1 package SQL::Abstract; # see doc at end of file
5 use Module::Runtime ();
10 use Exporter 'import';
11 our @EXPORT_OK = qw(is_plain_value is_literal_value);
21 *SQL::Abstract::_ENV_::DETECT_AUTOGENERATED_STRINGIFICATION = $ENV{SQLA_ISVALUE_IGNORE_AUTOGENERATED_STRINGIFICATION}
27 #======================================================================
29 #======================================================================
31 our $VERSION = '1.86';
33 # This would confuse some packagers
34 $VERSION = eval $VERSION if $VERSION =~ /_/; # numify for warning-free dev releases
38 # special operators (-in, -between). May be extended/overridden by user.
39 # See section WHERE: BUILTIN SPECIAL OPERATORS below for implementation
40 my @BUILTIN_SPECIAL_OPS = (
41 {regex => qr/^ (?: not \s )? between $/ix, handler => sub { die "NOPE" }},
42 {regex => qr/^ is (?: \s+ not )? $/ix, handler => sub { die "NOPE" }},
43 {regex => qr/^ (?: not \s )? in $/ix, handler => sub { die "NOPE" }},
44 {regex => qr/^ ident $/ix, handler => sub { die "NOPE" }},
45 {regex => qr/^ value $/ix, handler => sub { die "NOPE" }},
48 #======================================================================
49 # DEBUGGING AND ERROR REPORTING
50 #======================================================================
53 return unless $_[0]->{debug}; shift; # a little faster
54 my $func = (caller(1))[3];
55 warn "[$func] ", @_, "\n";
59 my($func) = (caller(1))[3];
60 Carp::carp "[$func] Warning: ", @_;
64 my($func) = (caller(1))[3];
65 Carp::croak "[$func] Fatal: ", @_;
68 sub is_literal_value ($) {
69 ref $_[0] eq 'SCALAR' ? [ ${$_[0]} ]
70 : ( ref $_[0] eq 'REF' and ref ${$_[0]} eq 'ARRAY' ) ? [ @${ $_[0] } ]
74 sub is_undef_value ($) {
78 and exists $_[0]->{-value}
79 and not defined $_[0]->{-value}
83 # FIXME XSify - this can be done so much more efficiently
84 sub is_plain_value ($) {
86 ! length ref $_[0] ? \($_[0])
88 ref $_[0] eq 'HASH' and keys %{$_[0]} == 1
90 exists $_[0]->{-value}
91 ) ? \($_[0]->{-value})
93 # reuse @_ for even moar speedz
94 defined ( $_[1] = Scalar::Util::blessed $_[0] )
96 # deliberately not using Devel::OverloadInfo - the checks we are
97 # intersted in are much more limited than the fullblown thing, and
98 # this is a very hot piece of code
100 # simply using ->can('(""') can leave behind stub methods that
101 # break actually using the overload later (see L<perldiag/Stub
102 # found while resolving method "%s" overloading "%s" in package
103 # "%s"> and the source of overload::mycan())
105 # either has stringification which DBI SHOULD prefer out of the box
106 grep { *{ (qq[${_}::(""]) }{CODE} } @{ $_[2] = mro::get_linear_isa( $_[1] ) }
108 # has nummification or boolification, AND fallback is *not* disabled
110 SQL::Abstract::_ENV_::DETECT_AUTOGENERATED_STRINGIFICATION
113 grep { *{"${_}::(0+"}{CODE} } @{$_[2]}
115 grep { *{"${_}::(bool"}{CODE} } @{$_[2]}
119 # no fallback specified at all
120 ! ( ($_[3]) = grep { *{"${_}::()"}{CODE} } @{$_[2]} )
122 # fallback explicitly undef
123 ! defined ${"$_[3]::()"}
136 #======================================================================
138 #======================================================================
142 my $class = ref($self) || $self;
143 my %opt = (ref $_[0] eq 'HASH') ? %{$_[0]} : @_;
145 # choose our case by keeping an option around
146 delete $opt{case} if $opt{case} && $opt{case} ne 'lower';
148 # default logic for interpreting arrayrefs
149 $opt{logic} = $opt{logic} ? uc $opt{logic} : 'OR';
151 # how to return bind vars
152 $opt{bindtype} ||= 'normal';
154 # default comparison is "=", but can be overridden
157 # try to recognize which are the 'equality' and 'inequality' ops
158 # (temporary quickfix (in 2007), should go through a more seasoned API)
159 $opt{equality_op} = qr/^( \Q$opt{cmp}\E | \= )$/ix;
160 $opt{inequality_op} = qr/^( != | <> )$/ix;
162 $opt{like_op} = qr/^ (is_)?r?like $/xi;
163 $opt{not_like_op} = qr/^ (is_)?not_r?like $/xi;
166 $opt{sqltrue} ||= '1=1';
167 $opt{sqlfalse} ||= '0=1';
170 $opt{special_ops} ||= [];
172 if ($class->isa('DBIx::Class::SQLMaker')) {
173 $opt{warn_once_on_nest} = 1;
174 $opt{disable_old_special_ops} = 1;
178 $opt{unary_ops} ||= [];
180 # rudimentary sanity-check for user supplied bits treated as functions/operators
181 # If a purported function matches this regular expression, an exception is thrown.
182 # Literal SQL is *NOT* subject to this check, only functions (and column names
183 # when quoting is not in effect)
186 # need to guard against ()'s in column names too, but this will break tons of
187 # hacks... ideas anyone?
188 $opt{injection_guard} ||= qr/
194 $opt{expand_unary} = {};
197 not => '_expand_not',
198 bool => '_expand_bool',
199 and => '_expand_op_andor',
200 or => '_expand_op_andor',
201 nest => '_expand_nest',
202 bind => '_expand_bind',
204 not_in => '_expand_in',
205 row => '_expand_row',
206 between => '_expand_between',
207 not_between => '_expand_between',
209 (map +($_ => '_expand_op_is'), ('is', 'is_not')),
210 ident => '_expand_ident',
211 value => '_expand_value',
212 func => '_expand_func',
216 'between' => '_expand_between',
217 'not_between' => '_expand_between',
218 'in' => '_expand_in',
219 'not_in' => '_expand_in',
220 'nest' => '_expand_nest',
221 (map +($_ => '_expand_op_andor'), ('and', 'or')),
222 (map +($_ => '_expand_op_is'), ('is', 'is_not')),
223 'ident' => '_expand_ident',
224 'value' => '_expand_value',
228 (map +($_, "_render_$_"), qw(op func bind ident literal row)),
233 (map +($_ => '_render_op_between'), 'between', 'not_between'),
234 (map +($_ => '_render_op_in'), 'in', 'not_in'),
235 (map +($_ => '_render_unop_postfix'),
236 'is_null', 'is_not_null', 'asc', 'desc',
238 (not => '_render_unop_paren'),
239 (map +($_ => '_render_op_andor'), qw(and or)),
240 ',' => '_render_op_multop',
243 if ($opt{lazy_join_sql_parts}) {
244 my $mod = Module::Runtime::use_module('SQL::Abstract::Parts');
245 $opt{join_sql_parts} ||= sub { $mod->new(@_) };
248 $opt{join_sql_parts} ||= sub { join $_[0], @_[1..$#_] };
250 return bless \%opt, $class;
253 sub sqltrue { +{ -literal => [ $_[0]->{sqltrue} ] } }
254 sub sqlfalse { +{ -literal => [ $_[0]->{sqlfalse} ] } }
256 sub _assert_pass_injection_guard {
257 if ($_[1] =~ $_[0]->{injection_guard}) {
258 my $class = ref $_[0];
259 puke "Possible SQL injection attempt '$_[1]'. If this is indeed a part of the "
260 . "desired SQL use literal SQL ( \'...' or \[ '...' ] ) or supply your own "
261 . "{injection_guard} attribute to ${class}->new()"
266 #======================================================================
268 #======================================================================
272 my $table = $self->_table(shift);
273 my $data = shift || return;
278 my ($f_aqt, $v_aqt) = $self->_expand_insert_values($data);
280 my @parts = ([ $self->_sqlcase('insert into').' '.$table ]);
281 push @parts, $self->render_aqt($f_aqt) if $f_aqt;
282 push @parts, [ $self->_sqlcase('values') ], $self->render_aqt($v_aqt);
284 if ($options->{returning}) {
285 push @parts, [ $self->_insert_returning($options) ];
288 my ($sql, @bind) = @{ $self->join_query_parts(' ', @parts) };
289 return wantarray ? ($sql, @bind) : $sql;
292 sub _expand_insert_values {
293 my ($self, $data) = @_;
294 if (is_literal_value($data)) {
295 (undef, $self->expand_expr($data));
297 my ($fields, $values) = (
298 ref($data) eq 'HASH' ?
