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 $VERSION = '1.51';
20 # This would confuse some packagers
21 #$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_str, $v) = @_;
447 $op_str =~ /^ (AND|OR|NEST) ( \_? \d* ) $/xi
448 or puke "unknown operator: -$op_str";
450 my $op = uc($1); # uppercase, remove trailing digits
452 belch 'Use of [and|or|nest]_N modifiers is deprecated and will be removed in SQLA v2.0. '
453 . "You probably wanted ...-and => [ $op_str => COND1, $op_str => COND2 ... ]";
456 $self->_debug("OP(-$op) within hashref, recursing...");
458 $self->_SWITCH_refkind($v, {
461 return $self->_where_ARRAYREF($v, $op eq 'NEST' ? '' : $op);
466 return $self->_where_ARRAYREF([%$v], 'OR');
469 return $self->_where_HASHREF($v);
473 SCALARREF => sub { # literal SQL
475 or puke "-$op => \\\$scalar not supported, use -nest => ...";
479 ARRAYREFREF => sub { # literal SQL
481 or puke "-$op => \\[..] not supported, use -nest => ...";
485 SCALAR => sub { # permissively interpreted as SQL
487 or puke "-$op => 'scalar' not supported, use -nest => \\'scalar'";
488 belch "literal SQL should be -nest => \\'scalar' "
489 . "instead of -nest => 'scalar' ";
494 puke "-$op => undef not supported";
500 sub _where_hashpair_ARRAYREF {
501 my ($self, $k, $v) = @_;
504 my @v = @$v; # need copy because of shift below
505 $self->_debug("ARRAY($k) means distribute over elements");
507 # put apart first element if it is an operator (-and, -or)
508 my $op = ($v[0] =~ /^ - (?: AND|OR ) $/ix
512 my @distributed = map { {$k => $_} } @v;
515 $self->_debug("OP($op) reinjected into the distributed array");
516 unshift @distributed, $op;
519 my $logic = $op ? substr($op, 1) : '';
521 return $self->_recurse_where(\@distributed, $logic);
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) = @$$val;
567 $self->_assert_bindval_matches_bindtype(@sub_bind);
568 $sql = join ' ', $self->_convert($self->_quote($k)),
569 $self->_sqlcase($op),
574 UNDEF => sub { # CASE: col => {op => undef} : sql "IS (NOT)? NULL"
575 my $is = ($op =~ $self->{equality_op}) ? 'is' :
576 ($op =~ $self->{inequality_op}) ? 'is not' :
577 puke "unexpected operator '$op' with undef operand";
578 $sql = $self->_quote($k) . $self->_sqlcase(" $is null");
581 FALLBACK => sub { # CASE: col => {op => $scalar}
582 $sql = join ' ', $self->_convert($self->_quote($k)),
583 $self->_sqlcase($op),
584 $self->_convert('?');
585 @bind = $self->_bindtype($k, $val);
591 push @all_bind, @bind;
594 return $self->_join_sql_clauses('and', \@all_sql, \@all_bind);
599 sub _where_field_op_ARRAYREF {
600 my ($self, $k, $op, $vals) = @_;
603 $self->_debug("ARRAY($vals) means multiple elements: [ @$vals ]");
605 # LDNOTE : had planned to change the distribution logic when
606 # $op =~ $self->{inequality_op}, because of Morgan laws :
607 # with {field => {'!=' => [22, 33]}}, it would be ridiculous to generate
608 # WHERE field != 22 OR field != 33 : the user probably means
609 # WHERE field != 22 AND field != 33.
610 # To do this, replace the line below by :
611 # my $logic = ($op =~ $self->{inequality_op}) ? 'AND' : 'OR';
612 # return $self->_recurse_where([map { {$k => {$op, $_}} } @$vals], $logic);
614 # distribute $op over each member of @$vals
615 return $self->_recurse_where([map { {$k => {$op, $_}} } @$vals]);
618 # try to DWIM on equality operators
619 # LDNOTE : not 100% sure this is the correct thing to do ...
620 return ($self->{sqlfalse}) if $op =~ $self->{equality_op};
621 return ($self->{sqltrue}) if $op =~ $self->{inequality_op};
624 puke "operator '$op' applied on an empty array (field '$k')";
629 sub _where_hashpair_SCALARREF {
630 my ($self, $k, $v) = @_;
631 $self->_debug("SCALAR($k) means literal SQL: $$v");
632 my $sql = $self->_quote($k) . " " . $$v;
636 # literal SQL with bind
637 sub _where_hashpair_ARRAYREFREF {
638 my ($self, $k, $v) = @_;
639 $self->_debug("REF($k) means literal SQL: @${$v}");
640 my ($sql, @bind) = @${$v};
641 $self->_assert_bindval_matches_bindtype(@bind);
642 $sql = $self->_quote($k) . " " . $sql;
643 return ($sql, @bind );
646 # literal SQL without bind
647 sub _where_hashpair_SCALAR {
648 my ($self, $k, $v) = @_;
649 $self->_debug("NOREF($k) means simple key=val: $k $self->{cmp} $v");
650 my $sql = join ' ', $self->_convert($self->_quote($k)),
651 $self->_sqlcase($self->{cmp}),
652 $self->_convert('?');
653 my @bind = $self->_bindtype($k, $v);
654 return ( $sql, @bind);
658 sub _where_hashpair_UNDEF {
659 my ($self, $k, $v) = @_;
660 $self->_debug("UNDEF($k) means IS NULL");
661 my $sql = $self->_quote($k) . $self->_sqlcase(' is null');
665 #======================================================================
666 # WHERE: TOP-LEVEL OTHERS (SCALARREF, SCALAR, UNDEF)
667 #======================================================================
670 sub _where_SCALARREF {
671 my ($self, $where) = @_;
674 $self->_debug("SCALAR(*top) means literal SQL: $$where");
680 my ($self, $where) = @_;
683 $self->_debug("NOREF(*top) means literal SQL: $where");
694 #======================================================================
695 # WHERE: BUILTIN SPECIAL OPERATORS (-in, -between)
696 #======================================================================
699 sub _where_field_BETWEEN {
700 my ($self, $k, $op, $vals) = @_;
702 ref $vals eq 'ARRAY' && @$vals == 2
703 or puke "special op 'between' requires an arrayref of two values";
705 my ($label) = $self->_convert($self->_quote($k));
706 my ($placeholder) = $self->_convert('?');
707 my $and = $self->_sqlcase('and');
