1 package SQL::Abstract; # see doc at end of file
10 our @EXPORT_OK = qw(is_plain_value is_literal_value);
20 *SQL::Abstract::_ENV_::DETECT_AUTOGENERATED_STRINGIFICATION = $ENV{SQLA_ISVALUE_IGNORE_AUTOGENERATED_STRINGIFICATION}
26 #======================================================================
28 #======================================================================
30 our $VERSION = '1.86';
32 # This would confuse some packagers
33 $VERSION = eval $VERSION if $VERSION =~ /_/; # numify for warning-free dev releases
37 # special operators (-in, -between). May be extended/overridden by user.
38 # See section WHERE: BUILTIN SPECIAL OPERATORS below for implementation
39 my @BUILTIN_SPECIAL_OPS = (
40 {regex => qr/^ (?: not \s )? between $/ix, handler => '_where_field_BETWEEN'},
41 {regex => qr/^ (?: not \s )? in $/ix, handler => '_where_field_IN'},
42 {regex => qr/^ ident $/ix, handler => '_where_op_IDENT'},
43 {regex => qr/^ value $/ix, handler => '_where_op_VALUE'},
44 {regex => qr/^ is (?: \s+ not )? $/ix, handler => '_where_field_IS'},
47 # unaryish operators - key maps to handler
48 my @BUILTIN_UNARY_OPS = (
49 # the digits are backcompat stuff
50 { regex => qr/^ and (?: [_\s]? \d+ )? $/xi, handler => '_where_op_ANDOR' },
51 { regex => qr/^ or (?: [_\s]? \d+ )? $/xi, handler => '_where_op_ANDOR' },
52 { regex => qr/^ nest (?: [_\s]? \d+ )? $/xi, handler => '_where_op_NEST' },
53 { regex => qr/^ (?: not \s )? bool $/xi, handler => '_where_op_BOOL' },
54 { regex => qr/^ ident $/xi, handler => '_where_op_IDENT' },
55 { regex => qr/^ value $/xi, handler => '_where_op_VALUE' },
56 { regex => qr/^ op $/xi, handler => '_where_op_OP' },
57 { regex => qr/^ bind $/xi, handler => '_where_op_BIND' },
58 { regex => qr/^ literal $/xi, handler => '_where_op_LITERAL' },
61 #======================================================================
62 # DEBUGGING AND ERROR REPORTING
63 #======================================================================
66 return unless $_[0]->{debug}; shift; # a little faster
67 my $func = (caller(1))[3];
68 warn "[$func] ", @_, "\n";
72 my($func) = (caller(1))[3];
73 Carp::carp "[$func] Warning: ", @_;
77 my($func) = (caller(1))[3];
78 Carp::croak "[$func] Fatal: ", @_;
81 sub is_literal_value ($) {
82 ref $_[0] eq 'SCALAR' ? [ ${$_[0]} ]
83 : ( ref $_[0] eq 'REF' and ref ${$_[0]} eq 'ARRAY' ) ? [ @${ $_[0] } ]
87 # FIXME XSify - this can be done so much more efficiently
88 sub is_plain_value ($) {
90 ! length ref $_[0] ? \($_[0])
92 ref $_[0] eq 'HASH' and keys %{$_[0]} == 1
94 exists $_[0]->{-value}
95 ) ? \($_[0]->{-value})
97 # reuse @_ for even moar speedz
98 defined ( $_[1] = Scalar::Util::blessed $_[0] )
100 # deliberately not using Devel::OverloadInfo - the checks we are
101 # intersted in are much more limited than the fullblown thing, and
102 # this is a very hot piece of code
104 # simply using ->can('(""') can leave behind stub methods that
105 # break actually using the overload later (see L<perldiag/Stub
106 # found while resolving method "%s" overloading "%s" in package
107 # "%s"> and the source of overload::mycan())
109 # either has stringification which DBI SHOULD prefer out of the box
110 grep { *{ (qq[${_}::(""]) }{CODE} } @{ $_[2] = mro::get_linear_isa( $_[1] ) }
112 # has nummification or boolification, AND fallback is *not* disabled
114 SQL::Abstract::_ENV_::DETECT_AUTOGENERATED_STRINGIFICATION
117 grep { *{"${_}::(0+"}{CODE} } @{$_[2]}
119 grep { *{"${_}::(bool"}{CODE} } @{$_[2]}
123 # no fallback specified at all
124 ! ( ($_[3]) = grep { *{"${_}::()"}{CODE} } @{$_[2]} )
126 # fallback explicitly undef
127 ! defined ${"$_[3]::()"}
140 #======================================================================
142 #======================================================================
146 my $class = ref($self) || $self;
147 my %opt = (ref $_[0] eq 'HASH') ? %{$_[0]} : @_;
149 # choose our case by keeping an option around
150 delete $opt{case} if $opt{case} && $opt{case} ne 'lower';
152 # default logic for interpreting arrayrefs
153 $opt{logic} = $opt{logic} ? uc $opt{logic} : 'OR';
155 # how to return bind vars
156 $opt{bindtype} ||= 'normal';
158 # default comparison is "=", but can be overridden
161 # try to recognize which are the 'equality' and 'inequality' ops
162 # (temporary quickfix (in 2007), should go through a more seasoned API)
163 $opt{equality_op} = qr/^( \Q$opt{cmp}\E | \= )$/ix;
164 $opt{inequality_op} = qr/^( != | <> )$/ix;
166 $opt{like_op} = qr/^ (is\s+)? r?like $/xi;
167 $opt{not_like_op} = qr/^ (is\s+)? not \s+ r?like $/xi;
170 $opt{sqltrue} ||= '1=1';
171 $opt{sqlfalse} ||= '0=1';
174 $opt{special_ops} ||= [];
175 # regexes are applied in order, thus push after user-defines
176 push @{$opt{special_ops}}, @BUILTIN_SPECIAL_OPS;
179 $opt{unary_ops} ||= [];
180 push @{$opt{unary_ops}}, @BUILTIN_UNARY_OPS;
182 # rudimentary sanity-check for user supplied bits treated as functions/operators
183 # If a purported function matches this regular expression, an exception is thrown.
184 # Literal SQL is *NOT* subject to this check, only functions (and column names
185 # when quoting is not in effect)
188 # need to guard against ()'s in column names too, but this will break tons of
189 # hacks... ideas anyone?
190 $opt{injection_guard} ||= qr/
196 return bless \%opt, $class;
200 sub _assert_pass_injection_guard {
201 if ($_[1] =~ $_[0]->{injection_guard}) {
202 my $class = ref $_[0];
203 puke "Possible SQL injection attempt '$_[1]'. If this is indeed a part of the "
204 . "desired SQL use literal SQL ( \'...' or \[ '...' ] ) or supply your own "
205 . "{injection_guard} attribute to ${class}->new()"
210 #======================================================================
212 #======================================================================
216 my $table = $self->_table(shift);
217 my $data = shift || return;
220 my $method = $self->_METHOD_FOR_refkind("_insert", $data);
221 my ($sql, @bind) = $self->$method($data);
222 $sql = join " ", $self->_sqlcase('insert into'), $table, $sql;
224 if ($options->{returning}) {
225 my ($s, @b) = $self->_insert_returning($options);
230 return wantarray ? ($sql, @bind) : $sql;
233 # So that subclasses can override INSERT ... RETURNING separately from
234 # UPDATE and DELETE (e.g. DBIx::Class::SQLMaker::Oracle does this)
235 sub _insert_returning { shift->_returning(@_) }
238 my ($self, $options) = @_;
240 my $f = $options->{returning};
242 my $fieldlist = $self->_SWITCH_refkind($f, {
243 ARRAYREF => sub {join ', ', map { $self->_quote($_) } @$f;},
244 SCALAR => sub {$self->_quote($f)},
245 SCALARREF => sub {$$f},
247 return $self->_sqlcase(' returning ') . $fieldlist;
250 sub _insert_HASHREF { # explicit list of fields and then values
251 my ($self, $data) = @_;
253 my @fields = sort keys %$data;
255 my ($sql, @bind) = $self->_insert_values($data);
258 $_ = $self->_quote($_) foreach @fields;
259 $sql = "( ".join(", ", @fields).") ".$sql;
261 return ($sql, @bind);
264 sub _insert_ARRAYREF { # just generate values(?,?) part (no list of fields)
265 my ($self, $data) = @_;
267 # no names (arrayref) so can't generate bindtype
268 $self->{bindtype} ne 'columns'
269 or belch "can't do 'columns' bindtype when called with arrayref";
271 my (@values, @all_bind);
272 foreach my $value (@$data) {
273 my ($values, @bind) = $self->_insert_value(undef, $value);
274 push @values, $values;
275 push @all_bind, @bind;
277 my $sql = $self->_sqlcase('values')." ( ".join(", ", @values)." )";
278 return ($sql, @all_bind);
281 sub _insert_ARRAYREFREF { # literal SQL with bind
282 my ($self, $data) = @_;
284 my ($sql, @bind) = @${$data};
285 $self->_assert_bindval_matches_bindtype(@bind);
287 return ($sql, @bind);
291 sub _insert_SCALARREF { # literal SQL without bind
292 my ($self, $data) = @_;
298 my ($self, $data) = @_;
300 my (@values, @all_bind);
301 foreach my $column (sort keys %$data) {
302 my ($values, @bind) = $self->_insert_value($column, $data->{$column});
303 push @values, $values;
304 push @all_bind, @bind;
306 my $sql = $self->_sqlcase('values')." ( ".join(", ", @values)." )";
307 return ($sql, @all_bind);
311 my ($self, $column, $v) = @_;
313 my (@values, @all_bind);
314 $self->_SWITCH_refkind($v, {
317 if ($self->{array_datatypes}) { # if array datatype are activated
319 push @all_bind, $self->_bindtype($column, $v);
321 else { # else literal SQL with bind
322 my ($sql, @bind) = @$v;
323 $self->_assert_bindval_matches_bindtype(@bind);
325 push @all_bind, @bind;
329 ARRAYREFREF => sub { # literal SQL with bind
330 my ($sql, @bind) = @${$v};
331 $self->_assert_bindval_matches_bindtype(@bind);
333 push @all_bind, @bind;
336 # THINK: anything useful to do with a HASHREF ?
