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 $VERSION = '1.53';
20 # This would confuse some packagers
21 #$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} = $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 for my $k (sort keys %$where) {
426 my $v = $where->{$k};
428 # ($k => $v) is either a special op or a regular hashpair
429 my ($sql, @bind) = ($k =~ /^-(.+)/) ? $self->_where_op_in_hash($1, $v)
431 my $method = $self->_METHOD_FOR_refkind("_where_hashpair", $v);
432 $self->$method($k, $v);
435 push @sql_clauses, $sql;
436 push @all_bind, @bind;
439 return $self->_join_sql_clauses('and', \@sql_clauses, \@all_bind);
443 sub _where_op_in_hash {
444 my ($self, $op_str, $v) = @_;
446 $op_str =~ /^ (AND|OR|PAREN|NEST) ( \_? \d* ) $/xi
447 or puke "unknown operator: -$op_str";
449 my $op = uc($1); # uppercase, remove trailing digits
451 belch 'Use of op_N modifiers is deprecated and will be removed in SQLA v2.0. '
452 . "You probably wanted ...-and => [ $op_str => COND1, $op_str => COND2 ... ]";
456 belch 'The -nest modifier is deprecated in favor of -paren and will be removed in SQLA v2.0';
460 $self->_debug("OP(-$op) within hashref, recursing...");
462 $self->_SWITCH_refkind($v, {
465 return $self->_where_ARRAYREF($v, $op eq 'PAREN' ? '' : $op);
470 return $self->_where_ARRAYREF([ map { $_ => $v->{$_} } (sort keys %$v) ], 'OR');
473 return $self->_where_HASHREF($v);
477 SCALARREF => sub { # literal SQL
479 or puke "-$op => \\\$scalar not supported, use -paren => ...";
483 ARRAYREFREF => sub { # literal SQL
485 or puke "-$op => \\[..] not supported, use -paren => ...";
489 SCALAR => sub { # permissively interpreted as SQL
491 or puke "-$op => 'scalar' not supported, use -paren => \\'scalar'";
492 belch "literal SQL should be -paren => \\'scalar' "
493 . "instead of -paren => 'scalar' ";
498 puke "-$op => undef not supported";
504 sub _where_hashpair_ARRAYREF {
505 my ($self, $k, $v) = @_;
508 my @v = @$v; # need copy because of shift below
509 $self->_debug("ARRAY($k) means distribute over elements");
511 # put apart first element if it is an operator (-and, -or)
513 (defined $v[0] && $v[0] =~ /^ - (?: AND|OR ) $/ix)
517 my @distributed = map { {$k => $_} } @v;
520 $self->_debug("OP($op) reinjected into the distributed array");
521 unshift @distributed, $op;
524 my $logic = $op ? substr($op, 1) : '';
526 return $self->_recurse_where(\@distributed, $logic);
529 # LDNOTE : not sure of this one. What does "distribute over nothing" mean?
530 $self->_debug("empty ARRAY($k) means 0=1");
531 return ($self->{sqlfalse});
535 sub _where_hashpair_HASHREF {
536 my ($self, $k, $v, $logic) = @_;
539 my ($all_sql, @all_bind);
541 for my $op (sort keys %$v) {
544 # put the operator in canonical form
545 $op =~ s/^-//; # remove initial dash
546 $op =~ tr/_/ /; # underscores become spaces
547 $op =~ s/^\s+//; # no initial space
548 $op =~ s/\s+$//; # no final space
549 $op =~ s/\s+/ /; # multiple spaces become one
553 # CASE: special operators like -in or -between
554 my $special_op = first {$op =~ $_->{regex}} @{$self->{special_ops}};
556 ($sql, @bind) = $special_op->{handler}->($self, $k, $op, $val);
559 $self->_SWITCH_refkind($val, {
561 ARRAYREF => sub { # CASE: col => {op => \@vals}
562 ($sql, @bind) = $self->_where_field_op_ARRAYREF($k, $op, $val);
565 SCALARREF => sub { # CASE: col => {op => \$scalar} (literal SQL without bind)
566 $sql = join ' ', $self->_convert($self->_quote($k)),
567 $self->_sqlcase($op),
571 ARRAYREFREF => sub { # CASE: col => {op => \[$sql, @bind]} (literal SQL with bind)
572 my ($sub_sql, @sub_bind) = @$$val;
573 $self->_assert_bindval_matches_bindtype(@sub_bind);
574 $sql = join ' ', $self->_convert($self->_quote($k)),
575 $self->_sqlcase($op),
581 ($sql, @bind) = $self->_where_hashpair_HASHREF($k, $val, $op);
584 UNDEF => sub { # CASE: col => {op => undef} : sql "IS (NOT)? NULL"
585 my $is = ($op =~ $self->{equality_op}) ? 'is' :
586 ($op =~ $self->{inequality_op}) ? 'is not' :
587 puke "unexpected operator '$op' with undef operand";
588 $sql = $self->_quote($k) . $self->_sqlcase(" $is null");
591 FALLBACK => sub { # CASE: col => {op => $scalar}
592 $sql = join ' ', $self->_convert($self->_quote($k)),
593 $self->_sqlcase($op),
594 $self->_convert('?');
595 @bind = $self->_bindtype($k, $val);
600 ($all_sql) = (defined $all_sql and $all_sql) ? $self->_join_sql_clauses($logic, [$all_sql, $sql], []) : $sql;
601 push @all_bind, @bind;
603 return ($all_sql, @all_bind);
608 sub _where_field_op_ARRAYREF {
609 my ($self, $k, $op, $vals) = @_;
612 $self->_debug("ARRAY($vals) means multiple elements: [ @$vals ]");
614 # see if the first element is an -and/-or op
616 if ($vals->[0] =~ /^ - ( AND|OR ) $/ix) {
621 # distribute $op over each remaining member of @$vals, append logic if exists
622 return $self->_recurse_where([map { {$k => {$op, $_}} } @$vals], $logic);
624 # LDNOTE : had planned to change the distribution logic when
625 # $op =~ $self->{inequality_op}, because of Morgan laws :
626 # with {field => {'!=' => [22, 33]}}, it would be ridiculous to generate
627 # WHERE field != 22 OR field != 33 : the user probably means
628 # WHERE field != 22 AND field != 33.
629 # To do this, replace the above to roughly :
630 # my $logic = ($op =~ $self->{inequality_op}) ? 'AND' : 'OR';
631 # return $self->_recurse_where([map { {$k => {$op, $_}} } @$vals], $logic);
635 # try to DWIM on equality operators
636 # LDNOTE : not 100% sure this is the correct thing to do ...