708 $op = $self->_sqlcase($op);
710 my $sql = "( $label $op $placeholder $and $placeholder )";
711 my @bind = $self->_bindtype($k, @$vals);
716 sub _where_field_IN {
717 my ($self, $k, $op, $vals) = @_;
719 # backwards compatibility : if scalar, force into an arrayref
720 $vals = [$vals] if defined $vals && ! ref $vals;
722 my ($label) = $self->_convert($self->_quote($k));
723 my ($placeholder) = $self->_convert('?');
724 $op = $self->_sqlcase($op);
726 my ($sql, @bind) = $self->_SWITCH_refkind($vals, {
727 ARRAYREF => sub { # list of choices
728 if (@$vals) { # nonempty list
729 my $placeholders = join ", ", (($placeholder) x @$vals);
730 my $sql = "$label $op ( $placeholders )";
731 my @bind = $self->_bindtype($k, @$vals);
733 return ($sql, @bind);
735 else { # empty list : some databases won't understand "IN ()", so DWIM
736 my $sql = ($op =~ /\bnot\b/i) ? $self->{sqltrue} : $self->{sqlfalse};
741 ARRAYREFREF => sub { # literal SQL with bind
742 my ($sql, @bind) = @$$vals;
743 $self->_assert_bindval_matches_bindtype(@bind);
744 return ("$label $op ( $sql )", @bind);
748 puke "special op 'in' requires an arrayref (or arrayref-ref)";
752 return ($sql, @bind);
760 #======================================================================
762 #======================================================================
765 my ($self, $arg) = @_;
767 # construct list of ordering instructions
768 my @order = $self->_SWITCH_refkind($arg, {
771 map {$self->_SWITCH_refkind($_, {
772 SCALAR => sub {$self->_quote($_)},
774 SCALARREF => sub {$$_}, # literal SQL, no quoting
775 HASHREF => sub {$self->_order_by_hash($_)}
779 SCALAR => sub {$self->_quote($arg)},
781 SCALARREF => sub {$$arg}, # literal SQL, no quoting
782 HASHREF => sub {$self->_order_by_hash($arg)},
787 my $order = join ', ', @order;
788 return $order ? $self->_sqlcase(' order by')." $order" : '';
793 my ($self, $hash) = @_;
795 # get first pair in hash
796 my ($key, $val) = each %$hash;
798 # check if one pair was found and no other pair in hash
799 $key && !(each %$hash)
800 or puke "hash passed to _order_by must have exactly one key (-desc or -asc)";
802 my ($order) = ($key =~ /^-(desc|asc)/i)
803 or puke "invalid key in _order_by hash : $key";
805 return $self->_quote($val) ." ". $self->_sqlcase($order);
810 #======================================================================
811 # DATASOURCE (FOR NOW, JUST PLAIN TABLE OR LIST OF TABLES)
812 #======================================================================
817 $self->_SWITCH_refkind($from, {
818 ARRAYREF => sub {join ', ', map { $self->_quote($_) } @$from;},
819 SCALAR => sub {$self->_quote($from)},
820 SCALARREF => sub {$$from},
821 ARRAYREFREF => sub {join ', ', @$from;},
826 #======================================================================
828 #======================================================================
834 $label or puke "can't quote an empty label";
836 # left and right quote characters
837 my ($ql, $qr, @other) = $self->_SWITCH_refkind($self->{quote_char}, {
838 SCALAR => sub {($self->{quote_char}, $self->{quote_char})},
839 ARRAYREF => sub {@{$self->{quote_char}}},
843 or puke "quote_char must be an arrayref of 2 values";
845 # no quoting if no quoting chars
846 $ql or return $label;
848 # no quoting for literal SQL
849 return $$label if ref($label) eq 'SCALAR';
851 # separate table / column (if applicable)
852 my $sep = $self->{name_sep} || '';
853 my @to_quote = $sep ? split /\Q$sep\E/, $label : ($label);
855 # do the quoting, except for "*" or for `table`.*
856 my @quoted = map { $_ eq '*' ? $_: $ql.$_.$qr} @to_quote;
858 # reassemble and return.
859 return join $sep, @quoted;
863 # Conversion, if applicable
865 my ($self, $arg) = @_;
867 # LDNOTE : modified the previous implementation below because
868 # it was not consistent : the first "return" is always an array,
869 # the second "return" is context-dependent. Anyway, _convert
870 # seems always used with just a single argument, so make it a
872 # return @_ unless $self->{convert};
873 # my $conv = $self->_sqlcase($self->{convert});
874 # my @ret = map { $conv.'('.$_.')' } @_;
875 # return wantarray ? @ret : $ret[0];
876 if ($self->{convert}) {
877 my $conv = $self->_sqlcase($self->{convert});
878 $arg = $conv.'('.$arg.')';
886 my($col, @vals) = @_;
888 #LDNOTE : changed original implementation below because it did not make
889 # sense when bindtype eq 'columns' and @vals > 1.
890 # return $self->{bindtype} eq 'columns' ? [ $col, @vals ] : @vals;
892 return $self->{bindtype} eq 'columns' ? map {[$col, $_]} @vals : @vals;
895 # Dies if any element of @bind is not in [colname => value] format
896 # if bindtype is 'columns'.
897 sub _assert_bindval_matches_bindtype {
898 my ($self, @bind) = @_;
900 if ($self->{bindtype} eq 'columns') {
901 foreach my $val (@bind) {
902 if (!defined $val || ref($val) ne 'ARRAY' || @$val != 2) {
903 die "bindtype 'columns' selected, you need to pass: [column_name => bind_value]"
909 sub _join_sql_clauses {
910 my ($self, $logic, $clauses_aref, $bind_aref) = @_;
912 if (@$clauses_aref > 1) {
913 my $join = " " . $self->_sqlcase($logic) . " ";
914 my $sql = '( ' . join($join, @$clauses_aref) . ' )';
915 return ($sql, @$bind_aref);
917 elsif (@$clauses_aref) {
918 return ($clauses_aref->[0], @$bind_aref); # no parentheses
921 return (); # if no SQL, ignore @$bind_aref
926 # Fix SQL case, if so requested
930 # LDNOTE: if $self->{case} is true, then it contains 'lower', so we
931 # don't touch the argument ... crooked logic, but let's not change it!
