1 package SQL::Abstract; # see doc at end of file
3 # LDNOTE : this code is heavy refactoring from original SQLA.
4 # Several design decisions will need discussion during
5 # the test / diffusion / acceptance phase; those are marked with flag
6 # 'LDNOTE' (note by laurent.dami AT free.fr)
10 use warnings FATAL => 'all';
13 use Data::Query::Constants qw(
14 DQ_IDENTIFIER DQ_OPERATOR DQ_VALUE DQ_LITERAL DQ_JOIN DQ_SELECT DQ_ORDER
16 use Data::Query::ExprHelpers qw(perl_scalar_value);
18 #======================================================================
20 #======================================================================
22 our $VERSION = '1.72';
24 # This would confuse some packagers
25 $VERSION = eval $VERSION if $VERSION =~ /_/; # numify for warning-free dev releases
29 # special operators (-in, -between). May be extended/overridden by user.
30 # See section WHERE: BUILTIN SPECIAL OPERATORS below for implementation
31 my @BUILTIN_SPECIAL_OPS = ();
33 # unaryish operators - key maps to handler
34 my @BUILTIN_UNARY_OPS = (
35 # the digits are backcompat stuff
36 { regex => qr/^ and (?: [_\s]? \d+ )? $/xi, handler => '_where_op_ANDOR' },
37 { regex => qr/^ or (?: [_\s]? \d+ )? $/xi, handler => '_where_op_ANDOR' },
38 { regex => qr/^ nest (?: [_\s]? \d+ )? $/xi, handler => '_where_op_NEST' },
39 { regex => qr/^ (?: not \s )? bool $/xi, handler => '_where_op_BOOL' },
40 { regex => qr/^ ident $/xi, handler => '_where_op_IDENT' },
41 { regex => qr/^ value $/ix, handler => '_where_op_VALUE' },
44 #======================================================================
45 # DEBUGGING AND ERROR REPORTING
46 #======================================================================
49 return unless $_[0]->{debug}; shift; # a little faster
50 my $func = (caller(1))[3];
51 warn "[$func] ", @_, "\n";
55 my($func) = (caller(1))[3];
56 Carp::carp "[$func] Warning: ", @_;
60 my($func) = (caller(1))[3];
61 Carp::croak "[$func] Fatal: ", @_;
65 #======================================================================
67 #======================================================================
71 my $class = ref($self) || $self;
72 my %opt = (ref $_[0] eq 'HASH') ? %{$_[0]} : @_;
74 # choose our case by keeping an option around
75 delete $opt{case} if $opt{case} && $opt{case} ne 'lower';
77 # default logic for interpreting arrayrefs
78 $opt{logic} = $opt{logic} ? uc $opt{logic} : 'OR';
80 # how to return bind vars
81 # LDNOTE: changed nwiger code : why this 'delete' ??
82 # $opt{bindtype} ||= delete($opt{bind_type}) || 'normal';
83 $opt{bindtype} ||= 'normal';
85 # default comparison is "=", but can be overridden
88 # try to recognize which are the 'equality' and 'unequality' ops
89 # (temporary quickfix, should go through a more seasoned API)
90 $opt{equality_op} = qr/^(\Q$opt{cmp}\E|is|(is\s+)?like)$/i;
91 $opt{inequality_op} = qr/^(!=|<>|(is\s+)?not(\s+like)?)$/i;
94 $opt{sqltrue} ||= '1=1';
95 $opt{sqlfalse} ||= '0=1';
98 $opt{special_ops} ||= [];
99 # regexes are applied in order, thus push after user-defines
100 push @{$opt{special_ops}}, @BUILTIN_SPECIAL_OPS;
103 $opt{unary_ops} ||= [];
104 push @{$opt{unary_ops}}, @BUILTIN_UNARY_OPS;
106 # rudimentary saniy-check for user supplied bits treated as functions/operators
107 # If a purported function matches this regular expression, an exception is thrown.
108 # Literal SQL is *NOT* subject to this check, only functions (and column names
109 # when quoting is not in effect)
112 # need to guard against ()'s in column names too, but this will break tons of
113 # hacks... ideas anyone?
114 $opt{injection_guard} ||= qr/
120 $opt{name_sep} ||= '.';
122 $opt{renderer} ||= do {
123 require Data::Query::Renderer::SQL::Naive;
124 my ($always, $chars);
125 for ($opt{quote_char}) {
126 $chars = defined() ? (ref() ? $_ : [$_]) : ['',''];
129 Data::Query::Renderer::SQL::Naive->new({
130 quote_chars => $chars, always_quote => $always,
131 ($opt{case} ? (lc_keywords => 1) : ()), # always 'lower' if it exists
135 return bless \%opt, $class;
139 my ($self, $dq) = @_;
140 my ($sql, @bind) = @{$self->{renderer}->render($dq)};
142 ($self->{bindtype} eq 'normal'
143 ? ($sql, map $_->{value}, @bind)
144 : ($sql, map [ $_->{value_meta}, $_->{value} ], @bind)
150 my ($self, $literal) = @_;
152 ($literal, @bind) = @$literal if ref($literal) eq 'ARRAY';
157 (@bind ? (values => [ $self->_bind_to_dq(@bind) ]) : ()),
162 my ($self, @bind) = @_;
164 $self->{bindtype} eq 'normal'
165 ? map perl_scalar_value($_), @bind
167 $self->_assert_bindval_matches_bindtype(@bind);
168 map perl_scalar_value(reverse @$_), @bind
173 my ($self, $value) = @_;
174 $self->_maybe_convert_dq(perl_scalar_value($value, our $Cur_Col_Meta));
178 my ($self, $ident) = @_;
179 $self->_assert_pass_injection_guard($ident)
180 unless $self->{renderer}{always_quote};
181 $self->_maybe_convert_dq({
182 type => DQ_IDENTIFIER,
183 elements => [ split /\Q$self->{name_sep}/, $ident ],
187 sub _maybe_convert_dq {
188 my ($self, $dq) = @_;
189 if (my $c = $self->{where_convert}) {
192 operator => { 'SQL.Naive' => 'apply' },
194 { type => DQ_IDENTIFIER, elements => [ $self->_sqlcase($c) ] },
204 my ($self, $op, @args) = @_;
205 $self->_assert_pass_injection_guard($op);
208 operator => { 'SQL.Naive' => $op },
213 sub _assert_pass_injection_guard {
214 if ($_[1] =~ $_[0]->{injection_guard}) {
215 my $class = ref $_[0];
216 puke "Possible SQL injection attempt '$_[1]'. If this is indeed a part of the "
217 . "desired SQL use literal SQL ( \'...' or \[ '...' ] ) or supply your own "
218 . "{injection_guard} attribute to ${class}->new()"
223 #======================================================================
225 #======================================================================
229 my $table = $self->_table(shift);
230 my $data = shift || return;
233 my $method = $self->_METHOD_FOR_refkind("_insert", $data);
234 my ($sql, @bind) = $self->$method($data);
235 $sql = join " ", $self->_sqlcase('insert into'), $table, $sql;
237 if ($options->{returning}) {
238 my ($s, @b) = $self->_insert_returning ($options);
243 return wantarray ? ($sql, @bind) : $sql;
246 sub _insert_returning {
247 my ($self, $options) = @_;
249 my $f = $options->{returning};
251 my $fieldlist = $self->_SWITCH_refkind($f, {
252 ARRAYREF => sub {join ', ', map { $self->_quote($_) } @$f;},
253 SCALAR => sub {$self->_quote($f)},
254 SCALARREF => sub {$$f},
256 return $self->_sqlcase(' returning ') . $fieldlist;
259 sub _insert_HASHREF { # explicit list of fields and then values
260 my ($self, $data) = @_;
262 my @fields = sort keys %$data;
264 my ($sql, @bind) = $self->_insert_values($data);
267 $_ = $self->_quote($_) foreach @fields;
268 $sql = "( ".join(", ", @fields).") ".$sql;
270 return ($sql, @bind);
273 sub _insert_ARRAYREF { # just generate values(?,?) part (no list of fields)
274 my ($self, $data) = @_;
276 # no names (arrayref) so can't generate bindtype
277 $self->{bindtype} ne 'columns'
278 or belch "can't do 'columns' bindtype when called with arrayref";
280 # fold the list of values into a hash of column name - value pairs
281 # (where the column names are artificially generated, and their
282 # lexicographical ordering keep the ordering of the original list)
283 my $i = "a"; # incremented values will be in lexicographical order
284 my $data_in_hash = { map { ($i++ => $_) } @$data };
286 return $self->_insert_values($data_in_hash);
289 sub _insert_ARRAYREFREF { # literal SQL with bind
290 my ($self, $data) = @_;
292 my ($sql, @bind) = @${$data};
293 $self->_assert_bindval_matches_bindtype(@bind);
295 return ($sql, @bind);
299 sub _insert_SCALARREF { # literal SQL without bind
300 my ($self, $data) = @_;
306 my ($self, $data) = @_;
308 my (@values, @all_bind);
309 foreach my $column (sort keys %$data) {
310 my $v = $data->{$column};
312 $self->_SWITCH_refkind($v, {
315 if ($self->{array_datatypes}) { # if array datatype are activated
317 push @all_bind, $self->_bindtype($column, $v);
319 else { # else literal SQL with bind
320 my ($sql, @bind) = @$v;
321 $self->_assert_bindval_matches_bindtype(@bind);
323 push @all_bind, @bind;
327 ARRAYREFREF => sub { # literal SQL with bind
328 my ($sql, @bind) = @${$v};
329 $self->_assert_bindval_matches_bindtype(@bind);
331 push @all_bind, @bind;
334 # THINK : anything useful to do with a HASHREF ?