637 return ($self->{sqlfalse}) if $op =~ $self->{equality_op};
638 return ($self->{sqltrue}) if $op =~ $self->{inequality_op};
641 puke "operator '$op' applied on an empty array (field '$k')";
646 sub _where_hashpair_SCALARREF {
647 my ($self, $k, $v) = @_;
648 $self->_debug("SCALAR($k) means literal SQL: $$v");
649 my $sql = $self->_quote($k) . " " . $$v;
653 # literal SQL with bind
654 sub _where_hashpair_ARRAYREFREF {
655 my ($self, $k, $v) = @_;
656 $self->_debug("REF($k) means literal SQL: @${$v}");
657 my ($sql, @bind) = @${$v};
658 $self->_assert_bindval_matches_bindtype(@bind);
659 $sql = $self->_quote($k) . " " . $sql;
660 return ($sql, @bind );
663 # literal SQL without bind
664 sub _where_hashpair_SCALAR {
665 my ($self, $k, $v) = @_;
666 $self->_debug("NOREF($k) means simple key=val: $k $self->{cmp} $v");
667 my $sql = join ' ', $self->_convert($self->_quote($k)),
668 $self->_sqlcase($self->{cmp}),
669 $self->_convert('?');
670 my @bind = $self->_bindtype($k, $v);
671 return ( $sql, @bind);
675 sub _where_hashpair_UNDEF {
676 my ($self, $k, $v) = @_;
677 $self->_debug("UNDEF($k) means IS NULL");
678 my $sql = $self->_quote($k) . $self->_sqlcase(' is null');
682 #======================================================================
683 # WHERE: TOP-LEVEL OTHERS (SCALARREF, SCALAR, UNDEF)
684 #======================================================================
687 sub _where_SCALARREF {
688 my ($self, $where) = @_;
691 $self->_debug("SCALAR(*top) means literal SQL: $$where");
697 my ($self, $where) = @_;
700 $self->_debug("NOREF(*top) means literal SQL: $where");
711 #======================================================================
712 # WHERE: BUILTIN SPECIAL OPERATORS (-in, -between)
713 #======================================================================
716 sub _where_field_BETWEEN {
717 my ($self, $k, $op, $vals) = @_;
719 (ref $vals eq 'ARRAY' && @$vals == 2) or
720 (ref $vals eq 'REF' && (@$$vals == 1 || @$$vals == 2 || @$$vals == 3))
721 or puke "special op 'between' requires an arrayref of two values (or a scalarref or arrayrefref for literal SQL)";
723 my ($clause, @bind, $label, $and, $placeholder);
724 $label = $self->_convert($self->_quote($k));
725 $and = ' ' . $self->_sqlcase('and') . ' ';
726 $placeholder = $self->_convert('?');
727 $op = $self->_sqlcase($op);
729 if (ref $vals eq 'REF') {
730 ($clause, @bind) = @$$vals;
733 my (@all_sql, @all_bind);
735 foreach my $val (@$vals) {
736 my ($sql, @bind) = $self->_SWITCH_refkind($val, {
738 return ($placeholder, ($val));
741 return ($self->_convert($$val), ());
745 push @all_bind, @bind;
748 $clause = (join $and, @all_sql);
749 @bind = $self->_bindtype($k, @all_bind);
751 my $sql = "( $label $op $clause )";
756 sub _where_field_IN {
757 my ($self, $k, $op, $vals) = @_;
759 # backwards compatibility : if scalar, force into an arrayref
760 $vals = [$vals] if defined $vals && ! ref $vals;
762 my ($label) = $self->_convert($self->_quote($k));
763 my ($placeholder) = $self->_convert('?');
764 $op = $self->_sqlcase($op);
766 my ($sql, @bind) = $self->_SWITCH_refkind($vals, {
767 ARRAYREF => sub { # list of choices
768 if (@$vals) { # nonempty list
769 my $placeholders = join ", ", (($placeholder) x @$vals);
770 my $sql = "$label $op ( $placeholders )";
771 my @bind = $self->_bindtype($k, @$vals);
773 return ($sql, @bind);
775 else { # empty list : some databases won't understand "IN ()", so DWIM
776 my $sql = ($op =~ /\bnot\b/i) ? $self->{sqltrue} : $self->{sqlfalse};
781 ARRAYREFREF => sub { # literal SQL with bind
782 my ($sql, @bind) = @$$vals;
783 $self->_assert_bindval_matches_bindtype(@bind);
784 return ("$label $op ( $sql )", @bind);
788 puke "special op 'in' requires an arrayref (or arrayref-ref)";
792 return ($sql, @bind);
800 #======================================================================
802 #======================================================================
805 my ($self, $arg) = @_;
807 # construct list of ordering instructions
808 my @order = $self->_SWITCH_refkind($arg, {
811 map {$self->_SWITCH_refkind($_, {
812 SCALAR => sub {$self->_quote($_)},
814 SCALARREF => sub {$$_}, # literal SQL, no quoting
815 HASHREF => sub {$self->_order_by_hash($_)}
819 SCALAR => sub {$self->_quote($arg)},
821 SCALARREF => sub {$$arg}, # literal SQL, no quoting
822 HASHREF => sub {$self->_order_by_hash($arg)},
827 my $order = join ', ', @order;
828 return $order ? $self->_sqlcase(' order by')." $order" : '';
833 my ($self, $hash) = @_;
835 # get first pair in hash
836 my ($key, $val) = each %$hash;
838 # check if one pair was found and no other pair in hash
839 $key && !(each %$hash)
840 or puke "hash passed to _order_by must have exactly one key (-desc or -asc)";
842 my ($order) = ($key =~ /^-(desc|asc)/i)
843 or puke "invalid key in _order_by hash : $key";
845 $val = ref $val eq 'ARRAY' ? $val : [$val];
846 return join ', ', map { $self->_quote($_) . ' ' . $self->_sqlcase($order) } @$val;
851 #======================================================================
852 # DATASOURCE (FOR NOW, JUST PLAIN TABLE OR LIST OF TABLES)
853 #======================================================================
858 $self->_SWITCH_refkind($from, {
859 ARRAYREF => sub {join ', ', map { $self->_quote($_) } @$from;},
860 SCALAR => sub {$self->_quote($from)},
861 SCALARREF => sub {$$from},
862 ARRAYREFREF => sub {join ', ', @$from;},
867 #======================================================================
869 #======================================================================
875 $label or puke "can't quote an empty label";
877 # left and right quote characters
878 my ($ql, $qr, @other) = $self->_SWITCH_refkind($self->{quote_char}, {
879 SCALAR => sub {($self->{quote_char}, $self->{quote_char})},
880 ARRAYREF => sub {@{$self->{quote_char}}},
884 or puke "quote_char must be an arrayref of 2 values";
886 # no quoting if no quoting chars
887 $ql or return $label;
889 # no quoting for literal SQL
890 return $$label if ref($label) eq 'SCALAR';
892 # separate table / column (if applicable)
893 my $sep = $self->{name_sep} || '';
894 my @to_quote = $sep ? split /\Q$sep\E/, $label : ($label);
896 # do the quoting, except for "*" or for `table`.*
897 my @quoted = map { $_ eq '*' ? $_: $ql.$_.$qr} @to_quote;
899 # reassemble and return.
900 return join $sep, @quoted;
904 # Conversion, if applicable
906 my ($self, $arg) = @_;
908 # LDNOTE : modified the previous implementation below because
909 # it was not consistent : the first "return" is always an array,
910 # the second "return" is context-dependent. Anyway, _convert
911 # seems always used with just a single argument, so make it a
913 # return @_ unless $self->{convert};
914 # my $conv = $self->_sqlcase($self->{convert});
915 # my @ret = map { $conv.'('.$_.')' } @_;
916 # return wantarray ? @ret : $ret[0];
917 if ($self->{convert}) {
918 my $conv = $self->_sqlcase($self->{convert});
919 $arg = $conv.'('.$arg.')';
927 my($col, @vals) = @_;
929 #LDNOTE : changed original implementation below because it did not make
930 # sense when bindtype eq 'columns' and @vals > 1.
