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)
11 use List::Util qw/first/;
12 use Scalar::Util qw/blessed/;
14 #======================================================================
16 #======================================================================
18 our $VERSION = '1.58';
20 # This would confuse some packagers
21 #$VERSION = eval $VERSION; # numify for warning-free dev releases
25 # special operators (-in, -between). May be extended/overridden by user.
26 # See section WHERE: BUILTIN SPECIAL OPERATORS below for implementation
27 my @BUILTIN_SPECIAL_OPS = (
28 {regex => qr/^(not )?between$/i, handler => '_where_field_BETWEEN'},
29 {regex => qr/^(not )?in$/i, handler => '_where_field_IN'},
32 # unaryish operators - key maps to handler
33 my @BUILTIN_UNARY_OPS = (
34 # the digits are backcompat stuff
35 { regex => qr/^and (?: \s? \d+ )? $/xi, handler => '_where_op_ANDOR' },
36 { regex => qr/^or (?: \s? \d+ )? $/xi, handler => '_where_op_ANDOR' },
37 { regex => qr/^nest (?: \s? \d+ )? $/xi, handler => '_where_op_NEST' },
38 { regex => qr/^ (?: not \s )? bool $/xi, handler => '_where_op_BOOL' },
41 #======================================================================
42 # DEBUGGING AND ERROR REPORTING
43 #======================================================================
46 return unless $_[0]->{debug}; shift; # a little faster
47 my $func = (caller(1))[3];
48 warn "[$func] ", @_, "\n";
52 my($func) = (caller(1))[3];
53 carp "[$func] Warning: ", @_;
57 my($func) = (caller(1))[3];
58 croak "[$func] Fatal: ", @_;
62 #======================================================================
64 #======================================================================
68 my $class = ref($self) || $self;
69 my %opt = (ref $_[0] eq 'HASH') ? %{$_[0]} : @_;
71 # choose our case by keeping an option around
72 delete $opt{case} if $opt{case} && $opt{case} ne 'lower';
74 # default logic for interpreting arrayrefs
75 $opt{logic} = $opt{logic} ? uc $opt{logic} : 'OR';
77 # how to return bind vars
78 # LDNOTE: changed nwiger code : why this 'delete' ??
79 # $opt{bindtype} ||= delete($opt{bind_type}) || 'normal';
80 $opt{bindtype} ||= 'normal';
82 # default comparison is "=", but can be overridden
85 # try to recognize which are the 'equality' and 'unequality' ops
86 # (temporary quickfix, should go through a more seasoned API)
87 $opt{equality_op} = qr/^(\Q$opt{cmp}\E|is|(is\s+)?like)$/i;
88 $opt{inequality_op} = qr/^(!=|<>|(is\s+)?not(\s+like)?)$/i;
91 $opt{sqltrue} ||= '1=1';
92 $opt{sqlfalse} ||= '0=1';
95 $opt{special_ops} ||= [];
96 push @{$opt{special_ops}}, @BUILTIN_SPECIAL_OPS;
99 $opt{unary_ops} ||= [];
100 push @{$opt{unary_ops}}, @BUILTIN_UNARY_OPS;
102 return bless \%opt, $class;
107 #======================================================================
109 #======================================================================
113 my $table = $self->_table(shift);
114 my $data = shift || return;
116 my $method = $self->_METHOD_FOR_refkind("_insert", $data);
117 my ($sql, @bind) = $self->$method($data);
118 $sql = join " ", $self->_sqlcase('insert into'), $table, $sql;
119 return wantarray ? ($sql, @bind) : $sql;
122 sub _insert_HASHREF { # explicit list of fields and then values
123 my ($self, $data) = @_;
125 my @fields = sort keys %$data;
127 my ($sql, @bind) = $self->_insert_values($data);
130 $_ = $self->_quote($_) foreach @fields;
131 $sql = "( ".join(", ", @fields).") ".$sql;
133 return ($sql, @bind);
136 sub _insert_ARRAYREF { # just generate values(?,?) part (no list of fields)
137 my ($self, $data) = @_;
139 # no names (arrayref) so can't generate bindtype
140 $self->{bindtype} ne 'columns'
141 or belch "can't do 'columns' bindtype when called with arrayref";
143 # fold the list of values into a hash of column name - value pairs
144 # (where the column names are artificially generated, and their
145 # lexicographical ordering keep the ordering of the original list)
146 my $i = "a"; # incremented values will be in lexicographical order
147 my $data_in_hash = { map { ($i++ => $_) } @$data };
149 return $self->_insert_values($data_in_hash);
152 sub _insert_ARRAYREFREF { # literal SQL with bind
153 my ($self, $data) = @_;
155 my ($sql, @bind) = @${$data};
156 $self->_assert_bindval_matches_bindtype(@bind);
158 return ($sql, @bind);
162 sub _insert_SCALARREF { # literal SQL without bind
163 my ($self, $data) = @_;
169 my ($self, $data) = @_;
171 my (@values, @all_bind);
172 foreach my $column (sort keys %$data) {
173 my $v = $data->{$column};
175 $self->_SWITCH_refkind($v, {
178 if ($self->{array_datatypes}) { # if array datatype are activated
180 push @all_bind, $self->_bindtype($column, $v);
182 else { # else literal SQL with bind
183 my ($sql, @bind) = @$v;
184 $self->_assert_bindval_matches_bindtype(@bind);
186 push @all_bind, @bind;
190 ARRAYREFREF => sub { # literal SQL with bind
191 my ($sql, @bind) = @${$v};
192 $self->_assert_bindval_matches_bindtype(@bind);
194 push @all_bind, @bind;
197 # THINK : anything useful to do with a HASHREF ?
198 HASHREF => sub { # (nothing, but old SQLA passed it through)
199 #TODO in SQLA >= 2.0 it will die instead
200 belch "HASH ref as bind value in insert is not supported";
202 push @all_bind, $self->_bindtype($column, $v);
205 SCALARREF => sub { # literal SQL without bind
209 SCALAR_or_UNDEF => sub {
211 push @all_bind, $self->_bindtype($column, $v);
218 my $sql = $self->_sqlcase('values')." ( ".join(", ", @values)." )";
219 return ($sql, @all_bind);
224 #======================================================================
226 #======================================================================
231 my $table = $self->_table(shift);
232 my $data = shift || return;
235 # first build the 'SET' part of the sql statement
236 my (@set, @all_bind);
237 puke "Unsupported data type specified to \$sql->update"
238 unless ref $data eq 'HASH';
240 for my $k (sort keys %$data) {
243 my $label = $self->_quote($k);
245 $self->_SWITCH_refkind($v, {
247 if ($self->{array_datatypes}) { # array datatype
248 push @set, "$label = ?";
249 push @all_bind, $self->_bindtype($k, $v);
251 else { # literal SQL with bind
252 my ($sql, @bind) = @$v;
253 $self->_assert_bindval_matches_bindtype(@bind);
254 push @set, "$label = $sql";
255 push @all_bind, @bind;
258 ARRAYREFREF => sub { # literal SQL with bind
259 my ($sql, @bind) = @${$v};
260 $self->_assert_bindval_matches_bindtype(@bind);
261 push @set, "$label = $sql";
262 push @all_bind, @bind;
264 SCALARREF => sub { # literal SQL without bind
265 push @set, "$label = $$v";
267 SCALAR_or_UNDEF => sub {
268 push @set, "$label = ?";
269 push @all_bind, $self->_bindtype($k, $v);
275 my $sql = $self->_sqlcase('update') . " $table " . $self->_sqlcase('set ')
279 my($where_sql, @where_bind) = $self->where($where);
281 push @all_bind, @where_bind;
284 return wantarray ? ($sql, @all_bind) : $sql;
290 #======================================================================
292 #======================================================================
297 my $table = $self->_table(shift);
298 my $fields = shift || '*';
302 my($where_sql, @bind) = $self->where($where, $order);
304 my $f = (ref $fields eq 'ARRAY') ? join ', ', map { $self->_quote($_) } @$fields
306 my $sql = join(' ', $self->_sqlcase('select'), $f,
307 $self->_sqlcase('from'), $table)
310 return wantarray ? ($sql, @bind) : $sql;
313 #======================================================================
315 #======================================================================
320 my $table = $self->_table(shift);
324 my($where_sql, @bind) = $self->where($where);
325 my $sql = $self->_sqlcase('delete from') . " $table" . $where_sql;
327 return wantarray ? ($sql, @bind) : $sql;
331 #======================================================================
333 #======================================================================
337 # Finally, a separate routine just to handle WHERE clauses
339 my ($self, $where, $order) = @_;
342 my ($sql, @bind) = $self->_recurse_where($where);
343 $sql = $sql ? $self->_sqlcase(' where ') . "( $sql )" : '';
347 $sql .= $self->_order_by($order);
350 return wantarray ? ($sql, @bind) : $sql;
355 my ($self, $where, $logic) = @_;
357 # dispatch on appropriate method according to refkind of $where
358 my $method = $self->_METHOD_FOR_refkind("_where", $where);
361 my ($sql, @bind) = $self->$method($where, $logic);
363 # DBIx::Class directly calls _recurse_where in scalar context, so
364 # we must implement it, even if not in the official API
365 return wantarray ? ($sql, @bind) : $sql;
370 #======================================================================
371 # WHERE: top-level ARRAYREF
372 #======================================================================
375 sub _where_ARRAYREF {
376 my ($self, $where, $logic) = @_;
378 $logic = uc($logic || $self->{logic});
379 $logic eq 'AND' or $logic eq 'OR' or puke "unknown logic: $logic";
381 my @clauses = @$where;
383 my (@sql_clauses, @all_bind);
384 # need to use while() so can shift() for pairs
385 while (my $el = shift @clauses) {
387 # switch according to kind of $el and get corresponding ($sql, @bind)
388 my ($sql, @bind) = $self->_SWITCH_refkind($el, {
390 # skip empty elements, otherwise get invalid trailing AND stuff
391 ARRAYREF => sub {$self->_recurse_where($el) if @$el},
393 ARRAYREFREF => sub { @{${$el}} if @{${$el}}},
395 HASHREF => sub {$self->_recurse_where($el, 'and') if %$el},
396 # LDNOTE : previous SQLA code for hashrefs was creating a dirty
397 # side-effect: the first hashref within an array would change
398 # the global logic to 'AND'. So [ {cond1, cond2}, [cond3, cond4] ]
399 # was interpreted as "(cond1 AND cond2) OR (cond3 AND cond4)",
400 # whereas it should be "(cond1 AND cond2) OR (cond3 OR cond4)".
