d+ $//x ) {
575 belch 'Use of [and|or|nest]_N modifiers is deprecated and will be removed in SQLA v2.0. '
576 . "You probably wanted ...-and => [ $key => COND1, $key => COND2 ... ]";
578 if (my $exp = $self->{expand}{$key}) {
579 return $self->$exp($key, $value);
581 return $self->_expand_expr_hashpair($key, $value);
583 if (ref($expr) eq 'ARRAY') {
584 my $logic = '-'.lc($self->{logic});
585 return $self->_expand_op_andor($logic, $expr);
587 if (my $literal = is_literal_value($expr)) {
588 return +{ -literal => $literal };
590 if (!ref($expr) or Scalar::Util::blessed($expr)) {
591 if (my $d = our $Default_Scalar_To) {
592 return $self->_expand_expr({ $d => $expr });
594 return $self->_expand_value(-value => $expr);
599 sub _expand_expr_hashpair {
600 my ($self, $k, $v) = @_;
601 unless (defined($k) and length($k)) {
602 if (defined($k) and my $literal = is_literal_value($v)) {
603 belch 'Hash-pairs consisting of an empty string with a literal are deprecated, and will be removed in 2.0: use -and => [ $literal ] instead';
604 return { -literal => $literal };
606 puke "Supplying an empty left hand side argument is not supported";
608 $self->_assert_pass_injection_guard($k =~ /^-(.*)$/s) if $k =~ /^-/;
610 return $self->_expand_expr_hashpair_op($k, $v);
616 and exists $v->{-value}
617 and not defined $v->{-value}
620 return $self->_expand_expr({ $k => { $self->{cmp} => undef } });
622 if (!ref($v) or Scalar::Util::blessed($v)) {
623 my $d = our $Default_Scalar_To;
627 $self->_expand_ident(-ident => $k),
629 ? $self->_expand_expr($d => $v)
630 : { -bind => [ $k, $v ] }
635 if (ref($v) eq 'HASH') {
637 return $self->_expand_op_andor(-and => [
638 map +{ $k => { $_ => $v->{$_} } },
642 return undef unless keys %$v;
644 my $op = join ' ', split '_', (map lc, $vk =~ /^-?(.*)$/)[0];
645 $self->_assert_pass_injection_guard($op);
646 if ($op =~ s/ [_\s]? \d+ $//x ) {
647 belch 'Use of [and|or|nest]_N modifiers is deprecated and will be removed in SQLA v2.0. '
648 . "You probably wanted ...-and => [ -$op => COND1, -$op => COND2 ... ]";
650 if (my $x = $self->{expand_op}{$op}) {
651 local our $Cur_Col_Meta = $k;
652 return $self->$x($op, $vv, $k);
654 if ($op =~ /^is(?: not)?$/) {
655 puke "$op can only take undef as argument"
659 and exists($vv->{-value})
660 and !defined($vv->{-value})
662 return +{ -op => [ $op.' null', $self->_expand_ident(-ident => $k) ] };
664 if (my $us = List::Util::first { $op =~ $_->{regex} } @{$self->{special_ops}}) {
665 return { -op => [ $op, $self->_expand_ident(-ident => $k), $vv ] };
667 if (my $us = List::Util::first { $op =~ $_->{regex} } @{$self->{unary_ops}}) {
670 $self->_expand_ident(-ident => $k),
671 { -op => [ $op, $vv ] }
674 if (ref($vv) eq 'ARRAY') {
675 my ($logic, @values) = (
676 (defined($vv->[0]) and $vv->[0] =~ /^-(and|or)$/i)
681 $op =~ $self->{inequality_op}
682 or $op =~ $self->{not_like_op}
684 if (lc($logic) eq '-or' and @values > 1) {
685 belch "A multi-element arrayref as an argument to the inequality op '${\uc($op)}' "
686 . 'is technically equivalent to an always-true 1=1 (you probably wanted '
687 . "to say ...{ \$inequality_op => [ -and => \@values ] }... instead)"
692 # try to DWIM on equality operators
694 $op =~ $self->{equality_op} ? $self->sqlfalse
695 : $op =~ $self->{like_op} ? belch("Supplying an empty arrayref to '@{[ uc $op]}' is deprecated") && $self->sqlfalse
696 : $op =~ $self->{inequality_op} ? $self->sqltrue
697 : $op =~ $self->{not_like_op} ? belch("Supplying an empty arrayref to '@{[ uc $op]}' is deprecated") && $self->sqltrue
698 : puke "operator '$op' applied on an empty array (field '$k')";
700 return $self->_expand_op_andor($logic => [
701 map +{ $k => { $vk => $_ } },
709 and exists $vv->{-value}
710 and not defined $vv->{-value}
714 $op =~ /^not$/i ? 'is not' # legacy
715 : $op =~ $self->{equality_op} ? 'is'
716 : $op =~ $self->{like_op} ? belch("Supplying an undefined argument to '@{[ uc $op]}' is deprecated") && 'is'
717 : $op =~ $self->{inequality_op} ? 'is not'
718 : $op =~ $self->{not_like_op} ? belch("Supplying an undefined argument to '@{[ uc $op]}' is deprecated") && 'is not'
719 : puke "unexpected operator '$op' with undef operand";
720 return +{ -op => [ $is.' null', $self->_expand_ident(-ident => $k) ] };
722 local our $Cur_Col_Meta = $k;
725 $self->_expand_ident(-ident => $k),
726 $self->_expand_expr($vv)
729 if (ref($v) eq 'ARRAY') {
730 return $self->sqlfalse unless @$v;
731 $self->_debug("ARRAY($k) means distribute over elements");
733 $v->[0] =~ /^-(and|or)$/i
734 ? shift(@{$v = [ @$v ]})
735 : '-'.lc($self->{logic} || 'OR')
737 return $self->_expand_op_andor(
738 $logic => [ map +{ $k => $_ }, @$v ]
741 if (my $literal = is_literal_value($v)) {
743 belch 'Hash-pairs consisting of an empty string with a literal are deprecated, and will be removed in 2.0: use -and => [ $literal ] instead';
746 my ($sql, @bind) = @$literal;
747 if ($self->{bindtype} eq 'columns') {
749 $self->_assert_bindval_matches_bindtype($_);
752 return +{ -literal => [ $self->_quote($k).' '.$sql, @bind ] };
757 sub _expand_expr_hashpair_op {
758 my ($self, $k, $v) = @_;
759 if (my ($rest) = $k =~/^-not[_ ](.*)$/) {
762 $self->_expand_expr({ "-${rest}", $v })
767 $op =~ s/^-// if length($op) > 1;
769 # top level special ops are illegal in general
770 # note that, arguably, if it makes no sense at top level, it also
771 # makes no sense on the other side of an = sign or similar but DBIC
772 # gets disappointingly upset if I disallow it
774 (our $Expand_Depth) == 1
775 and List::Util::first { $op =~ $_->{regex} } @{$self->{special_ops}}
777 puke "Illegal use of top-level '-$op'"
779 if (my $us = List::Util::first { $op =~ $_->{regex} } @{$self->{unary_ops}}) {
780 return { -op => [ $op, $v ] };
783 if ($self->{render}{$k}) {
789 and (keys %$v)[0] =~ /^-/
791 my ($func) = $k =~ /^-(.*)$/;
792 if (List::Util::first { $func =~ $_->{regex} } @{$self->{special_ops}}) {
793 return +{ -op => [ $func, $self->_expand_expr($v) ] };
795 return +{ -func => [ $func, $self->_expand_expr($v) ] };
797 if (!ref($v) or is_literal_value($v)) {
798 return +{ -op => [ $k =~ /^-(.*)$/, $self->_expand_expr($v) ] };
804 my ($self, $op, $body) = @_;
805 unless (defined($body) or (ref($body) and ref($body) eq 'ARRAY')) {
806 puke "$op requires a single plain scalar argument (a quotable identifier) or an arrayref of identifier parts";
808 my @parts = map split(/\Q${\($self->{name_sep}||'.')}\E/, $_),
809 ref($body) ? @$body : $body;
810 return { -ident => $parts[-1] } if $self->{_dequalify_idents};
811 unless ($self->{quote_char}) {
812 $self->_assert_pass_injection_guard($_) for @parts;
814 return +{ -ident => \@parts };
818 +{ -bind => [ our $Cur_Col_Meta, $_[2] ] };
822 +{ -op => [ 'not', $_[0]->_expand_expr($_[2]) ] };
826 my ($self, undef, $v) = @_;
828 return $self->_expand_expr($v);
830 puke "-bool => undef not supported" unless defined($v);
831 return $self->_expand_ident(-ident => $v);
834 sub _expand_op_andor {
835 my ($self, $logic, $v, $k) = @_;
837 $v = [ map +{ $k, { $_ => $v->{$_} } },
840 my ($logop) = $logic =~ /^-?(.*)$/;
841 if (ref($v) eq 'HASH') {
844 map $self->_expand_expr({ $_ => $v->{$_} }),
848 if (ref($v) eq 'ARRAY') {
849 $logop eq 'and' or $logop eq 'or' or puke "unknown logic: $logop";
852 (ref($_) eq 'ARRAY' and @$_)
853 or (ref($_) eq 'HASH' and %$_)
859 while (my ($el) = splice @expr, 0, 1) {
860 puke "Supplying an empty left hand side argument is not supported in array-pairs"
861 unless defined($el) and length($el);
862 my $elref = ref($el);
864 local our $Expand_Depth = 0;
865 push(@res, grep defined, $self->_expand_expr({ $el, shift(@expr) }));
866 } elsif ($elref eq 'ARRAY') {
867 push(@res, grep defined, $self->_expand_expr($el)) if @$el;
868 } elsif (my $l = is_literal_value($el)) {
869 push @res, { -literal => $l };
870 } elsif ($elref eq 'HASH') {
871 local our $Expand_Depth = 0;
872 push @res, grep defined, $self->_expand_expr($el) if %$el;
878 # return $res[0] if @res == 1;
