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 { '-'.$logic => \@res };
582 sub _expand_expr_hashpair {
583 my ($self, $k, $v, $logic) = @_;
584 unless (defined($k) and length($k)) {
585 if (defined($k) and is_literal_value($v)) {
586 belch 'Hash-pairs consisting of an empty string with a literal are deprecated, and will be removed in 2.0: use -and => [ $literal ] instead';
589 puke "Supplying an empty left hand side argument is not supported";
593 return $self->_expand_expr($v);
597 return $self->_expand_expr($v);
599 puke "-bool => undef not supported" unless defined($v);
600 return { -ident => $v };
602 if (my ($rest) = $k =~/^-not[_ ](.*)$/) {
603 return $self->_expand_expr({ -not => { "-${rest}", $v } }, $logic);
606 unless (defined($v)) {
607 my $orig_op = my $op = $self->{cmp};
609 $op =~ /^not$/i ? 'is not' # legacy
610 : $op =~ $self->{equality_op} ? 'is'
611 : $op =~ $self->{like_op} ? belch("Supplying an undefined argument to '@{[ uc $op]}' is deprecated") && 'is'
612 : $op =~ $self->{inequality_op} ? 'is not'
613 : $op =~ $self->{not_like_op} ? belch("Supplying an undefined argument to '@{[ uc $op]}' is deprecated") && 'is not'
614 : puke "unexpected operator '$orig_op' with undef operand";
615 return +{ -op => [ $is.' null', { -ident => $k } ] };
622 { -bind => [ $k, $v ] }
626 if (ref($v) eq 'HASH' and keys %$v > 1) {
628 map $self->_expand_expr_hashpair($k => { $_ => $v->{$_} }),
632 if (ref($v) eq 'ARRAY') {
633 return $self->{sqlfalse} unless @$v;
634 $self->_debug("ARRAY($k) means distribute over elements");
636 $v->[0] =~ /^-((?:and|or))$/i
637 ? ($v = [ @{$v}[1..$#$v] ], $1)
638 : ($self->{logic} || 'or')
640 return +{ "-${this_logic}" => [ map $self->_expand_expr({ $k => $_ }, $this_logic), @$v ] };
642 if (my $literal = is_literal_value($v)) {
644 belch 'Hash-pairs consisting of an empty string with a literal are deprecated, and will be removed in 2.0: use -and => [ $literal ] instead';
647 my ($sql, @bind) = @$literal;
648 if ($self->{bindtype} eq 'columns') {
650 if (!defined $_ || ref($_) ne 'ARRAY' || @$_ != 2) {
651 puke "bindtype 'columns' selected, you need to pass: [column_name => bind_value]"
655 return +{ -literal => [ $self->_quote($k).' '.$sql, @bind ] };
662 my ($self, $where, $logic) = @_;
664 my $where_exp = $self->_expand_expr($where, $logic);
666 # dispatch on appropriate method according to refkind of $where
667 my $method = $self->_METHOD_FOR_refkind("_where", $where_exp);
669 my ($sql, @bind) = $self->$method($where_exp, $logic);
671 # DBIx::Class used to call _recurse_where in scalar context
672 # something else might too...
674 return ($sql, @bind);
677 belch "Calling _recurse_where in scalar context is deprecated and will go away before 2.0";
684 #======================================================================
685 # WHERE: top-level ARRAYREF
686 #======================================================================
689 sub _where_ARRAYREF {
690 my ($self, $where, $logic) = @_;
692 $logic = uc($logic || $self->{logic});
693 $logic eq 'AND' or $logic eq 'OR' or puke "unknown logic: $logic";
695 my @clauses = @$where;
697 my (@sql_clauses, @all_bind);
698 # need to use while() so can shift() for pairs
700 my $el = shift @clauses;
702 $el = undef if (defined $el and ! length $el);
704 # switch according to kind of $el and get corresponding ($sql, @bind)
705 my ($sql, @bind) = $self->_SWITCH_refkind($el, {
707 # skip empty elements, otherwise get invalid trailing AND stuff
708 ARRAYREF => sub {$self->_recurse_where($el) if @$el},
712 $self->_assert_bindval_matches_bindtype(@b);
716 HASHREF => sub {$self->_recurse_where($el, 'and') if %$el},
718 SCALARREF => sub { ($$el); },
721 # top-level arrayref with scalars, recurse in pairs
722 $self->_recurse_where({$el => shift(@clauses)})
725 UNDEF => sub {puke "Supplying an empty left hand side argument is not supported in array-pairs" },
729 push @sql_clauses, $sql;
730 push @all_bind, @bind;
734 return $self->_join_sql_clauses($logic, \@sql_clauses, \@all_bind);
737 #======================================================================
738 # WHERE: top-level ARRAYREFREF
739 #======================================================================
741 sub _where_ARRAYREFREF {
742 my ($self, $where) = @_;
743 my ($sql, @bind) = @$$where;
744 $self->_assert_bindval_matches_bindtype(@bind);
745 return ($sql, @bind);
748 #======================================================================
749 # WHERE: top-level HASHREF
750 #======================================================================
753 my ($self, $where) = @_;
754 my (@sql_clauses, @all_bind);
756 for my $k (sort keys %$where) {
757 my $v = $where->{$k};
759 # ($k => $v) is either a special unary op or a regular hashpair
760 my ($sql, @bind) = do {
762 # put the operator in canonical form
764 $op = substr $op, 1; # remove initial dash
765 $op =~ s/^\s+|\s+$//g;# remove leading/trailing space
766 $op =~ s/\s+/ /g; # compress whitespace
768 # so that -not_foo works correctly
769 $op =~ s/^not_/NOT /i;
771 $self->_debug("Unary OP(-$op) within hashref, recursing...");
772 my ($s, @b) = $self->_where_unary_op($op, $v);
774 # top level vs nested
775 # we assume that handled unary ops will take care of their ()s
777 List::Util::first {$op =~ $_->{regex}} @{$self->{unary_ops}}
779 ( defined $self->{_nested_func_lhs} and $self->{_nested_func_lhs} eq $k )
785 if (is_literal_value ($v) ) {
786 belch 'Hash-pairs consisting of an empty string with a literal are deprecated, and will be removed in 2.0: use -and => [ $literal ] instead';
789 puke "Supplying an empty left hand side argument is not supported in hash-pairs";
793 my $method = $self->_METHOD_FOR_refkind("_where_hashpair", $v);
794 $self->$method($k, $v);
798 push @sql_clauses, $sql;
799 push @all_bind, @bind;
802 return $self->_join_sql_clauses('and', \@sql_clauses, \@all_bind);
805 sub _where_unary_op {
806 my ($self, $op, $rhs) = @_;
808 # top level special ops are illegal in general
809 puke "Illegal use of top-level '-$op'"
810 if !(defined $self->{_nested_func_lhs})
811 and List::Util::first { $op =~ $_->{regex} } @{$self->{special_ops}}
812 and not List::Util::first { $op =~ $_->{regex} } @{$self->{unary_ops}};
814 if (my $op_entry = List::Util::first { $op =~ $_->{regex} } @{$self->{unary_ops}}) {
815 my $handler = $op_entry->{handler};
817 if (not ref $handler) {
818 if ($op =~ s/ [_\s]? \d+ $//x ) {
819 belch 'Use of [and|or|nest]_N modifiers is deprecated and will be removed in SQLA v2.0. '
820 . "You probably wanted ...-and => [ -$op => COND1, -$op => COND2 ... ]";
822 return $self->$handler($op, $rhs);
824 elsif (ref $handler eq 'CODE') {
825 return $handler->($self, $op, $rhs);
828 puke "Illegal handler for operator $op - expecting a method name or a coderef";
832 $self->_debug("Generic unary OP: $op - recursing as function");
834 $self->_assert_pass_injection_guard($op);
836 my ($sql, @bind) = $self->_SWITCH_refkind($rhs, {
838 puke "Illegal use of top-level '-$op'"
839 unless defined $self->{_nested_func_lhs};
842 $self->_convert('?'),
843 $self->_bindtype($self->{_nested_func_lhs}, $rhs)
847 $self->_recurse_where($rhs)
851 $sql = sprintf('%s %s',
852 $self->_sqlcase($op),
856 return ($sql, @bind);
859 sub _where_op_ANDOR {
860 my ($self, $op, $v) = @_;
862 $self->_SWITCH_refkind($v, {
864 return $self->_where_ARRAYREF($v, $op);
868 return ($op =~ /^or/i)
869 ? $self->_where_ARRAYREF([ map { $_ => $v->{$_} } (sort keys %$v) ], $op)
870 : $self->_where_HASHREF($v);
874 puke "-$op => \\\$scalar makes little sense, use " .
876 ? '[ \$scalar, \%rest_of_conditions ] instead'
877 : '-and => [ \$scalar, \%rest_of_conditions ] instead'
882 puke "-$op => \\[...] makes little sense, use " .
