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 = '1.49_04';
19 $VERSION = eval $VERSION; # numify for warning-free dev releases
24 # special operators (-in, -between). May be extended/overridden by user.
25 # See section WHERE: BUILTIN SPECIAL OPERATORS below for implementation
26 my @BUILTIN_SPECIAL_OPS = (
27 {regex => qr/^(not )?between$/i, handler => \&_where_field_BETWEEN},
28 {regex => qr/^(not )?in$/i, handler => \&_where_field_IN},
31 #======================================================================
32 # DEBUGGING AND ERROR REPORTING
33 #======================================================================
36 return unless $_[0]->{debug}; shift; # a little faster
37 my $func = (caller(1))[3];
38 warn "[$func] ", @_, "\n";
42 my($func) = (caller(1))[3];
43 carp "[$func] Warning: ", @_;
47 my($func) = (caller(1))[3];
48 croak "[$func] Fatal: ", @_;
52 #======================================================================
54 #======================================================================
58 my $class = ref($self) || $self;
59 my %opt = (ref $_[0] eq 'HASH') ? %{$_[0]} : @_;
61 # choose our case by keeping an option around
62 delete $opt{case} if $opt{case} && $opt{case} ne 'lower';
64 # default logic for interpreting arrayrefs
65 $opt{logic} = uc $opt{logic} || 'OR';
67 # how to return bind vars
68 # LDNOTE: changed nwiger code : why this 'delete' ??
69 # $opt{bindtype} ||= delete($opt{bind_type}) || 'normal';
70 $opt{bindtype} ||= 'normal';
72 # default comparison is "=", but can be overridden
75 # try to recognize which are the 'equality' and 'unequality' ops
76 # (temporary quickfix, should go through a more seasoned API)
77 $opt{equality_op} = qr/^(\Q$opt{cmp}\E|is|(is\s+)?like)$/i;
78 $opt{inequality_op} = qr/^(!=|<>|(is\s+)?not(\s+like)?)$/i;
81 $opt{sqltrue} ||= '1=1';
82 $opt{sqlfalse} ||= '0=1';
85 $opt{special_ops} ||= [];
86 push @{$opt{special_ops}}, @BUILTIN_SPECIAL_OPS;
88 return bless \%opt, $class;
93 #======================================================================
95 #======================================================================
99 my $table = $self->_table(shift);
100 my $data = shift || return;
102 my $method = $self->_METHOD_FOR_refkind("_insert", $data);
103 my ($sql, @bind) = $self->$method($data);
104 $sql = join " ", $self->_sqlcase('insert into'), $table, $sql;
105 return wantarray ? ($sql, @bind) : $sql;
108 sub _insert_HASHREF { # explicit list of fields and then values
109 my ($self, $data) = @_;
111 my @fields = sort keys %$data;
113 my ($sql, @bind) = $self->_insert_values($data);
116 $_ = $self->_quote($_) foreach @fields;
117 $sql = "( ".join(", ", @fields).") ".$sql;
119 return ($sql, @bind);
122 sub _insert_ARRAYREF { # just generate values(?,?) part (no list of fields)
123 my ($self, $data) = @_;
125 # no names (arrayref) so can't generate bindtype
126 $self->{bindtype} ne 'columns'
127 or belch "can't do 'columns' bindtype when called with arrayref";
129 # fold the list of values into a hash of column name - value pairs
130 # (where the column names are artificially generated, and their
131 # lexicographical ordering keep the ordering of the original list)
132 my $i = "a"; # incremented values will be in lexicographical order
133 my $data_in_hash = { map { ($i++ => $_) } @$data };
135 return $self->_insert_values($data_in_hash);
138 sub _insert_ARRAYREFREF { # literal SQL with bind
139 my ($self, $data) = @_;
141 my ($sql, @bind) = @${$data};
142 $self->_assert_bindval_matches_bindtype(@bind);
144 return ($sql, @bind);
148 sub _insert_SCALARREF { # literal SQL without bind
149 my ($self, $data) = @_;
155 my ($self, $data) = @_;
157 my (@values, @all_bind);
158 foreach my $column (sort keys %$data) {
159 my $v = $data->{$column};
161 $self->_SWITCH_refkind($v, {
164 if ($self->{array_datatypes}) { # if array datatype are activated
166 push @all_bind, $self->_bindtype($column, $v);
168 else { # else literal SQL with bind
169 my ($sql, @bind) = @$v;
170 $self->_assert_bindval_matches_bindtype(@bind);
172 push @all_bind, @bind;
176 ARRAYREFREF => sub { # literal SQL with bind
177 my ($sql, @bind) = @${$v};
178 $self->_assert_bindval_matches_bindtype(@bind);
180 push @all_bind, @bind;
183 # THINK : anything useful to do with a HASHREF ?
184 HASHREF => sub { # (nothing, but old SQLA passed it through)
185 #TODO in SQLA >= 2.0 it will die instead
186 belch "HASH ref as bind value in insert is not supported";
188 push @all_bind, $self->_bindtype($column, $v);
191 SCALARREF => sub { # literal SQL without bind
195 SCALAR_or_UNDEF => sub {
197 push @all_bind, $self->_bindtype($column, $v);
204 my $sql = $self->_sqlcase('values')." ( ".join(", ", @values)." )";
205 return ($sql, @all_bind);
210 #======================================================================
212 #======================================================================
217 my $table = $self->_table(shift);
218 my $data = shift || return;
221 # first build the 'SET' part of the sql statement
222 my (@set, @all_bind);
223 puke "Unsupported data type specified to \$sql->update"
224 unless ref $data eq 'HASH';
226 for my $k (sort keys %$data) {
229 my $label = $self->_quote($k);
231 $self->_SWITCH_refkind($v, {
233 if ($self->{array_datatypes}) { # array datatype
234 push @set, "$label = ?";
235 push @all_bind, $self->_bindtype($k, $v);
237 else { # literal SQL with bind
238 my ($sql, @bind) = @$v;
239 $self->_assert_bindval_matches_bindtype(@bind);
240 push @set, "$label = $sql";
241 push @all_bind, @bind;
244 ARRAYREFREF => sub { # literal SQL with bind
245 my ($sql, @bind) = @${$v};
246 $self->_assert_bindval_matches_bindtype(@bind);
247 push @set, "$label = $sql";
248 push @all_bind, @bind;
250 SCALARREF => sub { # literal SQL without bind
251 push @set, "$label = $$v";
253 SCALAR_or_UNDEF => sub {
254 push @set, "$label = ?";
255 push @all_bind, $self->_bindtype($k, $v);
261 my $sql = $self->_sqlcase('update') . " $table " . $self->_sqlcase('set ')
265 my($where_sql, @where_bind) = $self->where($where);
267 push @all_bind, @where_bind;
270 return wantarray ? ($sql, @all_bind) : $sql;
276 #======================================================================
278 #======================================================================
283 my $table = $self->_table(shift);
284 my $fields = shift || '*';
288 my($where_sql, @bind) = $self->where($where, $order);
290 my $f = (ref $fields eq 'ARRAY') ? join ', ', map { $self->_quote($_) } @$fields
292 my $sql = join(' ', $self->_sqlcase('select'), $f,
293 $self->_sqlcase('from'), $table)
296 return wantarray ? ($sql, @bind) : $sql;
299 #======================================================================
301 #======================================================================
306 my $table = $self->_table(shift);
310 my($where_sql, @bind) = $self->where($where);
311 my $sql = $self->_sqlcase('delete from') . " $table" . $where_sql;
313 return wantarray ? ($sql, @bind) : $sql;
317 #======================================================================
319 #======================================================================
323 # Finally, a separate routine just to handle WHERE clauses
325 my ($self, $where, $order) = @_;
328 my ($sql, @bind) = $self->_recurse_where($where);
329 $sql = $sql ? $self->_sqlcase(' where ') . "( $sql )" : '';
333 $sql .= $self->_order_by($order);
336 return wantarray ? ($sql, @bind) : $sql;
341 my ($self, $where, $logic) = @_;
343 # dispatch on appropriate method according to refkind of $where
344 my $method = $self->_METHOD_FOR_refkind("_where", $where);
347 my ($sql, @bind) = $self->$method($where, $logic);
349 # DBIx::Class directly calls _recurse_where in scalar context, so
350 # we must implement it, even if not in the official API
351 return wantarray ? ($sql, @bind) : $sql;
356 #======================================================================
357 # WHERE: top-level ARRAYREF
358 #======================================================================
361 sub _where_ARRAYREF {
362 my ($self, $where, $logic) = @_;
364 $logic = uc($logic || $self->{logic});
365 $logic eq 'AND' or $logic eq 'OR' or puke "unknown logic: $logic";
367 my @clauses = @$where;
369 # if the array starts with [-and|or => ...], recurse with that logic
370 my $first = $clauses[0] || '';
371 if ($first =~ /^-(and|or)/i) {
374 return $self->_where_ARRAYREF(\@clauses, $logic);
378 my (@sql_clauses, @all_bind);
380 # need to use while() so can shift() for pairs
381 while (my $el = shift @clauses) {
383 # switch according to kind of $el and get corresponding ($sql, @bind)
384 my ($sql, @bind) = $self->_SWITCH_refkind($el, {
386 # skip empty elements, otherwise get invalid trailing AND stuff
387 ARRAYREF => sub {$self->_recurse_where($el) if @$el},
389 ARRAYREFREF => sub { @{${$el}} if @{${$el}}},
391 HASHREF => sub {$self->_recurse_where($el, 'and') if %$el},
392 # LDNOTE : previous SQLA code for hashrefs was creating a dirty
393 # side-effect: the first hashref within an array would change
394 # the global logic to 'AND'. So [ {cond1, cond2}, [cond3, cond4] ]
395 # was interpreted as "(cond1 AND cond2) OR (cond3 AND cond4)",
396 # whereas it should be "(cond1 AND cond2) OR (cond3 OR cond4)".
