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 $k =~ /^-/;
610 if (my ($rest) = $k =~/^-not[_ ](.*)$/) {
613 $self->_expand_expr({ "-${rest}", $v })
618 $op =~ s/^-// if length($op) > 1;
620 # top level special ops are illegal in general
621 # note that, arguably, if it makes no sense at top level, it also
622 # makes no sense on the other side of an = sign or similar but DBIC
623 # gets disappointingly upset if I disallow it
625 (our $Expand_Depth) == 1
626 and List::Util::first { $op =~ $_->{regex} } @{$self->{special_ops}}
628 puke "Illegal use of top-level '-$op'"
630 if (my $us = List::Util::first { $op =~ $_->{regex} } @{$self->{unary_ops}}) {
631 return { -op => [ $op, $v ] };
634 if ($self->{render}{$k}) {
640 and (keys %$v)[0] =~ /^-/
642 my ($func) = $k =~ /^-(.*)$/;
643 if (List::Util::first { $func =~ $_->{regex} } @{$self->{special_ops}}) {
644 return +{ -op => [ $func, $self->_expand_expr($v) ] };
646 return +{ -func => [ $func, $self->_expand_expr($v) ] };
648 if (!ref($v) or is_literal_value($v)) {
649 return +{ -op => [ $k =~ /^-(.*)$/, $self->_expand_expr($v) ] };
656 and exists $v->{-value}
657 and not defined $v->{-value}
660 return $self->_expand_expr({ $k => { $self->{cmp} => undef } });
662 if (!ref($v) or Scalar::Util::blessed($v)) {
663 my $d = our $Default_Scalar_To;
667 $self->_expand_ident(-ident => $k),
669 ? $self->_expand_expr($d => $v)
670 : { -bind => [ $k, $v ] }
675 if (ref($v) eq 'HASH') {
677 return $self->_expand_op_andor(-and => [
678 map +{ $k => { $_ => $v->{$_} } },
682 return undef unless keys %$v;
684 my $op = join ' ', split '_', (map lc, $vk =~ /^-?(.*)$/)[0];
685 $self->_assert_pass_injection_guard($op);
686 if ($op =~ s/ [_\s]? \d+ $//x ) {
687 belch 'Use of [and|or|nest]_N modifiers is deprecated and will be removed in SQLA v2.0. '
688 . "You probably wanted ...-and => [ -$op => COND1, -$op => COND2 ... ]";
690 if (my $x = $self->{expand_op}{$op}) {
691 local our $Cur_Col_Meta = $k;
692 return $self->$x($op, $vv, $k);
694 if ($op eq 'value' and not defined($vv)) {
695 return $self->_expand_expr({ $k, undef }) unless defined($vv);
697 if ($op =~ /^is(?: not)?$/) {
698 puke "$op can only take undef as argument"
702 and exists($vv->{-value})
703 and !defined($vv->{-value})
705 return +{ -op => [ $op.' null', $self->_expand_ident(-ident => $k) ] };
707 if (my $us = List::Util::first { $op =~ $_->{regex} } @{$self->{special_ops}}) {
708 return { -op => [ $op, $self->_expand_ident(-ident => $k), $vv ] };
710 if (my $us = List::Util::first { $op =~ $_->{regex} } @{$self->{unary_ops}}) {
713 $self->_expand_ident(-ident => $k),
714 { -op => [ $op, $vv ] }
717 if (ref($vv) eq 'ARRAY') {
718 my ($logic, @values) = (
719 (defined($vv->[0]) and $vv->[0] =~ /^-(and|or)$/i)
724 $op =~ $self->{inequality_op}
725 or $op =~ $self->{not_like_op}
727 if (lc($logic) eq '-or' and @values > 1) {
728 belch "A multi-element arrayref as an argument to the inequality op '${\uc($op)}' "
729 . 'is technically equivalent to an always-true 1=1 (you probably wanted '
730 . "to say ...{ \$inequality_op => [ -and => \@values ] }... instead)"
735 # try to DWIM on equality operators
737 $op =~ $self->{equality_op} ? $self->sqlfalse
738 : $op =~ $self->{like_op} ? belch("Supplying an empty arrayref to '@{[ uc $op]}' is deprecated") && $self->sqlfalse
739 : $op =~ $self->{inequality_op} ? $self->sqltrue
740 : $op =~ $self->{not_like_op} ? belch("Supplying an empty arrayref to '@{[ uc $op]}' is deprecated") && $self->sqltrue
741 : puke "operator '$op' applied on an empty array (field '$k')";
743 return $self->_expand_op_andor($logic => [
744 map +{ $k => { $vk => $_ } },
752 and exists $vv->{-value}
753 and not defined $vv->{-value}
757 $op =~ /^not$/i ? 'is not' # legacy
758 : $op =~ $self->{equality_op} ? 'is'
759 : $op =~ $self->{like_op} ? belch("Supplying an undefined argument to '@{[ uc $op]}' is deprecated") && 'is'
760 : $op =~ $self->{inequality_op} ? 'is not'
761 : $op =~ $self->{not_like_op} ? belch("Supplying an undefined argument to '@{[ uc $op]}' is deprecated") && 'is not'
762 : puke "unexpected operator '$op' with undef operand";
763 return +{ -op => [ $is.' null', $self->_expand_ident(-ident => $k) ] };
765 local our $Cur_Col_Meta = $k;
768 $self->_expand_ident(-ident => $k),
769 $self->_expand_expr($vv)
772 if (ref($v) eq 'ARRAY') {
773 return $self->sqlfalse unless @$v;
774 $self->_debug("ARRAY($k) means distribute over elements");
776 $v->[0] =~ /^-(and|or)$/i
777 ? shift(@{$v = [ @$v ]})
778 : '-'.lc($self->{logic} || 'OR')
780 return $self->_expand_op_andor(
781 $logic => [ map +{ $k => $_ }, @$v ]
784 if (my $literal = is_literal_value($v)) {
786 belch 'Hash-pairs consisting of an empty string with a literal are deprecated, and will be removed in 2.0: use -and => [ $literal ] instead';
789 my ($sql, @bind) = @$literal;
790 if ($self->{bindtype} eq 'columns') {
792 $self->_assert_bindval_matches_bindtype($_);
795 return +{ -literal => [ $self->_quote($k).' '.$sql, @bind ] };
801 my ($self, $op, $body) = @_;
802 unless (defined($body) or (ref($body) and ref($body) eq 'ARRAY')) {
803 puke "$op requires a single plain scalar argument (a quotable identifier) or an arrayref of identifier parts";
805 my @parts = map split(/\Q${\($self->{name_sep}||'.')}\E/, $_),
806 ref($body) ? @$body : $body;
807 return { -ident => $parts[-1] } if $self->{_dequalify_idents};
808 unless ($self->{quote_char}) {
809 $self->_assert_pass_injection_guard($_) for @parts;
811 return +{ -ident => \@parts };
815 +{ -bind => [ our $Cur_Col_Meta, $_[2] ] };
819 +{ -op => [ 'not', $_[0]->_expand_expr($_[2]) ] };
823 my ($self, undef, $v) = @_;
825 return $self->_expand_expr($v);
827 puke "-bool => undef not supported" unless defined($v);
828 return $self->_expand_ident(-ident => $v);
831 sub _expand_op_andor {
832 my ($self, $logic, $v, $k) = @_;
834 $v = [ map +{ $k, { $_ => $v->{$_} } },
837 my ($logop) = $logic =~ /^-?(.*)$/;
838 if (ref($v) eq 'HASH') {
841 map $self->_expand_expr({ $_ => $v->{$_} }),
845 if (ref($v) eq 'ARRAY') {
846 $logop eq 'and' or $logop eq 'or' or puke "unknown logic: $logop";
849 (ref($_) eq 'ARRAY' and @$_)
850 or (ref($_) eq 'HASH' and %$_)
856 while (my ($el) = splice @expr, 0, 1) {
857 puke "Supplying an empty left hand side argument is not supported in array-pairs"
858 unless defined($el) and length($el);
859 my $elref = ref($el);
861 local our $Expand_Depth = 0;
862 push(@res, grep defined, $self->_expand_expr({ $el, shift(@expr) }));
863 } elsif ($elref eq 'ARRAY') {
864 push(@res, grep defined, $self->_expand_expr($el)) if @$el;
865 } elsif (my $l = is_literal_value($el)) {
866 push @res, { -literal => $l };
867 } elsif ($elref eq 'HASH') {
868 local our $Expand_Depth = 0;
869 push @res, grep defined, $self->_expand_expr($el) if %$el;
875 # return $res[0] if @res == 1;
876 return { -op => [ $logop, @res ] };
881 sub _expand_between {
882 my ($self, $op, $vv, $k) = @_;
883 local our $Cur_Col_Meta = $k;
884 my @rhs = map $self->_expand_expr($_),
885 ref($vv) eq 'ARRAY' ? @$vv : $vv;
887 (@rhs == 1 and ref($rhs[0]) eq 'HASH' and $rhs[0]->{-literal})
889 (@rhs == 2 and defined($rhs[0]) and defined($rhs[1]))
891 puke "Operator '${\uc($op)}' requires either an arrayref with two defined values or expressions, or a single literal scalarref/arrayref-ref";
895 $self->_expand_ident(-ident => $k),
901 my ($self, $op, $vv, $k) = @_;
902 if (my $literal = is_literal_value($vv)) {
903 my ($sql, @bind) = @$literal;
904 my $opened_sql = $self->_open_outer_paren($sql);
906 $op, $self->_expand_ident(-ident => $k),
907 [ { -literal => [ $opened_sql, @bind ] } ]
911 'SQL::Abstract before v1.75 used to generate incorrect SQL when the '
912 . "-${\uc($op)} operator was given an undef-containing list: !!!AUDIT YOUR CODE "
913 . 'AND DATA!!! (the upcoming Data::Query-based version of SQL::Abstract '
914 . 'will emit the logically correct SQL instead of raising this exception)'
916 puke("Argument passed to the '${\uc($op)}' operator can not be undefined")
918 my @rhs = map $self->_expand_expr($_),
919 map { ref($_) ? $_ : { -bind => [ $k, $_ ] } }
920 map { defined($_) ? $_: puke($undef_err) }
921 (ref($vv) eq 'ARRAY' ? @$vv : $vv);
922 return $self->${\($op =~ /^not/ ? 'sqltrue' : 'sqlfalse')} unless @rhs;
926 $self->_expand_ident(-ident => $k),
932 my ($self, $op, $v) = @_;
933 # DBIx::Class requires a nest warning to be emitted once but the private
934 # method it overrode to do so no longer exists
935 if ($self->{is_dbic_sqlmaker}) {
936 unless (our $Nest_Warned) {
938 "-nest in search conditions is deprecated, you most probably wanted:\n"
939 .q|{..., -and => [ \%cond0, \@cond1, \'cond2', \[ 'cond3', [ col => bind ] ], etc. ], ... }|
944 return $self->_expand_expr($v);
948 my ($self, $where, $logic) = @_;
950 # Special case: top level simple string treated as literal
952 my $where_exp = (ref($where)
953 ? $self->_expand_expr($where, $logic)
954 : { -literal => [ $where ] });
956 # dispatch expanded expression
958 my ($sql, @bind) = defined($where_exp) ? $self->render_aqt($where_exp) : (undef);
959 # DBIx::Class used to call _recurse_where in scalar context
960 # something else might too...
