1 package SQL::Abstract; # see doc at end of file
10 our @EXPORT_OK = qw(is_plain_value is_literal_value);
20 *SQL::Abstract::_ENV_::DETECT_AUTOGENERATED_STRINGIFICATION = $ENV{SQLA_ISVALUE_IGNORE_AUTOGENERATED_STRINGIFICATION}
26 #======================================================================
28 #======================================================================
30 our $VERSION = '1.87';
32 # This would confuse some packagers
33 $VERSION = eval $VERSION if $VERSION =~ /_/; # numify for warning-free dev releases
37 # special operators (-in, -between). May be extended/overridden by user.
38 # See section WHERE: BUILTIN SPECIAL OPERATORS below for implementation
39 my @BUILTIN_SPECIAL_OPS = (
40 {regex => qr/^ (?: not \s )? between $/ix, handler => sub { die "NOPE" }},
41 {regex => qr/^ is (?: \s+ not )? $/ix, handler => sub { die "NOPE" }},
42 {regex => qr/^ (?: not \s )? in $/ix, handler => sub { die "NOPE" }},
45 #======================================================================
46 # DEBUGGING AND ERROR REPORTING
47 #======================================================================
50 return unless $_[0]->{debug}; shift; # a little faster
51 my $func = (caller(1))[3];
52 warn "[$func] ", @_, "\n";
56 my($func) = (caller(1))[3];
57 Carp::carp "[$func] Warning: ", @_;
61 my($func) = (caller(1))[3];
62 Carp::croak "[$func] Fatal: ", @_;
65 sub is_literal_value ($) {
66 ref $_[0] eq 'SCALAR' ? [ ${$_[0]} ]
67 : ( ref $_[0] eq 'REF' and ref ${$_[0]} eq 'ARRAY' ) ? [ @${ $_[0] } ]
71 sub is_undef_value ($) {
75 and exists $_[0]->{-value}
76 and not defined $_[0]->{-value}
80 # FIXME XSify - this can be done so much more efficiently
81 sub is_plain_value ($) {
83 ! length ref $_[0] ? \($_[0])
85 ref $_[0] eq 'HASH' and keys %{$_[0]} == 1
87 exists $_[0]->{-value}
88 ) ? \($_[0]->{-value})
90 # reuse @_ for even moar speedz
91 defined ( $_[1] = Scalar::Util::blessed $_[0] )
93 # deliberately not using Devel::OverloadInfo - the checks we are
94 # intersted in are much more limited than the fullblown thing, and
95 # this is a very hot piece of code
97 # simply using ->can('(""') can leave behind stub methods that
98 # break actually using the overload later (see L<perldiag/Stub
99 # found while resolving method "%s" overloading "%s" in package
100 # "%s"> and the source of overload::mycan())
102 # either has stringification which DBI SHOULD prefer out of the box
103 grep { *{ (qq[${_}::(""]) }{CODE} } @{ $_[2] = mro::get_linear_isa( $_[1] ) }
105 # has nummification or boolification, AND fallback is *not* disabled
107 SQL::Abstract::_ENV_::DETECT_AUTOGENERATED_STRINGIFICATION
110 grep { *{"${_}::(0+"}{CODE} } @{$_[2]}
112 grep { *{"${_}::(bool"}{CODE} } @{$_[2]}
116 # no fallback specified at all
117 ! ( ($_[3]) = grep { *{"${_}::()"}{CODE} } @{$_[2]} )
119 # fallback explicitly undef
120 ! defined ${"$_[3]::()"}
133 #======================================================================
135 #======================================================================
139 my $class = ref($self) || $self;
140 my %opt = (ref $_[0] eq 'HASH') ? %{$_[0]} : @_;
142 # choose our case by keeping an option around
143 delete $opt{case} if $opt{case} && $opt{case} ne 'lower';
145 # default logic for interpreting arrayrefs
146 $opt{logic} = $opt{logic} ? uc $opt{logic} : 'OR';
148 # how to return bind vars
149 $opt{bindtype} ||= 'normal';
151 # default comparison is "=", but can be overridden
154 # try to recognize which are the 'equality' and 'inequality' ops
155 # (temporary quickfix (in 2007), should go through a more seasoned API)
156 $opt{equality_op} = qr/^( \Q$opt{cmp}\E | \= )$/ix;
157 $opt{inequality_op} = qr/^( != | <> )$/ix;
159 $opt{like_op} = qr/^ (is_)?r?like $/xi;
160 $opt{not_like_op} = qr/^ (is_)?not_r?like $/xi;
163 $opt{sqltrue} ||= '1=1';
164 $opt{sqlfalse} ||= '0=1';
167 $opt{special_ops} ||= [];
169 # regexes are applied in order, thus push after user-defines
170 push @{$opt{special_ops}}, @BUILTIN_SPECIAL_OPS;
172 if ($class->isa('DBIx::Class::SQLMaker')) {
173 push @{$opt{special_ops}}, our $DBIC_Compat_Op ||= {
174 regex => qr/^(?:ident|value|(?:not\s)?in)$/i, handler => sub { die "NOPE" }
176 $opt{is_dbic_sqlmaker} = 1;
180 $opt{unary_ops} ||= [];
182 # rudimentary sanity-check for user supplied bits treated as functions/operators
183 # If a purported function matches this regular expression, an exception is thrown.
184 # Literal SQL is *NOT* subject to this check, only functions (and column names
185 # when quoting is not in effect)
188 # need to guard against ()'s in column names too, but this will break tons of
189 # hacks... ideas anyone?
190 $opt{injection_guard} ||= qr/
196 $opt{expand_unary} = {};
199 -not => '_expand_not',
200 -bool => '_expand_bool',
201 -and => '_expand_op_andor',
202 -or => '_expand_op_andor',
203 -nest => '_expand_nest',
204 -bind => sub { shift; +{ @_ } },
206 -not_in => '_expand_in',
210 'between' => '_expand_between',
211 'not_between' => '_expand_between',
212 'in' => '_expand_in',
213 'not_in' => '_expand_in',
214 'nest' => '_expand_nest',
215 (map +($_ => '_expand_op_andor'), ('and', 'or')),
216 (map +($_ => '_expand_op_is'), ('is', 'is_not')),
219 # placeholder for _expand_unop system
221 my %unops = (-ident => '_expand_ident', -value => '_expand_value');
222 foreach my $name (keys %unops) {
223 $opt{expand}{$name} = $unops{$name};
224 my ($op) = $name =~ /^-(.*)$/;
225 $opt{expand_op}{$op} = sub {
226 my ($self, $op, $arg, $k) = @_;
227 return $self->_expand_expr_hashpair_cmp(
228 $k, { "-${op}" => $arg }
235 (map +("-$_", "_render_$_"), qw(op func bind ident literal list)),
240 (map +($_ => '_render_op_between'), 'between', 'not_between'),
241 (map +($_ => '_render_op_in'), 'in', 'not_in'),
242 (map +($_ => '_render_unop_postfix'),
243 'is_null', 'is_not_null', 'asc', 'desc',
245 (not => '_render_op_not'),
246 (map +($_ => '_render_op_andor'), qw(and or)),
249 return bless \%opt, $class;
252 sub sqltrue { +{ -literal => [ $_[0]->{sqltrue} ] } }
253 sub sqlfalse { +{ -literal => [ $_[0]->{sqlfalse} ] } }
255 sub _assert_pass_injection_guard {
256 if ($_[1] =~ $_[0]->{injection_guard}) {
257 my $class = ref $_[0];
258 puke "Possible SQL injection attempt '$_[1]'. If this is indeed a part of the "
259 . "desired SQL use literal SQL ( \'...' or \[ '...' ] ) or supply your own "
260 . "{injection_guard} attribute to ${class}->new()"
265 #======================================================================
267 #======================================================================
271 my $table = $self->_table(shift);
272 my $data = shift || return;
275 my $method = $self->_METHOD_FOR_refkind("_insert", $data);
276 my ($sql, @bind) = $self->$method($data);
277 $sql = join " ", $self->_sqlcase('insert into'), $table, $sql;
279 if ($options->{returning}) {
280 my ($s, @b) = $self->_insert_returning($options);
285 return wantarray ? ($sql, @bind) : $sql;
288 # So that subclasses can override INSERT ... RETURNING separately from
289 # UPDATE and DELETE (e.g. DBIx::Class::SQLMaker::Oracle does this)
290 sub _insert_returning { shift->_returning(@_) }
293 my ($self, $options) = @_;
295 my $f = $options->{returning};
297 my ($sql, @bind) = $self->render_aqt(
298 $self->_expand_maybe_list_expr($f, undef, -ident)
301 ? $self->_sqlcase(' returning ') . $sql
302 : ($self->_sqlcase(' returning ').$sql, @bind);
305 sub _insert_HASHREF { # explicit list of fields and then values
306 my ($self, $data) = @_;
308 my @fields = sort keys %$data;
310 my ($sql, @bind) = $self->_insert_values($data);
313 $_ = $self->_quote($_) foreach @fields;
314 $sql = "( ".join(", ", @fields).") ".$sql;
316 return ($sql, @bind);
319 sub _insert_ARRAYREF { # just generate values(?,?) part (no list of fields)
320 my ($self, $data) = @_;
322 # no names (arrayref) so can't generate bindtype
323 $self->{bindtype} ne 'columns'
324 or belch "can't do 'columns' bindtype when called with arrayref";
326 my (@values, @all_bind);
327 foreach my $value (@$data) {
328 my ($values, @bind) = $self->_insert_value(undef, $value);
329 push @values, $values;
330 push @all_bind, @bind;
332 my $sql = $self->_sqlcase('values')." ( ".join(", ", @values)." )";
333 return ($sql, @all_bind);
336 sub _insert_ARRAYREFREF { # literal SQL with bind
337 my ($self, $data) = @_;
339 my ($sql, @bind) = @${$data};
340 $self->_assert_bindval_matches_bindtype(@bind);
342 return ($sql, @bind);
346 sub _insert_SCALARREF { # literal SQL without bind
347 my ($self, $data) = @_;
353 my ($self, $data) = @_;
355 my (@values, @all_bind);
356 foreach my $column (sort keys %$data) {
357 my ($values, @bind) = $self->_insert_value($column, $data->{$column});
358 push @values, $values;
359 push @all_bind, @bind;
361 my $sql = $self->_sqlcase('values')." ( ".join(", ", @values)." )";
362 return ($sql, @all_bind);
366 my ($self, $column, $v) = @_;
368 return $self->render_aqt(
369 $self->_expand_insert_value($column, $v)
373 sub _expand_insert_value {
374 my ($self, $column, $v) = @_;
376 if (ref($v) eq 'ARRAY') {
377 if ($self->{array_datatypes}) {
378 return +{ -bind => [ $column, $v ] };
380 my ($sql, @bind) = @$v;
381 $self->_assert_bindval_matches_bindtype(@bind);
382 return +{ -literal => $v };
384 if (ref($v) eq 'HASH') {
385 if (grep !/^-/, keys %$v) {
386 belch "HASH ref as bind value in insert is not supported";
387 return +{ -bind => [ $column, $v ] };
391 return +{ -bind => [ $column, undef ] };
393 local our $Cur_Col_Meta = $column;
394 return $self->expand_expr($v);
399 #======================================================================
401 #======================================================================
406 my $table = $self->_table(shift);
407 my $data = shift || return;
411 # first build the 'SET' part of the sql statement
412 puke "Unsupported data type specified to \$sql->update"
413 unless ref $data eq 'HASH';
415 my ($sql, @all_bind) = $self->_update_set_values($data);
416 $sql = $self->_sqlcase('update ') . $table . $self->_sqlcase(' set ')
420 my($where_sql, @where_bind) = $self->where($where);
422 push @all_bind, @where_bind;
425 if ($options->{returning}) {
426 my ($returning_sql, @returning_bind) = $self->_update_returning($options);
427 $sql .= $returning_sql;
