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" }},
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',
211 func => '_expand_func',
215 'between' => '_expand_between',
216 'not_between' => '_expand_between',
217 'in' => '_expand_in',
218 'not_in' => '_expand_in',
219 'nest' => '_expand_nest',
220 (map +($_ => '_expand_op_andor'), ('and', 'or')),
221 (map +($_ => '_expand_op_is'), ('is', 'is_not')),
222 'ident' => '_expand_ident',
223 'value' => '_expand_value',
227 (map +($_, "_render_$_"), qw(op func bind ident literal row)),
232 (map +($_ => '_render_op_between'), 'between', 'not_between'),
233 (map +($_ => '_render_op_in'), 'in', 'not_in'),
234 (map +($_ => '_render_unop_postfix'),
235 'is_null', 'is_not_null', 'asc', 'desc',
237 (not => '_render_unop_paren'),
238 (map +($_ => '_render_op_andor'), qw(and or)),
239 ',' => '_render_op_multop',
242 return bless \%opt, $class;
245 sub sqltrue { +{ -literal => [ $_[0]->{sqltrue} ] } }
246 sub sqlfalse { +{ -literal => [ $_[0]->{sqlfalse} ] } }
248 sub _assert_pass_injection_guard {
249 if ($_[1] =~ $_[0]->{injection_guard}) {
250 my $class = ref $_[0];
251 puke "Possible SQL injection attempt '$_[1]'. If this is indeed a part of the "
252 . "desired SQL use literal SQL ( \'...' or \[ '...' ] ) or supply your own "
253 . "{injection_guard} attribute to ${class}->new()"
258 #======================================================================
260 #======================================================================
264 my $table = $self->_table(shift);
265 my $data = shift || return;
270 my ($f_aqt, $v_aqt) = $self->_expand_insert_values($data);
272 my @parts = ([ $self->_sqlcase('insert into').' '.$table ]);
273 push @parts, [ $self->render_aqt($f_aqt) ] if $f_aqt;
274 push @parts, [ $self->_sqlcase('values') ], [ $self->render_aqt($v_aqt) ];
276 if ($options->{returning}) {
277 push @parts, [ $self->_insert_returning($options) ];
280 return $self->join_query_parts(' ', @parts);
283 sub _expand_insert_values {
284 my ($self, $data) = @_;
285 if (is_literal_value($data)) {
286 (undef, $self->expand_expr($data));
288 my ($fields, $values) = (
289 ref($data) eq 'HASH' ?
290 ([ sort keys %$data ], [ @{$data}{sort keys %$data} ])
294 # no names (arrayref) means can't generate bindtype
295 !($fields) && $self->{bindtype} eq 'columns'
296 && belch "can't do 'columns' bindtype when called with arrayref";
300 ? $self->expand_expr({ -row => $fields }, -ident)
305 local our $Cur_Col_Meta = $fields->[$_];
306 $self->_expand_insert_value($values->[$_])
313 # So that subclasses can override INSERT ... RETURNING separately from
314 # UPDATE and DELETE (e.g. DBIx::Class::SQLMaker::Oracle does this)
315 sub _insert_returning { shift->_returning(@_) }
318 my ($self, $options) = @_;
320 my $f = $options->{returning};
322 my ($sql, @bind) = $self->render_aqt(
323 $self->_expand_maybe_list_expr($f, -ident)
325 return ($self->_sqlcase(' returning ').$sql, @bind);
328 sub _expand_insert_value {
331 my $k = our $Cur_Col_Meta;
333 if (ref($v) eq 'ARRAY') {
334 if ($self->{array_datatypes}) {
335 return +{ -bind => [ $k, $v ] };
337 my ($sql, @bind) = @$v;
338 $self->_assert_bindval_matches_bindtype(@bind);
339 return +{ -literal => $v };
341 if (ref($v) eq 'HASH') {
342 if (grep !/^-/, keys %$v) {
343 belch "HASH ref as bind value in insert is not supported";
344 return +{ -bind => [ $k, $v ] };
348 return +{ -bind => [ $k, undef ] };
350 return $self->expand_expr($v);
355 #======================================================================
357 #======================================================================
362 my $table = $self->_table(shift);
363 my $data = shift || return;
367 # first build the 'SET' part of the sql statement
368 puke "Unsupported data type specified to \$sql->update"
369 unless ref $data eq 'HASH';
371 my ($sql, @all_bind) = $self->_update_set_values($data);
372 $sql = $self->_sqlcase('update ') . $table . $self->_sqlcase(' set ')
376 my($where_sql, @where_bind) = $self->where($where);
378 push @all_bind, @where_bind;
381 if ($options->{returning}) {
382 my ($returning_sql, @returning_bind) = $self->_update_returning($options);
383 $sql .= $returning_sql;
384 push @all_bind, @returning_bind;
387 return wantarray ? ($sql, @all_bind) : $sql;
390 sub _update_set_values {
391 my ($self, $data) = @_;
393 return $self->render_aqt(
394 $self->_expand_update_set_values(undef, $data),
398 sub _expand_update_set_values {
399 my ($self, undef, $data) = @_;
400 $self->_expand_maybe_list_expr( [
403 $set = { -bind => $_ } unless defined $set;
404 +{ -op => [ '=', $self->_expand_ident(-ident => $k), $set ] };
410 ? ($self->{array_datatypes}
411 ? [ $k, +{ -bind => [ $k, $v ] } ]
412 : [ $k, +{ -literal => $v } ])
414 local our $Cur_Col_Meta = $k;
415 [ $k, $self->_expand_expr($v) ]
422 # So that subclasses can override UPDATE ... RETURNING separately from
424 sub _update_returning { shift->_returning(@_) }
428 #======================================================================
430 #======================================================================
435 my $table = $self->_table(shift);
436 my $fields = shift || '*';
440 my ($fields_sql, @bind) = $self->_select_fields($fields);
442 my ($where_sql, @where_bind) = $self->where($where, $order);
443 push @bind, @where_bind;
445 my $sql = join(' ', $self->_sqlcase('select'), $fields_sql,
446 $self->_sqlcase('from'), $table)
449 return wantarray ? ($sql, @bind) : $sql;
453 my ($self, $fields) = @_;
454 return $fields unless ref($fields);
455 return $self->render_aqt(
456 $self->_expand_maybe_list_expr($fields, '-ident')
460 #======================================================================
462 #======================================================================
467 my $table = $self->_table(shift);
471 my($where_sql, @bind) = $self->where($where);
472 my $sql = $self->_sqlcase('delete from ') . $table . $where_sql;
474 if ($options->{returning}) {
475 my ($returning_sql, @returning_bind) = $self->_delete_returning($options);
476 $sql .= $returning_sql;
477 push @bind, @returning_bind;
480 return wantarray ? ($sql, @bind) : $sql;
483 # So that subclasses can override DELETE ... RETURNING separately from
485 sub _delete_returning { shift->_returning(@_) }
489 #======================================================================
491 #======================================================================
495 # Finally, a separate routine just to handle WHERE clauses
497 my ($self, $where, $order) = @_;
499 local $self->{convert_where} = $self->{convert};
502 my ($sql, @bind) = defined($where)
503 ? $self->_recurse_where($where)
505 $sql = (defined $sql and length $sql) ? $self->_sqlcase(' where ') . "( $sql )" : '';
509 my ($order_sql, @order_bind) = $self->_order_by($order);
511 push @bind, @order_bind;
514 return wantarray ? ($sql, @bind) : $sql;
517 { our $Default_Scalar_To = -value }
520 my ($self, $expr, $default_scalar_to) = @_;
521 local our $Default_Scalar_To = $default_scalar_to if $default_scalar_to;
522 $self->_expand_expr($expr);
526 my ($self, $aqt) = @_;
527 my ($k, $v, @rest) = %$aqt;
529 die "Not a node type: $k" unless $k =~ s/^-//;
530 if (my $meth = $self->{render}{$k}) {
531 return $self->$meth($k, $v);
533 die "notreached: $k";
537 my ($self, $expr, $default_scalar_to) = @_;
538 my ($sql, @bind) = $self->render_aqt(
539 $self->expand_expr($expr, $default_scalar_to)
541 return ($sql, @bind);
545 my ($self, $raw) = @_;
546 s/^-(?=.)//, s/\s+/_/g for my $op = lc $raw;
551 my ($self, $expr) = @_;
552 our $Expand_Depth ||= 0; local $Expand_Depth = $Expand_Depth + 1;
553 return undef unless defined($expr);
554 if (ref($expr) eq 'HASH') {
555 return undef unless my $kc = keys %$expr;
557 return $self->_expand_op_andor(and => $expr);
559 my ($key, $value) = %$expr;
560 if ($key =~ /^-/ and $key =~ s/ [_\s]? \d+ $//x ) {
561 belch 'Use of [and|or|nest]_N modifiers is deprecated and will be removed in SQLA v2.0. '
562 . "You probably wanted ...-and => [ $key => COND1, $key => COND2 ... ]";
564 return $self->_expand_hashpair($key, $value);
566 if (ref($expr) eq 'ARRAY') {
567 return $self->_expand_op_andor(lc($self->{logic}), $expr);
569 if (my $literal = is_literal_value($expr)) {
570 return +{ -literal => $literal };
572 if (!ref($expr) or Scalar::Util::blessed($expr)) {
573 return $self->_expand_scalar($expr);
578 sub _expand_hashpair {
579 my ($self, $k, $v) = @_;
580 unless (defined($k) and length($k)) {
581 if (defined($k) and my $literal = is_literal_value($v)) {
582 belch 'Hash-pairs consisting of an empty string with a literal are deprecated, and will be removed in 2.0: use -and => [ $literal ] instead';
583 return { -literal => $literal };
585 puke "Supplying an empty left hand side argument is not supported";
588 return $self->_expand_hashpair_op($k, $v);
589 } elsif ($k =~ /^[^\w]/i) {
590 my ($lhs, @rhs) = @$v;
591 return $self->_expand_op(
592 -op, [ $k, $self->expand_expr($lhs, -ident), @rhs ]
595 return $self->_expand_hashpair_ident($k, $v);
598 sub _expand_hashpair_ident {
599 my ($self, $k, $v) = @_;
601 local our $Cur_Col_Meta = $k;
603 # hash with multiple or no elements is andor
