1 package SQL::Abstract; # see doc at end of file
10 our @EXPORT_OK = qw(is_plain_value is_literal_value);
20 *SQL::Abstract::_ENV_::DETECT_AUTOGENERATED_STRINGIFICATION = $ENV{SQLA_ISVALUE_IGNORE_AUTOGENERATED_STRINGIFICATION}
26 #======================================================================
28 #======================================================================
30 our $VERSION = '1.86';
32 # This would confuse some packagers
33 $VERSION = eval $VERSION if $VERSION =~ /_/; # numify for warning-free dev releases
37 # special operators (-in, -between). May be extended/overridden by user.
38 # See section WHERE: BUILTIN SPECIAL OPERATORS below for implementation
39 my @BUILTIN_SPECIAL_OPS = (
40 {regex => qr/^ (?: not \s )? between $/ix, handler => sub { die "NOPE" }},
41 {regex => qr/^ (?: 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";
609 return $self->_expand_expr_hashpair_op($k, $v);
611 return $self->_expand_expr_hashpair_ident($k, $v);
614 sub _expand_expr_hashpair_ident {
615 my ($self, $k, $v) = @_;
617 # undef needs to be re-sent with cmp to achieve IS/IS NOT NULL
623 and exists $v->{-value}
624 and not defined $v->{-value}
627 return $self->_expand_expr({ $k => { $self->{cmp} => undef } });
630 my $ik = $self->_expand_ident(-ident => $k);
632 # scalars and objects get expanded as whatever requested or values
634 if (!ref($v) or Scalar::Util::blessed($v)) {
635 my $d = our $Default_Scalar_To;
636 local our $Cur_Col_Meta = $k;
637 return $self->_expand_expr_hashpair_ident(
640 ? $self->_expand_expr($d => $v)
645 if (ref($v) eq 'HASH') {
647 return $self->_expand_op_andor(-and => $v, $k);
649 return undef unless keys %$v;
651 my $op = join ' ', split '_', (map lc, $vk =~ /^-?(.*)$/)[0];
652 $self->_assert_pass_injection_guard($op);
653 if ($op =~ s/ [_\s]? \d+ $//x ) {
654 return $self->_expand_expr($k, $v);
656 if (my $x = $self->{expand_op}{$op}) {
657 local our $Cur_Col_Meta = $k;
658 return $self->$x($op, $vv, $k);
660 if ($op =~ /^is(?: not)?$/) {
661 puke "$op can only take undef as argument"
665 and exists($vv->{-value})
666 and !defined($vv->{-value})
668 return +{ -op => [ $op.' null', $ik ] };
670 if (my $us = List::Util::first { $op =~ $_->{regex} } @{$self->{special_ops}}) {
671 return { -op => [ $op, $ik, $vv ] };
673 if (my $us = List::Util::first { $op =~ $_->{regex} } @{$self->{unary_ops}}) {
677 { -op => [ $op, $vv ] }
680 if (ref($vv) eq 'ARRAY') {
681 my ($logic, @values) = (
682 (defined($vv->[0]) and $vv->[0] =~ /^-(and|or)$/i)
687 $op =~ $self->{inequality_op}
688 or $op =~ $self->{not_like_op}
690 if (lc($logic) eq '-or' and @values > 1) {
691 belch "A multi-element arrayref as an argument to the inequality op '${\uc($op)}' "
692 . 'is technically equivalent to an always-true 1=1 (you probably wanted '
693 . "to say ...{ \$inequality_op => [ -and => \@values ] }... instead)"
698 # try to DWIM on equality operators
700 $op =~ $self->{equality_op} ? $self->sqlfalse
701 : $op =~ $self->{like_op} ? belch("Supplying an empty arrayref to '@{[ uc $op]}' is deprecated") && $self->sqlfalse
702 : $op =~ $self->{inequality_op} ? $self->sqltrue
703 : $op =~ $self->{not_like_op} ? belch("Supplying an empty arrayref to '@{[ uc $op]}' is deprecated") && $self->sqltrue
704 : puke "operator '$op' applied on an empty array (field '$k')";
706 return $self->_expand_op_andor($logic => [
715 and exists $vv->{-value}
716 and not defined $vv->{-value}
720 $op =~ /^not$/i ? 'is not' # legacy
721 : $op =~ $self->{equality_op} ? 'is'
722 : $op =~ $self->{like_op} ? belch("Supplying an undefined argument to '@{[ uc $op]}' is deprecated") && 'is'
723 : $op =~ $self->{inequality_op} ? 'is not'
724 : $op =~ $self->{not_like_op} ? belch("Supplying an undefined argument to '@{[ uc $op]}' is deprecated") && 'is not'
725 : puke "unexpected operator '$op' with undef operand";
726 return +{ -op => [ $is.' null', $ik ] };
728 local our $Cur_Col_Meta = $k;
732 $self->_expand_expr($vv)
735 if (ref($v) eq 'ARRAY') {
736 return $self->sqlfalse unless @$v;
737 $self->_debug("ARRAY($k) means distribute over elements");
739 $v->[0] =~ /^-(and|or)$/i
740 ? shift(@{$v = [ @$v ]})
741 : '-'.lc($self->{logic} || 'OR')
743 return $self->_expand_op_andor(
747 if (my $literal = is_literal_value($v)) {
749 belch 'Hash-pairs consisting of an empty string with a literal are deprecated, and will be removed in 2.0: use -and => [ $literal ] instead';
752 my ($sql, @bind) = @$literal;
753 if ($self->{bindtype} eq 'columns') {
755 $self->_assert_bindval_matches_bindtype($_);
758 return +{ -literal => [ $self->_quote($k).' '.$sql, @bind ] };
763 sub _expand_expr_hashpair_op {
764 my ($self, $k, $v) = @_;
767 $op =~ s/^-// if length($op) > 1;
768 $self->_assert_pass_injection_guard($op);
770 # Ops prefixed with -not_ get converted
772 if (my ($rest) = $op =~/^not[_ ](.*)$/) {
775 $self->_expand_expr({ "-${rest}", $v })
779 # the old special op system requires illegality for top-level use
782 (our $Expand_Depth) == 1
783 and List::Util::first { $op =~ $_->{regex} } @{$self->{special_ops}}
785 puke "Illegal use of top-level '-$op'"
788 # the old unary op system means we should touch nothing and let it work
790 if (my $us = List::Util::first { $op =~ $_->{regex} } @{$self->{unary_ops}}) {
791 return { -op => [ $op, $v ] };
794 # an explicit node type is currently assumed to be expanded (this is almost
795 # certainly wrong and there should be expansion anyway)
797 if ($self->{render}{$k}) {
801 # hashref RHS values get expanded and used as op/func args
806 and (keys %$v)[0] =~ /^-/
808 my ($func) = $k =~ /^-(.*)$/;
809 if (List::Util::first { $func =~ $_->{regex} } @{$self->{special_ops}}) {
810 return +{ -op => [ $func, $self->_expand_expr($v) ] };
812 return +{ -func => [ $func, $self->_expand_expr($v) ] };
815 # scalars and literals get simply expanded
817 if (!ref($v) or is_literal_value($v)) {
818 return +{ -op => [ $op, $self->_expand_expr($v) ] };
825 my ($self, $op, $body) = @_;
826 unless (defined($body) or (ref($body) and ref($body) eq 'ARRAY')) {
827 puke "$op requires a single plain scalar argument (a quotable identifier) or an arrayref of identifier parts";
829 my @parts = map split(/\Q${\($self->{name_sep}||'.')}\E/, $_),
830 ref($body) ? @$body : $body;
831 return { -ident => $parts[-1] } if $self->{_dequalify_idents};
832 unless ($self->{quote_char}) {
833 $self->_assert_pass_injection_guard($_) for @parts;
835 return +{ -ident => \@parts };
839 +{ -bind => [ our $Cur_Col_Meta, $_[2] ] };
843 +{ -op => [ 'not', $_[0]->_expand_expr($_[2]) ] };
847 my ($self, undef, $v) = @_;
849 return $self->_expand_expr($v);
851 puke "-bool => undef not supported" unless defined($v);
852 return $self->_expand_ident(-ident => $v);
855 sub _expand_op_andor {
856 my ($self, $logic, $v, $k) = @_;
858 $v = [ map +{ $k, $_ },
860 ? (map +{ $_ => $v->{$_} }, sort keys %$v)
864 my ($logop) = $logic =~ /^-?(.*)$/;
865 if (ref($v) eq 'HASH') {
868 map $self->_expand_expr({ $_ => $v->{$_} }),
872 if (ref($v) eq 'ARRAY') {
873 $logop eq 'and' or $logop eq 'or' or puke "unknown logic: $logop";
876 (ref($_) eq 'ARRAY' and @$_)
877 or (ref($_) eq 'HASH' and %$_)
883 while (my ($el) = splice @expr, 0, 1) {
884 puke "Supplying an empty left hand side argument is not supported in array-pairs"
885 unless defined($el) and length($el);
886 my $elref = ref($el);
888 local our $Expand_Depth = 0;
889 push(@res, grep defined, $self->_expand_expr({ $el, shift(@expr) }));
890 } elsif ($elref eq 'ARRAY') {
891 push(@res, grep defined, $self->_expand_expr($el)) if @$el;
892 } elsif (my $l = is_literal_value($el)) {
893 push @res, { -literal => $l };
894 } elsif ($elref eq 'HASH') {
895 local our $Expand_Depth = 0;
896 push @res, grep defined, $self->_expand_expr($el) if %$el;
902 # return $res[0] if @res == 1;
903 return { -op => [ $logop, @res ] };
908 sub _expand_between {
909 my ($self, $op, $vv, $k) = @_;
910 local our $Cur_Col_Meta = $k;
911 my @rhs = map $self->_expand_expr($_),
912 ref($vv) eq 'ARRAY' ? @$vv : $vv;
914 (@rhs == 1 and ref($rhs[0]) eq 'HASH' and $rhs[0]->{-literal})
916 (@rhs == 2 and defined($rhs[0]) and defined($rhs[1]))
918 puke "Operator '${\uc($op)}' requires either an arrayref with two defined values or expressions, or a single literal scalarref/arrayref-ref";
922 $self->_expand_ident(-ident => $k),
928 my ($self, $op, $vv, $k) = @_;
929 if (my $literal = is_literal_value($vv)) {
930 my ($sql, @bind) = @$literal;
931 my $opened_sql = $self->_open_outer_paren($sql);
933 $op, $self->_expand_ident(-ident => $k),
934 [ { -literal => [ $opened_sql, @bind ] } ]
938 'SQL::Abstract before v1.75 used to generate incorrect SQL when the '
939 . "-${\uc($op)} operator was given an undef-containing list: !!!AUDIT YOUR CODE "
940 . 'AND DATA!!! (the upcoming Data::Query-based version of SQL::Abstract '
941 . 'will emit the logically correct SQL instead of raising this exception)'
943 puke("Argument passed to the '${\uc($op)}' operator can not be undefined")
945 my @rhs = map $self->_expand_expr($_),
946 map { ref($_) ? $_ : { -bind => [ $k, $_ ] } }
947 map { defined($_) ? $_: puke($undef_err) }
948 (ref($vv) eq 'ARRAY' ? @$vv : $vv);
949 return $self->${\($op =~ /^not/ ? 'sqltrue' : 'sqlfalse')} unless @rhs;
953 $self->_expand_ident(-ident => $k),
959 my ($self, $op, $v) = @_;
960 # DBIx::Class requires a nest warning to be emitted once but the private
961 # method it overrode to do so no longer exists
962 if ($self->{is_dbic_sqlmaker}) {
963 unless (our $Nest_Warned) {
965 "-nest in search conditions is deprecated, you most probably wanted:\n"
966 .q|{..., -and => [ \%cond0, \@cond1, \'cond2', \[ 'cond3', [ col => bind ] ], etc. ], ... }|
971 return $self->_expand_expr($v);
975 my ($self, $where, $logic) = @_;
977 # Special case: top level simple string treated as literal
979 my $where_exp = (ref($where)
980 ? $self->_expand_expr($where, $logic)
981 : { -literal => [ $where ] });
983 # dispatch expanded expression
985 my ($sql, @bind) = defined($where_exp) ? $self->render_aqt($where_exp) : (undef);
986 # DBIx::Class used to call _recurse_where in scalar context
987 # something else might too...
