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 # LDNOTE : should deprecate {-or => [...]} and {-and => [...]}
456 # because they are misleading; the only proper way would be
457 # -nest => [-or => ...], -nest => [-and ...]
458 return $self->_where_ARRAYREF($v, $op eq 'NEST' ? '' : $op);
463 belch "-or => {...} should be -nest => [...]";
464 return $self->_where_ARRAYREF([%$v], 'OR');
467 return $self->_where_HASHREF($v);
471 SCALARREF => sub { # literal SQL
473 or puke "-$op => \\\$scalar not supported, use -nest => ...";
477 ARRAYREFREF => sub { # literal SQL
479 or puke "-$op => \\[..] not supported, use -nest => ...";
483 SCALAR => sub { # permissively interpreted as SQL
485 or puke "-$op => 'scalar' not supported, use -nest => \\'scalar'";
486 belch "literal SQL should be -nest => \\'scalar' "
487 . "instead of -nest => 'scalar' ";
492 puke "-$op => undef not supported";
498 sub _where_hashpair_ARRAYREF {
499 my ($self, $k, $v) = @_;
502 my @v = @$v; # need copy because of shift below
503 $self->_debug("ARRAY($k) means distribute over elements");
505 # put apart first element if it is an operator (-and, -or)
506 my $op = $v[0] =~ /^-/ ? shift @v : undef;
507 $self->_debug("OP($op) reinjected into the distributed array") if $op;
509 my @distributed = map { {$k => $_} } @v;
510 unshift @distributed, $op if $op;
511 my $logic = $op ? substr($op, 1) : '';
513 return $self->_recurse_where(\@distributed, $logic);
516 # LDNOTE : not sure of this one. What does "distribute over nothing" mean?
517 $self->_debug("empty ARRAY($k) means 0=1");
518 return ($self->{sqlfalse});
522 sub _where_hashpair_HASHREF {
523 my ($self, $k, $v) = @_;
525 my (@all_sql, @all_bind);
527 for my $op (sort keys %$v) {
530 # put the operator in canonical form
531 $op =~ s/^-//; # remove initial dash
532 $op =~ tr/_/ /; # underscores become spaces
533 $op =~ s/^\s+//; # no initial space
534 $op =~ s/\s+$//; # no final space
535 $op =~ s/\s+/ /; # multiple spaces become one
539 # CASE: special operators like -in or -between
540 my $special_op = first {$op =~ $_->{regex}} @{$self->{special_ops}};
542 ($sql, @bind) = $special_op->{handler}->($self, $k, $op, $val);
545 $self->_SWITCH_refkind($val, {
547 ARRAYREF => sub { # CASE: col => {op => \@vals}
548 ($sql, @bind) = $self->_where_field_op_ARRAYREF($k, $op, $val);
551 SCALARREF => sub { # CASE: col => {op => \$scalar} (literal SQL without bind)
552 $sql = join ' ', $self->_convert($self->_quote($k)),
553 $self->_sqlcase($op),
557 ARRAYREFREF => sub { # CASE: col => {op => \[$sql, @bind]} (literal SQL with bind)
558 my ($sub_sql, @sub_bind) = @$$val;
559 $self->_assert_bindval_matches_bindtype(@sub_bind);
560 $sql = join ' ', $self->_convert($self->_quote($k)),
561 $self->_sqlcase($op),
566 UNDEF => sub { # CASE: col => {op => undef} : sql "IS (NOT)? NULL"
567 my $is = ($op =~ $self->{equality_op}) ? 'is' :
568 ($op =~ $self->{inequality_op}) ? 'is not' :
569 puke "unexpected operator '$op' with undef operand";
570 $sql = $self->_quote($k) . $self->_sqlcase(" $is null");
573 FALLBACK => sub { # CASE: col => {op => $scalar}
574 $sql = join ' ', $self->_convert($self->_quote($k)),
575 $self->_sqlcase($op),
576 $self->_convert('?');
577 @bind = $self->_bindtype($k, $val);
583 push @all_bind, @bind;
586 return $self->_join_sql_clauses('and', \@all_sql, \@all_bind);
591 sub _where_field_op_ARRAYREF {
592 my ($self, $k, $op, $vals) = @_;
595 $self->_debug("ARRAY($vals) means multiple elements: [ @$vals ]");
599 # LDNOTE : change the distribution logic when
600 # $op =~ $self->{inequality_op}, because of Morgan laws :
601 # with {field => {'!=' => [22, 33]}}, it would be ridiculous to generate
602 # WHERE field != 22 OR field != 33 : the user probably means
603 # WHERE field != 22 AND field != 33.
604 my $logic = ($op =~ $self->{inequality_op}) ? 'AND' : 'OR';
606 # distribute $op over each member of @$vals
607 return $self->_recurse_where([map { {$k => {$op, $_}} } @$vals], $logic);
611 # try to DWIM on equality operators
612 # LDNOTE : not 100% sure this is the correct thing to do ...