402 SCALARREF => sub { ($$el); },
404 SCALAR => sub {# top-level arrayref with scalars, recurse in pairs
405 $self->_recurse_where({$el => shift(@clauses)})},
407 UNDEF => sub {puke "not supported : UNDEF in arrayref" },
411 push @sql_clauses, $sql;
412 push @all_bind, @bind;
416 return $self->_join_sql_clauses($logic, \@sql_clauses, \@all_bind);
419 #======================================================================
420 # WHERE: top-level ARRAYREFREF
421 #======================================================================
423 sub _where_ARRAYREFREF {
424 my ($self, $where) = @_;
425 my ($sql, @bind) = @{${$where}};
427 return ($sql, @bind);
430 #======================================================================
431 # WHERE: top-level HASHREF
432 #======================================================================
435 my ($self, $where) = @_;
436 my (@sql_clauses, @all_bind);
438 for my $k (sort keys %$where) {
439 my $v = $where->{$k};
441 # ($k => $v) is either a special op or a regular hashpair
442 my ($sql, @bind) = ($k =~ /^(-.+)/) ? $self->_where_op_in_hash($1, $v)
444 my $method = $self->_METHOD_FOR_refkind("_where_hashpair", $v);
445 $self->$method($k, $v);
448 push @sql_clauses, $sql;
449 push @all_bind, @bind;
452 return $self->_join_sql_clauses('and', \@sql_clauses, \@all_bind);
456 sub _where_op_in_hash {
457 my ($self, $orig_op, $v) = @_;
459 # put the operator in canonical form
461 $op =~ s/^-//; # remove initial dash
462 $op =~ s/[_\t ]+/ /g; # underscores and whitespace become single spaces
463 $op =~ s/^\s+|\s+$//g;# remove leading/trailing space
465 $self->_debug("OP(-$op) within hashref, recursing...");
467 my $op_entry = first {$op =~ $_->{regex}} @{$self->{unary_ops}};
468 my $handler = $op_entry->{handler};
470 puke "unknown operator: $orig_op";
472 elsif (not ref $handler) {
473 if ($op =~ s/\s?\d+$//) {
474 belch 'Use of [and|or|nest]_N modifiers is deprecated and will be removed in SQLA v2.0. '
475 . "You probably wanted ...-and => [ -$op => COND1, -$op => COND2 ... ]";
477 return $self->$handler ($op, $v);
479 elsif (ref $handler eq 'CODE') {
480 return $handler->($self, $op, $v);
483 puke "Illegal handler for operator $orig_op - expecting a method name or a coderef";
487 sub _where_op_ANDOR {
488 my ($self, $op, $v) = @_;
490 $self->_SWITCH_refkind($v, {
492 return $self->_where_ARRAYREF($v, $op);
496 return ( $op =~ /^or/i )
497 ? $self->_where_ARRAYREF( [ map { $_ => $v->{$_} } ( sort keys %$v ) ], $op )
498 : $self->_where_HASHREF($v);
502 puke "-$op => \\\$scalar not supported, use -nest => ...";
506 puke "-$op => \\[..] not supported, use -nest => ...";
509 SCALAR => sub { # permissively interpreted as SQL
510 puke "-$op => 'scalar' not supported, use -nest => \\'scalar'";
514 puke "-$op => undef not supported";
520 my ($self, $op, $v) = @_;
522 $self->_SWITCH_refkind($v, {
525 return $self->_where_ARRAYREF($v, '');
529 return $self->_where_HASHREF($v);
532 SCALARREF => sub { # literal SQL
536 ARRAYREFREF => sub { # literal SQL
540 SCALAR => sub { # permissively interpreted as SQL
541 belch "literal SQL should be -nest => \\'scalar' "
542 . "instead of -nest => 'scalar' ";
547 puke "-$op => undef not supported";
554 my ($self, $op, $v) = @_;
556 my ( $prefix, $suffix ) = ( $op =~ /\bnot\b/i )
559 $self->_SWITCH_refkind($v, {
561 my ( $sql, @bind ) = $self->_where_ARRAYREF($v, '');
562 return ( ($prefix . $sql . $suffix), @bind );
566 my ( $sql, @bind ) = @{ ${$v} };
567 return ( ($prefix . $sql . $suffix), @bind );
571 my ( $sql, @bind ) = $self->_where_HASHREF($v);
572 return ( ($prefix . $sql . $suffix), @bind );
575 SCALARREF => sub { # literal SQL
576 return ($prefix . $$v . $suffix);
579 SCALAR => sub { # interpreted as SQL column
580 return ($prefix . $self->_convert($self->_quote($v)) . $suffix);
584 puke "-$op => undef not supported";
590 sub _where_hashpair_ARRAYREF {
591 my ($self, $k, $v) = @_;
594 my @v = @$v; # need copy because of shift below
595 $self->_debug("ARRAY($k) means distribute over elements");
597 # put apart first element if it is an operator (-and, -or)
599 (defined $v[0] && $v[0] =~ /^ - (?: AND|OR ) $/ix)
603 my @distributed = map { {$k => $_} } @v;
606 $self->_debug("OP($op) reinjected into the distributed array");
607 unshift @distributed, $op;
610 my $logic = $op ? substr($op, 1) : '';
612 return $self->_recurse_where(\@distributed, $logic);
615 # LDNOTE : not sure of this one. What does "distribute over nothing" mean?
616 $self->_debug("empty ARRAY($k) means 0=1");
617 return ($self->{sqlfalse});
621 sub _where_hashpair_HASHREF {
622 my ($self, $k, $v, $logic) = @_;
625 my ($all_sql, @all_bind);
627 for my $orig_op (sort keys %$v) {
628 my $val = $v->{$orig_op};
630 # put the operator in canonical form
632 $op =~ s/^-//; # remove initial dash
633 $op =~ s/[_\t ]+/ /g; # underscores and whitespace become single spaces
634 $op =~ s/^\s+|\s+$//g;# remove leading/trailing space
638 # CASE: special operators like -in or -between
639 my $special_op = first {$op =~ $_->{regex}} @{$self->{special_ops}};
641 my $handler = $special_op->{handler};
643 puke "No handler supplied for special operator $orig_op";
645 elsif (not ref $handler) {
646 ($sql, @bind) = $self->$handler ($k, $op, $val);
648 elsif (ref $handler eq 'CODE') {
649 ($sql, @bind) = $handler->($self, $k, $op, $val);
652 puke "Illegal handler for special operator $orig_op - expecting a method name or a coderef";
656 $self->_SWITCH_refkind($val, {
658 ARRAYREF => sub { # CASE: col => {op => \@vals}
659 ($sql, @bind) = $self->_where_field_op_ARRAYREF($k, $op, $val);
662 SCALARREF => sub { # CASE: col => {op => \$scalar} (literal SQL without bind)
663 $sql = join ' ', $self->_convert($self->_quote($k)),
664 $self->_sqlcase($op),
668 ARRAYREFREF => sub { # CASE: col => {op => \[$sql, @bind]} (literal SQL with bind)
669 my ($sub_sql, @sub_bind) = @$$val;
670 $self->_assert_bindval_matches_bindtype(@sub_bind);
671 $sql = join ' ', $self->_convert($self->_quote($k)),
672 $self->_sqlcase($op),
678 ($sql, @bind) = $self->_where_hashpair_HASHREF($k, $val, $op);
681 UNDEF => sub { # CASE: col => {op => undef} : sql "IS (NOT)? NULL"
682 my $is = ($op =~ $self->{equality_op}) ? 'is' :
683 ($op =~ $self->{inequality_op}) ? 'is not' :
684 puke "unexpected operator '$orig_op' with undef operand";
685 $sql = $self->_quote($k) . $self->_sqlcase(" $is null");
688 FALLBACK => sub { # CASE: col => {op => $scalar}
689 $sql = join ' ', $self->_convert($self->_quote($k)),
690 $self->_sqlcase($op),
691 $self->_convert('?');
692 @bind = $self->_bindtype($k, $val);
697 ($all_sql) = (defined $all_sql and $all_sql) ? $self->_join_sql_clauses($logic, [$all_sql, $sql], []) : $sql;
698 push @all_bind, @bind;
700 return ($all_sql, @all_bind);
705 sub _where_field_op_ARRAYREF {
706 my ($self, $k, $op, $vals) = @_;
708 my @vals = @$vals; #always work on a copy
711 $self->_debug(sprintf '%s means multiple elements: [ %s ]',
713 join (', ', map { defined $_ ? "'$_'" : 'NULL' } @vals ),
716 # see if the first element is an -and/-or op
718 if (defined $vals[0] && $vals[0] =~ /^ - ( AND|OR ) $/ix) {
723 # distribute $op over each remaining member of @vals, append logic if exists
724 return $self->_recurse_where([map { {$k => {$op, $_}} } @vals], $logic);
726 # LDNOTE : had planned to change the distribution logic when
727 # $op =~ $self->{inequality_op}, because of Morgan laws :
728 # with {field => {'!=' => [22, 33]}}, it would be ridiculous to generate
729 # WHERE field != 22 OR field != 33 : the user probably means
730 # WHERE field != 22 AND field != 33.