428 push @all_bind, @returning_bind;
431 return wantarray ? ($sql, @all_bind) : $sql;
434 sub _update_set_values {
435 my ($self, $data) = @_;
437 return $self->render_aqt(
438 $self->_expand_update_set_values($data),
442 sub _expand_update_set_values {
443 my ($self, $data) = @_;
444 $self->_expand_maybe_list_expr( [
447 $set = { -bind => $_ } unless defined $set;
448 +{ -op => [ '=', $self->_expand_ident(-ident => $k), $set ] };
454 ? ($self->{array_datatypes}
455 ? [ $k, +{ -bind => [ $k, $v ] } ]
456 : [ $k, +{ -literal => $v } ])
458 local our $Cur_Col_Meta = $k;
459 [ $k, $self->_expand_expr($v) ]
466 # So that subclasses can override UPDATE ... RETURNING separately from
468 sub _update_returning { shift->_returning(@_) }
472 #======================================================================
474 #======================================================================
479 my $table = $self->_table(shift);
480 my $fields = shift || '*';
484 my ($fields_sql, @bind) = $self->_select_fields($fields);
486 my ($where_sql, @where_bind) = $self->where($where, $order);
487 push @bind, @where_bind;
489 my $sql = join(' ', $self->_sqlcase('select'), $fields_sql,
490 $self->_sqlcase('from'), $table)
493 return wantarray ? ($sql, @bind) : $sql;
497 my ($self, $fields) = @_;
498 return $fields unless ref($fields);
499 return $self->render_aqt(
500 $self->_expand_maybe_list_expr($fields, undef, '-ident')
504 #======================================================================
506 #======================================================================
511 my $table = $self->_table(shift);
515 my($where_sql, @bind) = $self->where($where);
516 my $sql = $self->_sqlcase('delete from ') . $table . $where_sql;
518 if ($options->{returning}) {
519 my ($returning_sql, @returning_bind) = $self->_delete_returning($options);
520 $sql .= $returning_sql;
521 push @bind, @returning_bind;
524 return wantarray ? ($sql, @bind) : $sql;
527 # So that subclasses can override DELETE ... RETURNING separately from
529 sub _delete_returning { shift->_returning(@_) }
533 #======================================================================
535 #======================================================================
539 # Finally, a separate routine just to handle WHERE clauses
541 my ($self, $where, $order) = @_;
543 local $self->{convert_where} = $self->{convert};
546 my ($sql, @bind) = defined($where)
547 ? $self->_recurse_where($where)
549 $sql = (defined $sql and length $sql) ? $self->_sqlcase(' where ') . "( $sql )" : '';
553 my ($order_sql, @order_bind) = $self->_order_by($order);
555 push @bind, @order_bind;
558 return wantarray ? ($sql, @bind) : $sql;
561 { our $Default_Scalar_To = -value }
564 my ($self, $expr, $default_scalar_to) = @_;
565 local our $Default_Scalar_To = $default_scalar_to if $default_scalar_to;
566 $self->_expand_expr($expr);
570 my ($self, $aqt) = @_;
571 my ($k, $v, @rest) = %$aqt;
573 if (my $meth = $self->{render}{$k}) {
574 return $self->$meth($v);
576 die "notreached: $k";
580 my ($self, $expr) = @_;
581 $self->render_aqt($self->expand_expr($expr));
585 my ($self, $raw) = @_;
586 s/^-(?=.)//, s/\s+/_/g for my $op = lc $raw;
591 my ($self, $expr) = @_;
592 our $Expand_Depth ||= 0; local $Expand_Depth = $Expand_Depth + 1;
593 return undef unless defined($expr);
594 if (ref($expr) eq 'HASH') {
595 return undef unless my $kc = keys %$expr;
597 return $self->_expand_op_andor(-and => $expr);
599 my ($key, $value) = %$expr;
600 if ($key =~ /^-/ and $key =~ s/ [_\s]? \d+ $//x ) {
601 belch 'Use of [and|or|nest]_N modifiers is deprecated and will be removed in SQLA v2.0. '
602 . "You probably wanted ...-and => [ $key => COND1, $key => COND2 ... ]";
604 if (my $exp = $self->{expand}{$key}) {
605 return $self->$exp($key, $value);
607 return $self->_expand_expr_hashpair($key, $value);
609 if (ref($expr) eq 'ARRAY') {
610 my $logic = '-'.lc($self->{logic});
611 return $self->_expand_op_andor($logic, $expr);
613 if (my $literal = is_literal_value($expr)) {
614 return +{ -literal => $literal };
616 if (!ref($expr) or Scalar::Util::blessed($expr)) {
617 return $self->_expand_expr_scalar($expr);
622 sub _expand_expr_hashpair {
623 my ($self, $k, $v) = @_;
624 unless (defined($k) and length($k)) {
625 if (defined($k) and my $literal = is_literal_value($v)) {
626 belch 'Hash-pairs consisting of an empty string with a literal are deprecated, and will be removed in 2.0: use -and => [ $literal ] instead';
627 return { -literal => $literal };
629 puke "Supplying an empty left hand side argument is not supported";
632 return $self->_expand_expr_hashpair_op($k, $v);
634 return $self->_expand_expr_hashpair_ident($k, $v);
637 sub _expand_expr_hashpair_ident {
638 my ($self, $k, $v) = @_;
640 local our $Cur_Col_Meta = $k;
642 # hash with multiple or no elements is andor
644 if (ref($v) eq 'HASH' and keys %$v != 1) {
645 return $self->_expand_op_andor(-and => $v, $k);
648 # undef needs to be re-sent with cmp to achieve IS/IS NOT NULL
650 if (is_undef_value($v)) {
651 return $self->_expand_expr_hashpair_cmp($k => undef);
654 # scalars and objects get expanded as whatever requested or values
656 if (!ref($v) or Scalar::Util::blessed($v)) {
657 return $self->_expand_expr_hashpair_scalar($k, $v);
660 # single key hashref is a hashtriple
662 if (ref($v) eq 'HASH') {
663 return $self->_expand_expr_hashtriple($k, %$v);
666 # arrayref needs re-engineering over the elements
668 if (ref($v) eq 'ARRAY') {
669 return $self->sqlfalse unless @$v;
670 $self->_debug("ARRAY($k) means distribute over elements");
672 $v->[0] =~ /^-(and|or)$/i
673 ? shift(@{$v = [ @$v ]})
674 : '-'.lc($self->{logic} || 'OR')
676 return $self->_expand_op_andor(
681 if (my $literal = is_literal_value($v)) {
683 belch 'Hash-pairs consisting of an empty string with a literal are deprecated, and will be removed in 2.0: use -and => [ $literal ] instead';
686 my ($sql, @bind) = @$literal;
687 if ($self->{bindtype} eq 'columns') {
689 $self->_assert_bindval_matches_bindtype($_);
692 return +{ -literal => [ $self->_quote($k).' '.$sql, @bind ] };
697 sub _expand_expr_scalar {
698 my ($self, $expr) = @_;
700 return $self->_expand_expr({ (our $Default_Scalar_To) => $expr });
703 sub _expand_expr_hashpair_scalar {
704 my ($self, $k, $v) = @_;
706 return $self->_expand_expr_hashpair_cmp(
707 $k, $self->_expand_expr_scalar($v),
711 sub _expand_expr_hashpair_op {
712 my ($self, $k, $v) = @_;
714 $self->_assert_pass_injection_guard($k =~ /\A-(.*)\Z/s);
716 my $op = $self->_normalize_op($k);
718 # Ops prefixed with -not_ get converted
720 if (my ($rest) = $op =~/^not_(.*)$/) {
723 $self->_expand_expr({ "-${rest}", $v })
729 my $op = join(' ', split '_', $op);
731 # the old special op system requires illegality for top-level use
734 (our $Expand_Depth) == 1
735 and List::Util::first { $op =~ $_->{regex} } @{$self->{special_ops}}
737 puke "Illegal use of top-level '-$op'"
740 # the old unary op system means we should touch nothing and let it work
742 if (my $us = List::Util::first { $op =~ $_->{regex} } @{$self->{unary_ops}}) {
743 return { -op => [ $op, $v ] };
747 # an explicit node type is currently assumed to be expanded (this is almost
748 # certainly wrong and there should be expansion anyway)
750 if ($self->{render}{$k}) {
754 # hashref RHS values get expanded and used as op/func args
759 and (keys %$v)[0] =~ /^-/
761 my ($func) = $k =~ /^-(.*)$/;
763 if (List::Util::first { $func =~ $_->{regex} } @{$self->{special_ops}}) {
764 return +{ -op => [ $func, $self->_expand_expr($v) ] };
767 return +{ -func => [ $func, $self->_expand_expr($v) ] };
770 # scalars and literals get simply expanded
772 if (!ref($v) or is_literal_value($v)) {
773 return +{ -op => [ $op, $self->_expand_expr($v) ] };
779 sub _expand_expr_hashpair_cmp {
780 my ($self, $k, $v) = @_;
781 $self->_expand_expr_hashtriple($k, $self->{cmp}, $v);
784 sub _expand_expr_hashtriple {
785 my ($self, $k, $vk, $vv) = @_;
787 my $ik = $self->_expand_ident(-ident => $k);
789 my $op = $self->_normalize_op($vk);
790 $self->_assert_pass_injection_guard($op);
792 if ($op =~ s/ _? \d+ $//x ) {
793 return $self->_expand_expr($k, { $vk, $vv });
795 if (my $x = $self->{expand_op}{$op}) {
796 local our $Cur_Col_Meta = $k;
797 return $self->$x($op, $vv, $k);
801 my $op = join(' ', split '_', $op);
803 if (my $us = List::Util::first { $op =~ $_->{regex} } @{$self->{special_ops}}) {
804 return { -op => [ $op, $ik, $vv ] };
806 if (my $us = List::Util::first { $op =~ $_->{regex} } @{$self->{unary_ops}}) {
810 { -op => [ $op, $vv ] }
814 if (ref($vv) eq 'ARRAY') {
816 my $logic = (defined($raw[0]) and $raw[0] =~ /^-(and|or)$/i)
817 ? shift @raw : '-or';
818 my @values = map +{ $vk => $_ }, @raw;
820 $op =~ $self->{inequality_op}
821 or $op =~ $self->{not_like_op}
823 if (lc($logic) eq '-or' and @values > 1) {
824 belch "A multi-element arrayref as an argument to the inequality op '${\uc(join ' ', split '_', $op)}' "
825 . 'is technically equivalent to an always-true 1=1 (you probably wanted '
826 . "to say ...{ \$inequality_op => [ -and => \@values ] }... instead)"
831 # try to DWIM on equality operators
832 return ($self->_dwim_op_to_is($op,
833 "Supplying an empty arrayref to '%s' is deprecated",
834 "operator '%s' applied on an empty array (field '$k')"
835 ) ? $self->sqlfalse : $self->sqltrue);
837 return $self->_expand_op_andor($logic => \@values, $k);
839 if (is_undef_value($vv)) {
840 my $is = ($self->_dwim_op_to_is($op,
841 "Supplying an undefined argument to '%s' is deprecated",
842 "unexpected operator '%s' with undef operand",
843 ) ? 'is' : 'is not');
845 return $self->_expand_expr_hashpair($k => { $is, undef });
847 local our $Cur_Col_Meta = $k;
851 $self->_expand_expr($vv)
856 my ($self, $raw, $empty, $fail) = @_;
858 my $op = $self->_normalize_op($raw);
860 if ($op =~ /^not$/i) {
863 if ($op =~ $self->{equality_op}) {
866 if ($op =~ $self->{like_op}) {
867 belch(sprintf $empty, uc(join ' ', split '_', $op));
870 if ($op =~ $self->{inequality_op}) {
873 if ($op =~ $self->{not_like_op}) {
874 belch(sprintf $empty, uc(join ' ', split '_', $op));
877 puke(sprintf $fail, $op);
881 my ($self, $op, $body) = @_;
882 unless (defined($body) or (ref($body) and ref($body) eq 'ARRAY')) {