613 return ($self->{sqlfalse}) if $op =~ $self->{equality_op};
614 return ($self->{sqltrue}) if $op =~ $self->{inequality_op};
617 puke "operator '$op' applied on an empty array (field '$k')";
622 sub _where_hashpair_SCALARREF {
623 my ($self, $k, $v) = @_;
624 $self->_debug("SCALAR($k) means literal SQL: $$v");
625 my $sql = $self->_quote($k) . " " . $$v;
629 # literal SQL with bind
630 sub _where_hashpair_ARRAYREFREF {
631 my ($self, $k, $v) = @_;
632 $self->_debug("REF($k) means literal SQL: @${$v}");
633 my ($sql, @bind) = @${$v};
634 $self->_assert_bindval_matches_bindtype(@bind);
635 $sql = $self->_quote($k) . " " . $sql;
636 return ($sql, @bind );
639 # literal SQL without bind
640 sub _where_hashpair_SCALAR {
641 my ($self, $k, $v) = @_;
642 $self->_debug("NOREF($k) means simple key=val: $k $self->{cmp} $v");
643 my $sql = join ' ', $self->_convert($self->_quote($k)),
644 $self->_sqlcase($self->{cmp}),
645 $self->_convert('?');
646 my @bind = $self->_bindtype($k, $v);
647 return ( $sql, @bind);
651 sub _where_hashpair_UNDEF {
652 my ($self, $k, $v) = @_;
653 $self->_debug("UNDEF($k) means IS NULL");
654 my $sql = $self->_quote($k) . $self->_sqlcase(' is null');
658 #======================================================================
659 # WHERE: TOP-LEVEL OTHERS (SCALARREF, SCALAR, UNDEF)
660 #======================================================================
663 sub _where_SCALARREF {
664 my ($self, $where) = @_;
667 $self->_debug("SCALAR(*top) means literal SQL: $$where");
673 my ($self, $where) = @_;
676 $self->_debug("NOREF(*top) means literal SQL: $where");
687 #======================================================================
688 # WHERE: BUILTIN SPECIAL OPERATORS (-in, -between)
689 #======================================================================
692 sub _where_field_BETWEEN {
693 my ($self, $k, $op, $vals) = @_;
695 ref $vals eq 'ARRAY' && @$vals == 2
696 or puke "special op 'between' requires an arrayref of two values";
698 my ($label) = $self->_convert($self->_quote($k));
699 my ($placeholder) = $self->_convert('?');
700 my $and = $self->_sqlcase('and');
701 $op = $self->_sqlcase($op);
703 my $sql = "( $label $op $placeholder $and $placeholder )";
704 my @bind = $self->_bindtype($k, @$vals);
709 sub _where_field_IN {
710 my ($self, $k, $op, $vals) = @_;
712 # backwards compatibility : if scalar, force into an arrayref
713 $vals = [$vals] if defined $vals && ! ref $vals;
715 my ($label) = $self->_convert($self->_quote($k));
716 my ($placeholder) = $self->_convert('?');
717 $op = $self->_sqlcase($op);
719 my ($sql, @bind) = $self->_SWITCH_refkind($vals, {
720 ARRAYREF => sub { # list of choices
721 if (@$vals) { # nonempty list
722 my $placeholders = join ", ", (($placeholder) x @$vals);
723 my $sql = "$label $op ( $placeholders )";
724 my @bind = $self->_bindtype($k, @$vals);
726 return ($sql, @bind);
728 else { # empty list : some databases won't understand "IN ()", so DWIM
729 my $sql = ($op =~ /\bnot\b/i) ? $self->{sqltrue} : $self->{sqlfalse};
734 ARRAYREFREF => sub { # literal SQL with bind
735 my ($sql, @bind) = @$$vals;
736 $self->_assert_bindval_matches_bindtype(@bind);
737 return ("$label $op ( $sql )", @bind);
741 puke "special op 'in' requires an arrayref (or arrayref-ref)";
745 return ($sql, @bind);
753 #======================================================================
755 #======================================================================
758 my ($self, $arg) = @_;
760 # construct list of ordering instructions
761 my @order = $self->_SWITCH_refkind($arg, {
764 map {$self->_SWITCH_refkind($_, {
765 SCALAR => sub {$self->_quote($_)},
767 SCALARREF => sub {$$_}, # literal SQL, no quoting
768 HASHREF => sub {$self->_order_by_hash($_)}
772 SCALAR => sub {$self->_quote($arg)},
774 SCALARREF => sub {$$arg}, # literal SQL, no quoting
775 HASHREF => sub {$self->_order_by_hash($arg)},
780 my $order = join ', ', @order;
781 return $order ? $self->_sqlcase(' order by')." $order" : '';
786 my ($self, $hash) = @_;
788 # get first pair in hash
789 my ($key, $val) = each %$hash;
791 # check if one pair was found and no other pair in hash
792 $key && !(each %$hash)
793 or puke "hash passed to _order_by must have exactly one key (-desc or -asc)";
795 my ($order) = ($key =~ /^-(desc|asc)/i)
796 or puke "invalid key in _order_by hash : $key";
798 return $self->_quote($val) ." ". $self->_sqlcase($order);
803 #======================================================================
804 # DATASOURCE (FOR NOW, JUST PLAIN TABLE OR LIST OF TABLES)
805 #======================================================================
810 $self->_SWITCH_refkind($from, {
811 ARRAYREF => sub {join ', ', map { $self->_quote($_) } @$from;},
812 SCALAR => sub {$self->_quote($from)},
813 SCALARREF => sub {$$from},
814 ARRAYREFREF => sub {join ', ', @$from;},
819 #======================================================================
821 #======================================================================
827 $label or puke "can't quote an empty label";
829 # left and right quote characters
830 my ($ql, $qr, @other) = $self->_SWITCH_refkind($self->{quote_char}, {
831 SCALAR => sub {($self->{quote_char}, $self->{quote_char})},
832 ARRAYREF => sub {@{$self->{quote_char}}},
836 or puke "quote_char must be an arrayref of 2 values";
838 # no quoting if no quoting chars
839 $ql or return $label;
841 # no quoting for literal SQL
842 return $$label if ref($label) eq 'SCALAR';
844 # separate table / column (if applicable)
845 my $sep = $self->{name_sep} || '';
846 my @to_quote = $sep ? split /\Q$sep\E/, $label : ($label);
848 # do the quoting, except for "*" or for `table`.*
849 my @quoted = map { $_ eq '*' ? $_: $ql.$_.$qr} @to_quote;
851 # reassemble and return.
852 return join $sep, @quoted;
856 # Conversion, if applicable
858 my ($self, $arg) = @_;
860 # LDNOTE : modified the previous implementation below because
861 # it was not consistent : the first "return" is always an array,
862 # the second "return" is context-dependent. Anyway, _convert
863 # seems always used with just a single argument, so make it a
865 # return @_ unless $self->{convert};
866 # my $conv = $self->_sqlcase($self->{convert});
867 # my @ret = map { $conv.'('.$_.')' } @_;
868 # return wantarray ? @ret : $ret[0];
869 if ($self->{convert}) {
870 my $conv = $self->_sqlcase($self->{convert});
871 $arg = $conv.'('.$arg.')';
879 my($col, @vals) = @_;
881 #LDNOTE : changed original implementation below because it did not make
882 # sense when bindtype eq 'columns' and @vals > 1.
883 # return $self->{bindtype} eq 'columns' ? [ $col, @vals ] : @vals;
885 return $self->{bindtype} eq 'columns' ? map {[$col, $_]} @vals : @vals;
888 # Dies if any element of @bind is not in [colname => value] format
889 # if bindtype is 'columns'.
890 sub _assert_bindval_matches_bindtype {
891 my ($self, @bind) = @_;
893 if ($self->{bindtype} eq 'columns') {
894 foreach my $val (@bind) {
895 if (!defined $val || ref($val) ne 'ARRAY' || @$val != 2) {
896 die "bindtype 'columns' selected, you need to pass: [column_name => bind_value]"
902 sub _join_sql_clauses {
903 my ($self, $logic, $clauses_aref, $bind_aref) = @_;
905 if (@$clauses_aref > 1) {
906 my $join = " " . $self->_sqlcase($logic) . " ";
907 my $sql = '( ' . join($join, @$clauses_aref) . ' )';
908 return ($sql, @$bind_aref);
910 elsif (@$clauses_aref) {
911 return ($clauses_aref->[0], @$bind_aref); # no parentheses
914 return (); # if no SQL, ignore @$bind_aref
919 # Fix SQL case, if so requested
923 # LDNOTE: if $self->{case} is true, then it contains 'lower', so we
924 # don't touch the argument ... crooked logic, but let's not change it!
