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, $logic) = @_;
534 my ($all_sql, @all_bind);
536 for my $op (sort keys %$v) {
539 # put the operator in canonical form
540 $op =~ s/^-//; # remove initial dash
541 $op =~ tr/_/ /; # underscores become spaces
542 $op =~ s/^\s+//; # no initial space
543 $op =~ s/\s+$//; # no final space
544 $op =~ s/\s+/ /; # multiple spaces become one
548 # CASE: special operators like -in or -between
549 my $special_op = first {$op =~ $_->{regex}} @{$self->{special_ops}};
551 ($sql, @bind) = $special_op->{handler}->($self, $k, $op, $val);
554 $self->_SWITCH_refkind($val, {
556 ARRAYREF => sub { # CASE: col => {op => \@vals}
557 ($sql, @bind) = $self->_where_field_op_ARRAYREF($k, $op, $val);
560 SCALARREF => sub { # CASE: col => {op => \$scalar} (literal SQL without bind)
561 $sql = join ' ', $self->_convert($self->_quote($k)),
562 $self->_sqlcase($op),
566 ARRAYREFREF => sub { # CASE: col => {op => \[$sql, @bind]} (literal SQL with bind)
567 my ($sub_sql, @sub_bind) = @$$val;
568 $self->_assert_bindval_matches_bindtype(@sub_bind);
569 $sql = join ' ', $self->_convert($self->_quote($k)),
570 $self->_sqlcase($op),
576 ($sql, @bind) = $self->_where_hashpair_HASHREF($k, $val, $op);
579 UNDEF => sub { # CASE: col => {op => undef} : sql "IS (NOT)? NULL"
580 my $is = ($op =~ $self->{equality_op}) ? 'is' :
581 ($op =~ $self->{inequality_op}) ? 'is not' :
582 puke "unexpected operator '$op' with undef operand";
583 $sql = $self->_quote($k) . $self->_sqlcase(" $is null");
586 FALLBACK => sub { # CASE: col => {op => $scalar}
587 $sql = join ' ', $self->_convert($self->_quote($k)),
588 $self->_sqlcase($op),
589 $self->_convert('?');
590 @bind = $self->_bindtype($k, $val);
595 ($all_sql) = (defined $all_sql and $all_sql) ? $self->_join_sql_clauses($logic, [$all_sql, $sql], []) : $sql;
596 push @all_bind, @bind;
598 return ($all_sql, @all_bind);
603 sub _where_field_op_ARRAYREF {
604 my ($self, $k, $op, $vals) = @_;
607 $self->_debug("ARRAY($vals) means multiple elements: [ @$vals ]");
609 # LDNOTE : had planned to change the distribution logic when
610 # $op =~ $self->{inequality_op}, because of Morgan laws :
611 # with {field => {'!=' => [22, 33]}}, it would be ridiculous to generate
612 # WHERE field != 22 OR field != 33 : the user probably means
613 # WHERE field != 22 AND field != 33.
614 # To do this, replace the line below by :
615 # my $logic = ($op =~ $self->{inequality_op}) ? 'AND' : 'OR';
616 # return $self->_recurse_where([map { {$k => {$op, $_}} } @$vals], $logic);
618 # distribute $op over each member of @$vals
619 return $self->_recurse_where([map { {$k => {$op, $_}} } @$vals]);
622 # try to DWIM on equality operators
623 # LDNOTE : not 100% sure this is the correct thing to do ...
624 return ($self->{sqlfalse}) if $op =~ $self->{equality_op};
625 return ($self->{sqltrue}) if $op =~ $self->{inequality_op};
628 puke "operator '$op' applied on an empty array (field '$k')";
633 sub _where_hashpair_SCALARREF {
634 my ($self, $k, $v) = @_;
635 $self->_debug("SCALAR($k) means literal SQL: $$v");
636 my $sql = $self->_quote($k) . " " . $$v;
640 # literal SQL with bind
641 sub _where_hashpair_ARRAYREFREF {
642 my ($self, $k, $v) = @_;
643 $self->_debug("REF($k) means literal SQL: @${$v}");
644 my ($sql, @bind) = @${$v};
645 $self->_assert_bindval_matches_bindtype(@bind);
646 $sql = $self->_quote($k) . " " . $sql;
647 return ($sql, @bind );
650 # literal SQL without bind
651 sub _where_hashpair_SCALAR {
652 my ($self, $k, $v) = @_;
653 $self->_debug("NOREF($k) means simple key=val: $k $self->{cmp} $v");
654 my $sql = join ' ', $self->_convert($self->_quote($k)),
655 $self->_sqlcase($self->{cmp}),
656 $self->_convert('?');
657 my @bind = $self->_bindtype($k, $v);
658 return ( $sql, @bind);
662 sub _where_hashpair_UNDEF {
663 my ($self, $k, $v) = @_;
664 $self->_debug("UNDEF($k) means IS NULL");
665 my $sql = $self->_quote($k) . $self->_sqlcase(' is null');
669 #======================================================================
670 # WHERE: TOP-LEVEL OTHERS (SCALARREF, SCALAR, UNDEF)
671 #======================================================================
674 sub _where_SCALARREF {
675 my ($self, $where) = @_;
678 $self->_debug("SCALAR(*top) means literal SQL: $$where");
684 my ($self, $where) = @_;
687 $self->_debug("NOREF(*top) means literal SQL: $where");
698 #======================================================================
699 # WHERE: BUILTIN SPECIAL OPERATORS (-in, -between)
700 #======================================================================
703 sub _where_field_BETWEEN {
704 my ($self, $k, $op, $vals) = @_;
706 ref $vals eq 'ARRAY' && @$vals == 2
707 or puke "special op 'between' requires an arrayref of two values";
709 my ($label) = $self->_convert($self->_quote($k));
710 my ($placeholder) = $self->_convert('?');
711 my $and = $self->_sqlcase('and');
712 $op = $self->_sqlcase($op);
714 my $sql = "( $label $op $placeholder $and $placeholder )";
715 my @bind = $self->_bindtype($k, @$vals);
720 sub _where_field_IN {
721 my ($self, $k, $op, $vals) = @_;
723 # backwards compatibility : if scalar, force into an arrayref
724 $vals = [$vals] if defined $vals && ! ref $vals;
726 my ($label) = $self->_convert($self->_quote($k));
727 my ($placeholder) = $self->_convert('?');
728 $op = $self->_sqlcase($op);
730 my ($sql, @bind) = $self->_SWITCH_refkind($vals, {
731 ARRAYREF => sub { # list of choices
732 if (@$vals) { # nonempty list
733 my $placeholders = join ", ", (($placeholder) x @$vals);
734 my $sql = "$label $op ( $placeholders )";
735 my @bind = $self->_bindtype($k, @$vals);
737 return ($sql, @bind);
739 else { # empty list : some databases won't understand "IN ()", so DWIM
740 my $sql = ($op =~ /\bnot\b/i) ? $self->{sqltrue} : $self->{sqlfalse};
745 ARRAYREFREF => sub { # literal SQL with bind
746 my ($sql, @bind) = @$$vals;
747 $self->_assert_bindval_matches_bindtype(@bind);
748 return ("$label $op ( $sql )", @bind);