337 HASHREF => sub { # (nothing, but old SQLA passed it through)
338 #TODO in SQLA >= 2.0 it will die instead
339 belch "HASH ref as bind value in insert is not supported";
341 push @all_bind, $self->_bindtype($column, $v);
344 SCALARREF => sub { # literal SQL without bind
348 SCALAR_or_UNDEF => sub {
350 push @all_bind, $self->_bindtype($column, $v);
355 my $sql = join(", ", @values);
356 return ($sql, @all_bind);
361 #======================================================================
363 #======================================================================
368 my $table = $self->_table(shift);
369 my $data = shift || return;
373 # first build the 'SET' part of the sql statement
374 puke "Unsupported data type specified to \$sql->update"
375 unless ref $data eq 'HASH';
377 my ($sql, @all_bind) = $self->_update_set_values($data);
378 $sql = $self->_sqlcase('update ') . $table . $self->_sqlcase(' set ')
382 my($where_sql, @where_bind) = $self->where($where);
384 push @all_bind, @where_bind;
387 if ($options->{returning}) {
388 my ($returning_sql, @returning_bind) = $self->_update_returning($options);
389 $sql .= $returning_sql;
390 push @all_bind, @returning_bind;
393 return wantarray ? ($sql, @all_bind) : $sql;
396 sub _update_set_values {
397 my ($self, $data) = @_;
399 my (@set, @all_bind);
400 for my $k (sort keys %$data) {
403 my $label = $self->_quote($k);
405 $self->_SWITCH_refkind($v, {
407 if ($self->{array_datatypes}) { # array datatype
408 push @set, "$label = ?";
409 push @all_bind, $self->_bindtype($k, $v);
411 else { # literal SQL with bind
412 my ($sql, @bind) = @$v;
413 $self->_assert_bindval_matches_bindtype(@bind);
414 push @set, "$label = $sql";
415 push @all_bind, @bind;
418 ARRAYREFREF => sub { # literal SQL with bind
419 my ($sql, @bind) = @${$v};
420 $self->_assert_bindval_matches_bindtype(@bind);
421 push @set, "$label = $sql";
422 push @all_bind, @bind;
424 SCALARREF => sub { # literal SQL without bind
425 push @set, "$label = $$v";
428 my ($op, $arg, @rest) = %$v;
430 puke 'Operator calls in update must be in the form { -op => $arg }'
431 if (@rest or not $op =~ /^\-(.+)/);
433 local $self->{_nested_func_lhs} = $k;
434 my ($sql, @bind) = $self->_where_unary_op($1, $arg);
436 push @set, "$label = $sql";
437 push @all_bind, @bind;
439 SCALAR_or_UNDEF => sub {
440 push @set, "$label = ?";
441 push @all_bind, $self->_bindtype($k, $v);
447 my $sql = join ', ', @set;
449 return ($sql, @all_bind);
452 # So that subclasses can override UPDATE ... RETURNING separately from
454 sub _update_returning { shift->_returning(@_) }
458 #======================================================================
460 #======================================================================
465 my $table = $self->_table(shift);
466 my $fields = shift || '*';
470 my ($fields_sql, @bind) = $self->_select_fields($fields);
472 my ($where_sql, @where_bind) = $self->where($where, $order);
473 push @bind, @where_bind;
475 my $sql = join(' ', $self->_sqlcase('select'), $fields_sql,
476 $self->_sqlcase('from'), $table)
479 return wantarray ? ($sql, @bind) : $sql;
483 my ($self, $fields) = @_;
484 return ref $fields eq 'ARRAY' ? join ', ', map { $self->_quote($_) } @$fields
488 #======================================================================
490 #======================================================================
495 my $table = $self->_table(shift);
499 my($where_sql, @bind) = $self->where($where);
500 my $sql = $self->_sqlcase('delete from ') . $table . $where_sql;
502 if ($options->{returning}) {
503 my ($returning_sql, @returning_bind) = $self->_delete_returning($options);
504 $sql .= $returning_sql;
505 push @bind, @returning_bind;
508 return wantarray ? ($sql, @bind) : $sql;
511 # So that subclasses can override DELETE ... RETURNING separately from
513 sub _delete_returning { shift->_returning(@_) }
517 #======================================================================
519 #======================================================================
523 # Finally, a separate routine just to handle WHERE clauses
525 my ($self, $where, $order) = @_;
528 my ($sql, @bind) = $self->_recurse_where($where);
529 $sql = (defined $sql and length $sql) ? $self->_sqlcase(' where ') . "( $sql )" : '';
533 my ($order_sql, @order_bind) = $self->_order_by($order);
535 push @bind, @order_bind;
538 return wantarray ? ($sql, @bind) : $sql;
542 my ($self, $expr, $logic) = @_;
543 if (ref($expr) eq 'HASH') {
544 if (keys %$expr > 1) {
546 return +{ "-${logic}" => [
547 map $self->_expand_expr_hashpair($_ => $expr->{$_}, $logic),
551 return $self->_expand_expr_hashpair(%$expr, $logic);
553 if (ref($expr) eq 'ARRAY') {
554 $logic = lc($logic || $self->{logic});
555 $logic eq 'and' or $logic eq 'or' or puke "unknown logic: $logic";
561 while (my ($el) = splice @expr, 0, 1) {
562 puke "Supplying an empty left hand side argument is not supported in array-pairs"
563 unless defined($el) and length($el);
564 my $elref = ref($el);
566 push(@res, $self->_expand_expr({ $el, shift(@expr) }));
567 } elsif ($elref eq 'ARRAY') {
568 push(@res, $self->_expand_expr($el)) if @$el;
569 } elsif (is_literal_value($el)) {
571 } elsif ($elref eq 'HASH') {
572 push @res, $self->_expand_expr($el);
577 return { '-'.$logic => \@res };
579 if (my $literal = is_literal_value($expr)) {
580 return +{ -literal => $literal };
585 sub _expand_expr_hashpair {
586 my ($self, $k, $v, $logic) = @_;
587 unless (defined($k) and length($k)) {
588 if (defined($k) and is_literal_value($v)) {
589 belch 'Hash-pairs consisting of an empty string with a literal are deprecated, and will be removed in 2.0: use -and => [ $literal ] instead';
592 puke "Supplying an empty left hand side argument is not supported";
596 return $self->_expand_expr($v);
600 return $self->_expand_expr($v);
602 puke "-bool => undef not supported" unless defined($v);
603 return { -ident => $v };
605 if (my ($rest) = $k =~/^-not[_ ](.*)$/) {
606 return $self->_expand_expr({ -not => { "-${rest}", $v } }, $logic);
609 unless (defined($v)) {
610 my $orig_op = my $op = $self->{cmp};
612 $op =~ /^not$/i ? 'is not' # legacy
613 : $op =~ $self->{equality_op} ? 'is'
614 : $op =~ $self->{like_op} ? belch("Supplying an undefined argument to '@{[ uc $op]}' is deprecated") && 'is'
615 : $op =~ $self->{inequality_op} ? 'is not'
616 : $op =~ $self->{not_like_op} ? belch("Supplying an undefined argument to '@{[ uc $op]}' is deprecated") && 'is not'
617 : puke "unexpected operator '$orig_op' with undef operand";
618 return +{ -op => [ $is.' null', { -ident => $k } ] };
625 { -bind => [ $k, $v ] }
629 if (ref($v) eq 'HASH' and keys %$v > 1) {
631 map $self->_expand_expr_hashpair($k => { $_ => $v->{$_} }),
635 if (ref($v) eq 'ARRAY') {
636 return $self->{sqlfalse} unless @$v;
637 $self->_debug("ARRAY($k) means distribute over elements");
639 $v->[0] =~ /^-((?:and|or))$/i
640 ? ($v = [ @{$v}[1..$#$v] ], $1)
641 : ($self->{logic} || 'or')
643 return +{ "-${this_logic}" => [ map $self->_expand_expr({ $k => $_ }, $this_logic), @$v ] };
645 if (my $literal = is_literal_value($v)) {
647 belch 'Hash-pairs consisting of an empty string with a literal are deprecated, and will be removed in 2.0: use -and => [ $literal ] instead';
650 my ($sql, @bind) = @$literal;
651 if ($self->{bindtype} eq 'columns') {
653 if (!defined $_ || ref($_) ne 'ARRAY' || @$_ != 2) {
654 puke "bindtype 'columns' selected, you need to pass: [column_name => bind_value]"
658 return +{ -literal => [ $self->_quote($k).' '.$sql, @bind ] };
665 my ($self, $where, $logic) = @_;
667 my $where_exp = $self->_expand_expr($where, $logic);
669 # dispatch on appropriate method according to refkind of $where
670 my $method = $self->_METHOD_FOR_refkind("_where", $where_exp);
672 my ($sql, @bind) = $self->$method($where_exp, $logic);
674 # DBIx::Class used to call _recurse_where in scalar context
675 # something else might too...
677 return ($sql, @bind);
680 belch "Calling _recurse_where in scalar context is deprecated and will go away before 2.0";
687 #======================================================================
688 # WHERE: top-level ARRAYREF
689 #======================================================================
692 sub _where_ARRAYREF {
693 my ($self, $where, $logic) = @_;
695 $logic = uc($logic || $self->{logic});
696 $logic eq 'AND' or $logic eq 'OR' or puke "unknown logic: $logic";
698 my @clauses = @$where;
700 my (@sql_clauses, @all_bind);
701 # need to use while() so can shift() for pairs
703 my $el = shift @clauses;
705 $el = undef if (defined $el and ! length $el);
707 # switch according to kind of $el and get corresponding ($sql, @bind)
708 my ($sql, @bind) = $self->_SWITCH_refkind($el, {
710 # skip empty elements, otherwise get invalid trailing AND stuff
711 ARRAYREF => sub {$self->_recurse_where($el) if @$el},
715 $self->_assert_bindval_matches_bindtype(@b);
719 HASHREF => sub {$self->_recurse_where($el, 'and') if %$el},
721 SCALARREF => sub { ($$el); },
724 # top-level arrayref with scalars, recurse in pairs
725 $self->_recurse_where({$el => shift(@clauses)})
728 UNDEF => sub {puke "Supplying an empty left hand side argument is not supported in array-pairs" },
732 push @sql_clauses, $sql;
733 push @all_bind, @bind;
737 return $self->_join_sql_clauses($logic, \@sql_clauses, \@all_bind);
740 #======================================================================
741 # WHERE: top-level ARRAYREFREF
742 #======================================================================
744 sub _where_ARRAYREFREF {
745 my ($self, $where) = @_;
746 my ($sql, @bind) = @$$where;
747 $self->_assert_bindval_matches_bindtype(@bind);
748 return ($sql, @bind);
751 #======================================================================
752 # WHERE: top-level HASHREF
753 #======================================================================
756 my ($self, $where) = @_;
757 my (@sql_clauses, @all_bind);
759 for my $k (sort keys %$where) {
760 my $v = $where->{$k};
762 # ($k => $v) is either a special unary op or a regular hashpair
763 my ($sql, @bind) = do {
765 # put the operator in canonical form
767 $op = substr $op, 1; # remove initial dash
768 $op =~ s/^\s+|\s+$//g;# remove leading/trailing space
769 $op =~ s/\s+/ /g; # compress whitespace
771 # so that -not_foo works correctly
772 $op =~ s/^not_/NOT /i;
774 $self->_debug("Unary OP(-$op) within hashref, recursing...");
775 my ($s, @b) = $self->_where_unary_op($op, $v);
777 # top level vs nested
778 # we assume that handled unary ops will take care of their ()s
780 List::Util::first {$op =~ $_->{regex}} @{$self->{unary_ops}}
782 ( defined $self->{_nested_func_lhs} and $self->{_nested_func_lhs} eq $k )
788 if (is_literal_value ($v) ) {
789 belch 'Hash-pairs consisting of an empty string with a literal are deprecated, and will be removed in 2.0: use -and => [ $literal ] instead';
792 puke "Supplying an empty left hand side argument is not supported in hash-pairs";
796 my $method = $self->_METHOD_FOR_refkind("_where_hashpair", $v);
797 $self->$method($k, $v);
801 push @sql_clauses, $sql;
802 push @all_bind, @bind;
805 return $self->_join_sql_clauses('and', \@sql_clauses, \@all_bind);
808 sub _where_unary_op {
809 my ($self, $op, $rhs) = @_;
811 # top level special ops are illegal in general
812 puke "Illegal use of top-level '-$op'"
813 if !(defined $self->{_nested_func_lhs})
814 and List::Util::first { $op =~ $_->{regex} } @{$self->{special_ops}}
815 and not List::Util::first { $op =~ $_->{regex} } @{$self->{unary_ops}};
817 if (my $op_entry = List::Util::first { $op =~ $_->{regex} } @{$self->{unary_ops}}) {
818 my $handler = $op_entry->{handler};
820 if (not ref $handler) {
821 if ($op =~ s/ [_\s]? \d+ $//x ) {
822 belch 'Use of [and|or|nest]_N modifiers is deprecated and will be removed in SQLA v2.0. '
823 . "You probably wanted ...-and => [ -$op => COND1, -$op => COND2 ... ]";
825 return $self->$handler($op, $rhs);
827 elsif (ref $handler eq 'CODE') {
828 return $handler->($self, $op, $rhs);
831 puke "Illegal handler for operator $op - expecting a method name or a coderef";
835 $self->_debug("Generic unary OP: $op - recursing as function");
837 $self->_assert_pass_injection_guard($op);
839 my ($sql, @bind) = $self->_SWITCH_refkind($rhs, {
841 puke "Illegal use of top-level '-$op'"
842 unless defined $self->{_nested_func_lhs};
845 $self->_convert('?'),
846 $self->_bindtype($self->{_nested_func_lhs}, $rhs)
850 $self->_recurse_where($rhs)
854 $sql = sprintf('%s %s',
855 $self->_sqlcase($op),
859 return ($sql, @bind);
862 sub _where_op_ANDOR {
863 my ($self, $op, $v) = @_;
865 $self->_SWITCH_refkind($v, {
867 return $self->_where_ARRAYREF($v, $op);
871 return ($op =~ /^or/i)
872 ? $self->_where_ARRAYREF([ map { $_ => $v->{$_} } (sort keys %$v) ], $op)
873 : $self->_where_HASHREF($v);
877 puke "-$op => \\\$scalar makes little sense, use " .
879 ? '[ \$scalar, \%rest_of_conditions ] instead'
880 : '-and => [ \$scalar, \%rest_of_conditions ] instead'
885 puke "-$op => \\[...] makes little sense, use " .