879 return { -op => [ $logop, @res ] };
884 sub _expand_between {
885 my ($self, $op, $vv, $k) = @_;
886 local our $Cur_Col_Meta = $k;
887 my @rhs = map $self->_expand_expr($_),
888 ref($vv) eq 'ARRAY' ? @$vv : $vv;
890 (@rhs == 1 and ref($rhs[0]) eq 'HASH' and $rhs[0]->{-literal})
892 (@rhs == 2 and defined($rhs[0]) and defined($rhs[1]))
894 puke "Operator '${\uc($op)}' requires either an arrayref with two defined values or expressions, or a single literal scalarref/arrayref-ref";
898 $self->_expand_ident(-ident => $k),
904 my ($self, $op, $vv, $k) = @_;
905 if (my $literal = is_literal_value($vv)) {
906 my ($sql, @bind) = @$literal;
907 my $opened_sql = $self->_open_outer_paren($sql);
909 $op, $self->_expand_ident(-ident => $k),
910 [ { -literal => [ $opened_sql, @bind ] } ]
914 'SQL::Abstract before v1.75 used to generate incorrect SQL when the '
915 . "-${\uc($op)} operator was given an undef-containing list: !!!AUDIT YOUR CODE "
916 . 'AND DATA!!! (the upcoming Data::Query-based version of SQL::Abstract '
917 . 'will emit the logically correct SQL instead of raising this exception)'
919 puke("Argument passed to the '${\uc($op)}' operator can not be undefined")
921 my @rhs = map $self->_expand_expr($_),
922 map { ref($_) ? $_ : { -bind => [ $k, $_ ] } }
923 map { defined($_) ? $_: puke($undef_err) }
924 (ref($vv) eq 'ARRAY' ? @$vv : $vv);
925 return $self->${\($op =~ /^not/ ? 'sqltrue' : 'sqlfalse')} unless @rhs;
929 $self->_expand_ident(-ident => $k),
935 my ($self, $op, $v) = @_;
936 # DBIx::Class requires a nest warning to be emitted once but the private
937 # method it overrode to do so no longer exists
938 if ($self->{is_dbic_sqlmaker}) {
939 unless (our $Nest_Warned) {
941 "-nest in search conditions is deprecated, you most probably wanted:\n"
942 .q|{..., -and => [ \%cond0, \@cond1, \'cond2', \[ 'cond3', [ col => bind ] ], etc. ], ... }|
947 return $self->_expand_expr($v);
951 my ($self, $where, $logic) = @_;
953 # Special case: top level simple string treated as literal
955 my $where_exp = (ref($where)
956 ? $self->_expand_expr($where, $logic)
957 : { -literal => [ $where ] });
959 # dispatch expanded expression
961 my ($sql, @bind) = defined($where_exp) ? $self->render_aqt($where_exp) : (undef);
962 # DBIx::Class used to call _recurse_where in scalar context
963 # something else might too...
965 return ($sql, @bind);
968 belch "Calling _recurse_where in scalar context is deprecated and will go away before 2.0";
974 my ($self, $ident) = @_;
976 return $self->_convert($self->_quote($ident));
980 my ($self, $list) = @_;
981 my @parts = grep length($_->[0]), map [ $self->render_aqt($_) ], @$list;
982 return join(', ', map $_->[0], @parts), map @{$_}[1..$#$_], @parts;
986 my ($self, $rest) = @_;
987 my ($func, @args) = @$rest;
991 push @arg_sql, shift @x;
993 } map [ $self->render_aqt($_) ], @args;
994 return ($self->_sqlcase($func).'('.join(', ', @arg_sql).')', @bind);
998 my ($self, $bind) = @_;
999 return ($self->_convert('?'), $self->_bindtype(@$bind));
1002 sub _render_literal {
1003 my ($self, $literal) = @_;
1004 $self->_assert_bindval_matches_bindtype(@{$literal}[1..$#$literal]);
1010 my ($self, $op, $args) = @_;
1011 my ($left, $low, $high) = @$args;
1012 my ($rhsql, @rhbind) = do {
1014 puke "Single arg to between must be a literal"
1015 unless $low->{-literal};
1018 my ($l, $h) = map [ $self->render_aqt($_) ], $low, $high;
1019 (join(' ', $l->[0], $self->_sqlcase('and'), $h->[0]),
1020 @{$l}[1..$#$l], @{$h}[1..$#$h])
1023 my ($lhsql, @lhbind) = $self->render_aqt($left);
1025 join(' ', '(', $lhsql, $self->_sqlcase($op), $rhsql, ')'),
1028 }), 'between', 'not between'),
1030 my ($self, $op, $args) = @_;
1031 my ($lhs, $rhs) = @$args;
1034 my ($sql, @bind) = $self->render_aqt($_);
1035 push @in_bind, @bind;
1038 my ($lhsql, @lbind) = $self->render_aqt($lhs);
1040 $lhsql.' '.$self->_sqlcase($op).' ( '
1041 .join(', ', @in_sql)
1045 }), 'in', 'not in'),
1046 (map +($_ => '_render_unop_postfix'),
1047 'is null', 'is not null', 'asc', 'desc',
1049 (not => '_render_op_not'),
1051 my ($self, $op, $args) = @_;
1052 my @parts = grep length($_->[0]), map [ $self->render_aqt($_) ], @$args;
1053 return '' unless @parts;
1054 return @{$parts[0]} if @parts == 1;
1055 my ($final_sql) = join(
1056 ' '.$self->_sqlcase($op).' ',
1061 map @{$_}[1..$#$_], @parts
1067 my ($self, $v) = @_;
1068 my ($op, @args) = @$v;
1069 if (my $r = $self->{render_op}{$op}) {
1070 return $self->$r($op, \@args);
1072 my $us = List::Util::first { $op =~ $_->{regex} } @{$self->{special_ops}};
1073 if ($us and @args > 1) {
1074 puke "Special op '${op}' requires first value to be identifier"
1075 unless my ($ident) = map $_->{-ident}, grep ref($_) eq 'HASH', $args[0];
1076 my $k = join(($self->{name_sep}||'.'), @$ident);
1077 local our $Expand_Depth = 1;
1078 return $self->${\($us->{handler})}($k, $op, $args[1]);
1080 if (my $us = List::Util::first { $op =~ $_->{regex} } @{$self->{unary_ops}}) {
1081 return $self->${\($us->{handler})}($op, $args[0]);
1084 return $self->_render_unop_prefix($op, \@args);
1086 my @parts = grep length($_->[0]), map [ $self->render_aqt($_) ], @args;
1087 return '' unless @parts;
1088 my ($final_sql) = join(
1089 ' '.$self->_sqlcase($op).' ',
1094 map @{$_}[1..$#$_], @parts
1100 sub _render_op_not {
1101 my ($self, $op, $v) = @_;
1102 my ($sql, @bind) = $self->_render_unop_prefix($op, $v);
1103 return "(${sql})", @bind;
1106 sub _render_unop_prefix {
1107 my ($self, $op, $v) = @_;
1108 my ($expr_sql, @bind) = $self->render_aqt($v->[0]);
1109 my $op_sql = $self->_sqlcase($op);
1110 return ("${op_sql} ${expr_sql}", @bind);
1113 sub _render_unop_postfix {
1114 my ($self, $op, $v) = @_;
1115 my ($expr_sql, @bind) = $self->render_aqt($v->[0]);
1116 my $op_sql = $self->_sqlcase($op);
1117 return ($expr_sql.' '.$op_sql, @bind);
1120 # Some databases (SQLite) treat col IN (1, 2) different from
1121 # col IN ( (1, 2) ). Use this to strip all outer parens while
1122 # adding them back in the corresponding method
1123 sub _open_outer_paren {
1124 my ($self, $sql) = @_;
1126 while (my ($inner) = $sql =~ /^ \s* \( (.*) \) \s* $/xs) {
1128 # there are closing parens inside, need the heavy duty machinery
1129 # to reevaluate the extraction starting from $sql (full reevaluation)
1130 if ($inner =~ /\)/) {
1131 require Text::Balanced;
1133 my (undef, $remainder) = do {
1134 # idiotic design - writes to $@ but *DOES NOT* throw exceptions
1136 Text::Balanced::extract_bracketed($sql, '()', qr/\s*/);
1139 # the entire expression needs to be a balanced bracketed thing
1140 # (after an extract no remainder sans trailing space)
1141 last if defined $remainder and $remainder =~ /\S/;
1151 #======================================================================
1153 #======================================================================
1155 sub _expand_order_by {
1156 my ($self, $arg) = @_;
1158 return unless defined($arg) and not (ref($arg) eq 'ARRAY' and !@$arg);
1160 my $expander = sub {
1161 my ($self, $dir, $expr) = @_;
1162 my @to_expand = ref($expr) eq 'ARRAY' ? @$expr : $expr;
1163 foreach my $arg (@to_expand) {
1167 and grep /^-(asc|desc)$/, keys %$arg
1169 puke "ordering direction hash passed to order by must have exactly one key (-asc or -desc)";
1173 defined($dir) ? { -op => [ $dir =~ /^-?(.*)$/ ,=> $_ ] } : $_
1175 map $self->expand_expr($_, -ident),
1176 map ref($_) eq 'ARRAY' ? @$_ : $_, @to_expand;
1177 return (@exp > 1 ? { -list => \@exp } : $exp[0]);
1180 local @{$self->{expand}}{qw(-asc -desc)} = (($expander) x 2);
1182 return $self->$expander(undef, $arg);
1186 my ($self, $arg) = @_;
1188 return '' unless defined(my $expanded = $self->_expand_order_by($arg));
1190 my ($sql, @bind) = $self->render_aqt($expanded);
1192 return '' unless length($sql);
1194 my $final_sql = $self->_sqlcase(' order by ').$sql;
1196 return wantarray ? ($final_sql, @bind) : $final_sql;
1199 # _order_by no longer needs to call this so doesn't but DBIC uses it.