931 # return $self->{bindtype} eq 'columns' ? [ $col, @vals ] : @vals;
933 return $self->{bindtype} eq 'columns' ? map {[$col, $_]} @vals : @vals;
936 # Dies if any element of @bind is not in [colname => value] format
937 # if bindtype is 'columns'.
938 sub _assert_bindval_matches_bindtype {
939 my ($self, @bind) = @_;
941 if ($self->{bindtype} eq 'columns') {
942 foreach my $val (@bind) {
943 if (!defined $val || ref($val) ne 'ARRAY' || @$val != 2) {
944 die "bindtype 'columns' selected, you need to pass: [column_name => bind_value]"
950 sub _join_sql_clauses {
951 my ($self, $logic, $clauses_aref, $bind_aref) = @_;
953 if (@$clauses_aref > 1) {
954 my $join = " " . $self->_sqlcase($logic) . " ";
955 my $sql = '( ' . join($join, @$clauses_aref) . ' )';
956 return ($sql, @$bind_aref);
958 elsif (@$clauses_aref) {
959 return ($clauses_aref->[0], @$bind_aref); # no parentheses
962 return (); # if no SQL, ignore @$bind_aref
967 # Fix SQL case, if so requested
971 # LDNOTE: if $self->{case} is true, then it contains 'lower', so we
972 # don't touch the argument ... crooked logic, but let's not change it!
973 return $self->{case} ? $_[0] : uc($_[0]);
977 #======================================================================
978 # DISPATCHING FROM REFKIND
979 #======================================================================
982 my ($self, $data) = @_;
988 # blessed objects are treated like scalars
989 $ref = (blessed $data) ? '' : ref $data;
990 $n_steps += 1 if $ref;
991 last if $ref ne 'REF';
995 my $base = $ref || (defined $data ? 'SCALAR' : 'UNDEF');
997 return $base . ('REF' x $n_steps);
1003 my ($self, $data) = @_;
1004 my @try = ($self->_refkind($data));
1005 push @try, 'SCALAR_or_UNDEF' if $try[0] eq 'SCALAR' || $try[0] eq 'UNDEF';
1006 push @try, 'FALLBACK';
1010 sub _METHOD_FOR_refkind {
1011 my ($self, $meth_prefix, $data) = @_;
1012 my $method = first {$_} map {$self->can($meth_prefix."_".$_)}
1013 $self->_try_refkind($data)
1014 or puke "cannot dispatch on '$meth_prefix' for ".$self->_refkind($data);
1019 sub _SWITCH_refkind {
1020 my ($self, $data, $dispatch_table) = @_;
1022 my $coderef = first {$_} map {$dispatch_table->{$_}}
1023 $self->_try_refkind($data)
1024 or puke "no dispatch entry for ".$self->_refkind($data);
1031 #======================================================================
1032 # VALUES, GENERATE, AUTOLOAD
1033 #======================================================================
1035 # LDNOTE: original code from nwiger, didn't touch code in that section
1036 # I feel the AUTOLOAD stuff should not be the default, it should
1037 # only be activated on explicit demand by user.
1041 my $data = shift || return;
1042 puke "Argument to ", __PACKAGE__, "->values must be a \\%hash"
1043 unless ref $data eq 'HASH';
1046 foreach my $k ( sort keys %$data ) {
1047 my $v = $data->{$k};
1048 $self->_SWITCH_refkind($v, {
1050 if ($self->{array_datatypes}) { # array datatype
1051 push @all_bind, $self->_bindtype($k, $v);
1053 else { # literal SQL with bind
1054 my ($sql, @bind) = @$v;
1055 $self->_assert_bindval_matches_bindtype(@bind);
1056 push @all_bind, @bind;
1059 ARRAYREFREF => sub { # literal SQL with bind
1060 my ($sql, @bind) = @${$v};
1061 $self->_assert_bindval_matches_bindtype(@bind);
1062 push @all_bind, @bind;
1064 SCALARREF => sub { # literal SQL without bind
1066 SCALAR_or_UNDEF => sub {
1067 push @all_bind, $self->_bindtype($k, $v);
1078 my(@sql, @sqlq, @sqlv);
1082 if ($ref eq 'HASH') {
1083 for my $k (sort keys %$_) {
1086 my $label = $self->_quote($k);
1087 if ($r eq 'ARRAY') {
1088 # literal SQL with bind
1089 my ($sql, @bind) = @$v;
1090 $self->_assert_bindval_matches_bindtype(@bind);
1091 push @sqlq, "$label = $sql";
1093 } elsif ($r eq 'SCALAR') {
1094 # literal SQL without bind
1095 push @sqlq, "$label = $$v";
1097 push @sqlq, "$label = ?";
1098 push @sqlv, $self->_bindtype($k, $v);
1101 push @sql, $self->_sqlcase('set'), join ', ', @sqlq;
1102 } elsif ($ref eq 'ARRAY') {
1103 # unlike insert(), assume these are ONLY the column names, i.e. for SQL
1106 if ($r eq 'ARRAY') { # literal SQL with bind
1107 my ($sql, @bind) = @$v;
1108 $self->_assert_bindval_matches_bindtype(@bind);
1111 } elsif ($r eq 'SCALAR') { # literal SQL without bind
1112 # embedded literal SQL
1119 push @sql, '(' . join(', ', @sqlq) . ')';
1120 } elsif ($ref eq 'SCALAR') {
1124 # strings get case twiddled
1125 push @sql, $self->_sqlcase($_);
1129 my $sql = join ' ', @sql;
1131 # this is pretty tricky
1132 # if ask for an array, return ($stmt, @bind)
1133 # otherwise, s/?/shift @sqlv/ to put it inline
1135 return ($sql, @sqlv);
1137 1 while $sql =~ s/\?/my $d = shift(@sqlv);
1138 ref $d ? $d->[1] : $d/e;
1147 # This allows us to check for a local, then _form, attr
1149 my($name) = $AUTOLOAD =~ /.*::(.+)/;
1150 return $self->generate($name, @_);
1161 SQL::Abstract - Generate SQL from Perl data structures
1167 my $sql = SQL::Abstract->new;
1169 my($stmt, @bind) = $sql->select($table, \@fields, \%where, \@order);
1171 my($stmt, @bind) = $sql->insert($table, \%fieldvals || \@values);
1173 my($stmt, @bind) = $sql->update($table, \%fieldvals, \%where);
1175 my($stmt, @bind) = $sql->delete($table, \%where);
1177 # Then, use these in your DBI statements
1178 my $sth = $dbh->prepare($stmt);
1179 $sth->execute(@bind);
1181 # Just generate the WHERE clause
1182 my($stmt, @bind) = $sql->where(\%where, \@order);
1184 # Return values in the same order, for hashed queries
1185 # See PERFORMANCE section for more details
1186 my @bind = $sql->values(\%fieldvals);
1190 This module was inspired by the excellent L<DBIx::Abstract>.
1191 However, in using that module I found that what I really wanted
1192 to do was generate SQL, but still retain complete control over my
1193 statement handles and use the DBI interface. So, I set out to
1194 create an abstract SQL generation module.