299 ([ sort keys %$data ], [ @{$data}{sort keys %$data} ])
303 # no names (arrayref) means can't generate bindtype
304 !($fields) && $self->{bindtype} eq 'columns'
305 && belch "can't do 'columns' bindtype when called with arrayref";
309 ? $self->expand_expr({ -row => $fields }, -ident)
314 local our $Cur_Col_Meta = $fields->[$_];
315 $self->_expand_insert_value($values->[$_])
322 # So that subclasses can override INSERT ... RETURNING separately from
323 # UPDATE and DELETE (e.g. DBIx::Class::SQLMaker::Oracle does this)
324 sub _insert_returning { shift->_returning(@_) }
327 my ($self, $options) = @_;
329 my $f = $options->{returning};
331 my ($sql, @bind) = @{ $self->render_aqt(
332 $self->_expand_maybe_list_expr($f, -ident)
334 return ($self->_sqlcase(' returning ').$sql, @bind);
337 sub _expand_insert_value {
340 my $k = our $Cur_Col_Meta;
342 if (ref($v) eq 'ARRAY') {
343 if ($self->{array_datatypes}) {
344 return +{ -bind => [ $k, $v ] };
346 my ($sql, @bind) = @$v;
347 $self->_assert_bindval_matches_bindtype(@bind);
348 return +{ -literal => $v };
350 if (ref($v) eq 'HASH') {
351 if (grep !/^-/, keys %$v) {
352 belch "HASH ref as bind value in insert is not supported";
353 return +{ -bind => [ $k, $v ] };
357 return +{ -bind => [ $k, undef ] };
359 return $self->expand_expr($v);
364 #======================================================================
366 #======================================================================
371 my $table = $self->_table(shift);
372 my $data = shift || return;
376 # first build the 'SET' part of the sql statement
377 puke "Unsupported data type specified to \$sql->update"
378 unless ref $data eq 'HASH';
380 my ($sql, @all_bind) = $self->_update_set_values($data);
381 $sql = $self->_sqlcase('update ') . $table . $self->_sqlcase(' set ')
385 my($where_sql, @where_bind) = $self->where($where);
387 push @all_bind, @where_bind;
390 if ($options->{returning}) {
391 my ($returning_sql, @returning_bind) = $self->_update_returning($options);
392 $sql .= $returning_sql;
393 push @all_bind, @returning_bind;
396 return wantarray ? ($sql, @all_bind) : $sql;
399 sub _update_set_values {
400 my ($self, $data) = @_;
402 return @{ $self->render_aqt(
403 $self->_expand_update_set_values(undef, $data),
407 sub _expand_update_set_values {
408 my ($self, undef, $data) = @_;
409 $self->_expand_maybe_list_expr( [
412 $set = { -bind => $_ } unless defined $set;
413 +{ -op => [ '=', { -ident => $k }, $set ] };
419 ? ($self->{array_datatypes}
420 ? [ $k, +{ -bind => [ $k, $v ] } ]
421 : [ $k, +{ -literal => $v } ])
423 local our $Cur_Col_Meta = $k;
424 [ $k, $self->_expand_expr($v) ]
431 # So that subclasses can override UPDATE ... RETURNING separately from
433 sub _update_returning { shift->_returning(@_) }
437 #======================================================================
439 #======================================================================
444 my $table = $self->_table(shift);
445 my $fields = shift || '*';
449 my ($fields_sql, @bind) = $self->_select_fields($fields);
451 my ($where_sql, @where_bind) = $self->where($where, $order);
452 push @bind, @where_bind;
454 my $sql = join(' ', $self->_sqlcase('select'), $fields_sql,
455 $self->_sqlcase('from'), $table)
458 return wantarray ? ($sql, @bind) : $sql;
462 my ($self, $fields) = @_;
463 return $fields unless ref($fields);
464 return @{ $self->render_aqt(
465 $self->_expand_maybe_list_expr($fields, '-ident')
469 #======================================================================
471 #======================================================================
476 my $table = $self->_table(shift);
480 my($where_sql, @bind) = $self->where($where);
481 my $sql = $self->_sqlcase('delete from ') . $table . $where_sql;
483 if ($options->{returning}) {
484 my ($returning_sql, @returning_bind) = $self->_delete_returning($options);
485 $sql .= $returning_sql;
486 push @bind, @returning_bind;
489 return wantarray ? ($sql, @bind) : $sql;
492 # So that subclasses can override DELETE ... RETURNING separately from
494 sub _delete_returning { shift->_returning(@_) }
498 #======================================================================
500 #======================================================================
504 # Finally, a separate routine just to handle WHERE clauses
506 my ($self, $where, $order) = @_;
508 local $self->{convert_where} = $self->{convert};
511 my ($sql, @bind) = defined($where)
512 ? $self->_recurse_where($where)
514 $sql = (defined $sql and length $sql) ? $self->_sqlcase(' where ') . "( $sql )" : '';
518 my ($order_sql, @order_bind) = $self->_order_by($order);
520 push @bind, @order_bind;
523 return wantarray ? ($sql, @bind) : $sql;
526 { our $Default_Scalar_To = -value }
529 my ($self, $expr, $default_scalar_to) = @_;
530 local our $Default_Scalar_To = $default_scalar_to if $default_scalar_to;
531 $self->_expand_expr($expr);
535 my ($self, $aqt) = @_;
536 my ($k, $v, @rest) = %$aqt;
538 die "Not a node type: $k" unless $k =~ s/^-//;
539 if (my $meth = $self->{render}{$k}) {
540 return $self->$meth($k, $v);
542 die "notreached: $k";
546 my ($self, $expr, $default_scalar_to) = @_;
547 return @{ $self->render_aqt(
548 $self->expand_expr($expr, $default_scalar_to)
553 my ($self, $raw) = @_;
555 return $op if grep $_->{$op}, @{$self}{qw(is_op expand_op render_op)};
556 s/^-(?=.)//, s/\s+/_/g for $op;
561 my ($self, $expr) = @_;
562 our $Expand_Depth ||= 0; local $Expand_Depth = $Expand_Depth + 1;
563 return undef unless defined($expr);
564 if (ref($expr) eq 'HASH') {
565 return undef unless my $kc = keys %$expr;
567 return $self->_expand_op_andor(and => $expr);
569 my ($key, $value) = %$expr;
570 if ($key =~ /^-/ and $key =~ s/ [_\s]? \d+ $//x ) {
571 belch 'Use of [and|or|nest]_N modifiers is deprecated and will be removed in SQLA v2.0. '
572 . "You probably wanted ...-and => [ $key => COND1, $key => COND2 ... ]";
574 return $self->_expand_hashpair($key, $value);
576 if (ref($expr) eq 'ARRAY') {
577 return $self->_expand_op_andor(lc($self->{logic}), $expr);
579 if (my $literal = is_literal_value($expr)) {
580 return +{ -literal => $literal };
582 if (!ref($expr) or Scalar::Util::blessed($expr)) {
583 return $self->_expand_scalar($expr);
588 sub _expand_hashpair {
589 my ($self, $k, $v) = @_;
590 unless (defined($k) and length($k)) {
591 if (defined($k) and my $literal = is_literal_value($v)) {
592 belch 'Hash-pairs consisting of an empty string with a literal are deprecated, and will be removed in 2.0: use -and => [ $literal ] instead';
593 return { -literal => $literal };
595 puke "Supplying an empty left hand side argument is not supported";
598 return $self->_expand_hashpair_op($k, $v);
599 } elsif ($k =~ /^[^\w]/i) {
600 my ($lhs, @rhs) = @$v;
601 return $self->_expand_op(
602 -op, [ $k, $self->expand_expr($lhs, -ident), @rhs ]
605 return $self->_expand_hashpair_ident($k, $v);
608 sub _expand_hashpair_ident {
609 my ($self, $k, $v) = @_;
611 local our $Cur_Col_Meta = $k;
613 # hash with multiple or no elements is andor
615 if (ref($v) eq 'HASH' and keys %$v != 1) {
616 return $self->_expand_op_andor(and => $v, $k);
619 # undef needs to be re-sent with cmp to achieve IS/IS NOT NULL
621 if (is_undef_value($v)) {
622 return $self->_expand_hashpair_cmp($k => undef);
625 # scalars and objects get expanded as whatever requested or values
627 if (!ref($v) or Scalar::Util::blessed($v)) {
628 return $self->_expand_hashpair_scalar($k, $v);
631 # single key hashref is a hashtriple
633 if (ref($v) eq 'HASH') {
634 return $self->_expand_hashtriple($k, %$v);
637 # arrayref needs re-engineering over the elements
639 if (ref($v) eq 'ARRAY') {
640 return $self->sqlfalse unless @$v;
641 $self->_debug("ARRAY($k) means distribute over elements");
643 $v->[0] =~ /^-(and|or)$/i
644 ? (shift(@{$v = [ @$v ]}), $1)
645 : lc($self->{logic} || 'OR')
647 return $self->_expand_op_andor(
652 if (my $literal = is_literal_value($v)) {
654 belch 'Hash-pairs consisting of an empty string with a literal are deprecated, and will be removed in 2.0: use -and => [ $literal ] instead';
657 my ($sql, @bind) = @$literal;
658 if ($self->{bindtype} eq 'columns') {
660 $self->_assert_bindval_matches_bindtype($_);
663 return +{ -literal => [ $self->_quote($k).' '.$sql, @bind ] };
669 my ($self, $expr) = @_;
671 return $self->_expand_expr({ (our $Default_Scalar_To) => $expr });
674 sub _expand_hashpair_scalar {
675 my ($self, $k, $v) = @_;
677 return $self->_expand_hashpair_cmp(
678 $k, $self->_expand_scalar($v),
682 sub _expand_hashpair_op {
683 my ($self, $k, $v) = @_;
685 $self->_assert_pass_injection_guard($k =~ /\A-(.*)\Z/s);
687 my $op = $self->_normalize_op($k);
689 if (my $exp = $self->{expand}{$op}) {
690 return $self->$exp($op, $v);
693 # Ops prefixed with -not_ get converted
695 if (my ($rest) = $op =~/^not_(.*)$/) {
698 $self->_expand_expr({ "-${rest}", $v })
704 my $op = join(' ', split '_', $op);
706 # the old special op system requires illegality for top-level use
709 (our $Expand_Depth) == 1
711 List::Util::first { $op =~ $_->{regex} } @{$self->{special_ops}}
713 $self->{disable_old_special_ops}
714 and List::Util::first { $op =~ $_->{regex} } @BUILTIN_SPECIAL_OPS
718 puke "Illegal use of top-level '-$op'"
721 # the old unary op system means we should touch nothing and let it work
723 if (my $us = List::Util::first { $op =~ $_->{regex} } @{$self->{unary_ops}}) {
724 return { -op => [ $op, $v ] };
728 # an explicit node type is currently assumed to be expanded (this is almost
729 # certainly wrong and there should be expansion anyway)
731 if ($self->{render}{$op}) {
735 my $type = $self->{unknown_unop_always_func} ? -func : -op;
742 and (keys %$v)[0] =~ /^-/
745 (List::Util::first { $op =~ $_->{regex} } @{$self->{special_ops}})
754 ($type eq -func and ref($v) eq 'ARRAY')
755 ? map $self->_expand_expr($_), @$v
756 : $self->_expand_expr($v)
760 sub _expand_hashpair_cmp {
761 my ($self, $k, $v) = @_;
762 $self->_expand_hashtriple($k, $self->{cmp}, $v);
765 sub _expand_hashtriple {
766 my ($self, $k, $vk, $vv) = @_;
768 my $ik = $self->_expand_expr({ -ident => $k });
770 my $op = $self->_normalize_op($vk);
771 $self->_assert_pass_injection_guard($op);
773 if ($op =~ s/ _? \d+ $//x ) {
774 return $self->_expand_expr($k, { $vk, $vv });
776 if (my $x = $self->{expand_op}{$op}) {
777 local our $Cur_Col_Meta = $k;
778 return $self->$x($op, $vv, $k);
782 my $op = join(' ', split '_', $op);
784 if (my $us = List::Util::first { $op =~ $_->{regex} } @{$self->{special_ops}}) {
785 return { -op => [ $op, $ik, $vv ] };
787 if (my $us = List::Util::first { $op =~ $_->{regex} } @{$self->{unary_ops}}) {
791 { -op => [ $op, $vv ] }
795 if (ref($vv) eq 'ARRAY') {
797 my $logic = (defined($raw[0]) and $raw[0] =~ /^-(and|or)$/i)
798 ? (shift(@raw), $1) : 'or';
799 my @values = map +{ $vk => $_ }, @raw;
801 $op =~ $self->{inequality_op}
802 or $op =~ $self->{not_like_op}
804 if (lc($logic) eq 'or' and @values > 1) {
805 belch "A multi-element arrayref as an argument to the inequality op '${\uc(join ' ', split '_', $op)}' "
806 . 'is technically equivalent to an always-true 1=1 (you probably wanted '
807 . "to say ...{ \$inequality_op => [ -and => \@values ] }... instead)"
812 # try to DWIM on equality operators
813 return ($self->_dwim_op_to_is($op,
814 "Supplying an empty arrayref to '%s' is deprecated",
815 "operator '%s' applied on an empty array (field '$k')"
816 ) ? $self->sqlfalse : $self->sqltrue);
818 return $self->_expand_op_andor($logic => \@values, $k);
820 if (is_undef_value($vv)) {
821 my $is = ($self->_dwim_op_to_is($op,