731 # To do this, replace the above to roughly :
732 # my $logic = ($op =~ $self->{inequality_op}) ? 'AND' : 'OR';
733 # return $self->_recurse_where([map { {$k => {$op, $_}} } @vals], $logic);
737 # try to DWIM on equality operators
738 # LDNOTE : not 100% sure this is the correct thing to do ...
739 return ($self->{sqlfalse}) if $op =~ $self->{equality_op};
740 return ($self->{sqltrue}) if $op =~ $self->{inequality_op};
743 puke "operator '$op' applied on an empty array (field '$k')";
748 sub _where_hashpair_SCALARREF {
749 my ($self, $k, $v) = @_;
750 $self->_debug("SCALAR($k) means literal SQL: $$v");
751 my $sql = $self->_quote($k) . " " . $$v;
755 # literal SQL with bind
756 sub _where_hashpair_ARRAYREFREF {
757 my ($self, $k, $v) = @_;
758 $self->_debug("REF($k) means literal SQL: @${$v}");
759 my ($sql, @bind) = @${$v};
760 $self->_assert_bindval_matches_bindtype(@bind);
761 $sql = $self->_quote($k) . " " . $sql;
762 return ($sql, @bind );
765 # literal SQL without bind
766 sub _where_hashpair_SCALAR {
767 my ($self, $k, $v) = @_;
768 $self->_debug("NOREF($k) means simple key=val: $k $self->{cmp} $v");
769 my $sql = join ' ', $self->_convert($self->_quote($k)),
770 $self->_sqlcase($self->{cmp}),
771 $self->_convert('?');
772 my @bind = $self->_bindtype($k, $v);
773 return ( $sql, @bind);
777 sub _where_hashpair_UNDEF {
778 my ($self, $k, $v) = @_;
779 $self->_debug("UNDEF($k) means IS NULL");
780 my $sql = $self->_quote($k) . $self->_sqlcase(' is null');
784 #======================================================================
785 # WHERE: TOP-LEVEL OTHERS (SCALARREF, SCALAR, UNDEF)
786 #======================================================================
789 sub _where_SCALARREF {
790 my ($self, $where) = @_;
793 $self->_debug("SCALAR(*top) means literal SQL: $$where");
799 my ($self, $where) = @_;
802 $self->_debug("NOREF(*top) means literal SQL: $where");
813 #======================================================================
814 # WHERE: BUILTIN SPECIAL OPERATORS (-in, -between)
815 #======================================================================
818 sub _where_field_BETWEEN {
819 my ($self, $k, $op, $vals) = @_;
821 (ref $vals eq 'ARRAY' && @$vals == 2) or
822 (ref $vals eq 'REF' && (@$$vals == 1 || @$$vals == 2 || @$$vals == 3))
823 or puke "special op 'between' requires an arrayref of two values (or a scalarref or arrayrefref for literal SQL)";
825 my ($clause, @bind, $label, $and, $placeholder);
826 $label = $self->_convert($self->_quote($k));
827 $and = ' ' . $self->_sqlcase('and') . ' ';
828 $placeholder = $self->_convert('?');
829 $op = $self->_sqlcase($op);
831 if (ref $vals eq 'REF') {
832 ($clause, @bind) = @$$vals;
835 my (@all_sql, @all_bind);
837 foreach my $val (@$vals) {
838 my ($sql, @bind) = $self->_SWITCH_refkind($val, {
840 return ($placeholder, ($val));
843 return ($self->_convert($$val), ());
847 push @all_bind, @bind;
850 $clause = (join $and, @all_sql);
851 @bind = $self->_bindtype($k, @all_bind);
853 my $sql = "( $label $op $clause )";
858 sub _where_field_IN {
859 my ($self, $k, $op, $vals) = @_;
861 # backwards compatibility : if scalar, force into an arrayref
862 $vals = [$vals] if defined $vals && ! ref $vals;
864 my ($label) = $self->_convert($self->_quote($k));
865 my ($placeholder) = $self->_convert('?');
866 $op = $self->_sqlcase($op);
868 my ($sql, @bind) = $self->_SWITCH_refkind($vals, {
869 ARRAYREF => sub { # list of choices
870 if (@$vals) { # nonempty list
871 my $placeholders = join ", ", (($placeholder) x @$vals);
872 my $sql = "$label $op ( $placeholders )";
873 my @bind = $self->_bindtype($k, @$vals);
875 return ($sql, @bind);
877 else { # empty list : some databases won't understand "IN ()", so DWIM
878 my $sql = ($op =~ /\bnot\b/i) ? $self->{sqltrue} : $self->{sqlfalse};
883 ARRAYREFREF => sub { # literal SQL with bind
884 my ($sql, @bind) = @$$vals;
885 $self->_assert_bindval_matches_bindtype(@bind);
886 return ("$label $op ( $sql )", @bind);
890 puke "special op 'in' requires an arrayref (or arrayref-ref)";
894 return ($sql, @bind);
900 #======================================================================
902 #======================================================================
905 my ($self, $arg) = @_;
908 for my $c ($self->_order_by_chunks ($arg) ) {
909 $self->_SWITCH_refkind ($c, {
910 SCALAR => sub { push @sql, $c },
911 ARRAYREF => sub { push @sql, shift @$c; push @bind, @$c },
917 $self->_sqlcase(' order by'),
923 return wantarray ? ($sql, @bind) : $sql;
926 sub _order_by_chunks {
927 my ($self, $arg) = @_;
929 return $self->_SWITCH_refkind($arg, {
932 map { $self->_order_by_chunks ($_ ) } @$arg;
935 ARRAYREFREF => sub { [ @$$arg ] },
937 SCALAR => sub {$self->_quote($arg)},
939 UNDEF => sub {return () },
941 SCALARREF => sub {$$arg}, # literal SQL, no quoting
944 # get first pair in hash
945 my ($key, $val) = each %$arg;
947 return () unless $key;
949 if ( (keys %$arg) > 1 or not $key =~ /^-(desc|asc)/i ) {
950 puke "hash passed to _order_by must have exactly one key (-desc or -asc)";
956 for my $c ($self->_order_by_chunks ($val)) {
959 $self->_SWITCH_refkind ($c, {
968 $sql = $sql . ' ' . $self->_sqlcase($direction);
970 push @ret, [ $sql, @bind];
979 #======================================================================
980 # DATASOURCE (FOR NOW, JUST PLAIN TABLE OR LIST OF TABLES)
981 #======================================================================
986 $self->_SWITCH_refkind($from, {
987 ARRAYREF => sub {join ', ', map { $self->_quote($_) } @$from;},
988 SCALAR => sub {$self->_quote($from)},
989 SCALARREF => sub {$$from},
990 ARRAYREFREF => sub {join ', ', @$from;},
995 #======================================================================
997 #======================================================================
1003 $label or puke "can't quote an empty label";
1005 # left and right quote characters
1006 my ($ql, $qr, @other) = $self->_SWITCH_refkind($self->{quote_char}, {
1007 SCALAR => sub {($self->{quote_char}, $self->{quote_char})},
1008 ARRAYREF => sub {@{$self->{quote_char}}},
1012 or puke "quote_char must be an arrayref of 2 values";
1014 # no quoting if no quoting chars
1015 $ql or return $label;
1017 # no quoting for literal SQL
1018 return $$label if ref($label) eq 'SCALAR';
1020 # separate table / column (if applicable)
1021 my $sep = $self->{name_sep} || '';
1022 my @to_quote = $sep ? split /\Q$sep\E/, $label : ($label);
1024 # do the quoting, except for "*" or for `table`.*
1025 my @quoted = map { $_ eq '*' ? $_: $ql.$_.$qr} @to_quote;
1027 # reassemble and return.