887 ? '[ \[...], \%rest_of_conditions ] instead'
888 : '-and => [ \[...], \%rest_of_conditions ] instead'
892 SCALAR => sub { # permissively interpreted as SQL
893 puke "-$op => \$value makes little sense, use -bool => \$value instead";
897 puke "-$op => undef not supported";
903 my ($self, $op, $v) = @_;
905 $self->_SWITCH_refkind($v, {
907 SCALAR => sub { # permissively interpreted as SQL
908 belch "literal SQL should be -nest => \\'scalar' "
909 . "instead of -nest => 'scalar' ";
914 puke "-$op => undef not supported";
918 $self->_recurse_where($v);
926 my ($self, $op, $v) = @_;
928 my ($s, @b) = $self->_SWITCH_refkind($v, {
929 SCALAR => sub { # interpreted as SQL column
930 $self->_convert($self->_quote($v));
934 puke "-$op => undef not supported";
938 $self->_recurse_where($v);
942 $s = "(NOT $s)" if $op =~ /^not/i;
947 sub _where_op_IDENT {
949 my ($op, $rhs) = splice @_, -2;
950 if (! defined $rhs or length ref $rhs) {
951 puke "-$op requires a single plain scalar argument (a quotable identifier)";
954 # in case we are called as a top level special op (no '=')
955 my $has_lhs = my $lhs = shift;
957 $_ = $self->_convert($self->_quote($_)) for ($lhs, $rhs);
965 sub _where_op_VALUE {
967 my ($op, $rhs) = splice @_, -2;
969 # in case we are called as a top level special op (no '=')
973 if (! defined $rhs) {
975 ? $self->_where_hashpair_HASHREF($lhs, { -is => undef })
982 (defined $lhs ? $lhs : $self->{_nested_func_lhs}),
989 $self->_convert($self->_quote($lhs)) . ' = ' . $self->_convert('?'),
993 $self->_convert('?'),
1000 my %unop_postfix = map +($_ => 1), 'is null', 'is not null';
1003 my ($self, undef, $v) = @_;
1004 my ($op, @args) = @$v;
1005 $op =~ s/^-// if length($op) > 1;
1006 local $self->{_nested_func_lhs};
1008 my ($expr_sql, @bind) = $self->_recurse_where($args[0]);
1009 my $final_op = join ' ', split '_', $op;
1010 my $op_sql = $self->_sqlcase($final_op);
1012 $unop_postfix{lc($final_op)}
1013 ? "${expr_sql} ${op_sql}"
1014 : "${op_sql} ${expr_sql}"
1016 return ($final_sql, @bind);
1017 } elsif (@args == 2) {
1018 my ($l, $r) = map [ $self->_recurse_where($_) ], @args;
1019 return ( $l->[0].' '.$self->_sqlcase(join ' ', split '_', $op).' '.$r->[0], @{$l}[1..$#$l], @{$r}[1..$#$r] );
1024 sub _where_op_BIND {
1025 my ($self, undef, $bind) = @_;
1026 return ($self->_convert('?'), $self->_bindtype(@$bind));
1029 sub _where_op_LITERAL {
1030 my ($self, undef, $literal) = @_;
1034 sub _where_hashpair_ARRAYREF {
1035 my ($self, $k, $v) = @_;
1038 my @v = @$v; # need copy because of shift below
1039 $self->_debug("ARRAY($k) means distribute over elements");
1041 # put apart first element if it is an operator (-and, -or)
1043 (defined $v[0] && $v[0] =~ /^ - (?: AND|OR ) $/ix)
1047 my @distributed = map { {$k => $_} } @v;
1050 $self->_debug("OP($op) reinjected into the distributed array");
1051 unshift @distributed, $op;
1054 my $logic = $op ? substr($op, 1) : '';
1056 return $self->_recurse_where(\@distributed, $logic);
1059 $self->_debug("empty ARRAY($k) means 0=1");
1060 return ($self->{sqlfalse});
1064 sub _where_hashpair_HASHREF {
1065 my ($self, $k, $v, $logic) = @_;
1068 local $self->{_nested_func_lhs} = defined $self->{_nested_func_lhs}
1069 ? $self->{_nested_func_lhs}
1073 my ($all_sql, @all_bind);
1075 for my $orig_op (sort keys %$v) {
1076 my $val = $v->{$orig_op};
1078 # put the operator in canonical form
1081 # FIXME - we need to phase out dash-less ops
1082 $op =~ s/^-//; # remove possible initial dash
1083 $op =~ s/^\s+|\s+$//g;# remove leading/trailing space
1084 $op =~ s/\s+/ /g; # compress whitespace
1086 $self->_assert_pass_injection_guard($op);
1089 $op =~ s/^is_not/IS NOT/i;
1091 # so that -not_foo works correctly
1092 $op =~ s/^not_/NOT /i;
1094 # another retarded special case: foo => { $op => { -value => undef } }
1095 if (ref $val eq 'HASH' and keys %$val == 1 and exists $val->{-value} and ! defined $val->{-value} ) {
1101 # CASE: col-value logic modifiers
1102 if ($orig_op =~ /^ \- (and|or) $/xi) {
1103 ($sql, @bind) = $self->_where_hashpair_HASHREF($k, $val, $1);
1105 # CASE: special operators like -in or -between
1106 elsif (my $special_op = List::Util::first { $op =~ $_->{regex} } @{$self->{special_ops}}) {
1107 my $handler = $special_op->{handler};
1109 puke "No handler supplied for special operator $orig_op";
1111 elsif (not ref $handler) {
1112 ($sql, @bind) = $self->$handler($k, $op, $val);
1114 elsif (ref $handler eq 'CODE') {
1115 ($sql, @bind) = $handler->($self, $k, $op, $val);
1118 puke "Illegal handler for special operator $orig_op - expecting a method name or a coderef";
1122 $self->_SWITCH_refkind($val, {
1124 ARRAYREF => sub { # CASE: col => {op => \@vals}
1125 ($sql, @bind) = $self->_where_field_op_ARRAYREF($k, $op, $val);
1128 ARRAYREFREF => sub { # CASE: col => {op => \[$sql, @bind]} (literal SQL with bind)
1129 my ($sub_sql, @sub_bind) = @$$val;
1130 $self->_assert_bindval_matches_bindtype(@sub_bind);
1131 $sql = join ' ', $self->_convert($self->_quote($k)),
1132 $self->_sqlcase($op),
1137 UNDEF => sub { # CASE: col => {op => undef} : sql "IS (NOT)? NULL"
1139 $op =~ /^not$/i ? 'is not' # legacy
1140 : $op =~ $self->{equality_op} ? 'is'
1141 : $op =~ $self->{like_op} ? belch("Supplying an undefined argument to '@{[ uc $op]}' is deprecated") && 'is'
1142 : $op =~ $self->{inequality_op} ? 'is not'
1143 : $op =~ $self->{not_like_op} ? belch("Supplying an undefined argument to '@{[ uc $op]}' is deprecated") && 'is not'
1144 : puke "unexpected operator '$orig_op' with undef operand";
1146 $sql = $self->_quote($k) . $self->_sqlcase(" $is null");
1149 FALLBACK => sub { # CASE: col => {op/func => $stuff}
1150 ($sql, @bind) = $self->_where_unary_op($op, $val);
1153 $self->_convert($self->_quote($k)),
1154 $self->{_nested_func_lhs} eq $k ? $sql : "($sql)", # top level vs nested
1160 ($all_sql) = (defined $all_sql and $all_sql) ? $self->_join_sql_clauses($logic, [$all_sql, $sql], []) : $sql;
1161 push @all_bind, @bind;
1163 return ($all_sql, @all_bind);
1166 sub _where_field_IS {
1167 my ($self, $k, $op, $v) = @_;
1169 my ($s) = $self->_SWITCH_refkind($v, {
1172 $self->_convert($self->_quote($k)),
1173 map { $self->_sqlcase($_)} ($op, 'null')
1176 puke "$op can only take undef as argument";
1183 sub _where_field_op_ARRAYREF {
1184 my ($self, $k, $op, $vals) = @_;
1186 my @vals = @$vals; #always work on a copy
1189 $self->_debug(sprintf '%s means multiple elements: [ %s ]',
1191 join(', ', map { defined $_ ? "'$_'" : 'NULL' } @vals ),
1194 # see if the first element is an -and/-or op
1196 if (defined $vals[0] && $vals[0] =~ /^ - (AND|OR) $/ix) {
1201 # a long standing API wart - an attempt to change this behavior during
1202 # the 1.50 series failed *spectacularly*. Warn instead and leave the
1207 (!$logic or $logic eq 'OR')
1209 ($op =~ $self->{inequality_op} or $op =~ $self->{not_like_op})
1212 belch "A multi-element arrayref as an argument to the inequality op '$o' "
1213 . 'is technically equivalent to an always-true 1=1 (you probably wanted '
1214 . "to say ...{ \$inequality_op => [ -and => \@values ] }... instead)"
1218 # distribute $op over each remaining member of @vals, append logic if exists
1219 return $self->_recurse_where([map { {$k => {$op, $_}} } @vals], $logic);
1223 # try to DWIM on equality operators
1225 $op =~ $self->{equality_op} ? $self->{sqlfalse}
1226 : $op =~ $self->{like_op} ? belch("Supplying an empty arrayref to '@{[ uc $op]}' is deprecated") && $self->{sqlfalse}
1227 : $op =~ $self->{inequality_op} ? $self->{sqltrue}
1228 : $op =~ $self->{not_like_op} ? belch("Supplying an empty arrayref to '@{[ uc $op]}' is deprecated") && $self->{sqltrue}
1229 : puke "operator '$op' applied on an empty array (field '$k')";
1234 sub _where_hashpair_SCALARREF {
1235 my ($self, $k, $v) = @_;
1236 $self->_debug("SCALAR($k) means literal SQL: $$v");
1237 my $sql = $self->_quote($k) . " " . $$v;
1241 # literal SQL with bind
1242 sub _where_hashpair_ARRAYREFREF {
1243 my ($self, $k, $v) = @_;
1244 $self->_debug("REF($k) means literal SQL: @${$v}");
1245 my ($sql, @bind) = @$$v;
1246 $self->_assert_bindval_matches_bindtype(@bind);
1247 $sql = $self->_quote($k) . " " . $sql;
1248 return ($sql, @bind );
1251 # literal SQL without bind
1252 sub _where_hashpair_SCALAR {
1253 my ($self, $k, $v) = @_;
1254 $self->_debug("NOREF($k) means simple key=val: $k $self->{cmp} $v");
1255 return ($self->_where_hashpair_HASHREF($k, { $self->{cmp} => $v }));
1259 sub _where_hashpair_UNDEF {
1260 my ($self, $k, $v) = @_;
1261 $self->_debug("UNDEF($k) means IS NULL");
1262 return $self->_where_hashpair_HASHREF($k, { -is => undef });
1265 #======================================================================
1266 # WHERE: TOP-LEVEL OTHERS (SCALARREF, SCALAR, UNDEF)
1267 #======================================================================
1270 sub _where_SCALARREF {
1271 my ($self, $where) = @_;
1274 $self->_debug("SCALAR(*top) means literal SQL: $$where");
1280 my ($self, $where) = @_;
1283 $self->_debug("NOREF(*top) means literal SQL: $where");
1294 #======================================================================
1295 # WHERE: BUILTIN SPECIAL OPERATORS (-in, -between)
1296 #======================================================================
1299 sub _where_field_BETWEEN {
1300 my ($self, $k, $op, $vals) = @_;
1302 my ($label, $and, $placeholder);
1303 $label = $self->_convert($self->_quote($k));
1304 $and = ' ' . $self->_sqlcase('and') . ' ';
1305 $placeholder = $self->_convert('?');
