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{name_sep} ||= '.';
115 $opt{renderer} ||= do {
116 require Data::Query::Renderer::SQL::Naive;
117 my ($always, $chars);
118 for ($opt{quote_char}) {
119 $chars = defined() ? (ref() ? $_ : [$_]) : ['',''];
122 Data::Query::Renderer::SQL::Naive->new({
123 quote_chars => $chars, always_quote => $always,
124 ($opt{case} ? (lc_keywords => 1) : ()), # always 'lower' if it exists
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, $literal) = @_;
148 ($literal, @bind) = @$literal if ref($literal) eq 'ARRAY';
153 (@bind ? (values => [ $self->_bind_to_dq(@bind) ]) : ()),
158 my ($self, @bind) = @_;
160 $self->{bindtype} eq 'normal'
161 ? map perl_scalar_value($_), @bind
163 $self->_assert_bindval_matches_bindtype(@bind);
164 map perl_scalar_value(reverse @$_), @bind
169 my ($self, $value) = @_;
170 $self->_maybe_convert_dq(perl_scalar_value($value, our $Cur_Col_Meta));
174 my ($self, $ident) = @_;
175 $self->_assert_pass_injection_guard($ident)
176 unless $self->{renderer}{always_quote};
177 $self->_maybe_convert_dq({
178 type => DQ_IDENTIFIER,
179 elements => [ split /\Q$self->{name_sep}/, $ident ],
183 sub _maybe_convert_dq {
184 my ($self, $dq) = @_;
185 if (my $c = $self->{where_convert}) {
188 operator => { 'SQL.Naive' => 'apply' },
190 { type => DQ_IDENTIFIER, elements => [ $self->_sqlcase($c) ] },
200 my ($self, $op, @args) = @_;
201 $self->_assert_pass_injection_guard($op);
204 operator => { 'SQL.Naive' => $op },
209 sub _assert_pass_injection_guard {
210 if ($_[1] =~ $_[0]->{injection_guard}) {
211 my $class = ref $_[0];
212 puke "Possible SQL injection attempt '$_[1]'. If this is indeed a part of the "
213 . "desired SQL use literal SQL ( \'...' or \[ '...' ] ) or supply your own "
214 . "{injection_guard} attribute to ${class}->new()"
219 #======================================================================
221 #======================================================================
225 $self->_render_dq($self->_insert_to_dq(@_));
229 my ($self, $table, $data, $options) = @_;
230 my (@names, @values);
231 if (ref($data) eq 'HASH') {
232 @names = sort keys %$data;
233 foreach my $k (@names) {
234 local our $Cur_Col_Meta = $k;
235 push @values, $self->_mutation_rhs_to_dq($data->{$k});
237 } elsif (ref($data) eq 'ARRAY') {
238 local our $Cur_Col_Meta;
239 @values = map $self->_mutation_rhs_to_dq($_), @$data;
241 die "Not handled yet";
244 if (my $r_source = $options->{returning}) {
246 map +(ref($_) ? $self->_expr_to_dq($_) : $self->_ident_to_dq($_)),
247 (ref($r_source) eq 'ARRAY' ? @$r_source : $r_source),
252 target => $self->_ident_to_dq($table),
253 (@names ? (names => [ map $self->_ident_to_dq($_), @names ]) : ()),
254 values => [ \@values ],
255 ($returning ? (returning => $returning) : ()),
259 sub _mutation_rhs_to_dq {
261 if (ref($v) eq 'ARRAY') {
262 if ($self->{array_datatypes}) {
263 return $self->_value_to_dq($v);
265 $v = \do { my $x = $v };
267 if (ref($v) eq 'HASH') {
268 my ($op, $arg, @rest) = %$v;
270 puke 'Operator calls in update/insert must be in the form { -op => $arg }'
271 if (@rest or not $op =~ /^\-(.+)/);
273 return $self->_expr_to_dq($v);
276 #======================================================================
278 #======================================================================
283 $self->_render_dq($self->_update_to_dq(@_));
287 my ($self, $table, $data, $where) = @_;
289 puke "Unsupported data type specified to \$sql->update"
290 unless ref $data eq 'HASH';
294 foreach my $k (sort keys %$data) {
296 local our $Cur_Col_Meta = $k;
297 push @set, [ $self->_ident_to_dq($k), $self->_mutation_rhs_to_dq($v) ];
302 target => $self->_ident_to_dq($table),
304 where => $self->_where_to_dq($where),
309 #======================================================================
311 #======================================================================
314 my ($self, $table, $where) = @_;
316 my $source_dq = $self->_table_to_dq($table);
318 if (my $where_dq = $self->_where_to_dq($where)) {
331 return $self->_render_dq($self->_select_to_dq(@_));
335 my ($self, $table, $fields, $where, $order) = @_;
338 my $source_dq = $self->_source_to_dq($table, $where);
343 map $self->_ident_to_dq($_),
344 ref($fields) eq 'ARRAY' ? @$fields : $fields
350 $final_dq = $self->_order_by_to_dq($order, undef, $final_dq);
356 #======================================================================
358 #======================================================================
363 $self->_render_dq($self->_delete_to_dq(@_));
367 my ($self, $table, $where) = @_;
370 target => $self->_table_to_dq($table),
371 where => $self->_where_to_dq($where),
376 #======================================================================
378 #======================================================================
382 # Finally, a separate routine just to handle WHERE clauses
384 my ($self, $where, $order) = @_;
390 ($sql, @bind) = $self->_recurse_where($where) if defined($where);
391 $sql = $sql ? $self->_sqlcase(' where ') . "( $sql )" : '';
395 $sql .= $self->_order_by($order);
398 return wantarray ? ($sql, @bind) : $sql;
402 my ($self, $where, $logic) = @_;
404 return $self->_render_dq($self->_where_to_dq($where, $logic));
408 my ($self, $where, $logic) = @_;
410 return undef unless defined($where);
412 # turn the convert misfeature on - only used in WHERE clauses
413 local $self->{where_convert} = $self->{convert};
415 return $self->_expr_to_dq($where, $logic);
419 my ($self, $where, $logic) = @_;
421 if (ref($where) eq 'ARRAY') {
422 return $self->_expr_to_dq_ARRAYREF($where, $logic);
423 } elsif (ref($where) eq 'HASH') {
424 return $self->_expr_to_dq_HASHREF($where, $logic);
426 ref($where) eq 'SCALAR'
427 or (ref($where) eq 'REF' and ref($$where) eq 'ARRAY')
429 return $self->_literal_to_dq($$where);
430 } elsif (!ref($where) or Scalar::Util::blessed($where)) {
431 return $self->_value_to_dq($where);
433 die "Can't handle $where";
436 sub _expr_to_dq_ARRAYREF {
437 my ($self, $where, $logic) = @_;
439 $logic = uc($logic || $self->{logic} || 'OR');
440 $logic eq 'AND' or $logic eq 'OR' or puke "unknown logic: $logic";
442 return unless @$where;
444 my ($first, @rest) = @$where;
446 return $self->_expr_to_dq($first) unless @rest;
450 $self->_where_hashpair_to_dq($first => shift(@rest));
452 $self->_expr_to_dq($first);
456 return $self->_expr_to_dq_ARRAYREF(\@rest, $logic) unless $first_dq;
459 $logic, $first_dq, $self->_expr_to_dq_ARRAYREF(\@rest, $logic)
463 sub _expr_to_dq_HASHREF {
464 my ($self, $where, $logic) = @_;
466 $logic = uc($logic) if $logic;
469 $self->_where_hashpair_to_dq($_ => $where->{$_}, $logic)
472 return $dq[0] unless @dq > 1;
474 my $final = pop(@dq);
476 foreach my $dq (reverse @dq) {
477 $final = $self->_op_to_dq($logic||'AND', $dq, $final);
483 sub _where_to_dq_SCALAR {
484 shift->_value_to_dq(@_);
487 sub _where_op_IDENT {
489 my ($op, $rhs) = splice @_, -2;
491 puke "-$op takes a single scalar argument (a quotable identifier)";
494 # in case we are called as a top level special op (no '=')
497 $_ = $self->_convert($self->_quote($_)) for ($lhs, $rhs);
505 sub _where_op_VALUE {
507 my ($op, $rhs) = splice @_, -2;
509 # in case we are called as a top level special op (no '=')
514 ($lhs || $self->{_nested_func_lhs}),
