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' },
60 #======================================================================
61 # DEBUGGING AND ERROR REPORTING
62 #======================================================================
65 return unless $_[0]->{debug}; shift; # a little faster
66 my $func = (caller(1))[3];
67 warn "[$func] ", @_, "\n";
71 my($func) = (caller(1))[3];
72 Carp::carp "[$func] Warning: ", @_;
76 my($func) = (caller(1))[3];
77 Carp::croak "[$func] Fatal: ", @_;
80 sub is_literal_value ($) {
81 ref $_[0] eq 'SCALAR' ? [ ${$_[0]} ]
82 : ( ref $_[0] eq 'REF' and ref ${$_[0]} eq 'ARRAY' ) ? [ @${ $_[0] } ]
86 # FIXME XSify - this can be done so much more efficiently
87 sub is_plain_value ($) {
89 ! length ref $_[0] ? \($_[0])
91 ref $_[0] eq 'HASH' and keys %{$_[0]} == 1
93 exists $_[0]->{-value}
94 ) ? \($_[0]->{-value})
96 # reuse @_ for even moar speedz
97 defined ( $_[1] = Scalar::Util::blessed $_[0] )
99 # deliberately not using Devel::OverloadInfo - the checks we are
100 # intersted in are much more limited than the fullblown thing, and
101 # this is a very hot piece of code
103 # simply using ->can('(""') can leave behind stub methods that
104 # break actually using the overload later (see L<perldiag/Stub
105 # found while resolving method "%s" overloading "%s" in package
106 # "%s"> and the source of overload::mycan())
108 # either has stringification which DBI SHOULD prefer out of the box
109 grep { *{ (qq[${_}::(""]) }{CODE} } @{ $_[2] = mro::get_linear_isa( $_[1] ) }
111 # has nummification or boolification, AND fallback is *not* disabled
113 SQL::Abstract::_ENV_::DETECT_AUTOGENERATED_STRINGIFICATION
116 grep { *{"${_}::(0+"}{CODE} } @{$_[2]}
118 grep { *{"${_}::(bool"}{CODE} } @{$_[2]}
122 # no fallback specified at all
123 ! ( ($_[3]) = grep { *{"${_}::()"}{CODE} } @{$_[2]} )
125 # fallback explicitly undef
126 ! defined ${"$_[3]::()"}
139 #======================================================================
141 #======================================================================
145 my $class = ref($self) || $self;
146 my %opt = (ref $_[0] eq 'HASH') ? %{$_[0]} : @_;
148 # choose our case by keeping an option around
149 delete $opt{case} if $opt{case} && $opt{case} ne 'lower';
151 # default logic for interpreting arrayrefs
152 $opt{logic} = $opt{logic} ? uc $opt{logic} : 'OR';
154 # how to return bind vars
155 $opt{bindtype} ||= 'normal';
157 # default comparison is "=", but can be overridden
160 # try to recognize which are the 'equality' and 'inequality' ops
161 # (temporary quickfix (in 2007), should go through a more seasoned API)
162 $opt{equality_op} = qr/^( \Q$opt{cmp}\E | \= )$/ix;
163 $opt{inequality_op} = qr/^( != | <> )$/ix;
165 $opt{like_op} = qr/^ (is\s+)? r?like $/xi;
166 $opt{not_like_op} = qr/^ (is\s+)? not \s+ r?like $/xi;
169 $opt{sqltrue} ||= '1=1';
170 $opt{sqlfalse} ||= '0=1';
173 $opt{special_ops} ||= [];
174 # regexes are applied in order, thus push after user-defines
175 push @{$opt{special_ops}}, @BUILTIN_SPECIAL_OPS;
178 $opt{unary_ops} ||= [];
179 push @{$opt{unary_ops}}, @BUILTIN_UNARY_OPS;
181 # rudimentary sanity-check for user supplied bits treated as functions/operators
182 # If a purported function matches this regular expression, an exception is thrown.
183 # Literal SQL is *NOT* subject to this check, only functions (and column names
184 # when quoting is not in effect)
187 # need to guard against ()'s in column names too, but this will break tons of
188 # hacks... ideas anyone?
189 $opt{injection_guard} ||= qr/
195 return bless \%opt, $class;
199 sub _assert_pass_injection_guard {
200 if ($_[1] =~ $_[0]->{injection_guard}) {
201 my $class = ref $_[0];
202 puke "Possible SQL injection attempt '$_[1]'. If this is indeed a part of the "
203 . "desired SQL use literal SQL ( \'...' or \[ '...' ] ) or supply your own "
204 . "{injection_guard} attribute to ${class}->new()"
209 #======================================================================
211 #======================================================================
215 my $table = $self->_table(shift);
216 my $data = shift || return;
219 my $method = $self->_METHOD_FOR_refkind("_insert", $data);
220 my ($sql, @bind) = $self->$method($data);
221 $sql = join " ", $self->_sqlcase('insert into'), $table, $sql;
223 if ($options->{returning}) {
224 my ($s, @b) = $self->_insert_returning($options);
229 return wantarray ? ($sql, @bind) : $sql;
232 # So that subclasses can override INSERT ... RETURNING separately from
233 # UPDATE and DELETE (e.g. DBIx::Class::SQLMaker::Oracle does this)
234 sub _insert_returning { shift->_returning(@_) }
237 my ($self, $options) = @_;
239 my $f = $options->{returning};
241 my $fieldlist = $self->_SWITCH_refkind($f, {
242 ARRAYREF => sub {join ', ', map { $self->_quote($_) } @$f;},
243 SCALAR => sub {$self->_quote($f)},
244 SCALARREF => sub {$$f},
246 return $self->_sqlcase(' returning ') . $fieldlist;
249 sub _insert_HASHREF { # explicit list of fields and then values
250 my ($self, $data) = @_;
252 my @fields = sort keys %$data;
254 my ($sql, @bind) = $self->_insert_values($data);
257 $_ = $self->_quote($_) foreach @fields;
258 $sql = "( ".join(", ", @fields).") ".$sql;
260 return ($sql, @bind);
263 sub _insert_ARRAYREF { # just generate values(?,?) part (no list of fields)
264 my ($self, $data) = @_;
266 # no names (arrayref) so can't generate bindtype
267 $self->{bindtype} ne 'columns'
268 or belch "can't do 'columns' bindtype when called with arrayref";
270 my (@values, @all_bind);
271 foreach my $value (@$data) {
272 my ($values, @bind) = $self->_insert_value(undef, $value);
273 push @values, $values;
274 push @all_bind, @bind;
276 my $sql = $self->_sqlcase('values')." ( ".join(", ", @values)." )";
277 return ($sql, @all_bind);
280 sub _insert_ARRAYREFREF { # literal SQL with bind
281 my ($self, $data) = @_;
283 my ($sql, @bind) = @${$data};
284 $self->_assert_bindval_matches_bindtype(@bind);
286 return ($sql, @bind);
290 sub _insert_SCALARREF { # literal SQL without bind
291 my ($self, $data) = @_;
297 my ($self, $data) = @_;
299 my (@values, @all_bind);
300 foreach my $column (sort keys %$data) {
301 my ($values, @bind) = $self->_insert_value($column, $data->{$column});
302 push @values, $values;
303 push @all_bind, @bind;
305 my $sql = $self->_sqlcase('values')." ( ".join(", ", @values)." )";
306 return ($sql, @all_bind);
310 my ($self, $column, $v) = @_;
312 my (@values, @all_bind);
313 $self->_SWITCH_refkind($v, {
316 if ($self->{array_datatypes}) { # if array datatype are activated
318 push @all_bind, $self->_bindtype($column, $v);
320 else { # else literal SQL with bind
321 my ($sql, @bind) = @$v;
322 $self->_assert_bindval_matches_bindtype(@bind);
324 push @all_bind, @bind;
328 ARRAYREFREF => sub { # literal SQL with bind
329 my ($sql, @bind) = @${$v};
330 $self->_assert_bindval_matches_bindtype(@bind);
332 push @all_bind, @bind;
335 # THINK: anything useful to do with a HASHREF ?
336 HASHREF => sub { # (nothing, but old SQLA passed it through)
337 #TODO in SQLA >= 2.0 it will die instead
338 belch "HASH ref as bind value in insert is not supported";
340 push @all_bind, $self->_bindtype($column, $v);
343 SCALARREF => sub { # literal SQL without bind
347 SCALAR_or_UNDEF => sub {
349 push @all_bind, $self->_bindtype($column, $v);
354 my $sql = join(", ", @values);
355 return ($sql, @all_bind);
360 #======================================================================
362 #======================================================================
367 my $table = $self->_table(shift);
368 my $data = shift || return;
372 # first build the 'SET' part of the sql statement
373 puke "Unsupported data type specified to \$sql->update"
374 unless ref $data eq 'HASH';
376 my ($sql, @all_bind) = $self->_update_set_values($data);
377 $sql = $self->_sqlcase('update ') . $table . $self->_sqlcase(' set ')
381 my($where_sql, @where_bind) = $self->where($where);
383 push @all_bind, @where_bind;
386 if ($options->{returning}) {
387 my ($returning_sql, @returning_bind) = $self->_update_returning($options);
388 $sql .= $returning_sql;
389 push @all_bind, @returning_bind;
392 return wantarray ? ($sql, @all_bind) : $sql;
395 sub _update_set_values {
396 my ($self, $data) = @_;
398 my (@set, @all_bind);
399 for my $k (sort keys %$data) {
402 my $label = $self->_quote($k);
404 $self->_SWITCH_refkind($v, {
406 if ($self->{array_datatypes}) { # array datatype
407 push @set, "$label = ?";
408 push @all_bind, $self->_bindtype($k, $v);
410 else { # literal SQL with bind
411 my ($sql, @bind) = @$v;
412 $self->_assert_bindval_matches_bindtype(@bind);
413 push @set, "$label = $sql";
414 push @all_bind, @bind;
417 ARRAYREFREF => sub { # literal SQL with bind
418 my ($sql, @bind) = @${$v};
419 $self->_assert_bindval_matches_bindtype(@bind);
420 push @set, "$label = $sql";
421 push @all_bind, @bind;
423 SCALARREF => sub { # literal SQL without bind
424 push @set, "$label = $$v";
427 my ($op, $arg, @rest) = %$v;
429 puke 'Operator calls in update must be in the form { -op => $arg }'
430 if (@rest or not $op =~ /^\-(.+)/);
432 local $self->{_nested_func_lhs} = $k;
433 my ($sql, @bind) = $self->_where_unary_op($1, $arg);
435 push @set, "$label = $sql";
436 push @all_bind, @bind;
438 SCALAR_or_UNDEF => sub {
439 push @set, "$label = ?";
440 push @all_bind, $self->_bindtype($k, $v);
446 my $sql = join ', ', @set;
448 return ($sql, @all_bind);
451 # So that subclasses can override UPDATE ... RETURNING separately from
453 sub _update_returning { shift->_returning(@_) }
457 #======================================================================
459 #======================================================================
464 my $table = $self->_table(shift);
465 my $fields = shift || '*';
