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/^ is (?: \s+ not )? $/ix, handler => sub { die "NOPE" }},
42 {regex => qr/^ (?: not \s )? in $/ix, handler => sub { die "NOPE" }},
43 {regex => qr/^ ident $/ix, handler => sub { die "NOPE" }},
44 {regex => qr/^ value $/ix, handler => sub { die "NOPE" }},
47 #======================================================================
48 # DEBUGGING AND ERROR REPORTING
49 #======================================================================
52 return unless $_[0]->{debug}; shift; # a little faster
53 my $func = (caller(1))[3];
54 warn "[$func] ", @_, "\n";
58 my($func) = (caller(1))[3];
59 Carp::carp "[$func] Warning: ", @_;
63 my($func) = (caller(1))[3];
64 Carp::croak "[$func] Fatal: ", @_;
67 sub is_literal_value ($) {
68 ref $_[0] eq 'SCALAR' ? [ ${$_[0]} ]
69 : ( ref $_[0] eq 'REF' and ref ${$_[0]} eq 'ARRAY' ) ? [ @${ $_[0] } ]
73 sub is_undef_value ($) {
77 and exists $_[0]->{-value}
78 and not defined $_[0]->{-value}
82 # FIXME XSify - this can be done so much more efficiently
83 sub is_plain_value ($) {
85 ! length ref $_[0] ? \($_[0])
87 ref $_[0] eq 'HASH' and keys %{$_[0]} == 1
89 exists $_[0]->{-value}
90 ) ? \($_[0]->{-value})
92 # reuse @_ for even moar speedz
93 defined ( $_[1] = Scalar::Util::blessed $_[0] )
95 # deliberately not using Devel::OverloadInfo - the checks we are
96 # intersted in are much more limited than the fullblown thing, and
97 # this is a very hot piece of code
99 # simply using ->can('(""') can leave behind stub methods that
100 # break actually using the overload later (see L<perldiag/Stub
101 # found while resolving method "%s" overloading "%s" in package
102 # "%s"> and the source of overload::mycan())
104 # either has stringification which DBI SHOULD prefer out of the box
105 grep { *{ (qq[${_}::(""]) }{CODE} } @{ $_[2] = mro::get_linear_isa( $_[1] ) }
107 # has nummification or boolification, AND fallback is *not* disabled
109 SQL::Abstract::_ENV_::DETECT_AUTOGENERATED_STRINGIFICATION
112 grep { *{"${_}::(0+"}{CODE} } @{$_[2]}
114 grep { *{"${_}::(bool"}{CODE} } @{$_[2]}
118 # no fallback specified at all
119 ! ( ($_[3]) = grep { *{"${_}::()"}{CODE} } @{$_[2]} )
121 # fallback explicitly undef
122 ! defined ${"$_[3]::()"}
135 #======================================================================
137 #======================================================================
141 my $class = ref($self) || $self;
142 my %opt = (ref $_[0] eq 'HASH') ? %{$_[0]} : @_;
144 # choose our case by keeping an option around
145 delete $opt{case} if $opt{case} && $opt{case} ne 'lower';
147 # default logic for interpreting arrayrefs
148 $opt{logic} = $opt{logic} ? uc $opt{logic} : 'OR';
150 # how to return bind vars
151 $opt{bindtype} ||= 'normal';
153 # default comparison is "=", but can be overridden
156 # try to recognize which are the 'equality' and 'inequality' ops
157 # (temporary quickfix (in 2007), should go through a more seasoned API)
158 $opt{equality_op} = qr/^( \Q$opt{cmp}\E | \= )$/ix;
159 $opt{inequality_op} = qr/^( != | <> )$/ix;
161 $opt{like_op} = qr/^ (is_)?r?like $/xi;
162 $opt{not_like_op} = qr/^ (is_)?not_r?like $/xi;
165 $opt{sqltrue} ||= '1=1';
166 $opt{sqlfalse} ||= '0=1';
169 $opt{special_ops} ||= [];
171 if ($class->isa('DBIx::Class::SQLMaker')) {
172 $opt{warn_once_on_nest} = 1;
173 $opt{disable_old_special_ops} = 1;
177 $opt{unary_ops} ||= [];
179 # rudimentary sanity-check for user supplied bits treated as functions/operators
180 # If a purported function matches this regular expression, an exception is thrown.
181 # Literal SQL is *NOT* subject to this check, only functions (and column names
182 # when quoting is not in effect)
185 # need to guard against ()'s in column names too, but this will break tons of
186 # hacks... ideas anyone?
187 $opt{injection_guard} ||= qr/
193 $opt{expand_unary} = {};
196 not => '_expand_not',
197 bool => '_expand_bool',
198 and => '_expand_op_andor',
199 or => '_expand_op_andor',
200 nest => '_expand_nest',
201 bind => '_expand_bind',
203 not_in => '_expand_in',
204 row => '_expand_row',
205 between => '_expand_between',
206 not_between => '_expand_between',
208 (map +($_ => '_expand_op_is'), ('is', 'is_not')),
209 ident => '_expand_ident',
210 value => '_expand_value',
214 'between' => '_expand_between',
215 'not_between' => '_expand_between',
216 'in' => '_expand_in',
217 'not_in' => '_expand_in',
218 'nest' => '_expand_nest',
219 (map +($_ => '_expand_op_andor'), ('and', 'or')),
220 (map +($_ => '_expand_op_is'), ('is', 'is_not')),
221 'ident' => '_expand_ident',
222 'value' => '_expand_value',
226 (map +($_, "_render_$_"), qw(op func bind ident literal row)),
231 (map +($_ => '_render_op_between'), 'between', 'not_between'),
232 (map +($_ => '_render_op_in'), 'in', 'not_in'),
233 (map +($_ => '_render_unop_postfix'),
234 'is_null', 'is_not_null', 'asc', 'desc',
236 (not => '_render_unop_paren'),
237 (map +($_ => '_render_op_andor'), qw(and or)),
238 ',' => '_render_op_multop',
241 return bless \%opt, $class;
244 sub sqltrue { +{ -literal => [ $_[0]->{sqltrue} ] } }
245 sub sqlfalse { +{ -literal => [ $_[0]->{sqlfalse} ] } }
247 sub _assert_pass_injection_guard {
248 if ($_[1] =~ $_[0]->{injection_guard}) {
249 my $class = ref $_[0];
250 puke "Possible SQL injection attempt '$_[1]'. If this is indeed a part of the "
251 . "desired SQL use literal SQL ( \'...' or \[ '...' ] ) or supply your own "
252 . "{injection_guard} attribute to ${class}->new()"
257 #======================================================================
259 #======================================================================
263 my $table = $self->_table(shift);
264 my $data = shift || return;
267 my $method = $self->_METHOD_FOR_refkind("_insert", $data);
268 my ($sql, @bind) = $self->$method($data);
269 $sql = join " ", $self->_sqlcase('insert into'), $table, $sql;
271 if ($options->{returning}) {
272 my ($s, @b) = $self->_insert_returning($options);
277 return wantarray ? ($sql, @bind) : $sql;
280 # So that subclasses can override INSERT ... RETURNING separately from
281 # UPDATE and DELETE (e.g. DBIx::Class::SQLMaker::Oracle does this)
282 sub _insert_returning { shift->_returning(@_) }
285 my ($self, $options) = @_;
287 my $f = $options->{returning};
289 my ($sql, @bind) = $self->render_aqt(
290 $self->_expand_maybe_list_expr($f, -ident)
293 ? $self->_sqlcase(' returning ') . $sql
294 : ($self->_sqlcase(' returning ').$sql, @bind);
297 sub _insert_HASHREF { # explicit list of fields and then values
298 my ($self, $data) = @_;
300 my @fields = sort keys %$data;
302 my ($sql, @bind) = $self->_insert_values($data);
305 $_ = $self->_quote($_) foreach @fields;
306 $sql = "( ".join(", ", @fields).") ".$sql;
308 return ($sql, @bind);
311 sub _insert_ARRAYREF { # just generate values(?,?) part (no list of fields)
312 my ($self, $data) = @_;
314 # no names (arrayref) so can't generate bindtype
315 $self->{bindtype} ne 'columns'
316 or belch "can't do 'columns' bindtype when called with arrayref";
318 my (@values, @all_bind);
319 foreach my $value (@$data) {
320 my ($values, @bind) = $self->_insert_value(undef, $value);
321 push @values, $values;
322 push @all_bind, @bind;
324 my $sql = $self->_sqlcase('values')." ( ".join(", ", @values)." )";
325 return ($sql, @all_bind);
328 sub _insert_ARRAYREFREF { # literal SQL with bind
329 my ($self, $data) = @_;
331 my ($sql, @bind) = @${$data};
332 $self->_assert_bindval_matches_bindtype(@bind);
334 return ($sql, @bind);
338 sub _insert_SCALARREF { # literal SQL without bind
339 my ($self, $data) = @_;
345 my ($self, $data) = @_;
347 my (@values, @all_bind);
348 foreach my $column (sort keys %$data) {
349 my ($values, @bind) = $self->_insert_value($column, $data->{$column});
350 push @values, $values;
351 push @all_bind, @bind;
353 my $sql = $self->_sqlcase('values')." ( ".join(", ", @values)." )";
354 return ($sql, @all_bind);
358 my ($self, $column, $v) = @_;
360 return $self->render_aqt(
361 $self->_expand_insert_value($column, $v)
365 sub _expand_insert_value {
366 my ($self, $column, $v) = @_;
368 if (ref($v) eq 'ARRAY') {
369 if ($self->{array_datatypes}) {
370 return +{ -bind => [ $column, $v ] };
372 my ($sql, @bind) = @$v;
373 $self->_assert_bindval_matches_bindtype(@bind);
374 return +{ -literal => $v };
376 if (ref($v) eq 'HASH') {
377 if (grep !/^-/, keys %$v) {
378 belch "HASH ref as bind value in insert is not supported";
379 return +{ -bind => [ $column, $v ] };
383 return +{ -bind => [ $column, undef ] };
385 local our $Cur_Col_Meta = $column;
386 return $self->expand_expr($v);
391 #======================================================================
393 #======================================================================
398 my $table = $self->_table(shift);
399 my $data = shift || return;
403 # first build the 'SET' part of the sql statement
404 puke "Unsupported data type specified to \$sql->update"
405 unless ref $data eq 'HASH';
407 my ($sql, @all_bind) = $self->_update_set_values($data);
408 $sql = $self->_sqlcase('update ') . $table . $self->_sqlcase(' set ')
412 my($where_sql, @where_bind) = $self->where($where);
414 push @all_bind, @where_bind;
417 if ($options->{returning}) {
418 my ($returning_sql, @returning_bind) = $self->_update_returning($options);
419 $sql .= $returning_sql;
420 push @all_bind, @returning_bind;
423 return wantarray ? ($sql, @all_bind) : $sql;
426 sub _update_set_values {
427 my ($self, $data) = @_;
429 return $self->render_aqt(
430 $self->_expand_update_set_values($data),
434 sub _expand_update_set_values {
435 my ($self, $data) = @_;
436 $self->_expand_maybe_list_expr( [
439 $set = { -bind => $_ } unless defined $set;
440 +{ -op => [ '=', $self->_expand_ident(-ident => $k), $set ] };
446 ? ($self->{array_datatypes}
447 ? [ $k, +{ -bind => [ $k, $v ] } ]
448 : [ $k, +{ -literal => $v } ])
450 local our $Cur_Col_Meta = $k;
451 [ $k, $self->_expand_expr($v) ]
458 # So that subclasses can override UPDATE ... RETURNING separately from
460 sub _update_returning { shift->_returning(@_) }
464 #======================================================================