605 if (ref($v) eq 'HASH' and keys %$v != 1) {
606 return $self->_expand_op_andor(and => $v, $k);
609 # undef needs to be re-sent with cmp to achieve IS/IS NOT NULL
611 if (is_undef_value($v)) {
612 return $self->_expand_hashpair_cmp($k => undef);
615 # scalars and objects get expanded as whatever requested or values
617 if (!ref($v) or Scalar::Util::blessed($v)) {
618 return $self->_expand_hashpair_scalar($k, $v);
621 # single key hashref is a hashtriple
623 if (ref($v) eq 'HASH') {
624 return $self->_expand_hashtriple($k, %$v);
627 # arrayref needs re-engineering over the elements
629 if (ref($v) eq 'ARRAY') {
630 return $self->sqlfalse unless @$v;
631 $self->_debug("ARRAY($k) means distribute over elements");
633 $v->[0] =~ /^-(and|or)$/i
634 ? (shift(@{$v = [ @$v ]}), $1)
635 : lc($self->{logic} || 'OR')
637 return $self->_expand_op_andor(
642 if (my $literal = is_literal_value($v)) {
644 belch 'Hash-pairs consisting of an empty string with a literal are deprecated, and will be removed in 2.0: use -and => [ $literal ] instead';
647 my ($sql, @bind) = @$literal;
648 if ($self->{bindtype} eq 'columns') {
650 $self->_assert_bindval_matches_bindtype($_);
653 return +{ -literal => [ $self->_quote($k).' '.$sql, @bind ] };
659 my ($self, $expr) = @_;
661 return $self->_expand_expr({ (our $Default_Scalar_To) => $expr });
664 sub _expand_hashpair_scalar {
665 my ($self, $k, $v) = @_;
667 return $self->_expand_hashpair_cmp(
668 $k, $self->_expand_scalar($v),
672 sub _expand_hashpair_op {
673 my ($self, $k, $v) = @_;
675 $self->_assert_pass_injection_guard($k =~ /\A-(.*)\Z/s);
677 my $op = $self->_normalize_op($k);
679 if (my $exp = $self->{expand}{$op}) {
680 return $self->$exp($op, $v);
683 # Ops prefixed with -not_ get converted
685 if (my ($rest) = $op =~/^not_(.*)$/) {
688 $self->_expand_expr({ "-${rest}", $v })
694 my $op = join(' ', split '_', $op);
696 # the old special op system requires illegality for top-level use
699 (our $Expand_Depth) == 1
701 List::Util::first { $op =~ $_->{regex} } @{$self->{special_ops}}
703 $self->{disable_old_special_ops}
704 and List::Util::first { $op =~ $_->{regex} } @BUILTIN_SPECIAL_OPS
708 puke "Illegal use of top-level '-$op'"
711 # the old unary op system means we should touch nothing and let it work
713 if (my $us = List::Util::first { $op =~ $_->{regex} } @{$self->{unary_ops}}) {
714 return { -op => [ $op, $v ] };
718 # an explicit node type is currently assumed to be expanded (this is almost
719 # certainly wrong and there should be expansion anyway)
721 if ($self->{render}{$op}) {
725 my $type = $self->{unknown_unop_always_func} ? -func : -op;
732 and (keys %$v)[0] =~ /^-/
735 (List::Util::first { $op =~ $_->{regex} } @{$self->{special_ops}})
744 ($type eq -func and ref($v) eq 'ARRAY')
745 ? map $self->_expand_expr($_), @$v
746 : $self->_expand_expr($v)
750 sub _expand_hashpair_cmp {
751 my ($self, $k, $v) = @_;
752 $self->_expand_hashtriple($k, $self->{cmp}, $v);
755 sub _expand_hashtriple {
756 my ($self, $k, $vk, $vv) = @_;
758 my $ik = $self->_expand_ident(-ident => $k);
760 my $op = $self->_normalize_op($vk);
761 $self->_assert_pass_injection_guard($op);
763 if ($op =~ s/ _? \d+ $//x ) {
764 return $self->_expand_expr($k, { $vk, $vv });
766 if (my $x = $self->{expand_op}{$op}) {
767 local our $Cur_Col_Meta = $k;
768 return $self->$x($op, $vv, $k);
772 my $op = join(' ', split '_', $op);
774 if (my $us = List::Util::first { $op =~ $_->{regex} } @{$self->{special_ops}}) {
775 return { -op => [ $op, $ik, $vv ] };
777 if (my $us = List::Util::first { $op =~ $_->{regex} } @{$self->{unary_ops}}) {
781 { -op => [ $op, $vv ] }
785 if (ref($vv) eq 'ARRAY') {
787 my $logic = (defined($raw[0]) and $raw[0] =~ /^-(and|or)$/i)
788 ? (shift(@raw), $1) : 'or';
789 my @values = map +{ $vk => $_ }, @raw;
791 $op =~ $self->{inequality_op}
792 or $op =~ $self->{not_like_op}
794 if (lc($logic) eq 'or' and @values > 1) {
795 belch "A multi-element arrayref as an argument to the inequality op '${\uc(join ' ', split '_', $op)}' "
796 . 'is technically equivalent to an always-true 1=1 (you probably wanted '
797 . "to say ...{ \$inequality_op => [ -and => \@values ] }... instead)"
802 # try to DWIM on equality operators
803 return ($self->_dwim_op_to_is($op,
804 "Supplying an empty arrayref to '%s' is deprecated",
805 "operator '%s' applied on an empty array (field '$k')"
806 ) ? $self->sqlfalse : $self->sqltrue);
808 return $self->_expand_op_andor($logic => \@values, $k);
810 if (is_undef_value($vv)) {
811 my $is = ($self->_dwim_op_to_is($op,
812 "Supplying an undefined argument to '%s' is deprecated",
813 "unexpected operator '%s' with undef operand",
814 ) ? 'is' : 'is not');
816 return $self->_expand_hashpair($k => { $is, undef });
818 local our $Cur_Col_Meta = $k;
822 $self->_expand_expr($vv)
827 my ($self, $raw, $empty, $fail) = @_;
829 my $op = $self->_normalize_op($raw);
831 if ($op =~ /^not$/i) {
834 if ($op =~ $self->{equality_op}) {
837 if ($op =~ $self->{like_op}) {
838 belch(sprintf $empty, uc(join ' ', split '_', $op));
841 if ($op =~ $self->{inequality_op}) {
844 if ($op =~ $self->{not_like_op}) {
845 belch(sprintf $empty, uc(join ' ', split '_', $op));
848 puke(sprintf $fail, $op);
852 my ($self, undef, $args) = @_;
853 my ($func, @args) = @$args;
854 return { -func => [ $func, map $self->expand_expr($_), @args ] };
858 my ($self, undef, $body, $k) = @_;
859 return $self->_expand_hashpair_cmp(
860 $k, { -ident => $body }
862 unless (defined($body) or (ref($body) and ref($body) eq 'ARRAY')) {
863 puke "-ident requires a single plain scalar argument (a quotable identifier) or an arrayref of identifier parts";
865 my @parts = map split(/\Q${\($self->{name_sep}||'.')}\E/, $_),
866 ref($body) ? @$body : $body;
867 return { -ident => $parts[-1] } if $self->{_dequalify_idents};
868 unless ($self->{quote_char}) {
869 $self->_assert_pass_injection_guard($_) for @parts;
871 return +{ -ident => \@parts };
875 return $_[0]->_expand_hashpair_cmp(
876 $_[3], { -value => $_[2] },
878 +{ -bind => [ our $Cur_Col_Meta, $_[2] ] };
882 +{ -op => [ 'not', $_[0]->_expand_expr($_[2]) ] };
886 my ($self, undef, $args) = @_;
887 +{ -row => [ map $self->expand_expr($_), @$args ] };
891 my ($self, undef, $args) = @_;
892 my ($op, @opargs) = @$args;
893 if (my $exp = $self->{expand_op}{$op}) {
894 return $self->$exp($op, \@opargs);
896 +{ -op => [ $op, map $self->expand_expr($_), @opargs ] };
900 my ($self, undef, $v) = @_;
902 return $self->_expand_expr($v);
904 puke "-bool => undef not supported" unless defined($v);
905 return $self->_expand_ident(-ident => $v);
908 sub _expand_op_andor {
909 my ($self, $logop, $v, $k) = @_;
911 $v = [ map +{ $k, $_ },
913 ? (map +{ $_ => $v->{$_} }, sort keys %$v)
917 if (ref($v) eq 'HASH') {
918 return undef unless keys %$v;
921 map $self->_expand_expr({ $_ => $v->{$_} }),
925 if (ref($v) eq 'ARRAY') {
926 $logop eq 'and' or $logop eq 'or' or puke "unknown logic: $logop";
929 (ref($_) eq 'ARRAY' and @$_)
930 or (ref($_) eq 'HASH' and %$_)
936 while (my ($el) = splice @expr, 0, 1) {
937 puke "Supplying an empty left hand side argument is not supported in array-pairs"
938 unless defined($el) and length($el);
939 my $elref = ref($el);
941 local our $Expand_Depth = 0;
942 push(@res, grep defined, $self->_expand_expr({ $el, shift(@expr) }));
943 } elsif ($elref eq 'ARRAY') {
944 push(@res, grep defined, $self->_expand_expr($el)) if @$el;
945 } elsif (my $l = is_literal_value($el)) {
946 push @res, { -literal => $l };
947 } elsif ($elref eq 'HASH') {
948 local our $Expand_Depth = 0;
949 push @res, grep defined, $self->_expand_expr($el) if %$el;
955 # return $res[0] if @res == 1;
956 return { -op => [ $logop, @res ] };
962 my ($self, $op, $vv, $k) = @_;
963 ($k, $vv) = @$vv unless defined $k;
964 puke "$op can only take undef as argument"
968 and exists($vv->{-value})
969 and !defined($vv->{-value})
971 return +{ -op => [ $op.'_null', $self->expand_expr($k, -ident) ] };
974 sub _expand_between {
975 my ($self, $op, $vv, $k) = @_;
976 $k = shift @{$vv = [ @$vv ]} unless defined $k;
977 my @rhs = map $self->_expand_expr($_),
978 ref($vv) eq 'ARRAY' ? @$vv : $vv;
980 (@rhs == 1 and ref($rhs[0]) eq 'HASH' and $rhs[0]->{-literal})
982 (@rhs == 2 and defined($rhs[0]) and defined($rhs[1]))
984 puke "Operator '${\uc($op)}' requires either an arrayref with two defined values or expressions, or a single literal scalarref/arrayref-ref";
988 $self->expand_expr(ref($k) ? $k : { -ident => $k }),
994 my ($self, $op, $vv, $k) = @_;
995 $k = shift @{$vv = [ @$vv ]} unless defined $k;
996 if (my $literal = is_literal_value($vv)) {
997 my ($sql, @bind) = @$literal;
998 my $opened_sql = $self->_open_outer_paren($sql);
1000 $op, $self->expand_expr($k, -ident),
1001 { -literal => [ $opened_sql, @bind ] }
1005 'SQL::Abstract before v1.75 used to generate incorrect SQL when the '
1006 . "-${\uc($op)} operator was given an undef-containing list: !!!AUDIT YOUR CODE "
1007 . 'AND DATA!!! (the upcoming Data::Query-based version of SQL::Abstract '
1008 . 'will emit the logically correct SQL instead of raising this exception)'