822 "Supplying an undefined argument to '%s' is deprecated",
823 "unexpected operator '%s' with undef operand",
824 ) ? 'is' : 'is not');
826 return $self->_expand_hashpair($k => { $is, undef });
828 local our $Cur_Col_Meta = $k;
832 $self->_expand_expr($vv)
837 my ($self, $raw, $empty, $fail) = @_;
839 my $op = $self->_normalize_op($raw);
841 if ($op =~ /^not$/i) {
844 if ($op =~ $self->{equality_op}) {
847 if ($op =~ $self->{like_op}) {
848 belch(sprintf $empty, uc(join ' ', split '_', $op));
851 if ($op =~ $self->{inequality_op}) {
854 if ($op =~ $self->{not_like_op}) {
855 belch(sprintf $empty, uc(join ' ', split '_', $op));
858 puke(sprintf $fail, $op);
862 my ($self, undef, $args) = @_;
863 my ($func, @args) = @$args;
864 return +{ -func => [ $func, map $self->expand_expr($_), @args ] };
868 my ($self, undef, $body, $k) = @_;
869 return $self->_expand_hashpair_cmp(
870 $k, { -ident => $body }
872 unless (defined($body) or (ref($body) and ref($body) eq 'ARRAY')) {
873 puke "-ident requires a single plain scalar argument (a quotable identifier) or an arrayref of identifier parts";
875 my @parts = map split(/\Q${\($self->{name_sep}||'.')}\E/, $_),
876 ref($body) ? @$body : $body;
877 return { -ident => $parts[-1] } if $self->{_dequalify_idents};
878 unless ($self->{quote_char}) {
879 $self->_assert_pass_injection_guard($_) for @parts;
881 return +{ -ident => \@parts };
885 return $_[0]->_expand_hashpair_cmp(
886 $_[3], { -value => $_[2] },
888 +{ -bind => [ our $Cur_Col_Meta, $_[2] ] };
892 +{ -op => [ 'not', $_[0]->_expand_expr($_[2]) ] };
896 my ($self, undef, $args) = @_;
897 +{ -row => [ map $self->expand_expr($_), @$args ] };
901 my ($self, undef, $args) = @_;
902 my ($op, @opargs) = @$args;
903 if (my $exp = $self->{expand_op}{$op}) {
904 return $self->$exp($op, \@opargs);
906 +{ -op => [ $op, map $self->expand_expr($_), @opargs ] };
910 my ($self, undef, $v) = @_;
912 return $self->_expand_expr($v);
914 puke "-bool => undef not supported" unless defined($v);
915 return $self->_expand_expr({ -ident => $v });
918 sub _expand_op_andor {
919 my ($self, $logop, $v, $k) = @_;
921 $v = [ map +{ $k, $_ },
923 ? (map +{ $_ => $v->{$_} }, sort keys %$v)
927 if (ref($v) eq 'HASH') {
928 return undef unless keys %$v;
931 map $self->_expand_expr({ $_ => $v->{$_} }),
935 if (ref($v) eq 'ARRAY') {
936 $logop eq 'and' or $logop eq 'or' or puke "unknown logic: $logop";
939 (ref($_) eq 'ARRAY' and @$_)
940 or (ref($_) eq 'HASH' and %$_)
946 while (my ($el) = splice @expr, 0, 1) {
947 puke "Supplying an empty left hand side argument is not supported in array-pairs"
948 unless defined($el) and length($el);
949 my $elref = ref($el);
951 local our $Expand_Depth = 0;
952 push(@res, grep defined, $self->_expand_expr({ $el, shift(@expr) }));
953 } elsif ($elref eq 'ARRAY') {
954 push(@res, grep defined, $self->_expand_expr($el)) if @$el;
955 } elsif (my $l = is_literal_value($el)) {
956 push @res, { -literal => $l };
957 } elsif ($elref eq 'HASH') {
958 local our $Expand_Depth = 0;
959 push @res, grep defined, $self->_expand_expr($el) if %$el;
965 # return $res[0] if @res == 1;
966 return { -op => [ $logop, @res ] };
972 my ($self, $op, $vv, $k) = @_;
973 ($k, $vv) = @$vv unless defined $k;
974 puke "$op can only take undef as argument"
978 and exists($vv->{-value})
979 and !defined($vv->{-value})
981 return +{ -op => [ $op.'_null', $self->expand_expr($k, -ident) ] };
984 sub _expand_between {
985 my ($self, $op, $vv, $k) = @_;
986 $k = shift @{$vv = [ @$vv ]} unless defined $k;
987 my @rhs = map $self->_expand_expr($_),
988 ref($vv) eq 'ARRAY' ? @$vv : $vv;
990 (@rhs == 1 and ref($rhs[0]) eq 'HASH' and $rhs[0]->{-literal})
992 (@rhs == 2 and defined($rhs[0]) and defined($rhs[1]))
994 puke "Operator '${\uc($op)}' requires either an arrayref with two defined values or expressions, or a single literal scalarref/arrayref-ref";
998 $self->expand_expr(ref($k) ? $k : { -ident => $k }),
1004 my ($self, $op, $vv, $k) = @_;
1005 $k = shift @{$vv = [ @$vv ]} unless defined $k;
1006 if (my $literal = is_literal_value($vv)) {
1007 my ($sql, @bind) = @$literal;
1008 my $opened_sql = $self->_open_outer_paren($sql);
1010 $op, $self->expand_expr($k, -ident),
1011 { -literal => [ $opened_sql, @bind ] }
1015 'SQL::Abstract before v1.75 used to generate incorrect SQL when the '
1016 . "-${\uc($op)} operator was given an undef-containing list: !!!AUDIT YOUR CODE "
1017 . 'AND DATA!!! (the upcoming Data::Query-based version of SQL::Abstract '
1018 . 'will emit the logically correct SQL instead of raising this exception)'
1020 puke("Argument passed to the '${\uc($op)}' operator can not be undefined")
1022 my @rhs = map $self->expand_expr($_, -value),
1023 map { defined($_) ? $_: puke($undef_err) }
1024 (ref($vv) eq 'ARRAY' ? @$vv : $vv);
1025 return $self->${\($op =~ /^not/ ? 'sqltrue' : 'sqlfalse')} unless @rhs;
1029 $self->expand_expr($k, -ident),
1035 my ($self, undef, $v) = @_;
1036 # DBIx::Class requires a nest warning to be emitted once but the private
1037 # method it overrode to do so no longer exists
1038 if ($self->{warn_once_on_nest}) {
1039 unless (our $Nest_Warned) {
1041 "-nest in search conditions is deprecated, you most probably wanted:\n"
1042 .q|{..., -and => [ \%cond0, \@cond1, \'cond2', \[ 'cond3', [ col => bind ] ], etc. ], ... }|
1047 return $self->_expand_expr($v);
1051 my ($self, undef, $bind) = @_;
1052 return { -bind => $bind };
1055 sub _recurse_where {
1056 my ($self, $where, $logic) = @_;
1058 # Special case: top level simple string treated as literal
1060 my $where_exp = (ref($where)
1061 ? $self->_expand_expr($where, $logic)
1062 : { -literal => [ $where ] });
1064 # dispatch expanded expression
1066 my ($sql, @bind) = defined($where_exp) ? @{ $self->render_aqt($where_exp) || [] } : ();
1067 # DBIx::Class used to call _recurse_where in scalar context
1068 # something else might too...
1070 return ($sql, @bind);
1073 belch "Calling _recurse_where in scalar context is deprecated and will go away before 2.0";
1079 my ($self, undef, $ident) = @_;
1081 return [ $self->_convert($self->_quote($ident)) ];
1085 my ($self, undef, $values) = @_;
1086 return $self->join_query_parts('',
1088 $self->_render_op(undef, [ ',', @$values ]),
1094 my ($self, undef, $rest) = @_;
1095 my ($func, @args) = @$rest;
1096 return $self->join_query_parts('',
1097 $self->_sqlcase($func),
1099 $self->join_query_parts(', ', @args),
1105 my ($self, undef, $bind) = @_;
1106 return [ $self->_convert('?'), $self->_bindtype(@$bind) ];
1109 sub _render_literal {
1110 my ($self, undef, $literal) = @_;
1111 $self->_assert_bindval_matches_bindtype(@{$literal}[1..$#$literal]);
1116 my ($self, undef, $v) = @_;
1117 my ($op, @args) = @$v;
1118 if (my $r = $self->{render_op}{$op}) {
1119 return $self->$r($op, \@args);
1124 my $op = join(' ', split '_', $op);
1126 my $ss = List::Util::first { $op =~ $_->{regex} } @{$self->{special_ops}};
1127 if ($ss and @args > 1) {
1128 puke "Special op '${op}' requires first value to be identifier"
1129 unless my ($ident) = map $_->{-ident}, grep ref($_) eq 'HASH', $args[0];
1130 my $k = join(($self->{name_sep}||'.'), @$ident);
1131 local our $Expand_Depth = 1;
1132 return [ $self->${\($ss->{handler})}($k, $op, $args[1]) ];
1134 if (my $us = List::Util::first { $op =~ $_->{regex} } @{$self->{unary_ops}}) {
1135 return [ $self->${\($us->{handler})}($op, $args[0]) ];
1138 return $self->_render_unop_paren($op, \@args);
1142 return $self->_render_unop_prefix($op, \@args);
1144 return $self->_render_op_multop($op, \@args);
1150 sub _render_op_between {
1151 my ($self, $op, $args) = @_;
1152 my ($left, $low, $high) = @$args;
1155 puke "Single arg to between must be a literal"
1156 unless $low->{-literal};
1159 +($low, $self->format_keyword('and'), $high);
1162 return $self->join_query_parts(' ',
1163 '(', $left, $self->format_keyword($op), @rh, ')',
1168 my ($self, $op, $args) = @_;
1169 my ($lhs, @rhs) = @$args;
1171 return $self->join_query_parts(' ',
1173 $self->format_keyword($op),
1175 $self->join_query_parts(', ', @rhs),
1180 sub _render_op_andor {
1181 my ($self, $op, $args) = @_;
1182 return undef unless @$args;
1183 return $self->join_query_parts('', $args->[0]) if @$args == 1;
1184 return $self->join_query_parts(
1185 ' ' => '(', $self->_render_op_multop($op, $args), ')'
1189 sub _render_op_multop {
1190 my ($self, $op, $args) = @_;
1192 return undef unless @parts;
1193 return $self->render_aqt($parts[0]) if @parts == 1;
1194 my $join = ($op eq ','
1196 : ' '.$self->format_keyword($op).' '
1198 return $self->join_query_parts($join, @parts);
1201 sub join_query_parts {
1202 my ($self, $join, @parts) = @_;
1205 ? $self->render_aqt($_)
1206 : ((ref($_) eq 'ARRAY') ? $_ : [ $_ ])
1209 $self->{join_sql_parts}->($join, map $_->[0], @final),
1210 (map @{$_}[1..$#$_], @final),
1214 sub _render_unop_paren {
1215 my ($self, $op, $v) = @_;
1216 return $self->join_query_parts('',
1217 '(', $self->_render_unop_prefix($op, $v), ')'
1221 sub _render_unop_prefix {
1222 my ($self, $op, $v) = @_;
1223 return $self->join_query_parts(' ',
1224 $self->_sqlcase($op), $v->[0]
1228 sub _render_unop_postfix {
1229 my ($self, $op, $v) = @_;
1230 return $self->join_query_parts(' ',
1231 $v->[0], $self->format_keyword($op),
1235 # Some databases (SQLite) treat col IN (1, 2) different from
1236 # col IN ( (1, 2) ). Use this to strip all outer parens while
1237 # adding them back in the corresponding method
1238 sub _open_outer_paren {
1239 my ($self, $sql) = @_;
1241 while (my ($inner) = $sql =~ /^ \s* \( (.*) \) \s* $/xs) {
1243 # there are closing parens inside, need the heavy duty machinery
1244 # to reevaluate the extraction starting from $sql (full reevaluation)
1245 if ($inner =~ /\)/) {
1246 require Text::Balanced;
1248 my (undef, $remainder) = do {
1249 # idiotic design - writes to $@ but *DOES NOT* throw exceptions
1251 Text::Balanced::extract_bracketed($sql, '()', qr/\s*/);
1254 # the entire expression needs to be a balanced bracketed thing
1255 # (after an extract no remainder sans trailing space)
1256 last if defined $remainder and $remainder =~ /\S/;
1266 #======================================================================
1268 #======================================================================
1270 sub _expand_order_by {
1271 my ($self, $arg) = @_;
1273 return unless defined($arg) and not (ref($arg) eq 'ARRAY' and !@$arg);
1275 return $self->_expand_maybe_list_expr($arg)
1276 if ref($arg) eq 'HASH' and ($arg->{-op}||[''])->[0] eq ',';
1278 my $expander = sub {
1279 my ($self, $dir, $expr) = @_;
1280 my @to_expand = ref($expr) eq 'ARRAY' ? @$expr : $expr;
1281 foreach my $arg (@to_expand) {
1285 and grep /^-(asc|desc)$/, keys %$arg
1287 puke "ordering direction hash passed to order by must have exactly one key (-asc or -desc)";
1291 defined($dir) ? { -op => [ $dir =~ /^-?(.*)$/ ,=> $_ ] } : $_
1293 map $self->expand_expr($_, -ident),
1294 map ref($_) eq 'ARRAY' ? @$_ : $_, @to_expand;
1295 return undef unless @exp;
1296 return undef if @exp == 1 and not defined($exp[0]);
1297 return +{ -op => [ ',', @exp ] };
1300 local @{$self->{expand}}{qw(asc desc)} = (($expander) x 2);
1302 return $self->$expander(undef, $arg);
1306 my ($self, $arg) = @_;
1308 return '' unless defined(my $expanded = $self->_expand_order_by($arg));
1310 my ($sql, @bind) = @{ $self->render_aqt($expanded) };
1312 return '' unless length($sql);
1314 my $final_sql = $self->_sqlcase(' order by ').$sql;
1316 return ($final_sql, @bind);
1319 # _order_by no longer needs to call this so doesn't but DBIC uses it.