466 #======================================================================
471 my $table = $self->_table(shift);
472 my $fields = shift || '*';
476 my ($fields_sql, @bind) = $self->_select_fields($fields);
478 my ($where_sql, @where_bind) = $self->where($where, $order);
479 push @bind, @where_bind;
481 my $sql = join(' ', $self->_sqlcase('select'), $fields_sql,
482 $self->_sqlcase('from'), $table)
485 return wantarray ? ($sql, @bind) : $sql;
489 my ($self, $fields) = @_;
490 return $fields unless ref($fields);
491 return $self->render_aqt(
492 $self->_expand_maybe_list_expr($fields, '-ident')
496 #======================================================================
498 #======================================================================
503 my $table = $self->_table(shift);
507 my($where_sql, @bind) = $self->where($where);
508 my $sql = $self->_sqlcase('delete from ') . $table . $where_sql;
510 if ($options->{returning}) {
511 my ($returning_sql, @returning_bind) = $self->_delete_returning($options);
512 $sql .= $returning_sql;
513 push @bind, @returning_bind;
516 return wantarray ? ($sql, @bind) : $sql;
519 # So that subclasses can override DELETE ... RETURNING separately from
521 sub _delete_returning { shift->_returning(@_) }
525 #======================================================================
527 #======================================================================
531 # Finally, a separate routine just to handle WHERE clauses
533 my ($self, $where, $order) = @_;
535 local $self->{convert_where} = $self->{convert};
538 my ($sql, @bind) = defined($where)
539 ? $self->_recurse_where($where)
541 $sql = (defined $sql and length $sql) ? $self->_sqlcase(' where ') . "( $sql )" : '';
545 my ($order_sql, @order_bind) = $self->_order_by($order);
547 push @bind, @order_bind;
550 return wantarray ? ($sql, @bind) : $sql;
553 { our $Default_Scalar_To = -value }
556 my ($self, $expr, $default_scalar_to) = @_;
557 local our $Default_Scalar_To = $default_scalar_to if $default_scalar_to;
558 $self->_expand_expr($expr);
562 my ($self, $aqt) = @_;
563 my ($k, $v, @rest) = %$aqt;
565 die "Not a node type: $k" unless $k =~ s/^-//;
566 if (my $meth = $self->{render}{$k}) {
567 return $self->$meth($v);
569 die "notreached: $k";
573 my ($self, $expr) = @_;
574 $self->render_aqt($self->expand_expr($expr));
578 my ($self, $raw) = @_;
579 s/^-(?=.)//, s/\s+/_/g for my $op = lc $raw;
584 my ($self, $expr) = @_;
585 our $Expand_Depth ||= 0; local $Expand_Depth = $Expand_Depth + 1;
586 return undef unless defined($expr);
587 if (ref($expr) eq 'HASH') {
588 return undef unless my $kc = keys %$expr;
590 return $self->_expand_op_andor(and => $expr);
592 my ($key, $value) = %$expr;
593 if ($key =~ /^-/ and $key =~ s/ [_\s]? \d+ $//x ) {
594 belch 'Use of [and|or|nest]_N modifiers is deprecated and will be removed in SQLA v2.0. '
595 . "You probably wanted ...-and => [ $key => COND1, $key => COND2 ... ]";
597 return $self->_expand_hashpair($key, $value);
599 if (ref($expr) eq 'ARRAY') {
600 return $self->_expand_op_andor(lc($self->{logic}), $expr);
602 if (my $literal = is_literal_value($expr)) {
603 return +{ -literal => $literal };
605 if (!ref($expr) or Scalar::Util::blessed($expr)) {
606 return $self->_expand_scalar($expr);
611 sub _expand_hashpair {
612 my ($self, $k, $v) = @_;
613 unless (defined($k) and length($k)) {
614 if (defined($k) and my $literal = is_literal_value($v)) {
615 belch 'Hash-pairs consisting of an empty string with a literal are deprecated, and will be removed in 2.0: use -and => [ $literal ] instead';
616 return { -literal => $literal };
618 puke "Supplying an empty left hand side argument is not supported";
621 return $self->_expand_hashpair_op($k, $v);
623 return $self->_expand_hashpair_ident($k, $v);
626 sub _expand_hashpair_ident {
627 my ($self, $k, $v) = @_;
629 local our $Cur_Col_Meta = $k;
631 # hash with multiple or no elements is andor
633 if (ref($v) eq 'HASH' and keys %$v != 1) {
634 return $self->_expand_op_andor(and => $v, $k);
637 # undef needs to be re-sent with cmp to achieve IS/IS NOT NULL
639 if (is_undef_value($v)) {
640 return $self->_expand_hashpair_cmp($k => undef);
643 # scalars and objects get expanded as whatever requested or values
645 if (!ref($v) or Scalar::Util::blessed($v)) {
646 return $self->_expand_hashpair_scalar($k, $v);
649 # single key hashref is a hashtriple
651 if (ref($v) eq 'HASH') {
652 return $self->_expand_hashtriple($k, %$v);
655 # arrayref needs re-engineering over the elements
657 if (ref($v) eq 'ARRAY') {
658 return $self->sqlfalse unless @$v;
659 $self->_debug("ARRAY($k) means distribute over elements");
661 $v->[0] =~ /^-(and|or)$/i
662 ? (shift(@{$v = [ @$v ]}), $1)
663 : lc($self->{logic} || 'OR')
665 return $self->_expand_op_andor(
670 if (my $literal = is_literal_value($v)) {
672 belch 'Hash-pairs consisting of an empty string with a literal are deprecated, and will be removed in 2.0: use -and => [ $literal ] instead';
675 my ($sql, @bind) = @$literal;
676 if ($self->{bindtype} eq 'columns') {
678 $self->_assert_bindval_matches_bindtype($_);
681 return +{ -literal => [ $self->_quote($k).' '.$sql, @bind ] };
687 my ($self, $expr) = @_;
689 return $self->_expand_expr({ (our $Default_Scalar_To) => $expr });
692 sub _expand_hashpair_scalar {
693 my ($self, $k, $v) = @_;
695 return $self->_expand_hashpair_cmp(
696 $k, $self->_expand_scalar($v),
700 sub _expand_hashpair_op {
701 my ($self, $k, $v) = @_;
703 $self->_assert_pass_injection_guard($k =~ /\A-(.*)\Z/s);
705 my $op = $self->_normalize_op($k);
707 if (my $exp = $self->{expand}{$op}) {
708 return $self->$exp($op, $v);
711 # Ops prefixed with -not_ get converted
713 if (my ($rest) = $op =~/^not_(.*)$/) {
716 $self->_expand_expr({ "-${rest}", $v })
722 my $op = join(' ', split '_', $op);
724 # the old special op system requires illegality for top-level use
727 (our $Expand_Depth) == 1
729 List::Util::first { $op =~ $_->{regex} } @{$self->{special_ops}}
731 $self->{disable_old_special_ops}
732 and List::Util::first { $op =~ $_->{regex} } @BUILTIN_SPECIAL_OPS
736 puke "Illegal use of top-level '-$op'"
739 # the old unary op system means we should touch nothing and let it work
741 if (my $us = List::Util::first { $op =~ $_->{regex} } @{$self->{unary_ops}}) {
742 return { -op => [ $op, $v ] };
746 # an explicit node type is currently assumed to be expanded (this is almost
747 # certainly wrong and there should be expansion anyway)
749 if ($self->{render}{$op}) {
753 # hashref RHS values get expanded and used as op/func args
758 and (keys %$v)[0] =~ /^-/
760 my ($func) = $k =~ /^-(.*)$/;
762 if (List::Util::first { $func =~ $_->{regex} } @{$self->{special_ops}}) {
763 return +{ -op => [ $func, $self->_expand_expr($v) ] };
766 return +{ -func => [ $func, $self->_expand_expr($v) ] };
769 # scalars and literals get simply expanded
771 if (!ref($v) or is_literal_value($v)) {
772 return +{ -op => [ $op, $self->_expand_expr($v) ] };
778 sub _expand_hashpair_cmp {
779 my ($self, $k, $v) = @_;
780 $self->_expand_hashtriple($k, $self->{cmp}, $v);
783 sub _expand_hashtriple {
784 my ($self, $k, $vk, $vv) = @_;
786 my $ik = $self->_expand_ident(-ident => $k);
788 my $op = $self->_normalize_op($vk);
789 $self->_assert_pass_injection_guard($op);
791 if ($op =~ s/ _? \d+ $//x ) {
792 return $self->_expand_expr($k, { $vk, $vv });
794 if (my $x = $self->{expand_op}{$op}) {
795 local our $Cur_Col_Meta = $k;
796 return $self->$x($op, $vv, $k);
800 my $op = join(' ', split '_', $op);
802 if (my $us = List::Util::first { $op =~ $_->{regex} } @{$self->{special_ops}}) {
803 return { -op => [ $op, $ik, $vv ] };
805 if (my $us = List::Util::first { $op =~ $_->{regex} } @{$self->{unary_ops}}) {
809 { -op => [ $op, $vv ] }
813 if (ref($vv) eq 'ARRAY') {
815 my $logic = (defined($raw[0]) and $raw[0] =~ /^-(and|or)$/i)
816 ? (shift(@raw), $1) : 'or';
817 my @values = map +{ $vk => $_ }, @raw;
819 $op =~ $self->{inequality_op}
820 or $op =~ $self->{not_like_op}
822 if (lc($logic) eq 'or' and @values > 1) {
823 belch "A multi-element arrayref as an argument to the inequality op '${\uc(join ' ', split '_', $op)}' "
824 . 'is technically equivalent to an always-true 1=1 (you probably wanted '
825 . "to say ...{ \$inequality_op => [ -and => \@values ] }... instead)"
830 # try to DWIM on equality operators
831 return ($self->_dwim_op_to_is($op,
832 "Supplying an empty arrayref to '%s' is deprecated",
833 "operator '%s' applied on an empty array (field '$k')"
834 ) ? $self->sqlfalse : $self->sqltrue);
836 return $self->_expand_op_andor($logic => \@values, $k);
838 if (is_undef_value($vv)) {
839 my $is = ($self->_dwim_op_to_is($op,
840 "Supplying an undefined argument to '%s' is deprecated",
841 "unexpected operator '%s' with undef operand",
842 ) ? 'is' : 'is not');
844 return $self->_expand_hashpair($k => { $is, undef });
846 local our $Cur_Col_Meta = $k;
850 $self->_expand_expr($vv)
855 my ($self, $raw, $empty, $fail) = @_;
857 my $op = $self->_normalize_op($raw);
859 if ($op =~ /^not$/i) {
862 if ($op =~ $self->{equality_op}) {
865 if ($op =~ $self->{like_op}) {
866 belch(sprintf $empty, uc(join ' ', split '_', $op));
869 if ($op =~ $self->{inequality_op}) {
872 if ($op =~ $self->{not_like_op}) {
873 belch(sprintf $empty, uc(join ' ', split '_', $op));
876 puke(sprintf $fail, $op);
880 my ($self, undef, $body, $k) = @_;
881 return $self->_expand_hashpair_cmp(
882 $k, { -ident => $body }
884 unless (defined($body) or (ref($body) and ref($body) eq 'ARRAY')) {
885 puke "-ident requires a single plain scalar argument (a quotable identifier) or an arrayref of identifier parts";
887 my @parts = map split(/\Q${\($self->{name_sep}||'.')}\E/, $_),
888 ref($body) ? @$body : $body;
889 return { -ident => $parts[-1] } if $self->{_dequalify_idents};
890 unless ($self->{quote_char}) {
891 $self->_assert_pass_injection_guard($_) for @parts;
893 return +{ -ident => \@parts };
897 return $_[0]->_expand_hashpair_cmp(
898 $_[3], { -value => $_[2] },
900 +{ -bind => [ our $Cur_Col_Meta, $_[2] ] };
904 +{ -op => [ 'not', $_[0]->_expand_expr($_[2]) ] };
