1 package SQL::Abstract; # see doc at end of file
3 # LDNOTE : this code is heavy refactoring from original SQLA.
4 # Several design decisions will need discussion during
5 # the test / diffusion / acceptance phase; those are marked with flag
6 # 'LDNOTE' (note by laurent.dami AT free.fr)
10 use warnings FATAL => 'all';
13 use Data::Query::Constants qw(
14 DQ_IDENTIFIER DQ_OPERATOR DQ_VALUE DQ_LITERAL DQ_JOIN DQ_SELECT DQ_ORDER
16 use Data::Query::ExprHelpers qw(perl_scalar_value);
18 #======================================================================
20 #======================================================================
22 our $VERSION = '1.72';
24 # This would confuse some packagers
25 $VERSION = eval $VERSION if $VERSION =~ /_/; # numify for warning-free dev releases
29 # special operators (-in, -between). May be extended/overridden by user.
30 # See section WHERE: BUILTIN SPECIAL OPERATORS below for implementation
31 my @BUILTIN_SPECIAL_OPS = (
32 {regex => qr/^ (?: not \s )? between $/ix, handler => '_where_field_BETWEEN'},
33 {regex => qr/^ (?: not \s )? in $/ix, handler => '_where_field_IN'},
34 {regex => qr/^ ident $/ix, handler => '_where_op_IDENT'},
35 {regex => qr/^ value $/ix, handler => '_where_op_VALUE'},
38 # unaryish operators - key maps to handler
39 my @BUILTIN_UNARY_OPS = (
40 # the digits are backcompat stuff
41 { regex => qr/^ and (?: [_\s]? \d+ )? $/xi, handler => '_where_op_ANDOR' },
42 { regex => qr/^ or (?: [_\s]? \d+ )? $/xi, handler => '_where_op_ANDOR' },
43 { regex => qr/^ nest (?: [_\s]? \d+ )? $/xi, handler => '_where_op_NEST' },
44 { regex => qr/^ (?: not \s )? bool $/xi, handler => '_where_op_BOOL' },
45 { regex => qr/^ ident $/xi, handler => '_where_op_IDENT' },
46 { regex => qr/^ value $/ix, handler => '_where_op_VALUE' },
49 #======================================================================
50 # DEBUGGING AND ERROR REPORTING
51 #======================================================================
54 return unless $_[0]->{debug}; shift; # a little faster
55 my $func = (caller(1))[3];
56 warn "[$func] ", @_, "\n";
60 my($func) = (caller(1))[3];
61 Carp::carp "[$func] Warning: ", @_;
65 my($func) = (caller(1))[3];
66 Carp::croak "[$func] Fatal: ", @_;
70 #======================================================================
72 #======================================================================
76 my $class = ref($self) || $self;
77 my %opt = (ref $_[0] eq 'HASH') ? %{$_[0]} : @_;
79 # choose our case by keeping an option around
80 delete $opt{case} if $opt{case} && $opt{case} ne 'lower';
82 # default logic for interpreting arrayrefs
83 $opt{logic} = $opt{logic} ? uc $opt{logic} : 'OR';
85 # how to return bind vars
86 # LDNOTE: changed nwiger code : why this 'delete' ??
87 # $opt{bindtype} ||= delete($opt{bind_type}) || 'normal';
88 $opt{bindtype} ||= 'normal';
90 # default comparison is "=", but can be overridden
93 # try to recognize which are the 'equality' and 'unequality' ops
94 # (temporary quickfix, should go through a more seasoned API)
95 $opt{equality_op} = qr/^(\Q$opt{cmp}\E|is|(is\s+)?like)$/i;
96 $opt{inequality_op} = qr/^(!=|<>|(is\s+)?not(\s+like)?)$/i;
99 $opt{sqltrue} ||= '1=1';
100 $opt{sqlfalse} ||= '0=1';
103 $opt{special_ops} ||= [];
104 # regexes are applied in order, thus push after user-defines
105 push @{$opt{special_ops}}, @BUILTIN_SPECIAL_OPS;
108 $opt{unary_ops} ||= [];
109 push @{$opt{unary_ops}}, @BUILTIN_UNARY_OPS;
111 # rudimentary saniy-check for user supplied bits treated as functions/operators
112 # If a purported function matches this regular expression, an exception is thrown.
113 # Literal SQL is *NOT* subject to this check, only functions (and column names
114 # when quoting is not in effect)
117 # need to guard against ()'s in column names too, but this will break tons of
118 # hacks... ideas anyone?
119 $opt{injection_guard} ||= qr/
125 $opt{name_sep} ||= '.';
127 $opt{renderer} ||= do {
128 require Data::Query::Renderer::SQL::Naive;
129 my ($always, $chars);
130 for ($opt{quote_char}) {
131 $chars = defined() ? (ref() ? $_ : [$_]) : ['',''];
134 Data::Query::Renderer::SQL::Naive->new({
135 quote_chars => $chars, always_quote => $always,
139 return bless \%opt, $class;
143 my ($self, $dq) = @_;
144 my ($sql, @bind) = @{$self->{renderer}->render($dq)};
146 ($self->{bindtype} eq 'normal'
147 ? ($sql, map $_->{value}, @bind)
148 : ($sql, map [ $_->{value_meta}, $_->{value} ], @bind)
154 my ($self, $literal) = @_;
156 ($literal, @bind) = @$literal if ref($literal) eq 'ARRAY';
161 (@bind ? (values => [ $self->_bind_to_dq(@bind) ]) : ()),
165 sub _literal_with_prepend_to_dq {
166 my ($self, $prepend, $literal) = @_;
168 $self->_literal_to_dq(
169 [ join(' ', $prepend, $literal->[0]), @{$literal}[1..$#$literal] ]
172 $self->_literal_to_dq(
173 join(' ', $prepend, $literal)
179 my ($self, @bind) = @_;
181 $self->{bindtype} eq 'normal'
182 ? map perl_scalar_value($_), @bind
184 $self->_assert_bindval_matches_bindtype(@bind);
185 map perl_scalar_value(reverse @$_), @bind
190 my ($self, $value) = @_;
191 perl_scalar_value($value, our $Cur_Col_Meta);
195 my ($self, $ident) = @_;
197 type => DQ_IDENTIFIER,
198 elements => [ split /\Q$self->{name_sep}/, $ident ],
202 sub _assert_pass_injection_guard {
203 if ($_[1] =~ $_[0]->{injection_guard}) {
204 my $class = ref $_[0];
205 puke "Possible SQL injection attempt '$_[1]'. If this is indeed a part of the "
206 . "desired SQL use literal SQL ( \'...' or \[ '...' ] ) or supply your own "
207 . "{injection_guard} attribute to ${class}->new()"
212 #======================================================================
214 #======================================================================
218 my $table = $self->_table(shift);
219 my $data = shift || return;
222 my $method = $self->_METHOD_FOR_refkind("_insert", $data);
223 my ($sql, @bind) = $self->$method($data);
224 $sql = join " ", $self->_sqlcase('insert into'), $table, $sql;
226 if ($options->{returning}) {
227 my ($s, @b) = $self->_insert_returning ($options);
232 return wantarray ? ($sql, @bind) : $sql;
235 sub _insert_returning {
236 my ($self, $options) = @_;
238 my $f = $options->{returning};
240 my $fieldlist = $self->_SWITCH_refkind($f, {
241 ARRAYREF => sub {join ', ', map { $self->_quote($_) } @$f;},
242 SCALAR => sub {$self->_quote($f)},
243 SCALARREF => sub {$$f},
245 return $self->_sqlcase(' returning ') . $fieldlist;
248 sub _insert_HASHREF { # explicit list of fields and then values
249 my ($self, $data) = @_;
251 my @fields = sort keys %$data;
253 my ($sql, @bind) = $self->_insert_values($data);
256 $_ = $self->_quote($_) foreach @fields;
257 $sql = "( ".join(", ", @fields).") ".$sql;
259 return ($sql, @bind);
262 sub _insert_ARRAYREF { # just generate values(?,?) part (no list of fields)
263 my ($self, $data) = @_;
265 # no names (arrayref) so can't generate bindtype
266 $self->{bindtype} ne 'columns'
267 or belch "can't do 'columns' bindtype when called with arrayref";
269 # fold the list of values into a hash of column name - value pairs
270 # (where the column names are artificially generated, and their
271 # lexicographical ordering keep the ordering of the original list)
272 my $i = "a"; # incremented values will be in lexicographical order
273 my $data_in_hash = { map { ($i++ => $_) } @$data };
275 return $self->_insert_values($data_in_hash);
278 sub _insert_ARRAYREFREF { # literal SQL with bind
279 my ($self, $data) = @_;
281 my ($sql, @bind) = @${$data};
282 $self->_assert_bindval_matches_bindtype(@bind);
284 return ($sql, @bind);
288 sub _insert_SCALARREF { # literal SQL without bind
289 my ($self, $data) = @_;
295 my ($self, $data) = @_;
297 my (@values, @all_bind);
298 foreach my $column (sort keys %$data) {
299 my $v = $data->{$column};
301 $self->_SWITCH_refkind($v, {
304 if ($self->{array_datatypes}) { # if array datatype are activated
306 push @all_bind, $self->_bindtype($column, $v);
308 else { # else literal SQL with bind
309 my ($sql, @bind) = @$v;
310 $self->_assert_bindval_matches_bindtype(@bind);
312 push @all_bind, @bind;
316 ARRAYREFREF => sub { # literal SQL with bind
317 my ($sql, @bind) = @${$v};
318 $self->_assert_bindval_matches_bindtype(@bind);
320 push @all_bind, @bind;
323 # THINK : anything useful to do with a HASHREF ?