1321 sub _order_by_chunks {
1322 my ($self, $arg) = @_;
1324 return () unless defined(my $expanded = $self->_expand_order_by($arg));
1326 my @res = $self->_chunkify_order_by($expanded);
1327 (ref() ? $_->[0] : $_) .= '' for @res;
1331 sub _chunkify_order_by {
1332 my ($self, $expanded) = @_;
1334 return grep length, @{ $self->render_aqt($expanded) }
1335 if $expanded->{-ident} or @{$expanded->{-literal}||[]} == 1;
1338 if (ref() eq 'HASH' and $_->{-op} and $_->{-op}[0] eq ',') {
1339 my ($comma, @list) = @{$_->{-op}};
1340 return map $self->_chunkify_order_by($_), @list;
1342 return $self->render_aqt($_);
1346 #======================================================================
1347 # DATASOURCE (FOR NOW, JUST PLAIN TABLE OR LIST OF TABLES)
1348 #======================================================================
1354 $self->_expand_maybe_list_expr($from, -ident)
1359 #======================================================================
1361 #======================================================================
1363 sub _expand_maybe_list_expr {
1364 my ($self, $expr, $default) = @_;
1366 ',', map $self->expand_expr($_, $default),
1367 @{$expr->{-op}}[1..$#{$expr->{-op}}]
1368 ] } if ref($expr) eq 'HASH' and ($expr->{-op}||[''])->[0] eq ',';
1369 return +{ -op => [ ',',
1370 map $self->expand_expr($_, $default),
1371 ref($expr) eq 'ARRAY' ? @$expr : $expr
1375 # highly optimized, as it's called way too often
1377 # my ($self, $label) = @_;
1379 return '' unless defined $_[1];
1380 return ${$_[1]} if ref($_[1]) eq 'SCALAR';
1381 puke 'Identifier cannot be hashref' if ref($_[1]) eq 'HASH';
1383 unless ($_[0]->{quote_char}) {
1384 if (ref($_[1]) eq 'ARRAY') {
1385 return join($_[0]->{name_sep}||'.', @{$_[1]});
1387 $_[0]->_assert_pass_injection_guard($_[1]);
1392 my $qref = ref $_[0]->{quote_char};
1394 !$qref ? ($_[0]->{quote_char}, $_[0]->{quote_char})
1395 : ($qref eq 'ARRAY') ? @{$_[0]->{quote_char}}
1396 : puke "Unsupported quote_char format: $_[0]->{quote_char}";
1398 my $esc = $_[0]->{escape_char} || $r;
1400 # parts containing * are naturally unquoted
1402 $_[0]->{name_sep}||'',
1406 : do { (my $n = $_) =~ s/(\Q$esc\E|\Q$r\E)/$esc$1/g; $l . $n . $r }
1408 (ref($_[1]) eq 'ARRAY'
1412 ? split (/\Q$_[0]->{name_sep}\E/, $_[1] )
1420 # Conversion, if applicable
1422 #my ($self, $arg) = @_;
1423 if (my $conv = $_[0]->{convert_where}) {
1424 return @{ $_[0]->join_query_parts('',
1425 $_[0]->format_keyword($conv),
1434 #my ($self, $col, @vals) = @_;
1435 # called often - tighten code
1436 return $_[0]->{bindtype} eq 'columns'
1437 ? map {[$_[1], $_]} @_[2 .. $#_]
1442 # Dies if any element of @bind is not in [colname => value] format
1443 # if bindtype is 'columns'.
1444 sub _assert_bindval_matches_bindtype {
1445 # my ($self, @bind) = @_;
1447 if ($self->{bindtype} eq 'columns') {
1449 if (!defined $_ || ref($_) ne 'ARRAY' || @$_ != 2) {
1450 puke "bindtype 'columns' selected, you need to pass: [column_name => bind_value]"
1456 sub _join_sql_clauses {
1457 my ($self, $logic, $clauses_aref, $bind_aref) = @_;
1459 if (@$clauses_aref > 1) {
1460 my $join = " " . $self->_sqlcase($logic) . " ";
1461 my $sql = '( ' . join($join, @$clauses_aref) . ' )';
1462 return ($sql, @$bind_aref);
1464 elsif (@$clauses_aref) {
1465 return ($clauses_aref->[0], @$bind_aref); # no parentheses
1468 return (); # if no SQL, ignore @$bind_aref
1473 # Fix SQL case, if so requested
1475 # LDNOTE: if $self->{case} is true, then it contains 'lower', so we
1476 # don't touch the argument ... crooked logic, but let's not change it!
1477 return $_[0]->{case} ? $_[1] : uc($_[1]);
1480 sub format_keyword { $_[0]->_sqlcase(join ' ', split '_', $_[1]) }
1482 #======================================================================
1483 # DISPATCHING FROM REFKIND
1484 #======================================================================
1487 my ($self, $data) = @_;
1489 return 'UNDEF' unless defined $data;
1491 # blessed objects are treated like scalars
1492 my $ref = (Scalar::Util::blessed $data) ? '' : ref $data;
1494 return 'SCALAR' unless $ref;
1497 while ($ref eq 'REF') {
1499 $ref = (Scalar::Util::blessed $data) ? '' : ref $data;
1503 return ($ref||'SCALAR') . ('REF' x $n_steps);
1507 my ($self, $data) = @_;
1508 my @try = ($self->_refkind($data));
1509 push @try, 'SCALAR_or_UNDEF' if $try[0] eq 'SCALAR' || $try[0] eq 'UNDEF';
1510 push @try, 'FALLBACK';
1514 sub _METHOD_FOR_refkind {
1515 my ($self, $meth_prefix, $data) = @_;
1518 for (@{$self->_try_refkind($data)}) {
1519 $method = $self->can($meth_prefix."_".$_)
1523 return $method || puke "cannot dispatch on '$meth_prefix' for ".$self->_refkind($data);
1527 sub _SWITCH_refkind {
1528 my ($self, $data, $dispatch_table) = @_;
1531 for (@{$self->_try_refkind($data)}) {
1532 $coderef = $dispatch_table->{$_}
1536 puke "no dispatch entry for ".$self->_refkind($data)
1545 #======================================================================
1546 # VALUES, GENERATE, AUTOLOAD
1547 #======================================================================
1549 # LDNOTE: original code from nwiger, didn't touch code in that section
1550 # I feel the AUTOLOAD stuff should not be the default, it should
1551 # only be activated on explicit demand by user.
1555 my $data = shift || return;
1556 puke "Argument to ", __PACKAGE__, "->values must be a \\%hash"
1557 unless ref $data eq 'HASH';
1560 foreach my $k (sort keys %$data) {
1561 my $v = $data->{$k};
1562 $self->_SWITCH_refkind($v, {
1564 if ($self->{array_datatypes}) { # array datatype
1565 push @all_bind, $self->_bindtype($k, $v);
1567 else { # literal SQL with bind
1568 my ($sql, @bind) = @$v;
1569 $self->_assert_bindval_matches_bindtype(@bind);
1570 push @all_bind, @bind;
1573 ARRAYREFREF => sub { # literal SQL with bind
1574 my ($sql, @bind) = @${$v};
1575 $self->_assert_bindval_matches_bindtype(@bind);
1576 push @all_bind, @bind;
1578 SCALARREF => sub { # literal SQL without bind
1580 SCALAR_or_UNDEF => sub {
1581 push @all_bind, $self->_bindtype($k, $v);
1592 my(@sql, @sqlq, @sqlv);
1596 if ($ref eq 'HASH') {
1597 for my $k (sort keys %$_) {
1600 my $label = $self->_quote($k);
1601 if ($r eq 'ARRAY') {
1602 # literal SQL with bind
1603 my ($sql, @bind) = @$v;
1604 $self->_assert_bindval_matches_bindtype(@bind);
1605 push @sqlq, "$label = $sql";
1607 } elsif ($r eq 'SCALAR') {
1608 # literal SQL without bind
1609 push @sqlq, "$label = $$v";
1611 push @sqlq, "$label = ?";
1612 push @sqlv, $self->_bindtype($k, $v);
1615 push @sql, $self->_sqlcase('set'), join ', ', @sqlq;
1616 } elsif ($ref eq 'ARRAY') {
1617 # unlike insert(), assume these are ONLY the column names, i.e. for SQL
1620 if ($r eq 'ARRAY') { # literal SQL with bind
1621 my ($sql, @bind) = @$v;
1622 $self->_assert_bindval_matches_bindtype(@bind);
1625 } elsif ($r eq 'SCALAR') { # literal SQL without bind
1626 # embedded literal SQL
1633 push @sql, '(' . join(', ', @sqlq) . ')';
1634 } elsif ($ref eq 'SCALAR') {
1638 # strings get case twiddled
1639 push @sql, $self->_sqlcase($_);
1643 my $sql = join ' ', @sql;
1645 # this is pretty tricky
1646 # if ask for an array, return ($stmt, @bind)
1647 # otherwise, s/?/shift @sqlv/ to put it inline
1649 return ($sql, @sqlv);
1651 1 while $sql =~ s/\?/my $d = shift(@sqlv);
1652 ref $d ? $d->[1] : $d/e;
1661 # This allows us to check for a local, then _form, attr
1663 my($name) = $AUTOLOAD =~ /.*::(.+)/;
1664 puke "AUTOLOAD invoked for method name ${name} and allow_autoload option not set" unless $self->{allow_autoload};
1665 return $self->generate($name, @_);
1676 SQL::Abstract - Generate SQL from Perl data structures
1682 my $sql = SQL::Abstract->new;
1684 my($stmt, @bind) = $sql->select($source, \@fields, \%where, $order);
1686 my($stmt, @bind) = $sql->insert($table, \%fieldvals || \@values);
1688 my($stmt, @bind) = $sql->update($table, \%fieldvals, \%where);
1690 my($stmt, @bind) = $sql->delete($table, \%where);
1692 # Then, use these in your DBI statements
1693 my $sth = $dbh->prepare($stmt);
1694 $sth->execute(@bind);
1696 # Just generate the WHERE clause
1697 my($stmt, @bind) = $sql->where(\%where, $order);
1699 # Return values in the same order, for hashed queries
1700 # See PERFORMANCE section for more details
1701 my @bind = $sql->values(\%fieldvals);
1705 This module was inspired by the excellent L<DBIx::Abstract>.
1706 However, in using that module I found that what I really wanted
1707 to do was generate SQL, but still retain complete control over my
1708 statement handles and use the DBI interface. So, I set out to
1709 create an abstract SQL generation module.
1711 While based on the concepts used by L<DBIx::Abstract>, there are
1712 several important differences, especially when it comes to WHERE
1713 clauses. I have modified the concepts used to make the SQL easier
1714 to generate from Perl data structures and, IMO, more intuitive.