1010 puke("Argument passed to the '${\uc($op)}' operator can not be undefined")
1012 my @rhs = map $self->expand_expr($_, -value),
1013 map { defined($_) ? $_: puke($undef_err) }
1014 (ref($vv) eq 'ARRAY' ? @$vv : $vv);
1015 return $self->${\($op =~ /^not/ ? 'sqltrue' : 'sqlfalse')} unless @rhs;
1019 $self->expand_expr($k, -ident),
1025 my ($self, undef, $v) = @_;
1026 # DBIx::Class requires a nest warning to be emitted once but the private
1027 # method it overrode to do so no longer exists
1028 if ($self->{warn_once_on_nest}) {
1029 unless (our $Nest_Warned) {
1031 "-nest in search conditions is deprecated, you most probably wanted:\n"
1032 .q|{..., -and => [ \%cond0, \@cond1, \'cond2', \[ 'cond3', [ col => bind ] ], etc. ], ... }|
1037 return $self->_expand_expr($v);
1041 my ($self, undef, $bind) = @_;
1042 return { -bind => $bind };
1045 sub _recurse_where {
1046 my ($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, undef, $ident) = @_;
1071 return $self->_convert($self->_quote($ident));
1075 my ($self, undef, $values) = @_;
1076 my ($sql, @bind) = $self->_render_op(undef, [ ',', @$values ]);
1077 return $self->join_query_parts('', [ '(' ], [ $sql, @bind ], [ ')' ]);
1081 my ($self, undef, $rest) = @_;
1082 my ($func, @args) = @$rest;
1083 return $self->join_query_parts('',
1084 [ $self->_sqlcase($func) ],
1086 [ $self->join_query_parts(', ', @args) ],
1092 my ($self, undef, $bind) = @_;
1093 return ($self->_convert('?'), $self->_bindtype(@$bind));
1096 sub _render_literal {
1097 my ($self, undef, $literal) = @_;
1098 $self->_assert_bindval_matches_bindtype(@{$literal}[1..$#$literal]);
1103 my ($self, undef, $v) = @_;
1104 my ($op, @args) = @$v;
1105 if (my $r = $self->{render_op}{$op}) {
1106 return $self->$r($op, \@args);
1111 my $op = join(' ', split '_', $op);
1113 my $ss = List::Util::first { $op =~ $_->{regex} } @{$self->{special_ops}};
1114 if ($ss and @args > 1) {
1115 puke "Special op '${op}' requires first value to be identifier"
1116 unless my ($ident) = map $_->{-ident}, grep ref($_) eq 'HASH', $args[0];
1117 my $k = join(($self->{name_sep}||'.'), @$ident);
1118 local our $Expand_Depth = 1;
1119 return $self->${\($ss->{handler})}($k, $op, $args[1]);
1121 if (my $us = List::Util::first { $op =~ $_->{regex} } @{$self->{unary_ops}}) {
1122 return $self->${\($us->{handler})}($op, $args[0]);
1125 return $self->_render_unop_paren($op, \@args);
1129 return $self->_render_unop_prefix($op, \@args);
1131 return $self->_render_op_multop($op, \@args);
1137 sub _render_op_between {
1138 my ($self, $op, $args) = @_;
1139 my ($left, $low, $high) = @$args;
1142 puke "Single arg to between must be a literal"
1143 unless $low->{-literal};
1146 +($low, $self->format_keyword('and'), $high);
1149 return $self->join_query_parts(' ',
1150 '(', $left, $self->format_keyword($op), @rh, ')',
1155 my ($self, $op, $args) = @_;
1156 my ($lhs, @rhs) = @$args;
1158 return $self->join_query_parts(' ',
1160 $self->format_keyword($op),
1162 [ $self->join_query_parts(', ', @rhs) ],
1167 sub _render_op_andor {
1168 my ($self, $op, $args) = @_;
1169 return '' unless @$args;
1170 return $self->join_query_parts('', $args->[0]) if @$args == 1;
1171 return $self->join_query_parts(
1172 ' ' => '(',[ $self->_render_op_multop($op, $args) ], ')'
1176 sub _render_op_multop {
1177 my ($self, $op, $args) = @_;
1178 my @parts = map [ $self->render_aqt($_) ], @$args;
1179 return '' unless @parts;
1180 return @{$parts[0]} if @parts == 1;
1181 my $join = ($op eq ','
1183 : ' '.$self->format_keyword($op).' '
1185 return $self->join_query_parts($join, @parts);
1188 sub join_query_parts {
1189 my ($self, $join, @parts) = @_;
1192 ? [ $self->render_aqt($_) ]
1193 : ref($_) eq 'ARRAY' ? $_ : [ $_ ]),
1196 join($join, map $_->[0], @final),
1197 (map @{$_}[1..$#$_], @final),
1201 sub _render_unop_paren {
1202 my ($self, $op, $v) = @_;
1203 return $self->join_query_parts('',
1204 '(', [ $self->_render_unop_prefix($op, $v) ], ')'
1208 sub _render_unop_prefix {
1209 my ($self, $op, $v) = @_;
1210 return $self->join_query_parts(' ',
1211 $self->_sqlcase($op), $v->[0]
1215 sub _render_unop_postfix {
1216 my ($self, $op, $v) = @_;
1217 return $self->join_query_parts(' ',
1218 $v->[0], $self->format_keyword($op),
1222 # Some databases (SQLite) treat col IN (1, 2) different from
1223 # col IN ( (1, 2) ). Use this to strip all outer parens while
1224 # adding them back in the corresponding method
1225 sub _open_outer_paren {
1226 my ($self, $sql) = @_;
1228 while (my ($inner) = $sql =~ /^ \s* \( (.*) \) \s* $/xs) {
1230 # there are closing parens inside, need the heavy duty machinery
1231 # to reevaluate the extraction starting from $sql (full reevaluation)
1232 if ($inner =~ /\)/) {
1233 require Text::Balanced;
1235 my (undef, $remainder) = do {
1236 # idiotic design - writes to $@ but *DOES NOT* throw exceptions
1238 Text::Balanced::extract_bracketed($sql, '()', qr/\s*/);
1241 # the entire expression needs to be a balanced bracketed thing
1242 # (after an extract no remainder sans trailing space)
1243 last if defined $remainder and $remainder =~ /\S/;
1253 #======================================================================
1255 #======================================================================
1257 sub _expand_order_by {
1258 my ($self, $arg) = @_;
1260 return unless defined($arg) and not (ref($arg) eq 'ARRAY' and !@$arg);
1262 return $self->_expand_maybe_list_expr($arg)
1263 if ref($arg) eq 'HASH' and ($arg->{-op}||[''])->[0] eq ',';
1265 my $expander = sub {
1266 my ($self, $dir, $expr) = @_;
1267 my @to_expand = ref($expr) eq 'ARRAY' ? @$expr : $expr;
1268 foreach my $arg (@to_expand) {
1272 and grep /^-(asc|desc)$/, keys %$arg
1274 puke "ordering direction hash passed to order by must have exactly one key (-asc or -desc)";
1278 defined($dir) ? { -op => [ $dir =~ /^-?(.*)$/ ,=> $_ ] } : $_
1280 map $self->expand_expr($_, -ident),
1281 map ref($_) eq 'ARRAY' ? @$_ : $_, @to_expand;
1282 return undef unless @exp;
1283 return undef if @exp == 1 and not defined($exp[0]);
1284 return +{ -op => [ ',', @exp ] };
1287 local @{$self->{expand}}{qw(asc desc)} = (($expander) x 2);
1289 return $self->$expander(undef, $arg);
1293 my ($self, $arg) = @_;
1295 return '' unless defined(my $expanded = $self->_expand_order_by($arg));
1297 my ($sql, @bind) = $self->render_aqt($expanded);
1299 return '' unless length($sql);
1301 my $final_sql = $self->_sqlcase(' order by ').$sql;
1303 return ($final_sql, @bind);
1306 # _order_by no longer needs to call this so doesn't but DBIC uses it.
1308 sub _order_by_chunks {
1309 my ($self, $arg) = @_;
1311 return () unless defined(my $expanded = $self->_expand_order_by($arg));
1313 return $self->_chunkify_order_by($expanded);
1316 sub _chunkify_order_by {
1317 my ($self, $expanded) = @_;
1319 return grep length, $self->render_aqt($expanded)
1320 if $expanded->{-ident} or @{$expanded->{-literal}||[]} == 1;
1323 if (ref() eq 'HASH' and $_->{-op} and $_->{-op}[0] eq ',') {
1324 my ($comma, @list) = @{$_->{-op}};
1325 return map $self->_chunkify_order_by($_), @list;
1327 return [ $self->render_aqt($_) ];
1331 #======================================================================
1332 # DATASOURCE (FOR NOW, JUST PLAIN TABLE OR LIST OF TABLES)
1333 #======================================================================
1339 $self->_expand_maybe_list_expr($from, -ident)
1344 #======================================================================
1346 #======================================================================
1348 sub _expand_maybe_list_expr {
1349 my ($self, $expr, $default) = @_;
1351 ',', map $self->expand_expr($_, $default),
1352 @{$expr->{-op}}[1..$#{$expr->{-op}}]
1353 ] } if ref($expr) eq 'HASH' and ($expr->{-op}||[''])->[0] eq ',';
1354 return +{ -op => [ ',',
1355 map $self->expand_expr($_, $default),
1356 ref($expr) eq 'ARRAY' ? @$expr : $expr
1360 # highly optimized, as it's called way too often
1362 # my ($self, $label) = @_;
1364 return '' unless defined $_[1];
1365 return ${$_[1]} if ref($_[1]) eq 'SCALAR';
1366 puke 'Identifier cannot be hashref' if ref($_[1]) eq 'HASH';
1368 unless ($_[0]->{quote_char}) {
1369 if (ref($_[1]) eq 'ARRAY') {
1370 return join($_[0]->{name_sep}||'.', @{$_[1]});
1372 $_[0]->_assert_pass_injection_guard($_[1]);
1377 my $qref = ref $_[0]->{quote_char};
1379 !$qref ? ($_[0]->{quote_char}, $_[0]->{quote_char})
1380 : ($qref eq 'ARRAY') ? @{$_[0]->{quote_char}}
1381 : puke "Unsupported quote_char format: $_[0]->{quote_char}";
1383 my $esc = $_[0]->{escape_char} || $r;
1385 # parts containing * are naturally unquoted
1387 $_[0]->{name_sep}||'',
1391 : do { (my $n = $_) =~ s/(\Q$esc\E|\Q$r\E)/$esc$1/g; $l . $n . $r }
1393 (ref($_[1]) eq 'ARRAY'
1397 ? split (/\Q$_[0]->{name_sep}\E/, $_[1] )
1405 # Conversion, if applicable
1407 #my ($self, $arg) = @_;
1408 if ($_[0]->{convert_where}) {
1409 return $_[0]->_sqlcase($_[0]->{convert_where}) .'(' . $_[1] . ')';
1416 #my ($self, $col, @vals) = @_;
1417 # called often - tighten code
1418 return $_[0]->{bindtype} eq 'columns'
1419 ? map {[$_[1], $_]} @_[2 .. $#_]
1424 # Dies if any element of @bind is not in [colname => value] format
1425 # if bindtype is 'columns'.