469 my ($fields_sql, @bind) = $self->_select_fields($fields);
471 my ($where_sql, @where_bind) = $self->where($where, $order);
472 push @bind, @where_bind;
474 my $sql = join(' ', $self->_sqlcase('select'), $fields_sql,
475 $self->_sqlcase('from'), $table)
478 return wantarray ? ($sql, @bind) : $sql;
482 my ($self, $fields) = @_;
483 return ref $fields eq 'ARRAY' ? join ', ', map { $self->_quote($_) } @$fields
487 #======================================================================
489 #======================================================================
494 my $table = $self->_table(shift);
498 my($where_sql, @bind) = $self->where($where);
499 my $sql = $self->_sqlcase('delete from ') . $table . $where_sql;
501 if ($options->{returning}) {
502 my ($returning_sql, @returning_bind) = $self->_delete_returning($options);
503 $sql .= $returning_sql;
504 push @bind, @returning_bind;
507 return wantarray ? ($sql, @bind) : $sql;
510 # So that subclasses can override DELETE ... RETURNING separately from
512 sub _delete_returning { shift->_returning(@_) }
516 #======================================================================
518 #======================================================================
522 # Finally, a separate routine just to handle WHERE clauses
524 my ($self, $where, $order) = @_;
527 my ($sql, @bind) = $self->_recurse_where($where);
528 $sql = (defined $sql and length $sql) ? $self->_sqlcase(' where ') . "( $sql )" : '';
532 my ($order_sql, @order_bind) = $self->_order_by($order);
534 push @bind, @order_bind;
537 return wantarray ? ($sql, @bind) : $sql;
541 my ($self, $expr, $logic) = @_;
542 if (ref($expr) eq 'HASH') {
543 if (keys %$expr > 1) {
545 return +{ "-${logic}" => [
546 map $self->_expand_expr_hashpair($_ => $expr->{$_}, $logic),
550 return $self->_expand_expr_hashpair(%$expr, $logic);
552 if (ref($expr) eq 'ARRAY') {
553 $logic = lc($logic || $self->{logic});
554 $logic eq 'and' or $logic eq 'or' or puke "unknown logic: $logic";
560 while (my ($el) = splice @expr, 0, 1) {
561 puke "Supplying an empty left hand side argument is not supported in array-pairs"
562 unless defined($el) and length($el);
563 my $elref = ref($el);
565 push(@res, $self->_expand_expr({ $el, shift(@expr) }));
566 } elsif ($elref eq 'ARRAY') {
567 push(@res, $self->_expand_expr($el)) if @$el;
568 } elsif (is_literal_value($el)) {
570 } elsif ($elref eq 'HASH') {
571 push @res, $self->_expand_expr($el);
576 return { '-'.$logic => \@res };
581 sub _expand_expr_hashpair {
582 my ($self, $k, $v, $logic) = @_;
583 unless (defined($k) and length($k)) {
584 if (defined($k) and is_literal_value($v)) {
585 belch 'Hash-pairs consisting of an empty string with a literal are deprecated, and will be removed in 2.0: use -and => [ $literal ] instead';
588 puke "Supplying an empty left hand side argument is not supported";
592 return $self->_expand_expr($v);
596 return $self->_expand_expr($v);
598 puke "-bool => undef not supported" unless defined($v);
599 return { -ident => $v };
601 if (my ($rest) = $k =~/^-not[_ ](.*)$/) {
602 return $self->_expand_expr({ -not => { "-${rest}", $v } }, $logic);
605 unless (defined($v)) {
606 my $orig_op = my $op = $self->{cmp};
608 $op =~ /^not$/i ? 'is not' # legacy
609 : $op =~ $self->{equality_op} ? 'is'
610 : $op =~ $self->{like_op} ? belch("Supplying an undefined argument to '@{[ uc $op]}' is deprecated") && 'is'
611 : $op =~ $self->{inequality_op} ? 'is not'
612 : $op =~ $self->{not_like_op} ? belch("Supplying an undefined argument to '@{[ uc $op]}' is deprecated") && 'is not'
613 : puke "unexpected operator '$orig_op' with undef operand";
614 return +{ -op => [ $is.' null', { -ident => $k } ] };
621 { -bind => [ $k, $v ] }
625 if (ref($v) eq 'ARRAY') {
626 return $self->{sqlfalse} unless @$v;
627 $self->_debug("ARRAY($k) means distribute over elements");
629 $v->[0] =~ /^-((?:and|or))$/i
630 ? ($v = [ @{$v}[1..$#$v] ], $1)
631 : ($self->{logic} || 'or')
633 return +{ "-${this_logic}" => [ map $self->_expand_expr({ $k => $_ }, $this_logic), @$v ] };
635 if (my $literal = is_literal_value($v)) {
637 belch 'Hash-pairs consisting of an empty string with a literal are deprecated, and will be removed in 2.0: use -and => [ $literal ] instead';
640 my ($sql, @bind) = @$literal;
641 return \[ $self->_quote($k).' '.$sql, @bind ];
648 my ($self, $where, $logic) = @_;
650 my $where_exp = $self->_expand_expr($where, $logic);
652 # dispatch on appropriate method according to refkind of $where
653 my $method = $self->_METHOD_FOR_refkind("_where", $where_exp);
655 my ($sql, @bind) = $self->$method($where_exp, $logic);
657 # DBIx::Class used to call _recurse_where in scalar context
658 # something else might too...
660 return ($sql, @bind);
663 belch "Calling _recurse_where in scalar context is deprecated and will go away before 2.0";
670 #======================================================================
671 # WHERE: top-level ARRAYREF
672 #======================================================================
675 sub _where_ARRAYREF {
676 my ($self, $where, $logic) = @_;
678 $logic = uc($logic || $self->{logic});
679 $logic eq 'AND' or $logic eq 'OR' or puke "unknown logic: $logic";
681 my @clauses = @$where;
683 my (@sql_clauses, @all_bind);
684 # need to use while() so can shift() for pairs
686 my $el = shift @clauses;
688 $el = undef if (defined $el and ! length $el);
690 # switch according to kind of $el and get corresponding ($sql, @bind)
691 my ($sql, @bind) = $self->_SWITCH_refkind($el, {
693 # skip empty elements, otherwise get invalid trailing AND stuff
694 ARRAYREF => sub {$self->_recurse_where($el) if @$el},
698 $self->_assert_bindval_matches_bindtype(@b);
702 HASHREF => sub {$self->_recurse_where($el, 'and') if %$el},
704 SCALARREF => sub { ($$el); },
707 # top-level arrayref with scalars, recurse in pairs
708 $self->_recurse_where({$el => shift(@clauses)})
711 UNDEF => sub {puke "Supplying an empty left hand side argument is not supported in array-pairs" },
715 push @sql_clauses, $sql;
716 push @all_bind, @bind;
720 return $self->_join_sql_clauses($logic, \@sql_clauses, \@all_bind);
723 #======================================================================
724 # WHERE: top-level ARRAYREFREF
725 #======================================================================
727 sub _where_ARRAYREFREF {
728 my ($self, $where) = @_;
729 my ($sql, @bind) = @$$where;
730 $self->_assert_bindval_matches_bindtype(@bind);
731 return ($sql, @bind);
734 #======================================================================
735 # WHERE: top-level HASHREF
736 #======================================================================
739 my ($self, $where) = @_;
740 my (@sql_clauses, @all_bind);
742 for my $k (sort keys %$where) {
743 my $v = $where->{$k};
745 # ($k => $v) is either a special unary op or a regular hashpair
746 my ($sql, @bind) = do {
748 # put the operator in canonical form
750 $op = substr $op, 1; # remove initial dash
751 $op =~ s/^\s+|\s+$//g;# remove leading/trailing space
752 $op =~ s/\s+/ /g; # compress whitespace
754 # so that -not_foo works correctly
755 $op =~ s/^not_/NOT /i;
757 $self->_debug("Unary OP(-$op) within hashref, recursing...");
758 my ($s, @b) = $self->_where_unary_op($op, $v);
760 # top level vs nested
761 # we assume that handled unary ops will take care of their ()s
763 List::Util::first {$op =~ $_->{regex}} @{$self->{unary_ops}}
765 ( defined $self->{_nested_func_lhs} and $self->{_nested_func_lhs} eq $k )
771 if (is_literal_value ($v) ) {
772 belch 'Hash-pairs consisting of an empty string with a literal are deprecated, and will be removed in 2.0: use -and => [ $literal ] instead';
775 puke "Supplying an empty left hand side argument is not supported in hash-pairs";
779 my $method = $self->_METHOD_FOR_refkind("_where_hashpair", $v);
780 $self->$method($k, $v);
784 push @sql_clauses, $sql;
785 push @all_bind, @bind;
788 return $self->_join_sql_clauses('and', \@sql_clauses, \@all_bind);
791 sub _where_unary_op {
792 my ($self, $op, $rhs) = @_;
794 # top level special ops are illegal in general
795 puke "Illegal use of top-level '-$op'"
796 if !(defined $self->{_nested_func_lhs})
797 and List::Util::first { $op =~ $_->{regex} } @{$self->{special_ops}}
798 and not List::Util::first { $op =~ $_->{regex} } @{$self->{unary_ops}};
800 if (my $op_entry = List::Util::first { $op =~ $_->{regex} } @{$self->{unary_ops}}) {
801 my $handler = $op_entry->{handler};
803 if (not ref $handler) {
804 if ($op =~ s/ [_\s]? \d+ $//x ) {
805 belch 'Use of [and|or|nest]_N modifiers is deprecated and will be removed in SQLA v2.0. '
806 . "You probably wanted ...-and => [ -$op => COND1, -$op => COND2 ... ]";
808 return $self->$handler($op, $rhs);
810 elsif (ref $handler eq 'CODE') {
811 return $handler->($self, $op, $rhs);
814 puke "Illegal handler for operator $op - expecting a method name or a coderef";
818 $self->_debug("Generic unary OP: $op - recursing as function");
820 $self->_assert_pass_injection_guard($op);
822 my ($sql, @bind) = $self->_SWITCH_refkind($rhs, {
824 puke "Illegal use of top-level '-$op'"
825 unless defined $self->{_nested_func_lhs};
828 $self->_convert('?'),
829 $self->_bindtype($self->{_nested_func_lhs}, $rhs)
833 $self->_recurse_where($rhs)
837 $sql = sprintf('%s %s',
838 $self->_sqlcase($op),
842 return ($sql, @bind);
845 sub _where_op_ANDOR {
846 my ($self, $op, $v) = @_;
848 $self->_SWITCH_refkind($v, {
850 return $self->_where_ARRAYREF($v, $op);
854 return ($op =~ /^or/i)
855 ? $self->_where_ARRAYREF([ map { $_ => $v->{$_} } (sort keys %$v) ], $op)
856 : $self->_where_HASHREF($v);
860 puke "-$op => \\\$scalar makes little sense, use " .
862 ? '[ \$scalar, \%rest_of_conditions ] instead'
863 : '-and => [ \$scalar, \%rest_of_conditions ] instead'
868 puke "-$op => \\[...] makes little sense, use " .