752 puke "special op 'in' requires an arrayref (or arrayref-ref)";
756 return ($sql, @bind);
764 #======================================================================
766 #======================================================================
769 my ($self, $arg) = @_;
771 # construct list of ordering instructions
772 my @order = $self->_SWITCH_refkind($arg, {
775 map {$self->_SWITCH_refkind($_, {
776 SCALAR => sub {$self->_quote($_)},
778 SCALARREF => sub {$$_}, # literal SQL, no quoting
779 HASHREF => sub {$self->_order_by_hash($_)}
783 SCALAR => sub {$self->_quote($arg)},
785 SCALARREF => sub {$$arg}, # literal SQL, no quoting
786 HASHREF => sub {$self->_order_by_hash($arg)},
791 my $order = join ', ', @order;
792 return $order ? $self->_sqlcase(' order by')." $order" : '';
797 my ($self, $hash) = @_;
799 # get first pair in hash
800 my ($key, $val) = each %$hash;
802 # check if one pair was found and no other pair in hash
803 $key && !(each %$hash)
804 or puke "hash passed to _order_by must have exactly one key (-desc or -asc)";
806 my ($order) = ($key =~ /^-(desc|asc)/i)
807 or puke "invalid key in _order_by hash : $key";
809 return $self->_quote($val) ." ". $self->_sqlcase($order);
814 #======================================================================
815 # DATASOURCE (FOR NOW, JUST PLAIN TABLE OR LIST OF TABLES)
816 #======================================================================
821 $self->_SWITCH_refkind($from, {
822 ARRAYREF => sub {join ', ', map { $self->_quote($_) } @$from;},
823 SCALAR => sub {$self->_quote($from)},
824 SCALARREF => sub {$$from},
825 ARRAYREFREF => sub {join ', ', @$from;},
830 #======================================================================
832 #======================================================================
838 $label or puke "can't quote an empty label";
840 # left and right quote characters
841 my ($ql, $qr, @other) = $self->_SWITCH_refkind($self->{quote_char}, {
842 SCALAR => sub {($self->{quote_char}, $self->{quote_char})},
843 ARRAYREF => sub {@{$self->{quote_char}}},
847 or puke "quote_char must be an arrayref of 2 values";
849 # no quoting if no quoting chars
850 $ql or return $label;
852 # no quoting for literal SQL
853 return $$label if ref($label) eq 'SCALAR';
855 # separate table / column (if applicable)
856 my $sep = $self->{name_sep} || '';
857 my @to_quote = $sep ? split /\Q$sep\E/, $label : ($label);
859 # do the quoting, except for "*" or for `table`.*
860 my @quoted = map { $_ eq '*' ? $_: $ql.$_.$qr} @to_quote;
862 # reassemble and return.
863 return join $sep, @quoted;
867 # Conversion, if applicable
869 my ($self, $arg) = @_;
871 # LDNOTE : modified the previous implementation below because
872 # it was not consistent : the first "return" is always an array,
873 # the second "return" is context-dependent. Anyway, _convert
874 # seems always used with just a single argument, so make it a
876 # return @_ unless $self->{convert};
877 # my $conv = $self->_sqlcase($self->{convert});
878 # my @ret = map { $conv.'('.$_.')' } @_;
879 # return wantarray ? @ret : $ret[0];
880 if ($self->{convert}) {
881 my $conv = $self->_sqlcase($self->{convert});
882 $arg = $conv.'('.$arg.')';
890 my($col, @vals) = @_;
892 #LDNOTE : changed original implementation below because it did not make
893 # sense when bindtype eq 'columns' and @vals > 1.
894 # return $self->{bindtype} eq 'columns' ? [ $col, @vals ] : @vals;
896 return $self->{bindtype} eq 'columns' ? map {[$col, $_]} @vals : @vals;
899 # Dies if any element of @bind is not in [colname => value] format
900 # if bindtype is 'columns'.
901 sub _assert_bindval_matches_bindtype {
902 my ($self, @bind) = @_;
904 if ($self->{bindtype} eq 'columns') {
905 foreach my $val (@bind) {
906 if (!defined $val || ref($val) ne 'ARRAY' || @$val != 2) {
907 die "bindtype 'columns' selected, you need to pass: [column_name => bind_value]"
913 sub _join_sql_clauses {
914 my ($self, $logic, $clauses_aref, $bind_aref) = @_;
916 if (@$clauses_aref > 1) {
917 my $join = " " . $self->_sqlcase($logic) . " ";
918 my $sql = '( ' . join($join, @$clauses_aref) . ' )';
919 return ($sql, @$bind_aref);
921 elsif (@$clauses_aref) {
922 return ($clauses_aref->[0], @$bind_aref); # no parentheses
925 return (); # if no SQL, ignore @$bind_aref
930 # Fix SQL case, if so requested
934 # LDNOTE: if $self->{case} is true, then it contains 'lower', so we
935 # don't touch the argument ... crooked logic, but let's not change it!
936 return $self->{case} ? $_[0] : uc($_[0]);
940 #======================================================================
941 # DISPATCHING FROM REFKIND
942 #======================================================================
945 my ($self, $data) = @_;
951 # blessed objects are treated like scalars
952 $ref = (blessed $data) ? '' : ref $data;
953 $n_steps += 1 if $ref;
954 last if $ref ne 'REF';
958 my $base = $ref || (defined $data ? 'SCALAR' : 'UNDEF');
960 return $base . ('REF' x $n_steps);
966 my ($self, $data) = @_;
967 my @try = ($self->_refkind($data));
968 push @try, 'SCALAR_or_UNDEF' if $try[0] eq 'SCALAR' || $try[0] eq 'UNDEF';
969 push @try, 'FALLBACK';
973 sub _METHOD_FOR_refkind {
974 my ($self, $meth_prefix, $data) = @_;
975 my $method = first {$_} map {$self->can($meth_prefix."_".$_)}
976 $self->_try_refkind($data)
977 or puke "cannot dispatch on '$meth_prefix' for ".$self->_refkind($data);
982 sub _SWITCH_refkind {
983 my ($self, $data, $dispatch_table) = @_;
985 my $coderef = first {$_} map {$dispatch_table->{$_}}
986 $self->_try_refkind($data)
987 or puke "no dispatch entry for ".$self->_refkind($data);
994 #======================================================================
995 # VALUES, GENERATE, AUTOLOAD
996 #======================================================================
998 # LDNOTE: original code from nwiger, didn't touch code in that section
999 # I feel the AUTOLOAD stuff should not be the default, it should
1000 # only be activated on explicit demand by user.
