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 $VERSION = '1.50';
20 # This would confuse some packagers
21 #$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 #======================================================================
33 # DEBUGGING AND ERROR REPORTING
34 #======================================================================
37 return unless $_[0]->{debug}; shift; # a little faster
38 my $func = (caller(1))[3];
39 warn "[$func] ", @_, "\n";
43 my($func) = (caller(1))[3];
44 carp "[$func] Warning: ", @_;
48 my($func) = (caller(1))[3];
49 croak "[$func] Fatal: ", @_;
53 #======================================================================
55 #======================================================================
59 my $class = ref($self) || $self;
60 my %opt = (ref $_[0] eq 'HASH') ? %{$_[0]} : @_;
62 # choose our case by keeping an option around
63 delete $opt{case} if $opt{case} && $opt{case} ne 'lower';
65 # default logic for interpreting arrayrefs
66 $opt{logic} = uc $opt{logic} || 'OR';
68 # how to return bind vars
69 # LDNOTE: changed nwiger code : why this 'delete' ??
70 # $opt{bindtype} ||= delete($opt{bind_type}) || 'normal';
71 $opt{bindtype} ||= 'normal';
73 # default comparison is "=", but can be overridden
76 # try to recognize which are the 'equality' and 'unequality' ops
77 # (temporary quickfix, should go through a more seasoned API)
78 $opt{equality_op} = qr/^(\Q$opt{cmp}\E|is|(is\s+)?like)$/i;
79 $opt{inequality_op} = qr/^(!=|<>|(is\s+)?not(\s+like)?)$/i;
82 $opt{sqltrue} ||= '1=1';
83 $opt{sqlfalse} ||= '0=1';
86 $opt{special_ops} ||= [];
87 push @{$opt{special_ops}}, @BUILTIN_SPECIAL_OPS;
89 return bless \%opt, $class;
94 #======================================================================
96 #======================================================================
100 my $table = $self->_table(shift);
101 my $data = shift || return;
103 my $method = $self->_METHOD_FOR_refkind("_insert", $data);
104 my ($sql, @bind) = $self->$method($data);
105 $sql = join " ", $self->_sqlcase('insert into'), $table, $sql;
106 return wantarray ? ($sql, @bind) : $sql;
109 sub _insert_HASHREF { # explicit list of fields and then values
110 my ($self, $data) = @_;
112 my @fields = sort keys %$data;
114 my ($sql, @bind) = $self->_insert_values($data);
117 $_ = $self->_quote($_) foreach @fields;
118 $sql = "( ".join(", ", @fields).") ".$sql;
120 return ($sql, @bind);
123 sub _insert_ARRAYREF { # just generate values(?,?) part (no list of fields)
124 my ($self, $data) = @_;
126 # no names (arrayref) so can't generate bindtype
127 $self->{bindtype} ne 'columns'
128 or belch "can't do 'columns' bindtype when called with arrayref";
130 # fold the list of values into a hash of column name - value pairs
131 # (where the column names are artificially generated, and their
132 # lexicographical ordering keep the ordering of the original list)
133 my $i = "a"; # incremented values will be in lexicographical order
134 my $data_in_hash = { map { ($i++ => $_) } @$data };
136 return $self->_insert_values($data_in_hash);
139 sub _insert_ARRAYREFREF { # literal SQL with bind
140 my ($self, $data) = @_;
142 my ($sql, @bind) = @${$data};
143 $self->_assert_bindval_matches_bindtype(@bind);
145 return ($sql, @bind);
149 sub _insert_SCALARREF { # literal SQL without bind
150 my ($self, $data) = @_;
156 my ($self, $data) = @_;
158 my (@values, @all_bind);
159 foreach my $column (sort keys %$data) {
160 my $v = $data->{$column};
162 $self->_SWITCH_refkind($v, {
165 if ($self->{array_datatypes}) { # if array datatype are activated
167 push @all_bind, $self->_bindtype($column, $v);
169 else { # else literal SQL with bind
170 my ($sql, @bind) = @$v;
171 $self->_assert_bindval_matches_bindtype(@bind);
173 push @all_bind, @bind;
177 ARRAYREFREF => sub { # literal SQL with bind
178 my ($sql, @bind) = @${$v};
179 $self->_assert_bindval_matches_bindtype(@bind);
181 push @all_bind, @bind;
184 # THINK : anything useful to do with a HASHREF ?
185 HASHREF => sub { # (nothing, but old SQLA passed it through)
186 #TODO in SQLA >= 2.0 it will die instead
187 belch "HASH ref as bind value in insert is not supported";
189 push @all_bind, $self->_bindtype($column, $v);
192 SCALARREF => sub { # literal SQL without bind
196 SCALAR_or_UNDEF => sub {
198 push @all_bind, $self->_bindtype($column, $v);
205 my $sql = $self->_sqlcase('values')." ( ".join(", ", @values)." )";
206 return ($sql, @all_bind);
211 #======================================================================
213 #======================================================================
218 my $table = $self->_table(shift);
219 my $data = shift || return;
222 # first build the 'SET' part of the sql statement
223 my (@set, @all_bind);
224 puke "Unsupported data type specified to \$sql->update"
225 unless ref $data eq 'HASH';
227 for my $k (sort keys %$data) {
230 my $label = $self->_quote($k);
232 $self->_SWITCH_refkind($v, {
234 if ($self->{array_datatypes}) { # array datatype
235 push @set, "$label = ?";
236 push @all_bind, $self->_bindtype($k, $v);
238 else { # literal SQL with bind
239 my ($sql, @bind) = @$v;
240 $self->_assert_bindval_matches_bindtype(@bind);
241 push @set, "$label = $sql";
242 push @all_bind, @bind;
245 ARRAYREFREF => sub { # literal SQL with bind
246 my ($sql, @bind) = @${$v};
247 $self->_assert_bindval_matches_bindtype(@bind);
248 push @set, "$label = $sql";
249 push @all_bind, @bind;
251 SCALARREF => sub { # literal SQL without bind
252 push @set, "$label = $$v";
254 SCALAR_or_UNDEF => sub {
255 push @set, "$label = ?";
256 push @all_bind, $self->_bindtype($k, $v);
262 my $sql = $self->_sqlcase('update') . " $table " . $self->_sqlcase('set ')
266 my($where_sql, @where_bind) = $self->where($where);
268 push @all_bind, @where_bind;
271 return wantarray ? ($sql, @all_bind) : $sql;
277 #======================================================================
279 #======================================================================
284 my $table = $self->_table(shift);
285 my $fields = shift || '*';
289 my($where_sql, @bind) = $self->where($where, $order);
291 my $f = (ref $fields eq 'ARRAY') ? join ', ', map { $self->_quote($_) } @$fields
293 my $sql = join(' ', $self->_sqlcase('select'), $f,
294 $self->_sqlcase('from'), $table)
297 return wantarray ? ($sql, @bind) : $sql;
300 #======================================================================
302 #======================================================================
307 my $table = $self->_table(shift);
311 my($where_sql, @bind) = $self->where($where);
312 my $sql = $self->_sqlcase('delete from') . " $table" . $where_sql;
314 return wantarray ? ($sql, @bind) : $sql;
318 #======================================================================
320 #======================================================================
324 # Finally, a separate routine just to handle WHERE clauses
326 my ($self, $where, $order) = @_;
329 my ($sql, @bind) = $self->_recurse_where($where);
330 $sql = $sql ? $self->_sqlcase(' where ') . "( $sql )" : '';
334 $sql .= $self->_order_by($order);
337 return wantarray ? ($sql, @bind) : $sql;
342 my ($self, $where, $logic) = @_;
344 # dispatch on appropriate method according to refkind of $where
345 my $method = $self->_METHOD_FOR_refkind("_where", $where);
348 my ($sql, @bind) = $self->$method($where, $logic);
350 # DBIx::Class directly calls _recurse_where in scalar context, so
351 # we must implement it, even if not in the official API
352 return wantarray ? ($sql, @bind) : $sql;
357 #======================================================================
358 # WHERE: top-level ARRAYREF
359 #======================================================================
362 sub _where_ARRAYREF {
363 my ($self, $where, $logic) = @_;
365 $logic = uc($logic || $self->{logic});
366 $logic eq 'AND' or $logic eq 'OR' or puke "unknown logic: $logic";
368 my @clauses = @$where;
370 my (@sql_clauses, @all_bind);
371 # need to use while() so can shift() for pairs
372 while (my $el = shift @clauses) {
374 # switch according to kind of $el and get corresponding ($sql, @bind)
375 my ($sql, @bind) = $self->_SWITCH_refkind($el, {
377 # skip empty elements, otherwise get invalid trailing AND stuff
378 ARRAYREF => sub {$self->_recurse_where($el) if @$el},
380 ARRAYREFREF => sub { @{${$el}} if @{${$el}}},
382 HASHREF => sub {$self->_recurse_where($el, 'and') if %$el},
383 # LDNOTE : previous SQLA code for hashrefs was creating a dirty
384 # side-effect: the first hashref within an array would change
385 # the global logic to 'AND'. So [ {cond1, cond2}, [cond3, cond4] ]
386 # was interpreted as "(cond1 AND cond2) OR (cond3 AND cond4)",
387 # whereas it should be "(cond1 AND cond2) OR (cond3 OR cond4)".