883 puke "$op requires a single plain scalar argument (a quotable identifier) or an arrayref of identifier parts";
885 my @parts = map split(/\Q${\($self->{name_sep}||'.')}\E/, $_),
886 ref($body) ? @$body : $body;
887 return { -ident => $parts[-1] } if $self->{_dequalify_idents};
888 unless ($self->{quote_char}) {
889 $self->_assert_pass_injection_guard($_) for @parts;
891 return +{ -ident => \@parts };
895 +{ -bind => [ our $Cur_Col_Meta, $_[2] ] };
899 +{ -op => [ 'not', $_[0]->_expand_expr($_[2]) ] };
903 my ($self, undef, $v) = @_;
905 return $self->_expand_expr($v);
907 puke "-bool => undef not supported" unless defined($v);
908 return $self->_expand_ident(-ident => $v);
911 sub _expand_op_andor {
912 my ($self, $logic, $v, $k) = @_;
914 $v = [ map +{ $k, $_ },
916 ? (map +{ $_ => $v->{$_} }, sort keys %$v)
920 my ($logop) = $logic =~ /^-?(.*)$/;
921 if (ref($v) eq 'HASH') {
922 return undef unless keys %$v;
925 map $self->_expand_expr({ $_ => $v->{$_} }),
929 if (ref($v) eq 'ARRAY') {
930 $logop eq 'and' or $logop eq 'or' or puke "unknown logic: $logop";
933 (ref($_) eq 'ARRAY' and @$_)
934 or (ref($_) eq 'HASH' and %$_)
940 while (my ($el) = splice @expr, 0, 1) {
941 puke "Supplying an empty left hand side argument is not supported in array-pairs"
942 unless defined($el) and length($el);
943 my $elref = ref($el);
945 local our $Expand_Depth = 0;
946 push(@res, grep defined, $self->_expand_expr({ $el, shift(@expr) }));
947 } elsif ($elref eq 'ARRAY') {
948 push(@res, grep defined, $self->_expand_expr($el)) if @$el;
949 } elsif (my $l = is_literal_value($el)) {
950 push @res, { -literal => $l };
951 } elsif ($elref eq 'HASH') {
952 local our $Expand_Depth = 0;
953 push @res, grep defined, $self->_expand_expr($el) if %$el;
959 # return $res[0] if @res == 1;
960 return { -op => [ $logop, @res ] };
966 my ($self, $op, $vv, $k) = @_;
967 puke "$op can only take undef as argument"
971 and exists($vv->{-value})
972 and !defined($vv->{-value})
974 return +{ -op => [ $op.'_null', $self->_expand_ident(-ident => $k) ] };
977 sub _expand_between {
978 my ($self, $op, $vv, $k) = @_;
979 local our $Cur_Col_Meta = $k;
980 my @rhs = map $self->_expand_expr($_),
981 ref($vv) eq 'ARRAY' ? @$vv : $vv;
983 (@rhs == 1 and ref($rhs[0]) eq 'HASH' and $rhs[0]->{-literal})
985 (@rhs == 2 and defined($rhs[0]) and defined($rhs[1]))
987 puke "Operator '${\uc($op)}' requires either an arrayref with two defined values or expressions, or a single literal scalarref/arrayref-ref";
991 $self->_expand_ident(-ident => $k),
997 my ($self, $raw, $vv, $k) = @_;
998 $k = shift @{$vv = [ @$vv ]} unless defined $k;
999 local our $Cur_Col_Meta = $k;
1000 my $op = $self->_normalize_op($raw);
1001 if (my $literal = is_literal_value($vv)) {
1002 my ($sql, @bind) = @$literal;
1003 my $opened_sql = $self->_open_outer_paren($sql);
1005 $op, $self->_expand_ident(-ident => $k),
1006 [ { -literal => [ $opened_sql, @bind ] } ]
1010 'SQL::Abstract before v1.75 used to generate incorrect SQL when the '
1011 . "-${\uc($op)} operator was given an undef-containing list: !!!AUDIT YOUR CODE "
1012 . 'AND DATA!!! (the upcoming Data::Query-based version of SQL::Abstract '
1013 . 'will emit the logically correct SQL instead of raising this exception)'
1015 puke("Argument passed to the '${\uc($op)}' operator can not be undefined")
1017 my @rhs = map $self->expand_expr($_, -value),
1018 map { defined($_) ? $_: puke($undef_err) }
1019 (ref($vv) eq 'ARRAY' ? @$vv : $vv);
1020 return $self->${\($op =~ /^not/ ? 'sqltrue' : 'sqlfalse')} unless @rhs;
1024 $self->_expand_ident(-ident => $k),
1030 my ($self, $op, $v) = @_;
1031 # DBIx::Class requires a nest warning to be emitted once but the private
1032 # method it overrode to do so no longer exists
1033 if ($self->{is_dbic_sqlmaker}) {
1034 unless (our $Nest_Warned) {
1036 "-nest in search conditions is deprecated, you most probably wanted:\n"
1037 .q|{..., -and => [ \%cond0, \@cond1, \'cond2', \[ 'cond3', [ col => bind ] ], etc. ], ... }|
1042 return $self->_expand_expr($v);
1045 sub _recurse_where {
1046 my ($self, $where, $logic) = @_;
1048 # Special case: top level simple string treated as literal
1050 my $where_exp = (ref($where)
1051 ? $self->_expand_expr($where, $logic)
1052 : { -literal => [ $where ] });
1054 # dispatch expanded expression
1056 my ($sql, @bind) = defined($where_exp) ? $self->render_aqt($where_exp) : (undef);
1057 # DBIx::Class used to call _recurse_where in scalar context
1058 # something else might too...
1060 return ($sql, @bind);
1063 belch "Calling _recurse_where in scalar context is deprecated and will go away before 2.0";
1069 my ($self, $ident) = @_;
1071 return $self->_convert($self->_quote($ident));
1075 my ($self, $list) = @_;
1076 my @parts = grep length($_->[0]), map [ $self->render_aqt($_) ], @$list;
1077 return join(', ', map $_->[0], @parts), map @{$_}[1..$#$_], @parts;
1081 my ($self, $rest) = @_;
1082 my ($func, @args) = @$rest;
1086 push @arg_sql, shift @x;
1088 } map [ $self->render_aqt($_) ], @args;
1089 return ($self->_sqlcase($func).'('.join(', ', @arg_sql).')', @bind);
1093 my ($self, $bind) = @_;
1094 return ($self->_convert('?'), $self->_bindtype(@$bind));
1097 sub _render_literal {
1098 my ($self, $literal) = @_;
1099 $self->_assert_bindval_matches_bindtype(@{$literal}[1..$#$literal]);
1104 my ($self, $v) = @_;
1105 my ($op, @args) = @$v;
1106 if (my $r = $self->{render_op}{$op}) {
1107 return $self->$r($op, \@args);
1112 my $op = join(' ', split '_', $op);
1114 my $us = List::Util::first { $op =~ $_->{regex} } @{$self->{special_ops}};
1115 if ($us and @args > 1) {
1116 puke "Special op '${op}' requires first value to be identifier"
1117 unless my ($ident) = map $_->{-ident}, grep ref($_) eq 'HASH', $args[0];
1118 my $k = join(($self->{name_sep}||'.'), @$ident);
1119 local our $Expand_Depth = 1;
1120 return $self->${\($us->{handler})}($k, $op, $args[1]);
1122 if (my $us = List::Util::first { $op =~ $_->{regex} } @{$self->{unary_ops}}) {
1123 return $self->${\($us->{handler})}($op, $args[0]);
1128 return $self->_render_unop_prefix($op, \@args);
1130 return $self->_render_op_multop($op, \@args);
1136 sub _render_op_between {
1137 my ($self, $op, $args) = @_;
1138 my ($left, $low, $high) = @$args;
1139 my ($rhsql, @rhbind) = do {
1141 puke "Single arg to between must be a literal"
1142 unless $low->{-literal};
1145 my ($l, $h) = map [ $self->render_aqt($_) ], $low, $high;
1146 (join(' ', $l->[0], $self->_sqlcase('and'), $h->[0]),
1147 @{$l}[1..$#$l], @{$h}[1..$#$h])
1150 my ($lhsql, @lhbind) = $self->render_aqt($left);
1154 $self->_sqlcase(join ' ', split '_', $op),
1162 my ($self, $op, $args) = @_;
1163 my ($lhs, $rhs) = @$args;
1166 my ($sql, @bind) = $self->render_aqt($_);
1167 push @in_bind, @bind;
1170 my ($lhsql, @lbind) = $self->render_aqt($lhs);
1172 $lhsql.' '.$self->_sqlcase(join ' ', split '_', $op).' ( '
1173 .join(', ', @in_sql)
1179 sub _render_op_andor {
1180 my ($self, $op, $args) = @_;
1181 my @parts = grep length($_->[0]), map [ $self->render_aqt($_) ], @$args;
1182 return '' unless @parts;
1183 return @{$parts[0]} if @parts == 1;
1184 my ($sql, @bind) = $self->_render_op_multop($op, $args);
1185 return '( '.$sql.' )', @bind;
1188 sub _render_op_multop {
1189 my ($self, $op, $args) = @_;
1190 my @parts = grep length($_->[0]), map [ $self->render_aqt($_) ], @$args;
1191 return '' unless @parts;
1192 return @{$parts[0]} if @parts == 1;
1193 my ($final_sql) = join(
1194 ' '.$self->_sqlcase(join ' ', split '_', $op).' ',
1199 map @{$_}[1..$#$_], @parts
1202 sub _render_op_not {
1203 my ($self, $op, $v) = @_;
1204 my ($sql, @bind) = $self->_render_unop_prefix($op, $v);
1205 return "(${sql})", @bind;
1208 sub _render_unop_prefix {
1209 my ($self, $op, $v) = @_;
1210 my ($expr_sql, @bind) = $self->render_aqt($v->[0]);
1212 my $op_sql = $self->_sqlcase($op); # join ' ', split '_', $op);
1213 return ("${op_sql} ${expr_sql}", @bind);
1216 sub _render_unop_postfix {
1217 my ($self, $op, $v) = @_;
1218 my ($expr_sql, @bind) = $self->render_aqt($v->[0]);
1219 my $op_sql = $self->_sqlcase(join ' ', split '_', $op);
1220 return ($expr_sql.' '.$op_sql, @bind);
1223 # Some databases (SQLite) treat col IN (1, 2) different from
1224 # col IN ( (1, 2) ). Use this to strip all outer parens while
1225 # adding them back in the corresponding method
1226 sub _open_outer_paren {
1227 my ($self, $sql) = @_;
1229 while (my ($inner) = $sql =~ /^ \s* \( (.*) \) \s* $/xs) {
1231 # there are closing parens inside, need the heavy duty machinery
1232 # to reevaluate the extraction starting from $sql (full reevaluation)
1233 if ($inner =~ /\)/) {
1234 require Text::Balanced;
1236 my (undef, $remainder) = do {
1237 # idiotic design - writes to $@ but *DOES NOT* throw exceptions
1239 Text::Balanced::extract_bracketed($sql, '()', qr/\s*/);
1242 # the entire expression needs to be a balanced bracketed thing
1243 # (after an extract no remainder sans trailing space)
1244 last if defined $remainder and $remainder =~ /\S/;
1254 #======================================================================
1256 #======================================================================
1258 sub _expand_order_by {
1259 my ($self, $arg) = @_;
1261 return unless defined($arg) and not (ref($arg) eq 'ARRAY' and !@$arg);
1263 my $expander = sub {
1264 my ($self, $dir, $expr) = @_;
1265 my @to_expand = ref($expr) eq 'ARRAY' ? @$expr : $expr;
1266 foreach my $arg (@to_expand) {
1270 and grep /^-(asc|desc)$/, keys %$arg
1272 puke "ordering direction hash passed to order by must have exactly one key (-asc or -desc)";
1276 defined($dir) ? { -op => [ $dir =~ /^-?(.*)$/ ,=> $_ ] } : $_
1278 map $self->expand_expr($_, -ident),
1279 map ref($_) eq 'ARRAY' ? @$_ : $_, @to_expand;
1280 return (@exp > 1 ? { -list => \@exp } : $exp[0]);
1283 local @{$self->{expand}}{qw(-asc -desc)} = (($expander) x 2);
1285 return $self->$expander(undef, $arg);
1289 my ($self, $arg) = @_;
1291 return '' unless defined(my $expanded = $self->_expand_order_by($arg));
1293 my ($sql, @bind) = $self->render_aqt($expanded);
1295 return '' unless length($sql);
1297 my $final_sql = $self->_sqlcase(' order by ').$sql;
1299 return wantarray ? ($final_sql, @bind) : $final_sql;
1302 # _order_by no longer needs to call this so doesn't but DBIC uses it.