1004 my $data = shift || return;
1005 puke "Argument to ", __PACKAGE__, "->values must be a \\%hash"
1006 unless ref $data eq 'HASH';
1009 foreach my $k ( sort keys %$data ) {
1010 my $v = $data->{$k};
1011 $self->_SWITCH_refkind($v, {
1013 if ($self->{array_datatypes}) { # array datatype
1014 push @all_bind, $self->_bindtype($k, $v);
1016 else { # literal SQL with bind
1017 my ($sql, @bind) = @$v;
1018 $self->_assert_bindval_matches_bindtype(@bind);
1019 push @all_bind, @bind;
1022 ARRAYREFREF => sub { # literal SQL with bind
1023 my ($sql, @bind) = @${$v};
1024 $self->_assert_bindval_matches_bindtype(@bind);
1025 push @all_bind, @bind;
1027 SCALARREF => sub { # literal SQL without bind
1029 SCALAR_or_UNDEF => sub {
1030 push @all_bind, $self->_bindtype($k, $v);
1041 my(@sql, @sqlq, @sqlv);
1045 if ($ref eq 'HASH') {
1046 for my $k (sort keys %$_) {
1049 my $label = $self->_quote($k);
1050 if ($r eq 'ARRAY') {
1051 # literal SQL with bind
1052 my ($sql, @bind) = @$v;
1053 $self->_assert_bindval_matches_bindtype(@bind);
1054 push @sqlq, "$label = $sql";
1056 } elsif ($r eq 'SCALAR') {
1057 # literal SQL without bind
1058 push @sqlq, "$label = $$v";
1060 push @sqlq, "$label = ?";
1061 push @sqlv, $self->_bindtype($k, $v);
1064 push @sql, $self->_sqlcase('set'), join ', ', @sqlq;
1065 } elsif ($ref eq 'ARRAY') {
1066 # unlike insert(), assume these are ONLY the column names, i.e. for SQL
1069 if ($r eq 'ARRAY') { # literal SQL with bind
1070 my ($sql, @bind) = @$v;
1071 $self->_assert_bindval_matches_bindtype(@bind);
1074 } elsif ($r eq 'SCALAR') { # literal SQL without bind
1075 # embedded literal SQL
1082 push @sql, '(' . join(', ', @sqlq) . ')';
1083 } elsif ($ref eq 'SCALAR') {
1087 # strings get case twiddled
1088 push @sql, $self->_sqlcase($_);
1092 my $sql = join ' ', @sql;
1094 # this is pretty tricky
1095 # if ask for an array, return ($stmt, @bind)
1096 # otherwise, s/?/shift @sqlv/ to put it inline
1098 return ($sql, @sqlv);
1100 1 while $sql =~ s/\?/my $d = shift(@sqlv);
1101 ref $d ? $d->[1] : $d/e;
1110 # This allows us to check for a local, then _form, attr
1112 my($name) = $AUTOLOAD =~ /.*::(.+)/;
1113 return $self->generate($name, @_);
1124 SQL::Abstract - Generate SQL from Perl data structures
1130 my $sql = SQL::Abstract->new;
1132 my($stmt, @bind) = $sql->select($table, \@fields, \%where, \@order);
1134 my($stmt, @bind) = $sql->insert($table, \%fieldvals || \@values);
1136 my($stmt, @bind) = $sql->update($table, \%fieldvals, \%where);
1138 my($stmt, @bind) = $sql->delete($table, \%where);
1140 # Then, use these in your DBI statements
1141 my $sth = $dbh->prepare($stmt);
1142 $sth->execute(@bind);
1144 # Just generate the WHERE clause
1145 my($stmt, @bind) = $sql->where(\%where, \@order);
1147 # Return values in the same order, for hashed queries
1148 # See PERFORMANCE section for more details
1149 my @bind = $sql->values(\%fieldvals);
1153 This module was inspired by the excellent L<DBIx::Abstract>.
1154 However, in using that module I found that what I really wanted
1155 to do was generate SQL, but still retain complete control over my
1156 statement handles and use the DBI interface. So, I set out to
1157 create an abstract SQL generation module.
1159 While based on the concepts used by L<DBIx::Abstract>, there are
1160 several important differences, especially when it comes to WHERE
1161 clauses. I have modified the concepts used to make the SQL easier
1162 to generate from Perl data structures and, IMO, more intuitive.
1163 The underlying idea is for this module to do what you mean, based
1164 on the data structures you provide it. The big advantage is that
1165 you don't have to modify your code every time your data changes,
1166 as this module figures it out.
1168 To begin with, an SQL INSERT is as easy as just specifying a hash
1169 of C<key=value> pairs:
1172 name => 'Jimbo Bobson',
1173 phone => '123-456-7890',
1174 address => '42 Sister Lane',
1175 city => 'St. Louis',
1176 state => 'Louisiana',
1179 The SQL can then be generated with this:
1181 my($stmt, @bind) = $sql->insert('people', \%data);
1183 Which would give you something like this:
1185 $stmt = "INSERT INTO people
1186 (address, city, name, phone, state)
1187 VALUES (?, ?, ?, ?, ?)";
1188 @bind = ('42 Sister Lane', 'St. Louis', 'Jimbo Bobson',
1189 '123-456-7890', 'Louisiana');
1191 These are then used directly in your DBI code:
1193 my $sth = $dbh->prepare($stmt);
1194 $sth->execute(@bind);
1196 =head2 Inserting and Updating Arrays
1198 If your database has array types (like for example Postgres),
1199 activate the special option C<< array_datatypes => 1 >>
1200 when creating the C<SQL::Abstract> object.
1201 Then you may use an arrayref to insert and update database array types:
1203 my $sql = SQL::Abstract->new(array_datatypes => 1);
1205 planets => [qw/Mercury Venus Earth Mars/]
1208 my($stmt, @bind) = $sql->insert('solar_system', \%data);
1212 $stmt = "INSERT INTO solar_system (planets) VALUES (?)"