1201 sub _order_by_chunks {
1202 my ($self, $arg) = @_;
1204 return () unless defined(my $expanded = $self->_expand_order_by($arg));
1206 return $self->_chunkify_order_by($expanded);
1209 sub _chunkify_order_by {
1210 my ($self, $expanded) = @_;
1212 return grep length, $self->render_aqt($expanded)
1213 if $expanded->{-ident} or @{$expanded->{-literal}||[]} == 1;
1216 if (ref() eq 'HASH' and my $l = $_->{-list}) {
1217 return map $self->_chunkify_order_by($_), @$l;
1219 return [ $self->render_aqt($_) ];
1223 #======================================================================
1224 # DATASOURCE (FOR NOW, JUST PLAIN TABLE OR LIST OF TABLES)
1225 #======================================================================
1231 $self->_expand_maybe_list_expr($from, undef, -ident)
1236 #======================================================================
1238 #======================================================================
1240 sub _expand_maybe_list_expr {
1241 my ($self, $expr, $logic, $default) = @_;
1243 if (ref($expr) eq 'ARRAY') {
1245 map $self->expand_expr($_, $default), @$expr
1252 return $self->expand_expr($e, $default);
1255 # highly optimized, as it's called way too often
1257 # my ($self, $label) = @_;
1259 return '' unless defined $_[1];
1260 return ${$_[1]} if ref($_[1]) eq 'SCALAR';
1261 puke 'Identifier cannot be hashref' if ref($_[1]) eq 'HASH';
1263 unless ($_[0]->{quote_char}) {
1264 if (ref($_[1]) eq 'ARRAY') {
1265 return join($_[0]->{name_sep}||'.', @{$_[1]});
1267 $_[0]->_assert_pass_injection_guard($_[1]);
1272 my $qref = ref $_[0]->{quote_char};
1274 !$qref ? ($_[0]->{quote_char}, $_[0]->{quote_char})
1275 : ($qref eq 'ARRAY') ? @{$_[0]->{quote_char}}
1276 : puke "Unsupported quote_char format: $_[0]->{quote_char}";
1278 my $esc = $_[0]->{escape_char} || $r;
1280 # parts containing * are naturally unquoted
1282 $_[0]->{name_sep}||'',
1286 : do { (my $n = $_) =~ s/(\Q$esc\E|\Q$r\E)/$esc$1/g; $l . $n . $r }
1288 (ref($_[1]) eq 'ARRAY'
1292 ? split (/\Q$_[0]->{name_sep}\E/, $_[1] )
1300 # Conversion, if applicable
1302 #my ($self, $arg) = @_;
1303 if ($_[0]->{convert_where}) {
1304 return $_[0]->_sqlcase($_[0]->{convert_where}) .'(' . $_[1] . ')';
1311 #my ($self, $col, @vals) = @_;
1312 # called often - tighten code
1313 return $_[0]->{bindtype} eq 'columns'
1314 ? map {[$_[1], $_]} @_[2 .. $#_]
1319 # Dies if any element of @bind is not in [colname => value] format
1320 # if bindtype is 'columns'.
1321 sub _assert_bindval_matches_bindtype {
1322 # my ($self, @bind) = @_;
1324 if ($self->{bindtype} eq 'columns') {
1326 if (!defined $_ || ref($_) ne 'ARRAY' || @$_ != 2) {
1327 puke "bindtype 'columns' selected, you need to pass: [column_name => bind_value]"
1333 sub _join_sql_clauses {
1334 my ($self, $logic, $clauses_aref, $bind_aref) = @_;
1336 if (@$clauses_aref > 1) {
1337 my $join = " " . $self->_sqlcase($logic) . " ";
1338 my $sql = '( ' . join($join, @$clauses_aref) . ' )';
1339 return ($sql, @$bind_aref);
1341 elsif (@$clauses_aref) {
1342 return ($clauses_aref->[0], @$bind_aref); # no parentheses
1345 return (); # if no SQL, ignore @$bind_aref
1350 # Fix SQL case, if so requested
1352 # LDNOTE: if $self->{case} is true, then it contains 'lower', so we
1353 # don't touch the argument ... crooked logic, but let's not change it!
1354 return $_[0]->{case} ? $_[1] : uc($_[1]);
1358 #======================================================================
1359 # DISPATCHING FROM REFKIND
1360 #======================================================================
1363 my ($self, $data) = @_;
1365 return 'UNDEF' unless defined $data;
1367 # blessed objects are treated like scalars
1368 my $ref = (Scalar::Util::blessed $data) ? '' : ref $data;
1370 return 'SCALAR' unless $ref;
1373 while ($ref eq 'REF') {
1375 $ref = (Scalar::Util::blessed $data) ? '' : ref $data;
1379 return ($ref||'SCALAR') . ('REF' x $n_steps);
1383 my ($self, $data) = @_;
1384 my @try = ($self->_refkind($data));
1385 push @try, 'SCALAR_or_UNDEF' if $try[0] eq 'SCALAR' || $try[0] eq 'UNDEF';
1386 push @try, 'FALLBACK';
1390 sub _METHOD_FOR_refkind {
1391 my ($self, $meth_prefix, $data) = @_;
1394 for (@{$self->_try_refkind($data)}) {
1395 $method = $self->can($meth_prefix."_".$_)
1399 return $method || puke "cannot dispatch on '$meth_prefix' for ".$self->_refkind($data);
1403 sub _SWITCH_refkind {
1404 my ($self, $data, $dispatch_table) = @_;
1407 for (@{$self->_try_refkind($data)}) {
1408 $coderef = $dispatch_table->{$_}
1412 puke "no dispatch entry for ".$self->_refkind($data)
1421 #======================================================================
1422 # VALUES, GENERATE, AUTOLOAD
1423 #======================================================================
1425 # LDNOTE: original code from nwiger, didn't touch code in that section
1426 # I feel the AUTOLOAD stuff should not be the default, it should
1427 # only be activated on explicit demand by user.
1431 my $data = shift || return;
1432 puke "Argument to ", __PACKAGE__, "->values must be a \\%hash"
1433 unless ref $data eq 'HASH';
1436 foreach my $k (sort keys %$data) {
1437 my $v = $data->{$k};
1438 $self->_SWITCH_refkind($v, {
1440 if ($self->{array_datatypes}) { # array datatype
1441 push @all_bind, $self->_bindtype($k, $v);
1443 else { # literal SQL with bind
1444 my ($sql, @bind) = @$v;
1445 $self->_assert_bindval_matches_bindtype(@bind);
1446 push @all_bind, @bind;
1449 ARRAYREFREF => sub { # literal SQL with bind
1450 my ($sql, @bind) = @${$v};
1451 $self->_assert_bindval_matches_bindtype(@bind);
1452 push @all_bind, @bind;
1454 SCALARREF => sub { # literal SQL without bind
1456 SCALAR_or_UNDEF => sub {
1457 push @all_bind, $self->_bindtype($k, $v);
1468 my(@sql, @sqlq, @sqlv);
1472 if ($ref eq 'HASH') {
1473 for my $k (sort keys %$_) {
1476 my $label = $self->_quote($k);
1477 if ($r eq 'ARRAY') {
1478 # literal SQL with bind
1479 my ($sql, @bind) = @$v;
1480 $self->_assert_bindval_matches_bindtype(@bind);
1481 push @sqlq, "$label = $sql";
1483 } elsif ($r eq 'SCALAR') {
1484 # literal SQL without bind
1485 push @sqlq, "$label = $$v";
1487 push @sqlq, "$label = ?";
1488 push @sqlv, $self->_bindtype($k, $v);
1491 push @sql, $self->_sqlcase('set'), join ', ', @sqlq;
1492 } elsif ($ref eq 'ARRAY') {
1493 # unlike insert(), assume these are ONLY the column names, i.e. for SQL
1496 if ($r eq 'ARRAY') { # literal SQL with bind
1497 my ($sql, @bind) = @$v;
1498 $self->_assert_bindval_matches_bindtype(@bind);
1501 } elsif ($r eq 'SCALAR') { # literal SQL without bind
1502 # embedded literal SQL
1509 push @sql, '(' . join(', ', @sqlq) . ')';
1510 } elsif ($ref eq 'SCALAR') {
1514 # strings get case twiddled
1515 push @sql, $self->_sqlcase($_);
1519 my $sql = join ' ', @sql;
1521 # this is pretty tricky
1522 # if ask for an array, return ($stmt, @bind)
1523 # otherwise, s/?/shift @sqlv/ to put it inline
1525 return ($sql, @sqlv);
1527 1 while $sql =~ s/\?/my $d = shift(@sqlv);
1528 ref $d ? $d->[1] : $d/e;
1537 # This allows us to check for a local, then _form, attr
1539 my($name) = $AUTOLOAD =~ /.*::(.+)/;
1540 return $self->generate($name, @_);
1551 SQL::Abstract - Generate SQL from Perl data structures
1557 my $sql = SQL::Abstract->new;
1559 my($stmt, @bind) = $sql->select($source, \@fields, \%where, $order);
1561 my($stmt, @bind) = $sql->insert($table, \%fieldvals || \@values);
1563 my($stmt, @bind) = $sql->update($table, \%fieldvals, \%where);
1565 my($stmt, @bind) = $sql->delete($table, \%where);
1567 # Then, use these in your DBI statements
1568 my $sth = $dbh->prepare($stmt);
1569 $sth->execute(@bind);
1571 # Just generate the WHERE clause
1572 my($stmt, @bind) = $sql->where(\%where, $order);
1574 # Return values in the same order, for hashed queries
1575 # See PERFORMANCE section for more details
1576 my @bind = $sql->values(\%fieldvals);
1580 This module was inspired by the excellent L<DBIx::Abstract>.
1581 However, in using that module I found that what I really wanted
1582 to do was generate SQL, but still retain complete control over my
1583 statement handles and use the DBI interface. So, I set out to
1584 create an abstract SQL generation module.