884 ? '[ \[...], \%rest_of_conditions ] instead'
885 : '-and => [ \[...], \%rest_of_conditions ] instead'
889 SCALAR => sub { # permissively interpreted as SQL
890 puke "-$op => \$value makes little sense, use -bool => \$value instead";
894 puke "-$op => undef not supported";
900 my ($self, $op, $v) = @_;
902 $self->_SWITCH_refkind($v, {
904 SCALAR => sub { # permissively interpreted as SQL
905 belch "literal SQL should be -nest => \\'scalar' "
906 . "instead of -nest => 'scalar' ";
911 puke "-$op => undef not supported";
915 $self->_recurse_where($v);
923 my ($self, $op, $v) = @_;
925 my ($s, @b) = $self->_SWITCH_refkind($v, {
926 SCALAR => sub { # interpreted as SQL column
927 $self->_convert($self->_quote($v));
931 puke "-$op => undef not supported";
935 $self->_recurse_where($v);
939 $s = "(NOT $s)" if $op =~ /^not/i;
944 sub _where_op_IDENT {
946 my ($op, $rhs) = splice @_, -2;
947 if (! defined $rhs or length ref $rhs) {
948 puke "-$op requires a single plain scalar argument (a quotable identifier)";
951 # in case we are called as a top level special op (no '=')
952 my $has_lhs = my $lhs = shift;
954 $_ = $self->_convert($self->_quote($_)) for ($lhs, $rhs);
962 sub _where_op_VALUE {
964 my ($op, $rhs) = splice @_, -2;
966 # in case we are called as a top level special op (no '=')
970 if (! defined $rhs) {
972 ? $self->_where_hashpair_HASHREF($lhs, { -is => undef })
979 (defined $lhs ? $lhs : $self->{_nested_func_lhs}),
986 $self->_convert($self->_quote($lhs)) . ' = ' . $self->_convert('?'),
990 $self->_convert('?'),
997 my %unop_postfix = map +($_ => 1), 'is null', 'is not null';
1000 my ($self, undef, $v) = @_;
1001 my ($op, @args) = @$v;
1002 $op =~ s/^-// if length($op) > 1;
1003 local $self->{_nested_func_lhs};
1005 my ($expr_sql, @bind) = $self->_recurse_where($args[0]);
1006 my $final_op = join ' ', split '_', $op;
1007 my $op_sql = $self->_sqlcase($final_op);
1009 $unop_postfix{lc($final_op)}
1010 ? "${expr_sql} ${op_sql}"
1011 : "${op_sql} ${expr_sql}"
1013 return ($final_sql, @bind);
1014 } elsif (@args == 2) {
1015 my ($l, $r) = map [ $self->_recurse_where($_) ], @args;
1016 return ( $l->[0].' '.$self->_sqlcase(join ' ', split '_', $op).' '.$r->[0], @{$l}[1..$#$l], @{$r}[1..$#$r] );
1021 sub _where_op_BIND {
1022 my ($self, undef, $bind) = @_;
1023 return ($self->_convert('?'), $self->_bindtype(@$bind));
1026 sub _where_op_LITERAL {
1027 my ($self, undef, $literal) = @_;
1031 sub _where_hashpair_ARRAYREF {
1032 my ($self, $k, $v) = @_;
1035 my @v = @$v; # need copy because of shift below
1036 $self->_debug("ARRAY($k) means distribute over elements");
1038 # put apart first element if it is an operator (-and, -or)
1040 (defined $v[0] && $v[0] =~ /^ - (?: AND|OR ) $/ix)
1044 my @distributed = map { {$k => $_} } @v;
1047 $self->_debug("OP($op) reinjected into the distributed array");
1048 unshift @distributed, $op;
1051 my $logic = $op ? substr($op, 1) : '';
1053 return $self->_recurse_where(\@distributed, $logic);
1056 $self->_debug("empty ARRAY($k) means 0=1");
1057 return ($self->{sqlfalse});
1061 sub _where_hashpair_HASHREF {
1062 my ($self, $k, $v, $logic) = @_;
1065 local $self->{_nested_func_lhs} = defined $self->{_nested_func_lhs}
1066 ? $self->{_nested_func_lhs}
1070 my ($all_sql, @all_bind);
1072 for my $orig_op (sort keys %$v) {
1073 my $val = $v->{$orig_op};
1075 # put the operator in canonical form
1078 # FIXME - we need to phase out dash-less ops
1079 $op =~ s/^-//; # remove possible initial dash
1080 $op =~ s/^\s+|\s+$//g;# remove leading/trailing space
1081 $op =~ s/\s+/ /g; # compress whitespace
1083 $self->_assert_pass_injection_guard($op);
1086 $op =~ s/^is_not/IS NOT/i;
1088 # so that -not_foo works correctly
1089 $op =~ s/^not_/NOT /i;
1091 # another retarded special case: foo => { $op => { -value => undef } }
1092 if (ref $val eq 'HASH' and keys %$val == 1 and exists $val->{-value} and ! defined $val->{-value} ) {
1098 # CASE: col-value logic modifiers
1099 if ($orig_op =~ /^ \- (and|or) $/xi) {
1100 ($sql, @bind) = $self->_where_hashpair_HASHREF($k, $val, $1);
1102 # CASE: special operators like -in or -between
1103 elsif (my $special_op = List::Util::first { $op =~ $_->{regex} } @{$self->{special_ops}}) {
1104 my $handler = $special_op->{handler};
1106 puke "No handler supplied for special operator $orig_op";
1108 elsif (not ref $handler) {
1109 ($sql, @bind) = $self->$handler($k, $op, $val);
1111 elsif (ref $handler eq 'CODE') {
1112 ($sql, @bind) = $handler->($self, $k, $op, $val);
1115 puke "Illegal handler for special operator $orig_op - expecting a method name or a coderef";
1119 $self->_SWITCH_refkind($val, {
1121 ARRAYREF => sub { # CASE: col => {op => \@vals}
1122 ($sql, @bind) = $self->_where_field_op_ARRAYREF($k, $op, $val);
1125 ARRAYREFREF => sub { # CASE: col => {op => \[$sql, @bind]} (literal SQL with bind)
1126 my ($sub_sql, @sub_bind) = @$$val;
1127 $self->_assert_bindval_matches_bindtype(@sub_bind);
1128 $sql = join ' ', $self->_convert($self->_quote($k)),
1129 $self->_sqlcase($op),
1134 UNDEF => sub { # CASE: col => {op => undef} : sql "IS (NOT)? NULL"
1136 $op =~ /^not$/i ? 'is not' # legacy
1137 : $op =~ $self->{equality_op} ? 'is'
1138 : $op =~ $self->{like_op} ? belch("Supplying an undefined argument to '@{[ uc $op]}' is deprecated") && 'is'
1139 : $op =~ $self->{inequality_op} ? 'is not'
1140 : $op =~ $self->{not_like_op} ? belch("Supplying an undefined argument to '@{[ uc $op]}' is deprecated") && 'is not'
1141 : puke "unexpected operator '$orig_op' with undef operand";
1143 $sql = $self->_quote($k) . $self->_sqlcase(" $is null");
1146 FALLBACK => sub { # CASE: col => {op/func => $stuff}
1147 ($sql, @bind) = $self->_where_unary_op($op, $val);
1150 $self->_convert($self->_quote($k)),
1151 $self->{_nested_func_lhs} eq $k ? $sql : "($sql)", # top level vs nested
1157 ($all_sql) = (defined $all_sql and $all_sql) ? $self->_join_sql_clauses($logic, [$all_sql, $sql], []) : $sql;
1158 push @all_bind, @bind;
1160 return ($all_sql, @all_bind);
1163 sub _where_field_IS {
1164 my ($self, $k, $op, $v) = @_;
1166 my ($s) = $self->_SWITCH_refkind($v, {
1169 $self->_convert($self->_quote($k)),
1170 map { $self->_sqlcase($_)} ($op, 'null')
1173 puke "$op can only take undef as argument";
1180 sub _where_field_op_ARRAYREF {
1181 my ($self, $k, $op, $vals) = @_;
1183 my @vals = @$vals; #always work on a copy
1186 $self->_debug(sprintf '%s means multiple elements: [ %s ]',
1188 join(', ', map { defined $_ ? "'$_'" : 'NULL' } @vals ),
1191 # see if the first element is an -and/-or op
1193 if (defined $vals[0] && $vals[0] =~ /^ - (AND|OR) $/ix) {
1198 # a long standing API wart - an attempt to change this behavior during
1199 # the 1.50 series failed *spectacularly*. Warn instead and leave the
1204 (!$logic or $logic eq 'OR')
1206 ($op =~ $self->{inequality_op} or $op =~ $self->{not_like_op})
1209 belch "A multi-element arrayref as an argument to the inequality op '$o' "
1210 . 'is technically equivalent to an always-true 1=1 (you probably wanted '
1211 . "to say ...{ \$inequality_op => [ -and => \@values ] }... instead)"
1215 # distribute $op over each remaining member of @vals, append logic if exists
1216 return $self->_recurse_where([map { {$k => {$op, $_}} } @vals], $logic);
1220 # try to DWIM on equality operators
1222 $op =~ $self->{equality_op} ? $self->{sqlfalse}
1223 : $op =~ $self->{like_op} ? belch("Supplying an empty arrayref to '@{[ uc $op]}' is deprecated") && $self->{sqlfalse}
1224 : $op =~ $self->{inequality_op} ? $self->{sqltrue}
1225 : $op =~ $self->{not_like_op} ? belch("Supplying an empty arrayref to '@{[ uc $op]}' is deprecated") && $self->{sqltrue}
1226 : puke "operator '$op' applied on an empty array (field '$k')";
1231 sub _where_hashpair_SCALARREF {
1232 my ($self, $k, $v) = @_;
1233 $self->_debug("SCALAR($k) means literal SQL: $$v");
1234 my $sql = $self->_quote($k) . " " . $$v;
1238 # literal SQL with bind
1239 sub _where_hashpair_ARRAYREFREF {
1240 my ($self, $k, $v) = @_;
1241 $self->_debug("REF($k) means literal SQL: @${$v}");
1242 my ($sql, @bind) = @$$v;
1243 $self->_assert_bindval_matches_bindtype(@bind);
1244 $sql = $self->_quote($k) . " " . $sql;
1245 return ($sql, @bind );
1248 # literal SQL without bind
1249 sub _where_hashpair_SCALAR {
1250 my ($self, $k, $v) = @_;
1251 $self->_debug("NOREF($k) means simple key=val: $k $self->{cmp} $v");
1252 return ($self->_where_hashpair_HASHREF($k, { $self->{cmp} => $v }));
1256 sub _where_hashpair_UNDEF {
1257 my ($self, $k, $v) = @_;
1258 $self->_debug("UNDEF($k) means IS NULL");
1259 return $self->_where_hashpair_HASHREF($k, { -is => undef });
1262 #======================================================================
1263 # WHERE: TOP-LEVEL OTHERS (SCALARREF, SCALAR, UNDEF)
1264 #======================================================================
1267 sub _where_SCALARREF {
1268 my ($self, $where) = @_;
1271 $self->_debug("SCALAR(*top) means literal SQL: $$where");
1277 my ($self, $where) = @_;
1280 $self->_debug("NOREF(*top) means literal SQL: $where");
1291 #======================================================================
1292 # WHERE: BUILTIN SPECIAL OPERATORS (-in, -between)
1293 #======================================================================
1296 sub _where_field_BETWEEN {
1297 my ($self, $k, $op, $vals) = @_;
1299 my ($label, $and, $placeholder);
1300 $label = $self->_convert($self->_quote($k));
1301 $and = ' ' . $self->_sqlcase('and') . ' ';
1302 $placeholder = $self->_convert('?');