989 return ($sql, @bind);
992 belch "Calling _recurse_where in scalar context is deprecated and will go away before 2.0";
998 my ($self, $ident) = @_;
1000 return $self->_convert($self->_quote($ident));
1004 my ($self, $list) = @_;
1005 my @parts = grep length($_->[0]), map [ $self->render_aqt($_) ], @$list;
1006 return join(', ', map $_->[0], @parts), map @{$_}[1..$#$_], @parts;
1010 my ($self, $rest) = @_;
1011 my ($func, @args) = @$rest;
1015 push @arg_sql, shift @x;
1017 } map [ $self->render_aqt($_) ], @args;
1018 return ($self->_sqlcase($func).'('.join(', ', @arg_sql).')', @bind);
1022 my ($self, $bind) = @_;
1023 return ($self->_convert('?'), $self->_bindtype(@$bind));
1026 sub _render_literal {
1027 my ($self, $literal) = @_;
1028 $self->_assert_bindval_matches_bindtype(@{$literal}[1..$#$literal]);
1034 my ($self, $op, $args) = @_;
1035 my ($left, $low, $high) = @$args;
1036 my ($rhsql, @rhbind) = do {
1038 puke "Single arg to between must be a literal"
1039 unless $low->{-literal};
1042 my ($l, $h) = map [ $self->render_aqt($_) ], $low, $high;
1043 (join(' ', $l->[0], $self->_sqlcase('and'), $h->[0]),
1044 @{$l}[1..$#$l], @{$h}[1..$#$h])
1047 my ($lhsql, @lhbind) = $self->render_aqt($left);
1049 join(' ', '(', $lhsql, $self->_sqlcase($op), $rhsql, ')'),
1052 }), 'between', 'not between'),
1054 my ($self, $op, $args) = @_;
1055 my ($lhs, $rhs) = @$args;
1058 my ($sql, @bind) = $self->render_aqt($_);
1059 push @in_bind, @bind;
1062 my ($lhsql, @lbind) = $self->render_aqt($lhs);
1064 $lhsql.' '.$self->_sqlcase($op).' ( '
1065 .join(', ', @in_sql)
1069 }), 'in', 'not in'),
1070 (map +($_ => '_render_unop_postfix'),
1071 'is null', 'is not null', 'asc', 'desc',
1073 (not => '_render_op_not'),
1075 my ($self, $op, $args) = @_;
1076 my @parts = grep length($_->[0]), map [ $self->render_aqt($_) ], @$args;
1077 return '' unless @parts;
1078 return @{$parts[0]} if @parts == 1;
1079 my ($final_sql) = join(
1080 ' '.$self->_sqlcase($op).' ',
1085 map @{$_}[1..$#$_], @parts
1091 my ($self, $v) = @_;
1092 my ($op, @args) = @$v;
1093 if (my $r = $self->{render_op}{$op}) {
1094 return $self->$r($op, \@args);
1096 my $us = List::Util::first { $op =~ $_->{regex} } @{$self->{special_ops}};
1097 if ($us and @args > 1) {
1098 puke "Special op '${op}' requires first value to be identifier"
1099 unless my ($ident) = map $_->{-ident}, grep ref($_) eq 'HASH', $args[0];
1100 my $k = join(($self->{name_sep}||'.'), @$ident);
1101 local our $Expand_Depth = 1;
1102 return $self->${\($us->{handler})}($k, $op, $args[1]);
1104 if (my $us = List::Util::first { $op =~ $_->{regex} } @{$self->{unary_ops}}) {
1105 return $self->${\($us->{handler})}($op, $args[0]);
1108 return $self->_render_unop_prefix($op, \@args);
1110 my @parts = grep length($_->[0]), map [ $self->render_aqt($_) ], @args;
1111 return '' unless @parts;
1112 my ($final_sql) = join(
1113 ' '.$self->_sqlcase($op).' ',
1118 map @{$_}[1..$#$_], @parts
1124 sub _render_op_not {
1125 my ($self, $op, $v) = @_;
1126 my ($sql, @bind) = $self->_render_unop_prefix($op, $v);
1127 return "(${sql})", @bind;
1130 sub _render_unop_prefix {
1131 my ($self, $op, $v) = @_;
1132 my ($expr_sql, @bind) = $self->render_aqt($v->[0]);
1133 my $op_sql = $self->_sqlcase($op);
1134 return ("${op_sql} ${expr_sql}", @bind);
1137 sub _render_unop_postfix {
1138 my ($self, $op, $v) = @_;
1139 my ($expr_sql, @bind) = $self->render_aqt($v->[0]);
1140 my $op_sql = $self->_sqlcase($op);
1141 return ($expr_sql.' '.$op_sql, @bind);
1144 # Some databases (SQLite) treat col IN (1, 2) different from
1145 # col IN ( (1, 2) ). Use this to strip all outer parens while
1146 # adding them back in the corresponding method
1147 sub _open_outer_paren {
1148 my ($self, $sql) = @_;
1150 while (my ($inner) = $sql =~ /^ \s* \( (.*) \) \s* $/xs) {
1152 # there are closing parens inside, need the heavy duty machinery
1153 # to reevaluate the extraction starting from $sql (full reevaluation)
1154 if ($inner =~ /\)/) {
1155 require Text::Balanced;
1157 my (undef, $remainder) = do {
1158 # idiotic design - writes to $@ but *DOES NOT* throw exceptions
1160 Text::Balanced::extract_bracketed($sql, '()', qr/\s*/);
1163 # the entire expression needs to be a balanced bracketed thing
1164 # (after an extract no remainder sans trailing space)
1165 last if defined $remainder and $remainder =~ /\S/;
1175 #======================================================================
1177 #======================================================================
1179 sub _expand_order_by {
1180 my ($self, $arg) = @_;
1182 return unless defined($arg) and not (ref($arg) eq 'ARRAY' and !@$arg);
1184 my $expander = sub {
1185 my ($self, $dir, $expr) = @_;
1186 my @to_expand = ref($expr) eq 'ARRAY' ? @$expr : $expr;
1187 foreach my $arg (@to_expand) {
1191 and grep /^-(asc|desc)$/, keys %$arg
1193 puke "ordering direction hash passed to order by must have exactly one key (-asc or -desc)";
1197 defined($dir) ? { -op => [ $dir =~ /^-?(.*)$/ ,=> $_ ] } : $_
1199 map $self->expand_expr($_, -ident),
1200 map ref($_) eq 'ARRAY' ? @$_ : $_, @to_expand;
1201 return (@exp > 1 ? { -list => \@exp } : $exp[0]);
1204 local @{$self->{expand}}{qw(-asc -desc)} = (($expander) x 2);
1206 return $self->$expander(undef, $arg);
1210 my ($self, $arg) = @_;
1212 return '' unless defined(my $expanded = $self->_expand_order_by($arg));
1214 my ($sql, @bind) = $self->render_aqt($expanded);
1216 return '' unless length($sql);
1218 my $final_sql = $self->_sqlcase(' order by ').$sql;
1220 return wantarray ? ($final_sql, @bind) : $final_sql;
1223 # _order_by no longer needs to call this so doesn't but DBIC uses it.