324 HASHREF => sub { # (nothing, but old SQLA passed it through)
325 #TODO in SQLA >= 2.0 it will die instead
326 belch "HASH ref as bind value in insert is not supported";
328 push @all_bind, $self->_bindtype($column, $v);
331 SCALARREF => sub { # literal SQL without bind
335 SCALAR_or_UNDEF => sub {
337 push @all_bind, $self->_bindtype($column, $v);
344 my $sql = $self->_sqlcase('values')." ( ".join(", ", @values)." )";
345 return ($sql, @all_bind);
350 #======================================================================
352 #======================================================================
357 my $table = $self->_table(shift);
358 my $data = shift || return;
361 # first build the 'SET' part of the sql statement
362 my (@set, @all_bind);
363 puke "Unsupported data type specified to \$sql->update"
364 unless ref $data eq 'HASH';
366 for my $k (sort keys %$data) {
369 my $label = $self->_quote($k);
371 $self->_SWITCH_refkind($v, {
373 if ($self->{array_datatypes}) { # array datatype
374 push @set, "$label = ?";
375 push @all_bind, $self->_bindtype($k, $v);
377 else { # literal SQL with bind
378 my ($sql, @bind) = @$v;
379 $self->_assert_bindval_matches_bindtype(@bind);
380 push @set, "$label = $sql";
381 push @all_bind, @bind;
384 ARRAYREFREF => sub { # literal SQL with bind
385 my ($sql, @bind) = @${$v};
386 $self->_assert_bindval_matches_bindtype(@bind);
387 push @set, "$label = $sql";
388 push @all_bind, @bind;
390 SCALARREF => sub { # literal SQL without bind
391 push @set, "$label = $$v";
394 my ($op, $arg, @rest) = %$v;
396 puke 'Operator calls in update must be in the form { -op => $arg }'
397 if (@rest or not $op =~ /^\-(.+)/);
399 local $self->{_nested_func_lhs} = $k;
400 my ($sql, @bind) = $self->_where_unary_op ($1, $arg);
402 push @set, "$label = $sql";
403 push @all_bind, @bind;
405 SCALAR_or_UNDEF => sub {
406 push @set, "$label = ?";
407 push @all_bind, $self->_bindtype($k, $v);
413 my $sql = $self->_sqlcase('update') . " $table " . $self->_sqlcase('set ')
417 my($where_sql, @where_bind) = $self->where($where);
419 push @all_bind, @where_bind;
422 return wantarray ? ($sql, @all_bind) : $sql;
428 #======================================================================
430 #======================================================================
436 my $fields = shift || '*';
440 my($where_sql, @bind) = $self->where($where, $order);
442 my $sql = $self->_render_dq({
445 map $self->_ident_to_dq($_),
446 ref($fields) eq 'ARRAY' ? @$fields : $fields
448 from => $self->_table_to_dq($table),
453 return wantarray ? ($sql, @bind) : $sql;
456 #======================================================================
458 #======================================================================
463 my $table = $self->_table(shift);
467 my($where_sql, @bind) = $self->where($where);
468 my $sql = $self->_sqlcase('delete from') . " $table" . $where_sql;
470 return wantarray ? ($sql, @bind) : $sql;
474 #======================================================================
476 #======================================================================
480 # Finally, a separate routine just to handle WHERE clauses
482 my ($self, $where, $order) = @_;
485 my ($sql, @bind) = $self->_recurse_where($where);
486 $sql = $sql ? $self->_sqlcase(' where ') . "( $sql )" : '';
490 $sql .= $self->_order_by($order);
493 return wantarray ? ($sql, @bind) : $sql;
498 my ($self, $where, $logic) = @_;
500 return $self->_render_dq($self->_where_to_dq($where, $logic));
504 my ($self, $where, $logic) = @_;
506 if (ref($where) eq 'ARRAY') {
507 return $self->_where_to_dq_ARRAYREF($where, $logic);
508 } elsif (ref($where) eq 'HASH') {
509 return $self->_where_to_dq_HASHREF($where, $logic);
511 ref($where) eq 'SCALAR'
512 or (ref($where) eq 'REF' and ref($$where) eq 'ARRAY')
514 return $self->_literal_to_dq($$where);
515 } elsif (!ref($where) or Scalar::Util::blessed($where)) {
516 return $self->_value_to_dq($where);
518 die "Can't handle $where";
521 sub _where_to_dq_ARRAYREF {
522 my ($self, $where, $logic) = @_;
524 $logic = uc($logic || 'OR');
525 $logic eq 'AND' or $logic eq 'OR' or puke "unknown logic: $logic";
527 return unless @$where;
529 my ($first, @rest) = @$where;
531 return $self->_where_to_dq($first) unless @rest;
535 $self->_where_hashpair_to_dq($first => shift(@rest));
537 $self->_where_to_dq($first);
541 return $self->_where_to_dq_ARRAYREF(\@rest, $logic) unless $first_dq;
545 operator => { 'SQL.Naive' => $logic },
546 args => [ $first_dq, $self->_where_to_dq_ARRAYREF(\@rest, $logic) ]
550 sub _where_to_dq_HASHREF {
551 my ($self, $where, $logic) = @_;
553 $logic = uc($logic || 'AND');
556 $self->_where_hashpair_to_dq($_ => $where->{$_})
559 return $dq[0] unless @dq > 1;
561 my $final = pop(@dq);
563 foreach my $dq (reverse @dq) {
566 operator => { 'SQL.Naive' => $logic },
567 args => [ $dq, $final ]
574 sub _where_to_dq_SCALAR {
575 shift->_value_to_dq(@_);
578 sub _where_op_IDENT {
580 my ($op, $rhs) = splice @_, -2;
582 puke "-$op takes a single scalar argument (a quotable identifier)";
585 # in case we are called as a top level special op (no '=')
588 $_ = $self->_convert($self->_quote($_)) for ($lhs, $rhs);
596 sub _where_op_VALUE {
598 my ($op, $rhs) = splice @_, -2;
600 # in case we are called as a top level special op (no '=')
605 ($lhs || $self->{_nested_func_lhs}),
612 $self->_convert($self->_quote($lhs)) . ' = ' . $self->_convert('?'),
616 $self->_convert('?'),
622 sub _where_hashpair_to_dq {
623 my ($self, $k, $v) = @_;
627 if ($op eq 'AND' or $op eq 'OR') {
628 return $self->_where_to_dq($v, $op);
629 } elsif ($op eq 'NEST') {
630 return $self->_where_to_dq($v);
631 } elsif ($op eq 'NOT') {
634 operator => { 'SQL.Naive' => 'NOT' },
635 args => [ $self->_where_to_dq($v) ]
637 } elsif ($op eq 'BOOL') {
638 return ref($v) ? $self->_where_to_dq($v) : $self->_ident_to_dq($v);
639 } elsif ($op eq 'NOT_BOOL') {
642 operator => { 'SQL.Naive' => 'NOT' },
643 args => [ ref($v) ? $self->_where_to_dq($v) : $self->_ident_to_dq($v) ]
647 if (ref($v) eq 'HASH' and keys(%$v) == 1 and (keys %$v)[0] =~ /-(.*)/) {
648 my ($inner) = values %$v;
651 operator => { 'SQL.Naive' => uc($1) },
653 (map $self->_where_to_dq($_),
654 (ref($inner) eq 'ARRAY' ? @$inner : $inner))
658 (map $self->_where_to_dq($_), (ref($v) eq 'ARRAY' ? @$v : $v))
663 operator => { 'SQL.Naive' => 'apply' },
665 $self->_ident_to_dq($op), @args
670 local our $Cur_Col_Meta = $k;
671 if (ref($v) eq 'ARRAY') {
673 return $self->_literal_to_dq($self->{sqlfalse});
674 } elsif (defined($v->[0]) && $v->[0] =~ /-(and|or)/i) {
675 return $self->_where_to_dq_ARRAYREF([
676 map +{ $k => $_ }, @{$v}[1..$#$v]
679 return $self->_where_to_dq_ARRAYREF([
680 map +{ $k => $_ }, @$v
682 } elsif (ref($v) eq 'SCALAR' or (ref($v) eq 'REF' and ref($$v) eq 'ARRAY')) {
683 return $self->_literal_with_prepend_to_dq($k, $$v);
685 my ($op, $rhs) = do {
686 if (ref($v) eq 'HASH') {
688 return $self->_where_to_dq_ARRAYREF([
689 map +{ $k => { $_ => $v->{$_} } }, keys %$v
692 (uc((keys %$v)[0]), (values %$v)[0]);
697 s/^-//, s/_/ /g for $op;
698 if ($op eq 'BETWEEN' or $op eq 'IN' or $op eq 'NOT IN' or $op eq 'NOT BETWEEN') {
699 if (ref($rhs) ne 'ARRAY') {
701 # have to add parens if none present because -in => \"SELECT ..."
702 # got documented. mst hates everything.