1303 $op = $self->_sqlcase($op);
1305 my $invalid_args = "Operator '$op' requires either an arrayref with two defined values or expressions, or a single literal scalarref/arrayref-ref";
1307 my ($clause, @bind) = $self->_SWITCH_refkind($vals, {
1308 ARRAYREFREF => sub {
1309 my ($s, @b) = @$$vals;
1310 $self->_assert_bindval_matches_bindtype(@b);
1317 puke $invalid_args if @$vals != 2;
1319 my (@all_sql, @all_bind);
1320 foreach my $val (@$vals) {
1321 my ($sql, @bind) = $self->_SWITCH_refkind($val, {
1323 return ($placeholder, $self->_bindtype($k, $val) );
1328 ARRAYREFREF => sub {
1329 my ($sql, @bind) = @$$val;
1330 $self->_assert_bindval_matches_bindtype(@bind);
1331 return ($sql, @bind);
1334 my ($func, $arg, @rest) = %$val;
1335 puke "Only simple { -func => arg } functions accepted as sub-arguments to BETWEEN"
1336 if (@rest or $func !~ /^ \- (.+)/x);
1337 $self->_where_unary_op($1 => $arg);
1343 push @all_sql, $sql;
1344 push @all_bind, @bind;
1348 (join $and, @all_sql),
1357 my $sql = "( $label $op $clause )";
1358 return ($sql, @bind)
1362 sub _where_field_IN {
1363 my ($self, $k, $op, $vals) = @_;
1365 # backwards compatibility: if scalar, force into an arrayref
1366 $vals = [$vals] if defined $vals && ! ref $vals;
1368 my ($label) = $self->_convert($self->_quote($k));
1369 my ($placeholder) = $self->_convert('?');
1370 $op = $self->_sqlcase($op);
1372 my ($sql, @bind) = $self->_SWITCH_refkind($vals, {
1373 ARRAYREF => sub { # list of choices
1374 if (@$vals) { # nonempty list
1375 my (@all_sql, @all_bind);
1377 for my $val (@$vals) {
1378 my ($sql, @bind) = $self->_SWITCH_refkind($val, {
1380 return ($placeholder, $val);
1385 ARRAYREFREF => sub {
1386 my ($sql, @bind) = @$$val;
1387 $self->_assert_bindval_matches_bindtype(@bind);
1388 return ($sql, @bind);
1391 my ($func, $arg, @rest) = %$val;
1392 puke "Only simple { -func => arg } functions accepted as sub-arguments to IN"
1393 if (@rest or $func !~ /^ \- (.+)/x);
1394 $self->_where_unary_op($1 => $arg);
1398 'SQL::Abstract before v1.75 used to generate incorrect SQL when the '
1399 . "-$op operator was given an undef-containing list: !!!AUDIT YOUR CODE "
1400 . 'AND DATA!!! (the upcoming Data::Query-based version of SQL::Abstract '
1401 . 'will emit the logically correct SQL instead of raising this exception)'
1405 push @all_sql, $sql;
1406 push @all_bind, @bind;
1410 sprintf('%s %s ( %s )',
1413 join(', ', @all_sql)
1415 $self->_bindtype($k, @all_bind),
1418 else { # empty list: some databases won't understand "IN ()", so DWIM
1419 my $sql = ($op =~ /\bnot\b/i) ? $self->{sqltrue} : $self->{sqlfalse};
1424 SCALARREF => sub { # literal SQL
1425 my $sql = $self->_open_outer_paren($$vals);
1426 return ("$label $op ( $sql )");
1428 ARRAYREFREF => sub { # literal SQL with bind
1429 my ($sql, @bind) = @$$vals;
1430 $self->_assert_bindval_matches_bindtype(@bind);
1431 $sql = $self->_open_outer_paren($sql);
1432 return ("$label $op ( $sql )", @bind);
1436 puke "Argument passed to the '$op' operator can not be undefined";
1440 puke "special op $op requires an arrayref (or scalarref/arrayref-ref)";
1444 return ($sql, @bind);
1447 # Some databases (SQLite) treat col IN (1, 2) different from
1448 # col IN ( (1, 2) ). Use this to strip all outer parens while
1449 # adding them back in the corresponding method
1450 sub _open_outer_paren {
1451 my ($self, $sql) = @_;
1453 while (my ($inner) = $sql =~ /^ \s* \( (.*) \) \s* $/xs) {
1455 # there are closing parens inside, need the heavy duty machinery
1456 # to reevaluate the extraction starting from $sql (full reevaluation)
1457 if ($inner =~ /\)/) {
1458 require Text::Balanced;
1460 my (undef, $remainder) = do {
1461 # idiotic design - writes to $@ but *DOES NOT* throw exceptions
1463 Text::Balanced::extract_bracketed($sql, '()', qr/\s*/);
1466 # the entire expression needs to be a balanced bracketed thing
1467 # (after an extract no remainder sans trailing space)
1468 last if defined $remainder and $remainder =~ /\S/;
1478 #======================================================================
1480 #======================================================================
1483 my ($self, $arg) = @_;
1486 for my $c ($self->_order_by_chunks($arg) ) {
1487 $self->_SWITCH_refkind($c, {
1488 SCALAR => sub { push @sql, $c },
1489 ARRAYREF => sub { push @sql, shift @$c; push @bind, @$c },
1495 $self->_sqlcase(' order by'),
1501 return wantarray ? ($sql, @bind) : $sql;
1504 sub _order_by_chunks {
1505 my ($self, $arg) = @_;
1507 return $self->_SWITCH_refkind($arg, {
1510 map { $self->_order_by_chunks($_ ) } @$arg;
1513 ARRAYREFREF => sub {
1514 my ($s, @b) = @$$arg;
1515 $self->_assert_bindval_matches_bindtype(@b);
1519 SCALAR => sub {$self->_quote($arg)},
1521 UNDEF => sub {return () },
1523 SCALARREF => sub {$$arg}, # literal SQL, no quoting
1526 # get first pair in hash
1527 my ($key, $val, @rest) = %$arg;
1529 return () unless $key;
1531 if (@rest or not $key =~ /^-(desc|asc)/i) {
1532 puke "hash passed to _order_by must have exactly one key (-desc or -asc)";
1538 for my $c ($self->_order_by_chunks($val)) {
1541 $self->_SWITCH_refkind($c, {
1546 ($sql, @bind) = @$c;
1550 $sql = $sql . ' ' . $self->_sqlcase($direction);
1552 push @ret, [ $sql, @bind];
1561 #======================================================================
1562 # DATASOURCE (FOR NOW, JUST PLAIN TABLE OR LIST OF TABLES)
1563 #======================================================================
1568 $self->_SWITCH_refkind($from, {
1569 ARRAYREF => sub {join ', ', map { $self->_quote($_) } @$from;},
1570 SCALAR => sub {$self->_quote($from)},
1571 SCALARREF => sub {$$from},
1576 #======================================================================
1578 #======================================================================
1580 # highly optimized, as it's called way too often
1582 # my ($self, $label) = @_;
1584 return '' unless defined $_[1];
1585 return ${$_[1]} if ref($_[1]) eq 'SCALAR';
1587 $_[0]->{quote_char} or
1588 ($_[0]->_assert_pass_injection_guard($_[1]), return $_[1]);
1590 my $qref = ref $_[0]->{quote_char};
1592 !$qref ? ($_[0]->{quote_char}, $_[0]->{quote_char})
1593 : ($qref eq 'ARRAY') ? @{$_[0]->{quote_char}}
1594 : puke "Unsupported quote_char format: $_[0]->{quote_char}";
1596 my $esc = $_[0]->{escape_char} || $r;
1598 # parts containing * are naturally unquoted
1599 return join($_[0]->{name_sep}||'', map
1600 +( $_ eq '*' ? $_ : do { (my $n = $_) =~ s/(\Q$esc\E|\Q$r\E)/$esc$1/g; $l . $n . $r } ),
1601 ( $_[0]->{name_sep} ? split (/\Q$_[0]->{name_sep}\E/, $_[1] ) : $_[1] )
1606 # Conversion, if applicable
1608 #my ($self, $arg) = @_;
1609 if ($_[0]->{convert}) {
1610 return $_[0]->_sqlcase($_[0]->{convert}) .'(' . $_[1] . ')';
1617 #my ($self, $col, @vals) = @_;
1618 # called often - tighten code
1619 return $_[0]->{bindtype} eq 'columns'
1620 ? map {[$_[1], $_]} @_[2 .. $#_]
1625 # Dies if any element of @bind is not in [colname => value] format
1626 # if bindtype is 'columns'.
1627 sub _assert_bindval_matches_bindtype {
1628 # my ($self, @bind) = @_;
1630 if ($self->{bindtype} eq 'columns') {
1632 if (!defined $_ || ref($_) ne 'ARRAY' || @$_ != 2) {
1633 puke "bindtype 'columns' selected, you need to pass: [column_name => bind_value]"
1639 sub _join_sql_clauses {
1640 my ($self, $logic, $clauses_aref, $bind_aref) = @_;
1642 if (@$clauses_aref > 1) {
1643 my $join = " " . $self->_sqlcase($logic) . " ";
1644 my $sql = '( ' . join($join, @$clauses_aref) . ' )';
1645 return ($sql, @$bind_aref);
1647 elsif (@$clauses_aref) {
1648 return ($clauses_aref->[0], @$bind_aref); # no parentheses
1651 return (); # if no SQL, ignore @$bind_aref
1656 # Fix SQL case, if so requested
1658 # LDNOTE: if $self->{case} is true, then it contains 'lower', so we
1659 # don't touch the argument ... crooked logic, but let's not change it!
1660 return $_[0]->{case} ? $_[1] : uc($_[1]);
1664 #======================================================================
1665 # DISPATCHING FROM REFKIND
1666 #======================================================================
1669 my ($self, $data) = @_;
1671 return 'UNDEF' unless defined $data;
1673 # blessed objects are treated like scalars
1674 my $ref = (Scalar::Util::blessed $data) ? '' : ref $data;
1676 return 'SCALAR' unless $ref;
1679 while ($ref eq 'REF') {
1681 $ref = (Scalar::Util::blessed $data) ? '' : ref $data;
1685 return ($ref||'SCALAR') . ('REF' x $n_steps);
1689 my ($self, $data) = @_;
1690 my @try = ($self->_refkind($data));
1691 push @try, 'SCALAR_or_UNDEF' if $try[0] eq 'SCALAR' || $try[0] eq 'UNDEF';
1692 push @try, 'FALLBACK';
1696 sub _METHOD_FOR_refkind {
1697 my ($self, $meth_prefix, $data) = @_;
1700 for (@{$self->_try_refkind($data)}) {
1701 $method = $self->can($meth_prefix."_".$_)
1705 return $method || puke "cannot dispatch on '$meth_prefix' for ".$self->_refkind($data);
1709 sub _SWITCH_refkind {
1710 my ($self, $data, $dispatch_table) = @_;
1713 for (@{$self->_try_refkind($data)}) {
1714 $coderef = $dispatch_table->{$_}
1718 puke "no dispatch entry for ".$self->_refkind($data)
1727 #======================================================================
1728 # VALUES, GENERATE, AUTOLOAD
1729 #======================================================================
1731 # LDNOTE: original code from nwiger, didn't touch code in that section
1732 # I feel the AUTOLOAD stuff should not be the default, it should
1733 # only be activated on explicit demand by user.