1196 While based on the concepts used by L<DBIx::Abstract>, there are
1197 several important differences, especially when it comes to WHERE
1198 clauses. I have modified the concepts used to make the SQL easier
1199 to generate from Perl data structures and, IMO, more intuitive.
1200 The underlying idea is for this module to do what you mean, based
1201 on the data structures you provide it. The big advantage is that
1202 you don't have to modify your code every time your data changes,
1203 as this module figures it out.
1205 To begin with, an SQL INSERT is as easy as just specifying a hash
1206 of C<key=value> pairs:
1209 name => 'Jimbo Bobson',
1210 phone => '123-456-7890',
1211 address => '42 Sister Lane',
1212 city => 'St. Louis',
1213 state => 'Louisiana',
1216 The SQL can then be generated with this:
1218 my($stmt, @bind) = $sql->insert('people', \%data);
1220 Which would give you something like this:
1222 $stmt = "INSERT INTO people
1223 (address, city, name, phone, state)
1224 VALUES (?, ?, ?, ?, ?)";
1225 @bind = ('42 Sister Lane', 'St. Louis', 'Jimbo Bobson',
1226 '123-456-7890', 'Louisiana');
1228 These are then used directly in your DBI code:
1230 my $sth = $dbh->prepare($stmt);
1231 $sth->execute(@bind);
1233 =head2 Inserting and Updating Arrays
1235 If your database has array types (like for example Postgres),
1236 activate the special option C<< array_datatypes => 1 >>
1237 when creating the C<SQL::Abstract> object.
1238 Then you may use an arrayref to insert and update database array types:
1240 my $sql = SQL::Abstract->new(array_datatypes => 1);
1242 planets => [qw/Mercury Venus Earth Mars/]
1245 my($stmt, @bind) = $sql->insert('solar_system', \%data);
1249 $stmt = "INSERT INTO solar_system (planets) VALUES (?)"
1251 @bind = (['Mercury', 'Venus', 'Earth', 'Mars']);
1254 =head2 Inserting and Updating SQL
1256 In order to apply SQL functions to elements of your C<%data> you may
1257 specify a reference to an arrayref for the given hash value. For example,
1258 if you need to execute the Oracle C<to_date> function on a value, you can
1259 say something like this:
1263 date_entered => \["to_date(?,'MM/DD/YYYY')", "03/02/2003"],
1266 The first value in the array is the actual SQL. Any other values are
1267 optional and would be included in the bind values array. This gives
1270 my($stmt, @bind) = $sql->insert('people', \%data);
1272 $stmt = "INSERT INTO people (name, date_entered)
1273 VALUES (?, to_date(?,'MM/DD/YYYY'))";
1274 @bind = ('Bill', '03/02/2003');
1276 An UPDATE is just as easy, all you change is the name of the function:
1278 my($stmt, @bind) = $sql->update('people', \%data);
1280 Notice that your C<%data> isn't touched; the module will generate
1281 the appropriately quirky SQL for you automatically. Usually you'll
1282 want to specify a WHERE clause for your UPDATE, though, which is
1283 where handling C<%where> hashes comes in handy...
1285 =head2 Complex where statements
1287 This module can generate pretty complicated WHERE statements
1288 easily. For example, simple C<key=value> pairs are taken to mean
1289 equality, and if you want to see if a field is within a set
1290 of values, you can use an arrayref. Let's say we wanted to
1291 SELECT some data based on this criteria:
1294 requestor => 'inna',
1295 worker => ['nwiger', 'rcwe', 'sfz'],
1296 status => { '!=', 'completed' }
1299 my($stmt, @bind) = $sql->select('tickets', '*', \%where);
1301 The above would give you something like this:
1303 $stmt = "SELECT * FROM tickets WHERE
1304 ( requestor = ? ) AND ( status != ? )
1305 AND ( worker = ? OR worker = ? OR worker = ? )";
1306 @bind = ('inna', 'completed', 'nwiger', 'rcwe', 'sfz');
1308 Which you could then use in DBI code like so:
1310 my $sth = $dbh->prepare($stmt);
1311 $sth->execute(@bind);
1317 The functions are simple. There's one for each major SQL operation,
1318 and a constructor you use first. The arguments are specified in a
1319 similar order to each function (table, then fields, then a where
1320 clause) to try and simplify things.
1325 =head2 new(option => 'value')
1327 The C<new()> function takes a list of options and values, and returns
1328 a new B<SQL::Abstract> object which can then be used to generate SQL
1329 through the methods below. The options accepted are:
1335 If set to 'lower', then SQL will be generated in all lowercase. By
1336 default SQL is generated in "textbook" case meaning something like:
1338 SELECT a_field FROM a_table WHERE some_field LIKE '%someval%'
1340 Any setting other than 'lower' is ignored.
1344 This determines what the default comparison operator is. By default
1345 it is C<=>, meaning that a hash like this:
1347 %where = (name => 'nwiger', email => 'nate@wiger.org');
1349 Will generate SQL like this:
1351 WHERE name = 'nwiger' AND email = 'nate@wiger.org'
1353 However, you may want loose comparisons by default, so if you set
1354 C<cmp> to C<like> you would get SQL such as:
1356 WHERE name like 'nwiger' AND email like 'nate@wiger.org'
1358 You can also override the comparsion on an individual basis - see
1359 the huge section on L</"WHERE CLAUSES"> at the bottom.
1361 =item sqltrue, sqlfalse
1363 Expressions for inserting boolean values within SQL statements.
1364 By default these are C<1=1> and C<1=0>. They are used
1365 by the special operators C<-in> and C<-not_in> for generating
1366 correct SQL even when the argument is an empty array (see below).
1370 This determines the default logical operator for multiple WHERE
1371 statements in arrays or hashes. If absent, the default logic is "or"
1372 for arrays, and "and" for hashes. This means that a WHERE
1376 event_date => {'>=', '2/13/99'},
1377 event_date => {'<=', '4/24/03'},
1380 will generate SQL like this:
1382 WHERE event_date >= '2/13/99' OR event_date <= '4/24/03'
1384 This is probably not what you want given this query, though (look
1385 at the dates). To change the "OR" to an "AND", simply specify:
1387 my $sql = SQL::Abstract->new(logic => 'and');
1389 Which will change the above C<WHERE> to:
1391 WHERE event_date >= '2/13/99' AND event_date <= '4/24/03'
1393 The logic can also be changed locally by inserting
1394 a modifier in front of an arrayref :
1396 @where = (-and => [event_date => {'>=', '2/13/99'},
1397 event_date => {'<=', '4/24/03'} ]);
1399 See the L</"WHERE CLAUSES"> section for explanations.
1403 This will automatically convert comparisons using the specified SQL
1404 function for both column and value. This is mostly used with an argument
1405 of C<upper> or C<lower>, so that the SQL will have the effect of
1406 case-insensitive "searches". For example, this:
1408 $sql = SQL::Abstract->new(convert => 'upper');
1409 %where = (keywords => 'MaKe iT CAse inSeNSItive');
1411 Will turn out the following SQL:
1413 WHERE upper(keywords) like upper('MaKe iT CAse inSeNSItive')
1415 The conversion can be C<upper()>, C<lower()>, or any other SQL function
1416 that can be applied symmetrically to fields (actually B<SQL::Abstract> does
1417 not validate this option; it will just pass through what you specify verbatim).
