1 package SQL::Abstract; # see doc at end of file
3 # LDNOTE : this code is heavy refactoring from original SQLA.
4 # Several design decisions will need discussion during
5 # the test / diffusion / acceptance phase; those are marked with flag
6 # 'LDNOTE' (note by laurent.dami AT free.fr)
11 use List::Util qw/first/;
12 use Scalar::Util qw/blessed/;
14 #======================================================================
16 #======================================================================
18 our $VERSION = '1.50';
20 # This would confuse some packagers
21 #$VERSION = eval $VERSION; # numify for warning-free dev releases
25 # special operators (-in, -between). May be extended/overridden by user.
26 # See section WHERE: BUILTIN SPECIAL OPERATORS below for implementation
27 my @BUILTIN_SPECIAL_OPS = (
28 {regex => qr/^(not )?between$/i, handler => \&_where_field_BETWEEN},
29 {regex => qr/^(not )?in$/i, handler => \&_where_field_IN},
32 #======================================================================
33 # DEBUGGING AND ERROR REPORTING
34 #======================================================================
37 return unless $_[0]->{debug}; shift; # a little faster
38 my $func = (caller(1))[3];
39 warn "[$func] ", @_, "\n";
43 my($func) = (caller(1))[3];
44 carp "[$func] Warning: ", @_;
48 my($func) = (caller(1))[3];
49 croak "[$func] Fatal: ", @_;
53 #======================================================================
55 #======================================================================
59 my $class = ref($self) || $self;
60 my %opt = (ref $_[0] eq 'HASH') ? %{$_[0]} : @_;
62 # choose our case by keeping an option around
63 delete $opt{case} if $opt{case} && $opt{case} ne 'lower';
65 # default logic for interpreting arrayrefs
66 $opt{logic} = uc $opt{logic} || 'OR';
68 # how to return bind vars
69 # LDNOTE: changed nwiger code : why this 'delete' ??
70 # $opt{bindtype} ||= delete($opt{bind_type}) || 'normal';
71 $opt{bindtype} ||= 'normal';
73 # default comparison is "=", but can be overridden
76 # try to recognize which are the 'equality' and 'unequality' ops
77 # (temporary quickfix, should go through a more seasoned API)
78 $opt{equality_op} = qr/^(\Q$opt{cmp}\E|is|(is\s+)?like)$/i;
79 $opt{inequality_op} = qr/^(!=|<>|(is\s+)?not(\s+like)?)$/i;
82 $opt{sqltrue} ||= '1=1';
83 $opt{sqlfalse} ||= '0=1';
86 $opt{special_ops} ||= [];
87 push @{$opt{special_ops}}, @BUILTIN_SPECIAL_OPS;
89 return bless \%opt, $class;
94 #======================================================================
96 #======================================================================
100 my $table = $self->_table(shift);
101 my $data = shift || return;
103 my $method = $self->_METHOD_FOR_refkind("_insert", $data);
104 my ($sql, @bind) = $self->$method($data);
105 $sql = join " ", $self->_sqlcase('insert into'), $table, $sql;
106 return wantarray ? ($sql, @bind) : $sql;
109 sub _insert_HASHREF { # explicit list of fields and then values
110 my ($self, $data) = @_;
112 my @fields = sort keys %$data;
114 my ($sql, @bind) = $self->_insert_values($data);
117 $_ = $self->_quote($_) foreach @fields;
118 $sql = "( ".join(", ", @fields).") ".$sql;
120 return ($sql, @bind);
123 sub _insert_ARRAYREF { # just generate values(?,?) part (no list of fields)
124 my ($self, $data) = @_;
126 # no names (arrayref) so can't generate bindtype
127 $self->{bindtype} ne 'columns'
128 or belch "can't do 'columns' bindtype when called with arrayref";
130 # fold the list of values into a hash of column name - value pairs
131 # (where the column names are artificially generated, and their
132 # lexicographical ordering keep the ordering of the original list)
133 my $i = "a"; # incremented values will be in lexicographical order
134 my $data_in_hash = { map { ($i++ => $_) } @$data };
136 return $self->_insert_values($data_in_hash);
139 sub _insert_ARRAYREFREF { # literal SQL with bind
140 my ($self, $data) = @_;
142 my ($sql, @bind) = @${$data};
143 $self->_assert_bindval_matches_bindtype(@bind);
145 return ($sql, @bind);
149 sub _insert_SCALARREF { # literal SQL without bind
150 my ($self, $data) = @_;
156 my ($self, $data) = @_;
158 my (@values, @all_bind);
159 foreach my $column (sort keys %$data) {
160 my $v = $data->{$column};
162 $self->_SWITCH_refkind($v, {
165 if ($self->{array_datatypes}) { # if array datatype are activated
167 push @all_bind, $self->_bindtype($column, $v);
169 else { # else literal SQL with bind
170 my ($sql, @bind) = @$v;
171 $self->_assert_bindval_matches_bindtype(@bind);
173 push @all_bind, @bind;
177 ARRAYREFREF => sub { # literal SQL with bind
178 my ($sql, @bind) = @${$v};
179 $self->_assert_bindval_matches_bindtype(@bind);
181 push @all_bind, @bind;
184 # THINK : anything useful to do with a HASHREF ?
185 HASHREF => sub { # (nothing, but old SQLA passed it through)
186 #TODO in SQLA >= 2.0 it will die instead
187 belch "HASH ref as bind value in insert is not supported";
189 push @all_bind, $self->_bindtype($column, $v);
192 SCALARREF => sub { # literal SQL without bind
196 SCALAR_or_UNDEF => sub {
198 push @all_bind, $self->_bindtype($column, $v);
205 my $sql = $self->_sqlcase('values')." ( ".join(", ", @values)." )";
206 return ($sql, @all_bind);
211 #======================================================================
213 #======================================================================
218 my $table = $self->_table(shift);
219 my $data = shift || return;
222 # first build the 'SET' part of the sql statement
223 my (@set, @all_bind);
224 puke "Unsupported data type specified to \$sql->update"
225 unless ref $data eq 'HASH';
227 for my $k (sort keys %$data) {
230 my $label = $self->_quote($k);
232 $self->_SWITCH_refkind($v, {
234 if ($self->{array_datatypes}) { # array datatype
235 push @set, "$label = ?";
236 push @all_bind, $self->_bindtype($k, $v);
238 else { # literal SQL with bind
239 my ($sql, @bind) = @$v;
240 $self->_assert_bindval_matches_bindtype(@bind);
241 push @set, "$label = $sql";
242 push @all_bind, @bind;
245 ARRAYREFREF => sub { # literal SQL with bind
246 my ($sql, @bind) = @${$v};
247 $self->_assert_bindval_matches_bindtype(@bind);
248 push @set, "$label = $sql";
249 push @all_bind, @bind;
251 SCALARREF => sub { # literal SQL without bind
252 push @set, "$label = $$v";
254 SCALAR_or_UNDEF => sub {
255 push @set, "$label = ?";
256 push @all_bind, $self->_bindtype($k, $v);
262 my $sql = $self->_sqlcase('update') . " $table " . $self->_sqlcase('set ')
266 my($where_sql, @where_bind) = $self->where($where);
268 push @all_bind, @where_bind;
271 return wantarray ? ($sql, @all_bind) : $sql;
277 #======================================================================
279 #======================================================================
284 my $table = $self->_table(shift);
285 my $fields = shift || '*';
289 my($where_sql, @bind) = $self->where($where, $order);
291 my $f = (ref $fields eq 'ARRAY') ? join ', ', map { $self->_quote($_) } @$fields
293 my $sql = join(' ', $self->_sqlcase('select'), $f,
294 $self->_sqlcase('from'), $table)
297 return wantarray ? ($sql, @bind) : $sql;
300 #======================================================================
302 #======================================================================
307 my $table = $self->_table(shift);
311 my($where_sql, @bind) = $self->where($where);
312 my $sql = $self->_sqlcase('delete from') . " $table" . $where_sql;
314 return wantarray ? ($sql, @bind) : $sql;
318 #======================================================================
320 #======================================================================
324 # Finally, a separate routine just to handle WHERE clauses
326 my ($self, $where, $order) = @_;
329 my ($sql, @bind) = $self->_recurse_where($where);
330 $sql = $sql ? $self->_sqlcase(' where ') . "( $sql )" : '';
334 $sql .= $self->_order_by($order);
337 return wantarray ? ($sql, @bind) : $sql;
342 my ($self, $where, $logic) = @_;
344 # dispatch on appropriate method according to refkind of $where
345 my $method = $self->_METHOD_FOR_refkind("_where", $where);
348 my ($sql, @bind) = $self->$method($where, $logic);
350 # DBIx::Class directly calls _recurse_where in scalar context, so
351 # we must implement it, even if not in the official API
352 return wantarray ? ($sql, @bind) : $sql;
357 #======================================================================
358 # WHERE: top-level ARRAYREF
359 #======================================================================
362 sub _where_ARRAYREF {
363 my ($self, $where, $logic) = @_;
365 $logic = uc($logic || $self->{logic});
366 $logic eq 'AND' or $logic eq 'OR' or puke "unknown logic: $logic";
368 my @clauses = @$where;
370 # if the array starts with [-and|or => ...], recurse with that logic
371 my $first = $clauses[0] || '';
372 if ($first =~ /^-(and|or)/i) {
375 return $self->_where_ARRAYREF(\@clauses, $logic);
379 my (@sql_clauses, @all_bind);
381 # need to use while() so can shift() for pairs
382 while (my $el = shift @clauses) {
384 # switch according to kind of $el and get corresponding ($sql, @bind)
385 my ($sql, @bind) = $self->_SWITCH_refkind($el, {
387 # skip empty elements, otherwise get invalid trailing AND stuff
388 ARRAYREF => sub {$self->_recurse_where($el) if @$el},
390 ARRAYREFREF => sub { @{${$el}} if @{${$el}}},
392 HASHREF => sub {$self->_recurse_where($el, 'and') if %$el},
393 # LDNOTE : previous SQLA code for hashrefs was creating a dirty
394 # side-effect: the first hashref within an array would change
395 # the global logic to 'AND'. So [ {cond1, cond2}, [cond3, cond4] ]
396 # was interpreted as "(cond1 AND cond2) OR (cond3 AND cond4)",
397 # whereas it should be "(cond1 AND cond2) OR (cond3 OR cond4)".