1737 my $data = shift || return;
1738 puke "Argument to ", __PACKAGE__, "->values must be a \\%hash"
1739 unless ref $data eq 'HASH';
1742 foreach my $k (sort keys %$data) {
1743 my $v = $data->{$k};
1744 $self->_SWITCH_refkind($v, {
1746 if ($self->{array_datatypes}) { # array datatype
1747 push @all_bind, $self->_bindtype($k, $v);
1749 else { # literal SQL with bind
1750 my ($sql, @bind) = @$v;
1751 $self->_assert_bindval_matches_bindtype(@bind);
1752 push @all_bind, @bind;
1755 ARRAYREFREF => sub { # literal SQL with bind
1756 my ($sql, @bind) = @${$v};
1757 $self->_assert_bindval_matches_bindtype(@bind);
1758 push @all_bind, @bind;
1760 SCALARREF => sub { # literal SQL without bind
1762 SCALAR_or_UNDEF => sub {
1763 push @all_bind, $self->_bindtype($k, $v);
1774 my(@sql, @sqlq, @sqlv);
1778 if ($ref eq 'HASH') {
1779 for my $k (sort keys %$_) {
1782 my $label = $self->_quote($k);
1783 if ($r eq 'ARRAY') {
1784 # literal SQL with bind
1785 my ($sql, @bind) = @$v;
1786 $self->_assert_bindval_matches_bindtype(@bind);
1787 push @sqlq, "$label = $sql";
1789 } elsif ($r eq 'SCALAR') {
1790 # literal SQL without bind
1791 push @sqlq, "$label = $$v";
1793 push @sqlq, "$label = ?";
1794 push @sqlv, $self->_bindtype($k, $v);
1797 push @sql, $self->_sqlcase('set'), join ', ', @sqlq;
1798 } elsif ($ref eq 'ARRAY') {
1799 # unlike insert(), assume these are ONLY the column names, i.e. for SQL
1802 if ($r eq 'ARRAY') { # literal SQL with bind
1803 my ($sql, @bind) = @$v;
1804 $self->_assert_bindval_matches_bindtype(@bind);
1807 } elsif ($r eq 'SCALAR') { # literal SQL without bind
1808 # embedded literal SQL
1815 push @sql, '(' . join(', ', @sqlq) . ')';
1816 } elsif ($ref eq 'SCALAR') {
1820 # strings get case twiddled
1821 push @sql, $self->_sqlcase($_);
1825 my $sql = join ' ', @sql;
1827 # this is pretty tricky
1828 # if ask for an array, return ($stmt, @bind)
1829 # otherwise, s/?/shift @sqlv/ to put it inline
1831 return ($sql, @sqlv);
1833 1 while $sql =~ s/\?/my $d = shift(@sqlv);
1834 ref $d ? $d->[1] : $d/e;
1843 # This allows us to check for a local, then _form, attr
1845 my($name) = $AUTOLOAD =~ /.*::(.+)/;
1846 return $self->generate($name, @_);
1857 SQL::Abstract - Generate SQL from Perl data structures
1863 my $sql = SQL::Abstract->new;
1865 my($stmt, @bind) = $sql->select($source, \@fields, \%where, $order);
1867 my($stmt, @bind) = $sql->insert($table, \%fieldvals || \@values);
1869 my($stmt, @bind) = $sql->update($table, \%fieldvals, \%where);
1871 my($stmt, @bind) = $sql->delete($table, \%where);
1873 # Then, use these in your DBI statements
1874 my $sth = $dbh->prepare($stmt);
1875 $sth->execute(@bind);
1877 # Just generate the WHERE clause
1878 my($stmt, @bind) = $sql->where(\%where, $order);
1880 # Return values in the same order, for hashed queries
1881 # See PERFORMANCE section for more details
1882 my @bind = $sql->values(\%fieldvals);
1886 This module was inspired by the excellent L<DBIx::Abstract>.
1887 However, in using that module I found that what I really wanted
1888 to do was generate SQL, but still retain complete control over my
1889 statement handles and use the DBI interface. So, I set out to
1890 create an abstract SQL generation module.
1892 While based on the concepts used by L<DBIx::Abstract>, there are
1893 several important differences, especially when it comes to WHERE
1894 clauses. I have modified the concepts used to make the SQL easier
1895 to generate from Perl data structures and, IMO, more intuitive.
1896 The underlying idea is for this module to do what you mean, based
1897 on the data structures you provide it. The big advantage is that
1898 you don't have to modify your code every time your data changes,
1899 as this module figures it out.
1901 To begin with, an SQL INSERT is as easy as just specifying a hash
1902 of C<key=value> pairs:
1905 name => 'Jimbo Bobson',
1906 phone => '123-456-7890',
1907 address => '42 Sister Lane',
1908 city => 'St. Louis',
1909 state => 'Louisiana',
1912 The SQL can then be generated with this:
1914 my($stmt, @bind) = $sql->insert('people', \%data);
1916 Which would give you something like this:
1918 $stmt = "INSERT INTO people
1919 (address, city, name, phone, state)
1920 VALUES (?, ?, ?, ?, ?)";
1921 @bind = ('42 Sister Lane', 'St. Louis', 'Jimbo Bobson',
1922 '123-456-7890', 'Louisiana');
1924 These are then used directly in your DBI code:
1926 my $sth = $dbh->prepare($stmt);
1927 $sth->execute(@bind);
1929 =head2 Inserting and Updating Arrays
1931 If your database has array types (like for example Postgres),
1932 activate the special option C<< array_datatypes => 1 >>
1933 when creating the C<SQL::Abstract> object.
1934 Then you may use an arrayref to insert and update database array types:
1936 my $sql = SQL::Abstract->new(array_datatypes => 1);
1938 planets => [qw/Mercury Venus Earth Mars/]
1941 my($stmt, @bind) = $sql->insert('solar_system', \%data);
1945 $stmt = "INSERT INTO solar_system (planets) VALUES (?)"
1947 @bind = (['Mercury', 'Venus', 'Earth', 'Mars']);
1950 =head2 Inserting and Updating SQL
1952 In order to apply SQL functions to elements of your C<%data> you may
1953 specify a reference to an arrayref for the given hash value. For example,
1954 if you need to execute the Oracle C<to_date> function on a value, you can
1955 say something like this:
1959 date_entered => \[ "to_date(?,'MM/DD/YYYY')", "03/02/2003" ],
1962 The first value in the array is the actual SQL. Any other values are
1963 optional and would be included in the bind values array. This gives
1966 my($stmt, @bind) = $sql->insert('people', \%data);
1968 $stmt = "INSERT INTO people (name, date_entered)
1969 VALUES (?, to_date(?,'MM/DD/YYYY'))";
1970 @bind = ('Bill', '03/02/2003');
1972 An UPDATE is just as easy, all you change is the name of the function:
1974 my($stmt, @bind) = $sql->update('people', \%data);
1976 Notice that your C<%data> isn't touched; the module will generate
1977 the appropriately quirky SQL for you automatically. Usually you'll
1978 want to specify a WHERE clause for your UPDATE, though, which is
1979 where handling C<%where> hashes comes in handy...
1981 =head2 Complex where statements
1983 This module can generate pretty complicated WHERE statements
1984 easily. For example, simple C<key=value> pairs are taken to mean
1985 equality, and if you want to see if a field is within a set
1986 of values, you can use an arrayref. Let's say we wanted to
1987 SELECT some data based on this criteria:
1990 requestor => 'inna',
1991 worker => ['nwiger', 'rcwe', 'sfz'],
1992 status => { '!=', 'completed' }
1995 my($stmt, @bind) = $sql->select('tickets', '*', \%where);
1997 The above would give you something like this:
1999 $stmt = "SELECT * FROM tickets WHERE
2000 ( requestor = ? ) AND ( status != ? )
2001 AND ( worker = ? OR worker = ? OR worker = ? )";
2002 @bind = ('inna', 'completed', 'nwiger', 'rcwe', 'sfz');
2004 Which you could then use in DBI code like so:
2006 my $sth = $dbh->prepare($stmt);
2007 $sth->execute(@bind);
2013 The methods are simple. There's one for every major SQL operation,
2014 and a constructor you use first. The arguments are specified in a
2015 similar order for each method (table, then fields, then a where
2016 clause) to try and simplify things.
2018 =head2 new(option => 'value')
2020 The C<new()> function takes a list of options and values, and returns
2021 a new B<SQL::Abstract> object which can then be used to generate SQL
2022 through the methods below. The options accepted are:
2028 If set to 'lower', then SQL will be generated in all lowercase. By
2029 default SQL is generated in "textbook" case meaning something like:
2031 SELECT a_field FROM a_table WHERE some_field LIKE '%someval%'
2033 Any setting other than 'lower' is ignored.
2037 This determines what the default comparison operator is. By default
2038 it is C<=>, meaning that a hash like this:
2040 %where = (name => 'nwiger', email => 'nate@wiger.org');
2042 Will generate SQL like this:
2044 WHERE name = 'nwiger' AND email = 'nate@wiger.org'
2046 However, you may want loose comparisons by default, so if you set
2047 C<cmp> to C<like> you would get SQL such as:
2049 WHERE name like 'nwiger' AND email like 'nate@wiger.org'
2051 You can also override the comparison on an individual basis - see
2052 the huge section on L</"WHERE CLAUSES"> at the bottom.
2054 =item sqltrue, sqlfalse
2056 Expressions for inserting boolean values within SQL statements.
2057 By default these are C<1=1> and C<1=0>. They are used
2058 by the special operators C<-in> and C<-not_in> for generating
2059 correct SQL even when the argument is an empty array (see below).