521 $self->_convert($self->_quote($lhs)) . ' = ' . $self->_convert('?'),
525 $self->_convert('?'),
531 sub _where_hashpair_to_dq {
532 my ($self, $k, $v, $logic) = @_;
534 if ($k =~ /^-(.*)/s) {
536 if ($op eq 'AND' or $op eq 'OR') {
537 return $self->_expr_to_dq($v, $op);
538 } elsif ($op eq 'NEST') {
539 return $self->_expr_to_dq($v);
540 } elsif ($op eq 'NOT') {
541 return $self->_op_to_dq(NOT => $self->_expr_to_dq($v));
542 } elsif ($op eq 'BOOL') {
543 return ref($v) ? $self->_expr_to_dq($v) : $self->_ident_to_dq($v);
544 } elsif ($op eq 'NOT_BOOL') {
545 return $self->_op_to_dq(
546 NOT => ref($v) ? $self->_expr_to_dq($v) : $self->_ident_to_dq($v)
548 } elsif ($op =~ /^(?:AND|OR|NEST)_?\d+/) {
549 die "Use of [and|or|nest]_N modifiers is no longer supported";
552 if (ref($v) eq 'HASH' and keys(%$v) == 1 and (keys %$v)[0] =~ /^-(.*)/s) {
554 my ($inner) = values %$v;
557 (map $self->_expr_to_dq($_),
558 (ref($inner) eq 'ARRAY' ? @$inner : $inner))
561 (map $self->_expr_to_dq($_), (ref($v) eq 'ARRAY' ? @$v : $v))
564 $self->_assert_pass_injection_guard($op);
565 return $self->_op_to_dq(
566 apply => $self->_ident_to_dq($op), @args
570 local our $Cur_Col_Meta = $k;
571 if (ref($v) eq 'ARRAY') {
573 return $self->_literal_to_dq($self->{sqlfalse});
574 } elsif (defined($v->[0]) && $v->[0] =~ /-(and|or)/i) {
575 return $self->_expr_to_dq_ARRAYREF([
576 map +{ $k => $_ }, @{$v}[1..$#$v]
579 return $self->_expr_to_dq_ARRAYREF([
580 map +{ $k => $_ }, @$v
582 } elsif (ref($v) eq 'SCALAR' or (ref($v) eq 'REF' and ref($$v) eq 'ARRAY')) {
586 parts => [ $self->_ident_to_dq($k), $self->_literal_to_dq($$v) ]
589 my ($op, $rhs) = do {
590 if (ref($v) eq 'HASH') {
592 return $self->_expr_to_dq_ARRAYREF([
593 map +{ $k => { $_ => $v->{$_} } }, sort keys %$v
596 my ($op, $value) = %$v;
597 s/^-//, s/_/ /g for $op;
598 if ($op =~ /^(and|or)$/i) {
599 return $self->_expr_to_dq({ $k => $value }, $op);
601 my $special_op = List::Util::first {$op =~ $_->{regex}}
602 @{$self->{special_ops}}
604 return $self->_literal_to_dq(
605 [ $self->${\$special_op->{handler}}($k, $op, $value) ]
607 } elsif ($op =~ /^(?:AND|OR|NEST)_?\d+$/i) {
608 die "Use of [and|or|nest]_N modifiers is no longer supported";
615 if ($op eq 'BETWEEN' or $op eq 'IN' or $op eq 'NOT IN' or $op eq 'NOT BETWEEN') {
616 if (ref($rhs) ne 'ARRAY') {
618 # have to add parens if none present because -in => \"SELECT ..."
619 # got documented. mst hates everything.
620 if (ref($rhs) eq 'SCALAR') {
622 1 while ($x =~ s/\A\s*\((.*)\)\s*\Z/$1/s);
625 my ($x, @rest) = @{$$rhs};
626 1 while ($x =~ s/\A\s*\((.*)\)\s*\Z/$1/s);
627 $rhs = \[ $x, @rest ];
630 return $self->_op_to_dq(
631 $op, $self->_ident_to_dq($k), $self->_literal_to_dq($$rhs)
634 return $self->_literal_to_dq($self->{sqlfalse}) unless @$rhs;
635 return $self->_op_to_dq(
636 $op, $self->_ident_to_dq($k), map $self->_expr_to_dq($_), @$rhs
638 } elsif ($op =~ s/^NOT (?!LIKE)//) {
639 return $self->_where_hashpair_to_dq(-not => { $k => { $op => $rhs } });
640 } elsif (!defined($rhs)) {
642 if ($op eq '=' or $op eq 'LIKE') {
644 } elsif ($op eq '!=') {
647 die "Can't do undef -> NULL transform for operator ${op}";
650 return $self->_op_to_dq($null_op, $self->_ident_to_dq($k));
652 if (ref($rhs) eq 'ARRAY') {
654 return $self->_literal_to_dq(
655 $op eq '!=' ? $self->{sqltrue} : $self->{sqlfalse}
657 } elsif (defined($rhs->[0]) and $rhs->[0] =~ /^-(and|or)$/i) {
658 return $self->_expr_to_dq_ARRAYREF([
659 map +{ $k => { $op => $_ } }, @{$rhs}[1..$#$rhs]
661 } elsif ($op =~ /^-(?:AND|OR|NEST)_?\d+/) {
662 die "Use of [and|or|nest]_N modifiers is no longer supported";
664 return $self->_expr_to_dq_ARRAYREF([
665 map +{ $k => { $op => $_ } }, @$rhs
668 return $self->_op_to_dq(
669 $op, $self->_ident_to_dq($k), $self->_expr_to_dq($rhs)
674 #======================================================================
676 #======================================================================
679 my ($self, $arg) = @_;
680 if (my $dq = $self->_order_by_to_dq($arg)) {
681 # SQLA generates ' ORDER BY foo'. The hilarity.
683 ? do { my @r = $self->_render_dq($dq); $r[0] = ' '.$r[0]; @r }
684 : ' '.$self->_render_dq($dq);
690 sub _order_by_to_dq {
691 my ($self, $arg, $dir, $from) = @_;
697 ($dir ? (direction => $dir) : ()),
698 ($from ? (from => $from) : ()),
702 $dq->{by} = $self->_ident_to_dq($arg);
703 } elsif (ref($arg) eq 'ARRAY') {
705 local our $Order_Inner unless our $Order_Recursing;
706 local $Order_Recursing = 1;
708 foreach my $member (@$arg) {
710 my $next = $self->_order_by_to_dq($member, $dir, $from);
712 $inner->{from} = $next if $inner;
713 $inner = $Order_Inner || $next;
715 $Order_Inner = $inner;
717 } elsif (ref($arg) eq 'REF' and ref($$arg) eq 'ARRAY') {
718 $dq->{by} = $self->_literal_to_dq($$arg);
719 } elsif (ref($arg) eq 'SCALAR') {
720 $dq->{by} = $self->_literal_to_dq($$arg);
721 } elsif (ref($arg) eq 'HASH') {
722 my ($key, $val, @rest) = %$arg;
726 if (@rest or not $key =~ /^-(desc|asc)/i) {
727 puke "hash passed to _order_by must have exactly one key (-desc or -asc)";
730 return $self->_order_by_to_dq($val, $dir, $from);
732 die "Can't handle $arg in _order_by_to_dq";
737 #======================================================================
738 # DATASOURCE (FOR NOW, JUST PLAIN TABLE OR LIST OF TABLES)
739 #======================================================================
742 my ($self, $from) = @_;
743 $self->_render_dq($self->_table_to_dq($from));
747 my ($self, $from) = @_;
748 if (ref($from) eq 'ARRAY') {
749 die "Empty FROM list" unless my @f = @$from;
750 my $dq = $self->_ident_to_dq(shift @f);
751 while (my $x = shift @f) {
754 join => [ $dq, $self->_ident_to_dq($x) ]
758 } elsif (ref($from) eq 'SCALAR') {
765 $self->_ident_to_dq($from);
770 #======================================================================
772 #======================================================================
774 # highly optimized, as it's called way too often
776 # my ($self, $label) = @_;
778 return '' unless defined $_[1];
779 return ${$_[1]} if ref($_[1]) eq 'SCALAR';
781 unless ($_[0]->{quote_char}) {
782 $_[0]->_assert_pass_injection_guard($_[1]);
786 my $qref = ref $_[0]->{quote_char};
789 ($l, $r) = ( $_[0]->{quote_char}, $_[0]->{quote_char} );
791 elsif ($qref eq 'ARRAY') {
792 ($l, $r) = @{$_[0]->{quote_char}};
795 puke "Unsupported quote_char format: $_[0]->{quote_char}";
798 # parts containing * are naturally unquoted
799 return join( $_[0]->{name_sep}||'', map
800 { $_ eq '*' ? $_ : $l . $_ . $r }
801 ( $_[0]->{name_sep} ? split (/\Q$_[0]->{name_sep}\E/, $_[1] ) : $_[1] )
806 # Conversion, if applicable
808 #my ($self, $arg) = @_;
810 # LDNOTE : modified the previous implementation below because
811 # it was not consistent : the first "return" is always an array,
812 # the second "return" is context-dependent. Anyway, _convert
813 # seems always used with just a single argument, so make it a
815 # return @_ unless $self->{convert};
816 # my $conv = $self->_sqlcase($self->{convert});
817 # my @ret = map { $conv.'('.$_.')' } @_;
818 # return wantarray ? @ret : $ret[0];
819 if ($_[0]->{convert}) {
820 return $_[0]->_sqlcase($_[0]->{convert}) .'(' . $_[1] . ')';
827 #my ($self, $col, @vals) = @_;
829 #LDNOTE : changed original implementation below because it did not make
830 # sense when bindtype eq 'columns' and @vals > 1.