870 ? '[ \[...], \%rest_of_conditions ] instead'
871 : '-and => [ \[...], \%rest_of_conditions ] instead'
875 SCALAR => sub { # permissively interpreted as SQL
876 puke "-$op => \$value makes little sense, use -bool => \$value instead";
880 puke "-$op => undef not supported";
886 my ($self, $op, $v) = @_;
888 $self->_SWITCH_refkind($v, {
890 SCALAR => sub { # permissively interpreted as SQL
891 belch "literal SQL should be -nest => \\'scalar' "
892 . "instead of -nest => 'scalar' ";
897 puke "-$op => undef not supported";
901 $self->_recurse_where($v);
909 my ($self, $op, $v) = @_;
911 my ($s, @b) = $self->_SWITCH_refkind($v, {
912 SCALAR => sub { # interpreted as SQL column
913 $self->_convert($self->_quote($v));
917 puke "-$op => undef not supported";
921 $self->_recurse_where($v);
925 $s = "(NOT $s)" if $op =~ /^not/i;
930 sub _where_op_IDENT {
932 my ($op, $rhs) = splice @_, -2;
933 if (! defined $rhs or length ref $rhs) {
934 puke "-$op requires a single plain scalar argument (a quotable identifier)";
937 # in case we are called as a top level special op (no '=')
938 my $has_lhs = my $lhs = shift;
940 $_ = $self->_convert($self->_quote($_)) for ($lhs, $rhs);
948 sub _where_op_VALUE {
950 my ($op, $rhs) = splice @_, -2;
952 # in case we are called as a top level special op (no '=')
956 if (! defined $rhs) {
958 ? $self->_where_hashpair_HASHREF($lhs, { -is => undef })
965 (defined $lhs ? $lhs : $self->{_nested_func_lhs}),
972 $self->_convert($self->_quote($lhs)) . ' = ' . $self->_convert('?'),
976 $self->_convert('?'),
983 my %unop_postfix = map +($_ => 1), 'is null', 'is not null';
986 my ($self, undef, $v) = @_;
987 my ($op, @args) = @$v;
988 $op =~ s/^-// if length($op) > 1;
989 local $self->{_nested_func_lhs};
991 my ($expr_sql, @bind) = $self->_recurse_where($args[0]);
992 my $final_op = join ' ', split '_', $op;
993 my $op_sql = $self->_sqlcase($final_op);
995 $unop_postfix{lc($final_op)}
996 ? "${expr_sql} ${op_sql}"
997 : "${op_sql} ${expr_sql}"
999 return ($final_sql, @bind);
1000 } elsif (@args == 2) {
1001 my ($l, $r) = map [ $self->_recurse_where($_) ], @args;
1002 return ( $l->[0].' '.$self->_sqlcase(join ' ', split '_', $op).' '.$r->[0], @{$l}[1..$#$l], @{$r}[1..$#$r] );
1007 sub _where_op_BIND {
1008 my ($self, undef, $bind) = @_;
1009 return ($self->_convert('?'), $self->_bindtype(@$bind));
1012 sub _where_hashpair_ARRAYREF {
1013 my ($self, $k, $v) = @_;
1016 my @v = @$v; # need copy because of shift below
1017 $self->_debug("ARRAY($k) means distribute over elements");
1019 # put apart first element if it is an operator (-and, -or)
1021 (defined $v[0] && $v[0] =~ /^ - (?: AND|OR ) $/ix)
1025 my @distributed = map { {$k => $_} } @v;
1028 $self->_debug("OP($op) reinjected into the distributed array");
1029 unshift @distributed, $op;
1032 my $logic = $op ? substr($op, 1) : '';
1034 return $self->_recurse_where(\@distributed, $logic);
1037 $self->_debug("empty ARRAY($k) means 0=1");
1038 return ($self->{sqlfalse});
1042 sub _where_hashpair_HASHREF {
1043 my ($self, $k, $v, $logic) = @_;
1046 local $self->{_nested_func_lhs} = defined $self->{_nested_func_lhs}
1047 ? $self->{_nested_func_lhs}
1051 my ($all_sql, @all_bind);
1053 for my $orig_op (sort keys %$v) {
1054 my $val = $v->{$orig_op};
1056 # put the operator in canonical form
1059 # FIXME - we need to phase out dash-less ops
1060 $op =~ s/^-//; # remove possible initial dash
1061 $op =~ s/^\s+|\s+$//g;# remove leading/trailing space
1062 $op =~ s/\s+/ /g; # compress whitespace
1064 $self->_assert_pass_injection_guard($op);
1067 $op =~ s/^is_not/IS NOT/i;
1069 # so that -not_foo works correctly
1070 $op =~ s/^not_/NOT /i;
1072 # another retarded special case: foo => { $op => { -value => undef } }
1073 if (ref $val eq 'HASH' and keys %$val == 1 and exists $val->{-value} and ! defined $val->{-value} ) {
1079 # CASE: col-value logic modifiers
1080 if ($orig_op =~ /^ \- (and|or) $/xi) {
1081 ($sql, @bind) = $self->_where_hashpair_HASHREF($k, $val, $1);
1083 # CASE: special operators like -in or -between
1084 elsif (my $special_op = List::Util::first { $op =~ $_->{regex} } @{$self->{special_ops}}) {
1085 my $handler = $special_op->{handler};
1087 puke "No handler supplied for special operator $orig_op";
1089 elsif (not ref $handler) {
1090 ($sql, @bind) = $self->$handler($k, $op, $val);
1092 elsif (ref $handler eq 'CODE') {
1093 ($sql, @bind) = $handler->($self, $k, $op, $val);
1096 puke "Illegal handler for special operator $orig_op - expecting a method name or a coderef";
1100 $self->_SWITCH_refkind($val, {
1102 ARRAYREF => sub { # CASE: col => {op => \@vals}
1103 ($sql, @bind) = $self->_where_field_op_ARRAYREF($k, $op, $val);
1106 ARRAYREFREF => sub { # CASE: col => {op => \[$sql, @bind]} (literal SQL with bind)
1107 my ($sub_sql, @sub_bind) = @$$val;
1108 $self->_assert_bindval_matches_bindtype(@sub_bind);
1109 $sql = join ' ', $self->_convert($self->_quote($k)),
1110 $self->_sqlcase($op),
1115 UNDEF => sub { # CASE: col => {op => undef} : sql "IS (NOT)? NULL"
1117 $op =~ /^not$/i ? 'is not' # legacy
1118 : $op =~ $self->{equality_op} ? 'is'
1119 : $op =~ $self->{like_op} ? belch("Supplying an undefined argument to '@{[ uc $op]}' is deprecated") && 'is'
1120 : $op =~ $self->{inequality_op} ? 'is not'
1121 : $op =~ $self->{not_like_op} ? belch("Supplying an undefined argument to '@{[ uc $op]}' is deprecated") && 'is not'
1122 : puke "unexpected operator '$orig_op' with undef operand";
1124 $sql = $self->_quote($k) . $self->_sqlcase(" $is null");
1127 FALLBACK => sub { # CASE: col => {op/func => $stuff}
1128 ($sql, @bind) = $self->_where_unary_op($op, $val);
1131 $self->_convert($self->_quote($k)),
1132 $self->{_nested_func_lhs} eq $k ? $sql : "($sql)", # top level vs nested
1138 ($all_sql) = (defined $all_sql and $all_sql) ? $self->_join_sql_clauses($logic, [$all_sql, $sql], []) : $sql;
1139 push @all_bind, @bind;
1141 return ($all_sql, @all_bind);
1144 sub _where_field_IS {
1145 my ($self, $k, $op, $v) = @_;
1147 my ($s) = $self->_SWITCH_refkind($v, {
1150 $self->_convert($self->_quote($k)),
1151 map { $self->_sqlcase($_)} ($op, 'null')
1154 puke "$op can only take undef as argument";
1161 sub _where_field_op_ARRAYREF {
1162 my ($self, $k, $op, $vals) = @_;
1164 my @vals = @$vals; #always work on a copy
1167 $self->_debug(sprintf '%s means multiple elements: [ %s ]',
1169 join(', ', map { defined $_ ? "'$_'" : 'NULL' } @vals ),
1172 # see if the first element is an -and/-or op
1174 if (defined $vals[0] && $vals[0] =~ /^ - (AND|OR) $/ix) {
1179 # a long standing API wart - an attempt to change this behavior during
1180 # the 1.50 series failed *spectacularly*. Warn instead and leave the
1185 (!$logic or $logic eq 'OR')
1187 ($op =~ $self->{inequality_op} or $op =~ $self->{not_like_op})
1190 belch "A multi-element arrayref as an argument to the inequality op '$o' "
1191 . 'is technically equivalent to an always-true 1=1 (you probably wanted '
1192 . "to say ...{ \$inequality_op => [ -and => \@values ] }... instead)"
1196 # distribute $op over each remaining member of @vals, append logic if exists
1197 return $self->_recurse_where([map { {$k => {$op, $_}} } @vals], $logic);
1201 # try to DWIM on equality operators
1203 $op =~ $self->{equality_op} ? $self->{sqlfalse}
1204 : $op =~ $self->{like_op} ? belch("Supplying an empty arrayref to '@{[ uc $op]}' is deprecated") && $self->{sqlfalse}
1205 : $op =~ $self->{inequality_op} ? $self->{sqltrue}
1206 : $op =~ $self->{not_like_op} ? belch("Supplying an empty arrayref to '@{[ uc $op]}' is deprecated") && $self->{sqltrue}
1207 : puke "operator '$op' applied on an empty array (field '$k')";
1212 sub _where_hashpair_SCALARREF {
1213 my ($self, $k, $v) = @_;
1214 $self->_debug("SCALAR($k) means literal SQL: $$v");
1215 my $sql = $self->_quote($k) . " " . $$v;
1219 # literal SQL with bind
1220 sub _where_hashpair_ARRAYREFREF {
1221 my ($self, $k, $v) = @_;
1222 $self->_debug("REF($k) means literal SQL: @${$v}");
1223 my ($sql, @bind) = @$$v;
1224 $self->_assert_bindval_matches_bindtype(@bind);
1225 $sql = $self->_quote($k) . " " . $sql;
1226 return ($sql, @bind );
1229 # literal SQL without bind
1230 sub _where_hashpair_SCALAR {
1231 my ($self, $k, $v) = @_;
1232 $self->_debug("NOREF($k) means simple key=val: $k $self->{cmp} $v");
1233 return ($self->_where_hashpair_HASHREF($k, { $self->{cmp} => $v }));
1237 sub _where_hashpair_UNDEF {
1238 my ($self, $k, $v) = @_;
1239 $self->_debug("UNDEF($k) means IS NULL");
1240 return $self->_where_hashpair_HASHREF($k, { -is => undef });
1243 #======================================================================
1244 # WHERE: TOP-LEVEL OTHERS (SCALARREF, SCALAR, UNDEF)
1245 #======================================================================
1248 sub _where_SCALARREF {
1249 my ($self, $where) = @_;
1252 $self->_debug("SCALAR(*top) means literal SQL: $$where");
1258 my ($self, $where) = @_;
1261 $self->_debug("NOREF(*top) means literal SQL: $where");
1272 #======================================================================
1273 # WHERE: BUILTIN SPECIAL OPERATORS (-in, -between)
1274 #======================================================================
1277 sub _where_field_BETWEEN {
1278 my ($self, $k, $op, $vals) = @_;
1280 my ($label, $and, $placeholder);
1281 $label = $self->_convert($self->_quote($k));
1282 $and = ' ' . $self->_sqlcase('and') . ' ';
1283 $placeholder = $self->_convert('?');
1284 $op = $self->_sqlcase($op);
1286 my $invalid_args = "Operator '$op' requires either an arrayref with two defined values or expressions, or a single literal scalarref/arrayref-ref";
1288 my ($clause, @bind) = $self->_SWITCH_refkind($vals, {
1289 ARRAYREFREF => sub {
1290 my ($s, @b) = @$$vals;
1291 $self->_assert_bindval_matches_bindtype(@b);
1298 puke $invalid_args if @$vals != 2;
1300 my (@all_sql, @all_bind);
1301 foreach my $val (@$vals) {
1302 my ($sql, @bind) = $self->_SWITCH_refkind($val, {
1304 return ($placeholder, $self->_bindtype($k, $val) );
1309 ARRAYREFREF => sub {
1310 my ($sql, @bind) = @$$val;
1311 $self->_assert_bindval_matches_bindtype(@bind);
1312 return ($sql, @bind);
1315 my ($func, $arg, @rest) = %$val;
1316 puke "Only simple { -func => arg } functions accepted as sub-arguments to BETWEEN"
1317 if (@rest or $func !~ /^ \- (.+)/x);
1318 $self->_where_unary_op($1 => $arg);
1324 push @all_sql, $sql;
1325 push @all_bind, @bind;
1329 (join $and, @all_sql),
1338 my $sql = "( $label $op $clause )";
1339 return ($sql, @bind)
1343 sub _where_field_IN {
1344 my ($self, $k, $op, $vals) = @_;
1346 # backwards compatibility: if scalar, force into an arrayref
1347 $vals = [$vals] if defined $vals && ! ref $vals;
1349 my ($label) = $self->_convert($self->_quote($k));
1350 my ($placeholder) = $self->_convert('?');
1351 $op = $self->_sqlcase($op);
1353 my ($sql, @bind) = $self->_SWITCH_refkind($vals, {
1354 ARRAYREF => sub { # list of choices
1355 if (@$vals) { # nonempty list
1356 my (@all_sql, @all_bind);
1358 for my $val (@$vals) {
1359 my ($sql, @bind) = $self->_SWITCH_refkind($val, {