1421 This is a kludge because many databases suck. For example, you can't
1422 just bind values using DBI's C<execute()> for Oracle C<CLOB> or C<BLOB> fields.
1423 Instead, you have to use C<bind_param()>:
1425 $sth->bind_param(1, 'reg data');
1426 $sth->bind_param(2, $lots, {ora_type => ORA_CLOB});
1428 The problem is, B<SQL::Abstract> will normally just return a C<@bind> array,
1429 which loses track of which field each slot refers to. Fear not.
1431 If you specify C<bindtype> in new, you can determine how C<@bind> is returned.
1432 Currently, you can specify either C<normal> (default) or C<columns>. If you
1433 specify C<columns>, you will get an array that looks like this:
1435 my $sql = SQL::Abstract->new(bindtype => 'columns');
1436 my($stmt, @bind) = $sql->insert(...);
1439 [ 'column1', 'value1' ],
1440 [ 'column2', 'value2' ],
1441 [ 'column3', 'value3' ],
1444 You can then iterate through this manually, using DBI's C<bind_param()>.
1446 $sth->prepare($stmt);
1449 my($col, $data) = @$_;
1450 if ($col eq 'details' || $col eq 'comments') {
1451 $sth->bind_param($i, $data, {ora_type => ORA_CLOB});
1452 } elsif ($col eq 'image') {
1453 $sth->bind_param($i, $data, {ora_type => ORA_BLOB});
1455 $sth->bind_param($i, $data);
1459 $sth->execute; # execute without @bind now
1461 Now, why would you still use B<SQL::Abstract> if you have to do this crap?
1462 Basically, the advantage is still that you don't have to care which fields
1463 are or are not included. You could wrap that above C<for> loop in a simple
1464 sub called C<bind_fields()> or something and reuse it repeatedly. You still
1465 get a layer of abstraction over manual SQL specification.
1467 Note that if you set L</bindtype> to C<columns>, the C<\[$sql, @bind]>
1468 construct (see L</Literal SQL with placeholders and bind values (subqueries)>)
1469 will expect the bind values in this format.
1473 This is the character that a table or column name will be quoted
1474 with. By default this is an empty string, but you could set it to
1475 the character C<`>, to generate SQL like this:
1477 SELECT `a_field` FROM `a_table` WHERE `some_field` LIKE '%someval%'
1479 Alternatively, you can supply an array ref of two items, the first being the left
1480 hand quote character, and the second the right hand quote character. For
1481 example, you could supply C<['[',']']> for SQL Server 2000 compliant quotes
1482 that generates SQL like this:
1484 SELECT [a_field] FROM [a_table] WHERE [some_field] LIKE '%someval%'
1486 Quoting is useful if you have tables or columns names that are reserved
1487 words in your database's SQL dialect.
1491 This is the character that separates a table and column name. It is
1492 necessary to specify this when the C<quote_char> option is selected,
1493 so that tables and column names can be individually quoted like this:
1495 SELECT `table`.`one_field` FROM `table` WHERE `table`.`other_field` = 1
1497 =item array_datatypes
1499 When this option is true, arrayrefs in INSERT or UPDATE are
1500 interpreted as array datatypes and are passed directly
1502 When this option is false, arrayrefs are interpreted
1503 as literal SQL, just like refs to arrayrefs
1504 (but this behavior is for backwards compatibility; when writing
1505 new queries, use the "reference to arrayref" syntax
1511 Takes a reference to a list of "special operators"
1512 to extend the syntax understood by L<SQL::Abstract>.
1513 See section L</"SPECIAL OPERATORS"> for details.
1519 =head2 insert($table, \@values || \%fieldvals)
1521 This is the simplest function. You simply give it a table name
1522 and either an arrayref of values or hashref of field/value pairs.
1523 It returns an SQL INSERT statement and a list of bind values.
1524 See the sections on L</"Inserting and Updating Arrays"> and
1525 L</"Inserting and Updating SQL"> for information on how to insert
1526 with those data types.
1528 =head2 update($table, \%fieldvals, \%where)
1530 This takes a table, hashref of field/value pairs, and an optional
1531 hashref L<WHERE clause|/WHERE CLAUSES>. It returns an SQL UPDATE function and a list
1533 See the sections on L</"Inserting and Updating Arrays"> and
1534 L</"Inserting and Updating SQL"> for information on how to insert
1535 with those data types.
1537 =head2 select($source, $fields, $where, $order)
1539 This returns a SQL SELECT statement and associated list of bind values, as
1540 specified by the arguments :
1546 Specification of the 'FROM' part of the statement.
1547 The argument can be either a plain scalar (interpreted as a table
1548 name, will be quoted), or an arrayref (interpreted as a list
1549 of table names, joined by commas, quoted), or a scalarref
1550 (literal table name, not quoted), or a ref to an arrayref
1551 (list of literal table names, joined by commas, not quoted).
1555 Specification of the list of fields to retrieve from
1557 The argument can be either an arrayref (interpreted as a list
1558 of field names, will be joined by commas and quoted), or a
1559 plain scalar (literal SQL, not quoted).
1560 Please observe that this API is not as flexible as for
1561 the first argument C<$table>, for backwards compatibility reasons.
1565 Optional argument to specify the WHERE part of the query.
1566 The argument is most often a hashref, but can also be
1567 an arrayref or plain scalar --
1568 see section L<WHERE clause|/"WHERE CLAUSES"> for details.
1572 Optional argument to specify the ORDER BY part of the query.
1573 The argument can be a scalar, a hashref or an arrayref
1574 -- see section L<ORDER BY clause|/"ORDER BY CLAUSES">
1580 =head2 delete($table, \%where)
1582 This takes a table name and optional hashref L<WHERE clause|/WHERE CLAUSES>.
1583 It returns an SQL DELETE statement and list of bind values.
1585 =head2 where(\%where, \@order)
1587 This is used to generate just the WHERE clause. For example,
1588 if you have an arbitrary data structure and know what the
1589 rest of your SQL is going to look like, but want an easy way
1590 to produce a WHERE clause, use this. It returns an SQL WHERE
1591 clause and list of bind values.
1594 =head2 values(\%data)
1596 This just returns the values from the hash C<%data>, in the same
1597 order that would be returned from any of the other above queries.
1598 Using this allows you to markedly speed up your queries if you
1599 are affecting lots of rows. See below under the L</"PERFORMANCE"> section.
1601 =head2 generate($any, 'number', $of, \@data, $struct, \%types)
1603 Warning: This is an experimental method and subject to change.
1605 This returns arbitrarily generated SQL. It's a really basic shortcut.
1606 It will return two different things, depending on return context:
1608 my($stmt, @bind) = $sql->generate('create table', \$table, \@fields);
1609 my $stmt_and_val = $sql->generate('create table', \$table, \@fields);
1611 These would return the following:
1613 # First calling form
1614 $stmt = "CREATE TABLE test (?, ?)";
1615 @bind = (field1, field2);
1617 # Second calling form
1618 $stmt_and_val = "CREATE TABLE test (field1, field2)";
1620 Depending on what you're trying to do, it's up to you to choose the correct
1621 format. In this example, the second form is what you would want.