335 HASHREF => sub { # (nothing, but old SQLA passed it through)
336 #TODO in SQLA >= 2.0 it will die instead
337 belch "HASH ref as bind value in insert is not supported";
339 push @all_bind, $self->_bindtype($column, $v);
342 SCALARREF => sub { # literal SQL without bind
346 SCALAR_or_UNDEF => sub {
348 push @all_bind, $self->_bindtype($column, $v);
355 my $sql = $self->_sqlcase('values')." ( ".join(", ", @values)." )";
356 return ($sql, @all_bind);
361 #======================================================================
363 #======================================================================
368 my $table = $self->_table(shift);
369 my $data = shift || return;
372 # first build the 'SET' part of the sql statement
373 my (@set, @all_bind);
374 puke "Unsupported data type specified to \$sql->update"
375 unless ref $data eq 'HASH';
377 for my $k (sort keys %$data) {
380 my $label = $self->_quote($k);
382 $self->_SWITCH_refkind($v, {
384 if ($self->{array_datatypes}) { # array datatype
385 push @set, "$label = ?";
386 push @all_bind, $self->_bindtype($k, $v);
388 else { # literal SQL with bind
389 my ($sql, @bind) = @$v;
390 $self->_assert_bindval_matches_bindtype(@bind);
391 push @set, "$label = $sql";
392 push @all_bind, @bind;
395 ARRAYREFREF => sub { # literal SQL with bind
396 my ($sql, @bind) = @${$v};
397 $self->_assert_bindval_matches_bindtype(@bind);
398 push @set, "$label = $sql";
399 push @all_bind, @bind;
401 SCALARREF => sub { # literal SQL without bind
402 push @set, "$label = $$v";
405 my ($op, $arg, @rest) = %$v;
407 puke 'Operator calls in update must be in the form { -op => $arg }'
408 if (@rest or not $op =~ /^\-(.+)/);
410 local $self->{_nested_func_lhs} = $k;
411 local our $Cur_Col_Meta = $k;
412 my ($sql, @bind) = $self->_render_dq($self->_where_to_dq({ $op => $arg }));
414 push @set, "$label = $sql";
415 push @all_bind, @bind;
417 SCALAR_or_UNDEF => sub {
418 push @set, "$label = ?";
419 push @all_bind, $self->_bindtype($k, $v);
425 my $sql = $self->_sqlcase('update') . " $table " . $self->_sqlcase('set ')
429 my($where_sql, @where_bind) = $self->where($where);
431 push @all_bind, @where_bind;
434 return wantarray ? ($sql, @all_bind) : $sql;
440 #======================================================================
442 #======================================================================
448 my $fields = shift || '*';
452 my($where_sql, @bind) = $self->where($where, $order);
454 my $sql = $self->_render_dq({
457 map $self->_ident_to_dq($_),
458 ref($fields) eq 'ARRAY' ? @$fields : $fields
460 from => $self->_table_to_dq($table),
465 return wantarray ? ($sql, @bind) : $sql;
468 #======================================================================
470 #======================================================================
475 my $table = $self->_table(shift);
479 my($where_sql, @bind) = $self->where($where);
480 my $sql = $self->_sqlcase('delete from') . " $table" . $where_sql;
482 return wantarray ? ($sql, @bind) : $sql;
486 #======================================================================
488 #======================================================================
492 # Finally, a separate routine just to handle WHERE clauses
494 my ($self, $where, $order) = @_;
500 ($sql, @bind) = $self->_recurse_where($where) if defined($where);
501 $sql = $sql ? $self->_sqlcase(' where ') . "( $sql )" : '';
505 $sql .= $self->_order_by($order);
508 return wantarray ? ($sql, @bind) : $sql;
513 my ($self, $where, $logic) = @_;
515 # turn the convert misfeature on - only used in WHERE clauses
516 local $self->{where_convert} = $self->{convert};
518 return $self->_render_dq($self->_where_to_dq($where, $logic));
522 my ($self, $where, $logic) = @_;
524 if (ref($where) eq 'ARRAY') {
525 return $self->_where_to_dq_ARRAYREF($where, $logic);
526 } elsif (ref($where) eq 'HASH') {
527 return $self->_where_to_dq_HASHREF($where, $logic);
529 ref($where) eq 'SCALAR'
530 or (ref($where) eq 'REF' and ref($$where) eq 'ARRAY')
532 return $self->_literal_to_dq($$where);
533 } elsif (!ref($where) or Scalar::Util::blessed($where)) {
534 return $self->_value_to_dq($where);
536 die "Can't handle $where";
539 sub _where_to_dq_ARRAYREF {
540 my ($self, $where, $logic) = @_;
542 $logic = uc($logic || $self->{logic} || 'OR');
543 $logic eq 'AND' or $logic eq 'OR' or puke "unknown logic: $logic";
545 return unless @$where;
547 my ($first, @rest) = @$where;
549 return $self->_where_to_dq($first) unless @rest;
553 $self->_where_hashpair_to_dq($first => shift(@rest));
555 $self->_where_to_dq($first);
559 return $self->_where_to_dq_ARRAYREF(\@rest, $logic) unless $first_dq;
562 $logic, $first_dq, $self->_where_to_dq_ARRAYREF(\@rest, $logic)
566 sub _where_to_dq_HASHREF {
567 my ($self, $where, $logic) = @_;
569 $logic = uc($logic) if $logic;
572 $self->_where_hashpair_to_dq($_ => $where->{$_}, $logic)
575 return $dq[0] unless @dq > 1;
577 my $final = pop(@dq);
579 foreach my $dq (reverse @dq) {
580 $final = $self->_op_to_dq($logic||'AND', $dq, $final);
586 sub _where_to_dq_SCALAR {
587 shift->_value_to_dq(@_);
590 sub _where_op_IDENT {
592 my ($op, $rhs) = splice @_, -2;
594 puke "-$op takes a single scalar argument (a quotable identifier)";
597 # in case we are called as a top level special op (no '=')
600 $_ = $self->_convert($self->_quote($_)) for ($lhs, $rhs);
608 sub _where_op_VALUE {
610 my ($op, $rhs) = splice @_, -2;
612 # in case we are called as a top level special op (no '=')
617 ($lhs || $self->{_nested_func_lhs}),
624 $self->_convert($self->_quote($lhs)) . ' = ' . $self->_convert('?'),
628 $self->_convert('?'),
634 sub _where_hashpair_to_dq {
635 my ($self, $k, $v, $logic) = @_;
637 if ($k =~ /^-(.*)/s) {
639 if ($op eq 'AND' or $op eq 'OR') {
640 return $self->_where_to_dq($v, $op);
641 } elsif ($op eq 'NEST') {
642 return $self->_where_to_dq($v);
643 } elsif ($op eq 'NOT') {
644 return $self->_op_to_dq(NOT => $self->_where_to_dq($v));
645 } elsif ($op eq 'BOOL') {
646 return ref($v) ? $self->_where_to_dq($v) : $self->_ident_to_dq($v);
647 } elsif ($op eq 'NOT_BOOL') {
648 return $self->_op_to_dq(
649 NOT => ref($v) ? $self->_where_to_dq($v) : $self->_ident_to_dq($v)
651 } elsif ($op =~ /^(?:AND|OR|NEST)_?\d+/) {
652 die "Use of [and|or|nest]_N modifiers is no longer supported";
655 if (ref($v) eq 'HASH' and keys(%$v) == 1 and (keys %$v)[0] =~ /^-(.*)/s) {
657 my ($inner) = values %$v;
660 (map $self->_where_to_dq($_),
661 (ref($inner) eq 'ARRAY' ? @$inner : $inner))
664 (map $self->_where_to_dq($_), (ref($v) eq 'ARRAY' ? @$v : $v))
667 $self->_assert_pass_injection_guard($op);
668 return $self->_op_to_dq(
669 apply => $self->_ident_to_dq($op), @args
673 local our $Cur_Col_Meta = $k;
674 if (ref($v) eq 'ARRAY') {
676 return $self->_literal_to_dq($self->{sqlfalse});
677 } elsif (defined($v->[0]) && $v->[0] =~ /-(and|or)/i) {
678 return $self->_where_to_dq_ARRAYREF([
679 map +{ $k => $_ }, @{$v}[1..$#$v]
682 return $self->_where_to_dq_ARRAYREF([
683 map +{ $k => $_ }, @$v
685 } elsif (ref($v) eq 'SCALAR' or (ref($v) eq 'REF' and ref($$v) eq 'ARRAY')) {
689 parts => [ $self->_ident_to_dq($k), $self->_literal_to_dq($$v) ]
692 my ($op, $rhs) = do {
693 if (ref($v) eq 'HASH') {
695 return $self->_where_to_dq_ARRAYREF([
696 map +{ $k => { $_ => $v->{$_} } }, sort keys %$v
699 my ($op, $value) = %$v;
700 s/^-//, s/_/ /g for $op;
701 if ($op =~ /^(and|or)$/i) {
702 return $self->_where_to_dq({ $k => $value }, $op);
704 my $special_op = List::Util::first {$op =~ $_->{regex}}
705 @{$self->{special_ops}}
707 return $self->_literal_to_dq(
708 [ $self->${\$special_op->{handler}}($k, $op, $value) ]
710 } elsif ($op =~ /^(?:AND|OR|NEST)_?\d+$/i) {
711 die "Use of [and|or|nest]_N modifiers is no longer supported";
718 if ($op eq 'BETWEEN' or $op eq 'IN' or $op eq 'NOT IN' or $op eq 'NOT BETWEEN') {
719 if (ref($rhs) ne 'ARRAY') {
721 # have to add parens if none present because -in => \"SELECT ..."