932 return $self->{case} ? $_[0] : uc($_[0]);
936 #======================================================================
937 # DISPATCHING FROM REFKIND
938 #======================================================================
941 my ($self, $data) = @_;
947 # blessed objects are treated like scalars
948 $ref = (blessed $data) ? '' : ref $data;
949 $n_steps += 1 if $ref;
950 last if $ref ne 'REF';
954 my $base = $ref || (defined $data ? 'SCALAR' : 'UNDEF');
956 return $base . ('REF' x $n_steps);
962 my ($self, $data) = @_;
963 my @try = ($self->_refkind($data));
964 push @try, 'SCALAR_or_UNDEF' if $try[0] eq 'SCALAR' || $try[0] eq 'UNDEF';
965 push @try, 'FALLBACK';
969 sub _METHOD_FOR_refkind {
970 my ($self, $meth_prefix, $data) = @_;
971 my $method = first {$_} map {$self->can($meth_prefix."_".$_)}
972 $self->_try_refkind($data)
973 or puke "cannot dispatch on '$meth_prefix' for ".$self->_refkind($data);
978 sub _SWITCH_refkind {
979 my ($self, $data, $dispatch_table) = @_;
981 my $coderef = first {$_} map {$dispatch_table->{$_}}
982 $self->_try_refkind($data)
983 or puke "no dispatch entry for ".$self->_refkind($data);
990 #======================================================================
991 # VALUES, GENERATE, AUTOLOAD
992 #======================================================================
994 # LDNOTE: original code from nwiger, didn't touch code in that section
995 # I feel the AUTOLOAD stuff should not be the default, it should
996 # only be activated on explicit demand by user.
1000 my $data = shift || return;
1001 puke "Argument to ", __PACKAGE__, "->values must be a \\%hash"
1002 unless ref $data eq 'HASH';
1005 foreach my $k ( sort keys %$data ) {
1006 my $v = $data->{$k};
1007 $self->_SWITCH_refkind($v, {
1009 if ($self->{array_datatypes}) { # array datatype
1010 push @all_bind, $self->_bindtype($k, $v);
1012 else { # literal SQL with bind
1013 my ($sql, @bind) = @$v;
1014 $self->_assert_bindval_matches_bindtype(@bind);
1015 push @all_bind, @bind;
1018 ARRAYREFREF => sub { # literal SQL with bind
1019 my ($sql, @bind) = @${$v};
1020 $self->_assert_bindval_matches_bindtype(@bind);
1021 push @all_bind, @bind;
1023 SCALARREF => sub { # literal SQL without bind
1025 SCALAR_or_UNDEF => sub {
1026 push @all_bind, $self->_bindtype($k, $v);
1037 my(@sql, @sqlq, @sqlv);
1041 if ($ref eq 'HASH') {
1042 for my $k (sort keys %$_) {
1045 my $label = $self->_quote($k);
1046 if ($r eq 'ARRAY') {
1047 # literal SQL with bind
1048 my ($sql, @bind) = @$v;
1049 $self->_assert_bindval_matches_bindtype(@bind);
1050 push @sqlq, "$label = $sql";
1052 } elsif ($r eq 'SCALAR') {
1053 # literal SQL without bind
1054 push @sqlq, "$label = $$v";
1056 push @sqlq, "$label = ?";
1057 push @sqlv, $self->_bindtype($k, $v);
1060 push @sql, $self->_sqlcase('set'), join ', ', @sqlq;
1061 } elsif ($ref eq 'ARRAY') {
1062 # unlike insert(), assume these are ONLY the column names, i.e. for SQL
1065 if ($r eq 'ARRAY') { # literal SQL with bind
1066 my ($sql, @bind) = @$v;
1067 $self->_assert_bindval_matches_bindtype(@bind);
1070 } elsif ($r eq 'SCALAR') { # literal SQL without bind
1071 # embedded literal SQL
1078 push @sql, '(' . join(', ', @sqlq) . ')';
1079 } elsif ($ref eq 'SCALAR') {
1083 # strings get case twiddled
1084 push @sql, $self->_sqlcase($_);
1088 my $sql = join ' ', @sql;
1090 # this is pretty tricky
1091 # if ask for an array, return ($stmt, @bind)
1092 # otherwise, s/?/shift @sqlv/ to put it inline
1094 return ($sql, @sqlv);
1096 1 while $sql =~ s/\?/my $d = shift(@sqlv);
1097 ref $d ? $d->[1] : $d/e;
1106 # This allows us to check for a local, then _form, attr
1108 my($name) = $AUTOLOAD =~ /.*::(.+)/;
1109 return $self->generate($name, @_);
1120 SQL::Abstract - Generate SQL from Perl data structures
1126 my $sql = SQL::Abstract->new;
1128 my($stmt, @bind) = $sql->select($table, \@fields, \%where, \@order);
1130 my($stmt, @bind) = $sql->insert($table, \%fieldvals || \@values);
1132 my($stmt, @bind) = $sql->update($table, \%fieldvals, \%where);
1134 my($stmt, @bind) = $sql->delete($table, \%where);
1136 # Then, use these in your DBI statements
1137 my $sth = $dbh->prepare($stmt);
1138 $sth->execute(@bind);
1140 # Just generate the WHERE clause
1141 my($stmt, @bind) = $sql->where(\%where, \@order);
1143 # Return values in the same order, for hashed queries
1144 # See PERFORMANCE section for more details
1145 my @bind = $sql->values(\%fieldvals);
1149 This module was inspired by the excellent L<DBIx::Abstract>.
1150 However, in using that module I found that what I really wanted
1151 to do was generate SQL, but still retain complete control over my
1152 statement handles and use the DBI interface. So, I set out to
1153 create an abstract SQL generation module.
1155 While based on the concepts used by L<DBIx::Abstract>, there are
1156 several important differences, especially when it comes to WHERE
1157 clauses. I have modified the concepts used to make the SQL easier
1158 to generate from Perl data structures and, IMO, more intuitive.
1159 The underlying idea is for this module to do what you mean, based
1160 on the data structures you provide it. The big advantage is that
1161 you don't have to modify your code every time your data changes,
1162 as this module figures it out.