1028 return join $sep, @quoted;
1032 # Conversion, if applicable
1034 my ($self, $arg) = @_;
1036 # LDNOTE : modified the previous implementation below because
1037 # it was not consistent : the first "return" is always an array,
1038 # the second "return" is context-dependent. Anyway, _convert
1039 # seems always used with just a single argument, so make it a
1041 # return @_ unless $self->{convert};
1042 # my $conv = $self->_sqlcase($self->{convert});
1043 # my @ret = map { $conv.'('.$_.')' } @_;
1044 # return wantarray ? @ret : $ret[0];
1045 if ($self->{convert}) {
1046 my $conv = $self->_sqlcase($self->{convert});
1047 $arg = $conv.'('.$arg.')';
1055 my($col, @vals) = @_;
1057 #LDNOTE : changed original implementation below because it did not make
1058 # sense when bindtype eq 'columns' and @vals > 1.
1059 # return $self->{bindtype} eq 'columns' ? [ $col, @vals ] : @vals;
1061 return $self->{bindtype} eq 'columns' ? map {[$col, $_]} @vals : @vals;
1064 # Dies if any element of @bind is not in [colname => value] format
1065 # if bindtype is 'columns'.
1066 sub _assert_bindval_matches_bindtype {
1067 my ($self, @bind) = @_;
1069 if ($self->{bindtype} eq 'columns') {
1070 foreach my $val (@bind) {
1071 if (!defined $val || ref($val) ne 'ARRAY' || @$val != 2) {
1072 die "bindtype 'columns' selected, you need to pass: [column_name => bind_value]"
1078 sub _join_sql_clauses {
1079 my ($self, $logic, $clauses_aref, $bind_aref) = @_;
1081 if (@$clauses_aref > 1) {
1082 my $join = " " . $self->_sqlcase($logic) . " ";
1083 my $sql = '( ' . join($join, @$clauses_aref) . ' )';
1084 return ($sql, @$bind_aref);
1086 elsif (@$clauses_aref) {
1087 return ($clauses_aref->[0], @$bind_aref); # no parentheses
1090 return (); # if no SQL, ignore @$bind_aref
1095 # Fix SQL case, if so requested
1099 # LDNOTE: if $self->{case} is true, then it contains 'lower', so we
1100 # don't touch the argument ... crooked logic, but let's not change it!
1101 return $self->{case} ? $_[0] : uc($_[0]);
1105 #======================================================================
1106 # DISPATCHING FROM REFKIND
1107 #======================================================================
1110 my ($self, $data) = @_;
1116 # blessed objects are treated like scalars
1117 $ref = (blessed $data) ? '' : ref $data;
1118 $n_steps += 1 if $ref;
1119 last if $ref ne 'REF';
1123 my $base = $ref || (defined $data ? 'SCALAR' : 'UNDEF');
1125 return $base . ('REF' x $n_steps);
1131 my ($self, $data) = @_;
1132 my @try = ($self->_refkind($data));
1133 push @try, 'SCALAR_or_UNDEF' if $try[0] eq 'SCALAR' || $try[0] eq 'UNDEF';
1134 push @try, 'FALLBACK';
1138 sub _METHOD_FOR_refkind {
1139 my ($self, $meth_prefix, $data) = @_;
1140 my $method = first {$_} map {$self->can($meth_prefix."_".$_)}
1141 $self->_try_refkind($data)
1142 or puke "cannot dispatch on '$meth_prefix' for ".$self->_refkind($data);
1147 sub _SWITCH_refkind {
1148 my ($self, $data, $dispatch_table) = @_;
1150 my $coderef = first {$_} map {$dispatch_table->{$_}}
1151 $self->_try_refkind($data)
1152 or puke "no dispatch entry for ".$self->_refkind($data);
1159 #======================================================================
1160 # VALUES, GENERATE, AUTOLOAD
1161 #======================================================================
1163 # LDNOTE: original code from nwiger, didn't touch code in that section
1164 # I feel the AUTOLOAD stuff should not be the default, it should
1165 # only be activated on explicit demand by user.
1169 my $data = shift || return;
1170 puke "Argument to ", __PACKAGE__, "->values must be a \\%hash"
1171 unless ref $data eq 'HASH';
1174 foreach my $k ( sort keys %$data ) {
1175 my $v = $data->{$k};
1176 $self->_SWITCH_refkind($v, {
1178 if ($self->{array_datatypes}) { # array datatype
1179 push @all_bind, $self->_bindtype($k, $v);
1181 else { # literal SQL with bind
1182 my ($sql, @bind) = @$v;
1183 $self->_assert_bindval_matches_bindtype(@bind);
1184 push @all_bind, @bind;
1187 ARRAYREFREF => sub { # literal SQL with bind
1188 my ($sql, @bind) = @${$v};
1189 $self->_assert_bindval_matches_bindtype(@bind);
1190 push @all_bind, @bind;
1192 SCALARREF => sub { # literal SQL without bind
1194 SCALAR_or_UNDEF => sub {
1195 push @all_bind, $self->_bindtype($k, $v);
1206 my(@sql, @sqlq, @sqlv);
1210 if ($ref eq 'HASH') {
1211 for my $k (sort keys %$_) {
1214 my $label = $self->_quote($k);
1215 if ($r eq 'ARRAY') {
1216 # literal SQL with bind
1217 my ($sql, @bind) = @$v;
1218 $self->_assert_bindval_matches_bindtype(@bind);
1219 push @sqlq, "$label = $sql";
1221 } elsif ($r eq 'SCALAR') {
1222 # literal SQL without bind
1223 push @sqlq, "$label = $$v";
1225 push @sqlq, "$label = ?";
1226 push @sqlv, $self->_bindtype($k, $v);
1229 push @sql, $self->_sqlcase('set'), join ', ', @sqlq;
1230 } elsif ($ref eq 'ARRAY') {
1231 # unlike insert(), assume these are ONLY the column names, i.e. for SQL
1234 if ($r eq 'ARRAY') { # literal SQL with bind
1235 my ($sql, @bind) = @$v;
1236 $self->_assert_bindval_matches_bindtype(@bind);
1239 } elsif ($r eq 'SCALAR') { # literal SQL without bind
1240 # embedded literal SQL
1247 push @sql, '(' . join(', ', @sqlq) . ')';
1248 } elsif ($ref eq 'SCALAR') {
1252 # strings get case twiddled
1253 push @sql, $self->_sqlcase($_);
1257 my $sql = join ' ', @sql;
1259 # this is pretty tricky
1260 # if ask for an array, return ($stmt, @bind)
1261 # otherwise, s/?/shift @sqlv/ to put it inline
1263 return ($sql, @sqlv);
1265 1 while $sql =~ s/\?/my $d = shift(@sqlv);
1266 ref $d ? $d->[1] : $d/e;
1275 # This allows us to check for a local, then _form, attr
1277 my($name) = $AUTOLOAD =~ /.*::(.+)/;
1278 return $self->generate($name, @_);
1289 SQL::Abstract - Generate SQL from Perl data structures
1295 my $sql = SQL::Abstract->new;
1297 my($stmt, @bind) = $sql->select($table, \@fields, \%where, \@order);
1299 my($stmt, @bind) = $sql->insert($table, \%fieldvals || \@values);
1301 my($stmt, @bind) = $sql->update($table, \%fieldvals, \%where);
1303 my($stmt, @bind) = $sql->delete($table, \%where);
1305 # Then, use these in your DBI statements
1306 my $sth = $dbh->prepare($stmt);
1307 $sth->execute(@bind);
1309 # Just generate the WHERE clause
1310 my($stmt, @bind) = $sql->where(\%where, \@order);
1312 # Return values in the same order, for hashed queries
1313 # See PERFORMANCE section for more details
1314 my @bind = $sql->values(\%fieldvals);
1318 This module was inspired by the excellent L<DBIx::Abstract>.
1319 However, in using that module I found that what I really wanted
1320 to do was generate SQL, but still retain complete control over my
1321 statement handles and use the DBI interface. So, I set out to
1322 create an abstract SQL generation module.
1324 While based on the concepts used by L<DBIx::Abstract>, there are
1325 several important differences, especially when it comes to WHERE
1326 clauses. I have modified the concepts used to make the SQL easier
1327 to generate from Perl data structures and, IMO, more intuitive.
1328 The underlying idea is for this module to do what you mean, based
1329 on the data structures you provide it. The big advantage is that
1330 you don't have to modify your code every time your data changes,
1331 as this module figures it out.
1333 To begin with, an SQL INSERT is as easy as just specifying a hash
1334 of C<key=value> pairs:
1337 name => 'Jimbo Bobson',
1338 phone => '123-456-7890',
1339 address => '42 Sister Lane',
1340 city => 'St. Louis',
1341 state => 'Louisiana',
1344 The SQL can then be generated with this:
1346 my($stmt, @bind) = $sql->insert('people', \%data);
1348 Which would give you something like this:
1350 $stmt = "INSERT INTO people
1351 (address, city, name, phone, state)
1352 VALUES (?, ?, ?, ?, ?)";
1353 @bind = ('42 Sister Lane', 'St. Louis', 'Jimbo Bobson',
1354 '123-456-7890', 'Louisiana');
1356 These are then used directly in your DBI code:
1358 my $sth = $dbh->prepare($stmt);
1359 $sth->execute(@bind);
1361 =head2 Inserting and Updating Arrays
1363 If your database has array types (like for example Postgres),
1364 activate the special option C<< array_datatypes => 1 >>
1365 when creating the C<SQL::Abstract> object.