1625 $sql->generate('alter session', { nls_date_format => 'MM/YY' });
1629 ALTER SESSION SET nls_date_format = 'MM/YY'
1631 You get the idea. Strings get their case twiddled, but everything
1632 else remains verbatim.
1637 =head1 WHERE CLAUSES
1641 This module uses a variation on the idea from L<DBIx::Abstract>. It
1642 is B<NOT>, repeat I<not> 100% compatible. B<The main logic of this
1643 module is that things in arrays are OR'ed, and things in hashes
1646 The easiest way to explain is to show lots of examples. After
1647 each C<%where> hash shown, it is assumed you used:
1649 my($stmt, @bind) = $sql->where(\%where);
1651 However, note that the C<%where> hash can be used directly in any
1652 of the other functions as well, as described above.
1654 =head2 Key-value pairs
1656 So, let's get started. To begin, a simple hash:
1660 status => 'completed'
1663 Is converted to SQL C<key = val> statements:
1665 $stmt = "WHERE user = ? AND status = ?";
1666 @bind = ('nwiger', 'completed');
1668 One common thing I end up doing is having a list of values that
1669 a field can be in. To do this, simply specify a list inside of
1674 status => ['assigned', 'in-progress', 'pending'];
1677 This simple code will create the following:
1679 $stmt = "WHERE user = ? AND ( status = ? OR status = ? OR status = ? )";
1680 @bind = ('nwiger', 'assigned', 'in-progress', 'pending');
1682 A field associated to an empty arrayref will be considered a
1683 logical false and will generate 0=1.
1685 =head2 Specific comparison operators
1687 If you want to specify a different type of operator for your comparison,
1688 you can use a hashref for a given column:
1692 status => { '!=', 'completed' }
1695 Which would generate:
1697 $stmt = "WHERE user = ? AND status != ?";
1698 @bind = ('nwiger', 'completed');
1700 To test against multiple values, just enclose the values in an arrayref:
1702 status => { '=', ['assigned', 'in-progress', 'pending'] };
1704 Which would give you:
1706 "WHERE status = ? OR status = ? OR status = ?"
1709 The hashref can also contain multiple pairs, in which case it is expanded
1710 into an C<AND> of its elements:
1714 status => { '!=', 'completed', -not_like => 'pending%' }
1717 # Or more dynamically, like from a form
1718 $where{user} = 'nwiger';
1719 $where{status}{'!='} = 'completed';
1720 $where{status}{'-not_like'} = 'pending%';
1722 # Both generate this
1723 $stmt = "WHERE user = ? AND status != ? AND status NOT LIKE ?";
1724 @bind = ('nwiger', 'completed', 'pending%');
1727 To get an OR instead, you can combine it with the arrayref idea:
1731 priority => [ {'=', 2}, {'!=', 1} ]
1734 Which would generate:
1736 $stmt = "WHERE user = ? AND priority = ? OR priority != ?";
1737 @bind = ('nwiger', '2', '1');
1739 If you want to include literal SQL (with or without bind values), just use a
1740 scalar reference or array reference as the value:
1743 date_entered => { '>' => \["to_date(?, 'MM/DD/YYYY')", "11/26/2008"] },
1744 date_expires => { '<' => \"now()" }
1747 Which would generate:
1749 $stmt = "WHERE date_entered > "to_date(?, 'MM/DD/YYYY') AND date_expires < now()";
1750 @bind = ('11/26/2008');
1753 =head2 Logic and nesting operators
1755 In the example above,
1756 there is a subtle trap if you want to say something like
1757 this (notice the C<AND>):
1759 WHERE priority != ? AND priority != ?
1761 Because, in Perl you I<can't> do this:
1763 priority => { '!=', 2, '!=', 1 }
1765 As the second C<!=> key will obliterate the first. The solution
1766 is to use the special C<-modifier> form inside an arrayref:
1768 priority => [ -and => {'!=', 2},
1772 Normally, these would be joined by C<OR>, but the modifier tells it
1773 to use C<AND> instead. (Hint: You can use this in conjunction with the
1774 C<logic> option to C<new()> in order to change the way your queries
1775 work by default.) B<Important:> Note that the C<-modifier> goes
1776 B<INSIDE> the arrayref, as an extra first element. This will
1777 B<NOT> do what you think it might:
1779 priority => -and => [{'!=', 2}, {'!=', 1}] # WRONG!
1781 Here is a quick list of equivalencies, since there is some overlap:
1784 status => {'!=', 'completed', 'not like', 'pending%' }
1785 status => [ -and => {'!=', 'completed'}, {'not like', 'pending%'}]
1788 status => {'=', ['assigned', 'in-progress']}
1789 status => [ -or => {'=', 'assigned'}, {'=', 'in-progress'}]
1790 status => [ {'=', 'assigned'}, {'=', 'in-progress'} ]
1794 =head2 Special operators : IN, BETWEEN, etc.
1796 You can also use the hashref format to compare a list of fields using the
1797 C<IN> comparison operator, by specifying the list as an arrayref:
1800 status => 'completed',
1801 reportid => { -in => [567, 2335, 2] }
1804 Which would generate:
1806 $stmt = "WHERE status = ? AND reportid IN (?,?,?)";
1807 @bind = ('completed', '567', '2335', '2');
1809 The reverse operator C<-not_in> generates SQL C<NOT IN> and is used in
1812 If the argument to C<-in> is an empty array, 'sqlfalse' is generated
1813 (by default : C<1=0>). Similarly, C<< -not_in => [] >> generates
1814 'sqltrue' (by default : C<1=1>).
1818 Another pair of operators is C<-between> and C<-not_between>,
1819 used with an arrayref of two values:
1823 completion_date => {
1824 -not_between => ['2002-10-01', '2003-02-06']
1830 WHERE user = ? AND completion_date NOT BETWEEN ( ? AND ? )
1832 These are the two builtin "special operators"; but the
1833 list can be expanded : see section L</"SPECIAL OPERATORS"> below.
1835 =head2 Nested conditions, -and/-or prefixes
1837 So far, we've seen how multiple conditions are joined with a top-level
1838 C<AND>. We can change this by putting the different conditions we want in
1839 hashes and then putting those hashes in an array. For example:
1844 status => { -like => ['pending%', 'dispatched'] },
1848 status => 'unassigned',
1852 This data structure would create the following:
1854 $stmt = "WHERE ( user = ? AND ( status LIKE ? OR status LIKE ? ) )
1855 OR ( user = ? AND status = ? ) )";
1856 @bind = ('nwiger', 'pending', 'dispatched', 'robot', 'unassigned');
1859 There is also a special C<-paren>
1860 operator which adds an additional set of parens, to create a subquery.