1304 sub _order_by_chunks {
1305 my ($self, $arg) = @_;
1307 return () unless defined(my $expanded = $self->_expand_order_by($arg));
1309 return $self->_chunkify_order_by($expanded);
1312 sub _chunkify_order_by {
1313 my ($self, $expanded) = @_;
1315 return grep length, $self->render_aqt($expanded)
1316 if $expanded->{-ident} or @{$expanded->{-literal}||[]} == 1;
1319 if (ref() eq 'HASH' and my $l = $_->{-list}) {
1320 return map $self->_chunkify_order_by($_), @$l;
1322 return [ $self->render_aqt($_) ];
1326 #======================================================================
1327 # DATASOURCE (FOR NOW, JUST PLAIN TABLE OR LIST OF TABLES)
1328 #======================================================================
1334 $self->_expand_maybe_list_expr($from, undef, -ident)
1339 #======================================================================
1341 #======================================================================
1343 sub _expand_maybe_list_expr {
1344 my ($self, $expr, $logic, $default) = @_;
1346 if (ref($expr) eq 'ARRAY') {
1348 map $self->expand_expr($_, $default), @$expr
1355 return $self->expand_expr($e, $default);
1358 # highly optimized, as it's called way too often
1360 # my ($self, $label) = @_;
1362 return '' unless defined $_[1];
1363 return ${$_[1]} if ref($_[1]) eq 'SCALAR';
1364 puke 'Identifier cannot be hashref' if ref($_[1]) eq 'HASH';
1366 unless ($_[0]->{quote_char}) {
1367 if (ref($_[1]) eq 'ARRAY') {
1368 return join($_[0]->{name_sep}||'.', @{$_[1]});
1370 $_[0]->_assert_pass_injection_guard($_[1]);
1375 my $qref = ref $_[0]->{quote_char};
1377 !$qref ? ($_[0]->{quote_char}, $_[0]->{quote_char})
1378 : ($qref eq 'ARRAY') ? @{$_[0]->{quote_char}}
1379 : puke "Unsupported quote_char format: $_[0]->{quote_char}";
1381 my $esc = $_[0]->{escape_char} || $r;
1383 # parts containing * are naturally unquoted
1385 $_[0]->{name_sep}||'',
1389 : do { (my $n = $_) =~ s/(\Q$esc\E|\Q$r\E)/$esc$1/g; $l . $n . $r }
1391 (ref($_[1]) eq 'ARRAY'
1395 ? split (/\Q$_[0]->{name_sep}\E/, $_[1] )
1403 # Conversion, if applicable
1405 #my ($self, $arg) = @_;
1406 if ($_[0]->{convert_where}) {
1407 return $_[0]->_sqlcase($_[0]->{convert_where}) .'(' . $_[1] . ')';
1414 #my ($self, $col, @vals) = @_;
1415 # called often - tighten code
1416 return $_[0]->{bindtype} eq 'columns'
1417 ? map {[$_[1], $_]} @_[2 .. $#_]
1422 # Dies if any element of @bind is not in [colname => value] format
1423 # if bindtype is 'columns'.
1424 sub _assert_bindval_matches_bindtype {
1425 # my ($self, @bind) = @_;
1427 if ($self->{bindtype} eq 'columns') {
1429 if (!defined $_ || ref($_) ne 'ARRAY' || @$_ != 2) {
1430 puke "bindtype 'columns' selected, you need to pass: [column_name => bind_value]"
1436 sub _join_sql_clauses {
1437 my ($self, $logic, $clauses_aref, $bind_aref) = @_;
1439 if (@$clauses_aref > 1) {
1440 my $join = " " . $self->_sqlcase($logic) . " ";
1441 my $sql = '( ' . join($join, @$clauses_aref) . ' )';
1442 return ($sql, @$bind_aref);
1444 elsif (@$clauses_aref) {
1445 return ($clauses_aref->[0], @$bind_aref); # no parentheses
1448 return (); # if no SQL, ignore @$bind_aref
1453 # Fix SQL case, if so requested
1455 # LDNOTE: if $self->{case} is true, then it contains 'lower', so we
1456 # don't touch the argument ... crooked logic, but let's not change it!
1457 return $_[0]->{case} ? $_[1] : uc($_[1]);
1461 #======================================================================
1462 # DISPATCHING FROM REFKIND
1463 #======================================================================
1466 my ($self, $data) = @_;
1468 return 'UNDEF' unless defined $data;
1470 # blessed objects are treated like scalars
1471 my $ref = (Scalar::Util::blessed $data) ? '' : ref $data;
1473 return 'SCALAR' unless $ref;
1476 while ($ref eq 'REF') {
1478 $ref = (Scalar::Util::blessed $data) ? '' : ref $data;
1482 return ($ref||'SCALAR') . ('REF' x $n_steps);
1486 my ($self, $data) = @_;
1487 my @try = ($self->_refkind($data));
1488 push @try, 'SCALAR_or_UNDEF' if $try[0] eq 'SCALAR' || $try[0] eq 'UNDEF';
1489 push @try, 'FALLBACK';
1493 sub _METHOD_FOR_refkind {
1494 my ($self, $meth_prefix, $data) = @_;
1497 for (@{$self->_try_refkind($data)}) {
1498 $method = $self->can($meth_prefix."_".$_)
1502 return $method || puke "cannot dispatch on '$meth_prefix' for ".$self->_refkind($data);
1506 sub _SWITCH_refkind {
1507 my ($self, $data, $dispatch_table) = @_;
1510 for (@{$self->_try_refkind($data)}) {
1511 $coderef = $dispatch_table->{$_}
1515 puke "no dispatch entry for ".$self->_refkind($data)
1524 #======================================================================
1525 # VALUES, GENERATE, AUTOLOAD
1526 #======================================================================
1528 # LDNOTE: original code from nwiger, didn't touch code in that section
1529 # I feel the AUTOLOAD stuff should not be the default, it should
1530 # only be activated on explicit demand by user.
1534 my $data = shift || return;
1535 puke "Argument to ", __PACKAGE__, "->values must be a \\%hash"
1536 unless ref $data eq 'HASH';
1539 foreach my $k (sort keys %$data) {
1540 my $v = $data->{$k};
1541 $self->_SWITCH_refkind($v, {
1543 if ($self->{array_datatypes}) { # array datatype
1544 push @all_bind, $self->_bindtype($k, $v);
1546 else { # literal SQL with bind
1547 my ($sql, @bind) = @$v;
1548 $self->_assert_bindval_matches_bindtype(@bind);
1549 push @all_bind, @bind;
1552 ARRAYREFREF => sub { # literal SQL with bind
1553 my ($sql, @bind) = @${$v};
1554 $self->_assert_bindval_matches_bindtype(@bind);
1555 push @all_bind, @bind;
1557 SCALARREF => sub { # literal SQL without bind
1559 SCALAR_or_UNDEF => sub {
1560 push @all_bind, $self->_bindtype($k, $v);
1571 my(@sql, @sqlq, @sqlv);
1575 if ($ref eq 'HASH') {
1576 for my $k (sort keys %$_) {
1579 my $label = $self->_quote($k);
1580 if ($r eq 'ARRAY') {
1581 # literal SQL with bind
1582 my ($sql, @bind) = @$v;
1583 $self->_assert_bindval_matches_bindtype(@bind);
1584 push @sqlq, "$label = $sql";
1586 } elsif ($r eq 'SCALAR') {
1587 # literal SQL without bind
1588 push @sqlq, "$label = $$v";
1590 push @sqlq, "$label = ?";
1591 push @sqlv, $self->_bindtype($k, $v);
1594 push @sql, $self->_sqlcase('set'), join ', ', @sqlq;
1595 } elsif ($ref eq 'ARRAY') {
1596 # unlike insert(), assume these are ONLY the column names, i.e. for SQL
1599 if ($r eq 'ARRAY') { # literal SQL with bind
1600 my ($sql, @bind) = @$v;
1601 $self->_assert_bindval_matches_bindtype(@bind);
1604 } elsif ($r eq 'SCALAR') { # literal SQL without bind
1605 # embedded literal SQL
1612 push @sql, '(' . join(', ', @sqlq) . ')';
1613 } elsif ($ref eq 'SCALAR') {
1617 # strings get case twiddled
1618 push @sql, $self->_sqlcase($_);
1622 my $sql = join ' ', @sql;
1624 # this is pretty tricky
1625 # if ask for an array, return ($stmt, @bind)
1626 # otherwise, s/?/shift @sqlv/ to put it inline
1628 return ($sql, @sqlv);
1630 1 while $sql =~ s/\?/my $d = shift(@sqlv);
1631 ref $d ? $d->[1] : $d/e;
1640 # This allows us to check for a local, then _form, attr
1642 my($name) = $AUTOLOAD =~ /.*::(.+)/;
1643 return $self->generate($name, @_);
1654 SQL::Abstract - Generate SQL from Perl data structures
1660 my $sql = SQL::Abstract->new;
1662 my($stmt, @bind) = $sql->select($source, \@fields, \%where, $order);
1664 my($stmt, @bind) = $sql->insert($table, \%fieldvals || \@values);
1666 my($stmt, @bind) = $sql->update($table, \%fieldvals, \%where);
1668 my($stmt, @bind) = $sql->delete($table, \%where);
1670 # Then, use these in your DBI statements
1671 my $sth = $dbh->prepare($stmt);
1672 $sth->execute(@bind);
1674 # Just generate the WHERE clause
1675 my($stmt, @bind) = $sql->where(\%where, $order);
1677 # Return values in the same order, for hashed queries
1678 # See PERFORMANCE section for more details
1679 my @bind = $sql->values(\%fieldvals);
1683 This module was inspired by the excellent L<DBIx::Abstract>.
1684 However, in using that module I found that what I really wanted
1685 to do was generate SQL, but still retain complete control over my
1686 statement handles and use the DBI interface. So, I set out to
1687 create an abstract SQL generation module.
1689 While based on the concepts used by L<DBIx::Abstract>, there are
1690 several important differences, especially when it comes to WHERE
1691 clauses. I have modified the concepts used to make the SQL easier
1692 to generate from Perl data structures and, IMO, more intuitive.
1693 The underlying idea is for this module to do what you mean, based
1694 on the data structures you provide it. The big advantage is that
1695 you don't have to modify your code every time your data changes,
1696 as this module figures it out.