703 if (ref($rhs) eq 'SCALAR') {
705 $x = "($x)" unless $x =~ /^\s*\(/;
708 my ($x, @rest) = @{$$rhs};
709 $x = "($x)" unless $x =~ /^\s*\(/;
710 $rhs = \[ $x, @rest ];
713 return $self->_literal_with_prepend_to_dq("$k $op", $$rhs);
715 return $self->_literal_to_dq($self->{sqlfalse}) unless @$rhs;
718 operator => { 'SQL.Naive' => $op },
719 args => [ $self->_ident_to_dq($k), map $self->_where_to_dq($_), @$rhs ]
721 } elsif ($op =~ s/^NOT (?!LIKE)//) {
722 return $self->_where_hashpair_to_dq(-not => { $k => { $op => $rhs } });
723 } elsif (!defined($rhs)) {
725 if ($op eq '=' or $op eq 'LIKE') {
727 } elsif ($op eq '!=') {
730 die "Can't do undef -> NULL transform for operator ${op}";
735 operator => { 'SQL.Naive' => $null_op },
736 args => [ $self->_ident_to_dq($k) ]
739 if (ref($rhs) eq 'ARRAY') {
741 return $self->_literal_to_dq(
742 $op eq '!=' ? $self->{sqltrue} : $self->{sqlfalse}
744 } elsif (defined($rhs->[0]) and $rhs->[0] =~ /-(and|or)/i) {
745 return $self->_where_to_dq_ARRAYREF([
746 map +{ $k => { $op => $_ } }, @{$rhs}[1..$#$rhs]
749 return $self->_where_to_dq_ARRAYREF([
750 map +{ $k => { $op => $_ } }, @$rhs
755 operator => { 'SQL.Naive' => $op },
756 args => [ $self->_ident_to_dq($k), $self->_where_to_dq($rhs) ]
761 #======================================================================
763 #======================================================================
766 my ($self, $arg) = @_;
767 if (my $dq = $self->_order_by_to_dq($arg)) {
768 # SQLA generates ' ORDER BY foo'. The hilarity.
770 ? do { my @r = $self->_render_dq($dq); $r[0] = ' '.$r[0]; @r }
771 : ' '.$self->_render_dq($dq);
777 sub _order_by_to_dq {
778 my ($self, $arg, $dir) = @_;
784 ($dir ? (direction => $dir) : ()),
788 $dq->{by} = $self->_ident_to_dq($arg);
789 } elsif (ref($arg) eq 'ARRAY') {
791 local our $Order_Inner unless our $Order_Recursing;
792 local $Order_Recursing = 1;
794 foreach my $member (@$arg) {
796 my $next = $self->_order_by_to_dq($member, $dir);
798 $inner->{from} = $next if $inner;
799 $inner = $Order_Inner || $next;
801 $Order_Inner = $inner;
803 } elsif (ref($arg) eq 'REF' and ref($$arg) eq 'ARRAY') {
804 $dq->{by} = $self->_literal_to_dq($$arg);
805 } elsif (ref($arg) eq 'SCALAR') {
806 $dq->{by} = $self->_literal_to_dq($$arg);
807 } elsif (ref($arg) eq 'HASH') {
808 my ($key, $val, @rest) = %$arg;
812 if (@rest or not $key =~ /^-(desc|asc)/i) {
813 puke "hash passed to _order_by must have exactly one key (-desc or -asc)";
816 return $self->_order_by_to_dq($val, $dir);
818 die "Can't handle $arg in _order_by_to_dq";
823 #======================================================================
824 # DATASOURCE (FOR NOW, JUST PLAIN TABLE OR LIST OF TABLES)
825 #======================================================================
828 my ($self, $from) = @_;
829 $self->_render_dq($self->_table_to_dq($from));
833 my ($self, $from) = @_;
834 $self->_SWITCH_refkind($from, {
836 die "Empty FROM list" unless my @f = @$from;
837 my $dq = $self->_ident_to_dq(shift @f);
838 while (my $x = shift @f) {
841 join => [ $dq, $self->_ident_to_dq($x) ]
846 SCALAR => sub { $self->_ident_to_dq($from) },
858 #======================================================================
860 #======================================================================
862 # highly optimized, as it's called way too often
864 # my ($self, $label) = @_;
866 return '' unless defined $_[1];
867 return ${$_[1]} if ref($_[1]) eq 'SCALAR';
869 unless ($_[0]->{quote_char}) {
870 $_[0]->_assert_pass_injection_guard($_[1]);
874 my $qref = ref $_[0]->{quote_char};
877 ($l, $r) = ( $_[0]->{quote_char}, $_[0]->{quote_char} );
879 elsif ($qref eq 'ARRAY') {
880 ($l, $r) = @{$_[0]->{quote_char}};
883 puke "Unsupported quote_char format: $_[0]->{quote_char}";
886 # parts containing * are naturally unquoted
887 return join( $_[0]->{name_sep}||'', map
888 { $_ eq '*' ? $_ : $l . $_ . $r }
889 ( $_[0]->{name_sep} ? split (/\Q$_[0]->{name_sep}\E/, $_[1] ) : $_[1] )
894 # Conversion, if applicable
896 #my ($self, $arg) = @_;
898 # LDNOTE : modified the previous implementation below because
899 # it was not consistent : the first "return" is always an array,
900 # the second "return" is context-dependent. Anyway, _convert
901 # seems always used with just a single argument, so make it a
903 # return @_ unless $self->{convert};
904 # my $conv = $self->_sqlcase($self->{convert});
905 # my @ret = map { $conv.'('.$_.')' } @_;
906 # return wantarray ? @ret : $ret[0];
907 if ($_[0]->{convert}) {
908 return $_[0]->_sqlcase($_[0]->{convert}) .'(' . $_[1] . ')';
915 #my ($self, $col, @vals) = @_;
917 #LDNOTE : changed original implementation below because it did not make
918 # sense when bindtype eq 'columns' and @vals > 1.
919 # return $self->{bindtype} eq 'columns' ? [ $col, @vals ] : @vals;
921 # called often - tighten code
922 return $_[0]->{bindtype} eq 'columns'
923 ? map {[$_[1], $_]} @_[2 .. $#_]
928 # Dies if any element of @bind is not in [colname => value] format
929 # if bindtype is 'columns'.
930 sub _assert_bindval_matches_bindtype {
931 # my ($self, @bind) = @_;
933 if ($self->{bindtype} eq 'columns') {
935 if (!defined $_ || ref($_) ne 'ARRAY' || @$_ != 2) {
936 puke "bindtype 'columns' selected, you need to pass: [column_name => bind_value]"
942 sub _join_sql_clauses {
943 my ($self, $logic, $clauses_aref, $bind_aref) = @_;
945 if (@$clauses_aref > 1) {
946 my $join = " " . $self->_sqlcase($logic) . " ";
947 my $sql = '( ' . join($join, @$clauses_aref) . ' )';
948 return ($sql, @$bind_aref);
950 elsif (@$clauses_aref) {
951 return ($clauses_aref->[0], @$bind_aref); # no parentheses
954 return (); # if no SQL, ignore @$bind_aref
959 # Fix SQL case, if so requested
961 # LDNOTE: if $self->{case} is true, then it contains 'lower', so we
962 # don't touch the argument ... crooked logic, but let's not change it!
963 return $_[0]->{case} ? $_[1] : uc($_[1]);
967 #======================================================================
968 # DISPATCHING FROM REFKIND
969 #======================================================================
972 my ($self, $data) = @_;
974 return 'UNDEF' unless defined $data;
976 # blessed objects are treated like scalars
977 my $ref = (Scalar::Util::blessed $data) ? '' : ref $data;
979 return 'SCALAR' unless $ref;
982 while ($ref eq 'REF') {
984 $ref = (Scalar::Util::blessed $data) ? '' : ref $data;
988 return ($ref||'SCALAR') . ('REF' x $n_steps);
992 my ($self, $data) = @_;
993 my @try = ($self->_refkind($data));
994 push @try, 'SCALAR_or_UNDEF' if $try[0] eq 'SCALAR' || $try[0] eq 'UNDEF';
995 push @try, 'FALLBACK';
999 sub _METHOD_FOR_refkind {
1000 my ($self, $meth_prefix, $data) = @_;
1003 for (@{$self->_try_refkind($data)}) {
1004 $method = $self->can($meth_prefix."_".$_)
1008 return $method || puke "cannot dispatch on '$meth_prefix' for ".$self->_refkind($data);
1012 sub _SWITCH_refkind {
1013 my ($self, $data, $dispatch_table) = @_;
1016 for (@{$self->_try_refkind($data)}) {
1017 $coderef = $dispatch_table->{$_}
1021 puke "no dispatch entry for ".$self->_refkind($data)
1030 #======================================================================
1031 # VALUES, GENERATE, AUTOLOAD
1032 #======================================================================
1034 # LDNOTE: original code from nwiger, didn't touch code in that section
1035 # I feel the AUTOLOAD stuff should not be the default, it should
1036 # only be activated on explicit demand by user.