1586 While based on the concepts used by L<DBIx::Abstract>, there are
1587 several important differences, especially when it comes to WHERE
1588 clauses. I have modified the concepts used to make the SQL easier
1589 to generate from Perl data structures and, IMO, more intuitive.
1590 The underlying idea is for this module to do what you mean, based
1591 on the data structures you provide it. The big advantage is that
1592 you don't have to modify your code every time your data changes,
1593 as this module figures it out.
1595 To begin with, an SQL INSERT is as easy as just specifying a hash
1596 of C<key=value> pairs:
1599 name => 'Jimbo Bobson',
1600 phone => '123-456-7890',
1601 address => '42 Sister Lane',
1602 city => 'St. Louis',
1603 state => 'Louisiana',
1606 The SQL can then be generated with this:
1608 my($stmt, @bind) = $sql->insert('people', \%data);
1610 Which would give you something like this:
1612 $stmt = "INSERT INTO people
1613 (address, city, name, phone, state)
1614 VALUES (?, ?, ?, ?, ?)";
1615 @bind = ('42 Sister Lane', 'St. Louis', 'Jimbo Bobson',
1616 '123-456-7890', 'Louisiana');
1618 These are then used directly in your DBI code:
1620 my $sth = $dbh->prepare($stmt);
1621 $sth->execute(@bind);
1623 =head2 Inserting and Updating Arrays
1625 If your database has array types (like for example Postgres),
1626 activate the special option C<< array_datatypes => 1 >>
1627 when creating the C<SQL::Abstract> object.
1628 Then you may use an arrayref to insert and update database array types:
1630 my $sql = SQL::Abstract->new(array_datatypes => 1);
1632 planets => [qw/Mercury Venus Earth Mars/]
1635 my($stmt, @bind) = $sql->insert('solar_system', \%data);
1639 $stmt = "INSERT INTO solar_system (planets) VALUES (?)"
1641 @bind = (['Mercury', 'Venus', 'Earth', 'Mars']);
1644 =head2 Inserting and Updating SQL
1646 In order to apply SQL functions to elements of your C<%data> you may
1647 specify a reference to an arrayref for the given hash value. For example,
1648 if you need to execute the Oracle C<to_date> function on a value, you can
1649 say something like this:
1653 date_entered => \[ "to_date(?,'MM/DD/YYYY')", "03/02/2003" ],
1656 The first value in the array is the actual SQL. Any other values are
1657 optional and would be included in the bind values array. This gives
1660 my($stmt, @bind) = $sql->insert('people', \%data);
1662 $stmt = "INSERT INTO people (name, date_entered)
1663 VALUES (?, to_date(?,'MM/DD/YYYY'))";
1664 @bind = ('Bill', '03/02/2003');
1666 An UPDATE is just as easy, all you change is the name of the function:
1668 my($stmt, @bind) = $sql->update('people', \%data);
1670 Notice that your C<%data> isn't touched; the module will generate
1671 the appropriately quirky SQL for you automatically. Usually you'll
1672 want to specify a WHERE clause for your UPDATE, though, which is
1673 where handling C<%where> hashes comes in handy...
1675 =head2 Complex where statements
1677 This module can generate pretty complicated WHERE statements
1678 easily. For example, simple C<key=value> pairs are taken to mean
1679 equality, and if you want to see if a field is within a set
1680 of values, you can use an arrayref. Let's say we wanted to
1681 SELECT some data based on this criteria:
1684 requestor => 'inna',
1685 worker => ['nwiger', 'rcwe', 'sfz'],
1686 status => { '!=', 'completed' }
1689 my($stmt, @bind) = $sql->select('tickets', '*', \%where);
1691 The above would give you something like this:
1693 $stmt = "SELECT * FROM tickets WHERE
1694 ( requestor = ? ) AND ( status != ? )
1695 AND ( worker = ? OR worker = ? OR worker = ? )";
1696 @bind = ('inna', 'completed', 'nwiger', 'rcwe', 'sfz');
1698 Which you could then use in DBI code like so:
1700 my $sth = $dbh->prepare($stmt);
1701 $sth->execute(@bind);
1707 The methods are simple. There's one for every major SQL operation,
1708 and a constructor you use first. The arguments are specified in a
1709 similar order for each method (table, then fields, then a where
1710 clause) to try and simplify things.
1712 =head2 new(option => 'value')
1714 The C<new()> function takes a list of options and values, and returns
1715 a new B<SQL::Abstract> object which can then be used to generate SQL
1716 through the methods below. The options accepted are:
1722 If set to 'lower', then SQL will be generated in all lowercase. By
1723 default SQL is generated in "textbook" case meaning something like:
1725 SELECT a_field FROM a_table WHERE some_field LIKE '%someval%'
1727 Any setting other than 'lower' is ignored.
1731 This determines what the default comparison operator is. By default
1732 it is C<=>, meaning that a hash like this:
1734 %where = (name => 'nwiger', email => 'nate@wiger.org');
1736 Will generate SQL like this:
1738 WHERE name = 'nwiger' AND email = 'nate@wiger.org'
1740 However, you may want loose comparisons by default, so if you set
1741 C<cmp> to C<like> you would get SQL such as:
1743 WHERE name like 'nwiger' AND email like 'nate@wiger.org'
1745 You can also override the comparison on an individual basis - see
1746 the huge section on L</"WHERE CLAUSES"> at the bottom.
1748 =item sqltrue, sqlfalse
1750 Expressions for inserting boolean values within SQL statements.
1751 By default these are C<1=1> and C<1=0>. They are used
1752 by the special operators C<-in> and C<-not_in> for generating
1753 correct SQL even when the argument is an empty array (see below).
1757 This determines the default logical operator for multiple WHERE
1758 statements in arrays or hashes. If absent, the default logic is "or"
1759 for arrays, and "and" for hashes. This means that a WHERE
1763 event_date => {'>=', '2/13/99'},
1764 event_date => {'<=', '4/24/03'},
1767 will generate SQL like this:
1769 WHERE event_date >= '2/13/99' OR event_date <= '4/24/03'
1771 This is probably not what you want given this query, though (look
1772 at the dates). To change the "OR" to an "AND", simply specify:
1774 my $sql = SQL::Abstract->new(logic => 'and');
1776 Which will change the above C<WHERE> to:
1778 WHERE event_date >= '2/13/99' AND event_date <= '4/24/03'
1780 The logic can also be changed locally by inserting
1781 a modifier in front of an arrayref:
1783 @where = (-and => [event_date => {'>=', '2/13/99'},
1784 event_date => {'<=', '4/24/03'} ]);
1786 See the L</"WHERE CLAUSES"> section for explanations.
1790 This will automatically convert comparisons using the specified SQL
1791 function for both column and value. This is mostly used with an argument
1792 of C<upper> or C<lower>, so that the SQL will have the effect of
1793 case-insensitive "searches". For example, this:
1795 $sql = SQL::Abstract->new(convert => 'upper');
1796 %where = (keywords => 'MaKe iT CAse inSeNSItive');
1798 Will turn out the following SQL:
1800 WHERE upper(keywords) like upper('MaKe iT CAse inSeNSItive')
1802 The conversion can be C<upper()>, C<lower()>, or any other SQL function
1803 that can be applied symmetrically to fields (actually B<SQL::Abstract> does
1804 not validate this option; it will just pass through what you specify verbatim).
1808 This is a kludge because many databases suck. For example, you can't
1809 just bind values using DBI's C<execute()> for Oracle C<CLOB> or C<BLOB> fields.
1810 Instead, you have to use C<bind_param()>:
1812 $sth->bind_param(1, 'reg data');
1813 $sth->bind_param(2, $lots, {ora_type => ORA_CLOB});
1815 The problem is, B<SQL::Abstract> will normally just return a C<@bind> array,
1816 which loses track of which field each slot refers to. Fear not.
1818 If you specify C<bindtype> in new, you can determine how C<@bind> is returned.
1819 Currently, you can specify either C<normal> (default) or C<columns>. If you
1820 specify C<columns>, you will get an array that looks like this:
1822 my $sql = SQL::Abstract->new(bindtype => 'columns');
1823 my($stmt, @bind) = $sql->insert(...);
1826 [ 'column1', 'value1' ],
1827 [ 'column2', 'value2' ],
1828 [ 'column3', 'value3' ],
1831 You can then iterate through this manually, using DBI's C<bind_param()>.
1833 $sth->prepare($stmt);
1836 my($col, $data) = @$_;
1837 if ($col eq 'details' || $col eq 'comments') {
1838 $sth->bind_param($i, $data, {ora_type => ORA_CLOB});
1839 } elsif ($col eq 'image') {
1840 $sth->bind_param($i, $data, {ora_type => ORA_BLOB});
1842 $sth->bind_param($i, $data);
1846 $sth->execute; # execute without @bind now
1848 Now, why would you still use B<SQL::Abstract> if you have to do this crap?
1849 Basically, the advantage is still that you don't have to care which fields
1850 are or are not included. You could wrap that above C<for> loop in a simple
1851 sub called C<bind_fields()> or something and reuse it repeatedly. You still
1852 get a layer of abstraction over manual SQL specification.
