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|NEST) ( \_? \d* ) $/xi
447 or puke "unknown operator: -$op_str";
449 my $op = uc($1); # uppercase, remove trailing digits
451 belch 'Use of [and|or|nest]_N modifiers is deprecated and will be removed in SQLA v2.0. '
452 . "You probably wanted ...-and => [ $op_str => COND1, $op_str => COND2 ... ]";
455 $self->_debug("OP(-$op) within hashref, recursing...");
457 $self->_SWITCH_refkind($v, {
460 return $self->_where_ARRAYREF($v, $op eq 'NEST' ? '' : $op);
465 return $self->_where_ARRAYREF([ map { $_ => $v->{$_} } (sort keys %$v) ], 'OR');
468 return $self->_where_HASHREF($v);
472 SCALARREF => sub { # literal SQL
474 or puke "-$op => \\\$scalar not supported, use -nest => ...";
478 ARRAYREFREF => sub { # literal SQL
480 or puke "-$op => \\[..] not supported, use -nest => ...";
484 SCALAR => sub { # permissively interpreted as SQL
486 or puke "-$op => 'scalar' not supported, use -nest => \\'scalar'";
487 belch "literal SQL should be -nest => \\'scalar' "
488 . "instead of -nest => 'scalar' ";
493 puke "-$op => undef not supported";
499 sub _where_hashpair_ARRAYREF {
500 my ($self, $k, $v) = @_;
503 my @v = @$v; # need copy because of shift below
504 $self->_debug("ARRAY($k) means distribute over elements");
506 # put apart first element if it is an operator (-and, -or)
508 (defined $v[0] && $v[0] =~ /^ - (?: AND|OR ) $/ix)
512 my @distributed = map { {$k => $_} } @v;
515 $self->_debug("OP($op) reinjected into the distributed array");
516 unshift @distributed, $op;
519 my $logic = $op ? substr($op, 1) : '';
521 return $self->_recurse_where(\@distributed, $logic);
524 # LDNOTE : not sure of this one. What does "distribute over nothing" mean?
525 $self->_debug("empty ARRAY($k) means 0=1");
526 return ($self->{sqlfalse});
530 sub _where_hashpair_HASHREF {
531 my ($self, $k, $v, $logic) = @_;
534 my ($all_sql, @all_bind);
536 for my $op (sort keys %$v) {
539 # put the operator in canonical form
540 $op =~ s/^-//; # remove initial dash
541 $op =~ tr/_/ /; # underscores become spaces
542 $op =~ s/^\s+//; # no initial space
543 $op =~ s/\s+$//; # no final space
544 $op =~ s/\s+/ /; # multiple spaces become one
548 # CASE: special operators like -in or -between
549 my $special_op = first {$op =~ $_->{regex}} @{$self->{special_ops}};
551 my $handler = $special_op->{handler};
553 puke "No handler supplied for special operator matching $special_op->{regex}";
555 elsif (not ref $handler) {
556 ($sql, @bind) = $self->$handler ($k, $op, $val);
558 elsif (ref $handler eq 'CODE') {
559 ($sql, @bind) = $handler->($self, $k, $op, $val);
562 puke "Illegal handler for special operator matching $special_op->{regex} - expecting a method name or a coderef";
566 $self->_SWITCH_refkind($val, {
568 ARRAYREF => sub { # CASE: col => {op => \@vals}
569 ($sql, @bind) = $self->_where_field_op_ARRAYREF($k, $op, $val);
572 SCALARREF => sub { # CASE: col => {op => \$scalar} (literal SQL without bind)
573 $sql = join ' ', $self->_convert($self->_quote($k)),
574 $self->_sqlcase($op),
578 ARRAYREFREF => sub { # CASE: col => {op => \[$sql, @bind]} (literal SQL with bind)
579 my ($sub_sql, @sub_bind) = @$$val;
580 $self->_assert_bindval_matches_bindtype(@sub_bind);
581 $sql = join ' ', $self->_convert($self->_quote($k)),