962 return ($sql, @bind);
965 belch "Calling _recurse_where in scalar context is deprecated and will go away before 2.0";
971 my ($self, $ident) = @_;
973 return $self->_convert($self->_quote($ident));
977 my ($self, $list) = @_;
978 my @parts = grep length($_->[0]), map [ $self->render_aqt($_) ], @$list;
979 return join(', ', map $_->[0], @parts), map @{$_}[1..$#$_], @parts;
983 my ($self, $rest) = @_;
984 my ($func, @args) = @$rest;
988 push @arg_sql, shift @x;
990 } map [ $self->render_aqt($_) ], @args;
991 return ($self->_sqlcase($func).'('.join(', ', @arg_sql).')', @bind);
995 my ($self, $bind) = @_;
996 return ($self->_convert('?'), $self->_bindtype(@$bind));
999 sub _render_literal {
1000 my ($self, $literal) = @_;
1001 $self->_assert_bindval_matches_bindtype(@{$literal}[1..$#$literal]);
1007 my ($self, $op, $args) = @_;
1008 my ($left, $low, $high) = @$args;
1009 my ($rhsql, @rhbind) = do {
1011 puke "Single arg to between must be a literal"
1012 unless $low->{-literal};
1015 my ($l, $h) = map [ $self->render_aqt($_) ], $low, $high;
1016 (join(' ', $l->[0], $self->_sqlcase('and'), $h->[0]),
1017 @{$l}[1..$#$l], @{$h}[1..$#$h])
1020 my ($lhsql, @lhbind) = $self->render_aqt($left);
1022 join(' ', '(', $lhsql, $self->_sqlcase($op), $rhsql, ')'),
1025 }), 'between', 'not between'),
1027 my ($self, $op, $args) = @_;
1028 my ($lhs, $rhs) = @$args;
1031 my ($sql, @bind) = $self->render_aqt($_);
1032 push @in_bind, @bind;
1035 my ($lhsql, @lbind) = $self->render_aqt($lhs);
1037 $lhsql.' '.$self->_sqlcase($op).' ( '
1038 .join(', ', @in_sql)
1042 }), 'in', 'not in'),
1043 (map +($_ => '_render_unop_postfix'),
1044 'is null', 'is not null', 'asc', 'desc',
1046 (not => '_render_op_not'),
1048 my ($self, $op, $args) = @_;
1049 my @parts = grep length($_->[0]), map [ $self->render_aqt($_) ], @$args;
1050 return '' unless @parts;
1051 return @{$parts[0]} if @parts == 1;
1052 my ($final_sql) = join(
1053 ' '.$self->_sqlcase($op).' ',
1058 map @{$_}[1..$#$_], @parts
1064 my ($self, $v) = @_;
1065 my ($op, @args) = @$v;
1066 if (my $r = $self->{render_op}{$op}) {
1067 return $self->$r($op, \@args);
1069 my $us = List::Util::first { $op =~ $_->{regex} } @{$self->{special_ops}};
1070 if ($us and @args > 1) {
1071 puke "Special op '${op}' requires first value to be identifier"
1072 unless my ($ident) = map $_->{-ident}, grep ref($_) eq 'HASH', $args[0];
1073 my $k = join(($self->{name_sep}||'.'), @$ident);
1074 local our $Expand_Depth = 1;
1075 return $self->${\($us->{handler})}($k, $op, $args[1]);
1077 if (my $us = List::Util::first { $op =~ $_->{regex} } @{$self->{unary_ops}}) {
1078 return $self->${\($us->{handler})}($op, $args[0]);
1081 return $self->_render_unop_prefix($op, \@args);
1083 my @parts = grep length($_->[0]), map [ $self->render_aqt($_) ], @args;
1084 return '' unless @parts;
1085 my ($final_sql) = join(
1086 ' '.$self->_sqlcase($op).' ',
1091 map @{$_}[1..$#$_], @parts
1097 sub _render_op_not {
1098 my ($self, $op, $v) = @_;
1099 my ($sql, @bind) = $self->_render_unop_prefix($op, $v);
1100 return "(${sql})", @bind;
1103 sub _render_unop_prefix {
1104 my ($self, $op, $v) = @_;
1105 my ($expr_sql, @bind) = $self->render_aqt($v->[0]);
1106 my $op_sql = $self->_sqlcase($op);
1107 return ("${op_sql} ${expr_sql}", @bind);
1110 sub _render_unop_postfix {
1111 my ($self, $op, $v) = @_;
1112 my ($expr_sql, @bind) = $self->render_aqt($v->[0]);
1113 my $op_sql = $self->_sqlcase($op);
1114 return ($expr_sql.' '.$op_sql, @bind);
1117 # Some databases (SQLite) treat col IN (1, 2) different from
1118 # col IN ( (1, 2) ). Use this to strip all outer parens while
1119 # adding them back in the corresponding method
1120 sub _open_outer_paren {
1121 my ($self, $sql) = @_;
1123 while (my ($inner) = $sql =~ /^ \s* \( (.*) \) \s* $/xs) {
1125 # there are closing parens inside, need the heavy duty machinery
1126 # to reevaluate the extraction starting from $sql (full reevaluation)
1127 if ($inner =~ /\)/) {
1128 require Text::Balanced;
1130 my (undef, $remainder) = do {
1131 # idiotic design - writes to $@ but *DOES NOT* throw exceptions
1133 Text::Balanced::extract_bracketed($sql, '()', qr/\s*/);
1136 # the entire expression needs to be a balanced bracketed thing
1137 # (after an extract no remainder sans trailing space)
1138 last if defined $remainder and $remainder =~ /\S/;
1148 #======================================================================
1150 #======================================================================
1152 sub _expand_order_by {
1153 my ($self, $arg) = @_;
1155 return unless defined($arg) and not (ref($arg) eq 'ARRAY' and !@$arg);
1157 my $expander = sub {
1158 my ($self, $dir, $expr) = @_;
1159 my @to_expand = ref($expr) eq 'ARRAY' ? @$expr : $expr;
1160 foreach my $arg (@to_expand) {
1164 and grep /^-(asc|desc)$/, keys %$arg
1166 puke "ordering direction hash passed to order by must have exactly one key (-asc or -desc)";
1170 defined($dir) ? { -op => [ $dir =~ /^-?(.*)$/ ,=> $_ ] } : $_
1172 map $self->expand_expr($_, -ident),
1173 map ref($_) eq 'ARRAY' ? @$_ : $_, @to_expand;
1174 return (@exp > 1 ? { -list => \@exp } : $exp[0]);
1177 local @{$self->{expand}}{qw(-asc -desc)} = (($expander) x 2);
1179 return $self->$expander(undef, $arg);
1183 my ($self, $arg) = @_;
1185 return '' unless defined(my $expanded = $self->_expand_order_by($arg));
1187 my ($sql, @bind) = $self->render_aqt($expanded);
1189 return '' unless length($sql);
1191 my $final_sql = $self->_sqlcase(' order by ').$sql;
1193 return wantarray ? ($final_sql, @bind) : $final_sql;
1196 # _order_by no longer needs to call this so doesn't but DBIC uses it.