399 SCALARREF => sub { ($$el); },
401 SCALAR => sub {# top-level arrayref with scalars, recurse in pairs
402 $self->_recurse_where({$el => shift(@clauses)})},
404 UNDEF => sub {puke "not supported : UNDEF in arrayref" },
408 push @sql_clauses, $sql;
409 push @all_bind, @bind;
413 return $self->_join_sql_clauses($logic, \@sql_clauses, \@all_bind);
416 #======================================================================
417 # WHERE: top-level ARRAYREFREF
418 #======================================================================
420 sub _where_ARRAYREFREF {
421 my ($self, $where) = @_;
422 my ($sql, @bind) = @{${$where}};
424 return ($sql, @bind);
427 #======================================================================
428 # WHERE: top-level HASHREF
429 #======================================================================
432 my ($self, $where) = @_;
433 my (@sql_clauses, @all_bind);
435 # LDNOTE : don't really know why we need to sort keys
436 for my $k (sort keys %$where) {
437 my $v = $where->{$k};
439 # ($k => $v) is either a special op or a regular hashpair
440 my ($sql, @bind) = ($k =~ /^-(.+)/) ? $self->_where_op_in_hash($1, $v)
442 my $method = $self->_METHOD_FOR_refkind("_where_hashpair", $v);
443 $self->$method($k, $v);
446 push @sql_clauses, $sql;
447 push @all_bind, @bind;
450 return $self->_join_sql_clauses('and', \@sql_clauses, \@all_bind);
454 sub _where_op_in_hash {
455 my ($self, $op, $v) = @_;
457 $op =~ /^(AND|OR|NEST)[_\d]*/i
458 or puke "unknown operator: -$op";
459 $op = uc($1); # uppercase, remove trailing digits
460 $self->_debug("OP(-$op) within hashref, recursing...");
462 $self->_SWITCH_refkind($v, {
465 # LDNOTE : should deprecate {-or => [...]} and {-and => [...]}
466 # because they are misleading; the only proper way would be
467 # -nest => [-or => ...], -nest => [-and ...]
468 return $self->_where_ARRAYREF($v, $op eq 'NEST' ? '' : $op);
473 belch "-or => {...} should be -nest => [...]";
474 return $self->_where_ARRAYREF([%$v], 'OR');
477 return $self->_where_HASHREF($v);
481 SCALARREF => sub { # literal SQL
483 or puke "-$op => \\\$scalar not supported, use -nest => ...";
487 ARRAYREFREF => sub { # literal SQL
489 or puke "-$op => \\[..] not supported, use -nest => ...";
493 SCALAR => sub { # permissively interpreted as SQL
495 or puke "-$op => 'scalar' not supported, use -nest => \\'scalar'";
496 belch "literal SQL should be -nest => \\'scalar' "
497 . "instead of -nest => 'scalar' ";
502 puke "-$op => undef not supported";
508 sub _where_hashpair_ARRAYREF {
509 my ($self, $k, $v) = @_;
512 my @v = @$v; # need copy because of shift below
513 $self->_debug("ARRAY($k) means distribute over elements");
515 # put apart first element if it is an operator (-and, -or)
516 my $op = $v[0] =~ /^-/ ? shift @v : undef;
517 $self->_debug("OP($op) reinjected into the distributed array") if $op;
519 my @distributed = map { {$k => $_} } @v;
520 unshift @distributed, $op if $op;
522 return $self->_recurse_where(\@distributed);
525 # LDNOTE : not sure of this one. What does "distribute over nothing" mean?
526 $self->_debug("empty ARRAY($k) means 0=1");
527 return ($self->{sqlfalse});
531 sub _where_hashpair_HASHREF {
532 my ($self, $k, $v) = @_;
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) = @$$val;
568 $self->_assert_bindval_matches_bindtype(@sub_bind);
569 $sql = join ' ', $self->_convert($self->_quote($k)),
570 $self->_sqlcase($op),
575 UNDEF => sub { # CASE: col => {op => undef} : sql "IS (NOT)? NULL"
576 my $is = ($op =~ $self->{equality_op}) ? 'is' :
577 ($op =~ $self->{inequality_op}) ? 'is not' :
578 puke "unexpected operator '$op' with undef operand";
579 $sql = $self->_quote($k) . $self->_sqlcase(" $is null");
582 FALLBACK => sub { # CASE: col => {op => $scalar}
583 $sql = join ' ', $self->_convert($self->_quote($k)),
584 $self->_sqlcase($op),
585 $self->_convert('?');
586 @bind = $self->_bindtype($k, $val);
592 push @all_bind, @bind;
595 return $self->_join_sql_clauses('and', \@all_sql, \@all_bind);
600 sub _where_field_op_ARRAYREF {
601 my ($self, $k, $op, $vals) = @_;
604 $self->_debug("ARRAY($vals) means multiple elements: [ @$vals ]");
608 # LDNOTE : change the distribution logic when
609 # $op =~ $self->{inequality_op}, because of Morgan laws :
610 # with {field => {'!=' => [22, 33]}}, it would be ridiculous to generate
611 # WHERE field != 22 OR field != 33 : the user probably means
612 # WHERE field != 22 AND field != 33.
613 my $logic = ($op =~ $self->{inequality_op}) ? 'AND' : 'OR';
615 # distribute $op over each member of @$vals
616 return $self->_recurse_where([map { {$k => {$op, $_}} } @$vals], $logic);
620 # try to DWIM on equality operators
621 # LDNOTE : not 100% sure this is the correct thing to do ...