1426 sub _assert_bindval_matches_bindtype {
1427 # my ($self, @bind) = @_;
1429 if ($self->{bindtype} eq 'columns') {
1431 if (!defined $_ || ref($_) ne 'ARRAY' || @$_ != 2) {
1432 puke "bindtype 'columns' selected, you need to pass: [column_name => bind_value]"
1438 sub _join_sql_clauses {
1439 my ($self, $logic, $clauses_aref, $bind_aref) = @_;
1441 if (@$clauses_aref > 1) {
1442 my $join = " " . $self->_sqlcase($logic) . " ";
1443 my $sql = '( ' . join($join, @$clauses_aref) . ' )';
1444 return ($sql, @$bind_aref);
1446 elsif (@$clauses_aref) {
1447 return ($clauses_aref->[0], @$bind_aref); # no parentheses
1450 return (); # if no SQL, ignore @$bind_aref
1455 # Fix SQL case, if so requested
1457 # LDNOTE: if $self->{case} is true, then it contains 'lower', so we
1458 # don't touch the argument ... crooked logic, but let's not change it!
1459 return $_[0]->{case} ? $_[1] : uc($_[1]);
1462 sub format_keyword { $_[0]->_sqlcase(join ' ', split '_', $_[1]) }
1464 #======================================================================
1465 # DISPATCHING FROM REFKIND
1466 #======================================================================
1469 my ($self, $data) = @_;
1471 return 'UNDEF' unless defined $data;
1473 # blessed objects are treated like scalars
1474 my $ref = (Scalar::Util::blessed $data) ? '' : ref $data;
1476 return 'SCALAR' unless $ref;
1479 while ($ref eq 'REF') {
1481 $ref = (Scalar::Util::blessed $data) ? '' : ref $data;
1485 return ($ref||'SCALAR') . ('REF' x $n_steps);
1489 my ($self, $data) = @_;
1490 my @try = ($self->_refkind($data));
1491 push @try, 'SCALAR_or_UNDEF' if $try[0] eq 'SCALAR' || $try[0] eq 'UNDEF';
1492 push @try, 'FALLBACK';
1496 sub _METHOD_FOR_refkind {
1497 my ($self, $meth_prefix, $data) = @_;
1500 for (@{$self->_try_refkind($data)}) {
1501 $method = $self->can($meth_prefix."_".$_)
1505 return $method || puke "cannot dispatch on '$meth_prefix' for ".$self->_refkind($data);
1509 sub _SWITCH_refkind {
1510 my ($self, $data, $dispatch_table) = @_;
1513 for (@{$self->_try_refkind($data)}) {
1514 $coderef = $dispatch_table->{$_}
1518 puke "no dispatch entry for ".$self->_refkind($data)
1527 #======================================================================
1528 # VALUES, GENERATE, AUTOLOAD
1529 #======================================================================
1531 # LDNOTE: original code from nwiger, didn't touch code in that section
1532 # I feel the AUTOLOAD stuff should not be the default, it should
1533 # only be activated on explicit demand by user.
1537 my $data = shift || return;
1538 puke "Argument to ", __PACKAGE__, "->values must be a \\%hash"
1539 unless ref $data eq 'HASH';
1542 foreach my $k (sort keys %$data) {
1543 my $v = $data->{$k};
1544 $self->_SWITCH_refkind($v, {
1546 if ($self->{array_datatypes}) { # array datatype
1547 push @all_bind, $self->_bindtype($k, $v);
1549 else { # literal SQL with bind
1550 my ($sql, @bind) = @$v;
1551 $self->_assert_bindval_matches_bindtype(@bind);
1552 push @all_bind, @bind;
1555 ARRAYREFREF => sub { # literal SQL with bind
1556 my ($sql, @bind) = @${$v};
1557 $self->_assert_bindval_matches_bindtype(@bind);
1558 push @all_bind, @bind;
1560 SCALARREF => sub { # literal SQL without bind
1562 SCALAR_or_UNDEF => sub {
1563 push @all_bind, $self->_bindtype($k, $v);
1574 my(@sql, @sqlq, @sqlv);
1578 if ($ref eq 'HASH') {
1579 for my $k (sort keys %$_) {
1582 my $label = $self->_quote($k);
1583 if ($r eq 'ARRAY') {
1584 # literal SQL with bind
1585 my ($sql, @bind) = @$v;
1586 $self->_assert_bindval_matches_bindtype(@bind);
1587 push @sqlq, "$label = $sql";
1589 } elsif ($r eq 'SCALAR') {
1590 # literal SQL without bind
1591 push @sqlq, "$label = $$v";
1593 push @sqlq, "$label = ?";
1594 push @sqlv, $self->_bindtype($k, $v);
1597 push @sql, $self->_sqlcase('set'), join ', ', @sqlq;
1598 } elsif ($ref eq 'ARRAY') {
1599 # unlike insert(), assume these are ONLY the column names, i.e. for SQL
1602 if ($r eq 'ARRAY') { # literal SQL with bind
1603 my ($sql, @bind) = @$v;
1604 $self->_assert_bindval_matches_bindtype(@bind);
1607 } elsif ($r eq 'SCALAR') { # literal SQL without bind
1608 # embedded literal SQL
1615 push @sql, '(' . join(', ', @sqlq) . ')';
1616 } elsif ($ref eq 'SCALAR') {
1620 # strings get case twiddled
1621 push @sql, $self->_sqlcase($_);
1625 my $sql = join ' ', @sql;
1627 # this is pretty tricky
1628 # if ask for an array, return ($stmt, @bind)
1629 # otherwise, s/?/shift @sqlv/ to put it inline
1631 return ($sql, @sqlv);
1633 1 while $sql =~ s/\?/my $d = shift(@sqlv);
1634 ref $d ? $d->[1] : $d/e;
1643 # This allows us to check for a local, then _form, attr
1645 my($name) = $AUTOLOAD =~ /.*::(.+)/;
1646 puke "AUTOLOAD invoked for method name ${name} and allow_autoload option not set" unless $self->{allow_autoload};
1647 return $self->generate($name, @_);
1658 SQL::Abstract - Generate SQL from Perl data structures
1664 my $sql = SQL::Abstract->new;
1666 my($stmt, @bind) = $sql->select($source, \@fields, \%where, $order);
1668 my($stmt, @bind) = $sql->insert($table, \%fieldvals || \@values);
1670 my($stmt, @bind) = $sql->update($table, \%fieldvals, \%where);
1672 my($stmt, @bind) = $sql->delete($table, \%where);
1674 # Then, use these in your DBI statements
1675 my $sth = $dbh->prepare($stmt);
1676 $sth->execute(@bind);
1678 # Just generate the WHERE clause
1679 my($stmt, @bind) = $sql->where(\%where, $order);
1681 # Return values in the same order, for hashed queries
1682 # See PERFORMANCE section for more details
1683 my @bind = $sql->values(\%fieldvals);
1687 This module was inspired by the excellent L<DBIx::Abstract>.
1688 However, in using that module I found that what I really wanted
1689 to do was generate SQL, but still retain complete control over my
1690 statement handles and use the DBI interface. So, I set out to
1691 create an abstract SQL generation module.
1693 While based on the concepts used by L<DBIx::Abstract>, there are
1694 several important differences, especially when it comes to WHERE
1695 clauses. I have modified the concepts used to make the SQL easier
1696 to generate from Perl data structures and, IMO, more intuitive.
1697 The underlying idea is for this module to do what you mean, based
1698 on the data structures you provide it. The big advantage is that
1699 you don't have to modify your code every time your data changes,
1700 as this module figures it out.
1702 To begin with, an SQL INSERT is as easy as just specifying a hash
1703 of C<key=value> pairs:
1706 name => 'Jimbo Bobson',
1707 phone => '123-456-7890',
1708 address => '42 Sister Lane',
1709 city => 'St. Louis',
1710 state => 'Louisiana',
1713 The SQL can then be generated with this:
1715 my($stmt, @bind) = $sql->insert('people', \%data);
1717 Which would give you something like this:
1719 $stmt = "INSERT INTO people
1720 (address, city, name, phone, state)
1721 VALUES (?, ?, ?, ?, ?)";
1722 @bind = ('42 Sister Lane', 'St. Louis', 'Jimbo Bobson',
1723 '123-456-7890', 'Louisiana');
1725 These are then used directly in your DBI code:
1727 my $sth = $dbh->prepare($stmt);
1728 $sth->execute(@bind);
1730 =head2 Inserting and Updating Arrays
1732 If your database has array types (like for example Postgres),
1733 activate the special option C<< array_datatypes => 1 >>
1734 when creating the C<SQL::Abstract> object.
1735 Then you may use an arrayref to insert and update database array types:
1737 my $sql = SQL::Abstract->new(array_datatypes => 1);
1739 planets => [qw/Mercury Venus Earth Mars/]
1742 my($stmt, @bind) = $sql->insert('solar_system', \%data);
1746 $stmt = "INSERT INTO solar_system (planets) VALUES (?)"
1748 @bind = (['Mercury', 'Venus', 'Earth', 'Mars']);
1751 =head2 Inserting and Updating SQL
1753 In order to apply SQL functions to elements of your C<%data> you may
1754 specify a reference to an arrayref for the given hash value. For example,
1755 if you need to execute the Oracle C<to_date> function on a value, you can
1756 say something like this:
1760 date_entered => \[ "to_date(?,'MM/DD/YYYY')", "03/02/2003" ],
1763 The first value in the array is the actual SQL. Any other values are
1764 optional and would be included in the bind values array. This gives
1767 my($stmt, @bind) = $sql->insert('people', \%data);
1769 $stmt = "INSERT INTO people (name, date_entered)
1770 VALUES (?, to_date(?,'MM/DD/YYYY'))";
1771 @bind = ('Bill', '03/02/2003');
1773 An UPDATE is just as easy, all you change is the name of the function:
1775 my($stmt, @bind) = $sql->update('people', \%data);
1777 Notice that your C<%data> isn't touched; the module will generate
1778 the appropriately quirky SQL for you automatically. Usually you'll
1779 want to specify a WHERE clause for your UPDATE, though, which is
1780 where handling C<%where> hashes comes in handy...
1782 =head2 Complex where statements
1784 This module can generate pretty complicated WHERE statements
1785 easily. For example, simple C<key=value> pairs are taken to mean
1786 equality, and if you want to see if a field is within a set
1787 of values, you can use an arrayref. Let's say we wanted to
1788 SELECT some data based on this criteria:
1791 requestor => 'inna',
1792 worker => ['nwiger', 'rcwe', 'sfz'],
1793 status => { '!=', 'completed' }
1796 my($stmt, @bind) = $sql->select('tickets', '*', \%where);
1798 The above would give you something like this:
1800 $stmt = "SELECT * FROM tickets WHERE
1801 ( requestor = ? ) AND ( status != ? )
1802 AND ( worker = ? OR worker = ? OR worker = ? )";
1803 @bind = ('inna', 'completed', 'nwiger', 'rcwe', 'sfz');
1805 Which you could then use in DBI code like so:
1807 my $sth = $dbh->prepare($stmt);
1808 $sth->execute(@bind);
1814 The methods are simple. There's one for every major SQL operation,
1815 and a constructor you use first. The arguments are specified in a
1816 similar order for each method (table, then fields, then a where
1817 clause) to try and simplify things.