398 SCALARREF => sub { ($$el); },
400 SCALAR => sub {# top-level arrayref with scalars, recurse in pairs
401 $self->_recurse_where({$el => shift(@clauses)})},
403 UNDEF => sub {puke "not supported : UNDEF in arrayref" },
407 push @sql_clauses, $sql;
408 push @all_bind, @bind;
412 return $self->_join_sql_clauses($logic, \@sql_clauses, \@all_bind);
415 #======================================================================
416 # WHERE: top-level ARRAYREFREF
417 #======================================================================
419 sub _where_ARRAYREFREF {
420 my ($self, $where) = @_;
421 my ($sql, @bind) = @{${$where}};
423 return ($sql, @bind);
426 #======================================================================
427 # WHERE: top-level HASHREF
428 #======================================================================
431 my ($self, $where) = @_;
432 my (@sql_clauses, @all_bind);
434 # LDNOTE : don't really know why we need to sort keys
435 for my $k (sort keys %$where) {
436 my $v = $where->{$k};
438 # ($k => $v) is either a special op or a regular hashpair
439 my ($sql, @bind) = ($k =~ /^-(.+)/) ? $self->_where_op_in_hash($1, $v)
441 my $method = $self->_METHOD_FOR_refkind("_where_hashpair", $v);
442 $self->$method($k, $v);
445 push @sql_clauses, $sql;
446 push @all_bind, @bind;
449 return $self->_join_sql_clauses('and', \@sql_clauses, \@all_bind);
453 sub _where_op_in_hash {
454 my ($self, $op, $v) = @_;
456 $op =~ /^(AND|OR|NEST)[_\d]*/i
457 or puke "unknown operator: -$op";
458 $op = uc($1); # uppercase, remove trailing digits
459 $self->_debug("OP(-$op) within hashref, recursing...");
461 $self->_SWITCH_refkind($v, {
464 # LDNOTE : should deprecate {-or => [...]} and {-and => [...]}
465 # because they are misleading; the only proper way would be
466 # -nest => [-or => ...], -nest => [-and ...]
467 return $self->_where_ARRAYREF($v, $op eq 'NEST' ? '' : $op);
472 belch "-or => {...} should be -nest => [...]";
473 return $self->_where_ARRAYREF([%$v], 'OR');
476 return $self->_where_HASHREF($v);
480 SCALARREF => sub { # literal SQL
482 or puke "-$op => \\\$scalar not supported, use -nest => ...";
486 ARRAYREFREF => sub { # literal SQL
488 or puke "-$op => \\[..] not supported, use -nest => ...";
492 SCALAR => sub { # permissively interpreted as SQL
494 or puke "-$op => 'scalar' not supported, use -nest => \\'scalar'";
495 belch "literal SQL should be -nest => \\'scalar' "
496 . "instead of -nest => 'scalar' ";
501 puke "-$op => undef not supported";
507 sub _where_hashpair_ARRAYREF {
508 my ($self, $k, $v) = @_;
511 my @v = @$v; # need copy because of shift below
512 $self->_debug("ARRAY($k) means distribute over elements");
514 # put apart first element if it is an operator (-and, -or)
515 my $op = $v[0] =~ /^-/ ? shift @v : undef;
516 $self->_debug("OP($op) reinjected into the distributed array") if $op;
518 my @distributed = map { {$k => $_} } @v;
519 unshift @distributed, $op if $op;
521 return $self->_recurse_where(\@distributed);
524 # LDNOTE : not sure of this one. What does "distribute over nothing" mean?
525 $self->_debug("empty ARRAY($k) means 0=1");
526 return ($self->{sqlfalse});
530 sub _where_hashpair_HASHREF {
531 my ($self, $k, $v) = @_;
533 my (@all_sql, @all_bind);
535 for my $op (sort keys %$v) {
538 # put the operator in canonical form
539 $op =~ s/^-//; # remove initial dash
540 $op =~ tr/_/ /; # underscores become spaces
541 $op =~ s/^\s+//; # no initial space
542 $op =~ s/\s+$//; # no final space
543 $op =~ s/\s+/ /; # multiple spaces become one
547 # CASE: special operators like -in or -between
548 my $special_op = first {$op =~ $_->{regex}} @{$self->{special_ops}};
550 ($sql, @bind) = $special_op->{handler}->($self, $k, $op, $val);
553 $self->_SWITCH_refkind($val, {
555 ARRAYREF => sub { # CASE: col => {op => \@vals}
556 ($sql, @bind) = $self->_where_field_op_ARRAYREF($k, $op, $val);
559 SCALARREF => sub { # CASE: col => {op => \$scalar} (literal SQL without bind)
560 $sql = join ' ', $self->_convert($self->_quote($k)),
561 $self->_sqlcase($op),
565 ARRAYREFREF => sub { # CASE: col => {op => \[$sql, @bind]} (literal SQL with bind)
566 my ($sub_sql, @sub_bind) = @$$val;
567 $self->_assert_bindval_matches_bindtype(@sub_bind);
568 $sql = join ' ', $self->_convert($self->_quote($k)),
569 $self->_sqlcase($op),
574 UNDEF => sub { # CASE: col => {op => undef} : sql "IS (NOT)? NULL"
575 my $is = ($op =~ $self->{equality_op}) ? 'is' :
576 ($op =~ $self->{inequality_op}) ? 'is not' :
577 puke "unexpected operator '$op' with undef operand";
578 $sql = $self->_quote($k) . $self->_sqlcase(" $is null");
581 FALLBACK => sub { # CASE: col => {op => $scalar}
582 $sql = join ' ', $self->_convert($self->_quote($k)),
583 $self->_sqlcase($op),
584 $self->_convert('?');
585 @bind = $self->_bindtype($k, $val);
591 push @all_bind, @bind;
594 return $self->_join_sql_clauses('and', \@all_sql, \@all_bind);
599 sub _where_field_op_ARRAYREF {
600 my ($self, $k, $op, $vals) = @_;
603 $self->_debug("ARRAY($vals) means multiple elements: [ @$vals ]");
607 # LDNOTE : change the distribution logic when
608 # $op =~ $self->{inequality_op}, because of Morgan laws :
609 # with {field => {'!=' => [22, 33]}}, it would be ridiculous to generate
610 # WHERE field != 22 OR field != 33 : the user probably means
611 # WHERE field != 22 AND field != 33.
612 my $logic = ($op =~ $self->{inequality_op}) ? 'AND' : 'OR';
614 # distribute $op over each member of @$vals
615 return $self->_recurse_where([map { {$k => {$op, $_}} } @$vals], $logic);
619 # try to DWIM on equality operators
620 # LDNOTE : not 100% sure this is the correct thing to do ...