1854 Note that if you set L</bindtype> to C<columns>, the C<\[ $sql, @bind ]>
1855 construct (see L</Literal SQL with placeholders and bind values (subqueries)>)
1856 will expect the bind values in this format.
1860 This is the character that a table or column name will be quoted
1861 with. By default this is an empty string, but you could set it to
1862 the character C<`>, to generate SQL like this:
1864 SELECT `a_field` FROM `a_table` WHERE `some_field` LIKE '%someval%'
1866 Alternatively, you can supply an array ref of two items, the first being the left
1867 hand quote character, and the second the right hand quote character. For
1868 example, you could supply C<['[',']']> for SQL Server 2000 compliant quotes
1869 that generates SQL like this:
1871 SELECT [a_field] FROM [a_table] WHERE [some_field] LIKE '%someval%'
1873 Quoting is useful if you have tables or columns names that are reserved
1874 words in your database's SQL dialect.
1878 This is the character that will be used to escape L</quote_char>s appearing
1879 in an identifier before it has been quoted.
1881 The parameter default in case of a single L</quote_char> character is the quote
1884 When opening-closing-style quoting is used (L</quote_char> is an arrayref)
1885 this parameter defaults to the B<closing (right)> L</quote_char>. Occurrences
1886 of the B<opening (left)> L</quote_char> within the identifier are currently left
1887 untouched. The default for opening-closing-style quotes may change in future
1888 versions, thus you are B<strongly encouraged> to specify the escape character
1893 This is the character that separates a table and column name. It is
1894 necessary to specify this when the C<quote_char> option is selected,
1895 so that tables and column names can be individually quoted like this:
1897 SELECT `table`.`one_field` FROM `table` WHERE `table`.`other_field` = 1
1899 =item injection_guard
1901 A regular expression C<qr/.../> that is applied to any C<-function> and unquoted
1902 column name specified in a query structure. This is a safety mechanism to avoid
1903 injection attacks when mishandling user input e.g.:
1905 my %condition_as_column_value_pairs = get_values_from_user();
1906 $sqla->select( ... , \%condition_as_column_value_pairs );
1908 If the expression matches an exception is thrown. Note that literal SQL
1909 supplied via C<\'...'> or C<\['...']> is B<not> checked in any way.
1911 Defaults to checking for C<;> and the C<GO> keyword (TransactSQL)
1913 =item array_datatypes
1915 When this option is true, arrayrefs in INSERT or UPDATE are
1916 interpreted as array datatypes and are passed directly
1918 When this option is false, arrayrefs are interpreted
1919 as literal SQL, just like refs to arrayrefs
1920 (but this behavior is for backwards compatibility; when writing
1921 new queries, use the "reference to arrayref" syntax
1927 Takes a reference to a list of "special operators"
1928 to extend the syntax understood by L<SQL::Abstract>.
1929 See section L</"SPECIAL OPERATORS"> for details.
1933 Takes a reference to a list of "unary operators"
1934 to extend the syntax understood by L<SQL::Abstract>.
1935 See section L</"UNARY OPERATORS"> for details.
1941 =head2 insert($table, \@values || \%fieldvals, \%options)
1943 This is the simplest function. You simply give it a table name
1944 and either an arrayref of values or hashref of field/value pairs.
1945 It returns an SQL INSERT statement and a list of bind values.
1946 See the sections on L</"Inserting and Updating Arrays"> and
1947 L</"Inserting and Updating SQL"> for information on how to insert
1948 with those data types.
1950 The optional C<\%options> hash reference may contain additional
1951 options to generate the insert SQL. Currently supported options
1958 Takes either a scalar of raw SQL fields, or an array reference of
1959 field names, and adds on an SQL C<RETURNING> statement at the end.
1960 This allows you to return data generated by the insert statement
1961 (such as row IDs) without performing another C<SELECT> statement.
1962 Note, however, this is not part of the SQL standard and may not
1963 be supported by all database engines.
1967 =head2 update($table, \%fieldvals, \%where, \%options)
1969 This takes a table, hashref of field/value pairs, and an optional
1970 hashref L<WHERE clause|/WHERE CLAUSES>. It returns an SQL UPDATE function and a list
1972 See the sections on L</"Inserting and Updating Arrays"> and
1973 L</"Inserting and Updating SQL"> for information on how to insert
1974 with those data types.
1976 The optional C<\%options> hash reference may contain additional
1977 options to generate the update SQL. Currently supported options
1984 See the C<returning> option to
1985 L<insert|/insert($table, \@values || \%fieldvals, \%options)>.
1989 =head2 select($source, $fields, $where, $order)
1991 This returns a SQL SELECT statement and associated list of bind values, as
1992 specified by the arguments:
1998 Specification of the 'FROM' part of the statement.
1999 The argument can be either a plain scalar (interpreted as a table
2000 name, will be quoted), or an arrayref (interpreted as a list
2001 of table names, joined by commas, quoted), or a scalarref
2002 (literal SQL, not quoted).
2006 Specification of the list of fields to retrieve from
2008 The argument can be either an arrayref (interpreted as a list
2009 of field names, will be joined by commas and quoted), or a
2010 plain scalar (literal SQL, not quoted).
2011 Please observe that this API is not as flexible as that of
2012 the first argument C<$source>, for backwards compatibility reasons.
2016 Optional argument to specify the WHERE part of the query.
2017 The argument is most often a hashref, but can also be
2018 an arrayref or plain scalar --
2019 see section L<WHERE clause|/"WHERE CLAUSES"> for details.
2023 Optional argument to specify the ORDER BY part of the query.
2024 The argument can be a scalar, a hashref or an arrayref
2025 -- see section L<ORDER BY clause|/"ORDER BY CLAUSES">
2031 =head2 delete($table, \%where, \%options)
2033 This takes a table name and optional hashref L<WHERE clause|/WHERE CLAUSES>.
2034 It returns an SQL DELETE statement and list of bind values.
2036 The optional C<\%options> hash reference may contain additional
2037 options to generate the delete SQL. Currently supported options
2044 See the C<returning> option to
2045 L<insert|/insert($table, \@values || \%fieldvals, \%options)>.
2049 =head2 where(\%where, $order)
2051 This is used to generate just the WHERE clause. For example,
2052 if you have an arbitrary data structure and know what the
2053 rest of your SQL is going to look like, but want an easy way
2054 to produce a WHERE clause, use this. It returns an SQL WHERE
2055 clause and list of bind values.
2058 =head2 values(\%data)
2060 This just returns the values from the hash C<%data>, in the same
2061 order that would be returned from any of the other above queries.
2062 Using this allows you to markedly speed up your queries if you
2063 are affecting lots of rows. See below under the L</"PERFORMANCE"> section.
2065 =head2 generate($any, 'number', $of, \@data, $struct, \%types)
2067 Warning: This is an experimental method and subject to change.
2069 This returns arbitrarily generated SQL. It's a really basic shortcut.
2070 It will return two different things, depending on return context:
2072 my($stmt, @bind) = $sql->generate('create table', \$table, \@fields);
2073 my $stmt_and_val = $sql->generate('create table', \$table, \@fields);
2075 These would return the following:
2077 # First calling form
2078 $stmt = "CREATE TABLE test (?, ?)";
2079 @bind = (field1, field2);
2081 # Second calling form
2082 $stmt_and_val = "CREATE TABLE test (field1, field2)";
2084 Depending on what you're trying to do, it's up to you to choose the correct
2085 format. In this example, the second form is what you would want.
2089 $sql->generate('alter session', { nls_date_format => 'MM/YY' });
2093 ALTER SESSION SET nls_date_format = 'MM/YY'
2095 You get the idea. Strings get their case twiddled, but everything
2096 else remains verbatim.
2098 =head1 EXPORTABLE FUNCTIONS
2100 =head2 is_plain_value
2102 Determines if the supplied argument is a plain value as understood by this
2107 =item * The value is C<undef>
2109 =item * The value is a non-reference
2111 =item * The value is an object with stringification overloading
2113 =item * The value is of the form C<< { -value => $anything } >>
2117 On failure returns C<undef>, on success returns a B<scalar> reference
2118 to the original supplied argument.
2124 The stringification overloading detection is rather advanced: it takes
2125 into consideration not only the presence of a C<""> overload, but if that
2126 fails also checks for enabled
2127 L<autogenerated versions of C<"">|overload/Magic Autogeneration>, based
2128 on either C<0+> or C<bool>.
2130 Unfortunately testing in the field indicates that this
2131 detection B<< may tickle a latent bug in perl versions before 5.018 >>,
2132 but only when very large numbers of stringifying objects are involved.
2133 At the time of writing ( Sep 2014 ) there is no clear explanation of
2134 the direct cause, nor is there a manageably small test case that reliably
2135 reproduces the problem.
2137 If you encounter any of the following exceptions in B<random places within
2138 your application stack> - this module may be to blame:
2140 Operation "ne": no method found,
2141 left argument in overloaded package <something>,
2142 right argument in overloaded package <something>
2146 Stub found while resolving method "???" overloading """" in package <something>
2148 If you fall victim to the above - please attempt to reduce the problem
2149 to something that could be sent to the L<SQL::Abstract developers
2150 |DBIx::Class/GETTING HELP/SUPPORT>
2151 (either publicly or privately). As a workaround in the meantime you can
2152 set C<$ENV{SQLA_ISVALUE_IGNORE_AUTOGENERATED_STRINGIFICATION}> to a true
2153 value, which will most likely eliminate your problem (at the expense of
2154 not being able to properly detect exotic forms of stringification).