1198 sub _order_by_chunks {
1199 my ($self, $arg) = @_;
1201 return () unless defined(my $expanded = $self->_expand_order_by($arg));
1203 return $self->_chunkify_order_by($expanded);
1206 sub _chunkify_order_by {
1207 my ($self, $expanded) = @_;
1209 return grep length, $self->render_aqt($expanded)
1210 if $expanded->{-ident} or @{$expanded->{-literal}||[]} == 1;
1213 if (ref() eq 'HASH' and my $l = $_->{-list}) {
1214 return map $self->_chunkify_order_by($_), @$l;
1216 return [ $self->render_aqt($_) ];
1220 #======================================================================
1221 # DATASOURCE (FOR NOW, JUST PLAIN TABLE OR LIST OF TABLES)
1222 #======================================================================
1228 $self->_expand_maybe_list_expr($from, undef, -ident)
1233 #======================================================================
1235 #======================================================================
1237 sub _expand_maybe_list_expr {
1238 my ($self, $expr, $logic, $default) = @_;
1240 if (ref($expr) eq 'ARRAY') {
1242 map $self->expand_expr($_, $default), @$expr
1249 return $self->expand_expr($e, $default);
1252 # highly optimized, as it's called way too often
1254 # my ($self, $label) = @_;
1256 return '' unless defined $_[1];
1257 return ${$_[1]} if ref($_[1]) eq 'SCALAR';
1258 puke 'Identifier cannot be hashref' if ref($_[1]) eq 'HASH';
1260 unless ($_[0]->{quote_char}) {
1261 if (ref($_[1]) eq 'ARRAY') {
1262 return join($_[0]->{name_sep}||'.', @{$_[1]});
1264 $_[0]->_assert_pass_injection_guard($_[1]);
1269 my $qref = ref $_[0]->{quote_char};
1271 !$qref ? ($_[0]->{quote_char}, $_[0]->{quote_char})
1272 : ($qref eq 'ARRAY') ? @{$_[0]->{quote_char}}
1273 : puke "Unsupported quote_char format: $_[0]->{quote_char}";
1275 my $esc = $_[0]->{escape_char} || $r;
1277 # parts containing * are naturally unquoted
1279 $_[0]->{name_sep}||'',
1283 : do { (my $n = $_) =~ s/(\Q$esc\E|\Q$r\E)/$esc$1/g; $l . $n . $r }
1285 (ref($_[1]) eq 'ARRAY'
1289 ? split (/\Q$_[0]->{name_sep}\E/, $_[1] )
1297 # Conversion, if applicable
1299 #my ($self, $arg) = @_;
1300 if ($_[0]->{convert_where}) {
1301 return $_[0]->_sqlcase($_[0]->{convert_where}) .'(' . $_[1] . ')';
1308 #my ($self, $col, @vals) = @_;
1309 # called often - tighten code
1310 return $_[0]->{bindtype} eq 'columns'
1311 ? map {[$_[1], $_]} @_[2 .. $#_]
1316 # Dies if any element of @bind is not in [colname => value] format
1317 # if bindtype is 'columns'.
1318 sub _assert_bindval_matches_bindtype {
1319 # my ($self, @bind) = @_;
1321 if ($self->{bindtype} eq 'columns') {
1323 if (!defined $_ || ref($_) ne 'ARRAY' || @$_ != 2) {
1324 puke "bindtype 'columns' selected, you need to pass: [column_name => bind_value]"
1330 sub _join_sql_clauses {
1331 my ($self, $logic, $clauses_aref, $bind_aref) = @_;
1333 if (@$clauses_aref > 1) {
1334 my $join = " " . $self->_sqlcase($logic) . " ";
1335 my $sql = '( ' . join($join, @$clauses_aref) . ' )';
1336 return ($sql, @$bind_aref);
1338 elsif (@$clauses_aref) {
1339 return ($clauses_aref->[0], @$bind_aref); # no parentheses
1342 return (); # if no SQL, ignore @$bind_aref
1347 # Fix SQL case, if so requested
1349 # LDNOTE: if $self->{case} is true, then it contains 'lower', so we
1350 # don't touch the argument ... crooked logic, but let's not change it!
1351 return $_[0]->{case} ? $_[1] : uc($_[1]);
1355 #======================================================================
1356 # DISPATCHING FROM REFKIND
1357 #======================================================================
1360 my ($self, $data) = @_;
1362 return 'UNDEF' unless defined $data;
1364 # blessed objects are treated like scalars
1365 my $ref = (Scalar::Util::blessed $data) ? '' : ref $data;
1367 return 'SCALAR' unless $ref;
1370 while ($ref eq 'REF') {
1372 $ref = (Scalar::Util::blessed $data) ? '' : ref $data;
1376 return ($ref||'SCALAR') . ('REF' x $n_steps);
1380 my ($self, $data) = @_;
1381 my @try = ($self->_refkind($data));
1382 push @try, 'SCALAR_or_UNDEF' if $try[0] eq 'SCALAR' || $try[0] eq 'UNDEF';
1383 push @try, 'FALLBACK';
1387 sub _METHOD_FOR_refkind {
1388 my ($self, $meth_prefix, $data) = @_;
1391 for (@{$self->_try_refkind($data)}) {
1392 $method = $self->can($meth_prefix."_".$_)
1396 return $method || puke "cannot dispatch on '$meth_prefix' for ".$self->_refkind($data);
1400 sub _SWITCH_refkind {
1401 my ($self, $data, $dispatch_table) = @_;
1404 for (@{$self->_try_refkind($data)}) {
1405 $coderef = $dispatch_table->{$_}
1409 puke "no dispatch entry for ".$self->_refkind($data)
1418 #======================================================================
1419 # VALUES, GENERATE, AUTOLOAD
1420 #======================================================================
1422 # LDNOTE: original code from nwiger, didn't touch code in that section
1423 # I feel the AUTOLOAD stuff should not be the default, it should
1424 # only be activated on explicit demand by user.
1428 my $data = shift || return;
1429 puke "Argument to ", __PACKAGE__, "->values must be a \\%hash"
1430 unless ref $data eq 'HASH';
1433 foreach my $k (sort keys %$data) {
1434 my $v = $data->{$k};
1435 $self->_SWITCH_refkind($v, {
1437 if ($self->{array_datatypes}) { # array datatype
1438 push @all_bind, $self->_bindtype($k, $v);
1440 else { # literal SQL with bind
1441 my ($sql, @bind) = @$v;
1442 $self->_assert_bindval_matches_bindtype(@bind);
1443 push @all_bind, @bind;
1446 ARRAYREFREF => sub { # literal SQL with bind
1447 my ($sql, @bind) = @${$v};
1448 $self->_assert_bindval_matches_bindtype(@bind);
1449 push @all_bind, @bind;
1451 SCALARREF => sub { # literal SQL without bind
1453 SCALAR_or_UNDEF => sub {
1454 push @all_bind, $self->_bindtype($k, $v);
1465 my(@sql, @sqlq, @sqlv);
1469 if ($ref eq 'HASH') {
1470 for my $k (sort keys %$_) {
1473 my $label = $self->_quote($k);
1474 if ($r eq 'ARRAY') {
1475 # literal SQL with bind
1476 my ($sql, @bind) = @$v;
1477 $self->_assert_bindval_matches_bindtype(@bind);
1478 push @sqlq, "$label = $sql";
1480 } elsif ($r eq 'SCALAR') {
1481 # literal SQL without bind
1482 push @sqlq, "$label = $$v";
1484 push @sqlq, "$label = ?";
1485 push @sqlv, $self->_bindtype($k, $v);
1488 push @sql, $self->_sqlcase('set'), join ', ', @sqlq;
1489 } elsif ($ref eq 'ARRAY') {
1490 # unlike insert(), assume these are ONLY the column names, i.e. for SQL
1493 if ($r eq 'ARRAY') { # literal SQL with bind
1494 my ($sql, @bind) = @$v;
1495 $self->_assert_bindval_matches_bindtype(@bind);
1498 } elsif ($r eq 'SCALAR') { # literal SQL without bind
1499 # embedded literal SQL
1506 push @sql, '(' . join(', ', @sqlq) . ')';
1507 } elsif ($ref eq 'SCALAR') {
1511 # strings get case twiddled
1512 push @sql, $self->_sqlcase($_);
1516 my $sql = join ' ', @sql;
1518 # this is pretty tricky
1519 # if ask for an array, return ($stmt, @bind)
1520 # otherwise, s/?/shift @sqlv/ to put it inline
1522 return ($sql, @sqlv);
1524 1 while $sql =~ s/\?/my $d = shift(@sqlv);
1525 ref $d ? $d->[1] : $d/e;
1534 # This allows us to check for a local, then _form, attr
1536 my($name) = $AUTOLOAD =~ /.*::(.+)/;
1537 return $self->generate($name, @_);
1548 SQL::Abstract - Generate SQL from Perl data structures
1554 my $sql = SQL::Abstract->new;
1556 my($stmt, @bind) = $sql->select($source, \@fields, \%where, $order);
1558 my($stmt, @bind) = $sql->insert($table, \%fieldvals || \@values);
1560 my($stmt, @bind) = $sql->update($table, \%fieldvals, \%where);
1562 my($stmt, @bind) = $sql->delete($table, \%where);
1564 # Then, use these in your DBI statements
1565 my $sth = $dbh->prepare($stmt);
1566 $sth->execute(@bind);
1568 # Just generate the WHERE clause
1569 my($stmt, @bind) = $sql->where(\%where, $order);
1571 # Return values in the same order, for hashed queries
1572 # See PERFORMANCE section for more details
1573 my @bind = $sql->values(\%fieldvals);
1577 This module was inspired by the excellent L<DBIx::Abstract>.
1578 However, in using that module I found that what I really wanted
1579 to do was generate SQL, but still retain complete control over my
1580 statement handles and use the DBI interface. So, I set out to
1581 create an abstract SQL generation module.