1698 To begin with, an SQL INSERT is as easy as just specifying a hash
1699 of C<key=value> pairs:
1702 name => 'Jimbo Bobson',
1703 phone => '123-456-7890',
1704 address => '42 Sister Lane',
1705 city => 'St. Louis',
1706 state => 'Louisiana',
1709 The SQL can then be generated with this:
1711 my($stmt, @bind) = $sql->insert('people', \%data);
1713 Which would give you something like this:
1715 $stmt = "INSERT INTO people
1716 (address, city, name, phone, state)
1717 VALUES (?, ?, ?, ?, ?)";
1718 @bind = ('42 Sister Lane', 'St. Louis', 'Jimbo Bobson',
1719 '123-456-7890', 'Louisiana');
1721 These are then used directly in your DBI code:
1723 my $sth = $dbh->prepare($stmt);
1724 $sth->execute(@bind);
1726 =head2 Inserting and Updating Arrays
1728 If your database has array types (like for example Postgres),
1729 activate the special option C<< array_datatypes => 1 >>
1730 when creating the C<SQL::Abstract> object.
1731 Then you may use an arrayref to insert and update database array types:
1733 my $sql = SQL::Abstract->new(array_datatypes => 1);
1735 planets => [qw/Mercury Venus Earth Mars/]
1738 my($stmt, @bind) = $sql->insert('solar_system', \%data);
1742 $stmt = "INSERT INTO solar_system (planets) VALUES (?)"
1744 @bind = (['Mercury', 'Venus', 'Earth', 'Mars']);
1747 =head2 Inserting and Updating SQL
1749 In order to apply SQL functions to elements of your C<%data> you may
1750 specify a reference to an arrayref for the given hash value. For example,
1751 if you need to execute the Oracle C<to_date> function on a value, you can
1752 say something like this:
1756 date_entered => \[ "to_date(?,'MM/DD/YYYY')", "03/02/2003" ],
1759 The first value in the array is the actual SQL. Any other values are
1760 optional and would be included in the bind values array. This gives
1763 my($stmt, @bind) = $sql->insert('people', \%data);
1765 $stmt = "INSERT INTO people (name, date_entered)
1766 VALUES (?, to_date(?,'MM/DD/YYYY'))";
1767 @bind = ('Bill', '03/02/2003');
1769 An UPDATE is just as easy, all you change is the name of the function:
1771 my($stmt, @bind) = $sql->update('people', \%data);
1773 Notice that your C<%data> isn't touched; the module will generate
1774 the appropriately quirky SQL for you automatically. Usually you'll
1775 want to specify a WHERE clause for your UPDATE, though, which is
1776 where handling C<%where> hashes comes in handy...
1778 =head2 Complex where statements
1780 This module can generate pretty complicated WHERE statements
1781 easily. For example, simple C<key=value> pairs are taken to mean
1782 equality, and if you want to see if a field is within a set
1783 of values, you can use an arrayref. Let's say we wanted to
1784 SELECT some data based on this criteria:
1787 requestor => 'inna',
1788 worker => ['nwiger', 'rcwe', 'sfz'],
1789 status => { '!=', 'completed' }
1792 my($stmt, @bind) = $sql->select('tickets', '*', \%where);
1794 The above would give you something like this:
1796 $stmt = "SELECT * FROM tickets WHERE
1797 ( requestor = ? ) AND ( status != ? )
1798 AND ( worker = ? OR worker = ? OR worker = ? )";
1799 @bind = ('inna', 'completed', 'nwiger', 'rcwe', 'sfz');
1801 Which you could then use in DBI code like so:
1803 my $sth = $dbh->prepare($stmt);
1804 $sth->execute(@bind);
1810 The methods are simple. There's one for every major SQL operation,
1811 and a constructor you use first. The arguments are specified in a
1812 similar order for each method (table, then fields, then a where
1813 clause) to try and simplify things.
1815 =head2 new(option => 'value')
1817 The C<new()> function takes a list of options and values, and returns
1818 a new B<SQL::Abstract> object which can then be used to generate SQL
1819 through the methods below. The options accepted are:
1825 If set to 'lower', then SQL will be generated in all lowercase. By
1826 default SQL is generated in "textbook" case meaning something like:
1828 SELECT a_field FROM a_table WHERE some_field LIKE '%someval%'
1830 Any setting other than 'lower' is ignored.
1834 This determines what the default comparison operator is. By default
1835 it is C<=>, meaning that a hash like this:
1837 %where = (name => 'nwiger', email => 'nate@wiger.org');
1839 Will generate SQL like this:
1841 WHERE name = 'nwiger' AND email = 'nate@wiger.org'
1843 However, you may want loose comparisons by default, so if you set
1844 C<cmp> to C<like> you would get SQL such as:
1846 WHERE name like 'nwiger' AND email like 'nate@wiger.org'
1848 You can also override the comparison on an individual basis - see
1849 the huge section on L</"WHERE CLAUSES"> at the bottom.
1851 =item sqltrue, sqlfalse
1853 Expressions for inserting boolean values within SQL statements.
1854 By default these are C<1=1> and C<1=0>. They are used
1855 by the special operators C<-in> and C<-not_in> for generating
1856 correct SQL even when the argument is an empty array (see below).
1860 This determines the default logical operator for multiple WHERE
1861 statements in arrays or hashes. If absent, the default logic is "or"
1862 for arrays, and "and" for hashes. This means that a WHERE
1866 event_date => {'>=', '2/13/99'},
1867 event_date => {'<=', '4/24/03'},
1870 will generate SQL like this:
1872 WHERE event_date >= '2/13/99' OR event_date <= '4/24/03'
1874 This is probably not what you want given this query, though (look
1875 at the dates). To change the "OR" to an "AND", simply specify:
1877 my $sql = SQL::Abstract->new(logic => 'and');
1879 Which will change the above C<WHERE> to:
1881 WHERE event_date >= '2/13/99' AND event_date <= '4/24/03'
1883 The logic can also be changed locally by inserting
1884 a modifier in front of an arrayref:
1886 @where = (-and => [event_date => {'>=', '2/13/99'},
1887 event_date => {'<=', '4/24/03'} ]);
1889 See the L</"WHERE CLAUSES"> section for explanations.
1893 This will automatically convert comparisons using the specified SQL
1894 function for both column and value. This is mostly used with an argument
1895 of C<upper> or C<lower>, so that the SQL will have the effect of
1896 case-insensitive "searches". For example, this:
1898 $sql = SQL::Abstract->new(convert => 'upper');
1899 %where = (keywords => 'MaKe iT CAse inSeNSItive');
1901 Will turn out the following SQL:
1903 WHERE upper(keywords) like upper('MaKe iT CAse inSeNSItive')
1905 The conversion can be C<upper()>, C<lower()>, or any other SQL function
1906 that can be applied symmetrically to fields (actually B<SQL::Abstract> does
1907 not validate this option; it will just pass through what you specify verbatim).
1911 This is a kludge because many databases suck. For example, you can't
1912 just bind values using DBI's C<execute()> for Oracle C<CLOB> or C<BLOB> fields.
1913 Instead, you have to use C<bind_param()>:
1915 $sth->bind_param(1, 'reg data');
1916 $sth->bind_param(2, $lots, {ora_type => ORA_CLOB});
1918 The problem is, B<SQL::Abstract> will normally just return a C<@bind> array,
1919 which loses track of which field each slot refers to. Fear not.
1921 If you specify C<bindtype> in new, you can determine how C<@bind> is returned.
1922 Currently, you can specify either C<normal> (default) or C<columns>. If you
1923 specify C<columns>, you will get an array that looks like this:
1925 my $sql = SQL::Abstract->new(bindtype => 'columns');
1926 my($stmt, @bind) = $sql->insert(...);
1929 [ 'column1', 'value1' ],
1930 [ 'column2', 'value2' ],
1931 [ 'column3', 'value3' ],
1934 You can then iterate through this manually, using DBI's C<bind_param()>.
1936 $sth->prepare($stmt);
1939 my($col, $data) = @$_;
1940 if ($col eq 'details' || $col eq 'comments') {
1941 $sth->bind_param($i, $data, {ora_type => ORA_CLOB});
1942 } elsif ($col eq 'image') {
1943 $sth->bind_param($i, $data, {ora_type => ORA_BLOB});
1945 $sth->bind_param($i, $data);
1949 $sth->execute; # execute without @bind now
1951 Now, why would you still use B<SQL::Abstract> if you have to do this crap?
1952 Basically, the advantage is still that you don't have to care which fields
1953 are or are not included. You could wrap that above C<for> loop in a simple
1954 sub called C<bind_fields()> or something and reuse it repeatedly. You still
1955 get a layer of abstraction over manual SQL specification.
1957 Note that if you set L</bindtype> to C<columns>, the C<\[ $sql, @bind ]>
1958 construct (see L</Literal SQL with placeholders and bind values (subqueries)>)
1959 will expect the bind values in this format.
1963 This is the character that a table or column name will be quoted
1964 with. By default this is an empty string, but you could set it to
1965 the character C<`>, to generate SQL like this:
1967 SELECT `a_field` FROM `a_table` WHERE `some_field` LIKE '%someval%'
1969 Alternatively, you can supply an array ref of two items, the first being the left
1970 hand quote character, and the second the right hand quote character. For
1971 example, you could supply C<['[',']']> for SQL Server 2000 compliant quotes
1972 that generates SQL like this:
1974 SELECT [a_field] FROM [a_table] WHERE [some_field] LIKE '%someval%'
1976 Quoting is useful if you have tables or columns names that are reserved
1977 words in your database's SQL dialect.
1981 This is the character that will be used to escape L</quote_char>s appearing
1982 in an identifier before it has been quoted.
1984 The parameter default in case of a single L</quote_char> character is the quote
1987 When opening-closing-style quoting is used (L</quote_char> is an arrayref)
1988 this parameter defaults to the B<closing (right)> L</quote_char>. Occurrences
1989 of the B<opening (left)> L</quote_char> within the identifier are currently left
1990 untouched. The default for opening-closing-style quotes may change in future
1991 versions, thus you are B<strongly encouraged> to specify the escape character
1996 This is the character that separates a table and column name. It is
1997 necessary to specify this when the C<quote_char> option is selected,
1998 so that tables and column names can be individually quoted like this:
2000 SELECT `table`.`one_field` FROM `table` WHERE `table`.`other_field` = 1
2002 =item injection_guard
2004 A regular expression C<qr/.../> that is applied to any C<-function> and unquoted
2005 column name specified in a query structure. This is a safety mechanism to avoid
2006 injection attacks when mishandling user input e.g.:
2008 my %condition_as_column_value_pairs = get_values_from_user();
2009 $sqla->select( ... , \%condition_as_column_value_pairs );
2011 If the expression matches an exception is thrown. Note that literal SQL
2012 supplied via C<\'...'> or C<\['...']> is B<not> checked in any way.
2014 Defaults to checking for C<;> and the C<GO> keyword (TransactSQL)
2016 =item array_datatypes
2018 When this option is true, arrayrefs in INSERT or UPDATE are
2019 interpreted as array datatypes and are passed directly
2021 When this option is false, arrayrefs are interpreted
2022 as literal SQL, just like refs to arrayrefs
2023 (but this behavior is for backwards compatibility; when writing
2024 new queries, use the "reference to arrayref" syntax
2030 Takes a reference to a list of "special operators"
2031 to extend the syntax understood by L<SQL::Abstract>.
2032 See section L</"SPECIAL OPERATORS"> for details.
2036 Takes a reference to a list of "unary operators"
2037 to extend the syntax understood by L<SQL::Abstract>.
2038 See section L</"UNARY OPERATORS"> for details.
2044 =head2 insert($table, \@values || \%fieldvals, \%options)
2046 This is the simplest function. You simply give it a table name
2047 and either an arrayref of values or hashref of field/value pairs.
