1 package SQL::Abstract; # see doc at end of file
3 # LDNOTE : this code is heavy refactoring from original SQLA.
4 # Several design decisions will need discussion during
5 # the test / diffusion / acceptance phase; those are marked with flag
6 # 'LDNOTE' (note by laurent.dami AT free.fr)
11 use List::Util qw/first/;
12 use Scalar::Util qw/blessed/;
14 #======================================================================
16 #======================================================================
18 our $VERSION = '1.50';
20 # This would confuse some packagers
21 #$VERSION = eval $VERSION; # numify for warning-free dev releases
25 # special operators (-in, -between). May be extended/overridden by user.
26 # See section WHERE: BUILTIN SPECIAL OPERATORS below for implementation
27 my @BUILTIN_SPECIAL_OPS = (
28 {regex => qr/^(not )?between$/i, handler => \&_where_field_BETWEEN},
29 {regex => qr/^(not )?in$/i, handler => \&_where_field_IN},
32 #======================================================================
33 # DEBUGGING AND ERROR REPORTING
34 #======================================================================
37 return unless $_[0]->{debug}; shift; # a little faster
38 my $func = (caller(1))[3];
39 warn "[$func] ", @_, "\n";
43 my($func) = (caller(1))[3];
44 carp "[$func] Warning: ", @_;
48 my($func) = (caller(1))[3];
49 croak "[$func] Fatal: ", @_;
53 #======================================================================
55 #======================================================================
59 my $class = ref($self) || $self;
60 my %opt = (ref $_[0] eq 'HASH') ? %{$_[0]} : @_;
62 # choose our case by keeping an option around
63 delete $opt{case} if $opt{case} && $opt{case} ne 'lower';
65 # default logic for interpreting arrayrefs
66 $opt{logic} = uc $opt{logic} || 'OR';
68 # how to return bind vars
69 # LDNOTE: changed nwiger code : why this 'delete' ??
70 # $opt{bindtype} ||= delete($opt{bind_type}) || 'normal';
71 $opt{bindtype} ||= 'normal';
73 # default comparison is "=", but can be overridden
76 # try to recognize which are the 'equality' and 'unequality' ops
77 # (temporary quickfix, should go through a more seasoned API)
78 $opt{equality_op} = qr/^(\Q$opt{cmp}\E|is|(is\s+)?like)$/i;
79 $opt{inequality_op} = qr/^(!=|<>|(is\s+)?not(\s+like)?)$/i;
82 $opt{sqltrue} ||= '1=1';
83 $opt{sqlfalse} ||= '0=1';
86 $opt{special_ops} ||= [];
87 push @{$opt{special_ops}}, @BUILTIN_SPECIAL_OPS;
89 return bless \%opt, $class;
94 #======================================================================
96 #======================================================================
100 my $table = $self->_table(shift);
101 my $data = shift || return;
103 my $method = $self->_METHOD_FOR_refkind("_insert", $data);
104 my ($sql, @bind) = $self->$method($data);
105 $sql = join " ", $self->_sqlcase('insert into'), $table, $sql;
106 return wantarray ? ($sql, @bind) : $sql;
109 sub _insert_HASHREF { # explicit list of fields and then values
110 my ($self, $data) = @_;
112 my @fields = sort keys %$data;
114 my ($sql, @bind) = $self->_insert_values($data);
117 $_ = $self->_quote($_) foreach @fields;
118 $sql = "( ".join(", ", @fields).") ".$sql;
120 return ($sql, @bind);
123 sub _insert_ARRAYREF { # just generate values(?,?) part (no list of fields)
124 my ($self, $data) = @_;
126 # no names (arrayref) so can't generate bindtype
127 $self->{bindtype} ne 'columns'
128 or belch "can't do 'columns' bindtype when called with arrayref";
130 # fold the list of values into a hash of column name - value pairs
131 # (where the column names are artificially generated, and their
132 # lexicographical ordering keep the ordering of the original list)