1040 my $data = shift || return;
1041 puke "Argument to ", __PACKAGE__, "->values must be a \\%hash"
1042 unless ref $data eq 'HASH';
1045 foreach my $k ( sort keys %$data ) {
1046 my $v = $data->{$k};
1047 $self->_SWITCH_refkind($v, {
1049 if ($self->{array_datatypes}) { # array datatype
1050 push @all_bind, $self->_bindtype($k, $v);
1052 else { # literal SQL with bind
1053 my ($sql, @bind) = @$v;
1054 $self->_assert_bindval_matches_bindtype(@bind);
1055 push @all_bind, @bind;
1058 ARRAYREFREF => sub { # literal SQL with bind
1059 my ($sql, @bind) = @${$v};
1060 $self->_assert_bindval_matches_bindtype(@bind);
1061 push @all_bind, @bind;
1063 SCALARREF => sub { # literal SQL without bind
1065 SCALAR_or_UNDEF => sub {
1066 push @all_bind, $self->_bindtype($k, $v);
1077 my(@sql, @sqlq, @sqlv);
1081 if ($ref eq 'HASH') {
1082 for my $k (sort keys %$_) {
1085 my $label = $self->_quote($k);
1086 if ($r eq 'ARRAY') {
1087 # literal SQL with bind
1088 my ($sql, @bind) = @$v;
1089 $self->_assert_bindval_matches_bindtype(@bind);
1090 push @sqlq, "$label = $sql";
1092 } elsif ($r eq 'SCALAR') {
1093 # literal SQL without bind
1094 push @sqlq, "$label = $$v";
1096 push @sqlq, "$label = ?";
1097 push @sqlv, $self->_bindtype($k, $v);
1100 push @sql, $self->_sqlcase('set'), join ', ', @sqlq;
1101 } elsif ($ref eq 'ARRAY') {
1102 # unlike insert(), assume these are ONLY the column names, i.e. for SQL
1105 if ($r eq 'ARRAY') { # literal SQL with bind
1106 my ($sql, @bind) = @$v;
1107 $self->_assert_bindval_matches_bindtype(@bind);
1110 } elsif ($r eq 'SCALAR') { # literal SQL without bind
1111 # embedded literal SQL
1118 push @sql, '(' . join(', ', @sqlq) . ')';
1119 } elsif ($ref eq 'SCALAR') {
1123 # strings get case twiddled
1124 push @sql, $self->_sqlcase($_);
1128 my $sql = join ' ', @sql;
1130 # this is pretty tricky
1131 # if ask for an array, return ($stmt, @bind)
1132 # otherwise, s/?/shift @sqlv/ to put it inline
1134 return ($sql, @sqlv);
1136 1 while $sql =~ s/\?/my $d = shift(@sqlv);
1137 ref $d ? $d->[1] : $d/e;
1146 # This allows us to check for a local, then _form, attr
1148 my($name) = $AUTOLOAD =~ /.*::(.+)/;
1149 return $self->generate($name, @_);
1160 SQL::Abstract - Generate SQL from Perl data structures
1166 my $sql = SQL::Abstract->new;
1168 my($stmt, @bind) = $sql->select($table, \@fields, \%where, \@order);
1170 my($stmt, @bind) = $sql->insert($table, \%fieldvals || \@values);
1172 my($stmt, @bind) = $sql->update($table, \%fieldvals, \%where);
1174 my($stmt, @bind) = $sql->delete($table, \%where);
1176 # Then, use these in your DBI statements
1177 my $sth = $dbh->prepare($stmt);
1178 $sth->execute(@bind);
1180 # Just generate the WHERE clause
1181 my($stmt, @bind) = $sql->where(\%where, \@order);
1183 # Return values in the same order, for hashed queries
1184 # See PERFORMANCE section for more details
1185 my @bind = $sql->values(\%fieldvals);
1189 This module was inspired by the excellent L<DBIx::Abstract>.
1190 However, in using that module I found that what I really wanted
1191 to do was generate SQL, but still retain complete control over my
1192 statement handles and use the DBI interface. So, I set out to
1193 create an abstract SQL generation module.
1195 While based on the concepts used by L<DBIx::Abstract>, there are
1196 several important differences, especially when it comes to WHERE
1197 clauses. I have modified the concepts used to make the SQL easier
1198 to generate from Perl data structures and, IMO, more intuitive.
1199 The underlying idea is for this module to do what you mean, based
1200 on the data structures you provide it. The big advantage is that
1201 you don't have to modify your code every time your data changes,
1202 as this module figures it out.
1204 To begin with, an SQL INSERT is as easy as just specifying a hash
1205 of C<key=value> pairs:
1208 name => 'Jimbo Bobson',
1209 phone => '123-456-7890',
1210 address => '42 Sister Lane',
1211 city => 'St. Louis',
1212 state => 'Louisiana',
1215 The SQL can then be generated with this:
1217 my($stmt, @bind) = $sql->insert('people', \%data);
1219 Which would give you something like this:
1221 $stmt = "INSERT INTO people
1222 (address, city, name, phone, state)
1223 VALUES (?, ?, ?, ?, ?)";
1224 @bind = ('42 Sister Lane', 'St. Louis', 'Jimbo Bobson',
1225 '123-456-7890', 'Louisiana');
1227 These are then used directly in your DBI code:
1229 my $sth = $dbh->prepare($stmt);
1230 $sth->execute(@bind);
1232 =head2 Inserting and Updating Arrays
1234 If your database has array types (like for example Postgres),
1235 activate the special option C<< array_datatypes => 1 >>
1236 when creating the C<SQL::Abstract> object.
1237 Then you may use an arrayref to insert and update database array types:
1239 my $sql = SQL::Abstract->new(array_datatypes => 1);
1241 planets => [qw/Mercury Venus Earth Mars/]
1244 my($stmt, @bind) = $sql->insert('solar_system', \%data);
1248 $stmt = "INSERT INTO solar_system (planets) VALUES (?)"
1250 @bind = (['Mercury', 'Venus', 'Earth', 'Mars']);
1253 =head2 Inserting and Updating SQL
1255 In order to apply SQL functions to elements of your C<%data> you may
1256 specify a reference to an arrayref for the given hash value. For example,
1257 if you need to execute the Oracle C<to_date> function on a value, you can
1258 say something like this:
1262 date_entered => \["to_date(?,'MM/DD/YYYY')", "03/02/2003"],
1265 The first value in the array is the actual SQL. Any other values are
1266 optional and would be included in the bind values array. This gives
1269 my($stmt, @bind) = $sql->insert('people', \%data);
1271 $stmt = "INSERT INTO people (name, date_entered)
1272 VALUES (?, to_date(?,'MM/DD/YYYY'))";
1273 @bind = ('Bill', '03/02/2003');
1275 An UPDATE is just as easy, all you change is the name of the function:
1277 my($stmt, @bind) = $sql->update('people', \%data);
1279 Notice that your C<%data> isn't touched; the module will generate
1280 the appropriately quirky SQL for you automatically. Usually you'll
1281 want to specify a WHERE clause for your UPDATE, though, which is
1282 where handling C<%where> hashes comes in handy...
1284 =head2 Complex where statements
1286 This module can generate pretty complicated WHERE statements
1287 easily. For example, simple C<key=value> pairs are taken to mean
1288 equality, and if you want to see if a field is within a set
1289 of values, you can use an arrayref. Let's say we wanted to
1290 SELECT some data based on this criteria:
1293 requestor => 'inna',
1294 worker => ['nwiger', 'rcwe', 'sfz'],
1295 status => { '!=', 'completed' }
1298 my($stmt, @bind) = $sql->select('tickets', '*', \%where);
1300 The above would give you something like this:
1302 $stmt = "SELECT * FROM tickets WHERE
1303 ( requestor = ? ) AND ( status != ? )
1304 AND ( worker = ? OR worker = ? OR worker = ? )";
1305 @bind = ('inna', 'completed', 'nwiger', 'rcwe', 'sfz');
1307 Which you could then use in DBI code like so:
1309 my $sth = $dbh->prepare($stmt);
1310 $sth->execute(@bind);
1316 The functions are simple. There's one for each major SQL operation,
1317 and a constructor you use first. The arguments are specified in a
1318 similar order to each function (table, then fields, then a where
1319 clause) to try and simplify things.
1324 =head2 new(option => 'value')
1326 The C<new()> function takes a list of options and values, and returns
1327 a new B<SQL::Abstract> object which can then be used to generate SQL
1328 through the methods below. The options accepted are:
1334 If set to 'lower', then SQL will be generated in all lowercase. By
1335 default SQL is generated in "textbook" case meaning something like:
1337 SELECT a_field FROM a_table WHERE some_field LIKE '%someval%'
1339 Any setting other than 'lower' is ignored.
1343 This determines what the default comparison operator is. By default
1344 it is C<=>, meaning that a hash like this:
1346 %where = (name => 'nwiger', email => 'nate@wiger.org');
1348 Will generate SQL like this:
1350 WHERE name = 'nwiger' AND email = 'nate@wiger.org'
1352 However, you may want loose comparisons by default, so if you set
1353 C<cmp> to C<like> you would get SQL such as:
1355 WHERE name like 'nwiger' AND email like 'nate@wiger.org'
1357 You can also override the comparsion on an individual basis - see
1358 the huge section on L</"WHERE CLAUSES"> at the bottom.
1360 =item sqltrue, sqlfalse
1362 Expressions for inserting boolean values within SQL statements.
1363 By default these are C<1=1> and C<1=0>. They are used
1364 by the special operators C<-in> and C<-not_in> for generating
1365 correct SQL even when the argument is an empty array (see below).