1583 While based on the concepts used by L<DBIx::Abstract>, there are
1584 several important differences, especially when it comes to WHERE
1585 clauses. I have modified the concepts used to make the SQL easier
1586 to generate from Perl data structures and, IMO, more intuitive.
1587 The underlying idea is for this module to do what you mean, based
1588 on the data structures you provide it. The big advantage is that
1589 you don't have to modify your code every time your data changes,
1590 as this module figures it out.
1592 To begin with, an SQL INSERT is as easy as just specifying a hash
1593 of C<key=value> pairs:
1596 name => 'Jimbo Bobson',
1597 phone => '123-456-7890',
1598 address => '42 Sister Lane',
1599 city => 'St. Louis',
1600 state => 'Louisiana',
1603 The SQL can then be generated with this:
1605 my($stmt, @bind) = $sql->insert('people', \%data);
1607 Which would give you something like this:
1609 $stmt = "INSERT INTO people
1610 (address, city, name, phone, state)
1611 VALUES (?, ?, ?, ?, ?)";
1612 @bind = ('42 Sister Lane', 'St. Louis', 'Jimbo Bobson',
1613 '123-456-7890', 'Louisiana');
1615 These are then used directly in your DBI code:
1617 my $sth = $dbh->prepare($stmt);
1618 $sth->execute(@bind);
1620 =head2 Inserting and Updating Arrays
1622 If your database has array types (like for example Postgres),
1623 activate the special option C<< array_datatypes => 1 >>
1624 when creating the C<SQL::Abstract> object.
1625 Then you may use an arrayref to insert and update database array types:
1627 my $sql = SQL::Abstract->new(array_datatypes => 1);
1629 planets => [qw/Mercury Venus Earth Mars/]
1632 my($stmt, @bind) = $sql->insert('solar_system', \%data);
1636 $stmt = "INSERT INTO solar_system (planets) VALUES (?)"
1638 @bind = (['Mercury', 'Venus', 'Earth', 'Mars']);
1641 =head2 Inserting and Updating SQL
1643 In order to apply SQL functions to elements of your C<%data> you may
1644 specify a reference to an arrayref for the given hash value. For example,
1645 if you need to execute the Oracle C<to_date> function on a value, you can
1646 say something like this:
1650 date_entered => \[ "to_date(?,'MM/DD/YYYY')", "03/02/2003" ],
1653 The first value in the array is the actual SQL. Any other values are
1654 optional and would be included in the bind values array. This gives
1657 my($stmt, @bind) = $sql->insert('people', \%data);
1659 $stmt = "INSERT INTO people (name, date_entered)
1660 VALUES (?, to_date(?,'MM/DD/YYYY'))";
1661 @bind = ('Bill', '03/02/2003');
1663 An UPDATE is just as easy, all you change is the name of the function:
1665 my($stmt, @bind) = $sql->update('people', \%data);
1667 Notice that your C<%data> isn't touched; the module will generate
1668 the appropriately quirky SQL for you automatically. Usually you'll
1669 want to specify a WHERE clause for your UPDATE, though, which is
1670 where handling C<%where> hashes comes in handy...
1672 =head2 Complex where statements
1674 This module can generate pretty complicated WHERE statements
1675 easily. For example, simple C<key=value> pairs are taken to mean
1676 equality, and if you want to see if a field is within a set
1677 of values, you can use an arrayref. Let's say we wanted to
1678 SELECT some data based on this criteria:
1681 requestor => 'inna',
1682 worker => ['nwiger', 'rcwe', 'sfz'],
1683 status => { '!=', 'completed' }
1686 my($stmt, @bind) = $sql->select('tickets', '*', \%where);
1688 The above would give you something like this:
1690 $stmt = "SELECT * FROM tickets WHERE
1691 ( requestor = ? ) AND ( status != ? )
1692 AND ( worker = ? OR worker = ? OR worker = ? )";
1693 @bind = ('inna', 'completed', 'nwiger', 'rcwe', 'sfz');
1695 Which you could then use in DBI code like so:
1697 my $sth = $dbh->prepare($stmt);
1698 $sth->execute(@bind);
1704 The methods are simple. There's one for every major SQL operation,
1705 and a constructor you use first. The arguments are specified in a
1706 similar order for each method (table, then fields, then a where
1707 clause) to try and simplify things.
1709 =head2 new(option => 'value')
1711 The C<new()> function takes a list of options and values, and returns
1712 a new B<SQL::Abstract> object which can then be used to generate SQL
1713 through the methods below. The options accepted are:
1719 If set to 'lower', then SQL will be generated in all lowercase. By
1720 default SQL is generated in "textbook" case meaning something like:
1722 SELECT a_field FROM a_table WHERE some_field LIKE '%someval%'
1724 Any setting other than 'lower' is ignored.
1728 This determines what the default comparison operator is. By default
1729 it is C<=>, meaning that a hash like this:
1731 %where = (name => 'nwiger', email => 'nate@wiger.org');
1733 Will generate SQL like this:
1735 WHERE name = 'nwiger' AND email = 'nate@wiger.org'
1737 However, you may want loose comparisons by default, so if you set
1738 C<cmp> to C<like> you would get SQL such as:
1740 WHERE name like 'nwiger' AND email like 'nate@wiger.org'
1742 You can also override the comparison on an individual basis - see
1743 the huge section on L</"WHERE CLAUSES"> at the bottom.
1745 =item sqltrue, sqlfalse
1747 Expressions for inserting boolean values within SQL statements.
1748 By default these are C<1=1> and C<1=0>. They are used
1749 by the special operators C<-in> and C<-not_in> for generating
1750 correct SQL even when the argument is an empty array (see below).
1754 This determines the default logical operator for multiple WHERE
1755 statements in arrays or hashes. If absent, the default logic is "or"
1756 for arrays, and "and" for hashes. This means that a WHERE
1760 event_date => {'>=', '2/13/99'},
1761 event_date => {'<=', '4/24/03'},
1764 will generate SQL like this:
1766 WHERE event_date >= '2/13/99' OR event_date <= '4/24/03'
1768 This is probably not what you want given this query, though (look
1769 at the dates). To change the "OR" to an "AND", simply specify:
1771 my $sql = SQL::Abstract->new(logic => 'and');
1773 Which will change the above C<WHERE> to:
1775 WHERE event_date >= '2/13/99' AND event_date <= '4/24/03'
1777 The logic can also be changed locally by inserting
1778 a modifier in front of an arrayref:
1780 @where = (-and => [event_date => {'>=', '2/13/99'},
1781 event_date => {'<=', '4/24/03'} ]);
1783 See the L</"WHERE CLAUSES"> section for explanations.
1787 This will automatically convert comparisons using the specified SQL
1788 function for both column and value. This is mostly used with an argument
1789 of C<upper> or C<lower>, so that the SQL will have the effect of
1790 case-insensitive "searches". For example, this:
1792 $sql = SQL::Abstract->new(convert => 'upper');
1793 %where = (keywords => 'MaKe iT CAse inSeNSItive');
1795 Will turn out the following SQL:
1797 WHERE upper(keywords) like upper('MaKe iT CAse inSeNSItive')
1799 The conversion can be C<upper()>, C<lower()>, or any other SQL function
1800 that can be applied symmetrically to fields (actually B<SQL::Abstract> does
1801 not validate this option; it will just pass through what you specify verbatim).
1805 This is a kludge because many databases suck. For example, you can't
1806 just bind values using DBI's C<execute()> for Oracle C<CLOB> or C<BLOB> fields.
1807 Instead, you have to use C<bind_param()>:
1809 $sth->bind_param(1, 'reg data');
1810 $sth->bind_param(2, $lots, {ora_type => ORA_CLOB});
1812 The problem is, B<SQL::Abstract> will normally just return a C<@bind> array,
1813 which loses track of which field each slot refers to. Fear not.
1815 If you specify C<bindtype> in new, you can determine how C<@bind> is returned.
1816 Currently, you can specify either C<normal> (default) or C<columns>. If you
1817 specify C<columns>, you will get an array that looks like this:
1819 my $sql = SQL::Abstract->new(bindtype => 'columns');
1820 my($stmt, @bind) = $sql->insert(...);
1823 [ 'column1', 'value1' ],
1824 [ 'column2', 'value2' ],
1825 [ 'column3', 'value3' ],
1828 You can then iterate through this manually, using DBI's C<bind_param()>.
1830 $sth->prepare($stmt);
1833 my($col, $data) = @$_;
1834 if ($col eq 'details' || $col eq 'comments') {
1835 $sth->bind_param($i, $data, {ora_type => ORA_CLOB});
1836 } elsif ($col eq 'image') {
1837 $sth->bind_param($i, $data, {ora_type => ORA_BLOB});
1839 $sth->bind_param($i, $data);
1843 $sth->execute; # execute without @bind now
1845 Now, why would you still use B<SQL::Abstract> if you have to do this crap?
1846 Basically, the advantage is still that you don't have to care which fields
1847 are or are not included. You could wrap that above C<for> loop in a simple
1848 sub called C<bind_fields()> or something and reuse it repeatedly. You still
1849 get a layer of abstraction over manual SQL specification.
1851 Note that if you set L</bindtype> to C<columns>, the C<\[ $sql, @bind ]>
1852 construct (see L</Literal SQL with placeholders and bind values (subqueries)>)
1853 will expect the bind values in this format.