1306 $op = $self->_sqlcase($op);
1308 my $invalid_args = "Operator '$op' requires either an arrayref with two defined values or expressions, or a single literal scalarref/arrayref-ref";
1310 my ($clause, @bind) = $self->_SWITCH_refkind($vals, {
1311 ARRAYREFREF => sub {
1312 my ($s, @b) = @$$vals;
1313 $self->_assert_bindval_matches_bindtype(@b);
1320 puke $invalid_args if @$vals != 2;
1322 my (@all_sql, @all_bind);
1323 foreach my $val (@$vals) {
1324 my ($sql, @bind) = $self->_SWITCH_refkind($val, {
1326 return ($placeholder, $self->_bindtype($k, $val) );
1331 ARRAYREFREF => sub {
1332 my ($sql, @bind) = @$$val;
1333 $self->_assert_bindval_matches_bindtype(@bind);
1334 return ($sql, @bind);
1337 my ($func, $arg, @rest) = %$val;
1338 puke "Only simple { -func => arg } functions accepted as sub-arguments to BETWEEN"
1339 if (@rest or $func !~ /^ \- (.+)/x);
1340 $self->_where_unary_op($1 => $arg);
1346 push @all_sql, $sql;
1347 push @all_bind, @bind;
1351 (join $and, @all_sql),
1360 my $sql = "( $label $op $clause )";
1361 return ($sql, @bind)
1365 sub _where_field_IN {
1366 my ($self, $k, $op, $vals) = @_;
1368 # backwards compatibility: if scalar, force into an arrayref
1369 $vals = [$vals] if defined $vals && ! ref $vals;
1371 my ($label) = $self->_convert($self->_quote($k));
1372 my ($placeholder) = $self->_convert('?');
1373 $op = $self->_sqlcase($op);
1375 my ($sql, @bind) = $self->_SWITCH_refkind($vals, {
1376 ARRAYREF => sub { # list of choices
1377 if (@$vals) { # nonempty list
1378 my (@all_sql, @all_bind);
1380 for my $val (@$vals) {
1381 my ($sql, @bind) = $self->_SWITCH_refkind($val, {
1383 return ($placeholder, $val);
1388 ARRAYREFREF => sub {
1389 my ($sql, @bind) = @$$val;
1390 $self->_assert_bindval_matches_bindtype(@bind);
1391 return ($sql, @bind);
1394 my ($func, $arg, @rest) = %$val;
1395 puke "Only simple { -func => arg } functions accepted as sub-arguments to IN"
1396 if (@rest or $func !~ /^ \- (.+)/x);
1397 $self->_where_unary_op($1 => $arg);
1401 'SQL::Abstract before v1.75 used to generate incorrect SQL when the '
1402 . "-$op operator was given an undef-containing list: !!!AUDIT YOUR CODE "
1403 . 'AND DATA!!! (the upcoming Data::Query-based version of SQL::Abstract '
1404 . 'will emit the logically correct SQL instead of raising this exception)'
1408 push @all_sql, $sql;
1409 push @all_bind, @bind;
1413 sprintf('%s %s ( %s )',
1416 join(', ', @all_sql)
1418 $self->_bindtype($k, @all_bind),
1421 else { # empty list: some databases won't understand "IN ()", so DWIM
1422 my $sql = ($op =~ /\bnot\b/i) ? $self->{sqltrue} : $self->{sqlfalse};
1427 SCALARREF => sub { # literal SQL
1428 my $sql = $self->_open_outer_paren($$vals);
1429 return ("$label $op ( $sql )");
1431 ARRAYREFREF => sub { # literal SQL with bind
1432 my ($sql, @bind) = @$$vals;
1433 $self->_assert_bindval_matches_bindtype(@bind);
1434 $sql = $self->_open_outer_paren($sql);
1435 return ("$label $op ( $sql )", @bind);
1439 puke "Argument passed to the '$op' operator can not be undefined";
1443 puke "special op $op requires an arrayref (or scalarref/arrayref-ref)";
1447 return ($sql, @bind);
1450 # Some databases (SQLite) treat col IN (1, 2) different from
1451 # col IN ( (1, 2) ). Use this to strip all outer parens while
1452 # adding them back in the corresponding method
1453 sub _open_outer_paren {
1454 my ($self, $sql) = @_;
1456 while (my ($inner) = $sql =~ /^ \s* \( (.*) \) \s* $/xs) {
1458 # there are closing parens inside, need the heavy duty machinery
1459 # to reevaluate the extraction starting from $sql (full reevaluation)
1460 if ($inner =~ /\)/) {
1461 require Text::Balanced;
1463 my (undef, $remainder) = do {
1464 # idiotic design - writes to $@ but *DOES NOT* throw exceptions
1466 Text::Balanced::extract_bracketed($sql, '()', qr/\s*/);
1469 # the entire expression needs to be a balanced bracketed thing
1470 # (after an extract no remainder sans trailing space)
1471 last if defined $remainder and $remainder =~ /\S/;
1481 #======================================================================
1483 #======================================================================
1486 my ($self, $arg) = @_;
1489 for my $c ($self->_order_by_chunks($arg) ) {
1490 $self->_SWITCH_refkind($c, {
1491 SCALAR => sub { push @sql, $c },
1492 ARRAYREF => sub { push @sql, shift @$c; push @bind, @$c },
1498 $self->_sqlcase(' order by'),
1504 return wantarray ? ($sql, @bind) : $sql;
1507 sub _order_by_chunks {
1508 my ($self, $arg) = @_;
1510 return $self->_SWITCH_refkind($arg, {
1513 map { $self->_order_by_chunks($_ ) } @$arg;
1516 ARRAYREFREF => sub {
1517 my ($s, @b) = @$$arg;
1518 $self->_assert_bindval_matches_bindtype(@b);
1522 SCALAR => sub {$self->_quote($arg)},
1524 UNDEF => sub {return () },
1526 SCALARREF => sub {$$arg}, # literal SQL, no quoting
1529 # get first pair in hash
1530 my ($key, $val, @rest) = %$arg;
1532 return () unless $key;
1534 if (@rest or not $key =~ /^-(desc|asc)/i) {
1535 puke "hash passed to _order_by must have exactly one key (-desc or -asc)";
1541 for my $c ($self->_order_by_chunks($val)) {
1544 $self->_SWITCH_refkind($c, {
1549 ($sql, @bind) = @$c;
1553 $sql = $sql . ' ' . $self->_sqlcase($direction);
1555 push @ret, [ $sql, @bind];
1564 #======================================================================
1565 # DATASOURCE (FOR NOW, JUST PLAIN TABLE OR LIST OF TABLES)
1566 #======================================================================
1571 $self->_SWITCH_refkind($from, {
1572 ARRAYREF => sub {join ', ', map { $self->_quote($_) } @$from;},
1573 SCALAR => sub {$self->_quote($from)},
1574 SCALARREF => sub {$$from},
1579 #======================================================================
1581 #======================================================================
1583 # highly optimized, as it's called way too often
1585 # my ($self, $label) = @_;
1587 return '' unless defined $_[1];
1588 return ${$_[1]} if ref($_[1]) eq 'SCALAR';
1590 $_[0]->{quote_char} or
1591 ($_[0]->_assert_pass_injection_guard($_[1]), return $_[1]);
1593 my $qref = ref $_[0]->{quote_char};
1595 !$qref ? ($_[0]->{quote_char}, $_[0]->{quote_char})
1596 : ($qref eq 'ARRAY') ? @{$_[0]->{quote_char}}
1597 : puke "Unsupported quote_char format: $_[0]->{quote_char}";
1599 my $esc = $_[0]->{escape_char} || $r;
1601 # parts containing * are naturally unquoted
1602 return join($_[0]->{name_sep}||'', map
1603 +( $_ eq '*' ? $_ : do { (my $n = $_) =~ s/(\Q$esc\E|\Q$r\E)/$esc$1/g; $l . $n . $r } ),
1604 ( $_[0]->{name_sep} ? split (/\Q$_[0]->{name_sep}\E/, $_[1] ) : $_[1] )
1609 # Conversion, if applicable
1611 #my ($self, $arg) = @_;
1612 if ($_[0]->{convert}) {
1613 return $_[0]->_sqlcase($_[0]->{convert}) .'(' . $_[1] . ')';
1620 #my ($self, $col, @vals) = @_;
1621 # called often - tighten code
1622 return $_[0]->{bindtype} eq 'columns'
1623 ? map {[$_[1], $_]} @_[2 .. $#_]
1628 # Dies if any element of @bind is not in [colname => value] format
1629 # if bindtype is 'columns'.
1630 sub _assert_bindval_matches_bindtype {
1631 # my ($self, @bind) = @_;
1633 if ($self->{bindtype} eq 'columns') {
1635 if (!defined $_ || ref($_) ne 'ARRAY' || @$_ != 2) {
1636 puke "bindtype 'columns' selected, you need to pass: [column_name => bind_value]"
1642 sub _join_sql_clauses {
1643 my ($self, $logic, $clauses_aref, $bind_aref) = @_;
1645 if (@$clauses_aref > 1) {
1646 my $join = " " . $self->_sqlcase($logic) . " ";
1647 my $sql = '( ' . join($join, @$clauses_aref) . ' )';
1648 return ($sql, @$bind_aref);
1650 elsif (@$clauses_aref) {
1651 return ($clauses_aref->[0], @$bind_aref); # no parentheses
1654 return (); # if no SQL, ignore @$bind_aref
1659 # Fix SQL case, if so requested
1661 # LDNOTE: if $self->{case} is true, then it contains 'lower', so we
1662 # don't touch the argument ... crooked logic, but let's not change it!
1663 return $_[0]->{case} ? $_[1] : uc($_[1]);
1667 #======================================================================
1668 # DISPATCHING FROM REFKIND
1669 #======================================================================
1672 my ($self, $data) = @_;
1674 return 'UNDEF' unless defined $data;
1676 # blessed objects are treated like scalars
1677 my $ref = (Scalar::Util::blessed $data) ? '' : ref $data;
1679 return 'SCALAR' unless $ref;
1682 while ($ref eq 'REF') {
1684 $ref = (Scalar::Util::blessed $data) ? '' : ref $data;
1688 return ($ref||'SCALAR') . ('REF' x $n_steps);
1692 my ($self, $data) = @_;
1693 my @try = ($self->_refkind($data));
1694 push @try, 'SCALAR_or_UNDEF' if $try[0] eq 'SCALAR' || $try[0] eq 'UNDEF';
1695 push @try, 'FALLBACK';
1699 sub _METHOD_FOR_refkind {
1700 my ($self, $meth_prefix, $data) = @_;
1703 for (@{$self->_try_refkind($data)}) {
1704 $method = $self->can($meth_prefix."_".$_)
1708 return $method || puke "cannot dispatch on '$meth_prefix' for ".$self->_refkind($data);
1712 sub _SWITCH_refkind {
1713 my ($self, $data, $dispatch_table) = @_;
1716 for (@{$self->_try_refkind($data)}) {
1717 $coderef = $dispatch_table->{$_}
1721 puke "no dispatch entry for ".$self->_refkind($data)
1730 #======================================================================
1731 # VALUES, GENERATE, AUTOLOAD
1732 #======================================================================
1734 # LDNOTE: original code from nwiger, didn't touch code in that section
1735 # I feel the AUTOLOAD stuff should not be the default, it should
1736 # only be activated on explicit demand by user.