1225 sub _order_by_chunks {
1226 my ($self, $arg) = @_;
1228 return () unless defined(my $expanded = $self->_expand_order_by($arg));
1230 return $self->_chunkify_order_by($expanded);
1233 sub _chunkify_order_by {
1234 my ($self, $expanded) = @_;
1236 return grep length, $self->render_aqt($expanded)
1237 if $expanded->{-ident} or @{$expanded->{-literal}||[]} == 1;
1240 if (ref() eq 'HASH' and my $l = $_->{-list}) {
1241 return map $self->_chunkify_order_by($_), @$l;
1243 return [ $self->render_aqt($_) ];
1247 #======================================================================
1248 # DATASOURCE (FOR NOW, JUST PLAIN TABLE OR LIST OF TABLES)
1249 #======================================================================
1255 $self->_expand_maybe_list_expr($from, undef, -ident)
1260 #======================================================================
1262 #======================================================================
1264 sub _expand_maybe_list_expr {
1265 my ($self, $expr, $logic, $default) = @_;
1267 if (ref($expr) eq 'ARRAY') {
1269 map $self->expand_expr($_, $default), @$expr
1276 return $self->expand_expr($e, $default);
1279 # highly optimized, as it's called way too often
1281 # my ($self, $label) = @_;
1283 return '' unless defined $_[1];
1284 return ${$_[1]} if ref($_[1]) eq 'SCALAR';
1285 puke 'Identifier cannot be hashref' if ref($_[1]) eq 'HASH';
1287 unless ($_[0]->{quote_char}) {
1288 if (ref($_[1]) eq 'ARRAY') {
1289 return join($_[0]->{name_sep}||'.', @{$_[1]});
1291 $_[0]->_assert_pass_injection_guard($_[1]);
1296 my $qref = ref $_[0]->{quote_char};
1298 !$qref ? ($_[0]->{quote_char}, $_[0]->{quote_char})
1299 : ($qref eq 'ARRAY') ? @{$_[0]->{quote_char}}
1300 : puke "Unsupported quote_char format: $_[0]->{quote_char}";
1302 my $esc = $_[0]->{escape_char} || $r;
1304 # parts containing * are naturally unquoted
1306 $_[0]->{name_sep}||'',
1310 : do { (my $n = $_) =~ s/(\Q$esc\E|\Q$r\E)/$esc$1/g; $l . $n . $r }
1312 (ref($_[1]) eq 'ARRAY'
1316 ? split (/\Q$_[0]->{name_sep}\E/, $_[1] )
1324 # Conversion, if applicable
1326 #my ($self, $arg) = @_;
1327 if ($_[0]->{convert_where}) {
1328 return $_[0]->_sqlcase($_[0]->{convert_where}) .'(' . $_[1] . ')';
1335 #my ($self, $col, @vals) = @_;
1336 # called often - tighten code
1337 return $_[0]->{bindtype} eq 'columns'
1338 ? map {[$_[1], $_]} @_[2 .. $#_]
1343 # Dies if any element of @bind is not in [colname => value] format
1344 # if bindtype is 'columns'.
1345 sub _assert_bindval_matches_bindtype {
1346 # my ($self, @bind) = @_;
1348 if ($self->{bindtype} eq 'columns') {
1350 if (!defined $_ || ref($_) ne 'ARRAY' || @$_ != 2) {
1351 puke "bindtype 'columns' selected, you need to pass: [column_name => bind_value]"
1357 sub _join_sql_clauses {
1358 my ($self, $logic, $clauses_aref, $bind_aref) = @_;
1360 if (@$clauses_aref > 1) {
1361 my $join = " " . $self->_sqlcase($logic) . " ";
1362 my $sql = '( ' . join($join, @$clauses_aref) . ' )';
1363 return ($sql, @$bind_aref);
1365 elsif (@$clauses_aref) {
1366 return ($clauses_aref->[0], @$bind_aref); # no parentheses
1369 return (); # if no SQL, ignore @$bind_aref
1374 # Fix SQL case, if so requested
1376 # LDNOTE: if $self->{case} is true, then it contains 'lower', so we
1377 # don't touch the argument ... crooked logic, but let's not change it!
1378 return $_[0]->{case} ? $_[1] : uc($_[1]);
1382 #======================================================================
1383 # DISPATCHING FROM REFKIND
1384 #======================================================================
1387 my ($self, $data) = @_;
1389 return 'UNDEF' unless defined $data;
1391 # blessed objects are treated like scalars
1392 my $ref = (Scalar::Util::blessed $data) ? '' : ref $data;
1394 return 'SCALAR' unless $ref;
1397 while ($ref eq 'REF') {
1399 $ref = (Scalar::Util::blessed $data) ? '' : ref $data;
1403 return ($ref||'SCALAR') . ('REF' x $n_steps);
1407 my ($self, $data) = @_;
1408 my @try = ($self->_refkind($data));
1409 push @try, 'SCALAR_or_UNDEF' if $try[0] eq 'SCALAR' || $try[0] eq 'UNDEF';
1410 push @try, 'FALLBACK';
1414 sub _METHOD_FOR_refkind {
1415 my ($self, $meth_prefix, $data) = @_;
1418 for (@{$self->_try_refkind($data)}) {
1419 $method = $self->can($meth_prefix."_".$_)
1423 return $method || puke "cannot dispatch on '$meth_prefix' for ".$self->_refkind($data);
1427 sub _SWITCH_refkind {
1428 my ($self, $data, $dispatch_table) = @_;
1431 for (@{$self->_try_refkind($data)}) {
1432 $coderef = $dispatch_table->{$_}
1436 puke "no dispatch entry for ".$self->_refkind($data)
1445 #======================================================================
1446 # VALUES, GENERATE, AUTOLOAD
1447 #======================================================================
1449 # LDNOTE: original code from nwiger, didn't touch code in that section
1450 # I feel the AUTOLOAD stuff should not be the default, it should
1451 # only be activated on explicit demand by user.
1455 my $data = shift || return;
1456 puke "Argument to ", __PACKAGE__, "->values must be a \\%hash"
1457 unless ref $data eq 'HASH';
1460 foreach my $k (sort keys %$data) {
1461 my $v = $data->{$k};
1462 $self->_SWITCH_refkind($v, {
1464 if ($self->{array_datatypes}) { # array datatype
1465 push @all_bind, $self->_bindtype($k, $v);
1467 else { # literal SQL with bind
1468 my ($sql, @bind) = @$v;
1469 $self->_assert_bindval_matches_bindtype(@bind);
1470 push @all_bind, @bind;
1473 ARRAYREFREF => sub { # literal SQL with bind
1474 my ($sql, @bind) = @${$v};
1475 $self->_assert_bindval_matches_bindtype(@bind);
1476 push @all_bind, @bind;
1478 SCALARREF => sub { # literal SQL without bind
1480 SCALAR_or_UNDEF => sub {
1481 push @all_bind, $self->_bindtype($k, $v);
1492 my(@sql, @sqlq, @sqlv);
1496 if ($ref eq 'HASH') {
1497 for my $k (sort keys %$_) {
1500 my $label = $self->_quote($k);
1501 if ($r eq 'ARRAY') {
1502 # literal SQL with bind
1503 my ($sql, @bind) = @$v;
1504 $self->_assert_bindval_matches_bindtype(@bind);
1505 push @sqlq, "$label = $sql";
1507 } elsif ($r eq 'SCALAR') {
1508 # literal SQL without bind
1509 push @sqlq, "$label = $$v";
1511 push @sqlq, "$label = ?";
1512 push @sqlv, $self->_bindtype($k, $v);
1515 push @sql, $self->_sqlcase('set'), join ', ', @sqlq;
1516 } elsif ($ref eq 'ARRAY') {
1517 # unlike insert(), assume these are ONLY the column names, i.e. for SQL
1520 if ($r eq 'ARRAY') { # literal SQL with bind
1521 my ($sql, @bind) = @$v;
1522 $self->_assert_bindval_matches_bindtype(@bind);
1525 } elsif ($r eq 'SCALAR') { # literal SQL without bind
1526 # embedded literal SQL
1533 push @sql, '(' . join(', ', @sqlq) . ')';
1534 } elsif ($ref eq 'SCALAR') {
1538 # strings get case twiddled
1539 push @sql, $self->_sqlcase($_);
1543 my $sql = join ' ', @sql;
1545 # this is pretty tricky
1546 # if ask for an array, return ($stmt, @bind)
1547 # otherwise, s/?/shift @sqlv/ to put it inline
1549 return ($sql, @sqlv);
1551 1 while $sql =~ s/\?/my $d = shift(@sqlv);
1552 ref $d ? $d->[1] : $d/e;
1561 # This allows us to check for a local, then _form, attr
1563 my($name) = $AUTOLOAD =~ /.*::(.+)/;
1564 return $self->generate($name, @_);
1575 SQL::Abstract - Generate SQL from Perl data structures
1581 my $sql = SQL::Abstract->new;
1583 my($stmt, @bind) = $sql->select($source, \@fields, \%where, $order);
1585 my($stmt, @bind) = $sql->insert($table, \%fieldvals || \@values);
1587 my($stmt, @bind) = $sql->update($table, \%fieldvals, \%where);
1589 my($stmt, @bind) = $sql->delete($table, \%where);
1591 # Then, use these in your DBI statements
1592 my $sth = $dbh->prepare($stmt);
1593 $sth->execute(@bind);
1595 # Just generate the WHERE clause
1596 my($stmt, @bind) = $sql->where(\%where, $order);
1598 # Return values in the same order, for hashed queries
1599 # See PERFORMANCE section for more details
1600 my @bind = $sql->values(\%fieldvals);
1604 This module was inspired by the excellent L<DBIx::Abstract>.
1605 However, in using that module I found that what I really wanted
1606 to do was generate SQL, but still retain complete control over my
1607 statement handles and use the DBI interface. So, I set out to
1608 create an abstract SQL generation module.