831 # return $self->{bindtype} eq 'columns' ? [ $col, @vals ] : @vals;
833 # called often - tighten code
834 return $_[0]->{bindtype} eq 'columns'
835 ? map {[$_[1], $_]} @_[2 .. $#_]
840 # Dies if any element of @bind is not in [colname => value] format
841 # if bindtype is 'columns'.
842 sub _assert_bindval_matches_bindtype {
843 # my ($self, @bind) = @_;
845 if ($self->{bindtype} eq 'columns') {
847 if (!defined $_ || ref($_) ne 'ARRAY' || @$_ != 2) {
848 puke "bindtype 'columns' selected, you need to pass: [column_name => bind_value]"
854 sub _join_sql_clauses {
855 my ($self, $logic, $clauses_aref, $bind_aref) = @_;
857 if (@$clauses_aref > 1) {
858 my $join = " " . $self->_sqlcase($logic) . " ";
859 my $sql = '( ' . join($join, @$clauses_aref) . ' )';
860 return ($sql, @$bind_aref);
862 elsif (@$clauses_aref) {
863 return ($clauses_aref->[0], @$bind_aref); # no parentheses
866 return (); # if no SQL, ignore @$bind_aref
871 # Fix SQL case, if so requested
873 # LDNOTE: if $self->{case} is true, then it contains 'lower', so we
874 # don't touch the argument ... crooked logic, but let's not change it!
875 return $_[0]->{case} ? $_[1] : uc($_[1]);
879 #======================================================================
880 # DISPATCHING FROM REFKIND
881 #======================================================================
884 my ($self, $data) = @_;
886 return 'UNDEF' unless defined $data;
888 # blessed objects are treated like scalars
889 my $ref = (Scalar::Util::blessed $data) ? '' : ref $data;
891 return 'SCALAR' unless $ref;
894 while ($ref eq 'REF') {
896 $ref = (Scalar::Util::blessed $data) ? '' : ref $data;
900 return ($ref||'SCALAR') . ('REF' x $n_steps);
904 my ($self, $data) = @_;
905 my @try = ($self->_refkind($data));
906 push @try, 'SCALAR_or_UNDEF' if $try[0] eq 'SCALAR' || $try[0] eq 'UNDEF';
907 push @try, 'FALLBACK';
911 sub _METHOD_FOR_refkind {
912 my ($self, $meth_prefix, $data) = @_;
915 for (@{$self->_try_refkind($data)}) {
916 $method = $self->can($meth_prefix."_".$_)
920 return $method || puke "cannot dispatch on '$meth_prefix' for ".$self->_refkind($data);
924 sub _SWITCH_refkind {
925 my ($self, $data, $dispatch_table) = @_;
928 for (@{$self->_try_refkind($data)}) {
929 $coderef = $dispatch_table->{$_}
933 puke "no dispatch entry for ".$self->_refkind($data)
942 #======================================================================
943 # VALUES, GENERATE, AUTOLOAD
944 #======================================================================
946 # LDNOTE: original code from nwiger, didn't touch code in that section
947 # I feel the AUTOLOAD stuff should not be the default, it should
948 # only be activated on explicit demand by user.
952 my $data = shift || return;
953 puke "Argument to ", __PACKAGE__, "->values must be a \\%hash"
954 unless ref $data eq 'HASH';
957 foreach my $k ( sort keys %$data ) {
959 $self->_SWITCH_refkind($v, {
961 if ($self->{array_datatypes}) { # array datatype
962 push @all_bind, $self->_bindtype($k, $v);
964 else { # literal SQL with bind
965 my ($sql, @bind) = @$v;
966 $self->_assert_bindval_matches_bindtype(@bind);
967 push @all_bind, @bind;
970 ARRAYREFREF => sub { # literal SQL with bind
971 my ($sql, @bind) = @${$v};
972 $self->_assert_bindval_matches_bindtype(@bind);
973 push @all_bind, @bind;
975 SCALARREF => sub { # literal SQL without bind
977 SCALAR_or_UNDEF => sub {
978 push @all_bind, $self->_bindtype($k, $v);
989 my(@sql, @sqlq, @sqlv);
993 if ($ref eq 'HASH') {
994 for my $k (sort keys %$_) {
997 my $label = $self->_quote($k);
999 # literal SQL with bind
1000 my ($sql, @bind) = @$v;
1001 $self->_assert_bindval_matches_bindtype(@bind);
1002 push @sqlq, "$label = $sql";
1004 } elsif ($r eq 'SCALAR') {
1005 # literal SQL without bind
1006 push @sqlq, "$label = $$v";
1008 push @sqlq, "$label = ?";
1009 push @sqlv, $self->_bindtype($k, $v);
1012 push @sql, $self->_sqlcase('set'), join ', ', @sqlq;
1013 } elsif ($ref eq 'ARRAY') {
1014 # unlike insert(), assume these are ONLY the column names, i.e. for SQL
1017 if ($r eq 'ARRAY') { # literal SQL with bind
1018 my ($sql, @bind) = @$v;
1019 $self->_assert_bindval_matches_bindtype(@bind);
1022 } elsif ($r eq 'SCALAR') { # literal SQL without bind
1023 # embedded literal SQL
1030 push @sql, '(' . join(', ', @sqlq) . ')';
1031 } elsif ($ref eq 'SCALAR') {
1035 # strings get case twiddled
1036 push @sql, $self->_sqlcase($_);
1040 my $sql = join ' ', @sql;
1042 # this is pretty tricky
1043 # if ask for an array, return ($stmt, @bind)
1044 # otherwise, s/?/shift @sqlv/ to put it inline
1046 return ($sql, @sqlv);
1048 1 while $sql =~ s/\?/my $d = shift(@sqlv);
1049 ref $d ? $d->[1] : $d/e;
1058 # # This allows us to check for a local, then _form, attr
1060 # my($name) = $AUTOLOAD =~ /.*::(.+)/;
1061 # return $self->generate($name, @_);
1072 SQL::Abstract - Generate SQL from Perl data structures
1078 my $sql = SQL::Abstract->new;
1080 my($stmt, @bind) = $sql->select($table, \@fields, \%where, \@order);
1082 my($stmt, @bind) = $sql->insert($table, \%fieldvals || \@values);
1084 my($stmt, @bind) = $sql->update($table, \%fieldvals, \%where);
1086 my($stmt, @bind) = $sql->delete($table, \%where);
1088 # Then, use these in your DBI statements
1089 my $sth = $dbh->prepare($stmt);
1090 $sth->execute(@bind);
1092 # Just generate the WHERE clause
1093 my($stmt, @bind) = $sql->where(\%where, \@order);
1095 # Return values in the same order, for hashed queries
1096 # See PERFORMANCE section for more details
1097 my @bind = $sql->values(\%fieldvals);
1101 This module was inspired by the excellent L<DBIx::Abstract>.
1102 However, in using that module I found that what I really wanted
1103 to do was generate SQL, but still retain complete control over my
1104 statement handles and use the DBI interface. So, I set out to
1105 create an abstract SQL generation module.
1107 While based on the concepts used by L<DBIx::Abstract>, there are
1108 several important differences, especially when it comes to WHERE
1109 clauses. I have modified the concepts used to make the SQL easier
1110 to generate from Perl data structures and, IMO, more intuitive.
1111 The underlying idea is for this module to do what you mean, based
1112 on the data structures you provide it. The big advantage is that
1113 you don't have to modify your code every time your data changes,
1114 as this module figures it out.