1361 return ($placeholder, $val);
1366 ARRAYREFREF => sub {
1367 my ($sql, @bind) = @$$val;
1368 $self->_assert_bindval_matches_bindtype(@bind);
1369 return ($sql, @bind);
1372 my ($func, $arg, @rest) = %$val;
1373 puke "Only simple { -func => arg } functions accepted as sub-arguments to IN"
1374 if (@rest or $func !~ /^ \- (.+)/x);
1375 $self->_where_unary_op($1 => $arg);
1379 'SQL::Abstract before v1.75 used to generate incorrect SQL when the '
1380 . "-$op operator was given an undef-containing list: !!!AUDIT YOUR CODE "
1381 . 'AND DATA!!! (the upcoming Data::Query-based version of SQL::Abstract '
1382 . 'will emit the logically correct SQL instead of raising this exception)'
1386 push @all_sql, $sql;
1387 push @all_bind, @bind;
1391 sprintf('%s %s ( %s )',
1394 join(', ', @all_sql)
1396 $self->_bindtype($k, @all_bind),
1399 else { # empty list: some databases won't understand "IN ()", so DWIM
1400 my $sql = ($op =~ /\bnot\b/i) ? $self->{sqltrue} : $self->{sqlfalse};
1405 SCALARREF => sub { # literal SQL
1406 my $sql = $self->_open_outer_paren($$vals);
1407 return ("$label $op ( $sql )");
1409 ARRAYREFREF => sub { # literal SQL with bind
1410 my ($sql, @bind) = @$$vals;
1411 $self->_assert_bindval_matches_bindtype(@bind);
1412 $sql = $self->_open_outer_paren($sql);
1413 return ("$label $op ( $sql )", @bind);
1417 puke "Argument passed to the '$op' operator can not be undefined";
1421 puke "special op $op requires an arrayref (or scalarref/arrayref-ref)";
1425 return ($sql, @bind);
1428 # Some databases (SQLite) treat col IN (1, 2) different from
1429 # col IN ( (1, 2) ). Use this to strip all outer parens while
1430 # adding them back in the corresponding method
1431 sub _open_outer_paren {
1432 my ($self, $sql) = @_;
1434 while (my ($inner) = $sql =~ /^ \s* \( (.*) \) \s* $/xs) {
1436 # there are closing parens inside, need the heavy duty machinery
1437 # to reevaluate the extraction starting from $sql (full reevaluation)
1438 if ($inner =~ /\)/) {
1439 require Text::Balanced;
1441 my (undef, $remainder) = do {
1442 # idiotic design - writes to $@ but *DOES NOT* throw exceptions
1444 Text::Balanced::extract_bracketed($sql, '()', qr/\s*/);
1447 # the entire expression needs to be a balanced bracketed thing
1448 # (after an extract no remainder sans trailing space)
1449 last if defined $remainder and $remainder =~ /\S/;
1459 #======================================================================
1461 #======================================================================
1464 my ($self, $arg) = @_;
1467 for my $c ($self->_order_by_chunks($arg) ) {
1468 $self->_SWITCH_refkind($c, {
1469 SCALAR => sub { push @sql, $c },
1470 ARRAYREF => sub { push @sql, shift @$c; push @bind, @$c },
1476 $self->_sqlcase(' order by'),
1482 return wantarray ? ($sql, @bind) : $sql;
1485 sub _order_by_chunks {
1486 my ($self, $arg) = @_;
1488 return $self->_SWITCH_refkind($arg, {
1491 map { $self->_order_by_chunks($_ ) } @$arg;
1494 ARRAYREFREF => sub {
1495 my ($s, @b) = @$$arg;
1496 $self->_assert_bindval_matches_bindtype(@b);
1500 SCALAR => sub {$self->_quote($arg)},
1502 UNDEF => sub {return () },
1504 SCALARREF => sub {$$arg}, # literal SQL, no quoting
1507 # get first pair in hash
1508 my ($key, $val, @rest) = %$arg;
1510 return () unless $key;
1512 if (@rest or not $key =~ /^-(desc|asc)/i) {
1513 puke "hash passed to _order_by must have exactly one key (-desc or -asc)";
1519 for my $c ($self->_order_by_chunks($val)) {
1522 $self->_SWITCH_refkind($c, {
1527 ($sql, @bind) = @$c;
1531 $sql = $sql . ' ' . $self->_sqlcase($direction);
1533 push @ret, [ $sql, @bind];
1542 #======================================================================
1543 # DATASOURCE (FOR NOW, JUST PLAIN TABLE OR LIST OF TABLES)
1544 #======================================================================
1549 $self->_SWITCH_refkind($from, {
1550 ARRAYREF => sub {join ', ', map { $self->_quote($_) } @$from;},
1551 SCALAR => sub {$self->_quote($from)},
1552 SCALARREF => sub {$$from},
1557 #======================================================================
1559 #======================================================================
1561 # highly optimized, as it's called way too often
1563 # my ($self, $label) = @_;
1565 return '' unless defined $_[1];
1566 return ${$_[1]} if ref($_[1]) eq 'SCALAR';
1568 $_[0]->{quote_char} or
1569 ($_[0]->_assert_pass_injection_guard($_[1]), return $_[1]);
1571 my $qref = ref $_[0]->{quote_char};
1573 !$qref ? ($_[0]->{quote_char}, $_[0]->{quote_char})
1574 : ($qref eq 'ARRAY') ? @{$_[0]->{quote_char}}
1575 : puke "Unsupported quote_char format: $_[0]->{quote_char}";
1577 my $esc = $_[0]->{escape_char} || $r;
1579 # parts containing * are naturally unquoted
1580 return join($_[0]->{name_sep}||'', map
1581 +( $_ eq '*' ? $_ : do { (my $n = $_) =~ s/(\Q$esc\E|\Q$r\E)/$esc$1/g; $l . $n . $r } ),
1582 ( $_[0]->{name_sep} ? split (/\Q$_[0]->{name_sep}\E/, $_[1] ) : $_[1] )
1587 # Conversion, if applicable
1589 #my ($self, $arg) = @_;
1590 if ($_[0]->{convert}) {
1591 return $_[0]->_sqlcase($_[0]->{convert}) .'(' . $_[1] . ')';
1598 #my ($self, $col, @vals) = @_;
1599 # called often - tighten code
1600 return $_[0]->{bindtype} eq 'columns'
1601 ? map {[$_[1], $_]} @_[2 .. $#_]
1606 # Dies if any element of @bind is not in [colname => value] format
1607 # if bindtype is 'columns'.
1608 sub _assert_bindval_matches_bindtype {
1609 # my ($self, @bind) = @_;
1611 if ($self->{bindtype} eq 'columns') {
1613 if (!defined $_ || ref($_) ne 'ARRAY' || @$_ != 2) {
1614 puke "bindtype 'columns' selected, you need to pass: [column_name => bind_value]"
1620 sub _join_sql_clauses {
1621 my ($self, $logic, $clauses_aref, $bind_aref) = @_;
1623 if (@$clauses_aref > 1) {
1624 my $join = " " . $self->_sqlcase($logic) . " ";
1625 my $sql = '( ' . join($join, @$clauses_aref) . ' )';
1626 return ($sql, @$bind_aref);
1628 elsif (@$clauses_aref) {
1629 return ($clauses_aref->[0], @$bind_aref); # no parentheses
1632 return (); # if no SQL, ignore @$bind_aref
1637 # Fix SQL case, if so requested
1639 # LDNOTE: if $self->{case} is true, then it contains 'lower', so we
1640 # don't touch the argument ... crooked logic, but let's not change it!
1641 return $_[0]->{case} ? $_[1] : uc($_[1]);
1645 #======================================================================
1646 # DISPATCHING FROM REFKIND
1647 #======================================================================
1650 my ($self, $data) = @_;
1652 return 'UNDEF' unless defined $data;
1654 # blessed objects are treated like scalars
1655 my $ref = (Scalar::Util::blessed $data) ? '' : ref $data;
1657 return 'SCALAR' unless $ref;
1660 while ($ref eq 'REF') {
1662 $ref = (Scalar::Util::blessed $data) ? '' : ref $data;
1666 return ($ref||'SCALAR') . ('REF' x $n_steps);
1670 my ($self, $data) = @_;
1671 my @try = ($self->_refkind($data));
1672 push @try, 'SCALAR_or_UNDEF' if $try[0] eq 'SCALAR' || $try[0] eq 'UNDEF';
1673 push @try, 'FALLBACK';
1677 sub _METHOD_FOR_refkind {
1678 my ($self, $meth_prefix, $data) = @_;
1681 for (@{$self->_try_refkind($data)}) {
1682 $method = $self->can($meth_prefix."_".$_)
1686 return $method || puke "cannot dispatch on '$meth_prefix' for ".$self->_refkind($data);
1690 sub _SWITCH_refkind {
1691 my ($self, $data, $dispatch_table) = @_;
1694 for (@{$self->_try_refkind($data)}) {
1695 $coderef = $dispatch_table->{$_}
1699 puke "no dispatch entry for ".$self->_refkind($data)
1708 #======================================================================
1709 # VALUES, GENERATE, AUTOLOAD
1710 #======================================================================
1712 # LDNOTE: original code from nwiger, didn't touch code in that section
1713 # I feel the AUTOLOAD stuff should not be the default, it should
1714 # only be activated on explicit demand by user.
1718 my $data = shift || return;
1719 puke "Argument to ", __PACKAGE__, "->values must be a \\%hash"
1720 unless ref $data eq 'HASH';
1723 foreach my $k (sort keys %$data) {
1724 my $v = $data->{$k};
1725 $self->_SWITCH_refkind($v, {
1727 if ($self->{array_datatypes}) { # array datatype
1728 push @all_bind, $self->_bindtype($k, $v);
1730 else { # literal SQL with bind
1731 my ($sql, @bind) = @$v;
1732 $self->_assert_bindval_matches_bindtype(@bind);
1733 push @all_bind, @bind;
1736 ARRAYREFREF => sub { # literal SQL with bind
1737 my ($sql, @bind) = @${$v};
1738 $self->_assert_bindval_matches_bindtype(@bind);
1739 push @all_bind, @bind;
1741 SCALARREF => sub { # literal SQL without bind
1743 SCALAR_or_UNDEF => sub {
1744 push @all_bind, $self->_bindtype($k, $v);
1755 my(@sql, @sqlq, @sqlv);
1759 if ($ref eq 'HASH') {
1760 for my $k (sort keys %$_) {
1763 my $label = $self->_quote($k);
1764 if ($r eq 'ARRAY') {
1765 # literal SQL with bind
1766 my ($sql, @bind) = @$v;
1767 $self->_assert_bindval_matches_bindtype(@bind);
1768 push @sqlq, "$label = $sql";
1770 } elsif ($r eq 'SCALAR') {
1771 # literal SQL without bind
1772 push @sqlq, "$label = $$v";
1774 push @sqlq, "$label = ?";
1775 push @sqlv, $self->_bindtype($k, $v);
1778 push @sql, $self->_sqlcase('set'), join ', ', @sqlq;
1779 } elsif ($ref eq 'ARRAY') {
1780 # unlike insert(), assume these are ONLY the column names, i.e. for SQL
1783 if ($r eq 'ARRAY') { # literal SQL with bind
1784 my ($sql, @bind) = @$v;
1785 $self->_assert_bindval_matches_bindtype(@bind);
1788 } elsif ($r eq 'SCALAR') { # literal SQL without bind
1789 # embedded literal SQL
1796 push @sql, '(' . join(', ', @sqlq) . ')';
1797 } elsif ($ref eq 'SCALAR') {
1801 # strings get case twiddled
1802 push @sql, $self->_sqlcase($_);
1806 my $sql = join ' ', @sql;
1808 # this is pretty tricky
1809 # if ask for an array, return ($stmt, @bind)
1810 # otherwise, s/?/shift @sqlv/ to put it inline
1812 return ($sql, @sqlv);
1814 1 while $sql =~ s/\?/my $d = shift(@sqlv);
1815 ref $d ? $d->[1] : $d/e;
1824 # This allows us to check for a local, then _form, attr
1826 my($name) = $AUTOLOAD =~ /.*::(.+)/;
1827 return $self->generate($name, @_);
1838 SQL::Abstract - Generate SQL from Perl data structures
1844 my $sql = SQL::Abstract->new;
1846 my($stmt, @bind) = $sql->select($source, \@fields, \%where, $order);
1848 my($stmt, @bind) = $sql->insert($table, \%fieldvals || \@values);
1850 my($stmt, @bind) = $sql->update($table, \%fieldvals, \%where);
1852 my($stmt, @bind) = $sql->delete($table, \%where);
1854 # Then, use these in your DBI statements
1855 my $sth = $dbh->prepare($stmt);
1856 $sth->execute(@bind);
1858 # Just generate the WHERE clause
1859 my($stmt, @bind) = $sql->where(\%where, $order);
1861 # Return values in the same order, for hashed queries
1862 # See PERFORMANCE section for more details
1863 my @bind = $sql->values(\%fieldvals);
1867 This module was inspired by the excellent L<DBIx::Abstract>.
1868 However, in using that module I found that what I really wanted
1869 to do was generate SQL, but still retain complete control over my
1870 statement handles and use the DBI interface. So, I set out to
1871 create an abstract SQL generation module.