389 SCALARREF => sub { ($$el); },
391 SCALAR => sub {# top-level arrayref with scalars, recurse in pairs
392 $self->_recurse_where({$el => shift(@clauses)})},
394 UNDEF => sub {puke "not supported : UNDEF in arrayref" },
398 push @sql_clauses, $sql;
399 push @all_bind, @bind;
403 return $self->_join_sql_clauses($logic, \@sql_clauses, \@all_bind);
406 #======================================================================
407 # WHERE: top-level ARRAYREFREF
408 #======================================================================
410 sub _where_ARRAYREFREF {
411 my ($self, $where) = @_;
412 my ($sql, @bind) = @{${$where}};
414 return ($sql, @bind);
417 #======================================================================
418 # WHERE: top-level HASHREF
419 #======================================================================
422 my ($self, $where) = @_;
423 my (@sql_clauses, @all_bind);
425 # LDNOTE : don't really know why we need to sort keys
426 for my $k (sort keys %$where) {
427 my $v = $where->{$k};
429 # ($k => $v) is either a special op or a regular hashpair
430 my ($sql, @bind) = ($k =~ /^-(.+)/) ? $self->_where_op_in_hash($1, $v)
432 my $method = $self->_METHOD_FOR_refkind("_where_hashpair", $v);
433 $self->$method($k, $v);
436 push @sql_clauses, $sql;
437 push @all_bind, @bind;
440 return $self->_join_sql_clauses('and', \@sql_clauses, \@all_bind);
444 sub _where_op_in_hash {
445 my ($self, $op, $v) = @_;
447 $op =~ /^(AND|OR|NEST)[_\d]*/i
448 or puke "unknown operator: -$op";
449 $op = uc($1); # uppercase, remove trailing digits
450 $self->_debug("OP(-$op) within hashref, recursing...");
452 $self->_SWITCH_refkind($v, {
455 return $self->_where_ARRAYREF($v, $op eq 'NEST' ? '' : $op);
460 return $self->_where_ARRAYREF([%$v], 'OR');
463 return $self->_where_HASHREF($v);
467 SCALARREF => sub { # literal SQL
469 or puke "-$op => \\\$scalar not supported, use -nest => ...";
473 ARRAYREFREF => sub { # literal SQL
475 or puke "-$op => \\[..] not supported, use -nest => ...";
479 SCALAR => sub { # permissively interpreted as SQL
481 or puke "-$op => 'scalar' not supported, use -nest => \\'scalar'";
482 belch "literal SQL should be -nest => \\'scalar' "
483 . "instead of -nest => 'scalar' ";
488 puke "-$op => undef not supported";
494 sub _where_hashpair_ARRAYREF {
495 my ($self, $k, $v) = @_;
498 my @v = @$v; # need copy because of shift below
499 $self->_debug("ARRAY($k) means distribute over elements");
501 # put apart first element if it is an operator (-and, -or)
502 my $op = ($v[0] =~ /^ - (?: AND|OR ) $/ix
506 my @distributed = map { {$k => $_} } @v;
509 $self->_debug("OP($op) reinjected into the distributed array");
510 unshift @distributed, $op;
513 my $logic = $op ? substr($op, 1) : '';
515 return $self->_recurse_where(\@distributed, $logic);
518 # LDNOTE : not sure of this one. What does "distribute over nothing" mean?
519 $self->_debug("empty ARRAY($k) means 0=1");
520 return ($self->{sqlfalse});
524 sub _where_hashpair_HASHREF {
525 my ($self, $k, $v) = @_;
527 my (@all_sql, @all_bind);
529 for my $op (sort keys %$v) {
532 # put the operator in canonical form
533 $op =~ s/^-//; # remove initial dash
534 $op =~ tr/_/ /; # underscores become spaces
535 $op =~ s/^\s+//; # no initial space
536 $op =~ s/\s+$//; # no final space
537 $op =~ s/\s+/ /; # multiple spaces become one
541 # CASE: special operators like -in or -between
542 my $special_op = first {$op =~ $_->{regex}} @{$self->{special_ops}};
544 ($sql, @bind) = $special_op->{handler}->($self, $k, $op, $val);
547 $self->_SWITCH_refkind($val, {
549 ARRAYREF => sub { # CASE: col => {op => \@vals}
550 ($sql, @bind) = $self->_where_field_op_ARRAYREF($k, $op, $val);
553 SCALARREF => sub { # CASE: col => {op => \$scalar} (literal SQL without bind)
554 $sql = join ' ', $self->_convert($self->_quote($k)),
555 $self->_sqlcase($op),
559 ARRAYREFREF => sub { # CASE: col => {op => \[$sql, @bind]} (literal SQL with bind)
560 my ($sub_sql, @sub_bind) = @$$val;
561 $self->_assert_bindval_matches_bindtype(@sub_bind);
562 $sql = join ' ', $self->_convert($self->_quote($k)),
563 $self->_sqlcase($op),
568 UNDEF => sub { # CASE: col => {op => undef} : sql "IS (NOT)? NULL"
569 my $is = ($op =~ $self->{equality_op}) ? 'is' :
570 ($op =~ $self->{inequality_op}) ? 'is not' :
571 puke "unexpected operator '$op' with undef operand";
572 $sql = $self->_quote($k) . $self->_sqlcase(" $is null");
575 FALLBACK => sub { # CASE: col => {op => $scalar}
576 $sql = join ' ', $self->_convert($self->_quote($k)),
577 $self->_sqlcase($op),
578 $self->_convert('?');
579 @bind = $self->_bindtype($k, $val);
585 push @all_bind, @bind;
588 return $self->_join_sql_clauses('and', \@all_sql, \@all_bind);
593 sub _where_field_op_ARRAYREF {
594 my ($self, $k, $op, $vals) = @_;
597 $self->_debug("ARRAY($vals) means multiple elements: [ @$vals ]");
599 # LDNOTE : had planned to change the distribution logic when
600 # $op =~ $self->{inequality_op}, because of Morgan laws :
601 # with {field => {'!=' => [22, 33]}}, it would be ridiculous to generate
602 # WHERE field != 22 OR field != 33 : the user probably means
603 # WHERE field != 22 AND field != 33.
604 # To do this, replace the line below by :
605 # my $logic = ($op =~ $self->{inequality_op}) ? 'AND' : 'OR';
606 # return $self->_recurse_where([map { {$k => {$op, $_}} } @$vals], $logic);
608 # distribute $op over each member of @$vals
609 return $self->_recurse_where([map { {$k => {$op, $_}} } @$vals]);
612 # try to DWIM on equality operators
613 # LDNOTE : not 100% sure this is the correct thing to do ...