1715 The underlying idea is for this module to do what you mean, based
1716 on the data structures you provide it. The big advantage is that
1717 you don't have to modify your code every time your data changes,
1718 as this module figures it out.
1720 To begin with, an SQL INSERT is as easy as just specifying a hash
1721 of C<key=value> pairs:
1724 name => 'Jimbo Bobson',
1725 phone => '123-456-7890',
1726 address => '42 Sister Lane',
1727 city => 'St. Louis',
1728 state => 'Louisiana',
1731 The SQL can then be generated with this:
1733 my($stmt, @bind) = $sql->insert('people', \%data);
1735 Which would give you something like this:
1737 $stmt = "INSERT INTO people
1738 (address, city, name, phone, state)
1739 VALUES (?, ?, ?, ?, ?)";
1740 @bind = ('42 Sister Lane', 'St. Louis', 'Jimbo Bobson',
1741 '123-456-7890', 'Louisiana');
1743 These are then used directly in your DBI code:
1745 my $sth = $dbh->prepare($stmt);
1746 $sth->execute(@bind);
1748 =head2 Inserting and Updating Arrays
1750 If your database has array types (like for example Postgres),
1751 activate the special option C<< array_datatypes => 1 >>
1752 when creating the C<SQL::Abstract> object.
1753 Then you may use an arrayref to insert and update database array types:
1755 my $sql = SQL::Abstract->new(array_datatypes => 1);
1757 planets => [qw/Mercury Venus Earth Mars/]
1760 my($stmt, @bind) = $sql->insert('solar_system', \%data);
1764 $stmt = "INSERT INTO solar_system (planets) VALUES (?)"
1766 @bind = (['Mercury', 'Venus', 'Earth', 'Mars']);
1769 =head2 Inserting and Updating SQL
1771 In order to apply SQL functions to elements of your C<%data> you may
1772 specify a reference to an arrayref for the given hash value. For example,
1773 if you need to execute the Oracle C<to_date> function on a value, you can
1774 say something like this:
1778 date_entered => \[ "to_date(?,'MM/DD/YYYY')", "03/02/2003" ],
1781 The first value in the array is the actual SQL. Any other values are
1782 optional and would be included in the bind values array. This gives
1785 my($stmt, @bind) = $sql->insert('people', \%data);
1787 $stmt = "INSERT INTO people (name, date_entered)
1788 VALUES (?, to_date(?,'MM/DD/YYYY'))";
1789 @bind = ('Bill', '03/02/2003');
1791 An UPDATE is just as easy, all you change is the name of the function:
1793 my($stmt, @bind) = $sql->update('people', \%data);
1795 Notice that your C<%data> isn't touched; the module will generate
1796 the appropriately quirky SQL for you automatically. Usually you'll
1797 want to specify a WHERE clause for your UPDATE, though, which is
1798 where handling C<%where> hashes comes in handy...
1800 =head2 Complex where statements
1802 This module can generate pretty complicated WHERE statements
1803 easily. For example, simple C<key=value> pairs are taken to mean
1804 equality, and if you want to see if a field is within a set
1805 of values, you can use an arrayref. Let's say we wanted to
1806 SELECT some data based on this criteria:
1809 requestor => 'inna',
1810 worker => ['nwiger', 'rcwe', 'sfz'],
1811 status => { '!=', 'completed' }
1814 my($stmt, @bind) = $sql->select('tickets', '*', \%where);
1816 The above would give you something like this:
1818 $stmt = "SELECT * FROM tickets WHERE
1819 ( requestor = ? ) AND ( status != ? )
1820 AND ( worker = ? OR worker = ? OR worker = ? )";
1821 @bind = ('inna', 'completed', 'nwiger', 'rcwe', 'sfz');
1823 Which you could then use in DBI code like so:
1825 my $sth = $dbh->prepare($stmt);
1826 $sth->execute(@bind);
1832 The methods are simple. There's one for every major SQL operation,
1833 and a constructor you use first. The arguments are specified in a
1834 similar order for each method (table, then fields, then a where
1835 clause) to try and simplify things.
1837 =head2 new(option => 'value')
1839 The C<new()> function takes a list of options and values, and returns
1840 a new B<SQL::Abstract> object which can then be used to generate SQL
1841 through the methods below. The options accepted are:
1847 If set to 'lower', then SQL will be generated in all lowercase. By
1848 default SQL is generated in "textbook" case meaning something like:
1850 SELECT a_field FROM a_table WHERE some_field LIKE '%someval%'
1852 Any setting other than 'lower' is ignored.
1856 This determines what the default comparison operator is. By default
1857 it is C<=>, meaning that a hash like this:
1859 %where = (name => 'nwiger', email => 'nate@wiger.org');
1861 Will generate SQL like this:
1863 WHERE name = 'nwiger' AND email = 'nate@wiger.org'
1865 However, you may want loose comparisons by default, so if you set
1866 C<cmp> to C<like> you would get SQL such as:
1868 WHERE name like 'nwiger' AND email like 'nate@wiger.org'
1870 You can also override the comparison on an individual basis - see
1871 the huge section on L</"WHERE CLAUSES"> at the bottom.
1873 =item sqltrue, sqlfalse
1875 Expressions for inserting boolean values within SQL statements.
1876 By default these are C<1=1> and C<1=0>. They are used
1877 by the special operators C<-in> and C<-not_in> for generating
1878 correct SQL even when the argument is an empty array (see below).
1882 This determines the default logical operator for multiple WHERE
1883 statements in arrays or hashes. If absent, the default logic is "or"
1884 for arrays, and "and" for hashes. This means that a WHERE
1888 event_date => {'>=', '2/13/99'},
1889 event_date => {'<=', '4/24/03'},
1892 will generate SQL like this:
1894 WHERE event_date >= '2/13/99' OR event_date <= '4/24/03'
1896 This is probably not what you want given this query, though (look
1897 at the dates). To change the "OR" to an "AND", simply specify:
1899 my $sql = SQL::Abstract->new(logic => 'and');
1901 Which will change the above C<WHERE> to:
1903 WHERE event_date >= '2/13/99' AND event_date <= '4/24/03'
1905 The logic can also be changed locally by inserting
1906 a modifier in front of an arrayref:
1908 @where = (-and => [event_date => {'>=', '2/13/99'},
1909 event_date => {'<=', '4/24/03'} ]);
1911 See the L</"WHERE CLAUSES"> section for explanations.
1915 This will automatically convert comparisons using the specified SQL
1916 function for both column and value. This is mostly used with an argument
1917 of C<upper> or C<lower>, so that the SQL will have the effect of
1918 case-insensitive "searches". For example, this:
1920 $sql = SQL::Abstract->new(convert => 'upper');
1921 %where = (keywords => 'MaKe iT CAse inSeNSItive');
1923 Will turn out the following SQL:
1925 WHERE upper(keywords) like upper('MaKe iT CAse inSeNSItive')
1927 The conversion can be C<upper()>, C<lower()>, or any other SQL function
1928 that can be applied symmetrically to fields (actually B<SQL::Abstract> does
1929 not validate this option; it will just pass through what you specify verbatim).
1933 This is a kludge because many databases suck. For example, you can't
1934 just bind values using DBI's C<execute()> for Oracle C<CLOB> or C<BLOB> fields.
1935 Instead, you have to use C<bind_param()>:
1937 $sth->bind_param(1, 'reg data');
1938 $sth->bind_param(2, $lots, {ora_type => ORA_CLOB});
1940 The problem is, B<SQL::Abstract> will normally just return a C<@bind> array,
1941 which loses track of which field each slot refers to. Fear not.
1943 If you specify C<bindtype> in new, you can determine how C<@bind> is returned.
1944 Currently, you can specify either C<normal> (default) or C<columns>. If you
1945 specify C<columns>, you will get an array that looks like this:
1947 my $sql = SQL::Abstract->new(bindtype => 'columns');
1948 my($stmt, @bind) = $sql->insert(...);
1951 [ 'column1', 'value1' ],
1952 [ 'column2', 'value2' ],
1953 [ 'column3', 'value3' ],
1956 You can then iterate through this manually, using DBI's C<bind_param()>.
1958 $sth->prepare($stmt);
1961 my($col, $data) = @$_;
1962 if ($col eq 'details' || $col eq 'comments') {
1963 $sth->bind_param($i, $data, {ora_type => ORA_CLOB});
1964 } elsif ($col eq 'image') {
1965 $sth->bind_param($i, $data, {ora_type => ORA_BLOB});
1967 $sth->bind_param($i, $data);
1971 $sth->execute; # execute without @bind now
1973 Now, why would you still use B<SQL::Abstract> if you have to do this crap?
1974 Basically, the advantage is still that you don't have to care which fields
1975 are or are not included. You could wrap that above C<for> loop in a simple
1976 sub called C<bind_fields()> or something and reuse it repeatedly. You still
1977 get a layer of abstraction over manual SQL specification.
1979 Note that if you set L</bindtype> to C<columns>, the C<\[ $sql, @bind ]>
1980 construct (see L</Literal SQL with placeholders and bind values (subqueries)>)
1981 will expect the bind values in this format.
1985 This is the character that a table or column name will be quoted
1986 with. By default this is an empty string, but you could set it to
1987 the character C<`>, to generate SQL like this:
1989 SELECT `a_field` FROM `a_table` WHERE `some_field` LIKE '%someval%'
1991 Alternatively, you can supply an array ref of two items, the first being the left
1992 hand quote character, and the second the right hand quote character. For
1993 example, you could supply C<['[',']']> for SQL Server 2000 compliant quotes
1994 that generates SQL like this:
1996 SELECT [a_field] FROM [a_table] WHERE [some_field] LIKE '%someval%'
1998 Quoting is useful if you have tables or columns names that are reserved
1999 words in your database's SQL dialect.
2003 This is the character that will be used to escape L</quote_char>s appearing
2004 in an identifier before it has been quoted.
2006 The parameter default in case of a single L</quote_char> character is the quote
2009 When opening-closing-style quoting is used (L</quote_char> is an arrayref)
2010 this parameter defaults to the B<closing (right)> L</quote_char>. Occurrences
2011 of the B<opening (left)> L</quote_char> within the identifier are currently left
2012 untouched. The default for opening-closing-style quotes may change in future
2013 versions, thus you are B<strongly encouraged> to specify the escape character
2018 This is the character that separates a table and column name. It is
2019 necessary to specify this when the C<quote_char> option is selected,
2020 so that tables and column names can be individually quoted like this:
2022 SELECT `table`.`one_field` FROM `table` WHERE `table`.`other_field` = 1
2024 =item injection_guard
2026 A regular expression C<qr/.../> that is applied to any C<-function> and unquoted
2027 column name specified in a query structure. This is a safety mechanism to avoid
2028 injection attacks when mishandling user input e.g.:
2030 my %condition_as_column_value_pairs = get_values_from_user();
2031 $sqla->select( ... , \%condition_as_column_value_pairs );
2033 If the expression matches an exception is thrown. Note that literal SQL
2034 supplied via C<\'...'> or C<\['...']> is B<not> checked in any way.
2036 Defaults to checking for C<;> and the C<GO> keyword (TransactSQL)
2038 =item array_datatypes
2040 When this option is true, arrayrefs in INSERT or UPDATE are
2041 interpreted as array datatypes and are passed directly
2043 When this option is false, arrayrefs are interpreted
2044 as literal SQL, just like refs to arrayrefs
2045 (but this behavior is for backwards compatibility; when writing
2046 new queries, use the "reference to arrayref" syntax
2052 Takes a reference to a list of "special operators"
2053 to extend the syntax understood by L<SQL::Abstract>.
2054 See section L</"SPECIAL OPERATORS"> for details.
2058 Takes a reference to a list of "unary operators"
2059 to extend the syntax understood by L<SQL::Abstract>.