1740 my $data = shift || return;
1741 puke "Argument to ", __PACKAGE__, "->values must be a \\%hash"
1742 unless ref $data eq 'HASH';
1745 foreach my $k (sort keys %$data) {
1746 my $v = $data->{$k};
1747 $self->_SWITCH_refkind($v, {
1749 if ($self->{array_datatypes}) { # array datatype
1750 push @all_bind, $self->_bindtype($k, $v);
1752 else { # literal SQL with bind
1753 my ($sql, @bind) = @$v;
1754 $self->_assert_bindval_matches_bindtype(@bind);
1755 push @all_bind, @bind;
1758 ARRAYREFREF => sub { # literal SQL with bind
1759 my ($sql, @bind) = @${$v};
1760 $self->_assert_bindval_matches_bindtype(@bind);
1761 push @all_bind, @bind;
1763 SCALARREF => sub { # literal SQL without bind
1765 SCALAR_or_UNDEF => sub {
1766 push @all_bind, $self->_bindtype($k, $v);
1777 my(@sql, @sqlq, @sqlv);
1781 if ($ref eq 'HASH') {
1782 for my $k (sort keys %$_) {
1785 my $label = $self->_quote($k);
1786 if ($r eq 'ARRAY') {
1787 # literal SQL with bind
1788 my ($sql, @bind) = @$v;
1789 $self->_assert_bindval_matches_bindtype(@bind);
1790 push @sqlq, "$label = $sql";
1792 } elsif ($r eq 'SCALAR') {
1793 # literal SQL without bind
1794 push @sqlq, "$label = $$v";
1796 push @sqlq, "$label = ?";
1797 push @sqlv, $self->_bindtype($k, $v);
1800 push @sql, $self->_sqlcase('set'), join ', ', @sqlq;
1801 } elsif ($ref eq 'ARRAY') {
1802 # unlike insert(), assume these are ONLY the column names, i.e. for SQL
1805 if ($r eq 'ARRAY') { # literal SQL with bind
1806 my ($sql, @bind) = @$v;
1807 $self->_assert_bindval_matches_bindtype(@bind);
1810 } elsif ($r eq 'SCALAR') { # literal SQL without bind
1811 # embedded literal SQL
1818 push @sql, '(' . join(', ', @sqlq) . ')';
1819 } elsif ($ref eq 'SCALAR') {
1823 # strings get case twiddled
1824 push @sql, $self->_sqlcase($_);
1828 my $sql = join ' ', @sql;
1830 # this is pretty tricky
1831 # if ask for an array, return ($stmt, @bind)
1832 # otherwise, s/?/shift @sqlv/ to put it inline
1834 return ($sql, @sqlv);
1836 1 while $sql =~ s/\?/my $d = shift(@sqlv);
1837 ref $d ? $d->[1] : $d/e;
1846 # This allows us to check for a local, then _form, attr
1848 my($name) = $AUTOLOAD =~ /.*::(.+)/;
1849 return $self->generate($name, @_);
1860 SQL::Abstract - Generate SQL from Perl data structures
1866 my $sql = SQL::Abstract->new;
1868 my($stmt, @bind) = $sql->select($source, \@fields, \%where, $order);
1870 my($stmt, @bind) = $sql->insert($table, \%fieldvals || \@values);
1872 my($stmt, @bind) = $sql->update($table, \%fieldvals, \%where);
1874 my($stmt, @bind) = $sql->delete($table, \%where);
1876 # Then, use these in your DBI statements
1877 my $sth = $dbh->prepare($stmt);
1878 $sth->execute(@bind);
1880 # Just generate the WHERE clause
1881 my($stmt, @bind) = $sql->where(\%where, $order);
1883 # Return values in the same order, for hashed queries
1884 # See PERFORMANCE section for more details
1885 my @bind = $sql->values(\%fieldvals);
1889 This module was inspired by the excellent L<DBIx::Abstract>.
1890 However, in using that module I found that what I really wanted
1891 to do was generate SQL, but still retain complete control over my
1892 statement handles and use the DBI interface. So, I set out to
1893 create an abstract SQL generation module.
1895 While based on the concepts used by L<DBIx::Abstract>, there are
1896 several important differences, especially when it comes to WHERE
1897 clauses. I have modified the concepts used to make the SQL easier
1898 to generate from Perl data structures and, IMO, more intuitive.
1899 The underlying idea is for this module to do what you mean, based
1900 on the data structures you provide it. The big advantage is that
1901 you don't have to modify your code every time your data changes,
1902 as this module figures it out.
1904 To begin with, an SQL INSERT is as easy as just specifying a hash
1905 of C<key=value> pairs:
1908 name => 'Jimbo Bobson',
1909 phone => '123-456-7890',
1910 address => '42 Sister Lane',
1911 city => 'St. Louis',
1912 state => 'Louisiana',
1915 The SQL can then be generated with this:
1917 my($stmt, @bind) = $sql->insert('people', \%data);
1919 Which would give you something like this:
1921 $stmt = "INSERT INTO people
1922 (address, city, name, phone, state)
1923 VALUES (?, ?, ?, ?, ?)";
1924 @bind = ('42 Sister Lane', 'St. Louis', 'Jimbo Bobson',
1925 '123-456-7890', 'Louisiana');
1927 These are then used directly in your DBI code:
1929 my $sth = $dbh->prepare($stmt);
1930 $sth->execute(@bind);
1932 =head2 Inserting and Updating Arrays
1934 If your database has array types (like for example Postgres),
1935 activate the special option C<< array_datatypes => 1 >>
1936 when creating the C<SQL::Abstract> object.
1937 Then you may use an arrayref to insert and update database array types:
1939 my $sql = SQL::Abstract->new(array_datatypes => 1);
1941 planets => [qw/Mercury Venus Earth Mars/]
1944 my($stmt, @bind) = $sql->insert('solar_system', \%data);
1948 $stmt = "INSERT INTO solar_system (planets) VALUES (?)"
1950 @bind = (['Mercury', 'Venus', 'Earth', 'Mars']);
1953 =head2 Inserting and Updating SQL
1955 In order to apply SQL functions to elements of your C<%data> you may
1956 specify a reference to an arrayref for the given hash value. For example,
1957 if you need to execute the Oracle C<to_date> function on a value, you can
1958 say something like this:
1962 date_entered => \[ "to_date(?,'MM/DD/YYYY')", "03/02/2003" ],
1965 The first value in the array is the actual SQL. Any other values are
1966 optional and would be included in the bind values array. This gives
1969 my($stmt, @bind) = $sql->insert('people', \%data);
1971 $stmt = "INSERT INTO people (name, date_entered)
1972 VALUES (?, to_date(?,'MM/DD/YYYY'))";
1973 @bind = ('Bill', '03/02/2003');
1975 An UPDATE is just as easy, all you change is the name of the function:
1977 my($stmt, @bind) = $sql->update('people', \%data);
1979 Notice that your C<%data> isn't touched; the module will generate
1980 the appropriately quirky SQL for you automatically. Usually you'll
1981 want to specify a WHERE clause for your UPDATE, though, which is
1982 where handling C<%where> hashes comes in handy...
1984 =head2 Complex where statements
1986 This module can generate pretty complicated WHERE statements
1987 easily. For example, simple C<key=value> pairs are taken to mean
1988 equality, and if you want to see if a field is within a set
1989 of values, you can use an arrayref. Let's say we wanted to
1990 SELECT some data based on this criteria:
1993 requestor => 'inna',
1994 worker => ['nwiger', 'rcwe', 'sfz'],
1995 status => { '!=', 'completed' }
1998 my($stmt, @bind) = $sql->select('tickets', '*', \%where);
2000 The above would give you something like this:
2002 $stmt = "SELECT * FROM tickets WHERE
2003 ( requestor = ? ) AND ( status != ? )
2004 AND ( worker = ? OR worker = ? OR worker = ? )";
2005 @bind = ('inna', 'completed', 'nwiger', 'rcwe', 'sfz');
2007 Which you could then use in DBI code like so:
2009 my $sth = $dbh->prepare($stmt);
2010 $sth->execute(@bind);
2016 The methods are simple. There's one for every major SQL operation,
2017 and a constructor you use first. The arguments are specified in a
2018 similar order for each method (table, then fields, then a where
2019 clause) to try and simplify things.
2021 =head2 new(option => 'value')
2023 The C<new()> function takes a list of options and values, and returns
2024 a new B<SQL::Abstract> object which can then be used to generate SQL
2025 through the methods below. The options accepted are:
2031 If set to 'lower', then SQL will be generated in all lowercase. By
2032 default SQL is generated in "textbook" case meaning something like:
2034 SELECT a_field FROM a_table WHERE some_field LIKE '%someval%'
2036 Any setting other than 'lower' is ignored.
2040 This determines what the default comparison operator is. By default
2041 it is C<=>, meaning that a hash like this:
2043 %where = (name => 'nwiger', email => 'nate@wiger.org');
2045 Will generate SQL like this:
2047 WHERE name = 'nwiger' AND email = 'nate@wiger.org'
2049 However, you may want loose comparisons by default, so if you set
2050 C<cmp> to C<like> you would get SQL such as:
2052 WHERE name like 'nwiger' AND email like 'nate@wiger.org'
2054 You can also override the comparison on an individual basis - see
2055 the huge section on L</"WHERE CLAUSES"> at the bottom.
2057 =item sqltrue, sqlfalse
2059 Expressions for inserting boolean values within SQL statements.
2060 By default these are C<1=1> and C<1=0>. They are used
2061 by the special operators C<-in> and C<-not_in> for generating
2062 correct SQL even when the argument is an empty array (see below).
2066 This determines the default logical operator for multiple WHERE
2067 statements in arrays or hashes. If absent, the default logic is "or"
2068 for arrays, and "and" for hashes. This means that a WHERE
2072 event_date => {'>=', '2/13/99'},
2073 event_date => {'<=', '4/24/03'},
2076 will generate SQL like this:
2078 WHERE event_date >= '2/13/99' OR event_date <= '4/24/03'
2080 This is probably not what you want given this query, though (look
2081 at the dates). To change the "OR" to an "AND", simply specify:
2083 my $sql = SQL::Abstract->new(logic => 'and');
2085 Which will change the above C<WHERE> to:
2087 WHERE event_date >= '2/13/99' AND event_date <= '4/24/03'
2089 The logic can also be changed locally by inserting
2090 a modifier in front of an arrayref:
2092 @where = (-and => [event_date => {'>=', '2/13/99'},
2093 event_date => {'<=', '4/24/03'} ]);
2095 See the L</"WHERE CLAUSES"> section for explanations.
2099 This will automatically convert comparisons using the specified SQL
2100 function for both column and value. This is mostly used with an argument
2101 of C<upper> or C<lower>, so that the SQL will have the effect of
2102 case-insensitive "searches". For example, this:
2104 $sql = SQL::Abstract->new(convert => 'upper');
2105 %where = (keywords => 'MaKe iT CAse inSeNSItive');
2107 Will turn out the following SQL:
2109 WHERE upper(keywords) like upper('MaKe iT CAse inSeNSItive')
2111 The conversion can be C<upper()>, C<lower()>, or any other SQL function
2112 that can be applied symmetrically to fields (actually B<SQL::Abstract> does
2113 not validate this option; it will just pass through what you specify verbatim).
2117 This is a kludge because many databases suck. For example, you can't
2118 just bind values using DBI's C<execute()> for Oracle C<CLOB> or C<BLOB> fields.
2119 Instead, you have to use C<bind_param()>:
2121 $sth->bind_param(1, 'reg data');
2122 $sth->bind_param(2, $lots, {ora_type => ORA_CLOB});
2124 The problem is, B<SQL::Abstract> will normally just return a C<@bind> array,
2125 which loses track of which field each slot refers to. Fear not.
2127 If you specify C<bindtype> in new, you can determine how C<@bind> is returned.
2128 Currently, you can specify either C<normal> (default) or C<columns>. If you
2129 specify C<columns>, you will get an array that looks like this:
2131 my $sql = SQL::Abstract->new(bindtype => 'columns');
2132 my($stmt, @bind) = $sql->insert(...);
2135 [ 'column1', 'value1' ],
2136 [ 'column2', 'value2' ],
2137 [ 'column3', 'value3' ],
2140 You can then iterate through this manually, using DBI's C<bind_param()>.
2142 $sth->prepare($stmt);
2145 my($col, $data) = @$_;
2146 if ($col eq 'details' || $col eq 'comments') {
2147 $sth->bind_param($i, $data, {ora_type => ORA_CLOB});
2148 } elsif ($col eq 'image') {
2149 $sth->bind_param($i, $data, {ora_type => ORA_BLOB});
2151 $sth->bind_param($i, $data);
2155 $sth->execute; # execute without @bind now
2157 Now, why would you still use B<SQL::Abstract> if you have to do this crap?
2158 Basically, the advantage is still that you don't have to care which fields
2159 are or are not included. You could wrap that above C<for> loop in a simple
2160 sub called C<bind_fields()> or something and reuse it repeatedly. You still
2161 get a layer of abstraction over manual SQL specification.
2163 Note that if you set L</bindtype> to C<columns>, the C<\[ $sql, @bind ]>
2164 construct (see L</Literal SQL with placeholders and bind values (subqueries)>)
2165 will expect the bind values in this format.
2169 This is the character that a table or column name will be quoted
2170 with. By default this is an empty string, but you could set it to
2171 the character C<`>, to generate SQL like this:
2173 SELECT `a_field` FROM `a_table` WHERE `some_field` LIKE '%someval%'
2175 Alternatively, you can supply an array ref of two items, the first being the left
2176 hand quote character, and the second the right hand quote character. For
2177 example, you could supply C<['[',']']> for SQL Server 2000 compliant quotes
2178 that generates SQL like this:
2180 SELECT [a_field] FROM [a_table] WHERE [some_field] LIKE '%someval%'
2182 Quoting is useful if you have tables or columns names that are reserved
2183 words in your database's SQL dialect.
2187 This is the character that will be used to escape L</quote_char>s appearing
2188 in an identifier before it has been quoted.