1116 To begin with, an SQL INSERT is as easy as just specifying a hash
1117 of C<key=value> pairs:
1120 name => 'Jimbo Bobson',
1121 phone => '123-456-7890',
1122 address => '42 Sister Lane',
1123 city => 'St. Louis',
1124 state => 'Louisiana',
1127 The SQL can then be generated with this:
1129 my($stmt, @bind) = $sql->insert('people', \%data);
1131 Which would give you something like this:
1133 $stmt = "INSERT INTO people
1134 (address, city, name, phone, state)
1135 VALUES (?, ?, ?, ?, ?)";
1136 @bind = ('42 Sister Lane', 'St. Louis', 'Jimbo Bobson',
1137 '123-456-7890', 'Louisiana');
1139 These are then used directly in your DBI code:
1141 my $sth = $dbh->prepare($stmt);
1142 $sth->execute(@bind);
1144 =head2 Inserting and Updating Arrays
1146 If your database has array types (like for example Postgres),
1147 activate the special option C<< array_datatypes => 1 >>
1148 when creating the C<SQL::Abstract> object.
1149 Then you may use an arrayref to insert and update database array types:
1151 my $sql = SQL::Abstract->new(array_datatypes => 1);
1153 planets => [qw/Mercury Venus Earth Mars/]
1156 my($stmt, @bind) = $sql->insert('solar_system', \%data);
1160 $stmt = "INSERT INTO solar_system (planets) VALUES (?)"
1162 @bind = (['Mercury', 'Venus', 'Earth', 'Mars']);
1165 =head2 Inserting and Updating SQL
1167 In order to apply SQL functions to elements of your C<%data> you may
1168 specify a reference to an arrayref for the given hash value. For example,
1169 if you need to execute the Oracle C<to_date> function on a value, you can
1170 say something like this:
1174 date_entered => \["to_date(?,'MM/DD/YYYY')", "03/02/2003"],
1177 The first value in the array is the actual SQL. Any other values are
1178 optional and would be included in the bind values array. This gives
1181 my($stmt, @bind) = $sql->insert('people', \%data);
1183 $stmt = "INSERT INTO people (name, date_entered)
1184 VALUES (?, to_date(?,'MM/DD/YYYY'))";
1185 @bind = ('Bill', '03/02/2003');
1187 An UPDATE is just as easy, all you change is the name of the function:
1189 my($stmt, @bind) = $sql->update('people', \%data);
1191 Notice that your C<%data> isn't touched; the module will generate
1192 the appropriately quirky SQL for you automatically. Usually you'll
1193 want to specify a WHERE clause for your UPDATE, though, which is
1194 where handling C<%where> hashes comes in handy...
1196 =head2 Complex where statements
1198 This module can generate pretty complicated WHERE statements
1199 easily. For example, simple C<key=value> pairs are taken to mean
1200 equality, and if you want to see if a field is within a set
1201 of values, you can use an arrayref. Let's say we wanted to
1202 SELECT some data based on this criteria:
1205 requestor => 'inna',
1206 worker => ['nwiger', 'rcwe', 'sfz'],
1207 status => { '!=', 'completed' }
1210 my($stmt, @bind) = $sql->select('tickets', '*', \%where);
1212 The above would give you something like this:
1214 $stmt = "SELECT * FROM tickets WHERE
1215 ( requestor = ? ) AND ( status != ? )
1216 AND ( worker = ? OR worker = ? OR worker = ? )";
1217 @bind = ('inna', 'completed', 'nwiger', 'rcwe', 'sfz');
1219 Which you could then use in DBI code like so:
1221 my $sth = $dbh->prepare($stmt);
1222 $sth->execute(@bind);
1228 The functions are simple. There's one for each major SQL operation,
1229 and a constructor you use first. The arguments are specified in a
1230 similar order to each function (table, then fields, then a where
1231 clause) to try and simplify things.
1236 =head2 new(option => 'value')
1238 The C<new()> function takes a list of options and values, and returns
1239 a new B<SQL::Abstract> object which can then be used to generate SQL
1240 through the methods below. The options accepted are:
1246 If set to 'lower', then SQL will be generated in all lowercase. By
1247 default SQL is generated in "textbook" case meaning something like:
1249 SELECT a_field FROM a_table WHERE some_field LIKE '%someval%'
1251 Any setting other than 'lower' is ignored.
1255 This determines what the default comparison operator is. By default
1256 it is C<=>, meaning that a hash like this:
1258 %where = (name => 'nwiger', email => 'nate@wiger.org');
1260 Will generate SQL like this:
1262 WHERE name = 'nwiger' AND email = 'nate@wiger.org'
1264 However, you may want loose comparisons by default, so if you set
1265 C<cmp> to C<like> you would get SQL such as:
1267 WHERE name like 'nwiger' AND email like 'nate@wiger.org'
1269 You can also override the comparsion on an individual basis - see
1270 the huge section on L</"WHERE CLAUSES"> at the bottom.
1272 =item sqltrue, sqlfalse
1274 Expressions for inserting boolean values within SQL statements.
1275 By default these are C<1=1> and C<1=0>. They are used
1276 by the special operators C<-in> and C<-not_in> for generating
1277 correct SQL even when the argument is an empty array (see below).
1281 This determines the default logical operator for multiple WHERE
1282 statements in arrays or hashes. If absent, the default logic is "or"
1283 for arrays, and "and" for hashes. This means that a WHERE
1287 event_date => {'>=', '2/13/99'},
1288 event_date => {'<=', '4/24/03'},
1291 will generate SQL like this:
1293 WHERE event_date >= '2/13/99' OR event_date <= '4/24/03'
1295 This is probably not what you want given this query, though (look
1296 at the dates). To change the "OR" to an "AND", simply specify:
1298 my $sql = SQL::Abstract->new(logic => 'and');
1300 Which will change the above C<WHERE> to:
1302 WHERE event_date >= '2/13/99' AND event_date <= '4/24/03'
1304 The logic can also be changed locally by inserting
1305 a modifier in front of an arrayref :
1307 @where = (-and => [event_date => {'>=', '2/13/99'},
1308 event_date => {'<=', '4/24/03'} ]);
1310 See the L</"WHERE CLAUSES"> section for explanations.
1314 This will automatically convert comparisons using the specified SQL
1315 function for both column and value. This is mostly used with an argument
1316 of C<upper> or C<lower>, so that the SQL will have the effect of
1317 case-insensitive "searches". For example, this:
1319 $sql = SQL::Abstract->new(convert => 'upper');
1320 %where = (keywords => 'MaKe iT CAse inSeNSItive');
1322 Will turn out the following SQL:
1324 WHERE upper(keywords) like upper('MaKe iT CAse inSeNSItive')
1326 The conversion can be C<upper()>, C<lower()>, or any other SQL function
1327 that can be applied symmetrically to fields (actually B<SQL::Abstract> does
1328 not validate this option; it will just pass through what you specify verbatim).
1332 This is a kludge because many databases suck. For example, you can't
1333 just bind values using DBI's C<execute()> for Oracle C<CLOB> or C<BLOB> fields.
1334 Instead, you have to use C<bind_param()>:
1336 $sth->bind_param(1, 'reg data');
1337 $sth->bind_param(2, $lots, {ora_type => ORA_CLOB});
1339 The problem is, B<SQL::Abstract> will normally just return a C<@bind> array,
1340 which loses track of which field each slot refers to. Fear not.
1342 If you specify C<bindtype> in new, you can determine how C<@bind> is returned.
1343 Currently, you can specify either C<normal> (default) or C<columns>. If you
1344 specify C<columns>, you will get an array that looks like this:
1346 my $sql = SQL::Abstract->new(bindtype => 'columns');
1347 my($stmt, @bind) = $sql->insert(...);
1350 [ 'column1', 'value1' ],
1351 [ 'column2', 'value2' ],
1352 [ 'column3', 'value3' ],
1355 You can then iterate through this manually, using DBI's C<bind_param()>.
1357 $sth->prepare($stmt);
1360 my($col, $data) = @$_;
1361 if ($col eq 'details' || $col eq 'comments') {
1362 $sth->bind_param($i, $data, {ora_type => ORA_CLOB});
1363 } elsif ($col eq 'image') {
1364 $sth->bind_param($i, $data, {ora_type => ORA_BLOB});
1366 $sth->bind_param($i, $data);
1370 $sth->execute; # execute without @bind now
1372 Now, why would you still use B<SQL::Abstract> if you have to do this crap?
1373 Basically, the advantage is still that you don't have to care which fields
1374 are or are not included. You could wrap that above C<for> loop in a simple
1375 sub called C<bind_fields()> or something and reuse it repeatedly. You still
1376 get a layer of abstraction over manual SQL specification.