622 return ($self->{sqlfalse}) if $op =~ $self->{equality_op};
623 return ($self->{sqltrue}) if $op =~ $self->{inequality_op};
626 puke "operator '$op' applied on an empty array (field '$k')";
631 sub _where_hashpair_SCALARREF {
632 my ($self, $k, $v) = @_;
633 $self->_debug("SCALAR($k) means literal SQL: $$v");
634 my $sql = $self->_quote($k) . " " . $$v;
638 # literal SQL with bind
639 sub _where_hashpair_ARRAYREFREF {
640 my ($self, $k, $v) = @_;
641 $self->_debug("REF($k) means literal SQL: @${$v}");
642 my ($sql, @bind) = @${$v};
643 $self->_assert_bindval_matches_bindtype(@bind);
644 $sql = $self->_quote($k) . " " . $sql;
645 return ($sql, @bind );
648 # literal SQL without bind
649 sub _where_hashpair_SCALAR {
650 my ($self, $k, $v) = @_;
651 $self->_debug("NOREF($k) means simple key=val: $k $self->{cmp} $v");
652 my $sql = join ' ', $self->_convert($self->_quote($k)),
653 $self->_sqlcase($self->{cmp}),
654 $self->_convert('?');
655 my @bind = $self->_bindtype($k, $v);
656 return ( $sql, @bind);
660 sub _where_hashpair_UNDEF {
661 my ($self, $k, $v) = @_;
662 $self->_debug("UNDEF($k) means IS NULL");
663 my $sql = $self->_quote($k) . $self->_sqlcase(' is null');
667 #======================================================================
668 # WHERE: TOP-LEVEL OTHERS (SCALARREF, SCALAR, UNDEF)
669 #======================================================================
672 sub _where_SCALARREF {
673 my ($self, $where) = @_;
676 $self->_debug("SCALAR(*top) means literal SQL: $$where");
682 my ($self, $where) = @_;
685 $self->_debug("NOREF(*top) means literal SQL: $where");
696 #======================================================================
697 # WHERE: BUILTIN SPECIAL OPERATORS (-in, -between)
698 #======================================================================
701 sub _where_field_BETWEEN {
702 my ($self, $k, $op, $vals) = @_;
704 ref $vals eq 'ARRAY' && @$vals == 2
705 or puke "special op 'between' requires an arrayref of two values";
707 my ($label) = $self->_convert($self->_quote($k));
708 my ($placeholder) = $self->_convert('?');
709 my $and = $self->_sqlcase('and');
710 $op = $self->_sqlcase($op);
712 my $sql = "( $label $op $placeholder $and $placeholder )";
713 my @bind = $self->_bindtype($k, @$vals);
718 sub _where_field_IN {
719 my ($self, $k, $op, $vals) = @_;
721 # backwards compatibility : if scalar, force into an arrayref
722 $vals = [$vals] if defined $vals && ! ref $vals;
724 my ($label) = $self->_convert($self->_quote($k));
725 my ($placeholder) = $self->_convert('?');
726 $op = $self->_sqlcase($op);
728 my ($sql, @bind) = $self->_SWITCH_refkind($vals, {
729 ARRAYREF => sub { # list of choices
730 if (@$vals) { # nonempty list
731 my $placeholders = join ", ", (($placeholder) x @$vals);
732 my $sql = "$label $op ( $placeholders )";
733 my @bind = $self->_bindtype($k, @$vals);
735 return ($sql, @bind);
737 else { # empty list : some databases won't understand "IN ()", so DWIM
738 my $sql = ($op =~ /\bnot\b/i) ? $self->{sqltrue} : $self->{sqlfalse};
743 ARRAYREFREF => sub { # literal SQL with bind
744 my ($sql, @bind) = @$$vals;
745 $self->_assert_bindval_matches_bindtype(@bind);
746 return ("$label $op ( $sql )", @bind);
750 puke "special op 'in' requires an arrayref (or arrayref-ref)";
754 return ($sql, @bind);
762 #======================================================================
764 #======================================================================
767 my ($self, $arg) = @_;
769 # construct list of ordering instructions
770 my @order = $self->_SWITCH_refkind($arg, {
773 map {$self->_SWITCH_refkind($_, {
774 SCALAR => sub {$self->_quote($_)},
776 SCALARREF => sub {$$_}, # literal SQL, no quoting
777 HASHREF => sub {$self->_order_by_hash($_)}
781 SCALAR => sub {$self->_quote($arg)},
783 SCALARREF => sub {$$arg}, # literal SQL, no quoting
784 HASHREF => sub {$self->_order_by_hash($arg)},
789 my $order = join ', ', @order;
790 return $order ? $self->_sqlcase(' order by')." $order" : '';
795 my ($self, $hash) = @_;
797 # get first pair in hash
798 my ($key, $val) = each %$hash;
800 # check if one pair was found and no other pair in hash
801 $key && !(each %$hash)
802 or puke "hash passed to _order_by must have exactly one key (-desc or -asc)";
804 my ($order) = ($key =~ /^-(desc|asc)/i)
805 or puke "invalid key in _order_by hash : $key";
807 return $self->_quote($val) ." ". $self->_sqlcase($order);
812 #======================================================================
813 # DATASOURCE (FOR NOW, JUST PLAIN TABLE OR LIST OF TABLES)
814 #======================================================================
819 $self->_SWITCH_refkind($from, {
820 ARRAYREF => sub {join ', ', map { $self->_quote($_) } @$from;},
821 SCALAR => sub {$self->_quote($from)},
822 SCALARREF => sub {$$from},
823 ARRAYREFREF => sub {join ', ', @$from;},
828 #======================================================================
830 #======================================================================
836 $label or puke "can't quote an empty label";
838 # left and right quote characters
839 my ($ql, $qr, @other) = $self->_SWITCH_refkind($self->{quote_char}, {
840 SCALAR => sub {($self->{quote_char}, $self->{quote_char})},
841 ARRAYREF => sub {@{$self->{quote_char}}},
845 or puke "quote_char must be an arrayref of 2 values";
847 # no quoting if no quoting chars
848 $ql or return $label;
850 # no quoting for literal SQL
851 return $$label if ref($label) eq 'SCALAR';
853 # separate table / column (if applicable)
854 my $sep = $self->{name_sep} || '';
855 my @to_quote = $sep ? split /\Q$sep\E/, $label : ($label);
857 # do the quoting, except for "*" or for `table`.*
858 my @quoted = map { $_ eq '*' ? $_: $ql.$_.$qr} @to_quote;
860 # reassemble and return.
861 return join $sep, @quoted;
865 # Conversion, if applicable
867 my ($self, $arg) = @_;
869 # LDNOTE : modified the previous implementation below because
870 # it was not consistent : the first "return" is always an array,
871 # the second "return" is context-dependent. Anyway, _convert
872 # seems always used with just a single argument, so make it a
874 # return @_ unless $self->{convert};
875 # my $conv = $self->_sqlcase($self->{convert});
876 # my @ret = map { $conv.'('.$_.')' } @_;
877 # return wantarray ? @ret : $ret[0];
878 if ($self->{convert}) {
879 my $conv = $self->_sqlcase($self->{convert});
880 $arg = $conv.'('.$arg.')';
888 my($col, @vals) = @_;
890 #LDNOTE : changed original implementation below because it did not make
891 # sense when bindtype eq 'columns' and @vals > 1.
892 # return $self->{bindtype} eq 'columns' ? [ $col, @vals ] : @vals;
894 return $self->{bindtype} eq 'columns' ? map {[$col, $_]} @vals : @vals;
897 # Dies if any element of @bind is not in [colname => value] format
898 # if bindtype is 'columns'.
899 sub _assert_bindval_matches_bindtype {
900 my ($self, @bind) = @_;
902 if ($self->{bindtype} eq 'columns') {
903 foreach my $val (@bind) {
904 if (!defined $val || ref($val) ne 'ARRAY' || @$val != 2) {
905 die "bindtype 'columns' selected, you need to pass: [column_name => bind_value]"
911 sub _join_sql_clauses {
912 my ($self, $logic, $clauses_aref, $bind_aref) = @_;
914 if (@$clauses_aref > 1) {
915 my $join = " " . $self->_sqlcase($logic) . " ";
916 my $sql = '( ' . join($join, @$clauses_aref) . ' )';
917 return ($sql, @$bind_aref);
919 elsif (@$clauses_aref) {
920 return ($clauses_aref->[0], @$bind_aref); # no parentheses
923 return (); # if no SQL, ignore @$bind_aref
928 # Fix SQL case, if so requested
932 # LDNOTE: if $self->{case} is true, then it contains 'lower', so we
933 # don't touch the argument ... crooked logic, but let's not change it!