1857 This is the character that a table or column name will be quoted
1858 with. By default this is an empty string, but you could set it to
1859 the character C<`>, to generate SQL like this:
1861 SELECT `a_field` FROM `a_table` WHERE `some_field` LIKE '%someval%'
1863 Alternatively, you can supply an array ref of two items, the first being the left
1864 hand quote character, and the second the right hand quote character. For
1865 example, you could supply C<['[',']']> for SQL Server 2000 compliant quotes
1866 that generates SQL like this:
1868 SELECT [a_field] FROM [a_table] WHERE [some_field] LIKE '%someval%'
1870 Quoting is useful if you have tables or columns names that are reserved
1871 words in your database's SQL dialect.
1875 This is the character that will be used to escape L</quote_char>s appearing
1876 in an identifier before it has been quoted.
1878 The parameter default in case of a single L</quote_char> character is the quote
1881 When opening-closing-style quoting is used (L</quote_char> is an arrayref)
1882 this parameter defaults to the B<closing (right)> L</quote_char>. Occurrences
1883 of the B<opening (left)> L</quote_char> within the identifier are currently left
1884 untouched. The default for opening-closing-style quotes may change in future
1885 versions, thus you are B<strongly encouraged> to specify the escape character
1890 This is the character that separates a table and column name. It is
1891 necessary to specify this when the C<quote_char> option is selected,
1892 so that tables and column names can be individually quoted like this:
1894 SELECT `table`.`one_field` FROM `table` WHERE `table`.`other_field` = 1
1896 =item injection_guard
1898 A regular expression C<qr/.../> that is applied to any C<-function> and unquoted
1899 column name specified in a query structure. This is a safety mechanism to avoid
1900 injection attacks when mishandling user input e.g.:
1902 my %condition_as_column_value_pairs = get_values_from_user();
1903 $sqla->select( ... , \%condition_as_column_value_pairs );
1905 If the expression matches an exception is thrown. Note that literal SQL
1906 supplied via C<\'...'> or C<\['...']> is B<not> checked in any way.
1908 Defaults to checking for C<;> and the C<GO> keyword (TransactSQL)
1910 =item array_datatypes
1912 When this option is true, arrayrefs in INSERT or UPDATE are
1913 interpreted as array datatypes and are passed directly
1915 When this option is false, arrayrefs are interpreted
1916 as literal SQL, just like refs to arrayrefs
1917 (but this behavior is for backwards compatibility; when writing
1918 new queries, use the "reference to arrayref" syntax
1924 Takes a reference to a list of "special operators"
1925 to extend the syntax understood by L<SQL::Abstract>.
1926 See section L</"SPECIAL OPERATORS"> for details.
1930 Takes a reference to a list of "unary operators"
1931 to extend the syntax understood by L<SQL::Abstract>.
1932 See section L</"UNARY OPERATORS"> for details.
1938 =head2 insert($table, \@values || \%fieldvals, \%options)
1940 This is the simplest function. You simply give it a table name
1941 and either an arrayref of values or hashref of field/value pairs.
1942 It returns an SQL INSERT statement and a list of bind values.
1943 See the sections on L</"Inserting and Updating Arrays"> and
1944 L</"Inserting and Updating SQL"> for information on how to insert
1945 with those data types.
1947 The optional C<\%options> hash reference may contain additional
1948 options to generate the insert SQL. Currently supported options
1955 Takes either a scalar of raw SQL fields, or an array reference of
1956 field names, and adds on an SQL C<RETURNING> statement at the end.
1957 This allows you to return data generated by the insert statement
1958 (such as row IDs) without performing another C<SELECT> statement.
1959 Note, however, this is not part of the SQL standard and may not
1960 be supported by all database engines.
1964 =head2 update($table, \%fieldvals, \%where, \%options)
1966 This takes a table, hashref of field/value pairs, and an optional
1967 hashref L<WHERE clause|/WHERE CLAUSES>. It returns an SQL UPDATE function and a list
1969 See the sections on L</"Inserting and Updating Arrays"> and
1970 L</"Inserting and Updating SQL"> for information on how to insert
1971 with those data types.
1973 The optional C<\%options> hash reference may contain additional
1974 options to generate the update SQL. Currently supported options
1981 See the C<returning> option to
1982 L<insert|/insert($table, \@values || \%fieldvals, \%options)>.
1986 =head2 select($source, $fields, $where, $order)
1988 This returns a SQL SELECT statement and associated list of bind values, as
1989 specified by the arguments:
1995 Specification of the 'FROM' part of the statement.
1996 The argument can be either a plain scalar (interpreted as a table
1997 name, will be quoted), or an arrayref (interpreted as a list
1998 of table names, joined by commas, quoted), or a scalarref
1999 (literal SQL, not quoted).
2003 Specification of the list of fields to retrieve from
2005 The argument can be either an arrayref (interpreted as a list
2006 of field names, will be joined by commas and quoted), or a
2007 plain scalar (literal SQL, not quoted).
2008 Please observe that this API is not as flexible as that of
2009 the first argument C<$source>, for backwards compatibility reasons.
2013 Optional argument to specify the WHERE part of the query.
2014 The argument is most often a hashref, but can also be
2015 an arrayref or plain scalar --
2016 see section L<WHERE clause|/"WHERE CLAUSES"> for details.
2020 Optional argument to specify the ORDER BY part of the query.
2021 The argument can be a scalar, a hashref or an arrayref
2022 -- see section L<ORDER BY clause|/"ORDER BY CLAUSES">
2028 =head2 delete($table, \%where, \%options)
2030 This takes a table name and optional hashref L<WHERE clause|/WHERE CLAUSES>.
2031 It returns an SQL DELETE statement and list of bind values.
2033 The optional C<\%options> hash reference may contain additional
2034 options to generate the delete SQL. Currently supported options
2041 See the C<returning> option to
2042 L<insert|/insert($table, \@values || \%fieldvals, \%options)>.
2046 =head2 where(\%where, $order)
2048 This is used to generate just the WHERE clause. For example,
2049 if you have an arbitrary data structure and know what the
2050 rest of your SQL is going to look like, but want an easy way
2051 to produce a WHERE clause, use this. It returns an SQL WHERE
2052 clause and list of bind values.
2055 =head2 values(\%data)
2057 This just returns the values from the hash C<%data>, in the same
2058 order that would be returned from any of the other above queries.
2059 Using this allows you to markedly speed up your queries if you
2060 are affecting lots of rows. See below under the L</"PERFORMANCE"> section.
2062 =head2 generate($any, 'number', $of, \@data, $struct, \%types)
2064 Warning: This is an experimental method and subject to change.
2066 This returns arbitrarily generated SQL. It's a really basic shortcut.
2067 It will return two different things, depending on return context:
2069 my($stmt, @bind) = $sql->generate('create table', \$table, \@fields);
2070 my $stmt_and_val = $sql->generate('create table', \$table, \@fields);
2072 These would return the following:
2074 # First calling form
2075 $stmt = "CREATE TABLE test (?, ?)";
2076 @bind = (field1, field2);
2078 # Second calling form
2079 $stmt_and_val = "CREATE TABLE test (field1, field2)";
2081 Depending on what you're trying to do, it's up to you to choose the correct
2082 format. In this example, the second form is what you would want.
2086 $sql->generate('alter session', { nls_date_format => 'MM/YY' });
2090 ALTER SESSION SET nls_date_format = 'MM/YY'
2092 You get the idea. Strings get their case twiddled, but everything
2093 else remains verbatim.
2095 =head1 EXPORTABLE FUNCTIONS
2097 =head2 is_plain_value
2099 Determines if the supplied argument is a plain value as understood by this
2104 =item * The value is C<undef>
2106 =item * The value is a non-reference
2108 =item * The value is an object with stringification overloading
2110 =item * The value is of the form C<< { -value => $anything } >>
2114 On failure returns C<undef>, on success returns a B<scalar> reference
2115 to the original supplied argument.
2121 The stringification overloading detection is rather advanced: it takes
2122 into consideration not only the presence of a C<""> overload, but if that
2123 fails also checks for enabled
2124 L<autogenerated versions of C<"">|overload/Magic Autogeneration>, based
2125 on either C<0+> or C<bool>.
2127 Unfortunately testing in the field indicates that this
2128 detection B<< may tickle a latent bug in perl versions before 5.018 >>,
2129 but only when very large numbers of stringifying objects are involved.
2130 At the time of writing ( Sep 2014 ) there is no clear explanation of
2131 the direct cause, nor is there a manageably small test case that reliably
2132 reproduces the problem.
2134 If you encounter any of the following exceptions in B<random places within
2135 your application stack> - this module may be to blame:
2137 Operation "ne": no method found,
2138 left argument in overloaded package <something>,
2139 right argument in overloaded package <something>
2143 Stub found while resolving method "???" overloading """" in package <something>
2145 If you fall victim to the above - please attempt to reduce the problem
2146 to something that could be sent to the L<SQL::Abstract developers
2147 |DBIx::Class/GETTING HELP/SUPPORT>
2148 (either publicly or privately). As a workaround in the meantime you can
2149 set C<$ENV{SQLA_ISVALUE_IGNORE_AUTOGENERATED_STRINGIFICATION}> to a true
2150 value, which will most likely eliminate your problem (at the expense of
2151 not being able to properly detect exotic forms of stringification).
2153 This notice and environment variable will be removed in a future version,
2154 as soon as the underlying problem is found and a reliable workaround is
2159 =head2 is_literal_value
2161 Determines if the supplied argument is a literal value as understood by this
2166 =item * C<\$sql_string>
2168 =item * C<\[ $sql_string, @bind_values ]>
2172 On failure returns C<undef>, on success returns an B<array> reference
2173 containing the unpacked version of the supplied literal SQL and bind values.
