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)) {
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) ]
648 operator => { 'SQL.Naive' => 'apply' },
650 $self->_ident_to_dq($op),
651 (map $self->_where_to_dq($_), (ref($v) eq 'ARRAY' ? @$v : $v))
656 local our $Cur_Col_Meta = $k;
657 if (ref($v) eq 'ARRAY') {
659 return $self->_literal_to_dq($self->{sqlfalse});
660 } elsif (defined($v->[0]) && $v->[0] =~ /-(and|or)/i) {
661 return $self->_where_to_dq_ARRAYREF([
662 map +{ $k => $_ }, @{$v}[1..$#$v]
665 return $self->_where_to_dq_ARRAYREF([
666 map +{ $k => $_ }, @$v
668 } elsif (ref($v) eq 'SCALAR' or (ref($v) eq 'REF' and ref($$v) eq 'ARRAY')) {
669 return $self->_literal_with_prepend_to_dq($k, $$v);
671 my ($op, $rhs) = do {
672 if (ref($v) eq 'HASH') {
674 return $self->_where_to_dq_ARRAYREF([
675 map +{ $k => { $_ => $v->{$_} } }, keys %$v
678 (uc((keys %$v)[0]), (values %$v)[0]);
683 s/^-//, s/_/ /g for $op;
684 if ($op eq 'BETWEEN' or $op eq 'IN' or $op eq 'NOT IN' or $op eq 'NOT BETWEEN') {
685 if (ref($rhs) ne 'ARRAY') {
687 # have to add parens if none present because -in => \"SELECT ..."
688 # got documented. mst hates everything.
689 if (ref($rhs) eq 'SCALAR') {
691 $x = "($x)" unless $x =~ /^\s*\(/;
694 my ($x, @rest) = @{$$rhs};
695 $x = "($x)" unless $x =~ /^\s*\(/;
696 $rhs = \[ $x, @rest ];
699 return $self->_literal_with_prepend_to_dq("$k $op", $$rhs);
701 return $self->_literal_to_dq($self->{sqlfalse}) unless @$rhs;
704 operator => { 'SQL.Naive' => $op },
705 args => [ $self->_ident_to_dq($k), map $self->_where_to_dq($_), @$rhs ]
707 } elsif ($op =~ s/^NOT (?!LIKE)//) {
708 return $self->_where_hashpair_to_dq(-not => { $k => { $op => $rhs } });
709 } elsif (!defined($rhs)) {
711 if ($op eq '=' or $op eq 'LIKE') {
713 } elsif ($op eq '!=') {
716 die "Can't do undef -> NULL transform for operator ${op}";
721 operator => { 'SQL.Naive' => $null_op },
722 args => [ $self->_ident_to_dq($k) ]
725 if (ref($rhs) eq 'ARRAY') {
727 return $self->_literal_to_dq(
728 $op eq '!=' ? $self->{sqltrue} : $self->{sqlfalse}
730 } elsif (defined($rhs->[0]) and $rhs->[0] =~ /-(and|or)/i) {
731 return $self->_where_to_dq_ARRAYREF([
732 map +{ $k => { $op => $_ } }, @{$rhs}[1..$#$rhs]
735 return $self->_where_to_dq_ARRAYREF([
736 map +{ $k => { $op => $_ } }, @$rhs
741 operator => { 'SQL.Naive' => $op },
742 args => [ $self->_ident_to_dq($k), $self->_where_to_dq($rhs) ]
747 #======================================================================
749 #======================================================================
752 my ($self, $arg) = @_;
753 if (my $dq = $self->_order_by_to_dq($arg)) {
754 # SQLA generates ' ORDER BY foo'. The hilarity.
756 ? do { my @r = $self->_render_dq($dq); $r[0] = ' '.$r[0]; @r }
757 : ' '.$self->_render_dq($dq);
763 sub _order_by_to_dq {
764 my ($self, $arg, $dir) = @_;
770 ($dir ? (direction => $dir) : ()),
774 $dq->{by} = $self->_ident_to_dq($arg);
775 } elsif (ref($arg) eq 'ARRAY') {
777 local our $Order_Inner unless our $Order_Recursing;
778 local $Order_Recursing = 1;
780 foreach my $member (@$arg) {
782 my $next = $self->_order_by_to_dq($member, $dir);
784 $inner->{from} = $next if $inner;
785 $inner = $Order_Inner || $next;
787 $Order_Inner = $inner;
789 } elsif (ref($arg) eq 'REF' and ref($$arg) eq 'ARRAY') {
790 $dq->{by} = $self->_literal_to_dq($$arg);
791 } elsif (ref($arg) eq 'SCALAR') {
792 $dq->{by} = $self->_literal_to_dq($$arg);
793 } elsif (ref($arg) eq 'HASH') {
794 my ($key, $val, @rest) = %$arg;
798 if (@rest or not $key =~ /^-(desc|asc)/i) {
799 puke "hash passed to _order_by must have exactly one key (-desc or -asc)";
802 return $self->_order_by_to_dq($val, $dir);
804 die "Can't handle $arg in _order_by_to_dq";
809 #======================================================================
810 # DATASOURCE (FOR NOW, JUST PLAIN TABLE OR LIST OF TABLES)
811 #======================================================================
814 my ($self, $from) = @_;
815 $self->_render_dq($self->_table_to_dq($from));
819 my ($self, $from) = @_;
820 $self->_SWITCH_refkind($from, {
822 die "Empty FROM list" unless my @f = @$from;
823 my $dq = $self->_ident_to_dq(shift @f);
824 while (my $x = shift @f) {
827 join => [ $dq, $self->_ident_to_dq($x) ]
832 SCALAR => sub { $self->_ident_to_dq($from) },
844 #======================================================================
846 #======================================================================
848 # highly optimized, as it's called way too often
850 # my ($self, $label) = @_;
852 return '' unless defined $_[1];
853 return ${$_[1]} if ref($_[1]) eq 'SCALAR';
855 unless ($_[0]->{quote_char}) {
856 $_[0]->_assert_pass_injection_guard($_[1]);
860 my $qref = ref $_[0]->{quote_char};
863 ($l, $r) = ( $_[0]->{quote_char}, $_[0]->{quote_char} );
865 elsif ($qref eq 'ARRAY') {
866 ($l, $r) = @{$_[0]->{quote_char}};
869 puke "Unsupported quote_char format: $_[0]->{quote_char}";
872 # parts containing * are naturally unquoted
873 return join( $_[0]->{name_sep}||'', map
874 { $_ eq '*' ? $_ : $l . $_ . $r }
875 ( $_[0]->{name_sep} ? split (/\Q$_[0]->{name_sep}\E/, $_[1] ) : $_[1] )
880 # Conversion, if applicable
882 #my ($self, $arg) = @_;
884 # LDNOTE : modified the previous implementation below because
885 # it was not consistent : the first "return" is always an array,
886 # the second "return" is context-dependent. Anyway, _convert
887 # seems always used with just a single argument, so make it a
889 # return @_ unless $self->{convert};
890 # my $conv = $self->_sqlcase($self->{convert});
891 # my @ret = map { $conv.'('.$_.')' } @_;
892 # return wantarray ? @ret : $ret[0];
893 if ($_[0]->{convert}) {
894 return $_[0]->_sqlcase($_[0]->{convert}) .'(' . $_[1] . ')';
901 #my ($self, $col, @vals) = @_;
903 #LDNOTE : changed original implementation below because it did not make
904 # sense when bindtype eq 'columns' and @vals > 1.