1164 To begin with, an SQL INSERT is as easy as just specifying a hash
1165 of C<key=value> pairs:
1168 name => 'Jimbo Bobson',
1169 phone => '123-456-7890',
1170 address => '42 Sister Lane',
1171 city => 'St. Louis',
1172 state => 'Louisiana',
1175 The SQL can then be generated with this:
1177 my($stmt, @bind) = $sql->insert('people', \%data);
1179 Which would give you something like this:
1181 $stmt = "INSERT INTO people
1182 (address, city, name, phone, state)
1183 VALUES (?, ?, ?, ?, ?)";
1184 @bind = ('42 Sister Lane', 'St. Louis', 'Jimbo Bobson',
1185 '123-456-7890', 'Louisiana');
1187 These are then used directly in your DBI code:
1189 my $sth = $dbh->prepare($stmt);
1190 $sth->execute(@bind);
1192 =head2 Inserting and Updating Arrays
1194 If your database has array types (like for example Postgres),
1195 activate the special option C<< array_datatypes => 1 >>
1196 when creating the C<SQL::Abstract> object.
1197 Then you may use an arrayref to insert and update database array types:
1199 my $sql = SQL::Abstract->new(array_datatypes => 1);
1201 planets => [qw/Mercury Venus Earth Mars/]
1204 my($stmt, @bind) = $sql->insert('solar_system', \%data);
1208 $stmt = "INSERT INTO solar_system (planets) VALUES (?)"
1210 @bind = (['Mercury', 'Venus', 'Earth', 'Mars']);
1213 =head2 Inserting and Updating SQL
1215 In order to apply SQL functions to elements of your C<%data> you may
1216 specify a reference to an arrayref for the given hash value. For example,
1217 if you need to execute the Oracle C<to_date> function on a value, you can
1218 say something like this:
1222 date_entered => \["to_date(?,'MM/DD/YYYY')", "03/02/2003"],
1225 The first value in the array is the actual SQL. Any other values are
1226 optional and would be included in the bind values array. This gives
1229 my($stmt, @bind) = $sql->insert('people', \%data);
1231 $stmt = "INSERT INTO people (name, date_entered)
1232 VALUES (?, to_date(?,'MM/DD/YYYY'))";
1233 @bind = ('Bill', '03/02/2003');
1235 An UPDATE is just as easy, all you change is the name of the function:
1237 my($stmt, @bind) = $sql->update('people', \%data);
1239 Notice that your C<%data> isn't touched; the module will generate
1240 the appropriately quirky SQL for you automatically. Usually you'll
1241 want to specify a WHERE clause for your UPDATE, though, which is
1242 where handling C<%where> hashes comes in handy...
1244 =head2 Complex where statements
1246 This module can generate pretty complicated WHERE statements
1247 easily. For example, simple C<key=value> pairs are taken to mean
1248 equality, and if you want to see if a field is within a set
1249 of values, you can use an arrayref. Let's say we wanted to
1250 SELECT some data based on this criteria:
1253 requestor => 'inna',
1254 worker => ['nwiger', 'rcwe', 'sfz'],
1255 status => { '!=', 'completed' }
1258 my($stmt, @bind) = $sql->select('tickets', '*', \%where);
1260 The above would give you something like this:
1262 $stmt = "SELECT * FROM tickets WHERE
1263 ( requestor = ? ) AND ( status != ? )
1264 AND ( worker = ? OR worker = ? OR worker = ? )";
1265 @bind = ('inna', 'completed', 'nwiger', 'rcwe', 'sfz');
1267 Which you could then use in DBI code like so:
1269 my $sth = $dbh->prepare($stmt);
1270 $sth->execute(@bind);
1276 The functions are simple. There's one for each major SQL operation,
1277 and a constructor you use first. The arguments are specified in a
1278 similar order to each function (table, then fields, then a where
1279 clause) to try and simplify things.
1284 =head2 new(option => 'value')
1286 The C<new()> function takes a list of options and values, and returns
1287 a new B<SQL::Abstract> object which can then be used to generate SQL
1288 through the methods below. The options accepted are:
1294 If set to 'lower', then SQL will be generated in all lowercase. By
1295 default SQL is generated in "textbook" case meaning something like:
1297 SELECT a_field FROM a_table WHERE some_field LIKE '%someval%'
1299 Any setting other than 'lower' is ignored.
1303 This determines what the default comparison operator is. By default
1304 it is C<=>, meaning that a hash like this:
1306 %where = (name => 'nwiger', email => 'nate@wiger.org');
1308 Will generate SQL like this:
1310 WHERE name = 'nwiger' AND email = 'nate@wiger.org'
1312 However, you may want loose comparisons by default, so if you set
1313 C<cmp> to C<like> you would get SQL such as:
1315 WHERE name like 'nwiger' AND email like 'nate@wiger.org'
1317 You can also override the comparsion on an individual basis - see
1318 the huge section on L</"WHERE CLAUSES"> at the bottom.
1320 =item sqltrue, sqlfalse
1322 Expressions for inserting boolean values within SQL statements.
1323 By default these are C<1=1> and C<1=0>. They are used
1324 by the special operators C<-in> and C<-not_in> for generating
1325 correct SQL even when the argument is an empty array (see below).
1329 This determines the default logical operator for multiple WHERE
1330 statements in arrays or hashes. If absent, the default logic is "or"
1331 for arrays, and "and" for hashes. This means that a WHERE
1335 event_date => {'>=', '2/13/99'},
1336 event_date => {'<=', '4/24/03'},
1339 will generate SQL like this:
1341 WHERE event_date >= '2/13/99' OR event_date <= '4/24/03'
1343 This is probably not what you want given this query, though (look
1344 at the dates). To change the "OR" to an "AND", simply specify:
1346 my $sql = SQL::Abstract->new(logic => 'and');
1348 Which will change the above C<WHERE> to:
1350 WHERE event_date >= '2/13/99' AND event_date <= '4/24/03'
1352 The logic can also be changed locally by inserting
1353 a modifier in front of an arrayref :
1355 @where = (-and => [event_date => {'>=', '2/13/99'},
1356 event_date => {'<=', '4/24/03'} ]);
1358 See the L</"WHERE CLAUSES"> section for explanations.
1362 This will automatically convert comparisons using the specified SQL
1363 function for both column and value. This is mostly used with an argument
1364 of C<upper> or C<lower>, so that the SQL will have the effect of
1365 case-insensitive "searches". For example, this:
1367 $sql = SQL::Abstract->new(convert => 'upper');
1368 %where = (keywords => 'MaKe iT CAse inSeNSItive');
1370 Will turn out the following SQL:
1372 WHERE upper(keywords) like upper('MaKe iT CAse inSeNSItive')
1374 The conversion can be C<upper()>, C<lower()>, or any other SQL function
1375 that can be applied symmetrically to fields (actually B<SQL::Abstract> does
1376 not validate this option; it will just pass through what you specify verbatim).