2175 =head1 WHERE CLAUSES
2179 This module uses a variation on the idea from L<DBIx::Abstract>. It
2180 is B<NOT>, repeat I<not> 100% compatible. B<The main logic of this
2181 module is that things in arrays are OR'ed, and things in hashes
2184 The easiest way to explain is to show lots of examples. After
2185 each C<%where> hash shown, it is assumed you used:
2187 my($stmt, @bind) = $sql->where(\%where);
2189 However, note that the C<%where> hash can be used directly in any
2190 of the other functions as well, as described above.
2192 =head2 Key-value pairs
2194 So, let's get started. To begin, a simple hash:
2198 status => 'completed'
2201 Is converted to SQL C<key = val> statements:
2203 $stmt = "WHERE user = ? AND status = ?";
2204 @bind = ('nwiger', 'completed');
2206 One common thing I end up doing is having a list of values that
2207 a field can be in. To do this, simply specify a list inside of
2212 status => ['assigned', 'in-progress', 'pending'];
2215 This simple code will create the following:
2217 $stmt = "WHERE user = ? AND ( status = ? OR status = ? OR status = ? )";
2218 @bind = ('nwiger', 'assigned', 'in-progress', 'pending');
2220 A field associated to an empty arrayref will be considered a
2221 logical false and will generate 0=1.
2223 =head2 Tests for NULL values
2225 If the value part is C<undef> then this is converted to SQL <IS NULL>
2234 $stmt = "WHERE user = ? AND status IS NULL";
2237 To test if a column IS NOT NULL:
2241 status => { '!=', undef },
2244 =head2 Specific comparison operators
2246 If you want to specify a different type of operator for your comparison,
2247 you can use a hashref for a given column:
2251 status => { '!=', 'completed' }
2254 Which would generate:
2256 $stmt = "WHERE user = ? AND status != ?";
2257 @bind = ('nwiger', 'completed');
2259 To test against multiple values, just enclose the values in an arrayref:
2261 status => { '=', ['assigned', 'in-progress', 'pending'] };
2263 Which would give you:
2265 "WHERE status = ? OR status = ? OR status = ?"
2268 The hashref can also contain multiple pairs, in which case it is expanded
2269 into an C<AND> of its elements:
2273 status => { '!=', 'completed', -not_like => 'pending%' }
2276 # Or more dynamically, like from a form
2277 $where{user} = 'nwiger';
2278 $where{status}{'!='} = 'completed';
2279 $where{status}{'-not_like'} = 'pending%';
2281 # Both generate this
2282 $stmt = "WHERE user = ? AND status != ? AND status NOT LIKE ?";
2283 @bind = ('nwiger', 'completed', 'pending%');
2286 To get an OR instead, you can combine it with the arrayref idea:
2290 priority => [ { '=', 2 }, { '>', 5 } ]
2293 Which would generate:
2295 $stmt = "WHERE ( priority = ? OR priority > ? ) AND user = ?";
2296 @bind = ('2', '5', 'nwiger');
2298 If you want to include literal SQL (with or without bind values), just use a
2299 scalar reference or reference to an arrayref as the value:
2302 date_entered => { '>' => \["to_date(?, 'MM/DD/YYYY')", "11/26/2008"] },
2303 date_expires => { '<' => \"now()" }
2306 Which would generate:
2308 $stmt = "WHERE date_entered > to_date(?, 'MM/DD/YYYY') AND date_expires < now()";
2309 @bind = ('11/26/2008');
2312 =head2 Logic and nesting operators
2314 In the example above,
2315 there is a subtle trap if you want to say something like
2316 this (notice the C<AND>):
2318 WHERE priority != ? AND priority != ?
2320 Because, in Perl you I<can't> do this:
2322 priority => { '!=' => 2, '!=' => 1 }
2324 As the second C<!=> key will obliterate the first. The solution
2325 is to use the special C<-modifier> form inside an arrayref:
2327 priority => [ -and => {'!=', 2},
2331 Normally, these would be joined by C<OR>, but the modifier tells it
2332 to use C<AND> instead. (Hint: You can use this in conjunction with the
2333 C<logic> option to C<new()> in order to change the way your queries
2334 work by default.) B<Important:> Note that the C<-modifier> goes
2335 B<INSIDE> the arrayref, as an extra first element. This will
2336 B<NOT> do what you think it might:
2338 priority => -and => [{'!=', 2}, {'!=', 1}] # WRONG!
2340 Here is a quick list of equivalencies, since there is some overlap:
2343 status => {'!=', 'completed', 'not like', 'pending%' }
2344 status => [ -and => {'!=', 'completed'}, {'not like', 'pending%'}]
2347 status => {'=', ['assigned', 'in-progress']}
2348 status => [ -or => {'=', 'assigned'}, {'=', 'in-progress'}]
2349 status => [ {'=', 'assigned'}, {'=', 'in-progress'} ]
2353 =head2 Special operators: IN, BETWEEN, etc.
2355 You can also use the hashref format to compare a list of fields using the
2356 C<IN> comparison operator, by specifying the list as an arrayref:
2359 status => 'completed',
2360 reportid => { -in => [567, 2335, 2] }
2363 Which would generate:
2365 $stmt = "WHERE status = ? AND reportid IN (?,?,?)";
2366 @bind = ('completed', '567', '2335', '2');
2368 The reverse operator C<-not_in> generates SQL C<NOT IN> and is used in
2371 If the argument to C<-in> is an empty array, 'sqlfalse' is generated
2372 (by default: C<1=0>). Similarly, C<< -not_in => [] >> generates
2373 'sqltrue' (by default: C<1=1>).
2375 In addition to the array you can supply a chunk of literal sql or
2376 literal sql with bind:
2379 customer => { -in => \[
2380 'SELECT cust_id FROM cust WHERE balance > ?',
2383 status => { -in => \'SELECT status_codes FROM states' },
2389 customer IN ( SELECT cust_id FROM cust WHERE balance > ? )
2390 AND status IN ( SELECT status_codes FROM states )
2394 Finally, if the argument to C<-in> is not a reference, it will be
2395 treated as a single-element array.
2397 Another pair of operators is C<-between> and C<-not_between>,
2398 used with an arrayref of two values:
2402 completion_date => {
2403 -not_between => ['2002-10-01', '2003-02-06']
2409 WHERE user = ? AND completion_date NOT BETWEEN ( ? AND ? )
2411 Just like with C<-in> all plausible combinations of literal SQL
2415 start0 => { -between => [ 1, 2 ] },
2416 start1 => { -between => \["? AND ?", 1, 2] },
2417 start2 => { -between => \"lower(x) AND upper(y)" },
2418 start3 => { -between => [
2420 \["upper(?)", 'stuff' ],
2427 ( start0 BETWEEN ? AND ? )
2428 AND ( start1 BETWEEN ? AND ? )
2429 AND ( start2 BETWEEN lower(x) AND upper(y) )
2430 AND ( start3 BETWEEN lower(x) AND upper(?) )
2432 @bind = (1, 2, 1, 2, 'stuff');
2435 These are the two builtin "special operators"; but the
2436 list can be expanded: see section L</"SPECIAL OPERATORS"> below.
2438 =head2 Unary operators: bool
2440 If you wish to test against boolean columns or functions within your
2441 database you can use the C<-bool> and C<-not_bool> operators. For
2442 example to test the column C<is_user> being true and the column
2443 C<is_enabled> being false you would use:-
2447 -not_bool => 'is_enabled',
2452 WHERE is_user AND NOT is_enabled
2454 If a more complex combination is required, testing more conditions,
2455 then you should use the and/or operators:-
2460 -not_bool => { two=> { -rlike => 'bar' } },
2461 -not_bool => { three => [ { '=', 2 }, { '>', 5 } ] },
2472 (NOT ( three = ? OR three > ? ))
2475 =head2 Nested conditions, -and/-or prefixes
2477 So far, we've seen how multiple conditions are joined with a top-level
2478 C<AND>. We can change this by putting the different conditions we want in
2479 hashes and then putting those hashes in an array. For example:
2484 status => { -like => ['pending%', 'dispatched'] },
2488 status => 'unassigned',
2492 This data structure would create the following:
2494 $stmt = "WHERE ( user = ? AND ( status LIKE ? OR status LIKE ? ) )
2495 OR ( user = ? AND status = ? ) )";
2496 @bind = ('nwiger', 'pending', 'dispatched', 'robot', 'unassigned');
2499 Clauses in hashrefs or arrayrefs can be prefixed with an C<-and> or C<-or>
2500 to change the logic inside:
2506 -and => [ workhrs => {'>', 20}, geo => 'ASIA' ],
2507 -or => { workhrs => {'<', 50}, geo => 'EURO' },
2514 $stmt = "WHERE ( user = ?