905 # return $self->{bindtype} eq 'columns' ? [ $col, @vals ] : @vals;
907 # called often - tighten code
908 return $_[0]->{bindtype} eq 'columns'
909 ? map {[$_[1], $_]} @_[2 .. $#_]
914 # Dies if any element of @bind is not in [colname => value] format
915 # if bindtype is 'columns'.
916 sub _assert_bindval_matches_bindtype {
917 # my ($self, @bind) = @_;
919 if ($self->{bindtype} eq 'columns') {
921 if (!defined $_ || ref($_) ne 'ARRAY' || @$_ != 2) {
922 puke "bindtype 'columns' selected, you need to pass: [column_name => bind_value]"
928 sub _join_sql_clauses {
929 my ($self, $logic, $clauses_aref, $bind_aref) = @_;
931 if (@$clauses_aref > 1) {
932 my $join = " " . $self->_sqlcase($logic) . " ";
933 my $sql = '( ' . join($join, @$clauses_aref) . ' )';
934 return ($sql, @$bind_aref);
936 elsif (@$clauses_aref) {
937 return ($clauses_aref->[0], @$bind_aref); # no parentheses
940 return (); # if no SQL, ignore @$bind_aref
945 # Fix SQL case, if so requested
947 # LDNOTE: if $self->{case} is true, then it contains 'lower', so we
948 # don't touch the argument ... crooked logic, but let's not change it!
949 return $_[0]->{case} ? $_[1] : uc($_[1]);
953 #======================================================================
954 # DISPATCHING FROM REFKIND
955 #======================================================================
958 my ($self, $data) = @_;
960 return 'UNDEF' unless defined $data;
962 # blessed objects are treated like scalars
963 my $ref = (Scalar::Util::blessed $data) ? '' : ref $data;
965 return 'SCALAR' unless $ref;
968 while ($ref eq 'REF') {
970 $ref = (Scalar::Util::blessed $data) ? '' : ref $data;
974 return ($ref||'SCALAR') . ('REF' x $n_steps);
978 my ($self, $data) = @_;
979 my @try = ($self->_refkind($data));
980 push @try, 'SCALAR_or_UNDEF' if $try[0] eq 'SCALAR' || $try[0] eq 'UNDEF';
981 push @try, 'FALLBACK';
985 sub _METHOD_FOR_refkind {
986 my ($self, $meth_prefix, $data) = @_;
989 for (@{$self->_try_refkind($data)}) {
990 $method = $self->can($meth_prefix."_".$_)
994 return $method || puke "cannot dispatch on '$meth_prefix' for ".$self->_refkind($data);
998 sub _SWITCH_refkind {
999 my ($self, $data, $dispatch_table) = @_;
1002 for (@{$self->_try_refkind($data)}) {
1003 $coderef = $dispatch_table->{$_}
1007 puke "no dispatch entry for ".$self->_refkind($data)
1016 #======================================================================
1017 # VALUES, GENERATE, AUTOLOAD
1018 #======================================================================
1020 # LDNOTE: original code from nwiger, didn't touch code in that section
1021 # I feel the AUTOLOAD stuff should not be the default, it should
1022 # only be activated on explicit demand by user.
1026 my $data = shift || return;
1027 puke "Argument to ", __PACKAGE__, "->values must be a \\%hash"
1028 unless ref $data eq 'HASH';
1031 foreach my $k ( sort keys %$data ) {
1032 my $v = $data->{$k};
1033 $self->_SWITCH_refkind($v, {
1035 if ($self->{array_datatypes}) { # array datatype
1036 push @all_bind, $self->_bindtype($k, $v);
1038 else { # literal SQL with bind
1039 my ($sql, @bind) = @$v;
1040 $self->_assert_bindval_matches_bindtype(@bind);
1041 push @all_bind, @bind;
1044 ARRAYREFREF => sub { # literal SQL with bind
1045 my ($sql, @bind) = @${$v};
1046 $self->_assert_bindval_matches_bindtype(@bind);
1047 push @all_bind, @bind;
1049 SCALARREF => sub { # literal SQL without bind
1051 SCALAR_or_UNDEF => sub {
1052 push @all_bind, $self->_bindtype($k, $v);
1063 my(@sql, @sqlq, @sqlv);
1067 if ($ref eq 'HASH') {
1068 for my $k (sort keys %$_) {
1071 my $label = $self->_quote($k);
1072 if ($r eq 'ARRAY') {
1073 # literal SQL with bind
1074 my ($sql, @bind) = @$v;
1075 $self->_assert_bindval_matches_bindtype(@bind);
1076 push @sqlq, "$label = $sql";
1078 } elsif ($r eq 'SCALAR') {
1079 # literal SQL without bind
1080 push @sqlq, "$label = $$v";
1082 push @sqlq, "$label = ?";
1083 push @sqlv, $self->_bindtype($k, $v);
1086 push @sql, $self->_sqlcase('set'), join ', ', @sqlq;
1087 } elsif ($ref eq 'ARRAY') {
1088 # unlike insert(), assume these are ONLY the column names, i.e. for SQL
1091 if ($r eq 'ARRAY') { # literal SQL with bind
1092 my ($sql, @bind) = @$v;
1093 $self->_assert_bindval_matches_bindtype(@bind);
1096 } elsif ($r eq 'SCALAR') { # literal SQL without bind
1097 # embedded literal SQL
1104 push @sql, '(' . join(', ', @sqlq) . ')';
1105 } elsif ($ref eq 'SCALAR') {
1109 # strings get case twiddled
1110 push @sql, $self->_sqlcase($_);
1114 my $sql = join ' ', @sql;
1116 # this is pretty tricky
1117 # if ask for an array, return ($stmt, @bind)
1118 # otherwise, s/?/shift @sqlv/ to put it inline
1120 return ($sql, @sqlv);
1122 1 while $sql =~ s/\?/my $d = shift(@sqlv);
1123 ref $d ? $d->[1] : $d/e;
1132 # This allows us to check for a local, then _form, attr
1134 my($name) = $AUTOLOAD =~ /.*::(.+)/;
1135 return $self->generate($name, @_);
1146 SQL::Abstract - Generate SQL from Perl data structures
1152 my $sql = SQL::Abstract->new;
1154 my($stmt, @bind) = $sql->select($table, \@fields, \%where, \@order);
1156 my($stmt, @bind) = $sql->insert($table, \%fieldvals || \@values);
1158 my($stmt, @bind) = $sql->update($table, \%fieldvals, \%where);
1160 my($stmt, @bind) = $sql->delete($table, \%where);
1162 # Then, use these in your DBI statements
1163 my $sth = $dbh->prepare($stmt);
1164 $sth->execute(@bind);
1166 # Just generate the WHERE clause
1167 my($stmt, @bind) = $sql->where(\%where, \@order);
1169 # Return values in the same order, for hashed queries
1170 # See PERFORMANCE section for more details
1171 my @bind = $sql->values(\%fieldvals);
1175 This module was inspired by the excellent L<DBIx::Abstract>.
1176 However, in using that module I found that what I really wanted
1177 to do was generate SQL, but still retain complete control over my
1178 statement handles and use the DBI interface. So, I set out to
1179 create an abstract SQL generation module.