908 my ($self, undef, $args) = @_;
909 +{ -row => [ map $self->expand_expr($_), @$args ] };
913 my ($self, undef, $args) = @_;
914 my ($op, @opargs) = @$args;
915 +{ -op => [ $op, map $self->expand_expr($_), @opargs ] };
919 my ($self, undef, $v) = @_;
921 return $self->_expand_expr($v);
923 puke "-bool => undef not supported" unless defined($v);
924 return $self->_expand_ident(-ident => $v);
927 sub _expand_op_andor {
928 my ($self, $logop, $v, $k) = @_;
930 $v = [ map +{ $k, $_ },
932 ? (map +{ $_ => $v->{$_} }, sort keys %$v)
936 if (ref($v) eq 'HASH') {
937 return undef unless keys %$v;
940 map $self->_expand_expr({ $_ => $v->{$_} }),
944 if (ref($v) eq 'ARRAY') {
945 $logop eq 'and' or $logop eq 'or' or puke "unknown logic: $logop";
948 (ref($_) eq 'ARRAY' and @$_)
949 or (ref($_) eq 'HASH' and %$_)
955 while (my ($el) = splice @expr, 0, 1) {
956 puke "Supplying an empty left hand side argument is not supported in array-pairs"
957 unless defined($el) and length($el);
958 my $elref = ref($el);
960 local our $Expand_Depth = 0;
961 push(@res, grep defined, $self->_expand_expr({ $el, shift(@expr) }));
962 } elsif ($elref eq 'ARRAY') {
963 push(@res, grep defined, $self->_expand_expr($el)) if @$el;
964 } elsif (my $l = is_literal_value($el)) {
965 push @res, { -literal => $l };
966 } elsif ($elref eq 'HASH') {
967 local our $Expand_Depth = 0;
968 push @res, grep defined, $self->_expand_expr($el) if %$el;
974 # return $res[0] if @res == 1;
975 return { -op => [ $logop, @res ] };
981 my ($self, $op, $vv, $k) = @_;
982 ($k, $vv) = @$vv unless defined $k;
983 puke "$op can only take undef as argument"
987 and exists($vv->{-value})
988 and !defined($vv->{-value})
990 return +{ -op => [ $op.'_null', $self->expand_expr($k, -ident) ] };
993 sub _expand_between {
994 my ($self, $op, $vv, $k) = @_;
995 $k = shift @{$vv = [ @$vv ]} unless defined $k;
996 my @rhs = map $self->_expand_expr($_),
997 ref($vv) eq 'ARRAY' ? @$vv : $vv;
999 (@rhs == 1 and ref($rhs[0]) eq 'HASH' and $rhs[0]->{-literal})
1001 (@rhs == 2 and defined($rhs[0]) and defined($rhs[1]))
1003 puke "Operator '${\uc($op)}' requires either an arrayref with two defined values or expressions, or a single literal scalarref/arrayref-ref";
1007 $self->expand_expr(ref($k) ? $k : { -ident => $k }),
1013 my ($self, $op, $vv, $k) = @_;
1014 $k = shift @{$vv = [ @$vv ]} unless defined $k;
1015 if (my $literal = is_literal_value($vv)) {
1016 my ($sql, @bind) = @$literal;
1017 my $opened_sql = $self->_open_outer_paren($sql);
1019 $op, $self->expand_expr($k, -ident),
1020 [ { -literal => [ $opened_sql, @bind ] } ]
1024 'SQL::Abstract before v1.75 used to generate incorrect SQL when the '
1025 . "-${\uc($op)} operator was given an undef-containing list: !!!AUDIT YOUR CODE "
1026 . 'AND DATA!!! (the upcoming Data::Query-based version of SQL::Abstract '
1027 . 'will emit the logically correct SQL instead of raising this exception)'
1029 puke("Argument passed to the '${\uc($op)}' operator can not be undefined")
1031 my @rhs = map $self->expand_expr($_, -value),
1032 map { defined($_) ? $_: puke($undef_err) }
1033 (ref($vv) eq 'ARRAY' ? @$vv : $vv);
1034 return $self->${\($op =~ /^not/ ? 'sqltrue' : 'sqlfalse')} unless @rhs;
1038 $self->expand_expr($k, -ident),
1044 my ($self, undef, $v) = @_;
1045 # DBIx::Class requires a nest warning to be emitted once but the private
1046 # method it overrode to do so no longer exists
1047 if ($self->{warn_once_on_nest}) {
1048 unless (our $Nest_Warned) {
1050 "-nest in search conditions is deprecated, you most probably wanted:\n"
1051 .q|{..., -and => [ \%cond0, \@cond1, \'cond2', \[ 'cond3', [ col => bind ] ], etc. ], ... }|
1056 return $self->_expand_expr($v);
1060 my ($self, undef, $bind) = @_;
1061 return { -bind => $bind };
1064 sub _recurse_where {
1065 my ($self, $where, $logic) = @_;
1067 # Special case: top level simple string treated as literal
1069 my $where_exp = (ref($where)
1070 ? $self->_expand_expr($where, $logic)
1071 : { -literal => [ $where ] });
1073 # dispatch expanded expression
1075 my ($sql, @bind) = defined($where_exp) ? $self->render_aqt($where_exp) : (undef);
1076 # DBIx::Class used to call _recurse_where in scalar context
1077 # something else might too...
1079 return ($sql, @bind);
1082 belch "Calling _recurse_where in scalar context is deprecated and will go away before 2.0";
1088 my ($self, $ident) = @_;
1090 return $self->_convert($self->_quote($ident));
1094 my ($self, $values) = @_;
1095 my ($sql, @bind) = $self->_render_op([ ',', @$values ]);
1096 return "($sql)", @bind;
1100 my ($self, $rest) = @_;
1101 my ($func, @args) = @$rest;
1105 push @arg_sql, shift @x;
1107 } map [ $self->render_aqt($_) ], @args;
1108 return ($self->_sqlcase($func).'('.join(', ', @arg_sql).')', @bind);
1112 my ($self, $bind) = @_;
1113 return ($self->_convert('?'), $self->_bindtype(@$bind));
1116 sub _render_literal {
1117 my ($self, $literal) = @_;
1118 $self->_assert_bindval_matches_bindtype(@{$literal}[1..$#$literal]);
1123 my ($self, $v) = @_;
1124 my ($op, @args) = @$v;
1125 if (my $r = $self->{render_op}{$op}) {
1126 return $self->$r($op, \@args);
1131 my $op = join(' ', split '_', $op);
1133 my $ss = List::Util::first { $op =~ $_->{regex} } @{$self->{special_ops}};
1134 if ($ss and @args > 1) {
1135 puke "Special op '${op}' requires first value to be identifier"
1136 unless my ($ident) = map $_->{-ident}, grep ref($_) eq 'HASH', $args[0];
1137 my $k = join(($self->{name_sep}||'.'), @$ident);
1138 local our $Expand_Depth = 1;
1139 return $self->${\($ss->{handler})}($k, $op, $args[1]);
1141 if (my $us = List::Util::first { $op =~ $_->{regex} } @{$self->{unary_ops}}) {
1142 return $self->${\($us->{handler})}($op, $args[0]);
1145 return $self->_render_unop_paren($op, \@args);
1149 return $self->_render_unop_prefix($op, \@args);
1151 return $self->_render_op_multop($op, \@args);
1157 sub _render_op_between {
1158 my ($self, $op, $args) = @_;
1159 my ($left, $low, $high) = @$args;
1160 my ($rhsql, @rhbind) = do {
1162 puke "Single arg to between must be a literal"
1163 unless $low->{-literal};
1166 my ($l, $h) = map [ $self->render_aqt($_) ], $low, $high;
1167 (join(' ', $l->[0], $self->_sqlcase('and'), $h->[0]),
1168 @{$l}[1..$#$l], @{$h}[1..$#$h])
1171 my ($lhsql, @lhbind) = $self->render_aqt($left);
1175 $self->_sqlcase(join ' ', split '_', $op),
1183 my ($self, $op, $args) = @_;
1184 my ($lhs, $rhs) = @$args;
1187 my ($sql, @bind) = $self->render_aqt($_);
1188 push @in_bind, @bind;
1191 my ($lhsql, @lbind) = $self->render_aqt($lhs);
1193 $lhsql.' '.$self->_sqlcase(join ' ', split '_', $op).' ( '
1194 .join(', ', @in_sql)
1200 sub _render_op_andor {
1201 my ($self, $op, $args) = @_;
1202 my @parts = grep length($_->[0]), map [ $self->render_aqt($_) ], @$args;
1203 return '' unless @parts;
1204 return @{$parts[0]} if @parts == 1;
1205 my ($sql, @bind) = $self->_render_op_multop($op, $args);
1206 return '( '.$sql.' )', @bind;
1209 sub _render_op_multop {
1210 my ($self, $op, $args) = @_;
1211 my @parts = grep length($_->[0]), map [ $self->render_aqt($_) ], @$args;
1212 return '' unless @parts;
1213 return @{$parts[0]} if @parts == 1;
1214 my ($final_sql) = join(
1215 ($op eq ',' ? '' : ' ').$self->_sqlcase(join ' ', split '_', $op).' ',
1220 map @{$_}[1..$#$_], @parts
1223 sub _render_unop_paren {
1224 my ($self, $op, $v) = @_;
1225 my ($sql, @bind) = $self->_render_unop_prefix($op, $v);
1226 return "(${sql})", @bind;
1229 sub _render_unop_prefix {
1230 my ($self, $op, $v) = @_;
1231 my ($expr_sql, @bind) = $self->render_aqt($v->[0]);
1233 my $op_sql = $self->_sqlcase($op); # join ' ', split '_', $op);
1234 return ("${op_sql} ${expr_sql}", @bind);
1237 sub _render_unop_postfix {
1238 my ($self, $op, $v) = @_;
1239 my ($expr_sql, @bind) = $self->render_aqt($v->[0]);
1240 my $op_sql = $self->_sqlcase(join ' ', split '_', $op);
1241 return ($expr_sql.' '.$op_sql, @bind);
1244 # Some databases (SQLite) treat col IN (1, 2) different from
1245 # col IN ( (1, 2) ). Use this to strip all outer parens while
1246 # adding them back in the corresponding method
1247 sub _open_outer_paren {
1248 my ($self, $sql) = @_;
1250 while (my ($inner) = $sql =~ /^ \s* \( (.*) \) \s* $/xs) {
1252 # there are closing parens inside, need the heavy duty machinery
1253 # to reevaluate the extraction starting from $sql (full reevaluation)
1254 if ($inner =~ /\)/) {
1255 require Text::Balanced;
1257 my (undef, $remainder) = do {
1258 # idiotic design - writes to $@ but *DOES NOT* throw exceptions
1260 Text::Balanced::extract_bracketed($sql, '()', qr/\s*/);
1263 # the entire expression needs to be a balanced bracketed thing
1264 # (after an extract no remainder sans trailing space)
1265 last if defined $remainder and $remainder =~ /\S/;
1275 #======================================================================
1277 #======================================================================
1279 sub _expand_order_by {
1280 my ($self, $arg) = @_;
1282 return unless defined($arg) and not (ref($arg) eq 'ARRAY' and !@$arg);
1284 my $expander = sub {
1285 my ($self, $dir, $expr) = @_;
1286 my @to_expand = ref($expr) eq 'ARRAY' ? @$expr : $expr;
1287 foreach my $arg (@to_expand) {
1291 and grep /^-(asc|desc)$/, keys %$arg
1293 puke "ordering direction hash passed to order by must have exactly one key (-asc or -desc)";
1297 defined($dir) ? { -op => [ $dir =~ /^-?(.*)$/ ,=> $_ ] } : $_
1299 map $self->expand_expr($_, -ident),
1300 map ref($_) eq 'ARRAY' ? @$_ : $_, @to_expand;
1301 return undef unless @exp;
1302 return undef if @exp == 1 and not defined($exp[0]);
1303 return +{ -op => [ ',', @exp ] };
1306 local @{$self->{expand}}{qw(asc desc)} = (($expander) x 2);
1308 return $self->$expander(undef, $arg);
1312 my ($self, $arg) = @_;
1314 return '' unless defined(my $expanded = $self->_expand_order_by($arg));
1316 my ($sql, @bind) = $self->render_aqt($expanded);
1318 return '' unless length($sql);
1320 my $final_sql = $self->_sqlcase(' order by ').$sql;
1322 return wantarray ? ($final_sql, @bind) : $final_sql;
1325 # _order_by no longer needs to call this so doesn't but DBIC uses it.
