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
15 DQ_WHERE DQ_DELETE DQ_UPDATE DQ_INSERT
17 use Data::Query::ExprHelpers qw(perl_scalar_value);
19 #======================================================================
21 #======================================================================
23 our $VERSION = '1.72';
25 # This would confuse some packagers
26 $VERSION = eval $VERSION if $VERSION =~ /_/; # numify for warning-free dev releases
30 # special operators (-in, -between). May be extended/overridden by user.
31 # See section WHERE: BUILTIN SPECIAL OPERATORS below for implementation
32 my @BUILTIN_SPECIAL_OPS = ();
34 # unaryish operators - key maps to handler
35 my @BUILTIN_UNARY_OPS = ();
37 #======================================================================
38 # DEBUGGING AND ERROR REPORTING
39 #======================================================================
42 return unless $_[0]->{debug}; shift; # a little faster
43 my $func = (caller(1))[3];
44 warn "[$func] ", @_, "\n";
48 my($func) = (caller(1))[3];
49 Carp::carp "[$func] Warning: ", @_;
53 my($func) = (caller(1))[3];
54 Carp::croak "[$func] Fatal: ", @_;
58 #======================================================================
60 #======================================================================
64 my $class = ref($self) || $self;
65 my %opt = (ref $_[0] eq 'HASH') ? %{$_[0]} : @_;
67 # choose our case by keeping an option around
68 delete $opt{case} if $opt{case} && $opt{case} ne 'lower';
70 # default logic for interpreting arrayrefs
71 $opt{logic} = $opt{logic} ? uc $opt{logic} : 'OR';
73 # how to return bind vars
74 # LDNOTE: changed nwiger code : why this 'delete' ??
75 # $opt{bindtype} ||= delete($opt{bind_type}) || 'normal';
76 $opt{bindtype} ||= 'normal';
78 # default comparison is "=", but can be overridden
81 # try to recognize which are the 'equality' and 'unequality' ops
82 # (temporary quickfix, should go through a more seasoned API)
83 $opt{equality_op} = qr/^(\Q$opt{cmp}\E|is|(is\s+)?like)$/i;
84 $opt{inequality_op} = qr/^(!=|<>|(is\s+)?not(\s+like)?)$/i;
87 $opt{sqltrue} ||= '1=1';
88 $opt{sqlfalse} ||= '0=1';
91 $opt{special_ops} ||= [];
92 # regexes are applied in order, thus push after user-defines
93 push @{$opt{special_ops}}, @BUILTIN_SPECIAL_OPS;
96 $opt{unary_ops} ||= [];
97 push @{$opt{unary_ops}}, @BUILTIN_UNARY_OPS;
99 # rudimentary saniy-check for user supplied bits treated as functions/operators
100 # If a purported function matches this regular expression, an exception is thrown.
101 # Literal SQL is *NOT* subject to this check, only functions (and column names
102 # when quoting is not in effect)
105 # need to guard against ()'s in column names too, but this will break tons of
106 # hacks... ideas anyone?
107 $opt{injection_guard} ||= qr/
113 $opt{renderer} ||= do {
114 require Data::Query::Renderer::SQL::Naive;
115 my ($always, $chars);
116 for ($opt{quote_char}) {
117 $chars = defined() ? (ref() ? $_ : [$_]) : ['',''];
119 Data::Query::Renderer::SQL::Naive->new({
120 quote_chars => $chars, always_quote => 1,
121 ($opt{name_sep} ? (identifier_sep => $opt{name_sep}) : ()),
122 ($opt{case} ? (lc_keywords => 1) : ()), # always 'lower' if it exists
126 $opt{name_sep} ||= '.';
128 return bless \%opt, $class;
132 my ($self, $dq) = @_;
136 my ($sql, @bind) = @{$self->{renderer}->render($dq)};
138 ($self->{bindtype} eq 'normal'
139 ? ($sql, map $_->{value}, @bind)
140 : ($sql, map [ $_->{value_meta}, $_->{value} ], @bind)
146 my ($self, $type, @args) = @_;
147 $self->_render_dq($self->${\"_${type}_to_dq"}(@args));
151 my ($self, $literal) = @_;
153 ($literal, @bind) = @$literal if ref($literal) eq 'ARRAY';
158 (@bind ? (values => [ $self->_bind_to_dq(@bind) ]) : ()),
163 my ($self, @bind) = @_;
165 $self->{bindtype} eq 'normal'
166 ? map perl_scalar_value($_), @bind
168 $self->_assert_bindval_matches_bindtype(@bind);
169 map perl_scalar_value(reverse @$_), @bind
174 my ($self, $value) = @_;
175 $self->_maybe_convert_dq(perl_scalar_value($value, our $Cur_Col_Meta));
179 my ($self, $ident) = @_;
180 $self->_assert_pass_injection_guard($ident)
181 unless $self->{renderer}{quote_chars}[0] && $self->{renderer}{always_quote};
182 $self->_maybe_convert_dq({
183 type => DQ_IDENTIFIER,
184 elements => [ split /\Q${\$self->{renderer}->identifier_sep}/, $ident ],
188 sub _maybe_convert_dq {
189 my ($self, $dq) = @_;
190 if (my $c = $self->{where_convert}) {
193 operator => { 'SQL.Naive' => 'apply' },
195 { type => DQ_IDENTIFIER, elements => [ $self->_sqlcase($c) ] },
205 my ($self, $op, @args) = @_;
206 $self->_assert_pass_injection_guard($op);
209 operator => { 'SQL.Naive' => $op },
214 sub _assert_pass_injection_guard {
215 if ($_[1] =~ $_[0]->{injection_guard}) {
216 my $class = ref $_[0];
217 puke "Possible SQL injection attempt '$_[1]'. If this is indeed a part of the "
218 . "desired SQL use literal SQL ( \'...' or \[ '...' ] ) or supply your own "
219 . "{injection_guard} attribute to ${class}->new()"
224 #======================================================================
226 #======================================================================
228 sub insert { shift->_render_sqla(insert => @_) }
231 my ($self, $table, $data, $options) = @_;
232 my (@names, @values);
233 if (ref($data) eq 'HASH') {
234 @names = sort keys %$data;
235 foreach my $k (@names) {
236 local our $Cur_Col_Meta = $k;
237 push @values, $self->_mutation_rhs_to_dq($data->{$k});
239 } elsif (ref($data) eq 'ARRAY') {
240 local our $Cur_Col_Meta;
241 @values = map $self->_mutation_rhs_to_dq($_), @$data;
243 die "Not handled yet";
246 if (my $r_source = $options->{returning}) {
248 map +(ref($_) ? $self->_expr_to_dq($_) : $self->_ident_to_dq($_)),
249 (ref($r_source) eq 'ARRAY' ? @$r_source : $r_source),
254 target => $self->_table_to_dq($table),
255 (@names ? (names => [ map $self->_ident_to_dq($_), @names ]) : ()),
256 values => [ \@values ],
257 ($returning ? (returning => $returning) : ()),
261 sub _mutation_rhs_to_dq {
263 if (ref($v) eq 'ARRAY') {
264 if ($self->{array_datatypes}) {
265 return $self->_value_to_dq($v);
267 $v = \do { my $x = $v };
269 if (ref($v) eq 'HASH') {
270 my ($op, $arg, @rest) = %$v;
272 puke 'Operator calls in update/insert must be in the form { -op => $arg }'
273 if (@rest or not $op =~ /^\-(.+)/);
275 return $self->_expr_to_dq($v);
278 #======================================================================
280 #======================================================================
283 sub update { shift->_render_sqla(update => @_) }
286 my ($self, $table, $data, $where) = @_;
288 puke "Unsupported data type specified to \$sql->update"
289 unless ref $data eq 'HASH';
293 foreach my $k (sort keys %$data) {
295 local our $Cur_Col_Meta = $k;
296 push @set, [ $self->_ident_to_dq($k), $self->_mutation_rhs_to_dq($v) ];
301 target => $self->_table_to_dq($table),
303 where => $self->_where_to_dq($where),
308 #======================================================================
310 #======================================================================
313 my ($self, $table, $where) = @_;
315 my $source_dq = $self->_table_to_dq($table);
317 if (my $where_dq = $self->_where_to_dq($where)) {
328 sub select { shift->_render_sqla(select => @_) }
331 my ($self, $table, $fields, $where, $order) = @_;
334 my $source_dq = $self->_source_to_dq($table, $where);
339 map $self->_ident_to_dq($_),
340 ref($fields) eq 'ARRAY' ? @$fields : $fields
346 $final_dq = $self->_order_by_to_dq($order, undef, $final_dq);
352 #======================================================================
354 #======================================================================
357 sub delete { shift->_render_sqla(delete => @_) }