1181 While based on the concepts used by L<DBIx::Abstract>, there are
1182 several important differences, especially when it comes to WHERE
1183 clauses. I have modified the concepts used to make the SQL easier
1184 to generate from Perl data structures and, IMO, more intuitive.
1185 The underlying idea is for this module to do what you mean, based
1186 on the data structures you provide it. The big advantage is that
1187 you don't have to modify your code every time your data changes,
1188 as this module figures it out.
1190 To begin with, an SQL INSERT is as easy as just specifying a hash
1191 of C<key=value> pairs:
1194 name => 'Jimbo Bobson',
1195 phone => '123-456-7890',
1196 address => '42 Sister Lane',
1197 city => 'St. Louis',
1198 state => 'Louisiana',
1201 The SQL can then be generated with this:
1203 my($stmt, @bind) = $sql->insert('people', \%data);
1205 Which would give you something like this:
1207 $stmt = "INSERT INTO people
1208 (address, city, name, phone, state)
1209 VALUES (?, ?, ?, ?, ?)";
1210 @bind = ('42 Sister Lane', 'St. Louis', 'Jimbo Bobson',
1211 '123-456-7890', 'Louisiana');
1213 These are then used directly in your DBI code:
1215 my $sth = $dbh->prepare($stmt);
1216 $sth->execute(@bind);
1218 =head2 Inserting and Updating Arrays
1220 If your database has array types (like for example Postgres),
1221 activate the special option C<< array_datatypes => 1 >>
1222 when creating the C<SQL::Abstract> object.
1223 Then you may use an arrayref to insert and update database array types:
1225 my $sql = SQL::Abstract->new(array_datatypes => 1);
1227 planets => [qw/Mercury Venus Earth Mars/]
1230 my($stmt, @bind) = $sql->insert('solar_system', \%data);
1234 $stmt = "INSERT INTO solar_system (planets) VALUES (?)"
1236 @bind = (['Mercury', 'Venus', 'Earth', 'Mars']);
1239 =head2 Inserting and Updating SQL
1241 In order to apply SQL functions to elements of your C<%data> you may
1242 specify a reference to an arrayref for the given hash value. For example,
1243 if you need to execute the Oracle C<to_date> function on a value, you can
1244 say something like this:
1248 date_entered => \["to_date(?,'MM/DD/YYYY')", "03/02/2003"],
1251 The first value in the array is the actual SQL. Any other values are
1252 optional and would be included in the bind values array. This gives
1255 my($stmt, @bind) = $sql->insert('people', \%data);
1257 $stmt = "INSERT INTO people (name, date_entered)
1258 VALUES (?, to_date(?,'MM/DD/YYYY'))";
1259 @bind = ('Bill', '03/02/2003');
1261 An UPDATE is just as easy, all you change is the name of the function:
1263 my($stmt, @bind) = $sql->update('people', \%data);
1265 Notice that your C<%data> isn't touched; the module will generate
1266 the appropriately quirky SQL for you automatically. Usually you'll
1267 want to specify a WHERE clause for your UPDATE, though, which is
1268 where handling C<%where> hashes comes in handy...
1270 =head2 Complex where statements
1272 This module can generate pretty complicated WHERE statements
1273 easily. For example, simple C<key=value> pairs are taken to mean
1274 equality, and if you want to see if a field is within a set
1275 of values, you can use an arrayref. Let's say we wanted to
1276 SELECT some data based on this criteria:
1279 requestor => 'inna',
1280 worker => ['nwiger', 'rcwe', 'sfz'],
1281 status => { '!=', 'completed' }
1284 my($stmt, @bind) = $sql->select('tickets', '*', \%where);
1286 The above would give you something like this:
1288 $stmt = "SELECT * FROM tickets WHERE
1289 ( requestor = ? ) AND ( status != ? )
1290 AND ( worker = ? OR worker = ? OR worker = ? )";
1291 @bind = ('inna', 'completed', 'nwiger', 'rcwe', 'sfz');
1293 Which you could then use in DBI code like so:
1295 my $sth = $dbh->prepare($stmt);
1296 $sth->execute(@bind);
1302 The functions are simple. There's one for each major SQL operation,
1303 and a constructor you use first. The arguments are specified in a
1304 similar order to each function (table, then fields, then a where
1305 clause) to try and simplify things.
1310 =head2 new(option => 'value')
1312 The C<new()> function takes a list of options and values, and returns
1313 a new B<SQL::Abstract> object which can then be used to generate SQL
1314 through the methods below. The options accepted are:
1320 If set to 'lower', then SQL will be generated in all lowercase. By
1321 default SQL is generated in "textbook" case meaning something like:
1323 SELECT a_field FROM a_table WHERE some_field LIKE '%someval%'
1325 Any setting other than 'lower' is ignored.
1329 This determines what the default comparison operator is. By default
1330 it is C<=>, meaning that a hash like this:
1332 %where = (name => 'nwiger', email => 'nate@wiger.org');
1334 Will generate SQL like this:
1336 WHERE name = 'nwiger' AND email = 'nate@wiger.org'
1338 However, you may want loose comparisons by default, so if you set
1339 C<cmp> to C<like> you would get SQL such as:
1341 WHERE name like 'nwiger' AND email like 'nate@wiger.org'
1343 You can also override the comparsion on an individual basis - see
1344 the huge section on L</"WHERE CLAUSES"> at the bottom.
1346 =item sqltrue, sqlfalse
1348 Expressions for inserting boolean values within SQL statements.
1349 By default these are C<1=1> and C<1=0>. They are used
1350 by the special operators C<-in> and C<-not_in> for generating
1351 correct SQL even when the argument is an empty array (see below).
1355 This determines the default logical operator for multiple WHERE
1356 statements in arrays or hashes. If absent, the default logic is "or"
1357 for arrays, and "and" for hashes. This means that a WHERE
1361 event_date => {'>=', '2/13/99'},
1362 event_date => {'<=', '4/24/03'},
1365 will generate SQL like this:
1367 WHERE event_date >= '2/13/99' OR event_date <= '4/24/03'
1369 This is probably not what you want given this query, though (look
1370 at the dates). To change the "OR" to an "AND", simply specify:
1372 my $sql = SQL::Abstract->new(logic => 'and');
1374 Which will change the above C<WHERE> to:
1376 WHERE event_date >= '2/13/99' AND event_date <= '4/24/03'
1378 The logic can also be changed locally by inserting
1379 a modifier in front of an arrayref :
1381 @where = (-and => [event_date => {'>=', '2/13/99'},
1382 event_date => {'<=', '4/24/03'} ]);
1384 See the L</"WHERE CLAUSES"> section for explanations.
1388 This will automatically convert comparisons using the specified SQL
1389 function for both column and value. This is mostly used with an argument
1390 of C<upper> or C<lower>, so that the SQL will have the effect of
1391 case-insensitive "searches". For example, this:
1393 $sql = SQL::Abstract->new(convert => 'upper');
1394 %where = (keywords => 'MaKe iT CAse inSeNSItive');
1396 Will turn out the following SQL:
1398 WHERE upper(keywords) like upper('MaKe iT CAse inSeNSItive')
1400 The conversion can be C<upper()>, C<lower()>, or any other SQL function
1401 that can be applied symmetrically to fields (actually B<SQL::Abstract> does
1402 not validate this option; it will just pass through what you specify verbatim).