614 return ($self->{sqlfalse}) if $op =~ $self->{equality_op};
615 return ($self->{sqltrue}) if $op =~ $self->{inequality_op};
618 puke "operator '$op' applied on an empty array (field '$k')";
623 sub _where_hashpair_SCALARREF {
624 my ($self, $k, $v) = @_;
625 $self->_debug("SCALAR($k) means literal SQL: $$v");
626 my $sql = $self->_quote($k) . " " . $$v;
630 # literal SQL with bind
631 sub _where_hashpair_ARRAYREFREF {
632 my ($self, $k, $v) = @_;
633 $self->_debug("REF($k) means literal SQL: @${$v}");
634 my ($sql, @bind) = @${$v};
635 $self->_assert_bindval_matches_bindtype(@bind);
636 $sql = $self->_quote($k) . " " . $sql;
637 return ($sql, @bind );
640 # literal SQL without bind
641 sub _where_hashpair_SCALAR {
642 my ($self, $k, $v) = @_;
643 $self->_debug("NOREF($k) means simple key=val: $k $self->{cmp} $v");
644 my $sql = join ' ', $self->_convert($self->_quote($k)),
645 $self->_sqlcase($self->{cmp}),
646 $self->_convert('?');
647 my @bind = $self->_bindtype($k, $v);
648 return ( $sql, @bind);
652 sub _where_hashpair_UNDEF {
653 my ($self, $k, $v) = @_;
654 $self->_debug("UNDEF($k) means IS NULL");
655 my $sql = $self->_quote($k) . $self->_sqlcase(' is null');
659 #======================================================================
660 # WHERE: TOP-LEVEL OTHERS (SCALARREF, SCALAR, UNDEF)
661 #======================================================================
664 sub _where_SCALARREF {
665 my ($self, $where) = @_;
668 $self->_debug("SCALAR(*top) means literal SQL: $$where");
674 my ($self, $where) = @_;
677 $self->_debug("NOREF(*top) means literal SQL: $where");
688 #======================================================================
689 # WHERE: BUILTIN SPECIAL OPERATORS (-in, -between)
690 #======================================================================
693 sub _where_field_BETWEEN {
694 my ($self, $k, $op, $vals) = @_;
696 ref $vals eq 'ARRAY' && @$vals == 2
697 or puke "special op 'between' requires an arrayref of two values";
699 my ($label) = $self->_convert($self->_quote($k));
700 my ($placeholder) = $self->_convert('?');
701 my $and = $self->_sqlcase('and');
702 $op = $self->_sqlcase($op);
704 my $sql = "( $label $op $placeholder $and $placeholder )";
705 my @bind = $self->_bindtype($k, @$vals);
710 sub _where_field_IN {
711 my ($self, $k, $op, $vals) = @_;
713 # backwards compatibility : if scalar, force into an arrayref
714 $vals = [$vals] if defined $vals && ! ref $vals;
716 my ($label) = $self->_convert($self->_quote($k));
717 my ($placeholder) = $self->_convert('?');
718 $op = $self->_sqlcase($op);
720 my ($sql, @bind) = $self->_SWITCH_refkind($vals, {
721 ARRAYREF => sub { # list of choices
722 if (@$vals) { # nonempty list
723 my $placeholders = join ", ", (($placeholder) x @$vals);
724 my $sql = "$label $op ( $placeholders )";
725 my @bind = $self->_bindtype($k, @$vals);
727 return ($sql, @bind);
729 else { # empty list : some databases won't understand "IN ()", so DWIM
730 my $sql = ($op =~ /\bnot\b/i) ? $self->{sqltrue} : $self->{sqlfalse};
735 ARRAYREFREF => sub { # literal SQL with bind
736 my ($sql, @bind) = @$$vals;
737 $self->_assert_bindval_matches_bindtype(@bind);
738 return ("$label $op ( $sql )", @bind);
742 puke "special op 'in' requires an arrayref (or arrayref-ref)";
746 return ($sql, @bind);
754 #======================================================================
756 #======================================================================
759 my ($self, $arg) = @_;
761 # construct list of ordering instructions
762 my @order = $self->_SWITCH_refkind($arg, {
765 map {$self->_SWITCH_refkind($_, {
766 SCALAR => sub {$self->_quote($_)},
768 SCALARREF => sub {$$_}, # literal SQL, no quoting
769 HASHREF => sub {$self->_order_by_hash($_)}
773 SCALAR => sub {$self->_quote($arg)},
775 SCALARREF => sub {$$arg}, # literal SQL, no quoting
776 HASHREF => sub {$self->_order_by_hash($arg)},
781 my $order = join ', ', @order;
782 return $order ? $self->_sqlcase(' order by')." $order" : '';
787 my ($self, $hash) = @_;
789 # get first pair in hash
790 my ($key, $val) = each %$hash;
792 # check if one pair was found and no other pair in hash
793 $key && !(each %$hash)
794 or puke "hash passed to _order_by must have exactly one key (-desc or -asc)";
796 my ($order) = ($key =~ /^-(desc|asc)/i)
797 or puke "invalid key in _order_by hash : $key";
799 return $self->_quote($val) ." ". $self->_sqlcase($order);
804 #======================================================================
805 # DATASOURCE (FOR NOW, JUST PLAIN TABLE OR LIST OF TABLES)
806 #======================================================================
811 $self->_SWITCH_refkind($from, {
812 ARRAYREF => sub {join ', ', map { $self->_quote($_) } @$from;},
813 SCALAR => sub {$self->_quote($from)},
814 SCALARREF => sub {$$from},
815 ARRAYREFREF => sub {join ', ', @$from;},
820 #======================================================================
822 #======================================================================
828 $label or puke "can't quote an empty label";
830 # left and right quote characters
831 my ($ql, $qr, @other) = $self->_SWITCH_refkind($self->{quote_char}, {
832 SCALAR => sub {($self->{quote_char}, $self->{quote_char})},
833 ARRAYREF => sub {@{$self->{quote_char}}},
837 or puke "quote_char must be an arrayref of 2 values";
839 # no quoting if no quoting chars
840 $ql or return $label;
842 # no quoting for literal SQL
843 return $$label if ref($label) eq 'SCALAR';
845 # separate table / column (if applicable)
846 my $sep = $self->{name_sep} || '';
847 my @to_quote = $sep ? split /\Q$sep\E/, $label : ($label);
849 # do the quoting, except for "*" or for `table`.*
850 my @quoted = map { $_ eq '*' ? $_: $ql.$_.$qr} @to_quote;
852 # reassemble and return.
853 return join $sep, @quoted;
857 # Conversion, if applicable
859 my ($self, $arg) = @_;
861 # LDNOTE : modified the previous implementation below because
862 # it was not consistent : the first "return" is always an array,
863 # the second "return" is context-dependent. Anyway, _convert
864 # seems always used with just a single argument, so make it a
866 # return @_ unless $self->{convert};
867 # my $conv = $self->_sqlcase($self->{convert});
868 # my @ret = map { $conv.'('.$_.')' } @_;
869 # return wantarray ? @ret : $ret[0];
870 if ($self->{convert}) {
871 my $conv = $self->_sqlcase($self->{convert});
872 $arg = $conv.'('.$arg.')';
880 my($col, @vals) = @_;
882 #LDNOTE : changed original implementation below because it did not make
883 # sense when bindtype eq 'columns' and @vals > 1.
884 # return $self->{bindtype} eq 'columns' ? [ $col, @vals ] : @vals;
886 return $self->{bindtype} eq 'columns' ? map {[$col, $_]} @vals : @vals;
889 # Dies if any element of @bind is not in [colname => value] format
890 # if bindtype is 'columns'.
891 sub _assert_bindval_matches_bindtype {
892 my ($self, @bind) = @_;
894 if ($self->{bindtype} eq 'columns') {
895 foreach my $val (@bind) {
896 if (!defined $val || ref($val) ne 'ARRAY' || @$val != 2) {
897 die "bindtype 'columns' selected, you need to pass: [column_name => bind_value]"
903 sub _join_sql_clauses {
904 my ($self, $logic, $clauses_aref, $bind_aref) = @_;
906 if (@$clauses_aref > 1) {
907 my $join = " " . $self->_sqlcase($logic) . " ";
908 my $sql = '( ' . join($join, @$clauses_aref) . ' )';
909 return ($sql, @$bind_aref);
911 elsif (@$clauses_aref) {
912 return ($clauses_aref->[0], @$bind_aref); # no parentheses
915 return (); # if no SQL, ignore @$bind_aref
920 # Fix SQL case, if so requested
924 # LDNOTE: if $self->{case} is true, then it contains 'lower', so we
925 # don't touch the argument ... crooked logic, but let's not change it!