2063 This determines the default logical operator for multiple WHERE
2064 statements in arrays or hashes. If absent, the default logic is "or"
2065 for arrays, and "and" for hashes. This means that a WHERE
2069 event_date => {'>=', '2/13/99'},
2070 event_date => {'<=', '4/24/03'},
2073 will generate SQL like this:
2075 WHERE event_date >= '2/13/99' OR event_date <= '4/24/03'
2077 This is probably not what you want given this query, though (look
2078 at the dates). To change the "OR" to an "AND", simply specify:
2080 my $sql = SQL::Abstract->new(logic => 'and');
2082 Which will change the above C<WHERE> to:
2084 WHERE event_date >= '2/13/99' AND event_date <= '4/24/03'
2086 The logic can also be changed locally by inserting
2087 a modifier in front of an arrayref:
2089 @where = (-and => [event_date => {'>=', '2/13/99'},
2090 event_date => {'<=', '4/24/03'} ]);
2092 See the L</"WHERE CLAUSES"> section for explanations.
2096 This will automatically convert comparisons using the specified SQL
2097 function for both column and value. This is mostly used with an argument
2098 of C<upper> or C<lower>, so that the SQL will have the effect of
2099 case-insensitive "searches". For example, this:
2101 $sql = SQL::Abstract->new(convert => 'upper');
2102 %where = (keywords => 'MaKe iT CAse inSeNSItive');
2104 Will turn out the following SQL:
2106 WHERE upper(keywords) like upper('MaKe iT CAse inSeNSItive')
2108 The conversion can be C<upper()>, C<lower()>, or any other SQL function
2109 that can be applied symmetrically to fields (actually B<SQL::Abstract> does
2110 not validate this option; it will just pass through what you specify verbatim).
2114 This is a kludge because many databases suck. For example, you can't
2115 just bind values using DBI's C<execute()> for Oracle C<CLOB> or C<BLOB> fields.
2116 Instead, you have to use C<bind_param()>:
2118 $sth->bind_param(1, 'reg data');
2119 $sth->bind_param(2, $lots, {ora_type => ORA_CLOB});
2121 The problem is, B<SQL::Abstract> will normally just return a C<@bind> array,
2122 which loses track of which field each slot refers to. Fear not.
2124 If you specify C<bindtype> in new, you can determine how C<@bind> is returned.
2125 Currently, you can specify either C<normal> (default) or C<columns>. If you
2126 specify C<columns>, you will get an array that looks like this:
2128 my $sql = SQL::Abstract->new(bindtype => 'columns');
2129 my($stmt, @bind) = $sql->insert(...);
2132 [ 'column1', 'value1' ],
2133 [ 'column2', 'value2' ],
2134 [ 'column3', 'value3' ],
2137 You can then iterate through this manually, using DBI's C<bind_param()>.
2139 $sth->prepare($stmt);
2142 my($col, $data) = @$_;
2143 if ($col eq 'details' || $col eq 'comments') {
2144 $sth->bind_param($i, $data, {ora_type => ORA_CLOB});
2145 } elsif ($col eq 'image') {
2146 $sth->bind_param($i, $data, {ora_type => ORA_BLOB});
2148 $sth->bind_param($i, $data);
2152 $sth->execute; # execute without @bind now
2154 Now, why would you still use B<SQL::Abstract> if you have to do this crap?
2155 Basically, the advantage is still that you don't have to care which fields
2156 are or are not included. You could wrap that above C<for> loop in a simple
2157 sub called C<bind_fields()> or something and reuse it repeatedly. You still
2158 get a layer of abstraction over manual SQL specification.
2160 Note that if you set L</bindtype> to C<columns>, the C<\[ $sql, @bind ]>
2161 construct (see L</Literal SQL with placeholders and bind values (subqueries)>)
2162 will expect the bind values in this format.
2166 This is the character that a table or column name will be quoted
2167 with. By default this is an empty string, but you could set it to
2168 the character C<`>, to generate SQL like this:
2170 SELECT `a_field` FROM `a_table` WHERE `some_field` LIKE '%someval%'
2172 Alternatively, you can supply an array ref of two items, the first being the left
2173 hand quote character, and the second the right hand quote character. For
2174 example, you could supply C<['[',']']> for SQL Server 2000 compliant quotes
2175 that generates SQL like this:
2177 SELECT [a_field] FROM [a_table] WHERE [some_field] LIKE '%someval%'
2179 Quoting is useful if you have tables or columns names that are reserved
2180 words in your database's SQL dialect.
2184 This is the character that will be used to escape L</quote_char>s appearing
2185 in an identifier before it has been quoted.
2187 The parameter default in case of a single L</quote_char> character is the quote
2190 When opening-closing-style quoting is used (L</quote_char> is an arrayref)
2191 this parameter defaults to the B<closing (right)> L</quote_char>. Occurrences
2192 of the B<opening (left)> L</quote_char> within the identifier are currently left
2193 untouched. The default for opening-closing-style quotes may change in future
2194 versions, thus you are B<strongly encouraged> to specify the escape character
2199 This is the character that separates a table and column name. It is
2200 necessary to specify this when the C<quote_char> option is selected,
2201 so that tables and column names can be individually quoted like this:
2203 SELECT `table`.`one_field` FROM `table` WHERE `table`.`other_field` = 1
2205 =item injection_guard
2207 A regular expression C<qr/.../> that is applied to any C<-function> and unquoted
2208 column name specified in a query structure. This is a safety mechanism to avoid
2209 injection attacks when mishandling user input e.g.:
2211 my %condition_as_column_value_pairs = get_values_from_user();
2212 $sqla->select( ... , \%condition_as_column_value_pairs );
2214 If the expression matches an exception is thrown. Note that literal SQL
2215 supplied via C<\'...'> or C<\['...']> is B<not> checked in any way.
2217 Defaults to checking for C<;> and the C<GO> keyword (TransactSQL)
2219 =item array_datatypes
2221 When this option is true, arrayrefs in INSERT or UPDATE are
2222 interpreted as array datatypes and are passed directly
2224 When this option is false, arrayrefs are interpreted
2225 as literal SQL, just like refs to arrayrefs
2226 (but this behavior is for backwards compatibility; when writing
2227 new queries, use the "reference to arrayref" syntax
2233 Takes a reference to a list of "special operators"
2234 to extend the syntax understood by L<SQL::Abstract>.
2235 See section L</"SPECIAL OPERATORS"> for details.
2239 Takes a reference to a list of "unary operators"
2240 to extend the syntax understood by L<SQL::Abstract>.
2241 See section L</"UNARY OPERATORS"> for details.
2247 =head2 insert($table, \@values || \%fieldvals, \%options)
2249 This is the simplest function. You simply give it a table name
2250 and either an arrayref of values or hashref of field/value pairs.
2251 It returns an SQL INSERT statement and a list of bind values.
2252 See the sections on L</"Inserting and Updating Arrays"> and
2253 L</"Inserting and Updating SQL"> for information on how to insert
2254 with those data types.
2256 The optional C<\%options> hash reference may contain additional
2257 options to generate the insert SQL. Currently supported options
2264 Takes either a scalar of raw SQL fields, or an array reference of
2265 field names, and adds on an SQL C<RETURNING> statement at the end.
2266 This allows you to return data generated by the insert statement
2267 (such as row IDs) without performing another C<SELECT> statement.
2268 Note, however, this is not part of the SQL standard and may not
2269 be supported by all database engines.
2273 =head2 update($table, \%fieldvals, \%where, \%options)
2275 This takes a table, hashref of field/value pairs, and an optional
2276 hashref L<WHERE clause|/WHERE CLAUSES>. It returns an SQL UPDATE function and a list
2278 See the sections on L</"Inserting and Updating Arrays"> and
2279 L</"Inserting and Updating SQL"> for information on how to insert
2280 with those data types.
2282 The optional C<\%options> hash reference may contain additional
2283 options to generate the update SQL. Currently supported options
2290 See the C<returning> option to
2291 L<insert|/insert($table, \@values || \%fieldvals, \%options)>.
2295 =head2 select($source, $fields, $where, $order)
2297 This returns a SQL SELECT statement and associated list of bind values, as
2298 specified by the arguments:
2304 Specification of the 'FROM' part of the statement.
2305 The argument can be either a plain scalar (interpreted as a table
2306 name, will be quoted), or an arrayref (interpreted as a list
2307 of table names, joined by commas, quoted), or a scalarref
2308 (literal SQL, not quoted).
2312 Specification of the list of fields to retrieve from
2314 The argument can be either an arrayref (interpreted as a list
2315 of field names, will be joined by commas and quoted), or a
2316 plain scalar (literal SQL, not quoted).
2317 Please observe that this API is not as flexible as that of
2318 the first argument C<$source>, for backwards compatibility reasons.
2322 Optional argument to specify the WHERE part of the query.
2323 The argument is most often a hashref, but can also be
2324 an arrayref or plain scalar --
2325 see section L<WHERE clause|/"WHERE CLAUSES"> for details.
2329 Optional argument to specify the ORDER BY part of the query.
2330 The argument can be a scalar, a hashref or an arrayref
2331 -- see section L<ORDER BY clause|/"ORDER BY CLAUSES">
2337 =head2 delete($table, \%where, \%options)
2339 This takes a table name and optional hashref L<WHERE clause|/WHERE CLAUSES>.
2340 It returns an SQL DELETE statement and list of bind values.
2342 The optional C<\%options> hash reference may contain additional
2343 options to generate the delete SQL. Currently supported options
2350 See the C<returning> option to
2351 L<insert|/insert($table, \@values || \%fieldvals, \%options)>.
2355 =head2 where(\%where, $order)
2357 This is used to generate just the WHERE clause. For example,
2358 if you have an arbitrary data structure and know what the
2359 rest of your SQL is going to look like, but want an easy way
2360 to produce a WHERE clause, use this. It returns an SQL WHERE
2361 clause and list of bind values.
2364 =head2 values(\%data)
2366 This just returns the values from the hash C<%data>, in the same
2367 order that would be returned from any of the other above queries.
2368 Using this allows you to markedly speed up your queries if you
2369 are affecting lots of rows. See below under the L</"PERFORMANCE"> section.
2371 =head2 generate($any, 'number', $of, \@data, $struct, \%types)
2373 Warning: This is an experimental method and subject to change.
2375 This returns arbitrarily generated SQL. It's a really basic shortcut.
2376 It will return two different things, depending on return context:
2378 my($stmt, @bind) = $sql->generate('create table', \$table, \@fields);
2379 my $stmt_and_val = $sql->generate('create table', \$table, \@fields);
2381 These would return the following:
2383 # First calling form
2384 $stmt = "CREATE TABLE test (?, ?)";
2385 @bind = (field1, field2);
2387 # Second calling form
2388 $stmt_and_val = "CREATE TABLE test (field1, field2)";
2390 Depending on what you're trying to do, it's up to you to choose the correct
2391 format. In this example, the second form is what you would want.