1406 This is a kludge because many databases suck. For example, you can't
1407 just bind values using DBI's C<execute()> for Oracle C<CLOB> or C<BLOB> fields.
1408 Instead, you have to use C<bind_param()>:
1410 $sth->bind_param(1, 'reg data');
1411 $sth->bind_param(2, $lots, {ora_type => ORA_CLOB});
1413 The problem is, B<SQL::Abstract> will normally just return a C<@bind> array,
1414 which loses track of which field each slot refers to. Fear not.
1416 If you specify C<bindtype> in new, you can determine how C<@bind> is returned.
1417 Currently, you can specify either C<normal> (default) or C<columns>. If you
1418 specify C<columns>, you will get an array that looks like this:
1420 my $sql = SQL::Abstract->new(bindtype => 'columns');
1421 my($stmt, @bind) = $sql->insert(...);
1424 [ 'column1', 'value1' ],
1425 [ 'column2', 'value2' ],
1426 [ 'column3', 'value3' ],
1429 You can then iterate through this manually, using DBI's C<bind_param()>.
1431 $sth->prepare($stmt);
1434 my($col, $data) = @$_;
1435 if ($col eq 'details' || $col eq 'comments') {
1436 $sth->bind_param($i, $data, {ora_type => ORA_CLOB});
1437 } elsif ($col eq 'image') {
1438 $sth->bind_param($i, $data, {ora_type => ORA_BLOB});
1440 $sth->bind_param($i, $data);
1444 $sth->execute; # execute without @bind now
1446 Now, why would you still use B<SQL::Abstract> if you have to do this crap?
1447 Basically, the advantage is still that you don't have to care which fields
1448 are or are not included. You could wrap that above C<for> loop in a simple
1449 sub called C<bind_fields()> or something and reuse it repeatedly. You still
1450 get a layer of abstraction over manual SQL specification.
1452 Note that if you set L</bindtype> to C<columns>, the C<\[$sql, @bind]>
1453 construct (see L</Literal SQL with placeholders and bind values (subqueries)>)
1454 will expect the bind values in this format.
1458 This is the character that a table or column name will be quoted
1459 with. By default this is an empty string, but you could set it to
1460 the character C<`>, to generate SQL like this:
1462 SELECT `a_field` FROM `a_table` WHERE `some_field` LIKE '%someval%'
1464 Alternatively, you can supply an array ref of two items, the first being the left
1465 hand quote character, and the second the right hand quote character. For
1466 example, you could supply C<['[',']']> for SQL Server 2000 compliant quotes
1467 that generates SQL like this:
1469 SELECT [a_field] FROM [a_table] WHERE [some_field] LIKE '%someval%'
1471 Quoting is useful if you have tables or columns names that are reserved
1472 words in your database's SQL dialect.
1476 This is the character that separates a table and column name. It is
1477 necessary to specify this when the C<quote_char> option is selected,
1478 so that tables and column names can be individually quoted like this:
1480 SELECT `table`.`one_field` FROM `table` WHERE `table`.`other_field` = 1
1482 =item injection_guard
1484 A regular expression C<qr/.../> that is applied to any C<-function> and unquoted
1485 column name specified in a query structure. This is a safety mechanism to avoid
1486 injection attacks when mishandling user input e.g.:
1488 my %condition_as_column_value_pairs = get_values_from_user();
1489 $sqla->select( ... , \%condition_as_column_value_pairs );
1491 If the expression matches an exception is thrown. Note that literal SQL
1492 supplied via C<\'...'> or C<\['...']> is B<not> checked in any way.
1494 Defaults to checking for C<;> and the C<GO> keyword (TransactSQL)
1496 =item array_datatypes
1498 When this option is true, arrayrefs in INSERT or UPDATE are
1499 interpreted as array datatypes and are passed directly
1501 When this option is false, arrayrefs are interpreted
1502 as literal SQL, just like refs to arrayrefs
1503 (but this behavior is for backwards compatibility; when writing
1504 new queries, use the "reference to arrayref" syntax
1510 Takes a reference to a list of "special operators"
1511 to extend the syntax understood by L<SQL::Abstract>.
1512 See section L</"SPECIAL OPERATORS"> for details.
1516 Takes a reference to a list of "unary operators"
1517 to extend the syntax understood by L<SQL::Abstract>.
1518 See section L</"UNARY OPERATORS"> for details.
1524 =head2 insert($table, \@values || \%fieldvals, \%options)
1526 This is the simplest function. You simply give it a table name
1527 and either an arrayref of values or hashref of field/value pairs.
1528 It returns an SQL INSERT statement and a list of bind values.
1529 See the sections on L</"Inserting and Updating Arrays"> and
1530 L</"Inserting and Updating SQL"> for information on how to insert
1531 with those data types.
1533 The optional C<\%options> hash reference may contain additional
1534 options to generate the insert SQL. Currently supported options
1541 Takes either a scalar of raw SQL fields, or an array reference of
1542 field names, and adds on an SQL C<RETURNING> statement at the end.
1543 This allows you to return data generated by the insert statement
1544 (such as row IDs) without performing another C<SELECT> statement.
1545 Note, however, this is not part of the SQL standard and may not
1546 be supported by all database engines.
1550 =head2 update($table, \%fieldvals, \%where)
1552 This takes a table, hashref of field/value pairs, and an optional
1553 hashref L<WHERE clause|/WHERE CLAUSES>. It returns an SQL UPDATE function and a list
1555 See the sections on L</"Inserting and Updating Arrays"> and
1556 L</"Inserting and Updating SQL"> for information on how to insert
1557 with those data types.
1559 =head2 select($source, $fields, $where, $order)
1561 This returns a SQL SELECT statement and associated list of bind values, as
1562 specified by the arguments :
1568 Specification of the 'FROM' part of the statement.
1569 The argument can be either a plain scalar (interpreted as a table
1570 name, will be quoted), or an arrayref (interpreted as a list
1571 of table names, joined by commas, quoted), or a scalarref
1572 (literal table name, not quoted), or a ref to an arrayref
1573 (list of literal table names, joined by commas, not quoted).
1577 Specification of the list of fields to retrieve from
1579 The argument can be either an arrayref (interpreted as a list
1580 of field names, will be joined by commas and quoted), or a
1581 plain scalar (literal SQL, not quoted).
1582 Please observe that this API is not as flexible as for
1583 the first argument C<$table>, for backwards compatibility reasons.
1587 Optional argument to specify the WHERE part of the query.
1588 The argument is most often a hashref, but can also be
1589 an arrayref or plain scalar --
1590 see section L<WHERE clause|/"WHERE CLAUSES"> for details.