360 my ($self, $table, $where) = @_;
363 target => $self->_table_to_dq($table),
364 where => $self->_where_to_dq($where),
369 #======================================================================
371 #======================================================================
375 # Finally, a separate routine just to handle WHERE clauses
377 my ($self, $where, $order) = @_;
383 ($sql, @bind) = $self->_recurse_where($where) if defined($where);
384 $sql = $sql ? $self->_sqlcase(' where ') . "( $sql )" : '';
388 $sql .= $self->_order_by($order);
391 return wantarray ? ($sql, @bind) : $sql;
394 sub _recurse_where { shift->_render_sqla(where => @_) }
397 my ($self, $where, $logic) = @_;
399 return undef unless defined($where);
401 # turn the convert misfeature on - only used in WHERE clauses
402 local $self->{where_convert} = $self->{convert};
404 return $self->_expr_to_dq($where, $logic);
408 my ($self, $where, $logic) = @_;
410 if (ref($where) eq 'ARRAY') {
411 return $self->_expr_to_dq_ARRAYREF($where, $logic);
412 } elsif (ref($where) eq 'HASH') {
413 return $self->_expr_to_dq_HASHREF($where, $logic);
415 ref($where) eq 'SCALAR'
416 or (ref($where) eq 'REF' and ref($$where) eq 'ARRAY')
418 return $self->_literal_to_dq($$where);
419 } elsif (!ref($where) or Scalar::Util::blessed($where)) {
420 return $self->_value_to_dq($where);
422 die "Can't handle $where";
425 sub _expr_to_dq_ARRAYREF {
426 my ($self, $where, $logic) = @_;
428 $logic = uc($logic || $self->{logic} || 'OR');
429 $logic eq 'AND' or $logic eq 'OR' or puke "unknown logic: $logic";
431 return unless @$where;
433 my ($first, @rest) = @$where;
435 return $self->_expr_to_dq($first) unless @rest;
439 $self->_where_hashpair_to_dq($first => shift(@rest));
441 $self->_expr_to_dq($first);
445 return $self->_expr_to_dq_ARRAYREF(\@rest, $logic) unless $first_dq;
448 $logic, $first_dq, $self->_expr_to_dq_ARRAYREF(\@rest, $logic)
452 sub _expr_to_dq_HASHREF {
453 my ($self, $where, $logic) = @_;
455 $logic = uc($logic) if $logic;
458 $self->_where_hashpair_to_dq($_ => $where->{$_}, $logic)
461 return $dq[0] unless @dq > 1;
463 my $final = pop(@dq);
465 foreach my $dq (reverse @dq) {
466 $final = $self->_op_to_dq($logic||'AND', $dq, $final);
472 sub _where_to_dq_SCALAR {
473 shift->_value_to_dq(@_);
477 my ($self, $op, $v) = @_;
478 my @args = map $self->_expr_to_dq($_), (ref($v) eq 'ARRAY' ? @$v : $v);
480 # Ok. Welcome to stupid compat code land. An SQLA expr that would in the
481 # absence of this piece of crazy render to:
487 # { -a => { -b => { -c => $x } } }
489 # actually needs to render to:
493 # because SQL sucks, and databases are hateful, and SQLA is Just That DWIM.
495 # However, we don't want to catch 'A(x)' and turn it into 'A x'
497 # So the way we deal with this is to go through all our arguments, and
498 # then if the argument is -also- an apply, i.e. at least 'B', we check
499 # its arguments - and if there's only one of them, and that isn't an apply,
500 # then we convert to the bareword form. The end result should be:
503 # A( B( x ) ) -> A( B x )
504 # A( B( C( x ) ) ) -> A( B( C x ) )
505 # A( B( x + y ) ) -> A( B( x + y ) )
506 # A( B( x, y ) ) -> A( B( x, y ) )
508 # If this turns out not to be quite right, please add additional tests
509 # to either 01generate.t or 02where.t *and* update this comment.
511 foreach my $arg (@args) {
513 $arg->{type} eq DQ_OPERATOR and $arg->{operator}{'SQL.Naive'} eq 'apply'
514 and @{$arg->{args}} == 2 and $arg->{args}[1]{type} ne DQ_OPERATOR
516 $arg->{operator}{'SQL.Naive'} = (shift @{$arg->{args}})->{elements}->[0];
519 $self->_assert_pass_injection_guard($op);
520 return $self->_op_to_dq(
521 apply => $self->_ident_to_dq($op), @args
525 sub _where_hashpair_to_dq {
526 my ($self, $k, $v, $logic) = @_;
528 if ($k =~ /^-(.*)/s) {
530 if ($op eq 'AND' or $op eq 'OR') {
531 return $self->_expr_to_dq($v, $op);
532 } elsif ($op eq 'NEST') {
533 return $self->_expr_to_dq($v);
534 } elsif ($op eq 'NOT') {
535 return $self->_op_to_dq(NOT => $self->_expr_to_dq($v));
536 } elsif ($op eq 'BOOL') {
537 return ref($v) ? $self->_expr_to_dq($v) : $self->_ident_to_dq($v);
538 } elsif ($op eq 'NOT_BOOL') {
539 return $self->_op_to_dq(
540 NOT => ref($v) ? $self->_expr_to_dq($v) : $self->_ident_to_dq($v)
542 } elsif ($op eq 'IDENT') {
543 return $self->_ident_to_dq($v);
544 } elsif ($op eq 'VALUE') {
545 return $self->_value_to_dq($v);
546 } elsif ($op =~ /^(?:AND|OR|NEST)_?\d+/) {
547 die "Use of [and|or|nest]_N modifiers is no longer supported";
549 return $self->_apply_to_dq($op, $v);
552 local our $Cur_Col_Meta = $k;
553 if (ref($v) eq 'ARRAY') {
555 return $self->_literal_to_dq($self->{sqlfalse});
556 } elsif (defined($v->[0]) && $v->[0] =~ /-(and|or)/i) {
557 return $self->_expr_to_dq_ARRAYREF([
558 map +{ $k => $_ }, @{$v}[1..$#$v]
561 return $self->_expr_to_dq_ARRAYREF([
562 map +{ $k => $_ }, @$v
564 } elsif (ref($v) eq 'SCALAR' or (ref($v) eq 'REF' and ref($$v) eq 'ARRAY')) {
568 parts => [ $self->_ident_to_dq($k), $self->_literal_to_dq($$v) ]
571 my ($op, $rhs) = do {
572 if (ref($v) eq 'HASH') {
574 return $self->_expr_to_dq_ARRAYREF([
575 map +{ $k => { $_ => $v->{$_} } }, sort keys %$v
578 my ($op, $value) = %$v;
579 s/^-//, s/_/ /g for $op;
580 if ($op =~ /^(and|or)$/i) {
581 return $self->_expr_to_dq({ $k => $value }, $op);
583 my $special_op = List::Util::first {$op =~ $_->{regex}}
584 @{$self->{special_ops}}
586 return $self->_literal_to_dq(
587 [ $self->${\$special_op->{handler}}($k, $op, $value) ]
589 } elsif ($op =~ /^(?:AND|OR|NEST)_?\d+$/i) {
590 die "Use of [and|or|nest]_N modifiers is no longer supported";
597 if ($op eq 'BETWEEN' or $op eq 'IN' or $op eq 'NOT IN' or $op eq 'NOT BETWEEN') {
598 if (ref($rhs) ne 'ARRAY') {
600 # have to add parens if none present because -in => \"SELECT ..."
601 # got documented. mst hates everything.
602 if (ref($rhs) eq 'SCALAR') {
604 1 while ($x =~ s/\A\s*\((.*)\)\s*\Z/$1/s);
607 my ($x, @rest) = @{$$rhs};
608 1 while ($x =~ s/\A\s*\((.*)\)\s*\Z/$1/s);
609 $rhs = \[ $x, @rest ];
612 return $self->_op_to_dq(
613 $op, $self->_ident_to_dq($k), $self->_literal_to_dq($$rhs)
616 return $self->_literal_to_dq($self->{sqlfalse}) unless @$rhs;
617 return $self->_op_to_dq(
618 $op, $self->_ident_to_dq($k), map $self->_expr_to_dq($_), @$rhs
620 } elsif ($op =~ s/^NOT (?!LIKE)//) {
621 return $self->_where_hashpair_to_dq(-not => { $k => { $op => $rhs } });
622 } elsif ($op eq 'IDENT') {
623 return $self->_op_to_dq(
624 $self->{cmp}, $self->_ident_to_dq($k), $self->_ident_to_dq($rhs)
626 } elsif ($op eq 'VALUE') {
627 return $self->_op_to_dq(
628 $self->{cmp}, $self->_ident_to_dq($k), $self->_value_to_dq($rhs)
630 } elsif (!defined($rhs)) {
632 if ($op eq '=' or $op eq 'LIKE') {
634 } elsif ($op eq '!=') {
637 die "Can't do undef -> NULL transform for operator ${op}";
640 return $self->_op_to_dq($null_op, $self->_ident_to_dq($k));
642 if (ref($rhs) eq 'ARRAY') {
644 return $self->_literal_to_dq(
645 $op eq '!=' ? $self->{sqltrue} : $self->{sqlfalse}
647 } elsif (defined($rhs->[0]) and $rhs->[0] =~ /^-(and|or)$/i) {
648 return $self->_expr_to_dq_ARRAYREF([
649 map +{ $k => { $op => $_ } }, @{$rhs}[1..$#$rhs]
651 } elsif ($op =~ /^-(?:AND|OR|NEST)_?\d+/) {
652 die "Use of [and|or|nest]_N modifiers is no longer supported";
654 return $self->_expr_to_dq_ARRAYREF([
655 map +{ $k => { $op => $_ } }, @$rhs
658 return $self->_op_to_dq(
659 $op, $self->_ident_to_dq($k), $self->_expr_to_dq($rhs)
664 #======================================================================
666 #======================================================================
669 my ($self, $arg) = @_;
670 if (my $dq = $self->_order_by_to_dq($arg)) {
671 # SQLA generates ' ORDER BY foo'. The hilarity.