2395 $sql->generate('alter session', { nls_date_format => 'MM/YY' });
2399 ALTER SESSION SET nls_date_format = 'MM/YY'
2401 You get the idea. Strings get their case twiddled, but everything
2402 else remains verbatim.
2404 =head1 EXPORTABLE FUNCTIONS
2406 =head2 is_plain_value
2408 Determines if the supplied argument is a plain value as understood by this
2413 =item * The value is C<undef>
2415 =item * The value is a non-reference
2417 =item * The value is an object with stringification overloading
2419 =item * The value is of the form C<< { -value => $anything } >>
2423 On failure returns C<undef>, on success returns a B<scalar> reference
2424 to the original supplied argument.
2430 The stringification overloading detection is rather advanced: it takes
2431 into consideration not only the presence of a C<""> overload, but if that
2432 fails also checks for enabled
2433 L<autogenerated versions of C<"">|overload/Magic Autogeneration>, based
2434 on either C<0+> or C<bool>.
2436 Unfortunately testing in the field indicates that this
2437 detection B<< may tickle a latent bug in perl versions before 5.018 >>,
2438 but only when very large numbers of stringifying objects are involved.
2439 At the time of writing ( Sep 2014 ) there is no clear explanation of
2440 the direct cause, nor is there a manageably small test case that reliably
2441 reproduces the problem.
2443 If you encounter any of the following exceptions in B<random places within
2444 your application stack> - this module may be to blame:
2446 Operation "ne": no method found,
2447 left argument in overloaded package <something>,
2448 right argument in overloaded package <something>
2452 Stub found while resolving method "???" overloading """" in package <something>
2454 If you fall victim to the above - please attempt to reduce the problem
2455 to something that could be sent to the L<SQL::Abstract developers
2456 |DBIx::Class/GETTING HELP/SUPPORT>
2457 (either publicly or privately). As a workaround in the meantime you can
2458 set C<$ENV{SQLA_ISVALUE_IGNORE_AUTOGENERATED_STRINGIFICATION}> to a true
2459 value, which will most likely eliminate your problem (at the expense of
2460 not being able to properly detect exotic forms of stringification).
2462 This notice and environment variable will be removed in a future version,
2463 as soon as the underlying problem is found and a reliable workaround is
2468 =head2 is_literal_value
2470 Determines if the supplied argument is a literal value as understood by this
2475 =item * C<\$sql_string>
2477 =item * C<\[ $sql_string, @bind_values ]>
2481 On failure returns C<undef>, on success returns an B<array> reference
2482 containing the unpacked version of the supplied literal SQL and bind values.
2484 =head1 WHERE CLAUSES
2488 This module uses a variation on the idea from L<DBIx::Abstract>. It
2489 is B<NOT>, repeat I<not> 100% compatible. B<The main logic of this
2490 module is that things in arrays are OR'ed, and things in hashes
2493 The easiest way to explain is to show lots of examples. After
2494 each C<%where> hash shown, it is assumed you used:
2496 my($stmt, @bind) = $sql->where(\%where);
2498 However, note that the C<%where> hash can be used directly in any
2499 of the other functions as well, as described above.
2501 =head2 Key-value pairs
2503 So, let's get started. To begin, a simple hash:
2507 status => 'completed'
2510 Is converted to SQL C<key = val> statements:
2512 $stmt = "WHERE user = ? AND status = ?";
2513 @bind = ('nwiger', 'completed');
2515 One common thing I end up doing is having a list of values that
2516 a field can be in. To do this, simply specify a list inside of
2521 status => ['assigned', 'in-progress', 'pending'];
2524 This simple code will create the following:
2526 $stmt = "WHERE user = ? AND ( status = ? OR status = ? OR status = ? )";
2527 @bind = ('nwiger', 'assigned', 'in-progress', 'pending');
2529 A field associated to an empty arrayref will be considered a
2530 logical false and will generate 0=1.
2532 =head2 Tests for NULL values
2534 If the value part is C<undef> then this is converted to SQL <IS NULL>
2543 $stmt = "WHERE user = ? AND status IS NULL";
2546 To test if a column IS NOT NULL:
2550 status => { '!=', undef },
2553 =head2 Specific comparison operators
2555 If you want to specify a different type of operator for your comparison,
2556 you can use a hashref for a given column:
2560 status => { '!=', 'completed' }
2563 Which would generate:
2565 $stmt = "WHERE user = ? AND status != ?";
2566 @bind = ('nwiger', 'completed');
2568 To test against multiple values, just enclose the values in an arrayref:
2570 status => { '=', ['assigned', 'in-progress', 'pending'] };
2572 Which would give you:
2574 "WHERE status = ? OR status = ? OR status = ?"
2577 The hashref can also contain multiple pairs, in which case it is expanded
2578 into an C<AND> of its elements:
2582 status => { '!=', 'completed', -not_like => 'pending%' }
2585 # Or more dynamically, like from a form
2586 $where{user} = 'nwiger';
2587 $where{status}{'!='} = 'completed';
2588 $where{status}{'-not_like'} = 'pending%';
2590 # Both generate this
2591 $stmt = "WHERE user = ? AND status != ? AND status NOT LIKE ?";
2592 @bind = ('nwiger', 'completed', 'pending%');
2595 To get an OR instead, you can combine it with the arrayref idea:
2599 priority => [ { '=', 2 }, { '>', 5 } ]
2602 Which would generate:
2604 $stmt = "WHERE ( priority = ? OR priority > ? ) AND user = ?";
2605 @bind = ('2', '5', 'nwiger');
2607 If you want to include literal SQL (with or without bind values), just use a
2608 scalar reference or reference to an arrayref as the value:
2611 date_entered => { '>' => \["to_date(?, 'MM/DD/YYYY')", "11/26/2008"] },
2612 date_expires => { '<' => \"now()" }
2615 Which would generate:
2617 $stmt = "WHERE date_entered > to_date(?, 'MM/DD/YYYY') AND date_expires < now()";
2618 @bind = ('11/26/2008');
2621 =head2 Logic and nesting operators
2623 In the example above,
2624 there is a subtle trap if you want to say something like
2625 this (notice the C<AND>):
2627 WHERE priority != ? AND priority != ?
2629 Because, in Perl you I<can't> do this:
2631 priority => { '!=' => 2, '!=' => 1 }
2633 As the second C<!=> key will obliterate the first. The solution
2634 is to use the special C<-modifier> form inside an arrayref:
2636 priority => [ -and => {'!=', 2},
2640 Normally, these would be joined by C<OR>, but the modifier tells it
2641 to use C<AND> instead. (Hint: You can use this in conjunction with the
2642 C<logic> option to C<new()> in order to change the way your queries
2643 work by default.) B<Important:> Note that the C<-modifier> goes
2644 B<INSIDE> the arrayref, as an extra first element. This will
2645 B<NOT> do what you think it might:
2647 priority => -and => [{'!=', 2}, {'!=', 1}] # WRONG!
2649 Here is a quick list of equivalencies, since there is some overlap:
2652 status => {'!=', 'completed', 'not like', 'pending%' }
2653 status => [ -and => {'!=', 'completed'}, {'not like', 'pending%'}]
2656 status => {'=', ['assigned', 'in-progress']}
2657 status => [ -or => {'=', 'assigned'}, {'=', 'in-progress'}]
2658 status => [ {'=', 'assigned'}, {'=', 'in-progress'} ]
2662 =head2 Special operators: IN, BETWEEN, etc.
2664 You can also use the hashref format to compare a list of fields using the
2665 C<IN> comparison operator, by specifying the list as an arrayref:
2668 status => 'completed',
2669 reportid => { -in => [567, 2335, 2] }
2672 Which would generate:
2674 $stmt = "WHERE status = ? AND reportid IN (?,?,?)";
2675 @bind = ('completed', '567', '2335', '2');
2677 The reverse operator C<-not_in> generates SQL C<NOT IN> and is used in
2680 If the argument to C<-in> is an empty array, 'sqlfalse' is generated
2681 (by default: C<1=0>). Similarly, C<< -not_in => [] >> generates
2682 'sqltrue' (by default: C<1=1>).
2684 In addition to the array you can supply a chunk of literal sql or
2685 literal sql with bind:
2688 customer => { -in => \[
2689 'SELECT cust_id FROM cust WHERE balance > ?',
2692 status => { -in => \'SELECT status_codes FROM states' },
2698 customer IN ( SELECT cust_id FROM cust WHERE balance > ? )
2699 AND status IN ( SELECT status_codes FROM states )
2703 Finally, if the argument to C<-in> is not a reference, it will be
2704 treated as a single-element array.
2706 Another pair of operators is C<-between> and C<-not_between>,
2707 used with an arrayref of two values:
2711 completion_date => {
2712 -not_between => ['2002-10-01', '2003-02-06']
2718 WHERE user = ? AND completion_date NOT BETWEEN ( ? AND ? )
2720 Just like with C<-in> all plausible combinations of literal SQL
2724 start0 => { -between => [ 1, 2 ] },
2725 start1 => { -between => \["? AND ?", 1, 2] },
2726 start2 => { -between => \"lower(x) AND upper(y)" },
2727 start3 => { -between => [
2729 \["upper(?)", 'stuff' ],
2736 ( start0 BETWEEN ? AND ? )
2737 AND ( start1 BETWEEN ? AND ? )
2738 AND ( start2 BETWEEN lower(x) AND upper(y) )
2739 AND ( start3 BETWEEN lower(x) AND upper(?) )
2741 @bind = (1, 2, 1, 2, 'stuff');
2744 These are the two builtin "special operators"; but the
2745 list can be expanded: see section L</"SPECIAL OPERATORS"> below.