1610 While based on the concepts used by L<DBIx::Abstract>, there are
1611 several important differences, especially when it comes to WHERE
1612 clauses. I have modified the concepts used to make the SQL easier
1613 to generate from Perl data structures and, IMO, more intuitive.
1614 The underlying idea is for this module to do what you mean, based
1615 on the data structures you provide it. The big advantage is that
1616 you don't have to modify your code every time your data changes,
1617 as this module figures it out.
1619 To begin with, an SQL INSERT is as easy as just specifying a hash
1620 of C<key=value> pairs:
1623 name => 'Jimbo Bobson',
1624 phone => '123-456-7890',
1625 address => '42 Sister Lane',
1626 city => 'St. Louis',
1627 state => 'Louisiana',
1630 The SQL can then be generated with this:
1632 my($stmt, @bind) = $sql->insert('people', \%data);
1634 Which would give you something like this:
1636 $stmt = "INSERT INTO people
1637 (address, city, name, phone, state)
1638 VALUES (?, ?, ?, ?, ?)";
1639 @bind = ('42 Sister Lane', 'St. Louis', 'Jimbo Bobson',
1640 '123-456-7890', 'Louisiana');
1642 These are then used directly in your DBI code:
1644 my $sth = $dbh->prepare($stmt);
1645 $sth->execute(@bind);
1647 =head2 Inserting and Updating Arrays
1649 If your database has array types (like for example Postgres),
1650 activate the special option C<< array_datatypes => 1 >>
1651 when creating the C<SQL::Abstract> object.
1652 Then you may use an arrayref to insert and update database array types:
1654 my $sql = SQL::Abstract->new(array_datatypes => 1);
1656 planets => [qw/Mercury Venus Earth Mars/]
1659 my($stmt, @bind) = $sql->insert('solar_system', \%data);
1663 $stmt = "INSERT INTO solar_system (planets) VALUES (?)"
1665 @bind = (['Mercury', 'Venus', 'Earth', 'Mars']);
1668 =head2 Inserting and Updating SQL
1670 In order to apply SQL functions to elements of your C<%data> you may
1671 specify a reference to an arrayref for the given hash value. For example,
1672 if you need to execute the Oracle C<to_date> function on a value, you can
1673 say something like this:
1677 date_entered => \[ "to_date(?,'MM/DD/YYYY')", "03/02/2003" ],
1680 The first value in the array is the actual SQL. Any other values are
1681 optional and would be included in the bind values array. This gives
1684 my($stmt, @bind) = $sql->insert('people', \%data);
1686 $stmt = "INSERT INTO people (name, date_entered)
1687 VALUES (?, to_date(?,'MM/DD/YYYY'))";
1688 @bind = ('Bill', '03/02/2003');
1690 An UPDATE is just as easy, all you change is the name of the function:
1692 my($stmt, @bind) = $sql->update('people', \%data);
1694 Notice that your C<%data> isn't touched; the module will generate
1695 the appropriately quirky SQL for you automatically. Usually you'll
1696 want to specify a WHERE clause for your UPDATE, though, which is
1697 where handling C<%where> hashes comes in handy...
1699 =head2 Complex where statements
1701 This module can generate pretty complicated WHERE statements
1702 easily. For example, simple C<key=value> pairs are taken to mean
1703 equality, and if you want to see if a field is within a set
1704 of values, you can use an arrayref. Let's say we wanted to
1705 SELECT some data based on this criteria:
1708 requestor => 'inna',
1709 worker => ['nwiger', 'rcwe', 'sfz'],
1710 status => { '!=', 'completed' }
1713 my($stmt, @bind) = $sql->select('tickets', '*', \%where);
1715 The above would give you something like this:
1717 $stmt = "SELECT * FROM tickets WHERE
1718 ( requestor = ? ) AND ( status != ? )
1719 AND ( worker = ? OR worker = ? OR worker = ? )";
1720 @bind = ('inna', 'completed', 'nwiger', 'rcwe', 'sfz');
1722 Which you could then use in DBI code like so:
1724 my $sth = $dbh->prepare($stmt);
1725 $sth->execute(@bind);
1731 The methods are simple. There's one for every major SQL operation,
1732 and a constructor you use first. The arguments are specified in a
1733 similar order for each method (table, then fields, then a where
1734 clause) to try and simplify things.
1736 =head2 new(option => 'value')
1738 The C<new()> function takes a list of options and values, and returns
1739 a new B<SQL::Abstract> object which can then be used to generate SQL
1740 through the methods below. The options accepted are:
1746 If set to 'lower', then SQL will be generated in all lowercase. By
1747 default SQL is generated in "textbook" case meaning something like:
1749 SELECT a_field FROM a_table WHERE some_field LIKE '%someval%'
1751 Any setting other than 'lower' is ignored.
1755 This determines what the default comparison operator is. By default
1756 it is C<=>, meaning that a hash like this:
1758 %where = (name => 'nwiger', email => 'nate@wiger.org');
1760 Will generate SQL like this:
1762 WHERE name = 'nwiger' AND email = 'nate@wiger.org'
1764 However, you may want loose comparisons by default, so if you set
1765 C<cmp> to C<like> you would get SQL such as:
1767 WHERE name like 'nwiger' AND email like 'nate@wiger.org'
1769 You can also override the comparison on an individual basis - see
1770 the huge section on L</"WHERE CLAUSES"> at the bottom.
1772 =item sqltrue, sqlfalse
1774 Expressions for inserting boolean values within SQL statements.
1775 By default these are C<1=1> and C<1=0>. They are used
1776 by the special operators C<-in> and C<-not_in> for generating
1777 correct SQL even when the argument is an empty array (see below).
1781 This determines the default logical operator for multiple WHERE
1782 statements in arrays or hashes. If absent, the default logic is "or"
1783 for arrays, and "and" for hashes. This means that a WHERE
1787 event_date => {'>=', '2/13/99'},
1788 event_date => {'<=', '4/24/03'},
1791 will generate SQL like this:
1793 WHERE event_date >= '2/13/99' OR event_date <= '4/24/03'
1795 This is probably not what you want given this query, though (look
1796 at the dates). To change the "OR" to an "AND", simply specify:
1798 my $sql = SQL::Abstract->new(logic => 'and');
1800 Which will change the above C<WHERE> to:
1802 WHERE event_date >= '2/13/99' AND event_date <= '4/24/03'
1804 The logic can also be changed locally by inserting
1805 a modifier in front of an arrayref:
1807 @where = (-and => [event_date => {'>=', '2/13/99'},
1808 event_date => {'<=', '4/24/03'} ]);
1810 See the L</"WHERE CLAUSES"> section for explanations.
1814 This will automatically convert comparisons using the specified SQL
1815 function for both column and value. This is mostly used with an argument
1816 of C<upper> or C<lower>, so that the SQL will have the effect of
1817 case-insensitive "searches". For example, this:
1819 $sql = SQL::Abstract->new(convert => 'upper');
1820 %where = (keywords => 'MaKe iT CAse inSeNSItive');
1822 Will turn out the following SQL:
1824 WHERE upper(keywords) like upper('MaKe iT CAse inSeNSItive')
1826 The conversion can be C<upper()>, C<lower()>, or any other SQL function
1827 that can be applied symmetrically to fields (actually B<SQL::Abstract> does
1828 not validate this option; it will just pass through what you specify verbatim).
1832 This is a kludge because many databases suck. For example, you can't
1833 just bind values using DBI's C<execute()> for Oracle C<CLOB> or C<BLOB> fields.
1834 Instead, you have to use C<bind_param()>:
1836 $sth->bind_param(1, 'reg data');
1837 $sth->bind_param(2, $lots, {ora_type => ORA_CLOB});
1839 The problem is, B<SQL::Abstract> will normally just return a C<@bind> array,
1840 which loses track of which field each slot refers to. Fear not.
1842 If you specify C<bindtype> in new, you can determine how C<@bind> is returned.
1843 Currently, you can specify either C<normal> (default) or C<columns>. If you
1844 specify C<columns>, you will get an array that looks like this:
1846 my $sql = SQL::Abstract->new(bindtype => 'columns');
1847 my($stmt, @bind) = $sql->insert(...);
1850 [ 'column1', 'value1' ],
1851 [ 'column2', 'value2' ],
1852 [ 'column3', 'value3' ],
1855 You can then iterate through this manually, using DBI's C<bind_param()>.
1857 $sth->prepare($stmt);
1860 my($col, $data) = @$_;
1861 if ($col eq 'details' || $col eq 'comments') {
1862 $sth->bind_param($i, $data, {ora_type => ORA_CLOB});
1863 } elsif ($col eq 'image') {
1864 $sth->bind_param($i, $data, {ora_type => ORA_BLOB});
1866 $sth->bind_param($i, $data);
1870 $sth->execute; # execute without @bind now
1872 Now, why would you still use B<SQL::Abstract> if you have to do this crap?
1873 Basically, the advantage is still that you don't have to care which fields
1874 are or are not included. You could wrap that above C<for> loop in a simple
1875 sub called C<bind_fields()> or something and reuse it repeatedly. You still
1876 get a layer of abstraction over manual SQL specification.