1366 Then you may use an arrayref to insert and update database array types:
1368 my $sql = SQL::Abstract->new(array_datatypes => 1);
1370 planets => [qw/Mercury Venus Earth Mars/]
1373 my($stmt, @bind) = $sql->insert('solar_system', \%data);
1377 $stmt = "INSERT INTO solar_system (planets) VALUES (?)"
1379 @bind = (['Mercury', 'Venus', 'Earth', 'Mars']);
1382 =head2 Inserting and Updating SQL
1384 In order to apply SQL functions to elements of your C<%data> you may
1385 specify a reference to an arrayref for the given hash value. For example,
1386 if you need to execute the Oracle C<to_date> function on a value, you can
1387 say something like this:
1391 date_entered => \["to_date(?,'MM/DD/YYYY')", "03/02/2003"],
1394 The first value in the array is the actual SQL. Any other values are
1395 optional and would be included in the bind values array. This gives
1398 my($stmt, @bind) = $sql->insert('people', \%data);
1400 $stmt = "INSERT INTO people (name, date_entered)
1401 VALUES (?, to_date(?,'MM/DD/YYYY'))";
1402 @bind = ('Bill', '03/02/2003');
1404 An UPDATE is just as easy, all you change is the name of the function:
1406 my($stmt, @bind) = $sql->update('people', \%data);
1408 Notice that your C<%data> isn't touched; the module will generate
1409 the appropriately quirky SQL for you automatically. Usually you'll
1410 want to specify a WHERE clause for your UPDATE, though, which is
1411 where handling C<%where> hashes comes in handy...
1413 =head2 Complex where statements
1415 This module can generate pretty complicated WHERE statements
1416 easily. For example, simple C<key=value> pairs are taken to mean
1417 equality, and if you want to see if a field is within a set
1418 of values, you can use an arrayref. Let's say we wanted to
1419 SELECT some data based on this criteria:
1422 requestor => 'inna',
1423 worker => ['nwiger', 'rcwe', 'sfz'],
1424 status => { '!=', 'completed' }
1427 my($stmt, @bind) = $sql->select('tickets', '*', \%where);
1429 The above would give you something like this:
1431 $stmt = "SELECT * FROM tickets WHERE
1432 ( requestor = ? ) AND ( status != ? )
1433 AND ( worker = ? OR worker = ? OR worker = ? )";
1434 @bind = ('inna', 'completed', 'nwiger', 'rcwe', 'sfz');
1436 Which you could then use in DBI code like so:
1438 my $sth = $dbh->prepare($stmt);
1439 $sth->execute(@bind);
1445 The functions are simple. There's one for each major SQL operation,
1446 and a constructor you use first. The arguments are specified in a
1447 similar order to each function (table, then fields, then a where
1448 clause) to try and simplify things.
1453 =head2 new(option => 'value')
1455 The C<new()> function takes a list of options and values, and returns
1456 a new B<SQL::Abstract> object which can then be used to generate SQL
1457 through the methods below. The options accepted are:
1463 If set to 'lower', then SQL will be generated in all lowercase. By
1464 default SQL is generated in "textbook" case meaning something like:
1466 SELECT a_field FROM a_table WHERE some_field LIKE '%someval%'
1468 Any setting other than 'lower' is ignored.
1472 This determines what the default comparison operator is. By default
1473 it is C<=>, meaning that a hash like this:
1475 %where = (name => 'nwiger', email => 'nate@wiger.org');
1477 Will generate SQL like this:
1479 WHERE name = 'nwiger' AND email = 'nate@wiger.org'
1481 However, you may want loose comparisons by default, so if you set
1482 C<cmp> to C<like> you would get SQL such as:
1484 WHERE name like 'nwiger' AND email like 'nate@wiger.org'
1486 You can also override the comparsion on an individual basis - see
1487 the huge section on L</"WHERE CLAUSES"> at the bottom.
1489 =item sqltrue, sqlfalse
1491 Expressions for inserting boolean values within SQL statements.
1492 By default these are C<1=1> and C<1=0>. They are used
1493 by the special operators C<-in> and C<-not_in> for generating
1494 correct SQL even when the argument is an empty array (see below).
1498 This determines the default logical operator for multiple WHERE
1499 statements in arrays or hashes. If absent, the default logic is "or"
1500 for arrays, and "and" for hashes. This means that a WHERE
1504 event_date => {'>=', '2/13/99'},
1505 event_date => {'<=', '4/24/03'},
1508 will generate SQL like this:
1510 WHERE event_date >= '2/13/99' OR event_date <= '4/24/03'
1512 This is probably not what you want given this query, though (look
1513 at the dates). To change the "OR" to an "AND", simply specify:
1515 my $sql = SQL::Abstract->new(logic => 'and');
1517 Which will change the above C<WHERE> to:
1519 WHERE event_date >= '2/13/99' AND event_date <= '4/24/03'
1521 The logic can also be changed locally by inserting
1522 a modifier in front of an arrayref :
1524 @where = (-and => [event_date => {'>=', '2/13/99'},
1525 event_date => {'<=', '4/24/03'} ]);
1527 See the L</"WHERE CLAUSES"> section for explanations.
1531 This will automatically convert comparisons using the specified SQL
1532 function for both column and value. This is mostly used with an argument
1533 of C<upper> or C<lower>, so that the SQL will have the effect of
1534 case-insensitive "searches". For example, this:
1536 $sql = SQL::Abstract->new(convert => 'upper');
1537 %where = (keywords => 'MaKe iT CAse inSeNSItive');
1539 Will turn out the following SQL:
1541 WHERE upper(keywords) like upper('MaKe iT CAse inSeNSItive')
1543 The conversion can be C<upper()>, C<lower()>, or any other SQL function
1544 that can be applied symmetrically to fields (actually B<SQL::Abstract> does
1545 not validate this option; it will just pass through what you specify verbatim).
1549 This is a kludge because many databases suck. For example, you can't
1550 just bind values using DBI's C<execute()> for Oracle C<CLOB> or C<BLOB> fields.
1551 Instead, you have to use C<bind_param()>:
1553 $sth->bind_param(1, 'reg data');
1554 $sth->bind_param(2, $lots, {ora_type => ORA_CLOB});
1556 The problem is, B<SQL::Abstract> will normally just return a C<@bind> array,
1557 which loses track of which field each slot refers to. Fear not.
1559 If you specify C<bindtype> in new, you can determine how C<@bind> is returned.
1560 Currently, you can specify either C<normal> (default) or C<columns>. If you
1561 specify C<columns>, you will get an array that looks like this:
1563 my $sql = SQL::Abstract->new(bindtype => 'columns');
1564 my($stmt, @bind) = $sql->insert(...);
1567 [ 'column1', 'value1' ],
1568 [ 'column2', 'value2' ],
1569 [ 'column3', 'value3' ],
1572 You can then iterate through this manually, using DBI's C<bind_param()>.
1574 $sth->prepare($stmt);
1577 my($col, $data) = @$_;
1578 if ($col eq 'details' || $col eq 'comments') {
1579 $sth->bind_param($i, $data, {ora_type => ORA_CLOB});
1580 } elsif ($col eq 'image') {
1581 $sth->bind_param($i, $data, {ora_type => ORA_BLOB});
1583 $sth->bind_param($i, $data);
1587 $sth->execute; # execute without @bind now
1589 Now, why would you still use B<SQL::Abstract> if you have to do this crap?
1590 Basically, the advantage is still that you don't have to care which fields
1591 are or are not included. You could wrap that above C<for> loop in a simple
1592 sub called C<bind_fields()> or something and reuse it repeatedly. You still
1593 get a layer of abstraction over manual SQL specification.
1595 Note that if you set L</bindtype> to C<columns>, the C<\[$sql, @bind]>
1596 construct (see L</Literal SQL with placeholders and bind values (subqueries)>)
1597 will expect the bind values in this format.
1601 This is the character that a table or column name will be quoted
1602 with. By default this is an empty string, but you could set it to
1603 the character C<`>, to generate SQL like this:
1605 SELECT `a_field` FROM `a_table` WHERE `some_field` LIKE '%someval%'
1607 Alternatively, you can supply an array ref of two items, the first being the left
1608 hand quote character, and the second the right hand quote character. For
1609 example, you could supply C<['[',']']> for SQL Server 2000 compliant quotes
1610 that generates SQL like this:
1612 SELECT [a_field] FROM [a_table] WHERE [some_field] LIKE '%someval%'
1614 Quoting is useful if you have tables or columns names that are reserved
1615 words in your database's SQL dialect.
1619 This is the character that separates a table and column name. It is
1620 necessary to specify this when the C<quote_char> option is selected,
1621 so that tables and column names can be individually quoted like this:
1623 SELECT `table`.`one_field` FROM `table` WHERE `table`.`other_field` = 1
1625 =item array_datatypes
1627 When this option is true, arrayrefs in INSERT or UPDATE are
1628 interpreted as array datatypes and are passed directly
1630 When this option is false, arrayrefs are interpreted
1631 as literal SQL, just like refs to arrayrefs
1632 (but this behavior is for backwards compatibility; when writing
1633 new queries, use the "reference to arrayref" syntax
1639 Takes a reference to a list of "special operators"
1640 to extend the syntax understood by L<SQL::Abstract>.