1369 This determines the default logical operator for multiple WHERE
1370 statements in arrays or hashes. If absent, the default logic is "or"
1371 for arrays, and "and" for hashes. This means that a WHERE
1375 event_date => {'>=', '2/13/99'},
1376 event_date => {'<=', '4/24/03'},
1379 will generate SQL like this:
1381 WHERE event_date >= '2/13/99' OR event_date <= '4/24/03'
1383 This is probably not what you want given this query, though (look
1384 at the dates). To change the "OR" to an "AND", simply specify:
1386 my $sql = SQL::Abstract->new(logic => 'and');
1388 Which will change the above C<WHERE> to:
1390 WHERE event_date >= '2/13/99' AND event_date <= '4/24/03'
1392 The logic can also be changed locally by inserting
1393 a modifier in front of an arrayref :
1395 @where = (-and => [event_date => {'>=', '2/13/99'},
1396 event_date => {'<=', '4/24/03'} ]);
1398 See the L</"WHERE CLAUSES"> section for explanations.
1402 This will automatically convert comparisons using the specified SQL
1403 function for both column and value. This is mostly used with an argument
1404 of C<upper> or C<lower>, so that the SQL will have the effect of
1405 case-insensitive "searches". For example, this:
1407 $sql = SQL::Abstract->new(convert => 'upper');
1408 %where = (keywords => 'MaKe iT CAse inSeNSItive');
1410 Will turn out the following SQL:
1412 WHERE upper(keywords) like upper('MaKe iT CAse inSeNSItive')
1414 The conversion can be C<upper()>, C<lower()>, or any other SQL function
1415 that can be applied symmetrically to fields (actually B<SQL::Abstract> does
1416 not validate this option; it will just pass through what you specify verbatim).
1420 This is a kludge because many databases suck. For example, you can't
1421 just bind values using DBI's C<execute()> for Oracle C<CLOB> or C<BLOB> fields.
1422 Instead, you have to use C<bind_param()>:
1424 $sth->bind_param(1, 'reg data');
1425 $sth->bind_param(2, $lots, {ora_type => ORA_CLOB});
1427 The problem is, B<SQL::Abstract> will normally just return a C<@bind> array,
1428 which loses track of which field each slot refers to. Fear not.
1430 If you specify C<bindtype> in new, you can determine how C<@bind> is returned.
1431 Currently, you can specify either C<normal> (default) or C<columns>. If you
1432 specify C<columns>, you will get an array that looks like this:
1434 my $sql = SQL::Abstract->new(bindtype => 'columns');
1435 my($stmt, @bind) = $sql->insert(...);
1438 [ 'column1', 'value1' ],
1439 [ 'column2', 'value2' ],
1440 [ 'column3', 'value3' ],
1443 You can then iterate through this manually, using DBI's C<bind_param()>.
1445 $sth->prepare($stmt);
1448 my($col, $data) = @$_;
1449 if ($col eq 'details' || $col eq 'comments') {
1450 $sth->bind_param($i, $data, {ora_type => ORA_CLOB});
1451 } elsif ($col eq 'image') {
1452 $sth->bind_param($i, $data, {ora_type => ORA_BLOB});
1454 $sth->bind_param($i, $data);
1458 $sth->execute; # execute without @bind now
1460 Now, why would you still use B<SQL::Abstract> if you have to do this crap?
1461 Basically, the advantage is still that you don't have to care which fields
1462 are or are not included. You could wrap that above C<for> loop in a simple
1463 sub called C<bind_fields()> or something and reuse it repeatedly. You still
1464 get a layer of abstraction over manual SQL specification.
1466 Note that if you set L</bindtype> to C<columns>, the C<\[$sql, @bind]>
1467 construct (see L</Literal SQL with placeholders and bind values (subqueries)>)
1468 will expect the bind values in this format.
1472 This is the character that a table or column name will be quoted
1473 with. By default this is an empty string, but you could set it to
1474 the character C<`>, to generate SQL like this:
1476 SELECT `a_field` FROM `a_table` WHERE `some_field` LIKE '%someval%'
1478 Alternatively, you can supply an array ref of two items, the first being the left
1479 hand quote character, and the second the right hand quote character. For
1480 example, you could supply C<['[',']']> for SQL Server 2000 compliant quotes
1481 that generates SQL like this:
1483 SELECT [a_field] FROM [a_table] WHERE [some_field] LIKE '%someval%'
1485 Quoting is useful if you have tables or columns names that are reserved
1486 words in your database's SQL dialect.
1490 This is the character that separates a table and column name. It is
1491 necessary to specify this when the C<quote_char> option is selected,
1492 so that tables and column names can be individually quoted like this:
1494 SELECT `table`.`one_field` FROM `table` WHERE `table`.`other_field` = 1
1496 =item injection_guard
1498 A regular expression C<qr/.../> that is applied to any C<-function> and unquoted
1499 column name specified in a query structure. This is a safety mechanism to avoid
1500 injection attacks when mishandling user input e.g.:
1502 my %condition_as_column_value_pairs = get_values_from_user();
1503 $sqla->select( ... , \%condition_as_column_value_pairs );
1505 If the expression matches an exception is thrown. Note that literal SQL
1506 supplied via C<\'...'> or C<\['...']> is B<not> checked in any way.
1508 Defaults to checking for C<;> and the C<GO> keyword (TransactSQL)
1510 =item array_datatypes
1512 When this option is true, arrayrefs in INSERT or UPDATE are
1513 interpreted as array datatypes and are passed directly
1515 When this option is false, arrayrefs are interpreted
1516 as literal SQL, just like refs to arrayrefs
1517 (but this behavior is for backwards compatibility; when writing
1518 new queries, use the "reference to arrayref" syntax
1524 Takes a reference to a list of "special operators"
1525 to extend the syntax understood by L<SQL::Abstract>.
1526 See section L</"SPECIAL OPERATORS"> for details.
1530 Takes a reference to a list of "unary operators"
1531 to extend the syntax understood by L<SQL::Abstract>.
1532 See section L</"UNARY OPERATORS"> for details.
1538 =head2 insert($table, \@values || \%fieldvals, \%options)
1540 This is the simplest function. You simply give it a table name
1541 and either an arrayref of values or hashref of field/value pairs.
1542 It returns an SQL INSERT statement and a list of bind values.
1543 See the sections on L</"Inserting and Updating Arrays"> and
1544 L</"Inserting and Updating SQL"> for information on how to insert
1545 with those data types.
1547 The optional C<\%options> hash reference may contain additional
1548 options to generate the insert SQL. Currently supported options
1555 Takes either a scalar of raw SQL fields, or an array reference of
1556 field names, and adds on an SQL C<RETURNING> statement at the end.
1557 This allows you to return data generated by the insert statement
1558 (such as row IDs) without performing another C<SELECT> statement.
1559 Note, however, this is not part of the SQL standard and may not
1560 be supported by all database engines.
1564 =head2 update($table, \%fieldvals, \%where)
1566 This takes a table, hashref of field/value pairs, and an optional
1567 hashref L<WHERE clause|/WHERE CLAUSES>. It returns an SQL UPDATE function and a list
1569 See the sections on L</"Inserting and Updating Arrays"> and
1570 L</"Inserting and Updating SQL"> for information on how to insert
1571 with those data types.
1573 =head2 select($source, $fields, $where, $order)
1575 This returns a SQL SELECT statement and associated list of bind values, as
1576 specified by the arguments :
1582 Specification of the 'FROM' part of the statement.
1583 The argument can be either a plain scalar (interpreted as a table
1584 name, will be quoted), or an arrayref (interpreted as a list
1585 of table names, joined by commas, quoted), or a scalarref
1586 (literal table name, not quoted), or a ref to an arrayref
1587 (list of literal table names, joined by commas, not quoted).
1591 Specification of the list of fields to retrieve from
1593 The argument can be either an arrayref (interpreted as a list
1594 of field names, will be joined by commas and quoted), or a
1595 plain scalar (literal SQL, not quoted).
1596 Please observe that this API is not as flexible as for
1597 the first argument C<$table>, for backwards compatibility reasons.
1601 Optional argument to specify the WHERE part of the query.
1602 The argument is most often a hashref, but can also be
1603 an arrayref or plain scalar --
1604 see section L<WHERE clause|/"WHERE CLAUSES"> for details.
1608 Optional argument to specify the ORDER BY part of the query.
1609 The argument can be a scalar, a hashref or an arrayref
1610 -- see section L<ORDER BY clause|/"ORDER BY CLAUSES">
1616 =head2 delete($table, \%where)
1618 This takes a table name and optional hashref L<WHERE clause|/WHERE CLAUSES>.
1619 It returns an SQL DELETE statement and list of bind values.
1621 =head2 where(\%where, \@order)
1623 This is used to generate just the WHERE clause. For example,
1624 if you have an arbitrary data structure and know what the
1625 rest of your SQL is going to look like, but want an easy way
1626 to produce a WHERE clause, use this. It returns an SQL WHERE
1627 clause and list of bind values.
1630 =head2 values(\%data)
1632 This just returns the values from the hash C<%data>, in the same
1633 order that would be returned from any of the other above queries.
1634 Using this allows you to markedly speed up your queries if you
1635 are affecting lots of rows. See below under the L</"PERFORMANCE"> section.