1873 While based on the concepts used by L<DBIx::Abstract>, there are
1874 several important differences, especially when it comes to WHERE
1875 clauses. I have modified the concepts used to make the SQL easier
1876 to generate from Perl data structures and, IMO, more intuitive.
1877 The underlying idea is for this module to do what you mean, based
1878 on the data structures you provide it. The big advantage is that
1879 you don't have to modify your code every time your data changes,
1880 as this module figures it out.
1882 To begin with, an SQL INSERT is as easy as just specifying a hash
1883 of C<key=value> pairs:
1886 name => 'Jimbo Bobson',
1887 phone => '123-456-7890',
1888 address => '42 Sister Lane',
1889 city => 'St. Louis',
1890 state => 'Louisiana',
1893 The SQL can then be generated with this:
1895 my($stmt, @bind) = $sql->insert('people', \%data);
1897 Which would give you something like this:
1899 $stmt = "INSERT INTO people
1900 (address, city, name, phone, state)
1901 VALUES (?, ?, ?, ?, ?)";
1902 @bind = ('42 Sister Lane', 'St. Louis', 'Jimbo Bobson',
1903 '123-456-7890', 'Louisiana');
1905 These are then used directly in your DBI code:
1907 my $sth = $dbh->prepare($stmt);
1908 $sth->execute(@bind);
1910 =head2 Inserting and Updating Arrays
1912 If your database has array types (like for example Postgres),
1913 activate the special option C<< array_datatypes => 1 >>
1914 when creating the C<SQL::Abstract> object.
1915 Then you may use an arrayref to insert and update database array types:
1917 my $sql = SQL::Abstract->new(array_datatypes => 1);
1919 planets => [qw/Mercury Venus Earth Mars/]
1922 my($stmt, @bind) = $sql->insert('solar_system', \%data);
1926 $stmt = "INSERT INTO solar_system (planets) VALUES (?)"
1928 @bind = (['Mercury', 'Venus', 'Earth', 'Mars']);
1931 =head2 Inserting and Updating SQL
1933 In order to apply SQL functions to elements of your C<%data> you may
1934 specify a reference to an arrayref for the given hash value. For example,
1935 if you need to execute the Oracle C<to_date> function on a value, you can
1936 say something like this:
1940 date_entered => \[ "to_date(?,'MM/DD/YYYY')", "03/02/2003" ],
1943 The first value in the array is the actual SQL. Any other values are
1944 optional and would be included in the bind values array. This gives
1947 my($stmt, @bind) = $sql->insert('people', \%data);
1949 $stmt = "INSERT INTO people (name, date_entered)
1950 VALUES (?, to_date(?,'MM/DD/YYYY'))";
1951 @bind = ('Bill', '03/02/2003');
1953 An UPDATE is just as easy, all you change is the name of the function:
1955 my($stmt, @bind) = $sql->update('people', \%data);
1957 Notice that your C<%data> isn't touched; the module will generate
1958 the appropriately quirky SQL for you automatically. Usually you'll
1959 want to specify a WHERE clause for your UPDATE, though, which is
1960 where handling C<%where> hashes comes in handy...
1962 =head2 Complex where statements
1964 This module can generate pretty complicated WHERE statements
1965 easily. For example, simple C<key=value> pairs are taken to mean
1966 equality, and if you want to see if a field is within a set
1967 of values, you can use an arrayref. Let's say we wanted to
1968 SELECT some data based on this criteria:
1971 requestor => 'inna',
1972 worker => ['nwiger', 'rcwe', 'sfz'],
1973 status => { '!=', 'completed' }
1976 my($stmt, @bind) = $sql->select('tickets', '*', \%where);
1978 The above would give you something like this:
1980 $stmt = "SELECT * FROM tickets WHERE
1981 ( requestor = ? ) AND ( status != ? )
1982 AND ( worker = ? OR worker = ? OR worker = ? )";
1983 @bind = ('inna', 'completed', 'nwiger', 'rcwe', 'sfz');
1985 Which you could then use in DBI code like so:
1987 my $sth = $dbh->prepare($stmt);
1988 $sth->execute(@bind);
1994 The methods are simple. There's one for every major SQL operation,
1995 and a constructor you use first. The arguments are specified in a
1996 similar order for each method (table, then fields, then a where
1997 clause) to try and simplify things.
1999 =head2 new(option => 'value')
2001 The C<new()> function takes a list of options and values, and returns
2002 a new B<SQL::Abstract> object which can then be used to generate SQL
2003 through the methods below. The options accepted are:
2009 If set to 'lower', then SQL will be generated in all lowercase. By
2010 default SQL is generated in "textbook" case meaning something like:
2012 SELECT a_field FROM a_table WHERE some_field LIKE '%someval%'
2014 Any setting other than 'lower' is ignored.
2018 This determines what the default comparison operator is. By default
2019 it is C<=>, meaning that a hash like this:
2021 %where = (name => 'nwiger', email => 'nate@wiger.org');
2023 Will generate SQL like this:
2025 WHERE name = 'nwiger' AND email = 'nate@wiger.org'
2027 However, you may want loose comparisons by default, so if you set
2028 C<cmp> to C<like> you would get SQL such as:
2030 WHERE name like 'nwiger' AND email like 'nate@wiger.org'
2032 You can also override the comparison on an individual basis - see
2033 the huge section on L</"WHERE CLAUSES"> at the bottom.
2035 =item sqltrue, sqlfalse
2037 Expressions for inserting boolean values within SQL statements.
2038 By default these are C<1=1> and C<1=0>. They are used
2039 by the special operators C<-in> and C<-not_in> for generating
2040 correct SQL even when the argument is an empty array (see below).
2044 This determines the default logical operator for multiple WHERE
2045 statements in arrays or hashes. If absent, the default logic is "or"
2046 for arrays, and "and" for hashes. This means that a WHERE
2050 event_date => {'>=', '2/13/99'},
2051 event_date => {'<=', '4/24/03'},
2054 will generate SQL like this:
2056 WHERE event_date >= '2/13/99' OR event_date <= '4/24/03'
2058 This is probably not what you want given this query, though (look
2059 at the dates). To change the "OR" to an "AND", simply specify:
2061 my $sql = SQL::Abstract->new(logic => 'and');
2063 Which will change the above C<WHERE> to:
2065 WHERE event_date >= '2/13/99' AND event_date <= '4/24/03'
2067 The logic can also be changed locally by inserting
2068 a modifier in front of an arrayref:
2070 @where = (-and => [event_date => {'>=', '2/13/99'},
2071 event_date => {'<=', '4/24/03'} ]);
2073 See the L</"WHERE CLAUSES"> section for explanations.
2077 This will automatically convert comparisons using the specified SQL
2078 function for both column and value. This is mostly used with an argument
2079 of C<upper> or C<lower>, so that the SQL will have the effect of
2080 case-insensitive "searches". For example, this:
2082 $sql = SQL::Abstract->new(convert => 'upper');
2083 %where = (keywords => 'MaKe iT CAse inSeNSItive');
2085 Will turn out the following SQL:
2087 WHERE upper(keywords) like upper('MaKe iT CAse inSeNSItive')
2089 The conversion can be C<upper()>, C<lower()>, or any other SQL function
2090 that can be applied symmetrically to fields (actually B<SQL::Abstract> does
2091 not validate this option; it will just pass through what you specify verbatim).
2095 This is a kludge because many databases suck. For example, you can't
2096 just bind values using DBI's C<execute()> for Oracle C<CLOB> or C<BLOB> fields.
2097 Instead, you have to use C<bind_param()>:
2099 $sth->bind_param(1, 'reg data');
2100 $sth->bind_param(2, $lots, {ora_type => ORA_CLOB});
2102 The problem is, B<SQL::Abstract> will normally just return a C<@bind> array,
2103 which loses track of which field each slot refers to. Fear not.
2105 If you specify C<bindtype> in new, you can determine how C<@bind> is returned.
2106 Currently, you can specify either C<normal> (default) or C<columns>. If you
2107 specify C<columns>, you will get an array that looks like this:
2109 my $sql = SQL::Abstract->new(bindtype => 'columns');
2110 my($stmt, @bind) = $sql->insert(...);
2113 [ 'column1', 'value1' ],
2114 [ 'column2', 'value2' ],
2115 [ 'column3', 'value3' ],
2118 You can then iterate through this manually, using DBI's C<bind_param()>.
2120 $sth->prepare($stmt);
2123 my($col, $data) = @$_;
2124 if ($col eq 'details' || $col eq 'comments') {
2125 $sth->bind_param($i, $data, {ora_type => ORA_CLOB});
2126 } elsif ($col eq 'image') {
2127 $sth->bind_param($i, $data, {ora_type => ORA_BLOB});
2129 $sth->bind_param($i, $data);
2133 $sth->execute; # execute without @bind now
2135 Now, why would you still use B<SQL::Abstract> if you have to do this crap?
2136 Basically, the advantage is still that you don't have to care which fields
2137 are or are not included. You could wrap that above C<for> loop in a simple
2138 sub called C<bind_fields()> or something and reuse it repeatedly. You still
2139 get a layer of abstraction over manual SQL specification.
2141 Note that if you set L</bindtype> to C<columns>, the C<\[ $sql, @bind ]>
2142 construct (see L</Literal SQL with placeholders and bind values (subqueries)>)
2143 will expect the bind values in this format.
2147 This is the character that a table or column name will be quoted
2148 with. By default this is an empty string, but you could set it to
2149 the character C<`>, to generate SQL like this:
2151 SELECT `a_field` FROM `a_table` WHERE `some_field` LIKE '%someval%'
2153 Alternatively, you can supply an array ref of two items, the first being the left
2154 hand quote character, and the second the right hand quote character. For
2155 example, you could supply C<['[',']']> for SQL Server 2000 compliant quotes
2156 that generates SQL like this:
2158 SELECT [a_field] FROM [a_table] WHERE [some_field] LIKE '%someval%'
2160 Quoting is useful if you have tables or columns names that are reserved
2161 words in your database's SQL dialect.
2165 This is the character that will be used to escape L</quote_char>s appearing
2166 in an identifier before it has been quoted.
2168 The parameter default in case of a single L</quote_char> character is the quote
2171 When opening-closing-style quoting is used (L</quote_char> is an arrayref)
2172 this parameter defaults to the B<closing (right)> L</quote_char>. Occurrences
2173 of the B<opening (left)> L</quote_char> within the identifier are currently left
2174 untouched. The default for opening-closing-style quotes may change in future
2175 versions, thus you are B<strongly encouraged> to specify the escape character
2180 This is the character that separates a table and column name. It is
2181 necessary to specify this when the C<quote_char> option is selected,
2182 so that tables and column names can be individually quoted like this:
2184 SELECT `table`.`one_field` FROM `table` WHERE `table`.`other_field` = 1
2186 =item injection_guard
2188 A regular expression C<qr/.../> that is applied to any C<-function> and unquoted
2189 column name specified in a query structure. This is a safety mechanism to avoid
2190 injection attacks when mishandling user input e.g.:
2192 my %condition_as_column_value_pairs = get_values_from_user();
2193 $sqla->select( ... , \%condition_as_column_value_pairs );
2195 If the expression matches an exception is thrown. Note that literal SQL
2196 supplied via C<\'...'> or C<\['...']> is B<not> checked in any way.