2190 The parameter default in case of a single L</quote_char> character is the quote
2193 When opening-closing-style quoting is used (L</quote_char> is an arrayref)
2194 this parameter defaults to the B<closing (right)> L</quote_char>. Occurrences
2195 of the B<opening (left)> L</quote_char> within the identifier are currently left
2196 untouched. The default for opening-closing-style quotes may change in future
2197 versions, thus you are B<strongly encouraged> to specify the escape character
2202 This is the character that separates a table and column name. It is
2203 necessary to specify this when the C<quote_char> option is selected,
2204 so that tables and column names can be individually quoted like this:
2206 SELECT `table`.`one_field` FROM `table` WHERE `table`.`other_field` = 1
2208 =item injection_guard
2210 A regular expression C<qr/.../> that is applied to any C<-function> and unquoted
2211 column name specified in a query structure. This is a safety mechanism to avoid
2212 injection attacks when mishandling user input e.g.:
2214 my %condition_as_column_value_pairs = get_values_from_user();
2215 $sqla->select( ... , \%condition_as_column_value_pairs );
2217 If the expression matches an exception is thrown. Note that literal SQL
2218 supplied via C<\'...'> or C<\['...']> is B<not> checked in any way.
2220 Defaults to checking for C<;> and the C<GO> keyword (TransactSQL)
2222 =item array_datatypes
2224 When this option is true, arrayrefs in INSERT or UPDATE are
2225 interpreted as array datatypes and are passed directly
2227 When this option is false, arrayrefs are interpreted
2228 as literal SQL, just like refs to arrayrefs
2229 (but this behavior is for backwards compatibility; when writing
2230 new queries, use the "reference to arrayref" syntax
2236 Takes a reference to a list of "special operators"
2237 to extend the syntax understood by L<SQL::Abstract>.
2238 See section L</"SPECIAL OPERATORS"> for details.
2242 Takes a reference to a list of "unary operators"
2243 to extend the syntax understood by L<SQL::Abstract>.
2244 See section L</"UNARY OPERATORS"> for details.
2250 =head2 insert($table, \@values || \%fieldvals, \%options)
2252 This is the simplest function. You simply give it a table name
2253 and either an arrayref of values or hashref of field/value pairs.
2254 It returns an SQL INSERT statement and a list of bind values.
2255 See the sections on L</"Inserting and Updating Arrays"> and
2256 L</"Inserting and Updating SQL"> for information on how to insert
2257 with those data types.
2259 The optional C<\%options> hash reference may contain additional
2260 options to generate the insert SQL. Currently supported options
2267 Takes either a scalar of raw SQL fields, or an array reference of
2268 field names, and adds on an SQL C<RETURNING> statement at the end.
2269 This allows you to return data generated by the insert statement
2270 (such as row IDs) without performing another C<SELECT> statement.
2271 Note, however, this is not part of the SQL standard and may not
2272 be supported by all database engines.
2276 =head2 update($table, \%fieldvals, \%where, \%options)
2278 This takes a table, hashref of field/value pairs, and an optional
2279 hashref L<WHERE clause|/WHERE CLAUSES>. It returns an SQL UPDATE function and a list
2281 See the sections on L</"Inserting and Updating Arrays"> and
2282 L</"Inserting and Updating SQL"> for information on how to insert
2283 with those data types.
2285 The optional C<\%options> hash reference may contain additional
2286 options to generate the update SQL. Currently supported options
2293 See the C<returning> option to
2294 L<insert|/insert($table, \@values || \%fieldvals, \%options)>.
2298 =head2 select($source, $fields, $where, $order)
2300 This returns a SQL SELECT statement and associated list of bind values, as
2301 specified by the arguments:
2307 Specification of the 'FROM' part of the statement.
2308 The argument can be either a plain scalar (interpreted as a table
2309 name, will be quoted), or an arrayref (interpreted as a list
2310 of table names, joined by commas, quoted), or a scalarref
2311 (literal SQL, not quoted).
2315 Specification of the list of fields to retrieve from
2317 The argument can be either an arrayref (interpreted as a list
2318 of field names, will be joined by commas and quoted), or a
2319 plain scalar (literal SQL, not quoted).
2320 Please observe that this API is not as flexible as that of
2321 the first argument C<$source>, for backwards compatibility reasons.
2325 Optional argument to specify the WHERE part of the query.
2326 The argument is most often a hashref, but can also be
2327 an arrayref or plain scalar --
2328 see section L<WHERE clause|/"WHERE CLAUSES"> for details.
2332 Optional argument to specify the ORDER BY part of the query.
2333 The argument can be a scalar, a hashref or an arrayref
2334 -- see section L<ORDER BY clause|/"ORDER BY CLAUSES">
2340 =head2 delete($table, \%where, \%options)
2342 This takes a table name and optional hashref L<WHERE clause|/WHERE CLAUSES>.
2343 It returns an SQL DELETE statement and list of bind values.
2345 The optional C<\%options> hash reference may contain additional
2346 options to generate the delete SQL. Currently supported options
2353 See the C<returning> option to
2354 L<insert|/insert($table, \@values || \%fieldvals, \%options)>.
2358 =head2 where(\%where, $order)
2360 This is used to generate just the WHERE clause. For example,
2361 if you have an arbitrary data structure and know what the
2362 rest of your SQL is going to look like, but want an easy way
2363 to produce a WHERE clause, use this. It returns an SQL WHERE
2364 clause and list of bind values.
2367 =head2 values(\%data)
2369 This just returns the values from the hash C<%data>, in the same
2370 order that would be returned from any of the other above queries.
2371 Using this allows you to markedly speed up your queries if you
2372 are affecting lots of rows. See below under the L</"PERFORMANCE"> section.
2374 =head2 generate($any, 'number', $of, \@data, $struct, \%types)
2376 Warning: This is an experimental method and subject to change.
2378 This returns arbitrarily generated SQL. It's a really basic shortcut.
2379 It will return two different things, depending on return context:
2381 my($stmt, @bind) = $sql->generate('create table', \$table, \@fields);
2382 my $stmt_and_val = $sql->generate('create table', \$table, \@fields);
2384 These would return the following:
2386 # First calling form
2387 $stmt = "CREATE TABLE test (?, ?)";
2388 @bind = (field1, field2);
2390 # Second calling form
2391 $stmt_and_val = "CREATE TABLE test (field1, field2)";
2393 Depending on what you're trying to do, it's up to you to choose the correct
2394 format. In this example, the second form is what you would want.
2398 $sql->generate('alter session', { nls_date_format => 'MM/YY' });
2402 ALTER SESSION SET nls_date_format = 'MM/YY'
2404 You get the idea. Strings get their case twiddled, but everything
2405 else remains verbatim.
2407 =head1 EXPORTABLE FUNCTIONS
2409 =head2 is_plain_value
2411 Determines if the supplied argument is a plain value as understood by this
2416 =item * The value is C<undef>
2418 =item * The value is a non-reference
2420 =item * The value is an object with stringification overloading
2422 =item * The value is of the form C<< { -value => $anything } >>
2426 On failure returns C<undef>, on success returns a B<scalar> reference
2427 to the original supplied argument.
2433 The stringification overloading detection is rather advanced: it takes
2434 into consideration not only the presence of a C<""> overload, but if that
2435 fails also checks for enabled
2436 L<autogenerated versions of C<"">|overload/Magic Autogeneration>, based
2437 on either C<0+> or C<bool>.
2439 Unfortunately testing in the field indicates that this
2440 detection B<< may tickle a latent bug in perl versions before 5.018 >>,
2441 but only when very large numbers of stringifying objects are involved.
2442 At the time of writing ( Sep 2014 ) there is no clear explanation of
2443 the direct cause, nor is there a manageably small test case that reliably
2444 reproduces the problem.
2446 If you encounter any of the following exceptions in B<random places within
2447 your application stack> - this module may be to blame:
2449 Operation "ne": no method found,
2450 left argument in overloaded package <something>,
2451 right argument in overloaded package <something>
2455 Stub found while resolving method "???" overloading """" in package <something>
2457 If you fall victim to the above - please attempt to reduce the problem
2458 to something that could be sent to the L<SQL::Abstract developers
2459 |DBIx::Class/GETTING HELP/SUPPORT>
2460 (either publicly or privately). As a workaround in the meantime you can
2461 set C<$ENV{SQLA_ISVALUE_IGNORE_AUTOGENERATED_STRINGIFICATION}> to a true
2462 value, which will most likely eliminate your problem (at the expense of
2463 not being able to properly detect exotic forms of stringification).
2465 This notice and environment variable will be removed in a future version,
2466 as soon as the underlying problem is found and a reliable workaround is
2471 =head2 is_literal_value
2473 Determines if the supplied argument is a literal value as understood by this
2478 =item * C<\$sql_string>
2480 =item * C<\[ $sql_string, @bind_values ]>
2484 On failure returns C<undef>, on success returns an B<array> reference
2485 containing the unpacked version of the supplied literal SQL and bind values.
2487 =head1 WHERE CLAUSES
2491 This module uses a variation on the idea from L<DBIx::Abstract>. It
2492 is B<NOT>, repeat I<not> 100% compatible. B<The main logic of this
2493 module is that things in arrays are OR'ed, and things in hashes
2496 The easiest way to explain is to show lots of examples. After
2497 each C<%where> hash shown, it is assumed you used:
2499 my($stmt, @bind) = $sql->where(\%where);
2501 However, note that the C<%where> hash can be used directly in any
2502 of the other functions as well, as described above.
2504 =head2 Key-value pairs
2506 So, let's get started. To begin, a simple hash:
2510 status => 'completed'
2513 Is converted to SQL C<key = val> statements:
2515 $stmt = "WHERE user = ? AND status = ?";
2516 @bind = ('nwiger', 'completed');
2518 One common thing I end up doing is having a list of values that
2519 a field can be in. To do this, simply specify a list inside of
2524 status => ['assigned', 'in-progress', 'pending'];
2527 This simple code will create the following:
2529 $stmt = "WHERE user = ? AND ( status = ? OR status = ? OR status = ? )";
2530 @bind = ('nwiger', 'assigned', 'in-progress', 'pending');
2532 A field associated to an empty arrayref will be considered a
2533 logical false and will generate 0=1.
2535 =head2 Tests for NULL values
2537 If the value part is C<undef> then this is converted to SQL <IS NULL>
2546 $stmt = "WHERE user = ? AND status IS NULL";
2549 To test if a column IS NOT NULL:
2553 status => { '!=', undef },
2556 =head2 Specific comparison operators
2558 If you want to specify a different type of operator for your comparison,
2559 you can use a hashref for a given column:
2563 status => { '!=', 'completed' }
2566 Which would generate:
2568 $stmt = "WHERE user = ? AND status != ?";
2569 @bind = ('nwiger', 'completed');
2571 To test against multiple values, just enclose the values in an arrayref:
2573 status => { '=', ['assigned', 'in-progress', 'pending'] };
2575 Which would give you:
2577 "WHERE status = ? OR status = ? OR status = ?"
2580 The hashref can also contain multiple pairs, in which case it is expanded
2581 into an C<AND> of its elements:
2585 status => { '!=', 'completed', -not_like => 'pending%' }
2588 # Or more dynamically, like from a form
2589 $where{user} = 'nwiger';
2590 $where{status}{'!='} = 'completed';
2591 $where{status}{'-not_like'} = 'pending%';
2593 # Both generate this
2594 $stmt = "WHERE user = ? AND status != ? AND status NOT LIKE ?";
2595 @bind = ('nwiger', 'completed', 'pending%');
2598 To get an OR instead, you can combine it with the arrayref idea:
2602 priority => [ { '=', 2 }, { '>', 5 } ]
2605 Which would generate:
2607 $stmt = "WHERE ( priority = ? OR priority > ? ) AND user = ?";
2608 @bind = ('2', '5', 'nwiger');
2610 If you want to include literal SQL (with or without bind values), just use a
2611 scalar reference or reference to an arrayref as the value:
2614 date_entered => { '>' => \["to_date(?, 'MM/DD/YYYY')", "11/26/2008"] },
2615 date_expires => { '<' => \"now()" }
2618 Which would generate:
2620 $stmt = "WHERE date_entered > to_date(?, 'MM/DD/YYYY') AND date_expires < now()";
2621 @bind = ('11/26/2008');
2624 =head2 Logic and nesting operators
2626 In the example above,
2627 there is a subtle trap if you want to say something like
2628 this (notice the C<AND>):
2630 WHERE priority != ? AND priority != ?
2632 Because, in Perl you I<can't> do this:
2634 priority => { '!=' => 2, '!=' => 1 }
2636 As the second C<!=> key will obliterate the first. The solution
2637 is to use the special C<-modifier> form inside an arrayref:
2639 priority => [ -and => {'!=', 2},
2643 Normally, these would be joined by C<OR>, but the modifier tells it
2644 to use C<AND> instead. (Hint: You can use this in conjunction with the
2645 C<logic> option to C<new()> in order to change the way your queries
2646 work by default.) B<Important:> Note that the C<-modifier> goes
2647 B<INSIDE> the arrayref, as an extra first element. This will
2648 B<NOT> do what you think it might:
2650 priority => -and => [{'!=', 2}, {'!=', 1}] # WRONG!