2515 AND ( ( workhrs > ? AND geo = ? )
2516 OR ( workhrs < ? OR geo = ? ) ) )";
2517 @bind = ('nwiger', '20', 'ASIA', '50', 'EURO');
2519 =head3 Algebraic inconsistency, for historical reasons
2521 C<Important note>: when connecting several conditions, the C<-and->|C<-or>
2522 operator goes C<outside> of the nested structure; whereas when connecting
2523 several constraints on one column, the C<-and> operator goes
2524 C<inside> the arrayref. Here is an example combining both features:
2527 -and => [a => 1, b => 2],
2528 -or => [c => 3, d => 4],
2529 e => [-and => {-like => 'foo%'}, {-like => '%bar'} ]
2534 WHERE ( ( ( a = ? AND b = ? )
2535 OR ( c = ? OR d = ? )
2536 OR ( e LIKE ? AND e LIKE ? ) ) )
2538 This difference in syntax is unfortunate but must be preserved for
2539 historical reasons. So be careful: the two examples below would
2540 seem algebraically equivalent, but they are not
2543 { -like => 'foo%' },
2544 { -like => '%bar' },
2546 # yields: WHERE ( ( col LIKE ? AND col LIKE ? ) )
2549 { col => { -like => 'foo%' } },
2550 { col => { -like => '%bar' } },
2552 # yields: WHERE ( ( col LIKE ? OR col LIKE ? ) )
2555 =head2 Literal SQL and value type operators
2557 The basic premise of SQL::Abstract is that in WHERE specifications the "left
2558 side" is a column name and the "right side" is a value (normally rendered as
2559 a placeholder). This holds true for both hashrefs and arrayref pairs as you
2560 see in the L</WHERE CLAUSES> examples above. Sometimes it is necessary to
2561 alter this behavior. There are several ways of doing so.
2565 This is a virtual operator that signals the string to its right side is an
2566 identifier (a column name) and not a value. For example to compare two
2567 columns you would write:
2570 priority => { '<', 2 },
2571 requestor => { -ident => 'submitter' },
2576 $stmt = "WHERE priority < ? AND requestor = submitter";
2579 If you are maintaining legacy code you may see a different construct as
2580 described in L</Deprecated usage of Literal SQL>, please use C<-ident> in new
2585 This is a virtual operator that signals that the construct to its right side
2586 is a value to be passed to DBI. This is for example necessary when you want
2587 to write a where clause against an array (for RDBMS that support such
2588 datatypes). For example:
2591 array => { -value => [1, 2, 3] }
2596 $stmt = 'WHERE array = ?';
2597 @bind = ([1, 2, 3]);
2599 Note that if you were to simply say:
2605 the result would probably not be what you wanted:
2607 $stmt = 'WHERE array = ? OR array = ? OR array = ?';
2612 Finally, sometimes only literal SQL will do. To include a random snippet
2613 of SQL verbatim, you specify it as a scalar reference. Consider this only
2614 as a last resort. Usually there is a better way. For example:
2617 priority => { '<', 2 },
2618 requestor => { -in => \'(SELECT name FROM hitmen)' },
2623 $stmt = "WHERE priority < ? AND requestor IN (SELECT name FROM hitmen)"
2626 Note that in this example, you only get one bind parameter back, since
2627 the verbatim SQL is passed as part of the statement.
2631 Never use untrusted input as a literal SQL argument - this is a massive
2632 security risk (there is no way to check literal snippets for SQL
2633 injections and other nastyness). If you need to deal with untrusted input
2634 use literal SQL with placeholders as described next.
2636 =head3 Literal SQL with placeholders and bind values (subqueries)
2638 If the literal SQL to be inserted has placeholders and bind values,
2639 use a reference to an arrayref (yes this is a double reference --
2640 not so common, but perfectly legal Perl). For example, to find a date
2641 in Postgres you can use something like this:
2644 date_column => \[ "= date '2008-09-30' - ?::integer", 10 ]
2649 $stmt = "WHERE ( date_column = date '2008-09-30' - ?::integer )"
2652 Note that you must pass the bind values in the same format as they are returned
2653 by L<where|/where(\%where, $order)>. This means that if you set L</bindtype>
2654 to C<columns>, you must provide the bind values in the
2655 C<< [ column_meta => value ] >> format, where C<column_meta> is an opaque
2656 scalar value; most commonly the column name, but you can use any scalar value
2657 (including references and blessed references), L<SQL::Abstract> will simply
2658 pass it through intact. So if C<bindtype> is set to C<columns> the above
2659 example will look like:
2662 date_column => \[ "= date '2008-09-30' - ?::integer", [ {} => 10 ] ]
2665 Literal SQL is especially useful for nesting parenthesized clauses in the
2666 main SQL query. Here is a first example:
2668 my ($sub_stmt, @sub_bind) = ("SELECT c1 FROM t1 WHERE c2 < ? AND c3 LIKE ?",
2672 bar => \["IN ($sub_stmt)" => @sub_bind],
2677 $stmt = "WHERE (foo = ? AND bar IN (SELECT c1 FROM t1
2678 WHERE c2 < ? AND c3 LIKE ?))";
2679 @bind = (1234, 100, "foo%");
2681 Other subquery operators, like for example C<"E<gt> ALL"> or C<"NOT IN">,
2682 are expressed in the same way. Of course the C<$sub_stmt> and
2683 its associated bind values can be generated through a former call
2686 my ($sub_stmt, @sub_bind)
2687 = $sql->select("t1", "c1", {c2 => {"<" => 100},
2688 c3 => {-like => "foo%"}});
2691 bar => \["> ALL ($sub_stmt)" => @sub_bind],
2694 In the examples above, the subquery was used as an operator on a column;
2695 but the same principle also applies for a clause within the main C<%where>
2696 hash, like an EXISTS subquery:
2698 my ($sub_stmt, @sub_bind)
2699 = $sql->select("t1", "*", {c1 => 1, c2 => \"> t0.c0"});
2700 my %where = ( -and => [
2702 \["EXISTS ($sub_stmt)" => @sub_bind],
2707 $stmt = "WHERE (foo = ? AND EXISTS (SELECT * FROM t1
2708 WHERE c1 = ? AND c2 > t0.c0))";
2712 Observe that the condition on C<c2> in the subquery refers to
2713 column C<t0.c0> of the main query: this is I<not> a bind
2714 value, so we have to express it through a scalar ref.
2715 Writing C<< c2 => {">" => "t0.c0"} >> would have generated
2716 C<< c2 > ? >> with bind value C<"t0.c0"> ... not exactly
2717 what we wanted here.
2719 Finally, here is an example where a subquery is used
2720 for expressing unary negation:
2722 my ($sub_stmt, @sub_bind)
2723 = $sql->where({age => [{"<" => 10}, {">" => 20}]});
2724 $sub_stmt =~ s/^ where //i; # don't want "WHERE" in the subclause
2726 lname => {like => '%son%'},
2727 \["NOT ($sub_stmt)" => @sub_bind],
2732 $stmt = "lname LIKE ? AND NOT ( age < ? OR age > ? )"
2733 @bind = ('%son%', 10, 20)
2735 =head3 Deprecated usage of Literal SQL
2737 Below are some examples of archaic use of literal SQL. It is shown only as
2738 reference for those who deal with legacy code. Each example has a much
2739 better, cleaner and safer alternative that users should opt for in new code.
2745 my %where = ( requestor => \'IS NOT NULL' )
2747 $stmt = "WHERE requestor IS NOT NULL"
2749 This used to be the way of generating NULL comparisons, before the handling
2750 of C<undef> got formalized. For new code please use the superior syntax as
2751 described in L</Tests for NULL values>.
2755 my %where = ( requestor => \'= submitter' )
2757 $stmt = "WHERE requestor = submitter"
2759 This used to be the only way to compare columns. Use the superior L</-ident>
2760 method for all new code. For example an identifier declared in such a way
2761 will be properly quoted if L</quote_char> is properly set, while the legacy
2762 form will remain as supplied.
2766 my %where = ( is_ready => \"", completed => { '>', '2012-12-21' } )
2768 $stmt = "WHERE completed > ? AND is_ready"
2769 @bind = ('2012-12-21')
2771 Using an empty string literal used to be the only way to express a boolean.
2772 For all new code please use the much more readable
2773 L<-bool|/Unary operators: bool> operator.
2779 These pages could go on for a while, since the nesting of the data
2780 structures this module can handle are pretty much unlimited (the
2781 module implements the C<WHERE> expansion as a recursive function
2782 internally). Your best bet is to "play around" with the module a
2783 little to see how the data structures behave, and choose the best
2784 format for your data based on that.
2786 And of course, all the values above will probably be replaced with
2787 variables gotten from forms or the command line. After all, if you
2788 knew everything ahead of time, you wouldn't have to worry about
2789 dynamically-generating SQL and could just hardwire it into your
2792 =head1 ORDER BY CLAUSES
2794 Some functions take an order by clause. This can either be a scalar (just a
2795 column name), a hashref of C<< { -desc => 'col' } >> or C<< { -asc => 'col' }
2796 >>, a scalarref, an arrayref-ref, or an arrayref of any of the previous
2799 Given | Will Generate
2800 ---------------------------------------------------------------
2802 'colA' | ORDER BY colA
2804 [qw/colA colB/] | ORDER BY colA, colB
2806 {-asc => 'colA'} | ORDER BY colA ASC
2808 {-desc => 'colB'} | ORDER BY colB DESC
2810 ['colA', {-asc => 'colB'}] | ORDER BY colA, colB ASC
2812 { -asc => [qw/colA colB/] } | ORDER BY colA ASC, colB ASC
2814 \'colA DESC' | ORDER BY colA DESC
2816 \[ 'FUNC(colA, ?)', $x ] | ORDER BY FUNC(colA, ?)
2817 | /* ...with $x bound to ? */
2820 { -asc => 'colA' }, | colA ASC,
2821 { -desc => [qw/colB/] }, | colB DESC,
2822 { -asc => [qw/colC colD/] },| colC ASC, colD ASC,
2823 \'colE DESC', | colE DESC,
2824 \[ 'FUNC(colF, ?)', $x ], | FUNC(colF, ?)