2747 =head2 Unary operators: bool
2749 If you wish to test against boolean columns or functions within your
2750 database you can use the C<-bool> and C<-not_bool> operators. For
2751 example to test the column C<is_user> being true and the column
2752 C<is_enabled> being false you would use:-
2756 -not_bool => 'is_enabled',
2761 WHERE is_user AND NOT is_enabled
2763 If a more complex combination is required, testing more conditions,
2764 then you should use the and/or operators:-
2769 -not_bool => { two=> { -rlike => 'bar' } },
2770 -not_bool => { three => [ { '=', 2 }, { '>', 5 } ] },
2781 (NOT ( three = ? OR three > ? ))
2784 =head2 Nested conditions, -and/-or prefixes
2786 So far, we've seen how multiple conditions are joined with a top-level
2787 C<AND>. We can change this by putting the different conditions we want in
2788 hashes and then putting those hashes in an array. For example:
2793 status => { -like => ['pending%', 'dispatched'] },
2797 status => 'unassigned',
2801 This data structure would create the following:
2803 $stmt = "WHERE ( user = ? AND ( status LIKE ? OR status LIKE ? ) )
2804 OR ( user = ? AND status = ? ) )";
2805 @bind = ('nwiger', 'pending', 'dispatched', 'robot', 'unassigned');
2808 Clauses in hashrefs or arrayrefs can be prefixed with an C<-and> or C<-or>
2809 to change the logic inside:
2815 -and => [ workhrs => {'>', 20}, geo => 'ASIA' ],
2816 -or => { workhrs => {'<', 50}, geo => 'EURO' },
2823 $stmt = "WHERE ( user = ?
2824 AND ( ( workhrs > ? AND geo = ? )
2825 OR ( workhrs < ? OR geo = ? ) ) )";
2826 @bind = ('nwiger', '20', 'ASIA', '50', 'EURO');
2828 =head3 Algebraic inconsistency, for historical reasons
2830 C<Important note>: when connecting several conditions, the C<-and->|C<-or>
2831 operator goes C<outside> of the nested structure; whereas when connecting
2832 several constraints on one column, the C<-and> operator goes
2833 C<inside> the arrayref. Here is an example combining both features:
2836 -and => [a => 1, b => 2],
2837 -or => [c => 3, d => 4],
2838 e => [-and => {-like => 'foo%'}, {-like => '%bar'} ]
2843 WHERE ( ( ( a = ? AND b = ? )
2844 OR ( c = ? OR d = ? )
2845 OR ( e LIKE ? AND e LIKE ? ) ) )
2847 This difference in syntax is unfortunate but must be preserved for
2848 historical reasons. So be careful: the two examples below would
2849 seem algebraically equivalent, but they are not
2852 { -like => 'foo%' },
2853 { -like => '%bar' },
2855 # yields: WHERE ( ( col LIKE ? AND col LIKE ? ) )
2858 { col => { -like => 'foo%' } },
2859 { col => { -like => '%bar' } },
2861 # yields: WHERE ( ( col LIKE ? OR col LIKE ? ) )
2864 =head2 Literal SQL and value type operators
2866 The basic premise of SQL::Abstract is that in WHERE specifications the "left
2867 side" is a column name and the "right side" is a value (normally rendered as
2868 a placeholder). This holds true for both hashrefs and arrayref pairs as you
2869 see in the L</WHERE CLAUSES> examples above. Sometimes it is necessary to
2870 alter this behavior. There are several ways of doing so.
2874 This is a virtual operator that signals the string to its right side is an
2875 identifier (a column name) and not a value. For example to compare two
2876 columns you would write:
2879 priority => { '<', 2 },
2880 requestor => { -ident => 'submitter' },
2885 $stmt = "WHERE priority < ? AND requestor = submitter";
2888 If you are maintaining legacy code you may see a different construct as
2889 described in L</Deprecated usage of Literal SQL>, please use C<-ident> in new
2894 This is a virtual operator that signals that the construct to its right side
2895 is a value to be passed to DBI. This is for example necessary when you want
2896 to write a where clause against an array (for RDBMS that support such
2897 datatypes). For example:
2900 array => { -value => [1, 2, 3] }
2905 $stmt = 'WHERE array = ?';
2906 @bind = ([1, 2, 3]);
2908 Note that if you were to simply say:
2914 the result would probably not be what you wanted:
2916 $stmt = 'WHERE array = ? OR array = ? OR array = ?';
2921 Finally, sometimes only literal SQL will do. To include a random snippet
2922 of SQL verbatim, you specify it as a scalar reference. Consider this only
2923 as a last resort. Usually there is a better way. For example:
2926 priority => { '<', 2 },
2927 requestor => { -in => \'(SELECT name FROM hitmen)' },
2932 $stmt = "WHERE priority < ? AND requestor IN (SELECT name FROM hitmen)"
2935 Note that in this example, you only get one bind parameter back, since
2936 the verbatim SQL is passed as part of the statement.
2940 Never use untrusted input as a literal SQL argument - this is a massive
2941 security risk (there is no way to check literal snippets for SQL
2942 injections and other nastyness). If you need to deal with untrusted input
2943 use literal SQL with placeholders as described next.
2945 =head3 Literal SQL with placeholders and bind values (subqueries)
2947 If the literal SQL to be inserted has placeholders and bind values,
2948 use a reference to an arrayref (yes this is a double reference --
2949 not so common, but perfectly legal Perl). For example, to find a date
2950 in Postgres you can use something like this:
2953 date_column => \[ "= date '2008-09-30' - ?::integer", 10 ]
2958 $stmt = "WHERE ( date_column = date '2008-09-30' - ?::integer )"
2961 Note that you must pass the bind values in the same format as they are returned
2962 by L<where|/where(\%where, $order)>. This means that if you set L</bindtype>
2963 to C<columns>, you must provide the bind values in the
2964 C<< [ column_meta => value ] >> format, where C<column_meta> is an opaque
2965 scalar value; most commonly the column name, but you can use any scalar value
2966 (including references and blessed references), L<SQL::Abstract> will simply
2967 pass it through intact. So if C<bindtype> is set to C<columns> the above
2968 example will look like:
2971 date_column => \[ "= date '2008-09-30' - ?::integer", [ {} => 10 ] ]
2974 Literal SQL is especially useful for nesting parenthesized clauses in the
2975 main SQL query. Here is a first example:
2977 my ($sub_stmt, @sub_bind) = ("SELECT c1 FROM t1 WHERE c2 < ? AND c3 LIKE ?",
2981 bar => \["IN ($sub_stmt)" => @sub_bind],
2986 $stmt = "WHERE (foo = ? AND bar IN (SELECT c1 FROM t1
2987 WHERE c2 < ? AND c3 LIKE ?))";
2988 @bind = (1234, 100, "foo%");
2990 Other subquery operators, like for example C<"E<gt> ALL"> or C<"NOT IN">,
2991 are expressed in the same way. Of course the C<$sub_stmt> and
2992 its associated bind values can be generated through a former call
2995 my ($sub_stmt, @sub_bind)
2996 = $sql->select("t1", "c1", {c2 => {"<" => 100},
2997 c3 => {-like => "foo%"}});
3000 bar => \["> ALL ($sub_stmt)" => @sub_bind],
3003 In the examples above, the subquery was used as an operator on a column;
3004 but the same principle also applies for a clause within the main C<%where>
3005 hash, like an EXISTS subquery:
3007 my ($sub_stmt, @sub_bind)
3008 = $sql->select("t1", "*", {c1 => 1, c2 => \"> t0.c0"});
3009 my %where = ( -and => [
3011 \["EXISTS ($sub_stmt)" => @sub_bind],
3016 $stmt = "WHERE (foo = ? AND EXISTS (SELECT * FROM t1
3017 WHERE c1 = ? AND c2 > t0.c0))";
3021 Observe that the condition on C<c2> in the subquery refers to
3022 column C<t0.c0> of the main query: this is I<not> a bind
3023 value, so we have to express it through a scalar ref.
3024 Writing C<< c2 => {">" => "t0.c0"} >> would have generated
3025 C<< c2 > ? >> with bind value C<"t0.c0"> ... not exactly
3026 what we wanted here.
3028 Finally, here is an example where a subquery is used
3029 for expressing unary negation:
3031 my ($sub_stmt, @sub_bind)
3032 = $sql->where({age => [{"<" => 10}, {">" => 20}]});
3033 $sub_stmt =~ s/^ where //i; # don't want "WHERE" in the subclause
3035 lname => {like => '%son%'},
3036 \["NOT ($sub_stmt)" => @sub_bind],
3041 $stmt = "lname LIKE ? AND NOT ( age < ? OR age > ? )"
3042 @bind = ('%son%', 10, 20)
3044 =head3 Deprecated usage of Literal SQL
3046 Below are some examples of archaic use of literal SQL. It is shown only as
3047 reference for those who deal with legacy code. Each example has a much
3048 better, cleaner and safer alternative that users should opt for in new code.
3054 my %where = ( requestor => \'IS NOT NULL' )
3056 $stmt = "WHERE requestor IS NOT NULL"
3058 This used to be the way of generating NULL comparisons, before the handling
3059 of C<undef> got formalized. For new code please use the superior syntax as
3060 described in L</Tests for NULL values>.
3064 my %where = ( requestor => \'= submitter' )
3066 $stmt = "WHERE requestor = submitter"
3068 This used to be the only way to compare columns. Use the superior L</-ident>
3069 method for all new code. For example an identifier declared in such a way
3070 will be properly quoted if L</quote_char> is properly set, while the legacy
3071 form will remain as supplied.
3075 my %where = ( is_ready => \"", completed => { '>', '2012-12-21' } )
3077 $stmt = "WHERE completed > ? AND is_ready"
3078 @bind = ('2012-12-21')
3080 Using an empty string literal used to be the only way to express a boolean.
3081 For all new code please use the much more readable
3082 L<-bool|/Unary operators: bool> operator.
3088 These pages could go on for a while, since the nesting of the data
3089 structures this module can handle are pretty much unlimited (the
3090 module implements the C<WHERE> expansion as a recursive function
3091 internally). Your best bet is to "play around" with the module a
3092 little to see how the data structures behave, and choose the best
3093 format for your data based on that.
3095 And of course, all the values above will probably be replaced with
3096 variables gotten from forms or the command line. After all, if you
3097 knew everything ahead of time, you wouldn't have to worry about
3098 dynamically-generating SQL and could just hardwire it into your
3101 =head1 ORDER BY CLAUSES
3103 Some functions take an order by clause. This can either be a scalar (just a
3104 column name), a hashref of C<< { -desc => 'col' } >> or C<< { -asc => 'col' }
3105 >>, a scalarref, an arrayref-ref, or an arrayref of any of the previous
3108 Given | Will Generate
3109 ---------------------------------------------------------------
3111 'colA' | ORDER BY colA
3113 [qw/colA colB/] | ORDER BY colA, colB
3115 {-asc => 'colA'} | ORDER BY colA ASC
3117 {-desc => 'colB'} | ORDER BY colB DESC
3119 ['colA', {-asc => 'colB'}] | ORDER BY colA, colB ASC
3121 { -asc => [qw/colA colB/] } | ORDER BY colA ASC, colB ASC
3123 \'colA DESC' | ORDER BY colA DESC
3125 \[ 'FUNC(colA, ?)', $x ] | ORDER BY FUNC(colA, ?)