2652 Here is a quick list of equivalencies, since there is some overlap:
2655 status => {'!=', 'completed', 'not like', 'pending%' }
2656 status => [ -and => {'!=', 'completed'}, {'not like', 'pending%'}]
2659 status => {'=', ['assigned', 'in-progress']}
2660 status => [ -or => {'=', 'assigned'}, {'=', 'in-progress'}]
2661 status => [ {'=', 'assigned'}, {'=', 'in-progress'} ]
2665 =head2 Special operators: IN, BETWEEN, etc.
2667 You can also use the hashref format to compare a list of fields using the
2668 C<IN> comparison operator, by specifying the list as an arrayref:
2671 status => 'completed',
2672 reportid => { -in => [567, 2335, 2] }
2675 Which would generate:
2677 $stmt = "WHERE status = ? AND reportid IN (?,?,?)";
2678 @bind = ('completed', '567', '2335', '2');
2680 The reverse operator C<-not_in> generates SQL C<NOT IN> and is used in
2683 If the argument to C<-in> is an empty array, 'sqlfalse' is generated
2684 (by default: C<1=0>). Similarly, C<< -not_in => [] >> generates
2685 'sqltrue' (by default: C<1=1>).
2687 In addition to the array you can supply a chunk of literal sql or
2688 literal sql with bind:
2691 customer => { -in => \[
2692 'SELECT cust_id FROM cust WHERE balance > ?',
2695 status => { -in => \'SELECT status_codes FROM states' },
2701 customer IN ( SELECT cust_id FROM cust WHERE balance > ? )
2702 AND status IN ( SELECT status_codes FROM states )
2706 Finally, if the argument to C<-in> is not a reference, it will be
2707 treated as a single-element array.
2709 Another pair of operators is C<-between> and C<-not_between>,
2710 used with an arrayref of two values:
2714 completion_date => {
2715 -not_between => ['2002-10-01', '2003-02-06']
2721 WHERE user = ? AND completion_date NOT BETWEEN ( ? AND ? )
2723 Just like with C<-in> all plausible combinations of literal SQL
2727 start0 => { -between => [ 1, 2 ] },
2728 start1 => { -between => \["? AND ?", 1, 2] },
2729 start2 => { -between => \"lower(x) AND upper(y)" },
2730 start3 => { -between => [
2732 \["upper(?)", 'stuff' ],
2739 ( start0 BETWEEN ? AND ? )
2740 AND ( start1 BETWEEN ? AND ? )
2741 AND ( start2 BETWEEN lower(x) AND upper(y) )
2742 AND ( start3 BETWEEN lower(x) AND upper(?) )
2744 @bind = (1, 2, 1, 2, 'stuff');
2747 These are the two builtin "special operators"; but the
2748 list can be expanded: see section L</"SPECIAL OPERATORS"> below.
2750 =head2 Unary operators: bool
2752 If you wish to test against boolean columns or functions within your
2753 database you can use the C<-bool> and C<-not_bool> operators. For
2754 example to test the column C<is_user> being true and the column
2755 C<is_enabled> being false you would use:-
2759 -not_bool => 'is_enabled',
2764 WHERE is_user AND NOT is_enabled
2766 If a more complex combination is required, testing more conditions,
2767 then you should use the and/or operators:-
2772 -not_bool => { two=> { -rlike => 'bar' } },
2773 -not_bool => { three => [ { '=', 2 }, { '>', 5 } ] },
2784 (NOT ( three = ? OR three > ? ))
2787 =head2 Nested conditions, -and/-or prefixes
2789 So far, we've seen how multiple conditions are joined with a top-level
2790 C<AND>. We can change this by putting the different conditions we want in
2791 hashes and then putting those hashes in an array. For example:
2796 status => { -like => ['pending%', 'dispatched'] },
2800 status => 'unassigned',
2804 This data structure would create the following:
2806 $stmt = "WHERE ( user = ? AND ( status LIKE ? OR status LIKE ? ) )
2807 OR ( user = ? AND status = ? ) )";
2808 @bind = ('nwiger', 'pending', 'dispatched', 'robot', 'unassigned');
2811 Clauses in hashrefs or arrayrefs can be prefixed with an C<-and> or C<-or>
2812 to change the logic inside:
2818 -and => [ workhrs => {'>', 20}, geo => 'ASIA' ],
2819 -or => { workhrs => {'<', 50}, geo => 'EURO' },
2826 $stmt = "WHERE ( user = ?
2827 AND ( ( workhrs > ? AND geo = ? )
2828 OR ( workhrs < ? OR geo = ? ) ) )";
2829 @bind = ('nwiger', '20', 'ASIA', '50', 'EURO');
2831 =head3 Algebraic inconsistency, for historical reasons
2833 C<Important note>: when connecting several conditions, the C<-and->|C<-or>
2834 operator goes C<outside> of the nested structure; whereas when connecting
2835 several constraints on one column, the C<-and> operator goes
2836 C<inside> the arrayref. Here is an example combining both features:
2839 -and => [a => 1, b => 2],
2840 -or => [c => 3, d => 4],
2841 e => [-and => {-like => 'foo%'}, {-like => '%bar'} ]
2846 WHERE ( ( ( a = ? AND b = ? )
2847 OR ( c = ? OR d = ? )
2848 OR ( e LIKE ? AND e LIKE ? ) ) )
2850 This difference in syntax is unfortunate but must be preserved for
2851 historical reasons. So be careful: the two examples below would
2852 seem algebraically equivalent, but they are not
2855 { -like => 'foo%' },
2856 { -like => '%bar' },
2858 # yields: WHERE ( ( col LIKE ? AND col LIKE ? ) )
2861 { col => { -like => 'foo%' } },
2862 { col => { -like => '%bar' } },
2864 # yields: WHERE ( ( col LIKE ? OR col LIKE ? ) )
2867 =head2 Literal SQL and value type operators
2869 The basic premise of SQL::Abstract is that in WHERE specifications the "left
2870 side" is a column name and the "right side" is a value (normally rendered as
2871 a placeholder). This holds true for both hashrefs and arrayref pairs as you
2872 see in the L</WHERE CLAUSES> examples above. Sometimes it is necessary to
2873 alter this behavior. There are several ways of doing so.
2877 This is a virtual operator that signals the string to its right side is an
2878 identifier (a column name) and not a value. For example to compare two
2879 columns you would write:
2882 priority => { '<', 2 },
2883 requestor => { -ident => 'submitter' },
2888 $stmt = "WHERE priority < ? AND requestor = submitter";
2891 If you are maintaining legacy code you may see a different construct as
2892 described in L</Deprecated usage of Literal SQL>, please use C<-ident> in new
2897 This is a virtual operator that signals that the construct to its right side
2898 is a value to be passed to DBI. This is for example necessary when you want
2899 to write a where clause against an array (for RDBMS that support such
2900 datatypes). For example:
2903 array => { -value => [1, 2, 3] }
2908 $stmt = 'WHERE array = ?';
2909 @bind = ([1, 2, 3]);
2911 Note that if you were to simply say:
2917 the result would probably not be what you wanted:
2919 $stmt = 'WHERE array = ? OR array = ? OR array = ?';
2924 Finally, sometimes only literal SQL will do. To include a random snippet
2925 of SQL verbatim, you specify it as a scalar reference. Consider this only
2926 as a last resort. Usually there is a better way. For example:
2929 priority => { '<', 2 },
2930 requestor => { -in => \'(SELECT name FROM hitmen)' },
2935 $stmt = "WHERE priority < ? AND requestor IN (SELECT name FROM hitmen)"
2938 Note that in this example, you only get one bind parameter back, since
2939 the verbatim SQL is passed as part of the statement.
2943 Never use untrusted input as a literal SQL argument - this is a massive
2944 security risk (there is no way to check literal snippets for SQL
2945 injections and other nastyness). If you need to deal with untrusted input
2946 use literal SQL with placeholders as described next.
2948 =head3 Literal SQL with placeholders and bind values (subqueries)
2950 If the literal SQL to be inserted has placeholders and bind values,
2951 use a reference to an arrayref (yes this is a double reference --
2952 not so common, but perfectly legal Perl). For example, to find a date
2953 in Postgres you can use something like this:
2956 date_column => \[ "= date '2008-09-30' - ?::integer", 10 ]
2961 $stmt = "WHERE ( date_column = date '2008-09-30' - ?::integer )"
2964 Note that you must pass the bind values in the same format as they are returned
2965 by L<where|/where(\%where, $order)>. This means that if you set L</bindtype>
2966 to C<columns>, you must provide the bind values in the
2967 C<< [ column_meta => value ] >> format, where C<column_meta> is an opaque
2968 scalar value; most commonly the column name, but you can use any scalar value
2969 (including references and blessed references), L<SQL::Abstract> will simply
2970 pass it through intact. So if C<bindtype> is set to C<columns> the above
2971 example will look like:
2974 date_column => \[ "= date '2008-09-30' - ?::integer", [ {} => 10 ] ]
2977 Literal SQL is especially useful for nesting parenthesized clauses in the
2978 main SQL query. Here is a first example:
2980 my ($sub_stmt, @sub_bind) = ("SELECT c1 FROM t1 WHERE c2 < ? AND c3 LIKE ?",
2984 bar => \["IN ($sub_stmt)" => @sub_bind],
2989 $stmt = "WHERE (foo = ? AND bar IN (SELECT c1 FROM t1
2990 WHERE c2 < ? AND c3 LIKE ?))";
2991 @bind = (1234, 100, "foo%");
2993 Other subquery operators, like for example C<"E<gt> ALL"> or C<"NOT IN">,
2994 are expressed in the same way. Of course the C<$sub_stmt> and
2995 its associated bind values can be generated through a former call
2998 my ($sub_stmt, @sub_bind)
2999 = $sql->select("t1", "c1", {c2 => {"<" => 100},
3000 c3 => {-like => "foo%"}});
3003 bar => \["> ALL ($sub_stmt)" => @sub_bind],
3006 In the examples above, the subquery was used as an operator on a column;
3007 but the same principle also applies for a clause within the main C<%where>
3008 hash, like an EXISTS subquery:
3010 my ($sub_stmt, @sub_bind)
3011 = $sql->select("t1", "*", {c1 => 1, c2 => \"> t0.c0"});
3012 my %where = ( -and => [
3014 \["EXISTS ($sub_stmt)" => @sub_bind],
3019 $stmt = "WHERE (foo = ? AND EXISTS (SELECT * FROM t1
3020 WHERE c1 = ? AND c2 > t0.c0))";
3024 Observe that the condition on C<c2> in the subquery refers to
3025 column C<t0.c0> of the main query: this is I<not> a bind
3026 value, so we have to express it through a scalar ref.
3027 Writing C<< c2 => {">" => "t0.c0"} >> would have generated
3028 C<< c2 > ? >> with bind value C<"t0.c0"> ... not exactly
3029 what we wanted here.
3031 Finally, here is an example where a subquery is used
3032 for expressing unary negation:
3034 my ($sub_stmt, @sub_bind)
3035 = $sql->where({age => [{"<" => 10}, {">" => 20}]});
3036 $sub_stmt =~ s/^ where //i; # don't want "WHERE" in the subclause
3038 lname => {like => '%son%'},
3039 \["NOT ($sub_stmt)" => @sub_bind],
3044 $stmt = "lname LIKE ? AND NOT ( age < ? OR age > ? )"
3045 @bind = ('%son%', 10, 20)
3047 =head3 Deprecated usage of Literal SQL
3049 Below are some examples of archaic use of literal SQL. It is shown only as
3050 reference for those who deal with legacy code. Each example has a much
3051 better, cleaner and safer alternative that users should opt for in new code.