2060 See section L</"UNARY OPERATORS"> for details.
2066 =head2 insert($table, \@values || \%fieldvals, \%options)
2068 This is the simplest function. You simply give it a table name
2069 and either an arrayref of values or hashref of field/value pairs.
2070 It returns an SQL INSERT statement and a list of bind values.
2071 See the sections on L</"Inserting and Updating Arrays"> and
2072 L</"Inserting and Updating SQL"> for information on how to insert
2073 with those data types.
2075 The optional C<\%options> hash reference may contain additional
2076 options to generate the insert SQL. Currently supported options
2083 Takes either a scalar of raw SQL fields, or an array reference of
2084 field names, and adds on an SQL C<RETURNING> statement at the end.
2085 This allows you to return data generated by the insert statement
2086 (such as row IDs) without performing another C<SELECT> statement.
2087 Note, however, this is not part of the SQL standard and may not
2088 be supported by all database engines.
2092 =head2 update($table, \%fieldvals, \%where, \%options)
2094 This takes a table, hashref of field/value pairs, and an optional
2095 hashref L<WHERE clause|/WHERE CLAUSES>. It returns an SQL UPDATE function and a list
2097 See the sections on L</"Inserting and Updating Arrays"> and
2098 L</"Inserting and Updating SQL"> for information on how to insert
2099 with those data types.
2101 The optional C<\%options> hash reference may contain additional
2102 options to generate the update SQL. Currently supported options
2109 See the C<returning> option to
2110 L<insert|/insert($table, \@values || \%fieldvals, \%options)>.
2114 =head2 select($source, $fields, $where, $order)
2116 This returns a SQL SELECT statement and associated list of bind values, as
2117 specified by the arguments:
2123 Specification of the 'FROM' part of the statement.
2124 The argument can be either a plain scalar (interpreted as a table
2125 name, will be quoted), or an arrayref (interpreted as a list
2126 of table names, joined by commas, quoted), or a scalarref
2127 (literal SQL, not quoted).
2131 Specification of the list of fields to retrieve from
2133 The argument can be either an arrayref (interpreted as a list
2134 of field names, will be joined by commas and quoted), or a
2135 plain scalar (literal SQL, not quoted).
2136 Please observe that this API is not as flexible as that of
2137 the first argument C<$source>, for backwards compatibility reasons.
2141 Optional argument to specify the WHERE part of the query.
2142 The argument is most often a hashref, but can also be
2143 an arrayref or plain scalar --
2144 see section L<WHERE clause|/"WHERE CLAUSES"> for details.
2148 Optional argument to specify the ORDER BY part of the query.
2149 The argument can be a scalar, a hashref or an arrayref
2150 -- see section L<ORDER BY clause|/"ORDER BY CLAUSES">
2156 =head2 delete($table, \%where, \%options)
2158 This takes a table name and optional hashref L<WHERE clause|/WHERE CLAUSES>.
2159 It returns an SQL DELETE statement and list of bind values.
2161 The optional C<\%options> hash reference may contain additional
2162 options to generate the delete SQL. Currently supported options
2169 See the C<returning> option to
2170 L<insert|/insert($table, \@values || \%fieldvals, \%options)>.
2174 =head2 where(\%where, $order)
2176 This is used to generate just the WHERE clause. For example,
2177 if you have an arbitrary data structure and know what the
2178 rest of your SQL is going to look like, but want an easy way
2179 to produce a WHERE clause, use this. It returns an SQL WHERE
2180 clause and list of bind values.
2183 =head2 values(\%data)
2185 This just returns the values from the hash C<%data>, in the same
2186 order that would be returned from any of the other above queries.
2187 Using this allows you to markedly speed up your queries if you
2188 are affecting lots of rows. See below under the L</"PERFORMANCE"> section.
2190 =head2 generate($any, 'number', $of, \@data, $struct, \%types)
2192 Warning: This is an experimental method and subject to change.
2194 This returns arbitrarily generated SQL. It's a really basic shortcut.
2195 It will return two different things, depending on return context:
2197 my($stmt, @bind) = $sql->generate('create table', \$table, \@fields);
2198 my $stmt_and_val = $sql->generate('create table', \$table, \@fields);
2200 These would return the following:
2202 # First calling form
2203 $stmt = "CREATE TABLE test (?, ?)";
2204 @bind = (field1, field2);
2206 # Second calling form
2207 $stmt_and_val = "CREATE TABLE test (field1, field2)";
2209 Depending on what you're trying to do, it's up to you to choose the correct
2210 format. In this example, the second form is what you would want.
2214 $sql->generate('alter session', { nls_date_format => 'MM/YY' });
2218 ALTER SESSION SET nls_date_format = 'MM/YY'
2220 You get the idea. Strings get their case twiddled, but everything
2221 else remains verbatim.
2223 =head1 EXPORTABLE FUNCTIONS
2225 =head2 is_plain_value
2227 Determines if the supplied argument is a plain value as understood by this
2232 =item * The value is C<undef>
2234 =item * The value is a non-reference
2236 =item * The value is an object with stringification overloading
2238 =item * The value is of the form C<< { -value => $anything } >>
2242 On failure returns C<undef>, on success returns a B<scalar> reference
2243 to the original supplied argument.
2249 The stringification overloading detection is rather advanced: it takes
2250 into consideration not only the presence of a C<""> overload, but if that
2251 fails also checks for enabled
2252 L<autogenerated versions of C<"">|overload/Magic Autogeneration>, based
2253 on either C<0+> or C<bool>.
2255 Unfortunately testing in the field indicates that this
2256 detection B<< may tickle a latent bug in perl versions before 5.018 >>,
2257 but only when very large numbers of stringifying objects are involved.
2258 At the time of writing ( Sep 2014 ) there is no clear explanation of
2259 the direct cause, nor is there a manageably small test case that reliably
2260 reproduces the problem.
2262 If you encounter any of the following exceptions in B<random places within
2263 your application stack> - this module may be to blame:
2265 Operation "ne": no method found,
2266 left argument in overloaded package <something>,
2267 right argument in overloaded package <something>
2271 Stub found while resolving method "???" overloading """" in package <something>
2273 If you fall victim to the above - please attempt to reduce the problem
2274 to something that could be sent to the L<SQL::Abstract developers
2275 |DBIx::Class/GETTING HELP/SUPPORT>
2276 (either publicly or privately). As a workaround in the meantime you can
2277 set C<$ENV{SQLA_ISVALUE_IGNORE_AUTOGENERATED_STRINGIFICATION}> to a true
2278 value, which will most likely eliminate your problem (at the expense of
2279 not being able to properly detect exotic forms of stringification).
2281 This notice and environment variable will be removed in a future version,
2282 as soon as the underlying problem is found and a reliable workaround is
2287 =head2 is_literal_value
2289 Determines if the supplied argument is a literal value as understood by this
2294 =item * C<\$sql_string>
2296 =item * C<\[ $sql_string, @bind_values ]>
2300 On failure returns C<undef>, on success returns an B<array> reference
2301 containing the unpacked version of the supplied literal SQL and bind values.
2303 =head1 WHERE CLAUSES
2307 This module uses a variation on the idea from L<DBIx::Abstract>. It
2308 is B<NOT>, repeat I<not> 100% compatible. B<The main logic of this
2309 module is that things in arrays are OR'ed, and things in hashes
2312 The easiest way to explain is to show lots of examples. After
2313 each C<%where> hash shown, it is assumed you used:
2315 my($stmt, @bind) = $sql->where(\%where);
2317 However, note that the C<%where> hash can be used directly in any
2318 of the other functions as well, as described above.
2320 =head2 Key-value pairs
2322 So, let's get started. To begin, a simple hash:
2326 status => 'completed'
2329 Is converted to SQL C<key = val> statements:
2331 $stmt = "WHERE user = ? AND status = ?";
2332 @bind = ('nwiger', 'completed');
2334 One common thing I end up doing is having a list of values that
2335 a field can be in. To do this, simply specify a list inside of
2340 status => ['assigned', 'in-progress', 'pending'];
2343 This simple code will create the following:
2345 $stmt = "WHERE user = ? AND ( status = ? OR status = ? OR status = ? )";
2346 @bind = ('nwiger', 'assigned', 'in-progress', 'pending');
2348 A field associated to an empty arrayref will be considered a
2349 logical false and will generate 0=1.
2351 =head2 Tests for NULL values
2353 If the value part is C<undef> then this is converted to SQL <IS NULL>
2362 $stmt = "WHERE user = ? AND status IS NULL";
2365 To test if a column IS NOT NULL:
2369 status => { '!=', undef },
2372 =head2 Specific comparison operators
2374 If you want to specify a different type of operator for your comparison,
2375 you can use a hashref for a given column:
2379 status => { '!=', 'completed' }
2382 Which would generate:
2384 $stmt = "WHERE user = ? AND status != ?";
2385 @bind = ('nwiger', 'completed');
2387 To test against multiple values, just enclose the values in an arrayref:
2389 status => { '=', ['assigned', 'in-progress', 'pending'] };
2391 Which would give you:
2393 "WHERE status = ? OR status = ? OR status = ?"
2396 The hashref can also contain multiple pairs, in which case it is expanded
2397 into an C<AND> of its elements:
2401 status => { '!=', 'completed', -not_like => 'pending%' }
2404 # Or more dynamically, like from a form
2405 $where{user} = 'nwiger';
2406 $where{status}{'!='} = 'completed';
2407 $where{status}{'-not_like'} = 'pending%';
2409 # Both generate this
2410 $stmt = "WHERE user = ? AND status != ? AND status NOT LIKE ?";
2411 @bind = ('nwiger', 'completed', 'pending%');
2414 To get an OR instead, you can combine it with the arrayref idea:
2418 priority => [ { '=', 2 }, { '>', 5 } ]
2421 Which would generate:
2423 $stmt = "WHERE ( priority = ? OR priority > ? ) AND user = ?";
2424 @bind = ('2', '5', 'nwiger');
2426 If you want to include literal SQL (with or without bind values), just use a
2427 scalar reference or reference to an arrayref as the value:
2430 date_entered => { '>' => \["to_date(?, 'MM/DD/YYYY')", "11/26/2008"] },
2431 date_expires => { '<' => \"now()" }
2434 Which would generate:
2436 $stmt = "WHERE date_entered > to_date(?, 'MM/DD/YYYY') AND date_expires < now()";
2437 @bind = ('11/26/2008');
2440 =head2 Logic and nesting operators
2442 In the example above,
2443 there is a subtle trap if you want to say something like
2444 this (notice the C<AND>):
2446 WHERE priority != ? AND priority != ?
2448 Because, in Perl you I<can't> do this:
2450 priority => { '!=' => 2, '!=' => 1 }
2452 As the second C<!=> key will obliterate the first. The solution
2453 is to use the special C<-modifier> form inside an arrayref:
2455 priority => [ -and => {'!=', 2},
2459 Normally, these would be joined by C<OR>, but the modifier tells it
2460 to use C<AND> instead. (Hint: You can use this in conjunction with the
2461 C<logic> option to C<new()> in order to change the way your queries
2462 work by default.) B<Important:> Note that the C<-modifier> goes
2463 B<INSIDE> the arrayref, as an extra first element. This will
2464 B<NOT> do what you think it might:
2466 priority => -and => [{'!=', 2}, {'!=', 1}] # WRONG!
2468 Here is a quick list of equivalencies, since there is some overlap:
2471 status => {'!=', 'completed', 'not like', 'pending%' }
2472 status => [ -and => {'!=', 'completed'}, {'not like', 'pending%'}]
2475 status => {'=', ['assigned', 'in-progress']}
2476 status => [ -or => {'=', 'assigned'}, {'=', 'in-progress'}]
2477 status => [ {'=', 'assigned'}, {'=', 'in-progress'} ]
2481 =head2 Special operators: IN, BETWEEN, etc.