1637 =head2 generate($any, 'number', $of, \@data, $struct, \%types)
1639 Warning: This is an experimental method and subject to change.
1641 This returns arbitrarily generated SQL. It's a really basic shortcut.
1642 It will return two different things, depending on return context:
1644 my($stmt, @bind) = $sql->generate('create table', \$table, \@fields);
1645 my $stmt_and_val = $sql->generate('create table', \$table, \@fields);
1647 These would return the following:
1649 # First calling form
1650 $stmt = "CREATE TABLE test (?, ?)";
1651 @bind = (field1, field2);
1653 # Second calling form
1654 $stmt_and_val = "CREATE TABLE test (field1, field2)";
1656 Depending on what you're trying to do, it's up to you to choose the correct
1657 format. In this example, the second form is what you would want.
1661 $sql->generate('alter session', { nls_date_format => 'MM/YY' });
1665 ALTER SESSION SET nls_date_format = 'MM/YY'
1667 You get the idea. Strings get their case twiddled, but everything
1668 else remains verbatim.
1670 =head1 WHERE CLAUSES
1674 This module uses a variation on the idea from L<DBIx::Abstract>. It
1675 is B<NOT>, repeat I<not> 100% compatible. B<The main logic of this
1676 module is that things in arrays are OR'ed, and things in hashes
1679 The easiest way to explain is to show lots of examples. After
1680 each C<%where> hash shown, it is assumed you used:
1682 my($stmt, @bind) = $sql->where(\%where);
1684 However, note that the C<%where> hash can be used directly in any
1685 of the other functions as well, as described above.
1687 =head2 Key-value pairs
1689 So, let's get started. To begin, a simple hash:
1693 status => 'completed'
1696 Is converted to SQL C<key = val> statements:
1698 $stmt = "WHERE user = ? AND status = ?";
1699 @bind = ('nwiger', 'completed');
1701 One common thing I end up doing is having a list of values that
1702 a field can be in. To do this, simply specify a list inside of
1707 status => ['assigned', 'in-progress', 'pending'];
1710 This simple code will create the following:
1712 $stmt = "WHERE user = ? AND ( status = ? OR status = ? OR status = ? )";
1713 @bind = ('nwiger', 'assigned', 'in-progress', 'pending');
1715 A field associated to an empty arrayref will be considered a
1716 logical false and will generate 0=1.
1718 =head2 Tests for NULL values
1720 If the value part is C<undef> then this is converted to SQL <IS NULL>
1729 $stmt = "WHERE user = ? AND status IS NULL";
1732 To test if a column IS NOT NULL:
1736 status => { '!=', undef },
1739 =head2 Specific comparison operators
1741 If you want to specify a different type of operator for your comparison,
1742 you can use a hashref for a given column:
1746 status => { '!=', 'completed' }
1749 Which would generate:
1751 $stmt = "WHERE user = ? AND status != ?";
1752 @bind = ('nwiger', 'completed');
1754 To test against multiple values, just enclose the values in an arrayref:
1756 status => { '=', ['assigned', 'in-progress', 'pending'] };
1758 Which would give you:
1760 "WHERE status = ? OR status = ? OR status = ?"
1763 The hashref can also contain multiple pairs, in which case it is expanded
1764 into an C<AND> of its elements:
1768 status => { '!=', 'completed', -not_like => 'pending%' }
1771 # Or more dynamically, like from a form
1772 $where{user} = 'nwiger';
1773 $where{status}{'!='} = 'completed';
1774 $where{status}{'-not_like'} = 'pending%';
1776 # Both generate this
1777 $stmt = "WHERE user = ? AND status != ? AND status NOT LIKE ?";
1778 @bind = ('nwiger', 'completed', 'pending%');
1781 To get an OR instead, you can combine it with the arrayref idea:
1785 priority => [ { '=', 2 }, { '>', 5 } ]
1788 Which would generate:
1790 $stmt = "WHERE ( priority = ? OR priority > ? ) AND user = ?";
1791 @bind = ('2', '5', 'nwiger');
1793 If you want to include literal SQL (with or without bind values), just use a
1794 scalar reference or array reference as the value:
1797 date_entered => { '>' => \["to_date(?, 'MM/DD/YYYY')", "11/26/2008"] },
1798 date_expires => { '<' => \"now()" }
1801 Which would generate:
1803 $stmt = "WHERE date_entered > "to_date(?, 'MM/DD/YYYY') AND date_expires < now()";
1804 @bind = ('11/26/2008');
1807 =head2 Logic and nesting operators
1809 In the example above,
1810 there is a subtle trap if you want to say something like
1811 this (notice the C<AND>):
1813 WHERE priority != ? AND priority != ?
1815 Because, in Perl you I<can't> do this:
1817 priority => { '!=', 2, '!=', 1 }
1819 As the second C<!=> key will obliterate the first. The solution
1820 is to use the special C<-modifier> form inside an arrayref:
1822 priority => [ -and => {'!=', 2},
1826 Normally, these would be joined by C<OR>, but the modifier tells it
1827 to use C<AND> instead. (Hint: You can use this in conjunction with the
1828 C<logic> option to C<new()> in order to change the way your queries
1829 work by default.) B<Important:> Note that the C<-modifier> goes
1830 B<INSIDE> the arrayref, as an extra first element. This will
1831 B<NOT> do what you think it might:
1833 priority => -and => [{'!=', 2}, {'!=', 1}] # WRONG!
1835 Here is a quick list of equivalencies, since there is some overlap:
1838 status => {'!=', 'completed', 'not like', 'pending%' }
1839 status => [ -and => {'!=', 'completed'}, {'not like', 'pending%'}]
1842 status => {'=', ['assigned', 'in-progress']}
1843 status => [ -or => {'=', 'assigned'}, {'=', 'in-progress'}]
1844 status => [ {'=', 'assigned'}, {'=', 'in-progress'} ]
1848 =head2 Special operators : IN, BETWEEN, etc.
1850 You can also use the hashref format to compare a list of fields using the
1851 C<IN> comparison operator, by specifying the list as an arrayref:
1854 status => 'completed',
1855 reportid => { -in => [567, 2335, 2] }
1858 Which would generate:
1860 $stmt = "WHERE status = ? AND reportid IN (?,?,?)";
1861 @bind = ('completed', '567', '2335', '2');
1863 The reverse operator C<-not_in> generates SQL C<NOT IN> and is used in
1866 If the argument to C<-in> is an empty array, 'sqlfalse' is generated
1867 (by default : C<1=0>). Similarly, C<< -not_in => [] >> generates
1868 'sqltrue' (by default : C<1=1>).
1870 In addition to the array you can supply a chunk of literal sql or
1871 literal sql with bind:
1874 customer => { -in => \[
1875 'SELECT cust_id FROM cust WHERE balance > ?',
1878 status => { -in => \'SELECT status_codes FROM states' },
1884 customer IN ( SELECT cust_id FROM cust WHERE balance > ? )
1885 AND status IN ( SELECT status_codes FROM states )
1891 Another pair of operators is C<-between> and C<-not_between>,
1892 used with an arrayref of two values:
1896 completion_date => {
1897 -not_between => ['2002-10-01', '2003-02-06']
1903 WHERE user = ? AND completion_date NOT BETWEEN ( ? AND ? )
1905 Just like with C<-in> all plausible combinations of literal SQL
1909 start0 => { -between => [ 1, 2 ] },
1910 start1 => { -between => \["? AND ?", 1, 2] },
1911 start2 => { -between => \"lower(x) AND upper(y)" },
1912 start3 => { -between => [
1914 \["upper(?)", 'stuff' ],
1921 ( start0 BETWEEN ? AND ? )
1922 AND ( start1 BETWEEN ? AND ? )
1923 AND ( start2 BETWEEN lower(x) AND upper(y) )
1924 AND ( start3 BETWEEN lower(x) AND upper(?) )
1926 @bind = (1, 2, 1, 2, 'stuff');
1929 These are the two builtin "special operators"; but the
1930 list can be expanded : see section L</"SPECIAL OPERATORS"> below.