673 ? do { my @r = $self->_render_dq($dq); $r[0] = ' '.$r[0]; @r }
674 : ' '.$self->_render_dq($dq);
680 sub _order_by_to_dq {
681 my ($self, $arg, $dir, $from) = @_;
687 ($dir ? (direction => $dir) : ()),
688 ($from ? (from => $from) : ()),
692 $dq->{by} = $self->_ident_to_dq($arg);
693 } elsif (ref($arg) eq 'ARRAY') {
695 local our $Order_Inner unless our $Order_Recursing;
696 local $Order_Recursing = 1;
698 foreach my $member (@$arg) {
700 my $next = $self->_order_by_to_dq($member, $dir, $from);
702 $inner->{from} = $next if $inner;
703 $inner = $Order_Inner || $next;
705 $Order_Inner = $inner;
707 } elsif (ref($arg) eq 'REF' and ref($$arg) eq 'ARRAY') {
708 $dq->{by} = $self->_literal_to_dq($$arg);
709 } elsif (ref($arg) eq 'SCALAR') {
710 $dq->{by} = $self->_literal_to_dq($$arg);
711 } elsif (ref($arg) eq 'HASH') {
712 my ($key, $val, @rest) = %$arg;
716 if (@rest or not $key =~ /^-(desc|asc)/i) {
717 puke "hash passed to _order_by must have exactly one key (-desc or -asc)";
720 return $self->_order_by_to_dq($val, $dir, $from);
722 die "Can't handle $arg in _order_by_to_dq";
727 #======================================================================
728 # DATASOURCE (FOR NOW, JUST PLAIN TABLE OR LIST OF TABLES)
729 #======================================================================
731 sub _table { shift->_render_sqla(table => @_) }
734 my ($self, $from) = @_;
735 if (ref($from) eq 'ARRAY') {
736 die "Empty FROM list" unless my @f = @$from;
737 my $dq = $self->_table_to_dq(shift @f);
738 while (my $x = shift @f) {
741 join => [ $dq, $self->_table_to_dq($x) ]
745 } elsif (ref($from) eq 'SCALAR') {
752 $self->_ident_to_dq($from);
757 #======================================================================
759 #======================================================================
761 # highly optimized, as it's called way too often
763 # my ($self, $label) = @_;
765 return '' unless defined $_[1];
766 return ${$_[1]} if ref($_[1]) eq 'SCALAR';
768 unless ($_[0]->{quote_char}) {
769 $_[0]->_assert_pass_injection_guard($_[1]);
773 my $qref = ref $_[0]->{quote_char};
776 ($l, $r) = ( $_[0]->{quote_char}, $_[0]->{quote_char} );
778 elsif ($qref eq 'ARRAY') {
779 ($l, $r) = @{$_[0]->{quote_char}};
782 puke "Unsupported quote_char format: $_[0]->{quote_char}";
785 # parts containing * are naturally unquoted
786 return join( $_[0]->{name_sep}||'', map
787 { $_ eq '*' ? $_ : $l . $_ . $r }
788 ( $_[0]->{name_sep} ? split (/\Q$_[0]->{name_sep}\E/, $_[1] ) : $_[1] )
793 # Conversion, if applicable
795 #my ($self, $arg) = @_;
797 # LDNOTE : modified the previous implementation below because
798 # it was not consistent : the first "return" is always an array,
799 # the second "return" is context-dependent. Anyway, _convert
800 # seems always used with just a single argument, so make it a
802 # return @_ unless $self->{convert};
803 # my $conv = $self->_sqlcase($self->{convert});
804 # my @ret = map { $conv.'('.$_.')' } @_;
805 # return wantarray ? @ret : $ret[0];
806 if ($_[0]->{convert}) {
807 return $_[0]->_sqlcase($_[0]->{convert}) .'(' . $_[1] . ')';
814 #my ($self, $col, @vals) = @_;
816 #LDNOTE : changed original implementation below because it did not make
817 # sense when bindtype eq 'columns' and @vals > 1.
818 # return $self->{bindtype} eq 'columns' ? [ $col, @vals ] : @vals;
820 # called often - tighten code
821 return $_[0]->{bindtype} eq 'columns'
822 ? map {[$_[1], $_]} @_[2 .. $#_]
827 # Dies if any element of @bind is not in [colname => value] format
828 # if bindtype is 'columns'.
829 sub _assert_bindval_matches_bindtype {
830 # my ($self, @bind) = @_;
832 if ($self->{bindtype} eq 'columns') {
834 if (!defined $_ || ref($_) ne 'ARRAY' || @$_ != 2) {
835 puke "bindtype 'columns' selected, you need to pass: [column_name => bind_value]"
841 # Fix SQL case, if so requested
843 # LDNOTE: if $self->{case} is true, then it contains 'lower', so we
844 # don't touch the argument ... crooked logic, but let's not change it!
845 return $_[0]->{case} ? $_[1] : uc($_[1]);
848 #======================================================================
849 # VALUES, GENERATE, AUTOLOAD
850 #======================================================================
852 # LDNOTE: original code from nwiger, didn't touch code in that section
853 # I feel the AUTOLOAD stuff should not be the default, it should
854 # only be activated on explicit demand by user.
858 my $data = shift || return;
859 puke "Argument to ", __PACKAGE__, "->values must be a \\%hash"
860 unless ref $data eq 'HASH';
863 foreach my $k ( sort keys %$data ) {
865 local our $Cur_Col_Meta = $k;
866 my ($sql, @bind) = $self->_render_sqla(
869 push @all_bind, @bind;
878 my(@sql, @sqlq, @sqlv);
882 if ($ref eq 'HASH') {
883 for my $k (sort keys %$_) {
886 my $label = $self->_quote($k);
888 # literal SQL with bind
889 my ($sql, @bind) = @$v;
890 $self->_assert_bindval_matches_bindtype(@bind);
891 push @sqlq, "$label = $sql";
893 } elsif ($r eq 'SCALAR') {
894 # literal SQL without bind
895 push @sqlq, "$label = $$v";
897 push @sqlq, "$label = ?";
898 push @sqlv, $self->_bindtype($k, $v);
901 push @sql, $self->_sqlcase('set'), join ', ', @sqlq;
902 } elsif ($ref eq 'ARRAY') {
903 # unlike insert(), assume these are ONLY the column names, i.e. for SQL
906 if ($r eq 'ARRAY') { # literal SQL with bind
907 my ($sql, @bind) = @$v;
908 $self->_assert_bindval_matches_bindtype(@bind);
911 } elsif ($r eq 'SCALAR') { # literal SQL without bind
912 # embedded literal SQL
919 push @sql, '(' . join(', ', @sqlq) . ')';
920 } elsif ($ref eq 'SCALAR') {
924 # strings get case twiddled
925 push @sql, $self->_sqlcase($_);
929 my $sql = join ' ', @sql;
931 # this is pretty tricky
932 # if ask for an array, return ($stmt, @bind)
933 # otherwise, s/?/shift @sqlv/ to put it inline
935 return ($sql, @sqlv);
937 1 while $sql =~ s/\?/my $d = shift(@sqlv);
938 ref $d ? $d->[1] : $d/e;
947 # # This allows us to check for a local, then _form, attr
949 # my($name) = $AUTOLOAD =~ /.*::(.+)/;
950 # return $self->generate($name, @_);
961 SQL::Abstract - Generate SQL from Perl data structures
967 my $sql = SQL::Abstract->new;
969 my($stmt, @bind) = $sql->select($table, \@fields, \%where, \@order);
971 my($stmt, @bind) = $sql->insert($table, \%fieldvals || \@values);
973 my($stmt, @bind) = $sql->update($table, \%fieldvals, \%where);
975 my($stmt, @bind) = $sql->delete($table, \%where);
977 # Then, use these in your DBI statements
978 my $sth = $dbh->prepare($stmt);
979 $sth->execute(@bind);
981 # Just generate the WHERE clause
982 my($stmt, @bind) = $sql->where(\%where, \@order);
984 # Return values in the same order, for hashed queries
985 # See PERFORMANCE section for more details
986 my @bind = $sql->values(\%fieldvals);
990 This module was inspired by the excellent L<DBIx::Abstract>.
991 However, in using that module I found that what I really wanted
992 to do was generate SQL, but still retain complete control over my
993 statement handles and use the DBI interface. So, I set out to
994 create an abstract SQL generation module.