1819 =head2 new(option => 'value')
1821 The C<new()> function takes a list of options and values, and returns
1822 a new B<SQL::Abstract> object which can then be used to generate SQL
1823 through the methods below. The options accepted are:
1829 If set to 'lower', then SQL will be generated in all lowercase. By
1830 default SQL is generated in "textbook" case meaning something like:
1832 SELECT a_field FROM a_table WHERE some_field LIKE '%someval%'
1834 Any setting other than 'lower' is ignored.
1838 This determines what the default comparison operator is. By default
1839 it is C<=>, meaning that a hash like this:
1841 %where = (name => 'nwiger', email => 'nate@wiger.org');
1843 Will generate SQL like this:
1845 WHERE name = 'nwiger' AND email = 'nate@wiger.org'
1847 However, you may want loose comparisons by default, so if you set
1848 C<cmp> to C<like> you would get SQL such as:
1850 WHERE name like 'nwiger' AND email like 'nate@wiger.org'
1852 You can also override the comparison on an individual basis - see
1853 the huge section on L</"WHERE CLAUSES"> at the bottom.
1855 =item sqltrue, sqlfalse
1857 Expressions for inserting boolean values within SQL statements.
1858 By default these are C<1=1> and C<1=0>. They are used
1859 by the special operators C<-in> and C<-not_in> for generating
1860 correct SQL even when the argument is an empty array (see below).
1864 This determines the default logical operator for multiple WHERE
1865 statements in arrays or hashes. If absent, the default logic is "or"
1866 for arrays, and "and" for hashes. This means that a WHERE
1870 event_date => {'>=', '2/13/99'},
1871 event_date => {'<=', '4/24/03'},
1874 will generate SQL like this:
1876 WHERE event_date >= '2/13/99' OR event_date <= '4/24/03'
1878 This is probably not what you want given this query, though (look
1879 at the dates). To change the "OR" to an "AND", simply specify:
1881 my $sql = SQL::Abstract->new(logic => 'and');
1883 Which will change the above C<WHERE> to:
1885 WHERE event_date >= '2/13/99' AND event_date <= '4/24/03'
1887 The logic can also be changed locally by inserting
1888 a modifier in front of an arrayref:
1890 @where = (-and => [event_date => {'>=', '2/13/99'},
1891 event_date => {'<=', '4/24/03'} ]);
1893 See the L</"WHERE CLAUSES"> section for explanations.
1897 This will automatically convert comparisons using the specified SQL
1898 function for both column and value. This is mostly used with an argument
1899 of C<upper> or C<lower>, so that the SQL will have the effect of
1900 case-insensitive "searches". For example, this:
1902 $sql = SQL::Abstract->new(convert => 'upper');
1903 %where = (keywords => 'MaKe iT CAse inSeNSItive');
1905 Will turn out the following SQL:
1907 WHERE upper(keywords) like upper('MaKe iT CAse inSeNSItive')
1909 The conversion can be C<upper()>, C<lower()>, or any other SQL function
1910 that can be applied symmetrically to fields (actually B<SQL::Abstract> does
1911 not validate this option; it will just pass through what you specify verbatim).
1915 This is a kludge because many databases suck. For example, you can't
1916 just bind values using DBI's C<execute()> for Oracle C<CLOB> or C<BLOB> fields.
1917 Instead, you have to use C<bind_param()>:
1919 $sth->bind_param(1, 'reg data');
1920 $sth->bind_param(2, $lots, {ora_type => ORA_CLOB});
1922 The problem is, B<SQL::Abstract> will normally just return a C<@bind> array,
1923 which loses track of which field each slot refers to. Fear not.
1925 If you specify C<bindtype> in new, you can determine how C<@bind> is returned.
1926 Currently, you can specify either C<normal> (default) or C<columns>. If you
1927 specify C<columns>, you will get an array that looks like this:
1929 my $sql = SQL::Abstract->new(bindtype => 'columns');
1930 my($stmt, @bind) = $sql->insert(...);
1933 [ 'column1', 'value1' ],
1934 [ 'column2', 'value2' ],
1935 [ 'column3', 'value3' ],
1938 You can then iterate through this manually, using DBI's C<bind_param()>.
1940 $sth->prepare($stmt);
1943 my($col, $data) = @$_;
1944 if ($col eq 'details' || $col eq 'comments') {
1945 $sth->bind_param($i, $data, {ora_type => ORA_CLOB});
1946 } elsif ($col eq 'image') {
1947 $sth->bind_param($i, $data, {ora_type => ORA_BLOB});
1949 $sth->bind_param($i, $data);
1953 $sth->execute; # execute without @bind now
1955 Now, why would you still use B<SQL::Abstract> if you have to do this crap?
1956 Basically, the advantage is still that you don't have to care which fields
1957 are or are not included. You could wrap that above C<for> loop in a simple
1958 sub called C<bind_fields()> or something and reuse it repeatedly. You still
1959 get a layer of abstraction over manual SQL specification.
1961 Note that if you set L</bindtype> to C<columns>, the C<\[ $sql, @bind ]>
1962 construct (see L</Literal SQL with placeholders and bind values (subqueries)>)
1963 will expect the bind values in this format.
1967 This is the character that a table or column name will be quoted
1968 with. By default this is an empty string, but you could set it to
1969 the character C<`>, to generate SQL like this:
1971 SELECT `a_field` FROM `a_table` WHERE `some_field` LIKE '%someval%'
1973 Alternatively, you can supply an array ref of two items, the first being the left
1974 hand quote character, and the second the right hand quote character. For
1975 example, you could supply C<['[',']']> for SQL Server 2000 compliant quotes
1976 that generates SQL like this:
1978 SELECT [a_field] FROM [a_table] WHERE [some_field] LIKE '%someval%'
1980 Quoting is useful if you have tables or columns names that are reserved
1981 words in your database's SQL dialect.
1985 This is the character that will be used to escape L</quote_char>s appearing
1986 in an identifier before it has been quoted.
1988 The parameter default in case of a single L</quote_char> character is the quote
1991 When opening-closing-style quoting is used (L</quote_char> is an arrayref)
1992 this parameter defaults to the B<closing (right)> L</quote_char>. Occurrences
1993 of the B<opening (left)> L</quote_char> within the identifier are currently left
1994 untouched. The default for opening-closing-style quotes may change in future
1995 versions, thus you are B<strongly encouraged> to specify the escape character
2000 This is the character that separates a table and column name. It is
2001 necessary to specify this when the C<quote_char> option is selected,
2002 so that tables and column names can be individually quoted like this:
2004 SELECT `table`.`one_field` FROM `table` WHERE `table`.`other_field` = 1
2006 =item injection_guard
2008 A regular expression C<qr/.../> that is applied to any C<-function> and unquoted
2009 column name specified in a query structure. This is a safety mechanism to avoid
2010 injection attacks when mishandling user input e.g.:
2012 my %condition_as_column_value_pairs = get_values_from_user();
2013 $sqla->select( ... , \%condition_as_column_value_pairs );
2015 If the expression matches an exception is thrown. Note that literal SQL
2016 supplied via C<\'...'> or C<\['...']> is B<not> checked in any way.
2018 Defaults to checking for C<;> and the C<GO> keyword (TransactSQL)
2020 =item array_datatypes
2022 When this option is true, arrayrefs in INSERT or UPDATE are
2023 interpreted as array datatypes and are passed directly
2025 When this option is false, arrayrefs are interpreted
2026 as literal SQL, just like refs to arrayrefs
2027 (but this behavior is for backwards compatibility; when writing
2028 new queries, use the "reference to arrayref" syntax
2034 Takes a reference to a list of "special operators"
2035 to extend the syntax understood by L<SQL::Abstract>.
2036 See section L</"SPECIAL OPERATORS"> for details.
2040 Takes a reference to a list of "unary operators"
2041 to extend the syntax understood by L<SQL::Abstract>.
2042 See section L</"UNARY OPERATORS"> for details.
2048 =head2 insert($table, \@values || \%fieldvals, \%options)
2050 This is the simplest function. You simply give it a table name
2051 and either an arrayref of values or hashref of field/value pairs.
2052 It returns an SQL INSERT statement and a list of bind values.
2053 See the sections on L</"Inserting and Updating Arrays"> and
2054 L</"Inserting and Updating SQL"> for information on how to insert
2055 with those data types.
2057 The optional C<\%options> hash reference may contain additional
2058 options to generate the insert SQL. Currently supported options
2065 Takes either a scalar of raw SQL fields, or an array reference of
2066 field names, and adds on an SQL C<RETURNING> statement at the end.
2067 This allows you to return data generated by the insert statement
2068 (such as row IDs) without performing another C<SELECT> statement.
2069 Note, however, this is not part of the SQL standard and may not
2070 be supported by all database engines.
2074 =head2 update($table, \%fieldvals, \%where, \%options)
2076 This takes a table, hashref of field/value pairs, and an optional
2077 hashref L<WHERE clause|/WHERE CLAUSES>. It returns an SQL UPDATE function and a list
2079 See the sections on L</"Inserting and Updating Arrays"> and
2080 L</"Inserting and Updating SQL"> for information on how to insert
2081 with those data types.
2083 The optional C<\%options> hash reference may contain additional
2084 options to generate the update SQL. Currently supported options
2091 See the C<returning> option to
2092 L<insert|/insert($table, \@values || \%fieldvals, \%options)>.
2096 =head2 select($source, $fields, $where, $order)
2098 This returns a SQL SELECT statement and associated list of bind values, as
2099 specified by the arguments:
2105 Specification of the 'FROM' part of the statement.
2106 The argument can be either a plain scalar (interpreted as a table
2107 name, will be quoted), or an arrayref (interpreted as a list
2108 of table names, joined by commas, quoted), or a scalarref
2109 (literal SQL, not quoted).
2113 Specification of the list of fields to retrieve from
2115 The argument can be either an arrayref (interpreted as a list
2116 of field names, will be joined by commas and quoted), or a
2117 plain scalar (literal SQL, not quoted).
2118 Please observe that this API is not as flexible as that of
2119 the first argument C<$source>, for backwards compatibility reasons.
2123 Optional argument to specify the WHERE part of the query.
2124 The argument is most often a hashref, but can also be
2125 an arrayref or plain scalar --
2126 see section L<WHERE clause|/"WHERE CLAUSES"> for details.