926 return $self->{case} ? $_[0] : uc($_[0]);
930 #======================================================================
931 # DISPATCHING FROM REFKIND
932 #======================================================================
935 my ($self, $data) = @_;
941 # blessed objects are treated like scalars
942 $ref = (blessed $data) ? '' : ref $data;
943 $n_steps += 1 if $ref;
944 last if $ref ne 'REF';
948 my $base = $ref || (defined $data ? 'SCALAR' : 'UNDEF');
950 return $base . ('REF' x $n_steps);
956 my ($self, $data) = @_;
957 my @try = ($self->_refkind($data));
958 push @try, 'SCALAR_or_UNDEF' if $try[0] eq 'SCALAR' || $try[0] eq 'UNDEF';
959 push @try, 'FALLBACK';
963 sub _METHOD_FOR_refkind {
964 my ($self, $meth_prefix, $data) = @_;
965 my $method = first {$_} map {$self->can($meth_prefix."_".$_)}
966 $self->_try_refkind($data)
967 or puke "cannot dispatch on '$meth_prefix' for ".$self->_refkind($data);
972 sub _SWITCH_refkind {
973 my ($self, $data, $dispatch_table) = @_;
975 my $coderef = first {$_} map {$dispatch_table->{$_}}
976 $self->_try_refkind($data)
977 or puke "no dispatch entry for ".$self->_refkind($data);
984 #======================================================================
985 # VALUES, GENERATE, AUTOLOAD
986 #======================================================================
988 # LDNOTE: original code from nwiger, didn't touch code in that section
989 # I feel the AUTOLOAD stuff should not be the default, it should
990 # only be activated on explicit demand by user.
994 my $data = shift || return;
995 puke "Argument to ", __PACKAGE__, "->values must be a \\%hash"
996 unless ref $data eq 'HASH';
999 foreach my $k ( sort keys %$data ) {
1000 my $v = $data->{$k};
1001 $self->_SWITCH_refkind($v, {
1003 if ($self->{array_datatypes}) { # array datatype
1004 push @all_bind, $self->_bindtype($k, $v);
1006 else { # literal SQL with bind
1007 my ($sql, @bind) = @$v;
1008 $self->_assert_bindval_matches_bindtype(@bind);
1009 push @all_bind, @bind;
1012 ARRAYREFREF => sub { # literal SQL with bind
1013 my ($sql, @bind) = @${$v};
1014 $self->_assert_bindval_matches_bindtype(@bind);
1015 push @all_bind, @bind;
1017 SCALARREF => sub { # literal SQL without bind
1019 SCALAR_or_UNDEF => sub {
1020 push @all_bind, $self->_bindtype($k, $v);
1031 my(@sql, @sqlq, @sqlv);
1035 if ($ref eq 'HASH') {
1036 for my $k (sort keys %$_) {
1039 my $label = $self->_quote($k);
1040 if ($r eq 'ARRAY') {
1041 # literal SQL with bind
1042 my ($sql, @bind) = @$v;
1043 $self->_assert_bindval_matches_bindtype(@bind);
1044 push @sqlq, "$label = $sql";
1046 } elsif ($r eq 'SCALAR') {
1047 # literal SQL without bind
1048 push @sqlq, "$label = $$v";
1050 push @sqlq, "$label = ?";
1051 push @sqlv, $self->_bindtype($k, $v);
1054 push @sql, $self->_sqlcase('set'), join ', ', @sqlq;
1055 } elsif ($ref eq 'ARRAY') {
1056 # unlike insert(), assume these are ONLY the column names, i.e. for SQL
1059 if ($r eq 'ARRAY') { # literal SQL with bind
1060 my ($sql, @bind) = @$v;
1061 $self->_assert_bindval_matches_bindtype(@bind);
1064 } elsif ($r eq 'SCALAR') { # literal SQL without bind
1065 # embedded literal SQL
1072 push @sql, '(' . join(', ', @sqlq) . ')';
1073 } elsif ($ref eq 'SCALAR') {
1077 # strings get case twiddled
1078 push @sql, $self->_sqlcase($_);
1082 my $sql = join ' ', @sql;
1084 # this is pretty tricky
1085 # if ask for an array, return ($stmt, @bind)
1086 # otherwise, s/?/shift @sqlv/ to put it inline
1088 return ($sql, @sqlv);
1090 1 while $sql =~ s/\?/my $d = shift(@sqlv);
1091 ref $d ? $d->[1] : $d/e;
1100 # This allows us to check for a local, then _form, attr
1102 my($name) = $AUTOLOAD =~ /.*::(.+)/;
1103 return $self->generate($name, @_);
1114 SQL::Abstract - Generate SQL from Perl data structures
1120 my $sql = SQL::Abstract->new;
1122 my($stmt, @bind) = $sql->select($table, \@fields, \%where, \@order);
1124 my($stmt, @bind) = $sql->insert($table, \%fieldvals || \@values);
1126 my($stmt, @bind) = $sql->update($table, \%fieldvals, \%where);
1128 my($stmt, @bind) = $sql->delete($table, \%where);
1130 # Then, use these in your DBI statements
1131 my $sth = $dbh->prepare($stmt);
1132 $sth->execute(@bind);
1134 # Just generate the WHERE clause
1135 my($stmt, @bind) = $sql->where(\%where, \@order);
1137 # Return values in the same order, for hashed queries
1138 # See PERFORMANCE section for more details
1139 my @bind = $sql->values(\%fieldvals);
1143 This module was inspired by the excellent L<DBIx::Abstract>.
1144 However, in using that module I found that what I really wanted
1145 to do was generate SQL, but still retain complete control over my
1146 statement handles and use the DBI interface. So, I set out to
1147 create an abstract SQL generation module.
1149 While based on the concepts used by L<DBIx::Abstract>, there are
1150 several important differences, especially when it comes to WHERE
1151 clauses. I have modified the concepts used to make the SQL easier
1152 to generate from Perl data structures and, IMO, more intuitive.
1153 The underlying idea is for this module to do what you mean, based
1154 on the data structures you provide it. The big advantage is that
1155 you don't have to modify your code every time your data changes,
1156 as this module figures it out.
1158 To begin with, an SQL INSERT is as easy as just specifying a hash
1159 of C<key=value> pairs:
1162 name => 'Jimbo Bobson',
1163 phone => '123-456-7890',
1164 address => '42 Sister Lane',
1165 city => 'St. Louis',
1166 state => 'Louisiana',
1169 The SQL can then be generated with this:
1171 my($stmt, @bind) = $sql->insert('people', \%data);
1173 Which would give you something like this:
1175 $stmt = "INSERT INTO people
1176 (address, city, name, phone, state)
1177 VALUES (?, ?, ?, ?, ?)";
1178 @bind = ('42 Sister Lane', 'St. Louis', 'Jimbo Bobson',
1179 '123-456-7890', 'Louisiana');
1181 These are then used directly in your DBI code:
1183 my $sth = $dbh->prepare($stmt);
1184 $sth->execute(@bind);
1186 =head2 Inserting and Updating Arrays
1188 If your database has array types (like for example Postgres),
1189 activate the special option C<< array_datatypes => 1 >>
1190 when creating the C<SQL::Abstract> object.
1191 Then you may use an arrayref to insert and update database array types:
1193 my $sql = SQL::Abstract->new(array_datatypes => 1);
1195 planets => [qw/Mercury Venus Earth Mars/]
1198 my($stmt, @bind) = $sql->insert('solar_system', \%data);
1202 $stmt = "INSERT INTO solar_system (planets) VALUES (?)"
1204 @bind = (['Mercury', 'Venus', 'Earth', 'Mars']);
1207 =head2 Inserting and Updating SQL
1209 In order to apply SQL functions to elements of your C<%data> you may
1210 specify a reference to an arrayref for the given hash value. For example,
1211 if you need to execute the Oracle C<to_date> function on a value, you can
1212 say something like this:
1216 date_entered => \["to_date(?,'MM/DD/YYYY')", "03/02/2003"],
1219 The first value in the array is the actual SQL. Any other values are
1220 optional and would be included in the bind values array. This gives
1223 my($stmt, @bind) = $sql->insert('people', \%data);
1225 $stmt = "INSERT INTO people (name, date_entered)
1226 VALUES (?, to_date(?,'MM/DD/YYYY'))";
1227 @bind = ('Bill', '03/02/2003');
1229 An UPDATE is just as easy, all you change is the name of the function:
1231 my($stmt, @bind) = $sql->update('people', \%data);
1233 Notice that your C<%data> isn't touched; the module will generate
1234 the appropriately quirky SQL for you automatically. Usually you'll
1235 want to specify a WHERE clause for your UPDATE, though, which is
1236 where handling C<%where> hashes comes in handy...