133 my $i = "a"; # incremented values will be in lexicographical order
134 my $data_in_hash = { map { ($i++ => $_) } @$data };
136 return $self->_insert_values($data_in_hash);
139 sub _insert_ARRAYREFREF { # literal SQL with bind
140 my ($self, $data) = @_;
142 my ($sql, @bind) = @${$data};
143 $self->_assert_bindval_matches_bindtype(@bind);
145 return ($sql, @bind);
149 sub _insert_SCALARREF { # literal SQL without bind
150 my ($self, $data) = @_;
156 my ($self, $data) = @_;
158 my (@values, @all_bind);
159 foreach my $column (sort keys %$data) {
160 my $v = $data->{$column};
162 $self->_SWITCH_refkind($v, {
165 if ($self->{array_datatypes}) { # if array datatype are activated
167 push @all_bind, $self->_bindtype($column, $v);
169 else { # else literal SQL with bind
170 my ($sql, @bind) = @$v;
171 $self->_assert_bindval_matches_bindtype(@bind);
173 push @all_bind, @bind;
177 ARRAYREFREF => sub { # literal SQL with bind
178 my ($sql, @bind) = @${$v};
179 $self->_assert_bindval_matches_bindtype(@bind);
181 push @all_bind, @bind;
184 # THINK : anything useful to do with a HASHREF ?
185 HASHREF => sub { # (nothing, but old SQLA passed it through)
186 #TODO in SQLA >= 2.0 it will die instead
187 belch "HASH ref as bind value in insert is not supported";
189 push @all_bind, $self->_bindtype($column, $v);
192 SCALARREF => sub { # literal SQL without bind
196 SCALAR_or_UNDEF => sub {
198 push @all_bind, $self->_bindtype($column, $v);
205 my $sql = $self->_sqlcase('values')." ( ".join(", ", @values)." )";
206 return ($sql, @all_bind);
211 #======================================================================
213 #======================================================================
218 my $table = $self->_table(shift);
219 my $data = shift || return;
222 # first build the 'SET' part of the sql statement
223 my (@set, @all_bind);
224 puke "Unsupported data type specified to \$sql->update"
225 unless ref $data eq 'HASH';
227 for my $k (sort keys %$data) {
230 my $label = $self->_quote($k);
232 $self->_SWITCH_refkind($v, {
234 if ($self->{array_datatypes}) { # array datatype
235 push @set, "$label = ?";
236 push @all_bind, $self->_bindtype($k, $v);
238 else { # literal SQL with bind
239 my ($sql, @bind) = @$v;
240 $self->_assert_bindval_matches_bindtype(@bind);
241 push @set, "$label = $sql";
242 push @all_bind, @bind;
245 ARRAYREFREF => sub { # literal SQL with bind
246 my ($sql, @bind) = @${$v};
247 $self->_assert_bindval_matches_bindtype(@bind);
248 push @set, "$label = $sql";
249 push @all_bind, @bind;
251 SCALARREF => sub { # literal SQL without bind
252 push @set, "$label = $$v";
254 SCALAR_or_UNDEF => sub {
255 push @set, "$label = ?";
256 push @all_bind, $self->_bindtype($k, $v);
262 my $sql = $self->_sqlcase('update') . " $table " . $self->_sqlcase('set ')
266 my($where_sql, @where_bind) = $self->where($where);
268 push @all_bind, @where_bind;
271 return wantarray ? ($sql, @all_bind) : $sql;
277 #======================================================================
279 #======================================================================
284 my $table = $self->_table(shift);
285 my $fields = shift || '*';
289 my($where_sql, @bind) = $self->where($where, $order);
291 my $f = (ref $fields eq 'ARRAY') ? join ', ', map { $self->_quote($_) } @$fields
293 my $sql = join(' ', $self->_sqlcase('select'), $f,
294 $self->_sqlcase('from'), $table)
297 return wantarray ? ($sql, @bind) : $sql;
300 #======================================================================
302 #======================================================================
307 my $table = $self->_table(shift);
311 my($where_sql, @bind) = $self->where($where);