2130 Optional argument to specify the ORDER BY part of the query.
2131 The argument can be a scalar, a hashref or an arrayref
2132 -- see section L<ORDER BY clause|/"ORDER BY CLAUSES">
2138 =head2 delete($table, \%where, \%options)
2140 This takes a table name and optional hashref L<WHERE clause|/WHERE CLAUSES>.
2141 It returns an SQL DELETE statement and list of bind values.
2143 The optional C<\%options> hash reference may contain additional
2144 options to generate the delete SQL. Currently supported options
2151 See the C<returning> option to
2152 L<insert|/insert($table, \@values || \%fieldvals, \%options)>.
2156 =head2 where(\%where, $order)
2158 This is used to generate just the WHERE clause. For example,
2159 if you have an arbitrary data structure and know what the
2160 rest of your SQL is going to look like, but want an easy way
2161 to produce a WHERE clause, use this. It returns an SQL WHERE
2162 clause and list of bind values.
2165 =head2 values(\%data)
2167 This just returns the values from the hash C<%data>, in the same
2168 order that would be returned from any of the other above queries.
2169 Using this allows you to markedly speed up your queries if you
2170 are affecting lots of rows. See below under the L</"PERFORMANCE"> section.
2172 =head2 generate($any, 'number', $of, \@data, $struct, \%types)
2174 Warning: This is an experimental method and subject to change.
2176 This returns arbitrarily generated SQL. It's a really basic shortcut.
2177 It will return two different things, depending on return context:
2179 my($stmt, @bind) = $sql->generate('create table', \$table, \@fields);
2180 my $stmt_and_val = $sql->generate('create table', \$table, \@fields);
2182 These would return the following:
2184 # First calling form
2185 $stmt = "CREATE TABLE test (?, ?)";
2186 @bind = (field1, field2);
2188 # Second calling form
2189 $stmt_and_val = "CREATE TABLE test (field1, field2)";
2191 Depending on what you're trying to do, it's up to you to choose the correct
2192 format. In this example, the second form is what you would want.
2196 $sql->generate('alter session', { nls_date_format => 'MM/YY' });
2200 ALTER SESSION SET nls_date_format = 'MM/YY'
2202 You get the idea. Strings get their case twiddled, but everything
2203 else remains verbatim.
2205 =head1 EXPORTABLE FUNCTIONS
2207 =head2 is_plain_value
2209 Determines if the supplied argument is a plain value as understood by this
2214 =item * The value is C<undef>
2216 =item * The value is a non-reference
2218 =item * The value is an object with stringification overloading
2220 =item * The value is of the form C<< { -value => $anything } >>
2224 On failure returns C<undef>, on success returns a B<scalar> reference
2225 to the original supplied argument.
2231 The stringification overloading detection is rather advanced: it takes
2232 into consideration not only the presence of a C<""> overload, but if that
2233 fails also checks for enabled
2234 L<autogenerated versions of C<"">|overload/Magic Autogeneration>, based
2235 on either C<0+> or C<bool>.
2237 Unfortunately testing in the field indicates that this
2238 detection B<< may tickle a latent bug in perl versions before 5.018 >>,
2239 but only when very large numbers of stringifying objects are involved.
2240 At the time of writing ( Sep 2014 ) there is no clear explanation of
2241 the direct cause, nor is there a manageably small test case that reliably
2242 reproduces the problem.
2244 If you encounter any of the following exceptions in B<random places within
2245 your application stack> - this module may be to blame:
2247 Operation "ne": no method found,
2248 left argument in overloaded package <something>,
2249 right argument in overloaded package <something>
2253 Stub found while resolving method "???" overloading """" in package <something>
2255 If you fall victim to the above - please attempt to reduce the problem
2256 to something that could be sent to the L<SQL::Abstract developers
2257 |DBIx::Class/GETTING HELP/SUPPORT>
2258 (either publicly or privately). As a workaround in the meantime you can
2259 set C<$ENV{SQLA_ISVALUE_IGNORE_AUTOGENERATED_STRINGIFICATION}> to a true
2260 value, which will most likely eliminate your problem (at the expense of
2261 not being able to properly detect exotic forms of stringification).
2263 This notice and environment variable will be removed in a future version,
2264 as soon as the underlying problem is found and a reliable workaround is
2269 =head2 is_literal_value
2271 Determines if the supplied argument is a literal value as understood by this
2276 =item * C<\$sql_string>
2278 =item * C<\[ $sql_string, @bind_values ]>
2282 On failure returns C<undef>, on success returns an B<array> reference
2283 containing the unpacked version of the supplied literal SQL and bind values.
2285 =head1 WHERE CLAUSES
2289 This module uses a variation on the idea from L<DBIx::Abstract>. It
2290 is B<NOT>, repeat I<not> 100% compatible. B<The main logic of this
2291 module is that things in arrays are OR'ed, and things in hashes
2294 The easiest way to explain is to show lots of examples. After
2295 each C<%where> hash shown, it is assumed you used:
2297 my($stmt, @bind) = $sql->where(\%where);
2299 However, note that the C<%where> hash can be used directly in any
2300 of the other functions as well, as described above.
2302 =head2 Key-value pairs
2304 So, let's get started. To begin, a simple hash:
2308 status => 'completed'
2311 Is converted to SQL C<key = val> statements:
2313 $stmt = "WHERE user = ? AND status = ?";
2314 @bind = ('nwiger', 'completed');
2316 One common thing I end up doing is having a list of values that
2317 a field can be in. To do this, simply specify a list inside of
2322 status => ['assigned', 'in-progress', 'pending'];
2325 This simple code will create the following:
2327 $stmt = "WHERE user = ? AND ( status = ? OR status = ? OR status = ? )";
2328 @bind = ('nwiger', 'assigned', 'in-progress', 'pending');
2330 A field associated to an empty arrayref will be considered a
2331 logical false and will generate 0=1.
2333 =head2 Tests for NULL values
2335 If the value part is C<undef> then this is converted to SQL <IS NULL>
2344 $stmt = "WHERE user = ? AND status IS NULL";
2347 To test if a column IS NOT NULL:
2351 status => { '!=', undef },
2354 =head2 Specific comparison operators
2356 If you want to specify a different type of operator for your comparison,
2357 you can use a hashref for a given column:
2361 status => { '!=', 'completed' }
2364 Which would generate:
2366 $stmt = "WHERE user = ? AND status != ?";
2367 @bind = ('nwiger', 'completed');
2369 To test against multiple values, just enclose the values in an arrayref:
2371 status => { '=', ['assigned', 'in-progress', 'pending'] };
2373 Which would give you:
2375 "WHERE status = ? OR status = ? OR status = ?"
2378 The hashref can also contain multiple pairs, in which case it is expanded
2379 into an C<AND> of its elements:
2383 status => { '!=', 'completed', -not_like => 'pending%' }
2386 # Or more dynamically, like from a form
2387 $where{user} = 'nwiger';
2388 $where{status}{'!='} = 'completed';
2389 $where{status}{'-not_like'} = 'pending%';
2391 # Both generate this
2392 $stmt = "WHERE user = ? AND status != ? AND status NOT LIKE ?";
2393 @bind = ('nwiger', 'completed', 'pending%');
2396 To get an OR instead, you can combine it with the arrayref idea:
2400 priority => [ { '=', 2 }, { '>', 5 } ]
2403 Which would generate:
2405 $stmt = "WHERE ( priority = ? OR priority > ? ) AND user = ?";
2406 @bind = ('2', '5', 'nwiger');
2408 If you want to include literal SQL (with or without bind values), just use a
2409 scalar reference or reference to an arrayref as the value:
2412 date_entered => { '>' => \["to_date(?, 'MM/DD/YYYY')", "11/26/2008"] },
2413 date_expires => { '<' => \"now()" }
2416 Which would generate:
2418 $stmt = "WHERE date_entered > to_date(?, 'MM/DD/YYYY') AND date_expires < now()";
2419 @bind = ('11/26/2008');
2422 =head2 Logic and nesting operators
2424 In the example above,
2425 there is a subtle trap if you want to say something like
2426 this (notice the C<AND>):
2428 WHERE priority != ? AND priority != ?
2430 Because, in Perl you I<can't> do this:
2432 priority => { '!=' => 2, '!=' => 1 }
2434 As the second C<!=> key will obliterate the first. The solution
2435 is to use the special C<-modifier> form inside an arrayref:
2437 priority => [ -and => {'!=', 2},
2441 Normally, these would be joined by C<OR>, but the modifier tells it
2442 to use C<AND> instead. (Hint: You can use this in conjunction with the
2443 C<logic> option to C<new()> in order to change the way your queries
2444 work by default.) B<Important:> Note that the C<-modifier> goes
2445 B<INSIDE> the arrayref, as an extra first element. This will
2446 B<NOT> do what you think it might:
2448 priority => -and => [{'!=', 2}, {'!=', 1}] # WRONG!
2450 Here is a quick list of equivalencies, since there is some overlap:
2453 status => {'!=', 'completed', 'not like', 'pending%' }
2454 status => [ -and => {'!=', 'completed'}, {'not like', 'pending%'}]
2457 status => {'=', ['assigned', 'in-progress']}
2458 status => [ -or => {'=', 'assigned'}, {'=', 'in-progress'}]
2459 status => [ {'=', 'assigned'}, {'=', 'in-progress'} ]
2463 =head2 Special operators: IN, BETWEEN, etc.
2465 You can also use the hashref format to compare a list of fields using the
2466 C<IN> comparison operator, by specifying the list as an arrayref:
2469 status => 'completed',
2470 reportid => { -in => [567, 2335, 2] }
2473 Which would generate:
2475 $stmt = "WHERE status = ? AND reportid IN (?,?,?)";
2476 @bind = ('completed', '567', '2335', '2');
2478 The reverse operator C<-not_in> generates SQL C<NOT IN> and is used in
2481 If the argument to C<-in> is an empty array, 'sqlfalse' is generated
2482 (by default: C<1=0>). Similarly, C<< -not_in => [] >> generates
2483 'sqltrue' (by default: C<1=1>).
2485 In addition to the array you can supply a chunk of literal sql or
2486 literal sql with bind:
2489 customer => { -in => \[
2490 'SELECT cust_id FROM cust WHERE balance > ?',
2493 status => { -in => \'SELECT status_codes FROM states' },
2499 customer IN ( SELECT cust_id FROM cust WHERE balance > ? )
2500 AND status IN ( SELECT status_codes FROM states )
2504 Finally, if the argument to C<-in> is not a reference, it will be
2505 treated as a single-element array.
2507 Another pair of operators is C<-between> and C<-not_between>,
2508 used with an arrayref of two values:
2512 completion_date => {
2513 -not_between => ['2002-10-01', '2003-02-06']
2519 WHERE user = ? AND completion_date NOT BETWEEN ( ? AND ? )
2521 Just like with C<-in> all plausible combinations of literal SQL
2525 start0 => { -between => [ 1, 2 ] },
2526 start1 => { -between => \["? AND ?", 1, 2] },
2527 start2 => { -between => \"lower(x) AND upper(y)" },
2528 start3 => { -between => [
2530 \["upper(?)", 'stuff' ],
2537 ( start0 BETWEEN ? AND ? )
2538 AND ( start1 BETWEEN ? AND ? )
2539 AND ( start2 BETWEEN lower(x) AND upper(y) )
2540 AND ( start3 BETWEEN lower(x) AND upper(?) )
2542 @bind = (1, 2, 1, 2, 'stuff');
2545 These are the two builtin "special operators"; but the
2546 list can be expanded: see section L</"SPECIAL OPERATORS"> below.