934 return $self->{case} ? $_[0] : uc($_[0]);
938 #======================================================================
939 # DISPATCHING FROM REFKIND
940 #======================================================================
943 my ($self, $data) = @_;
949 # blessed objects are treated like scalars
950 $ref = (blessed $data) ? '' : ref $data;
951 $n_steps += 1 if $ref;
952 last if $ref ne 'REF';
956 my $base = $ref || (defined $data ? 'SCALAR' : 'UNDEF');
958 return $base . ('REF' x $n_steps);
964 my ($self, $data) = @_;
965 my @try = ($self->_refkind($data));
966 push @try, 'SCALAR_or_UNDEF' if $try[0] eq 'SCALAR' || $try[0] eq 'UNDEF';
967 push @try, 'FALLBACK';
971 sub _METHOD_FOR_refkind {
972 my ($self, $meth_prefix, $data) = @_;
973 my $method = first {$_} map {$self->can($meth_prefix."_".$_)}
974 $self->_try_refkind($data)
975 or puke "cannot dispatch on '$meth_prefix' for ".$self->_refkind($data);
980 sub _SWITCH_refkind {
981 my ($self, $data, $dispatch_table) = @_;
983 my $coderef = first {$_} map {$dispatch_table->{$_}}
984 $self->_try_refkind($data)
985 or puke "no dispatch entry for ".$self->_refkind($data);
992 #======================================================================
993 # VALUES, GENERATE, AUTOLOAD
994 #======================================================================
996 # LDNOTE: original code from nwiger, didn't touch code in that section
997 # I feel the AUTOLOAD stuff should not be the default, it should
998 # only be activated on explicit demand by user.
1002 my $data = shift || return;
1003 puke "Argument to ", __PACKAGE__, "->values must be a \\%hash"
1004 unless ref $data eq 'HASH';
1007 foreach my $k ( sort keys %$data ) {
1008 my $v = $data->{$k};
1009 $self->_SWITCH_refkind($v, {
1011 if ($self->{array_datatypes}) { # array datatype
1012 push @all_bind, $self->_bindtype($k, $v);
1014 else { # literal SQL with bind
1015 my ($sql, @bind) = @$v;
1016 $self->_assert_bindval_matches_bindtype(@bind);
1017 push @all_bind, @bind;
1020 ARRAYREFREF => sub { # literal SQL with bind
1021 my ($sql, @bind) = @${$v};
1022 $self->_assert_bindval_matches_bindtype(@bind);
1023 push @all_bind, @bind;
1025 SCALARREF => sub { # literal SQL without bind
1027 SCALAR_or_UNDEF => sub {
1028 push @all_bind, $self->_bindtype($k, $v);
1039 my(@sql, @sqlq, @sqlv);
1043 if ($ref eq 'HASH') {
1044 for my $k (sort keys %$_) {
1047 my $label = $self->_quote($k);
1048 if ($r eq 'ARRAY') {
1049 # literal SQL with bind
1050 my ($sql, @bind) = @$v;
1051 $self->_assert_bindval_matches_bindtype(@bind);
1052 push @sqlq, "$label = $sql";
1054 } elsif ($r eq 'SCALAR') {
1055 # literal SQL without bind
1056 push @sqlq, "$label = $$v";
1058 push @sqlq, "$label = ?";
1059 push @sqlv, $self->_bindtype($k, $v);
1062 push @sql, $self->_sqlcase('set'), join ', ', @sqlq;
1063 } elsif ($ref eq 'ARRAY') {
1064 # unlike insert(), assume these are ONLY the column names, i.e. for SQL
1067 if ($r eq 'ARRAY') { # literal SQL with bind
1068 my ($sql, @bind) = @$v;
1069 $self->_assert_bindval_matches_bindtype(@bind);
1072 } elsif ($r eq 'SCALAR') { # literal SQL without bind
1073 # embedded literal SQL
1080 push @sql, '(' . join(', ', @sqlq) . ')';
1081 } elsif ($ref eq 'SCALAR') {
1085 # strings get case twiddled
1086 push @sql, $self->_sqlcase($_);
1090 my $sql = join ' ', @sql;
1092 # this is pretty tricky
1093 # if ask for an array, return ($stmt, @bind)
1094 # otherwise, s/?/shift @sqlv/ to put it inline
1096 return ($sql, @sqlv);
1098 1 while $sql =~ s/\?/my $d = shift(@sqlv);
1099 ref $d ? $d->[1] : $d/e;
1108 # This allows us to check for a local, then _form, attr
1110 my($name) = $AUTOLOAD =~ /.*::(.+)/;
1111 return $self->generate($name, @_);
1122 SQL::Abstract - Generate SQL from Perl data structures
1128 my $sql = SQL::Abstract->new;
1130 my($stmt, @bind) = $sql->select($table, \@fields, \%where, \@order);
1132 my($stmt, @bind) = $sql->insert($table, \%fieldvals || \@values);
1134 my($stmt, @bind) = $sql->update($table, \%fieldvals, \%where);
1136 my($stmt, @bind) = $sql->delete($table, \%where);
1138 # Then, use these in your DBI statements
1139 my $sth = $dbh->prepare($stmt);
1140 $sth->execute(@bind);
1142 # Just generate the WHERE clause
1143 my($stmt, @bind) = $sql->where(\%where, \@order);
1145 # Return values in the same order, for hashed queries
1146 # See PERFORMANCE section for more details
1147 my @bind = $sql->values(\%fieldvals);
1151 This module was inspired by the excellent L<DBIx::Abstract>.
1152 However, in using that module I found that what I really wanted
1153 to do was generate SQL, but still retain complete control over my
1154 statement handles and use the DBI interface. So, I set out to
1155 create an abstract SQL generation module.
1157 While based on the concepts used by L<DBIx::Abstract>, there are
1158 several important differences, especially when it comes to WHERE
1159 clauses. I have modified the concepts used to make the SQL easier
1160 to generate from Perl data structures and, IMO, more intuitive.
1161 The underlying idea is for this module to do what you mean, based
1162 on the data structures you provide it. The big advantage is that
1163 you don't have to modify your code every time your data changes,
1164 as this module figures it out.
1166 To begin with, an SQL INSERT is as easy as just specifying a hash
1167 of C<key=value> pairs:
1170 name => 'Jimbo Bobson',
1171 phone => '123-456-7890',
1172 address => '42 Sister Lane',
1173 city => 'St. Louis',
1174 state => 'Louisiana',
1177 The SQL can then be generated with this:
1179 my($stmt, @bind) = $sql->insert('people', \%data);
1181 Which would give you something like this:
1183 $stmt = "INSERT INTO people
1184 (address, city, name, phone, state)
1185 VALUES (?, ?, ?, ?, ?)";
1186 @bind = ('42 Sister Lane', 'St. Louis', 'Jimbo Bobson',
1187 '123-456-7890', 'Louisiana');
1189 These are then used directly in your DBI code:
1191 my $sth = $dbh->prepare($stmt);
1192 $sth->execute(@bind);
1194 =head2 Inserting and Updating Arrays
1196 If your database has array types (like for example Postgres),
1197 activate the special option C<< array_datatypes => 1 >>
1198 when creating the C<SQL::Abstract> object.
1199 Then you may use an arrayref to insert and update database array types:
1201 my $sql = SQL::Abstract->new(array_datatypes => 1);
1203 planets => [qw/Mercury Venus Earth Mars/]
1206 my($stmt, @bind) = $sql->insert('solar_system', \%data);
1210 $stmt = "INSERT INTO solar_system (planets) VALUES (?)"