1380 This is a kludge because many databases suck. For example, you can't
1381 just bind values using DBI's C<execute()> for Oracle C<CLOB> or C<BLOB> fields.
1382 Instead, you have to use C<bind_param()>:
1384 $sth->bind_param(1, 'reg data');
1385 $sth->bind_param(2, $lots, {ora_type => ORA_CLOB});
1387 The problem is, B<SQL::Abstract> will normally just return a C<@bind> array,
1388 which loses track of which field each slot refers to. Fear not.
1390 If you specify C<bindtype> in new, you can determine how C<@bind> is returned.
1391 Currently, you can specify either C<normal> (default) or C<columns>. If you
1392 specify C<columns>, you will get an array that looks like this:
1394 my $sql = SQL::Abstract->new(bindtype => 'columns');
1395 my($stmt, @bind) = $sql->insert(...);
1398 [ 'column1', 'value1' ],
1399 [ 'column2', 'value2' ],
1400 [ 'column3', 'value3' ],
1403 You can then iterate through this manually, using DBI's C<bind_param()>.
1405 $sth->prepare($stmt);
1408 my($col, $data) = @$_;
1409 if ($col eq 'details' || $col eq 'comments') {
1410 $sth->bind_param($i, $data, {ora_type => ORA_CLOB});
1411 } elsif ($col eq 'image') {
1412 $sth->bind_param($i, $data, {ora_type => ORA_BLOB});
1414 $sth->bind_param($i, $data);
1418 $sth->execute; # execute without @bind now
1420 Now, why would you still use B<SQL::Abstract> if you have to do this crap?
1421 Basically, the advantage is still that you don't have to care which fields
1422 are or are not included. You could wrap that above C<for> loop in a simple
1423 sub called C<bind_fields()> or something and reuse it repeatedly. You still
1424 get a layer of abstraction over manual SQL specification.
1426 Note that if you set L</bindtype> to C<columns>, the C<\[$sql, @bind]>
1427 construct (see L</Literal SQL with placeholders and bind values (subqueries)>)
1428 will expect the bind values in this format.
1432 This is the character that a table or column name will be quoted
1433 with. By default this is an empty string, but you could set it to
1434 the character C<`>, to generate SQL like this:
1436 SELECT `a_field` FROM `a_table` WHERE `some_field` LIKE '%someval%'
1438 Alternatively, you can supply an array ref of two items, the first being the left
1439 hand quote character, and the second the right hand quote character. For
1440 example, you could supply C<['[',']']> for SQL Server 2000 compliant quotes
1441 that generates SQL like this:
1443 SELECT [a_field] FROM [a_table] WHERE [some_field] LIKE '%someval%'
1445 Quoting is useful if you have tables or columns names that are reserved
1446 words in your database's SQL dialect.
1450 This is the character that separates a table and column name. It is
1451 necessary to specify this when the C<quote_char> option is selected,
1452 so that tables and column names can be individually quoted like this:
1454 SELECT `table`.`one_field` FROM `table` WHERE `table`.`other_field` = 1
1456 =item array_datatypes
1458 When this option is true, arrayrefs in INSERT or UPDATE are
1459 interpreted as array datatypes and are passed directly
1461 When this option is false, arrayrefs are interpreted
1462 as literal SQL, just like refs to arrayrefs
1463 (but this behavior is for backwards compatibility; when writing
1464 new queries, use the "reference to arrayref" syntax
1470 Takes a reference to a list of "special operators"
1471 to extend the syntax understood by L<SQL::Abstract>.
1472 See section L</"SPECIAL OPERATORS"> for details.
1478 =head2 insert($table, \@values || \%fieldvals)
1480 This is the simplest function. You simply give it a table name
1481 and either an arrayref of values or hashref of field/value pairs.
1482 It returns an SQL INSERT statement and a list of bind values.
1483 See the sections on L</"Inserting and Updating Arrays"> and
1484 L</"Inserting and Updating SQL"> for information on how to insert
1485 with those data types.
1487 =head2 update($table, \%fieldvals, \%where)
1489 This takes a table, hashref of field/value pairs, and an optional
1490 hashref L<WHERE clause|/WHERE CLAUSES>. It returns an SQL UPDATE function and a list
1492 See the sections on L</"Inserting and Updating Arrays"> and
1493 L</"Inserting and Updating SQL"> for information on how to insert
1494 with those data types.
1496 =head2 select($source, $fields, $where, $order)
1498 This returns a SQL SELECT statement and associated list of bind values, as
1499 specified by the arguments :
1505 Specification of the 'FROM' part of the statement.
1506 The argument can be either a plain scalar (interpreted as a table
1507 name, will be quoted), or an arrayref (interpreted as a list
1508 of table names, joined by commas, quoted), or a scalarref
1509 (literal table name, not quoted), or a ref to an arrayref
1510 (list of literal table names, joined by commas, not quoted).
1514 Specification of the list of fields to retrieve from
1516 The argument can be either an arrayref (interpreted as a list
1517 of field names, will be joined by commas and quoted), or a
1518 plain scalar (literal SQL, not quoted).
1519 Please observe that this API is not as flexible as for
1520 the first argument C<$table>, for backwards compatibility reasons.
1524 Optional argument to specify the WHERE part of the query.
1525 The argument is most often a hashref, but can also be
1526 an arrayref or plain scalar --
1527 see section L<WHERE clause|/"WHERE CLAUSES"> for details.
1531 Optional argument to specify the ORDER BY part of the query.
1532 The argument can be a scalar, a hashref or an arrayref
1533 -- see section L<ORDER BY clause|/"ORDER BY CLAUSES">
1539 =head2 delete($table, \%where)
1541 This takes a table name and optional hashref L<WHERE clause|/WHERE CLAUSES>.
1542 It returns an SQL DELETE statement and list of bind values.
1544 =head2 where(\%where, \@order)
1546 This is used to generate just the WHERE clause. For example,
1547 if you have an arbitrary data structure and know what the
1548 rest of your SQL is going to look like, but want an easy way
1549 to produce a WHERE clause, use this. It returns an SQL WHERE
1550 clause and list of bind values.
1553 =head2 values(\%data)
1555 This just returns the values from the hash C<%data>, in the same
1556 order that would be returned from any of the other above queries.
1557 Using this allows you to markedly speed up your queries if you
1558 are affecting lots of rows. See below under the L</"PERFORMANCE"> section.