1878 Note that if you set L</bindtype> to C<columns>, the C<\[ $sql, @bind ]>
1879 construct (see L</Literal SQL with placeholders and bind values (subqueries)>)
1880 will expect the bind values in this format.
1884 This is the character that a table or column name will be quoted
1885 with. By default this is an empty string, but you could set it to
1886 the character C<`>, to generate SQL like this:
1888 SELECT `a_field` FROM `a_table` WHERE `some_field` LIKE '%someval%'
1890 Alternatively, you can supply an array ref of two items, the first being the left
1891 hand quote character, and the second the right hand quote character. For
1892 example, you could supply C<['[',']']> for SQL Server 2000 compliant quotes
1893 that generates SQL like this:
1895 SELECT [a_field] FROM [a_table] WHERE [some_field] LIKE '%someval%'
1897 Quoting is useful if you have tables or columns names that are reserved
1898 words in your database's SQL dialect.
1902 This is the character that will be used to escape L</quote_char>s appearing
1903 in an identifier before it has been quoted.
1905 The parameter default in case of a single L</quote_char> character is the quote
1908 When opening-closing-style quoting is used (L</quote_char> is an arrayref)
1909 this parameter defaults to the B<closing (right)> L</quote_char>. Occurrences
1910 of the B<opening (left)> L</quote_char> within the identifier are currently left
1911 untouched. The default for opening-closing-style quotes may change in future
1912 versions, thus you are B<strongly encouraged> to specify the escape character
1917 This is the character that separates a table and column name. It is
1918 necessary to specify this when the C<quote_char> option is selected,
1919 so that tables and column names can be individually quoted like this:
1921 SELECT `table`.`one_field` FROM `table` WHERE `table`.`other_field` = 1
1923 =item injection_guard
1925 A regular expression C<qr/.../> that is applied to any C<-function> and unquoted
1926 column name specified in a query structure. This is a safety mechanism to avoid
1927 injection attacks when mishandling user input e.g.:
1929 my %condition_as_column_value_pairs = get_values_from_user();
1930 $sqla->select( ... , \%condition_as_column_value_pairs );
1932 If the expression matches an exception is thrown. Note that literal SQL
1933 supplied via C<\'...'> or C<\['...']> is B<not> checked in any way.
1935 Defaults to checking for C<;> and the C<GO> keyword (TransactSQL)
1937 =item array_datatypes
1939 When this option is true, arrayrefs in INSERT or UPDATE are
1940 interpreted as array datatypes and are passed directly
1942 When this option is false, arrayrefs are interpreted
1943 as literal SQL, just like refs to arrayrefs
1944 (but this behavior is for backwards compatibility; when writing
1945 new queries, use the "reference to arrayref" syntax
1951 Takes a reference to a list of "special operators"
1952 to extend the syntax understood by L<SQL::Abstract>.
1953 See section L</"SPECIAL OPERATORS"> for details.
1957 Takes a reference to a list of "unary operators"
1958 to extend the syntax understood by L<SQL::Abstract>.
1959 See section L</"UNARY OPERATORS"> for details.
1965 =head2 insert($table, \@values || \%fieldvals, \%options)
1967 This is the simplest function. You simply give it a table name
1968 and either an arrayref of values or hashref of field/value pairs.
1969 It returns an SQL INSERT statement and a list of bind values.
1970 See the sections on L</"Inserting and Updating Arrays"> and
1971 L</"Inserting and Updating SQL"> for information on how to insert
1972 with those data types.
1974 The optional C<\%options> hash reference may contain additional
1975 options to generate the insert SQL. Currently supported options
1982 Takes either a scalar of raw SQL fields, or an array reference of
1983 field names, and adds on an SQL C<RETURNING> statement at the end.
1984 This allows you to return data generated by the insert statement
1985 (such as row IDs) without performing another C<SELECT> statement.
1986 Note, however, this is not part of the SQL standard and may not
1987 be supported by all database engines.
1991 =head2 update($table, \%fieldvals, \%where, \%options)
1993 This takes a table, hashref of field/value pairs, and an optional
1994 hashref L<WHERE clause|/WHERE CLAUSES>. It returns an SQL UPDATE function and a list
1996 See the sections on L</"Inserting and Updating Arrays"> and
1997 L</"Inserting and Updating SQL"> for information on how to insert
1998 with those data types.
2000 The optional C<\%options> hash reference may contain additional
2001 options to generate the update SQL. Currently supported options
2008 See the C<returning> option to
2009 L<insert|/insert($table, \@values || \%fieldvals, \%options)>.
2013 =head2 select($source, $fields, $where, $order)
2015 This returns a SQL SELECT statement and associated list of bind values, as
2016 specified by the arguments:
2022 Specification of the 'FROM' part of the statement.
2023 The argument can be either a plain scalar (interpreted as a table
2024 name, will be quoted), or an arrayref (interpreted as a list
2025 of table names, joined by commas, quoted), or a scalarref
2026 (literal SQL, not quoted).
2030 Specification of the list of fields to retrieve from
2032 The argument can be either an arrayref (interpreted as a list
2033 of field names, will be joined by commas and quoted), or a
2034 plain scalar (literal SQL, not quoted).
2035 Please observe that this API is not as flexible as that of
2036 the first argument C<$source>, for backwards compatibility reasons.
2040 Optional argument to specify the WHERE part of the query.
2041 The argument is most often a hashref, but can also be
2042 an arrayref or plain scalar --
2043 see section L<WHERE clause|/"WHERE CLAUSES"> for details.
2047 Optional argument to specify the ORDER BY part of the query.
2048 The argument can be a scalar, a hashref or an arrayref
2049 -- see section L<ORDER BY clause|/"ORDER BY CLAUSES">
2055 =head2 delete($table, \%where, \%options)
2057 This takes a table name and optional hashref L<WHERE clause|/WHERE CLAUSES>.
2058 It returns an SQL DELETE statement and list of bind values.
2060 The optional C<\%options> hash reference may contain additional
2061 options to generate the delete SQL. Currently supported options
2068 See the C<returning> option to
2069 L<insert|/insert($table, \@values || \%fieldvals, \%options)>.
2073 =head2 where(\%where, $order)
2075 This is used to generate just the WHERE clause. For example,
2076 if you have an arbitrary data structure and know what the
2077 rest of your SQL is going to look like, but want an easy way
2078 to produce a WHERE clause, use this. It returns an SQL WHERE
2079 clause and list of bind values.
2082 =head2 values(\%data)
2084 This just returns the values from the hash C<%data>, in the same
2085 order that would be returned from any of the other above queries.
2086 Using this allows you to markedly speed up your queries if you
2087 are affecting lots of rows. See below under the L</"PERFORMANCE"> section.
2089 =head2 generate($any, 'number', $of, \@data, $struct, \%types)
2091 Warning: This is an experimental method and subject to change.
2093 This returns arbitrarily generated SQL. It's a really basic shortcut.
2094 It will return two different things, depending on return context:
2096 my($stmt, @bind) = $sql->generate('create table', \$table, \@fields);
2097 my $stmt_and_val = $sql->generate('create table', \$table, \@fields);
2099 These would return the following:
2101 # First calling form
2102 $stmt = "CREATE TABLE test (?, ?)";
2103 @bind = (field1, field2);
2105 # Second calling form
2106 $stmt_and_val = "CREATE TABLE test (field1, field2)";
2108 Depending on what you're trying to do, it's up to you to choose the correct
2109 format. In this example, the second form is what you would want.
2113 $sql->generate('alter session', { nls_date_format => 'MM/YY' });
2117 ALTER SESSION SET nls_date_format = 'MM/YY'
2119 You get the idea. Strings get their case twiddled, but everything
2120 else remains verbatim.
2122 =head1 EXPORTABLE FUNCTIONS
2124 =head2 is_plain_value
2126 Determines if the supplied argument is a plain value as understood by this
2131 =item * The value is C<undef>
2133 =item * The value is a non-reference
2135 =item * The value is an object with stringification overloading
2137 =item * The value is of the form C<< { -value => $anything } >>
2141 On failure returns C<undef>, on success returns a B<scalar> reference
2142 to the original supplied argument.
2148 The stringification overloading detection is rather advanced: it takes
2149 into consideration not only the presence of a C<""> overload, but if that
2150 fails also checks for enabled
2151 L<autogenerated versions of C<"">|overload/Magic Autogeneration>, based
2152 on either C<0+> or C<bool>.
2154 Unfortunately testing in the field indicates that this
2155 detection B<< may tickle a latent bug in perl versions before 5.018 >>,
2156 but only when very large numbers of stringifying objects are involved.
2157 At the time of writing ( Sep 2014 ) there is no clear explanation of
2158 the direct cause, nor is there a manageably small test case that reliably
2159 reproduces the problem.
2161 If you encounter any of the following exceptions in B<random places within
2162 your application stack> - this module may be to blame:
2164 Operation "ne": no method found,
2165 left argument in overloaded package <something>,
2166 right argument in overloaded package <something>
2170 Stub found while resolving method "???" overloading """" in package <something>
2172 If you fall victim to the above - please attempt to reduce the problem
2173 to something that could be sent to the L<SQL::Abstract developers
2174 |DBIx::Class/GETTING HELP/SUPPORT>
2175 (either publicly or privately). As a workaround in the meantime you can
2176 set C<$ENV{SQLA_ISVALUE_IGNORE_AUTOGENERATED_STRINGIFICATION}> to a true
2177 value, which will most likely eliminate your problem (at the expense of
2178 not being able to properly detect exotic forms of stringification).