1212 @bind = (['Mercury', 'Venus', 'Earth', 'Mars']);
1215 =head2 Inserting and Updating SQL
1217 In order to apply SQL functions to elements of your C<%data> you may
1218 specify a reference to an arrayref for the given hash value. For example,
1219 if you need to execute the Oracle C<to_date> function on a value, you can
1220 say something like this:
1224 date_entered => \["to_date(?,'MM/DD/YYYY')", "03/02/2003"],
1227 The first value in the array is the actual SQL. Any other values are
1228 optional and would be included in the bind values array. This gives
1231 my($stmt, @bind) = $sql->insert('people', \%data);
1233 $stmt = "INSERT INTO people (name, date_entered)
1234 VALUES (?, to_date(?,'MM/DD/YYYY'))";
1235 @bind = ('Bill', '03/02/2003');
1237 An UPDATE is just as easy, all you change is the name of the function:
1239 my($stmt, @bind) = $sql->update('people', \%data);
1241 Notice that your C<%data> isn't touched; the module will generate
1242 the appropriately quirky SQL for you automatically. Usually you'll
1243 want to specify a WHERE clause for your UPDATE, though, which is
1244 where handling C<%where> hashes comes in handy...
1246 =head2 Complex where statements
1248 This module can generate pretty complicated WHERE statements
1249 easily. For example, simple C<key=value> pairs are taken to mean
1250 equality, and if you want to see if a field is within a set
1251 of values, you can use an arrayref. Let's say we wanted to
1252 SELECT some data based on this criteria:
1255 requestor => 'inna',
1256 worker => ['nwiger', 'rcwe', 'sfz'],
1257 status => { '!=', 'completed' }
1260 my($stmt, @bind) = $sql->select('tickets', '*', \%where);
1262 The above would give you something like this:
1264 $stmt = "SELECT * FROM tickets WHERE
1265 ( requestor = ? ) AND ( status != ? )
1266 AND ( worker = ? OR worker = ? OR worker = ? )";
1267 @bind = ('inna', 'completed', 'nwiger', 'rcwe', 'sfz');
1269 Which you could then use in DBI code like so:
1271 my $sth = $dbh->prepare($stmt);
1272 $sth->execute(@bind);
1278 The functions are simple. There's one for each major SQL operation,
1279 and a constructor you use first. The arguments are specified in a
1280 similar order to each function (table, then fields, then a where
1281 clause) to try and simplify things.
1286 =head2 new(option => 'value')
1288 The C<new()> function takes a list of options and values, and returns
1289 a new B<SQL::Abstract> object which can then be used to generate SQL
1290 through the methods below. The options accepted are:
1296 If set to 'lower', then SQL will be generated in all lowercase. By
1297 default SQL is generated in "textbook" case meaning something like:
1299 SELECT a_field FROM a_table WHERE some_field LIKE '%someval%'
1301 Any setting other than 'lower' is ignored.
1305 This determines what the default comparison operator is. By default
1306 it is C<=>, meaning that a hash like this:
1308 %where = (name => 'nwiger', email => 'nate@wiger.org');
1310 Will generate SQL like this:
1312 WHERE name = 'nwiger' AND email = 'nate@wiger.org'
1314 However, you may want loose comparisons by default, so if you set
1315 C<cmp> to C<like> you would get SQL such as:
1317 WHERE name like 'nwiger' AND email like 'nate@wiger.org'
1319 You can also override the comparsion on an individual basis - see
1320 the huge section on L</"WHERE CLAUSES"> at the bottom.
1322 =item sqltrue, sqlfalse
1324 Expressions for inserting boolean values within SQL statements.
1325 By default these are C<1=1> and C<1=0>.
1329 This determines the default logical operator for multiple WHERE
1330 statements in arrays. By default it is "or", meaning that a WHERE
1334 event_date => {'>=', '2/13/99'},
1335 event_date => {'<=', '4/24/03'},
1338 Will generate SQL like this:
1340 WHERE event_date >= '2/13/99' OR event_date <= '4/24/03'
1342 This is probably not what you want given this query, though (look
1343 at the dates). To change the "OR" to an "AND", simply specify:
1345 my $sql = SQL::Abstract->new(logic => 'and');
1347 Which will change the above C<WHERE> to:
1349 WHERE event_date >= '2/13/99' AND event_date <= '4/24/03'
1351 The logic can also be changed locally by inserting
1352 an extra first element in the array :
1354 @where = (-and => event_date => {'>=', '2/13/99'},
1355 event_date => {'<=', '4/24/03'} );
1357 See the L</"WHERE CLAUSES"> section for explanations.
1361 This will automatically convert comparisons using the specified SQL
1362 function for both column and value. This is mostly used with an argument
1363 of C<upper> or C<lower>, so that the SQL will have the effect of
1364 case-insensitive "searches". For example, this:
1366 $sql = SQL::Abstract->new(convert => 'upper');
1367 %where = (keywords => 'MaKe iT CAse inSeNSItive');
1369 Will turn out the following SQL:
1371 WHERE upper(keywords) like upper('MaKe iT CAse inSeNSItive')
1373 The conversion can be C<upper()>, C<lower()>, or any other SQL function
1374 that can be applied symmetrically to fields (actually B<SQL::Abstract> does
1375 not validate this option; it will just pass through what you specify verbatim).
1379 This is a kludge because many databases suck. For example, you can't
1380 just bind values using DBI's C<execute()> for Oracle C<CLOB> or C<BLOB> fields.
1381 Instead, you have to use C<bind_param()>:
1383 $sth->bind_param(1, 'reg data');
1384 $sth->bind_param(2, $lots, {ora_type => ORA_CLOB});
1386 The problem is, B<SQL::Abstract> will normally just return a C<@bind> array,
1387 which loses track of which field each slot refers to. Fear not.
1389 If you specify C<bindtype> in new, you can determine how C<@bind> is returned.
1390 Currently, you can specify either C<normal> (default) or C<columns>. If you
1391 specify C<columns>, you will get an array that looks like this:
1393 my $sql = SQL::Abstract->new(bindtype => 'columns');
1394 my($stmt, @bind) = $sql->insert(...);
1397 [ 'column1', 'value1' ],
1398 [ 'column2', 'value2' ],
1399 [ 'column3', 'value3' ],
1402 You can then iterate through this manually, using DBI's C<bind_param()>.
1404 $sth->prepare($stmt);
1407 my($col, $data) = @$_;
1408 if ($col eq 'details' || $col eq 'comments') {
1409 $sth->bind_param($i, $data, {ora_type => ORA_CLOB});
1410 } elsif ($col eq 'image') {
1411 $sth->bind_param($i, $data, {ora_type => ORA_BLOB});
1413 $sth->bind_param($i, $data);
1417 $sth->execute; # execute without @bind now
1419 Now, why would you still use B<SQL::Abstract> if you have to do this crap?
1420 Basically, the advantage is still that you don't have to care which fields
1421 are or are not included. You could wrap that above C<for> loop in a simple
1422 sub called C<bind_fields()> or something and reuse it repeatedly. You still
1423 get a layer of abstraction over manual SQL specification.
1425 Note that if you set L</bindtype> to C<columns>, the C<\[$sql, @bind]>
1426 construct (see L</Literal SQL with placeholders and bind values (subqueries)>)
1427 will expect the bind values in this format.
1431 This is the character that a table or column name will be quoted
1432 with. By default this is an empty string, but you could set it to
1433 the character C<`>, to generate SQL like this:
1435 SELECT `a_field` FROM `a_table` WHERE `some_field` LIKE '%someval%'
1437 Alternatively, you can supply an array ref of two items, the first being the left
1438 hand quote character, and the second the right hand quote character. For
1439 example, you could supply C<['[',']']> for SQL Server 2000 compliant quotes
1440 that generates SQL like this:
1442 SELECT [a_field] FROM [a_table] WHERE [some_field] LIKE '%someval%'
1444 Quoting is useful if you have tables or columns names that are reserved
1445 words in your database's SQL dialect.
1449 This is the character that separates a table and column name. It is
1450 necessary to specify this when the C<quote_char> option is selected,
1451 so that tables and column names can be individually quoted like this:
1453 SELECT `table`.`one_field` FROM `table` WHERE `table`.`other_field` = 1
1455 =item array_datatypes
1457 When this option is true, arrayrefs in INSERT or UPDATE are
1458 interpreted as array datatypes and are passed directly
1460 When this option is false, arrayrefs are interpreted
1461 as literal SQL, just like refs to arrayrefs
1462 (but this behavior is for backwards compatibility; when writing
1463 new queries, use the "reference to arrayref" syntax
1469 Takes a reference to a list of "special operators"
1470 to extend the syntax understood by L<SQL::Abstract>.