1214 @bind = (['Mercury', 'Venus', 'Earth', 'Mars']);
1217 =head2 Inserting and Updating SQL
1219 In order to apply SQL functions to elements of your C<%data> you may
1220 specify a reference to an arrayref for the given hash value. For example,
1221 if you need to execute the Oracle C<to_date> function on a value, you can
1222 say something like this:
1226 date_entered => \["to_date(?,'MM/DD/YYYY')", "03/02/2003"],
1229 The first value in the array is the actual SQL. Any other values are
1230 optional and would be included in the bind values array. This gives
1233 my($stmt, @bind) = $sql->insert('people', \%data);
1235 $stmt = "INSERT INTO people (name, date_entered)
1236 VALUES (?, to_date(?,'MM/DD/YYYY'))";
1237 @bind = ('Bill', '03/02/2003');
1239 An UPDATE is just as easy, all you change is the name of the function:
1241 my($stmt, @bind) = $sql->update('people', \%data);
1243 Notice that your C<%data> isn't touched; the module will generate
1244 the appropriately quirky SQL for you automatically. Usually you'll
1245 want to specify a WHERE clause for your UPDATE, though, which is
1246 where handling C<%where> hashes comes in handy...
1248 =head2 Complex where statements
1250 This module can generate pretty complicated WHERE statements
1251 easily. For example, simple C<key=value> pairs are taken to mean
1252 equality, and if you want to see if a field is within a set
1253 of values, you can use an arrayref. Let's say we wanted to
1254 SELECT some data based on this criteria:
1257 requestor => 'inna',
1258 worker => ['nwiger', 'rcwe', 'sfz'],
1259 status => { '!=', 'completed' }
1262 my($stmt, @bind) = $sql->select('tickets', '*', \%where);
1264 The above would give you something like this:
1266 $stmt = "SELECT * FROM tickets WHERE
1267 ( requestor = ? ) AND ( status != ? )
1268 AND ( worker = ? OR worker = ? OR worker = ? )";
1269 @bind = ('inna', 'completed', 'nwiger', 'rcwe', 'sfz');
1271 Which you could then use in DBI code like so:
1273 my $sth = $dbh->prepare($stmt);
1274 $sth->execute(@bind);
1280 The functions are simple. There's one for each major SQL operation,
1281 and a constructor you use first. The arguments are specified in a
1282 similar order to each function (table, then fields, then a where
1283 clause) to try and simplify things.
1288 =head2 new(option => 'value')
1290 The C<new()> function takes a list of options and values, and returns
1291 a new B<SQL::Abstract> object which can then be used to generate SQL
1292 through the methods below. The options accepted are:
1298 If set to 'lower', then SQL will be generated in all lowercase. By
1299 default SQL is generated in "textbook" case meaning something like:
1301 SELECT a_field FROM a_table WHERE some_field LIKE '%someval%'
1303 Any setting other than 'lower' is ignored.
1307 This determines what the default comparison operator is. By default
1308 it is C<=>, meaning that a hash like this:
1310 %where = (name => 'nwiger', email => 'nate@wiger.org');
1312 Will generate SQL like this:
1314 WHERE name = 'nwiger' AND email = 'nate@wiger.org'
1316 However, you may want loose comparisons by default, so if you set
1317 C<cmp> to C<like> you would get SQL such as:
1319 WHERE name like 'nwiger' AND email like 'nate@wiger.org'
1321 You can also override the comparsion on an individual basis - see
1322 the huge section on L</"WHERE CLAUSES"> at the bottom.
1324 =item sqltrue, sqlfalse
1326 Expressions for inserting boolean values within SQL statements.
1327 By default these are C<1=1> and C<1=0>. They are used
1328 by the special operators C<-in> and C<-not_in> for generating
1329 correct SQL even when the argument is an empty array (see below).
1333 This determines the default logical operator for multiple WHERE
1334 statements in arrays or hashes. If absent, the default logic is "or"
1335 for arrays, and "and" for hashes. This means that a WHERE
1339 event_date => {'>=', '2/13/99'},
1340 event_date => {'<=', '4/24/03'},
1343 will generate SQL like this:
1345 WHERE event_date >= '2/13/99' OR event_date <= '4/24/03'
1347 This is probably not what you want given this query, though (look
1348 at the dates). To change the "OR" to an "AND", simply specify:
1350 my $sql = SQL::Abstract->new(logic => 'and');
1352 Which will change the above C<WHERE> to:
1354 WHERE event_date >= '2/13/99' AND event_date <= '4/24/03'
1356 The logic can also be changed locally by inserting
1357 a modifier in front of an arrayref :
1359 @where = (-and => [event_date => {'>=', '2/13/99'},
1360 event_date => {'<=', '4/24/03'} ]);
1362 See the L</"WHERE CLAUSES"> section for explanations.
1366 This will automatically convert comparisons using the specified SQL
1367 function for both column and value. This is mostly used with an argument
1368 of C<upper> or C<lower>, so that the SQL will have the effect of
1369 case-insensitive "searches". For example, this:
1371 $sql = SQL::Abstract->new(convert => 'upper');
1372 %where = (keywords => 'MaKe iT CAse inSeNSItive');
1374 Will turn out the following SQL:
1376 WHERE upper(keywords) like upper('MaKe iT CAse inSeNSItive')
1378 The conversion can be C<upper()>, C<lower()>, or any other SQL function
1379 that can be applied symmetrically to fields (actually B<SQL::Abstract> does
1380 not validate this option; it will just pass through what you specify verbatim).
1384 This is a kludge because many databases suck. For example, you can't
1385 just bind values using DBI's C<execute()> for Oracle C<CLOB> or C<BLOB> fields.
1386 Instead, you have to use C<bind_param()>:
1388 $sth->bind_param(1, 'reg data');
1389 $sth->bind_param(2, $lots, {ora_type => ORA_CLOB});
1391 The problem is, B<SQL::Abstract> will normally just return a C<@bind> array,
1392 which loses track of which field each slot refers to. Fear not.
1394 If you specify C<bindtype> in new, you can determine how C<@bind> is returned.
1395 Currently, you can specify either C<normal> (default) or C<columns>. If you
1396 specify C<columns>, you will get an array that looks like this:
1398 my $sql = SQL::Abstract->new(bindtype => 'columns');
1399 my($stmt, @bind) = $sql->insert(...);
1402 [ 'column1', 'value1' ],
1403 [ 'column2', 'value2' ],
1404 [ 'column3', 'value3' ],
1407 You can then iterate through this manually, using DBI's C<bind_param()>.
1409 $sth->prepare($stmt);
1412 my($col, $data) = @$_;
1413 if ($col eq 'details' || $col eq 'comments') {
1414 $sth->bind_param($i, $data, {ora_type => ORA_CLOB});
1415 } elsif ($col eq 'image') {
1416 $sth->bind_param($i, $data, {ora_type => ORA_BLOB});
1418 $sth->bind_param($i, $data);
1422 $sth->execute; # execute without @bind now
1424 Now, why would you still use B<SQL::Abstract> if you have to do this crap?
1425 Basically, the advantage is still that you don't have to care which fields
1426 are or are not included. You could wrap that above C<for> loop in a simple
1427 sub called C<bind_fields()> or something and reuse it repeatedly. You still
1428 get a layer of abstraction over manual SQL specification.