621 return ($self->{sqlfalse}) if $op =~ $self->{equality_op};
622 return ($self->{sqltrue}) if $op =~ $self->{inequality_op};
625 puke "operator '$op' applied on an empty array (field '$k')";
630 sub _where_hashpair_SCALARREF {
631 my ($self, $k, $v) = @_;
632 $self->_debug("SCALAR($k) means literal SQL: $$v");
633 my $sql = $self->_quote($k) . " " . $$v;
637 # literal SQL with bind
638 sub _where_hashpair_ARRAYREFREF {
639 my ($self, $k, $v) = @_;
640 $self->_debug("REF($k) means literal SQL: @${$v}");
641 my ($sql, @bind) = @${$v};
642 $self->_assert_bindval_matches_bindtype(@bind);
643 $sql = $self->_quote($k) . " " . $sql;
644 return ($sql, @bind );
647 # literal SQL without bind
648 sub _where_hashpair_SCALAR {
649 my ($self, $k, $v) = @_;
650 $self->_debug("NOREF($k) means simple key=val: $k $self->{cmp} $v");
651 my $sql = join ' ', $self->_convert($self->_quote($k)),
652 $self->_sqlcase($self->{cmp}),
653 $self->_convert('?');
654 my @bind = $self->_bindtype($k, $v);
655 return ( $sql, @bind);
659 sub _where_hashpair_UNDEF {
660 my ($self, $k, $v) = @_;
661 $self->_debug("UNDEF($k) means IS NULL");
662 my $sql = $self->_quote($k) . $self->_sqlcase(' is null');
666 #======================================================================
667 # WHERE: TOP-LEVEL OTHERS (SCALARREF, SCALAR, UNDEF)
668 #======================================================================
671 sub _where_SCALARREF {
672 my ($self, $where) = @_;
675 $self->_debug("SCALAR(*top) means literal SQL: $$where");
681 my ($self, $where) = @_;
684 $self->_debug("NOREF(*top) means literal SQL: $where");
695 #======================================================================
696 # WHERE: BUILTIN SPECIAL OPERATORS (-in, -between)
697 #======================================================================
700 sub _where_field_BETWEEN {
701 my ($self, $k, $op, $vals) = @_;
703 ref $vals eq 'ARRAY' && @$vals == 2
704 or puke "special op 'between' requires an arrayref of two values";
706 my ($label) = $self->_convert($self->_quote($k));
707 my ($placeholder) = $self->_convert('?');
708 my $and = $self->_sqlcase('and');
709 $op = $self->_sqlcase($op);
711 my $sql = "( $label $op $placeholder $and $placeholder )";
712 my @bind = $self->_bindtype($k, @$vals);
717 sub _where_field_IN {
718 my ($self, $k, $op, $vals) = @_;
720 # backwards compatibility : if scalar, force into an arrayref
721 $vals = [$vals] if defined $vals && ! ref $vals;
723 my ($label) = $self->_convert($self->_quote($k));
724 my ($placeholder) = $self->_convert('?');
725 $op = $self->_sqlcase($op);
727 my ($sql, @bind) = $self->_SWITCH_refkind($vals, {
728 ARRAYREF => sub { # list of choices
729 if (@$vals) { # nonempty list
730 my $placeholders = join ", ", (($placeholder) x @$vals);
731 my $sql = "$label $op ( $placeholders )";
732 my @bind = $self->_bindtype($k, @$vals);
734 return ($sql, @bind);
736 else { # empty list : some databases won't understand "IN ()", so DWIM
737 my $sql = ($op =~ /\bnot\b/i) ? $self->{sqltrue} : $self->{sqlfalse};
742 ARRAYREFREF => sub { # literal SQL with bind
743 my ($sql, @bind) = @$$vals;
744 $self->_assert_bindval_matches_bindtype(@bind);
745 return ("$label $op ( $sql )", @bind);
749 puke "special op 'in' requires an arrayref (or arrayref-ref)";
753 return ($sql, @bind);
761 #======================================================================
763 #======================================================================
766 my ($self, $arg) = @_;
768 # construct list of ordering instructions
769 my @order = $self->_SWITCH_refkind($arg, {
772 map {$self->_SWITCH_refkind($_, {
773 SCALAR => sub {$self->_quote($_)},
775 SCALARREF => sub {$$_}, # literal SQL, no quoting
776 HASHREF => sub {$self->_order_by_hash($_)}
780 SCALAR => sub {$self->_quote($arg)},
782 SCALARREF => sub {$$arg}, # literal SQL, no quoting
783 HASHREF => sub {$self->_order_by_hash($arg)},
788 my $order = join ', ', @order;
789 return $order ? $self->_sqlcase(' order by')." $order" : '';
794 my ($self, $hash) = @_;
796 # get first pair in hash
797 my ($key, $val) = each %$hash;
799 # check if one pair was found and no other pair in hash
800 $key && !(each %$hash)
801 or puke "hash passed to _order_by must have exactly one key (-desc or -asc)";
803 my ($order) = ($key =~ /^-(desc|asc)/i)
804 or puke "invalid key in _order_by hash : $key";
806 return $self->_quote($val) ." ". $self->_sqlcase($order);
811 #======================================================================
812 # DATASOURCE (FOR NOW, JUST PLAIN TABLE OR LIST OF TABLES)
813 #======================================================================
818 $self->_SWITCH_refkind($from, {
819 ARRAYREF => sub {join ', ', map { $self->_quote($_) } @$from;},
820 SCALAR => sub {$self->_quote($from)},
821 SCALARREF => sub {$$from},
822 ARRAYREFREF => sub {join ', ', @$from;},
827 #======================================================================
829 #======================================================================
835 $label or puke "can't quote an empty label";
837 # left and right quote characters
838 my ($ql, $qr, @other) = $self->_SWITCH_refkind($self->{quote_char}, {
839 SCALAR => sub {($self->{quote_char}, $self->{quote_char})},
840 ARRAYREF => sub {@{$self->{quote_char}}},
844 or puke "quote_char must be an arrayref of 2 values";
846 # no quoting if no quoting chars
847 $ql or return $label;
849 # no quoting for literal SQL
850 return $$label if ref($label) eq 'SCALAR';
852 # separate table / column (if applicable)
853 my $sep = $self->{name_sep} || '';
854 my @to_quote = $sep ? split /\Q$sep\E/, $label : ($label);
856 # do the quoting, except for "*" or for `table`.*
857 my @quoted = map { $_ eq '*' ? $_: $ql.$_.$qr} @to_quote;
859 # reassemble and return.
860 return join $sep, @quoted;
864 # Conversion, if applicable
866 my ($self, $arg) = @_;
868 # LDNOTE : modified the previous implementation below because
869 # it was not consistent : the first "return" is always an array,
870 # the second "return" is context-dependent. Anyway, _convert
871 # seems always used with just a single argument, so make it a
873 # return @_ unless $self->{convert};
874 # my $conv = $self->_sqlcase($self->{convert});
875 # my @ret = map { $conv.'('.$_.')' } @_;
876 # return wantarray ? @ret : $ret[0];
877 if ($self->{convert}) {
878 my $conv = $self->_sqlcase($self->{convert});
879 $arg = $conv.'('.$arg.')';
887 my($col, @vals) = @_;
889 #LDNOTE : changed original implementation below because it did not make
890 # sense when bindtype eq 'columns' and @vals > 1.
891 # return $self->{bindtype} eq 'columns' ? [ $col, @vals ] : @vals;
893 return $self->{bindtype} eq 'columns' ? map {[$col, $_]} @vals : @vals;
896 # Dies if any element of @bind is not in [colname => value] format
897 # if bindtype is 'columns'.