1238 =head2 Complex where statements
1240 This module can generate pretty complicated WHERE statements
1241 easily. For example, simple C<key=value> pairs are taken to mean
1242 equality, and if you want to see if a field is within a set
1243 of values, you can use an arrayref. Let's say we wanted to
1244 SELECT some data based on this criteria:
1247 requestor => 'inna',
1248 worker => ['nwiger', 'rcwe', 'sfz'],
1249 status => { '!=', 'completed' }
1252 my($stmt, @bind) = $sql->select('tickets', '*', \%where);
1254 The above would give you something like this:
1256 $stmt = "SELECT * FROM tickets WHERE
1257 ( requestor = ? ) AND ( status != ? )
1258 AND ( worker = ? OR worker = ? OR worker = ? )";
1259 @bind = ('inna', 'completed', 'nwiger', 'rcwe', 'sfz');
1261 Which you could then use in DBI code like so:
1263 my $sth = $dbh->prepare($stmt);
1264 $sth->execute(@bind);
1270 The functions are simple. There's one for each major SQL operation,
1271 and a constructor you use first. The arguments are specified in a
1272 similar order to each function (table, then fields, then a where
1273 clause) to try and simplify things.
1278 =head2 new(option => 'value')
1280 The C<new()> function takes a list of options and values, and returns
1281 a new B<SQL::Abstract> object which can then be used to generate SQL
1282 through the methods below. The options accepted are:
1288 If set to 'lower', then SQL will be generated in all lowercase. By
1289 default SQL is generated in "textbook" case meaning something like:
1291 SELECT a_field FROM a_table WHERE some_field LIKE '%someval%'
1293 Any setting other than 'lower' is ignored.
1297 This determines what the default comparison operator is. By default
1298 it is C<=>, meaning that a hash like this:
1300 %where = (name => 'nwiger', email => 'nate@wiger.org');
1302 Will generate SQL like this:
1304 WHERE name = 'nwiger' AND email = 'nate@wiger.org'
1306 However, you may want loose comparisons by default, so if you set
1307 C<cmp> to C<like> you would get SQL such as:
1309 WHERE name like 'nwiger' AND email like 'nate@wiger.org'
1311 You can also override the comparsion on an individual basis - see
1312 the huge section on L</"WHERE CLAUSES"> at the bottom.
1314 =item sqltrue, sqlfalse
1316 Expressions for inserting boolean values within SQL statements.
1317 By default these are C<1=1> and C<1=0>.
1321 This determines the default logical operator for multiple WHERE
1322 statements in arrays or hashes. If absent, the default logic is "or"
1323 for arrays, and "and" for hashes. This means that a WHERE
1327 event_date => {'>=', '2/13/99'},
1328 event_date => {'<=', '4/24/03'},
1331 will generate SQL like this:
1333 WHERE event_date >= '2/13/99' OR event_date <= '4/24/03'
1335 This is probably not what you want given this query, though (look
1336 at the dates). To change the "OR" to an "AND", simply specify:
1338 my $sql = SQL::Abstract->new(logic => 'and');
1340 Which will change the above C<WHERE> to:
1342 WHERE event_date >= '2/13/99' AND event_date <= '4/24/03'
1344 The logic can also be changed locally by inserting
1345 a modifier in front of an arrayref :
1347 @where = (-and => [event_date => {'>=', '2/13/99'},
1348 event_date => {'<=', '4/24/03'} ]);
1350 See the L</"WHERE CLAUSES"> section for explanations.
1354 This will automatically convert comparisons using the specified SQL
1355 function for both column and value. This is mostly used with an argument
1356 of C<upper> or C<lower>, so that the SQL will have the effect of
1357 case-insensitive "searches". For example, this:
1359 $sql = SQL::Abstract->new(convert => 'upper');
1360 %where = (keywords => 'MaKe iT CAse inSeNSItive');
1362 Will turn out the following SQL:
1364 WHERE upper(keywords) like upper('MaKe iT CAse inSeNSItive')
1366 The conversion can be C<upper()>, C<lower()>, or any other SQL function
1367 that can be applied symmetrically to fields (actually B<SQL::Abstract> does
1368 not validate this option; it will just pass through what you specify verbatim).
1372 This is a kludge because many databases suck. For example, you can't
1373 just bind values using DBI's C<execute()> for Oracle C<CLOB> or C<BLOB> fields.
1374 Instead, you have to use C<bind_param()>:
1376 $sth->bind_param(1, 'reg data');
1377 $sth->bind_param(2, $lots, {ora_type => ORA_CLOB});
1379 The problem is, B<SQL::Abstract> will normally just return a C<@bind> array,
1380 which loses track of which field each slot refers to. Fear not.
1382 If you specify C<bindtype> in new, you can determine how C<@bind> is returned.
1383 Currently, you can specify either C<normal> (default) or C<columns>. If you
1384 specify C<columns>, you will get an array that looks like this:
1386 my $sql = SQL::Abstract->new(bindtype => 'columns');
1387 my($stmt, @bind) = $sql->insert(...);
1390 [ 'column1', 'value1' ],
1391 [ 'column2', 'value2' ],
1392 [ 'column3', 'value3' ],
1395 You can then iterate through this manually, using DBI's C<bind_param()>.
1397 $sth->prepare($stmt);
1400 my($col, $data) = @$_;
1401 if ($col eq 'details' || $col eq 'comments') {
1402 $sth->bind_param($i, $data, {ora_type => ORA_CLOB});
1403 } elsif ($col eq 'image') {
1404 $sth->bind_param($i, $data, {ora_type => ORA_BLOB});
1406 $sth->bind_param($i, $data);
1410 $sth->execute; # execute without @bind now
1412 Now, why would you still use B<SQL::Abstract> if you have to do this crap?
1413 Basically, the advantage is still that you don't have to care which fields
1414 are or are not included. You could wrap that above C<for> loop in a simple
1415 sub called C<bind_fields()> or something and reuse it repeatedly. You still
1416 get a layer of abstraction over manual SQL specification.
1418 Note that if you set L</bindtype> to C<columns>, the C<\[$sql, @bind]>
1419 construct (see L</Literal SQL with placeholders and bind values (subqueries)>)
1420 will expect the bind values in this format.
1424 This is the character that a table or column name will be quoted
1425 with. By default this is an empty string, but you could set it to
1426 the character C<`>, to generate SQL like this:
1428 SELECT `a_field` FROM `a_table` WHERE `some_field` LIKE '%someval%'
1430 Alternatively, you can supply an array ref of two items, the first being the left
1431 hand quote character, and the second the right hand quote character. For
1432 example, you could supply C<['[',']']> for SQL Server 2000 compliant quotes
1433 that generates SQL like this:
1435 SELECT [a_field] FROM [a_table] WHERE [some_field] LIKE '%someval%'
1437 Quoting is useful if you have tables or columns names that are reserved
1438 words in your database's SQL dialect.
1442 This is the character that separates a table and column name. It is
1443 necessary to specify this when the C<quote_char> option is selected,
1444 so that tables and column names can be individually quoted like this:
1446 SELECT `table`.`one_field` FROM `table` WHERE `table`.`other_field` = 1
1448 =item array_datatypes
1450 When this option is true, arrayrefs in INSERT or UPDATE are
1451 interpreted as array datatypes and are passed directly
1453 When this option is false, arrayrefs are interpreted
1454 as literal SQL, just like refs to arrayrefs
1455 (but this behavior is for backwards compatibility; when writing
1456 new queries, use the "reference to arrayref" syntax
1462 Takes a reference to a list of "special operators"
1463 to extend the syntax understood by L<SQL::Abstract>.