1430 Note that if you set L</bindtype> to C<columns>, the C<\[$sql, @bind]>
1431 construct (see L</Literal SQL with placeholders and bind values (subqueries)>)
1432 will expect the bind values in this format.
1436 This is the character that a table or column name will be quoted
1437 with. By default this is an empty string, but you could set it to
1438 the character C<`>, to generate SQL like this:
1440 SELECT `a_field` FROM `a_table` WHERE `some_field` LIKE '%someval%'
1442 Alternatively, you can supply an array ref of two items, the first being the left
1443 hand quote character, and the second the right hand quote character. For
1444 example, you could supply C<['[',']']> for SQL Server 2000 compliant quotes
1445 that generates SQL like this:
1447 SELECT [a_field] FROM [a_table] WHERE [some_field] LIKE '%someval%'
1449 Quoting is useful if you have tables or columns names that are reserved
1450 words in your database's SQL dialect.
1454 This is the character that separates a table and column name. It is
1455 necessary to specify this when the C<quote_char> option is selected,
1456 so that tables and column names can be individually quoted like this:
1458 SELECT `table`.`one_field` FROM `table` WHERE `table`.`other_field` = 1
1460 =item array_datatypes
1462 When this option is true, arrayrefs in INSERT or UPDATE are
1463 interpreted as array datatypes and are passed directly
1465 When this option is false, arrayrefs are interpreted
1466 as literal SQL, just like refs to arrayrefs
1467 (but this behavior is for backwards compatibility; when writing
1468 new queries, use the "reference to arrayref" syntax
1474 Takes a reference to a list of "special operators"
1475 to extend the syntax understood by L<SQL::Abstract>.
1476 See section L</"SPECIAL OPERATORS"> for details.
1482 =head2 insert($table, \@values || \%fieldvals)
1484 This is the simplest function. You simply give it a table name
1485 and either an arrayref of values or hashref of field/value pairs.
1486 It returns an SQL INSERT statement and a list of bind values.
1487 See the sections on L</"Inserting and Updating Arrays"> and
1488 L</"Inserting and Updating SQL"> for information on how to insert
1489 with those data types.
1491 =head2 update($table, \%fieldvals, \%where)
1493 This takes a table, hashref of field/value pairs, and an optional
1494 hashref L<WHERE clause|/WHERE CLAUSES>. It returns an SQL UPDATE function and a list
1496 See the sections on L</"Inserting and Updating Arrays"> and
1497 L</"Inserting and Updating SQL"> for information on how to insert
1498 with those data types.
1500 =head2 select($source, $fields, $where, $order)
1502 This returns a SQL SELECT statement and associated list of bind values, as
1503 specified by the arguments :
1509 Specification of the 'FROM' part of the statement.
1510 The argument can be either a plain scalar (interpreted as a table
1511 name, will be quoted), or an arrayref (interpreted as a list
1512 of table names, joined by commas, quoted), or a scalarref
1513 (literal table name, not quoted), or a ref to an arrayref
1514 (list of literal table names, joined by commas, not quoted).
1518 Specification of the list of fields to retrieve from
1520 The argument can be either an arrayref (interpreted as a list
1521 of field names, will be joined by commas and quoted), or a
1522 plain scalar (literal SQL, not quoted).
1523 Please observe that this API is not as flexible as for
1524 the first argument C<$table>, for backwards compatibility reasons.
1528 Optional argument to specify the WHERE part of the query.
1529 The argument is most often a hashref, but can also be
1530 an arrayref or plain scalar --
1531 see section L<WHERE clause|/"WHERE CLAUSES"> for details.
1535 Optional argument to specify the ORDER BY part of the query.
1536 The argument can be a scalar, a hashref or an arrayref
1537 -- see section L<ORDER BY clause|/"ORDER BY CLAUSES">
1543 =head2 delete($table, \%where)
1545 This takes a table name and optional hashref L<WHERE clause|/WHERE CLAUSES>.
1546 It returns an SQL DELETE statement and list of bind values.
1548 =head2 where(\%where, \@order)
1550 This is used to generate just the WHERE clause. For example,
1551 if you have an arbitrary data structure and know what the
1552 rest of your SQL is going to look like, but want an easy way
1553 to produce a WHERE clause, use this. It returns an SQL WHERE
1554 clause and list of bind values.
1557 =head2 values(\%data)
1559 This just returns the values from the hash C<%data>, in the same
1560 order that would be returned from any of the other above queries.
1561 Using this allows you to markedly speed up your queries if you
1562 are affecting lots of rows. See below under the L</"PERFORMANCE"> section.
1564 =head2 generate($any, 'number', $of, \@data, $struct, \%types)
1566 Warning: This is an experimental method and subject to change.
1568 This returns arbitrarily generated SQL. It's a really basic shortcut.
1569 It will return two different things, depending on return context:
1571 my($stmt, @bind) = $sql->generate('create table', \$table, \@fields);
1572 my $stmt_and_val = $sql->generate('create table', \$table, \@fields);
1574 These would return the following:
1576 # First calling form
1577 $stmt = "CREATE TABLE test (?, ?)";
1578 @bind = (field1, field2);
1580 # Second calling form
1581 $stmt_and_val = "CREATE TABLE test (field1, field2)";
1583 Depending on what you're trying to do, it's up to you to choose the correct
1584 format. In this example, the second form is what you would want.
1588 $sql->generate('alter session', { nls_date_format => 'MM/YY' });
1592 ALTER SESSION SET nls_date_format = 'MM/YY'
1594 You get the idea. Strings get their case twiddled, but everything
1595 else remains verbatim.
1600 =head1 WHERE CLAUSES
1604 This module uses a variation on the idea from L<DBIx::Abstract>. It
1605 is B<NOT>, repeat I<not> 100% compatible. B<The main logic of this
1606 module is that things in arrays are OR'ed, and things in hashes
1609 The easiest way to explain is to show lots of examples. After
1610 each C<%where> hash shown, it is assumed you used:
1612 my($stmt, @bind) = $sql->where(\%where);
1614 However, note that the C<%where> hash can be used directly in any
1615 of the other functions as well, as described above.
1617 =head2 Key-value pairs
1619 So, let's get started. To begin, a simple hash:
1623 status => 'completed'
1626 Is converted to SQL C<key = val> statements:
1628 $stmt = "WHERE user = ? AND status = ?";
1629 @bind = ('nwiger', 'completed');
1631 One common thing I end up doing is having a list of values that
1632 a field can be in. To do this, simply specify a list inside of
1637 status => ['assigned', 'in-progress', 'pending'];
1640 This simple code will create the following:
1642 $stmt = "WHERE user = ? AND ( status = ? OR status = ? OR status = ? )";
1643 @bind = ('nwiger', 'assigned', 'in-progress', 'pending');
1645 A field associated to an empty arrayref will be considered a
1646 logical false and will generate 0=1.