1641 See section L</"SPECIAL OPERATORS"> for details.
1645 Takes a reference to a list of "unary operators"
1646 to extend the syntax understood by L<SQL::Abstract>.
1647 See section L</"UNARY OPERATORS"> for details.
1653 =head2 insert($table, \@values || \%fieldvals)
1655 This is the simplest function. You simply give it a table name
1656 and either an arrayref of values or hashref of field/value pairs.
1657 It returns an SQL INSERT statement and a list of bind values.
1658 See the sections on L</"Inserting and Updating Arrays"> and
1659 L</"Inserting and Updating SQL"> for information on how to insert
1660 with those data types.
1662 =head2 update($table, \%fieldvals, \%where)
1664 This takes a table, hashref of field/value pairs, and an optional
1665 hashref L<WHERE clause|/WHERE CLAUSES>. It returns an SQL UPDATE function and a list
1667 See the sections on L</"Inserting and Updating Arrays"> and
1668 L</"Inserting and Updating SQL"> for information on how to insert
1669 with those data types.
1671 =head2 select($source, $fields, $where, $order)
1673 This returns a SQL SELECT statement and associated list of bind values, as
1674 specified by the arguments :
1680 Specification of the 'FROM' part of the statement.
1681 The argument can be either a plain scalar (interpreted as a table
1682 name, will be quoted), or an arrayref (interpreted as a list
1683 of table names, joined by commas, quoted), or a scalarref
1684 (literal table name, not quoted), or a ref to an arrayref
1685 (list of literal table names, joined by commas, not quoted).
1689 Specification of the list of fields to retrieve from
1691 The argument can be either an arrayref (interpreted as a list
1692 of field names, will be joined by commas and quoted), or a
1693 plain scalar (literal SQL, not quoted).
1694 Please observe that this API is not as flexible as for
1695 the first argument C<$table>, for backwards compatibility reasons.
1699 Optional argument to specify the WHERE part of the query.
1700 The argument is most often a hashref, but can also be
1701 an arrayref or plain scalar --
1702 see section L<WHERE clause|/"WHERE CLAUSES"> for details.
1706 Optional argument to specify the ORDER BY part of the query.
1707 The argument can be a scalar, a hashref or an arrayref
1708 -- see section L<ORDER BY clause|/"ORDER BY CLAUSES">
1714 =head2 delete($table, \%where)
1716 This takes a table name and optional hashref L<WHERE clause|/WHERE CLAUSES>.
1717 It returns an SQL DELETE statement and list of bind values.
1719 =head2 where(\%where, \@order)
1721 This is used to generate just the WHERE clause. For example,
1722 if you have an arbitrary data structure and know what the
1723 rest of your SQL is going to look like, but want an easy way
1724 to produce a WHERE clause, use this. It returns an SQL WHERE
1725 clause and list of bind values.
1728 =head2 values(\%data)
1730 This just returns the values from the hash C<%data>, in the same
1731 order that would be returned from any of the other above queries.
1732 Using this allows you to markedly speed up your queries if you
1733 are affecting lots of rows. See below under the L</"PERFORMANCE"> section.
1735 =head2 generate($any, 'number', $of, \@data, $struct, \%types)
1737 Warning: This is an experimental method and subject to change.
1739 This returns arbitrarily generated SQL. It's a really basic shortcut.
1740 It will return two different things, depending on return context:
1742 my($stmt, @bind) = $sql->generate('create table', \$table, \@fields);
1743 my $stmt_and_val = $sql->generate('create table', \$table, \@fields);
1745 These would return the following:
1747 # First calling form
1748 $stmt = "CREATE TABLE test (?, ?)";
1749 @bind = (field1, field2);
1751 # Second calling form
1752 $stmt_and_val = "CREATE TABLE test (field1, field2)";
1754 Depending on what you're trying to do, it's up to you to choose the correct
1755 format. In this example, the second form is what you would want.
1759 $sql->generate('alter session', { nls_date_format => 'MM/YY' });
1763 ALTER SESSION SET nls_date_format = 'MM/YY'
1765 You get the idea. Strings get their case twiddled, but everything
1766 else remains verbatim.
1771 =head1 WHERE CLAUSES
1775 This module uses a variation on the idea from L<DBIx::Abstract>. It
1776 is B<NOT>, repeat I<not> 100% compatible. B<The main logic of this
1777 module is that things in arrays are OR'ed, and things in hashes
1780 The easiest way to explain is to show lots of examples. After
1781 each C<%where> hash shown, it is assumed you used:
1783 my($stmt, @bind) = $sql->where(\%where);
1785 However, note that the C<%where> hash can be used directly in any
1786 of the other functions as well, as described above.
1788 =head2 Key-value pairs
1790 So, let's get started. To begin, a simple hash:
1794 status => 'completed'
1797 Is converted to SQL C<key = val> statements:
1799 $stmt = "WHERE user = ? AND status = ?";
1800 @bind = ('nwiger', 'completed');
1802 One common thing I end up doing is having a list of values that
1803 a field can be in. To do this, simply specify a list inside of
1808 status => ['assigned', 'in-progress', 'pending'];
1811 This simple code will create the following:
1813 $stmt = "WHERE user = ? AND ( status = ? OR status = ? OR status = ? )";
1814 @bind = ('nwiger', 'assigned', 'in-progress', 'pending');
1816 A field associated to an empty arrayref will be considered a
1817 logical false and will generate 0=1.
1819 =head2 Specific comparison operators
1821 If you want to specify a different type of operator for your comparison,
1822 you can use a hashref for a given column:
1826 status => { '!=', 'completed' }
1829 Which would generate:
1831 $stmt = "WHERE user = ? AND status != ?";
1832 @bind = ('nwiger', 'completed');
1834 To test against multiple values, just enclose the values in an arrayref:
1836 status => { '=', ['assigned', 'in-progress', 'pending'] };
1838 Which would give you:
1840 "WHERE status = ? OR status = ? OR status = ?"
1843 The hashref can also contain multiple pairs, in which case it is expanded
1844 into an C<AND> of its elements:
1848 status => { '!=', 'completed', -not_like => 'pending%' }
1851 # Or more dynamically, like from a form
1852 $where{user} = 'nwiger';
1853 $where{status}{'!='} = 'completed';
1854 $where{status}{'-not_like'} = 'pending%';
1856 # Both generate this
1857 $stmt = "WHERE user = ? AND status != ? AND status NOT LIKE ?";
1858 @bind = ('nwiger', 'completed', 'pending%');
1861 To get an OR instead, you can combine it with the arrayref idea:
1865 priority => [ {'=', 2}, {'!=', 1} ]
1868 Which would generate:
1870 $stmt = "WHERE user = ? AND priority = ? OR priority != ?";
1871 @bind = ('nwiger', '2', '1');
1873 If you want to include literal SQL (with or without bind values), just use a
1874 scalar reference or array reference as the value:
1877 date_entered => { '>' => \["to_date(?, 'MM/DD/YYYY')", "11/26/2008"] },
1878 date_expires => { '<' => \"now()" }
1881 Which would generate:
1883 $stmt = "WHERE date_entered > "to_date(?, 'MM/DD/YYYY') AND date_expires < now()";
1884 @bind = ('11/26/2008');
1887 =head2 Logic and nesting operators
1889 In the example above,
1890 there is a subtle trap if you want to say something like
1891 this (notice the C<AND>):
1893 WHERE priority != ? AND priority != ?
1895 Because, in Perl you I<can't> do this:
1897 priority => { '!=', 2, '!=', 1 }
1899 As the second C<!=> key will obliterate the first. The solution
1900 is to use the special C<-modifier> form inside an arrayref:
1902 priority => [ -and => {'!=', 2},
1906 Normally, these would be joined by C<OR>, but the modifier tells it
1907 to use C<AND> instead. (Hint: You can use this in conjunction with the
1908 C<logic> option to C<new()> in order to change the way your queries
1909 work by default.) B<Important:> Note that the C<-modifier> goes
1910 B<INSIDE> the arrayref, as an extra first element. This will
1911 B<NOT> do what you think it might:
1913 priority => -and => [{'!=', 2}, {'!=', 1}] # WRONG!
1915 Here is a quick list of equivalencies, since there is some overlap:
1918 status => {'!=', 'completed', 'not like', 'pending%' }
1919 status => [ -and => {'!=', 'completed'}, {'not like', 'pending%'}]
1922 status => {'=', ['assigned', 'in-progress']}
1923 status => [ -or => {'=', 'assigned'}, {'=', 'in-progress'}]
1924 status => [ {'=', 'assigned'}, {'=', 'in-progress'} ]
1928 =head2 Special operators : IN, BETWEEN, etc.
1930 You can also use the hashref format to compare a list of fields using the
1931 C<IN> comparison operator, by specifying the list as an arrayref:
1934 status => 'completed',
1935 reportid => { -in => [567, 2335, 2] }
1938 Which would generate:
1940 $stmt = "WHERE status = ? AND reportid IN (?,?,?)";
1941 @bind = ('completed', '567', '2335', '2');
1943 The reverse operator C<-not_in> generates SQL C<NOT IN> and is used in
1946 If the argument to C<-in> is an empty array, 'sqlfalse' is generated
1947 (by default : C<1=0>). Similarly, C<< -not_in => [] >> generates
1948 'sqltrue' (by default : C<1=1>).