2048 It returns an SQL INSERT statement and a list of bind values.
2049 See the sections on L</"Inserting and Updating Arrays"> and
2050 L</"Inserting and Updating SQL"> for information on how to insert
2051 with those data types.
2053 The optional C<\%options> hash reference may contain additional
2054 options to generate the insert SQL. Currently supported options
2061 Takes either a scalar of raw SQL fields, or an array reference of
2062 field names, and adds on an SQL C<RETURNING> statement at the end.
2063 This allows you to return data generated by the insert statement
2064 (such as row IDs) without performing another C<SELECT> statement.
2065 Note, however, this is not part of the SQL standard and may not
2066 be supported by all database engines.
2070 =head2 update($table, \%fieldvals, \%where, \%options)
2072 This takes a table, hashref of field/value pairs, and an optional
2073 hashref L<WHERE clause|/WHERE CLAUSES>. It returns an SQL UPDATE function and a list
2075 See the sections on L</"Inserting and Updating Arrays"> and
2076 L</"Inserting and Updating SQL"> for information on how to insert
2077 with those data types.
2079 The optional C<\%options> hash reference may contain additional
2080 options to generate the update SQL. Currently supported options
2087 See the C<returning> option to
2088 L<insert|/insert($table, \@values || \%fieldvals, \%options)>.
2092 =head2 select($source, $fields, $where, $order)
2094 This returns a SQL SELECT statement and associated list of bind values, as
2095 specified by the arguments:
2101 Specification of the 'FROM' part of the statement.
2102 The argument can be either a plain scalar (interpreted as a table
2103 name, will be quoted), or an arrayref (interpreted as a list
2104 of table names, joined by commas, quoted), or a scalarref
2105 (literal SQL, not quoted).
2109 Specification of the list of fields to retrieve from
2111 The argument can be either an arrayref (interpreted as a list
2112 of field names, will be joined by commas and quoted), or a
2113 plain scalar (literal SQL, not quoted).
2114 Please observe that this API is not as flexible as that of
2115 the first argument C<$source>, for backwards compatibility reasons.
2119 Optional argument to specify the WHERE part of the query.
2120 The argument is most often a hashref, but can also be
2121 an arrayref or plain scalar --
2122 see section L<WHERE clause|/"WHERE CLAUSES"> for details.
2126 Optional argument to specify the ORDER BY part of the query.
2127 The argument can be a scalar, a hashref or an arrayref
2128 -- see section L<ORDER BY clause|/"ORDER BY CLAUSES">
2134 =head2 delete($table, \%where, \%options)
2136 This takes a table name and optional hashref L<WHERE clause|/WHERE CLAUSES>.
2137 It returns an SQL DELETE statement and list of bind values.
2139 The optional C<\%options> hash reference may contain additional
2140 options to generate the delete SQL. Currently supported options
2147 See the C<returning> option to
2148 L<insert|/insert($table, \@values || \%fieldvals, \%options)>.
2152 =head2 where(\%where, $order)
2154 This is used to generate just the WHERE clause. For example,
2155 if you have an arbitrary data structure and know what the
2156 rest of your SQL is going to look like, but want an easy way
2157 to produce a WHERE clause, use this. It returns an SQL WHERE
2158 clause and list of bind values.
2161 =head2 values(\%data)
2163 This just returns the values from the hash C<%data>, in the same
2164 order that would be returned from any of the other above queries.
2165 Using this allows you to markedly speed up your queries if you
2166 are affecting lots of rows. See below under the L</"PERFORMANCE"> section.
2168 =head2 generate($any, 'number', $of, \@data, $struct, \%types)
2170 Warning: This is an experimental method and subject to change.
2172 This returns arbitrarily generated SQL. It's a really basic shortcut.
2173 It will return two different things, depending on return context:
2175 my($stmt, @bind) = $sql->generate('create table', \$table, \@fields);
2176 my $stmt_and_val = $sql->generate('create table', \$table, \@fields);
2178 These would return the following:
2180 # First calling form
2181 $stmt = "CREATE TABLE test (?, ?)";
2182 @bind = (field1, field2);
2184 # Second calling form
2185 $stmt_and_val = "CREATE TABLE test (field1, field2)";
2187 Depending on what you're trying to do, it's up to you to choose the correct
2188 format. In this example, the second form is what you would want.
2192 $sql->generate('alter session', { nls_date_format => 'MM/YY' });
2196 ALTER SESSION SET nls_date_format = 'MM/YY'
2198 You get the idea. Strings get their case twiddled, but everything
2199 else remains verbatim.
2201 =head1 EXPORTABLE FUNCTIONS
2203 =head2 is_plain_value
2205 Determines if the supplied argument is a plain value as understood by this
2210 =item * The value is C<undef>
2212 =item * The value is a non-reference
2214 =item * The value is an object with stringification overloading
2216 =item * The value is of the form C<< { -value => $anything } >>
2220 On failure returns C<undef>, on success returns a B<scalar> reference
2221 to the original supplied argument.
2227 The stringification overloading detection is rather advanced: it takes
2228 into consideration not only the presence of a C<""> overload, but if that
2229 fails also checks for enabled
2230 L<autogenerated versions of C<"">|overload/Magic Autogeneration>, based
2231 on either C<0+> or C<bool>.
2233 Unfortunately testing in the field indicates that this
2234 detection B<< may tickle a latent bug in perl versions before 5.018 >>,
2235 but only when very large numbers of stringifying objects are involved.
2236 At the time of writing ( Sep 2014 ) there is no clear explanation of
2237 the direct cause, nor is there a manageably small test case that reliably
2238 reproduces the problem.
2240 If you encounter any of the following exceptions in B<random places within
2241 your application stack> - this module may be to blame:
2243 Operation "ne": no method found,
2244 left argument in overloaded package <something>,
2245 right argument in overloaded package <something>
2249 Stub found while resolving method "???" overloading """" in package <something>
2251 If you fall victim to the above - please attempt to reduce the problem
2252 to something that could be sent to the L<SQL::Abstract developers
2253 |DBIx::Class/GETTING HELP/SUPPORT>
2254 (either publicly or privately). As a workaround in the meantime you can
2255 set C<$ENV{SQLA_ISVALUE_IGNORE_AUTOGENERATED_STRINGIFICATION}> to a true
2256 value, which will most likely eliminate your problem (at the expense of
2257 not being able to properly detect exotic forms of stringification).
2259 This notice and environment variable will be removed in a future version,
2260 as soon as the underlying problem is found and a reliable workaround is
2265 =head2 is_literal_value
2267 Determines if the supplied argument is a literal value as understood by this
2272 =item * C<\$sql_string>
2274 =item * C<\[ $sql_string, @bind_values ]>
2278 On failure returns C<undef>, on success returns an B<array> reference
2279 containing the unpacked version of the supplied literal SQL and bind values.
2281 =head1 WHERE CLAUSES
2285 This module uses a variation on the idea from L<DBIx::Abstract>. It
2286 is B<NOT>, repeat I<not> 100% compatible. B<The main logic of this
2287 module is that things in arrays are OR'ed, and things in hashes
2290 The easiest way to explain is to show lots of examples. After
2291 each C<%where> hash shown, it is assumed you used:
2293 my($stmt, @bind) = $sql->where(\%where);
2295 However, note that the C<%where> hash can be used directly in any
2296 of the other functions as well, as described above.
2298 =head2 Key-value pairs
2300 So, let's get started. To begin, a simple hash:
2304 status => 'completed'
2307 Is converted to SQL C<key = val> statements:
2309 $stmt = "WHERE user = ? AND status = ?";
2310 @bind = ('nwiger', 'completed');
2312 One common thing I end up doing is having a list of values that
2313 a field can be in. To do this, simply specify a list inside of
2318 status => ['assigned', 'in-progress', 'pending'];
2321 This simple code will create the following:
2323 $stmt = "WHERE user = ? AND ( status = ? OR status = ? OR status = ? )";
2324 @bind = ('nwiger', 'assigned', 'in-progress', 'pending');
2326 A field associated to an empty arrayref will be considered a
2327 logical false and will generate 0=1.
2329 =head2 Tests for NULL values
2331 If the value part is C<undef> then this is converted to SQL <IS NULL>
2340 $stmt = "WHERE user = ? AND status IS NULL";
2343 To test if a column IS NOT NULL:
2347 status => { '!=', undef },
2350 =head2 Specific comparison operators
2352 If you want to specify a different type of operator for your comparison,
2353 you can use a hashref for a given column:
2357 status => { '!=', 'completed' }
2360 Which would generate:
2362 $stmt = "WHERE user = ? AND status != ?";
2363 @bind = ('nwiger', 'completed');
2365 To test against multiple values, just enclose the values in an arrayref:
2367 status => { '=', ['assigned', 'in-progress', 'pending'] };
2369 Which would give you:
2371 "WHERE status = ? OR status = ? OR status = ?"
2374 The hashref can also contain multiple pairs, in which case it is expanded
2375 into an C<AND> of its elements:
2379 status => { '!=', 'completed', -not_like => 'pending%' }
2382 # Or more dynamically, like from a form
2383 $where{user} = 'nwiger';
2384 $where{status}{'!='} = 'completed';
2385 $where{status}{'-not_like'} = 'pending%';
2387 # Both generate this
2388 $stmt = "WHERE user = ? AND status != ? AND status NOT LIKE ?";
2389 @bind = ('nwiger', 'completed', 'pending%');
2392 To get an OR instead, you can combine it with the arrayref idea:
2396 priority => [ { '=', 2 }, { '>', 5 } ]
2399 Which would generate:
2401 $stmt = "WHERE ( priority = ? OR priority > ? ) AND user = ?";
2402 @bind = ('2', '5', 'nwiger');
2404 If you want to include literal SQL (with or without bind values), just use a
2405 scalar reference or reference to an arrayref as the value:
2408 date_entered => { '>' => \["to_date(?, 'MM/DD/YYYY')", "11/26/2008"] },
2409 date_expires => { '<' => \"now()" }
2412 Which would generate:
2414 $stmt = "WHERE date_entered > to_date(?, 'MM/DD/YYYY') AND date_expires < now()";
2415 @bind = ('11/26/2008');
2418 =head2 Logic and nesting operators
2420 In the example above,
2421 there is a subtle trap if you want to say something like
2422 this (notice the C<AND>):
2424 WHERE priority != ? AND priority != ?
2426 Because, in Perl you I<can't> do this:
2428 priority => { '!=' => 2, '!=' => 1 }
2430 As the second C<!=> key will obliterate the first. The solution
2431 is to use the special C<-modifier> form inside an arrayref:
2433 priority => [ -and => {'!=', 2},
2437 Normally, these would be joined by C<OR>, but the modifier tells it
2438 to use C<AND> instead. (Hint: You can use this in conjunction with the
2439 C<logic> option to C<new()> in order to change the way your queries
2440 work by default.) B<Important:> Note that the C<-modifier> goes
2441 B<INSIDE> the arrayref, as an extra first element. This will
2442 B<NOT> do what you think it might:
2444 priority => -and => [{'!=', 2}, {'!=', 1}] # WRONG!
2446 Here is a quick list of equivalencies, since there is some overlap:
2449 status => {'!=', 'completed', 'not like', 'pending%' }
2450 status => [ -and => {'!=', 'completed'}, {'not like', 'pending%'}]
2453 status => {'=', ['assigned', 'in-progress']}
2454 status => [ -or => {'=', 'assigned'}, {'=', 'in-progress'}]
2455 status => [ {'=', 'assigned'}, {'=', 'in-progress'} ]
2459 =head2 Special operators: IN, BETWEEN, etc.