722 # got documented. mst hates everything.
723 if (ref($rhs) eq 'SCALAR') {
725 1 while ($x =~ s/\A\s*\((.*)\)\s*\Z/$1/s);
728 my ($x, @rest) = @{$$rhs};
729 1 while ($x =~ s/\A\s*\((.*)\)\s*\Z/$1/s);
730 $rhs = \[ $x, @rest ];
733 return $self->_op_to_dq(
734 $op, $self->_ident_to_dq($k), $self->_literal_to_dq($$rhs)
737 return $self->_literal_to_dq($self->{sqlfalse}) unless @$rhs;
738 return $self->_op_to_dq(
739 $op, $self->_ident_to_dq($k), map $self->_where_to_dq($_), @$rhs
741 } elsif ($op =~ s/^NOT (?!LIKE)//) {
742 return $self->_where_hashpair_to_dq(-not => { $k => { $op => $rhs } });
743 } elsif (!defined($rhs)) {
745 if ($op eq '=' or $op eq 'LIKE') {
747 } elsif ($op eq '!=') {
750 die "Can't do undef -> NULL transform for operator ${op}";
753 return $self->_op_to_dq($null_op, $self->_ident_to_dq($k));
755 if (ref($rhs) eq 'ARRAY') {
757 return $self->_literal_to_dq(
758 $op eq '!=' ? $self->{sqltrue} : $self->{sqlfalse}
760 } elsif (defined($rhs->[0]) and $rhs->[0] =~ /^-(and|or)$/i) {
761 return $self->_where_to_dq_ARRAYREF([
762 map +{ $k => { $op => $_ } }, @{$rhs}[1..$#$rhs]
764 } elsif ($op =~ /^-(?:AND|OR|NEST)_?\d+/) {
765 die "Use of [and|or|nest]_N modifiers is no longer supported";
767 return $self->_where_to_dq_ARRAYREF([
768 map +{ $k => { $op => $_ } }, @$rhs
771 return $self->_op_to_dq(
772 $op, $self->_ident_to_dq($k), $self->_where_to_dq($rhs)
777 #======================================================================
779 #======================================================================
782 my ($self, $arg) = @_;
783 if (my $dq = $self->_order_by_to_dq($arg)) {
784 # SQLA generates ' ORDER BY foo'. The hilarity.
786 ? do { my @r = $self->_render_dq($dq); $r[0] = ' '.$r[0]; @r }
787 : ' '.$self->_render_dq($dq);
793 sub _order_by_to_dq {
794 my ($self, $arg, $dir) = @_;
800 ($dir ? (direction => $dir) : ()),
804 $dq->{by} = $self->_ident_to_dq($arg);
805 } elsif (ref($arg) eq 'ARRAY') {
807 local our $Order_Inner unless our $Order_Recursing;
808 local $Order_Recursing = 1;
810 foreach my $member (@$arg) {
812 my $next = $self->_order_by_to_dq($member, $dir);
814 $inner->{from} = $next if $inner;
815 $inner = $Order_Inner || $next;
817 $Order_Inner = $inner;
819 } elsif (ref($arg) eq 'REF' and ref($$arg) eq 'ARRAY') {
820 $dq->{by} = $self->_literal_to_dq($$arg);
821 } elsif (ref($arg) eq 'SCALAR') {
822 $dq->{by} = $self->_literal_to_dq($$arg);
823 } elsif (ref($arg) eq 'HASH') {
824 my ($key, $val, @rest) = %$arg;
828 if (@rest or not $key =~ /^-(desc|asc)/i) {
829 puke "hash passed to _order_by must have exactly one key (-desc or -asc)";
832 return $self->_order_by_to_dq($val, $dir);
834 die "Can't handle $arg in _order_by_to_dq";
839 #======================================================================
840 # DATASOURCE (FOR NOW, JUST PLAIN TABLE OR LIST OF TABLES)
841 #======================================================================
844 my ($self, $from) = @_;
845 $self->_render_dq($self->_table_to_dq($from));
849 my ($self, $from) = @_;
850 $self->_SWITCH_refkind($from, {
852 die "Empty FROM list" unless my @f = @$from;
853 my $dq = $self->_ident_to_dq(shift @f);
854 while (my $x = shift @f) {
857 join => [ $dq, $self->_ident_to_dq($x) ]
862 SCALAR => sub { $self->_ident_to_dq($from) },
874 #======================================================================
876 #======================================================================
878 # highly optimized, as it's called way too often
880 # my ($self, $label) = @_;
882 return '' unless defined $_[1];
883 return ${$_[1]} if ref($_[1]) eq 'SCALAR';
885 unless ($_[0]->{quote_char}) {
886 $_[0]->_assert_pass_injection_guard($_[1]);
890 my $qref = ref $_[0]->{quote_char};
893 ($l, $r) = ( $_[0]->{quote_char}, $_[0]->{quote_char} );
895 elsif ($qref eq 'ARRAY') {
896 ($l, $r) = @{$_[0]->{quote_char}};
899 puke "Unsupported quote_char format: $_[0]->{quote_char}";
902 # parts containing * are naturally unquoted
903 return join( $_[0]->{name_sep}||'', map
904 { $_ eq '*' ? $_ : $l . $_ . $r }
905 ( $_[0]->{name_sep} ? split (/\Q$_[0]->{name_sep}\E/, $_[1] ) : $_[1] )
910 # Conversion, if applicable
912 #my ($self, $arg) = @_;
914 # LDNOTE : modified the previous implementation below because
915 # it was not consistent : the first "return" is always an array,
916 # the second "return" is context-dependent. Anyway, _convert
917 # seems always used with just a single argument, so make it a
919 # return @_ unless $self->{convert};
920 # my $conv = $self->_sqlcase($self->{convert});
921 # my @ret = map { $conv.'('.$_.')' } @_;
922 # return wantarray ? @ret : $ret[0];
923 if ($_[0]->{convert}) {
924 return $_[0]->_sqlcase($_[0]->{convert}) .'(' . $_[1] . ')';
931 #my ($self, $col, @vals) = @_;
933 #LDNOTE : changed original implementation below because it did not make
934 # sense when bindtype eq 'columns' and @vals > 1.
935 # return $self->{bindtype} eq 'columns' ? [ $col, @vals ] : @vals;
937 # called often - tighten code
938 return $_[0]->{bindtype} eq 'columns'
939 ? map {[$_[1], $_]} @_[2 .. $#_]
944 # Dies if any element of @bind is not in [colname => value] format
945 # if bindtype is 'columns'.
946 sub _assert_bindval_matches_bindtype {
947 # my ($self, @bind) = @_;
949 if ($self->{bindtype} eq 'columns') {
951 if (!defined $_ || ref($_) ne 'ARRAY' || @$_ != 2) {
952 puke "bindtype 'columns' selected, you need to pass: [column_name => bind_value]"
958 sub _join_sql_clauses {
959 my ($self, $logic, $clauses_aref, $bind_aref) = @_;
961 if (@$clauses_aref > 1) {
962 my $join = " " . $self->_sqlcase($logic) . " ";
963 my $sql = '( ' . join($join, @$clauses_aref) . ' )';
964 return ($sql, @$bind_aref);
966 elsif (@$clauses_aref) {
967 return ($clauses_aref->[0], @$bind_aref); # no parentheses
970 return (); # if no SQL, ignore @$bind_aref
975 # Fix SQL case, if so requested
977 # LDNOTE: if $self->{case} is true, then it contains 'lower', so we
978 # don't touch the argument ... crooked logic, but let's not change it!
979 return $_[0]->{case} ? $_[1] : uc($_[1]);
983 #======================================================================
984 # DISPATCHING FROM REFKIND
985 #======================================================================
988 my ($self, $data) = @_;
990 return 'UNDEF' unless defined $data;
992 # blessed objects are treated like scalars
993 my $ref = (Scalar::Util::blessed $data) ? '' : ref $data;
995 return 'SCALAR' unless $ref;
998 while ($ref eq 'REF') {
1000 $ref = (Scalar::Util::blessed $data) ? '' : ref $data;
1004 return ($ref||'SCALAR') . ('REF' x $n_steps);
1008 my ($self, $data) = @_;
1009 my @try = ($self->_refkind($data));
1010 push @try, 'SCALAR_or_UNDEF' if $try[0] eq 'SCALAR' || $try[0] eq 'UNDEF';
1011 push @try, 'FALLBACK';
1015 sub _METHOD_FOR_refkind {
1016 my ($self, $meth_prefix, $data) = @_;
1019 for (@{$self->_try_refkind($data)}) {
1020 $method = $self->can($meth_prefix."_".$_)
1024 return $method || puke "cannot dispatch on '$meth_prefix' for ".$self->_refkind($data);
1028 sub _SWITCH_refkind {
1029 my ($self, $data, $dispatch_table) = @_;
1032 for (@{$self->_try_refkind($data)}) {
1033 $coderef = $dispatch_table->{$_}
1037 puke "no dispatch entry for ".$self->_refkind($data)
1046 #======================================================================
1047 # VALUES, GENERATE, AUTOLOAD
1048 #======================================================================
1050 # LDNOTE: original code from nwiger, didn't touch code in that section
1051 # I feel the AUTOLOAD stuff should not be the default, it should
1052 # only be activated on explicit demand by user.