2483 You can also use the hashref format to compare a list of fields using the
2484 C<IN> comparison operator, by specifying the list as an arrayref:
2487 status => 'completed',
2488 reportid => { -in => [567, 2335, 2] }
2491 Which would generate:
2493 $stmt = "WHERE status = ? AND reportid IN (?,?,?)";
2494 @bind = ('completed', '567', '2335', '2');
2496 The reverse operator C<-not_in> generates SQL C<NOT IN> and is used in
2499 If the argument to C<-in> is an empty array, 'sqlfalse' is generated
2500 (by default: C<1=0>). Similarly, C<< -not_in => [] >> generates
2501 'sqltrue' (by default: C<1=1>).
2503 In addition to the array you can supply a chunk of literal sql or
2504 literal sql with bind:
2507 customer => { -in => \[
2508 'SELECT cust_id FROM cust WHERE balance > ?',
2511 status => { -in => \'SELECT status_codes FROM states' },
2517 customer IN ( SELECT cust_id FROM cust WHERE balance > ? )
2518 AND status IN ( SELECT status_codes FROM states )
2522 Finally, if the argument to C<-in> is not a reference, it will be
2523 treated as a single-element array.
2525 Another pair of operators is C<-between> and C<-not_between>,
2526 used with an arrayref of two values:
2530 completion_date => {
2531 -not_between => ['2002-10-01', '2003-02-06']
2537 WHERE user = ? AND completion_date NOT BETWEEN ( ? AND ? )
2539 Just like with C<-in> all plausible combinations of literal SQL
2543 start0 => { -between => [ 1, 2 ] },
2544 start1 => { -between => \["? AND ?", 1, 2] },
2545 start2 => { -between => \"lower(x) AND upper(y)" },
2546 start3 => { -between => [
2548 \["upper(?)", 'stuff' ],
2555 ( start0 BETWEEN ? AND ? )
2556 AND ( start1 BETWEEN ? AND ? )
2557 AND ( start2 BETWEEN lower(x) AND upper(y) )
2558 AND ( start3 BETWEEN lower(x) AND upper(?) )
2560 @bind = (1, 2, 1, 2, 'stuff');
2563 These are the two builtin "special operators"; but the
2564 list can be expanded: see section L</"SPECIAL OPERATORS"> below.
2566 =head2 Unary operators: bool
2568 If you wish to test against boolean columns or functions within your
2569 database you can use the C<-bool> and C<-not_bool> operators. For
2570 example to test the column C<is_user> being true and the column
2571 C<is_enabled> being false you would use:-
2575 -not_bool => 'is_enabled',
2580 WHERE is_user AND NOT is_enabled
2582 If a more complex combination is required, testing more conditions,
2583 then you should use the and/or operators:-
2588 -not_bool => { two=> { -rlike => 'bar' } },
2589 -not_bool => { three => [ { '=', 2 }, { '>', 5 } ] },
2600 (NOT ( three = ? OR three > ? ))
2603 =head2 Nested conditions, -and/-or prefixes
2605 So far, we've seen how multiple conditions are joined with a top-level
2606 C<AND>. We can change this by putting the different conditions we want in
2607 hashes and then putting those hashes in an array. For example:
2612 status => { -like => ['pending%', 'dispatched'] },
2616 status => 'unassigned',
2620 This data structure would create the following:
2622 $stmt = "WHERE ( user = ? AND ( status LIKE ? OR status LIKE ? ) )
2623 OR ( user = ? AND status = ? ) )";
2624 @bind = ('nwiger', 'pending', 'dispatched', 'robot', 'unassigned');
2627 Clauses in hashrefs or arrayrefs can be prefixed with an C<-and> or C<-or>
2628 to change the logic inside:
2634 -and => [ workhrs => {'>', 20}, geo => 'ASIA' ],
2635 -or => { workhrs => {'<', 50}, geo => 'EURO' },
2642 $stmt = "WHERE ( user = ?
2643 AND ( ( workhrs > ? AND geo = ? )
2644 OR ( workhrs < ? OR geo = ? ) ) )";
2645 @bind = ('nwiger', '20', 'ASIA', '50', 'EURO');
2647 =head3 Algebraic inconsistency, for historical reasons
2649 C<Important note>: when connecting several conditions, the C<-and->|C<-or>
2650 operator goes C<outside> of the nested structure; whereas when connecting
2651 several constraints on one column, the C<-and> operator goes
2652 C<inside> the arrayref. Here is an example combining both features:
2655 -and => [a => 1, b => 2],
2656 -or => [c => 3, d => 4],
2657 e => [-and => {-like => 'foo%'}, {-like => '%bar'} ]
2662 WHERE ( ( ( a = ? AND b = ? )
2663 OR ( c = ? OR d = ? )
2664 OR ( e LIKE ? AND e LIKE ? ) ) )
2666 This difference in syntax is unfortunate but must be preserved for
2667 historical reasons. So be careful: the two examples below would
2668 seem algebraically equivalent, but they are not
2671 { -like => 'foo%' },
2672 { -like => '%bar' },
2674 # yields: WHERE ( ( col LIKE ? AND col LIKE ? ) )
2677 { col => { -like => 'foo%' } },
2678 { col => { -like => '%bar' } },
2680 # yields: WHERE ( ( col LIKE ? OR col LIKE ? ) )
2683 =head2 Literal SQL and value type operators
2685 The basic premise of SQL::Abstract is that in WHERE specifications the "left
2686 side" is a column name and the "right side" is a value (normally rendered as
2687 a placeholder). This holds true for both hashrefs and arrayref pairs as you
2688 see in the L</WHERE CLAUSES> examples above. Sometimes it is necessary to
2689 alter this behavior. There are several ways of doing so.
2693 This is a virtual operator that signals the string to its right side is an
2694 identifier (a column name) and not a value. For example to compare two
2695 columns you would write:
2698 priority => { '<', 2 },
2699 requestor => { -ident => 'submitter' },
2704 $stmt = "WHERE priority < ? AND requestor = submitter";
2707 If you are maintaining legacy code you may see a different construct as
2708 described in L</Deprecated usage of Literal SQL>, please use C<-ident> in new
2713 This is a virtual operator that signals that the construct to its right side
2714 is a value to be passed to DBI. This is for example necessary when you want
2715 to write a where clause against an array (for RDBMS that support such
2716 datatypes). For example:
2719 array => { -value => [1, 2, 3] }
2724 $stmt = 'WHERE array = ?';
2725 @bind = ([1, 2, 3]);
2727 Note that if you were to simply say:
2733 the result would probably not be what you wanted:
2735 $stmt = 'WHERE array = ? OR array = ? OR array = ?';
2740 Finally, sometimes only literal SQL will do. To include a random snippet
2741 of SQL verbatim, you specify it as a scalar reference. Consider this only
2742 as a last resort. Usually there is a better way. For example:
2745 priority => { '<', 2 },
2746 requestor => { -in => \'(SELECT name FROM hitmen)' },
2751 $stmt = "WHERE priority < ? AND requestor IN (SELECT name FROM hitmen)"
2754 Note that in this example, you only get one bind parameter back, since
2755 the verbatim SQL is passed as part of the statement.
2759 Never use untrusted input as a literal SQL argument - this is a massive
2760 security risk (there is no way to check literal snippets for SQL
2761 injections and other nastyness). If you need to deal with untrusted input
2762 use literal SQL with placeholders as described next.
2764 =head3 Literal SQL with placeholders and bind values (subqueries)
2766 If the literal SQL to be inserted has placeholders and bind values,
2767 use a reference to an arrayref (yes this is a double reference --
2768 not so common, but perfectly legal Perl). For example, to find a date
2769 in Postgres you can use something like this:
2772 date_column => \[ "= date '2008-09-30' - ?::integer", 10 ]
2777 $stmt = "WHERE ( date_column = date '2008-09-30' - ?::integer )"
2780 Note that you must pass the bind values in the same format as they are returned
2781 by L<where|/where(\%where, $order)>. This means that if you set L</bindtype>
2782 to C<columns>, you must provide the bind values in the
2783 C<< [ column_meta => value ] >> format, where C<column_meta> is an opaque
2784 scalar value; most commonly the column name, but you can use any scalar value
2785 (including references and blessed references), L<SQL::Abstract> will simply
2786 pass it through intact. So if C<bindtype> is set to C<columns> the above
2787 example will look like:
2790 date_column => \[ "= date '2008-09-30' - ?::integer", [ {} => 10 ] ]
2793 Literal SQL is especially useful for nesting parenthesized clauses in the
2794 main SQL query. Here is a first example:
2796 my ($sub_stmt, @sub_bind) = ("SELECT c1 FROM t1 WHERE c2 < ? AND c3 LIKE ?",
2800 bar => \["IN ($sub_stmt)" => @sub_bind],
2805 $stmt = "WHERE (foo = ? AND bar IN (SELECT c1 FROM t1
2806 WHERE c2 < ? AND c3 LIKE ?))";
2807 @bind = (1234, 100, "foo%");
2809 Other subquery operators, like for example C<"E<gt> ALL"> or C<"NOT IN">,
2810 are expressed in the same way. Of course the C<$sub_stmt> and
2811 its associated bind values can be generated through a former call
2814 my ($sub_stmt, @sub_bind)
2815 = $sql->select("t1", "c1", {c2 => {"<" => 100},
2816 c3 => {-like => "foo%"}});
2819 bar => \["> ALL ($sub_stmt)" => @sub_bind],
2822 In the examples above, the subquery was used as an operator on a column;
2823 but the same principle also applies for a clause within the main C<%where>
2824 hash, like an EXISTS subquery:
2826 my ($sub_stmt, @sub_bind)
2827 = $sql->select("t1", "*", {c1 => 1, c2 => \"> t0.c0"});
2828 my %where = ( -and => [
2830 \["EXISTS ($sub_stmt)" => @sub_bind],
2835 $stmt = "WHERE (foo = ? AND EXISTS (SELECT * FROM t1
2836 WHERE c1 = ? AND c2 > t0.c0))";
2840 Observe that the condition on C<c2> in the subquery refers to
2841 column C<t0.c0> of the main query: this is I<not> a bind
2842 value, so we have to express it through a scalar ref.
2843 Writing C<< c2 => {">" => "t0.c0"} >> would have generated
2844 C<< c2 > ? >> with bind value C<"t0.c0"> ... not exactly
2845 what we wanted here.
2847 Finally, here is an example where a subquery is used
2848 for expressing unary negation:
2850 my ($sub_stmt, @sub_bind)
2851 = $sql->where({age => [{"<" => 10}, {">" => 20}]});
2852 $sub_stmt =~ s/^ where //i; # don't want "WHERE" in the subclause
2854 lname => {like => '%son%'},
2855 \["NOT ($sub_stmt)" => @sub_bind],
2860 $stmt = "lname LIKE ? AND NOT ( age < ? OR age > ? )"
2861 @bind = ('%son%', 10, 20)
2863 =head3 Deprecated usage of Literal SQL
2865 Below are some examples of archaic use of literal SQL. It is shown only as
2866 reference for those who deal with legacy code. Each example has a much
2867 better, cleaner and safer alternative that users should opt for in new code.
2873 my %where = ( requestor => \'IS NOT NULL' )
2875 $stmt = "WHERE requestor IS NOT NULL"
2877 This used to be the way of generating NULL comparisons, before the handling
2878 of C<undef> got formalized. For new code please use the superior syntax as
2879 described in L</Tests for NULL values>.
2883 my %where = ( requestor => \'= submitter' )
2885 $stmt = "WHERE requestor = submitter"
2887 This used to be the only way to compare columns. Use the superior L</-ident>
2888 method for all new code. For example an identifier declared in such a way
2889 will be properly quoted if L</quote_char> is properly set, while the legacy
2890 form will remain as supplied.