312 my $sql = $self->_sqlcase('delete from') . " $table" . $where_sql;
314 return wantarray ? ($sql, @bind) : $sql;
318 #======================================================================
320 #======================================================================
324 # Finally, a separate routine just to handle WHERE clauses
326 my ($self, $where, $order) = @_;
329 my ($sql, @bind) = $self->_recurse_where($where);
330 $sql = $sql ? $self->_sqlcase(' where ') . "( $sql )" : '';
334 $sql .= $self->_order_by($order);
337 return wantarray ? ($sql, @bind) : $sql;
342 my ($self, $where, $logic) = @_;
344 # dispatch on appropriate method according to refkind of $where
345 my $method = $self->_METHOD_FOR_refkind("_where", $where);
348 my ($sql, @bind) = $self->$method($where, $logic);
350 # DBIx::Class directly calls _recurse_where in scalar context, so
351 # we must implement it, even if not in the official API
352 return wantarray ? ($sql, @bind) : $sql;
357 #======================================================================
358 # WHERE: top-level ARRAYREF
359 #======================================================================
362 sub _where_ARRAYREF {
363 my ($self, $where, $logic) = @_;
365 $logic = uc($logic || $self->{logic});
366 $logic eq 'AND' or $logic eq 'OR' or puke "unknown logic: $logic";
368 my @clauses = @$where;
370 my (@sql_clauses, @all_bind);
371 # need to use while() so can shift() for pairs
372 while (my $el = shift @clauses) {
374 # switch according to kind of $el and get corresponding ($sql, @bind)
375 my ($sql, @bind) = $self->_SWITCH_refkind($el, {
377 # skip empty elements, otherwise get invalid trailing AND stuff
378 ARRAYREF => sub {$self->_recurse_where($el) if @$el},
380 ARRAYREFREF => sub { @{${$el}} if @{${$el}}},
382 HASHREF => sub {$self->_recurse_where($el, 'and') if %$el},
383 # LDNOTE : previous SQLA code for hashrefs was creating a dirty
384 # side-effect: the first hashref within an array would change
385 # the global logic to 'AND'. So [ {cond1, cond2}, [cond3, cond4] ]
386 # was interpreted as "(cond1 AND cond2) OR (cond3 AND cond4)",
387 # whereas it should be "(cond1 AND cond2) OR (cond3 OR cond4)".
389 SCALARREF => sub { ($$el); },
391 SCALAR => sub {# top-level arrayref with scalars, recurse in pairs
392 $self->_recurse_where({$el => shift(@clauses)})},
394 UNDEF => sub {puke "not supported : UNDEF in arrayref" },
398 push @sql_clauses, $sql;
399 push @all_bind, @bind;
403 return $self->_join_sql_clauses($logic, \@sql_clauses, \@all_bind);
406 #======================================================================
407 # WHERE: top-level ARRAYREFREF
408 #======================================================================
410 sub _where_ARRAYREFREF {
411 my ($self, $where) = @_;
412 my ($sql, @bind) = @{${$where}};
414 return ($sql, @bind);
417 #======================================================================
418 # WHERE: top-level HASHREF
419 #======================================================================
422 my ($self, $where) = @_;
423 my (@sql_clauses, @all_bind);
425 # LDNOTE : don't really know why we need to sort keys
426 for my $k (sort keys %$where) {
427 my $v = $where->{$k};
429 # ($k => $v) is either a special op or a regular hashpair
430 my ($sql, @bind) = ($k =~ /^-(.+)/) ? $self->_where_op_in_hash($1, $v)
432 my $method = $self->_METHOD_FOR_refkind("_where_hashpair", $v);
433 $self->$method($k, $v);
436 push @sql_clauses, $sql;
437 push @all_bind, @bind;
440 return $self->_join_sql_clauses('and', \@sql_clauses, \@all_bind);
444 sub _where_op_in_hash {
445 my ($self, $op_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>.