1932 =head2 Unary operators: bool
1934 If you wish to test against boolean columns or functions within your
1935 database you can use the C<-bool> and C<-not_bool> operators. For
1936 example to test the column C<is_user> being true and the column
1937 C<is_enabled> being false you would use:-
1941 -not_bool => 'is_enabled',
1946 WHERE is_user AND NOT is_enabled
1948 If a more complex combination is required, testing more conditions,
1949 then you should use the and/or operators:-
1956 -not_bool => 'four',
1962 WHERE one AND two AND three AND NOT four
1965 =head2 Nested conditions, -and/-or prefixes
1967 So far, we've seen how multiple conditions are joined with a top-level
1968 C<AND>. We can change this by putting the different conditions we want in
1969 hashes and then putting those hashes in an array. For example:
1974 status => { -like => ['pending%', 'dispatched'] },
1978 status => 'unassigned',
1982 This data structure would create the following:
1984 $stmt = "WHERE ( user = ? AND ( status LIKE ? OR status LIKE ? ) )
1985 OR ( user = ? AND status = ? ) )";
1986 @bind = ('nwiger', 'pending', 'dispatched', 'robot', 'unassigned');
1989 Clauses in hashrefs or arrayrefs can be prefixed with an C<-and> or C<-or>
1990 to change the logic inside :
1996 -and => [ workhrs => {'>', 20}, geo => 'ASIA' ],
1997 -or => { workhrs => {'<', 50}, geo => 'EURO' },
2004 WHERE ( user = ? AND (
2005 ( workhrs > ? AND geo = ? )
2006 OR ( workhrs < ? OR geo = ? )
2009 =head3 Algebraic inconsistency, for historical reasons
2011 C<Important note>: when connecting several conditions, the C<-and->|C<-or>
2012 operator goes C<outside> of the nested structure; whereas when connecting
2013 several constraints on one column, the C<-and> operator goes
2014 C<inside> the arrayref. Here is an example combining both features :
2017 -and => [a => 1, b => 2],
2018 -or => [c => 3, d => 4],
2019 e => [-and => {-like => 'foo%'}, {-like => '%bar'} ]
2024 WHERE ( ( ( a = ? AND b = ? )
2025 OR ( c = ? OR d = ? )
2026 OR ( e LIKE ? AND e LIKE ? ) ) )
2028 This difference in syntax is unfortunate but must be preserved for
2029 historical reasons. So be careful : the two examples below would
2030 seem algebraically equivalent, but they are not
2032 {col => [-and => {-like => 'foo%'}, {-like => '%bar'}]}
2033 # yields : WHERE ( ( col LIKE ? AND col LIKE ? ) )
2035 [-and => {col => {-like => 'foo%'}, {col => {-like => '%bar'}}]]
2036 # yields : WHERE ( ( col LIKE ? OR col LIKE ? ) )
2039 =head2 Literal SQL and value type operators
2041 The basic premise of SQL::Abstract is that in WHERE specifications the "left
2042 side" is a column name and the "right side" is a value (normally rendered as
2043 a placeholder). This holds true for both hashrefs and arrayref pairs as you
2044 see in the L</WHERE CLAUSES> examples above. Sometimes it is necessary to
2045 alter this behavior. There are several ways of doing so.
2049 This is a virtual operator that signals the string to its right side is an
2050 identifier (a column name) and not a value. For example to compare two
2051 columns you would write:
2054 priority => { '<', 2 },
2055 requestor => { -ident => 'submitter' },
2060 $stmt = "WHERE priority < ? AND requestor = submitter";
2063 If you are maintaining legacy code you may see a different construct as
2064 described in L</Deprecated usage of Literal SQL>, please use C<-ident> in new
2069 This is a virtual operator that signals that the construct to its right side
2070 is a value to be passed to DBI. This is for example necessary when you want
2071 to write a where clause against an array (for RDBMS that support such
2072 datatypes). For example:
2075 array => { -value => [1, 2, 3] }
2080 $stmt = 'WHERE array = ?';
2081 @bind = ([1, 2, 3]);
2083 Note that if you were to simply say:
2089 the result would porbably be not what you wanted:
2091 $stmt = 'WHERE array = ? OR array = ? OR array = ?';
2096 Finally, sometimes only literal SQL will do. To include a random snippet
2097 of SQL verbatim, you specify it as a scalar reference. Consider this only
2098 as a last resort. Usually there is a better way. For example:
2101 priority => { '<', 2 },
2102 requestor => { -in => \'(SELECT name FROM hitmen)' },
2107 $stmt = "WHERE priority < ? AND requestor IN (SELECT name FROM hitmen)"
2110 Note that in this example, you only get one bind parameter back, since
2111 the verbatim SQL is passed as part of the statement.
2115 Never use untrusted input as a literal SQL argument - this is a massive
2116 security risk (there is no way to check literal snippets for SQL
2117 injections and other nastyness). If you need to deal with untrusted input
2118 use literal SQL with placeholders as described next.
2120 =head3 Literal SQL with placeholders and bind values (subqueries)
2122 If the literal SQL to be inserted has placeholders and bind values,
2123 use a reference to an arrayref (yes this is a double reference --
2124 not so common, but perfectly legal Perl). For example, to find a date
2125 in Postgres you can use something like this:
2128 date_column => \[q/= date '2008-09-30' - ?::integer/, 10/]
2133 $stmt = "WHERE ( date_column = date '2008-09-30' - ?::integer )"
2136 Note that you must pass the bind values in the same format as they are returned
2137 by L</where>. That means that if you set L</bindtype> to C<columns>, you must
2138 provide the bind values in the C<< [ column_meta => value ] >> format, where
2139 C<column_meta> is an opaque scalar value; most commonly the column name, but
2140 you can use any scalar value (including references and blessed references),
2141 L<SQL::Abstract> will simply pass it through intact. So if C<bindtype> is set
2142 to C<columns> the above example will look like:
2145 date_column => \[q/= date '2008-09-30' - ?::integer/, [ dummy => 10 ]/]
2148 Literal SQL is especially useful for nesting parenthesized clauses in the
2149 main SQL query. Here is a first example :
2151 my ($sub_stmt, @sub_bind) = ("SELECT c1 FROM t1 WHERE c2 < ? AND c3 LIKE ?",
2155 bar => \["IN ($sub_stmt)" => @sub_bind],
2160 $stmt = "WHERE (foo = ? AND bar IN (SELECT c1 FROM t1
2161 WHERE c2 < ? AND c3 LIKE ?))";
2162 @bind = (1234, 100, "foo%");
2164 Other subquery operators, like for example C<"E<gt> ALL"> or C<"NOT IN">,
2165 are expressed in the same way. Of course the C<$sub_stmt> and
2166 its associated bind values can be generated through a former call
2169 my ($sub_stmt, @sub_bind)
2170 = $sql->select("t1", "c1", {c2 => {"<" => 100},
2171 c3 => {-like => "foo%"}});
2174 bar => \["> ALL ($sub_stmt)" => @sub_bind],
2177 In the examples above, the subquery was used as an operator on a column;
2178 but the same principle also applies for a clause within the main C<%where>
2179 hash, like an EXISTS subquery :
2181 my ($sub_stmt, @sub_bind)
2182 = $sql->select("t1", "*", {c1 => 1, c2 => \"> t0.c0"});
2183 my %where = ( -and => [
2185 \["EXISTS ($sub_stmt)" => @sub_bind],
2190 $stmt = "WHERE (foo = ? AND EXISTS (SELECT * FROM t1
2191 WHERE c1 = ? AND c2 > t0.c0))";
2195 Observe that the condition on C<c2> in the subquery refers to
2196 column C<t0.c0> of the main query : this is I<not> a bind
2197 value, so we have to express it through a scalar ref.
2198 Writing C<< c2 => {">" => "t0.c0"} >> would have generated
2199 C<< c2 > ? >> with bind value C<"t0.c0"> ... not exactly
2200 what we wanted here.
2202 Finally, here is an example where a subquery is used
2203 for expressing unary negation:
2205 my ($sub_stmt, @sub_bind)
2206 = $sql->where({age => [{"<" => 10}, {">" => 20}]});
2207 $sub_stmt =~ s/^ where //i; # don't want "WHERE" in the subclause
2209 lname => {like => '%son%'},
2210 \["NOT ($sub_stmt)" => @sub_bind],
2215 $stmt = "lname LIKE ? AND NOT ( age < ? OR age > ? )"
2216 @bind = ('%son%', 10, 20)
2218 =head3 Deprecated usage of Literal SQL
2220 Below are some examples of archaic use of literal SQL. It is shown only as
2221 reference for those who deal with legacy code. Each example has a much
2222 better, cleaner and safer alternative that users should opt for in new code.
2228 my %where = ( requestor => \'IS NOT NULL' )
2230 $stmt = "WHERE requestor IS NOT NULL"
2232 This used to be the way of generating NULL comparisons, before the handling
2233 of C<undef> got formalized. For new code please use the superior syntax as
2234 described in L</Tests for NULL values>.
2238 my %where = ( requestor => \'= submitter' )
2240 $stmt = "WHERE requestor = submitter"
2242 This used to be the only way to compare columns. Use the superior L</-ident>
2243 method for all new code. For example an identifier declared in such a way
2244 will be properly quoted if L</quote_char> is properly set, while the legacy
2245 form will remain as supplied.
2249 my %where = ( is_ready => \"", completed => { '>', '2012-12-21' } )
2251 $stmt = "WHERE completed > ? AND is_ready"
2252 @bind = ('2012-12-21')
2254 Using an empty string literal used to be the only way to express a boolean.
2255 For all new code please use the much more readable
2256 L<-bool|/Unary operators: bool> operator.
2262 These pages could go on for a while, since the nesting of the data
2263 structures this module can handle are pretty much unlimited (the
2264 module implements the C<WHERE> expansion as a recursive function
2265 internally). Your best bet is to "play around" with the module a
2266 little to see how the data structures behave, and choose the best
2267 format for your data based on that.
2269 And of course, all the values above will probably be replaced with
2270 variables gotten from forms or the command line. After all, if you
2271 knew everything ahead of time, you wouldn't have to worry about
2272 dynamically-generating SQL and could just hardwire it into your
2275 =head1 ORDER BY CLAUSES
2277 Some functions take an order by clause. This can either be a scalar (just a
2278 column name,) a hash of C<< { -desc => 'col' } >> or C<< { -asc => 'col' } >>,
2279 or an array of either of the two previous forms. Examples:
2281 Given | Will Generate
2282 ----------------------------------------------------------
2284 \'colA DESC' | ORDER BY colA DESC
2286 'colA' | ORDER BY colA
2288 [qw/colA colB/] | ORDER BY colA, colB
2290 {-asc => 'colA'} | ORDER BY colA ASC
2292 {-desc => 'colB'} | ORDER BY colB DESC
2294 ['colA', {-asc => 'colB'}] | ORDER BY colA, colB ASC
2296 { -asc => [qw/colA colB/] } | ORDER BY colA ASC, colB ASC
2299 { -asc => 'colA' }, | ORDER BY colA ASC, colB DESC,
2300 { -desc => [qw/colB/], | colC ASC, colD ASC
2301 { -asc => [qw/colC colD/],|
2303 ===========================================================
2307 =head1 SPECIAL OPERATORS
2309 my $sqlmaker = SQL::Abstract->new(special_ops => [
2313 my ($self, $field, $op, $arg) = @_;
2319 handler => 'method_name',
2323 A "special operator" is a SQL syntactic clause that can be
2324 applied to a field, instead of a usual binary operator.