1471 See section L</"SPECIAL OPERATORS"> for details.
1477 =head2 insert($table, \@values || \%fieldvals)
1479 This is the simplest function. You simply give it a table name
1480 and either an arrayref of values or hashref of field/value pairs.
1481 It returns an SQL INSERT statement and a list of bind values.
1482 See the sections on L</"Inserting and Updating Arrays"> and
1483 L</"Inserting and Updating SQL"> for information on how to insert
1484 with those data types.
1486 =head2 update($table, \%fieldvals, \%where)
1488 This takes a table, hashref of field/value pairs, and an optional
1489 hashref L<WHERE clause|/WHERE CLAUSES>. It returns an SQL UPDATE function and a list
1491 See the sections on L</"Inserting and Updating Arrays"> and
1492 L</"Inserting and Updating SQL"> for information on how to insert
1493 with those data types.
1495 =head2 select($source, $fields, $where, $order)
1497 This returns a SQL SELECT statement and associated list of bind values, as
1498 specified by the arguments :
1504 Specification of the 'FROM' part of the statement.
1505 The argument can be either a plain scalar (interpreted as a table
1506 name, will be quoted), or an arrayref (interpreted as a list
1507 of table names, joined by commas, quoted), or a scalarref
1508 (literal table name, not quoted), or a ref to an arrayref
1509 (list of literal table names, joined by commas, not quoted).
1513 Specification of the list of fields to retrieve from
1515 The argument can be either an arrayref (interpreted as a list
1516 of field names, will be joined by commas and quoted), or a
1517 plain scalar (literal SQL, not quoted).
1518 Please observe that this API is not as flexible as for
1519 the first argument C<$table>, for backwards compatibility reasons.
1523 Optional argument to specify the WHERE part of the query.
1524 The argument is most often a hashref, but can also be
1525 an arrayref or plain scalar --
1526 see section L<WHERE clause|/"WHERE CLAUSES"> for details.
1530 Optional argument to specify the ORDER BY part of the query.
1531 The argument can be a scalar, a hashref or an arrayref
1532 -- see section L<ORDER BY clause|/"ORDER BY CLAUSES">
1538 =head2 delete($table, \%where)
1540 This takes a table name and optional hashref L<WHERE clause|/WHERE CLAUSES>.
1541 It returns an SQL DELETE statement and list of bind values.
1543 =head2 where(\%where, \@order)
1545 This is used to generate just the WHERE clause. For example,
1546 if you have an arbitrary data structure and know what the
1547 rest of your SQL is going to look like, but want an easy way
1548 to produce a WHERE clause, use this. It returns an SQL WHERE
1549 clause and list of bind values.
1552 =head2 values(\%data)
1554 This just returns the values from the hash C<%data>, in the same
1555 order that would be returned from any of the other above queries.
1556 Using this allows you to markedly speed up your queries if you
1557 are affecting lots of rows. See below under the L</"PERFORMANCE"> section.
1559 =head2 generate($any, 'number', $of, \@data, $struct, \%types)
1561 Warning: This is an experimental method and subject to change.
1563 This returns arbitrarily generated SQL. It's a really basic shortcut.
1564 It will return two different things, depending on return context:
1566 my($stmt, @bind) = $sql->generate('create table', \$table, \@fields);
1567 my $stmt_and_val = $sql->generate('create table', \$table, \@fields);
1569 These would return the following:
1571 # First calling form
1572 $stmt = "CREATE TABLE test (?, ?)";
1573 @bind = (field1, field2);
1575 # Second calling form
1576 $stmt_and_val = "CREATE TABLE test (field1, field2)";
1578 Depending on what you're trying to do, it's up to you to choose the correct
1579 format. In this example, the second form is what you would want.
1583 $sql->generate('alter session', { nls_date_format => 'MM/YY' });
1587 ALTER SESSION SET nls_date_format = 'MM/YY'
1589 You get the idea. Strings get their case twiddled, but everything
1590 else remains verbatim.
1595 =head1 WHERE CLAUSES
1599 This module uses a variation on the idea from L<DBIx::Abstract>. It
1600 is B<NOT>, repeat I<not> 100% compatible. B<The main logic of this
1601 module is that things in arrays are OR'ed, and things in hashes
1604 The easiest way to explain is to show lots of examples. After
1605 each C<%where> hash shown, it is assumed you used:
1607 my($stmt, @bind) = $sql->where(\%where);
1609 However, note that the C<%where> hash can be used directly in any
1610 of the other functions as well, as described above.
1612 =head2 Key-value pairs
1614 So, let's get started. To begin, a simple hash:
1618 status => 'completed'
1621 Is converted to SQL C<key = val> statements:
1623 $stmt = "WHERE user = ? AND status = ?";
1624 @bind = ('nwiger', 'completed');
1626 One common thing I end up doing is having a list of values that
1627 a field can be in. To do this, simply specify a list inside of
1632 status => ['assigned', 'in-progress', 'pending'];
1635 This simple code will create the following:
1637 $stmt = "WHERE user = ? AND ( status = ? OR status = ? OR status = ? )";
1638 @bind = ('nwiger', 'assigned', 'in-progress', 'pending');
1640 An empty arrayref will be considered a logical false and
1643 =head2 Key-value pairs
1645 If you want to specify a different type of operator for your comparison,
1646 you can use a hashref for a given column:
1650 status => { '!=', 'completed' }
1653 Which would generate:
1655 $stmt = "WHERE user = ? AND status != ?";
1656 @bind = ('nwiger', 'completed');
1658 To test against multiple values, just enclose the values in an arrayref:
1660 status => { '!=', ['assigned', 'in-progress', 'pending'] };
1662 Which would give you:
1664 "WHERE status != ? AND status != ? AND status != ?"
1666 Notice that since the operator was recognized as being a 'negative'
1667 operator, the arrayref was interpreted with 'AND' logic (because
1668 of Morgan's laws). By contrast, the reverse
1670 status => { '=', ['assigned', 'in-progress', 'pending'] };
1674 "WHERE status = ? OR status = ? OR status = ?"
1677 The hashref can also contain multiple pairs, in which case it is expanded
1678 into an C<AND> of its elements:
1682 status => { '!=', 'completed', -not_like => 'pending%' }
1685 # Or more dynamically, like from a form
1686 $where{user} = 'nwiger';
1687 $where{status}{'!='} = 'completed';
1688 $where{status}{'-not_like'} = 'pending%';
1690 # Both generate this
1691 $stmt = "WHERE user = ? AND status != ? AND status NOT LIKE ?";
1692 @bind = ('nwiger', 'completed', 'pending%');
1695 To get an OR instead, you can combine it with the arrayref idea:
1699 priority => [ {'=', 2}, {'!=', 1} ]
1702 Which would generate:
1704 $stmt = "WHERE user = ? AND priority = ? OR priority != ?";
1705 @bind = ('nwiger', '2', '1');
1707 If you want to include literal SQL (with or without bind values), just use a
1708 scalar reference or array reference as the value:
1711 date_entered => { '>' => \["to_date(?, 'MM/DD/YYYY')", "11/26/2008"] },
1712 date_expires => { '<' => \"now()" }
1715 Which would generate:
1717 $stmt = "WHERE date_entered > "to_date(?, 'MM/DD/YYYY') AND date_expires < now()";
1718 @bind = ('11/26/2008');
1721 =head2 Logic and nesting operators
1723 In the example above,
1724 there is a subtle trap if you want to say something like
1725 this (notice the C<AND>):
1727 WHERE priority != ? AND priority != ?
1729 Because, in Perl you I<can't> do this:
1731 priority => { '!=', 2, '!=', 1 }
1733 As the second C<!=> key will obliterate the first. The solution
1734 is to use the special C<-modifier> form inside an arrayref:
1736 priority => [ -and => {'!=', 2},
1740 Normally, these would be joined by C<OR>, but the modifier tells it
1741 to use C<AND> instead. (Hint: You can use this in conjunction with the
1742 C<logic> option to C<new()> in order to change the way your queries
1743 work by default.) B<Important:> Note that the C<-modifier> goes
1744 B<INSIDE> the arrayref, as an extra first element. This will
1745 B<NOT> do what you think it might:
1747 priority => -and => [{'!=', 2}, {'!=', 1}] # WRONG!