898 sub _assert_bindval_matches_bindtype {
899 my ($self, @bind) = @_;
901 if ($self->{bindtype} eq 'columns') {
902 foreach my $val (@bind) {
903 if (!defined $val || ref($val) ne 'ARRAY' || @$val != 2) {
904 die "bindtype 'columns' selected, you need to pass: [column_name => bind_value]"
910 sub _join_sql_clauses {
911 my ($self, $logic, $clauses_aref, $bind_aref) = @_;
913 if (@$clauses_aref > 1) {
914 my $join = " " . $self->_sqlcase($logic) . " ";
915 my $sql = '( ' . join($join, @$clauses_aref) . ' )';
916 return ($sql, @$bind_aref);
918 elsif (@$clauses_aref) {
919 return ($clauses_aref->[0], @$bind_aref); # no parentheses
922 return (); # if no SQL, ignore @$bind_aref
927 # Fix SQL case, if so requested
931 # LDNOTE: if $self->{case} is true, then it contains 'lower', so we
932 # don't touch the argument ... crooked logic, but let's not change it!
933 return $self->{case} ? $_[0] : uc($_[0]);
937 #======================================================================
938 # DISPATCHING FROM REFKIND
939 #======================================================================
942 my ($self, $data) = @_;
948 # blessed objects are treated like scalars
949 $ref = (blessed $data) ? '' : ref $data;
950 $n_steps += 1 if $ref;
951 last if $ref ne 'REF';
955 my $base = $ref || (defined $data ? 'SCALAR' : 'UNDEF');
957 return $base . ('REF' x $n_steps);
963 my ($self, $data) = @_;
964 my @try = ($self->_refkind($data));
965 push @try, 'SCALAR_or_UNDEF' if $try[0] eq 'SCALAR' || $try[0] eq 'UNDEF';
966 push @try, 'FALLBACK';
970 sub _METHOD_FOR_refkind {
971 my ($self, $meth_prefix, $data) = @_;
972 my $method = first {$_} map {$self->can($meth_prefix."_".$_)}
973 $self->_try_refkind($data)
974 or puke "cannot dispatch on '$meth_prefix' for ".$self->_refkind($data);
979 sub _SWITCH_refkind {
980 my ($self, $data, $dispatch_table) = @_;
982 my $coderef = first {$_} map {$dispatch_table->{$_}}
983 $self->_try_refkind($data)
984 or puke "no dispatch entry for ".$self->_refkind($data);
991 #======================================================================
992 # VALUES, GENERATE, AUTOLOAD
993 #======================================================================
995 # LDNOTE: original code from nwiger, didn't touch code in that section
996 # I feel the AUTOLOAD stuff should not be the default, it should
997 # only be activated on explicit demand by user.
1001 my $data = shift || return;
1002 puke "Argument to ", __PACKAGE__, "->values must be a \\%hash"
1003 unless ref $data eq 'HASH';
1006 foreach my $k ( sort keys %$data ) {
1007 my $v = $data->{$k};
1008 $self->_SWITCH_refkind($v, {
1010 if ($self->{array_datatypes}) { # array datatype
1011 push @all_bind, $self->_bindtype($k, $v);
1013 else { # literal SQL with bind
1014 my ($sql, @bind) = @$v;
1015 $self->_assert_bindval_matches_bindtype(@bind);
1016 push @all_bind, @bind;
1019 ARRAYREFREF => sub { # literal SQL with bind
1020 my ($sql, @bind) = @${$v};
1021 $self->_assert_bindval_matches_bindtype(@bind);
1022 push @all_bind, @bind;
1024 SCALARREF => sub { # literal SQL without bind
1026 SCALAR_or_UNDEF => sub {
1027 push @all_bind, $self->_bindtype($k, $v);
1038 my(@sql, @sqlq, @sqlv);
1042 if ($ref eq 'HASH') {
1043 for my $k (sort keys %$_) {
1046 my $label = $self->_quote($k);
1047 if ($r eq 'ARRAY') {
1048 # literal SQL with bind
1049 my ($sql, @bind) = @$v;
1050 $self->_assert_bindval_matches_bindtype(@bind);
1051 push @sqlq, "$label = $sql";
1053 } elsif ($r eq 'SCALAR') {
1054 # literal SQL without bind
1055 push @sqlq, "$label = $$v";
1057 push @sqlq, "$label = ?";
1058 push @sqlv, $self->_bindtype($k, $v);
1061 push @sql, $self->_sqlcase('set'), join ', ', @sqlq;
1062 } elsif ($ref eq 'ARRAY') {
1063 # unlike insert(), assume these are ONLY the column names, i.e. for SQL
1066 if ($r eq 'ARRAY') { # literal SQL with bind
1067 my ($sql, @bind) = @$v;
1068 $self->_assert_bindval_matches_bindtype(@bind);
1071 } elsif ($r eq 'SCALAR') { # literal SQL without bind
1072 # embedded literal SQL
1079 push @sql, '(' . join(', ', @sqlq) . ')';
1080 } elsif ($ref eq 'SCALAR') {
1084 # strings get case twiddled
1085 push @sql, $self->_sqlcase($_);
1089 my $sql = join ' ', @sql;
1091 # this is pretty tricky
1092 # if ask for an array, return ($stmt, @bind)
1093 # otherwise, s/?/shift @sqlv/ to put it inline
1095 return ($sql, @sqlv);
1097 1 while $sql =~ s/\?/my $d = shift(@sqlv);
1098 ref $d ? $d->[1] : $d/e;
1107 # This allows us to check for a local, then _form, attr
1109 my($name) = $AUTOLOAD =~ /.*::(.+)/;
1110 return $self->generate($name, @_);
1121 SQL::Abstract - Generate SQL from Perl data structures
1127 my $sql = SQL::Abstract->new;
1129 my($stmt, @bind) = $sql->select($table, \@fields, \%where, \@order);
1131 my($stmt, @bind) = $sql->insert($table, \%fieldvals || \@values);
1133 my($stmt, @bind) = $sql->update($table, \%fieldvals, \%where);
1135 my($stmt, @bind) = $sql->delete($table, \%where);
1137 # Then, use these in your DBI statements
1138 my $sth = $dbh->prepare($stmt);
1139 $sth->execute(@bind);
1141 # Just generate the WHERE clause
1142 my($stmt, @bind) = $sql->where(\%where, \@order);
1144 # Return values in the same order, for hashed queries
1145 # See PERFORMANCE section for more details
1146 my @bind = $sql->values(\%fieldvals);
1150 This module was inspired by the excellent L<DBIx::Abstract>.
1151 However, in using that module I found that what I really wanted
1152 to do was generate SQL, but still retain complete control over my
1153 statement handles and use the DBI interface. So, I set out to
1154 create an abstract SQL generation module.
1156 While based on the concepts used by L<DBIx::Abstract>, there are
1157 several important differences, especially when it comes to WHERE
1158 clauses. I have modified the concepts used to make the SQL easier
1159 to generate from Perl data structures and, IMO, more intuitive.
1160 The underlying idea is for this module to do what you mean, based
1161 on the data structures you provide it. The big advantage is that
1162 you don't have to modify your code every time your data changes,
1163 as this module figures it out.
1165 To begin with, an SQL INSERT is as easy as just specifying a hash
1166 of C<key=value> pairs:
1169 name => 'Jimbo Bobson',
1170 phone => '123-456-7890',
1171 address => '42 Sister Lane',
1172 city => 'St. Louis',
1173 state => 'Louisiana',
1176 The SQL can then be generated with this:
1178 my($stmt, @bind) = $sql->insert('people', \%data);
1180 Which would give you something like this:
1182 $stmt = "INSERT INTO people
1183 (address, city, name, phone, state)
1184 VALUES (?, ?, ?, ?, ?)";
1185 @bind = ('42 Sister Lane', 'St. Louis', 'Jimbo Bobson',
1186 '123-456-7890', 'Louisiana');
1188 These are then used directly in your DBI code:
1190 my $sth = $dbh->prepare($stmt);
1191 $sth->execute(@bind);
1193 =head2 Inserting and Updating Arrays
1195 If your database has array types (like for example Postgres),
1196 activate the special option C<< array_datatypes => 1 >>
1197 when creating the C<SQL::Abstract> object.
1198 Then you may use an arrayref to insert and update database array types:
1200 my $sql = SQL::Abstract->new(array_datatypes => 1);
1202 planets => [qw/Mercury Venus Earth Mars/]
1205 my($stmt, @bind) = $sql->insert('solar_system', \%data);
1209 $stmt = "INSERT INTO solar_system (planets) VALUES (?)"