2198 Defaults to checking for C<;> and the C<GO> keyword (TransactSQL)
2200 =item array_datatypes
2202 When this option is true, arrayrefs in INSERT or UPDATE are
2203 interpreted as array datatypes and are passed directly
2205 When this option is false, arrayrefs are interpreted
2206 as literal SQL, just like refs to arrayrefs
2207 (but this behavior is for backwards compatibility; when writing
2208 new queries, use the "reference to arrayref" syntax
2214 Takes a reference to a list of "special operators"
2215 to extend the syntax understood by L<SQL::Abstract>.
2216 See section L</"SPECIAL OPERATORS"> for details.
2220 Takes a reference to a list of "unary operators"
2221 to extend the syntax understood by L<SQL::Abstract>.
2222 See section L</"UNARY OPERATORS"> for details.
2228 =head2 insert($table, \@values || \%fieldvals, \%options)
2230 This is the simplest function. You simply give it a table name
2231 and either an arrayref of values or hashref of field/value pairs.
2232 It returns an SQL INSERT statement and a list of bind values.
2233 See the sections on L</"Inserting and Updating Arrays"> and
2234 L</"Inserting and Updating SQL"> for information on how to insert
2235 with those data types.
2237 The optional C<\%options> hash reference may contain additional
2238 options to generate the insert SQL. Currently supported options
2245 Takes either a scalar of raw SQL fields, or an array reference of
2246 field names, and adds on an SQL C<RETURNING> statement at the end.
2247 This allows you to return data generated by the insert statement
2248 (such as row IDs) without performing another C<SELECT> statement.
2249 Note, however, this is not part of the SQL standard and may not
2250 be supported by all database engines.
2254 =head2 update($table, \%fieldvals, \%where, \%options)
2256 This takes a table, hashref of field/value pairs, and an optional
2257 hashref L<WHERE clause|/WHERE CLAUSES>. It returns an SQL UPDATE function and a list
2259 See the sections on L</"Inserting and Updating Arrays"> and
2260 L</"Inserting and Updating SQL"> for information on how to insert
2261 with those data types.
2263 The optional C<\%options> hash reference may contain additional
2264 options to generate the update SQL. Currently supported options
2271 See the C<returning> option to
2272 L<insert|/insert($table, \@values || \%fieldvals, \%options)>.
2276 =head2 select($source, $fields, $where, $order)
2278 This returns a SQL SELECT statement and associated list of bind values, as
2279 specified by the arguments:
2285 Specification of the 'FROM' part of the statement.
2286 The argument can be either a plain scalar (interpreted as a table
2287 name, will be quoted), or an arrayref (interpreted as a list
2288 of table names, joined by commas, quoted), or a scalarref
2289 (literal SQL, not quoted).
2293 Specification of the list of fields to retrieve from
2295 The argument can be either an arrayref (interpreted as a list
2296 of field names, will be joined by commas and quoted), or a
2297 plain scalar (literal SQL, not quoted).
2298 Please observe that this API is not as flexible as that of
2299 the first argument C<$source>, for backwards compatibility reasons.
2303 Optional argument to specify the WHERE part of the query.
2304 The argument is most often a hashref, but can also be
2305 an arrayref or plain scalar --
2306 see section L<WHERE clause|/"WHERE CLAUSES"> for details.
2310 Optional argument to specify the ORDER BY part of the query.
2311 The argument can be a scalar, a hashref or an arrayref
2312 -- see section L<ORDER BY clause|/"ORDER BY CLAUSES">
2318 =head2 delete($table, \%where, \%options)
2320 This takes a table name and optional hashref L<WHERE clause|/WHERE CLAUSES>.
2321 It returns an SQL DELETE statement and list of bind values.
2323 The optional C<\%options> hash reference may contain additional
2324 options to generate the delete SQL. Currently supported options
2331 See the C<returning> option to
2332 L<insert|/insert($table, \@values || \%fieldvals, \%options)>.
2336 =head2 where(\%where, $order)
2338 This is used to generate just the WHERE clause. For example,
2339 if you have an arbitrary data structure and know what the
2340 rest of your SQL is going to look like, but want an easy way
2341 to produce a WHERE clause, use this. It returns an SQL WHERE
2342 clause and list of bind values.
2345 =head2 values(\%data)
2347 This just returns the values from the hash C<%data>, in the same
2348 order that would be returned from any of the other above queries.
2349 Using this allows you to markedly speed up your queries if you
2350 are affecting lots of rows. See below under the L</"PERFORMANCE"> section.
2352 =head2 generate($any, 'number', $of, \@data, $struct, \%types)
2354 Warning: This is an experimental method and subject to change.
2356 This returns arbitrarily generated SQL. It's a really basic shortcut.
2357 It will return two different things, depending on return context:
2359 my($stmt, @bind) = $sql->generate('create table', \$table, \@fields);
2360 my $stmt_and_val = $sql->generate('create table', \$table, \@fields);
2362 These would return the following:
2364 # First calling form
2365 $stmt = "CREATE TABLE test (?, ?)";
2366 @bind = (field1, field2);
2368 # Second calling form
2369 $stmt_and_val = "CREATE TABLE test (field1, field2)";
2371 Depending on what you're trying to do, it's up to you to choose the correct
2372 format. In this example, the second form is what you would want.
2376 $sql->generate('alter session', { nls_date_format => 'MM/YY' });
2380 ALTER SESSION SET nls_date_format = 'MM/YY'
2382 You get the idea. Strings get their case twiddled, but everything
2383 else remains verbatim.
2385 =head1 EXPORTABLE FUNCTIONS
2387 =head2 is_plain_value
2389 Determines if the supplied argument is a plain value as understood by this
2394 =item * The value is C<undef>
2396 =item * The value is a non-reference
2398 =item * The value is an object with stringification overloading
2400 =item * The value is of the form C<< { -value => $anything } >>
2404 On failure returns C<undef>, on success returns a B<scalar> reference
2405 to the original supplied argument.
2411 The stringification overloading detection is rather advanced: it takes
2412 into consideration not only the presence of a C<""> overload, but if that
2413 fails also checks for enabled
2414 L<autogenerated versions of C<"">|overload/Magic Autogeneration>, based
2415 on either C<0+> or C<bool>.
2417 Unfortunately testing in the field indicates that this
2418 detection B<< may tickle a latent bug in perl versions before 5.018 >>,
2419 but only when very large numbers of stringifying objects are involved.
2420 At the time of writing ( Sep 2014 ) there is no clear explanation of
2421 the direct cause, nor is there a manageably small test case that reliably
2422 reproduces the problem.
2424 If you encounter any of the following exceptions in B<random places within
2425 your application stack> - this module may be to blame:
2427 Operation "ne": no method found,
2428 left argument in overloaded package <something>,
2429 right argument in overloaded package <something>
2433 Stub found while resolving method "???" overloading """" in package <something>
2435 If you fall victim to the above - please attempt to reduce the problem
2436 to something that could be sent to the L<SQL::Abstract developers
2437 |DBIx::Class/GETTING HELP/SUPPORT>
2438 (either publicly or privately). As a workaround in the meantime you can
2439 set C<$ENV{SQLA_ISVALUE_IGNORE_AUTOGENERATED_STRINGIFICATION}> to a true
2440 value, which will most likely eliminate your problem (at the expense of
2441 not being able to properly detect exotic forms of stringification).
2443 This notice and environment variable will be removed in a future version,
2444 as soon as the underlying problem is found and a reliable workaround is
2449 =head2 is_literal_value
2451 Determines if the supplied argument is a literal value as understood by this
2456 =item * C<\$sql_string>
2458 =item * C<\[ $sql_string, @bind_values ]>
2462 On failure returns C<undef>, on success returns an B<array> reference
2463 containing the unpacked version of the supplied literal SQL and bind values.
2465 =head1 WHERE CLAUSES
2469 This module uses a variation on the idea from L<DBIx::Abstract>. It
2470 is B<NOT>, repeat I<not> 100% compatible. B<The main logic of this
2471 module is that things in arrays are OR'ed, and things in hashes
2474 The easiest way to explain is to show lots of examples. After
2475 each C<%where> hash shown, it is assumed you used:
2477 my($stmt, @bind) = $sql->where(\%where);
2479 However, note that the C<%where> hash can be used directly in any
2480 of the other functions as well, as described above.
2482 =head2 Key-value pairs
2484 So, let's get started. To begin, a simple hash:
2488 status => 'completed'
2491 Is converted to SQL C<key = val> statements:
2493 $stmt = "WHERE user = ? AND status = ?";
2494 @bind = ('nwiger', 'completed');
2496 One common thing I end up doing is having a list of values that
2497 a field can be in. To do this, simply specify a list inside of
2502 status => ['assigned', 'in-progress', 'pending'];
2505 This simple code will create the following:
2507 $stmt = "WHERE user = ? AND ( status = ? OR status = ? OR status = ? )";
2508 @bind = ('nwiger', 'assigned', 'in-progress', 'pending');
2510 A field associated to an empty arrayref will be considered a
2511 logical false and will generate 0=1.
2513 =head2 Tests for NULL values
2515 If the value part is C<undef> then this is converted to SQL <IS NULL>
2524 $stmt = "WHERE user = ? AND status IS NULL";
2527 To test if a column IS NOT NULL:
2531 status => { '!=', undef },
2534 =head2 Specific comparison operators
2536 If you want to specify a different type of operator for your comparison,
2537 you can use a hashref for a given column:
2541 status => { '!=', 'completed' }
2544 Which would generate:
2546 $stmt = "WHERE user = ? AND status != ?";
2547 @bind = ('nwiger', 'completed');
2549 To test against multiple values, just enclose the values in an arrayref:
2551 status => { '=', ['assigned', 'in-progress', 'pending'] };
2553 Which would give you:
2555 "WHERE status = ? OR status = ? OR status = ?"
2558 The hashref can also contain multiple pairs, in which case it is expanded
2559 into an C<AND> of its elements:
2563 status => { '!=', 'completed', -not_like => 'pending%' }
2566 # Or more dynamically, like from a form
2567 $where{user} = 'nwiger';
2568 $where{status}{'!='} = 'completed';
2569 $where{status}{'-not_like'} = 'pending%';
2571 # Both generate this
2572 $stmt = "WHERE user = ? AND status != ? AND status NOT LIKE ?";
2573 @bind = ('nwiger', 'completed', 'pending%');
2576 To get an OR instead, you can combine it with the arrayref idea:
2580 priority => [ { '=', 2 }, { '>', 5 } ]
2583 Which would generate:
2585 $stmt = "WHERE ( priority = ? OR priority > ? ) AND user = ?";
2586 @bind = ('2', '5', 'nwiger');
2588 If you want to include literal SQL (with or without bind values), just use a
2589 scalar reference or reference to an arrayref as the value:
2592 date_entered => { '>' => \["to_date(?, 'MM/DD/YYYY')", "11/26/2008"] },
2593 date_expires => { '<' => \"now()" }
2596 Which would generate:
2598 $stmt = "WHERE date_entered > to_date(?, 'MM/DD/YYYY') AND date_expires < now()";
2599 @bind = ('11/26/2008');
2602 =head2 Logic and nesting operators
2604 In the example above,
2605 there is a subtle trap if you want to say something like
2606 this (notice the C<AND>):
2608 WHERE priority != ? AND priority != ?
2610 Because, in Perl you I<can't> do this:
2612 priority => { '!=' => 2, '!=' => 1 }
2614 As the second C<!=> key will obliterate the first. The solution
2615 is to use the special C<-modifier> form inside an arrayref:
2617 priority => [ -and => {'!=', 2},
2621 Normally, these would be joined by C<OR>, but the modifier tells it
2622 to use C<AND> instead. (Hint: You can use this in conjunction with the
2623 C<logic> option to C<new()> in order to change the way your queries
2624 work by default.) B<Important:> Note that the C<-modifier> goes
2625 B<INSIDE> the arrayref, as an extra first element. This will
2626 B<NOT> do what you think it might:
2628 priority => -and => [{'!=', 2}, {'!=', 1}] # WRONG!