1378 Note that if you set L</bindtype> to C<columns>, the C<\[$sql, @bind]>
1379 construct (see L</Literal SQL with placeholders and bind values (subqueries)>)
1380 will expect the bind values in this format.
1384 This is the character that a table or column name will be quoted
1385 with. By default this is an empty string, but you could set it to
1386 the character C<`>, to generate SQL like this:
1388 SELECT `a_field` FROM `a_table` WHERE `some_field` LIKE '%someval%'
1390 Alternatively, you can supply an array ref of two items, the first being the left
1391 hand quote character, and the second the right hand quote character. For
1392 example, you could supply C<['[',']']> for SQL Server 2000 compliant quotes
1393 that generates SQL like this:
1395 SELECT [a_field] FROM [a_table] WHERE [some_field] LIKE '%someval%'
1397 Quoting is useful if you have tables or columns names that are reserved
1398 words in your database's SQL dialect.
1402 This is the character that separates a table and column name. It is
1403 necessary to specify this when the C<quote_char> option is selected,
1404 so that tables and column names can be individually quoted like this:
1406 SELECT `table`.`one_field` FROM `table` WHERE `table`.`other_field` = 1
1408 =item injection_guard
1410 A regular expression C<qr/.../> that is applied to any C<-function> and unquoted
1411 column name specified in a query structure. This is a safety mechanism to avoid
1412 injection attacks when mishandling user input e.g.:
1414 my %condition_as_column_value_pairs = get_values_from_user();
1415 $sqla->select( ... , \%condition_as_column_value_pairs );
1417 If the expression matches an exception is thrown. Note that literal SQL
1418 supplied via C<\'...'> or C<\['...']> is B<not> checked in any way.
1420 Defaults to checking for C<;> and the C<GO> keyword (TransactSQL)
1422 =item array_datatypes
1424 When this option is true, arrayrefs in INSERT or UPDATE are
1425 interpreted as array datatypes and are passed directly
1427 When this option is false, arrayrefs are interpreted
1428 as literal SQL, just like refs to arrayrefs
1429 (but this behavior is for backwards compatibility; when writing
1430 new queries, use the "reference to arrayref" syntax
1436 Takes a reference to a list of "special operators"
1437 to extend the syntax understood by L<SQL::Abstract>.
1438 See section L</"SPECIAL OPERATORS"> for details.
1442 Takes a reference to a list of "unary operators"
1443 to extend the syntax understood by L<SQL::Abstract>.
1444 See section L</"UNARY OPERATORS"> for details.
1450 =head2 insert($table, \@values || \%fieldvals, \%options)
1452 This is the simplest function. You simply give it a table name
1453 and either an arrayref of values or hashref of field/value pairs.
1454 It returns an SQL INSERT statement and a list of bind values.
1455 See the sections on L</"Inserting and Updating Arrays"> and
1456 L</"Inserting and Updating SQL"> for information on how to insert
1457 with those data types.
1459 The optional C<\%options> hash reference may contain additional
1460 options to generate the insert SQL. Currently supported options
1467 Takes either a scalar of raw SQL fields, or an array reference of
1468 field names, and adds on an SQL C<RETURNING> statement at the end.
1469 This allows you to return data generated by the insert statement
1470 (such as row IDs) without performing another C<SELECT> statement.
1471 Note, however, this is not part of the SQL standard and may not
1472 be supported by all database engines.
1476 =head2 update($table, \%fieldvals, \%where)
1478 This takes a table, hashref of field/value pairs, and an optional
1479 hashref L<WHERE clause|/WHERE CLAUSES>. It returns an SQL UPDATE function and a list
1481 See the sections on L</"Inserting and Updating Arrays"> and
1482 L</"Inserting and Updating SQL"> for information on how to insert
1483 with those data types.
1485 =head2 select($source, $fields, $where, $order)
1487 This returns a SQL SELECT statement and associated list of bind values, as
1488 specified by the arguments :
1494 Specification of the 'FROM' part of the statement.
1495 The argument can be either a plain scalar (interpreted as a table
1496 name, will be quoted), or an arrayref (interpreted as a list
1497 of table names, joined by commas, quoted), or a scalarref
1498 (literal table name, not quoted), or a ref to an arrayref
1499 (list of literal table names, joined by commas, not quoted).
1503 Specification of the list of fields to retrieve from
1505 The argument can be either an arrayref (interpreted as a list
1506 of field names, will be joined by commas and quoted), or a
1507 plain scalar (literal SQL, not quoted).
1508 Please observe that this API is not as flexible as for
1509 the first argument C<$table>, for backwards compatibility reasons.
1513 Optional argument to specify the WHERE part of the query.
1514 The argument is most often a hashref, but can also be
1515 an arrayref or plain scalar --
1516 see section L<WHERE clause|/"WHERE CLAUSES"> for details.
1520 Optional argument to specify the ORDER BY part of the query.
1521 The argument can be a scalar, a hashref or an arrayref
1522 -- see section L<ORDER BY clause|/"ORDER BY CLAUSES">
1528 =head2 delete($table, \%where)
1530 This takes a table name and optional hashref L<WHERE clause|/WHERE CLAUSES>.
1531 It returns an SQL DELETE statement and list of bind values.
1533 =head2 where(\%where, \@order)
1535 This is used to generate just the WHERE clause. For example,
1536 if you have an arbitrary data structure and know what the
1537 rest of your SQL is going to look like, but want an easy way
1538 to produce a WHERE clause, use this. It returns an SQL WHERE
1539 clause and list of bind values.
1542 =head2 values(\%data)
1544 This just returns the values from the hash C<%data>, in the same
1545 order that would be returned from any of the other above queries.
1546 Using this allows you to markedly speed up your queries if you
1547 are affecting lots of rows. See below under the L</"PERFORMANCE"> section.
1549 =head2 generate($any, 'number', $of, \@data, $struct, \%types)
1551 Warning: This is an experimental method and subject to change.
1553 This returns arbitrarily generated SQL. It's a really basic shortcut.
1554 It will return two different things, depending on return context:
1556 my($stmt, @bind) = $sql->generate('create table', \$table, \@fields);
1557 my $stmt_and_val = $sql->generate('create table', \$table, \@fields);
1559 These would return the following:
1561 # First calling form
1562 $stmt = "CREATE TABLE test (?, ?)";
1563 @bind = (field1, field2);
1565 # Second calling form
1566 $stmt_and_val = "CREATE TABLE test (field1, field2)";
1568 Depending on what you're trying to do, it's up to you to choose the correct
1569 format. In this example, the second form is what you would want.
1573 $sql->generate('alter session', { nls_date_format => 'MM/YY' });
1577 ALTER SESSION SET nls_date_format = 'MM/YY'
1579 You get the idea. Strings get their case twiddled, but everything
1580 else remains verbatim.
1582 =head1 WHERE CLAUSES
1586 This module uses a variation on the idea from L<DBIx::Abstract>. It
1587 is B<NOT>, repeat I<not> 100% compatible. B<The main logic of this
1588 module is that things in arrays are OR'ed, and things in hashes
1591 The easiest way to explain is to show lots of examples. After
1592 each C<%where> hash shown, it is assumed you used:
1594 my($stmt, @bind) = $sql->where(\%where);
1596 However, note that the C<%where> hash can be used directly in any
1597 of the other functions as well, as described above.
1599 =head2 Key-value pairs
1601 So, let's get started. To begin, a simple hash:
1605 status => 'completed'
1608 Is converted to SQL C<key = val> statements:
1610 $stmt = "WHERE user = ? AND status = ?";
1611 @bind = ('nwiger', 'completed');
1613 One common thing I end up doing is having a list of values that
1614 a field can be in. To do this, simply specify a list inside of
1619 status => ['assigned', 'in-progress', 'pending'];
1622 This simple code will create the following:
1624 $stmt = "WHERE user = ? AND ( status = ? OR status = ? OR status = ? )";
1625 @bind = ('nwiger', 'assigned', 'in-progress', 'pending');
1627 A field associated to an empty arrayref will be considered a
1628 logical false and will generate 0=1.