925 return $self->{case} ? $_[0] : uc($_[0]);
929 #======================================================================
930 # DISPATCHING FROM REFKIND
931 #======================================================================
934 my ($self, $data) = @_;
940 # blessed objects are treated like scalars
941 $ref = (blessed $data) ? '' : ref $data;
942 $n_steps += 1 if $ref;
943 last if $ref ne 'REF';
947 my $base = $ref || (defined $data ? 'SCALAR' : 'UNDEF');
949 return $base . ('REF' x $n_steps);
955 my ($self, $data) = @_;
956 my @try = ($self->_refkind($data));
957 push @try, 'SCALAR_or_UNDEF' if $try[0] eq 'SCALAR' || $try[0] eq 'UNDEF';
958 push @try, 'FALLBACK';
962 sub _METHOD_FOR_refkind {
963 my ($self, $meth_prefix, $data) = @_;
964 my $method = first {$_} map {$self->can($meth_prefix."_".$_)}
965 $self->_try_refkind($data)
966 or puke "cannot dispatch on '$meth_prefix' for ".$self->_refkind($data);
971 sub _SWITCH_refkind {
972 my ($self, $data, $dispatch_table) = @_;
974 my $coderef = first {$_} map {$dispatch_table->{$_}}
975 $self->_try_refkind($data)
976 or puke "no dispatch entry for ".$self->_refkind($data);
983 #======================================================================
984 # VALUES, GENERATE, AUTOLOAD
985 #======================================================================
987 # LDNOTE: original code from nwiger, didn't touch code in that section
988 # I feel the AUTOLOAD stuff should not be the default, it should
989 # only be activated on explicit demand by user.
993 my $data = shift || return;
994 puke "Argument to ", __PACKAGE__, "->values must be a \\%hash"
995 unless ref $data eq 'HASH';
998 foreach my $k ( sort keys %$data ) {
1000 $self->_SWITCH_refkind($v, {
1002 if ($self->{array_datatypes}) { # array datatype
1003 push @all_bind, $self->_bindtype($k, $v);
1005 else { # literal SQL with bind
1006 my ($sql, @bind) = @$v;
1007 $self->_assert_bindval_matches_bindtype(@bind);
1008 push @all_bind, @bind;
1011 ARRAYREFREF => sub { # literal SQL with bind
1012 my ($sql, @bind) = @${$v};
1013 $self->_assert_bindval_matches_bindtype(@bind);
1014 push @all_bind, @bind;
1016 SCALARREF => sub { # literal SQL without bind
1018 SCALAR_or_UNDEF => sub {
1019 push @all_bind, $self->_bindtype($k, $v);
1030 my(@sql, @sqlq, @sqlv);
1034 if ($ref eq 'HASH') {
1035 for my $k (sort keys %$_) {
1038 my $label = $self->_quote($k);
1039 if ($r eq 'ARRAY') {
1040 # literal SQL with bind
1041 my ($sql, @bind) = @$v;
1042 $self->_assert_bindval_matches_bindtype(@bind);
1043 push @sqlq, "$label = $sql";
1045 } elsif ($r eq 'SCALAR') {
1046 # literal SQL without bind
1047 push @sqlq, "$label = $$v";
1049 push @sqlq, "$label = ?";
1050 push @sqlv, $self->_bindtype($k, $v);
1053 push @sql, $self->_sqlcase('set'), join ', ', @sqlq;
1054 } elsif ($ref eq 'ARRAY') {
1055 # unlike insert(), assume these are ONLY the column names, i.e. for SQL
1058 if ($r eq 'ARRAY') { # literal SQL with bind
1059 my ($sql, @bind) = @$v;
1060 $self->_assert_bindval_matches_bindtype(@bind);
1063 } elsif ($r eq 'SCALAR') { # literal SQL without bind
1064 # embedded literal SQL
1071 push @sql, '(' . join(', ', @sqlq) . ')';
1072 } elsif ($ref eq 'SCALAR') {
1076 # strings get case twiddled
1077 push @sql, $self->_sqlcase($_);
1081 my $sql = join ' ', @sql;
1083 # this is pretty tricky
1084 # if ask for an array, return ($stmt, @bind)
1085 # otherwise, s/?/shift @sqlv/ to put it inline
1087 return ($sql, @sqlv);
1089 1 while $sql =~ s/\?/my $d = shift(@sqlv);
1090 ref $d ? $d->[1] : $d/e;
1099 # This allows us to check for a local, then _form, attr
1101 my($name) = $AUTOLOAD =~ /.*::(.+)/;
1102 return $self->generate($name, @_);
1113 SQL::Abstract - Generate SQL from Perl data structures
1119 my $sql = SQL::Abstract->new;
1121 my($stmt, @bind) = $sql->select($table, \@fields, \%where, \@order);
1123 my($stmt, @bind) = $sql->insert($table, \%fieldvals || \@values);
1125 my($stmt, @bind) = $sql->update($table, \%fieldvals, \%where);
1127 my($stmt, @bind) = $sql->delete($table, \%where);
1129 # Then, use these in your DBI statements
1130 my $sth = $dbh->prepare($stmt);
1131 $sth->execute(@bind);
1133 # Just generate the WHERE clause
1134 my($stmt, @bind) = $sql->where(\%where, \@order);
1136 # Return values in the same order, for hashed queries
1137 # See PERFORMANCE section for more details
1138 my @bind = $sql->values(\%fieldvals);
1142 This module was inspired by the excellent L<DBIx::Abstract>.
1143 However, in using that module I found that what I really wanted
1144 to do was generate SQL, but still retain complete control over my
1145 statement handles and use the DBI interface. So, I set out to
1146 create an abstract SQL generation module.
1148 While based on the concepts used by L<DBIx::Abstract>, there are
1149 several important differences, especially when it comes to WHERE
1150 clauses. I have modified the concepts used to make the SQL easier
1151 to generate from Perl data structures and, IMO, more intuitive.
1152 The underlying idea is for this module to do what you mean, based
1153 on the data structures you provide it. The big advantage is that
1154 you don't have to modify your code every time your data changes,
1155 as this module figures it out.
1157 To begin with, an SQL INSERT is as easy as just specifying a hash
1158 of C<key=value> pairs:
1161 name => 'Jimbo Bobson',
1162 phone => '123-456-7890',
1163 address => '42 Sister Lane',
1164 city => 'St. Louis',
1165 state => 'Louisiana',
1168 The SQL can then be generated with this:
1170 my($stmt, @bind) = $sql->insert('people', \%data);
1172 Which would give you something like this:
1174 $stmt = "INSERT INTO people
1175 (address, city, name, phone, state)
1176 VALUES (?, ?, ?, ?, ?)";
1177 @bind = ('42 Sister Lane', 'St. Louis', 'Jimbo Bobson',
1178 '123-456-7890', 'Louisiana');
1180 These are then used directly in your DBI code:
1182 my $sth = $dbh->prepare($stmt);
1183 $sth->execute(@bind);
1185 =head2 Inserting and Updating Arrays
1187 If your database has array types (like for example Postgres),
1188 activate the special option C<< array_datatypes => 1 >>
1189 when creating the C<SQL::Abstract> object.