582 $self->_sqlcase($op),
588 ($sql, @bind) = $self->_where_hashpair_HASHREF($k, $val, $op);
591 UNDEF => sub { # CASE: col => {op => undef} : sql "IS (NOT)? NULL"
592 my $is = ($op =~ $self->{equality_op}) ? 'is' :
593 ($op =~ $self->{inequality_op}) ? 'is not' :
594 puke "unexpected operator '$op' with undef operand";
595 $sql = $self->_quote($k) . $self->_sqlcase(" $is null");
598 FALLBACK => sub { # CASE: col => {op => $scalar}
599 $sql = join ' ', $self->_convert($self->_quote($k)),
600 $self->_sqlcase($op),
601 $self->_convert('?');
602 @bind = $self->_bindtype($k, $val);
607 ($all_sql) = (defined $all_sql and $all_sql) ? $self->_join_sql_clauses($logic, [$all_sql, $sql], []) : $sql;
608 push @all_bind, @bind;
610 return ($all_sql, @all_bind);
615 sub _where_field_op_ARRAYREF {
616 my ($self, $k, $op, $vals) = @_;
619 $self->_debug("ARRAY($vals) means multiple elements: [ @$vals ]");
621 # see if the first element is an -and/-or op
623 if ($vals->[0] =~ /^ - ( AND|OR ) $/ix) {
628 # distribute $op over each remaining member of @$vals, append logic if exists
629 return $self->_recurse_where([map { {$k => {$op, $_}} } @$vals], $logic);
631 # LDNOTE : had planned to change the distribution logic when
632 # $op =~ $self->{inequality_op}, because of Morgan laws :
633 # with {field => {'!=' => [22, 33]}}, it would be ridiculous to generate
634 # WHERE field != 22 OR field != 33 : the user probably means
635 # WHERE field != 22 AND field != 33.
636 # To do this, replace the above to roughly :
637 # my $logic = ($op =~ $self->{inequality_op}) ? 'AND' : 'OR';
638 # return $self->_recurse_where([map { {$k => {$op, $_}} } @$vals], $logic);
642 # try to DWIM on equality operators
643 # LDNOTE : not 100% sure this is the correct thing to do ...
644 return ($self->{sqlfalse}) if $op =~ $self->{equality_op};
645 return ($self->{sqltrue}) if $op =~ $self->{inequality_op};
648 puke "operator '$op' applied on an empty array (field '$k')";
653 sub _where_hashpair_SCALARREF {
654 my ($self, $k, $v) = @_;
655 $self->_debug("SCALAR($k) means literal SQL: $$v");
656 my $sql = $self->_quote($k) . " " . $$v;
660 # literal SQL with bind
661 sub _where_hashpair_ARRAYREFREF {
662 my ($self, $k, $v) = @_;
663 $self->_debug("REF($k) means literal SQL: @${$v}");
664 my ($sql, @bind) = @${$v};
665 $self->_assert_bindval_matches_bindtype(@bind);
666 $sql = $self->_quote($k) . " " . $sql;
667 return ($sql, @bind );
670 # literal SQL without bind
671 sub _where_hashpair_SCALAR {
672 my ($self, $k, $v) = @_;
673 $self->_debug("NOREF($k) means simple key=val: $k $self->{cmp} $v");
674 my $sql = join ' ', $self->_convert($self->_quote($k)),
675 $self->_sqlcase($self->{cmp}),
676 $self->_convert('?');
677 my @bind = $self->_bindtype($k, $v);
678 return ( $sql, @bind);
682 sub _where_hashpair_UNDEF {
683 my ($self, $k, $v) = @_;
684 $self->_debug("UNDEF($k) means IS NULL");
685 my $sql = $self->_quote($k) . $self->_sqlcase(' is null');
689 #======================================================================
690 # WHERE: TOP-LEVEL OTHERS (SCALARREF, SCALAR, UNDEF)
691 #======================================================================
694 sub _where_SCALARREF {
695 my ($self, $where) = @_;
698 $self->_debug("SCALAR(*top) means literal SQL: $$where");
704 my ($self, $where) = @_;