1464 See section L</"SPECIAL OPERATORS"> for details.
1470 =head2 insert($table, \@values || \%fieldvals)
1472 This is the simplest function. You simply give it a table name
1473 and either an arrayref of values or hashref of field/value pairs.
1474 It returns an SQL INSERT statement and a list of bind values.
1475 See the sections on L</"Inserting and Updating Arrays"> and
1476 L</"Inserting and Updating SQL"> for information on how to insert
1477 with those data types.
1479 =head2 update($table, \%fieldvals, \%where)
1481 This takes a table, hashref of field/value pairs, and an optional
1482 hashref L<WHERE clause|/WHERE CLAUSES>. It returns an SQL UPDATE function and a list
1484 See the sections on L</"Inserting and Updating Arrays"> and
1485 L</"Inserting and Updating SQL"> for information on how to insert
1486 with those data types.
1488 =head2 select($source, $fields, $where, $order)
1490 This returns a SQL SELECT statement and associated list of bind values, as
1491 specified by the arguments :
1497 Specification of the 'FROM' part of the statement.
1498 The argument can be either a plain scalar (interpreted as a table
1499 name, will be quoted), or an arrayref (interpreted as a list
1500 of table names, joined by commas, quoted), or a scalarref
1501 (literal table name, not quoted), or a ref to an arrayref
1502 (list of literal table names, joined by commas, not quoted).
1506 Specification of the list of fields to retrieve from
1508 The argument can be either an arrayref (interpreted as a list
1509 of field names, will be joined by commas and quoted), or a
1510 plain scalar (literal SQL, not quoted).
1511 Please observe that this API is not as flexible as for
1512 the first argument C<$table>, for backwards compatibility reasons.
1516 Optional argument to specify the WHERE part of the query.
1517 The argument is most often a hashref, but can also be
1518 an arrayref or plain scalar --
1519 see section L<WHERE clause|/"WHERE CLAUSES"> for details.
1523 Optional argument to specify the ORDER BY part of the query.
1524 The argument can be a scalar, a hashref or an arrayref
1525 -- see section L<ORDER BY clause|/"ORDER BY CLAUSES">
1531 =head2 delete($table, \%where)
1533 This takes a table name and optional hashref L<WHERE clause|/WHERE CLAUSES>.
1534 It returns an SQL DELETE statement and list of bind values.
1536 =head2 where(\%where, \@order)
1538 This is used to generate just the WHERE clause. For example,
1539 if you have an arbitrary data structure and know what the
1540 rest of your SQL is going to look like, but want an easy way
1541 to produce a WHERE clause, use this. It returns an SQL WHERE
1542 clause and list of bind values.
1545 =head2 values(\%data)
1547 This just returns the values from the hash C<%data>, in the same
1548 order that would be returned from any of the other above queries.
1549 Using this allows you to markedly speed up your queries if you
1550 are affecting lots of rows. See below under the L</"PERFORMANCE"> section.
1552 =head2 generate($any, 'number', $of, \@data, $struct, \%types)
1554 Warning: This is an experimental method and subject to change.
1556 This returns arbitrarily generated SQL. It's a really basic shortcut.
1557 It will return two different things, depending on return context:
1559 my($stmt, @bind) = $sql->generate('create table', \$table, \@fields);
1560 my $stmt_and_val = $sql->generate('create table', \$table, \@fields);
1562 These would return the following:
1564 # First calling form
1565 $stmt = "CREATE TABLE test (?, ?)";
1566 @bind = (field1, field2);
1568 # Second calling form
1569 $stmt_and_val = "CREATE TABLE test (field1, field2)";
1571 Depending on what you're trying to do, it's up to you to choose the correct
1572 format. In this example, the second form is what you would want.
1576 $sql->generate('alter session', { nls_date_format => 'MM/YY' });
1580 ALTER SESSION SET nls_date_format = 'MM/YY'
1582 You get the idea. Strings get their case twiddled, but everything
1583 else remains verbatim.
1588 =head1 WHERE CLAUSES
1592 This module uses a variation on the idea from L<DBIx::Abstract>. It
1593 is B<NOT>, repeat I<not> 100% compatible. B<The main logic of this
1594 module is that things in arrays are OR'ed, and things in hashes
1597 The easiest way to explain is to show lots of examples. After
1598 each C<%where> hash shown, it is assumed you used:
1600 my($stmt, @bind) = $sql->where(\%where);
1602 However, note that the C<%where> hash can be used directly in any
1603 of the other functions as well, as described above.
1605 =head2 Key-value pairs
1607 So, let's get started. To begin, a simple hash:
1611 status => 'completed'
1614 Is converted to SQL C<key = val> statements:
1616 $stmt = "WHERE user = ? AND status = ?";
1617 @bind = ('nwiger', 'completed');
1619 One common thing I end up doing is having a list of values that
1620 a field can be in. To do this, simply specify a list inside of
1625 status => ['assigned', 'in-progress', 'pending'];
1628 This simple code will create the following:
1630 $stmt = "WHERE user = ? AND ( status = ? OR status = ? OR status = ? )";
1631 @bind = ('nwiger', 'assigned', 'in-progress', 'pending');
1633 A field associated to an empty arrayref will be considered a
1634 logical false and will generate 0=1.
1636 =head2 Key-value pairs
1638 If you want to specify a different type of operator for your comparison,
1639 you can use a hashref for a given column:
1643 status => { '!=', 'completed' }
1646 Which would generate:
1648 $stmt = "WHERE user = ? AND status != ?";
1649 @bind = ('nwiger', 'completed');
1651 To test against multiple values, just enclose the values in an arrayref:
1653 status => { '=', ['assigned', 'in-progress', 'pending'] };
1655 Which would give you:
1657 "WHERE status = ? OR status = ? OR status = ?"
1660 The hashref can also contain multiple pairs, in which case it is expanded
1661 into an C<AND> of its elements:
1665 status => { '!=', 'completed', -not_like => 'pending%' }
1668 # Or more dynamically, like from a form
1669 $where{user} = 'nwiger';
1670 $where{status}{'!='} = 'completed';
1671 $where{status}{'-not_like'} = 'pending%';
1673 # Both generate this
1674 $stmt = "WHERE user = ? AND status != ? AND status NOT LIKE ?";
1675 @bind = ('nwiger', 'completed', 'pending%');
1678 To get an OR instead, you can combine it with the arrayref idea:
1682 priority => [ {'=', 2}, {'!=', 1} ]
1685 Which would generate:
1687 $stmt = "WHERE user = ? AND priority = ? OR priority != ?";
1688 @bind = ('nwiger', '2', '1');
1690 If you want to include literal SQL (with or without bind values), just use a
1691 scalar reference or array reference as the value:
1694 date_entered => { '>' => \["to_date(?, 'MM/DD/YYYY')", "11/26/2008"] },
1695 date_expires => { '<' => \"now()" }
1698 Which would generate:
1700 $stmt = "WHERE date_entered > "to_date(?, 'MM/DD/YYYY') AND date_expires < now()";
1701 @bind = ('11/26/2008');
1704 =head2 Logic and nesting operators
1706 In the example above,
1707 there is a subtle trap if you want to say something like
1708 this (notice the C<AND>):
1710 WHERE priority != ? AND priority != ?
1712 Because, in Perl you I<can't> do this:
1714 priority => { '!=', 2, '!=', 1 }
1716 As the second C<!=> key will obliterate the first. The solution
1717 is to use the special C<-modifier> form inside an arrayref:
1719 priority => [ -and => {'!=', 2},
1723 Normally, these would be joined by C<OR>, but the modifier tells it
1724 to use C<AND> instead. (Hint: You can use this in conjunction with the
1725 C<logic> option to C<new()> in order to change the way your queries
1726 work by default.) B<Important:> Note that the C<-modifier> goes
1727 B<INSIDE> the arrayref, as an extra first element. This will
1728 B<NOT> do what you think it might:
1730 priority => -and => [{'!=', 2}, {'!=', 1}] # WRONG!