1952 Another pair of operators is C<-between> and C<-not_between>,
1953 used with an arrayref of two values:
1957 completion_date => {
1958 -not_between => ['2002-10-01', '2003-02-06']
1964 WHERE user = ? AND completion_date NOT BETWEEN ( ? AND ? )
1966 These are the two builtin "special operators"; but the
1967 list can be expanded : see section L</"SPECIAL OPERATORS"> below.
1969 =head2 Unary operators: bool
1971 If you wish to test against boolean columns or functions within your
1972 database you can use the C<-bool> and C<-not_bool> operators. For
1973 example to test the column C<is_user> being true and the column
1974 <is_enabled> being false you would use:-
1978 -not_bool => 'is_enabled',
1983 WHERE is_user AND NOT is_enabled
1985 If a more complex combination is required, testing more conditions,
1986 then you should use the and/or operators:-
1993 -not_bool => 'four',
1999 WHERE one AND two AND three AND NOT four
2002 =head2 Nested conditions, -and/-or prefixes
2004 So far, we've seen how multiple conditions are joined with a top-level
2005 C<AND>. We can change this by putting the different conditions we want in
2006 hashes and then putting those hashes in an array. For example:
2011 status => { -like => ['pending%', 'dispatched'] },
2015 status => 'unassigned',
2019 This data structure would create the following:
2021 $stmt = "WHERE ( user = ? AND ( status LIKE ? OR status LIKE ? ) )
2022 OR ( user = ? AND status = ? ) )";
2023 @bind = ('nwiger', 'pending', 'dispatched', 'robot', 'unassigned');
2026 There is also a special C<-nest>
2027 operator which adds an additional set of parens, to create a subquery.
2028 For example, to get something like this:
2030 $stmt = "WHERE user = ? AND ( workhrs > ? OR geo = ? )";
2031 @bind = ('nwiger', '20', 'ASIA');
2037 -nest => [ workhrs => {'>', 20}, geo => 'ASIA' ],
2041 Finally, clauses in hashrefs or arrayrefs can be
2042 prefixed with an C<-and> or C<-or> to change the logic
2049 -and => [workhrs => {'>', 20}, geo => 'ASIA' ],
2050 -and => [workhrs => {'<', 50}, geo => 'EURO' ]
2057 WHERE ( user = ? AND
2058 ( ( workhrs > ? AND geo = ? )
2059 OR ( workhrs < ? AND geo = ? ) ) )
2062 =head2 Algebraic inconsistency, for historical reasons
2064 C<Important note>: when connecting several conditions, the C<-and->|C<-or>
2065 operator goes C<outside> of the nested structure; whereas when connecting
2066 several constraints on one column, the C<-and> operator goes
2067 C<inside> the arrayref. Here is an example combining both features :
2070 -and => [a => 1, b => 2],
2071 -or => [c => 3, d => 4],
2072 e => [-and => {-like => 'foo%'}, {-like => '%bar'} ]
2077 WHERE ( ( ( a = ? AND b = ? )
2078 OR ( c = ? OR d = ? )
2079 OR ( e LIKE ? AND e LIKE ? ) ) )
2081 This difference in syntax is unfortunate but must be preserved for
2082 historical reasons. So be careful : the two examples below would
2083 seem algebraically equivalent, but they are not
2085 {col => [-and => {-like => 'foo%'}, {-like => '%bar'}]}
2086 # yields : WHERE ( ( col LIKE ? AND col LIKE ? ) )
2088 [-and => {col => {-like => 'foo%'}, {col => {-like => '%bar'}}]]
2089 # yields : WHERE ( ( col LIKE ? OR col LIKE ? ) )
2094 Finally, sometimes only literal SQL will do. If you want to include
2095 literal SQL verbatim, you can specify it as a scalar reference, namely:
2097 my $inn = 'is Not Null';
2099 priority => { '<', 2 },
2105 $stmt = "WHERE priority < ? AND requestor is Not Null";
2108 Note that in this example, you only get one bind parameter back, since
2109 the verbatim SQL is passed as part of the statement.
2111 Of course, just to prove a point, the above can also be accomplished
2115 priority => { '<', 2 },
2116 requestor => { '!=', undef },
2122 Conditions on boolean columns can be expressed in the same way, passing
2123 a reference to an empty string, however using liternal SQL in this way
2124 is deprecated - the preferred method is to use the boolean operators -
2125 see L</"Unary operators: bool"> :
2128 priority => { '<', 2 },
2134 $stmt = "WHERE priority < ? AND is_ready";
2138 =head2 Literal SQL with placeholders and bind values (subqueries)
2140 If the literal SQL to be inserted has placeholders and bind values,
2141 use a reference to an arrayref (yes this is a double reference --
2142 not so common, but perfectly legal Perl). For example, to find a date
2143 in Postgres you can use something like this:
2146 date_column => \[q/= date '2008-09-30' - ?::integer/, 10/]
2151 $stmt = "WHERE ( date_column = date '2008-09-30' - ?::integer )"
2154 Note that you must pass the bind values in the same format as they are returned
2155 by L</where>. That means that if you set L</bindtype> to C<columns>, you must
2156 provide the bind values in the C<< [ column_meta => value ] >> format, where
2157 C<column_meta> is an opaque scalar value; most commonly the column name, but
2158 you can use any scalar value (including references and blessed references),
2159 L<SQL::Abstract> will simply pass it through intact. So if C<bindtype> is set
2160 to C<columns> the above example will look like:
2163 date_column => \[q/= date '2008-09-30' - ?::integer/, [ dummy => 10 ]/]
2166 Literal SQL is especially useful for nesting parenthesized clauses in the
2167 main SQL query. Here is a first example :
2169 my ($sub_stmt, @sub_bind) = ("SELECT c1 FROM t1 WHERE c2 < ? AND c3 LIKE ?",
2173 bar => \["IN ($sub_stmt)" => @sub_bind],
2178 $stmt = "WHERE (foo = ? AND bar IN (SELECT c1 FROM t1
2179 WHERE c2 < ? AND c3 LIKE ?))";
2180 @bind = (1234, 100, "foo%");
2182 Other subquery operators, like for example C<"E<gt> ALL"> or C<"NOT IN">,
2183 are expressed in the same way. Of course the C<$sub_stmt> and
2184 its associated bind values can be generated through a former call
2187 my ($sub_stmt, @sub_bind)
2188 = $sql->select("t1", "c1", {c2 => {"<" => 100},
2189 c3 => {-like => "foo%"}});
2192 bar => \["> ALL ($sub_stmt)" => @sub_bind],
2195 In the examples above, the subquery was used as an operator on a column;
2196 but the same principle also applies for a clause within the main C<%where>
2197 hash, like an EXISTS subquery :
2199 my ($sub_stmt, @sub_bind)
2200 = $sql->select("t1", "*", {c1 => 1, c2 => \"> t0.c0"});
2203 -nest => \["EXISTS ($sub_stmt)" => @sub_bind],
2208 $stmt = "WHERE (foo = ? AND EXISTS (SELECT * FROM t1
2209 WHERE c1 = ? AND c2 > t0.c0))";
2213 Observe that the condition on C<c2> in the subquery refers to
2214 column C<t0.c0> of the main query : this is I<not> a bind
2215 value, so we have to express it through a scalar ref.
2216 Writing C<< c2 => {">" => "t0.c0"} >> would have generated
2217 C<< c2 > ? >> with bind value C<"t0.c0"> ... not exactly
2218 what we wanted here.
2220 Another use of the subquery technique is when some SQL clauses need
2221 parentheses, as it often occurs with some proprietary SQL extensions
2222 like for example fulltext expressions, geospatial expressions,
2223 NATIVE clauses, etc. Here is an example of a fulltext query in MySQL :
2226 -nest => \["MATCH (col1, col2) AGAINST (?)" => qw/apples/]
2229 Finally, here is an example where a subquery is used
2230 for expressing unary negation:
2232 my ($sub_stmt, @sub_bind)
2233 = $sql->where({age => [{"<" => 10}, {">" => 20}]});
2234 $sub_stmt =~ s/^ where //i; # don't want "WHERE" in the subclause
2236 lname => {like => '%son%'},
2237 -nest => \["NOT ($sub_stmt)" => @sub_bind],
2242 $stmt = "lname LIKE ? AND NOT ( age < ? OR age > ? )"
2243 @bind = ('%son%', 10, 20)
2249 These pages could go on for a while, since the nesting of the data
2250 structures this module can handle are pretty much unlimited (the
2251 module implements the C<WHERE> expansion as a recursive function
2252 internally). Your best bet is to "play around" with the module a
2253 little to see how the data structures behave, and choose the best
2254 format for your data based on that.