1648 =head2 Specific comparison operators
1650 If you want to specify a different type of operator for your comparison,
1651 you can use a hashref for a given column:
1655 status => { '!=', 'completed' }
1658 Which would generate:
1660 $stmt = "WHERE user = ? AND status != ?";
1661 @bind = ('nwiger', 'completed');
1663 To test against multiple values, just enclose the values in an arrayref:
1665 status => { '=', ['assigned', 'in-progress', 'pending'] };
1667 Which would give you:
1669 "WHERE status = ? OR status = ? OR status = ?"
1672 The hashref can also contain multiple pairs, in which case it is expanded
1673 into an C<AND> of its elements:
1677 status => { '!=', 'completed', -not_like => 'pending%' }
1680 # Or more dynamically, like from a form
1681 $where{user} = 'nwiger';
1682 $where{status}{'!='} = 'completed';
1683 $where{status}{'-not_like'} = 'pending%';
1685 # Both generate this
1686 $stmt = "WHERE user = ? AND status != ? AND status NOT LIKE ?";
1687 @bind = ('nwiger', 'completed', 'pending%');
1690 To get an OR instead, you can combine it with the arrayref idea:
1694 priority => [ {'=', 2}, {'!=', 1} ]
1697 Which would generate:
1699 $stmt = "WHERE user = ? AND priority = ? OR priority != ?";
1700 @bind = ('nwiger', '2', '1');
1702 If you want to include literal SQL (with or without bind values), just use a
1703 scalar reference or array reference as the value:
1706 date_entered => { '>' => \["to_date(?, 'MM/DD/YYYY')", "11/26/2008"] },
1707 date_expires => { '<' => \"now()" }
1710 Which would generate:
1712 $stmt = "WHERE date_entered > "to_date(?, 'MM/DD/YYYY') AND date_expires < now()";
1713 @bind = ('11/26/2008');
1716 =head2 Logic and nesting operators
1718 In the example above,
1719 there is a subtle trap if you want to say something like
1720 this (notice the C<AND>):
1722 WHERE priority != ? AND priority != ?
1724 Because, in Perl you I<can't> do this:
1726 priority => { '!=', 2, '!=', 1 }
1728 As the second C<!=> key will obliterate the first. The solution
1729 is to use the special C<-modifier> form inside an arrayref:
1731 priority => [ -and => {'!=', 2},
1735 Normally, these would be joined by C<OR>, but the modifier tells it
1736 to use C<AND> instead. (Hint: You can use this in conjunction with the
1737 C<logic> option to C<new()> in order to change the way your queries
1738 work by default.) B<Important:> Note that the C<-modifier> goes
1739 B<INSIDE> the arrayref, as an extra first element. This will
1740 B<NOT> do what you think it might:
1742 priority => -and => [{'!=', 2}, {'!=', 1}] # WRONG!
1744 Here is a quick list of equivalencies, since there is some overlap:
1747 status => {'!=', 'completed', 'not like', 'pending%' }
1748 status => [ -and => {'!=', 'completed'}, {'not like', 'pending%'}]
1751 status => {'=', ['assigned', 'in-progress']}
1752 status => [ -or => {'=', 'assigned'}, {'=', 'in-progress'}]
1753 status => [ {'=', 'assigned'}, {'=', 'in-progress'} ]
1757 =head2 Special operators : IN, BETWEEN, etc.
1759 You can also use the hashref format to compare a list of fields using the
1760 C<IN> comparison operator, by specifying the list as an arrayref:
1763 status => 'completed',
1764 reportid => { -in => [567, 2335, 2] }
1767 Which would generate:
1769 $stmt = "WHERE status = ? AND reportid IN (?,?,?)";
1770 @bind = ('completed', '567', '2335', '2');
1772 The reverse operator C<-not_in> generates SQL C<NOT IN> and is used in
1775 If the argument to C<-in> is an empty array, 'sqlfalse' is generated
1776 (by default : C<1=0>). Similarly, C<< -not_in => [] >> generates
1777 'sqltrue' (by default : C<1=1>).
1781 Another pair of operators is C<-between> and C<-not_between>,
1782 used with an arrayref of two values:
1786 completion_date => {
1787 -not_between => ['2002-10-01', '2003-02-06']
1793 WHERE user = ? AND completion_date NOT BETWEEN ( ? AND ? )
1795 These are the two builtin "special operators"; but the
1796 list can be expanded : see section L</"SPECIAL OPERATORS"> below.
1798 =head2 Nested conditions, -and/-or prefixes
1800 So far, we've seen how multiple conditions are joined with a top-level
1801 C<AND>. We can change this by putting the different conditions we want in
1802 hashes and then putting those hashes in an array. For example:
1807 status => { -like => ['pending%', 'dispatched'] },
1811 status => 'unassigned',
1815 This data structure would create the following:
1817 $stmt = "WHERE ( user = ? AND ( status LIKE ? OR status LIKE ? ) )
1818 OR ( user = ? AND status = ? ) )";
1819 @bind = ('nwiger', 'pending', 'dispatched', 'robot', 'unassigned');
1822 There is also a special C<-nest>
1823 operator which adds an additional set of parens, to create a subquery.
1824 For example, to get something like this:
1826 $stmt = "WHERE user = ? AND ( workhrs > ? OR geo = ? )";
1827 @bind = ('nwiger', '20', 'ASIA');
1833 -nest => [ workhrs => {'>', 20}, geo => 'ASIA' ],
1837 Finally, clauses in hashrefs or arrayrefs can be
1838 prefixed with an C<-and> or C<-or> to change the logic
1845 -and => [workhrs => {'>', 20}, geo => 'ASIA' ],
1846 -and => [workhrs => {'<', 50}, geo => 'EURO' ]
1853 WHERE ( user = ? AND
1854 ( ( workhrs > ? AND geo = ? )
1855 OR ( workhrs < ? AND geo = ? ) ) )
1858 =head2 Algebraic inconsistency, for historical reasons
1860 C<Important note>: when connecting several conditions, the C<-and->|C<-or>
1861 operator goes C<outside> of the nested structure; whereas when connecting
1862 several constraints on one column, the C<-and> operator goes
1863 C<inside> the arrayref. Here is an example combining both features :
1866 -and => [a => 1, b => 2],
1867 -or => [c => 3, d => 4],
1868 e => [-and => {-like => 'foo%'}, {-like => '%bar'} ]
1873 WHERE ( ( ( a = ? AND b = ? )
1874 OR ( c = ? OR d = ? )
1875 OR ( e LIKE ? AND e LIKE ? ) ) )
1877 This difference in syntax is unfortunate but must be preserved for
1878 historical reasons. So be careful : the two examples below would
1879 seem algebraically equivalent, but they are not
1881 {col => [-and => {-like => 'foo%'}, {-like => '%bar'}]}
1882 # yields : WHERE ( ( col LIKE ? AND col LIKE ? ) )
1884 [-and => {col => {-like => 'foo%'}, {col => {-like => '%bar'}}]]
1885 # yields : WHERE ( ( col LIKE ? OR col LIKE ? ) )
1890 Finally, sometimes only literal SQL will do. If you want to include
1891 literal SQL verbatim, you can specify it as a scalar reference, namely:
1893 my $inn = 'is Not Null';
1895 priority => { '<', 2 },
1901 $stmt = "WHERE priority < ? AND requestor is Not Null";
1904 Note that in this example, you only get one bind parameter back, since
1905 the verbatim SQL is passed as part of the statement.