2180 This notice and environment variable will be removed in a future version,
2181 as soon as the underlying problem is found and a reliable workaround is
2186 =head2 is_literal_value
2188 Determines if the supplied argument is a literal value as understood by this
2193 =item * C<\$sql_string>
2195 =item * C<\[ $sql_string, @bind_values ]>
2199 On failure returns C<undef>, on success returns an B<array> reference
2200 containing the unpacked version of the supplied literal SQL and bind values.
2202 =head1 WHERE CLAUSES
2206 This module uses a variation on the idea from L<DBIx::Abstract>. It
2207 is B<NOT>, repeat I<not> 100% compatible. B<The main logic of this
2208 module is that things in arrays are OR'ed, and things in hashes
2211 The easiest way to explain is to show lots of examples. After
2212 each C<%where> hash shown, it is assumed you used:
2214 my($stmt, @bind) = $sql->where(\%where);
2216 However, note that the C<%where> hash can be used directly in any
2217 of the other functions as well, as described above.
2219 =head2 Key-value pairs
2221 So, let's get started. To begin, a simple hash:
2225 status => 'completed'
2228 Is converted to SQL C<key = val> statements:
2230 $stmt = "WHERE user = ? AND status = ?";
2231 @bind = ('nwiger', 'completed');
2233 One common thing I end up doing is having a list of values that
2234 a field can be in. To do this, simply specify a list inside of
2239 status => ['assigned', 'in-progress', 'pending'];
2242 This simple code will create the following:
2244 $stmt = "WHERE user = ? AND ( status = ? OR status = ? OR status = ? )";
2245 @bind = ('nwiger', 'assigned', 'in-progress', 'pending');
2247 A field associated to an empty arrayref will be considered a
2248 logical false and will generate 0=1.
2250 =head2 Tests for NULL values
2252 If the value part is C<undef> then this is converted to SQL <IS NULL>
2261 $stmt = "WHERE user = ? AND status IS NULL";
2264 To test if a column IS NOT NULL:
2268 status => { '!=', undef },
2271 =head2 Specific comparison operators
2273 If you want to specify a different type of operator for your comparison,
2274 you can use a hashref for a given column:
2278 status => { '!=', 'completed' }
2281 Which would generate:
2283 $stmt = "WHERE user = ? AND status != ?";
2284 @bind = ('nwiger', 'completed');
2286 To test against multiple values, just enclose the values in an arrayref:
2288 status => { '=', ['assigned', 'in-progress', 'pending'] };
2290 Which would give you:
2292 "WHERE status = ? OR status = ? OR status = ?"
2295 The hashref can also contain multiple pairs, in which case it is expanded
2296 into an C<AND> of its elements:
2300 status => { '!=', 'completed', -not_like => 'pending%' }
2303 # Or more dynamically, like from a form
2304 $where{user} = 'nwiger';
2305 $where{status}{'!='} = 'completed';
2306 $where{status}{'-not_like'} = 'pending%';
2308 # Both generate this
2309 $stmt = "WHERE user = ? AND status != ? AND status NOT LIKE ?";
2310 @bind = ('nwiger', 'completed', 'pending%');
2313 To get an OR instead, you can combine it with the arrayref idea:
2317 priority => [ { '=', 2 }, { '>', 5 } ]
2320 Which would generate:
2322 $stmt = "WHERE ( priority = ? OR priority > ? ) AND user = ?";
2323 @bind = ('2', '5', 'nwiger');
2325 If you want to include literal SQL (with or without bind values), just use a
2326 scalar reference or reference to an arrayref as the value:
2329 date_entered => { '>' => \["to_date(?, 'MM/DD/YYYY')", "11/26/2008"] },
2330 date_expires => { '<' => \"now()" }
2333 Which would generate:
2335 $stmt = "WHERE date_entered > to_date(?, 'MM/DD/YYYY') AND date_expires < now()";
2336 @bind = ('11/26/2008');
2339 =head2 Logic and nesting operators
2341 In the example above,
2342 there is a subtle trap if you want to say something like
2343 this (notice the C<AND>):
2345 WHERE priority != ? AND priority != ?
2347 Because, in Perl you I<can't> do this:
2349 priority => { '!=' => 2, '!=' => 1 }
2351 As the second C<!=> key will obliterate the first. The solution
2352 is to use the special C<-modifier> form inside an arrayref:
2354 priority => [ -and => {'!=', 2},
2358 Normally, these would be joined by C<OR>, but the modifier tells it
2359 to use C<AND> instead. (Hint: You can use this in conjunction with the
2360 C<logic> option to C<new()> in order to change the way your queries
2361 work by default.) B<Important:> Note that the C<-modifier> goes
2362 B<INSIDE> the arrayref, as an extra first element. This will
2363 B<NOT> do what you think it might:
2365 priority => -and => [{'!=', 2}, {'!=', 1}] # WRONG!
2367 Here is a quick list of equivalencies, since there is some overlap:
2370 status => {'!=', 'completed', 'not like', 'pending%' }
2371 status => [ -and => {'!=', 'completed'}, {'not like', 'pending%'}]
2374 status => {'=', ['assigned', 'in-progress']}
2375 status => [ -or => {'=', 'assigned'}, {'=', 'in-progress'}]
2376 status => [ {'=', 'assigned'}, {'=', 'in-progress'} ]
2380 =head2 Special operators: IN, BETWEEN, etc.
2382 You can also use the hashref format to compare a list of fields using the
2383 C<IN> comparison operator, by specifying the list as an arrayref:
2386 status => 'completed',
2387 reportid => { -in => [567, 2335, 2] }
2390 Which would generate:
2392 $stmt = "WHERE status = ? AND reportid IN (?,?,?)";
2393 @bind = ('completed', '567', '2335', '2');
2395 The reverse operator C<-not_in> generates SQL C<NOT IN> and is used in
2398 If the argument to C<-in> is an empty array, 'sqlfalse' is generated
2399 (by default: C<1=0>). Similarly, C<< -not_in => [] >> generates
2400 'sqltrue' (by default: C<1=1>).
2402 In addition to the array you can supply a chunk of literal sql or
2403 literal sql with bind:
2406 customer => { -in => \[
2407 'SELECT cust_id FROM cust WHERE balance > ?',
2410 status => { -in => \'SELECT status_codes FROM states' },
2416 customer IN ( SELECT cust_id FROM cust WHERE balance > ? )
2417 AND status IN ( SELECT status_codes FROM states )
2421 Finally, if the argument to C<-in> is not a reference, it will be
2422 treated as a single-element array.
2424 Another pair of operators is C<-between> and C<-not_between>,
2425 used with an arrayref of two values:
2429 completion_date => {
2430 -not_between => ['2002-10-01', '2003-02-06']
2436 WHERE user = ? AND completion_date NOT BETWEEN ( ? AND ? )
2438 Just like with C<-in> all plausible combinations of literal SQL
2442 start0 => { -between => [ 1, 2 ] },
2443 start1 => { -between => \["? AND ?", 1, 2] },
2444 start2 => { -between => \"lower(x) AND upper(y)" },
2445 start3 => { -between => [
2447 \["upper(?)", 'stuff' ],
2454 ( start0 BETWEEN ? AND ? )
2455 AND ( start1 BETWEEN ? AND ? )
2456 AND ( start2 BETWEEN lower(x) AND upper(y) )
2457 AND ( start3 BETWEEN lower(x) AND upper(?) )
2459 @bind = (1, 2, 1, 2, 'stuff');
2462 These are the two builtin "special operators"; but the
2463 list can be expanded: see section L</"SPECIAL OPERATORS"> below.
2465 =head2 Unary operators: bool
2467 If you wish to test against boolean columns or functions within your
2468 database you can use the C<-bool> and C<-not_bool> operators. For
2469 example to test the column C<is_user> being true and the column
2470 C<is_enabled> being false you would use:-
2474 -not_bool => 'is_enabled',
2479 WHERE is_user AND NOT is_enabled
2481 If a more complex combination is required, testing more conditions,
2482 then you should use the and/or operators:-
2487 -not_bool => { two=> { -rlike => 'bar' } },
2488 -not_bool => { three => [ { '=', 2 }, { '>', 5 } ] },
2499 (NOT ( three = ? OR three > ? ))
2502 =head2 Nested conditions, -and/-or prefixes
2504 So far, we've seen how multiple conditions are joined with a top-level
2505 C<AND>. We can change this by putting the different conditions we want in
2506 hashes and then putting those hashes in an array. For example:
2511 status => { -like => ['pending%', 'dispatched'] },
2515 status => 'unassigned',
2519 This data structure would create the following:
2521 $stmt = "WHERE ( user = ? AND ( status LIKE ? OR status LIKE ? ) )
2522 OR ( user = ? AND status = ? ) )";
2523 @bind = ('nwiger', 'pending', 'dispatched', 'robot', 'unassigned');
2526 Clauses in hashrefs or arrayrefs can be prefixed with an C<-and> or C<-or>
2527 to change the logic inside:
2533 -and => [ workhrs => {'>', 20}, geo => 'ASIA' ],
2534 -or => { workhrs => {'<', 50}, geo => 'EURO' },
2541 $stmt = "WHERE ( user = ?