2630 Here is a quick list of equivalencies, since there is some overlap:
2633 status => {'!=', 'completed', 'not like', 'pending%' }
2634 status => [ -and => {'!=', 'completed'}, {'not like', 'pending%'}]
2637 status => {'=', ['assigned', 'in-progress']}
2638 status => [ -or => {'=', 'assigned'}, {'=', 'in-progress'}]
2639 status => [ {'=', 'assigned'}, {'=', 'in-progress'} ]
2643 =head2 Special operators: IN, BETWEEN, etc.
2645 You can also use the hashref format to compare a list of fields using the
2646 C<IN> comparison operator, by specifying the list as an arrayref:
2649 status => 'completed',
2650 reportid => { -in => [567, 2335, 2] }
2653 Which would generate:
2655 $stmt = "WHERE status = ? AND reportid IN (?,?,?)";
2656 @bind = ('completed', '567', '2335', '2');
2658 The reverse operator C<-not_in> generates SQL C<NOT IN> and is used in
2661 If the argument to C<-in> is an empty array, 'sqlfalse' is generated
2662 (by default: C<1=0>). Similarly, C<< -not_in => [] >> generates
2663 'sqltrue' (by default: C<1=1>).
2665 In addition to the array you can supply a chunk of literal sql or
2666 literal sql with bind:
2669 customer => { -in => \[
2670 'SELECT cust_id FROM cust WHERE balance > ?',
2673 status => { -in => \'SELECT status_codes FROM states' },
2679 customer IN ( SELECT cust_id FROM cust WHERE balance > ? )
2680 AND status IN ( SELECT status_codes FROM states )
2684 Finally, if the argument to C<-in> is not a reference, it will be
2685 treated as a single-element array.
2687 Another pair of operators is C<-between> and C<-not_between>,
2688 used with an arrayref of two values:
2692 completion_date => {
2693 -not_between => ['2002-10-01', '2003-02-06']
2699 WHERE user = ? AND completion_date NOT BETWEEN ( ? AND ? )
2701 Just like with C<-in> all plausible combinations of literal SQL
2705 start0 => { -between => [ 1, 2 ] },
2706 start1 => { -between => \["? AND ?", 1, 2] },
2707 start2 => { -between => \"lower(x) AND upper(y)" },
2708 start3 => { -between => [
2710 \["upper(?)", 'stuff' ],
2717 ( start0 BETWEEN ? AND ? )
2718 AND ( start1 BETWEEN ? AND ? )
2719 AND ( start2 BETWEEN lower(x) AND upper(y) )
2720 AND ( start3 BETWEEN lower(x) AND upper(?) )
2722 @bind = (1, 2, 1, 2, 'stuff');
2725 These are the two builtin "special operators"; but the
2726 list can be expanded: see section L</"SPECIAL OPERATORS"> below.
2728 =head2 Unary operators: bool
2730 If you wish to test against boolean columns or functions within your
2731 database you can use the C<-bool> and C<-not_bool> operators. For
2732 example to test the column C<is_user> being true and the column
2733 C<is_enabled> being false you would use:-
2737 -not_bool => 'is_enabled',
2742 WHERE is_user AND NOT is_enabled
2744 If a more complex combination is required, testing more conditions,
2745 then you should use the and/or operators:-
2750 -not_bool => { two=> { -rlike => 'bar' } },
2751 -not_bool => { three => [ { '=', 2 }, { '>', 5 } ] },
2762 (NOT ( three = ? OR three > ? ))
2765 =head2 Nested conditions, -and/-or prefixes
2767 So far, we've seen how multiple conditions are joined with a top-level
2768 C<AND>. We can change this by putting the different conditions we want in
2769 hashes and then putting those hashes in an array. For example:
2774 status => { -like => ['pending%', 'dispatched'] },
2778 status => 'unassigned',
2782 This data structure would create the following:
2784 $stmt = "WHERE ( user = ? AND ( status LIKE ? OR status LIKE ? ) )
2785 OR ( user = ? AND status = ? ) )";
2786 @bind = ('nwiger', 'pending', 'dispatched', 'robot', 'unassigned');
2789 Clauses in hashrefs or arrayrefs can be prefixed with an C<-and> or C<-or>
2790 to change the logic inside:
2796 -and => [ workhrs => {'>', 20}, geo => 'ASIA' ],
2797 -or => { workhrs => {'<', 50}, geo => 'EURO' },
2804 $stmt = "WHERE ( user = ?
2805 AND ( ( workhrs > ? AND geo = ? )
2806 OR ( workhrs < ? OR geo = ? ) ) )";
2807 @bind = ('nwiger', '20', 'ASIA', '50', 'EURO');
2809 =head3 Algebraic inconsistency, for historical reasons
2811 C<Important note>: when connecting several conditions, the C<-and->|C<-or>
2812 operator goes C<outside> of the nested structure; whereas when connecting
2813 several constraints on one column, the C<-and> operator goes
2814 C<inside> the arrayref. Here is an example combining both features:
2817 -and => [a => 1, b => 2],
2818 -or => [c => 3, d => 4],
2819 e => [-and => {-like => 'foo%'}, {-like => '%bar'} ]
2824 WHERE ( ( ( a = ? AND b = ? )
2825 OR ( c = ? OR d = ? )
2826 OR ( e LIKE ? AND e LIKE ? ) ) )
2828 This difference in syntax is unfortunate but must be preserved for
2829 historical reasons. So be careful: the two examples below would
2830 seem algebraically equivalent, but they are not
2833 { -like => 'foo%' },
2834 { -like => '%bar' },
2836 # yields: WHERE ( ( col LIKE ? AND col LIKE ? ) )
2839 { col => { -like => 'foo%' } },
2840 { col => { -like => '%bar' } },
2842 # yields: WHERE ( ( col LIKE ? OR col LIKE ? ) )
2845 =head2 Literal SQL and value type operators
2847 The basic premise of SQL::Abstract is that in WHERE specifications the "left
2848 side" is a column name and the "right side" is a value (normally rendered as
2849 a placeholder). This holds true for both hashrefs and arrayref pairs as you
2850 see in the L</WHERE CLAUSES> examples above. Sometimes it is necessary to
2851 alter this behavior. There are several ways of doing so.
2855 This is a virtual operator that signals the string to its right side is an
2856 identifier (a column name) and not a value. For example to compare two
2857 columns you would write:
2860 priority => { '<', 2 },
2861 requestor => { -ident => 'submitter' },
2866 $stmt = "WHERE priority < ? AND requestor = submitter";
2869 If you are maintaining legacy code you may see a different construct as
2870 described in L</Deprecated usage of Literal SQL>, please use C<-ident> in new
2875 This is a virtual operator that signals that the construct to its right side
2876 is a value to be passed to DBI. This is for example necessary when you want
2877 to write a where clause against an array (for RDBMS that support such
2878 datatypes). For example:
2881 array => { -value => [1, 2, 3] }
2886 $stmt = 'WHERE array = ?';
2887 @bind = ([1, 2, 3]);
2889 Note that if you were to simply say:
2895 the result would probably not be what you wanted:
2897 $stmt = 'WHERE array = ? OR array = ? OR array = ?';
2902 Finally, sometimes only literal SQL will do. To include a random snippet
2903 of SQL verbatim, you specify it as a scalar reference. Consider this only
2904 as a last resort. Usually there is a better way. For example:
2907 priority => { '<', 2 },
2908 requestor => { -in => \'(SELECT name FROM hitmen)' },
2913 $stmt = "WHERE priority < ? AND requestor IN (SELECT name FROM hitmen)"
2916 Note that in this example, you only get one bind parameter back, since
2917 the verbatim SQL is passed as part of the statement.
2921 Never use untrusted input as a literal SQL argument - this is a massive
2922 security risk (there is no way to check literal snippets for SQL
2923 injections and other nastyness). If you need to deal with untrusted input
2924 use literal SQL with placeholders as described next.
2926 =head3 Literal SQL with placeholders and bind values (subqueries)
2928 If the literal SQL to be inserted has placeholders and bind values,
2929 use a reference to an arrayref (yes this is a double reference --
2930 not so common, but perfectly legal Perl). For example, to find a date
2931 in Postgres you can use something like this:
2934 date_column => \[ "= date '2008-09-30' - ?::integer", 10 ]
2939 $stmt = "WHERE ( date_column = date '2008-09-30' - ?::integer )"
2942 Note that you must pass the bind values in the same format as they are returned
2943 by L<where|/where(\%where, $order)>. This means that if you set L</bindtype>
2944 to C<columns>, you must provide the bind values in the
2945 C<< [ column_meta => value ] >> format, where C<column_meta> is an opaque
2946 scalar value; most commonly the column name, but you can use any scalar value
2947 (including references and blessed references), L<SQL::Abstract> will simply
2948 pass it through intact. So if C<bindtype> is set to C<columns> the above
2949 example will look like:
2952 date_column => \[ "= date '2008-09-30' - ?::integer", [ {} => 10 ] ]
2955 Literal SQL is especially useful for nesting parenthesized clauses in the
2956 main SQL query. Here is a first example:
2958 my ($sub_stmt, @sub_bind) = ("SELECT c1 FROM t1 WHERE c2 < ? AND c3 LIKE ?",
2962 bar => \["IN ($sub_stmt)" => @sub_bind],
2967 $stmt = "WHERE (foo = ? AND bar IN (SELECT c1 FROM t1
2968 WHERE c2 < ? AND c3 LIKE ?))";
2969 @bind = (1234, 100, "foo%");
2971 Other subquery operators, like for example C<"E<gt> ALL"> or C<"NOT IN">,
2972 are expressed in the same way. Of course the C<$sub_stmt> and
2973 its associated bind values can be generated through a former call
2976 my ($sub_stmt, @sub_bind)
2977 = $sql->select("t1", "c1", {c2 => {"<" => 100},
2978 c3 => {-like => "foo%"}});
2981 bar => \["> ALL ($sub_stmt)" => @sub_bind],
2984 In the examples above, the subquery was used as an operator on a column;
2985 but the same principle also applies for a clause within the main C<%where>
2986 hash, like an EXISTS subquery:
2988 my ($sub_stmt, @sub_bind)
2989 = $sql->select("t1", "*", {c1 => 1, c2 => \"> t0.c0"});
2990 my %where = ( -and => [
2992 \["EXISTS ($sub_stmt)" => @sub_bind],
2997 $stmt = "WHERE (foo = ? AND EXISTS (SELECT * FROM t1
2998 WHERE c1 = ? AND c2 > t0.c0))";
3002 Observe that the condition on C<c2> in the subquery refers to
3003 column C<t0.c0> of the main query: this is I<not> a bind
3004 value, so we have to express it through a scalar ref.
3005 Writing C<< c2 => {">" => "t0.c0"} >> would have generated
3006 C<< c2 > ? >> with bind value C<"t0.c0"> ... not exactly
3007 what we wanted here.
3009 Finally, here is an example where a subquery is used
3010 for expressing unary negation:
3012 my ($sub_stmt, @sub_bind)
3013 = $sql->where({age => [{"<" => 10}, {">" => 20}]});
3014 $sub_stmt =~ s/^ where //i; # don't want "WHERE" in the subclause
3016 lname => {like => '%son%'},
3017 \["NOT ($sub_stmt)" => @sub_bind],
3022 $stmt = "lname LIKE ? AND NOT ( age < ? OR age > ? )"
3023 @bind = ('%son%', 10, 20)
3025 =head3 Deprecated usage of Literal SQL
3027 Below are some examples of archaic use of literal SQL. It is shown only as
3028 reference for those who deal with legacy code. Each example has a much
3029 better, cleaner and safer alternative that users should opt for in new code.
3035 my %where = ( requestor => \'IS NOT NULL' )
3037 $stmt = "WHERE requestor IS NOT NULL"
3039 This used to be the way of generating NULL comparisons, before the handling
3040 of C<undef> got formalized. For new code please use the superior syntax as
3041 described in L</Tests for NULL values>.