1056 my $data = shift || return;
1057 puke "Argument to ", __PACKAGE__, "->values must be a \\%hash"
1058 unless ref $data eq 'HASH';
1061 foreach my $k ( sort keys %$data ) {
1062 my $v = $data->{$k};
1063 $self->_SWITCH_refkind($v, {
1065 if ($self->{array_datatypes}) { # array datatype
1066 push @all_bind, $self->_bindtype($k, $v);
1068 else { # literal SQL with bind
1069 my ($sql, @bind) = @$v;
1070 $self->_assert_bindval_matches_bindtype(@bind);
1071 push @all_bind, @bind;
1074 ARRAYREFREF => sub { # literal SQL with bind
1075 my ($sql, @bind) = @${$v};
1076 $self->_assert_bindval_matches_bindtype(@bind);
1077 push @all_bind, @bind;
1079 SCALARREF => sub { # literal SQL without bind
1081 SCALAR_or_UNDEF => sub {
1082 push @all_bind, $self->_bindtype($k, $v);
1093 my(@sql, @sqlq, @sqlv);
1097 if ($ref eq 'HASH') {
1098 for my $k (sort keys %$_) {
1101 my $label = $self->_quote($k);
1102 if ($r eq 'ARRAY') {
1103 # literal SQL with bind
1104 my ($sql, @bind) = @$v;
1105 $self->_assert_bindval_matches_bindtype(@bind);
1106 push @sqlq, "$label = $sql";
1108 } elsif ($r eq 'SCALAR') {
1109 # literal SQL without bind
1110 push @sqlq, "$label = $$v";
1112 push @sqlq, "$label = ?";
1113 push @sqlv, $self->_bindtype($k, $v);
1116 push @sql, $self->_sqlcase('set'), join ', ', @sqlq;
1117 } elsif ($ref eq 'ARRAY') {
1118 # unlike insert(), assume these are ONLY the column names, i.e. for SQL
1121 if ($r eq 'ARRAY') { # literal SQL with bind
1122 my ($sql, @bind) = @$v;
1123 $self->_assert_bindval_matches_bindtype(@bind);
1126 } elsif ($r eq 'SCALAR') { # literal SQL without bind
1127 # embedded literal SQL
1134 push @sql, '(' . join(', ', @sqlq) . ')';
1135 } elsif ($ref eq 'SCALAR') {
1139 # strings get case twiddled
1140 push @sql, $self->_sqlcase($_);
1144 my $sql = join ' ', @sql;
1146 # this is pretty tricky
1147 # if ask for an array, return ($stmt, @bind)
1148 # otherwise, s/?/shift @sqlv/ to put it inline
1150 return ($sql, @sqlv);
1152 1 while $sql =~ s/\?/my $d = shift(@sqlv);
1153 ref $d ? $d->[1] : $d/e;
1162 # This allows us to check for a local, then _form, attr
1164 my($name) = $AUTOLOAD =~ /.*::(.+)/;
1165 return $self->generate($name, @_);
1176 SQL::Abstract - Generate SQL from Perl data structures
1182 my $sql = SQL::Abstract->new;
1184 my($stmt, @bind) = $sql->select($table, \@fields, \%where, \@order);
1186 my($stmt, @bind) = $sql->insert($table, \%fieldvals || \@values);
1188 my($stmt, @bind) = $sql->update($table, \%fieldvals, \%where);
1190 my($stmt, @bind) = $sql->delete($table, \%where);
1192 # Then, use these in your DBI statements
1193 my $sth = $dbh->prepare($stmt);
1194 $sth->execute(@bind);
1196 # Just generate the WHERE clause
1197 my($stmt, @bind) = $sql->where(\%where, \@order);
1199 # Return values in the same order, for hashed queries
1200 # See PERFORMANCE section for more details
1201 my @bind = $sql->values(\%fieldvals);
1205 This module was inspired by the excellent L<DBIx::Abstract>.
1206 However, in using that module I found that what I really wanted
1207 to do was generate SQL, but still retain complete control over my
1208 statement handles and use the DBI interface. So, I set out to
1209 create an abstract SQL generation module.
1211 While based on the concepts used by L<DBIx::Abstract>, there are
1212 several important differences, especially when it comes to WHERE
1213 clauses. I have modified the concepts used to make the SQL easier
1214 to generate from Perl data structures and, IMO, more intuitive.
1215 The underlying idea is for this module to do what you mean, based
1216 on the data structures you provide it. The big advantage is that
1217 you don't have to modify your code every time your data changes,
1218 as this module figures it out.
1220 To begin with, an SQL INSERT is as easy as just specifying a hash
1221 of C<key=value> pairs:
1224 name => 'Jimbo Bobson',
1225 phone => '123-456-7890',
1226 address => '42 Sister Lane',
1227 city => 'St. Louis',
1228 state => 'Louisiana',
1231 The SQL can then be generated with this:
1233 my($stmt, @bind) = $sql->insert('people', \%data);
1235 Which would give you something like this:
1237 $stmt = "INSERT INTO people
1238 (address, city, name, phone, state)
1239 VALUES (?, ?, ?, ?, ?)";
1240 @bind = ('42 Sister Lane', 'St. Louis', 'Jimbo Bobson',
1241 '123-456-7890', 'Louisiana');
1243 These are then used directly in your DBI code:
1245 my $sth = $dbh->prepare($stmt);
1246 $sth->execute(@bind);
1248 =head2 Inserting and Updating Arrays
1250 If your database has array types (like for example Postgres),
1251 activate the special option C<< array_datatypes => 1 >>
1252 when creating the C<SQL::Abstract> object.
1253 Then you may use an arrayref to insert and update database array types:
1255 my $sql = SQL::Abstract->new(array_datatypes => 1);
1257 planets => [qw/Mercury Venus Earth Mars/]
1260 my($stmt, @bind) = $sql->insert('solar_system', \%data);
1264 $stmt = "INSERT INTO solar_system (planets) VALUES (?)"
1266 @bind = (['Mercury', 'Venus', 'Earth', 'Mars']);
1269 =head2 Inserting and Updating SQL
1271 In order to apply SQL functions to elements of your C<%data> you may
1272 specify a reference to an arrayref for the given hash value. For example,
1273 if you need to execute the Oracle C<to_date> function on a value, you can
1274 say something like this:
1278 date_entered => \["to_date(?,'MM/DD/YYYY')", "03/02/2003"],
1281 The first value in the array is the actual SQL. Any other values are
1282 optional and would be included in the bind values array. This gives
1285 my($stmt, @bind) = $sql->insert('people', \%data);
1287 $stmt = "INSERT INTO people (name, date_entered)
1288 VALUES (?, to_date(?,'MM/DD/YYYY'))";
1289 @bind = ('Bill', '03/02/2003');
1291 An UPDATE is just as easy, all you change is the name of the function:
1293 my($stmt, @bind) = $sql->update('people', \%data);
1295 Notice that your C<%data> isn't touched; the module will generate
1296 the appropriately quirky SQL for you automatically. Usually you'll
1297 want to specify a WHERE clause for your UPDATE, though, which is
1298 where handling C<%where> hashes comes in handy...
1300 =head2 Complex where statements
1302 This module can generate pretty complicated WHERE statements
1303 easily. For example, simple C<key=value> pairs are taken to mean
1304 equality, and if you want to see if a field is within a set
1305 of values, you can use an arrayref. Let's say we wanted to
1306 SELECT some data based on this criteria:
1309 requestor => 'inna',
1310 worker => ['nwiger', 'rcwe', 'sfz'],
1311 status => { '!=', 'completed' }
1314 my($stmt, @bind) = $sql->select('tickets', '*', \%where);
1316 The above would give you something like this:
1318 $stmt = "SELECT * FROM tickets WHERE
1319 ( requestor = ? ) AND ( status != ? )
1320 AND ( worker = ? OR worker = ? OR worker = ? )";
1321 @bind = ('inna', 'completed', 'nwiger', 'rcwe', 'sfz');
1323 Which you could then use in DBI code like so:
1325 my $sth = $dbh->prepare($stmt);
1326 $sth->execute(@bind);
1332 The functions are simple. There's one for each major SQL operation,
1333 and a constructor you use first. The arguments are specified in a
1334 similar order to each function (table, then fields, then a where
1335 clause) to try and simplify things.
1340 =head2 new(option => 'value')
1342 The C<new()> function takes a list of options and values, and returns
1343 a new B<SQL::Abstract> object which can then be used to generate SQL
1344 through the methods below. The options accepted are:
1350 If set to 'lower', then SQL will be generated in all lowercase. By
1351 default SQL is generated in "textbook" case meaning something like:
1353 SELECT a_field FROM a_table WHERE some_field LIKE '%someval%'
1355 Any setting other than 'lower' is ignored.
1359 This determines what the default comparison operator is. By default
1360 it is C<=>, meaning that a hash like this:
1362 %where = (name => 'nwiger', email => 'nate@wiger.org');
1364 Will generate SQL like this:
1366 WHERE name = 'nwiger' AND email = 'nate@wiger.org'
1368 However, you may want loose comparisons by default, so if you set
1369 C<cmp> to C<like> you would get SQL such as:
1371 WHERE name like 'nwiger' AND email like 'nate@wiger.org'
1373 You can also override the comparsion on an individual basis - see
1374 the huge section on L</"WHERE CLAUSES"> at the bottom.