1327 sub _order_by_chunks {
1328 my ($self, $arg) = @_;
1330 return () unless defined(my $expanded = $self->_expand_order_by($arg));
1332 return $self->_chunkify_order_by($expanded);
1335 sub _chunkify_order_by {
1336 my ($self, $expanded) = @_;
1338 return grep length, $self->render_aqt($expanded)
1339 if $expanded->{-ident} or @{$expanded->{-literal}||[]} == 1;
1342 if (ref() eq 'HASH' and $_->{-op} and $_->{-op}[0] eq ',') {
1343 my ($comma, @list) = @{$_->{-op}};
1344 return map $self->_chunkify_order_by($_), @list;
1346 return [ $self->render_aqt($_) ];
1350 #======================================================================
1351 # DATASOURCE (FOR NOW, JUST PLAIN TABLE OR LIST OF TABLES)
1352 #======================================================================
1358 $self->_expand_maybe_list_expr($from, -ident)
1363 #======================================================================
1365 #======================================================================
1367 sub _expand_maybe_list_expr {
1368 my ($self, $expr, $default) = @_;
1369 return +{ -op => [ ',',
1370 map $self->expand_expr($_, $default),
1371 ref($expr) eq 'ARRAY' ? @$expr : $expr
1375 # highly optimized, as it's called way too often
1377 # my ($self, $label) = @_;
1379 return '' unless defined $_[1];
1380 return ${$_[1]} if ref($_[1]) eq 'SCALAR';
1381 puke 'Identifier cannot be hashref' if ref($_[1]) eq 'HASH';
1383 unless ($_[0]->{quote_char}) {
1384 if (ref($_[1]) eq 'ARRAY') {
1385 return join($_[0]->{name_sep}||'.', @{$_[1]});
1387 $_[0]->_assert_pass_injection_guard($_[1]);
1392 my $qref = ref $_[0]->{quote_char};
1394 !$qref ? ($_[0]->{quote_char}, $_[0]->{quote_char})
1395 : ($qref eq 'ARRAY') ? @{$_[0]->{quote_char}}
1396 : puke "Unsupported quote_char format: $_[0]->{quote_char}";
1398 my $esc = $_[0]->{escape_char} || $r;
1400 # parts containing * are naturally unquoted
1402 $_[0]->{name_sep}||'',
1406 : do { (my $n = $_) =~ s/(\Q$esc\E|\Q$r\E)/$esc$1/g; $l . $n . $r }
1408 (ref($_[1]) eq 'ARRAY'
1412 ? split (/\Q$_[0]->{name_sep}\E/, $_[1] )
1420 # Conversion, if applicable
1422 #my ($self, $arg) = @_;
1423 if ($_[0]->{convert_where}) {
1424 return $_[0]->_sqlcase($_[0]->{convert_where}) .'(' . $_[1] . ')';
1431 #my ($self, $col, @vals) = @_;
1432 # called often - tighten code
1433 return $_[0]->{bindtype} eq 'columns'
1434 ? map {[$_[1], $_]} @_[2 .. $#_]
1439 # Dies if any element of @bind is not in [colname => value] format
1440 # if bindtype is 'columns'.
1441 sub _assert_bindval_matches_bindtype {
1442 # my ($self, @bind) = @_;
1444 if ($self->{bindtype} eq 'columns') {
1446 if (!defined $_ || ref($_) ne 'ARRAY' || @$_ != 2) {
1447 puke "bindtype 'columns' selected, you need to pass: [column_name => bind_value]"
1453 sub _join_sql_clauses {
1454 my ($self, $logic, $clauses_aref, $bind_aref) = @_;
1456 if (@$clauses_aref > 1) {
1457 my $join = " " . $self->_sqlcase($logic) . " ";
1458 my $sql = '( ' . join($join, @$clauses_aref) . ' )';
1459 return ($sql, @$bind_aref);
1461 elsif (@$clauses_aref) {
1462 return ($clauses_aref->[0], @$bind_aref); # no parentheses
1465 return (); # if no SQL, ignore @$bind_aref
1470 # Fix SQL case, if so requested
1472 # LDNOTE: if $self->{case} is true, then it contains 'lower', so we
1473 # don't touch the argument ... crooked logic, but let's not change it!
1474 return $_[0]->{case} ? $_[1] : uc($_[1]);
1478 #======================================================================
1479 # DISPATCHING FROM REFKIND
1480 #======================================================================
1483 my ($self, $data) = @_;
1485 return 'UNDEF' unless defined $data;
1487 # blessed objects are treated like scalars
1488 my $ref = (Scalar::Util::blessed $data) ? '' : ref $data;
1490 return 'SCALAR' unless $ref;
1493 while ($ref eq 'REF') {
1495 $ref = (Scalar::Util::blessed $data) ? '' : ref $data;
1499 return ($ref||'SCALAR') . ('REF' x $n_steps);
1503 my ($self, $data) = @_;
1504 my @try = ($self->_refkind($data));
1505 push @try, 'SCALAR_or_UNDEF' if $try[0] eq 'SCALAR' || $try[0] eq 'UNDEF';
1506 push @try, 'FALLBACK';
1510 sub _METHOD_FOR_refkind {
1511 my ($self, $meth_prefix, $data) = @_;
1514 for (@{$self->_try_refkind($data)}) {
1515 $method = $self->can($meth_prefix."_".$_)
1519 return $method || puke "cannot dispatch on '$meth_prefix' for ".$self->_refkind($data);
1523 sub _SWITCH_refkind {
1524 my ($self, $data, $dispatch_table) = @_;
1527 for (@{$self->_try_refkind($data)}) {
1528 $coderef = $dispatch_table->{$_}
1532 puke "no dispatch entry for ".$self->_refkind($data)
1541 #======================================================================
1542 # VALUES, GENERATE, AUTOLOAD
1543 #======================================================================
1545 # LDNOTE: original code from nwiger, didn't touch code in that section
1546 # I feel the AUTOLOAD stuff should not be the default, it should
1547 # only be activated on explicit demand by user.
1551 my $data = shift || return;
1552 puke "Argument to ", __PACKAGE__, "->values must be a \\%hash"
1553 unless ref $data eq 'HASH';
1556 foreach my $k (sort keys %$data) {
1557 my $v = $data->{$k};
1558 $self->_SWITCH_refkind($v, {
1560 if ($self->{array_datatypes}) { # array datatype
1561 push @all_bind, $self->_bindtype($k, $v);
1563 else { # literal SQL with bind
1564 my ($sql, @bind) = @$v;
1565 $self->_assert_bindval_matches_bindtype(@bind);
1566 push @all_bind, @bind;
1569 ARRAYREFREF => sub { # literal SQL with bind
1570 my ($sql, @bind) = @${$v};
1571 $self->_assert_bindval_matches_bindtype(@bind);
1572 push @all_bind, @bind;
1574 SCALARREF => sub { # literal SQL without bind
1576 SCALAR_or_UNDEF => sub {
1577 push @all_bind, $self->_bindtype($k, $v);
1588 my(@sql, @sqlq, @sqlv);
1592 if ($ref eq 'HASH') {
1593 for my $k (sort keys %$_) {
1596 my $label = $self->_quote($k);
1597 if ($r eq 'ARRAY') {
1598 # literal SQL with bind
1599 my ($sql, @bind) = @$v;
1600 $self->_assert_bindval_matches_bindtype(@bind);
1601 push @sqlq, "$label = $sql";
1603 } elsif ($r eq 'SCALAR') {
1604 # literal SQL without bind
1605 push @sqlq, "$label = $$v";
1607 push @sqlq, "$label = ?";
1608 push @sqlv, $self->_bindtype($k, $v);
1611 push @sql, $self->_sqlcase('set'), join ', ', @sqlq;
1612 } elsif ($ref eq 'ARRAY') {
1613 # unlike insert(), assume these are ONLY the column names, i.e. for SQL
1616 if ($r eq 'ARRAY') { # literal SQL with bind
1617 my ($sql, @bind) = @$v;
1618 $self->_assert_bindval_matches_bindtype(@bind);
1621 } elsif ($r eq 'SCALAR') { # literal SQL without bind
1622 # embedded literal SQL
1629 push @sql, '(' . join(', ', @sqlq) . ')';
1630 } elsif ($ref eq 'SCALAR') {
1634 # strings get case twiddled
1635 push @sql, $self->_sqlcase($_);
1639 my $sql = join ' ', @sql;
1641 # this is pretty tricky
1642 # if ask for an array, return ($stmt, @bind)
1643 # otherwise, s/?/shift @sqlv/ to put it inline
1645 return ($sql, @sqlv);
1647 1 while $sql =~ s/\?/my $d = shift(@sqlv);
1648 ref $d ? $d->[1] : $d/e;
1657 # This allows us to check for a local, then _form, attr
1659 my($name) = $AUTOLOAD =~ /.*::(.+)/;
1660 return $self->generate($name, @_);
1671 SQL::Abstract - Generate SQL from Perl data structures
1677 my $sql = SQL::Abstract->new;
1679 my($stmt, @bind) = $sql->select($source, \@fields, \%where, $order);
1681 my($stmt, @bind) = $sql->insert($table, \%fieldvals || \@values);
1683 my($stmt, @bind) = $sql->update($table, \%fieldvals, \%where);
1685 my($stmt, @bind) = $sql->delete($table, \%where);
1687 # Then, use these in your DBI statements
1688 my $sth = $dbh->prepare($stmt);
1689 $sth->execute(@bind);
1691 # Just generate the WHERE clause
1692 my($stmt, @bind) = $sql->where(\%where, $order);
1694 # Return values in the same order, for hashed queries
1695 # See PERFORMANCE section for more details
1696 my @bind = $sql->values(\%fieldvals);
1700 This module was inspired by the excellent L<DBIx::Abstract>.
1701 However, in using that module I found that what I really wanted
1702 to do was generate SQL, but still retain complete control over my
1703 statement handles and use the DBI interface. So, I set out to
1704 create an abstract SQL generation module.
1706 While based on the concepts used by L<DBIx::Abstract>, there are
1707 several important differences, especially when it comes to WHERE
1708 clauses. I have modified the concepts used to make the SQL easier
1709 to generate from Perl data structures and, IMO, more intuitive.
1710 The underlying idea is for this module to do what you mean, based
1711 on the data structures you provide it. The big advantage is that
1712 you don't have to modify your code every time your data changes,
1713 as this module figures it out.