2825 ] | /* ...with $x bound to ? */
2826 ===============================================================
2830 =head1 SPECIAL OPERATORS
2832 my $sqlmaker = SQL::Abstract->new(special_ops => [
2836 my ($self, $field, $op, $arg) = @_;
2842 handler => 'method_name',
2846 A "special operator" is a SQL syntactic clause that can be
2847 applied to a field, instead of a usual binary operator.
2850 WHERE field IN (?, ?, ?)
2851 WHERE field BETWEEN ? AND ?
2852 WHERE MATCH(field) AGAINST (?, ?)
2854 Special operators IN and BETWEEN are fairly standard and therefore
2855 are builtin within C<SQL::Abstract> (as the overridable methods
2856 C<_where_field_IN> and C<_where_field_BETWEEN>). For other operators,
2857 like the MATCH .. AGAINST example above which is specific to MySQL,
2858 you can write your own operator handlers - supply a C<special_ops>
2859 argument to the C<new> method. That argument takes an arrayref of
2860 operator definitions; each operator definition is a hashref with two
2867 the regular expression to match the operator
2871 Either a coderef or a plain scalar method name. In both cases
2872 the expected return is C<< ($sql, @bind) >>.
2874 When supplied with a method name, it is simply called on the
2875 L<SQL::Abstract> object as:
2877 $self->$method_name($field, $op, $arg)
2881 $field is the LHS of the operator
2882 $op is the part that matched the handler regex
2885 When supplied with a coderef, it is called as:
2887 $coderef->($self, $field, $op, $arg)
2892 For example, here is an implementation
2893 of the MATCH .. AGAINST syntax for MySQL
2895 my $sqlmaker = SQL::Abstract->new(special_ops => [
2897 # special op for MySql MATCH (field) AGAINST(word1, word2, ...)
2898 {regex => qr/^match$/i,
2900 my ($self, $field, $op, $arg) = @_;
2901 $arg = [$arg] if not ref $arg;
2902 my $label = $self->_quote($field);
2903 my ($placeholder) = $self->_convert('?');
2904 my $placeholders = join ", ", (($placeholder) x @$arg);
2905 my $sql = $self->_sqlcase('match') . " ($label) "
2906 . $self->_sqlcase('against') . " ($placeholders) ";
2907 my @bind = $self->_bindtype($field, @$arg);
2908 return ($sql, @bind);
2915 =head1 UNARY OPERATORS
2917 my $sqlmaker = SQL::Abstract->new(unary_ops => [
2921 my ($self, $op, $arg) = @_;
2927 handler => 'method_name',
2931 A "unary operator" is a SQL syntactic clause that can be
2932 applied to a field - the operator goes before the field
2934 You can write your own operator handlers - supply a C<unary_ops>
2935 argument to the C<new> method. That argument takes an arrayref of
2936 operator definitions; each operator definition is a hashref with two
2943 the regular expression to match the operator
2947 Either a coderef or a plain scalar method name. In both cases
2948 the expected return is C<< $sql >>.
2950 When supplied with a method name, it is simply called on the
2951 L<SQL::Abstract> object as:
2953 $self->$method_name($op, $arg)
2957 $op is the part that matched the handler regex
2958 $arg is the RHS or argument of the operator
2960 When supplied with a coderef, it is called as:
2962 $coderef->($self, $op, $arg)
2970 Thanks to some benchmarking by Mark Stosberg, it turns out that
2971 this module is many orders of magnitude faster than using C<DBIx::Abstract>.
2972 I must admit this wasn't an intentional design issue, but it's a
2973 byproduct of the fact that you get to control your C<DBI> handles
2976 To maximize performance, use a code snippet like the following:
2978 # prepare a statement handle using the first row
2979 # and then reuse it for the rest of the rows
2981 for my $href (@array_of_hashrefs) {
2982 $stmt ||= $sql->insert('table', $href);
2983 $sth ||= $dbh->prepare($stmt);
2984 $sth->execute($sql->values($href));
2987 The reason this works is because the keys in your C<$href> are sorted
2988 internally by B<SQL::Abstract>. Thus, as long as your data retains
2989 the same structure, you only have to generate the SQL the first time
2990 around. On subsequent queries, simply use the C<values> function provided
2991 by this module to return your values in the correct order.
2993 However this depends on the values having the same type - if, for
2994 example, the values of a where clause may either have values
2995 (resulting in sql of the form C<column = ?> with a single bind
2996 value), or alternatively the values might be C<undef> (resulting in
2997 sql of the form C<column IS NULL> with no bind value) then the
2998 caching technique suggested will not work.
3002 If you use my C<CGI::FormBuilder> module at all, you'll hopefully
3003 really like this part (I do, at least). Building up a complex query
3004 can be as simple as the following:
3011 use CGI::FormBuilder;
3014 my $form = CGI::FormBuilder->new(...);
3015 my $sql = SQL::Abstract->new;
3017 if ($form->submitted) {
3018 my $field = $form->field;
3019 my $id = delete $field->{id};
3020 my($stmt, @bind) = $sql->update('table', $field, {id => $id});
3023 Of course, you would still have to connect using C<DBI> to run the
3024 query, but the point is that if you make your form look like your
3025 table, the actual query script can be extremely simplistic.
3027 If you're B<REALLY> lazy (I am), check out C<HTML::QuickTable> for
3028 a fast interface to returning and formatting data. I frequently
3029 use these three modules together to write complex database query
3030 apps in under 50 lines.
3032 =head1 HOW TO CONTRIBUTE
3034 Contributions are always welcome, in all usable forms (we especially
3035 welcome documentation improvements). The delivery methods include git-
3036 or unified-diff formatted patches, GitHub pull requests, or plain bug
3037 reports either via RT or the Mailing list. Contributors are generally
3038 granted full access to the official repository after their first several
3039 patches pass successful review.
3041 This project is maintained in a git repository. The code and related tools are
3042 accessible at the following locations:
3046 =item * Official repo: L<git://git.shadowcat.co.uk/dbsrgits/SQL-Abstract.git>
3048 =item * Official gitweb: L<http://git.shadowcat.co.uk/gitweb/gitweb.cgi?p=dbsrgits/SQL-Abstract.git>
3050 =item * GitHub mirror: L<https://github.com/dbsrgits/sql-abstract>
3052 =item * Authorized committers: L<ssh://dbsrgits@git.shadowcat.co.uk/SQL-Abstract.git>
3058 Version 1.50 was a major internal refactoring of C<SQL::Abstract>.
3059 Great care has been taken to preserve the I<published> behavior
3060 documented in previous versions in the 1.* family; however,
3061 some features that were previously undocumented, or behaved
3062 differently from the documentation, had to be changed in order
3063 to clarify the semantics. Hence, client code that was relying
3064 on some dark areas of C<SQL::Abstract> v1.*
3065 B<might behave differently> in v1.50.
3067 The main changes are:
3073 support for literal SQL through the C<< \ [ $sql, @bind ] >> syntax.
3077 support for the { operator => \"..." } construct (to embed literal SQL)
3081 support for the { operator => \["...", @bind] } construct (to embed literal SQL with bind values)
3085 optional support for L<array datatypes|/"Inserting and Updating Arrays">
3089 defensive programming: check arguments
3093 fixed bug with global logic, which was previously implemented
3094 through global variables yielding side-effects. Prior versions would
3095 interpret C<< [ {cond1, cond2}, [cond3, cond4] ] >>
3096 as C<< "(cond1 AND cond2) OR (cond3 AND cond4)" >>.
3097 Now this is interpreted
3098 as C<< "(cond1 AND cond2) OR (cond3 OR cond4)" >>.
3103 fixed semantics of _bindtype on array args
3107 dropped the C<_anoncopy> of the %where tree. No longer necessary,
3108 we just avoid shifting arrays within that tree.
3112 dropped the C<_modlogic> function
3116 =head1 ACKNOWLEDGEMENTS
3118 There are a number of individuals that have really helped out with
3119 this module. Unfortunately, most of them submitted bugs via CPAN
3120 so I have no idea who they are! But the people I do know are:
3122 Ash Berlin (order_by hash term support)
3123 Matt Trout (DBIx::Class support)
3124 Mark Stosberg (benchmarking)
3125 Chas Owens (initial "IN" operator support)
3126 Philip Collins (per-field SQL functions)
3127 Eric Kolve (hashref "AND" support)
3128 Mike Fragassi (enhancements to "BETWEEN" and "LIKE")
3129 Dan Kubb (support for "quote_char" and "name_sep")
3130 Guillermo Roditi (patch to cleanup "IN" and "BETWEEN", fix and tests for _order_by)
3131 Laurent Dami (internal refactoring, extensible list of special operators, literal SQL)
3132 Norbert Buchmuller (support for literal SQL in hashpair, misc. fixes & tests)
3133 Peter Rabbitson (rewrite of SQLA::Test, misc. fixes & tests)
3134 Oliver Charles (support for "RETURNING" after "INSERT")
3140 L<DBIx::Class>, L<DBIx::Abstract>, L<CGI::FormBuilder>, L<HTML::QuickTable>.
3144 Copyright (c) 2001-2007 Nathan Wiger <nwiger@cpan.org>. All Rights Reserved.
3146 This module is actively maintained by Matt Trout <mst@shadowcatsystems.co.uk>
3148 For support, your best bet is to try the C<DBIx::Class> users mailing list.
3149 While not an official support venue, C<DBIx::Class> makes heavy use of
3150 C<SQL::Abstract>, and as such list members there are very familiar with
3151 how to create queries.
3155 This module is free software; you may copy this under the same
3156 terms as perl itself (either the GNU General Public License or
3157 the Artistic License)