1376 =item sqltrue, sqlfalse
1378 Expressions for inserting boolean values within SQL statements.
1379 By default these are C<1=1> and C<1=0>. They are used
1380 by the special operators C<-in> and C<-not_in> for generating
1381 correct SQL even when the argument is an empty array (see below).
1385 This determines the default logical operator for multiple WHERE
1386 statements in arrays or hashes. If absent, the default logic is "or"
1387 for arrays, and "and" for hashes. This means that a WHERE
1391 event_date => {'>=', '2/13/99'},
1392 event_date => {'<=', '4/24/03'},
1395 will generate SQL like this:
1397 WHERE event_date >= '2/13/99' OR event_date <= '4/24/03'
1399 This is probably not what you want given this query, though (look
1400 at the dates). To change the "OR" to an "AND", simply specify:
1402 my $sql = SQL::Abstract->new(logic => 'and');
1404 Which will change the above C<WHERE> to:
1406 WHERE event_date >= '2/13/99' AND event_date <= '4/24/03'
1408 The logic can also be changed locally by inserting
1409 a modifier in front of an arrayref :
1411 @where = (-and => [event_date => {'>=', '2/13/99'},
1412 event_date => {'<=', '4/24/03'} ]);
1414 See the L</"WHERE CLAUSES"> section for explanations.
1418 This will automatically convert comparisons using the specified SQL
1419 function for both column and value. This is mostly used with an argument
1420 of C<upper> or C<lower>, so that the SQL will have the effect of
1421 case-insensitive "searches". For example, this:
1423 $sql = SQL::Abstract->new(convert => 'upper');
1424 %where = (keywords => 'MaKe iT CAse inSeNSItive');
1426 Will turn out the following SQL:
1428 WHERE upper(keywords) like upper('MaKe iT CAse inSeNSItive')
1430 The conversion can be C<upper()>, C<lower()>, or any other SQL function
1431 that can be applied symmetrically to fields (actually B<SQL::Abstract> does
1432 not validate this option; it will just pass through what you specify verbatim).
1436 This is a kludge because many databases suck. For example, you can't
1437 just bind values using DBI's C<execute()> for Oracle C<CLOB> or C<BLOB> fields.
1438 Instead, you have to use C<bind_param()>:
1440 $sth->bind_param(1, 'reg data');
1441 $sth->bind_param(2, $lots, {ora_type => ORA_CLOB});
1443 The problem is, B<SQL::Abstract> will normally just return a C<@bind> array,
1444 which loses track of which field each slot refers to. Fear not.
1446 If you specify C<bindtype> in new, you can determine how C<@bind> is returned.
1447 Currently, you can specify either C<normal> (default) or C<columns>. If you
1448 specify C<columns>, you will get an array that looks like this:
1450 my $sql = SQL::Abstract->new(bindtype => 'columns');
1451 my($stmt, @bind) = $sql->insert(...);
1454 [ 'column1', 'value1' ],
1455 [ 'column2', 'value2' ],
1456 [ 'column3', 'value3' ],
1459 You can then iterate through this manually, using DBI's C<bind_param()>.
1461 $sth->prepare($stmt);
1464 my($col, $data) = @$_;
1465 if ($col eq 'details' || $col eq 'comments') {
1466 $sth->bind_param($i, $data, {ora_type => ORA_CLOB});
1467 } elsif ($col eq 'image') {
1468 $sth->bind_param($i, $data, {ora_type => ORA_BLOB});
1470 $sth->bind_param($i, $data);
1474 $sth->execute; # execute without @bind now
1476 Now, why would you still use B<SQL::Abstract> if you have to do this crap?
1477 Basically, the advantage is still that you don't have to care which fields
1478 are or are not included. You could wrap that above C<for> loop in a simple
1479 sub called C<bind_fields()> or something and reuse it repeatedly. You still
1480 get a layer of abstraction over manual SQL specification.
1482 Note that if you set L</bindtype> to C<columns>, the C<\[$sql, @bind]>
1483 construct (see L</Literal SQL with placeholders and bind values (subqueries)>)
1484 will expect the bind values in this format.
1488 This is the character that a table or column name will be quoted
1489 with. By default this is an empty string, but you could set it to
1490 the character C<`>, to generate SQL like this:
1492 SELECT `a_field` FROM `a_table` WHERE `some_field` LIKE '%someval%'
1494 Alternatively, you can supply an array ref of two items, the first being the left
1495 hand quote character, and the second the right hand quote character. For
1496 example, you could supply C<['[',']']> for SQL Server 2000 compliant quotes
1497 that generates SQL like this:
1499 SELECT [a_field] FROM [a_table] WHERE [some_field] LIKE '%someval%'
1501 Quoting is useful if you have tables or columns names that are reserved
1502 words in your database's SQL dialect.
1506 This is the character that separates a table and column name. It is
1507 necessary to specify this when the C<quote_char> option is selected,
1508 so that tables and column names can be individually quoted like this:
1510 SELECT `table`.`one_field` FROM `table` WHERE `table`.`other_field` = 1
1512 =item injection_guard
1514 A regular expression C<qr/.../> that is applied to any C<-function> and unquoted
1515 column name specified in a query structure. This is a safety mechanism to avoid
1516 injection attacks when mishandling user input e.g.:
1518 my %condition_as_column_value_pairs = get_values_from_user();
1519 $sqla->select( ... , \%condition_as_column_value_pairs );
1521 If the expression matches an exception is thrown. Note that literal SQL
1522 supplied via C<\'...'> or C<\['...']> is B<not> checked in any way.
1524 Defaults to checking for C<;> and the C<GO> keyword (TransactSQL)
1526 =item array_datatypes
1528 When this option is true, arrayrefs in INSERT or UPDATE are
1529 interpreted as array datatypes and are passed directly
1531 When this option is false, arrayrefs are interpreted
1532 as literal SQL, just like refs to arrayrefs
1533 (but this behavior is for backwards compatibility; when writing
1534 new queries, use the "reference to arrayref" syntax
1540 Takes a reference to a list of "special operators"
1541 to extend the syntax understood by L<SQL::Abstract>.
1542 See section L</"SPECIAL OPERATORS"> for details.
1546 Takes a reference to a list of "unary operators"
1547 to extend the syntax understood by L<SQL::Abstract>.
1548 See section L</"UNARY OPERATORS"> for details.
1554 =head2 insert($table, \@values || \%fieldvals, \%options)
1556 This is the simplest function. You simply give it a table name
1557 and either an arrayref of values or hashref of field/value pairs.
1558 It returns an SQL INSERT statement and a list of bind values.
1559 See the sections on L</"Inserting and Updating Arrays"> and
1560 L</"Inserting and Updating SQL"> for information on how to insert
1561 with those data types.
1563 The optional C<\%options> hash reference may contain additional
1564 options to generate the insert SQL. Currently supported options
1571 Takes either a scalar of raw SQL fields, or an array reference of
1572 field names, and adds on an SQL C<RETURNING> statement at the end.
1573 This allows you to return data generated by the insert statement
1574 (such as row IDs) without performing another C<SELECT> statement.
1575 Note, however, this is not part of the SQL standard and may not
1576 be supported by all database engines.
1580 =head2 update($table, \%fieldvals, \%where)
1582 This takes a table, hashref of field/value pairs, and an optional
1583 hashref L<WHERE clause|/WHERE CLAUSES>. It returns an SQL UPDATE function and a list
1585 See the sections on L</"Inserting and Updating Arrays"> and
1586 L</"Inserting and Updating SQL"> for information on how to insert
1587 with those data types.
1589 =head2 select($source, $fields, $where, $order)
1591 This returns a SQL SELECT statement and associated list of bind values, as
1592 specified by the arguments :
1598 Specification of the 'FROM' part of the statement.
1599 The argument can be either a plain scalar (interpreted as a table
1600 name, will be quoted), or an arrayref (interpreted as a list
1601 of table names, joined by commas, quoted), or a scalarref
1602 (literal table name, not quoted), or a ref to an arrayref
1603 (list of literal table names, joined by commas, not quoted).
1607 Specification of the list of fields to retrieve from
1609 The argument can be either an arrayref (interpreted as a list
1610 of field names, will be joined by commas and quoted), or a
1611 plain scalar (literal SQL, not quoted).
1612 Please observe that this API is not as flexible as for
1613 the first argument C<$table>, for backwards compatibility reasons.
1617 Optional argument to specify the WHERE part of the query.
1618 The argument is most often a hashref, but can also be
1619 an arrayref or plain scalar --
1620 see section L<WHERE clause|/"WHERE CLAUSES"> for details.
1624 Optional argument to specify the ORDER BY part of the query.
1625 The argument can be a scalar, a hashref or an arrayref
1626 -- see section L<ORDER BY clause|/"ORDER BY CLAUSES">
1632 =head2 delete($table, \%where)
1634 This takes a table name and optional hashref L<WHERE clause|/WHERE CLAUSES>.
1635 It returns an SQL DELETE statement and list of bind values.
1637 =head2 where(\%where, \@order)
1639 This is used to generate just the WHERE clause. For example,
1640 if you have an arbitrary data structure and know what the
1641 rest of your SQL is going to look like, but want an easy way
1642 to produce a WHERE clause, use this. It returns an SQL WHERE
1643 clause and list of bind values.