1594 Optional argument to specify the ORDER BY part of the query.
1595 The argument can be a scalar, a hashref or an arrayref
1596 -- see section L<ORDER BY clause|/"ORDER BY CLAUSES">
1602 =head2 delete($table, \%where)
1604 This takes a table name and optional hashref L<WHERE clause|/WHERE CLAUSES>.
1605 It returns an SQL DELETE statement and list of bind values.
1607 =head2 where(\%where, \@order)
1609 This is used to generate just the WHERE clause. For example,
1610 if you have an arbitrary data structure and know what the
1611 rest of your SQL is going to look like, but want an easy way
1612 to produce a WHERE clause, use this. It returns an SQL WHERE
1613 clause and list of bind values.
1616 =head2 values(\%data)
1618 This just returns the values from the hash C<%data>, in the same
1619 order that would be returned from any of the other above queries.
1620 Using this allows you to markedly speed up your queries if you
1621 are affecting lots of rows. See below under the L</"PERFORMANCE"> section.
1623 =head2 generate($any, 'number', $of, \@data, $struct, \%types)
1625 Warning: This is an experimental method and subject to change.
1627 This returns arbitrarily generated SQL. It's a really basic shortcut.
1628 It will return two different things, depending on return context:
1630 my($stmt, @bind) = $sql->generate('create table', \$table, \@fields);
1631 my $stmt_and_val = $sql->generate('create table', \$table, \@fields);
1633 These would return the following:
1635 # First calling form
1636 $stmt = "CREATE TABLE test (?, ?)";
1637 @bind = (field1, field2);
1639 # Second calling form
1640 $stmt_and_val = "CREATE TABLE test (field1, field2)";
1642 Depending on what you're trying to do, it's up to you to choose the correct
1643 format. In this example, the second form is what you would want.
1647 $sql->generate('alter session', { nls_date_format => 'MM/YY' });
1651 ALTER SESSION SET nls_date_format = 'MM/YY'
1653 You get the idea. Strings get their case twiddled, but everything
1654 else remains verbatim.
1656 =head1 WHERE CLAUSES
1660 This module uses a variation on the idea from L<DBIx::Abstract>. It
1661 is B<NOT>, repeat I<not> 100% compatible. B<The main logic of this
1662 module is that things in arrays are OR'ed, and things in hashes
1665 The easiest way to explain is to show lots of examples. After
1666 each C<%where> hash shown, it is assumed you used:
1668 my($stmt, @bind) = $sql->where(\%where);
1670 However, note that the C<%where> hash can be used directly in any
1671 of the other functions as well, as described above.
1673 =head2 Key-value pairs
1675 So, let's get started. To begin, a simple hash:
1679 status => 'completed'
1682 Is converted to SQL C<key = val> statements:
1684 $stmt = "WHERE user = ? AND status = ?";
1685 @bind = ('nwiger', 'completed');
1687 One common thing I end up doing is having a list of values that
1688 a field can be in. To do this, simply specify a list inside of
1693 status => ['assigned', 'in-progress', 'pending'];
1696 This simple code will create the following:
1698 $stmt = "WHERE user = ? AND ( status = ? OR status = ? OR status = ? )";
1699 @bind = ('nwiger', 'assigned', 'in-progress', 'pending');
1701 A field associated to an empty arrayref will be considered a
1702 logical false and will generate 0=1.
1704 =head2 Tests for NULL values
1706 If the value part is C<undef> then this is converted to SQL <IS NULL>
1715 $stmt = "WHERE user = ? AND status IS NULL";
1718 To test if a column IS NOT NULL:
1722 status => { '!=', undef },
1725 =head2 Specific comparison operators
1727 If you want to specify a different type of operator for your comparison,
1728 you can use a hashref for a given column:
1732 status => { '!=', 'completed' }
1735 Which would generate:
1737 $stmt = "WHERE user = ? AND status != ?";
1738 @bind = ('nwiger', 'completed');
1740 To test against multiple values, just enclose the values in an arrayref:
1742 status => { '=', ['assigned', 'in-progress', 'pending'] };
1744 Which would give you:
1746 "WHERE status = ? OR status = ? OR status = ?"
1749 The hashref can also contain multiple pairs, in which case it is expanded
1750 into an C<AND> of its elements:
1754 status => { '!=', 'completed', -not_like => 'pending%' }
1757 # Or more dynamically, like from a form
1758 $where{user} = 'nwiger';
1759 $where{status}{'!='} = 'completed';
1760 $where{status}{'-not_like'} = 'pending%';
1762 # Both generate this
1763 $stmt = "WHERE user = ? AND status != ? AND status NOT LIKE ?";
1764 @bind = ('nwiger', 'completed', 'pending%');
1767 To get an OR instead, you can combine it with the arrayref idea:
1771 priority => [ { '=', 2 }, { '>', 5 } ]
1774 Which would generate:
1776 $stmt = "WHERE ( priority = ? OR priority > ? ) AND user = ?";
1777 @bind = ('2', '5', 'nwiger');
1779 If you want to include literal SQL (with or without bind values), just use a
1780 scalar reference or array reference as the value:
1783 date_entered => { '>' => \["to_date(?, 'MM/DD/YYYY')", "11/26/2008"] },
1784 date_expires => { '<' => \"now()" }
1787 Which would generate:
1789 $stmt = "WHERE date_entered > "to_date(?, 'MM/DD/YYYY') AND date_expires < now()";
1790 @bind = ('11/26/2008');
1793 =head2 Logic and nesting operators
1795 In the example above,
1796 there is a subtle trap if you want to say something like
1797 this (notice the C<AND>):
1799 WHERE priority != ? AND priority != ?
1801 Because, in Perl you I<can't> do this:
1803 priority => { '!=', 2, '!=', 1 }
1805 As the second C<!=> key will obliterate the first. The solution
1806 is to use the special C<-modifier> form inside an arrayref:
1808 priority => [ -and => {'!=', 2},
1812 Normally, these would be joined by C<OR>, but the modifier tells it
1813 to use C<AND> instead. (Hint: You can use this in conjunction with the
1814 C<logic> option to C<new()> in order to change the way your queries
1815 work by default.) B<Important:> Note that the C<-modifier> goes
1816 B<INSIDE> the arrayref, as an extra first element. This will
1817 B<NOT> do what you think it might:
1819 priority => -and => [{'!=', 2}, {'!=', 1}] # WRONG!
1821 Here is a quick list of equivalencies, since there is some overlap:
1824 status => {'!=', 'completed', 'not like', 'pending%' }
1825 status => [ -and => {'!=', 'completed'}, {'not like', 'pending%'}]
1828 status => {'=', ['assigned', 'in-progress']}
1829 status => [ -or => {'=', 'assigned'}, {'=', 'in-progress'}]
1830 status => [ {'=', 'assigned'}, {'=', 'in-progress'} ]
1834 =head2 Special operators : IN, BETWEEN, etc.
1836 You can also use the hashref format to compare a list of fields using the
1837 C<IN> comparison operator, by specifying the list as an arrayref:
1840 status => 'completed',
1841 reportid => { -in => [567, 2335, 2] }
1844 Which would generate:
1846 $stmt = "WHERE status = ? AND reportid IN (?,?,?)";
1847 @bind = ('completed', '567', '2335', '2');
1849 The reverse operator C<-not_in> generates SQL C<NOT IN> and is used in
1852 If the argument to C<-in> is an empty array, 'sqlfalse' is generated
1853 (by default : C<1=0>). Similarly, C<< -not_in => [] >> generates
1854 'sqltrue' (by default : C<1=1>).