1732 Here is a quick list of equivalencies, since there is some overlap:
1735 status => {'!=', 'completed', 'not like', 'pending%' }
1736 status => [ -and => {'!=', 'completed'}, {'not like', 'pending%'}]
1739 status => {'=', ['assigned', 'in-progress']}
1740 status => [ -or => {'=', 'assigned'}, {'=', 'in-progress'}]
1741 status => [ {'=', 'assigned'}, {'=', 'in-progress'} ]
1743 In addition to C<-and> and C<-or>, there is also a special C<-nest>
1744 operator which adds an additional set of parens, to create a subquery.
1745 For example, to get something like this:
1747 $stmt = "WHERE user = ? AND ( workhrs > ? OR geo = ? )";
1748 @bind = ('nwiger', '20', 'ASIA');
1754 -nest => [ workhrs => {'>', 20}, geo => 'ASIA' ],
1757 If you need several nested subexpressions, you can number
1758 the C<-nest> branches :
1768 =head2 Special operators : IN, BETWEEN, etc.
1770 You can also use the hashref format to compare a list of fields using the
1771 C<IN> comparison operator, by specifying the list as an arrayref:
1774 status => 'completed',
1775 reportid => { -in => [567, 2335, 2] }
1778 Which would generate:
1780 $stmt = "WHERE status = ? AND reportid IN (?,?,?)";
1781 @bind = ('completed', '567', '2335', '2');
1783 The reverse operator C<-not_in> generates SQL C<NOT IN> and is used in
1786 Another pair of operators is C<-between> and C<-not_between>,
1787 used with an arrayref of two values:
1791 completion_date => {
1792 -not_between => ['2002-10-01', '2003-02-06']
1798 WHERE user = ? AND completion_date NOT BETWEEN ( ? AND ? )
1800 These are the two builtin "special operators"; but the
1801 list can be expanded : see section L</"SPECIAL OPERATORS"> below.
1803 =head2 Nested conditions
1805 So far, we've seen how multiple conditions are joined with a top-level
1806 C<AND>. We can change this by putting the different conditions we want in
1807 hashes and then putting those hashes in an array. For example:
1812 status => { -like => ['pending%', 'dispatched'] },
1816 status => 'unassigned',
1820 This data structure would create the following:
1822 $stmt = "WHERE ( user = ? AND ( status LIKE ? OR status LIKE ? ) )
1823 OR ( user = ? AND status = ? ) )";
1824 @bind = ('nwiger', 'pending', 'dispatched', 'robot', 'unassigned');
1826 This can be combined with the C<-nest> operator to properly group
1827 SQL statements. Furthermore, hashrefs or arrayrefs can be
1828 prefixed with an C<-and> or C<-or> to change the logic
1835 -and => [workhrs => {'>', 20}, geo => 'ASIA' ],
1836 -and => [workhrs => {'<', 50}, geo => 'EURO' ]
1843 WHERE ( user = ? AND
1844 ( ( workhrs > ? AND geo = ? )
1845 OR ( workhrs < ? AND geo = ? ) ) )
1847 C<Important note>: when connecting several conditions, the C<-and->|C<-or>
1848 operator goes C<outside> of the nested structure; whereas when connecting
1849 several constraints on one column, the C<-and> operator goes
1850 C<inside> the arrayref. Here is an example combining both features :
1853 -and => [a => 1, b => 2],
1854 -or => [c => 3, d => 4],
1855 e => [-and => {-like => 'foo%'}, {-like => '%bar'} ]
1860 WHERE ( ( ( a = ? AND b = ? )
1861 OR ( c = ? OR d = ? )
1862 OR ( e LIKE ? AND e LIKE ? ) ) )
1867 Finally, sometimes only literal SQL will do. If you want to include
1868 literal SQL verbatim, you can specify it as a scalar reference, namely:
1870 my $inn = 'is Not Null';
1872 priority => { '<', 2 },
1878 $stmt = "WHERE priority < ? AND requestor is Not Null";
1881 Note that in this example, you only get one bind parameter back, since
1882 the verbatim SQL is passed as part of the statement.
1884 Of course, just to prove a point, the above can also be accomplished
1888 priority => { '<', 2 },
1889 requestor => { '!=', undef },
1895 Conditions on boolean columns can be expressed in the
1896 same way, passing a reference to an empty string :
1899 priority => { '<', 2 },
1905 $stmt = "WHERE priority < ? AND is_ready";
1909 =head2 Literal SQL with placeholders and bind values (subqueries)
1911 If the literal SQL to be inserted has placeholders and bind values,
1912 use a reference to an arrayref (yes this is a double reference --
1913 not so common, but perfectly legal Perl). For example, to find a date
1914 in Postgres you can use something like this:
1917 date_column => \[q/= date '2008-09-30' - ?::integer/, 10/]
1922 $stmt = "WHERE ( date_column = date '2008-09-30' - ?::integer )"
1925 Note that you must pass the bind values in the same format as they are returned
1926 by L</where>. That means that if you set L</bindtype> to C<columns>, you must
1927 provide the bind values in the C<< [ column_meta => value ] >> format, where
1928 C<column_meta> is an opaque scalar value; most commonly the column name, but
1929 you can use any scalar value (including references and blessed references),
1930 L<SQL::Abstract> will simply pass it through intact. So if C<bindtype> is set
1931 to C<columns> the above example will look like:
1934 date_column => \[q/= date '2008-09-30' - ?::integer/, [ dummy => 10 ]/]
1937 Literal SQL is especially useful for nesting parenthesized clauses in the
1938 main SQL query. Here is a first example :
1940 my ($sub_stmt, @sub_bind) = ("SELECT c1 FROM t1 WHERE c2 < ? AND c3 LIKE ?",
1944 bar => \["IN ($sub_stmt)" => @sub_bind],
1949 $stmt = "WHERE (foo = ? AND bar IN (SELECT c1 FROM t1
1950 WHERE c2 < ? AND c3 LIKE ?))";
1951 @bind = (1234, 100, "foo%");
1953 Other subquery operators, like for example C<"E<gt> ALL"> or C<"NOT IN">,
1954 are expressed in the same way. Of course the C<$sub_stmt> and
1955 its associated bind values can be generated through a former call
1958 my ($sub_stmt, @sub_bind)
1959 = $sql->select("t1", "c1", {c2 => {"<" => 100},
1960 c3 => {-like => "foo%"}});
1963 bar => \["> ALL ($sub_stmt)" => @sub_bind],
1966 In the examples above, the subquery was used as an operator on a column;
1967 but the same principle also applies for a clause within the main C<%where>
1968 hash, like an EXISTS subquery :
1970 my ($sub_stmt, @sub_bind)
1971 = $sql->select("t1", "*", {c1 => 1, c2 => \"> t0.c0"});
1974 -nest => \["EXISTS ($sub_stmt)" => @sub_bind],
1979 $stmt = "WHERE (foo = ? AND EXISTS (SELECT * FROM t1
1980 WHERE c1 = ? AND c2 > t0.c0))";
1984 Observe that the condition on C<c2> in the subquery refers to
1985 column C<t0.c0> of the main query : this is I<not> a bind
1986 value, so we have to express it through a scalar ref.
1987 Writing C<< c2 => {">" => "t0.c0"} >> would have generated
1988 C<< c2 > ? >> with bind value C<"t0.c0"> ... not exactly
1989 what we wanted here.
1991 Another use of the subquery technique is when some SQL clauses need
1992 parentheses, as it often occurs with some proprietary SQL extensions
1993 like for example fulltext expressions, geospatial expressions,
1994 NATIVE clauses, etc. Here is an example of a fulltext query in MySQL :
1997 -nest => \["MATCH (col1, col2) AGAINST (?)" => qw/apples/]
2000 Finally, here is an example where a subquery is used
2001 for expressing unary negation:
2003 my ($sub_stmt, @sub_bind)
2004 = $sql->where({age => [{"<" => 10}, {">" => 20}]});
2005 $sub_stmt =~ s/^ where //i; # don't want "WHERE" in the subclause
2007 lname => {like => '%son%'},
2008 -nest => \["NOT ($sub_stmt)" => @sub_bind],
2013 $stmt = "lname LIKE ? AND NOT ( age < ? OR age > ? )"
2014 @bind = ('%son%', 10, 20)
2020 These pages could go on for a while, since the nesting of the data
2021 structures this module can handle are pretty much unlimited (the
2022 module implements the C<WHERE> expansion as a recursive function
2023 internally). Your best bet is to "play around" with the module a
2024 little to see how the data structures behave, and choose the best
2025 format for your data based on that.