1646 =head2 values(\%data)
1648 This just returns the values from the hash C<%data>, in the same
1649 order that would be returned from any of the other above queries.
1650 Using this allows you to markedly speed up your queries if you
1651 are affecting lots of rows. See below under the L</"PERFORMANCE"> section.
1653 =head2 generate($any, 'number', $of, \@data, $struct, \%types)
1655 Warning: This is an experimental method and subject to change.
1657 This returns arbitrarily generated SQL. It's a really basic shortcut.
1658 It will return two different things, depending on return context:
1660 my($stmt, @bind) = $sql->generate('create table', \$table, \@fields);
1661 my $stmt_and_val = $sql->generate('create table', \$table, \@fields);
1663 These would return the following:
1665 # First calling form
1666 $stmt = "CREATE TABLE test (?, ?)";
1667 @bind = (field1, field2);
1669 # Second calling form
1670 $stmt_and_val = "CREATE TABLE test (field1, field2)";
1672 Depending on what you're trying to do, it's up to you to choose the correct
1673 format. In this example, the second form is what you would want.
1677 $sql->generate('alter session', { nls_date_format => 'MM/YY' });
1681 ALTER SESSION SET nls_date_format = 'MM/YY'
1683 You get the idea. Strings get their case twiddled, but everything
1684 else remains verbatim.
1686 =head1 WHERE CLAUSES
1690 This module uses a variation on the idea from L<DBIx::Abstract>. It
1691 is B<NOT>, repeat I<not> 100% compatible. B<The main logic of this
1692 module is that things in arrays are OR'ed, and things in hashes
1695 The easiest way to explain is to show lots of examples. After
1696 each C<%where> hash shown, it is assumed you used:
1698 my($stmt, @bind) = $sql->where(\%where);
1700 However, note that the C<%where> hash can be used directly in any
1701 of the other functions as well, as described above.
1703 =head2 Key-value pairs
1705 So, let's get started. To begin, a simple hash:
1709 status => 'completed'
1712 Is converted to SQL C<key = val> statements:
1714 $stmt = "WHERE user = ? AND status = ?";
1715 @bind = ('nwiger', 'completed');
1717 One common thing I end up doing is having a list of values that
1718 a field can be in. To do this, simply specify a list inside of
1723 status => ['assigned', 'in-progress', 'pending'];
1726 This simple code will create the following:
1728 $stmt = "WHERE user = ? AND ( status = ? OR status = ? OR status = ? )";
1729 @bind = ('nwiger', 'assigned', 'in-progress', 'pending');
1731 A field associated to an empty arrayref will be considered a
1732 logical false and will generate 0=1.
1734 =head2 Tests for NULL values
1736 If the value part is C<undef> then this is converted to SQL <IS NULL>
1745 $stmt = "WHERE user = ? AND status IS NULL";
1748 To test if a column IS NOT NULL:
1752 status => { '!=', undef },
1755 =head2 Specific comparison operators
1757 If you want to specify a different type of operator for your comparison,
1758 you can use a hashref for a given column:
1762 status => { '!=', 'completed' }
1765 Which would generate:
1767 $stmt = "WHERE user = ? AND status != ?";
1768 @bind = ('nwiger', 'completed');
1770 To test against multiple values, just enclose the values in an arrayref:
1772 status => { '=', ['assigned', 'in-progress', 'pending'] };
1774 Which would give you:
1776 "WHERE status = ? OR status = ? OR status = ?"
1779 The hashref can also contain multiple pairs, in which case it is expanded
1780 into an C<AND> of its elements:
1784 status => { '!=', 'completed', -not_like => 'pending%' }
1787 # Or more dynamically, like from a form
1788 $where{user} = 'nwiger';
1789 $where{status}{'!='} = 'completed';
1790 $where{status}{'-not_like'} = 'pending%';
1792 # Both generate this
1793 $stmt = "WHERE user = ? AND status != ? AND status NOT LIKE ?";
1794 @bind = ('nwiger', 'completed', 'pending%');
1797 To get an OR instead, you can combine it with the arrayref idea:
1801 priority => [ { '=', 2 }, { '>', 5 } ]
1804 Which would generate:
1806 $stmt = "WHERE ( priority = ? OR priority > ? ) AND user = ?";
1807 @bind = ('2', '5', 'nwiger');
1809 If you want to include literal SQL (with or without bind values), just use a
1810 scalar reference or array reference as the value:
1813 date_entered => { '>' => \["to_date(?, 'MM/DD/YYYY')", "11/26/2008"] },
1814 date_expires => { '<' => \"now()" }
1817 Which would generate:
1819 $stmt = "WHERE date_entered > "to_date(?, 'MM/DD/YYYY') AND date_expires < now()";
1820 @bind = ('11/26/2008');
1823 =head2 Logic and nesting operators
1825 In the example above,
1826 there is a subtle trap if you want to say something like
1827 this (notice the C<AND>):
1829 WHERE priority != ? AND priority != ?
1831 Because, in Perl you I<can't> do this:
1833 priority => { '!=', 2, '!=', 1 }
1835 As the second C<!=> key will obliterate the first. The solution
1836 is to use the special C<-modifier> form inside an arrayref:
1838 priority => [ -and => {'!=', 2},
1842 Normally, these would be joined by C<OR>, but the modifier tells it
1843 to use C<AND> instead. (Hint: You can use this in conjunction with the
1844 C<logic> option to C<new()> in order to change the way your queries
1845 work by default.) B<Important:> Note that the C<-modifier> goes
1846 B<INSIDE> the arrayref, as an extra first element. This will
1847 B<NOT> do what you think it might:
1849 priority => -and => [{'!=', 2}, {'!=', 1}] # WRONG!
1851 Here is a quick list of equivalencies, since there is some overlap:
1854 status => {'!=', 'completed', 'not like', 'pending%' }
1855 status => [ -and => {'!=', 'completed'}, {'not like', 'pending%'}]
1858 status => {'=', ['assigned', 'in-progress']}
1859 status => [ -or => {'=', 'assigned'}, {'=', 'in-progress'}]
1860 status => [ {'=', 'assigned'}, {'=', 'in-progress'} ]
1864 =head2 Special operators : IN, BETWEEN, etc.
1866 You can also use the hashref format to compare a list of fields using the
1867 C<IN> comparison operator, by specifying the list as an arrayref:
1870 status => 'completed',
1871 reportid => { -in => [567, 2335, 2] }
1874 Which would generate:
1876 $stmt = "WHERE status = ? AND reportid IN (?,?,?)";
1877 @bind = ('completed', '567', '2335', '2');
1879 The reverse operator C<-not_in> generates SQL C<NOT IN> and is used in
1882 If the argument to C<-in> is an empty array, 'sqlfalse' is generated
1883 (by default : C<1=0>). Similarly, C<< -not_in => [] >> generates
1884 'sqltrue' (by default : C<1=1>).
1886 In addition to the array you can supply a chunk of literal sql or
1887 literal sql with bind:
1890 customer => { -in => \[
1891 'SELECT cust_id FROM cust WHERE balance > ?',
1894 status => { -in => \'SELECT status_codes FROM states' },
1900 customer IN ( SELECT cust_id FROM cust WHERE balance > ? )
1901 AND status IN ( SELECT status_codes FROM states )
1907 Another pair of operators is C<-between> and C<-not_between>,
1908 used with an arrayref of two values:
1912 completion_date => {
1913 -not_between => ['2002-10-01', '2003-02-06']
1919 WHERE user = ? AND completion_date NOT BETWEEN ( ? AND ? )
1921 Just like with C<-in> all plausible combinations of literal SQL
1925 start0 => { -between => [ 1, 2 ] },
1926 start1 => { -between => \["? AND ?", 1, 2] },
1927 start2 => { -between => \"lower(x) AND upper(y)" },
1928 start3 => { -between => [
1930 \["upper(?)", 'stuff' ],
1937 ( start0 BETWEEN ? AND ? )
1938 AND ( start1 BETWEEN ? AND ? )
1939 AND ( start2 BETWEEN lower(x) AND upper(y) )
1940 AND ( start3 BETWEEN lower(x) AND upper(?) )
1942 @bind = (1, 2, 1, 2, 'stuff');
1945 These are the two builtin "special operators"; but the
1946 list can be expanded : see section L</"SPECIAL OPERATORS"> below.