996 While based on the concepts used by L<DBIx::Abstract>, there are
997 several important differences, especially when it comes to WHERE
998 clauses. I have modified the concepts used to make the SQL easier
999 to generate from Perl data structures and, IMO, more intuitive.
1000 The underlying idea is for this module to do what you mean, based
1001 on the data structures you provide it. The big advantage is that
1002 you don't have to modify your code every time your data changes,
1003 as this module figures it out.
1005 To begin with, an SQL INSERT is as easy as just specifying a hash
1006 of C<key=value> pairs:
1009 name => 'Jimbo Bobson',
1010 phone => '123-456-7890',
1011 address => '42 Sister Lane',
1012 city => 'St. Louis',
1013 state => 'Louisiana',
1016 The SQL can then be generated with this:
1018 my($stmt, @bind) = $sql->insert('people', \%data);
1020 Which would give you something like this:
1022 $stmt = "INSERT INTO people
1023 (address, city, name, phone, state)
1024 VALUES (?, ?, ?, ?, ?)";
1025 @bind = ('42 Sister Lane', 'St. Louis', 'Jimbo Bobson',
1026 '123-456-7890', 'Louisiana');
1028 These are then used directly in your DBI code:
1030 my $sth = $dbh->prepare($stmt);
1031 $sth->execute(@bind);
1033 =head2 Inserting and Updating Arrays
1035 If your database has array types (like for example Postgres),
1036 activate the special option C<< array_datatypes => 1 >>
1037 when creating the C<SQL::Abstract> object.
1038 Then you may use an arrayref to insert and update database array types:
1040 my $sql = SQL::Abstract->new(array_datatypes => 1);
1042 planets => [qw/Mercury Venus Earth Mars/]
1045 my($stmt, @bind) = $sql->insert('solar_system', \%data);
1049 $stmt = "INSERT INTO solar_system (planets) VALUES (?)"
1051 @bind = (['Mercury', 'Venus', 'Earth', 'Mars']);
1054 =head2 Inserting and Updating SQL
1056 In order to apply SQL functions to elements of your C<%data> you may
1057 specify a reference to an arrayref for the given hash value. For example,
1058 if you need to execute the Oracle C<to_date> function on a value, you can
1059 say something like this:
1063 date_entered => \["to_date(?,'MM/DD/YYYY')", "03/02/2003"],
1066 The first value in the array is the actual SQL. Any other values are
1067 optional and would be included in the bind values array. This gives
1070 my($stmt, @bind) = $sql->insert('people', \%data);
1072 $stmt = "INSERT INTO people (name, date_entered)
1073 VALUES (?, to_date(?,'MM/DD/YYYY'))";
1074 @bind = ('Bill', '03/02/2003');
1076 An UPDATE is just as easy, all you change is the name of the function:
1078 my($stmt, @bind) = $sql->update('people', \%data);
1080 Notice that your C<%data> isn't touched; the module will generate
1081 the appropriately quirky SQL for you automatically. Usually you'll
1082 want to specify a WHERE clause for your UPDATE, though, which is
1083 where handling C<%where> hashes comes in handy...
1085 =head2 Complex where statements
1087 This module can generate pretty complicated WHERE statements
1088 easily. For example, simple C<key=value> pairs are taken to mean
1089 equality, and if you want to see if a field is within a set
1090 of values, you can use an arrayref. Let's say we wanted to
1091 SELECT some data based on this criteria:
1094 requestor => 'inna',
1095 worker => ['nwiger', 'rcwe', 'sfz'],
1096 status => { '!=', 'completed' }
1099 my($stmt, @bind) = $sql->select('tickets', '*', \%where);
1101 The above would give you something like this:
1103 $stmt = "SELECT * FROM tickets WHERE
1104 ( requestor = ? ) AND ( status != ? )
1105 AND ( worker = ? OR worker = ? OR worker = ? )";
1106 @bind = ('inna', 'completed', 'nwiger', 'rcwe', 'sfz');
1108 Which you could then use in DBI code like so:
1110 my $sth = $dbh->prepare($stmt);
1111 $sth->execute(@bind);
1117 The functions are simple. There's one for each major SQL operation,
1118 and a constructor you use first. The arguments are specified in a
1119 similar order to each function (table, then fields, then a where
1120 clause) to try and simplify things.
1125 =head2 new(option => 'value')
1127 The C<new()> function takes a list of options and values, and returns
1128 a new B<SQL::Abstract> object which can then be used to generate SQL
1129 through the methods below. The options accepted are:
1135 If set to 'lower', then SQL will be generated in all lowercase. By
1136 default SQL is generated in "textbook" case meaning something like:
1138 SELECT a_field FROM a_table WHERE some_field LIKE '%someval%'
1140 Any setting other than 'lower' is ignored.
1144 This determines what the default comparison operator is. By default
1145 it is C<=>, meaning that a hash like this:
1147 %where = (name => 'nwiger', email => 'nate@wiger.org');
1149 Will generate SQL like this:
1151 WHERE name = 'nwiger' AND email = 'nate@wiger.org'
1153 However, you may want loose comparisons by default, so if you set
1154 C<cmp> to C<like> you would get SQL such as:
1156 WHERE name like 'nwiger' AND email like 'nate@wiger.org'
1158 You can also override the comparsion on an individual basis - see
1159 the huge section on L</"WHERE CLAUSES"> at the bottom.
1161 =item sqltrue, sqlfalse
1163 Expressions for inserting boolean values within SQL statements.
1164 By default these are C<1=1> and C<1=0>. They are used
1165 by the special operators C<-in> and C<-not_in> for generating
1166 correct SQL even when the argument is an empty array (see below).
1170 This determines the default logical operator for multiple WHERE
1171 statements in arrays or hashes. If absent, the default logic is "or"
1172 for arrays, and "and" for hashes. This means that a WHERE
1176 event_date => {'>=', '2/13/99'},
1177 event_date => {'<=', '4/24/03'},
1180 will generate SQL like this:
1182 WHERE event_date >= '2/13/99' OR event_date <= '4/24/03'
1184 This is probably not what you want given this query, though (look
1185 at the dates). To change the "OR" to an "AND", simply specify:
1187 my $sql = SQL::Abstract->new(logic => 'and');
1189 Which will change the above C<WHERE> to:
1191 WHERE event_date >= '2/13/99' AND event_date <= '4/24/03'
1193 The logic can also be changed locally by inserting
1194 a modifier in front of an arrayref :
1196 @where = (-and => [event_date => {'>=', '2/13/99'},
1197 event_date => {'<=', '4/24/03'} ]);
1199 See the L</"WHERE CLAUSES"> section for explanations.
1203 This will automatically convert comparisons using the specified SQL
1204 function for both column and value. This is mostly used with an argument
1205 of C<upper> or C<lower>, so that the SQL will have the effect of
1206 case-insensitive "searches". For example, this:
1208 $sql = SQL::Abstract->new(convert => 'upper');
1209 %where = (keywords => 'MaKe iT CAse inSeNSItive');
1211 Will turn out the following SQL:
1213 WHERE upper(keywords) like upper('MaKe iT CAse inSeNSItive')
1215 The conversion can be C<upper()>, C<lower()>, or any other SQL function
1216 that can be applied symmetrically to fields (actually B<SQL::Abstract> does
1217 not validate this option; it will just pass through what you specify verbatim).
1221 This is a kludge because many databases suck. For example, you can't
1222 just bind values using DBI's C<execute()> for Oracle C<CLOB> or C<BLOB> fields.
1223 Instead, you have to use C<bind_param()>:
1225 $sth->bind_param(1, 'reg data');
1226 $sth->bind_param(2, $lots, {ora_type => ORA_CLOB});
1228 The problem is, B<SQL::Abstract> will normally just return a C<@bind> array,
1229 which loses track of which field each slot refers to. Fear not.