2156 This notice and environment variable will be removed in a future version,
2157 as soon as the underlying problem is found and a reliable workaround is
2162 =head2 is_literal_value
2164 Determines if the supplied argument is a literal value as understood by this
2169 =item * C<\$sql_string>
2171 =item * C<\[ $sql_string, @bind_values ]>
2175 On failure returns C<undef>, on success returns an B<array> reference
2176 containing the unpacked version of the supplied literal SQL and bind values.
2178 =head1 WHERE CLAUSES
2182 This module uses a variation on the idea from L<DBIx::Abstract>. It
2183 is B<NOT>, repeat I<not> 100% compatible. B<The main logic of this
2184 module is that things in arrays are OR'ed, and things in hashes
2187 The easiest way to explain is to show lots of examples. After
2188 each C<%where> hash shown, it is assumed you used:
2190 my($stmt, @bind) = $sql->where(\%where);
2192 However, note that the C<%where> hash can be used directly in any
2193 of the other functions as well, as described above.
2195 =head2 Key-value pairs
2197 So, let's get started. To begin, a simple hash:
2201 status => 'completed'
2204 Is converted to SQL C<key = val> statements:
2206 $stmt = "WHERE user = ? AND status = ?";
2207 @bind = ('nwiger', 'completed');
2209 One common thing I end up doing is having a list of values that
2210 a field can be in. To do this, simply specify a list inside of
2215 status => ['assigned', 'in-progress', 'pending'];
2218 This simple code will create the following:
2220 $stmt = "WHERE user = ? AND ( status = ? OR status = ? OR status = ? )";
2221 @bind = ('nwiger', 'assigned', 'in-progress', 'pending');
2223 A field associated to an empty arrayref will be considered a
2224 logical false and will generate 0=1.
2226 =head2 Tests for NULL values
2228 If the value part is C<undef> then this is converted to SQL <IS NULL>
2237 $stmt = "WHERE user = ? AND status IS NULL";
2240 To test if a column IS NOT NULL:
2244 status => { '!=', undef },
2247 =head2 Specific comparison operators
2249 If you want to specify a different type of operator for your comparison,
2250 you can use a hashref for a given column:
2254 status => { '!=', 'completed' }
2257 Which would generate:
2259 $stmt = "WHERE user = ? AND status != ?";
2260 @bind = ('nwiger', 'completed');
2262 To test against multiple values, just enclose the values in an arrayref:
2264 status => { '=', ['assigned', 'in-progress', 'pending'] };
2266 Which would give you:
2268 "WHERE status = ? OR status = ? OR status = ?"
2271 The hashref can also contain multiple pairs, in which case it is expanded
2272 into an C<AND> of its elements:
2276 status => { '!=', 'completed', -not_like => 'pending%' }
2279 # Or more dynamically, like from a form
2280 $where{user} = 'nwiger';
2281 $where{status}{'!='} = 'completed';
2282 $where{status}{'-not_like'} = 'pending%';
2284 # Both generate this
2285 $stmt = "WHERE user = ? AND status != ? AND status NOT LIKE ?";
2286 @bind = ('nwiger', 'completed', 'pending%');
2289 To get an OR instead, you can combine it with the arrayref idea:
2293 priority => [ { '=', 2 }, { '>', 5 } ]
2296 Which would generate:
2298 $stmt = "WHERE ( priority = ? OR priority > ? ) AND user = ?";
2299 @bind = ('2', '5', 'nwiger');
2301 If you want to include literal SQL (with or without bind values), just use a
2302 scalar reference or reference to an arrayref as the value:
2305 date_entered => { '>' => \["to_date(?, 'MM/DD/YYYY')", "11/26/2008"] },
2306 date_expires => { '<' => \"now()" }
2309 Which would generate:
2311 $stmt = "WHERE date_entered > to_date(?, 'MM/DD/YYYY') AND date_expires < now()";
2312 @bind = ('11/26/2008');
2315 =head2 Logic and nesting operators
2317 In the example above,
2318 there is a subtle trap if you want to say something like
2319 this (notice the C<AND>):
2321 WHERE priority != ? AND priority != ?
2323 Because, in Perl you I<can't> do this:
2325 priority => { '!=' => 2, '!=' => 1 }
2327 As the second C<!=> key will obliterate the first. The solution
2328 is to use the special C<-modifier> form inside an arrayref:
2330 priority => [ -and => {'!=', 2},
2334 Normally, these would be joined by C<OR>, but the modifier tells it
2335 to use C<AND> instead. (Hint: You can use this in conjunction with the
2336 C<logic> option to C<new()> in order to change the way your queries
2337 work by default.) B<Important:> Note that the C<-modifier> goes
2338 B<INSIDE> the arrayref, as an extra first element. This will
2339 B<NOT> do what you think it might:
2341 priority => -and => [{'!=', 2}, {'!=', 1}] # WRONG!
2343 Here is a quick list of equivalencies, since there is some overlap:
2346 status => {'!=', 'completed', 'not like', 'pending%' }
2347 status => [ -and => {'!=', 'completed'}, {'not like', 'pending%'}]
2350 status => {'=', ['assigned', 'in-progress']}
2351 status => [ -or => {'=', 'assigned'}, {'=', 'in-progress'}]
2352 status => [ {'=', 'assigned'}, {'=', 'in-progress'} ]
2356 =head2 Special operators: IN, BETWEEN, etc.
2358 You can also use the hashref format to compare a list of fields using the
2359 C<IN> comparison operator, by specifying the list as an arrayref:
2362 status => 'completed',
2363 reportid => { -in => [567, 2335, 2] }
2366 Which would generate:
2368 $stmt = "WHERE status = ? AND reportid IN (?,?,?)";
2369 @bind = ('completed', '567', '2335', '2');
2371 The reverse operator C<-not_in> generates SQL C<NOT IN> and is used in
2374 If the argument to C<-in> is an empty array, 'sqlfalse' is generated
2375 (by default: C<1=0>). Similarly, C<< -not_in => [] >> generates
2376 'sqltrue' (by default: C<1=1>).
2378 In addition to the array you can supply a chunk of literal sql or
2379 literal sql with bind:
2382 customer => { -in => \[
2383 'SELECT cust_id FROM cust WHERE balance > ?',
2386 status => { -in => \'SELECT status_codes FROM states' },
2392 customer IN ( SELECT cust_id FROM cust WHERE balance > ? )
2393 AND status IN ( SELECT status_codes FROM states )
2397 Finally, if the argument to C<-in> is not a reference, it will be
2398 treated as a single-element array.
2400 Another pair of operators is C<-between> and C<-not_between>,
2401 used with an arrayref of two values:
2405 completion_date => {
2406 -not_between => ['2002-10-01', '2003-02-06']
2412 WHERE user = ? AND completion_date NOT BETWEEN ( ? AND ? )
2414 Just like with C<-in> all plausible combinations of literal SQL
2418 start0 => { -between => [ 1, 2 ] },
2419 start1 => { -between => \["? AND ?", 1, 2] },
2420 start2 => { -between => \"lower(x) AND upper(y)" },
2421 start3 => { -between => [
2423 \["upper(?)", 'stuff' ],
2430 ( start0 BETWEEN ? AND ? )
2431 AND ( start1 BETWEEN ? AND ? )
2432 AND ( start2 BETWEEN lower(x) AND upper(y) )
2433 AND ( start3 BETWEEN lower(x) AND upper(?) )
2435 @bind = (1, 2, 1, 2, 'stuff');
2438 These are the two builtin "special operators"; but the
2439 list can be expanded: see section L</"SPECIAL OPERATORS"> below.
2441 =head2 Unary operators: bool
2443 If you wish to test against boolean columns or functions within your
2444 database you can use the C<-bool> and C<-not_bool> operators. For
2445 example to test the column C<is_user> being true and the column
2446 C<is_enabled> being false you would use:-
2450 -not_bool => 'is_enabled',
2455 WHERE is_user AND NOT is_enabled
2457 If a more complex combination is required, testing more conditions,
2458 then you should use the and/or operators:-
2463 -not_bool => { two=> { -rlike => 'bar' } },
2464 -not_bool => { three => [ { '=', 2 }, { '>', 5 } ] },
2475 (NOT ( three = ? OR three > ? ))
2478 =head2 Nested conditions, -and/-or prefixes
2480 So far, we've seen how multiple conditions are joined with a top-level
2481 C<AND>. We can change this by putting the different conditions we want in
2482 hashes and then putting those hashes in an array. For example:
2487 status => { -like => ['pending%', 'dispatched'] },
2491 status => 'unassigned',
2495 This data structure would create the following:
2497 $stmt = "WHERE ( user = ? AND ( status LIKE ? OR status LIKE ? ) )
2498 OR ( user = ? AND status = ? ) )";
2499 @bind = ('nwiger', 'pending', 'dispatched', 'robot', 'unassigned');
2502 Clauses in hashrefs or arrayrefs can be prefixed with an C<-and> or C<-or>
2503 to change the logic inside:
2509 -and => [ workhrs => {'>', 20}, geo => 'ASIA' ],
2510 -or => { workhrs => {'<', 50}, geo => 'EURO' },
2517 $stmt = "WHERE ( user = ?