1861 For example, to get something like this:
1863 $stmt = "WHERE user = ? AND ( workhrs > ? OR geo = ? )";
1864 @bind = ('nwiger', '20', 'ASIA');
1870 -paren => [ workhrs => {'>', 20}, geo => 'ASIA' ],
1874 Finally, clauses in hashrefs or arrayrefs can be
1875 prefixed with an C<-and> or C<-or> to change the logic
1882 -and => [workhrs => {'>', 20}, geo => 'ASIA' ],
1883 -and => [workhrs => {'<', 50}, geo => 'EURO' ]
1890 WHERE ( user = ? AND
1891 ( ( workhrs > ? AND geo = ? )
1892 OR ( workhrs < ? AND geo = ? ) ) )
1895 =head2 Algebraic inconsistency, for historical reasons
1897 C<Important note>: when connecting several conditions, the C<-and->|C<-or>
1898 operator goes C<outside> of the nested structure; whereas when connecting
1899 several constraints on one column, the C<-and> operator goes
1900 C<inside> the arrayref. Here is an example combining both features :
1903 -and => [a => 1, b => 2],
1904 -or => [c => 3, d => 4],
1905 e => [-and => {-like => 'foo%'}, {-like => '%bar'} ]
1910 WHERE ( ( ( a = ? AND b = ? )
1911 OR ( c = ? OR d = ? )
1912 OR ( e LIKE ? AND e LIKE ? ) ) )
1914 This difference in syntax is unfortunate but must be preserved for
1915 historical reasons. So be careful : the two examples below would
1916 seem algebraically equivalent, but they are not
1918 {col => [-and => {-like => 'foo%'}, {-like => '%bar'}]}
1919 # yields : WHERE ( ( col LIKE ? AND col LIKE ? ) )
1921 [-and => {col => {-like => 'foo%'}, {col => {-like => '%bar'}}]]
1922 # yields : WHERE ( ( col LIKE ? OR col LIKE ? ) )
1927 Finally, sometimes only literal SQL will do. If you want to include
1928 literal SQL verbatim, you can specify it as a scalar reference, namely:
1930 my $inn = 'is Not Null';
1932 priority => { '<', 2 },
1938 $stmt = "WHERE priority < ? AND requestor is Not Null";
1941 Note that in this example, you only get one bind parameter back, since
1942 the verbatim SQL is passed as part of the statement.
1944 Of course, just to prove a point, the above can also be accomplished
1948 priority => { '<', 2 },
1949 requestor => { '!=', undef },
1955 Conditions on boolean columns can be expressed in the
1956 same way, passing a reference to an empty string :
1959 priority => { '<', 2 },
1965 $stmt = "WHERE priority < ? AND is_ready";
1969 =head2 Literal SQL with placeholders and bind values (subqueries)
1971 If the literal SQL to be inserted has placeholders and bind values,
1972 use a reference to an arrayref (yes this is a double reference --
1973 not so common, but perfectly legal Perl). For example, to find a date
1974 in Postgres you can use something like this:
1977 date_column => \[q/= date '2008-09-30' - ?::integer/, 10/]
1982 $stmt = "WHERE ( date_column = date '2008-09-30' - ?::integer )"
1985 Note that you must pass the bind values in the same format as they are returned
1986 by L</where>. That means that if you set L</bindtype> to C<columns>, you must
1987 provide the bind values in the C<< [ column_meta => value ] >> format, where
1988 C<column_meta> is an opaque scalar value; most commonly the column name, but
1989 you can use any scalar value (including references and blessed references),
1990 L<SQL::Abstract> will simply pass it through intact. So if C<bindtype> is set
1991 to C<columns> the above example will look like:
1994 date_column => \[q/= date '2008-09-30' - ?::integer/, [ dummy => 10 ]/]
1997 Literal SQL is especially useful for nesting parenthesized clauses in the
1998 main SQL query. Here is a first example :
2000 my ($sub_stmt, @sub_bind) = ("SELECT c1 FROM t1 WHERE c2 < ? AND c3 LIKE ?",
2004 bar => \["IN ($sub_stmt)" => @sub_bind],
2009 $stmt = "WHERE (foo = ? AND bar IN (SELECT c1 FROM t1
2010 WHERE c2 < ? AND c3 LIKE ?))";
2011 @bind = (1234, 100, "foo%");
2013 Other subquery operators, like for example C<"E<gt> ALL"> or C<"NOT IN">,
2014 are expressed in the same way. Of course the C<$sub_stmt> and
2015 its associated bind values can be generated through a former call
2018 my ($sub_stmt, @sub_bind)
2019 = $sql->select("t1", "c1", {c2 => {"<" => 100},
2020 c3 => {-like => "foo%"}});
2023 bar => \["> ALL ($sub_stmt)" => @sub_bind],
2026 In the examples above, the subquery was used as an operator on a column;
2027 but the same principle also applies for a clause within the main C<%where>
2028 hash, like an EXISTS subquery :
2030 my ($sub_stmt, @sub_bind)
2031 = $sql->select("t1", "*", {c1 => 1, c2 => \"> t0.c0"});
2034 -paren => \["EXISTS ($sub_stmt)" => @sub_bind],
2039 $stmt = "WHERE (foo = ? AND EXISTS (SELECT * FROM t1
2040 WHERE c1 = ? AND c2 > t0.c0))";
2044 Observe that the condition on C<c2> in the subquery refers to
2045 column C<t0.c0> of the main query : this is I<not> a bind
2046 value, so we have to express it through a scalar ref.
2047 Writing C<< c2 => {">" => "t0.c0"} >> would have generated
2048 C<< c2 > ? >> with bind value C<"t0.c0"> ... not exactly
2049 what we wanted here.
2051 Another use of the subquery technique is when some SQL clauses need
2052 parentheses, as it often occurs with some proprietary SQL extensions
2053 like for example fulltext expressions, geospatial expressions,
2054 NATIVE clauses, etc. Here is an example of a fulltext query in MySQL :
2057 -paren => \["MATCH (col1, col2) AGAINST (?)" => qw/apples/]
2060 Finally, here is an example where a subquery is used
2061 for expressing unary negation:
2063 my ($sub_stmt, @sub_bind)
2064 = $sql->where({age => [{"<" => 10}, {">" => 20}]});
2065 $sub_stmt =~ s/^ where //i; # don't want "WHERE" in the subclause
2067 lname => {like => '%son%'},
2068 -paren => \["NOT ($sub_stmt)" => @sub_bind],
2073 $stmt = "lname LIKE ? AND NOT ( age < ? OR age > ? )"
2074 @bind = ('%son%', 10, 20)
2080 These pages could go on for a while, since the nesting of the data
2081 structures this module can handle are pretty much unlimited (the
2082 module implements the C<WHERE> expansion as a recursive function
2083 internally). Your best bet is to "play around" with the module a
2084 little to see how the data structures behave, and choose the best
2085 format for your data based on that.
2087 And of course, all the values above will probably be replaced with
2088 variables gotten from forms or the command line. After all, if you
2089 knew everything ahead of time, you wouldn't have to worry about
2090 dynamically-generating SQL and could just hardwire it into your
2096 =head1 ORDER BY CLAUSES
2098 Some functions take an order by clause. This can either be a scalar (just a
2099 column name,) a hash of C<< { -desc => 'col' } >> or C<< { -asc => 'col' } >>,
2100 or an array of either of the two previous forms. Examples:
2102 Given | Will Generate
2103 ----------------------------------------------------------
2105 \'colA DESC' | ORDER BY colA DESC
2107 'colA' | ORDER BY colA
2109 [qw/colA colB/] | ORDER BY colA, colB
2111 {-asc => 'colA'} | ORDER BY colA ASC
2113 {-desc => 'colB'} | ORDER BY colB DESC
2115 ['colA', {-asc => 'colB'}] | ORDER BY colA, colB ASC
2117 { -asc => [qw/colA colB] } | ORDER BY colA ASC, colB ASC
2120 { -asc => 'colA' }, | ORDER BY colA ASC, colB DESC,
2121 { -desc => [qw/colB/], | colC ASC, colD ASC
2122 { -asc => [qw/colC colD/],|
2124 ===========================================================
2128 =head1 SPECIAL OPERATORS
2130 my $sqlmaker = SQL::Abstract->new(special_ops => [
2133 my ($self, $field, $op, $arg) = @_;
2139 A "special operator" is a SQL syntactic clause that can be
2140 applied to a field, instead of a usual binary operator.