1560 =head2 generate($any, 'number', $of, \@data, $struct, \%types)
1562 Warning: This is an experimental method and subject to change.
1564 This returns arbitrarily generated SQL. It's a really basic shortcut.
1565 It will return two different things, depending on return context:
1567 my($stmt, @bind) = $sql->generate('create table', \$table, \@fields);
1568 my $stmt_and_val = $sql->generate('create table', \$table, \@fields);
1570 These would return the following:
1572 # First calling form
1573 $stmt = "CREATE TABLE test (?, ?)";
1574 @bind = (field1, field2);
1576 # Second calling form
1577 $stmt_and_val = "CREATE TABLE test (field1, field2)";
1579 Depending on what you're trying to do, it's up to you to choose the correct
1580 format. In this example, the second form is what you would want.
1584 $sql->generate('alter session', { nls_date_format => 'MM/YY' });
1588 ALTER SESSION SET nls_date_format = 'MM/YY'
1590 You get the idea. Strings get their case twiddled, but everything
1591 else remains verbatim.
1596 =head1 WHERE CLAUSES
1600 This module uses a variation on the idea from L<DBIx::Abstract>. It
1601 is B<NOT>, repeat I<not> 100% compatible. B<The main logic of this
1602 module is that things in arrays are OR'ed, and things in hashes
1605 The easiest way to explain is to show lots of examples. After
1606 each C<%where> hash shown, it is assumed you used:
1608 my($stmt, @bind) = $sql->where(\%where);
1610 However, note that the C<%where> hash can be used directly in any
1611 of the other functions as well, as described above.
1613 =head2 Key-value pairs
1615 So, let's get started. To begin, a simple hash:
1619 status => 'completed'
1622 Is converted to SQL C<key = val> statements:
1624 $stmt = "WHERE user = ? AND status = ?";
1625 @bind = ('nwiger', 'completed');
1627 One common thing I end up doing is having a list of values that
1628 a field can be in. To do this, simply specify a list inside of
1633 status => ['assigned', 'in-progress', 'pending'];
1636 This simple code will create the following:
1638 $stmt = "WHERE user = ? AND ( status = ? OR status = ? OR status = ? )";
1639 @bind = ('nwiger', 'assigned', 'in-progress', 'pending');
1641 A field associated to an empty arrayref will be considered a
1642 logical false and will generate 0=1.
1644 =head2 Specific comparison operators
1646 If you want to specify a different type of operator for your comparison,
1647 you can use a hashref for a given column:
1651 status => { '!=', 'completed' }
1654 Which would generate:
1656 $stmt = "WHERE user = ? AND status != ?";
1657 @bind = ('nwiger', 'completed');
1659 To test against multiple values, just enclose the values in an arrayref:
1661 status => { '=', ['assigned', 'in-progress', 'pending'] };
1663 Which would give you:
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'} ]
1753 =head2 Special operators : IN, BETWEEN, etc.
1755 You can also use the hashref format to compare a list of fields using the
1756 C<IN> comparison operator, by specifying the list as an arrayref:
1759 status => 'completed',
1760 reportid => { -in => [567, 2335, 2] }
1763 Which would generate:
1765 $stmt = "WHERE status = ? AND reportid IN (?,?,?)";
1766 @bind = ('completed', '567', '2335', '2');
1768 The reverse operator C<-not_in> generates SQL C<NOT IN> and is used in
1771 If the argument to C<-in> is an empty array, 'sqlfalse' is generated
1772 (by default : C<1=0>). Similarly, C<< -not_in => [] >> generates
1773 'sqltrue' (by default : C<1=1>).
1777 Another pair of operators is C<-between> and C<-not_between>,
1778 used with an arrayref of two values:
1782 completion_date => {
1783 -not_between => ['2002-10-01', '2003-02-06']
1789 WHERE user = ? AND completion_date NOT BETWEEN ( ? AND ? )
1791 These are the two builtin "special operators"; but the
1792 list can be expanded : see section L</"SPECIAL OPERATORS"> below.
1794 =head2 Nested conditions, -and/-or prefixes
1796 So far, we've seen how multiple conditions are joined with a top-level
1797 C<AND>. We can change this by putting the different conditions we want in
1798 hashes and then putting those hashes in an array. For example:
1803 status => { -like => ['pending%', 'dispatched'] },
1807 status => 'unassigned',
1811 This data structure would create the following:
1813 $stmt = "WHERE ( user = ? AND ( status LIKE ? OR status LIKE ? ) )
1814 OR ( user = ? AND status = ? ) )";
1815 @bind = ('nwiger', 'pending', 'dispatched', 'robot', 'unassigned');
1818 There is also a special C<-nest>
1819 operator which adds an additional set of parens, to create a subquery.
1820 For example, to get something like this:
1822 $stmt = "WHERE user = ? AND ( workhrs > ? OR geo = ? )";
1823 @bind = ('nwiger', '20', 'ASIA');
1829 -nest => [ workhrs => {'>', 20}, geo => 'ASIA' ],
1833 Finally, clauses in hashrefs or arrayrefs can be
1834 prefixed with an C<-and> or C<-or> to change the logic
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 = ? ) ) )
1854 =head2 Algebraic inconsistency, for historical reasons
1856 C<Important note>: when connecting several conditions, the C<-and->|C<-or>
1857 operator goes C<outside> of the nested structure; whereas when connecting
1858 several constraints on one column, the C<-and> operator goes
1859 C<inside> the arrayref. Here is an example combining both features :
1862 -and => [a => 1, b => 2],
1863 -or => [c => 3, d => 4],
1864 e => [-and => {-like => 'foo%'}, {-like => '%bar'} ]
1869 WHERE ( ( ( a = ? AND b = ? )
1870 OR ( c = ? OR d = ? )
1871 OR ( e LIKE ? AND e LIKE ? ) ) )
1873 This difference in syntax is unfortunate but must be preserved for
1874 historical reasons. So be careful : the two examples below would
1875 seem algebraically equivalent, but they are not
1877 {col => [-and => {-like => 'foo%'}, {-like => '%bar'}]}
1878 # yields : WHERE ( ( col LIKE ? AND col LIKE ? ) )
1880 [-and => {col => {-like => 'foo%'}, {col => {-like => '%bar'}}]]
1881 # yields : WHERE ( ( col LIKE ? OR col LIKE ? ) )
1886 Finally, sometimes only literal SQL will do. If you want to include
1887 literal SQL verbatim, you can specify it as a scalar reference, namely:
1889 my $inn = 'is Not Null';
1891 priority => { '<', 2 },
1897 $stmt = "WHERE priority < ? AND requestor is Not Null";
1900 Note that in this example, you only get one bind parameter back, since
1901 the verbatim SQL is passed as part of the statement.