3045 my %where = ( requestor => \'= submitter' )
3047 $stmt = "WHERE requestor = submitter"
3049 This used to be the only way to compare columns. Use the superior L</-ident>
3050 method for all new code. For example an identifier declared in such a way
3051 will be properly quoted if L</quote_char> is properly set, while the legacy
3052 form will remain as supplied.
3056 my %where = ( is_ready => \"", completed => { '>', '2012-12-21' } )
3058 $stmt = "WHERE completed > ? AND is_ready"
3059 @bind = ('2012-12-21')
3061 Using an empty string literal used to be the only way to express a boolean.
3062 For all new code please use the much more readable
3063 L<-bool|/Unary operators: bool> operator.
3069 These pages could go on for a while, since the nesting of the data
3070 structures this module can handle are pretty much unlimited (the
3071 module implements the C<WHERE> expansion as a recursive function
3072 internally). Your best bet is to "play around" with the module a
3073 little to see how the data structures behave, and choose the best
3074 format for your data based on that.
3076 And of course, all the values above will probably be replaced with
3077 variables gotten from forms or the command line. After all, if you
3078 knew everything ahead of time, you wouldn't have to worry about
3079 dynamically-generating SQL and could just hardwire it into your
3082 =head1 ORDER BY CLAUSES
3084 Some functions take an order by clause. This can either be a scalar (just a
3085 column name), a hashref of C<< { -desc => 'col' } >> or C<< { -asc => 'col' }
3086 >>, a scalarref, an arrayref-ref, or an arrayref of any of the previous
3089 Given | Will Generate
3090 ---------------------------------------------------------------
3092 'colA' | ORDER BY colA
3094 [qw/colA colB/] | ORDER BY colA, colB
3096 {-asc => 'colA'} | ORDER BY colA ASC
3098 {-desc => 'colB'} | ORDER BY colB DESC
3100 ['colA', {-asc => 'colB'}] | ORDER BY colA, colB ASC
3102 { -asc => [qw/colA colB/] } | ORDER BY colA ASC, colB ASC
3104 \'colA DESC' | ORDER BY colA DESC
3106 \[ 'FUNC(colA, ?)', $x ] | ORDER BY FUNC(colA, ?)
3107 | /* ...with $x bound to ? */
3110 { -asc => 'colA' }, | colA ASC,
3111 { -desc => [qw/colB/] }, | colB DESC,
3112 { -asc => [qw/colC colD/] },| colC ASC, colD ASC,
3113 \'colE DESC', | colE DESC,
3114 \[ 'FUNC(colF, ?)', $x ], | FUNC(colF, ?)
3115 ] | /* ...with $x bound to ? */
3116 ===============================================================
3120 =head1 SPECIAL OPERATORS
3122 my $sqlmaker = SQL::Abstract->new(special_ops => [
3126 my ($self, $field, $op, $arg) = @_;
3132 handler => 'method_name',
3136 A "special operator" is a SQL syntactic clause that can be
3137 applied to a field, instead of a usual binary operator.
3140 WHERE field IN (?, ?, ?)
3141 WHERE field BETWEEN ? AND ?
3142 WHERE MATCH(field) AGAINST (?, ?)
3144 Special operators IN and BETWEEN are fairly standard and therefore
3145 are builtin within C<SQL::Abstract> (as the overridable methods
3146 C<_where_field_IN> and C<_where_field_BETWEEN>). For other operators,
3147 like the MATCH .. AGAINST example above which is specific to MySQL,
3148 you can write your own operator handlers - supply a C<special_ops>
3149 argument to the C<new> method. That argument takes an arrayref of
3150 operator definitions; each operator definition is a hashref with two
3157 the regular expression to match the operator
3161 Either a coderef or a plain scalar method name. In both cases
3162 the expected return is C<< ($sql, @bind) >>.
3164 When supplied with a method name, it is simply called on the
3165 L<SQL::Abstract> object as:
3167 $self->$method_name($field, $op, $arg)
3171 $field is the LHS of the operator
3172 $op is the part that matched the handler regex
3175 When supplied with a coderef, it is called as:
3177 $coderef->($self, $field, $op, $arg)
3182 For example, here is an implementation
3183 of the MATCH .. AGAINST syntax for MySQL
3185 my $sqlmaker = SQL::Abstract->new(special_ops => [
3187 # special op for MySql MATCH (field) AGAINST(word1, word2, ...)
3188 {regex => qr/^match$/i,
3190 my ($self, $field, $op, $arg) = @_;
3191 $arg = [$arg] if not ref $arg;
3192 my $label = $self->_quote($field);
3193 my ($placeholder) = $self->_convert('?');
3194 my $placeholders = join ", ", (($placeholder) x @$arg);
3195 my $sql = $self->_sqlcase('match') . " ($label) "
3196 . $self->_sqlcase('against') . " ($placeholders) ";
3197 my @bind = $self->_bindtype($field, @$arg);
3198 return ($sql, @bind);
3205 =head1 UNARY OPERATORS
3207 my $sqlmaker = SQL::Abstract->new(unary_ops => [
3211 my ($self, $op, $arg) = @_;
3217 handler => 'method_name',
3221 A "unary operator" is a SQL syntactic clause that can be
3222 applied to a field - the operator goes before the field
3224 You can write your own operator handlers - supply a C<unary_ops>
3225 argument to the C<new> method. That argument takes an arrayref of
3226 operator definitions; each operator definition is a hashref with two
3233 the regular expression to match the operator
3237 Either a coderef or a plain scalar method name. In both cases
3238 the expected return is C<< $sql >>.
3240 When supplied with a method name, it is simply called on the
3241 L<SQL::Abstract> object as:
3243 $self->$method_name($op, $arg)
3247 $op is the part that matched the handler regex
3248 $arg is the RHS or argument of the operator
3250 When supplied with a coderef, it is called as:
3252 $coderef->($self, $op, $arg)
3260 Thanks to some benchmarking by Mark Stosberg, it turns out that
3261 this module is many orders of magnitude faster than using C<DBIx::Abstract>.
3262 I must admit this wasn't an intentional design issue, but it's a
3263 byproduct of the fact that you get to control your C<DBI> handles
3266 To maximize performance, use a code snippet like the following:
3268 # prepare a statement handle using the first row
3269 # and then reuse it for the rest of the rows
3271 for my $href (@array_of_hashrefs) {
3272 $stmt ||= $sql->insert('table', $href);
3273 $sth ||= $dbh->prepare($stmt);
3274 $sth->execute($sql->values($href));
3277 The reason this works is because the keys in your C<$href> are sorted
3278 internally by B<SQL::Abstract>. Thus, as long as your data retains
3279 the same structure, you only have to generate the SQL the first time
3280 around. On subsequent queries, simply use the C<values> function provided
3281 by this module to return your values in the correct order.
3283 However this depends on the values having the same type - if, for
3284 example, the values of a where clause may either have values
3285 (resulting in sql of the form C<column = ?> with a single bind
3286 value), or alternatively the values might be C<undef> (resulting in
3287 sql of the form C<column IS NULL> with no bind value) then the
3288 caching technique suggested will not work.
3292 If you use my C<CGI::FormBuilder> module at all, you'll hopefully
3293 really like this part (I do, at least). Building up a complex query
3294 can be as simple as the following:
3301 use CGI::FormBuilder;
3304 my $form = CGI::FormBuilder->new(...);
3305 my $sql = SQL::Abstract->new;
3307 if ($form->submitted) {
3308 my $field = $form->field;
3309 my $id = delete $field->{id};
3310 my($stmt, @bind) = $sql->update('table', $field, {id => $id});
3313 Of course, you would still have to connect using C<DBI> to run the
3314 query, but the point is that if you make your form look like your
3315 table, the actual query script can be extremely simplistic.
3317 If you're B<REALLY> lazy (I am), check out C<HTML::QuickTable> for
3318 a fast interface to returning and formatting data. I frequently
3319 use these three modules together to write complex database query
3320 apps in under 50 lines.
3322 =head1 HOW TO CONTRIBUTE
3324 Contributions are always welcome, in all usable forms (we especially
3325 welcome documentation improvements). The delivery methods include git-
3326 or unified-diff formatted patches, GitHub pull requests, or plain bug
3327 reports either via RT or the Mailing list. Contributors are generally
3328 granted full access to the official repository after their first several
3329 patches pass successful review.
3331 This project is maintained in a git repository. The code and related tools are
3332 accessible at the following locations:
3336 =item * Official repo: L<git://git.shadowcat.co.uk/dbsrgits/SQL-Abstract.git>
3338 =item * Official gitweb: L<http://git.shadowcat.co.uk/gitweb/gitweb.cgi?p=dbsrgits/SQL-Abstract.git>
3340 =item * GitHub mirror: L<https://github.com/dbsrgits/sql-abstract>
3342 =item * Authorized committers: L<ssh://dbsrgits@git.shadowcat.co.uk/SQL-Abstract.git>
3348 Version 1.50 was a major internal refactoring of C<SQL::Abstract>.
3349 Great care has been taken to preserve the I<published> behavior
3350 documented in previous versions in the 1.* family; however,
3351 some features that were previously undocumented, or behaved
3352 differently from the documentation, had to be changed in order
3353 to clarify the semantics. Hence, client code that was relying
3354 on some dark areas of C<SQL::Abstract> v1.*
3355 B<might behave differently> in v1.50.
3357 The main changes are:
3363 support for literal SQL through the C<< \ [ $sql, @bind ] >> syntax.
3367 support for the { operator => \"..." } construct (to embed literal SQL)
3371 support for the { operator => \["...", @bind] } construct (to embed literal SQL with bind values)
3375 optional support for L<array datatypes|/"Inserting and Updating Arrays">
3379 defensive programming: check arguments
3383 fixed bug with global logic, which was previously implemented
3384 through global variables yielding side-effects. Prior versions would
3385 interpret C<< [ {cond1, cond2}, [cond3, cond4] ] >>
3386 as C<< "(cond1 AND cond2) OR (cond3 AND cond4)" >>.
3387 Now this is interpreted
3388 as C<< "(cond1 AND cond2) OR (cond3 OR cond4)" >>.
3393 fixed semantics of _bindtype on array args
3397 dropped the C<_anoncopy> of the %where tree. No longer necessary,
3398 we just avoid shifting arrays within that tree.
3402 dropped the C<_modlogic> function
3406 =head1 ACKNOWLEDGEMENTS
3408 There are a number of individuals that have really helped out with
3409 this module. Unfortunately, most of them submitted bugs via CPAN
3410 so I have no idea who they are! But the people I do know are:
3412 Ash Berlin (order_by hash term support)
3413 Matt Trout (DBIx::Class support)
3414 Mark Stosberg (benchmarking)
3415 Chas Owens (initial "IN" operator support)
3416 Philip Collins (per-field SQL functions)
3417 Eric Kolve (hashref "AND" support)
3418 Mike Fragassi (enhancements to "BETWEEN" and "LIKE")
3419 Dan Kubb (support for "quote_char" and "name_sep")
3420 Guillermo Roditi (patch to cleanup "IN" and "BETWEEN", fix and tests for _order_by)
3421 Laurent Dami (internal refactoring, extensible list of special operators, literal SQL)
3422 Norbert Buchmuller (support for literal SQL in hashpair, misc. fixes & tests)
3423 Peter Rabbitson (rewrite of SQLA::Test, misc. fixes & tests)
3424 Oliver Charles (support for "RETURNING" after "INSERT")
3430 L<DBIx::Class>, L<DBIx::Abstract>, L<CGI::FormBuilder>, L<HTML::QuickTable>.
3434 Copyright (c) 2001-2007 Nathan Wiger <nwiger@cpan.org>. All Rights Reserved.
3436 This module is actively maintained by Matt Trout <mst@shadowcatsystems.co.uk>
3438 For support, your best bet is to try the C<DBIx::Class> users mailing list.
3439 While not an official support venue, C<DBIx::Class> makes heavy use of
3440 C<SQL::Abstract>, and as such list members there are very familiar with
3441 how to create queries.
3445 This module is free software; you may copy this under the same
3446 terms as perl itself (either the GNU General Public License or
3447 the Artistic License)