2548 =head2 Unary operators: bool
2550 If you wish to test against boolean columns or functions within your
2551 database you can use the C<-bool> and C<-not_bool> operators. For
2552 example to test the column C<is_user> being true and the column
2553 C<is_enabled> being false you would use:-
2557 -not_bool => 'is_enabled',
2562 WHERE is_user AND NOT is_enabled
2564 If a more complex combination is required, testing more conditions,
2565 then you should use the and/or operators:-
2570 -not_bool => { two=> { -rlike => 'bar' } },
2571 -not_bool => { three => [ { '=', 2 }, { '>', 5 } ] },
2582 (NOT ( three = ? OR three > ? ))
2585 =head2 Nested conditions, -and/-or prefixes
2587 So far, we've seen how multiple conditions are joined with a top-level
2588 C<AND>. We can change this by putting the different conditions we want in
2589 hashes and then putting those hashes in an array. For example:
2594 status => { -like => ['pending%', 'dispatched'] },
2598 status => 'unassigned',
2602 This data structure would create the following:
2604 $stmt = "WHERE ( user = ? AND ( status LIKE ? OR status LIKE ? ) )
2605 OR ( user = ? AND status = ? ) )";
2606 @bind = ('nwiger', 'pending', 'dispatched', 'robot', 'unassigned');
2609 Clauses in hashrefs or arrayrefs can be prefixed with an C<-and> or C<-or>
2610 to change the logic inside:
2616 -and => [ workhrs => {'>', 20}, geo => 'ASIA' ],
2617 -or => { workhrs => {'<', 50}, geo => 'EURO' },
2624 $stmt = "WHERE ( user = ?
2625 AND ( ( workhrs > ? AND geo = ? )
2626 OR ( workhrs < ? OR geo = ? ) ) )";
2627 @bind = ('nwiger', '20', 'ASIA', '50', 'EURO');
2629 =head3 Algebraic inconsistency, for historical reasons
2631 C<Important note>: when connecting several conditions, the C<-and->|C<-or>
2632 operator goes C<outside> of the nested structure; whereas when connecting
2633 several constraints on one column, the C<-and> operator goes
2634 C<inside> the arrayref. Here is an example combining both features:
2637 -and => [a => 1, b => 2],
2638 -or => [c => 3, d => 4],
2639 e => [-and => {-like => 'foo%'}, {-like => '%bar'} ]
2644 WHERE ( ( ( a = ? AND b = ? )
2645 OR ( c = ? OR d = ? )
2646 OR ( e LIKE ? AND e LIKE ? ) ) )
2648 This difference in syntax is unfortunate but must be preserved for
2649 historical reasons. So be careful: the two examples below would
2650 seem algebraically equivalent, but they are not
2653 { -like => 'foo%' },
2654 { -like => '%bar' },
2656 # yields: WHERE ( ( col LIKE ? AND col LIKE ? ) )
2659 { col => { -like => 'foo%' } },
2660 { col => { -like => '%bar' } },
2662 # yields: WHERE ( ( col LIKE ? OR col LIKE ? ) )
2665 =head2 Literal SQL and value type operators
2667 The basic premise of SQL::Abstract is that in WHERE specifications the "left
2668 side" is a column name and the "right side" is a value (normally rendered as
2669 a placeholder). This holds true for both hashrefs and arrayref pairs as you
2670 see in the L</WHERE CLAUSES> examples above. Sometimes it is necessary to
2671 alter this behavior. There are several ways of doing so.
2675 This is a virtual operator that signals the string to its right side is an
2676 identifier (a column name) and not a value. For example to compare two
2677 columns you would write:
2680 priority => { '<', 2 },
2681 requestor => { -ident => 'submitter' },
2686 $stmt = "WHERE priority < ? AND requestor = submitter";
2689 If you are maintaining legacy code you may see a different construct as
2690 described in L</Deprecated usage of Literal SQL>, please use C<-ident> in new
2695 This is a virtual operator that signals that the construct to its right side
2696 is a value to be passed to DBI. This is for example necessary when you want
2697 to write a where clause against an array (for RDBMS that support such
2698 datatypes). For example:
2701 array => { -value => [1, 2, 3] }
2706 $stmt = 'WHERE array = ?';
2707 @bind = ([1, 2, 3]);
2709 Note that if you were to simply say:
2715 the result would probably not be what you wanted:
2717 $stmt = 'WHERE array = ? OR array = ? OR array = ?';
2722 Finally, sometimes only literal SQL will do. To include a random snippet
2723 of SQL verbatim, you specify it as a scalar reference. Consider this only
2724 as a last resort. Usually there is a better way. For example:
2727 priority => { '<', 2 },
2728 requestor => { -in => \'(SELECT name FROM hitmen)' },
2733 $stmt = "WHERE priority < ? AND requestor IN (SELECT name FROM hitmen)"
2736 Note that in this example, you only get one bind parameter back, since
2737 the verbatim SQL is passed as part of the statement.
2741 Never use untrusted input as a literal SQL argument - this is a massive
2742 security risk (there is no way to check literal snippets for SQL
2743 injections and other nastyness). If you need to deal with untrusted input
2744 use literal SQL with placeholders as described next.
2746 =head3 Literal SQL with placeholders and bind values (subqueries)
2748 If the literal SQL to be inserted has placeholders and bind values,
2749 use a reference to an arrayref (yes this is a double reference --
2750 not so common, but perfectly legal Perl). For example, to find a date
2751 in Postgres you can use something like this:
2754 date_column => \[ "= date '2008-09-30' - ?::integer", 10 ]
2759 $stmt = "WHERE ( date_column = date '2008-09-30' - ?::integer )"
2762 Note that you must pass the bind values in the same format as they are returned
2763 by L<where|/where(\%where, $order)>. This means that if you set L</bindtype>
2764 to C<columns>, you must provide the bind values in the
2765 C<< [ column_meta => value ] >> format, where C<column_meta> is an opaque
2766 scalar value; most commonly the column name, but you can use any scalar value
2767 (including references and blessed references), L<SQL::Abstract> will simply
2768 pass it through intact. So if C<bindtype> is set to C<columns> the above
2769 example will look like:
2772 date_column => \[ "= date '2008-09-30' - ?::integer", [ {} => 10 ] ]
2775 Literal SQL is especially useful for nesting parenthesized clauses in the
2776 main SQL query. Here is a first example:
2778 my ($sub_stmt, @sub_bind) = ("SELECT c1 FROM t1 WHERE c2 < ? AND c3 LIKE ?",
2782 bar => \["IN ($sub_stmt)" => @sub_bind],
2787 $stmt = "WHERE (foo = ? AND bar IN (SELECT c1 FROM t1
2788 WHERE c2 < ? AND c3 LIKE ?))";
2789 @bind = (1234, 100, "foo%");
2791 Other subquery operators, like for example C<"E<gt> ALL"> or C<"NOT IN">,
2792 are expressed in the same way. Of course the C<$sub_stmt> and
2793 its associated bind values can be generated through a former call
2796 my ($sub_stmt, @sub_bind)
2797 = $sql->select("t1", "c1", {c2 => {"<" => 100},
2798 c3 => {-like => "foo%"}});
2801 bar => \["> ALL ($sub_stmt)" => @sub_bind],
2804 In the examples above, the subquery was used as an operator on a column;
2805 but the same principle also applies for a clause within the main C<%where>
2806 hash, like an EXISTS subquery:
2808 my ($sub_stmt, @sub_bind)
2809 = $sql->select("t1", "*", {c1 => 1, c2 => \"> t0.c0"});
2810 my %where = ( -and => [
2812 \["EXISTS ($sub_stmt)" => @sub_bind],
2817 $stmt = "WHERE (foo = ? AND EXISTS (SELECT * FROM t1
2818 WHERE c1 = ? AND c2 > t0.c0))";
2822 Observe that the condition on C<c2> in the subquery refers to
2823 column C<t0.c0> of the main query: this is I<not> a bind
2824 value, so we have to express it through a scalar ref.
2825 Writing C<< c2 => {">" => "t0.c0"} >> would have generated
2826 C<< c2 > ? >> with bind value C<"t0.c0"> ... not exactly
2827 what we wanted here.
2829 Finally, here is an example where a subquery is used
2830 for expressing unary negation:
2832 my ($sub_stmt, @sub_bind)
2833 = $sql->where({age => [{"<" => 10}, {">" => 20}]});
2834 $sub_stmt =~ s/^ where //i; # don't want "WHERE" in the subclause
2836 lname => {like => '%son%'},
2837 \["NOT ($sub_stmt)" => @sub_bind],
2842 $stmt = "lname LIKE ? AND NOT ( age < ? OR age > ? )"
2843 @bind = ('%son%', 10, 20)
2845 =head3 Deprecated usage of Literal SQL
2847 Below are some examples of archaic use of literal SQL. It is shown only as
2848 reference for those who deal with legacy code. Each example has a much
2849 better, cleaner and safer alternative that users should opt for in new code.
2855 my %where = ( requestor => \'IS NOT NULL' )
2857 $stmt = "WHERE requestor IS NOT NULL"
2859 This used to be the way of generating NULL comparisons, before the handling
2860 of C<undef> got formalized. For new code please use the superior syntax as
2861 described in L</Tests for NULL values>.
2865 my %where = ( requestor => \'= submitter' )
2867 $stmt = "WHERE requestor = submitter"
2869 This used to be the only way to compare columns. Use the superior L</-ident>
2870 method for all new code. For example an identifier declared in such a way
2871 will be properly quoted if L</quote_char> is properly set, while the legacy
2872 form will remain as supplied.
2876 my %where = ( is_ready => \"", completed => { '>', '2012-12-21' } )
2878 $stmt = "WHERE completed > ? AND is_ready"
2879 @bind = ('2012-12-21')
2881 Using an empty string literal used to be the only way to express a boolean.
2882 For all new code please use the much more readable
2883 L<-bool|/Unary operators: bool> operator.
2889 These pages could go on for a while, since the nesting of the data
2890 structures this module can handle are pretty much unlimited (the
2891 module implements the C<WHERE> expansion as a recursive function
2892 internally). Your best bet is to "play around" with the module a
2893 little to see how the data structures behave, and choose the best
2894 format for your data based on that.