1903 Of course, just to prove a point, the above can also be accomplished
1907 priority => { '<', 2 },
1908 requestor => { '!=', undef },
1914 Conditions on boolean columns can be expressed in the
1915 same way, passing a reference to an empty string :
1918 priority => { '<', 2 },
1924 $stmt = "WHERE priority < ? AND is_ready";
1928 =head2 Literal SQL with placeholders and bind values (subqueries)
1930 If the literal SQL to be inserted has placeholders and bind values,
1931 use a reference to an arrayref (yes this is a double reference --
1932 not so common, but perfectly legal Perl). For example, to find a date
1933 in Postgres you can use something like this:
1936 date_column => \[q/= date '2008-09-30' - ?::integer/, 10/]
1941 $stmt = "WHERE ( date_column = date '2008-09-30' - ?::integer )"
1944 Note that you must pass the bind values in the same format as they are returned
1945 by L</where>. That means that if you set L</bindtype> to C<columns>, you must
1946 provide the bind values in the C<< [ column_meta => value ] >> format, where
1947 C<column_meta> is an opaque scalar value; most commonly the column name, but
1948 you can use any scalar value (including references and blessed references),
1949 L<SQL::Abstract> will simply pass it through intact. So if C<bindtype> is set
1950 to C<columns> the above example will look like:
1953 date_column => \[q/= date '2008-09-30' - ?::integer/, [ dummy => 10 ]/]
1956 Literal SQL is especially useful for nesting parenthesized clauses in the
1957 main SQL query. Here is a first example :
1959 my ($sub_stmt, @sub_bind) = ("SELECT c1 FROM t1 WHERE c2 < ? AND c3 LIKE ?",
1963 bar => \["IN ($sub_stmt)" => @sub_bind],
1968 $stmt = "WHERE (foo = ? AND bar IN (SELECT c1 FROM t1
1969 WHERE c2 < ? AND c3 LIKE ?))";
1970 @bind = (1234, 100, "foo%");
1972 Other subquery operators, like for example C<"E<gt> ALL"> or C<"NOT IN">,
1973 are expressed in the same way. Of course the C<$sub_stmt> and
1974 its associated bind values can be generated through a former call
1977 my ($sub_stmt, @sub_bind)
1978 = $sql->select("t1", "c1", {c2 => {"<" => 100},
1979 c3 => {-like => "foo%"}});
1982 bar => \["> ALL ($sub_stmt)" => @sub_bind],
1985 In the examples above, the subquery was used as an operator on a column;
1986 but the same principle also applies for a clause within the main C<%where>
1987 hash, like an EXISTS subquery :
1989 my ($sub_stmt, @sub_bind)
1990 = $sql->select("t1", "*", {c1 => 1, c2 => \"> t0.c0"});
1993 -nest => \["EXISTS ($sub_stmt)" => @sub_bind],
1998 $stmt = "WHERE (foo = ? AND EXISTS (SELECT * FROM t1
1999 WHERE c1 = ? AND c2 > t0.c0))";
2003 Observe that the condition on C<c2> in the subquery refers to
2004 column C<t0.c0> of the main query : this is I<not> a bind
2005 value, so we have to express it through a scalar ref.
2006 Writing C<< c2 => {">" => "t0.c0"} >> would have generated
2007 C<< c2 > ? >> with bind value C<"t0.c0"> ... not exactly
2008 what we wanted here.
2010 Another use of the subquery technique is when some SQL clauses need
2011 parentheses, as it often occurs with some proprietary SQL extensions
2012 like for example fulltext expressions, geospatial expressions,
2013 NATIVE clauses, etc. Here is an example of a fulltext query in MySQL :
2016 -nest => \["MATCH (col1, col2) AGAINST (?)" => qw/apples/]
2019 Finally, here is an example where a subquery is used
2020 for expressing unary negation:
2022 my ($sub_stmt, @sub_bind)
2023 = $sql->where({age => [{"<" => 10}, {">" => 20}]});
2024 $sub_stmt =~ s/^ where //i; # don't want "WHERE" in the subclause
2026 lname => {like => '%son%'},
2027 -nest => \["NOT ($sub_stmt)" => @sub_bind],
2032 $stmt = "lname LIKE ? AND NOT ( age < ? OR age > ? )"
2033 @bind = ('%son%', 10, 20)
2039 These pages could go on for a while, since the nesting of the data
2040 structures this module can handle are pretty much unlimited (the
2041 module implements the C<WHERE> expansion as a recursive function
2042 internally). Your best bet is to "play around" with the module a
2043 little to see how the data structures behave, and choose the best
2044 format for your data based on that.
2046 And of course, all the values above will probably be replaced with
2047 variables gotten from forms or the command line. After all, if you
2048 knew everything ahead of time, you wouldn't have to worry about
2049 dynamically-generating SQL and could just hardwire it into your
2055 =head1 ORDER BY CLAUSES
2057 Some functions take an order by clause. This can either be a scalar (just a
2058 column name,) a hash of C<< { -desc => 'col' } >> or C<< { -asc => 'col' } >>,
2059 or an array of either of the two previous forms. Examples:
2061 Given | Will Generate
2062 ----------------------------------------------------------
2063 \'colA DESC' | ORDER BY colA DESC
2064 'colA' | ORDER BY colA
2065 [qw/colA colB/] | ORDER BY colA, colB
2066 {-asc => 'colA'} | ORDER BY colA ASC
2067 {-desc => 'colB'} | ORDER BY colB DESC
2069 {-asc => 'colA'}, | ORDER BY colA ASC, colB DESC
2072 [colA => {-asc => 'colB'}] | ORDER BY colA, colB ASC
2073 ==========================================================
2077 =head1 SPECIAL OPERATORS
2079 my $sqlmaker = SQL::Abstract->new(special_ops => [
2082 my ($self, $field, $op, $arg) = @_;
2088 A "special operator" is a SQL syntactic clause that can be
2089 applied to a field, instead of a usual binary operator.