1231 If you specify C<bindtype> in new, you can determine how C<@bind> is returned.
1232 Currently, you can specify either C<normal> (default) or C<columns>. If you
1233 specify C<columns>, you will get an array that looks like this:
1235 my $sql = SQL::Abstract->new(bindtype => 'columns');
1236 my($stmt, @bind) = $sql->insert(...);
1239 [ 'column1', 'value1' ],
1240 [ 'column2', 'value2' ],
1241 [ 'column3', 'value3' ],
1244 You can then iterate through this manually, using DBI's C<bind_param()>.
1246 $sth->prepare($stmt);
1249 my($col, $data) = @$_;
1250 if ($col eq 'details' || $col eq 'comments') {
1251 $sth->bind_param($i, $data, {ora_type => ORA_CLOB});
1252 } elsif ($col eq 'image') {
1253 $sth->bind_param($i, $data, {ora_type => ORA_BLOB});
1255 $sth->bind_param($i, $data);
1259 $sth->execute; # execute without @bind now
1261 Now, why would you still use B<SQL::Abstract> if you have to do this crap?
1262 Basically, the advantage is still that you don't have to care which fields
1263 are or are not included. You could wrap that above C<for> loop in a simple
1264 sub called C<bind_fields()> or something and reuse it repeatedly. You still
1265 get a layer of abstraction over manual SQL specification.
1267 Note that if you set L</bindtype> to C<columns>, the C<\[$sql, @bind]>
1268 construct (see L</Literal SQL with placeholders and bind values (subqueries)>)
1269 will expect the bind values in this format.
1273 This is the character that a table or column name will be quoted
1274 with. By default this is an empty string, but you could set it to
1275 the character C<`>, to generate SQL like this:
1277 SELECT `a_field` FROM `a_table` WHERE `some_field` LIKE '%someval%'
1279 Alternatively, you can supply an array ref of two items, the first being the left
1280 hand quote character, and the second the right hand quote character. For
1281 example, you could supply C<['[',']']> for SQL Server 2000 compliant quotes
1282 that generates SQL like this:
1284 SELECT [a_field] FROM [a_table] WHERE [some_field] LIKE '%someval%'
1286 Quoting is useful if you have tables or columns names that are reserved
1287 words in your database's SQL dialect.
1291 This is the character that separates a table and column name. It is
1292 necessary to specify this when the C<quote_char> option is selected,
1293 so that tables and column names can be individually quoted like this:
1295 SELECT `table`.`one_field` FROM `table` WHERE `table`.`other_field` = 1
1297 =item injection_guard
1299 A regular expression C<qr/.../> that is applied to any C<-function> and unquoted
1300 column name specified in a query structure. This is a safety mechanism to avoid
1301 injection attacks when mishandling user input e.g.:
1303 my %condition_as_column_value_pairs = get_values_from_user();
1304 $sqla->select( ... , \%condition_as_column_value_pairs );
1306 If the expression matches an exception is thrown. Note that literal SQL
1307 supplied via C<\'...'> or C<\['...']> is B<not> checked in any way.
1309 Defaults to checking for C<;> and the C<GO> keyword (TransactSQL)
1311 =item array_datatypes
1313 When this option is true, arrayrefs in INSERT or UPDATE are
1314 interpreted as array datatypes and are passed directly
1316 When this option is false, arrayrefs are interpreted
1317 as literal SQL, just like refs to arrayrefs
1318 (but this behavior is for backwards compatibility; when writing
1319 new queries, use the "reference to arrayref" syntax
1325 Takes a reference to a list of "special operators"
1326 to extend the syntax understood by L<SQL::Abstract>.
1327 See section L</"SPECIAL OPERATORS"> for details.
1331 Takes a reference to a list of "unary operators"
1332 to extend the syntax understood by L<SQL::Abstract>.
1333 See section L</"UNARY OPERATORS"> for details.
1339 =head2 insert($table, \@values || \%fieldvals, \%options)
1341 This is the simplest function. You simply give it a table name
1342 and either an arrayref of values or hashref of field/value pairs.
1343 It returns an SQL INSERT statement and a list of bind values.
1344 See the sections on L</"Inserting and Updating Arrays"> and
1345 L</"Inserting and Updating SQL"> for information on how to insert
1346 with those data types.
1348 The optional C<\%options> hash reference may contain additional
1349 options to generate the insert SQL. Currently supported options
1356 Takes either a scalar of raw SQL fields, or an array reference of
1357 field names, and adds on an SQL C<RETURNING> statement at the end.
1358 This allows you to return data generated by the insert statement
1359 (such as row IDs) without performing another C<SELECT> statement.
1360 Note, however, this is not part of the SQL standard and may not
1361 be supported by all database engines.
1365 =head2 update($table, \%fieldvals, \%where)
1367 This takes a table, hashref of field/value pairs, and an optional
1368 hashref L<WHERE clause|/WHERE CLAUSES>. It returns an SQL UPDATE function and a list
1370 See the sections on L</"Inserting and Updating Arrays"> and
1371 L</"Inserting and Updating SQL"> for information on how to insert
1372 with those data types.
1374 =head2 select($source, $fields, $where, $order)
1376 This returns a SQL SELECT statement and associated list of bind values, as
1377 specified by the arguments :
1383 Specification of the 'FROM' part of the statement.
1384 The argument can be either a plain scalar (interpreted as a table
1385 name, will be quoted), or an arrayref (interpreted as a list
1386 of table names, joined by commas, quoted), or a scalarref
1387 (literal table name, not quoted), or a ref to an arrayref
1388 (list of literal table names, joined by commas, not quoted).
1392 Specification of the list of fields to retrieve from
1394 The argument can be either an arrayref (interpreted as a list
1395 of field names, will be joined by commas and quoted), or a
1396 plain scalar (literal SQL, not quoted).
1397 Please observe that this API is not as flexible as for
1398 the first argument C<$table>, for backwards compatibility reasons.
1402 Optional argument to specify the WHERE part of the query.
1403 The argument is most often a hashref, but can also be
1404 an arrayref or plain scalar --
1405 see section L<WHERE clause|/"WHERE CLAUSES"> for details.
1409 Optional argument to specify the ORDER BY part of the query.
1410 The argument can be a scalar, a hashref or an arrayref
1411 -- see section L<ORDER BY clause|/"ORDER BY CLAUSES">
1417 =head2 delete($table, \%where)
1419 This takes a table name and optional hashref L<WHERE clause|/WHERE CLAUSES>.
1420 It returns an SQL DELETE statement and list of bind values.
1422 =head2 where(\%where, \@order)
1424 This is used to generate just the WHERE clause. For example,
1425 if you have an arbitrary data structure and know what the
1426 rest of your SQL is going to look like, but want an easy way
1427 to produce a WHERE clause, use this. It returns an SQL WHERE
1428 clause and list of bind values.
1431 =head2 values(\%data)
1433 This just returns the values from the hash C<%data>, in the same
1434 order that would be returned from any of the other above queries.
1435 Using this allows you to markedly speed up your queries if you
1436 are affecting lots of rows. See below under the L</"PERFORMANCE"> section.
1438 =head2 generate($any, 'number', $of, \@data, $struct, \%types)
1440 Warning: This is an experimental method and subject to change.
1442 This returns arbitrarily generated SQL. It's a really basic shortcut.
1443 It will return two different things, depending on return context:
1445 my($stmt, @bind) = $sql->generate('create table', \$table, \@fields);
1446 my $stmt_and_val = $sql->generate('create table', \$table, \@fields);
1448 These would return the following:
1450 # First calling form
1451 $stmt = "CREATE TABLE test (?, ?)";
1452 @bind = (field1, field2);
1454 # Second calling form
1455 $stmt_and_val = "CREATE TABLE test (field1, field2)";
1457 Depending on what you're trying to do, it's up to you to choose the correct
1458 format. In this example, the second form is what you would want.
1462 $sql->generate('alter session', { nls_date_format => 'MM/YY' });
1466 ALTER SESSION SET nls_date_format = 'MM/YY'
1468 You get the idea. Strings get their case twiddled, but everything
1469 else remains verbatim.
1471 =head1 WHERE CLAUSES
1475 This module uses a variation on the idea from L<DBIx::Abstract>. It
1476 is B<NOT>, repeat I<not> 100% compatible. B<The main logic of this
1477 module is that things in arrays are OR'ed, and things in hashes
1480 The easiest way to explain is to show lots of examples. After
1481 each C<%where> hash shown, it is assumed you used:
1483 my($stmt, @bind) = $sql->where(\%where);
1485 However, note that the C<%where> hash can be used directly in any
1486 of the other functions as well, as described above.