2518 AND ( ( workhrs > ? AND geo = ? )
2519 OR ( workhrs < ? OR geo = ? ) ) )";
2520 @bind = ('nwiger', '20', 'ASIA', '50', 'EURO');
2522 =head3 Algebraic inconsistency, for historical reasons
2524 C<Important note>: when connecting several conditions, the C<-and->|C<-or>
2525 operator goes C<outside> of the nested structure; whereas when connecting
2526 several constraints on one column, the C<-and> operator goes
2527 C<inside> the arrayref. Here is an example combining both features:
2530 -and => [a => 1, b => 2],
2531 -or => [c => 3, d => 4],
2532 e => [-and => {-like => 'foo%'}, {-like => '%bar'} ]
2537 WHERE ( ( ( a = ? AND b = ? )
2538 OR ( c = ? OR d = ? )
2539 OR ( e LIKE ? AND e LIKE ? ) ) )
2541 This difference in syntax is unfortunate but must be preserved for
2542 historical reasons. So be careful: the two examples below would
2543 seem algebraically equivalent, but they are not
2546 { -like => 'foo%' },
2547 { -like => '%bar' },
2549 # yields: WHERE ( ( col LIKE ? AND col LIKE ? ) )
2552 { col => { -like => 'foo%' } },
2553 { col => { -like => '%bar' } },
2555 # yields: WHERE ( ( col LIKE ? OR col LIKE ? ) )
2558 =head2 Literal SQL and value type operators
2560 The basic premise of SQL::Abstract is that in WHERE specifications the "left
2561 side" is a column name and the "right side" is a value (normally rendered as
2562 a placeholder). This holds true for both hashrefs and arrayref pairs as you
2563 see in the L</WHERE CLAUSES> examples above. Sometimes it is necessary to
2564 alter this behavior. There are several ways of doing so.
2568 This is a virtual operator that signals the string to its right side is an
2569 identifier (a column name) and not a value. For example to compare two
2570 columns you would write:
2573 priority => { '<', 2 },
2574 requestor => { -ident => 'submitter' },
2579 $stmt = "WHERE priority < ? AND requestor = submitter";
2582 If you are maintaining legacy code you may see a different construct as
2583 described in L</Deprecated usage of Literal SQL>, please use C<-ident> in new
2588 This is a virtual operator that signals that the construct to its right side
2589 is a value to be passed to DBI. This is for example necessary when you want
2590 to write a where clause against an array (for RDBMS that support such
2591 datatypes). For example:
2594 array => { -value => [1, 2, 3] }
2599 $stmt = 'WHERE array = ?';
2600 @bind = ([1, 2, 3]);
2602 Note that if you were to simply say:
2608 the result would probably not be what you wanted:
2610 $stmt = 'WHERE array = ? OR array = ? OR array = ?';
2615 Finally, sometimes only literal SQL will do. To include a random snippet
2616 of SQL verbatim, you specify it as a scalar reference. Consider this only
2617 as a last resort. Usually there is a better way. For example:
2620 priority => { '<', 2 },
2621 requestor => { -in => \'(SELECT name FROM hitmen)' },
2626 $stmt = "WHERE priority < ? AND requestor IN (SELECT name FROM hitmen)"
2629 Note that in this example, you only get one bind parameter back, since
2630 the verbatim SQL is passed as part of the statement.
2634 Never use untrusted input as a literal SQL argument - this is a massive
2635 security risk (there is no way to check literal snippets for SQL
2636 injections and other nastyness). If you need to deal with untrusted input
2637 use literal SQL with placeholders as described next.
2639 =head3 Literal SQL with placeholders and bind values (subqueries)
2641 If the literal SQL to be inserted has placeholders and bind values,
2642 use a reference to an arrayref (yes this is a double reference --
2643 not so common, but perfectly legal Perl). For example, to find a date
2644 in Postgres you can use something like this:
2647 date_column => \[ "= date '2008-09-30' - ?::integer", 10 ]
2652 $stmt = "WHERE ( date_column = date '2008-09-30' - ?::integer )"
2655 Note that you must pass the bind values in the same format as they are returned
2656 by L<where|/where(\%where, $order)>. This means that if you set L</bindtype>
2657 to C<columns>, you must provide the bind values in the
2658 C<< [ column_meta => value ] >> format, where C<column_meta> is an opaque
2659 scalar value; most commonly the column name, but you can use any scalar value
2660 (including references and blessed references), L<SQL::Abstract> will simply
2661 pass it through intact. So if C<bindtype> is set to C<columns> the above
2662 example will look like:
2665 date_column => \[ "= date '2008-09-30' - ?::integer", [ {} => 10 ] ]
2668 Literal SQL is especially useful for nesting parenthesized clauses in the
2669 main SQL query. Here is a first example:
2671 my ($sub_stmt, @sub_bind) = ("SELECT c1 FROM t1 WHERE c2 < ? AND c3 LIKE ?",
2675 bar => \["IN ($sub_stmt)" => @sub_bind],
2680 $stmt = "WHERE (foo = ? AND bar IN (SELECT c1 FROM t1
2681 WHERE c2 < ? AND c3 LIKE ?))";
2682 @bind = (1234, 100, "foo%");
2684 Other subquery operators, like for example C<"E<gt> ALL"> or C<"NOT IN">,
2685 are expressed in the same way. Of course the C<$sub_stmt> and
2686 its associated bind values can be generated through a former call
2689 my ($sub_stmt, @sub_bind)
2690 = $sql->select("t1", "c1", {c2 => {"<" => 100},
2691 c3 => {-like => "foo%"}});
2694 bar => \["> ALL ($sub_stmt)" => @sub_bind],
2697 In the examples above, the subquery was used as an operator on a column;
2698 but the same principle also applies for a clause within the main C<%where>
2699 hash, like an EXISTS subquery:
2701 my ($sub_stmt, @sub_bind)
2702 = $sql->select("t1", "*", {c1 => 1, c2 => \"> t0.c0"});
2703 my %where = ( -and => [
2705 \["EXISTS ($sub_stmt)" => @sub_bind],
2710 $stmt = "WHERE (foo = ? AND EXISTS (SELECT * FROM t1
2711 WHERE c1 = ? AND c2 > t0.c0))";
2715 Observe that the condition on C<c2> in the subquery refers to
2716 column C<t0.c0> of the main query: this is I<not> a bind
2717 value, so we have to express it through a scalar ref.
2718 Writing C<< c2 => {">" => "t0.c0"} >> would have generated
2719 C<< c2 > ? >> with bind value C<"t0.c0"> ... not exactly
2720 what we wanted here.
2722 Finally, here is an example where a subquery is used
2723 for expressing unary negation:
2725 my ($sub_stmt, @sub_bind)
2726 = $sql->where({age => [{"<" => 10}, {">" => 20}]});
2727 $sub_stmt =~ s/^ where //i; # don't want "WHERE" in the subclause
2729 lname => {like => '%son%'},
2730 \["NOT ($sub_stmt)" => @sub_bind],
2735 $stmt = "lname LIKE ? AND NOT ( age < ? OR age > ? )"
2736 @bind = ('%son%', 10, 20)
2738 =head3 Deprecated usage of Literal SQL
2740 Below are some examples of archaic use of literal SQL. It is shown only as
2741 reference for those who deal with legacy code. Each example has a much
2742 better, cleaner and safer alternative that users should opt for in new code.
2748 my %where = ( requestor => \'IS NOT NULL' )
2750 $stmt = "WHERE requestor IS NOT NULL"
2752 This used to be the way of generating NULL comparisons, before the handling
2753 of C<undef> got formalized. For new code please use the superior syntax as
2754 described in L</Tests for NULL values>.
2758 my %where = ( requestor => \'= submitter' )
2760 $stmt = "WHERE requestor = submitter"
2762 This used to be the only way to compare columns. Use the superior L</-ident>
2763 method for all new code. For example an identifier declared in such a way
2764 will be properly quoted if L</quote_char> is properly set, while the legacy
2765 form will remain as supplied.
2769 my %where = ( is_ready => \"", completed => { '>', '2012-12-21' } )
2771 $stmt = "WHERE completed > ? AND is_ready"
2772 @bind = ('2012-12-21')
2774 Using an empty string literal used to be the only way to express a boolean.
2775 For all new code please use the much more readable
2776 L<-bool|/Unary operators: bool> operator.
2782 These pages could go on for a while, since the nesting of the data
2783 structures this module can handle are pretty much unlimited (the
2784 module implements the C<WHERE> expansion as a recursive function
2785 internally). Your best bet is to "play around" with the module a
2786 little to see how the data structures behave, and choose the best
2787 format for your data based on that.
2789 And of course, all the values above will probably be replaced with
2790 variables gotten from forms or the command line. After all, if you
2791 knew everything ahead of time, you wouldn't have to worry about
2792 dynamically-generating SQL and could just hardwire it into your
2795 =head1 ORDER BY CLAUSES
2797 Some functions take an order by clause. This can either be a scalar (just a
2798 column name), a hashref of C<< { -desc => 'col' } >> or C<< { -asc => 'col' }
2799 >>, a scalarref, an arrayref-ref, or an arrayref of any of the previous
2802 Given | Will Generate
2803 ---------------------------------------------------------------
2805 'colA' | ORDER BY colA
2807 [qw/colA colB/] | ORDER BY colA, colB
2809 {-asc => 'colA'} | ORDER BY colA ASC
2811 {-desc => 'colB'} | ORDER BY colB DESC
2813 ['colA', {-asc => 'colB'}] | ORDER BY colA, colB ASC
2815 { -asc => [qw/colA colB/] } | ORDER BY colA ASC, colB ASC
2817 \'colA DESC' | ORDER BY colA DESC
2819 \[ 'FUNC(colA, ?)', $x ] | ORDER BY FUNC(colA, ?)
2820 | /* ...with $x bound to ? */
2823 { -asc => 'colA' }, | colA ASC,
2824 { -desc => [qw/colB/] }, | colB DESC,
2825 { -asc => [qw/colC colD/] },| colC ASC, colD ASC,
2826 \'colE DESC', | colE DESC,
2827 \[ 'FUNC(colF, ?)', $x ], | FUNC(colF, ?)