2461 You can also use the hashref format to compare a list of fields using the
2462 C<IN> comparison operator, by specifying the list as an arrayref:
2465 status => 'completed',
2466 reportid => { -in => [567, 2335, 2] }
2469 Which would generate:
2471 $stmt = "WHERE status = ? AND reportid IN (?,?,?)";
2472 @bind = ('completed', '567', '2335', '2');
2474 The reverse operator C<-not_in> generates SQL C<NOT IN> and is used in
2477 If the argument to C<-in> is an empty array, 'sqlfalse' is generated
2478 (by default: C<1=0>). Similarly, C<< -not_in => [] >> generates
2479 'sqltrue' (by default: C<1=1>).
2481 In addition to the array you can supply a chunk of literal sql or
2482 literal sql with bind:
2485 customer => { -in => \[
2486 'SELECT cust_id FROM cust WHERE balance > ?',
2489 status => { -in => \'SELECT status_codes FROM states' },
2495 customer IN ( SELECT cust_id FROM cust WHERE balance > ? )
2496 AND status IN ( SELECT status_codes FROM states )
2500 Finally, if the argument to C<-in> is not a reference, it will be
2501 treated as a single-element array.
2503 Another pair of operators is C<-between> and C<-not_between>,
2504 used with an arrayref of two values:
2508 completion_date => {
2509 -not_between => ['2002-10-01', '2003-02-06']
2515 WHERE user = ? AND completion_date NOT BETWEEN ( ? AND ? )
2517 Just like with C<-in> all plausible combinations of literal SQL
2521 start0 => { -between => [ 1, 2 ] },
2522 start1 => { -between => \["? AND ?", 1, 2] },
2523 start2 => { -between => \"lower(x) AND upper(y)" },
2524 start3 => { -between => [
2526 \["upper(?)", 'stuff' ],
2533 ( start0 BETWEEN ? AND ? )
2534 AND ( start1 BETWEEN ? AND ? )
2535 AND ( start2 BETWEEN lower(x) AND upper(y) )
2536 AND ( start3 BETWEEN lower(x) AND upper(?) )
2538 @bind = (1, 2, 1, 2, 'stuff');
2541 These are the two builtin "special operators"; but the
2542 list can be expanded: see section L</"SPECIAL OPERATORS"> below.
2544 =head2 Unary operators: bool
2546 If you wish to test against boolean columns or functions within your
2547 database you can use the C<-bool> and C<-not_bool> operators. For
2548 example to test the column C<is_user> being true and the column
2549 C<is_enabled> being false you would use:-
2553 -not_bool => 'is_enabled',
2558 WHERE is_user AND NOT is_enabled
2560 If a more complex combination is required, testing more conditions,
2561 then you should use the and/or operators:-
2566 -not_bool => { two=> { -rlike => 'bar' } },
2567 -not_bool => { three => [ { '=', 2 }, { '>', 5 } ] },
2578 (NOT ( three = ? OR three > ? ))
2581 =head2 Nested conditions, -and/-or prefixes
2583 So far, we've seen how multiple conditions are joined with a top-level
2584 C<AND>. We can change this by putting the different conditions we want in
2585 hashes and then putting those hashes in an array. For example:
2590 status => { -like => ['pending%', 'dispatched'] },
2594 status => 'unassigned',
2598 This data structure would create the following:
2600 $stmt = "WHERE ( user = ? AND ( status LIKE ? OR status LIKE ? ) )
2601 OR ( user = ? AND status = ? ) )";
2602 @bind = ('nwiger', 'pending', 'dispatched', 'robot', 'unassigned');
2605 Clauses in hashrefs or arrayrefs can be prefixed with an C<-and> or C<-or>
2606 to change the logic inside:
2612 -and => [ workhrs => {'>', 20}, geo => 'ASIA' ],
2613 -or => { workhrs => {'<', 50}, geo => 'EURO' },
2620 $stmt = "WHERE ( user = ?
2621 AND ( ( workhrs > ? AND geo = ? )
2622 OR ( workhrs < ? OR geo = ? ) ) )";
2623 @bind = ('nwiger', '20', 'ASIA', '50', 'EURO');
2625 =head3 Algebraic inconsistency, for historical reasons
2627 C<Important note>: when connecting several conditions, the C<-and->|C<-or>
2628 operator goes C<outside> of the nested structure; whereas when connecting
2629 several constraints on one column, the C<-and> operator goes
2630 C<inside> the arrayref. Here is an example combining both features:
2633 -and => [a => 1, b => 2],
2634 -or => [c => 3, d => 4],
2635 e => [-and => {-like => 'foo%'}, {-like => '%bar'} ]
2640 WHERE ( ( ( a = ? AND b = ? )
2641 OR ( c = ? OR d = ? )
2642 OR ( e LIKE ? AND e LIKE ? ) ) )
2644 This difference in syntax is unfortunate but must be preserved for
2645 historical reasons. So be careful: the two examples below would
2646 seem algebraically equivalent, but they are not
2649 { -like => 'foo%' },
2650 { -like => '%bar' },
2652 # yields: WHERE ( ( col LIKE ? AND col LIKE ? ) )
2655 { col => { -like => 'foo%' } },
2656 { col => { -like => '%bar' } },
2658 # yields: WHERE ( ( col LIKE ? OR col LIKE ? ) )
2661 =head2 Literal SQL and value type operators
2663 The basic premise of SQL::Abstract is that in WHERE specifications the "left
2664 side" is a column name and the "right side" is a value (normally rendered as
2665 a placeholder). This holds true for both hashrefs and arrayref pairs as you
2666 see in the L</WHERE CLAUSES> examples above. Sometimes it is necessary to
2667 alter this behavior. There are several ways of doing so.
2671 This is a virtual operator that signals the string to its right side is an
2672 identifier (a column name) and not a value. For example to compare two
2673 columns you would write:
2676 priority => { '<', 2 },
2677 requestor => { -ident => 'submitter' },
2682 $stmt = "WHERE priority < ? AND requestor = submitter";
2685 If you are maintaining legacy code you may see a different construct as
2686 described in L</Deprecated usage of Literal SQL>, please use C<-ident> in new
2691 This is a virtual operator that signals that the construct to its right side
2692 is a value to be passed to DBI. This is for example necessary when you want
2693 to write a where clause against an array (for RDBMS that support such
2694 datatypes). For example:
2697 array => { -value => [1, 2, 3] }
2702 $stmt = 'WHERE array = ?';
2703 @bind = ([1, 2, 3]);
2705 Note that if you were to simply say:
2711 the result would probably not be what you wanted:
2713 $stmt = 'WHERE array = ? OR array = ? OR array = ?';
2718 Finally, sometimes only literal SQL will do. To include a random snippet
2719 of SQL verbatim, you specify it as a scalar reference. Consider this only
2720 as a last resort. Usually there is a better way. For example:
2723 priority => { '<', 2 },
2724 requestor => { -in => \'(SELECT name FROM hitmen)' },
2729 $stmt = "WHERE priority < ? AND requestor IN (SELECT name FROM hitmen)"
2732 Note that in this example, you only get one bind parameter back, since
2733 the verbatim SQL is passed as part of the statement.
2737 Never use untrusted input as a literal SQL argument - this is a massive
2738 security risk (there is no way to check literal snippets for SQL
2739 injections and other nastyness). If you need to deal with untrusted input
2740 use literal SQL with placeholders as described next.
2742 =head3 Literal SQL with placeholders and bind values (subqueries)
2744 If the literal SQL to be inserted has placeholders and bind values,
2745 use a reference to an arrayref (yes this is a double reference --
2746 not so common, but perfectly legal Perl). For example, to find a date
2747 in Postgres you can use something like this:
2750 date_column => \[ "= date '2008-09-30' - ?::integer", 10 ]
2755 $stmt = "WHERE ( date_column = date '2008-09-30' - ?::integer )"
2758 Note that you must pass the bind values in the same format as they are returned
2759 by L<where|/where(\%where, $order)>. This means that if you set L</bindtype>
2760 to C<columns>, you must provide the bind values in the
2761 C<< [ column_meta => value ] >> format, where C<column_meta> is an opaque
2762 scalar value; most commonly the column name, but you can use any scalar value
2763 (including references and blessed references), L<SQL::Abstract> will simply
2764 pass it through intact. So if C<bindtype> is set to C<columns> the above
2765 example will look like:
2768 date_column => \[ "= date '2008-09-30' - ?::integer", [ {} => 10 ] ]
2771 Literal SQL is especially useful for nesting parenthesized clauses in the
2772 main SQL query. Here is a first example:
2774 my ($sub_stmt, @sub_bind) = ("SELECT c1 FROM t1 WHERE c2 < ? AND c3 LIKE ?",
2778 bar => \["IN ($sub_stmt)" => @sub_bind],
2783 $stmt = "WHERE (foo = ? AND bar IN (SELECT c1 FROM t1
2784 WHERE c2 < ? AND c3 LIKE ?))";
2785 @bind = (1234, 100, "foo%");
2787 Other subquery operators, like for example C<"E<gt> ALL"> or C<"NOT IN">,
2788 are expressed in the same way. Of course the C<$sub_stmt> and
2789 its associated bind values can be generated through a former call
2792 my ($sub_stmt, @sub_bind)
2793 = $sql->select("t1", "c1", {c2 => {"<" => 100},
2794 c3 => {-like => "foo%"}});
2797 bar => \["> ALL ($sub_stmt)" => @sub_bind],
2800 In the examples above, the subquery was used as an operator on a column;
2801 but the same principle also applies for a clause within the main C<%where>
2802 hash, like an EXISTS subquery:
2804 my ($sub_stmt, @sub_bind)
2805 = $sql->select("t1", "*", {c1 => 1, c2 => \"> t0.c0"});
2806 my %where = ( -and => [
2808 \["EXISTS ($sub_stmt)" => @sub_bind],
2813 $stmt = "WHERE (foo = ? AND EXISTS (SELECT * FROM t1
2814 WHERE c1 = ? AND c2 > t0.c0))";
2818 Observe that the condition on C<c2> in the subquery refers to
2819 column C<t0.c0> of the main query: this is I<not> a bind
2820 value, so we have to express it through a scalar ref.
2821 Writing C<< c2 => {">" => "t0.c0"} >> would have generated
2822 C<< c2 > ? >> with bind value C<"t0.c0"> ... not exactly
2823 what we wanted here.
2825 Finally, here is an example where a subquery is used
2826 for expressing unary negation:
2828 my ($sub_stmt, @sub_bind)
2829 = $sql->where({age => [{"<" => 10}, {">" => 20}]});
2830 $sub_stmt =~ s/^ where //i; # don't want "WHERE" in the subclause
2832 lname => {like => '%son%'},
2833 \["NOT ($sub_stmt)" => @sub_bind],
2838 $stmt = "lname LIKE ? AND NOT ( age < ? OR age > ? )"
2839 @bind = ('%son%', 10, 20)
2841 =head3 Deprecated usage of Literal SQL
2843 Below are some examples of archaic use of literal SQL. It is shown only as
2844 reference for those who deal with legacy code. Each example has a much
2845 better, cleaner and safer alternative that users should opt for in new code.
2851 my %where = ( requestor => \'IS NOT NULL' )
2853 $stmt = "WHERE requestor IS NOT NULL"
2855 This used to be the way of generating NULL comparisons, before the handling
2856 of C<undef> got formalized. For new code please use the superior syntax as
2857 described in L</Tests for NULL values>.
2861 my %where = ( requestor => \'= submitter' )
2863 $stmt = "WHERE requestor = submitter"
2865 This used to be the only way to compare columns. Use the superior L</-ident>
2866 method for all new code. For example an identifier declared in such a way
2867 will be properly quoted if L</quote_char> is properly set, while the legacy
2868 form will remain as supplied.