1630 =head2 Tests for NULL values
1632 If the value part is C<undef> then this is converted to SQL <IS NULL>
1641 $stmt = "WHERE user = ? AND status IS NULL";
1644 To test if a column IS NOT NULL:
1648 status => { '!=', undef },
1651 =head2 Specific comparison operators
1653 If you want to specify a different type of operator for your comparison,
1654 you can use a hashref for a given column:
1658 status => { '!=', 'completed' }
1661 Which would generate:
1663 $stmt = "WHERE user = ? AND status != ?";
1664 @bind = ('nwiger', 'completed');
1666 To test against multiple values, just enclose the values in an arrayref:
1668 status => { '=', ['assigned', 'in-progress', 'pending'] };
1670 Which would give you:
1672 "WHERE status = ? OR status = ? OR status = ?"
1675 The hashref can also contain multiple pairs, in which case it is expanded
1676 into an C<AND> of its elements:
1680 status => { '!=', 'completed', -not_like => 'pending%' }
1683 # Or more dynamically, like from a form
1684 $where{user} = 'nwiger';
1685 $where{status}{'!='} = 'completed';
1686 $where{status}{'-not_like'} = 'pending%';
1688 # Both generate this
1689 $stmt = "WHERE user = ? AND status != ? AND status NOT LIKE ?";
1690 @bind = ('nwiger', 'completed', 'pending%');
1693 To get an OR instead, you can combine it with the arrayref idea:
1697 priority => [ { '=', 2 }, { '>', 5 } ]
1700 Which would generate:
1702 $stmt = "WHERE ( priority = ? OR priority > ? ) AND user = ?";
1703 @bind = ('2', '5', 'nwiger');
1705 If you want to include literal SQL (with or without bind values), just use a
1706 scalar reference or array reference as the value:
1709 date_entered => { '>' => \["to_date(?, 'MM/DD/YYYY')", "11/26/2008"] },
1710 date_expires => { '<' => \"now()" }
1713 Which would generate:
1715 $stmt = "WHERE date_entered > "to_date(?, 'MM/DD/YYYY') AND date_expires < now()";
1716 @bind = ('11/26/2008');
1719 =head2 Logic and nesting operators
1721 In the example above,
1722 there is a subtle trap if you want to say something like
1723 this (notice the C<AND>):
1725 WHERE priority != ? AND priority != ?
1727 Because, in Perl you I<can't> do this:
1729 priority => { '!=', 2, '!=', 1 }
1731 As the second C<!=> key will obliterate the first. The solution
1732 is to use the special C<-modifier> form inside an arrayref:
1734 priority => [ -and => {'!=', 2},
1738 Normally, these would be joined by C<OR>, but the modifier tells it
1739 to use C<AND> instead. (Hint: You can use this in conjunction with the
1740 C<logic> option to C<new()> in order to change the way your queries
1741 work by default.) B<Important:> Note that the C<-modifier> goes
1742 B<INSIDE> the arrayref, as an extra first element. This will
1743 B<NOT> do what you think it might:
1745 priority => -and => [{'!=', 2}, {'!=', 1}] # WRONG!
1747 Here is a quick list of equivalencies, since there is some overlap:
1750 status => {'!=', 'completed', 'not like', 'pending%' }
1751 status => [ -and => {'!=', 'completed'}, {'not like', 'pending%'}]
1754 status => {'=', ['assigned', 'in-progress']}
1755 status => [ -or => {'=', 'assigned'}, {'=', 'in-progress'}]
1756 status => [ {'=', 'assigned'}, {'=', 'in-progress'} ]
1760 =head2 Special operators : IN, BETWEEN, etc.
1762 You can also use the hashref format to compare a list of fields using the
1763 C<IN> comparison operator, by specifying the list as an arrayref:
1766 status => 'completed',
1767 reportid => { -in => [567, 2335, 2] }
1770 Which would generate:
1772 $stmt = "WHERE status = ? AND reportid IN (?,?,?)";
1773 @bind = ('completed', '567', '2335', '2');
1775 The reverse operator C<-not_in> generates SQL C<NOT IN> and is used in
1778 If the argument to C<-in> is an empty array, 'sqlfalse' is generated
1779 (by default : C<1=0>). Similarly, C<< -not_in => [] >> generates
1780 'sqltrue' (by default : C<1=1>).
1782 In addition to the array you can supply a chunk of literal sql or
1783 literal sql with bind:
1786 customer => { -in => \[
1787 'SELECT cust_id FROM cust WHERE balance > ?',
1790 status => { -in => \'SELECT status_codes FROM states' },
1796 customer IN ( SELECT cust_id FROM cust WHERE balance > ? )
1797 AND status IN ( SELECT status_codes FROM states )
1803 Another pair of operators is C<-between> and C<-not_between>,
1804 used with an arrayref of two values:
1808 completion_date => {
1809 -not_between => ['2002-10-01', '2003-02-06']
1815 WHERE user = ? AND completion_date NOT BETWEEN ( ? AND ? )
1817 Just like with C<-in> all plausible combinations of literal SQL
1821 start0 => { -between => [ 1, 2 ] },
1822 start1 => { -between => \["? AND ?", 1, 2] },
1823 start2 => { -between => \"lower(x) AND upper(y)" },
1824 start3 => { -between => [
1826 \["upper(?)", 'stuff' ],
1833 ( start0 BETWEEN ? AND ? )
1834 AND ( start1 BETWEEN ? AND ? )
1835 AND ( start2 BETWEEN lower(x) AND upper(y) )
1836 AND ( start3 BETWEEN lower(x) AND upper(?) )
1838 @bind = (1, 2, 1, 2, 'stuff');
1841 These are the two builtin "special operators"; but the
1842 list can be expanded : see section L</"SPECIAL OPERATORS"> below.
1844 =head2 Unary operators: bool
1846 If you wish to test against boolean columns or functions within your
1847 database you can use the C<-bool> and C<-not_bool> operators. For
1848 example to test the column C<is_user> being true and the column
1849 C<is_enabled> being false you would use:-
1853 -not_bool => 'is_enabled',
1858 WHERE is_user AND NOT is_enabled
1860 If a more complex combination is required, testing more conditions,
1861 then you should use the and/or operators:-
1868 -not_bool => 'four',
1874 WHERE one AND two AND three AND NOT four
1877 =head2 Nested conditions, -and/-or prefixes
1879 So far, we've seen how multiple conditions are joined with a top-level
1880 C<AND>. We can change this by putting the different conditions we want in
1881 hashes and then putting those hashes in an array. For example:
1886 status => { -like => ['pending%', 'dispatched'] },
1890 status => 'unassigned',
1894 This data structure would create the following:
1896 $stmt = "WHERE ( user = ? AND ( status LIKE ? OR status LIKE ? ) )
1897 OR ( user = ? AND status = ? ) )";
1898 @bind = ('nwiger', 'pending', 'dispatched', 'robot', 'unassigned');
1901 Clauses in hashrefs or arrayrefs can be prefixed with an C<-and> or C<-or>
1902 to change the logic inside :
1908 -and => [ workhrs => {'>', 20}, geo => 'ASIA' ],
1909 -or => { workhrs => {'<', 50}, geo => 'EURO' },
1916 WHERE ( user = ? AND (
1917 ( workhrs > ? AND geo = ? )
1918 OR ( workhrs < ? OR geo = ? )
1921 =head3 Algebraic inconsistency, for historical reasons
1923 C<Important note>: when connecting several conditions, the C<-and->|C<-or>
1924 operator goes C<outside> of the nested structure; whereas when connecting
1925 several constraints on one column, the C<-and> operator goes
1926 C<inside> the arrayref. Here is an example combining both features :
1929 -and => [a => 1, b => 2],
1930 -or => [c => 3, d => 4],
1931 e => [-and => {-like => 'foo%'}, {-like => '%bar'} ]
1936 WHERE ( ( ( a = ? AND b = ? )
1937 OR ( c = ? OR d = ? )
1938 OR ( e LIKE ? AND e LIKE ? ) ) )
1940 This difference in syntax is unfortunate but must be preserved for
1941 historical reasons. So be careful : the two examples below would
1942 seem algebraically equivalent, but they are not
1944 {col => [-and => {-like => 'foo%'}, {-like => '%bar'}]}
1945 # yields : WHERE ( ( col LIKE ? AND col LIKE ? ) )
1947 [-and => {col => {-like => 'foo%'}, {col => {-like => '%bar'}}]]
1948 # yields : WHERE ( ( col LIKE ? OR col LIKE ? ) )
1951 =head2 Literal SQL and value type operators
1953 The basic premise of SQL::Abstract is that in WHERE specifications the "left
1954 side" is a column name and the "right side" is a value (normally rendered as
1955 a placeholder). This holds true for both hashrefs and arrayref pairs as you
1956 see in the L</WHERE CLAUSES> examples above. Sometimes it is necessary to
1957 alter this behavior. There are several ways of doing so.