2828 ] | /* ...with $x bound to ? */
2829 ===============================================================
2833 =head1 SPECIAL OPERATORS
2835 my $sqlmaker = SQL::Abstract->new(special_ops => [
2839 my ($self, $field, $op, $arg) = @_;
2845 handler => 'method_name',
2849 A "special operator" is a SQL syntactic clause that can be
2850 applied to a field, instead of a usual binary operator.
2853 WHERE field IN (?, ?, ?)
2854 WHERE field BETWEEN ? AND ?
2855 WHERE MATCH(field) AGAINST (?, ?)
2857 Special operators IN and BETWEEN are fairly standard and therefore
2858 are builtin within C<SQL::Abstract> (as the overridable methods
2859 C<_where_field_IN> and C<_where_field_BETWEEN>). For other operators,
2860 like the MATCH .. AGAINST example above which is specific to MySQL,
2861 you can write your own operator handlers - supply a C<special_ops>
2862 argument to the C<new> method. That argument takes an arrayref of
2863 operator definitions; each operator definition is a hashref with two
2870 the regular expression to match the operator
2874 Either a coderef or a plain scalar method name. In both cases
2875 the expected return is C<< ($sql, @bind) >>.
2877 When supplied with a method name, it is simply called on the
2878 L<SQL::Abstract> object as:
2880 $self->$method_name($field, $op, $arg)
2884 $field is the LHS of the operator
2885 $op is the part that matched the handler regex
2888 When supplied with a coderef, it is called as:
2890 $coderef->($self, $field, $op, $arg)
2895 For example, here is an implementation
2896 of the MATCH .. AGAINST syntax for MySQL
2898 my $sqlmaker = SQL::Abstract->new(special_ops => [
2900 # special op for MySql MATCH (field) AGAINST(word1, word2, ...)
2901 {regex => qr/^match$/i,
2903 my ($self, $field, $op, $arg) = @_;
2904 $arg = [$arg] if not ref $arg;
2905 my $label = $self->_quote($field);
2906 my ($placeholder) = $self->_convert('?');
2907 my $placeholders = join ", ", (($placeholder) x @$arg);
2908 my $sql = $self->_sqlcase('match') . " ($label) "
2909 . $self->_sqlcase('against') . " ($placeholders) ";
2910 my @bind = $self->_bindtype($field, @$arg);
2911 return ($sql, @bind);
2918 =head1 UNARY OPERATORS
2920 my $sqlmaker = SQL::Abstract->new(unary_ops => [
2924 my ($self, $op, $arg) = @_;
2930 handler => 'method_name',
2934 A "unary operator" is a SQL syntactic clause that can be
2935 applied to a field - the operator goes before the field
2937 You can write your own operator handlers - supply a C<unary_ops>
2938 argument to the C<new> method. That argument takes an arrayref of
2939 operator definitions; each operator definition is a hashref with two
2946 the regular expression to match the operator
2950 Either a coderef or a plain scalar method name. In both cases
2951 the expected return is C<< $sql >>.
2953 When supplied with a method name, it is simply called on the
2954 L<SQL::Abstract> object as:
2956 $self->$method_name($op, $arg)
2960 $op is the part that matched the handler regex
2961 $arg is the RHS or argument of the operator
2963 When supplied with a coderef, it is called as:
2965 $coderef->($self, $op, $arg)
2973 Thanks to some benchmarking by Mark Stosberg, it turns out that
2974 this module is many orders of magnitude faster than using C<DBIx::Abstract>.
2975 I must admit this wasn't an intentional design issue, but it's a
2976 byproduct of the fact that you get to control your C<DBI> handles
2979 To maximize performance, use a code snippet like the following:
2981 # prepare a statement handle using the first row
2982 # and then reuse it for the rest of the rows
2984 for my $href (@array_of_hashrefs) {
2985 $stmt ||= $sql->insert('table', $href);
2986 $sth ||= $dbh->prepare($stmt);
2987 $sth->execute($sql->values($href));
2990 The reason this works is because the keys in your C<$href> are sorted
2991 internally by B<SQL::Abstract>. Thus, as long as your data retains
2992 the same structure, you only have to generate the SQL the first time
2993 around. On subsequent queries, simply use the C<values> function provided
2994 by this module to return your values in the correct order.
2996 However this depends on the values having the same type - if, for
2997 example, the values of a where clause may either have values
2998 (resulting in sql of the form C<column = ?> with a single bind
2999 value), or alternatively the values might be C<undef> (resulting in
3000 sql of the form C<column IS NULL> with no bind value) then the
3001 caching technique suggested will not work.
3005 If you use my C<CGI::FormBuilder> module at all, you'll hopefully
3006 really like this part (I do, at least). Building up a complex query
3007 can be as simple as the following:
3014 use CGI::FormBuilder;
3017 my $form = CGI::FormBuilder->new(...);
3018 my $sql = SQL::Abstract->new;
3020 if ($form->submitted) {
3021 my $field = $form->field;
3022 my $id = delete $field->{id};
3023 my($stmt, @bind) = $sql->update('table', $field, {id => $id});
3026 Of course, you would still have to connect using C<DBI> to run the
3027 query, but the point is that if you make your form look like your
3028 table, the actual query script can be extremely simplistic.
3030 If you're B<REALLY> lazy (I am), check out C<HTML::QuickTable> for
3031 a fast interface to returning and formatting data. I frequently
3032 use these three modules together to write complex database query
3033 apps in under 50 lines.
3035 =head1 HOW TO CONTRIBUTE
3037 Contributions are always welcome, in all usable forms (we especially
3038 welcome documentation improvements). The delivery methods include git-
3039 or unified-diff formatted patches, GitHub pull requests, or plain bug
3040 reports either via RT or the Mailing list. Contributors are generally
3041 granted full access to the official repository after their first several
3042 patches pass successful review.
3044 This project is maintained in a git repository. The code and related tools are
3045 accessible at the following locations:
3049 =item * Official repo: L<git://git.shadowcat.co.uk/dbsrgits/SQL-Abstract.git>
3051 =item * Official gitweb: L<http://git.shadowcat.co.uk/gitweb/gitweb.cgi?p=dbsrgits/SQL-Abstract.git>
3053 =item * GitHub mirror: L<https://github.com/dbsrgits/sql-abstract>
3055 =item * Authorized committers: L<ssh://dbsrgits@git.shadowcat.co.uk/SQL-Abstract.git>
3061 Version 1.50 was a major internal refactoring of C<SQL::Abstract>.
3062 Great care has been taken to preserve the I<published> behavior
3063 documented in previous versions in the 1.* family; however,
3064 some features that were previously undocumented, or behaved
3065 differently from the documentation, had to be changed in order
3066 to clarify the semantics. Hence, client code that was relying
3067 on some dark areas of C<SQL::Abstract> v1.*
3068 B<might behave differently> in v1.50.
3070 The main changes are:
3076 support for literal SQL through the C<< \ [ $sql, @bind ] >> syntax.
3080 support for the { operator => \"..." } construct (to embed literal SQL)
3084 support for the { operator => \["...", @bind] } construct (to embed literal SQL with bind values)
3088 optional support for L<array datatypes|/"Inserting and Updating Arrays">
3092 defensive programming: check arguments
3096 fixed bug with global logic, which was previously implemented
3097 through global variables yielding side-effects. Prior versions would
3098 interpret C<< [ {cond1, cond2}, [cond3, cond4] ] >>
3099 as C<< "(cond1 AND cond2) OR (cond3 AND cond4)" >>.
3100 Now this is interpreted
3101 as C<< "(cond1 AND cond2) OR (cond3 OR cond4)" >>.
3106 fixed semantics of _bindtype on array args
3110 dropped the C<_anoncopy> of the %where tree. No longer necessary,
3111 we just avoid shifting arrays within that tree.
3115 dropped the C<_modlogic> function
3119 =head1 ACKNOWLEDGEMENTS
3121 There are a number of individuals that have really helped out with
3122 this module. Unfortunately, most of them submitted bugs via CPAN
3123 so I have no idea who they are! But the people I do know are:
3125 Ash Berlin (order_by hash term support)
3126 Matt Trout (DBIx::Class support)
3127 Mark Stosberg (benchmarking)
3128 Chas Owens (initial "IN" operator support)
3129 Philip Collins (per-field SQL functions)
3130 Eric Kolve (hashref "AND" support)
3131 Mike Fragassi (enhancements to "BETWEEN" and "LIKE")
3132 Dan Kubb (support for "quote_char" and "name_sep")
3133 Guillermo Roditi (patch to cleanup "IN" and "BETWEEN", fix and tests for _order_by)
3134 Laurent Dami (internal refactoring, extensible list of special operators, literal SQL)
3135 Norbert Buchmuller (support for literal SQL in hashpair, misc. fixes & tests)
3136 Peter Rabbitson (rewrite of SQLA::Test, misc. fixes & tests)
3137 Oliver Charles (support for "RETURNING" after "INSERT")
3143 L<DBIx::Class>, L<DBIx::Abstract>, L<CGI::FormBuilder>, L<HTML::QuickTable>.
3147 Copyright (c) 2001-2007 Nathan Wiger <nwiger@cpan.org>. All Rights Reserved.
3149 This module is actively maintained by Matt Trout <mst@shadowcatsystems.co.uk>
3151 For support, your best bet is to try the C<DBIx::Class> users mailing list.
3152 While not an official support venue, C<DBIx::Class> makes heavy use of
3153 C<SQL::Abstract>, and as such list members there are very familiar with
3154 how to create queries.
3158 This module is free software; you may copy this under the same
3159 terms as perl itself (either the GNU General Public License or
3160 the Artistic License)