1211 @bind = (['Mercury', 'Venus', 'Earth', 'Mars']);
1214 =head2 Inserting and Updating SQL
1216 In order to apply SQL functions to elements of your C<%data> you may
1217 specify a reference to an arrayref for the given hash value. For example,
1218 if you need to execute the Oracle C<to_date> function on a value, you can
1219 say something like this:
1223 date_entered => \["to_date(?,'MM/DD/YYYY')", "03/02/2003"],
1226 The first value in the array is the actual SQL. Any other values are
1227 optional and would be included in the bind values array. This gives
1230 my($stmt, @bind) = $sql->insert('people', \%data);
1232 $stmt = "INSERT INTO people (name, date_entered)
1233 VALUES (?, to_date(?,'MM/DD/YYYY'))";
1234 @bind = ('Bill', '03/02/2003');
1236 An UPDATE is just as easy, all you change is the name of the function:
1238 my($stmt, @bind) = $sql->update('people', \%data);
1240 Notice that your C<%data> isn't touched; the module will generate
1241 the appropriately quirky SQL for you automatically. Usually you'll
1242 want to specify a WHERE clause for your UPDATE, though, which is
1243 where handling C<%where> hashes comes in handy...
1245 =head2 Complex where statements
1247 This module can generate pretty complicated WHERE statements
1248 easily. For example, simple C<key=value> pairs are taken to mean
1249 equality, and if you want to see if a field is within a set
1250 of values, you can use an arrayref. Let's say we wanted to
1251 SELECT some data based on this criteria:
1254 requestor => 'inna',
1255 worker => ['nwiger', 'rcwe', 'sfz'],
1256 status => { '!=', 'completed' }
1259 my($stmt, @bind) = $sql->select('tickets', '*', \%where);
1261 The above would give you something like this:
1263 $stmt = "SELECT * FROM tickets WHERE
1264 ( requestor = ? ) AND ( status != ? )
1265 AND ( worker = ? OR worker = ? OR worker = ? )";
1266 @bind = ('inna', 'completed', 'nwiger', 'rcwe', 'sfz');
1268 Which you could then use in DBI code like so:
1270 my $sth = $dbh->prepare($stmt);
1271 $sth->execute(@bind);
1277 The functions are simple. There's one for each major SQL operation,
1278 and a constructor you use first. The arguments are specified in a
1279 similar order to each function (table, then fields, then a where
1280 clause) to try and simplify things.
1285 =head2 new(option => 'value')
1287 The C<new()> function takes a list of options and values, and returns
1288 a new B<SQL::Abstract> object which can then be used to generate SQL
1289 through the methods below. The options accepted are:
1295 If set to 'lower', then SQL will be generated in all lowercase. By
1296 default SQL is generated in "textbook" case meaning something like:
1298 SELECT a_field FROM a_table WHERE some_field LIKE '%someval%'
1300 Any setting other than 'lower' is ignored.
1304 This determines what the default comparison operator is. By default
1305 it is C<=>, meaning that a hash like this:
1307 %where = (name => 'nwiger', email => 'nate@wiger.org');
1309 Will generate SQL like this:
1311 WHERE name = 'nwiger' AND email = 'nate@wiger.org'
1313 However, you may want loose comparisons by default, so if you set
1314 C<cmp> to C<like> you would get SQL such as:
1316 WHERE name like 'nwiger' AND email like 'nate@wiger.org'
1318 You can also override the comparsion on an individual basis - see
1319 the huge section on L</"WHERE CLAUSES"> at the bottom.
1321 =item sqltrue, sqlfalse
1323 Expressions for inserting boolean values within SQL statements.
1324 By default these are C<1=1> and C<1=0>.
1328 This determines the default logical operator for multiple WHERE
1329 statements in arrays. By default it is "or", meaning that a WHERE
1333 event_date => {'>=', '2/13/99'},
1334 event_date => {'<=', '4/24/03'},
1337 Will generate SQL like this:
1339 WHERE event_date >= '2/13/99' OR event_date <= '4/24/03'
1341 This is probably not what you want given this query, though (look
1342 at the dates). To change the "OR" to an "AND", simply specify:
1344 my $sql = SQL::Abstract->new(logic => 'and');
1346 Which will change the above C<WHERE> to:
1348 WHERE event_date >= '2/13/99' AND event_date <= '4/24/03'
1350 The logic can also be changed locally by inserting
1351 an extra first element in the array :
1353 @where = (-and => event_date => {'>=', '2/13/99'},
1354 event_date => {'<=', '4/24/03'} );
1356 See the L</"WHERE CLAUSES"> section for explanations.
1360 This will automatically convert comparisons using the specified SQL
1361 function for both column and value. This is mostly used with an argument
1362 of C<upper> or C<lower>, so that the SQL will have the effect of
1363 case-insensitive "searches". For example, this:
1365 $sql = SQL::Abstract->new(convert => 'upper');
1366 %where = (keywords => 'MaKe iT CAse inSeNSItive');
1368 Will turn out the following SQL:
1370 WHERE upper(keywords) like upper('MaKe iT CAse inSeNSItive')
1372 The conversion can be C<upper()>, C<lower()>, or any other SQL function
1373 that can be applied symmetrically to fields (actually B<SQL::Abstract> does
1374 not validate this option; it will just pass through what you specify verbatim).
1378 This is a kludge because many databases suck. For example, you can't
1379 just bind values using DBI's C<execute()> for Oracle C<CLOB> or C<BLOB> fields.
1380 Instead, you have to use C<bind_param()>:
1382 $sth->bind_param(1, 'reg data');
1383 $sth->bind_param(2, $lots, {ora_type => ORA_CLOB});
1385 The problem is, B<SQL::Abstract> will normally just return a C<@bind> array,
1386 which loses track of which field each slot refers to. Fear not.
1388 If you specify C<bindtype> in new, you can determine how C<@bind> is returned.
1389 Currently, you can specify either C<normal> (default) or C<columns>. If you
1390 specify C<columns>, you will get an array that looks like this:
1392 my $sql = SQL::Abstract->new(bindtype => 'columns');
1393 my($stmt, @bind) = $sql->insert(...);
1396 [ 'column1', 'value1' ],
1397 [ 'column2', 'value2' ],
1398 [ 'column3', 'value3' ],
1401 You can then iterate through this manually, using DBI's C<bind_param()>.
1403 $sth->prepare($stmt);
1406 my($col, $data) = @$_;
1407 if ($col eq 'details' || $col eq 'comments') {
1408 $sth->bind_param($i, $data, {ora_type => ORA_CLOB});
1409 } elsif ($col eq 'image') {
1410 $sth->bind_param($i, $data, {ora_type => ORA_BLOB});
1412 $sth->bind_param($i, $data);
1416 $sth->execute; # execute without @bind now
1418 Now, why would you still use B<SQL::Abstract> if you have to do this crap?
1419 Basically, the advantage is still that you don't have to care which fields
1420 are or are not included. You could wrap that above C<for> loop in a simple
1421 sub called C<bind_fields()> or something and reuse it repeatedly. You still
1422 get a layer of abstraction over manual SQL specification.
1424 Note that if you set L</bindtype> to C<columns>, the C<\[$sql, @bind]>
1425 construct (see L</Literal SQL with placeholders and bind values (subqueries)>)
1426 will expect the bind values in this format.
1430 This is the character that a table or column name will be quoted
1431 with. By default this is an empty string, but you could set it to
1432 the character C<`>, to generate SQL like this:
1434 SELECT `a_field` FROM `a_table` WHERE `some_field` LIKE '%someval%'
1436 Alternatively, you can supply an array ref of two items, the first being the left
1437 hand quote character, and the second the right hand quote character. For
1438 example, you could supply C<['[',']']> for SQL Server 2000 compliant quotes
1439 that generates SQL like this:
1441 SELECT [a_field] FROM [a_table] WHERE [some_field] LIKE '%someval%'
1443 Quoting is useful if you have tables or columns names that are reserved
1444 words in your database's SQL dialect.