1856 In addition to the array you can supply a chunk of literal sql or
1857 literal sql with bind:
1860 customer => { -in => \[
1861 'SELECT cust_id FROM cust WHERE balance > ?',
1864 status => { -in => \'SELECT status_codes FROM states' },
1870 customer IN ( SELECT cust_id FROM cust WHERE balance > ? )
1871 AND status IN ( SELECT status_codes FROM states )
1877 Another pair of operators is C<-between> and C<-not_between>,
1878 used with an arrayref of two values:
1882 completion_date => {
1883 -not_between => ['2002-10-01', '2003-02-06']
1889 WHERE user = ? AND completion_date NOT BETWEEN ( ? AND ? )
1891 Just like with C<-in> all plausible combinations of literal SQL
1895 start0 => { -between => [ 1, 2 ] },
1896 start1 => { -between => \["? AND ?", 1, 2] },
1897 start2 => { -between => \"lower(x) AND upper(y)" },
1898 start3 => { -between => [
1900 \["upper(?)", 'stuff' ],
1907 ( start0 BETWEEN ? AND ? )
1908 AND ( start1 BETWEEN ? AND ? )
1909 AND ( start2 BETWEEN lower(x) AND upper(y) )
1910 AND ( start3 BETWEEN lower(x) AND upper(?) )
1912 @bind = (1, 2, 1, 2, 'stuff');
1915 These are the two builtin "special operators"; but the
1916 list can be expanded : see section L</"SPECIAL OPERATORS"> below.
1918 =head2 Unary operators: bool
1920 If you wish to test against boolean columns or functions within your
1921 database you can use the C<-bool> and C<-not_bool> operators. For
1922 example to test the column C<is_user> being true and the column
1923 C<is_enabled> being false you would use:-
1927 -not_bool => 'is_enabled',
1932 WHERE is_user AND NOT is_enabled
1934 If a more complex combination is required, testing more conditions,
1935 then you should use the and/or operators:-
1942 -not_bool => 'four',
1948 WHERE one AND two AND three AND NOT four
1951 =head2 Nested conditions, -and/-or prefixes
1953 So far, we've seen how multiple conditions are joined with a top-level
1954 C<AND>. We can change this by putting the different conditions we want in
1955 hashes and then putting those hashes in an array. For example:
1960 status => { -like => ['pending%', 'dispatched'] },
1964 status => 'unassigned',
1968 This data structure would create the following:
1970 $stmt = "WHERE ( user = ? AND ( status LIKE ? OR status LIKE ? ) )
1971 OR ( user = ? AND status = ? ) )";
1972 @bind = ('nwiger', 'pending', 'dispatched', 'robot', 'unassigned');
1975 Clauses in hashrefs or arrayrefs can be prefixed with an C<-and> or C<-or>
1976 to change the logic inside :
1982 -and => [ workhrs => {'>', 20}, geo => 'ASIA' ],
1983 -or => { workhrs => {'<', 50}, geo => 'EURO' },
1990 WHERE ( user = ? AND (
1991 ( workhrs > ? AND geo = ? )
1992 OR ( workhrs < ? OR geo = ? )
1995 =head3 Algebraic inconsistency, for historical reasons
1997 C<Important note>: when connecting several conditions, the C<-and->|C<-or>
1998 operator goes C<outside> of the nested structure; whereas when connecting
1999 several constraints on one column, the C<-and> operator goes
2000 C<inside> the arrayref. Here is an example combining both features :
2003 -and => [a => 1, b => 2],
2004 -or => [c => 3, d => 4],
2005 e => [-and => {-like => 'foo%'}, {-like => '%bar'} ]
2010 WHERE ( ( ( a = ? AND b = ? )
2011 OR ( c = ? OR d = ? )
2012 OR ( e LIKE ? AND e LIKE ? ) ) )
2014 This difference in syntax is unfortunate but must be preserved for
2015 historical reasons. So be careful : the two examples below would
2016 seem algebraically equivalent, but they are not
2018 {col => [-and => {-like => 'foo%'}, {-like => '%bar'}]}
2019 # yields : WHERE ( ( col LIKE ? AND col LIKE ? ) )
2021 [-and => {col => {-like => 'foo%'}, {col => {-like => '%bar'}}]]
2022 # yields : WHERE ( ( col LIKE ? OR col LIKE ? ) )
2025 =head2 Literal SQL and value type operators
2027 The basic premise of SQL::Abstract is that in WHERE specifications the "left
2028 side" is a column name and the "right side" is a value (normally rendered as
2029 a placeholder). This holds true for both hashrefs and arrayref pairs as you
2030 see in the L</WHERE CLAUSES> examples above. Sometimes it is necessary to
2031 alter this behavior. There are several ways of doing so.
2035 This is a virtual operator that signals the string to its right side is an
2036 identifier (a column name) and not a value. For example to compare two
2037 columns you would write:
2040 priority => { '<', 2 },
2041 requestor => { -ident => 'submitter' },
2046 $stmt = "WHERE priority < ? AND requestor = submitter";
2049 If you are maintaining legacy code you may see a different construct as
2050 described in L</Deprecated usage of Literal SQL>, please use C<-ident> in new
2055 This is a virtual operator that signals that the construct to its right side
2056 is a value to be passed to DBI. This is for example necessary when you want
2057 to write a where clause against an array (for RDBMS that support such
2058 datatypes). For example:
2061 array => { -value => [1, 2, 3] }
2066 $stmt = 'WHERE array = ?';
2067 @bind = ([1, 2, 3]);
2069 Note that if you were to simply say:
2075 the result would porbably be not what you wanted:
2077 $stmt = 'WHERE array = ? OR array = ? OR array = ?';
2082 Finally, sometimes only literal SQL will do. To include a random snippet
2083 of SQL verbatim, you specify it as a scalar reference. Consider this only
2084 as a last resort. Usually there is a better way. For example:
2087 priority => { '<', 2 },
2088 requestor => { -in => \'(SELECT name FROM hitmen)' },
2093 $stmt = "WHERE priority < ? AND requestor IN (SELECT name FROM hitmen)"
2096 Note that in this example, you only get one bind parameter back, since
2097 the verbatim SQL is passed as part of the statement.
2101 Never use untrusted input as a literal SQL argument - this is a massive
2102 security risk (there is no way to check literal snippets for SQL
2103 injections and other nastyness). If you need to deal with untrusted input
2104 use literal SQL with placeholders as described next.