2542 AND ( ( workhrs > ? AND geo = ? )
2543 OR ( workhrs < ? OR geo = ? ) ) )";
2544 @bind = ('nwiger', '20', 'ASIA', '50', 'EURO');
2546 =head3 Algebraic inconsistency, for historical reasons
2548 C<Important note>: when connecting several conditions, the C<-and->|C<-or>
2549 operator goes C<outside> of the nested structure; whereas when connecting
2550 several constraints on one column, the C<-and> operator goes
2551 C<inside> the arrayref. Here is an example combining both features:
2554 -and => [a => 1, b => 2],
2555 -or => [c => 3, d => 4],
2556 e => [-and => {-like => 'foo%'}, {-like => '%bar'} ]
2561 WHERE ( ( ( a = ? AND b = ? )
2562 OR ( c = ? OR d = ? )
2563 OR ( e LIKE ? AND e LIKE ? ) ) )
2565 This difference in syntax is unfortunate but must be preserved for
2566 historical reasons. So be careful: the two examples below would
2567 seem algebraically equivalent, but they are not
2570 { -like => 'foo%' },
2571 { -like => '%bar' },
2573 # yields: WHERE ( ( col LIKE ? AND col LIKE ? ) )
2576 { col => { -like => 'foo%' } },
2577 { col => { -like => '%bar' } },
2579 # yields: WHERE ( ( col LIKE ? OR col LIKE ? ) )
2582 =head2 Literal SQL and value type operators
2584 The basic premise of SQL::Abstract is that in WHERE specifications the "left
2585 side" is a column name and the "right side" is a value (normally rendered as
2586 a placeholder). This holds true for both hashrefs and arrayref pairs as you
2587 see in the L</WHERE CLAUSES> examples above. Sometimes it is necessary to
2588 alter this behavior. There are several ways of doing so.
2592 This is a virtual operator that signals the string to its right side is an
2593 identifier (a column name) and not a value. For example to compare two
2594 columns you would write:
2597 priority => { '<', 2 },
2598 requestor => { -ident => 'submitter' },
2603 $stmt = "WHERE priority < ? AND requestor = submitter";
2606 If you are maintaining legacy code you may see a different construct as
2607 described in L</Deprecated usage of Literal SQL>, please use C<-ident> in new
2612 This is a virtual operator that signals that the construct to its right side
2613 is a value to be passed to DBI. This is for example necessary when you want
2614 to write a where clause against an array (for RDBMS that support such
2615 datatypes). For example:
2618 array => { -value => [1, 2, 3] }
2623 $stmt = 'WHERE array = ?';
2624 @bind = ([1, 2, 3]);
2626 Note that if you were to simply say:
2632 the result would probably not be what you wanted:
2634 $stmt = 'WHERE array = ? OR array = ? OR array = ?';
2639 Finally, sometimes only literal SQL will do. To include a random snippet
2640 of SQL verbatim, you specify it as a scalar reference. Consider this only
2641 as a last resort. Usually there is a better way. For example:
2644 priority => { '<', 2 },
2645 requestor => { -in => \'(SELECT name FROM hitmen)' },
2650 $stmt = "WHERE priority < ? AND requestor IN (SELECT name FROM hitmen)"
2653 Note that in this example, you only get one bind parameter back, since
2654 the verbatim SQL is passed as part of the statement.
2658 Never use untrusted input as a literal SQL argument - this is a massive
2659 security risk (there is no way to check literal snippets for SQL
2660 injections and other nastyness). If you need to deal with untrusted input
2661 use literal SQL with placeholders as described next.
2663 =head3 Literal SQL with placeholders and bind values (subqueries)
2665 If the literal SQL to be inserted has placeholders and bind values,
2666 use a reference to an arrayref (yes this is a double reference --
2667 not so common, but perfectly legal Perl). For example, to find a date
2668 in Postgres you can use something like this:
2671 date_column => \[ "= date '2008-09-30' - ?::integer", 10 ]
2676 $stmt = "WHERE ( date_column = date '2008-09-30' - ?::integer )"
2679 Note that you must pass the bind values in the same format as they are returned
2680 by L<where|/where(\%where, $order)>. This means that if you set L</bindtype>
2681 to C<columns>, you must provide the bind values in the
2682 C<< [ column_meta => value ] >> format, where C<column_meta> is an opaque
2683 scalar value; most commonly the column name, but you can use any scalar value
2684 (including references and blessed references), L<SQL::Abstract> will simply
2685 pass it through intact. So if C<bindtype> is set to C<columns> the above
2686 example will look like:
2689 date_column => \[ "= date '2008-09-30' - ?::integer", [ {} => 10 ] ]
2692 Literal SQL is especially useful for nesting parenthesized clauses in the
2693 main SQL query. Here is a first example:
2695 my ($sub_stmt, @sub_bind) = ("SELECT c1 FROM t1 WHERE c2 < ? AND c3 LIKE ?",
2699 bar => \["IN ($sub_stmt)" => @sub_bind],
2704 $stmt = "WHERE (foo = ? AND bar IN (SELECT c1 FROM t1
2705 WHERE c2 < ? AND c3 LIKE ?))";
2706 @bind = (1234, 100, "foo%");
2708 Other subquery operators, like for example C<"E<gt> ALL"> or C<"NOT IN">,
2709 are expressed in the same way. Of course the C<$sub_stmt> and
2710 its associated bind values can be generated through a former call
2713 my ($sub_stmt, @sub_bind)
2714 = $sql->select("t1", "c1", {c2 => {"<" => 100},
2715 c3 => {-like => "foo%"}});
2718 bar => \["> ALL ($sub_stmt)" => @sub_bind],
2721 In the examples above, the subquery was used as an operator on a column;
2722 but the same principle also applies for a clause within the main C<%where>
2723 hash, like an EXISTS subquery:
2725 my ($sub_stmt, @sub_bind)
2726 = $sql->select("t1", "*", {c1 => 1, c2 => \"> t0.c0"});
2727 my %where = ( -and => [
2729 \["EXISTS ($sub_stmt)" => @sub_bind],
2734 $stmt = "WHERE (foo = ? AND EXISTS (SELECT * FROM t1
2735 WHERE c1 = ? AND c2 > t0.c0))";
2739 Observe that the condition on C<c2> in the subquery refers to
2740 column C<t0.c0> of the main query: this is I<not> a bind
2741 value, so we have to express it through a scalar ref.
2742 Writing C<< c2 => {">" => "t0.c0"} >> would have generated
2743 C<< c2 > ? >> with bind value C<"t0.c0"> ... not exactly
2744 what we wanted here.
2746 Finally, here is an example where a subquery is used
2747 for expressing unary negation:
2749 my ($sub_stmt, @sub_bind)
2750 = $sql->where({age => [{"<" => 10}, {">" => 20}]});
2751 $sub_stmt =~ s/^ where //i; # don't want "WHERE" in the subclause
2753 lname => {like => '%son%'},
2754 \["NOT ($sub_stmt)" => @sub_bind],
2759 $stmt = "lname LIKE ? AND NOT ( age < ? OR age > ? )"
2760 @bind = ('%son%', 10, 20)
2762 =head3 Deprecated usage of Literal SQL
2764 Below are some examples of archaic use of literal SQL. It is shown only as
2765 reference for those who deal with legacy code. Each example has a much
2766 better, cleaner and safer alternative that users should opt for in new code.
2772 my %where = ( requestor => \'IS NOT NULL' )
2774 $stmt = "WHERE requestor IS NOT NULL"
2776 This used to be the way of generating NULL comparisons, before the handling
2777 of C<undef> got formalized. For new code please use the superior syntax as
2778 described in L</Tests for NULL values>.
2782 my %where = ( requestor => \'= submitter' )
2784 $stmt = "WHERE requestor = submitter"
2786 This used to be the only way to compare columns. Use the superior L</-ident>
2787 method for all new code. For example an identifier declared in such a way
2788 will be properly quoted if L</quote_char> is properly set, while the legacy
2789 form will remain as supplied.
2793 my %where = ( is_ready => \"", completed => { '>', '2012-12-21' } )
2795 $stmt = "WHERE completed > ? AND is_ready"
2796 @bind = ('2012-12-21')
2798 Using an empty string literal used to be the only way to express a boolean.
2799 For all new code please use the much more readable
2800 L<-bool|/Unary operators: bool> operator.
2806 These pages could go on for a while, since the nesting of the data
2807 structures this module can handle are pretty much unlimited (the
2808 module implements the C<WHERE> expansion as a recursive function
2809 internally). Your best bet is to "play around" with the module a
2810 little to see how the data structures behave, and choose the best
2811 format for your data based on that.
2813 And of course, all the values above will probably be replaced with
2814 variables gotten from forms or the command line. After all, if you
2815 knew everything ahead of time, you wouldn't have to worry about
2816 dynamically-generating SQL and could just hardwire it into your
2819 =head1 ORDER BY CLAUSES
2821 Some functions take an order by clause. This can either be a scalar (just a
2822 column name), a hashref of C<< { -desc => 'col' } >> or C<< { -asc => 'col' }
2823 >>, a scalarref, an arrayref-ref, or an arrayref of any of the previous
2826 Given | Will Generate
2827 ---------------------------------------------------------------
2829 'colA' | ORDER BY colA
2831 [qw/colA colB/] | ORDER BY colA, colB
2833 {-asc => 'colA'} | ORDER BY colA ASC
2835 {-desc => 'colB'} | ORDER BY colB DESC
2837 ['colA', {-asc => 'colB'}] | ORDER BY colA, colB ASC
2839 { -asc => [qw/colA colB/] } | ORDER BY colA ASC, colB ASC
2841 \'colA DESC' | ORDER BY colA DESC
2843 \[ 'FUNC(colA, ?)', $x ] | ORDER BY FUNC(colA, ?)
2844 | /* ...with $x bound to ? */
2847 { -asc => 'colA' }, | colA ASC,
2848 { -desc => [qw/colB/] }, | colB DESC,
2849 { -asc => [qw/colC colD/] },| colC ASC, colD ASC,
2850 \'colE DESC', | colE DESC,
2851 \[ 'FUNC(colF, ?)', $x ], | FUNC(colF, ?)