1448 This is the character that separates a table and column name. It is
1449 necessary to specify this when the C<quote_char> option is selected,
1450 so that tables and column names can be individually quoted like this:
1452 SELECT `table`.`one_field` FROM `table` WHERE `table`.`other_field` = 1
1454 =item array_datatypes
1456 When this option is true, arrayrefs in INSERT or UPDATE are
1457 interpreted as array datatypes and are passed directly
1459 When this option is false, arrayrefs are interpreted
1460 as literal SQL, just like refs to arrayrefs
1461 (but this behavior is for backwards compatibility; when writing
1462 new queries, use the "reference to arrayref" syntax
1468 Takes a reference to a list of "special operators"
1469 to extend the syntax understood by L<SQL::Abstract>.
1470 See section L</"SPECIAL OPERATORS"> for details.
1476 =head2 insert($table, \@values || \%fieldvals)
1478 This is the simplest function. You simply give it a table name
1479 and either an arrayref of values or hashref of field/value pairs.
1480 It returns an SQL INSERT statement and a list of bind values.
1481 See the sections on L</"Inserting and Updating Arrays"> and
1482 L</"Inserting and Updating SQL"> for information on how to insert
1483 with those data types.
1485 =head2 update($table, \%fieldvals, \%where)
1487 This takes a table, hashref of field/value pairs, and an optional
1488 hashref L<WHERE clause|/WHERE CLAUSES>. It returns an SQL UPDATE function and a list
1490 See the sections on L</"Inserting and Updating Arrays"> and
1491 L</"Inserting and Updating SQL"> for information on how to insert
1492 with those data types.
1494 =head2 select($source, $fields, $where, $order)
1496 This returns a SQL SELECT statement and associated list of bind values, as
1497 specified by the arguments :
1503 Specification of the 'FROM' part of the statement.
1504 The argument can be either a plain scalar (interpreted as a table
1505 name, will be quoted), or an arrayref (interpreted as a list
1506 of table names, joined by commas, quoted), or a scalarref
1507 (literal table name, not quoted), or a ref to an arrayref
1508 (list of literal table names, joined by commas, not quoted).
1512 Specification of the list of fields to retrieve from
1514 The argument can be either an arrayref (interpreted as a list
1515 of field names, will be joined by commas and quoted), or a
1516 plain scalar (literal SQL, not quoted).
1517 Please observe that this API is not as flexible as for
1518 the first argument C<$table>, for backwards compatibility reasons.
1522 Optional argument to specify the WHERE part of the query.
1523 The argument is most often a hashref, but can also be
1524 an arrayref or plain scalar --
1525 see section L<WHERE clause|/"WHERE CLAUSES"> for details.
1529 Optional argument to specify the ORDER BY part of the query.
1530 The argument can be a scalar, a hashref or an arrayref
1531 -- see section L<ORDER BY clause|/"ORDER BY CLAUSES">
1537 =head2 delete($table, \%where)
1539 This takes a table name and optional hashref L<WHERE clause|/WHERE CLAUSES>.
1540 It returns an SQL DELETE statement and list of bind values.
1542 =head2 where(\%where, \@order)
1544 This is used to generate just the WHERE clause. For example,
1545 if you have an arbitrary data structure and know what the
1546 rest of your SQL is going to look like, but want an easy way
1547 to produce a WHERE clause, use this. It returns an SQL WHERE
1548 clause and list of bind values.
1551 =head2 values(\%data)
1553 This just returns the values from the hash C<%data>, in the same
1554 order that would be returned from any of the other above queries.
1555 Using this allows you to markedly speed up your queries if you
1556 are affecting lots of rows. See below under the L</"PERFORMANCE"> section.
1558 =head2 generate($any, 'number', $of, \@data, $struct, \%types)
1560 Warning: This is an experimental method and subject to change.
1562 This returns arbitrarily generated SQL. It's a really basic shortcut.
1563 It will return two different things, depending on return context:
1565 my($stmt, @bind) = $sql->generate('create table', \$table, \@fields);
1566 my $stmt_and_val = $sql->generate('create table', \$table, \@fields);
1568 These would return the following:
1570 # First calling form
1571 $stmt = "CREATE TABLE test (?, ?)";
1572 @bind = (field1, field2);
1574 # Second calling form
1575 $stmt_and_val = "CREATE TABLE test (field1, field2)";
1577 Depending on what you're trying to do, it's up to you to choose the correct
1578 format. In this example, the second form is what you would want.
1582 $sql->generate('alter session', { nls_date_format => 'MM/YY' });
1586 ALTER SESSION SET nls_date_format = 'MM/YY'
1588 You get the idea. Strings get their case twiddled, but everything
1589 else remains verbatim.
1594 =head1 WHERE CLAUSES
1598 This module uses a variation on the idea from L<DBIx::Abstract>. It
1599 is B<NOT>, repeat I<not> 100% compatible. B<The main logic of this
1600 module is that things in arrays are OR'ed, and things in hashes
1603 The easiest way to explain is to show lots of examples. After
1604 each C<%where> hash shown, it is assumed you used:
1606 my($stmt, @bind) = $sql->where(\%where);
1608 However, note that the C<%where> hash can be used directly in any
1609 of the other functions as well, as described above.
1611 =head2 Key-value pairs
1613 So, let's get started. To begin, a simple hash:
1617 status => 'completed'
1620 Is converted to SQL C<key = val> statements:
1622 $stmt = "WHERE user = ? AND status = ?";
1623 @bind = ('nwiger', 'completed');
1625 One common thing I end up doing is having a list of values that
1626 a field can be in. To do this, simply specify a list inside of
1631 status => ['assigned', 'in-progress', 'pending'];
1634 This simple code will create the following:
1636 $stmt = "WHERE user = ? AND ( status = ? OR status = ? OR status = ? )";
1637 @bind = ('nwiger', 'assigned', 'in-progress', 'pending');
1639 An empty arrayref will be considered a logical false and
1642 =head2 Key-value pairs
1644 If you want to specify a different type of operator for your comparison,
1645 you can use a hashref for a given column:
1649 status => { '!=', 'completed' }
1652 Which would generate:
1654 $stmt = "WHERE user = ? AND status != ?";
1655 @bind = ('nwiger', 'completed');
1657 To test against multiple values, just enclose the values in an arrayref:
1659 status => { '!=', ['assigned', 'in-progress', 'pending'] };
1661 Which would give you:
1663 "WHERE status != ? AND status != ? AND status != ?"
1665 Notice that since the operator was recognized as being a 'negative'
1666 operator, the arrayref was interpreted with 'AND' logic (because
1667 of Morgan's laws). By contrast, the reverse
1669 status => { '=', ['assigned', 'in-progress', 'pending'] };
1673 "WHERE status = ? OR status = ? OR status = ?"
1676 The hashref can also contain multiple pairs, in which case it is expanded
1677 into an C<AND> of its elements:
1681 status => { '!=', 'completed', -not_like => 'pending%' }
1684 # Or more dynamically, like from a form
1685 $where{user} = 'nwiger';
1686 $where{status}{'!='} = 'completed';
1687 $where{status}{'-not_like'} = 'pending%';
1689 # Both generate this
1690 $stmt = "WHERE user = ? AND status != ? AND status NOT LIKE ?";
1691 @bind = ('nwiger', 'completed', 'pending%');
1694 To get an OR instead, you can combine it with the arrayref idea:
1698 priority => [ {'=', 2}, {'!=', 1} ]
1701 Which would generate:
1703 $stmt = "WHERE user = ? AND priority = ? OR priority != ?";
1704 @bind = ('nwiger', '2', '1');
1706 If you want to include literal SQL (with or without bind values), just use a
1707 scalar reference or array reference as the value:
1710 date_entered => { '>' => \["to_date(?, 'MM/DD/YYYY')", "11/26/2008"] },
1711 date_expires => { '<' => \"now()" }
1714 Which would generate:
1716 $stmt = "WHERE date_entered > "to_date(?, 'MM/DD/YYYY') AND date_expires < now()";
1717 @bind = ('11/26/2008');
1720 =head2 Logic and nesting operators
1722 In the example above,
1723 there is a subtle trap if you want to say something like
1724 this (notice the C<AND>):
1726 WHERE priority != ? AND priority != ?
1728 Because, in Perl you I<can't> do this:
1730 priority => { '!=', 2, '!=', 1 }
1732 As the second C<!=> key will obliterate the first. The solution
1733 is to use the special C<-modifier> form inside an arrayref:
1735 priority => [ -and => {'!=', 2},
1739 Normally, these would be joined by C<OR>, but the modifier tells it
1740 to use C<AND> instead. (Hint: You can use this in conjunction with the
1741 C<logic> option to C<new()> in order to change the way your queries
1742 work by default.) B<Important:> Note that the C<-modifier> goes
1743 B<INSIDE> the arrayref, as an extra first element. This will
1744 B<NOT> do what you think it might:
1746 priority => -and => [{'!=', 2}, {'!=', 1}] # WRONG!