1327 This determines the default logical operator for multiple WHERE
1328 statements in arrays or hashes. If absent, the default logic is "or"
1329 for arrays, and "and" for hashes. This means 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 a modifier in front of an arrayref :
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 A field associated to an empty arrayref will be considered a
1640 logical false and will generate 0=1.
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 = ? OR status = ? OR status = ?"
1666 The hashref can also contain multiple pairs, in which case it is expanded
1667 into an C<AND> of its elements:
1671 status => { '!=', 'completed', -not_like => 'pending%' }
1674 # Or more dynamically, like from a form
1675 $where{user} = 'nwiger';
1676 $where{status}{'!='} = 'completed';
1677 $where{status}{'-not_like'} = 'pending%';
1679 # Both generate this
1680 $stmt = "WHERE user = ? AND status != ? AND status NOT LIKE ?";
1681 @bind = ('nwiger', 'completed', 'pending%');
1684 To get an OR instead, you can combine it with the arrayref idea:
1688 priority => [ {'=', 2}, {'!=', 1} ]
1691 Which would generate:
1693 $stmt = "WHERE user = ? AND priority = ? OR priority != ?";
1694 @bind = ('nwiger', '2', '1');
1696 If you want to include literal SQL (with or without bind values), just use a
1697 scalar reference or array reference as the value:
1700 date_entered => { '>' => \["to_date(?, 'MM/DD/YYYY')", "11/26/2008"] },
1701 date_expires => { '<' => \"now()" }
1704 Which would generate:
1706 $stmt = "WHERE date_entered > "to_date(?, 'MM/DD/YYYY') AND date_expires < now()";
1707 @bind = ('11/26/2008');
1710 =head2 Logic and nesting operators
1712 In the example above,
1713 there is a subtle trap if you want to say something like
1714 this (notice the C<AND>):
1716 WHERE priority != ? AND priority != ?
1718 Because, in Perl you I<can't> do this:
1720 priority => { '!=', 2, '!=', 1 }
1722 As the second C<!=> key will obliterate the first. The solution
1723 is to use the special C<-modifier> form inside an arrayref:
1725 priority => [ -and => {'!=', 2},
1729 Normally, these would be joined by C<OR>, but the modifier tells it
1730 to use C<AND> instead. (Hint: You can use this in conjunction with the
1731 C<logic> option to C<new()> in order to change the way your queries
1732 work by default.) B<Important:> Note that the C<-modifier> goes
1733 B<INSIDE> the arrayref, as an extra first element. This will
1734 B<NOT> do what you think it might:
1736 priority => -and => [{'!=', 2}, {'!=', 1}] # WRONG!
1738 Here is a quick list of equivalencies, since there is some overlap:
1741 status => {'!=', 'completed', 'not like', 'pending%' }
1742 status => [ -and => {'!=', 'completed'}, {'not like', 'pending%'}]
1745 status => {'=', ['assigned', 'in-progress']}
1746 status => [ -or => {'=', 'assigned'}, {'=', 'in-progress'}]
1747 status => [ {'=', 'assigned'}, {'=', 'in-progress'} ]
1751 =head2 Special operators : IN, BETWEEN, etc.
1753 You can also use the hashref format to compare a list of fields using the
1754 C<IN> comparison operator, by specifying the list as an arrayref:
1757 status => 'completed',
1758 reportid => { -in => [567, 2335, 2] }
1761 Which would generate:
1763 $stmt = "WHERE status = ? AND reportid IN (?,?,?)";
1764 @bind = ('completed', '567', '2335', '2');
1766 The reverse operator C<-not_in> generates SQL C<NOT IN> and is used in
1769 Another pair of operators is C<-between> and C<-not_between>,
1770 used with an arrayref of two values:
1774 completion_date => {
1775 -not_between => ['2002-10-01', '2003-02-06']
1781 WHERE user = ? AND completion_date NOT BETWEEN ( ? AND ? )
1783 These are the two builtin "special operators"; but the
1784 list can be expanded : see section L</"SPECIAL OPERATORS"> below.