2896 And of course, all the values above will probably be replaced with
2897 variables gotten from forms or the command line. After all, if you
2898 knew everything ahead of time, you wouldn't have to worry about
2899 dynamically-generating SQL and could just hardwire it into your
2902 =head1 ORDER BY CLAUSES
2904 Some functions take an order by clause. This can either be a scalar (just a
2905 column name), a hashref of C<< { -desc => 'col' } >> or C<< { -asc => 'col' }
2906 >>, a scalarref, an arrayref-ref, or an arrayref of any of the previous
2909 Given | Will Generate
2910 ---------------------------------------------------------------
2912 'colA' | ORDER BY colA
2914 [qw/colA colB/] | ORDER BY colA, colB
2916 {-asc => 'colA'} | ORDER BY colA ASC
2918 {-desc => 'colB'} | ORDER BY colB DESC
2920 ['colA', {-asc => 'colB'}] | ORDER BY colA, colB ASC
2922 { -asc => [qw/colA colB/] } | ORDER BY colA ASC, colB ASC
2924 \'colA DESC' | ORDER BY colA DESC
2926 \[ 'FUNC(colA, ?)', $x ] | ORDER BY FUNC(colA, ?)
2927 | /* ...with $x bound to ? */
2930 { -asc => 'colA' }, | colA ASC,
2931 { -desc => [qw/colB/] }, | colB DESC,
2932 { -asc => [qw/colC colD/] },| colC ASC, colD ASC,
2933 \'colE DESC', | colE DESC,
2934 \[ 'FUNC(colF, ?)', $x ], | FUNC(colF, ?)
2935 ] | /* ...with $x bound to ? */
2936 ===============================================================
2940 =head1 SPECIAL OPERATORS
2942 my $sqlmaker = SQL::Abstract->new(special_ops => [
2946 my ($self, $field, $op, $arg) = @_;
2952 handler => 'method_name',
2956 A "special operator" is a SQL syntactic clause that can be
2957 applied to a field, instead of a usual binary operator.
2960 WHERE field IN (?, ?, ?)
2961 WHERE field BETWEEN ? AND ?
2962 WHERE MATCH(field) AGAINST (?, ?)
2964 Special operators IN and BETWEEN are fairly standard and therefore
2965 are builtin within C<SQL::Abstract> (as the overridable methods
2966 C<_where_field_IN> and C<_where_field_BETWEEN>). For other operators,
2967 like the MATCH .. AGAINST example above which is specific to MySQL,
2968 you can write your own operator handlers - supply a C<special_ops>
2969 argument to the C<new> method. That argument takes an arrayref of
2970 operator definitions; each operator definition is a hashref with two
2977 the regular expression to match the operator
2981 Either a coderef or a plain scalar method name. In both cases
2982 the expected return is C<< ($sql, @bind) >>.
2984 When supplied with a method name, it is simply called on the
2985 L<SQL::Abstract> object as:
2987 $self->$method_name($field, $op, $arg)
2991 $field is the LHS of the operator
2992 $op is the part that matched the handler regex
2995 When supplied with a coderef, it is called as:
2997 $coderef->($self, $field, $op, $arg)
3002 For example, here is an implementation
3003 of the MATCH .. AGAINST syntax for MySQL
3005 my $sqlmaker = SQL::Abstract->new(special_ops => [
3007 # special op for MySql MATCH (field) AGAINST(word1, word2, ...)
3008 {regex => qr/^match$/i,
3010 my ($self, $field, $op, $arg) = @_;
3011 $arg = [$arg] if not ref $arg;
3012 my $label = $self->_quote($field);
3013 my ($placeholder) = $self->_convert('?');
3014 my $placeholders = join ", ", (($placeholder) x @$arg);
3015 my $sql = $self->_sqlcase('match') . " ($label) "
3016 . $self->_sqlcase('against') . " ($placeholders) ";
3017 my @bind = $self->_bindtype($field, @$arg);
3018 return ($sql, @bind);
3025 =head1 UNARY OPERATORS
3027 my $sqlmaker = SQL::Abstract->new(unary_ops => [
3031 my ($self, $op, $arg) = @_;
3037 handler => 'method_name',
3041 A "unary operator" is a SQL syntactic clause that can be
3042 applied to a field - the operator goes before the field
3044 You can write your own operator handlers - supply a C<unary_ops>
3045 argument to the C<new> method. That argument takes an arrayref of
3046 operator definitions; each operator definition is a hashref with two
3053 the regular expression to match the operator
3057 Either a coderef or a plain scalar method name. In both cases
3058 the expected return is C<< $sql >>.
3060 When supplied with a method name, it is simply called on the
3061 L<SQL::Abstract> object as:
3063 $self->$method_name($op, $arg)
3067 $op is the part that matched the handler regex
3068 $arg is the RHS or argument of the operator
3070 When supplied with a coderef, it is called as:
3072 $coderef->($self, $op, $arg)
3080 Thanks to some benchmarking by Mark Stosberg, it turns out that
3081 this module is many orders of magnitude faster than using C<DBIx::Abstract>.
3082 I must admit this wasn't an intentional design issue, but it's a
3083 byproduct of the fact that you get to control your C<DBI> handles
3086 To maximize performance, use a code snippet like the following:
3088 # prepare a statement handle using the first row
3089 # and then reuse it for the rest of the rows
3091 for my $href (@array_of_hashrefs) {
3092 $stmt ||= $sql->insert('table', $href);
3093 $sth ||= $dbh->prepare($stmt);
3094 $sth->execute($sql->values($href));
3097 The reason this works is because the keys in your C<$href> are sorted
3098 internally by B<SQL::Abstract>. Thus, as long as your data retains
3099 the same structure, you only have to generate the SQL the first time
3100 around. On subsequent queries, simply use the C<values> function provided
3101 by this module to return your values in the correct order.
3103 However this depends on the values having the same type - if, for
3104 example, the values of a where clause may either have values
3105 (resulting in sql of the form C<column = ?> with a single bind
3106 value), or alternatively the values might be C<undef> (resulting in
3107 sql of the form C<column IS NULL> with no bind value) then the
3108 caching technique suggested will not work.
3112 If you use my C<CGI::FormBuilder> module at all, you'll hopefully
3113 really like this part (I do, at least). Building up a complex query
3114 can be as simple as the following:
3121 use CGI::FormBuilder;
3124 my $form = CGI::FormBuilder->new(...);
3125 my $sql = SQL::Abstract->new;
3127 if ($form->submitted) {
3128 my $field = $form->field;
3129 my $id = delete $field->{id};
3130 my($stmt, @bind) = $sql->update('table', $field, {id => $id});
3133 Of course, you would still have to connect using C<DBI> to run the
3134 query, but the point is that if you make your form look like your
3135 table, the actual query script can be extremely simplistic.
3137 If you're B<REALLY> lazy (I am), check out C<HTML::QuickTable> for
3138 a fast interface to returning and formatting data. I frequently
3139 use these three modules together to write complex database query
3140 apps in under 50 lines.
3142 =head1 HOW TO CONTRIBUTE
3144 Contributions are always welcome, in all usable forms (we especially
3145 welcome documentation improvements). The delivery methods include git-
3146 or unified-diff formatted patches, GitHub pull requests, or plain bug
3147 reports either via RT or the Mailing list. Contributors are generally
3148 granted full access to the official repository after their first several
3149 patches pass successful review.
3151 This project is maintained in a git repository. The code and related tools are
3152 accessible at the following locations:
3156 =item * Official repo: L<git://git.shadowcat.co.uk/dbsrgits/SQL-Abstract.git>
3158 =item * Official gitweb: L<http://git.shadowcat.co.uk/gitweb/gitweb.cgi?p=dbsrgits/SQL-Abstract.git>
3160 =item * GitHub mirror: L<https://github.com/dbsrgits/sql-abstract>
3162 =item * Authorized committers: L<ssh://dbsrgits@git.shadowcat.co.uk/SQL-Abstract.git>
3168 Version 1.50 was a major internal refactoring of C<SQL::Abstract>.
3169 Great care has been taken to preserve the I<published> behavior
3170 documented in previous versions in the 1.* family; however,
3171 some features that were previously undocumented, or behaved
3172 differently from the documentation, had to be changed in order
3173 to clarify the semantics. Hence, client code that was relying
3174 on some dark areas of C<SQL::Abstract> v1.*
3175 B<might behave differently> in v1.50.
3177 The main changes are:
3183 support for literal SQL through the C<< \ [ $sql, @bind ] >> syntax.
3187 support for the { operator => \"..." } construct (to embed literal SQL)
3191 support for the { operator => \["...", @bind] } construct (to embed literal SQL with bind values)
3195 optional support for L<array datatypes|/"Inserting and Updating Arrays">
3199 defensive programming: check arguments
3203 fixed bug with global logic, which was previously implemented
3204 through global variables yielding side-effects. Prior versions would
3205 interpret C<< [ {cond1, cond2}, [cond3, cond4] ] >>
3206 as C<< "(cond1 AND cond2) OR (cond3 AND cond4)" >>.
3207 Now this is interpreted
3208 as C<< "(cond1 AND cond2) OR (cond3 OR cond4)" >>.
3213 fixed semantics of _bindtype on array args
3217 dropped the C<_anoncopy> of the %where tree. No longer necessary,
3218 we just avoid shifting arrays within that tree.
3222 dropped the C<_modlogic> function
3226 =head1 ACKNOWLEDGEMENTS
3228 There are a number of individuals that have really helped out with
3229 this module. Unfortunately, most of them submitted bugs via CPAN
3230 so I have no idea who they are! But the people I do know are:
3232 Ash Berlin (order_by hash term support)
3233 Matt Trout (DBIx::Class support)
3234 Mark Stosberg (benchmarking)
3235 Chas Owens (initial "IN" operator support)
3236 Philip Collins (per-field SQL functions)
3237 Eric Kolve (hashref "AND" support)
3238 Mike Fragassi (enhancements to "BETWEEN" and "LIKE")
3239 Dan Kubb (support for "quote_char" and "name_sep")
3240 Guillermo Roditi (patch to cleanup "IN" and "BETWEEN", fix and tests for _order_by)
3241 Laurent Dami (internal refactoring, extensible list of special operators, literal SQL)
3242 Norbert Buchmuller (support for literal SQL in hashpair, misc. fixes & tests)
3243 Peter Rabbitson (rewrite of SQLA::Test, misc. fixes & tests)
3244 Oliver Charles (support for "RETURNING" after "INSERT")
3250 L<DBIx::Class>, L<DBIx::Abstract>, L<CGI::FormBuilder>, L<HTML::QuickTable>.
3254 Copyright (c) 2001-2007 Nathan Wiger <nwiger@cpan.org>. All Rights Reserved.
3256 This module is actively maintained by Matt Trout <mst@shadowcatsystems.co.uk>
3258 For support, your best bet is to try the C<DBIx::Class> users mailing list.
3259 While not an official support venue, C<DBIx::Class> makes heavy use of
3260 C<SQL::Abstract>, and as such list members there are very familiar with
3261 how to create queries.
3265 This module is free software; you may copy this under the same
3266 terms as perl itself (either the GNU General Public License or
3267 the Artistic License)