2143 WHERE field IN (?, ?, ?)
2144 WHERE field BETWEEN ? AND ?
2145 WHERE MATCH(field) AGAINST (?, ?)
2147 Special operators IN and BETWEEN are fairly standard and therefore
2148 are builtin within C<SQL::Abstract>. For other operators,
2149 like the MATCH .. AGAINST example above which is
2150 specific to MySQL, you can write your own operator handlers :
2151 supply a C<special_ops> argument to the C<new> method.
2152 That argument takes an arrayref of operator definitions;
2153 each operator definition is a hashref with two entries
2159 the regular expression to match the operator
2163 coderef that will be called when meeting that operator
2164 in the input tree. The coderef will be called with
2165 arguments C<< ($self, $field, $op, $arg) >>, and
2166 should return a C<< ($sql, @bind) >> structure.
2170 For example, here is an implementation
2171 of the MATCH .. AGAINST syntax for MySQL
2173 my $sqlmaker = SQL::Abstract->new(special_ops => [
2175 # special op for MySql MATCH (field) AGAINST(word1, word2, ...)
2176 {regex => qr/^match$/i,
2178 my ($self, $field, $op, $arg) = @_;
2179 $arg = [$arg] if not ref $arg;
2180 my $label = $self->_quote($field);
2181 my ($placeholder) = $self->_convert('?');
2182 my $placeholders = join ", ", (($placeholder) x @$arg);
2183 my $sql = $self->_sqlcase('match') . " ($label) "
2184 . $self->_sqlcase('against') . " ($placeholders) ";
2185 my @bind = $self->_bindtype($field, @$arg);
2186 return ($sql, @bind);
2195 Thanks to some benchmarking by Mark Stosberg, it turns out that
2196 this module is many orders of magnitude faster than using C<DBIx::Abstract>.
2197 I must admit this wasn't an intentional design issue, but it's a
2198 byproduct of the fact that you get to control your C<DBI> handles
2201 To maximize performance, use a code snippet like the following:
2203 # prepare a statement handle using the first row
2204 # and then reuse it for the rest of the rows
2206 for my $href (@array_of_hashrefs) {
2207 $stmt ||= $sql->insert('table', $href);
2208 $sth ||= $dbh->prepare($stmt);
2209 $sth->execute($sql->values($href));
2212 The reason this works is because the keys in your C<$href> are sorted
2213 internally by B<SQL::Abstract>. Thus, as long as your data retains
2214 the same structure, you only have to generate the SQL the first time
2215 around. On subsequent queries, simply use the C<values> function provided
2216 by this module to return your values in the correct order.
2221 If you use my C<CGI::FormBuilder> module at all, you'll hopefully
2222 really like this part (I do, at least). Building up a complex query
2223 can be as simple as the following:
2227 use CGI::FormBuilder;
2230 my $form = CGI::FormBuilder->new(...);
2231 my $sql = SQL::Abstract->new;
2233 if ($form->submitted) {
2234 my $field = $form->field;
2235 my $id = delete $field->{id};
2236 my($stmt, @bind) = $sql->update('table', $field, {id => $id});
2239 Of course, you would still have to connect using C<DBI> to run the
2240 query, but the point is that if you make your form look like your
2241 table, the actual query script can be extremely simplistic.
2243 If you're B<REALLY> lazy (I am), check out C<HTML::QuickTable> for
2244 a fast interface to returning and formatting data. I frequently
2245 use these three modules together to write complex database query
2246 apps in under 50 lines.
2251 Version 1.50 was a major internal refactoring of C<SQL::Abstract>.
2252 Great care has been taken to preserve the I<published> behavior
2253 documented in previous versions in the 1.* family; however,
2254 some features that were previously undocumented, or behaved
2255 differently from the documentation, had to be changed in order
2256 to clarify the semantics. Hence, client code that was relying
2257 on some dark areas of C<SQL::Abstract> v1.*
2258 B<might behave differently> in v1.50.
2260 The main changes are :
2266 support for literal SQL through the C<< \ [$sql, bind] >> syntax.
2270 support for the { operator => \"..." } construct (to embed literal SQL)
2274 support for the { operator => \["...", @bind] } construct (to embed literal SQL with bind values)
2278 optional support for L<array datatypes|/"Inserting and Updating Arrays">
2282 defensive programming : check arguments
2286 fixed bug with global logic, which was previously implemented
2287 through global variables yielding side-effects. Prior versions would
2288 interpret C<< [ {cond1, cond2}, [cond3, cond4] ] >>
2289 as C<< "(cond1 AND cond2) OR (cond3 AND cond4)" >>.
2290 Now this is interpreted
2291 as C<< "(cond1 AND cond2) OR (cond3 OR cond4)" >>.
2296 fixed semantics of _bindtype on array args
2300 dropped the C<_anoncopy> of the %where tree. No longer necessary,
2301 we just avoid shifting arrays within that tree.
2305 dropped the C<_modlogic> function
2311 =head1 ACKNOWLEDGEMENTS
2313 There are a number of individuals that have really helped out with
2314 this module. Unfortunately, most of them submitted bugs via CPAN
2315 so I have no idea who they are! But the people I do know are:
2317 Ash Berlin (order_by hash term support)
2318 Matt Trout (DBIx::Class support)
2319 Mark Stosberg (benchmarking)
2320 Chas Owens (initial "IN" operator support)
2321 Philip Collins (per-field SQL functions)
2322 Eric Kolve (hashref "AND" support)
2323 Mike Fragassi (enhancements to "BETWEEN" and "LIKE")
2324 Dan Kubb (support for "quote_char" and "name_sep")
2325 Guillermo Roditi (patch to cleanup "IN" and "BETWEEN", fix and tests for _order_by)
2326 Laurent Dami (internal refactoring, multiple -paren, extensible list of special operators, literal SQL)
2327 Norbert Buchmuller (support for literal SQL in hashpair, misc. fixes & tests)
2328 Peter Rabbitson (rewrite of SQLA::Test, misc. fixes & tests)
2334 L<DBIx::Class>, L<DBIx::Abstract>, L<CGI::FormBuilder>, L<HTML::QuickTable>.
2338 Copyright (c) 2001-2007 Nathan Wiger <nwiger@cpan.org>. All Rights Reserved.
2340 This module is actively maintained by Matt Trout <mst@shadowcatsystems.co.uk>
2342 For support, your best bet is to try the C<DBIx::Class> users mailing list.
2343 While not an official support venue, C<DBIx::Class> makes heavy use of
2344 C<SQL::Abstract>, and as such list members there are very familiar with
2345 how to create queries.
2349 This module is free software; you may copy this under the terms of
2350 the GNU General Public License, or the Artistic License, copies of
2351 which should have accompanied your Perl kit.