1786 =head2 Nested conditions, -and/-or prefixes
1788 So far, we've seen how multiple conditions are joined with a top-level
1789 C<AND>. We can change this by putting the different conditions we want in
1790 hashes and then putting those hashes in an array. For example:
1795 status => { -like => ['pending%', 'dispatched'] },
1799 status => 'unassigned',
1803 This data structure would create the following:
1805 $stmt = "WHERE ( user = ? AND ( status LIKE ? OR status LIKE ? ) )
1806 OR ( user = ? AND status = ? ) )";
1807 @bind = ('nwiger', 'pending', 'dispatched', 'robot', 'unassigned');
1810 There is also a special C<-nest>
1811 operator which adds an additional set of parens, to create a subquery.
1812 For example, to get something like this:
1814 $stmt = "WHERE user = ? AND ( workhrs > ? OR geo = ? )";
1815 @bind = ('nwiger', '20', 'ASIA');
1821 -nest => [ workhrs => {'>', 20}, geo => 'ASIA' ],
1825 Finally, clauses in hashrefs or arrayrefs can be
1826 prefixed with an C<-and> or C<-or> to change the logic
1833 -and => [workhrs => {'>', 20}, geo => 'ASIA' ],
1834 -and => [workhrs => {'<', 50}, geo => 'EURO' ]
1841 WHERE ( user = ? AND
1842 ( ( workhrs > ? AND geo = ? )
1843 OR ( workhrs < ? AND geo = ? ) ) )
1846 =head2 Algebraic inconsistency, for historical reasons
1848 C<Important note>: when connecting several conditions, the C<-and->|C<-or>
1849 operator goes C<outside> of the nested structure; whereas when connecting
1850 several constraints on one column, the C<-and> operator goes
1851 C<inside> the arrayref. Here is an example combining both features :
1854 -and => [a => 1, b => 2],
1855 -or => [c => 3, d => 4],
1856 e => [-and => {-like => 'foo%'}, {-like => '%bar'} ]
1861 WHERE ( ( ( a = ? AND b = ? )
1862 OR ( c = ? OR d = ? )
1863 OR ( e LIKE ? AND e LIKE ? ) ) )
1865 This difference in syntax is unfortunate but must be preserved for
1866 historical reasons. So be careful : the two examples below would
1867 seem algebraically equivalent, but they are not
1869 {col => [-and => {-like => 'foo%'}, {-like => '%bar'}]}
1870 # yields : WHERE ( ( col LIKE ? AND col LIKE ? ) )
1872 [-and => {col => {-like => 'foo%'}, {col => {-like => '%bar'}}]]
1873 # yields : WHERE ( ( col LIKE ? OR col LIKE ? ) )
1878 Finally, sometimes only literal SQL will do. If you want to include
1879 literal SQL verbatim, you can specify it as a scalar reference, namely:
1881 my $inn = 'is Not Null';
1883 priority => { '<', 2 },
1889 $stmt = "WHERE priority < ? AND requestor is Not Null";
1892 Note that in this example, you only get one bind parameter back, since
1893 the verbatim SQL is passed as part of the statement.