2327 WHERE field IN (?, ?, ?)
2328 WHERE field BETWEEN ? AND ?
2329 WHERE MATCH(field) AGAINST (?, ?)
2331 Special operators IN and BETWEEN are fairly standard and therefore
2332 are builtin within C<SQL::Abstract> (as the overridable methods
2333 C<_where_field_IN> and C<_where_field_BETWEEN>). For other operators,
2334 like the MATCH .. AGAINST example above which is specific to MySQL,
2335 you can write your own operator handlers - supply a C<special_ops>
2336 argument to the C<new> method. That argument takes an arrayref of
2337 operator definitions; each operator definition is a hashref with two
2344 the regular expression to match the operator
2348 Either a coderef or a plain scalar method name. In both cases
2349 the expected return is C<< ($sql, @bind) >>.
2351 When supplied with a method name, it is simply called on the
2352 L<SQL::Abstract/> object as:
2354 $self->$method_name ($field, $op, $arg)
2358 $op is the part that matched the handler regex
2359 $field is the LHS of the operator
2362 When supplied with a coderef, it is called as:
2364 $coderef->($self, $field, $op, $arg)
2369 For example, here is an implementation
2370 of the MATCH .. AGAINST syntax for MySQL
2372 my $sqlmaker = SQL::Abstract->new(special_ops => [
2374 # special op for MySql MATCH (field) AGAINST(word1, word2, ...)
2375 {regex => qr/^match$/i,
2377 my ($self, $field, $op, $arg) = @_;
2378 $arg = [$arg] if not ref $arg;
2379 my $label = $self->_quote($field);
2380 my ($placeholder) = $self->_convert('?');
2381 my $placeholders = join ", ", (($placeholder) x @$arg);
2382 my $sql = $self->_sqlcase('match') . " ($label) "
2383 . $self->_sqlcase('against') . " ($placeholders) ";
2384 my @bind = $self->_bindtype($field, @$arg);
2385 return ($sql, @bind);
2392 =head1 UNARY OPERATORS
2394 my $sqlmaker = SQL::Abstract->new(unary_ops => [
2398 my ($self, $op, $arg) = @_;
2404 handler => 'method_name',
2408 A "unary operator" is a SQL syntactic clause that can be
2409 applied to a field - the operator goes before the field
2411 You can write your own operator handlers - supply a C<unary_ops>
2412 argument to the C<new> method. That argument takes an arrayref of
2413 operator definitions; each operator definition is a hashref with two
2420 the regular expression to match the operator
2424 Either a coderef or a plain scalar method name. In both cases
2425 the expected return is C<< $sql >>.
2427 When supplied with a method name, it is simply called on the
2428 L<SQL::Abstract/> object as:
2430 $self->$method_name ($op, $arg)
2434 $op is the part that matched the handler regex
2435 $arg is the RHS or argument of the operator
2437 When supplied with a coderef, it is called as:
2439 $coderef->($self, $op, $arg)
2447 Thanks to some benchmarking by Mark Stosberg, it turns out that
2448 this module is many orders of magnitude faster than using C<DBIx::Abstract>.
2449 I must admit this wasn't an intentional design issue, but it's a
2450 byproduct of the fact that you get to control your C<DBI> handles
2453 To maximize performance, use a code snippet like the following:
2455 # prepare a statement handle using the first row
2456 # and then reuse it for the rest of the rows
2458 for my $href (@array_of_hashrefs) {
2459 $stmt ||= $sql->insert('table', $href);
2460 $sth ||= $dbh->prepare($stmt);
2461 $sth->execute($sql->values($href));
2464 The reason this works is because the keys in your C<$href> are sorted
2465 internally by B<SQL::Abstract>. Thus, as long as your data retains
2466 the same structure, you only have to generate the SQL the first time
2467 around. On subsequent queries, simply use the C<values> function provided
2468 by this module to return your values in the correct order.
2470 However this depends on the values having the same type - if, for
2471 example, the values of a where clause may either have values
2472 (resulting in sql of the form C<column = ?> with a single bind
2473 value), or alternatively the values might be C<undef> (resulting in
2474 sql of the form C<column IS NULL> with no bind value) then the
2475 caching technique suggested will not work.
2479 If you use my C<CGI::FormBuilder> module at all, you'll hopefully
2480 really like this part (I do, at least). Building up a complex query
2481 can be as simple as the following:
2485 use CGI::FormBuilder;
2488 my $form = CGI::FormBuilder->new(...);
2489 my $sql = SQL::Abstract->new;
2491 if ($form->submitted) {
2492 my $field = $form->field;
2493 my $id = delete $field->{id};
2494 my($stmt, @bind) = $sql->update('table', $field, {id => $id});
2497 Of course, you would still have to connect using C<DBI> to run the
2498 query, but the point is that if you make your form look like your
2499 table, the actual query script can be extremely simplistic.
2501 If you're B<REALLY> lazy (I am), check out C<HTML::QuickTable> for
2502 a fast interface to returning and formatting data. I frequently
2503 use these three modules together to write complex database query
2504 apps in under 50 lines.
2510 =item * gitweb: L<http://git.shadowcat.co.uk/gitweb/gitweb.cgi?p=dbsrgits/SQL-Abstract.git>
2512 =item * git: L<git://git.shadowcat.co.uk/dbsrgits/SQL-Abstract.git>
2518 Version 1.50 was a major internal refactoring of C<SQL::Abstract>.
2519 Great care has been taken to preserve the I<published> behavior
2520 documented in previous versions in the 1.* family; however,
2521 some features that were previously undocumented, or behaved
2522 differently from the documentation, had to be changed in order
2523 to clarify the semantics. Hence, client code that was relying
2524 on some dark areas of C<SQL::Abstract> v1.*
2525 B<might behave differently> in v1.50.
2527 The main changes are :
2533 support for literal SQL through the C<< \ [$sql, bind] >> syntax.
2537 support for the { operator => \"..." } construct (to embed literal SQL)
2541 support for the { operator => \["...", @bind] } construct (to embed literal SQL with bind values)
2545 optional support for L<array datatypes|/"Inserting and Updating Arrays">
2549 defensive programming : check arguments
2553 fixed bug with global logic, which was previously implemented
2554 through global variables yielding side-effects. Prior versions would
2555 interpret C<< [ {cond1, cond2}, [cond3, cond4] ] >>
2556 as C<< "(cond1 AND cond2) OR (cond3 AND cond4)" >>.
2557 Now this is interpreted
2558 as C<< "(cond1 AND cond2) OR (cond3 OR cond4)" >>.
2563 fixed semantics of _bindtype on array args
2567 dropped the C<_anoncopy> of the %where tree. No longer necessary,
2568 we just avoid shifting arrays within that tree.
2572 dropped the C<_modlogic> function
2576 =head1 ACKNOWLEDGEMENTS
2578 There are a number of individuals that have really helped out with
2579 this module. Unfortunately, most of them submitted bugs via CPAN
2580 so I have no idea who they are! But the people I do know are:
2582 Ash Berlin (order_by hash term support)
2583 Matt Trout (DBIx::Class support)
2584 Mark Stosberg (benchmarking)
2585 Chas Owens (initial "IN" operator support)
2586 Philip Collins (per-field SQL functions)
2587 Eric Kolve (hashref "AND" support)
2588 Mike Fragassi (enhancements to "BETWEEN" and "LIKE")
2589 Dan Kubb (support for "quote_char" and "name_sep")
2590 Guillermo Roditi (patch to cleanup "IN" and "BETWEEN", fix and tests for _order_by)
2591 Laurent Dami (internal refactoring, extensible list of special operators, literal SQL)
2592 Norbert Buchmuller (support for literal SQL in hashpair, misc. fixes & tests)
2593 Peter Rabbitson (rewrite of SQLA::Test, misc. fixes & tests)
2594 Oliver Charles (support for "RETURNING" after "INSERT")
2600 L<DBIx::Class>, L<DBIx::Abstract>, L<CGI::FormBuilder>, L<HTML::QuickTable>.
2604 Copyright (c) 2001-2007 Nathan Wiger <nwiger@cpan.org>. All Rights Reserved.
2606 This module is actively maintained by Matt Trout <mst@shadowcatsystems.co.uk>
2608 For support, your best bet is to try the C<DBIx::Class> users mailing list.
2609 While not an official support venue, C<DBIx::Class> makes heavy use of
2610 C<SQL::Abstract>, and as such list members there are very familiar with
2611 how to create queries.
2615 This module is free software; you may copy this under the same
2616 terms as perl itself (either the GNU General Public License or
2617 the Artistic License)