1748 Here is a quick list of equivalencies, since there is some overlap:
1751 status => {'!=', 'completed', 'not like', 'pending%' }
1752 status => [ -and => {'!=', 'completed'}, {'not like', 'pending%'}]
1755 status => {'=', ['assigned', 'in-progress']}
1756 status => [ -or => {'=', 'assigned'}, {'=', 'in-progress'}]
1757 status => [ {'=', 'assigned'}, {'=', 'in-progress'} ]
1759 In addition to C<-and> and C<-or>, there is also a special C<-nest>
1760 operator which adds an additional set of parens, to create a subquery.
1761 For example, to get something like this:
1763 $stmt = "WHERE user = ? AND ( workhrs > ? OR geo = ? )";
1764 @bind = ('nwiger', '20', 'ASIA');
1770 -nest => [ workhrs => {'>', 20}, geo => 'ASIA' ],
1773 If you need several nested subexpressions, you can number
1774 the C<-nest> branches :
1784 =head2 Special operators : IN, BETWEEN, etc.
1786 You can also use the hashref format to compare a list of fields using the
1787 C<IN> comparison operator, by specifying the list as an arrayref:
1790 status => 'completed',
1791 reportid => { -in => [567, 2335, 2] }
1794 Which would generate:
1796 $stmt = "WHERE status = ? AND reportid IN (?,?,?)";
1797 @bind = ('completed', '567', '2335', '2');
1799 The reverse operator C<-not_in> generates SQL C<NOT IN> and is used in
1802 Another pair of operators is C<-between> and C<-not_between>,
1803 used with an arrayref of two values:
1807 completion_date => {
1808 -not_between => ['2002-10-01', '2003-02-06']
1814 WHERE user = ? AND completion_date NOT BETWEEN ( ? AND ? )
1816 These are the two builtin "special operators"; but the
1817 list can be expanded : see section L</"SPECIAL OPERATORS"> below.
1819 =head2 Nested conditions
1821 So far, we've seen how multiple conditions are joined with a top-level
1822 C<AND>. We can change this by putting the different conditions we want in
1823 hashes and then putting those hashes in an array. For example:
1828 status => { -like => ['pending%', 'dispatched'] },
1832 status => 'unassigned',
1836 This data structure would create the following:
1838 $stmt = "WHERE ( user = ? AND ( status LIKE ? OR status LIKE ? ) )
1839 OR ( user = ? AND status = ? ) )";
1840 @bind = ('nwiger', 'pending', 'dispatched', 'robot', 'unassigned');
1842 This can be combined with the C<-nest> operator to properly group
1849 ["-and", workhrs => {'>', 20}, geo => 'ASIA' ],
1850 ["-and", workhrs => {'<', 50}, geo => 'EURO' ]
1857 WHERE ( user = ? AND
1858 ( ( workhrs > ? AND geo = ? )
1859 OR ( workhrs < ? AND geo = ? ) ) )
1863 Finally, sometimes only literal SQL will do. If you want to include
1864 literal SQL verbatim, you can specify it as a scalar reference, namely:
1866 my $inn = 'is Not Null';
1868 priority => { '<', 2 },
1874 $stmt = "WHERE priority < ? AND requestor is Not Null";
1877 Note that in this example, you only get one bind parameter back, since
1878 the verbatim SQL is passed as part of the statement.
1880 Of course, just to prove a point, the above can also be accomplished
1884 priority => { '<', 2 },
1885 requestor => { '!=', undef },
1891 Conditions on boolean columns can be expressed in the
1892 same way, passing a reference to an empty string :
1895 priority => { '<', 2 },
1901 $stmt = "WHERE priority < ? AND is_ready";
1905 =head2 Literal SQL with placeholders and bind values (subqueries)
1907 If the literal SQL to be inserted has placeholders and bind values,
1908 use a reference to an arrayref (yes this is a double reference --
1909 not so common, but perfectly legal Perl). For example, to find a date
1910 in Postgres you can use something like this:
1913 date_column => \[q/= date '2008-09-30' - ?::integer/, 10/]
1918 $stmt = "WHERE ( date_column = date '2008-09-30' - ?::integer )"
1921 Note that you must pass the bind values in the same format as they are returned
1922 by C</where>. That means that if you set L</bindtype> to C<columns>, you must
1923 provide the bind values in the C<< [ column_meta => value ] >> format, where
1924 C<column_meta> is an opaque scalar value; most commonly the column name, but
1925 you can use any scalar scalar value (including references and blessed
1926 references), L<SQL::Abstract> will simply pass it through intact. So eg. the
1927 above example will look like:
1930 date_column => \[q/= date '2008-09-30' - ?::integer/, [ dummy => 10 ]/]
1933 Literal SQL is especially useful for nesting parenthesized clauses in the
1934 main SQL query. Here is a first example :
1936 my ($sub_stmt, @sub_bind) = ("SELECT c1 FROM t1 WHERE c2 < ? AND c3 LIKE ?",
1940 bar => \["IN ($sub_stmt)" => @sub_bind],
1945 $stmt = "WHERE (foo = ? AND bar IN (SELECT c1 FROM t1
1946 WHERE c2 < ? AND c3 LIKE ?))";
1947 @bind = (1234, 100, "foo%");
1949 Other subquery operators, like for example C<"E<gt> ALL"> or C<"NOT IN">,
1950 are expressed in the same way. Of course the C<$sub_stmt> and
1951 its associated bind values can be generated through a former call
1954 my ($sub_stmt, @sub_bind)
1955 = $sql->select("t1", "c1", {c2 => {"<" => 100},
1956 c3 => {-like => "foo%"}});
1959 bar => \["> ALL ($sub_stmt)" => @sub_bind],
1962 In the examples above, the subquery was used as an operator on a column;
1963 but the same principle also applies for a clause within the main C<%where>
1964 hash, like an EXISTS subquery :
1966 my ($sub_stmt, @sub_bind)
1967 = $sql->select("t1", "*", {c1 => 1, c2 => \"> t0.c0"});
1970 -nest => \["EXISTS ($sub_stmt)" => @sub_bind],
1975 $stmt = "WHERE (foo = ? AND EXISTS (SELECT * FROM t1
1976 WHERE c1 = ? AND c2 > t0.c0))";
1980 Observe that the condition on C<c2> in the subquery refers to
1981 column C<t0.c0> of the main query : this is I<not> a bind
1982 value, so we have to express it through a scalar ref.
1983 Writing C<< c2 => {">" => "t0.c0"} >> would have generated
1984 C<< c2 > ? >> with bind value C<"t0.c0"> ... not exactly
1985 what we wanted here.
1987 Another use of the subquery technique is when some SQL clauses need
1988 parentheses, as it often occurs with some proprietary SQL extensions
1989 like for example fulltext expressions, geospatial expressions,
1990 NATIVE clauses, etc. Here is an example of a fulltext query in MySQL :
1993 -nest => \["MATCH (col1, col2) AGAINST (?)" => qw/apples/]
1996 Finally, here is an example where a subquery is used
1997 for expressing unary negation:
1999 my ($sub_stmt, @sub_bind)
2000 = $sql->where({age => [{"<" => 10}, {">" => 20}]});
2001 $sub_stmt =~ s/^ where //i; # don't want "WHERE" in the subclause
2003 lname => {like => '%son%'},
2004 -nest => \["NOT ($sub_stmt)" => @sub_bind],
2009 $stmt = "lname LIKE ? AND NOT ( age < ? OR age > ? )"
2010 @bind = ('%son%', 10, 20)
2016 These pages could go on for a while, since the nesting of the data
2017 structures this module can handle are pretty much unlimited (the
2018 module implements the C<WHERE> expansion as a recursive function
2019 internally). Your best bet is to "play around" with the module a
2020 little to see how the data structures behave, and choose the best
2021 format for your data based on that.