1749 Here is a quick list of equivalencies, since there is some overlap:
1752 status => {'!=', 'completed', 'not like', 'pending%' }
1753 status => [ -and => {'!=', 'completed'}, {'not like', 'pending%'}]
1756 status => {'=', ['assigned', 'in-progress']}
1757 status => [ -or => {'=', 'assigned'}, {'=', 'in-progress'}]
1758 status => [ {'=', 'assigned'}, {'=', 'in-progress'} ]
1760 In addition to C<-and> and C<-or>, there is also a special C<-nest>
1761 operator which adds an additional set of parens, to create a subquery.
1762 For example, to get something like this:
1764 $stmt = "WHERE user = ? AND ( workhrs > ? OR geo = ? )";
1765 @bind = ('nwiger', '20', 'ASIA');
1771 -nest => [ workhrs => {'>', 20}, geo => 'ASIA' ],
1774 If you need several nested subexpressions, you can number
1775 the C<-nest> branches :
1785 =head2 Special operators : IN, BETWEEN, etc.
1787 You can also use the hashref format to compare a list of fields using the
1788 C<IN> comparison operator, by specifying the list as an arrayref:
1791 status => 'completed',
1792 reportid => { -in => [567, 2335, 2] }
1795 Which would generate:
1797 $stmt = "WHERE status = ? AND reportid IN (?,?,?)";
1798 @bind = ('completed', '567', '2335', '2');
1800 The reverse operator C<-not_in> generates SQL C<NOT IN> and is used in
1803 Another pair of operators is C<-between> and C<-not_between>,
1804 used with an arrayref of two values:
1808 completion_date => {
1809 -not_between => ['2002-10-01', '2003-02-06']
1815 WHERE user = ? AND completion_date NOT BETWEEN ( ? AND ? )
1817 These are the two builtin "special operators"; but the
1818 list can be expanded : see section L</"SPECIAL OPERATORS"> below.
1820 =head2 Nested conditions
1822 So far, we've seen how multiple conditions are joined with a top-level
1823 C<AND>. We can change this by putting the different conditions we want in
1824 hashes and then putting those hashes in an array. For example:
1829 status => { -like => ['pending%', 'dispatched'] },
1833 status => 'unassigned',
1837 This data structure would create the following:
1839 $stmt = "WHERE ( user = ? AND ( status LIKE ? OR status LIKE ? ) )
1840 OR ( user = ? AND status = ? ) )";
1841 @bind = ('nwiger', 'pending', 'dispatched', 'robot', 'unassigned');
1843 This can be combined with the C<-nest> operator to properly group
1850 ["-and", workhrs => {'>', 20}, geo => 'ASIA' ],
1851 ["-and", workhrs => {'<', 50}, geo => 'EURO' ]
1858 WHERE ( user = ? AND
1859 ( ( workhrs > ? AND geo = ? )
1860 OR ( workhrs < ? AND geo = ? ) ) )
1864 Finally, sometimes only literal SQL will do. If you want to include
1865 literal SQL verbatim, you can specify it as a scalar reference, namely:
1867 my $inn = 'is Not Null';
1869 priority => { '<', 2 },
1875 $stmt = "WHERE priority < ? AND requestor is Not Null";
1878 Note that in this example, you only get one bind parameter back, since
1879 the verbatim SQL is passed as part of the statement.
1881 Of course, just to prove a point, the above can also be accomplished
1885 priority => { '<', 2 },
1886 requestor => { '!=', undef },
1892 Conditions on boolean columns can be expressed in the
1893 same way, passing a reference to an empty string :
1896 priority => { '<', 2 },
1902 $stmt = "WHERE priority < ? AND is_ready";
1906 =head2 Literal SQL with placeholders and bind values (subqueries)
1908 If the literal SQL to be inserted has placeholders and bind values,
1909 use a reference to an arrayref (yes this is a double reference --
1910 not so common, but perfectly legal Perl). For example, to find a date
1911 in Postgres you can use something like this:
1914 date_column => \[q/= date '2008-09-30' - ?::integer/, 10/]
1919 $stmt = "WHERE ( date_column = date '2008-09-30' - ?::integer )"
1922 Note that you must pass the bind values in the same format as they are returned
1923 by C</where>. That means that if you set L</bindtype> to C<columns>, you must
1924 provide the bind values in the C<< [ column_meta => value ] >> format, where
1925 C<column_meta> is an opaque scalar value; most commonly the column name, but
1926 you can use any scalar scalar value (including references and blessed
1927 references), L<SQL::Abstract> will simply pass it through intact. So eg. the
1928 above example will look like:
1931 date_column => \[q/= date '2008-09-30' - ?::integer/, [ dummy => 10 ]/]
1934 Literal SQL is especially useful for nesting parenthesized clauses in the
1935 main SQL query. Here is a first example :
1937 my ($sub_stmt, @sub_bind) = ("SELECT c1 FROM t1 WHERE c2 < ? AND c3 LIKE ?",
1941 bar => \["IN ($sub_stmt)" => @sub_bind],
1946 $stmt = "WHERE (foo = ? AND bar IN (SELECT c1 FROM t1
1947 WHERE c2 < ? AND c3 LIKE ?))";
1948 @bind = (1234, 100, "foo%");
1950 Other subquery operators, like for example C<"E<gt> ALL"> or C<"NOT IN">,
1951 are expressed in the same way. Of course the C<$sub_stmt> and
1952 its associated bind values can be generated through a former call
1955 my ($sub_stmt, @sub_bind)
1956 = $sql->select("t1", "c1", {c2 => {"<" => 100},
1957 c3 => {-like => "foo%"}});
1960 bar => \["> ALL ($sub_stmt)" => @sub_bind],
1963 In the examples above, the subquery was used as an operator on a column;
1964 but the same principle also applies for a clause within the main C<%where>
1965 hash, like an EXISTS subquery :
1967 my ($sub_stmt, @sub_bind)
1968 = $sql->select("t1", "*", {c1 => 1, c2 => \"> t0.c0"});
1971 -nest => \["EXISTS ($sub_stmt)" => @sub_bind],
1976 $stmt = "WHERE (foo = ? AND EXISTS (SELECT * FROM t1
1977 WHERE c1 = ? AND c2 > t0.c0))";
1981 Observe that the condition on C<c2> in the subquery refers to
1982 column C<t0.c0> of the main query : this is I<not> a bind
1983 value, so we have to express it through a scalar ref.
1984 Writing C<< c2 => {">" => "t0.c0"} >> would have generated
1985 C<< c2 > ? >> with bind value C<"t0.c0"> ... not exactly
1986 what we wanted here.
1988 Another use of the subquery technique is when some SQL clauses need
1989 parentheses, as it often occurs with some proprietary SQL extensions
1990 like for example fulltext expressions, geospatial expressions,
1991 NATIVE clauses, etc. Here is an example of a fulltext query in MySQL :
1994 -nest => \["MATCH (col1, col2) AGAINST (?)" => qw/apples/]
1997 Finally, here is an example where a subquery is used
1998 for expressing unary negation:
2000 my ($sub_stmt, @sub_bind)
2001 = $sql->where({age => [{"<" => 10}, {">" => 20}]});
2002 $sub_stmt =~ s/^ where //i; # don't want "WHERE" in the subclause
2004 lname => {like => '%son%'},
2005 -nest => \["NOT ($sub_stmt)" => @sub_bind],
2010 $stmt = "lname LIKE ? AND NOT ( age < ? OR age > ? )"
2011 @bind = ('%son%', 10, 20)
2017 These pages could go on for a while, since the nesting of the data
2018 structures this module can handle are pretty much unlimited (the
2019 module implements the C<WHERE> expansion as a recursive function
2020 internally). Your best bet is to "play around" with the module a
2021 little to see how the data structures behave, and choose the best
2022 format for your data based on that.