1895 Of course, just to prove a point, the above can also be accomplished
1899 priority => { '<', 2 },
1900 requestor => { '!=', undef },
1906 Conditions on boolean columns can be expressed in the
1907 same way, passing a reference to an empty string :
1910 priority => { '<', 2 },
1916 $stmt = "WHERE priority < ? AND is_ready";
1920 =head2 Literal SQL with placeholders and bind values (subqueries)
1922 If the literal SQL to be inserted has placeholders and bind values,
1923 use a reference to an arrayref (yes this is a double reference --
1924 not so common, but perfectly legal Perl). For example, to find a date
1925 in Postgres you can use something like this:
1928 date_column => \[q/= date '2008-09-30' - ?::integer/, 10/]
1933 $stmt = "WHERE ( date_column = date '2008-09-30' - ?::integer )"
1936 Note that you must pass the bind values in the same format as they are returned
1937 by L</where>. That means that if you set L</bindtype> to C<columns>, you must
1938 provide the bind values in the C<< [ column_meta => value ] >> format, where
1939 C<column_meta> is an opaque scalar value; most commonly the column name, but
1940 you can use any scalar value (including references and blessed references),
1941 L<SQL::Abstract> will simply pass it through intact. So if C<bindtype> is set
1942 to C<columns> the above example will look like:
1945 date_column => \[q/= date '2008-09-30' - ?::integer/, [ dummy => 10 ]/]
1948 Literal SQL is especially useful for nesting parenthesized clauses in the
1949 main SQL query. Here is a first example :
1951 my ($sub_stmt, @sub_bind) = ("SELECT c1 FROM t1 WHERE c2 < ? AND c3 LIKE ?",
1955 bar => \["IN ($sub_stmt)" => @sub_bind],
1960 $stmt = "WHERE (foo = ? AND bar IN (SELECT c1 FROM t1
1961 WHERE c2 < ? AND c3 LIKE ?))";
1962 @bind = (1234, 100, "foo%");
1964 Other subquery operators, like for example C<"E<gt> ALL"> or C<"NOT IN">,
1965 are expressed in the same way. Of course the C<$sub_stmt> and
1966 its associated bind values can be generated through a former call
1969 my ($sub_stmt, @sub_bind)
1970 = $sql->select("t1", "c1", {c2 => {"<" => 100},
1971 c3 => {-like => "foo%"}});
1974 bar => \["> ALL ($sub_stmt)" => @sub_bind],
1977 In the examples above, the subquery was used as an operator on a column;
1978 but the same principle also applies for a clause within the main C<%where>
1979 hash, like an EXISTS subquery :
1981 my ($sub_stmt, @sub_bind)
1982 = $sql->select("t1", "*", {c1 => 1, c2 => \"> t0.c0"});
1985 -nest => \["EXISTS ($sub_stmt)" => @sub_bind],
1990 $stmt = "WHERE (foo = ? AND EXISTS (SELECT * FROM t1
1991 WHERE c1 = ? AND c2 > t0.c0))";
1995 Observe that the condition on C<c2> in the subquery refers to
1996 column C<t0.c0> of the main query : this is I<not> a bind
1997 value, so we have to express it through a scalar ref.
1998 Writing C<< c2 => {">" => "t0.c0"} >> would have generated
1999 C<< c2 > ? >> with bind value C<"t0.c0"> ... not exactly
2000 what we wanted here.
2002 Another use of the subquery technique is when some SQL clauses need
2003 parentheses, as it often occurs with some proprietary SQL extensions
2004 like for example fulltext expressions, geospatial expressions,
2005 NATIVE clauses, etc. Here is an example of a fulltext query in MySQL :
2008 -nest => \["MATCH (col1, col2) AGAINST (?)" => qw/apples/]
2011 Finally, here is an example where a subquery is used
2012 for expressing unary negation:
2014 my ($sub_stmt, @sub_bind)
2015 = $sql->where({age => [{"<" => 10}, {">" => 20}]});
2016 $sub_stmt =~ s/^ where //i; # don't want "WHERE" in the subclause
2018 lname => {like => '%son%'},
2019 -nest => \["NOT ($sub_stmt)" => @sub_bind],
2024 $stmt = "lname LIKE ? AND NOT ( age < ? OR age > ? )"
2025 @bind = ('%son%', 10, 20)
2031 These pages could go on for a while, since the nesting of the data
2032 structures this module can handle are pretty much unlimited (the
2033 module implements the C<WHERE> expansion as a recursive function
2034 internally). Your best bet is to "play around" with the module a
2035 little to see how the data structures behave, and choose the best
2036 format for your data based on that.