2894 my %where = ( is_ready => \"", completed => { '>', '2012-12-21' } )
2896 $stmt = "WHERE completed > ? AND is_ready"
2897 @bind = ('2012-12-21')
2899 Using an empty string literal used to be the only way to express a boolean.
2900 For all new code please use the much more readable
2901 L<-bool|/Unary operators: bool> operator.
2907 These pages could go on for a while, since the nesting of the data
2908 structures this module can handle are pretty much unlimited (the
2909 module implements the C<WHERE> expansion as a recursive function
2910 internally). Your best bet is to "play around" with the module a
2911 little to see how the data structures behave, and choose the best
2912 format for your data based on that.
2914 And of course, all the values above will probably be replaced with
2915 variables gotten from forms or the command line. After all, if you
2916 knew everything ahead of time, you wouldn't have to worry about
2917 dynamically-generating SQL and could just hardwire it into your
2920 =head1 ORDER BY CLAUSES
2922 Some functions take an order by clause. This can either be a scalar (just a
2923 column name), a hashref of C<< { -desc => 'col' } >> or C<< { -asc => 'col' }
2924 >>, a scalarref, an arrayref-ref, or an arrayref of any of the previous
2927 Given | Will Generate
2928 ---------------------------------------------------------------
2930 'colA' | ORDER BY colA
2932 [qw/colA colB/] | ORDER BY colA, colB
2934 {-asc => 'colA'} | ORDER BY colA ASC
2936 {-desc => 'colB'} | ORDER BY colB DESC
2938 ['colA', {-asc => 'colB'}] | ORDER BY colA, colB ASC
2940 { -asc => [qw/colA colB/] } | ORDER BY colA ASC, colB ASC
2942 \'colA DESC' | ORDER BY colA DESC
2944 \[ 'FUNC(colA, ?)', $x ] | ORDER BY FUNC(colA, ?)
2945 | /* ...with $x bound to ? */
2948 { -asc => 'colA' }, | colA ASC,
2949 { -desc => [qw/colB/] }, | colB DESC,
2950 { -asc => [qw/colC colD/] },| colC ASC, colD ASC,
2951 \'colE DESC', | colE DESC,
2952 \[ 'FUNC(colF, ?)', $x ], | FUNC(colF, ?)
2953 ] | /* ...with $x bound to ? */
2954 ===============================================================
2958 =head1 SPECIAL OPERATORS
2960 my $sqlmaker = SQL::Abstract->new(special_ops => [
2964 my ($self, $field, $op, $arg) = @_;
2970 handler => 'method_name',
2974 A "special operator" is a SQL syntactic clause that can be
2975 applied to a field, instead of a usual binary operator.
2978 WHERE field IN (?, ?, ?)
2979 WHERE field BETWEEN ? AND ?
2980 WHERE MATCH(field) AGAINST (?, ?)
2982 Special operators IN and BETWEEN are fairly standard and therefore
2983 are builtin within C<SQL::Abstract> (as the overridable methods
2984 C<_where_field_IN> and C<_where_field_BETWEEN>). For other operators,
2985 like the MATCH .. AGAINST example above which is specific to MySQL,
2986 you can write your own operator handlers - supply a C<special_ops>
2987 argument to the C<new> method. That argument takes an arrayref of
2988 operator definitions; each operator definition is a hashref with two
2995 the regular expression to match the operator
2999 Either a coderef or a plain scalar method name. In both cases
3000 the expected return is C<< ($sql, @bind) >>.
3002 When supplied with a method name, it is simply called on the
3003 L<SQL::Abstract> object as:
3005 $self->$method_name($field, $op, $arg)
3009 $field is the LHS of the operator
3010 $op is the part that matched the handler regex
3013 When supplied with a coderef, it is called as:
3015 $coderef->($self, $field, $op, $arg)
3020 For example, here is an implementation
3021 of the MATCH .. AGAINST syntax for MySQL
3023 my $sqlmaker = SQL::Abstract->new(special_ops => [
3025 # special op for MySql MATCH (field) AGAINST(word1, word2, ...)
3026 {regex => qr/^match$/i,
3028 my ($self, $field, $op, $arg) = @_;
3029 $arg = [$arg] if not ref $arg;
3030 my $label = $self->_quote($field);
3031 my ($placeholder) = $self->_convert('?');
3032 my $placeholders = join ", ", (($placeholder) x @$arg);
3033 my $sql = $self->_sqlcase('match') . " ($label) "
3034 . $self->_sqlcase('against') . " ($placeholders) ";
3035 my @bind = $self->_bindtype($field, @$arg);
3036 return ($sql, @bind);
3043 =head1 UNARY OPERATORS
3045 my $sqlmaker = SQL::Abstract->new(unary_ops => [
3049 my ($self, $op, $arg) = @_;
3055 handler => 'method_name',
3059 A "unary operator" is a SQL syntactic clause that can be
3060 applied to a field - the operator goes before the field
3062 You can write your own operator handlers - supply a C<unary_ops>
3063 argument to the C<new> method. That argument takes an arrayref of
3064 operator definitions; each operator definition is a hashref with two
3071 the regular expression to match the operator
3075 Either a coderef or a plain scalar method name. In both cases
3076 the expected return is C<< $sql >>.
3078 When supplied with a method name, it is simply called on the
3079 L<SQL::Abstract> object as:
3081 $self->$method_name($op, $arg)
3085 $op is the part that matched the handler regex
3086 $arg is the RHS or argument of the operator
3088 When supplied with a coderef, it is called as:
3090 $coderef->($self, $op, $arg)
3098 Thanks to some benchmarking by Mark Stosberg, it turns out that
3099 this module is many orders of magnitude faster than using C<DBIx::Abstract>.
3100 I must admit this wasn't an intentional design issue, but it's a
3101 byproduct of the fact that you get to control your C<DBI> handles
3104 To maximize performance, use a code snippet like the following:
3106 # prepare a statement handle using the first row
3107 # and then reuse it for the rest of the rows
3109 for my $href (@array_of_hashrefs) {
3110 $stmt ||= $sql->insert('table', $href);
3111 $sth ||= $dbh->prepare($stmt);
3112 $sth->execute($sql->values($href));
3115 The reason this works is because the keys in your C<$href> are sorted
3116 internally by B<SQL::Abstract>. Thus, as long as your data retains
3117 the same structure, you only have to generate the SQL the first time
3118 around. On subsequent queries, simply use the C<values> function provided
3119 by this module to return your values in the correct order.
3121 However this depends on the values having the same type - if, for
3122 example, the values of a where clause may either have values
3123 (resulting in sql of the form C<column = ?> with a single bind
3124 value), or alternatively the values might be C<undef> (resulting in
3125 sql of the form C<column IS NULL> with no bind value) then the
3126 caching technique suggested will not work.
3130 If you use my C<CGI::FormBuilder> module at all, you'll hopefully
3131 really like this part (I do, at least). Building up a complex query
3132 can be as simple as the following:
3139 use CGI::FormBuilder;
3142 my $form = CGI::FormBuilder->new(...);
3143 my $sql = SQL::Abstract->new;
3145 if ($form->submitted) {
3146 my $field = $form->field;
3147 my $id = delete $field->{id};
3148 my($stmt, @bind) = $sql->update('table', $field, {id => $id});
3151 Of course, you would still have to connect using C<DBI> to run the
3152 query, but the point is that if you make your form look like your
3153 table, the actual query script can be extremely simplistic.
3155 If you're B<REALLY> lazy (I am), check out C<HTML::QuickTable> for
3156 a fast interface to returning and formatting data. I frequently
3157 use these three modules together to write complex database query
3158 apps in under 50 lines.
3160 =head1 HOW TO CONTRIBUTE
3162 Contributions are always welcome, in all usable forms (we especially
3163 welcome documentation improvements). The delivery methods include git-
3164 or unified-diff formatted patches, GitHub pull requests, or plain bug
3165 reports either via RT or the Mailing list. Contributors are generally
3166 granted full access to the official repository after their first several
3167 patches pass successful review.
3169 This project is maintained in a git repository. The code and related tools are
3170 accessible at the following locations:
3174 =item * Official repo: L<git://git.shadowcat.co.uk/dbsrgits/SQL-Abstract.git>
3176 =item * Official gitweb: L<http://git.shadowcat.co.uk/gitweb/gitweb.cgi?p=dbsrgits/SQL-Abstract.git>
3178 =item * GitHub mirror: L<https://github.com/dbsrgits/sql-abstract>
3180 =item * Authorized committers: L<ssh://dbsrgits@git.shadowcat.co.uk/SQL-Abstract.git>
3186 Version 1.50 was a major internal refactoring of C<SQL::Abstract>.
3187 Great care has been taken to preserve the I<published> behavior
3188 documented in previous versions in the 1.* family; however,
3189 some features that were previously undocumented, or behaved
3190 differently from the documentation, had to be changed in order
3191 to clarify the semantics. Hence, client code that was relying
3192 on some dark areas of C<SQL::Abstract> v1.*
3193 B<might behave differently> in v1.50.
3195 The main changes are:
3201 support for literal SQL through the C<< \ [ $sql, @bind ] >> syntax.
3205 support for the { operator => \"..." } construct (to embed literal SQL)
3209 support for the { operator => \["...", @bind] } construct (to embed literal SQL with bind values)
3213 optional support for L<array datatypes|/"Inserting and Updating Arrays">
3217 defensive programming: check arguments
3221 fixed bug with global logic, which was previously implemented
3222 through global variables yielding side-effects. Prior versions would
3223 interpret C<< [ {cond1, cond2}, [cond3, cond4] ] >>
3224 as C<< "(cond1 AND cond2) OR (cond3 AND cond4)" >>.
3225 Now this is interpreted
3226 as C<< "(cond1 AND cond2) OR (cond3 OR cond4)" >>.
3231 fixed semantics of _bindtype on array args
3235 dropped the C<_anoncopy> of the %where tree. No longer necessary,
3236 we just avoid shifting arrays within that tree.
3240 dropped the C<_modlogic> function
3244 =head1 ACKNOWLEDGEMENTS
3246 There are a number of individuals that have really helped out with
3247 this module. Unfortunately, most of them submitted bugs via CPAN
3248 so I have no idea who they are! But the people I do know are:
3250 Ash Berlin (order_by hash term support)
3251 Matt Trout (DBIx::Class support)
3252 Mark Stosberg (benchmarking)
3253 Chas Owens (initial "IN" operator support)
3254 Philip Collins (per-field SQL functions)
3255 Eric Kolve (hashref "AND" support)
3256 Mike Fragassi (enhancements to "BETWEEN" and "LIKE")
3257 Dan Kubb (support for "quote_char" and "name_sep")
3258 Guillermo Roditi (patch to cleanup "IN" and "BETWEEN", fix and tests for _order_by)
3259 Laurent Dami (internal refactoring, extensible list of special operators, literal SQL)
3260 Norbert Buchmuller (support for literal SQL in hashpair, misc. fixes & tests)
3261 Peter Rabbitson (rewrite of SQLA::Test, misc. fixes & tests)
3262 Oliver Charles (support for "RETURNING" after "INSERT")
3268 L<DBIx::Class>, L<DBIx::Abstract>, L<CGI::FormBuilder>, L<HTML::QuickTable>.
3272 Copyright (c) 2001-2007 Nathan Wiger <nwiger@cpan.org>. All Rights Reserved.
3274 This module is actively maintained by Matt Trout <mst@shadowcatsystems.co.uk>
3276 For support, your best bet is to try the C<DBIx::Class> users mailing list.
3277 While not an official support venue, C<DBIx::Class> makes heavy use of
3278 C<SQL::Abstract>, and as such list members there are very familiar with
3279 how to create queries.
3283 This module is free software; you may copy this under the same
3284 terms as perl itself (either the GNU General Public License or
3285 the Artistic License)