1907 Of course, just to prove a point, the above can also be accomplished
1911 priority => { '<', 2 },
1912 requestor => { '!=', undef },
1918 Conditions on boolean columns can be expressed in the
1919 same way, passing a reference to an empty string :
1922 priority => { '<', 2 },
1928 $stmt = "WHERE priority < ? AND is_ready";
1932 =head2 Literal SQL with placeholders and bind values (subqueries)
1934 If the literal SQL to be inserted has placeholders and bind values,
1935 use a reference to an arrayref (yes this is a double reference --
1936 not so common, but perfectly legal Perl). For example, to find a date
1937 in Postgres you can use something like this:
1940 date_column => \[q/= date '2008-09-30' - ?::integer/, 10/]
1945 $stmt = "WHERE ( date_column = date '2008-09-30' - ?::integer )"
1948 Note that you must pass the bind values in the same format as they are returned
1949 by L</where>. That means that if you set L</bindtype> to C<columns>, you must
1950 provide the bind values in the C<< [ column_meta => value ] >> format, where
1951 C<column_meta> is an opaque scalar value; most commonly the column name, but
1952 you can use any scalar value (including references and blessed references),
1953 L<SQL::Abstract> will simply pass it through intact. So if C<bindtype> is set
1954 to C<columns> the above example will look like:
1957 date_column => \[q/= date '2008-09-30' - ?::integer/, [ dummy => 10 ]/]
1960 Literal SQL is especially useful for nesting parenthesized clauses in the
1961 main SQL query. Here is a first example :
1963 my ($sub_stmt, @sub_bind) = ("SELECT c1 FROM t1 WHERE c2 < ? AND c3 LIKE ?",
1967 bar => \["IN ($sub_stmt)" => @sub_bind],
1972 $stmt = "WHERE (foo = ? AND bar IN (SELECT c1 FROM t1
1973 WHERE c2 < ? AND c3 LIKE ?))";
1974 @bind = (1234, 100, "foo%");
1976 Other subquery operators, like for example C<"E<gt> ALL"> or C<"NOT IN">,
1977 are expressed in the same way. Of course the C<$sub_stmt> and
1978 its associated bind values can be generated through a former call
1981 my ($sub_stmt, @sub_bind)
1982 = $sql->select("t1", "c1", {c2 => {"<" => 100},
1983 c3 => {-like => "foo%"}});
1986 bar => \["> ALL ($sub_stmt)" => @sub_bind],
1989 In the examples above, the subquery was used as an operator on a column;
1990 but the same principle also applies for a clause within the main C<%where>
1991 hash, like an EXISTS subquery :
1993 my ($sub_stmt, @sub_bind)
1994 = $sql->select("t1", "*", {c1 => 1, c2 => \"> t0.c0"});
1997 -nest => \["EXISTS ($sub_stmt)" => @sub_bind],
2002 $stmt = "WHERE (foo = ? AND EXISTS (SELECT * FROM t1
2003 WHERE c1 = ? AND c2 > t0.c0))";
2007 Observe that the condition on C<c2> in the subquery refers to
2008 column C<t0.c0> of the main query : this is I<not> a bind
2009 value, so we have to express it through a scalar ref.
2010 Writing C<< c2 => {">" => "t0.c0"} >> would have generated
2011 C<< c2 > ? >> with bind value C<"t0.c0"> ... not exactly
2012 what we wanted here.
2014 Another use of the subquery technique is when some SQL clauses need
2015 parentheses, as it often occurs with some proprietary SQL extensions
2016 like for example fulltext expressions, geospatial expressions,
2017 NATIVE clauses, etc. Here is an example of a fulltext query in MySQL :
2020 -nest => \["MATCH (col1, col2) AGAINST (?)" => qw/apples/]
2023 Finally, here is an example where a subquery is used
2024 for expressing unary negation:
2026 my ($sub_stmt, @sub_bind)
2027 = $sql->where({age => [{"<" => 10}, {">" => 20}]});
2028 $sub_stmt =~ s/^ where //i; # don't want "WHERE" in the subclause
2030 lname => {like => '%son%'},
2031 -nest => \["NOT ($sub_stmt)" => @sub_bind],
2036 $stmt = "lname LIKE ? AND NOT ( age < ? OR age > ? )"
2037 @bind = ('%son%', 10, 20)
2043 These pages could go on for a while, since the nesting of the data
2044 structures this module can handle are pretty much unlimited (the
2045 module implements the C<WHERE> expansion as a recursive function
2046 internally). Your best bet is to "play around" with the module a
2047 little to see how the data structures behave, and choose the best
2048 format for your data based on that.
2050 And of course, all the values above will probably be replaced with
2051 variables gotten from forms or the command line. After all, if you
2052 knew everything ahead of time, you wouldn't have to worry about
2053 dynamically-generating SQL and could just hardwire it into your
2059 =head1 ORDER BY CLAUSES
2061 Some functions take an order by clause. This can either be a scalar (just a
2062 column name,) a hash of C<< { -desc => 'col' } >> or C<< { -asc => 'col' } >>,
2063 or an array of either of the two previous forms. Examples:
2065 Given | Will Generate
2066 ----------------------------------------------------------
2067 \'colA DESC' | ORDER BY colA DESC
2068 'colA' | ORDER BY colA
2069 [qw/colA colB/] | ORDER BY colA, colB
2070 {-asc => 'colA'} | ORDER BY colA ASC
2071 {-desc => 'colB'} | ORDER BY colB DESC
2073 {-asc => 'colA'}, | ORDER BY colA ASC, colB DESC
2076 [colA => {-asc => 'colB'}] | ORDER BY colA, colB ASC
2077 ==========================================================
2081 =head1 SPECIAL OPERATORS
2083 my $sqlmaker = SQL::Abstract->new(special_ops => [
2086 my ($self, $field, $op, $arg) = @_;
2092 A "special operator" is a SQL syntactic clause that can be
2093 applied to a field, instead of a usual binary operator.