1190 Then you may use an arrayref to insert and update database array types:
1192 my $sql = SQL::Abstract->new(array_datatypes => 1);
1194 planets => [qw/Mercury Venus Earth Mars/]
1197 my($stmt, @bind) = $sql->insert('solar_system', \%data);
1201 $stmt = "INSERT INTO solar_system (planets) VALUES (?)"
1203 @bind = (['Mercury', 'Venus', 'Earth', 'Mars']);
1206 =head2 Inserting and Updating SQL
1208 In order to apply SQL functions to elements of your C<%data> you may
1209 specify a reference to an arrayref for the given hash value. For example,
1210 if you need to execute the Oracle C<to_date> function on a value, you can
1211 say something like this:
1215 date_entered => \["to_date(?,'MM/DD/YYYY')", "03/02/2003"],
1218 The first value in the array is the actual SQL. Any other values are
1219 optional and would be included in the bind values array. This gives
1222 my($stmt, @bind) = $sql->insert('people', \%data);
1224 $stmt = "INSERT INTO people (name, date_entered)
1225 VALUES (?, to_date(?,'MM/DD/YYYY'))";
1226 @bind = ('Bill', '03/02/2003');
1228 An UPDATE is just as easy, all you change is the name of the function:
1230 my($stmt, @bind) = $sql->update('people', \%data);
1232 Notice that your C<%data> isn't touched; the module will generate
1233 the appropriately quirky SQL for you automatically. Usually you'll
1234 want to specify a WHERE clause for your UPDATE, though, which is
1235 where handling C<%where> hashes comes in handy...
1237 =head2 Complex where statements
1239 This module can generate pretty complicated WHERE statements
1240 easily. For example, simple C<key=value> pairs are taken to mean
1241 equality, and if you want to see if a field is within a set
1242 of values, you can use an arrayref. Let's say we wanted to
1243 SELECT some data based on this criteria:
1246 requestor => 'inna',
1247 worker => ['nwiger', 'rcwe', 'sfz'],
1248 status => { '!=', 'completed' }
1251 my($stmt, @bind) = $sql->select('tickets', '*', \%where);
1253 The above would give you something like this:
1255 $stmt = "SELECT * FROM tickets WHERE
1256 ( requestor = ? ) AND ( status != ? )
1257 AND ( worker = ? OR worker = ? OR worker = ? )";
1258 @bind = ('inna', 'completed', 'nwiger', 'rcwe', 'sfz');
1260 Which you could then use in DBI code like so:
1262 my $sth = $dbh->prepare($stmt);
1263 $sth->execute(@bind);
1269 The functions are simple. There's one for each major SQL operation,
1270 and a constructor you use first. The arguments are specified in a
1271 similar order to each function (table, then fields, then a where
1272 clause) to try and simplify things.
1277 =head2 new(option => 'value')
1279 The C<new()> function takes a list of options and values, and returns
1280 a new B<SQL::Abstract> object which can then be used to generate SQL
1281 through the methods below. The options accepted are:
1287 If set to 'lower', then SQL will be generated in all lowercase. By
1288 default SQL is generated in "textbook" case meaning something like:
1290 SELECT a_field FROM a_table WHERE some_field LIKE '%someval%'
1292 Any setting other than 'lower' is ignored.
1296 This determines what the default comparison operator is. By default
1297 it is C<=>, meaning that a hash like this:
1299 %where = (name => 'nwiger', email => 'nate@wiger.org');
1301 Will generate SQL like this:
1303 WHERE name = 'nwiger' AND email = 'nate@wiger.org'
1305 However, you may want loose comparisons by default, so if you set
1306 C<cmp> to C<like> you would get SQL such as:
1308 WHERE name like 'nwiger' AND email like 'nate@wiger.org'
1310 You can also override the comparsion on an individual basis - see
1311 the huge section on L</"WHERE CLAUSES"> at the bottom.
1313 =item sqltrue, sqlfalse
1315 Expressions for inserting boolean values within SQL statements.
1316 By default these are C<1=1> and C<1=0>.
1320 This determines the default logical operator for multiple WHERE
1321 statements in arrays. By default it is "or", meaning that a WHERE
1325 event_date => {'>=', '2/13/99'},
1326 event_date => {'<=', '4/24/03'},
1329 Will generate SQL like this:
1331 WHERE event_date >= '2/13/99' OR event_date <= '4/24/03'
1333 This is probably not what you want given this query, though (look
1334 at the dates). To change the "OR" to an "AND", simply specify:
1336 my $sql = SQL::Abstract->new(logic => 'and');
1338 Which will change the above C<WHERE> to:
1340 WHERE event_date >= '2/13/99' AND event_date <= '4/24/03'
1342 The logic can also be changed locally by inserting
1343 an extra first element in the array :
1345 @where = (-and => event_date => {'>=', '2/13/99'},
1346 event_date => {'<=', '4/24/03'} );
1348 See the L</"WHERE CLAUSES"> section for explanations.
1352 This will automatically convert comparisons using the specified SQL
1353 function for both column and value. This is mostly used with an argument
1354 of C<upper> or C<lower>, so that the SQL will have the effect of
1355 case-insensitive "searches". For example, this:
1357 $sql = SQL::Abstract->new(convert => 'upper');
1358 %where = (keywords => 'MaKe iT CAse inSeNSItive');
1360 Will turn out the following SQL:
1362 WHERE upper(keywords) like upper('MaKe iT CAse inSeNSItive')
1364 The conversion can be C<upper()>, C<lower()>, or any other SQL function
1365 that can be applied symmetrically to fields (actually B<SQL::Abstract> does
1366 not validate this option; it will just pass through what you specify verbatim).
1370 This is a kludge because many databases suck. For example, you can't
1371 just bind values using DBI's C<execute()> for Oracle C<CLOB> or C<BLOB> fields.
1372 Instead, you have to use C<bind_param()>:
1374 $sth->bind_param(1, 'reg data');
1375 $sth->bind_param(2, $lots, {ora_type => ORA_CLOB});
1377 The problem is, B<SQL::Abstract> will normally just return a C<@bind> array,
1378 which loses track of which field each slot refers to. Fear not.