2027 And of course, all the values above will probably be replaced with
2028 variables gotten from forms or the command line. After all, if you
2029 knew everything ahead of time, you wouldn't have to worry about
2030 dynamically-generating SQL and could just hardwire it into your
2036 =head1 ORDER BY CLAUSES
2038 Some functions take an order by clause. This can either be a scalar (just a
2039 column name,) a hash of C<< { -desc => 'col' } >> or C<< { -asc => 'col' } >>,
2040 or an array of either of the two previous forms. Examples:
2042 Given | Will Generate
2043 ----------------------------------------------------------
2044 \'colA DESC' | ORDER BY colA DESC
2045 'colA' | ORDER BY colA
2046 [qw/colA colB/] | ORDER BY colA, colB
2047 {-asc => 'colA'} | ORDER BY colA ASC
2048 {-desc => 'colB'} | ORDER BY colB DESC
2050 {-asc => 'colA'}, | ORDER BY colA ASC, colB DESC
2053 [colA => {-asc => 'colB'}] | ORDER BY colA, colB ASC
2054 ==========================================================
2058 =head1 SPECIAL OPERATORS
2060 my $sqlmaker = SQL::Abstract->new(special_ops => [
2063 my ($self, $field, $op, $arg) = @_;
2069 A "special operator" is a SQL syntactic clause that can be
2070 applied to a field, instead of a usual binary operator.
2073 WHERE field IN (?, ?, ?)
2074 WHERE field BETWEEN ? AND ?
2075 WHERE MATCH(field) AGAINST (?, ?)
2077 Special operators IN and BETWEEN are fairly standard and therefore
2078 are builtin within C<SQL::Abstract>. For other operators,
2079 like the MATCH .. AGAINST example above which is
2080 specific to MySQL, you can write your own operator handlers :
2081 supply a C<special_ops> argument to the C<new> method.
2082 That argument takes an arrayref of operator definitions;
2083 each operator definition is a hashref with two entries
2089 the regular expression to match the operator
2093 coderef that will be called when meeting that operator
2094 in the input tree. The coderef will be called with
2095 arguments C<< ($self, $field, $op, $arg) >>, and
2096 should return a C<< ($sql, @bind) >> structure.
2100 For example, here is an implementation
2101 of the MATCH .. AGAINST syntax for MySQL
2103 my $sqlmaker = SQL::Abstract->new(special_ops => [
2105 # special op for MySql MATCH (field) AGAINST(word1, word2, ...)
2106 {regex => qr/^match$/i,
2108 my ($self, $field, $op, $arg) = @_;
2109 $arg = [$arg] if not ref $arg;
2110 my $label = $self->_quote($field);
2111 my ($placeholder) = $self->_convert('?');
2112 my $placeholders = join ", ", (($placeholder) x @$arg);
2113 my $sql = $self->_sqlcase('match') . " ($label) "
2114 . $self->_sqlcase('against') . " ($placeholders) ";
2115 my @bind = $self->_bindtype($field, @$arg);
2116 return ($sql, @bind);
2125 Thanks to some benchmarking by Mark Stosberg, it turns out that
2126 this module is many orders of magnitude faster than using C<DBIx::Abstract>.
2127 I must admit this wasn't an intentional design issue, but it's a
2128 byproduct of the fact that you get to control your C<DBI> handles
2131 To maximize performance, use a code snippet like the following:
2133 # prepare a statement handle using the first row
2134 # and then reuse it for the rest of the rows
2136 for my $href (@array_of_hashrefs) {
2137 $stmt ||= $sql->insert('table', $href);
2138 $sth ||= $dbh->prepare($stmt);
2139 $sth->execute($sql->values($href));
2142 The reason this works is because the keys in your C<$href> are sorted
2143 internally by B<SQL::Abstract>. Thus, as long as your data retains
2144 the same structure, you only have to generate the SQL the first time
2145 around. On subsequent queries, simply use the C<values> function provided
2146 by this module to return your values in the correct order.
2151 If you use my C<CGI::FormBuilder> module at all, you'll hopefully
2152 really like this part (I do, at least). Building up a complex query
2153 can be as simple as the following:
2157 use CGI::FormBuilder;
2160 my $form = CGI::FormBuilder->new(...);
2161 my $sql = SQL::Abstract->new;
2163 if ($form->submitted) {
2164 my $field = $form->field;
2165 my $id = delete $field->{id};
2166 my($stmt, @bind) = $sql->update('table', $field, {id => $id});
2169 Of course, you would still have to connect using C<DBI> to run the
2170 query, but the point is that if you make your form look like your
2171 table, the actual query script can be extremely simplistic.
2173 If you're B<REALLY> lazy (I am), check out C<HTML::QuickTable> for
2174 a fast interface to returning and formatting data. I frequently
2175 use these three modules together to write complex database query
2176 apps in under 50 lines.
2181 Version 1.50 was a major internal refactoring of C<SQL::Abstract>.
2182 Great care has been taken to preserve the I<published> behavior
2183 documented in previous versions in the 1.* family; however,
2184 some features that were previously undocumented, or behaved
2185 differently from the documentation, had to be changed in order
2186 to clarify the semantics. Hence, client code that was relying
2187 on some dark areas of C<SQL::Abstract> v1.*
2188 B<might behave differently> in v1.50.
2190 The main changes are :
2196 support for literal SQL through the C<< \ [$sql, bind] >> syntax.
2200 support for the { operator => \"..." } construct (to embed literal SQL)
2204 support for the { operator => \["...", @bind] } construct (to embed literal SQL with bind values)
2208 added official support for -nest1, -nest2 or -nest_1, -nest_2, ...
2209 (undocumented in previous versions)
2213 optional support for L<array datatypes|/"Inserting and Updating Arrays">
2217 defensive programming : check arguments
2221 fixed bug with global logic, which was previously implemented
2222 through global variables yielding side-effects. Prior versions would
2223 interpret C<< [ {cond1, cond2}, [cond3, cond4] ] >>
2224 as C<< "(cond1 AND cond2) OR (cond3 AND cond4)" >>.
2225 Now this is interpreted
2226 as C<< "(cond1 AND cond2) OR (cond3 OR cond4)" >>.
2231 fixed semantics of _bindtype on array args
2235 dropped the C<_anoncopy> of the %where tree. No longer necessary,
2236 we just avoid shifting arrays within that tree.
2240 dropped the C<_modlogic> function
2246 =head1 ACKNOWLEDGEMENTS
2248 There are a number of individuals that have really helped out with
2249 this module. Unfortunately, most of them submitted bugs via CPAN
2250 so I have no idea who they are! But the people I do know are:
2252 Ash Berlin (order_by hash term support)
2253 Matt Trout (DBIx::Class support)
2254 Mark Stosberg (benchmarking)
2255 Chas Owens (initial "IN" operator support)
2256 Philip Collins (per-field SQL functions)
2257 Eric Kolve (hashref "AND" support)
2258 Mike Fragassi (enhancements to "BETWEEN" and "LIKE")
2259 Dan Kubb (support for "quote_char" and "name_sep")
2260 Guillermo Roditi (patch to cleanup "IN" and "BETWEEN", fix and tests for _order_by)
2261 Laurent Dami (internal refactoring, multiple -nest, extensible list of special operators, literal SQL)
2262 Norbert Buchmuller (support for literal SQL in hashpair, misc. fixes & tests)
2268 L<DBIx::Class>, L<DBIx::Abstract>, L<CGI::FormBuilder>, L<HTML::QuickTable>.
2272 Copyright (c) 2001-2007 Nathan Wiger <nwiger@cpan.org>. All Rights Reserved.
2274 This module is actively maintained by Matt Trout <mst@shadowcatsystems.co.uk>
2276 For support, your best bet is to try the C<DBIx::Class> users mailing list.
2277 While not an official support venue, C<DBIx::Class> makes heavy use of
2278 C<SQL::Abstract>, and as such list members there are very familiar with
2279 how to create queries.
2281 This module is free software; you may copy this under the terms of
2282 the GNU General Public License, or the Artistic License, copies of
2283 which should have accompanied your Perl kit.