1488 =head2 Key-value pairs
1490 So, let's get started. To begin, a simple hash:
1494 status => 'completed'
1497 Is converted to SQL C<key = val> statements:
1499 $stmt = "WHERE user = ? AND status = ?";
1500 @bind = ('nwiger', 'completed');
1502 One common thing I end up doing is having a list of values that
1503 a field can be in. To do this, simply specify a list inside of
1508 status => ['assigned', 'in-progress', 'pending'];
1511 This simple code will create the following:
1513 $stmt = "WHERE user = ? AND ( status = ? OR status = ? OR status = ? )";
1514 @bind = ('nwiger', 'assigned', 'in-progress', 'pending');
1516 A field associated to an empty arrayref will be considered a
1517 logical false and will generate 0=1.
1519 =head2 Tests for NULL values
1521 If the value part is C<undef> then this is converted to SQL <IS NULL>
1530 $stmt = "WHERE user = ? AND status IS NULL";
1533 To test if a column IS NOT NULL:
1537 status => { '!=', undef },
1540 =head2 Specific comparison operators
1542 If you want to specify a different type of operator for your comparison,
1543 you can use a hashref for a given column:
1547 status => { '!=', 'completed' }
1550 Which would generate:
1552 $stmt = "WHERE user = ? AND status != ?";
1553 @bind = ('nwiger', 'completed');
1555 To test against multiple values, just enclose the values in an arrayref:
1557 status => { '=', ['assigned', 'in-progress', 'pending'] };
1559 Which would give you:
1561 "WHERE status = ? OR status = ? OR status = ?"
1564 The hashref can also contain multiple pairs, in which case it is expanded
1565 into an C<AND> of its elements:
1569 status => { '!=', 'completed', -not_like => 'pending%' }
1572 # Or more dynamically, like from a form
1573 $where{user} = 'nwiger';
1574 $where{status}{'!='} = 'completed';
1575 $where{status}{'-not_like'} = 'pending%';
1577 # Both generate this
1578 $stmt = "WHERE user = ? AND status != ? AND status NOT LIKE ?";
1579 @bind = ('nwiger', 'completed', 'pending%');
1582 To get an OR instead, you can combine it with the arrayref idea:
1586 priority => [ { '=', 2 }, { '>', 5 } ]
1589 Which would generate:
1591 $stmt = "WHERE ( priority = ? OR priority > ? ) AND user = ?";
1592 @bind = ('2', '5', 'nwiger');
1594 If you want to include literal SQL (with or without bind values), just use a
1595 scalar reference or array reference as the value:
1598 date_entered => { '>' => \["to_date(?, 'MM/DD/YYYY')", "11/26/2008"] },
1599 date_expires => { '<' => \"now()" }
1602 Which would generate:
1604 $stmt = "WHERE date_entered > "to_date(?, 'MM/DD/YYYY') AND date_expires < now()";
1605 @bind = ('11/26/2008');
1608 =head2 Logic and nesting operators
1610 In the example above,
1611 there is a subtle trap if you want to say something like
1612 this (notice the C<AND>):
1614 WHERE priority != ? AND priority != ?
1616 Because, in Perl you I<can't> do this:
1618 priority => { '!=', 2, '!=', 1 }
1620 As the second C<!=> key will obliterate the first. The solution
1621 is to use the special C<-modifier> form inside an arrayref:
1623 priority => [ -and => {'!=', 2},
1627 Normally, these would be joined by C<OR>, but the modifier tells it
1628 to use C<AND> instead. (Hint: You can use this in conjunction with the
1629 C<logic> option to C<new()> in order to change the way your queries
1630 work by default.) B<Important:> Note that the C<-modifier> goes
1631 B<INSIDE> the arrayref, as an extra first element. This will
1632 B<NOT> do what you think it might:
1634 priority => -and => [{'!=', 2}, {'!=', 1}] # WRONG!
1636 Here is a quick list of equivalencies, since there is some overlap:
1639 status => {'!=', 'completed', 'not like', 'pending%' }
1640 status => [ -and => {'!=', 'completed'}, {'not like', 'pending%'}]
1643 status => {'=', ['assigned', 'in-progress']}
1644 status => [ -or => {'=', 'assigned'}, {'=', 'in-progress'}]
1645 status => [ {'=', 'assigned'}, {'=', 'in-progress'} ]
1649 =head2 Special operators : IN, BETWEEN, etc.
1651 You can also use the hashref format to compare a list of fields using the
1652 C<IN> comparison operator, by specifying the list as an arrayref:
1655 status => 'completed',
1656 reportid => { -in => [567, 2335, 2] }
1659 Which would generate:
1661 $stmt = "WHERE status = ? AND reportid IN (?,?,?)";
1662 @bind = ('completed', '567', '2335', '2');
1664 The reverse operator C<-not_in> generates SQL C<NOT IN> and is used in
1667 If the argument to C<-in> is an empty array, 'sqlfalse' is generated
1668 (by default : C<1=0>). Similarly, C<< -not_in => [] >> generates
1669 'sqltrue' (by default : C<1=1>).
1671 In addition to the array you can supply a chunk of literal sql or
1672 literal sql with bind:
1675 customer => { -in => \[
1676 'SELECT cust_id FROM cust WHERE balance > ?',
1679 status => { -in => \'SELECT status_codes FROM states' },
1685 customer IN ( SELECT cust_id FROM cust WHERE balance > ? )
1686 AND status IN ( SELECT status_codes FROM states )
1692 Another pair of operators is C<-between> and C<-not_between>,
1693 used with an arrayref of two values:
1697 completion_date => {
1698 -not_between => ['2002-10-01', '2003-02-06']
1704 WHERE user = ? AND completion_date NOT BETWEEN ( ? AND ? )
1706 Just like with C<-in> all plausible combinations of literal SQL
1710 start0 => { -between => [ 1, 2 ] },
1711 start1 => { -between => \["? AND ?", 1, 2] },
1712 start2 => { -between => \"lower(x) AND upper(y)" },
1713 start3 => { -between => [
1715 \["upper(?)", 'stuff' ],
1722 ( start0 BETWEEN ? AND ? )
1723 AND ( start1 BETWEEN ? AND ? )
1724 AND ( start2 BETWEEN lower(x) AND upper(y) )
1725 AND ( start3 BETWEEN lower(x) AND upper(?) )
1727 @bind = (1, 2, 1, 2, 'stuff');
1730 These are the two builtin "special operators"; but the
1731 list can be expanded : see section L</"SPECIAL OPERATORS"> below.
1733 =head2 Unary operators: bool
1735 If you wish to test against boolean columns or functions within your
1736 database you can use the C<-bool> and C<-not_bool> operators. For
1737 example to test the column C<is_user> being true and the column
1738 C<is_enabled> being false you would use:-
1742 -not_bool => 'is_enabled',
1747 WHERE is_user AND NOT is_enabled
1749 If a more complex combination is required, testing more conditions,
1750 then you should use the and/or operators:-
1757 -not_bool => 'four',
1763 WHERE one AND two AND three AND NOT four
1766 =head2 Nested conditions, -and/-or prefixes
1768 So far, we've seen how multiple conditions are joined with a top-level
1769 C<AND>. We can change this by putting the different conditions we want in
1770 hashes and then putting those hashes in an array. For example:
1775 status => { -like => ['pending%', 'dispatched'] },
1779 status => 'unassigned',
1783 This data structure would create the following:
1785 $stmt = "WHERE ( user = ? AND ( status LIKE ? OR status LIKE ? ) )
1786 OR ( user = ? AND status = ? ) )";
1787 @bind = ('nwiger', 'pending', 'dispatched', 'robot', 'unassigned');
1790 Clauses in hashrefs or arrayrefs can be prefixed with an C<-and> or C<-or>
1791 to change the logic inside :
1797 -and => [ workhrs => {'>', 20}, geo => 'ASIA' ],
1798 -or => { workhrs => {'<', 50}, geo => 'EURO' },
1805 WHERE ( user = ? AND (
1806 ( workhrs > ? AND geo = ? )
1807 OR ( workhrs < ? OR geo = ? )
1810 =head3 Algebraic inconsistency, for historical reasons
1812 C<Important note>: when connecting several conditions, the C<-and->|C<-or>
1813 operator goes C<outside> of the nested structure; whereas when connecting
1814 several constraints on one column, the C<-and> operator goes
1815 C<inside> the arrayref. Here is an example combining both features :
1818 -and => [a => 1, b => 2],
1819 -or => [c => 3, d => 4],
1820 e => [-and => {-like => 'foo%'}, {-like => '%bar'} ]
1825 WHERE ( ( ( a = ? AND b = ? )
1826 OR ( c = ? OR d = ? )
1827 OR ( e LIKE ? AND e LIKE ? ) ) )
1829 This difference in syntax is unfortunate but must be preserved for
1830 historical reasons. So be careful : the two examples below would
1831 seem algebraically equivalent, but they are not
1833 {col => [-and => {-like => 'foo%'}, {-like => '%bar'}]}
1834 # yields : WHERE ( ( col LIKE ? AND col LIKE ? ) )
1836 [-and => {col => {-like => 'foo%'}, {col => {-like => '%bar'}}]]
1837 # yields : WHERE ( ( col LIKE ? OR col LIKE ? ) )
1840 =head2 Literal SQL and value type operators
1842 The basic premise of SQL::Abstract is that in WHERE specifications the "left
1843 side" is a column name and the "right side" is a value (normally rendered as
1844 a placeholder). This holds true for both hashrefs and arrayref pairs as you
1845 see in the L</WHERE CLAUSES> examples above. Sometimes it is necessary to
1846 alter this behavior. There are several ways of doing so.