1948 =head2 Unary operators: bool
1950 If you wish to test against boolean columns or functions within your
1951 database you can use the C<-bool> and C<-not_bool> operators. For
1952 example to test the column C<is_user> being true and the column
1953 C<is_enabled> being false you would use:-
1957 -not_bool => 'is_enabled',
1962 WHERE is_user AND NOT is_enabled
1964 If a more complex combination is required, testing more conditions,
1965 then you should use the and/or operators:-
1972 -not_bool => 'four',
1978 WHERE one AND two AND three AND NOT four
1981 =head2 Nested conditions, -and/-or prefixes
1983 So far, we've seen how multiple conditions are joined with a top-level
1984 C<AND>. We can change this by putting the different conditions we want in
1985 hashes and then putting those hashes in an array. For example:
1990 status => { -like => ['pending%', 'dispatched'] },
1994 status => 'unassigned',
1998 This data structure would create the following:
2000 $stmt = "WHERE ( user = ? AND ( status LIKE ? OR status LIKE ? ) )
2001 OR ( user = ? AND status = ? ) )";
2002 @bind = ('nwiger', 'pending', 'dispatched', 'robot', 'unassigned');
2005 Clauses in hashrefs or arrayrefs can be prefixed with an C<-and> or C<-or>
2006 to change the logic inside :
2012 -and => [ workhrs => {'>', 20}, geo => 'ASIA' ],
2013 -or => { workhrs => {'<', 50}, geo => 'EURO' },
2020 WHERE ( user = ? AND (
2021 ( workhrs > ? AND geo = ? )
2022 OR ( workhrs < ? OR geo = ? )
2025 =head3 Algebraic inconsistency, for historical reasons
2027 C<Important note>: when connecting several conditions, the C<-and->|C<-or>
2028 operator goes C<outside> of the nested structure; whereas when connecting
2029 several constraints on one column, the C<-and> operator goes
2030 C<inside> the arrayref. Here is an example combining both features :
2033 -and => [a => 1, b => 2],
2034 -or => [c => 3, d => 4],
2035 e => [-and => {-like => 'foo%'}, {-like => '%bar'} ]
2040 WHERE ( ( ( a = ? AND b = ? )
2041 OR ( c = ? OR d = ? )
2042 OR ( e LIKE ? AND e LIKE ? ) ) )
2044 This difference in syntax is unfortunate but must be preserved for
2045 historical reasons. So be careful : the two examples below would
2046 seem algebraically equivalent, but they are not
2048 {col => [-and => {-like => 'foo%'}, {-like => '%bar'}]}
2049 # yields : WHERE ( ( col LIKE ? AND col LIKE ? ) )
2051 [-and => {col => {-like => 'foo%'}, {col => {-like => '%bar'}}]]
2052 # yields : WHERE ( ( col LIKE ? OR col LIKE ? ) )
2055 =head2 Literal SQL and value type operators
2057 The basic premise of SQL::Abstract is that in WHERE specifications the "left
2058 side" is a column name and the "right side" is a value (normally rendered as
2059 a placeholder). This holds true for both hashrefs and arrayref pairs as you
2060 see in the L</WHERE CLAUSES> examples above. Sometimes it is necessary to
2061 alter this behavior. There are several ways of doing so.
2065 This is a virtual operator that signals the string to its right side is an
2066 identifier (a column name) and not a value. For example to compare two
2067 columns you would write:
2070 priority => { '<', 2 },
2071 requestor => { -ident => 'submitter' },
2076 $stmt = "WHERE priority < ? AND requestor = submitter";
2079 If you are maintaining legacy code you may see a different construct as
2080 described in L</Deprecated usage of Literal SQL>, please use C<-ident> in new
2085 This is a virtual operator that signals that the construct to its right side
2086 is a value to be passed to DBI. This is for example necessary when you want
2087 to write a where clause against an array (for RDBMS that support such
2088 datatypes). For example:
2091 array => { -value => [1, 2, 3] }
2096 $stmt = 'WHERE array = ?';
2097 @bind = ([1, 2, 3]);
2099 Note that if you were to simply say:
2105 the result would porbably be not what you wanted:
2107 $stmt = 'WHERE array = ? OR array = ? OR array = ?';
2112 Finally, sometimes only literal SQL will do. To include a random snippet
2113 of SQL verbatim, you specify it as a scalar reference. Consider this only
2114 as a last resort. Usually there is a better way. For example:
2117 priority => { '<', 2 },
2118 requestor => { -in => \'(SELECT name FROM hitmen)' },
2123 $stmt = "WHERE priority < ? AND requestor IN (SELECT name FROM hitmen)"
2126 Note that in this example, you only get one bind parameter back, since
2127 the verbatim SQL is passed as part of the statement.
2131 Never use untrusted input as a literal SQL argument - this is a massive
2132 security risk (there is no way to check literal snippets for SQL
2133 injections and other nastyness). If you need to deal with untrusted input
2134 use literal SQL with placeholders as described next.
2136 =head3 Literal SQL with placeholders and bind values (subqueries)
2138 If the literal SQL to be inserted has placeholders and bind values,
2139 use a reference to an arrayref (yes this is a double reference --
2140 not so common, but perfectly legal Perl). For example, to find a date
2141 in Postgres you can use something like this:
2144 date_column => \[q/= date '2008-09-30' - ?::integer/, 10/]
2149 $stmt = "WHERE ( date_column = date '2008-09-30' - ?::integer )"
2152 Note that you must pass the bind values in the same format as they are returned
2153 by L</where>. That means that if you set L</bindtype> to C<columns>, you must
2154 provide the bind values in the C<< [ column_meta => value ] >> format, where
2155 C<column_meta> is an opaque scalar value; most commonly the column name, but
2156 you can use any scalar value (including references and blessed references),
2157 L<SQL::Abstract> will simply pass it through intact. So if C<bindtype> is set
2158 to C<columns> the above example will look like:
2161 date_column => \[q/= date '2008-09-30' - ?::integer/, [ dummy => 10 ]/]
2164 Literal SQL is especially useful for nesting parenthesized clauses in the
2165 main SQL query. Here is a first example :
2167 my ($sub_stmt, @sub_bind) = ("SELECT c1 FROM t1 WHERE c2 < ? AND c3 LIKE ?",
2171 bar => \["IN ($sub_stmt)" => @sub_bind],
2176 $stmt = "WHERE (foo = ? AND bar IN (SELECT c1 FROM t1
2177 WHERE c2 < ? AND c3 LIKE ?))";
2178 @bind = (1234, 100, "foo%");
2180 Other subquery operators, like for example C<"E<gt> ALL"> or C<"NOT IN">,
2181 are expressed in the same way. Of course the C<$sub_stmt> and
2182 its associated bind values can be generated through a former call
2185 my ($sub_stmt, @sub_bind)
2186 = $sql->select("t1", "c1", {c2 => {"<" => 100},
2187 c3 => {-like => "foo%"}});
2190 bar => \["> ALL ($sub_stmt)" => @sub_bind],
2193 In the examples above, the subquery was used as an operator on a column;
2194 but the same principle also applies for a clause within the main C<%where>
2195 hash, like an EXISTS subquery :
2197 my ($sub_stmt, @sub_bind)
2198 = $sql->select("t1", "*", {c1 => 1, c2 => \"> t0.c0"});
2199 my %where = ( -and => [
2201 \["EXISTS ($sub_stmt)" => @sub_bind],
2206 $stmt = "WHERE (foo = ? AND EXISTS (SELECT * FROM t1
2207 WHERE c1 = ? AND c2 > t0.c0))";
2211 Observe that the condition on C<c2> in the subquery refers to
2212 column C<t0.c0> of the main query : this is I<not> a bind
2213 value, so we have to express it through a scalar ref.
2214 Writing C<< c2 => {">" => "t0.c0"} >> would have generated
2215 C<< c2 > ? >> with bind value C<"t0.c0"> ... not exactly
2216 what we wanted here.
2218 Finally, here is an example where a subquery is used
2219 for expressing unary negation:
2221 my ($sub_stmt, @sub_bind)
2222 = $sql->where({age => [{"<" => 10}, {">" => 20}]});
2223 $sub_stmt =~ s/^ where //i; # don't want "WHERE" in the subclause
2225 lname => {like => '%son%'},
2226 \["NOT ($sub_stmt)" => @sub_bind],
2231 $stmt = "lname LIKE ? AND NOT ( age < ? OR age > ? )"
2232 @bind = ('%son%', 10, 20)
2234 =head3 Deprecated usage of Literal SQL
2236 Below are some examples of archaic use of literal SQL. It is shown only as
2237 reference for those who deal with legacy code. Each example has a much
2238 better, cleaner and safer alternative that users should opt for in new code.
2244 my %where = ( requestor => \'IS NOT NULL' )
2246 $stmt = "WHERE requestor IS NOT NULL"
2248 This used to be the way of generating NULL comparisons, before the handling
2249 of C<undef> got formalized. For new code please use the superior syntax as
2250 described in L</Tests for NULL values>.
2254 my %where = ( requestor => \'= submitter' )
2256 $stmt = "WHERE requestor = submitter"
2258 This used to be the only way to compare columns. Use the superior L</-ident>
2259 method for all new code. For example an identifier declared in such a way
2260 will be properly quoted if L</quote_char> is properly set, while the legacy
2261 form will remain as supplied.
2265 my %where = ( is_ready => \"", completed => { '>', '2012-12-21' } )
2267 $stmt = "WHERE completed > ? AND is_ready"
2268 @bind = ('2012-12-21')
2270 Using an empty string literal used to be the only way to express a boolean.
2271 For all new code please use the much more readable
2272 L<-bool|/Unary operators: bool> operator.
2278 These pages could go on for a while, since the nesting of the data
2279 structures this module can handle are pretty much unlimited (the
2280 module implements the C<WHERE> expansion as a recursive function
2281 internally). Your best bet is to "play around" with the module a
2282 little to see how the data structures behave, and choose the best
2283 format for your data based on that.