3126 | /* ...with $x bound to ? */
3129 { -asc => 'colA' }, | colA ASC,
3130 { -desc => [qw/colB/] }, | colB DESC,
3131 { -asc => [qw/colC colD/] },| colC ASC, colD ASC,
3132 \'colE DESC', | colE DESC,
3133 \[ 'FUNC(colF, ?)', $x ], | FUNC(colF, ?)
3134 ] | /* ...with $x bound to ? */
3135 ===============================================================
3139 =head1 SPECIAL OPERATORS
3141 my $sqlmaker = SQL::Abstract->new(special_ops => [
3145 my ($self, $field, $op, $arg) = @_;
3151 handler => 'method_name',
3155 A "special operator" is a SQL syntactic clause that can be
3156 applied to a field, instead of a usual binary operator.
3159 WHERE field IN (?, ?, ?)
3160 WHERE field BETWEEN ? AND ?
3161 WHERE MATCH(field) AGAINST (?, ?)
3163 Special operators IN and BETWEEN are fairly standard and therefore
3164 are builtin within C<SQL::Abstract> (as the overridable methods
3165 C<_where_field_IN> and C<_where_field_BETWEEN>). For other operators,
3166 like the MATCH .. AGAINST example above which is specific to MySQL,
3167 you can write your own operator handlers - supply a C<special_ops>
3168 argument to the C<new> method. That argument takes an arrayref of
3169 operator definitions; each operator definition is a hashref with two
3176 the regular expression to match the operator
3180 Either a coderef or a plain scalar method name. In both cases
3181 the expected return is C<< ($sql, @bind) >>.
3183 When supplied with a method name, it is simply called on the
3184 L<SQL::Abstract> object as:
3186 $self->$method_name($field, $op, $arg)
3190 $field is the LHS of the operator
3191 $op is the part that matched the handler regex
3194 When supplied with a coderef, it is called as:
3196 $coderef->($self, $field, $op, $arg)
3201 For example, here is an implementation
3202 of the MATCH .. AGAINST syntax for MySQL
3204 my $sqlmaker = SQL::Abstract->new(special_ops => [
3206 # special op for MySql MATCH (field) AGAINST(word1, word2, ...)
3207 {regex => qr/^match$/i,
3209 my ($self, $field, $op, $arg) = @_;
3210 $arg = [$arg] if not ref $arg;
3211 my $label = $self->_quote($field);
3212 my ($placeholder) = $self->_convert('?');
3213 my $placeholders = join ", ", (($placeholder) x @$arg);
3214 my $sql = $self->_sqlcase('match') . " ($label) "
3215 . $self->_sqlcase('against') . " ($placeholders) ";
3216 my @bind = $self->_bindtype($field, @$arg);
3217 return ($sql, @bind);
3224 =head1 UNARY OPERATORS
3226 my $sqlmaker = SQL::Abstract->new(unary_ops => [
3230 my ($self, $op, $arg) = @_;
3236 handler => 'method_name',
3240 A "unary operator" is a SQL syntactic clause that can be
3241 applied to a field - the operator goes before the field
3243 You can write your own operator handlers - supply a C<unary_ops>
3244 argument to the C<new> method. That argument takes an arrayref of
3245 operator definitions; each operator definition is a hashref with two
3252 the regular expression to match the operator
3256 Either a coderef or a plain scalar method name. In both cases
3257 the expected return is C<< $sql >>.
3259 When supplied with a method name, it is simply called on the
3260 L<SQL::Abstract> object as:
3262 $self->$method_name($op, $arg)
3266 $op is the part that matched the handler regex
3267 $arg is the RHS or argument of the operator
3269 When supplied with a coderef, it is called as:
3271 $coderef->($self, $op, $arg)
3279 Thanks to some benchmarking by Mark Stosberg, it turns out that
3280 this module is many orders of magnitude faster than using C<DBIx::Abstract>.
3281 I must admit this wasn't an intentional design issue, but it's a
3282 byproduct of the fact that you get to control your C<DBI> handles
3285 To maximize performance, use a code snippet like the following:
3287 # prepare a statement handle using the first row
3288 # and then reuse it for the rest of the rows
3290 for my $href (@array_of_hashrefs) {
3291 $stmt ||= $sql->insert('table', $href);
3292 $sth ||= $dbh->prepare($stmt);
3293 $sth->execute($sql->values($href));
3296 The reason this works is because the keys in your C<$href> are sorted
3297 internally by B<SQL::Abstract>. Thus, as long as your data retains
3298 the same structure, you only have to generate the SQL the first time
3299 around. On subsequent queries, simply use the C<values> function provided
3300 by this module to return your values in the correct order.
3302 However this depends on the values having the same type - if, for
3303 example, the values of a where clause may either have values
3304 (resulting in sql of the form C<column = ?> with a single bind
3305 value), or alternatively the values might be C<undef> (resulting in
3306 sql of the form C<column IS NULL> with no bind value) then the
3307 caching technique suggested will not work.
3311 If you use my C<CGI::FormBuilder> module at all, you'll hopefully
3312 really like this part (I do, at least). Building up a complex query
3313 can be as simple as the following:
3320 use CGI::FormBuilder;
3323 my $form = CGI::FormBuilder->new(...);
3324 my $sql = SQL::Abstract->new;
3326 if ($form->submitted) {
3327 my $field = $form->field;
3328 my $id = delete $field->{id};
3329 my($stmt, @bind) = $sql->update('table', $field, {id => $id});
3332 Of course, you would still have to connect using C<DBI> to run the
3333 query, but the point is that if you make your form look like your
3334 table, the actual query script can be extremely simplistic.
3336 If you're B<REALLY> lazy (I am), check out C<HTML::QuickTable> for
3337 a fast interface to returning and formatting data. I frequently
3338 use these three modules together to write complex database query
3339 apps in under 50 lines.
3341 =head1 HOW TO CONTRIBUTE
3343 Contributions are always welcome, in all usable forms (we especially
3344 welcome documentation improvements). The delivery methods include git-
3345 or unified-diff formatted patches, GitHub pull requests, or plain bug
3346 reports either via RT or the Mailing list. Contributors are generally
3347 granted full access to the official repository after their first several
3348 patches pass successful review.
3350 This project is maintained in a git repository. The code and related tools are
3351 accessible at the following locations:
3355 =item * Official repo: L<git://git.shadowcat.co.uk/dbsrgits/SQL-Abstract.git>
3357 =item * Official gitweb: L<http://git.shadowcat.co.uk/gitweb/gitweb.cgi?p=dbsrgits/SQL-Abstract.git>
3359 =item * GitHub mirror: L<https://github.com/dbsrgits/sql-abstract>
3361 =item * Authorized committers: L<ssh://dbsrgits@git.shadowcat.co.uk/SQL-Abstract.git>
3367 Version 1.50 was a major internal refactoring of C<SQL::Abstract>.
3368 Great care has been taken to preserve the I<published> behavior
3369 documented in previous versions in the 1.* family; however,
3370 some features that were previously undocumented, or behaved
3371 differently from the documentation, had to be changed in order
3372 to clarify the semantics. Hence, client code that was relying
3373 on some dark areas of C<SQL::Abstract> v1.*
3374 B<might behave differently> in v1.50.
3376 The main changes are:
3382 support for literal SQL through the C<< \ [ $sql, @bind ] >> syntax.
3386 support for the { operator => \"..." } construct (to embed literal SQL)
3390 support for the { operator => \["...", @bind] } construct (to embed literal SQL with bind values)
3394 optional support for L<array datatypes|/"Inserting and Updating Arrays">
3398 defensive programming: check arguments
3402 fixed bug with global logic, which was previously implemented
3403 through global variables yielding side-effects. Prior versions would
3404 interpret C<< [ {cond1, cond2}, [cond3, cond4] ] >>
3405 as C<< "(cond1 AND cond2) OR (cond3 AND cond4)" >>.
3406 Now this is interpreted
3407 as C<< "(cond1 AND cond2) OR (cond3 OR cond4)" >>.
3412 fixed semantics of _bindtype on array args
3416 dropped the C<_anoncopy> of the %where tree. No longer necessary,
3417 we just avoid shifting arrays within that tree.
3421 dropped the C<_modlogic> function
3425 =head1 ACKNOWLEDGEMENTS
3427 There are a number of individuals that have really helped out with
3428 this module. Unfortunately, most of them submitted bugs via CPAN
3429 so I have no idea who they are! But the people I do know are:
3431 Ash Berlin (order_by hash term support)
3432 Matt Trout (DBIx::Class support)
3433 Mark Stosberg (benchmarking)
3434 Chas Owens (initial "IN" operator support)
3435 Philip Collins (per-field SQL functions)
3436 Eric Kolve (hashref "AND" support)
3437 Mike Fragassi (enhancements to "BETWEEN" and "LIKE")
3438 Dan Kubb (support for "quote_char" and "name_sep")
3439 Guillermo Roditi (patch to cleanup "IN" and "BETWEEN", fix and tests for _order_by)
3440 Laurent Dami (internal refactoring, extensible list of special operators, literal SQL)
3441 Norbert Buchmuller (support for literal SQL in hashpair, misc. fixes & tests)
3442 Peter Rabbitson (rewrite of SQLA::Test, misc. fixes & tests)
3443 Oliver Charles (support for "RETURNING" after "INSERT")
3449 L<DBIx::Class>, L<DBIx::Abstract>, L<CGI::FormBuilder>, L<HTML::QuickTable>.
3453 Copyright (c) 2001-2007 Nathan Wiger <nwiger@cpan.org>. All Rights Reserved.
3455 This module is actively maintained by Matt Trout <mst@shadowcatsystems.co.uk>
3457 For support, your best bet is to try the C<DBIx::Class> users mailing list.
3458 While not an official support venue, C<DBIx::Class> makes heavy use of
3459 C<SQL::Abstract>, and as such list members there are very familiar with
3460 how to create queries.
3464 This module is free software; you may copy this under the same
3465 terms as perl itself (either the GNU General Public License or
3466 the Artistic License)