2096 WHERE field IN (?, ?, ?)
2097 WHERE field BETWEEN ? AND ?
2098 WHERE MATCH(field) AGAINST (?, ?)
2100 Special operators IN and BETWEEN are fairly standard and therefore
2101 are builtin within C<SQL::Abstract>. For other operators,
2102 like the MATCH .. AGAINST example above which is
2103 specific to MySQL, you can write your own operator handlers :
2104 supply a C<special_ops> argument to the C<new> method.
2105 That argument takes an arrayref of operator definitions;
2106 each operator definition is a hashref with two entries
2112 the regular expression to match the operator
2116 coderef that will be called when meeting that operator
2117 in the input tree. The coderef will be called with
2118 arguments C<< ($self, $field, $op, $arg) >>, and
2119 should return a C<< ($sql, @bind) >> structure.
2123 For example, here is an implementation
2124 of the MATCH .. AGAINST syntax for MySQL
2126 my $sqlmaker = SQL::Abstract->new(special_ops => [
2128 # special op for MySql MATCH (field) AGAINST(word1, word2, ...)
2129 {regex => qr/^match$/i,
2131 my ($self, $field, $op, $arg) = @_;
2132 $arg = [$arg] if not ref $arg;
2133 my $label = $self->_quote($field);
2134 my ($placeholder) = $self->_convert('?');
2135 my $placeholders = join ", ", (($placeholder) x @$arg);
2136 my $sql = $self->_sqlcase('match') . " ($label) "
2137 . $self->_sqlcase('against') . " ($placeholders) ";
2138 my @bind = $self->_bindtype($field, @$arg);
2139 return ($sql, @bind);
2148 Thanks to some benchmarking by Mark Stosberg, it turns out that
2149 this module is many orders of magnitude faster than using C<DBIx::Abstract>.
2150 I must admit this wasn't an intentional design issue, but it's a
2151 byproduct of the fact that you get to control your C<DBI> handles
2154 To maximize performance, use a code snippet like the following:
2156 # prepare a statement handle using the first row
2157 # and then reuse it for the rest of the rows
2159 for my $href (@array_of_hashrefs) {
2160 $stmt ||= $sql->insert('table', $href);
2161 $sth ||= $dbh->prepare($stmt);
2162 $sth->execute($sql->values($href));
2165 The reason this works is because the keys in your C<$href> are sorted
2166 internally by B<SQL::Abstract>. Thus, as long as your data retains
2167 the same structure, you only have to generate the SQL the first time
2168 around. On subsequent queries, simply use the C<values> function provided
2169 by this module to return your values in the correct order.
2174 If you use my C<CGI::FormBuilder> module at all, you'll hopefully
2175 really like this part (I do, at least). Building up a complex query
2176 can be as simple as the following:
2180 use CGI::FormBuilder;
2183 my $form = CGI::FormBuilder->new(...);
2184 my $sql = SQL::Abstract->new;
2186 if ($form->submitted) {
2187 my $field = $form->field;
2188 my $id = delete $field->{id};
2189 my($stmt, @bind) = $sql->update('table', $field, {id => $id});
2192 Of course, you would still have to connect using C<DBI> to run the
2193 query, but the point is that if you make your form look like your
2194 table, the actual query script can be extremely simplistic.
2196 If you're B<REALLY> lazy (I am), check out C<HTML::QuickTable> for
2197 a fast interface to returning and formatting data. I frequently
2198 use these three modules together to write complex database query
2199 apps in under 50 lines.
2204 Version 1.50 was a major internal refactoring of C<SQL::Abstract>.
2205 Great care has been taken to preserve the I<published> behavior
2206 documented in previous versions in the 1.* family; however,
2207 some features that were previously undocumented, or behaved
2208 differently from the documentation, had to be changed in order
2209 to clarify the semantics. Hence, client code that was relying
2210 on some dark areas of C<SQL::Abstract> v1.*
2211 B<might behave differently> in v1.50.
2213 The main changes are :
2219 support for literal SQL through the C<< \ [$sql, bind] >> syntax.
2223 support for the { operator => \"..." } construct (to embed literal SQL)
2227 support for the { operator => \["...", @bind] } construct (to embed literal SQL with bind values)
2231 optional support for L<array datatypes|/"Inserting and Updating Arrays">
2235 defensive programming : check arguments
2239 fixed bug with global logic, which was previously implemented
2240 through global variables yielding side-effects. Prior versions would
2241 interpret C<< [ {cond1, cond2}, [cond3, cond4] ] >>
2242 as C<< "(cond1 AND cond2) OR (cond3 AND cond4)" >>.
2243 Now this is interpreted
2244 as C<< "(cond1 AND cond2) OR (cond3 OR cond4)" >>.
2249 fixed semantics of _bindtype on array args
2253 dropped the C<_anoncopy> of the %where tree. No longer necessary,
2254 we just avoid shifting arrays within that tree.
2258 dropped the C<_modlogic> function
2264 =head1 ACKNOWLEDGEMENTS
2266 There are a number of individuals that have really helped out with
2267 this module. Unfortunately, most of them submitted bugs via CPAN
2268 so I have no idea who they are! But the people I do know are:
2270 Ash Berlin (order_by hash term support)
2271 Matt Trout (DBIx::Class support)
2272 Mark Stosberg (benchmarking)
2273 Chas Owens (initial "IN" operator support)
2274 Philip Collins (per-field SQL functions)
2275 Eric Kolve (hashref "AND" support)
2276 Mike Fragassi (enhancements to "BETWEEN" and "LIKE")
2277 Dan Kubb (support for "quote_char" and "name_sep")
2278 Guillermo Roditi (patch to cleanup "IN" and "BETWEEN", fix and tests for _order_by)
2279 Laurent Dami (internal refactoring, multiple -nest, extensible list of special operators, literal SQL)
2280 Norbert Buchmuller (support for literal SQL in hashpair, misc. fixes & tests)
2286 L<DBIx::Class>, L<DBIx::Abstract>, L<CGI::FormBuilder>, L<HTML::QuickTable>.
2290 Copyright (c) 2001-2007 Nathan Wiger <nwiger@cpan.org>. All Rights Reserved.
2292 This module is actively maintained by Matt Trout <mst@shadowcatsystems.co.uk>
2294 For support, your best bet is to try the C<DBIx::Class> users mailing list.
2295 While not an official support venue, C<DBIx::Class> makes heavy use of
2296 C<SQL::Abstract>, and as such list members there are very familiar with
2297 how to create queries.
2299 This module is free software; you may copy this under the terms of
2300 the GNU General Public License, or the Artistic License, copies of
2301 which should have accompanied your Perl kit.