1715 To begin with, an SQL INSERT is as easy as just specifying a hash
1716 of C<key=value> pairs:
1719 name => 'Jimbo Bobson',
1720 phone => '123-456-7890',
1721 address => '42 Sister Lane',
1722 city => 'St. Louis',
1723 state => 'Louisiana',
1726 The SQL can then be generated with this:
1728 my($stmt, @bind) = $sql->insert('people', \%data);
1730 Which would give you something like this:
1732 $stmt = "INSERT INTO people
1733 (address, city, name, phone, state)
1734 VALUES (?, ?, ?, ?, ?)";
1735 @bind = ('42 Sister Lane', 'St. Louis', 'Jimbo Bobson',
1736 '123-456-7890', 'Louisiana');
1738 These are then used directly in your DBI code:
1740 my $sth = $dbh->prepare($stmt);
1741 $sth->execute(@bind);
1743 =head2 Inserting and Updating Arrays
1745 If your database has array types (like for example Postgres),
1746 activate the special option C<< array_datatypes => 1 >>
1747 when creating the C<SQL::Abstract> object.
1748 Then you may use an arrayref to insert and update database array types:
1750 my $sql = SQL::Abstract->new(array_datatypes => 1);
1752 planets => [qw/Mercury Venus Earth Mars/]
1755 my($stmt, @bind) = $sql->insert('solar_system', \%data);
1759 $stmt = "INSERT INTO solar_system (planets) VALUES (?)"
1761 @bind = (['Mercury', 'Venus', 'Earth', 'Mars']);
1764 =head2 Inserting and Updating SQL
1766 In order to apply SQL functions to elements of your C<%data> you may
1767 specify a reference to an arrayref for the given hash value. For example,
1768 if you need to execute the Oracle C<to_date> function on a value, you can
1769 say something like this:
1773 date_entered => \[ "to_date(?,'MM/DD/YYYY')", "03/02/2003" ],
1776 The first value in the array is the actual SQL. Any other values are
1777 optional and would be included in the bind values array. This gives
1780 my($stmt, @bind) = $sql->insert('people', \%data);
1782 $stmt = "INSERT INTO people (name, date_entered)
1783 VALUES (?, to_date(?,'MM/DD/YYYY'))";
1784 @bind = ('Bill', '03/02/2003');
1786 An UPDATE is just as easy, all you change is the name of the function:
1788 my($stmt, @bind) = $sql->update('people', \%data);
1790 Notice that your C<%data> isn't touched; the module will generate
1791 the appropriately quirky SQL for you automatically. Usually you'll
1792 want to specify a WHERE clause for your UPDATE, though, which is
1793 where handling C<%where> hashes comes in handy...
1795 =head2 Complex where statements
1797 This module can generate pretty complicated WHERE statements
1798 easily. For example, simple C<key=value> pairs are taken to mean
1799 equality, and if you want to see if a field is within a set
1800 of values, you can use an arrayref. Let's say we wanted to
1801 SELECT some data based on this criteria:
1804 requestor => 'inna',
1805 worker => ['nwiger', 'rcwe', 'sfz'],
1806 status => { '!=', 'completed' }
1809 my($stmt, @bind) = $sql->select('tickets', '*', \%where);
1811 The above would give you something like this:
1813 $stmt = "SELECT * FROM tickets WHERE
1814 ( requestor = ? ) AND ( status != ? )
1815 AND ( worker = ? OR worker = ? OR worker = ? )";
1816 @bind = ('inna', 'completed', 'nwiger', 'rcwe', 'sfz');
1818 Which you could then use in DBI code like so:
1820 my $sth = $dbh->prepare($stmt);
1821 $sth->execute(@bind);
1827 The methods are simple. There's one for every major SQL operation,
1828 and a constructor you use first. The arguments are specified in a
1829 similar order for each method (table, then fields, then a where
1830 clause) to try and simplify things.
1832 =head2 new(option => 'value')
1834 The C<new()> function takes a list of options and values, and returns
1835 a new B<SQL::Abstract> object which can then be used to generate SQL
1836 through the methods below. The options accepted are:
1842 If set to 'lower', then SQL will be generated in all lowercase. By
1843 default SQL is generated in "textbook" case meaning something like:
1845 SELECT a_field FROM a_table WHERE some_field LIKE '%someval%'
1847 Any setting other than 'lower' is ignored.
1851 This determines what the default comparison operator is. By default
1852 it is C<=>, meaning that a hash like this:
1854 %where = (name => 'nwiger', email => 'nate@wiger.org');
1856 Will generate SQL like this:
1858 WHERE name = 'nwiger' AND email = 'nate@wiger.org'
1860 However, you may want loose comparisons by default, so if you set
1861 C<cmp> to C<like> you would get SQL such as:
1863 WHERE name like 'nwiger' AND email like 'nate@wiger.org'
1865 You can also override the comparison on an individual basis - see
1866 the huge section on L</"WHERE CLAUSES"> at the bottom.
1868 =item sqltrue, sqlfalse
1870 Expressions for inserting boolean values within SQL statements.
1871 By default these are C<1=1> and C<1=0>. They are used
1872 by the special operators C<-in> and C<-not_in> for generating
1873 correct SQL even when the argument is an empty array (see below).
1877 This determines the default logical operator for multiple WHERE
1878 statements in arrays or hashes. If absent, the default logic is "or"
1879 for arrays, and "and" for hashes. This means that a WHERE
1883 event_date => {'>=', '2/13/99'},
1884 event_date => {'<=', '4/24/03'},
1887 will generate SQL like this:
1889 WHERE event_date >= '2/13/99' OR event_date <= '4/24/03'
1891 This is probably not what you want given this query, though (look
1892 at the dates). To change the "OR" to an "AND", simply specify:
1894 my $sql = SQL::Abstract->new(logic => 'and');
1896 Which will change the above C<WHERE> to:
1898 WHERE event_date >= '2/13/99' AND event_date <= '4/24/03'
1900 The logic can also be changed locally by inserting
1901 a modifier in front of an arrayref:
1903 @where = (-and => [event_date => {'>=', '2/13/99'},
1904 event_date => {'<=', '4/24/03'} ]);
1906 See the L</"WHERE CLAUSES"> section for explanations.
1910 This will automatically convert comparisons using the specified SQL
1911 function for both column and value. This is mostly used with an argument
1912 of C<upper> or C<lower>, so that the SQL will have the effect of
1913 case-insensitive "searches". For example, this:
1915 $sql = SQL::Abstract->new(convert => 'upper');
1916 %where = (keywords => 'MaKe iT CAse inSeNSItive');
1918 Will turn out the following SQL:
1920 WHERE upper(keywords) like upper('MaKe iT CAse inSeNSItive')
1922 The conversion can be C<upper()>, C<lower()>, or any other SQL function
1923 that can be applied symmetrically to fields (actually B<SQL::Abstract> does
1924 not validate this option; it will just pass through what you specify verbatim).
1928 This is a kludge because many databases suck. For example, you can't
1929 just bind values using DBI's C<execute()> for Oracle C<CLOB> or C<BLOB> fields.
1930 Instead, you have to use C<bind_param()>:
1932 $sth->bind_param(1, 'reg data');
1933 $sth->bind_param(2, $lots, {ora_type => ORA_CLOB});
1935 The problem is, B<SQL::Abstract> will normally just return a C<@bind> array,
1936 which loses track of which field each slot refers to. Fear not.
1938 If you specify C<bindtype> in new, you can determine how C<@bind> is returned.
1939 Currently, you can specify either C<normal> (default) or C<columns>. If you
1940 specify C<columns>, you will get an array that looks like this:
1942 my $sql = SQL::Abstract->new(bindtype => 'columns');
1943 my($stmt, @bind) = $sql->insert(...);
1946 [ 'column1', 'value1' ],
1947 [ 'column2', 'value2' ],
1948 [ 'column3', 'value3' ],
1951 You can then iterate through this manually, using DBI's C<bind_param()>.
1953 $sth->prepare($stmt);
1956 my($col, $data) = @$_;
1957 if ($col eq 'details' || $col eq 'comments') {
1958 $sth->bind_param($i, $data, {ora_type => ORA_CLOB});
1959 } elsif ($col eq 'image') {
1960 $sth->bind_param($i, $data, {ora_type => ORA_BLOB});
1962 $sth->bind_param($i, $data);
1966 $sth->execute; # execute without @bind now
1968 Now, why would you still use B<SQL::Abstract> if you have to do this crap?
1969 Basically, the advantage is still that you don't have to care which fields
1970 are or are not included. You could wrap that above C<for> loop in a simple
1971 sub called C<bind_fields()> or something and reuse it repeatedly. You still
1972 get a layer of abstraction over manual SQL specification.
1974 Note that if you set L</bindtype> to C<columns>, the C<\[ $sql, @bind ]>
1975 construct (see L</Literal SQL with placeholders and bind values (subqueries)>)
1976 will expect the bind values in this format.
1980 This is the character that a table or column name will be quoted
1981 with. By default this is an empty string, but you could set it to
1982 the character C<`>, to generate SQL like this:
1984 SELECT `a_field` FROM `a_table` WHERE `some_field` LIKE '%someval%'
1986 Alternatively, you can supply an array ref of two items, the first being the left
1987 hand quote character, and the second the right hand quote character. For
1988 example, you could supply C<['[',']']> for SQL Server 2000 compliant quotes
1989 that generates SQL like this:
1991 SELECT [a_field] FROM [a_table] WHERE [some_field] LIKE '%someval%'
1993 Quoting is useful if you have tables or columns names that are reserved
1994 words in your database's SQL dialect.
1998 This is the character that will be used to escape L</quote_char>s appearing
1999 in an identifier before it has been quoted.
2001 The parameter default in case of a single L</quote_char> character is the quote
2004 When opening-closing-style quoting is used (L</quote_char> is an arrayref)
2005 this parameter defaults to the B<closing (right)> L</quote_char>. Occurrences
2006 of the B<opening (left)> L</quote_char> within the identifier are currently left
2007 untouched. The default for opening-closing-style quotes may change in future
2008 versions, thus you are B<strongly encouraged> to specify the escape character
2013 This is the character that separates a table and column name. It is
2014 necessary to specify this when the C<quote_char> option is selected,
2015 so that tables and column names can be individually quoted like this:
2017 SELECT `table`.`one_field` FROM `table` WHERE `table`.`other_field` = 1
2019 =item injection_guard
2021 A regular expression C<qr/.../> that is applied to any C<-function> and unquoted
2022 column name specified in a query structure. This is a safety mechanism to avoid
2023 injection attacks when mishandling user input e.g.:
2025 my %condition_as_column_value_pairs = get_values_from_user();
2026 $sqla->select( ... , \%condition_as_column_value_pairs );
2028 If the expression matches an exception is thrown. Note that literal SQL
2029 supplied via C<\'...'> or C<\['...']> is B<not> checked in any way.
2031 Defaults to checking for C<;> and the C<GO> keyword (TransactSQL)
2033 =item array_datatypes
2035 When this option is true, arrayrefs in INSERT or UPDATE are
2036 interpreted as array datatypes and are passed directly
2038 When this option is false, arrayrefs are interpreted
2039 as literal SQL, just like refs to arrayrefs
2040 (but this behavior is for backwards compatibility; when writing
2041 new queries, use the "reference to arrayref" syntax
2047 Takes a reference to a list of "special operators"
2048 to extend the syntax understood by L<SQL::Abstract>.
2049 See section L</"SPECIAL OPERATORS"> for details.
2053 Takes a reference to a list of "unary operators"
2054 to extend the syntax understood by L<SQL::Abstract>.
2055 See section L</"UNARY OPERATORS"> for details.
2061 =head2 insert($table, \@values || \%fieldvals, \%options)
2063 This is the simplest function. You simply give it a table name
2064 and either an arrayref of values or hashref of field/value pairs.