2106 =head3 Literal SQL with placeholders and bind values (subqueries)
2108 If the literal SQL to be inserted has placeholders and bind values,
2109 use a reference to an arrayref (yes this is a double reference --
2110 not so common, but perfectly legal Perl). For example, to find a date
2111 in Postgres you can use something like this:
2114 date_column => \[q/= date '2008-09-30' - ?::integer/, 10/]
2119 $stmt = "WHERE ( date_column = date '2008-09-30' - ?::integer )"
2122 Note that you must pass the bind values in the same format as they are returned
2123 by L</where>. That means that if you set L</bindtype> to C<columns>, you must
2124 provide the bind values in the C<< [ column_meta => value ] >> format, where
2125 C<column_meta> is an opaque scalar value; most commonly the column name, but
2126 you can use any scalar value (including references and blessed references),
2127 L<SQL::Abstract> will simply pass it through intact. So if C<bindtype> is set
2128 to C<columns> the above example will look like:
2131 date_column => \[q/= date '2008-09-30' - ?::integer/, [ dummy => 10 ]/]
2134 Literal SQL is especially useful for nesting parenthesized clauses in the
2135 main SQL query. Here is a first example :
2137 my ($sub_stmt, @sub_bind) = ("SELECT c1 FROM t1 WHERE c2 < ? AND c3 LIKE ?",
2141 bar => \["IN ($sub_stmt)" => @sub_bind],
2146 $stmt = "WHERE (foo = ? AND bar IN (SELECT c1 FROM t1
2147 WHERE c2 < ? AND c3 LIKE ?))";
2148 @bind = (1234, 100, "foo%");
2150 Other subquery operators, like for example C<"E<gt> ALL"> or C<"NOT IN">,
2151 are expressed in the same way. Of course the C<$sub_stmt> and
2152 its associated bind values can be generated through a former call
2155 my ($sub_stmt, @sub_bind)
2156 = $sql->select("t1", "c1", {c2 => {"<" => 100},
2157 c3 => {-like => "foo%"}});
2160 bar => \["> ALL ($sub_stmt)" => @sub_bind],
2163 In the examples above, the subquery was used as an operator on a column;
2164 but the same principle also applies for a clause within the main C<%where>
2165 hash, like an EXISTS subquery :
2167 my ($sub_stmt, @sub_bind)
2168 = $sql->select("t1", "*", {c1 => 1, c2 => \"> t0.c0"});
2169 my %where = ( -and => [
2171 \["EXISTS ($sub_stmt)" => @sub_bind],
2176 $stmt = "WHERE (foo = ? AND EXISTS (SELECT * FROM t1
2177 WHERE c1 = ? AND c2 > t0.c0))";
2181 Observe that the condition on C<c2> in the subquery refers to
2182 column C<t0.c0> of the main query : this is I<not> a bind
2183 value, so we have to express it through a scalar ref.
2184 Writing C<< c2 => {">" => "t0.c0"} >> would have generated
2185 C<< c2 > ? >> with bind value C<"t0.c0"> ... not exactly
2186 what we wanted here.
2188 Finally, here is an example where a subquery is used
2189 for expressing unary negation:
2191 my ($sub_stmt, @sub_bind)
2192 = $sql->where({age => [{"<" => 10}, {">" => 20}]});
2193 $sub_stmt =~ s/^ where //i; # don't want "WHERE" in the subclause
2195 lname => {like => '%son%'},
2196 \["NOT ($sub_stmt)" => @sub_bind],
2201 $stmt = "lname LIKE ? AND NOT ( age < ? OR age > ? )"
2202 @bind = ('%son%', 10, 20)
2204 =head3 Deprecated usage of Literal SQL
2206 Below are some examples of archaic use of literal SQL. It is shown only as
2207 reference for those who deal with legacy code. Each example has a much
2208 better, cleaner and safer alternative that users should opt for in new code.
2214 my %where = ( requestor => \'IS NOT NULL' )
2216 $stmt = "WHERE requestor IS NOT NULL"
2218 This used to be the way of generating NULL comparisons, before the handling
2219 of C<undef> got formalized. For new code please use the superior syntax as
2220 described in L</Tests for NULL values>.
2224 my %where = ( requestor => \'= submitter' )
2226 $stmt = "WHERE requestor = submitter"
2228 This used to be the only way to compare columns. Use the superior L</-ident>
2229 method for all new code. For example an identifier declared in such a way
2230 will be properly quoted if L</quote_char> is properly set, while the legacy
2231 form will remain as supplied.
2235 my %where = ( is_ready => \"", completed => { '>', '2012-12-21' } )
2237 $stmt = "WHERE completed > ? AND is_ready"
2238 @bind = ('2012-12-21')
2240 Using an empty string literal used to be the only way to express a boolean.
2241 For all new code please use the much more readable
2242 L<-bool|/Unary operators: bool> operator.
2248 These pages could go on for a while, since the nesting of the data
2249 structures this module can handle are pretty much unlimited (the
2250 module implements the C<WHERE> expansion as a recursive function
2251 internally). Your best bet is to "play around" with the module a
2252 little to see how the data structures behave, and choose the best
2253 format for your data based on that.
2255 And of course, all the values above will probably be replaced with
2256 variables gotten from forms or the command line. After all, if you
2257 knew everything ahead of time, you wouldn't have to worry about
2258 dynamically-generating SQL and could just hardwire it into your
2261 =head1 ORDER BY CLAUSES
2263 Some functions take an order by clause. This can either be a scalar (just a
2264 column name,) a hash of C<< { -desc => 'col' } >> or C<< { -asc => 'col' } >>,
2265 or an array of either of the two previous forms. Examples:
2267 Given | Will Generate
2268 ----------------------------------------------------------
2270 \'colA DESC' | ORDER BY colA DESC
2272 'colA' | ORDER BY colA
2274 [qw/colA colB/] | ORDER BY colA, colB
2276 {-asc => 'colA'} | ORDER BY colA ASC
2278 {-desc => 'colB'} | ORDER BY colB DESC
2280 ['colA', {-asc => 'colB'}] | ORDER BY colA, colB ASC
2282 { -asc => [qw/colA colB/] } | ORDER BY colA ASC, colB ASC
2285 { -asc => 'colA' }, | ORDER BY colA ASC, colB DESC,
2286 { -desc => [qw/colB/], | colC ASC, colD ASC
2287 { -asc => [qw/colC colD/],|
2289 ===========================================================
2293 =head1 SPECIAL OPERATORS
2295 my $sqlmaker = SQL::Abstract->new(special_ops => [
2299 my ($self, $field, $op, $arg) = @_;
2305 handler => 'method_name',
2309 A "special operator" is a SQL syntactic clause that can be
2310 applied to a field, instead of a usual binary operator.
2313 WHERE field IN (?, ?, ?)
2314 WHERE field BETWEEN ? AND ?
2315 WHERE MATCH(field) AGAINST (?, ?)
2317 Special operators IN and BETWEEN are fairly standard and therefore
2318 are builtin within C<SQL::Abstract> (as the overridable methods
2319 C<_where_field_IN> and C<_where_field_BETWEEN>). For other operators,
2320 like the MATCH .. AGAINST example above which is specific to MySQL,
2321 you can write your own operator handlers - supply a C<special_ops>
2322 argument to the C<new> method. That argument takes an arrayref of
2323 operator definitions; each operator definition is a hashref with two
2330 the regular expression to match the operator
2334 Either a coderef or a plain scalar method name. In both cases
2335 the expected return is C<< ($sql, @bind) >>.
2337 When supplied with a method name, it is simply called on the
2338 L<SQL::Abstract/> object as:
2340 $self->$method_name ($field, $op, $arg)
2344 $op is the part that matched the handler regex
2345 $field is the LHS of the operator
2348 When supplied with a coderef, it is called as:
2350 $coderef->($self, $field, $op, $arg)
2355 For example, here is an implementation
2356 of the MATCH .. AGAINST syntax for MySQL
2358 my $sqlmaker = SQL::Abstract->new(special_ops => [
2360 # special op for MySql MATCH (field) AGAINST(word1, word2, ...)