1380 If you specify C<bindtype> in new, you can determine how C<@bind> is returned.
1381 Currently, you can specify either C<normal> (default) or C<columns>. If you
1382 specify C<columns>, you will get an array that looks like this:
1384 my $sql = SQL::Abstract->new(bindtype => 'columns');
1385 my($stmt, @bind) = $sql->insert(...);
1388 [ 'column1', 'value1' ],
1389 [ 'column2', 'value2' ],
1390 [ 'column3', 'value3' ],
1393 You can then iterate through this manually, using DBI's C<bind_param()>.
1395 $sth->prepare($stmt);
1398 my($col, $data) = @$_;
1399 if ($col eq 'details' || $col eq 'comments') {
1400 $sth->bind_param($i, $data, {ora_type => ORA_CLOB});
1401 } elsif ($col eq 'image') {
1402 $sth->bind_param($i, $data, {ora_type => ORA_BLOB});
1404 $sth->bind_param($i, $data);
1408 $sth->execute; # execute without @bind now
1410 Now, why would you still use B<SQL::Abstract> if you have to do this crap?
1411 Basically, the advantage is still that you don't have to care which fields
1412 are or are not included. You could wrap that above C<for> loop in a simple
1413 sub called C<bind_fields()> or something and reuse it repeatedly. You still
1414 get a layer of abstraction over manual SQL specification.
1416 Note that if you set L</bindtype> to C<columns>, the C<\[$sql, @bind]>
1417 construct (see L</Literal SQL with placeholders and bind values (subqueries)>)
1418 will expect the bind values in this format.
1422 This is the character that a table or column name will be quoted
1423 with. By default this is an empty string, but you could set it to
1424 the character C<`>, to generate SQL like this:
1426 SELECT `a_field` FROM `a_table` WHERE `some_field` LIKE '%someval%'
1428 Alternatively, you can supply an array ref of two items, the first being the left
1429 hand quote character, and the second the right hand quote character. For
1430 example, you could supply C<['[',']']> for SQL Server 2000 compliant quotes
1431 that generates SQL like this:
1433 SELECT [a_field] FROM [a_table] WHERE [some_field] LIKE '%someval%'
1435 Quoting is useful if you have tables or columns names that are reserved
1436 words in your database's SQL dialect.
1440 This is the character that separates a table and column name. It is
1441 necessary to specify this when the C<quote_char> option is selected,
1442 so that tables and column names can be individually quoted like this:
1444 SELECT `table`.`one_field` FROM `table` WHERE `table`.`other_field` = 1
1446 =item array_datatypes
1448 When this option is true, arrayrefs in INSERT or UPDATE are
1449 interpreted as array datatypes and are passed directly
1451 When this option is false, arrayrefs are interpreted
1452 as literal SQL, just like refs to arrayrefs
1453 (but this behavior is for backwards compatibility; when writing
1454 new queries, use the "reference to arrayref" syntax
1460 Takes a reference to a list of "special operators"
1461 to extend the syntax understood by L<SQL::Abstract>.
1462 See section L</"SPECIAL OPERATORS"> for details.
1468 =head2 insert($table, \@values || \%fieldvals)
1470 This is the simplest function. You simply give it a table name
1471 and either an arrayref of values or hashref of field/value pairs.
1472 It returns an SQL INSERT statement and a list of bind values.
1473 See the sections on L</"Inserting and Updating Arrays"> and
1474 L</"Inserting and Updating SQL"> for information on how to insert
1475 with those data types.
1477 =head2 update($table, \%fieldvals, \%where)
1479 This takes a table, hashref of field/value pairs, and an optional
1480 hashref L<WHERE clause|/WHERE CLAUSES>. It returns an SQL UPDATE function and a list
1482 See the sections on L</"Inserting and Updating Arrays"> and
1483 L</"Inserting and Updating SQL"> for information on how to insert
1484 with those data types.
1486 =head2 select($source, $fields, $where, $order)
1488 This returns a SQL SELECT statement and associated list of bind values, as
1489 specified by the arguments :
1495 Specification of the 'FROM' part of the statement.
1496 The argument can be either a plain scalar (interpreted as a table
1497 name, will be quoted), or an arrayref (interpreted as a list
1498 of table names, joined by commas, quoted), or a scalarref
1499 (literal table name, not quoted), or a ref to an arrayref
1500 (list of literal table names, joined by commas, not quoted).
1504 Specification of the list of fields to retrieve from
1506 The argument can be either an arrayref (interpreted as a list
1507 of field names, will be joined by commas and quoted), or a
1508 plain scalar (literal SQL, not quoted).
1509 Please observe that this API is not as flexible as for
1510 the first argument C<$table>, for backwards compatibility reasons.
1514 Optional argument to specify the WHERE part of the query.
1515 The argument is most often a hashref, but can also be
1516 an arrayref or plain scalar --
1517 see section L<WHERE clause|/"WHERE CLAUSES"> for details.
1521 Optional argument to specify the ORDER BY part of the query.
1522 The argument can be a scalar, a hashref or an arrayref
1523 -- see section L<ORDER BY clause|/"ORDER BY CLAUSES">
1529 =head2 delete($table, \%where)
1531 This takes a table name and optional hashref L<WHERE clause|/WHERE CLAUSES>.
1532 It returns an SQL DELETE statement and list of bind values.
1534 =head2 where(\%where, \@order)
1536 This is used to generate just the WHERE clause. For example,
1537 if you have an arbitrary data structure and know what the
1538 rest of your SQL is going to look like, but want an easy way
1539 to produce a WHERE clause, use this. It returns an SQL WHERE
1540 clause and list of bind values.
1543 =head2 values(\%data)
1545 This just returns the values from the hash C<%data>, in the same
1546 order that would be returned from any of the other above queries.
1547 Using this allows you to markedly speed up your queries if you
1548 are affecting lots of rows. See below under the L</"PERFORMANCE"> section.
1550 =head2 generate($any, 'number', $of, \@data, $struct, \%types)
1552 Warning: This is an experimental method and subject to change.
1554 This returns arbitrarily generated SQL. It's a really basic shortcut.
1555 It will return two different things, depending on return context:
1557 my($stmt, @bind) = $sql->generate('create table', \$table, \@fields);
1558 my $stmt_and_val = $sql->generate('create table', \$table, \@fields);
1560 These would return the following:
1562 # First calling form
1563 $stmt = "CREATE TABLE test (?, ?)";
1564 @bind = (field1, field2);
1566 # Second calling form
1567 $stmt_and_val = "CREATE TABLE test (field1, field2)";
1569 Depending on what you're trying to do, it's up to you to choose the correct
1570 format. In this example, the second form is what you would want.
1574 $sql->generate('alter session', { nls_date_format => 'MM/YY' });
1578 ALTER SESSION SET nls_date_format = 'MM/YY'
1580 You get the idea. Strings get their case twiddled, but everything
1581 else remains verbatim.