3057 my %where = ( requestor => \'IS NOT NULL' )
3059 $stmt = "WHERE requestor IS NOT NULL"
3061 This used to be the way of generating NULL comparisons, before the handling
3062 of C<undef> got formalized. For new code please use the superior syntax as
3063 described in L</Tests for NULL values>.
3067 my %where = ( requestor => \'= submitter' )
3069 $stmt = "WHERE requestor = submitter"
3071 This used to be the only way to compare columns. Use the superior L</-ident>
3072 method for all new code. For example an identifier declared in such a way
3073 will be properly quoted if L</quote_char> is properly set, while the legacy
3074 form will remain as supplied.
3078 my %where = ( is_ready => \"", completed => { '>', '2012-12-21' } )
3080 $stmt = "WHERE completed > ? AND is_ready"
3081 @bind = ('2012-12-21')
3083 Using an empty string literal used to be the only way to express a boolean.
3084 For all new code please use the much more readable
3085 L<-bool|/Unary operators: bool> operator.
3091 These pages could go on for a while, since the nesting of the data
3092 structures this module can handle are pretty much unlimited (the
3093 module implements the C<WHERE> expansion as a recursive function
3094 internally). Your best bet is to "play around" with the module a
3095 little to see how the data structures behave, and choose the best
3096 format for your data based on that.
3098 And of course, all the values above will probably be replaced with
3099 variables gotten from forms or the command line. After all, if you
3100 knew everything ahead of time, you wouldn't have to worry about
3101 dynamically-generating SQL and could just hardwire it into your
3104 =head1 ORDER BY CLAUSES
3106 Some functions take an order by clause. This can either be a scalar (just a
3107 column name), a hashref of C<< { -desc => 'col' } >> or C<< { -asc => 'col' }
3108 >>, a scalarref, an arrayref-ref, or an arrayref of any of the previous
3111 Given | Will Generate
3112 ---------------------------------------------------------------
3114 'colA' | ORDER BY colA
3116 [qw/colA colB/] | ORDER BY colA, colB
3118 {-asc => 'colA'} | ORDER BY colA ASC
3120 {-desc => 'colB'} | ORDER BY colB DESC
3122 ['colA', {-asc => 'colB'}] | ORDER BY colA, colB ASC
3124 { -asc => [qw/colA colB/] } | ORDER BY colA ASC, colB ASC
3126 \'colA DESC' | ORDER BY colA DESC
3128 \[ 'FUNC(colA, ?)', $x ] | ORDER BY FUNC(colA, ?)
3129 | /* ...with $x bound to ? */
3132 { -asc => 'colA' }, | colA ASC,
3133 { -desc => [qw/colB/] }, | colB DESC,
3134 { -asc => [qw/colC colD/] },| colC ASC, colD ASC,
3135 \'colE DESC', | colE DESC,
3136 \[ 'FUNC(colF, ?)', $x ], | FUNC(colF, ?)
3137 ] | /* ...with $x bound to ? */
3138 ===============================================================
3142 =head1 SPECIAL OPERATORS
3144 my $sqlmaker = SQL::Abstract->new(special_ops => [
3148 my ($self, $field, $op, $arg) = @_;
3154 handler => 'method_name',
3158 A "special operator" is a SQL syntactic clause that can be
3159 applied to a field, instead of a usual binary operator.
3162 WHERE field IN (?, ?, ?)
3163 WHERE field BETWEEN ? AND ?
3164 WHERE MATCH(field) AGAINST (?, ?)
3166 Special operators IN and BETWEEN are fairly standard and therefore
3167 are builtin within C<SQL::Abstract> (as the overridable methods
3168 C<_where_field_IN> and C<_where_field_BETWEEN>). For other operators,
3169 like the MATCH .. AGAINST example above which is specific to MySQL,
3170 you can write your own operator handlers - supply a C<special_ops>
3171 argument to the C<new> method. That argument takes an arrayref of
3172 operator definitions; each operator definition is a hashref with two
3179 the regular expression to match the operator
3183 Either a coderef or a plain scalar method name. In both cases
3184 the expected return is C<< ($sql, @bind) >>.
3186 When supplied with a method name, it is simply called on the
3187 L<SQL::Abstract> object as:
3189 $self->$method_name($field, $op, $arg)
3193 $field is the LHS of the operator
3194 $op is the part that matched the handler regex
3197 When supplied with a coderef, it is called as:
3199 $coderef->($self, $field, $op, $arg)
3204 For example, here is an implementation
3205 of the MATCH .. AGAINST syntax for MySQL
3207 my $sqlmaker = SQL::Abstract->new(special_ops => [
3209 # special op for MySql MATCH (field) AGAINST(word1, word2, ...)
3210 {regex => qr/^match$/i,
3212 my ($self, $field, $op, $arg) = @_;
3213 $arg = [$arg] if not ref $arg;
3214 my $label = $self->_quote($field);
3215 my ($placeholder) = $self->_convert('?');
3216 my $placeholders = join ", ", (($placeholder) x @$arg);
3217 my $sql = $self->_sqlcase('match') . " ($label) "
3218 . $self->_sqlcase('against') . " ($placeholders) ";
3219 my @bind = $self->_bindtype($field, @$arg);
3220 return ($sql, @bind);
3227 =head1 UNARY OPERATORS
3229 my $sqlmaker = SQL::Abstract->new(unary_ops => [
3233 my ($self, $op, $arg) = @_;
3239 handler => 'method_name',
3243 A "unary operator" is a SQL syntactic clause that can be
3244 applied to a field - the operator goes before the field
3246 You can write your own operator handlers - supply a C<unary_ops>
3247 argument to the C<new> method. That argument takes an arrayref of
3248 operator definitions; each operator definition is a hashref with two
3255 the regular expression to match the operator
3259 Either a coderef or a plain scalar method name. In both cases
3260 the expected return is C<< $sql >>.
3262 When supplied with a method name, it is simply called on the
3263 L<SQL::Abstract> object as:
3265 $self->$method_name($op, $arg)
3269 $op is the part that matched the handler regex
3270 $arg is the RHS or argument of the operator
3272 When supplied with a coderef, it is called as:
3274 $coderef->($self, $op, $arg)
3282 Thanks to some benchmarking by Mark Stosberg, it turns out that
3283 this module is many orders of magnitude faster than using C<DBIx::Abstract>.
3284 I must admit this wasn't an intentional design issue, but it's a
3285 byproduct of the fact that you get to control your C<DBI> handles
3288 To maximize performance, use a code snippet like the following:
3290 # prepare a statement handle using the first row
3291 # and then reuse it for the rest of the rows
3293 for my $href (@array_of_hashrefs) {
3294 $stmt ||= $sql->insert('table', $href);
3295 $sth ||= $dbh->prepare($stmt);
3296 $sth->execute($sql->values($href));
3299 The reason this works is because the keys in your C<$href> are sorted
3300 internally by B<SQL::Abstract>. Thus, as long as your data retains
3301 the same structure, you only have to generate the SQL the first time
3302 around. On subsequent queries, simply use the C<values> function provided
3303 by this module to return your values in the correct order.
3305 However this depends on the values having the same type - if, for
3306 example, the values of a where clause may either have values
3307 (resulting in sql of the form C<column = ?> with a single bind
3308 value), or alternatively the values might be C<undef> (resulting in
3309 sql of the form C<column IS NULL> with no bind value) then the
3310 caching technique suggested will not work.
3314 If you use my C<CGI::FormBuilder> module at all, you'll hopefully
3315 really like this part (I do, at least). Building up a complex query
3316 can be as simple as the following:
3323 use CGI::FormBuilder;
3326 my $form = CGI::FormBuilder->new(...);
3327 my $sql = SQL::Abstract->new;
3329 if ($form->submitted) {
3330 my $field = $form->field;
3331 my $id = delete $field->{id};
3332 my($stmt, @bind) = $sql->update('table', $field, {id => $id});
3335 Of course, you would still have to connect using C<DBI> to run the
3336 query, but the point is that if you make your form look like your
3337 table, the actual query script can be extremely simplistic.
3339 If you're B<REALLY> lazy (I am), check out C<HTML::QuickTable> for
3340 a fast interface to returning and formatting data. I frequently
3341 use these three modules together to write complex database query
3342 apps in under 50 lines.
3344 =head1 HOW TO CONTRIBUTE
3346 Contributions are always welcome, in all usable forms (we especially
3347 welcome documentation improvements). The delivery methods include git-
3348 or unified-diff formatted patches, GitHub pull requests, or plain bug
3349 reports either via RT or the Mailing list. Contributors are generally
3350 granted full access to the official repository after their first several
3351 patches pass successful review.
3353 This project is maintained in a git repository. The code and related tools are
3354 accessible at the following locations:
3358 =item * Official repo: L<git://git.shadowcat.co.uk/dbsrgits/SQL-Abstract.git>
3360 =item * Official gitweb: L<http://git.shadowcat.co.uk/gitweb/gitweb.cgi?p=dbsrgits/SQL-Abstract.git>
3362 =item * GitHub mirror: L<https://github.com/dbsrgits/sql-abstract>
3364 =item * Authorized committers: L<ssh://dbsrgits@git.shadowcat.co.uk/SQL-Abstract.git>
3370 Version 1.50 was a major internal refactoring of C<SQL::Abstract>.
3371 Great care has been taken to preserve the I<published> behavior
3372 documented in previous versions in the 1.* family; however,
3373 some features that were previously undocumented, or behaved
3374 differently from the documentation, had to be changed in order
3375 to clarify the semantics. Hence, client code that was relying
3376 on some dark areas of C<SQL::Abstract> v1.*
3377 B<might behave differently> in v1.50.
3379 The main changes are:
3385 support for literal SQL through the C<< \ [ $sql, @bind ] >> syntax.
3389 support for the { operator => \"..." } construct (to embed literal SQL)
3393 support for the { operator => \["...", @bind] } construct (to embed literal SQL with bind values)
3397 optional support for L<array datatypes|/"Inserting and Updating Arrays">
3401 defensive programming: check arguments
3405 fixed bug with global logic, which was previously implemented
3406 through global variables yielding side-effects. Prior versions would
3407 interpret C<< [ {cond1, cond2}, [cond3, cond4] ] >>
3408 as C<< "(cond1 AND cond2) OR (cond3 AND cond4)" >>.
3409 Now this is interpreted
3410 as C<< "(cond1 AND cond2) OR (cond3 OR cond4)" >>.
3415 fixed semantics of _bindtype on array args
3419 dropped the C<_anoncopy> of the %where tree. No longer necessary,
3420 we just avoid shifting arrays within that tree.
3424 dropped the C<_modlogic> function
3428 =head1 ACKNOWLEDGEMENTS
3430 There are a number of individuals that have really helped out with
3431 this module. Unfortunately, most of them submitted bugs via CPAN
3432 so I have no idea who they are! But the people I do know are:
3434 Ash Berlin (order_by hash term support)
3435 Matt Trout (DBIx::Class support)
3436 Mark Stosberg (benchmarking)
3437 Chas Owens (initial "IN" operator support)
3438 Philip Collins (per-field SQL functions)
3439 Eric Kolve (hashref "AND" support)
3440 Mike Fragassi (enhancements to "BETWEEN" and "LIKE")
3441 Dan Kubb (support for "quote_char" and "name_sep")
3442 Guillermo Roditi (patch to cleanup "IN" and "BETWEEN", fix and tests for _order_by)
3443 Laurent Dami (internal refactoring, extensible list of special operators, literal SQL)
3444 Norbert Buchmuller (support for literal SQL in hashpair, misc. fixes & tests)
3445 Peter Rabbitson (rewrite of SQLA::Test, misc. fixes & tests)
3446 Oliver Charles (support for "RETURNING" after "INSERT")
3452 L<DBIx::Class>, L<DBIx::Abstract>, L<CGI::FormBuilder>, L<HTML::QuickTable>.
3456 Copyright (c) 2001-2007 Nathan Wiger <nwiger@cpan.org>. All Rights Reserved.
3458 This module is actively maintained by Matt Trout <mst@shadowcatsystems.co.uk>
3460 For support, your best bet is to try the C<DBIx::Class> users mailing list.
3461 While not an official support venue, C<DBIx::Class> makes heavy use of
3462 C<SQL::Abstract>, and as such list members there are very familiar with
3463 how to create queries.
3467 This module is free software; you may copy this under the same
3468 terms as perl itself (either the GNU General Public License or
3469 the Artistic License)