2285 And of course, all the values above will probably be replaced with
2286 variables gotten from forms or the command line. After all, if you
2287 knew everything ahead of time, you wouldn't have to worry about
2288 dynamically-generating SQL and could just hardwire it into your
2291 =head1 ORDER BY CLAUSES
2293 Some functions take an order by clause. This can either be a scalar (just a
2294 column name,) a hash of C<< { -desc => 'col' } >> or C<< { -asc => 'col' } >>,
2295 or an array of either of the two previous forms. Examples:
2297 Given | Will Generate
2298 ----------------------------------------------------------
2300 \'colA DESC' | ORDER BY colA DESC
2302 'colA' | ORDER BY colA
2304 [qw/colA colB/] | ORDER BY colA, colB
2306 {-asc => 'colA'} | ORDER BY colA ASC
2308 {-desc => 'colB'} | ORDER BY colB DESC
2310 ['colA', {-asc => 'colB'}] | ORDER BY colA, colB ASC
2312 { -asc => [qw/colA colB/] } | ORDER BY colA ASC, colB ASC
2315 { -asc => 'colA' }, | ORDER BY colA ASC, colB DESC,
2316 { -desc => [qw/colB/], | colC ASC, colD ASC
2317 { -asc => [qw/colC colD/],|
2319 ===========================================================
2323 =head1 SPECIAL OPERATORS
2325 my $sqlmaker = SQL::Abstract->new(special_ops => [
2329 my ($self, $field, $op, $arg) = @_;
2335 handler => 'method_name',
2339 A "special operator" is a SQL syntactic clause that can be
2340 applied to a field, instead of a usual binary operator.
2343 WHERE field IN (?, ?, ?)
2344 WHERE field BETWEEN ? AND ?
2345 WHERE MATCH(field) AGAINST (?, ?)
2347 Special operators IN and BETWEEN are fairly standard and therefore
2348 are builtin within C<SQL::Abstract> (as the overridable methods
2349 C<_where_field_IN> and C<_where_field_BETWEEN>). For other operators,
2350 like the MATCH .. AGAINST example above which is specific to MySQL,
2351 you can write your own operator handlers - supply a C<special_ops>
2352 argument to the C<new> method. That argument takes an arrayref of
2353 operator definitions; each operator definition is a hashref with two
2360 the regular expression to match the operator
2364 Either a coderef or a plain scalar method name. In both cases
2365 the expected return is C<< ($sql, @bind) >>.
2367 When supplied with a method name, it is simply called on the
2368 L<SQL::Abstract/> object as:
2370 $self->$method_name ($field, $op, $arg)
2374 $op is the part that matched the handler regex
2375 $field is the LHS of the operator
2378 When supplied with a coderef, it is called as:
2380 $coderef->($self, $field, $op, $arg)
2385 For example, here is an implementation
2386 of the MATCH .. AGAINST syntax for MySQL
2388 my $sqlmaker = SQL::Abstract->new(special_ops => [
2390 # special op for MySql MATCH (field) AGAINST(word1, word2, ...)
2391 {regex => qr/^match$/i,
2393 my ($self, $field, $op, $arg) = @_;
2394 $arg = [$arg] if not ref $arg;
2395 my $label = $self->_quote($field);
2396 my ($placeholder) = $self->_convert('?');
2397 my $placeholders = join ", ", (($placeholder) x @$arg);
2398 my $sql = $self->_sqlcase('match') . " ($label) "
2399 . $self->_sqlcase('against') . " ($placeholders) ";
2400 my @bind = $self->_bindtype($field, @$arg);
2401 return ($sql, @bind);
2408 =head1 UNARY OPERATORS
2410 my $sqlmaker = SQL::Abstract->new(unary_ops => [
2414 my ($self, $op, $arg) = @_;
2420 handler => 'method_name',
2424 A "unary operator" is a SQL syntactic clause that can be
2425 applied to a field - the operator goes before the field
2427 You can write your own operator handlers - supply a C<unary_ops>
2428 argument to the C<new> method. That argument takes an arrayref of
2429 operator definitions; each operator definition is a hashref with two
2436 the regular expression to match the operator
2440 Either a coderef or a plain scalar method name. In both cases
2441 the expected return is C<< $sql >>.
2443 When supplied with a method name, it is simply called on the
2444 L<SQL::Abstract/> object as:
2446 $self->$method_name ($op, $arg)
2450 $op is the part that matched the handler regex
2451 $arg is the RHS or argument of the operator
2453 When supplied with a coderef, it is called as:
2455 $coderef->($self, $op, $arg)
2463 Thanks to some benchmarking by Mark Stosberg, it turns out that
2464 this module is many orders of magnitude faster than using C<DBIx::Abstract>.
2465 I must admit this wasn't an intentional design issue, but it's a
2466 byproduct of the fact that you get to control your C<DBI> handles
2469 To maximize performance, use a code snippet like the following:
2471 # prepare a statement handle using the first row
2472 # and then reuse it for the rest of the rows
2474 for my $href (@array_of_hashrefs) {
2475 $stmt ||= $sql->insert('table', $href);
2476 $sth ||= $dbh->prepare($stmt);
2477 $sth->execute($sql->values($href));
2480 The reason this works is because the keys in your C<$href> are sorted
2481 internally by B<SQL::Abstract>. Thus, as long as your data retains
2482 the same structure, you only have to generate the SQL the first time
2483 around. On subsequent queries, simply use the C<values> function provided
2484 by this module to return your values in the correct order.
2486 However this depends on the values having the same type - if, for
2487 example, the values of a where clause may either have values
2488 (resulting in sql of the form C<column = ?> with a single bind
2489 value), or alternatively the values might be C<undef> (resulting in
2490 sql of the form C<column IS NULL> with no bind value) then the
2491 caching technique suggested will not work.
2495 If you use my C<CGI::FormBuilder> module at all, you'll hopefully
2496 really like this part (I do, at least). Building up a complex query
2497 can be as simple as the following:
2501 use CGI::FormBuilder;
2504 my $form = CGI::FormBuilder->new(...);
2505 my $sql = SQL::Abstract->new;
2507 if ($form->submitted) {
2508 my $field = $form->field;
2509 my $id = delete $field->{id};
2510 my($stmt, @bind) = $sql->update('table', $field, {id => $id});
2513 Of course, you would still have to connect using C<DBI> to run the
2514 query, but the point is that if you make your form look like your
2515 table, the actual query script can be extremely simplistic.
2517 If you're B<REALLY> lazy (I am), check out C<HTML::QuickTable> for
2518 a fast interface to returning and formatting data. I frequently
2519 use these three modules together to write complex database query
2520 apps in under 50 lines.
2526 =item * gitweb: L<http://git.shadowcat.co.uk/gitweb/gitweb.cgi?p=dbsrgits/SQL-Abstract.git>
2528 =item * git: L<git://git.shadowcat.co.uk/dbsrgits/SQL-Abstract.git>
2534 Version 1.50 was a major internal refactoring of C<SQL::Abstract>.
2535 Great care has been taken to preserve the I<published> behavior
2536 documented in previous versions in the 1.* family; however,
2537 some features that were previously undocumented, or behaved
2538 differently from the documentation, had to be changed in order
2539 to clarify the semantics. Hence, client code that was relying
2540 on some dark areas of C<SQL::Abstract> v1.*
2541 B<might behave differently> in v1.50.
2543 The main changes are :
2549 support for literal SQL through the C<< \ [$sql, bind] >> syntax.
2553 support for the { operator => \"..." } construct (to embed literal SQL)
2557 support for the { operator => \["...", @bind] } construct (to embed literal SQL with bind values)
2561 optional support for L<array datatypes|/"Inserting and Updating Arrays">
2565 defensive programming : check arguments
2569 fixed bug with global logic, which was previously implemented
2570 through global variables yielding side-effects. Prior versions would
2571 interpret C<< [ {cond1, cond2}, [cond3, cond4] ] >>
2572 as C<< "(cond1 AND cond2) OR (cond3 AND cond4)" >>.
2573 Now this is interpreted
2574 as C<< "(cond1 AND cond2) OR (cond3 OR cond4)" >>.
2579 fixed semantics of _bindtype on array args
2583 dropped the C<_anoncopy> of the %where tree. No longer necessary,
2584 we just avoid shifting arrays within that tree.
2588 dropped the C<_modlogic> function
2592 =head1 ACKNOWLEDGEMENTS
2594 There are a number of individuals that have really helped out with
2595 this module. Unfortunately, most of them submitted bugs via CPAN
2596 so I have no idea who they are! But the people I do know are:
2598 Ash Berlin (order_by hash term support)
2599 Matt Trout (DBIx::Class support)
2600 Mark Stosberg (benchmarking)
2601 Chas Owens (initial "IN" operator support)
2602 Philip Collins (per-field SQL functions)
2603 Eric Kolve (hashref "AND" support)
2604 Mike Fragassi (enhancements to "BETWEEN" and "LIKE")
2605 Dan Kubb (support for "quote_char" and "name_sep")
2606 Guillermo Roditi (patch to cleanup "IN" and "BETWEEN", fix and tests for _order_by)
2607 Laurent Dami (internal refactoring, extensible list of special operators, literal SQL)
2608 Norbert Buchmuller (support for literal SQL in hashpair, misc. fixes & tests)
2609 Peter Rabbitson (rewrite of SQLA::Test, misc. fixes & tests)
2610 Oliver Charles (support for "RETURNING" after "INSERT")
2616 L<DBIx::Class>, L<DBIx::Abstract>, L<CGI::FormBuilder>, L<HTML::QuickTable>.
2620 Copyright (c) 2001-2007 Nathan Wiger <nwiger@cpan.org>. All Rights Reserved.
2622 This module is actively maintained by Matt Trout <mst@shadowcatsystems.co.uk>
2624 For support, your best bet is to try the C<DBIx::Class> users mailing list.
2625 While not an official support venue, C<DBIx::Class> makes heavy use of
2626 C<SQL::Abstract>, and as such list members there are very familiar with
2627 how to create queries.
2631 This module is free software; you may copy this under the same
2632 terms as perl itself (either the GNU General Public License or
2633 the Artistic License)