2852 ] | /* ...with $x bound to ? */
2853 ===============================================================
2857 =head1 SPECIAL OPERATORS
2859 my $sqlmaker = SQL::Abstract->new(special_ops => [
2863 my ($self, $field, $op, $arg) = @_;
2869 handler => 'method_name',
2873 A "special operator" is a SQL syntactic clause that can be
2874 applied to a field, instead of a usual binary operator.
2877 WHERE field IN (?, ?, ?)
2878 WHERE field BETWEEN ? AND ?
2879 WHERE MATCH(field) AGAINST (?, ?)
2881 Special operators IN and BETWEEN are fairly standard and therefore
2882 are builtin within C<SQL::Abstract> (as the overridable methods
2883 C<_where_field_IN> and C<_where_field_BETWEEN>). For other operators,
2884 like the MATCH .. AGAINST example above which is specific to MySQL,
2885 you can write your own operator handlers - supply a C<special_ops>
2886 argument to the C<new> method. That argument takes an arrayref of
2887 operator definitions; each operator definition is a hashref with two
2894 the regular expression to match the operator
2898 Either a coderef or a plain scalar method name. In both cases
2899 the expected return is C<< ($sql, @bind) >>.
2901 When supplied with a method name, it is simply called on the
2902 L<SQL::Abstract> object as:
2904 $self->$method_name($field, $op, $arg)
2908 $field is the LHS of the operator
2909 $op is the part that matched the handler regex
2912 When supplied with a coderef, it is called as:
2914 $coderef->($self, $field, $op, $arg)
2919 For example, here is an implementation
2920 of the MATCH .. AGAINST syntax for MySQL
2922 my $sqlmaker = SQL::Abstract->new(special_ops => [
2924 # special op for MySql MATCH (field) AGAINST(word1, word2, ...)
2925 {regex => qr/^match$/i,
2927 my ($self, $field, $op, $arg) = @_;
2928 $arg = [$arg] if not ref $arg;
2929 my $label = $self->_quote($field);
2930 my ($placeholder) = $self->_convert('?');
2931 my $placeholders = join ", ", (($placeholder) x @$arg);
2932 my $sql = $self->_sqlcase('match') . " ($label) "
2933 . $self->_sqlcase('against') . " ($placeholders) ";
2934 my @bind = $self->_bindtype($field, @$arg);
2935 return ($sql, @bind);
2942 =head1 UNARY OPERATORS
2944 my $sqlmaker = SQL::Abstract->new(unary_ops => [
2948 my ($self, $op, $arg) = @_;
2954 handler => 'method_name',
2958 A "unary operator" is a SQL syntactic clause that can be
2959 applied to a field - the operator goes before the field
2961 You can write your own operator handlers - supply a C<unary_ops>
2962 argument to the C<new> method. That argument takes an arrayref of
2963 operator definitions; each operator definition is a hashref with two
2970 the regular expression to match the operator
2974 Either a coderef or a plain scalar method name. In both cases
2975 the expected return is C<< $sql >>.
2977 When supplied with a method name, it is simply called on the
2978 L<SQL::Abstract> object as:
2980 $self->$method_name($op, $arg)
2984 $op is the part that matched the handler regex
2985 $arg is the RHS or argument of the operator
2987 When supplied with a coderef, it is called as:
2989 $coderef->($self, $op, $arg)
2997 Thanks to some benchmarking by Mark Stosberg, it turns out that
2998 this module is many orders of magnitude faster than using C<DBIx::Abstract>.
2999 I must admit this wasn't an intentional design issue, but it's a
3000 byproduct of the fact that you get to control your C<DBI> handles
3003 To maximize performance, use a code snippet like the following:
3005 # prepare a statement handle using the first row
3006 # and then reuse it for the rest of the rows
3008 for my $href (@array_of_hashrefs) {
3009 $stmt ||= $sql->insert('table', $href);
3010 $sth ||= $dbh->prepare($stmt);
3011 $sth->execute($sql->values($href));
3014 The reason this works is because the keys in your C<$href> are sorted
3015 internally by B<SQL::Abstract>. Thus, as long as your data retains
3016 the same structure, you only have to generate the SQL the first time
3017 around. On subsequent queries, simply use the C<values> function provided
3018 by this module to return your values in the correct order.
3020 However this depends on the values having the same type - if, for
3021 example, the values of a where clause may either have values
3022 (resulting in sql of the form C<column = ?> with a single bind
3023 value), or alternatively the values might be C<undef> (resulting in
3024 sql of the form C<column IS NULL> with no bind value) then the
3025 caching technique suggested will not work.
3029 If you use my C<CGI::FormBuilder> module at all, you'll hopefully
3030 really like this part (I do, at least). Building up a complex query
3031 can be as simple as the following:
3038 use CGI::FormBuilder;
3041 my $form = CGI::FormBuilder->new(...);
3042 my $sql = SQL::Abstract->new;
3044 if ($form->submitted) {
3045 my $field = $form->field;
3046 my $id = delete $field->{id};
3047 my($stmt, @bind) = $sql->update('table', $field, {id => $id});
3050 Of course, you would still have to connect using C<DBI> to run the
3051 query, but the point is that if you make your form look like your
3052 table, the actual query script can be extremely simplistic.
3054 If you're B<REALLY> lazy (I am), check out C<HTML::QuickTable> for
3055 a fast interface to returning and formatting data. I frequently
3056 use these three modules together to write complex database query
3057 apps in under 50 lines.
3059 =head1 HOW TO CONTRIBUTE
3061 Contributions are always welcome, in all usable forms (we especially
3062 welcome documentation improvements). The delivery methods include git-
3063 or unified-diff formatted patches, GitHub pull requests, or plain bug
3064 reports either via RT or the Mailing list. Contributors are generally
3065 granted full access to the official repository after their first several
3066 patches pass successful review.
3068 This project is maintained in a git repository. The code and related tools are
3069 accessible at the following locations:
3073 =item * Official repo: L<git://git.shadowcat.co.uk/dbsrgits/SQL-Abstract.git>
3075 =item * Official gitweb: L<http://git.shadowcat.co.uk/gitweb/gitweb.cgi?p=dbsrgits/SQL-Abstract.git>
3077 =item * GitHub mirror: L<https://github.com/dbsrgits/sql-abstract>
3079 =item * Authorized committers: L<ssh://dbsrgits@git.shadowcat.co.uk/SQL-Abstract.git>
3085 Version 1.50 was a major internal refactoring of C<SQL::Abstract>.
3086 Great care has been taken to preserve the I<published> behavior
3087 documented in previous versions in the 1.* family; however,
3088 some features that were previously undocumented, or behaved
3089 differently from the documentation, had to be changed in order
3090 to clarify the semantics. Hence, client code that was relying
3091 on some dark areas of C<SQL::Abstract> v1.*
3092 B<might behave differently> in v1.50.
3094 The main changes are:
3100 support for literal SQL through the C<< \ [ $sql, @bind ] >> syntax.
3104 support for the { operator => \"..." } construct (to embed literal SQL)
3108 support for the { operator => \["...", @bind] } construct (to embed literal SQL with bind values)
3112 optional support for L<array datatypes|/"Inserting and Updating Arrays">
3116 defensive programming: check arguments
3120 fixed bug with global logic, which was previously implemented
3121 through global variables yielding side-effects. Prior versions would
3122 interpret C<< [ {cond1, cond2}, [cond3, cond4] ] >>
3123 as C<< "(cond1 AND cond2) OR (cond3 AND cond4)" >>.
3124 Now this is interpreted
3125 as C<< "(cond1 AND cond2) OR (cond3 OR cond4)" >>.
3130 fixed semantics of _bindtype on array args
3134 dropped the C<_anoncopy> of the %where tree. No longer necessary,
3135 we just avoid shifting arrays within that tree.
3139 dropped the C<_modlogic> function
3143 =head1 ACKNOWLEDGEMENTS
3145 There are a number of individuals that have really helped out with
3146 this module. Unfortunately, most of them submitted bugs via CPAN
3147 so I have no idea who they are! But the people I do know are:
3149 Ash Berlin (order_by hash term support)
3150 Matt Trout (DBIx::Class support)
3151 Mark Stosberg (benchmarking)
3152 Chas Owens (initial "IN" operator support)
3153 Philip Collins (per-field SQL functions)
3154 Eric Kolve (hashref "AND" support)
3155 Mike Fragassi (enhancements to "BETWEEN" and "LIKE")
3156 Dan Kubb (support for "quote_char" and "name_sep")
3157 Guillermo Roditi (patch to cleanup "IN" and "BETWEEN", fix and tests for _order_by)
3158 Laurent Dami (internal refactoring, extensible list of special operators, literal SQL)
3159 Norbert Buchmuller (support for literal SQL in hashpair, misc. fixes & tests)
3160 Peter Rabbitson (rewrite of SQLA::Test, misc. fixes & tests)
3161 Oliver Charles (support for "RETURNING" after "INSERT")
3167 L<DBIx::Class>, L<DBIx::Abstract>, L<CGI::FormBuilder>, L<HTML::QuickTable>.
3171 Copyright (c) 2001-2007 Nathan Wiger <nwiger@cpan.org>. All Rights Reserved.
3173 This module is actively maintained by Matt Trout <mst@shadowcatsystems.co.uk>
3175 For support, your best bet is to try the C<DBIx::Class> users mailing list.
3176 While not an official support venue, C<DBIx::Class> makes heavy use of
3177 C<SQL::Abstract>, and as such list members there are very familiar with
3178 how to create queries.
3182 This module is free software; you may copy this under the same
3183 terms as perl itself (either the GNU General Public License or
3184 the Artistic License)