2065 It returns an SQL INSERT statement and a list of bind values.
2066 See the sections on L</"Inserting and Updating Arrays"> and
2067 L</"Inserting and Updating SQL"> for information on how to insert
2068 with those data types.
2070 The optional C<\%options> hash reference may contain additional
2071 options to generate the insert SQL. Currently supported options
2078 Takes either a scalar of raw SQL fields, or an array reference of
2079 field names, and adds on an SQL C<RETURNING> statement at the end.
2080 This allows you to return data generated by the insert statement
2081 (such as row IDs) without performing another C<SELECT> statement.
2082 Note, however, this is not part of the SQL standard and may not
2083 be supported by all database engines.
2087 =head2 update($table, \%fieldvals, \%where, \%options)
2089 This takes a table, hashref of field/value pairs, and an optional
2090 hashref L<WHERE clause|/WHERE CLAUSES>. It returns an SQL UPDATE function and a list
2092 See the sections on L</"Inserting and Updating Arrays"> and
2093 L</"Inserting and Updating SQL"> for information on how to insert
2094 with those data types.
2096 The optional C<\%options> hash reference may contain additional
2097 options to generate the update SQL. Currently supported options
2104 See the C<returning> option to
2105 L<insert|/insert($table, \@values || \%fieldvals, \%options)>.
2109 =head2 select($source, $fields, $where, $order)
2111 This returns a SQL SELECT statement and associated list of bind values, as
2112 specified by the arguments:
2118 Specification of the 'FROM' part of the statement.
2119 The argument can be either a plain scalar (interpreted as a table
2120 name, will be quoted), or an arrayref (interpreted as a list
2121 of table names, joined by commas, quoted), or a scalarref
2122 (literal SQL, not quoted).
2126 Specification of the list of fields to retrieve from
2128 The argument can be either an arrayref (interpreted as a list
2129 of field names, will be joined by commas and quoted), or a
2130 plain scalar (literal SQL, not quoted).
2131 Please observe that this API is not as flexible as that of
2132 the first argument C<$source>, for backwards compatibility reasons.
2136 Optional argument to specify the WHERE part of the query.
2137 The argument is most often a hashref, but can also be
2138 an arrayref or plain scalar --
2139 see section L<WHERE clause|/"WHERE CLAUSES"> for details.
2143 Optional argument to specify the ORDER BY part of the query.
2144 The argument can be a scalar, a hashref or an arrayref
2145 -- see section L<ORDER BY clause|/"ORDER BY CLAUSES">
2151 =head2 delete($table, \%where, \%options)
2153 This takes a table name and optional hashref L<WHERE clause|/WHERE CLAUSES>.
2154 It returns an SQL DELETE statement and list of bind values.
2156 The optional C<\%options> hash reference may contain additional
2157 options to generate the delete SQL. Currently supported options
2164 See the C<returning> option to
2165 L<insert|/insert($table, \@values || \%fieldvals, \%options)>.
2169 =head2 where(\%where, $order)
2171 This is used to generate just the WHERE clause. For example,
2172 if you have an arbitrary data structure and know what the
2173 rest of your SQL is going to look like, but want an easy way
2174 to produce a WHERE clause, use this. It returns an SQL WHERE
2175 clause and list of bind values.
2178 =head2 values(\%data)
2180 This just returns the values from the hash C<%data>, in the same
2181 order that would be returned from any of the other above queries.
2182 Using this allows you to markedly speed up your queries if you
2183 are affecting lots of rows. See below under the L</"PERFORMANCE"> section.
2185 =head2 generate($any, 'number', $of, \@data, $struct, \%types)
2187 Warning: This is an experimental method and subject to change.
2189 This returns arbitrarily generated SQL. It's a really basic shortcut.
2190 It will return two different things, depending on return context:
2192 my($stmt, @bind) = $sql->generate('create table', \$table, \@fields);
2193 my $stmt_and_val = $sql->generate('create table', \$table, \@fields);
2195 These would return the following:
2197 # First calling form
2198 $stmt = "CREATE TABLE test (?, ?)";
2199 @bind = (field1, field2);
2201 # Second calling form
2202 $stmt_and_val = "CREATE TABLE test (field1, field2)";
2204 Depending on what you're trying to do, it's up to you to choose the correct
2205 format. In this example, the second form is what you would want.
2209 $sql->generate('alter session', { nls_date_format => 'MM/YY' });
2213 ALTER SESSION SET nls_date_format = 'MM/YY'
2215 You get the idea. Strings get their case twiddled, but everything
2216 else remains verbatim.
2218 =head1 EXPORTABLE FUNCTIONS
2220 =head2 is_plain_value
2222 Determines if the supplied argument is a plain value as understood by this
2227 =item * The value is C<undef>
2229 =item * The value is a non-reference
2231 =item * The value is an object with stringification overloading
2233 =item * The value is of the form C<< { -value => $anything } >>
2237 On failure returns C<undef>, on success returns a B<scalar> reference
2238 to the original supplied argument.
2244 The stringification overloading detection is rather advanced: it takes
2245 into consideration not only the presence of a C<""> overload, but if that
2246 fails also checks for enabled
2247 L<autogenerated versions of C<"">|overload/Magic Autogeneration>, based
2248 on either C<0+> or C<bool>.
2250 Unfortunately testing in the field indicates that this
2251 detection B<< may tickle a latent bug in perl versions before 5.018 >>,
2252 but only when very large numbers of stringifying objects are involved.
2253 At the time of writing ( Sep 2014 ) there is no clear explanation of
2254 the direct cause, nor is there a manageably small test case that reliably
2255 reproduces the problem.
2257 If you encounter any of the following exceptions in B<random places within
2258 your application stack> - this module may be to blame:
2260 Operation "ne": no method found,
2261 left argument in overloaded package <something>,
2262 right argument in overloaded package <something>
2266 Stub found while resolving method "???" overloading """" in package <something>
2268 If you fall victim to the above - please attempt to reduce the problem
2269 to something that could be sent to the L<SQL::Abstract developers
2270 |DBIx::Class/GETTING HELP/SUPPORT>
2271 (either publicly or privately). As a workaround in the meantime you can
2272 set C<$ENV{SQLA_ISVALUE_IGNORE_AUTOGENERATED_STRINGIFICATION}> to a true
2273 value, which will most likely eliminate your problem (at the expense of
2274 not being able to properly detect exotic forms of stringification).
2276 This notice and environment variable will be removed in a future version,
2277 as soon as the underlying problem is found and a reliable workaround is
2282 =head2 is_literal_value
2284 Determines if the supplied argument is a literal value as understood by this
2289 =item * C<\$sql_string>
2291 =item * C<\[ $sql_string, @bind_values ]>
2295 On failure returns C<undef>, on success returns an B<array> reference
2296 containing the unpacked version of the supplied literal SQL and bind values.
2298 =head1 WHERE CLAUSES
2302 This module uses a variation on the idea from L<DBIx::Abstract>. It
2303 is B<NOT>, repeat I<not> 100% compatible. B<The main logic of this
2304 module is that things in arrays are OR'ed, and things in hashes
2307 The easiest way to explain is to show lots of examples. After
2308 each C<%where> hash shown, it is assumed you used:
2310 my($stmt, @bind) = $sql->where(\%where);
2312 However, note that the C<%where> hash can be used directly in any
2313 of the other functions as well, as described above.
2315 =head2 Key-value pairs
2317 So, let's get started. To begin, a simple hash:
2321 status => 'completed'
2324 Is converted to SQL C<key = val> statements:
2326 $stmt = "WHERE user = ? AND status = ?";
2327 @bind = ('nwiger', 'completed');
2329 One common thing I end up doing is having a list of values that
2330 a field can be in. To do this, simply specify a list inside of
2335 status => ['assigned', 'in-progress', 'pending'];
2338 This simple code will create the following:
2340 $stmt = "WHERE user = ? AND ( status = ? OR status = ? OR status = ? )";
2341 @bind = ('nwiger', 'assigned', 'in-progress', 'pending');
2343 A field associated to an empty arrayref will be considered a
2344 logical false and will generate 0=1.
2346 =head2 Tests for NULL values
2348 If the value part is C<undef> then this is converted to SQL <IS NULL>
2357 $stmt = "WHERE user = ? AND status IS NULL";
2360 To test if a column IS NOT NULL:
2364 status => { '!=', undef },
2367 =head2 Specific comparison operators
2369 If you want to specify a different type of operator for your comparison,
2370 you can use a hashref for a given column:
2374 status => { '!=', 'completed' }
2377 Which would generate:
2379 $stmt = "WHERE user = ? AND status != ?";
2380 @bind = ('nwiger', 'completed');
2382 To test against multiple values, just enclose the values in an arrayref:
2384 status => { '=', ['assigned', 'in-progress', 'pending'] };
2386 Which would give you:
2388 "WHERE status = ? OR status = ? OR status = ?"
2391 The hashref can also contain multiple pairs, in which case it is expanded
2392 into an C<AND> of its elements:
2396 status => { '!=', 'completed', -not_like => 'pending%' }
2399 # Or more dynamically, like from a form
2400 $where{user} = 'nwiger';
2401 $where{status}{'!='} = 'completed';
2402 $where{status}{'-not_like'} = 'pending%';
2404 # Both generate this
2405 $stmt = "WHERE user = ? AND status != ? AND status NOT LIKE ?";
2406 @bind = ('nwiger', 'completed', 'pending%');
2409 To get an OR instead, you can combine it with the arrayref idea:
2413 priority => [ { '=', 2 }, { '>', 5 } ]
2416 Which would generate:
2418 $stmt = "WHERE ( priority = ? OR priority > ? ) AND user = ?";
2419 @bind = ('2', '5', 'nwiger');
2421 If you want to include literal SQL (with or without bind values), just use a
2422 scalar reference or reference to an arrayref as the value:
2425 date_entered => { '>' => \["to_date(?, 'MM/DD/YYYY')", "11/26/2008"] },
2426 date_expires => { '<' => \"now()" }
2429 Which would generate:
2431 $stmt = "WHERE date_entered > to_date(?, 'MM/DD/YYYY') AND date_expires < now()";
2432 @bind = ('11/26/2008');
2435 =head2 Logic and nesting operators
2437 In the example above,
2438 there is a subtle trap if you want to say something like
2439 this (notice the C<AND>):
2441 WHERE priority != ? AND priority != ?
2443 Because, in Perl you I<can't> do this:
2445 priority => { '!=' => 2, '!=' => 1 }
2447 As the second C<!=> key will obliterate the first. The solution
2448 is to use the special C<-modifier> form inside an arrayref:
2450 priority => [ -and => {'!=', 2},
2454 Normally, these would be joined by C<OR>, but the modifier tells it
2455 to use C<AND> instead. (Hint: You can use this in conjunction with the
2456 C<logic> option to C<new()> in order to change the way your queries
2457 work by default.) B<Important:> Note that the C<-modifier> goes
2458 B<INSIDE> the arrayref, as an extra first element. This will
2459 B<NOT> do what you think it might:
2461 priority => -and => [{'!=', 2}, {'!=', 1}] # WRONG!
2463 Here is a quick list of equivalencies, since there is some overlap:
2466 status => {'!=', 'completed', 'not like', 'pending%' }
2467 status => [ -and => {'!=', 'completed'}, {'not like', 'pending%'}]
2470 status => {'=', ['assigned', 'in-progress']}
2471 status => [ -or => {'=', 'assigned'}, {'=', 'in-progress'}]
2472 status => [ {'=', 'assigned'}, {'=', 'in-progress'} ]
2476 =head2 Special operators: IN, BETWEEN, etc.