1586 =head1 WHERE CLAUSES
1590 This module uses a variation on the idea from L<DBIx::Abstract>. It
1591 is B<NOT>, repeat I<not> 100% compatible. B<The main logic of this
1592 module is that things in arrays are OR'ed, and things in hashes
1595 The easiest way to explain is to show lots of examples. After
1596 each C<%where> hash shown, it is assumed you used:
1598 my($stmt, @bind) = $sql->where(\%where);
1600 However, note that the C<%where> hash can be used directly in any
1601 of the other functions as well, as described above.
1603 =head2 Key-value pairs
1605 So, let's get started. To begin, a simple hash:
1609 status => 'completed'
1612 Is converted to SQL C<key = val> statements:
1614 $stmt = "WHERE user = ? AND status = ?";
1615 @bind = ('nwiger', 'completed');
1617 One common thing I end up doing is having a list of values that
1618 a field can be in. To do this, simply specify a list inside of
1623 status => ['assigned', 'in-progress', 'pending'];
1626 This simple code will create the following:
1628 $stmt = "WHERE user = ? AND ( status = ? OR status = ? OR status = ? )";
1629 @bind = ('nwiger', 'assigned', 'in-progress', 'pending');
1631 An empty arrayref will be considered a logical false and
1634 =head2 Key-value pairs
1636 If you want to specify a different type of operator for your comparison,
1637 you can use a hashref for a given column:
1641 status => { '!=', 'completed' }
1644 Which would generate:
1646 $stmt = "WHERE user = ? AND status != ?";
1647 @bind = ('nwiger', 'completed');
1649 To test against multiple values, just enclose the values in an arrayref:
1651 status => { '!=', ['assigned', 'in-progress', 'pending'] };
1653 Which would give you:
1655 "WHERE status != ? AND status != ? AND status != ?"
1657 Notice that since the operator was recognized as being a 'negative'
1658 operator, the arrayref was interpreted with 'AND' logic (because
1659 of Morgan's laws). By contrast, the reverse
1661 status => { '=', ['assigned', 'in-progress', 'pending'] };
1665 "WHERE status = ? OR status = ? OR status = ?"
1668 The hashref can also contain multiple pairs, in which case it is expanded
1669 into an C<AND> of its elements:
1673 status => { '!=', 'completed', -not_like => 'pending%' }
1676 # Or more dynamically, like from a form
1677 $where{user} = 'nwiger';
1678 $where{status}{'!='} = 'completed';
1679 $where{status}{'-not_like'} = 'pending%';
1681 # Both generate this
1682 $stmt = "WHERE user = ? AND status != ? AND status NOT LIKE ?";
1683 @bind = ('nwiger', 'completed', 'pending%');
1686 To get an OR instead, you can combine it with the arrayref idea:
1690 priority => [ {'=', 2}, {'!=', 1} ]
1693 Which would generate:
1695 $stmt = "WHERE user = ? AND priority = ? OR priority != ?";
1696 @bind = ('nwiger', '2', '1');
1698 If you want to include literal SQL (with or without bind values), just use a
1699 scalar reference or array reference as the value:
1702 date_entered => { '>' => \["to_date(?, 'MM/DD/YYYY')", "11/26/2008"] },
1703 date_expires => { '<' => \"now()" }
1706 Which would generate:
1708 $stmt = "WHERE date_entered > "to_date(?, 'MM/DD/YYYY') AND date_expires < now()";
1709 @bind = ('11/26/2008');
1712 =head2 Logic and nesting operators
1714 In the example above,
1715 there is a subtle trap if you want to say something like
1716 this (notice the C<AND>):
1718 WHERE priority != ? AND priority != ?
1720 Because, in Perl you I<can't> do this:
1722 priority => { '!=', 2, '!=', 1 }
1724 As the second C<!=> key will obliterate the first. The solution
1725 is to use the special C<-modifier> form inside an arrayref:
1727 priority => [ -and => {'!=', 2},
1731 Normally, these would be joined by C<OR>, but the modifier tells it
1732 to use C<AND> instead. (Hint: You can use this in conjunction with the
1733 C<logic> option to C<new()> in order to change the way your queries
1734 work by default.) B<Important:> Note that the C<-modifier> goes
1735 B<INSIDE> the arrayref, as an extra first element. This will
1736 B<NOT> do what you think it might:
1738 priority => -and => [{'!=', 2}, {'!=', 1}] # WRONG!
1740 Here is a quick list of equivalencies, since there is some overlap:
1743 status => {'!=', 'completed', 'not like', 'pending%' }
1744 status => [ -and => {'!=', 'completed'}, {'not like', 'pending%'}]
1747 status => {'=', ['assigned', 'in-progress']}
1748 status => [ -or => {'=', 'assigned'}, {'=', 'in-progress'}]
1749 status => [ {'=', 'assigned'}, {'=', 'in-progress'} ]
1751 In addition to C<-and> and C<-or>, there is also a special C<-nest>
1752 operator which adds an additional set of parens, to create a subquery.
1753 For example, to get something like this:
1755 $stmt = "WHERE user = ? AND ( workhrs > ? OR geo = ? )";
1756 @bind = ('nwiger', '20', 'ASIA');
1762 -nest => [ workhrs => {'>', 20}, geo => 'ASIA' ],
1765 If you need several nested subexpressions, you can number
1766 the C<-nest> branches :
1776 =head2 Special operators : IN, BETWEEN, etc.
1778 You can also use the hashref format to compare a list of fields using the
1779 C<IN> comparison operator, by specifying the list as an arrayref:
1782 status => 'completed',
1783 reportid => { -in => [567, 2335, 2] }
1786 Which would generate:
1788 $stmt = "WHERE status = ? AND reportid IN (?,?,?)";
1789 @bind = ('completed', '567', '2335', '2');
1791 The reverse operator C<-not_in> generates SQL C<NOT IN> and is used in
1794 Another pair of operators is C<-between> and C<-not_between>,
1795 used with an arrayref of two values:
1799 completion_date => {
1800 -not_between => ['2002-10-01', '2003-02-06']
1806 WHERE user = ? AND completion_date NOT BETWEEN ( ? AND ? )
1808 These are the two builtin "special operators"; but the
1809 list can be expanded : see section L</"SPECIAL OPERATORS"> below.