1961 This is a virtual operator that signals the string to its right side is an
1962 identifier (a column name) and not a value. For example to compare two
1963 columns you would write:
1966 priority => { '<', 2 },
1967 requestor => { -ident => 'submitter' },
1972 $stmt = "WHERE priority < ? AND requestor = submitter";
1975 If you are maintaining legacy code you may see a different construct as
1976 described in L</Deprecated usage of Literal SQL>, please use C<-ident> in new
1981 This is a virtual operator that signals that the construct to its right side
1982 is a value to be passed to DBI. This is for example necessary when you want
1983 to write a where clause against an array (for RDBMS that support such
1984 datatypes). For example:
1987 array => { -value => [1, 2, 3] }
1992 $stmt = 'WHERE array = ?';
1993 @bind = ([1, 2, 3]);
1995 Note that if you were to simply say:
2001 the result would porbably be not what you wanted:
2003 $stmt = 'WHERE array = ? OR array = ? OR array = ?';
2008 Finally, sometimes only literal SQL will do. To include a random snippet
2009 of SQL verbatim, you specify it as a scalar reference. Consider this only
2010 as a last resort. Usually there is a better way. For example:
2013 priority => { '<', 2 },
2014 requestor => { -in => \'(SELECT name FROM hitmen)' },
2019 $stmt = "WHERE priority < ? AND requestor IN (SELECT name FROM hitmen)"
2022 Note that in this example, you only get one bind parameter back, since
2023 the verbatim SQL is passed as part of the statement.
2027 Never use untrusted input as a literal SQL argument - this is a massive
2028 security risk (there is no way to check literal snippets for SQL
2029 injections and other nastyness). If you need to deal with untrusted input
2030 use literal SQL with placeholders as described next.
2032 =head3 Literal SQL with placeholders and bind values (subqueries)
2034 If the literal SQL to be inserted has placeholders and bind values,
2035 use a reference to an arrayref (yes this is a double reference --
2036 not so common, but perfectly legal Perl). For example, to find a date
2037 in Postgres you can use something like this:
2040 date_column => \[q/= date '2008-09-30' - ?::integer/, 10/]
2045 $stmt = "WHERE ( date_column = date '2008-09-30' - ?::integer )"
2048 Note that you must pass the bind values in the same format as they are returned
2049 by L</where>. That means that if you set L</bindtype> to C<columns>, you must
2050 provide the bind values in the C<< [ column_meta => value ] >> format, where
2051 C<column_meta> is an opaque scalar value; most commonly the column name, but
2052 you can use any scalar value (including references and blessed references),
2053 L<SQL::Abstract> will simply pass it through intact. So if C<bindtype> is set
2054 to C<columns> the above example will look like:
2057 date_column => \[q/= date '2008-09-30' - ?::integer/, [ dummy => 10 ]/]
2060 Literal SQL is especially useful for nesting parenthesized clauses in the
2061 main SQL query. Here is a first example :
2063 my ($sub_stmt, @sub_bind) = ("SELECT c1 FROM t1 WHERE c2 < ? AND c3 LIKE ?",
2067 bar => \["IN ($sub_stmt)" => @sub_bind],
2072 $stmt = "WHERE (foo = ? AND bar IN (SELECT c1 FROM t1
2073 WHERE c2 < ? AND c3 LIKE ?))";
2074 @bind = (1234, 100, "foo%");
2076 Other subquery operators, like for example C<"E<gt> ALL"> or C<"NOT IN">,
2077 are expressed in the same way. Of course the C<$sub_stmt> and
2078 its associated bind values can be generated through a former call
2081 my ($sub_stmt, @sub_bind)
2082 = $sql->select("t1", "c1", {c2 => {"<" => 100},
2083 c3 => {-like => "foo%"}});
2086 bar => \["> ALL ($sub_stmt)" => @sub_bind],
2089 In the examples above, the subquery was used as an operator on a column;
2090 but the same principle also applies for a clause within the main C<%where>
2091 hash, like an EXISTS subquery :
2093 my ($sub_stmt, @sub_bind)
2094 = $sql->select("t1", "*", {c1 => 1, c2 => \"> t0.c0"});
2095 my %where = ( -and => [
2097 \["EXISTS ($sub_stmt)" => @sub_bind],
2102 $stmt = "WHERE (foo = ? AND EXISTS (SELECT * FROM t1
2103 WHERE c1 = ? AND c2 > t0.c0))";
2107 Observe that the condition on C<c2> in the subquery refers to
2108 column C<t0.c0> of the main query : this is I<not> a bind
2109 value, so we have to express it through a scalar ref.
2110 Writing C<< c2 => {">" => "t0.c0"} >> would have generated
2111 C<< c2 > ? >> with bind value C<"t0.c0"> ... not exactly
2112 what we wanted here.
2114 Finally, here is an example where a subquery is used
2115 for expressing unary negation:
2117 my ($sub_stmt, @sub_bind)
2118 = $sql->where({age => [{"<" => 10}, {">" => 20}]});
2119 $sub_stmt =~ s/^ where //i; # don't want "WHERE" in the subclause
2121 lname => {like => '%son%'},
2122 \["NOT ($sub_stmt)" => @sub_bind],
2127 $stmt = "lname LIKE ? AND NOT ( age < ? OR age > ? )"
2128 @bind = ('%son%', 10, 20)
2130 =head3 Deprecated usage of Literal SQL
2132 Below are some examples of archaic use of literal SQL. It is shown only as
2133 reference for those who deal with legacy code. Each example has a much
2134 better, cleaner and safer alternative that users should opt for in new code.
2140 my %where = ( requestor => \'IS NOT NULL' )
2142 $stmt = "WHERE requestor IS NOT NULL"
2144 This used to be the way of generating NULL comparisons, before the handling
2145 of C<undef> got formalized. For new code please use the superior syntax as
2146 described in L</Tests for NULL values>.
2150 my %where = ( requestor => \'= submitter' )
2152 $stmt = "WHERE requestor = submitter"
2154 This used to be the only way to compare columns. Use the superior L</-ident>
2155 method for all new code. For example an identifier declared in such a way
2156 will be properly quoted if L</quote_char> is properly set, while the legacy
2157 form will remain as supplied.
2161 my %where = ( is_ready => \"", completed => { '>', '2012-12-21' } )
2163 $stmt = "WHERE completed > ? AND is_ready"
2164 @bind = ('2012-12-21')
2166 Using an empty string literal used to be the only way to express a boolean.
2167 For all new code please use the much more readable
2168 L<-bool|/Unary operators: bool> operator.
2174 These pages could go on for a while, since the nesting of the data
2175 structures this module can handle are pretty much unlimited (the
2176 module implements the C<WHERE> expansion as a recursive function
2177 internally). Your best bet is to "play around" with the module a
2178 little to see how the data structures behave, and choose the best
2179 format for your data based on that.
2181 And of course, all the values above will probably be replaced with
2182 variables gotten from forms or the command line. After all, if you
2183 knew everything ahead of time, you wouldn't have to worry about
2184 dynamically-generating SQL and could just hardwire it into your
2187 =head1 ORDER BY CLAUSES
2189 Some functions take an order by clause. This can either be a scalar (just a
2190 column name,) a hash of C<< { -desc => 'col' } >> or C<< { -asc => 'col' } >>,
2191 or an array of either of the two previous forms. Examples:
2193 Given | Will Generate
2194 ----------------------------------------------------------
2196 \'colA DESC' | ORDER BY colA DESC
2198 'colA' | ORDER BY colA
2200 [qw/colA colB/] | ORDER BY colA, colB
2202 {-asc => 'colA'} | ORDER BY colA ASC
2204 {-desc => 'colB'} | ORDER BY colB DESC
2206 ['colA', {-asc => 'colB'}] | ORDER BY colA, colB ASC
2208 { -asc => [qw/colA colB/] } | ORDER BY colA ASC, colB ASC
2211 { -asc => 'colA' }, | ORDER BY colA ASC, colB DESC,
2212 { -desc => [qw/colB/], | colC ASC, colD ASC
2213 { -asc => [qw/colC colD/],|
2215 ===========================================================
2219 =head1 SPECIAL OPERATORS
2221 my $sqlmaker = SQL::Abstract->new(special_ops => [
2225 my ($self, $field, $op, $arg) = @_;
2231 handler => 'method_name',
2235 A "special operator" is a SQL syntactic clause that can be
2236 applied to a field, instead of a usual binary operator.