2478 You can also use the hashref format to compare a list of fields using the
2479 C<IN> comparison operator, by specifying the list as an arrayref:
2482 status => 'completed',
2483 reportid => { -in => [567, 2335, 2] }
2486 Which would generate:
2488 $stmt = "WHERE status = ? AND reportid IN (?,?,?)";
2489 @bind = ('completed', '567', '2335', '2');
2491 The reverse operator C<-not_in> generates SQL C<NOT IN> and is used in
2494 If the argument to C<-in> is an empty array, 'sqlfalse' is generated
2495 (by default: C<1=0>). Similarly, C<< -not_in => [] >> generates
2496 'sqltrue' (by default: C<1=1>).
2498 In addition to the array you can supply a chunk of literal sql or
2499 literal sql with bind:
2502 customer => { -in => \[
2503 'SELECT cust_id FROM cust WHERE balance > ?',
2506 status => { -in => \'SELECT status_codes FROM states' },
2512 customer IN ( SELECT cust_id FROM cust WHERE balance > ? )
2513 AND status IN ( SELECT status_codes FROM states )
2517 Finally, if the argument to C<-in> is not a reference, it will be
2518 treated as a single-element array.
2520 Another pair of operators is C<-between> and C<-not_between>,
2521 used with an arrayref of two values:
2525 completion_date => {
2526 -not_between => ['2002-10-01', '2003-02-06']
2532 WHERE user = ? AND completion_date NOT BETWEEN ( ? AND ? )
2534 Just like with C<-in> all plausible combinations of literal SQL
2538 start0 => { -between => [ 1, 2 ] },
2539 start1 => { -between => \["? AND ?", 1, 2] },
2540 start2 => { -between => \"lower(x) AND upper(y)" },
2541 start3 => { -between => [
2543 \["upper(?)", 'stuff' ],
2550 ( start0 BETWEEN ? AND ? )
2551 AND ( start1 BETWEEN ? AND ? )
2552 AND ( start2 BETWEEN lower(x) AND upper(y) )
2553 AND ( start3 BETWEEN lower(x) AND upper(?) )
2555 @bind = (1, 2, 1, 2, 'stuff');
2558 These are the two builtin "special operators"; but the
2559 list can be expanded: see section L</"SPECIAL OPERATORS"> below.
2561 =head2 Unary operators: bool
2563 If you wish to test against boolean columns or functions within your
2564 database you can use the C<-bool> and C<-not_bool> operators. For
2565 example to test the column C<is_user> being true and the column
2566 C<is_enabled> being false you would use:-
2570 -not_bool => 'is_enabled',
2575 WHERE is_user AND NOT is_enabled
2577 If a more complex combination is required, testing more conditions,
2578 then you should use the and/or operators:-
2583 -not_bool => { two=> { -rlike => 'bar' } },
2584 -not_bool => { three => [ { '=', 2 }, { '>', 5 } ] },
2595 (NOT ( three = ? OR three > ? ))
2598 =head2 Nested conditions, -and/-or prefixes
2600 So far, we've seen how multiple conditions are joined with a top-level
2601 C<AND>. We can change this by putting the different conditions we want in
2602 hashes and then putting those hashes in an array. For example:
2607 status => { -like => ['pending%', 'dispatched'] },
2611 status => 'unassigned',
2615 This data structure would create the following:
2617 $stmt = "WHERE ( user = ? AND ( status LIKE ? OR status LIKE ? ) )
2618 OR ( user = ? AND status = ? ) )";
2619 @bind = ('nwiger', 'pending', 'dispatched', 'robot', 'unassigned');
2622 Clauses in hashrefs or arrayrefs can be prefixed with an C<-and> or C<-or>
2623 to change the logic inside:
2629 -and => [ workhrs => {'>', 20}, geo => 'ASIA' ],
2630 -or => { workhrs => {'<', 50}, geo => 'EURO' },
2637 $stmt = "WHERE ( user = ?
2638 AND ( ( workhrs > ? AND geo = ? )
2639 OR ( workhrs < ? OR geo = ? ) ) )";
2640 @bind = ('nwiger', '20', 'ASIA', '50', 'EURO');
2642 =head3 Algebraic inconsistency, for historical reasons
2644 C<Important note>: when connecting several conditions, the C<-and->|C<-or>
2645 operator goes C<outside> of the nested structure; whereas when connecting
2646 several constraints on one column, the C<-and> operator goes
2647 C<inside> the arrayref. Here is an example combining both features:
2650 -and => [a => 1, b => 2],
2651 -or => [c => 3, d => 4],
2652 e => [-and => {-like => 'foo%'}, {-like => '%bar'} ]
2657 WHERE ( ( ( a = ? AND b = ? )
2658 OR ( c = ? OR d = ? )
2659 OR ( e LIKE ? AND e LIKE ? ) ) )
2661 This difference in syntax is unfortunate but must be preserved for
2662 historical reasons. So be careful: the two examples below would
2663 seem algebraically equivalent, but they are not
2666 { -like => 'foo%' },
2667 { -like => '%bar' },
2669 # yields: WHERE ( ( col LIKE ? AND col LIKE ? ) )
2672 { col => { -like => 'foo%' } },
2673 { col => { -like => '%bar' } },
2675 # yields: WHERE ( ( col LIKE ? OR col LIKE ? ) )
2678 =head2 Literal SQL and value type operators
2680 The basic premise of SQL::Abstract is that in WHERE specifications the "left
2681 side" is a column name and the "right side" is a value (normally rendered as
2682 a placeholder). This holds true for both hashrefs and arrayref pairs as you
2683 see in the L</WHERE CLAUSES> examples above. Sometimes it is necessary to
2684 alter this behavior. There are several ways of doing so.
2688 This is a virtual operator that signals the string to its right side is an
2689 identifier (a column name) and not a value. For example to compare two
2690 columns you would write:
2693 priority => { '<', 2 },
2694 requestor => { -ident => 'submitter' },
2699 $stmt = "WHERE priority < ? AND requestor = submitter";
2702 If you are maintaining legacy code you may see a different construct as
2703 described in L</Deprecated usage of Literal SQL>, please use C<-ident> in new
2708 This is a virtual operator that signals that the construct to its right side
2709 is a value to be passed to DBI. This is for example necessary when you want
2710 to write a where clause against an array (for RDBMS that support such
2711 datatypes). For example:
2714 array => { -value => [1, 2, 3] }
2719 $stmt = 'WHERE array = ?';
2720 @bind = ([1, 2, 3]);
2722 Note that if you were to simply say:
2728 the result would probably not be what you wanted:
2730 $stmt = 'WHERE array = ? OR array = ? OR array = ?';
2735 Finally, sometimes only literal SQL will do. To include a random snippet
2736 of SQL verbatim, you specify it as a scalar reference. Consider this only
2737 as a last resort. Usually there is a better way. For example:
2740 priority => { '<', 2 },
2741 requestor => { -in => \'(SELECT name FROM hitmen)' },
2746 $stmt = "WHERE priority < ? AND requestor IN (SELECT name FROM hitmen)"
2749 Note that in this example, you only get one bind parameter back, since
2750 the verbatim SQL is passed as part of the statement.
2754 Never use untrusted input as a literal SQL argument - this is a massive
2755 security risk (there is no way to check literal snippets for SQL
2756 injections and other nastyness). If you need to deal with untrusted input
2757 use literal SQL with placeholders as described next.
2759 =head3 Literal SQL with placeholders and bind values (subqueries)
2761 If the literal SQL to be inserted has placeholders and bind values,
2762 use a reference to an arrayref (yes this is a double reference --
2763 not so common, but perfectly legal Perl). For example, to find a date
2764 in Postgres you can use something like this:
2767 date_column => \[ "= date '2008-09-30' - ?::integer", 10 ]
2772 $stmt = "WHERE ( date_column = date '2008-09-30' - ?::integer )"
2775 Note that you must pass the bind values in the same format as they are returned
2776 by L<where|/where(\%where, $order)>. This means that if you set L</bindtype>
2777 to C<columns>, you must provide the bind values in the
2778 C<< [ column_meta => value ] >> format, where C<column_meta> is an opaque
2779 scalar value; most commonly the column name, but you can use any scalar value
2780 (including references and blessed references), L<SQL::Abstract> will simply
2781 pass it through intact. So if C<bindtype> is set to C<columns> the above
2782 example will look like:
2785 date_column => \[ "= date '2008-09-30' - ?::integer", [ {} => 10 ] ]
2788 Literal SQL is especially useful for nesting parenthesized clauses in the
2789 main SQL query. Here is a first example:
2791 my ($sub_stmt, @sub_bind) = ("SELECT c1 FROM t1 WHERE c2 < ? AND c3 LIKE ?",
2795 bar => \["IN ($sub_stmt)" => @sub_bind],
2800 $stmt = "WHERE (foo = ? AND bar IN (SELECT c1 FROM t1
2801 WHERE c2 < ? AND c3 LIKE ?))";
2802 @bind = (1234, 100, "foo%");
2804 Other subquery operators, like for example C<"E<gt> ALL"> or C<"NOT IN">,
2805 are expressed in the same way. Of course the C<$sub_stmt> and
2806 its associated bind values can be generated through a former call
2809 my ($sub_stmt, @sub_bind)
2810 = $sql->select("t1", "c1", {c2 => {"<" => 100},
2811 c3 => {-like => "foo%"}});
2814 bar => \["> ALL ($sub_stmt)" => @sub_bind],
2817 In the examples above, the subquery was used as an operator on a column;
2818 but the same principle also applies for a clause within the main C<%where>
2819 hash, like an EXISTS subquery:
2821 my ($sub_stmt, @sub_bind)
2822 = $sql->select("t1", "*", {c1 => 1, c2 => \"> t0.c0"});
2823 my %where = ( -and => [
2825 \["EXISTS ($sub_stmt)" => @sub_bind],
2830 $stmt = "WHERE (foo = ? AND EXISTS (SELECT * FROM t1
2831 WHERE c1 = ? AND c2 > t0.c0))";
2835 Observe that the condition on C<c2> in the subquery refers to
2836 column C<t0.c0> of the main query: this is I<not> a bind
2837 value, so we have to express it through a scalar ref.
2838 Writing C<< c2 => {">" => "t0.c0"} >> would have generated
2839 C<< c2 > ? >> with bind value C<"t0.c0"> ... not exactly
2840 what we wanted here.
2842 Finally, here is an example where a subquery is used
2843 for expressing unary negation:
2845 my ($sub_stmt, @sub_bind)
2846 = $sql->where({age => [{"<" => 10}, {">" => 20}]});
2847 $sub_stmt =~ s/^ where //i; # don't want "WHERE" in the subclause
2849 lname => {like => '%son%'},
2850 \["NOT ($sub_stmt)" => @sub_bind],
2855 $stmt = "lname LIKE ? AND NOT ( age < ? OR age > ? )"
2856 @bind = ('%son%', 10, 20)
2858 =head3 Deprecated usage of Literal SQL
2860 Below are some examples of archaic use of literal SQL. It is shown only as
2861 reference for those who deal with legacy code. Each example has a much
2862 better, cleaner and safer alternative that users should opt for in new code.
2868 my %where = ( requestor => \'IS NOT NULL' )
2870 $stmt = "WHERE requestor IS NOT NULL"
2872 This used to be the way of generating NULL comparisons, before the handling
2873 of C<undef> got formalized. For new code please use the superior syntax as
2874 described in L</Tests for NULL values>.
2878 my %where = ( requestor => \'= submitter' )
2880 $stmt = "WHERE requestor = submitter"
2882 This used to be the only way to compare columns. Use the superior L</-ident>
2883 method for all new code. For example an identifier declared in such a way
2884 will be properly quoted if L</quote_char> is properly set, while the legacy
2885 form will remain as supplied.