2038 And of course, all the values above will probably be replaced with
2039 variables gotten from forms or the command line. After all, if you
2040 knew everything ahead of time, you wouldn't have to worry about
2041 dynamically-generating SQL and could just hardwire it into your
2047 =head1 ORDER BY CLAUSES
2049 Some functions take an order by clause. This can either be a scalar (just a
2050 column name,) a hash of C<< { -desc => 'col' } >> or C<< { -asc => 'col' } >>,
2051 or an array of either of the two previous forms. Examples:
2053 Given | Will Generate
2054 ----------------------------------------------------------
2055 \'colA DESC' | ORDER BY colA DESC
2056 'colA' | ORDER BY colA
2057 [qw/colA colB/] | ORDER BY colA, colB
2058 {-asc => 'colA'} | ORDER BY colA ASC
2059 {-desc => 'colB'} | ORDER BY colB DESC
2061 {-asc => 'colA'}, | ORDER BY colA ASC, colB DESC
2064 [colA => {-asc => 'colB'}] | ORDER BY colA, colB ASC
2065 ==========================================================
2069 =head1 SPECIAL OPERATORS
2071 my $sqlmaker = SQL::Abstract->new(special_ops => [
2074 my ($self, $field, $op, $arg) = @_;
2080 A "special operator" is a SQL syntactic clause that can be
2081 applied to a field, instead of a usual binary operator.
2084 WHERE field IN (?, ?, ?)
2085 WHERE field BETWEEN ? AND ?
2086 WHERE MATCH(field) AGAINST (?, ?)
2088 Special operators IN and BETWEEN are fairly standard and therefore
2089 are builtin within C<SQL::Abstract>. For other operators,
2090 like the MATCH .. AGAINST example above which is
2091 specific to MySQL, you can write your own operator handlers :
2092 supply a C<special_ops> argument to the C<new> method.
2093 That argument takes an arrayref of operator definitions;
2094 each operator definition is a hashref with two entries
2100 the regular expression to match the operator
2104 coderef that will be called when meeting that operator
2105 in the input tree. The coderef will be called with
2106 arguments C<< ($self, $field, $op, $arg) >>, and
2107 should return a C<< ($sql, @bind) >> structure.
2111 For example, here is an implementation
2112 of the MATCH .. AGAINST syntax for MySQL
2114 my $sqlmaker = SQL::Abstract->new(special_ops => [
2116 # special op for MySql MATCH (field) AGAINST(word1, word2, ...)
2117 {regex => qr/^match$/i,
2119 my ($self, $field, $op, $arg) = @_;
2120 $arg = [$arg] if not ref $arg;
2121 my $label = $self->_quote($field);
2122 my ($placeholder) = $self->_convert('?');
2123 my $placeholders = join ", ", (($placeholder) x @$arg);
2124 my $sql = $self->_sqlcase('match') . " ($label) "
2125 . $self->_sqlcase('against') . " ($placeholders) ";
2126 my @bind = $self->_bindtype($field, @$arg);
2127 return ($sql, @bind);
2136 Thanks to some benchmarking by Mark Stosberg, it turns out that
2137 this module is many orders of magnitude faster than using C<DBIx::Abstract>.
2138 I must admit this wasn't an intentional design issue, but it's a
2139 byproduct of the fact that you get to control your C<DBI> handles
2142 To maximize performance, use a code snippet like the following:
2144 # prepare a statement handle using the first row
2145 # and then reuse it for the rest of the rows
2147 for my $href (@array_of_hashrefs) {
2148 $stmt ||= $sql->insert('table', $href);
2149 $sth ||= $dbh->prepare($stmt);
2150 $sth->execute($sql->values($href));
2153 The reason this works is because the keys in your C<$href> are sorted
2154 internally by B<SQL::Abstract>. Thus, as long as your data retains
2155 the same structure, you only have to generate the SQL the first time
2156 around. On subsequent queries, simply use the C<values> function provided
2157 by this module to return your values in the correct order.