1811 =head2 Nested conditions
1813 So far, we've seen how multiple conditions are joined with a top-level
1814 C<AND>. We can change this by putting the different conditions we want in
1815 hashes and then putting those hashes in an array. For example:
1820 status => { -like => ['pending%', 'dispatched'] },
1824 status => 'unassigned',
1828 This data structure would create the following:
1830 $stmt = "WHERE ( user = ? AND ( status LIKE ? OR status LIKE ? ) )
1831 OR ( user = ? AND status = ? ) )";
1832 @bind = ('nwiger', 'pending', 'dispatched', 'robot', 'unassigned');
1834 This can be combined with the C<-nest> operator to properly group
1841 ["-and", workhrs => {'>', 20}, geo => 'ASIA' ],
1842 ["-and", workhrs => {'<', 50}, geo => 'EURO' ]
1849 WHERE ( user = ? AND
1850 ( ( workhrs > ? AND geo = ? )
1851 OR ( workhrs < ? AND geo = ? ) ) )
1855 Finally, sometimes only literal SQL will do. If you want to include
1856 literal SQL verbatim, you can specify it as a scalar reference, namely:
1858 my $inn = 'is Not Null';
1860 priority => { '<', 2 },
1866 $stmt = "WHERE priority < ? AND requestor is Not Null";
1869 Note that in this example, you only get one bind parameter back, since
1870 the verbatim SQL is passed as part of the statement.
1872 Of course, just to prove a point, the above can also be accomplished
1876 priority => { '<', 2 },
1877 requestor => { '!=', undef },
1883 Conditions on boolean columns can be expressed in the
1884 same way, passing a reference to an empty string :
1887 priority => { '<', 2 },
1893 $stmt = "WHERE priority < ? AND is_ready";
1897 =head2 Literal SQL with placeholders and bind values (subqueries)
1899 If the literal SQL to be inserted has placeholders and bind values,
1900 use a reference to an arrayref (yes this is a double reference --
1901 not so common, but perfectly legal Perl). For example, to find a date
1902 in Postgres you can use something like this:
1905 date_column => \[q/= date '2008-09-30' - ?::integer/, 10/]
1910 $stmt = "WHERE ( date_column = date '2008-09-30' - ?::integer )"
1913 Note that you must pass the bind values in the same format as they are returned
1914 by L</where>. That means that if you set L</bindtype> to C<columns>, you must
1915 provide the bind values in the C<< [ column_meta => value ] >> format, where
1916 C<column_meta> is an opaque scalar value; most commonly the column name, but
1917 you can use any scalar value (including references and blessed references),
1918 L<SQL::Abstract> will simply pass it through intact. So if C<bindtype> is set
1919 to C<columns> the above example will look like:
1922 date_column => \[q/= date '2008-09-30' - ?::integer/, [ dummy => 10 ]/]
1925 Literal SQL is especially useful for nesting parenthesized clauses in the
1926 main SQL query. Here is a first example :
1928 my ($sub_stmt, @sub_bind) = ("SELECT c1 FROM t1 WHERE c2 < ? AND c3 LIKE ?",
1932 bar => \["IN ($sub_stmt)" => @sub_bind],
1937 $stmt = "WHERE (foo = ? AND bar IN (SELECT c1 FROM t1
1938 WHERE c2 < ? AND c3 LIKE ?))";
1939 @bind = (1234, 100, "foo%");
1941 Other subquery operators, like for example C<"E<gt> ALL"> or C<"NOT IN">,
1942 are expressed in the same way. Of course the C<$sub_stmt> and
1943 its associated bind values can be generated through a former call
1946 my ($sub_stmt, @sub_bind)
1947 = $sql->select("t1", "c1", {c2 => {"<" => 100},
1948 c3 => {-like => "foo%"}});
1951 bar => \["> ALL ($sub_stmt)" => @sub_bind],
1954 In the examples above, the subquery was used as an operator on a column;
1955 but the same principle also applies for a clause within the main C<%where>
1956 hash, like an EXISTS subquery :
1958 my ($sub_stmt, @sub_bind)
1959 = $sql->select("t1", "*", {c1 => 1, c2 => \"> t0.c0"});
1962 -nest => \["EXISTS ($sub_stmt)" => @sub_bind],
1967 $stmt = "WHERE (foo = ? AND EXISTS (SELECT * FROM t1
1968 WHERE c1 = ? AND c2 > t0.c0))";
1972 Observe that the condition on C<c2> in the subquery refers to
1973 column C<t0.c0> of the main query : this is I<not> a bind
1974 value, so we have to express it through a scalar ref.
1975 Writing C<< c2 => {">" => "t0.c0"} >> would have generated
1976 C<< c2 > ? >> with bind value C<"t0.c0"> ... not exactly
1977 what we wanted here.
1979 Another use of the subquery technique is when some SQL clauses need
1980 parentheses, as it often occurs with some proprietary SQL extensions
1981 like for example fulltext expressions, geospatial expressions,
1982 NATIVE clauses, etc. Here is an example of a fulltext query in MySQL :
1985 -nest => \["MATCH (col1, col2) AGAINST (?)" => qw/apples/]
1988 Finally, here is an example where a subquery is used
1989 for expressing unary negation:
1991 my ($sub_stmt, @sub_bind)
1992 = $sql->where({age => [{"<" => 10}, {">" => 20}]});
1993 $sub_stmt =~ s/^ where //i; # don't want "WHERE" in the subclause
1995 lname => {like => '%son%'},
1996 -nest => \["NOT ($sub_stmt)" => @sub_bind],
2001 $stmt = "lname LIKE ? AND NOT ( age < ? OR age > ? )"
2002 @bind = ('%son%', 10, 20)
2008 These pages could go on for a while, since the nesting of the data
2009 structures this module can handle are pretty much unlimited (the
2010 module implements the C<WHERE> expansion as a recursive function
2011 internally). Your best bet is to "play around" with the module a
2012 little to see how the data structures behave, and choose the best
2013 format for your data based on that.
2015 And of course, all the values above will probably be replaced with
2016 variables gotten from forms or the command line. After all, if you
2017 knew everything ahead of time, you wouldn't have to worry about
2018 dynamically-generating SQL and could just hardwire it into your
2024 =head1 ORDER BY CLAUSES
2026 Some functions take an order by clause. This can either be a scalar (just a
2027 column name,) a hash of C<< { -desc => 'col' } >> or C<< { -asc => 'col' } >>,
2028 or an array of either of the two previous forms. Examples:
2030 Given | Will Generate
2031 ----------------------------------------------------------
2032 \'colA DESC' | ORDER BY colA DESC
2033 'colA' | ORDER BY colA
2034 [qw/colA colB/] | ORDER BY colA, colB
2035 {-asc => 'colA'} | ORDER BY colA ASC
2036 {-desc => 'colB'} | ORDER BY colB DESC
2038 {-asc => 'colA'}, | ORDER BY colA ASC, colB DESC
2041 [colA => {-asc => 'colB'}] | ORDER BY colA, colB ASC
2042 ==========================================================
2046 =head1 SPECIAL OPERATORS
2048 my $sqlmaker = SQL::Abstract->new(special_ops => [
2051 my ($self, $field, $op, $arg) = @_;
2057 A "special operator" is a SQL syntactic clause that can be
2058 applied to a field, instead of a usual binary operator.