2889 my %where = ( is_ready => \"", completed => { '>', '2012-12-21' } )
2891 $stmt = "WHERE completed > ? AND is_ready"
2892 @bind = ('2012-12-21')
2894 Using an empty string literal used to be the only way to express a boolean.
2895 For all new code please use the much more readable
2896 L<-bool|/Unary operators: bool> operator.
2902 These pages could go on for a while, since the nesting of the data
2903 structures this module can handle are pretty much unlimited (the
2904 module implements the C<WHERE> expansion as a recursive function
2905 internally). Your best bet is to "play around" with the module a
2906 little to see how the data structures behave, and choose the best
2907 format for your data based on that.
2909 And of course, all the values above will probably be replaced with
2910 variables gotten from forms or the command line. After all, if you
2911 knew everything ahead of time, you wouldn't have to worry about
2912 dynamically-generating SQL and could just hardwire it into your
2915 =head1 ORDER BY CLAUSES
2917 Some functions take an order by clause. This can either be a scalar (just a
2918 column name), a hashref of C<< { -desc => 'col' } >> or C<< { -asc => 'col' }
2919 >>, a scalarref, an arrayref-ref, or an arrayref of any of the previous
2922 Given | Will Generate
2923 ---------------------------------------------------------------
2925 'colA' | ORDER BY colA
2927 [qw/colA colB/] | ORDER BY colA, colB
2929 {-asc => 'colA'} | ORDER BY colA ASC
2931 {-desc => 'colB'} | ORDER BY colB DESC
2933 ['colA', {-asc => 'colB'}] | ORDER BY colA, colB ASC
2935 { -asc => [qw/colA colB/] } | ORDER BY colA ASC, colB ASC
2937 \'colA DESC' | ORDER BY colA DESC
2939 \[ 'FUNC(colA, ?)', $x ] | ORDER BY FUNC(colA, ?)
2940 | /* ...with $x bound to ? */
2943 { -asc => 'colA' }, | colA ASC,
2944 { -desc => [qw/colB/] }, | colB DESC,
2945 { -asc => [qw/colC colD/] },| colC ASC, colD ASC,
2946 \'colE DESC', | colE DESC,
2947 \[ 'FUNC(colF, ?)', $x ], | FUNC(colF, ?)
2948 ] | /* ...with $x bound to ? */
2949 ===============================================================
2953 =head1 SPECIAL OPERATORS
2955 my $sqlmaker = SQL::Abstract->new(special_ops => [
2959 my ($self, $field, $op, $arg) = @_;
2965 handler => 'method_name',
2969 A "special operator" is a SQL syntactic clause that can be
2970 applied to a field, instead of a usual binary operator.
2973 WHERE field IN (?, ?, ?)
2974 WHERE field BETWEEN ? AND ?
2975 WHERE MATCH(field) AGAINST (?, ?)
2977 Special operators IN and BETWEEN are fairly standard and therefore
2978 are builtin within C<SQL::Abstract> (as the overridable methods
2979 C<_where_field_IN> and C<_where_field_BETWEEN>). For other operators,
2980 like the MATCH .. AGAINST example above which is specific to MySQL,
2981 you can write your own operator handlers - supply a C<special_ops>
2982 argument to the C<new> method. That argument takes an arrayref of
2983 operator definitions; each operator definition is a hashref with two
2990 the regular expression to match the operator
2994 Either a coderef or a plain scalar method name. In both cases
2995 the expected return is C<< ($sql, @bind) >>.
2997 When supplied with a method name, it is simply called on the
2998 L<SQL::Abstract> object as:
3000 $self->$method_name($field, $op, $arg)
3004 $field is the LHS of the operator
3005 $op is the part that matched the handler regex
3008 When supplied with a coderef, it is called as:
3010 $coderef->($self, $field, $op, $arg)
3015 For example, here is an implementation
3016 of the MATCH .. AGAINST syntax for MySQL
3018 my $sqlmaker = SQL::Abstract->new(special_ops => [
3020 # special op for MySql MATCH (field) AGAINST(word1, word2, ...)
3021 {regex => qr/^match$/i,
3023 my ($self, $field, $op, $arg) = @_;
3024 $arg = [$arg] if not ref $arg;
3025 my $label = $self->_quote($field);
3026 my ($placeholder) = $self->_convert('?');
3027 my $placeholders = join ", ", (($placeholder) x @$arg);
3028 my $sql = $self->_sqlcase('match') . " ($label) "
3029 . $self->_sqlcase('against') . " ($placeholders) ";
3030 my @bind = $self->_bindtype($field, @$arg);
3031 return ($sql, @bind);
3038 =head1 UNARY OPERATORS
3040 my $sqlmaker = SQL::Abstract->new(unary_ops => [
3044 my ($self, $op, $arg) = @_;
3050 handler => 'method_name',
3054 A "unary operator" is a SQL syntactic clause that can be
3055 applied to a field - the operator goes before the field
3057 You can write your own operator handlers - supply a C<unary_ops>
3058 argument to the C<new> method. That argument takes an arrayref of
3059 operator definitions; each operator definition is a hashref with two
3066 the regular expression to match the operator
3070 Either a coderef or a plain scalar method name. In both cases
3071 the expected return is C<< $sql >>.
3073 When supplied with a method name, it is simply called on the
3074 L<SQL::Abstract> object as:
3076 $self->$method_name($op, $arg)
3080 $op is the part that matched the handler regex
3081 $arg is the RHS or argument of the operator
3083 When supplied with a coderef, it is called as:
3085 $coderef->($self, $op, $arg)
3093 Thanks to some benchmarking by Mark Stosberg, it turns out that
3094 this module is many orders of magnitude faster than using C<DBIx::Abstract>.
3095 I must admit this wasn't an intentional design issue, but it's a
3096 byproduct of the fact that you get to control your C<DBI> handles
3099 To maximize performance, use a code snippet like the following:
3101 # prepare a statement handle using the first row
3102 # and then reuse it for the rest of the rows
3104 for my $href (@array_of_hashrefs) {
3105 $stmt ||= $sql->insert('table', $href);
3106 $sth ||= $dbh->prepare($stmt);
3107 $sth->execute($sql->values($href));
3110 The reason this works is because the keys in your C<$href> are sorted
3111 internally by B<SQL::Abstract>. Thus, as long as your data retains
3112 the same structure, you only have to generate the SQL the first time
3113 around. On subsequent queries, simply use the C<values> function provided
3114 by this module to return your values in the correct order.
3116 However this depends on the values having the same type - if, for
3117 example, the values of a where clause may either have values
3118 (resulting in sql of the form C<column = ?> with a single bind
3119 value), or alternatively the values might be C<undef> (resulting in
3120 sql of the form C<column IS NULL> with no bind value) then the
3121 caching technique suggested will not work.
3125 If you use my C<CGI::FormBuilder> module at all, you'll hopefully
3126 really like this part (I do, at least). Building up a complex query
3127 can be as simple as the following:
3134 use CGI::FormBuilder;
3137 my $form = CGI::FormBuilder->new(...);
3138 my $sql = SQL::Abstract->new;
3140 if ($form->submitted) {
3141 my $field = $form->field;
3142 my $id = delete $field->{id};
3143 my($stmt, @bind) = $sql->update('table', $field, {id => $id});
3146 Of course, you would still have to connect using C<DBI> to run the
3147 query, but the point is that if you make your form look like your
3148 table, the actual query script can be extremely simplistic.
3150 If you're B<REALLY> lazy (I am), check out C<HTML::QuickTable> for
3151 a fast interface to returning and formatting data. I frequently
3152 use these three modules together to write complex database query
3153 apps in under 50 lines.
3155 =head1 HOW TO CONTRIBUTE
3157 Contributions are always welcome, in all usable forms (we especially
3158 welcome documentation improvements). The delivery methods include git-
3159 or unified-diff formatted patches, GitHub pull requests, or plain bug
3160 reports either via RT or the Mailing list. Contributors are generally
3161 granted full access to the official repository after their first several
3162 patches pass successful review.
3164 This project is maintained in a git repository. The code and related tools are
3165 accessible at the following locations:
3169 =item * Official repo: L<git://git.shadowcat.co.uk/dbsrgits/SQL-Abstract.git>
3171 =item * Official gitweb: L<http://git.shadowcat.co.uk/gitweb/gitweb.cgi?p=dbsrgits/SQL-Abstract.git>
3173 =item * GitHub mirror: L<https://github.com/dbsrgits/sql-abstract>
3175 =item * Authorized committers: L<ssh://dbsrgits@git.shadowcat.co.uk/SQL-Abstract.git>
3181 Version 1.50 was a major internal refactoring of C<SQL::Abstract>.
3182 Great care has been taken to preserve the I<published> behavior
3183 documented in previous versions in the 1.* family; however,
3184 some features that were previously undocumented, or behaved
3185 differently from the documentation, had to be changed in order
3186 to clarify the semantics. Hence, client code that was relying
3187 on some dark areas of C<SQL::Abstract> v1.*
3188 B<might behave differently> in v1.50.
3190 The main changes are:
3196 support for literal SQL through the C<< \ [ $sql, @bind ] >> syntax.
3200 support for the { operator => \"..." } construct (to embed literal SQL)
3204 support for the { operator => \["...", @bind] } construct (to embed literal SQL with bind values)
3208 optional support for L<array datatypes|/"Inserting and Updating Arrays">
3212 defensive programming: check arguments
3216 fixed bug with global logic, which was previously implemented
3217 through global variables yielding side-effects. Prior versions would
3218 interpret C<< [ {cond1, cond2}, [cond3, cond4] ] >>
3219 as C<< "(cond1 AND cond2) OR (cond3 AND cond4)" >>.
3220 Now this is interpreted
3221 as C<< "(cond1 AND cond2) OR (cond3 OR cond4)" >>.
3226 fixed semantics of _bindtype on array args
3230 dropped the C<_anoncopy> of the %where tree. No longer necessary,
3231 we just avoid shifting arrays within that tree.
3235 dropped the C<_modlogic> function
3239 =head1 ACKNOWLEDGEMENTS
3241 There are a number of individuals that have really helped out with
3242 this module. Unfortunately, most of them submitted bugs via CPAN
3243 so I have no idea who they are! But the people I do know are:
3245 Ash Berlin (order_by hash term support)
3246 Matt Trout (DBIx::Class support)
3247 Mark Stosberg (benchmarking)
3248 Chas Owens (initial "IN" operator support)
3249 Philip Collins (per-field SQL functions)
3250 Eric Kolve (hashref "AND" support)
3251 Mike Fragassi (enhancements to "BETWEEN" and "LIKE")
3252 Dan Kubb (support for "quote_char" and "name_sep")
3253 Guillermo Roditi (patch to cleanup "IN" and "BETWEEN", fix and tests for _order_by)
3254 Laurent Dami (internal refactoring, extensible list of special operators, literal SQL)
3255 Norbert Buchmuller (support for literal SQL in hashpair, misc. fixes & tests)
3256 Peter Rabbitson (rewrite of SQLA::Test, misc. fixes & tests)
3257 Oliver Charles (support for "RETURNING" after "INSERT")
3263 L<DBIx::Class>, L<DBIx::Abstract>, L<CGI::FormBuilder>, L<HTML::QuickTable>.
3267 Copyright (c) 2001-2007 Nathan Wiger <nwiger@cpan.org>. All Rights Reserved.
3269 This module is actively maintained by Matt Trout <mst@shadowcatsystems.co.uk>
3271 For support, your best bet is to try the C<DBIx::Class> users mailing list.
3272 While not an official support venue, C<DBIx::Class> makes heavy use of
3273 C<SQL::Abstract>, and as such list members there are very familiar with
3274 how to create queries.
3278 This module is free software; you may copy this under the same
3279 terms as perl itself (either the GNU General Public License or
3280 the Artistic License)