2256 And of course, all the values above will probably be replaced with
2257 variables gotten from forms or the command line. After all, if you
2258 knew everything ahead of time, you wouldn't have to worry about
2259 dynamically-generating SQL and could just hardwire it into your
2265 =head1 ORDER BY CLAUSES
2267 Some functions take an order by clause. This can either be a scalar (just a
2268 column name,) a hash of C<< { -desc => 'col' } >> or C<< { -asc => 'col' } >>,
2269 or an array of either of the two previous forms. Examples:
2271 Given | Will Generate
2272 ----------------------------------------------------------
2274 \'colA DESC' | ORDER BY colA DESC
2276 'colA' | ORDER BY colA
2278 [qw/colA colB/] | ORDER BY colA, colB
2280 {-asc => 'colA'} | ORDER BY colA ASC
2282 {-desc => 'colB'} | ORDER BY colB DESC
2284 ['colA', {-asc => 'colB'}] | ORDER BY colA, colB ASC
2286 { -asc => [qw/colA colB] } | ORDER BY colA ASC, colB ASC
2289 { -asc => 'colA' }, | ORDER BY colA ASC, colB DESC,
2290 { -desc => [qw/colB/], | colC ASC, colD ASC
2291 { -asc => [qw/colC colD/],|
2293 ===========================================================
2297 =head1 SPECIAL OPERATORS
2299 my $sqlmaker = SQL::Abstract->new(special_ops => [
2303 my ($self, $field, $op, $arg) = @_;
2309 handler => 'method_name',
2313 A "special operator" is a SQL syntactic clause that can be
2314 applied to a field, instead of a usual binary operator.
2317 WHERE field IN (?, ?, ?)
2318 WHERE field BETWEEN ? AND ?
2319 WHERE MATCH(field) AGAINST (?, ?)
2321 Special operators IN and BETWEEN are fairly standard and therefore
2322 are builtin within C<SQL::Abstract> (as the overridable methods
2323 C<_where_field_IN> and C<_where_field_BETWEEN>). For other operators,
2324 like the MATCH .. AGAINST example above which is specific to MySQL,
2325 you can write your own operator handlers - supply a C<special_ops>
2326 argument to the C<new> method. That argument takes an arrayref of
2327 operator definitions; each operator definition is a hashref with two
2334 the regular expression to match the operator
2338 Either a coderef or a plain scalar method name. In both cases
2339 the expected return is C<< ($sql, @bind) >>.
2341 When supplied with a method name, it is simply called on the
2342 L<SQL::Abstract/> object as:
2344 $self->$method_name ($field, $op, $arg)
2348 $op is the part that matched the handler regex
2349 $field is the LHS of the operator
2352 When supplied with a coderef, it is called as:
2354 $coderef->($self, $field, $op, $arg)
2359 For example, here is an implementation
2360 of the MATCH .. AGAINST syntax for MySQL
2362 my $sqlmaker = SQL::Abstract->new(special_ops => [
2364 # special op for MySql MATCH (field) AGAINST(word1, word2, ...)
2365 {regex => qr/^match$/i,
2367 my ($self, $field, $op, $arg) = @_;
2368 $arg = [$arg] if not ref $arg;
2369 my $label = $self->_quote($field);
2370 my ($placeholder) = $self->_convert('?');
2371 my $placeholders = join ", ", (($placeholder) x @$arg);
2372 my $sql = $self->_sqlcase('match') . " ($label) "
2373 . $self->_sqlcase('against') . " ($placeholders) ";
2374 my @bind = $self->_bindtype($field, @$arg);
2375 return ($sql, @bind);
2382 =head1 UNARY OPERATORS
2384 my $sqlmaker = SQL::Abstract->new(unary_ops => [
2388 my ($self, $op, $arg) = @_;
2394 handler => 'method_name',
2398 A "unary operator" is a SQL syntactic clause that can be
2399 applied to a field - the operator goes before the field
2401 You can write your own operator handlers - supply a C<unary_ops>
2402 argument to the C<new> method. That argument takes an arrayref of
2403 operator definitions; each operator definition is a hashref with two
2410 the regular expression to match the operator
2414 Either a coderef or a plain scalar method name. In both cases
2415 the expected return is C<< $sql >>.
2417 When supplied with a method name, it is simply called on the
2418 L<SQL::Abstract/> object as:
2420 $self->$method_name ($op, $arg)
2424 $op is the part that matched the handler regex
2425 $arg is the RHS or argument of the operator
2427 When supplied with a coderef, it is called as:
2429 $coderef->($self, $op, $arg)
2437 Thanks to some benchmarking by Mark Stosberg, it turns out that
2438 this module is many orders of magnitude faster than using C<DBIx::Abstract>.
2439 I must admit this wasn't an intentional design issue, but it's a
2440 byproduct of the fact that you get to control your C<DBI> handles
2443 To maximize performance, use a code snippet like the following:
2445 # prepare a statement handle using the first row
2446 # and then reuse it for the rest of the rows
2448 for my $href (@array_of_hashrefs) {
2449 $stmt ||= $sql->insert('table', $href);
2450 $sth ||= $dbh->prepare($stmt);
2451 $sth->execute($sql->values($href));
2454 The reason this works is because the keys in your C<$href> are sorted
2455 internally by B<SQL::Abstract>. Thus, as long as your data retains
2456 the same structure, you only have to generate the SQL the first time
2457 around. On subsequent queries, simply use the C<values> function provided
2458 by this module to return your values in the correct order.
2463 If you use my C<CGI::FormBuilder> module at all, you'll hopefully
2464 really like this part (I do, at least). Building up a complex query
2465 can be as simple as the following:
2469 use CGI::FormBuilder;
2472 my $form = CGI::FormBuilder->new(...);
2473 my $sql = SQL::Abstract->new;
2475 if ($form->submitted) {
2476 my $field = $form->field;
2477 my $id = delete $field->{id};
2478 my($stmt, @bind) = $sql->update('table', $field, {id => $id});
2481 Of course, you would still have to connect using C<DBI> to run the
2482 query, but the point is that if you make your form look like your
2483 table, the actual query script can be extremely simplistic.
2485 If you're B<REALLY> lazy (I am), check out C<HTML::QuickTable> for
2486 a fast interface to returning and formatting data. I frequently
2487 use these three modules together to write complex database query
2488 apps in under 50 lines.
2493 Version 1.50 was a major internal refactoring of C<SQL::Abstract>.
2494 Great care has been taken to preserve the I<published> behavior
2495 documented in previous versions in the 1.* family; however,
2496 some features that were previously undocumented, or behaved
2497 differently from the documentation, had to be changed in order
2498 to clarify the semantics. Hence, client code that was relying
2499 on some dark areas of C<SQL::Abstract> v1.*
2500 B<might behave differently> in v1.50.
2502 The main changes are :
2508 support for literal SQL through the C<< \ [$sql, bind] >> syntax.
2512 support for the { operator => \"..." } construct (to embed literal SQL)
2516 support for the { operator => \["...", @bind] } construct (to embed literal SQL with bind values)
2520 optional support for L<array datatypes|/"Inserting and Updating Arrays">
2524 defensive programming : check arguments
2528 fixed bug with global logic, which was previously implemented
2529 through global variables yielding side-effects. Prior versions would
2530 interpret C<< [ {cond1, cond2}, [cond3, cond4] ] >>
2531 as C<< "(cond1 AND cond2) OR (cond3 AND cond4)" >>.
2532 Now this is interpreted
2533 as C<< "(cond1 AND cond2) OR (cond3 OR cond4)" >>.
2538 fixed semantics of _bindtype on array args
2542 dropped the C<_anoncopy> of the %where tree. No longer necessary,
2543 we just avoid shifting arrays within that tree.
2547 dropped the C<_modlogic> function
2553 =head1 ACKNOWLEDGEMENTS
2555 There are a number of individuals that have really helped out with
2556 this module. Unfortunately, most of them submitted bugs via CPAN
2557 so I have no idea who they are! But the people I do know are:
2559 Ash Berlin (order_by hash term support)
2560 Matt Trout (DBIx::Class support)
2561 Mark Stosberg (benchmarking)
2562 Chas Owens (initial "IN" operator support)
2563 Philip Collins (per-field SQL functions)
2564 Eric Kolve (hashref "AND" support)
2565 Mike Fragassi (enhancements to "BETWEEN" and "LIKE")
2566 Dan Kubb (support for "quote_char" and "name_sep")
2567 Guillermo Roditi (patch to cleanup "IN" and "BETWEEN", fix and tests for _order_by)
2568 Laurent Dami (internal refactoring, multiple -nest, extensible list of special operators, literal SQL)
2569 Norbert Buchmuller (support for literal SQL in hashpair, misc. fixes & tests)
2570 Peter Rabbitson (rewrite of SQLA::Test, misc. fixes & tests)
2576 L<DBIx::Class>, L<DBIx::Abstract>, L<CGI::FormBuilder>, L<HTML::QuickTable>.
2580 Copyright (c) 2001-2007 Nathan Wiger <nwiger@cpan.org>. All Rights Reserved.
2582 This module is actively maintained by Matt Trout <mst@shadowcatsystems.co.uk>
2584 For support, your best bet is to try the C<DBIx::Class> users mailing list.
2585 While not an official support venue, C<DBIx::Class> makes heavy use of
2586 C<SQL::Abstract>, and as such list members there are very familiar with
2587 how to create queries.
2591 This module is free software; you may copy this under the terms of
2592 the GNU General Public License, or the Artistic License, copies of
2593 which should have accompanied your Perl kit.