1850 This is a virtual operator that signals the string to its right side is an
1851 identifier (a column name) and not a value. For example to compare two
1852 columns you would write:
1855 priority => { '<', 2 },
1856 requestor => { -ident => 'submitter' },
1861 $stmt = "WHERE priority < ? AND requestor = submitter";
1864 If you are maintaining legacy code you may see a different construct as
1865 described in L</Deprecated usage of Literal SQL>, please use C<-ident> in new
1870 This is a virtual operator that signals that the construct to its right side
1871 is a value to be passed to DBI. This is for example necessary when you want
1872 to write a where clause against an array (for RDBMS that support such
1873 datatypes). For example:
1876 array => { -value => [1, 2, 3] }
1881 $stmt = 'WHERE array = ?';
1882 @bind = ([1, 2, 3]);
1884 Note that if you were to simply say:
1890 the result would porbably be not what you wanted:
1892 $stmt = 'WHERE array = ? OR array = ? OR array = ?';
1897 Finally, sometimes only literal SQL will do. To include a random snippet
1898 of SQL verbatim, you specify it as a scalar reference. Consider this only
1899 as a last resort. Usually there is a better way. For example:
1902 priority => { '<', 2 },
1903 requestor => { -in => \'(SELECT name FROM hitmen)' },
1908 $stmt = "WHERE priority < ? AND requestor IN (SELECT name FROM hitmen)"
1911 Note that in this example, you only get one bind parameter back, since
1912 the verbatim SQL is passed as part of the statement.
1916 Never use untrusted input as a literal SQL argument - this is a massive
1917 security risk (there is no way to check literal snippets for SQL
1918 injections and other nastyness). If you need to deal with untrusted input
1919 use literal SQL with placeholders as described next.
1921 =head3 Literal SQL with placeholders and bind values (subqueries)
1923 If the literal SQL to be inserted has placeholders and bind values,
1924 use a reference to an arrayref (yes this is a double reference --
1925 not so common, but perfectly legal Perl). For example, to find a date
1926 in Postgres you can use something like this:
1929 date_column => \[q/= date '2008-09-30' - ?::integer/, 10/]
1934 $stmt = "WHERE ( date_column = date '2008-09-30' - ?::integer )"
1937 Note that you must pass the bind values in the same format as they are returned
1938 by L</where>. That means that if you set L</bindtype> to C<columns>, you must
1939 provide the bind values in the C<< [ column_meta => value ] >> format, where
1940 C<column_meta> is an opaque scalar value; most commonly the column name, but
1941 you can use any scalar value (including references and blessed references),
1942 L<SQL::Abstract> will simply pass it through intact. So if C<bindtype> is set
1943 to C<columns> the above example will look like:
1946 date_column => \[q/= date '2008-09-30' - ?::integer/, [ dummy => 10 ]/]
1949 Literal SQL is especially useful for nesting parenthesized clauses in the
1950 main SQL query. Here is a first example :
1952 my ($sub_stmt, @sub_bind) = ("SELECT c1 FROM t1 WHERE c2 < ? AND c3 LIKE ?",
1956 bar => \["IN ($sub_stmt)" => @sub_bind],
1961 $stmt = "WHERE (foo = ? AND bar IN (SELECT c1 FROM t1
1962 WHERE c2 < ? AND c3 LIKE ?))";
1963 @bind = (1234, 100, "foo%");
1965 Other subquery operators, like for example C<"E<gt> ALL"> or C<"NOT IN">,
1966 are expressed in the same way. Of course the C<$sub_stmt> and
1967 its associated bind values can be generated through a former call
1970 my ($sub_stmt, @sub_bind)
1971 = $sql->select("t1", "c1", {c2 => {"<" => 100},
1972 c3 => {-like => "foo%"}});
1975 bar => \["> ALL ($sub_stmt)" => @sub_bind],
1978 In the examples above, the subquery was used as an operator on a column;
1979 but the same principle also applies for a clause within the main C<%where>
1980 hash, like an EXISTS subquery :
1982 my ($sub_stmt, @sub_bind)
1983 = $sql->select("t1", "*", {c1 => 1, c2 => \"> t0.c0"});
1984 my %where = ( -and => [
1986 \["EXISTS ($sub_stmt)" => @sub_bind],
1991 $stmt = "WHERE (foo = ? AND EXISTS (SELECT * FROM t1
1992 WHERE c1 = ? AND c2 > t0.c0))";
1996 Observe that the condition on C<c2> in the subquery refers to
1997 column C<t0.c0> of the main query : this is I<not> a bind
1998 value, so we have to express it through a scalar ref.
1999 Writing C<< c2 => {">" => "t0.c0"} >> would have generated
2000 C<< c2 > ? >> with bind value C<"t0.c0"> ... not exactly
2001 what we wanted here.
2003 Finally, here is an example where a subquery is used
2004 for expressing unary negation:
2006 my ($sub_stmt, @sub_bind)
2007 = $sql->where({age => [{"<" => 10}, {">" => 20}]});
2008 $sub_stmt =~ s/^ where //i; # don't want "WHERE" in the subclause
2010 lname => {like => '%son%'},
2011 \["NOT ($sub_stmt)" => @sub_bind],
2016 $stmt = "lname LIKE ? AND NOT ( age < ? OR age > ? )"
2017 @bind = ('%son%', 10, 20)
2019 =head3 Deprecated usage of Literal SQL
2021 Below are some examples of archaic use of literal SQL. It is shown only as
2022 reference for those who deal with legacy code. Each example has a much
2023 better, cleaner and safer alternative that users should opt for in new code.
2029 my %where = ( requestor => \'IS NOT NULL' )
2031 $stmt = "WHERE requestor IS NOT NULL"
2033 This used to be the way of generating NULL comparisons, before the handling
2034 of C<undef> got formalized. For new code please use the superior syntax as
2035 described in L</Tests for NULL values>.
2039 my %where = ( requestor => \'= submitter' )
2041 $stmt = "WHERE requestor = submitter"
2043 This used to be the only way to compare columns. Use the superior L</-ident>
2044 method for all new code. For example an identifier declared in such a way
2045 will be properly quoted if L</quote_char> is properly set, while the legacy
2046 form will remain as supplied.
2050 my %where = ( is_ready => \"", completed => { '>', '2012-12-21' } )
2052 $stmt = "WHERE completed > ? AND is_ready"
2053 @bind = ('2012-12-21')
2055 Using an empty string literal used to be the only way to express a boolean.
2056 For all new code please use the much more readable
2057 L<-bool|/Unary operators: bool> operator.
2063 These pages could go on for a while, since the nesting of the data
2064 structures this module can handle are pretty much unlimited (the
2065 module implements the C<WHERE> expansion as a recursive function
2066 internally). Your best bet is to "play around" with the module a
2067 little to see how the data structures behave, and choose the best
2068 format for your data based on that.
2070 And of course, all the values above will probably be replaced with
2071 variables gotten from forms or the command line. After all, if you
2072 knew everything ahead of time, you wouldn't have to worry about
2073 dynamically-generating SQL and could just hardwire it into your
2076 =head1 ORDER BY CLAUSES
2078 Some functions take an order by clause. This can either be a scalar (just a
2079 column name,) a hash of C<< { -desc => 'col' } >> or C<< { -asc => 'col' } >>,
2080 or an array of either of the two previous forms. Examples:
2082 Given | Will Generate
2083 ----------------------------------------------------------
2085 \'colA DESC' | ORDER BY colA DESC
2087 'colA' | ORDER BY colA
2089 [qw/colA colB/] | ORDER BY colA, colB
2091 {-asc => 'colA'} | ORDER BY colA ASC
2093 {-desc => 'colB'} | ORDER BY colB DESC
2095 ['colA', {-asc => 'colB'}] | ORDER BY colA, colB ASC
2097 { -asc => [qw/colA colB/] } | ORDER BY colA ASC, colB ASC
2100 { -asc => 'colA' }, | ORDER BY colA ASC, colB DESC,
2101 { -desc => [qw/colB/], | colC ASC, colD ASC
2102 { -asc => [qw/colC colD/],|
2104 ===========================================================
2108 =head1 SPECIAL OPERATORS
2110 my $sqlmaker = SQL::Abstract->new(special_ops => [
2114 my ($self, $field, $op, $arg) = @_;
2120 handler => 'method_name',
2124 A "special operator" is a SQL syntactic clause that can be
2125 applied to a field, instead of a usual binary operator.