2361 {regex => qr/^match$/i,
2363 my ($self, $field, $op, $arg) = @_;
2364 $arg = [$arg] if not ref $arg;
2365 my $label = $self->_quote($field);
2366 my ($placeholder) = $self->_convert('?');
2367 my $placeholders = join ", ", (($placeholder) x @$arg);
2368 my $sql = $self->_sqlcase('match') . " ($label) "
2369 . $self->_sqlcase('against') . " ($placeholders) ";
2370 my @bind = $self->_bindtype($field, @$arg);
2371 return ($sql, @bind);
2378 =head1 UNARY OPERATORS
2380 my $sqlmaker = SQL::Abstract->new(unary_ops => [
2384 my ($self, $op, $arg) = @_;
2390 handler => 'method_name',
2394 A "unary operator" is a SQL syntactic clause that can be
2395 applied to a field - the operator goes before the field
2397 You can write your own operator handlers - supply a C<unary_ops>
2398 argument to the C<new> method. That argument takes an arrayref of
2399 operator definitions; each operator definition is a hashref with two
2406 the regular expression to match the operator
2410 Either a coderef or a plain scalar method name. In both cases
2411 the expected return is C<< $sql >>.
2413 When supplied with a method name, it is simply called on the
2414 L<SQL::Abstract/> object as:
2416 $self->$method_name ($op, $arg)
2420 $op is the part that matched the handler regex
2421 $arg is the RHS or argument of the operator
2423 When supplied with a coderef, it is called as:
2425 $coderef->($self, $op, $arg)
2433 Thanks to some benchmarking by Mark Stosberg, it turns out that
2434 this module is many orders of magnitude faster than using C<DBIx::Abstract>.
2435 I must admit this wasn't an intentional design issue, but it's a
2436 byproduct of the fact that you get to control your C<DBI> handles
2439 To maximize performance, use a code snippet like the following:
2441 # prepare a statement handle using the first row
2442 # and then reuse it for the rest of the rows
2444 for my $href (@array_of_hashrefs) {
2445 $stmt ||= $sql->insert('table', $href);
2446 $sth ||= $dbh->prepare($stmt);
2447 $sth->execute($sql->values($href));
2450 The reason this works is because the keys in your C<$href> are sorted
2451 internally by B<SQL::Abstract>. Thus, as long as your data retains
2452 the same structure, you only have to generate the SQL the first time
2453 around. On subsequent queries, simply use the C<values> function provided
2454 by this module to return your values in the correct order.
2456 However this depends on the values having the same type - if, for
2457 example, the values of a where clause may either have values
2458 (resulting in sql of the form C<column = ?> with a single bind
2459 value), or alternatively the values might be C<undef> (resulting in
2460 sql of the form C<column IS NULL> with no bind value) then the
2461 caching technique suggested will not work.
2465 If you use my C<CGI::FormBuilder> module at all, you'll hopefully
2466 really like this part (I do, at least). Building up a complex query
2467 can be as simple as the following:
2471 use CGI::FormBuilder;
2474 my $form = CGI::FormBuilder->new(...);
2475 my $sql = SQL::Abstract->new;
2477 if ($form->submitted) {
2478 my $field = $form->field;
2479 my $id = delete $field->{id};
2480 my($stmt, @bind) = $sql->update('table', $field, {id => $id});
2483 Of course, you would still have to connect using C<DBI> to run the
2484 query, but the point is that if you make your form look like your
2485 table, the actual query script can be extremely simplistic.
2487 If you're B<REALLY> lazy (I am), check out C<HTML::QuickTable> for
2488 a fast interface to returning and formatting data. I frequently
2489 use these three modules together to write complex database query
2490 apps in under 50 lines.
2496 =item * gitweb: L<http://git.shadowcat.co.uk/gitweb/gitweb.cgi?p=dbsrgits/SQL-Abstract.git>
2498 =item * git: L<git://git.shadowcat.co.uk/dbsrgits/SQL-Abstract.git>
2504 Version 1.50 was a major internal refactoring of C<SQL::Abstract>.
2505 Great care has been taken to preserve the I<published> behavior
2506 documented in previous versions in the 1.* family; however,
2507 some features that were previously undocumented, or behaved
2508 differently from the documentation, had to be changed in order
2509 to clarify the semantics. Hence, client code that was relying
2510 on some dark areas of C<SQL::Abstract> v1.*
2511 B<might behave differently> in v1.50.
2513 The main changes are :
2519 support for literal SQL through the C<< \ [$sql, bind] >> syntax.
2523 support for the { operator => \"..." } construct (to embed literal SQL)
2527 support for the { operator => \["...", @bind] } construct (to embed literal SQL with bind values)
2531 optional support for L<array datatypes|/"Inserting and Updating Arrays">
2535 defensive programming : check arguments
2539 fixed bug with global logic, which was previously implemented
2540 through global variables yielding side-effects. Prior versions would
2541 interpret C<< [ {cond1, cond2}, [cond3, cond4] ] >>
2542 as C<< "(cond1 AND cond2) OR (cond3 AND cond4)" >>.
2543 Now this is interpreted
2544 as C<< "(cond1 AND cond2) OR (cond3 OR cond4)" >>.
2549 fixed semantics of _bindtype on array args
2553 dropped the C<_anoncopy> of the %where tree. No longer necessary,
2554 we just avoid shifting arrays within that tree.
2558 dropped the C<_modlogic> function
2562 =head1 ACKNOWLEDGEMENTS
2564 There are a number of individuals that have really helped out with
2565 this module. Unfortunately, most of them submitted bugs via CPAN
2566 so I have no idea who they are! But the people I do know are:
2568 Ash Berlin (order_by hash term support)
2569 Matt Trout (DBIx::Class support)
2570 Mark Stosberg (benchmarking)
2571 Chas Owens (initial "IN" operator support)
2572 Philip Collins (per-field SQL functions)
2573 Eric Kolve (hashref "AND" support)
2574 Mike Fragassi (enhancements to "BETWEEN" and "LIKE")
2575 Dan Kubb (support for "quote_char" and "name_sep")
2576 Guillermo Roditi (patch to cleanup "IN" and "BETWEEN", fix and tests for _order_by)
2577 Laurent Dami (internal refactoring, extensible list of special operators, literal SQL)
2578 Norbert Buchmuller (support for literal SQL in hashpair, misc. fixes & tests)
2579 Peter Rabbitson (rewrite of SQLA::Test, misc. fixes & tests)
2580 Oliver Charles (support for "RETURNING" after "INSERT")
2586 L<DBIx::Class>, L<DBIx::Abstract>, L<CGI::FormBuilder>, L<HTML::QuickTable>.
2590 Copyright (c) 2001-2007 Nathan Wiger <nwiger@cpan.org>. All Rights Reserved.
2592 This module is actively maintained by Matt Trout <mst@shadowcatsystems.co.uk>
2594 For support, your best bet is to try the C<DBIx::Class> users mailing list.
2595 While not an official support venue, C<DBIx::Class> makes heavy use of
2596 C<SQL::Abstract>, and as such list members there are very familiar with
2597 how to create queries.
2601 This module is free software; you may copy this under the same
2602 terms as perl itself (either the GNU General Public License or
2603 the Artistic License)