2023 And of course, all the values above will probably be replaced with
2024 variables gotten from forms or the command line. After all, if you
2025 knew everything ahead of time, you wouldn't have to worry about
2026 dynamically-generating SQL and could just hardwire it into your
2032 =head1 ORDER BY CLAUSES
2034 Some functions take an order by clause. This can either be a scalar (just a
2035 column name,) a hash of C<< { -desc => 'col' } >> or C<< { -asc => 'col' } >>,
2036 or an array of either of the two previous forms. Examples:
2038 Given | Will Generate
2039 ----------------------------------------------------------
2040 \'colA DESC' | ORDER BY colA DESC
2041 'colA' | ORDER BY colA
2042 [qw/colA colB/] | ORDER BY colA, colB
2043 {-asc => 'colA'} | ORDER BY colA ASC
2044 {-desc => 'colB'} | ORDER BY colB DESC
2046 {-asc => 'colA'}, | ORDER BY colA ASC, colB DESC
2049 [colA => {-asc => 'colB'}] | ORDER BY colA, colB ASC
2050 ==========================================================
2054 =head1 SPECIAL OPERATORS
2056 my $sqlmaker = SQL::Abstract->new(special_ops => [
2059 my ($self, $field, $op, $arg) = @_;
2065 A "special operator" is a SQL syntactic clause that can be
2066 applied to a field, instead of a usual binary operator.
2069 WHERE field IN (?, ?, ?)
2070 WHERE field BETWEEN ? AND ?
2071 WHERE MATCH(field) AGAINST (?, ?)
2073 Special operators IN and BETWEEN are fairly standard and therefore
2074 are builtin within C<SQL::Abstract>. For other operators,
2075 like the MATCH .. AGAINST example above which is
2076 specific to MySQL, you can write your own operator handlers :
2077 supply a C<special_ops> argument to the C<new> method.
2078 That argument takes an arrayref of operator definitions;
2079 each operator definition is a hashref with two entries
2085 the regular expression to match the operator
2089 coderef that will be called when meeting that operator
2090 in the input tree. The coderef will be called with
2091 arguments C<< ($self, $field, $op, $arg) >>, and
2092 should return a C<< ($sql, @bind) >> structure.
2096 For example, here is an implementation
2097 of the MATCH .. AGAINST syntax for MySQL
2099 my $sqlmaker = SQL::Abstract->new(special_ops => [
2101 # special op for MySql MATCH (field) AGAINST(word1, word2, ...)
2102 {regex => qr/^match$/i,
2104 my ($self, $field, $op, $arg) = @_;
2105 $arg = [$arg] if not ref $arg;
2106 my $label = $self->_quote($field);
2107 my ($placeholder) = $self->_convert('?');
2108 my $placeholders = join ", ", (($placeholder) x @$arg);
2109 my $sql = $self->_sqlcase('match') . " ($label) "
2110 . $self->_sqlcase('against') . " ($placeholders) ";
2111 my @bind = $self->_bindtype($field, @$arg);
2112 return ($sql, @bind);
2121 Thanks to some benchmarking by Mark Stosberg, it turns out that
2122 this module is many orders of magnitude faster than using C<DBIx::Abstract>.
2123 I must admit this wasn't an intentional design issue, but it's a
2124 byproduct of the fact that you get to control your C<DBI> handles
2127 To maximize performance, use a code snippet like the following:
2129 # prepare a statement handle using the first row
2130 # and then reuse it for the rest of the rows
2132 for my $href (@array_of_hashrefs) {
2133 $stmt ||= $sql->insert('table', $href);
2134 $sth ||= $dbh->prepare($stmt);
2135 $sth->execute($sql->values($href));
2138 The reason this works is because the keys in your C<$href> are sorted
2139 internally by B<SQL::Abstract>. Thus, as long as your data retains
2140 the same structure, you only have to generate the SQL the first time
2141 around. On subsequent queries, simply use the C<values> function provided
2142 by this module to return your values in the correct order.
2147 If you use my C<CGI::FormBuilder> module at all, you'll hopefully
2148 really like this part (I do, at least). Building up a complex query
2149 can be as simple as the following:
2153 use CGI::FormBuilder;
2156 my $form = CGI::FormBuilder->new(...);
2157 my $sql = SQL::Abstract->new;
2159 if ($form->submitted) {
2160 my $field = $form->field;
2161 my $id = delete $field->{id};
2162 my($stmt, @bind) = $sql->update('table', $field, {id => $id});
2165 Of course, you would still have to connect using C<DBI> to run the
2166 query, but the point is that if you make your form look like your
2167 table, the actual query script can be extremely simplistic.
2169 If you're B<REALLY> lazy (I am), check out C<HTML::QuickTable> for
2170 a fast interface to returning and formatting data. I frequently
2171 use these three modules together to write complex database query
2172 apps in under 50 lines.
2177 Version 1.50 was a major internal refactoring of C<SQL::Abstract>.
2178 Great care has been taken to preserve the I<published> behavior
2179 documented in previous versions in the 1.* family; however,
2180 some features that were previously undocumented, or behaved
2181 differently from the documentation, had to be changed in order
2182 to clarify the semantics. Hence, client code that was relying
2183 on some dark areas of C<SQL::Abstract> v1.*
2184 B<might behave differently> in v1.50.
2186 The main changes are :
2192 support for literal SQL through the C<< \ [$sql, bind] >> syntax.
2196 support for the { operator => \"..." } construct (to embed literal SQL)
2200 support for the { operator => \["...", @bind] } construct (to embed literal SQL with bind values)
2204 added -nest1, -nest2 or -nest_1, -nest_2, ...
2208 optional support for L<array datatypes|/"Inserting and Updating Arrays">
2212 defensive programming : check arguments
2216 fixed bug with global logic, which was previously implemented
2217 through global variables yielding side-effects. Prior versons would
2218 interpret C<< [ {cond1, cond2}, [cond3, cond4] ] >>
2219 as C<< "(cond1 AND cond2) OR (cond3 AND cond4)" >>.
2220 Now this is interpreted
2221 as C<< "(cond1 AND cond2) OR (cond3 OR cond4)" >>.
2225 C<-and> / C<-or> operators are no longer accepted
2226 in the middle of an arrayref : they are
2227 only admitted if in first position.
2231 changed logic for distributing an op over arrayrefs
2235 fixed semantics of _bindtype on array args
2239 dropped the C<_anoncopy> of the %where tree. No longer necessary,
2240 we just avoid shifting arrays within that tree.
2244 dropped the C<_modlogic> function
2250 =head1 ACKNOWLEDGEMENTS
2252 There are a number of individuals that have really helped out with
2253 this module. Unfortunately, most of them submitted bugs via CPAN
2254 so I have no idea who they are! But the people I do know are:
2256 Ash Berlin (order_by hash term support)
2257 Matt Trout (DBIx::Class support)
2258 Mark Stosberg (benchmarking)
2259 Chas Owens (initial "IN" operator support)
2260 Philip Collins (per-field SQL functions)
2261 Eric Kolve (hashref "AND" support)
2262 Mike Fragassi (enhancements to "BETWEEN" and "LIKE")
2263 Dan Kubb (support for "quote_char" and "name_sep")
2264 Guillermo Roditi (patch to cleanup "IN" and "BETWEEN", fix and tests for _order_by)
2265 Laurent Dami (internal refactoring, multiple -nest, extensible list of special operators, literal SQL)
2266 Norbert Buchmuller (support for literal SQL in hashpair, misc. fixes & tests)
2272 L<DBIx::Class>, L<DBIx::Abstract>, L<CGI::FormBuilder>, L<HTML::QuickTable>.
2276 Copyright (c) 2001-2007 Nathan Wiger <nwiger@cpan.org>. All Rights Reserved.
2278 This module is actively maintained by Matt Trout <mst@shadowcatsystems.co.uk>
2280 For support, your best bet is to try the C<DBIx::Class> users mailing list.
2281 While not an official support venue, C<DBIx::Class> makes heavy use of
2282 C<SQL::Abstract>, and as such list members there are very familiar with
2283 how to create queries.
2285 This module is free software; you may copy this under the terms of
2286 the GNU General Public License, or the Artistic License, copies of
2287 which should have accompanied your Perl kit.