2128 WHERE field IN (?, ?, ?)
2129 WHERE field BETWEEN ? AND ?
2130 WHERE MATCH(field) AGAINST (?, ?)
2132 Special operators IN and BETWEEN are fairly standard and therefore
2133 are builtin within C<SQL::Abstract> (as the overridable methods
2134 C<_where_field_IN> and C<_where_field_BETWEEN>). For other operators,
2135 like the MATCH .. AGAINST example above which is specific to MySQL,
2136 you can write your own operator handlers - supply a C<special_ops>
2137 argument to the C<new> method. That argument takes an arrayref of
2138 operator definitions; each operator definition is a hashref with two
2145 the regular expression to match the operator
2149 Either a coderef or a plain scalar method name. In both cases
2150 the expected return is C<< ($sql, @bind) >>.
2152 When supplied with a method name, it is simply called on the
2153 L<SQL::Abstract/> object as:
2155 $self->$method_name ($field, $op, $arg)
2159 $op is the part that matched the handler regex
2160 $field is the LHS of the operator
2163 When supplied with a coderef, it is called as:
2165 $coderef->($self, $field, $op, $arg)
2170 For example, here is an implementation
2171 of the MATCH .. AGAINST syntax for MySQL
2173 my $sqlmaker = SQL::Abstract->new(special_ops => [
2175 # special op for MySql MATCH (field) AGAINST(word1, word2, ...)
2176 {regex => qr/^match$/i,
2178 my ($self, $field, $op, $arg) = @_;
2179 $arg = [$arg] if not ref $arg;
2180 my $label = $self->_quote($field);
2181 my ($placeholder) = $self->_convert('?');
2182 my $placeholders = join ", ", (($placeholder) x @$arg);
2183 my $sql = $self->_sqlcase('match') . " ($label) "
2184 . $self->_sqlcase('against') . " ($placeholders) ";
2185 my @bind = $self->_bindtype($field, @$arg);
2186 return ($sql, @bind);
2193 =head1 UNARY OPERATORS
2195 my $sqlmaker = SQL::Abstract->new(unary_ops => [
2199 my ($self, $op, $arg) = @_;
2205 handler => 'method_name',
2209 A "unary operator" is a SQL syntactic clause that can be
2210 applied to a field - the operator goes before the field
2212 You can write your own operator handlers - supply a C<unary_ops>
2213 argument to the C<new> method. That argument takes an arrayref of
2214 operator definitions; each operator definition is a hashref with two
2221 the regular expression to match the operator
2225 Either a coderef or a plain scalar method name. In both cases
2226 the expected return is C<< $sql >>.
2228 When supplied with a method name, it is simply called on the
2229 L<SQL::Abstract/> object as:
2231 $self->$method_name ($op, $arg)
2235 $op is the part that matched the handler regex
2236 $arg is the RHS or argument of the operator
2238 When supplied with a coderef, it is called as:
2240 $coderef->($self, $op, $arg)
2248 Thanks to some benchmarking by Mark Stosberg, it turns out that
2249 this module is many orders of magnitude faster than using C<DBIx::Abstract>.
2250 I must admit this wasn't an intentional design issue, but it's a
2251 byproduct of the fact that you get to control your C<DBI> handles
2254 To maximize performance, use a code snippet like the following:
2256 # prepare a statement handle using the first row
2257 # and then reuse it for the rest of the rows
2259 for my $href (@array_of_hashrefs) {
2260 $stmt ||= $sql->insert('table', $href);
2261 $sth ||= $dbh->prepare($stmt);
2262 $sth->execute($sql->values($href));
2265 The reason this works is because the keys in your C<$href> are sorted
2266 internally by B<SQL::Abstract>. Thus, as long as your data retains
2267 the same structure, you only have to generate the SQL the first time
2268 around. On subsequent queries, simply use the C<values> function provided
2269 by this module to return your values in the correct order.
2271 However this depends on the values having the same type - if, for
2272 example, the values of a where clause may either have values
2273 (resulting in sql of the form C<column = ?> with a single bind
2274 value), or alternatively the values might be C<undef> (resulting in
2275 sql of the form C<column IS NULL> with no bind value) then the
2276 caching technique suggested will not work.
2280 If you use my C<CGI::FormBuilder> module at all, you'll hopefully
2281 really like this part (I do, at least). Building up a complex query
2282 can be as simple as the following:
2286 use CGI::FormBuilder;
2289 my $form = CGI::FormBuilder->new(...);
2290 my $sql = SQL::Abstract->new;
2292 if ($form->submitted) {
2293 my $field = $form->field;
2294 my $id = delete $field->{id};
2295 my($stmt, @bind) = $sql->update('table', $field, {id => $id});
2298 Of course, you would still have to connect using C<DBI> to run the
2299 query, but the point is that if you make your form look like your
2300 table, the actual query script can be extremely simplistic.
2302 If you're B<REALLY> lazy (I am), check out C<HTML::QuickTable> for
2303 a fast interface to returning and formatting data. I frequently
2304 use these three modules together to write complex database query
2305 apps in under 50 lines.
2311 =item * gitweb: L<http://git.shadowcat.co.uk/gitweb/gitweb.cgi?p=dbsrgits/SQL-Abstract.git>
2313 =item * git: L<git://git.shadowcat.co.uk/dbsrgits/SQL-Abstract.git>
2319 Version 1.50 was a major internal refactoring of C<SQL::Abstract>.
2320 Great care has been taken to preserve the I<published> behavior
2321 documented in previous versions in the 1.* family; however,
2322 some features that were previously undocumented, or behaved
2323 differently from the documentation, had to be changed in order
2324 to clarify the semantics. Hence, client code that was relying
2325 on some dark areas of C<SQL::Abstract> v1.*
2326 B<might behave differently> in v1.50.
2328 The main changes are :
2334 support for literal SQL through the C<< \ [$sql, bind] >> syntax.
2338 support for the { operator => \"..." } construct (to embed literal SQL)
2342 support for the { operator => \["...", @bind] } construct (to embed literal SQL with bind values)
2346 optional support for L<array datatypes|/"Inserting and Updating Arrays">
2350 defensive programming : check arguments
2354 fixed bug with global logic, which was previously implemented
2355 through global variables yielding side-effects. Prior versions would
2356 interpret C<< [ {cond1, cond2}, [cond3, cond4] ] >>
2357 as C<< "(cond1 AND cond2) OR (cond3 AND cond4)" >>.
2358 Now this is interpreted
2359 as C<< "(cond1 AND cond2) OR (cond3 OR cond4)" >>.
2364 fixed semantics of _bindtype on array args
2368 dropped the C<_anoncopy> of the %where tree. No longer necessary,
2369 we just avoid shifting arrays within that tree.
2373 dropped the C<_modlogic> function
2377 =head1 ACKNOWLEDGEMENTS
2379 There are a number of individuals that have really helped out with
2380 this module. Unfortunately, most of them submitted bugs via CPAN
2381 so I have no idea who they are! But the people I do know are:
2383 Ash Berlin (order_by hash term support)
2384 Matt Trout (DBIx::Class support)
2385 Mark Stosberg (benchmarking)
2386 Chas Owens (initial "IN" operator support)
2387 Philip Collins (per-field SQL functions)
2388 Eric Kolve (hashref "AND" support)
2389 Mike Fragassi (enhancements to "BETWEEN" and "LIKE")
2390 Dan Kubb (support for "quote_char" and "name_sep")
2391 Guillermo Roditi (patch to cleanup "IN" and "BETWEEN", fix and tests for _order_by)
2392 Laurent Dami (internal refactoring, extensible list of special operators, literal SQL)
2393 Norbert Buchmuller (support for literal SQL in hashpair, misc. fixes & tests)
2394 Peter Rabbitson (rewrite of SQLA::Test, misc. fixes & tests)
2395 Oliver Charles (support for "RETURNING" after "INSERT")
2401 L<DBIx::Class>, L<DBIx::Abstract>, L<CGI::FormBuilder>, L<HTML::QuickTable>.
2405 Copyright (c) 2001-2007 Nathan Wiger <nwiger@cpan.org>. All Rights Reserved.
2407 This module is actively maintained by Matt Trout <mst@shadowcatsystems.co.uk>
2409 For support, your best bet is to try the C<DBIx::Class> users mailing list.
2410 While not an official support venue, C<DBIx::Class> makes heavy use of
2411 C<SQL::Abstract>, and as such list members there are very familiar with
2412 how to create queries.
2416 This module is free software; you may copy this under the same
2417 terms as perl itself (either the GNU General Public License or
2418 the Artistic License)