2024 And of course, all the values above will probably be replaced with
2025 variables gotten from forms or the command line. After all, if you
2026 knew everything ahead of time, you wouldn't have to worry about
2027 dynamically-generating SQL and could just hardwire it into your
2033 =head1 ORDER BY CLAUSES
2035 Some functions take an order by clause. This can either be a scalar (just a
2036 column name,) a hash of C<< { -desc => 'col' } >> or C<< { -asc => 'col' } >>,
2037 or an array of either of the two previous forms. Examples:
2039 Given | Will Generate
2040 ----------------------------------------------------------
2041 \'colA DESC' | ORDER BY colA DESC
2042 'colA' | ORDER BY colA
2043 [qw/colA colB/] | ORDER BY colA, colB
2044 {-asc => 'colA'} | ORDER BY colA ASC
2045 {-desc => 'colB'} | ORDER BY colB DESC
2047 {-asc => 'colA'}, | ORDER BY colA ASC, colB DESC
2050 [colA => {-asc => 'colB'}] | ORDER BY colA, colB ASC
2051 ==========================================================
2055 =head1 SPECIAL OPERATORS
2057 my $sqlmaker = SQL::Abstract->new(special_ops => [
2060 my ($self, $field, $op, $arg) = @_;
2066 A "special operator" is a SQL syntactic clause that can be
2067 applied to a field, instead of a usual binary operator.
2070 WHERE field IN (?, ?, ?)
2071 WHERE field BETWEEN ? AND ?
2072 WHERE MATCH(field) AGAINST (?, ?)
2074 Special operators IN and BETWEEN are fairly standard and therefore
2075 are builtin within C<SQL::Abstract>. For other operators,
2076 like the MATCH .. AGAINST example above which is
2077 specific to MySQL, you can write your own operator handlers :
2078 supply a C<special_ops> argument to the C<new> method.
2079 That argument takes an arrayref of operator definitions;
2080 each operator definition is a hashref with two entries
2086 the regular expression to match the operator
2090 coderef that will be called when meeting that operator
2091 in the input tree. The coderef will be called with
2092 arguments C<< ($self, $field, $op, $arg) >>, and
2093 should return a C<< ($sql, @bind) >> structure.
2097 For example, here is an implementation
2098 of the MATCH .. AGAINST syntax for MySQL
2100 my $sqlmaker = SQL::Abstract->new(special_ops => [
2102 # special op for MySql MATCH (field) AGAINST(word1, word2, ...)
2103 {regex => qr/^match$/i,
2105 my ($self, $field, $op, $arg) = @_;
2106 $arg = [$arg] if not ref $arg;
2107 my $label = $self->_quote($field);
2108 my ($placeholder) = $self->_convert('?');
2109 my $placeholders = join ", ", (($placeholder) x @$arg);
2110 my $sql = $self->_sqlcase('match') . " ($label) "
2111 . $self->_sqlcase('against') . " ($placeholders) ";
2112 my @bind = $self->_bindtype($field, @$arg);
2113 return ($sql, @bind);
2122 Thanks to some benchmarking by Mark Stosberg, it turns out that
2123 this module is many orders of magnitude faster than using C<DBIx::Abstract>.
2124 I must admit this wasn't an intentional design issue, but it's a
2125 byproduct of the fact that you get to control your C<DBI> handles
2128 To maximize performance, use a code snippet like the following:
2130 # prepare a statement handle using the first row
2131 # and then reuse it for the rest of the rows
2133 for my $href (@array_of_hashrefs) {
2134 $stmt ||= $sql->insert('table', $href);
2135 $sth ||= $dbh->prepare($stmt);
2136 $sth->execute($sql->values($href));
2139 The reason this works is because the keys in your C<$href> are sorted
2140 internally by B<SQL::Abstract>. Thus, as long as your data retains
2141 the same structure, you only have to generate the SQL the first time
2142 around. On subsequent queries, simply use the C<values> function provided
2143 by this module to return your values in the correct order.
2148 If you use my C<CGI::FormBuilder> module at all, you'll hopefully
2149 really like this part (I do, at least). Building up a complex query
2150 can be as simple as the following:
2154 use CGI::FormBuilder;
2157 my $form = CGI::FormBuilder->new(...);
2158 my $sql = SQL::Abstract->new;
2160 if ($form->submitted) {
2161 my $field = $form->field;
2162 my $id = delete $field->{id};
2163 my($stmt, @bind) = $sql->update('table', $field, {id => $id});
2166 Of course, you would still have to connect using C<DBI> to run the
2167 query, but the point is that if you make your form look like your
2168 table, the actual query script can be extremely simplistic.
2170 If you're B<REALLY> lazy (I am), check out C<HTML::QuickTable> for
2171 a fast interface to returning and formatting data. I frequently
2172 use these three modules together to write complex database query
2173 apps in under 50 lines.
2178 Version 1.50 was a major internal refactoring of C<SQL::Abstract>.
2179 Great care has been taken to preserve the I<published> behavior
2180 documented in previous versions in the 1.* family; however,
2181 some features that were previously undocumented, or behaved
2182 differently from the documentation, had to be changed in order
2183 to clarify the semantics. Hence, client code that was relying
2184 on some dark areas of C<SQL::Abstract> v1.*
2185 B<might behave differently> in v1.50.
2187 The main changes are :
2193 support for literal SQL through the C<< \ [$sql, bind] >> syntax.
2197 support for the { operator => \"..." } construct (to embed literal SQL)
2201 support for the { operator => \["...", @bind] } construct (to embed literal SQL with bind values)
2205 added -nest1, -nest2 or -nest_1, -nest_2, ...
2209 optional support for L<array datatypes|/"Inserting and Updating Arrays">
2213 defensive programming : check arguments
2217 fixed bug with global logic, which was previously implemented
2218 through global variables yielding side-effects. Prior versons would
2219 interpret C<< [ {cond1, cond2}, [cond3, cond4] ] >>
2220 as C<< "(cond1 AND cond2) OR (cond3 AND cond4)" >>.
2221 Now this is interpreted
2222 as C<< "(cond1 AND cond2) OR (cond3 OR cond4)" >>.
2226 C<-and> / C<-or> operators are no longer accepted
2227 in the middle of an arrayref : they are
2228 only admitted if in first position.
2232 changed logic for distributing an op over arrayrefs
2236 fixed semantics of _bindtype on array args
2240 dropped the C<_anoncopy> of the %where tree. No longer necessary,
2241 we just avoid shifting arrays within that tree.
2245 dropped the C<_modlogic> function
2251 =head1 ACKNOWLEDGEMENTS
2253 There are a number of individuals that have really helped out with
2254 this module. Unfortunately, most of them submitted bugs via CPAN
2255 so I have no idea who they are! But the people I do know are:
2257 Ash Berlin (order_by hash term support)
2258 Matt Trout (DBIx::Class support)
2259 Mark Stosberg (benchmarking)
2260 Chas Owens (initial "IN" operator support)
2261 Philip Collins (per-field SQL functions)
2262 Eric Kolve (hashref "AND" support)
2263 Mike Fragassi (enhancements to "BETWEEN" and "LIKE")
2264 Dan Kubb (support for "quote_char" and "name_sep")
2265 Guillermo Roditi (patch to cleanup "IN" and "BETWEEN", fix and tests for _order_by)
2266 Laurent Dami (internal refactoring, multiple -nest, extensible list of special operators, literal SQL)
2267 Norbert Buchmuller (support for literal SQL in hashpair, misc. fixes & tests)
2273 L<DBIx::Class>, L<DBIx::Abstract>, L<CGI::FormBuilder>, L<HTML::QuickTable>.
2277 Copyright (c) 2001-2007 Nathan Wiger <nwiger@cpan.org>. All Rights Reserved.
2279 This module is actively maintained by Matt Trout <mst@shadowcatsystems.co.uk>
2281 For support, your best bet is to try the C<DBIx::Class> users mailing list.
2282 While not an official support venue, C<DBIx::Class> makes heavy use of
2283 C<SQL::Abstract>, and as such list members there are very familiar with
2284 how to create queries.
2286 This module is free software; you may copy this under the terms of
2287 the GNU General Public License, or the Artistic License, copies of
2288 which should have accompanied your Perl kit.