2061 WHERE field IN (?, ?, ?)
2062 WHERE field BETWEEN ? AND ?
2063 WHERE MATCH(field) AGAINST (?, ?)
2065 Special operators IN and BETWEEN are fairly standard and therefore
2066 are builtin within C<SQL::Abstract>. For other operators,
2067 like the MATCH .. AGAINST example above which is
2068 specific to MySQL, you can write your own operator handlers :
2069 supply a C<special_ops> argument to the C<new> method.
2070 That argument takes an arrayref of operator definitions;
2071 each operator definition is a hashref with two entries
2077 the regular expression to match the operator
2081 coderef that will be called when meeting that operator
2082 in the input tree. The coderef will be called with
2083 arguments C<< ($self, $field, $op, $arg) >>, and
2084 should return a C<< ($sql, @bind) >> structure.
2088 For example, here is an implementation
2089 of the MATCH .. AGAINST syntax for MySQL
2091 my $sqlmaker = SQL::Abstract->new(special_ops => [
2093 # special op for MySql MATCH (field) AGAINST(word1, word2, ...)
2094 {regex => qr/^match$/i,
2096 my ($self, $field, $op, $arg) = @_;
2097 $arg = [$arg] if not ref $arg;
2098 my $label = $self->_quote($field);
2099 my ($placeholder) = $self->_convert('?');
2100 my $placeholders = join ", ", (($placeholder) x @$arg);
2101 my $sql = $self->_sqlcase('match') . " ($label) "
2102 . $self->_sqlcase('against') . " ($placeholders) ";
2103 my @bind = $self->_bindtype($field, @$arg);
2104 return ($sql, @bind);
2113 Thanks to some benchmarking by Mark Stosberg, it turns out that
2114 this module is many orders of magnitude faster than using C<DBIx::Abstract>.
2115 I must admit this wasn't an intentional design issue, but it's a
2116 byproduct of the fact that you get to control your C<DBI> handles
2119 To maximize performance, use a code snippet like the following:
2121 # prepare a statement handle using the first row
2122 # and then reuse it for the rest of the rows
2124 for my $href (@array_of_hashrefs) {
2125 $stmt ||= $sql->insert('table', $href);
2126 $sth ||= $dbh->prepare($stmt);
2127 $sth->execute($sql->values($href));
2130 The reason this works is because the keys in your C<$href> are sorted
2131 internally by B<SQL::Abstract>. Thus, as long as your data retains
2132 the same structure, you only have to generate the SQL the first time
2133 around. On subsequent queries, simply use the C<values> function provided
2134 by this module to return your values in the correct order.
2139 If you use my C<CGI::FormBuilder> module at all, you'll hopefully
2140 really like this part (I do, at least). Building up a complex query
2141 can be as simple as the following:
2145 use CGI::FormBuilder;
2148 my $form = CGI::FormBuilder->new(...);
2149 my $sql = SQL::Abstract->new;
2151 if ($form->submitted) {
2152 my $field = $form->field;
2153 my $id = delete $field->{id};
2154 my($stmt, @bind) = $sql->update('table', $field, {id => $id});
2157 Of course, you would still have to connect using C<DBI> to run the
2158 query, but the point is that if you make your form look like your
2159 table, the actual query script can be extremely simplistic.
2161 If you're B<REALLY> lazy (I am), check out C<HTML::QuickTable> for
2162 a fast interface to returning and formatting data. I frequently
2163 use these three modules together to write complex database query
2164 apps in under 50 lines.
2169 Version 1.50 was a major internal refactoring of C<SQL::Abstract>.
2170 Great care has been taken to preserve the I<published> behavior
2171 documented in previous versions in the 1.* family; however,
2172 some features that were previously undocumented, or behaved
2173 differently from the documentation, had to be changed in order
2174 to clarify the semantics. Hence, client code that was relying
2175 on some dark areas of C<SQL::Abstract> v1.*
2176 B<might behave differently> in v1.50.
2178 The main changes are :
2184 support for literal SQL through the C<< \ [$sql, bind] >> syntax.
2188 support for the { operator => \"..." } construct (to embed literal SQL)
2192 support for the { operator => \["...", @bind] } construct (to embed literal SQL with bind values)
2196 added -nest1, -nest2 or -nest_1, -nest_2, ...
2200 optional support for L<array datatypes|/"Inserting and Updating Arrays">
2204 defensive programming : check arguments
2208 fixed bug with global logic, which was previously implemented
2209 through global variables yielding side-effects. Prior versons would
2210 interpret C<< [ {cond1, cond2}, [cond3, cond4] ] >>
2211 as C<< "(cond1 AND cond2) OR (cond3 AND cond4)" >>.
2212 Now this is interpreted
2213 as C<< "(cond1 AND cond2) OR (cond3 OR cond4)" >>.
2217 C<-and> / C<-or> operators are no longer accepted
2218 in the middle of an arrayref : they are
2219 only admitted if in first position.
2223 changed logic for distributing an op over arrayrefs
2227 fixed semantics of _bindtype on array args
2231 dropped the C<_anoncopy> of the %where tree. No longer necessary,
2232 we just avoid shifting arrays within that tree.
2236 dropped the C<_modlogic> function
2242 =head1 ACKNOWLEDGEMENTS
2244 There are a number of individuals that have really helped out with
2245 this module. Unfortunately, most of them submitted bugs via CPAN
2246 so I have no idea who they are! But the people I do know are:
2248 Ash Berlin (order_by hash term support)
2249 Matt Trout (DBIx::Class support)
2250 Mark Stosberg (benchmarking)
2251 Chas Owens (initial "IN" operator support)
2252 Philip Collins (per-field SQL functions)
2253 Eric Kolve (hashref "AND" support)
2254 Mike Fragassi (enhancements to "BETWEEN" and "LIKE")
2255 Dan Kubb (support for "quote_char" and "name_sep")
2256 Guillermo Roditi (patch to cleanup "IN" and "BETWEEN", fix and tests for _order_by)
2257 Laurent Dami (internal refactoring, multiple -nest, extensible list of special operators, literal SQL)
2258 Norbert Buchmuller (support for literal SQL in hashpair, misc. fixes & tests)
2264 L<DBIx::Class>, L<DBIx::Abstract>, L<CGI::FormBuilder>, L<HTML::QuickTable>.
2268 Copyright (c) 2001-2007 Nathan Wiger <nwiger@cpan.org>. All Rights Reserved.
2270 This module is actively maintained by Matt Trout <mst@shadowcatsystems.co.uk>
2272 For support, your best bet is to try the C<DBIx::Class> users mailing list.
2273 While not an official support venue, C<DBIx::Class> makes heavy use of
2274 C<SQL::Abstract>, and as such list members there are very familiar with
2275 how to create queries.
2277 This module is free software; you may copy this under the terms of
2278 the GNU General Public License, or the Artistic License, copies of
2279 which should have accompanied your Perl kit.