2239 WHERE field IN (?, ?, ?)
2240 WHERE field BETWEEN ? AND ?
2241 WHERE MATCH(field) AGAINST (?, ?)
2243 Special operators IN and BETWEEN are fairly standard and therefore
2244 are builtin within C<SQL::Abstract> (as the overridable methods
2245 C<_where_field_IN> and C<_where_field_BETWEEN>). For other operators,
2246 like the MATCH .. AGAINST example above which is specific to MySQL,
2247 you can write your own operator handlers - supply a C<special_ops>
2248 argument to the C<new> method. That argument takes an arrayref of
2249 operator definitions; each operator definition is a hashref with two
2256 the regular expression to match the operator
2260 Either a coderef or a plain scalar method name. In both cases
2261 the expected return is C<< ($sql, @bind) >>.
2263 When supplied with a method name, it is simply called on the
2264 L<SQL::Abstract/> object as:
2266 $self->$method_name ($field, $op, $arg)
2270 $op is the part that matched the handler regex
2271 $field is the LHS of the operator
2274 When supplied with a coderef, it is called as:
2276 $coderef->($self, $field, $op, $arg)
2281 For example, here is an implementation
2282 of the MATCH .. AGAINST syntax for MySQL
2284 my $sqlmaker = SQL::Abstract->new(special_ops => [
2286 # special op for MySql MATCH (field) AGAINST(word1, word2, ...)
2287 {regex => qr/^match$/i,
2289 my ($self, $field, $op, $arg) = @_;
2290 $arg = [$arg] if not ref $arg;
2291 my $label = $self->_quote($field);
2292 my ($placeholder) = $self->_convert('?');
2293 my $placeholders = join ", ", (($placeholder) x @$arg);
2294 my $sql = $self->_sqlcase('match') . " ($label) "
2295 . $self->_sqlcase('against') . " ($placeholders) ";
2296 my @bind = $self->_bindtype($field, @$arg);
2297 return ($sql, @bind);
2304 =head1 UNARY OPERATORS
2306 my $sqlmaker = SQL::Abstract->new(unary_ops => [
2310 my ($self, $op, $arg) = @_;
2316 handler => 'method_name',
2320 A "unary operator" is a SQL syntactic clause that can be
2321 applied to a field - the operator goes before the field
2323 You can write your own operator handlers - supply a C<unary_ops>
2324 argument to the C<new> method. That argument takes an arrayref of
2325 operator definitions; each operator definition is a hashref with two
2332 the regular expression to match the operator
2336 Either a coderef or a plain scalar method name. In both cases
2337 the expected return is C<< $sql >>.
2339 When supplied with a method name, it is simply called on the
2340 L<SQL::Abstract/> object as:
2342 $self->$method_name ($op, $arg)
2346 $op is the part that matched the handler regex
2347 $arg is the RHS or argument of the operator
2349 When supplied with a coderef, it is called as:
2351 $coderef->($self, $op, $arg)
2359 Thanks to some benchmarking by Mark Stosberg, it turns out that
2360 this module is many orders of magnitude faster than using C<DBIx::Abstract>.
2361 I must admit this wasn't an intentional design issue, but it's a
2362 byproduct of the fact that you get to control your C<DBI> handles
2365 To maximize performance, use a code snippet like the following:
2367 # prepare a statement handle using the first row
2368 # and then reuse it for the rest of the rows
2370 for my $href (@array_of_hashrefs) {
2371 $stmt ||= $sql->insert('table', $href);
2372 $sth ||= $dbh->prepare($stmt);
2373 $sth->execute($sql->values($href));
2376 The reason this works is because the keys in your C<$href> are sorted
2377 internally by B<SQL::Abstract>. Thus, as long as your data retains
2378 the same structure, you only have to generate the SQL the first time
2379 around. On subsequent queries, simply use the C<values> function provided
2380 by this module to return your values in the correct order.
2382 However this depends on the values having the same type - if, for
2383 example, the values of a where clause may either have values
2384 (resulting in sql of the form C<column = ?> with a single bind
2385 value), or alternatively the values might be C<undef> (resulting in
2386 sql of the form C<column IS NULL> with no bind value) then the
2387 caching technique suggested will not work.
2391 If you use my C<CGI::FormBuilder> module at all, you'll hopefully
2392 really like this part (I do, at least). Building up a complex query
2393 can be as simple as the following:
2397 use CGI::FormBuilder;
2400 my $form = CGI::FormBuilder->new(...);
2401 my $sql = SQL::Abstract->new;
2403 if ($form->submitted) {
2404 my $field = $form->field;
2405 my $id = delete $field->{id};
2406 my($stmt, @bind) = $sql->update('table', $field, {id => $id});
2409 Of course, you would still have to connect using C<DBI> to run the
2410 query, but the point is that if you make your form look like your
2411 table, the actual query script can be extremely simplistic.
2413 If you're B<REALLY> lazy (I am), check out C<HTML::QuickTable> for
2414 a fast interface to returning and formatting data. I frequently
2415 use these three modules together to write complex database query
2416 apps in under 50 lines.
2422 =item * gitweb: L<http://git.shadowcat.co.uk/gitweb/gitweb.cgi?p=dbsrgits/SQL-Abstract.git>
2424 =item * git: L<git://git.shadowcat.co.uk/dbsrgits/SQL-Abstract.git>
2430 Version 1.50 was a major internal refactoring of C<SQL::Abstract>.
2431 Great care has been taken to preserve the I<published> behavior
2432 documented in previous versions in the 1.* family; however,
2433 some features that were previously undocumented, or behaved
2434 differently from the documentation, had to be changed in order
2435 to clarify the semantics. Hence, client code that was relying
2436 on some dark areas of C<SQL::Abstract> v1.*
2437 B<might behave differently> in v1.50.
2439 The main changes are :
2445 support for literal SQL through the C<< \ [$sql, bind] >> syntax.
2449 support for the { operator => \"..." } construct (to embed literal SQL)
2453 support for the { operator => \["...", @bind] } construct (to embed literal SQL with bind values)
2457 optional support for L<array datatypes|/"Inserting and Updating Arrays">
2461 defensive programming : check arguments
2465 fixed bug with global logic, which was previously implemented
2466 through global variables yielding side-effects. Prior versions would
2467 interpret C<< [ {cond1, cond2}, [cond3, cond4] ] >>
2468 as C<< "(cond1 AND cond2) OR (cond3 AND cond4)" >>.
2469 Now this is interpreted
2470 as C<< "(cond1 AND cond2) OR (cond3 OR cond4)" >>.
2475 fixed semantics of _bindtype on array args
2479 dropped the C<_anoncopy> of the %where tree. No longer necessary,
2480 we just avoid shifting arrays within that tree.
2484 dropped the C<_modlogic> function
2488 =head1 ACKNOWLEDGEMENTS
2490 There are a number of individuals that have really helped out with
2491 this module. Unfortunately, most of them submitted bugs via CPAN
2492 so I have no idea who they are! But the people I do know are:
2494 Ash Berlin (order_by hash term support)
2495 Matt Trout (DBIx::Class support)
2496 Mark Stosberg (benchmarking)
2497 Chas Owens (initial "IN" operator support)
2498 Philip Collins (per-field SQL functions)
2499 Eric Kolve (hashref "AND" support)
2500 Mike Fragassi (enhancements to "BETWEEN" and "LIKE")
2501 Dan Kubb (support for "quote_char" and "name_sep")
2502 Guillermo Roditi (patch to cleanup "IN" and "BETWEEN", fix and tests for _order_by)
2503 Laurent Dami (internal refactoring, extensible list of special operators, literal SQL)
2504 Norbert Buchmuller (support for literal SQL in hashpair, misc. fixes & tests)
2505 Peter Rabbitson (rewrite of SQLA::Test, misc. fixes & tests)
2506 Oliver Charles (support for "RETURNING" after "INSERT")
2512 L<DBIx::Class>, L<DBIx::Abstract>, L<CGI::FormBuilder>, L<HTML::QuickTable>.
2516 Copyright (c) 2001-2007 Nathan Wiger <nwiger@cpan.org>. All Rights Reserved.
2518 This module is actively maintained by Matt Trout <mst@shadowcatsystems.co.uk>
2520 For support, your best bet is to try the C<DBIx::Class> users mailing list.
2521 While not an official support venue, C<DBIx::Class> makes heavy use of
2522 C<SQL::Abstract>, and as such list members there are very familiar with
2523 how to create queries.
2527 This module is free software; you may copy this under the same
2528 terms as perl itself (either the GNU General Public License or
2529 the Artistic License)