2092 WHERE field IN (?, ?, ?)
2093 WHERE field BETWEEN ? AND ?
2094 WHERE MATCH(field) AGAINST (?, ?)
2096 Special operators IN and BETWEEN are fairly standard and therefore
2097 are builtin within C<SQL::Abstract>. For other operators,
2098 like the MATCH .. AGAINST example above which is
2099 specific to MySQL, you can write your own operator handlers :
2100 supply a C<special_ops> argument to the C<new> method.
2101 That argument takes an arrayref of operator definitions;
2102 each operator definition is a hashref with two entries
2108 the regular expression to match the operator
2112 coderef that will be called when meeting that operator
2113 in the input tree. The coderef will be called with
2114 arguments C<< ($self, $field, $op, $arg) >>, and
2115 should return a C<< ($sql, @bind) >> structure.
2119 For example, here is an implementation
2120 of the MATCH .. AGAINST syntax for MySQL
2122 my $sqlmaker = SQL::Abstract->new(special_ops => [
2124 # special op for MySql MATCH (field) AGAINST(word1, word2, ...)
2125 {regex => qr/^match$/i,
2127 my ($self, $field, $op, $arg) = @_;
2128 $arg = [$arg] if not ref $arg;
2129 my $label = $self->_quote($field);
2130 my ($placeholder) = $self->_convert('?');
2131 my $placeholders = join ", ", (($placeholder) x @$arg);
2132 my $sql = $self->_sqlcase('match') . " ($label) "
2133 . $self->_sqlcase('against') . " ($placeholders) ";
2134 my @bind = $self->_bindtype($field, @$arg);
2135 return ($sql, @bind);
2144 Thanks to some benchmarking by Mark Stosberg, it turns out that
2145 this module is many orders of magnitude faster than using C<DBIx::Abstract>.
2146 I must admit this wasn't an intentional design issue, but it's a
2147 byproduct of the fact that you get to control your C<DBI> handles
2150 To maximize performance, use a code snippet like the following:
2152 # prepare a statement handle using the first row
2153 # and then reuse it for the rest of the rows
2155 for my $href (@array_of_hashrefs) {
2156 $stmt ||= $sql->insert('table', $href);
2157 $sth ||= $dbh->prepare($stmt);
2158 $sth->execute($sql->values($href));
2161 The reason this works is because the keys in your C<$href> are sorted
2162 internally by B<SQL::Abstract>. Thus, as long as your data retains
2163 the same structure, you only have to generate the SQL the first time
2164 around. On subsequent queries, simply use the C<values> function provided
2165 by this module to return your values in the correct order.
2170 If you use my C<CGI::FormBuilder> module at all, you'll hopefully
2171 really like this part (I do, at least). Building up a complex query
2172 can be as simple as the following:
2176 use CGI::FormBuilder;
2179 my $form = CGI::FormBuilder->new(...);
2180 my $sql = SQL::Abstract->new;
2182 if ($form->submitted) {
2183 my $field = $form->field;
2184 my $id = delete $field->{id};
2185 my($stmt, @bind) = $sql->update('table', $field, {id => $id});
2188 Of course, you would still have to connect using C<DBI> to run the
2189 query, but the point is that if you make your form look like your
2190 table, the actual query script can be extremely simplistic.
2192 If you're B<REALLY> lazy (I am), check out C<HTML::QuickTable> for
2193 a fast interface to returning and formatting data. I frequently
2194 use these three modules together to write complex database query
2195 apps in under 50 lines.
2200 Version 1.50 was a major internal refactoring of C<SQL::Abstract>.
2201 Great care has been taken to preserve the I<published> behavior
2202 documented in previous versions in the 1.* family; however,
2203 some features that were previously undocumented, or behaved
2204 differently from the documentation, had to be changed in order
2205 to clarify the semantics. Hence, client code that was relying
2206 on some dark areas of C<SQL::Abstract> v1.*
2207 B<might behave differently> in v1.50.
2209 The main changes are :
2215 support for literal SQL through the C<< \ [$sql, bind] >> syntax.
2219 support for the { operator => \"..." } construct (to embed literal SQL)
2223 support for the { operator => \["...", @bind] } construct (to embed literal SQL with bind values)
2227 optional support for L<array datatypes|/"Inserting and Updating Arrays">
2231 defensive programming : check arguments
2235 fixed bug with global logic, which was previously implemented
2236 through global variables yielding side-effects. Prior versions would
2237 interpret C<< [ {cond1, cond2}, [cond3, cond4] ] >>
2238 as C<< "(cond1 AND cond2) OR (cond3 AND cond4)" >>.
2239 Now this is interpreted
2240 as C<< "(cond1 AND cond2) OR (cond3 OR cond4)" >>.
2245 fixed semantics of _bindtype on array args
2249 dropped the C<_anoncopy> of the %where tree. No longer necessary,
2250 we just avoid shifting arrays within that tree.
2254 dropped the C<_modlogic> function
2260 =head1 ACKNOWLEDGEMENTS
2262 There are a number of individuals that have really helped out with
2263 this module. Unfortunately, most of them submitted bugs via CPAN
2264 so I have no idea who they are! But the people I do know are:
2266 Ash Berlin (order_by hash term support)
2267 Matt Trout (DBIx::Class support)
2268 Mark Stosberg (benchmarking)
2269 Chas Owens (initial "IN" operator support)
2270 Philip Collins (per-field SQL functions)
2271 Eric Kolve (hashref "AND" support)
2272 Mike Fragassi (enhancements to "BETWEEN" and "LIKE")
2273 Dan Kubb (support for "quote_char" and "name_sep")
2274 Guillermo Roditi (patch to cleanup "IN" and "BETWEEN", fix and tests for _order_by)
2275 Laurent Dami (internal refactoring, multiple -nest, extensible list of special operators, literal SQL)
2276 Norbert Buchmuller (support for literal SQL in hashpair, misc. fixes & tests)
2282 L<DBIx::Class>, L<DBIx::Abstract>, L<CGI::FormBuilder>, L<HTML::QuickTable>.
2286 Copyright (c) 2001-2007 Nathan Wiger <nwiger@cpan.org>. All Rights Reserved.
2288 This module is actively maintained by Matt Trout <mst@shadowcatsystems.co.uk>
2290 For support, your best bet is to try the C<DBIx::Class> users mailing list.
2291 While not an official support venue, C<DBIx::Class> makes heavy use of
2292 C<SQL::Abstract>, and as such list members there are very familiar with
2293 how to create queries.
2295 This module is free software; you may copy this under the terms of
2296 the GNU General Public License, or the Artistic License, copies of
2297 which should have accompanied your Perl kit.