2162 If you use my C<CGI::FormBuilder> module at all, you'll hopefully
2163 really like this part (I do, at least). Building up a complex query
2164 can be as simple as the following:
2168 use CGI::FormBuilder;
2171 my $form = CGI::FormBuilder->new(...);
2172 my $sql = SQL::Abstract->new;
2174 if ($form->submitted) {
2175 my $field = $form->field;
2176 my $id = delete $field->{id};
2177 my($stmt, @bind) = $sql->update('table', $field, {id => $id});
2180 Of course, you would still have to connect using C<DBI> to run the
2181 query, but the point is that if you make your form look like your
2182 table, the actual query script can be extremely simplistic.
2184 If you're B<REALLY> lazy (I am), check out C<HTML::QuickTable> for
2185 a fast interface to returning and formatting data. I frequently
2186 use these three modules together to write complex database query
2187 apps in under 50 lines.
2192 Version 1.50 was a major internal refactoring of C<SQL::Abstract>.
2193 Great care has been taken to preserve the I<published> behavior
2194 documented in previous versions in the 1.* family; however,
2195 some features that were previously undocumented, or behaved
2196 differently from the documentation, had to be changed in order
2197 to clarify the semantics. Hence, client code that was relying
2198 on some dark areas of C<SQL::Abstract> v1.*
2199 B<might behave differently> in v1.50.
2201 The main changes are :
2207 support for literal SQL through the C<< \ [$sql, bind] >> syntax.
2211 support for the { operator => \"..." } construct (to embed literal SQL)
2215 support for the { operator => \["...", @bind] } construct (to embed literal SQL with bind values)
2219 added official support for -nest1, -nest2 or -nest_1, -nest_2, ...
2220 (undocumented in previous versions)
2224 optional support for L<array datatypes|/"Inserting and Updating Arrays">
2228 defensive programming : check arguments
2232 fixed bug with global logic, which was previously implemented
2233 through global variables yielding side-effects. Prior versions would
2234 interpret C<< [ {cond1, cond2}, [cond3, cond4] ] >>
2235 as C<< "(cond1 AND cond2) OR (cond3 AND cond4)" >>.
2236 Now this is interpreted
2237 as C<< "(cond1 AND cond2) OR (cond3 OR cond4)" >>.
2242 fixed semantics of _bindtype on array args
2246 dropped the C<_anoncopy> of the %where tree. No longer necessary,
2247 we just avoid shifting arrays within that tree.
2251 dropped the C<_modlogic> function
2257 =head1 ACKNOWLEDGEMENTS
2259 There are a number of individuals that have really helped out with
2260 this module. Unfortunately, most of them submitted bugs via CPAN
2261 so I have no idea who they are! But the people I do know are:
2263 Ash Berlin (order_by hash term support)
2264 Matt Trout (DBIx::Class support)
2265 Mark Stosberg (benchmarking)
2266 Chas Owens (initial "IN" operator support)
2267 Philip Collins (per-field SQL functions)
2268 Eric Kolve (hashref "AND" support)
2269 Mike Fragassi (enhancements to "BETWEEN" and "LIKE")
2270 Dan Kubb (support for "quote_char" and "name_sep")
2271 Guillermo Roditi (patch to cleanup "IN" and "BETWEEN", fix and tests for _order_by)
2272 Laurent Dami (internal refactoring, multiple -nest, extensible list of special operators, literal SQL)
2273 Norbert Buchmuller (support for literal SQL in hashpair, misc. fixes & tests)
2279 L<DBIx::Class>, L<DBIx::Abstract>, L<CGI::FormBuilder>, L<HTML::QuickTable>.
2283 Copyright (c) 2001-2007 Nathan Wiger <nwiger@cpan.org>. All Rights Reserved.
2285 This module is actively maintained by Matt Trout <mst@shadowcatsystems.co.uk>
2287 For support, your best bet is to try the C<DBIx::Class> users mailing list.
2288 While not an official support venue, C<DBIx::Class> makes heavy use of
2289 C<SQL::Abstract>, and as such list members there are very familiar with
2290 how to create queries.
2292 This module is free software; you may copy this under the terms of
2293 the GNU General Public License, or the Artistic License, copies of
2294 which should have accompanied your Perl kit.