2872 my %where = ( is_ready => \"", completed => { '>', '2012-12-21' } )
2874 $stmt = "WHERE completed > ? AND is_ready"
2875 @bind = ('2012-12-21')
2877 Using an empty string literal used to be the only way to express a boolean.
2878 For all new code please use the much more readable
2879 L<-bool|/Unary operators: bool> operator.
2885 These pages could go on for a while, since the nesting of the data
2886 structures this module can handle are pretty much unlimited (the
2887 module implements the C<WHERE> expansion as a recursive function
2888 internally). Your best bet is to "play around" with the module a
2889 little to see how the data structures behave, and choose the best
2890 format for your data based on that.
2892 And of course, all the values above will probably be replaced with
2893 variables gotten from forms or the command line. After all, if you
2894 knew everything ahead of time, you wouldn't have to worry about
2895 dynamically-generating SQL and could just hardwire it into your
2898 =head1 ORDER BY CLAUSES
2900 Some functions take an order by clause. This can either be a scalar (just a
2901 column name), a hashref of C<< { -desc => 'col' } >> or C<< { -asc => 'col' }
2902 >>, a scalarref, an arrayref-ref, or an arrayref of any of the previous
2905 Given | Will Generate
2906 ---------------------------------------------------------------
2908 'colA' | ORDER BY colA
2910 [qw/colA colB/] | ORDER BY colA, colB
2912 {-asc => 'colA'} | ORDER BY colA ASC
2914 {-desc => 'colB'} | ORDER BY colB DESC
2916 ['colA', {-asc => 'colB'}] | ORDER BY colA, colB ASC
2918 { -asc => [qw/colA colB/] } | ORDER BY colA ASC, colB ASC
2920 \'colA DESC' | ORDER BY colA DESC
2922 \[ 'FUNC(colA, ?)', $x ] | ORDER BY FUNC(colA, ?)
2923 | /* ...with $x bound to ? */
2926 { -asc => 'colA' }, | colA ASC,
2927 { -desc => [qw/colB/] }, | colB DESC,
2928 { -asc => [qw/colC colD/] },| colC ASC, colD ASC,
2929 \'colE DESC', | colE DESC,
2930 \[ 'FUNC(colF, ?)', $x ], | FUNC(colF, ?)
2931 ] | /* ...with $x bound to ? */
2932 ===============================================================
2936 =head1 SPECIAL OPERATORS
2938 my $sqlmaker = SQL::Abstract->new(special_ops => [
2942 my ($self, $field, $op, $arg) = @_;
2948 handler => 'method_name',
2952 A "special operator" is a SQL syntactic clause that can be
2953 applied to a field, instead of a usual binary operator.
2956 WHERE field IN (?, ?, ?)
2957 WHERE field BETWEEN ? AND ?
2958 WHERE MATCH(field) AGAINST (?, ?)
2960 Special operators IN and BETWEEN are fairly standard and therefore
2961 are builtin within C<SQL::Abstract> (as the overridable methods
2962 C<_where_field_IN> and C<_where_field_BETWEEN>). For other operators,
2963 like the MATCH .. AGAINST example above which is specific to MySQL,
2964 you can write your own operator handlers - supply a C<special_ops>
2965 argument to the C<new> method. That argument takes an arrayref of
2966 operator definitions; each operator definition is a hashref with two
2973 the regular expression to match the operator
2977 Either a coderef or a plain scalar method name. In both cases
2978 the expected return is C<< ($sql, @bind) >>.
2980 When supplied with a method name, it is simply called on the
2981 L<SQL::Abstract> object as:
2983 $self->$method_name($field, $op, $arg)
2987 $field is the LHS of the operator
2988 $op is the part that matched the handler regex
2991 When supplied with a coderef, it is called as:
2993 $coderef->($self, $field, $op, $arg)
2998 For example, here is an implementation
2999 of the MATCH .. AGAINST syntax for MySQL
3001 my $sqlmaker = SQL::Abstract->new(special_ops => [
3003 # special op for MySql MATCH (field) AGAINST(word1, word2, ...)
3004 {regex => qr/^match$/i,
3006 my ($self, $field, $op, $arg) = @_;
3007 $arg = [$arg] if not ref $arg;
3008 my $label = $self->_quote($field);
3009 my ($placeholder) = $self->_convert('?');
3010 my $placeholders = join ", ", (($placeholder) x @$arg);
3011 my $sql = $self->_sqlcase('match') . " ($label) "
3012 . $self->_sqlcase('against') . " ($placeholders) ";
3013 my @bind = $self->_bindtype($field, @$arg);
3014 return ($sql, @bind);
3021 =head1 UNARY OPERATORS
3023 my $sqlmaker = SQL::Abstract->new(unary_ops => [
3027 my ($self, $op, $arg) = @_;
3033 handler => 'method_name',
3037 A "unary operator" is a SQL syntactic clause that can be
3038 applied to a field - the operator goes before the field
3040 You can write your own operator handlers - supply a C<unary_ops>
3041 argument to the C<new> method. That argument takes an arrayref of
3042 operator definitions; each operator definition is a hashref with two
3049 the regular expression to match the operator
3053 Either a coderef or a plain scalar method name. In both cases
3054 the expected return is C<< $sql >>.
3056 When supplied with a method name, it is simply called on the
3057 L<SQL::Abstract> object as:
3059 $self->$method_name($op, $arg)
3063 $op is the part that matched the handler regex
3064 $arg is the RHS or argument of the operator
3066 When supplied with a coderef, it is called as:
3068 $coderef->($self, $op, $arg)
3076 Thanks to some benchmarking by Mark Stosberg, it turns out that
3077 this module is many orders of magnitude faster than using C<DBIx::Abstract>.
3078 I must admit this wasn't an intentional design issue, but it's a
3079 byproduct of the fact that you get to control your C<DBI> handles
3082 To maximize performance, use a code snippet like the following:
3084 # prepare a statement handle using the first row
3085 # and then reuse it for the rest of the rows
3087 for my $href (@array_of_hashrefs) {
3088 $stmt ||= $sql->insert('table', $href);
3089 $sth ||= $dbh->prepare($stmt);
3090 $sth->execute($sql->values($href));
3093 The reason this works is because the keys in your C<$href> are sorted
3094 internally by B<SQL::Abstract>. Thus, as long as your data retains
3095 the same structure, you only have to generate the SQL the first time
3096 around. On subsequent queries, simply use the C<values> function provided
3097 by this module to return your values in the correct order.
3099 However this depends on the values having the same type - if, for
3100 example, the values of a where clause may either have values
3101 (resulting in sql of the form C<column = ?> with a single bind
3102 value), or alternatively the values might be C<undef> (resulting in
3103 sql of the form C<column IS NULL> with no bind value) then the
3104 caching technique suggested will not work.
3108 If you use my C<CGI::FormBuilder> module at all, you'll hopefully
3109 really like this part (I do, at least). Building up a complex query
3110 can be as simple as the following:
3117 use CGI::FormBuilder;
3120 my $form = CGI::FormBuilder->new(...);
3121 my $sql = SQL::Abstract->new;
3123 if ($form->submitted) {
3124 my $field = $form->field;
3125 my $id = delete $field->{id};
3126 my($stmt, @bind) = $sql->update('table', $field, {id => $id});
3129 Of course, you would still have to connect using C<DBI> to run the
3130 query, but the point is that if you make your form look like your
3131 table, the actual query script can be extremely simplistic.
3133 If you're B<REALLY> lazy (I am), check out C<HTML::QuickTable> for
3134 a fast interface to returning and formatting data. I frequently
3135 use these three modules together to write complex database query
3136 apps in under 50 lines.
3138 =head1 HOW TO CONTRIBUTE
3140 Contributions are always welcome, in all usable forms (we especially
3141 welcome documentation improvements). The delivery methods include git-
3142 or unified-diff formatted patches, GitHub pull requests, or plain bug
3143 reports either via RT or the Mailing list. Contributors are generally
3144 granted full access to the official repository after their first several
3145 patches pass successful review.
3147 This project is maintained in a git repository. The code and related tools are
3148 accessible at the following locations:
3152 =item * Official repo: L<git://git.shadowcat.co.uk/dbsrgits/SQL-Abstract.git>
3154 =item * Official gitweb: L<http://git.shadowcat.co.uk/gitweb/gitweb.cgi?p=dbsrgits/SQL-Abstract.git>
3156 =item * GitHub mirror: L<https://github.com/dbsrgits/sql-abstract>
3158 =item * Authorized committers: L<ssh://dbsrgits@git.shadowcat.co.uk/SQL-Abstract.git>
3164 Version 1.50 was a major internal refactoring of C<SQL::Abstract>.
3165 Great care has been taken to preserve the I<published> behavior
3166 documented in previous versions in the 1.* family; however,
3167 some features that were previously undocumented, or behaved
3168 differently from the documentation, had to be changed in order
3169 to clarify the semantics. Hence, client code that was relying
3170 on some dark areas of C<SQL::Abstract> v1.*
3171 B<might behave differently> in v1.50.
3173 The main changes are:
3179 support for literal SQL through the C<< \ [ $sql, @bind ] >> syntax.
3183 support for the { operator => \"..." } construct (to embed literal SQL)
3187 support for the { operator => \["...", @bind] } construct (to embed literal SQL with bind values)
3191 optional support for L<array datatypes|/"Inserting and Updating Arrays">
3195 defensive programming: check arguments
3199 fixed bug with global logic, which was previously implemented
3200 through global variables yielding side-effects. Prior versions would
3201 interpret C<< [ {cond1, cond2}, [cond3, cond4] ] >>
3202 as C<< "(cond1 AND cond2) OR (cond3 AND cond4)" >>.
3203 Now this is interpreted
3204 as C<< "(cond1 AND cond2) OR (cond3 OR cond4)" >>.
3209 fixed semantics of _bindtype on array args
3213 dropped the C<_anoncopy> of the %where tree. No longer necessary,
3214 we just avoid shifting arrays within that tree.
3218 dropped the C<_modlogic> function
3222 =head1 ACKNOWLEDGEMENTS
3224 There are a number of individuals that have really helped out with
3225 this module. Unfortunately, most of them submitted bugs via CPAN
3226 so I have no idea who they are! But the people I do know are:
3228 Ash Berlin (order_by hash term support)
3229 Matt Trout (DBIx::Class support)
3230 Mark Stosberg (benchmarking)
3231 Chas Owens (initial "IN" operator support)
3232 Philip Collins (per-field SQL functions)
3233 Eric Kolve (hashref "AND" support)
3234 Mike Fragassi (enhancements to "BETWEEN" and "LIKE")
3235 Dan Kubb (support for "quote_char" and "name_sep")
3236 Guillermo Roditi (patch to cleanup "IN" and "BETWEEN", fix and tests for _order_by)
3237 Laurent Dami (internal refactoring, extensible list of special operators, literal SQL)
3238 Norbert Buchmuller (support for literal SQL in hashpair, misc. fixes & tests)
3239 Peter Rabbitson (rewrite of SQLA::Test, misc. fixes & tests)
3240 Oliver Charles (support for "RETURNING" after "INSERT")
3246 L<DBIx::Class>, L<DBIx::Abstract>, L<CGI::FormBuilder>, L<HTML::QuickTable>.
3250 Copyright (c) 2001-2007 Nathan Wiger <nwiger@cpan.org>. All Rights Reserved.
3252 This module is actively maintained by Matt Trout <mst@shadowcatsystems.co.uk>
3254 For support, your best bet is to try the C<DBIx::Class> users mailing list.
3255 While not an official support venue, C<DBIx::Class> makes heavy use of
3256 C<SQL::Abstract>, and as such list members there are very familiar with
3257 how to create queries.
3261 This module is free software; you may copy this under the same
3262 terms as perl itself (either the GNU General Public License or
3263 the Artistic License)