707 $self->_debug("NOREF(*top) means literal SQL: $where");
718 #======================================================================
719 # WHERE: BUILTIN SPECIAL OPERATORS (-in, -between)
720 #======================================================================
723 sub _where_field_BETWEEN {
724 my ($self, $k, $op, $vals) = @_;
726 (ref $vals eq 'ARRAY' && @$vals == 2) or
727 (ref $vals eq 'REF' && (@$$vals == 1 || @$$vals == 2 || @$$vals == 3))
728 or puke "special op 'between' requires an arrayref of two values (or a scalarref or arrayrefref for literal SQL)";
730 my ($clause, @bind, $label, $and, $placeholder);
731 $label = $self->_convert($self->_quote($k));
732 $and = ' ' . $self->_sqlcase('and') . ' ';
733 $placeholder = $self->_convert('?');
734 $op = $self->_sqlcase($op);
736 if (ref $vals eq 'REF') {
737 ($clause, @bind) = @$$vals;
740 my (@all_sql, @all_bind);
742 foreach my $val (@$vals) {
743 my ($sql, @bind) = $self->_SWITCH_refkind($val, {
745 return ($placeholder, ($val));
748 return ($self->_convert($$val), ());
752 push @all_bind, @bind;
755 $clause = (join $and, @all_sql);
756 @bind = $self->_bindtype($k, @all_bind);
758 my $sql = "( $label $op $clause )";
763 sub _where_field_IN {
764 my ($self, $k, $op, $vals) = @_;
766 # backwards compatibility : if scalar, force into an arrayref
767 $vals = [$vals] if defined $vals && ! ref $vals;
769 my ($label) = $self->_convert($self->_quote($k));
770 my ($placeholder) = $self->_convert('?');
771 $op = $self->_sqlcase($op);
773 my ($sql, @bind) = $self->_SWITCH_refkind($vals, {
774 ARRAYREF => sub { # list of choices
775 if (@$vals) { # nonempty list
776 my $placeholders = join ", ", (($placeholder) x @$vals);
777 my $sql = "$label $op ( $placeholders )";
778 my @bind = $self->_bindtype($k, @$vals);
780 return ($sql, @bind);
782 else { # empty list : some databases won't understand "IN ()", so DWIM
783 my $sql = ($op =~ /\bnot\b/i) ? $self->{sqltrue} : $self->{sqlfalse};
788 ARRAYREFREF => sub { # literal SQL with bind
789 my ($sql, @bind) = @$$vals;
790 $self->_assert_bindval_matches_bindtype(@bind);
791 return ("$label $op ( $sql )", @bind);
795 puke "special op 'in' requires an arrayref (or arrayref-ref)";
799 return ($sql, @bind);
807 #======================================================================
809 #======================================================================
812 my ($self, $arg) = @_;
814 # construct list of ordering instructions
815 my @order = $self->_SWITCH_refkind($arg, {
818 map {$self->_SWITCH_refkind($_, {
819 SCALAR => sub {$self->_quote($_)},
821 SCALARREF => sub {$$_}, # literal SQL, no quoting
822 HASHREF => sub {$self->_order_by_hash($_)}
826 SCALAR => sub {$self->_quote($arg)},
828 SCALARREF => sub {$$arg}, # literal SQL, no quoting
829 HASHREF => sub {$self->_order_by_hash($arg)},
834 my $order = join ', ', @order;
835 return $order ? $self->_sqlcase(' order by')." $order" : '';
840 my ($self, $hash) = @_;
842 # get first pair in hash
843 my ($key, $val) = each %$hash;
845 # check if one pair was found and no other pair in hash
846 $key && !(each %$hash)
847 or puke "hash passed to _order_by must have exactly one key (-desc or -asc)";
849 my ($order) = ($key =~ /^-(desc|asc)/i)
850 or puke "invalid key in _order_by hash : $key";
852 $val = ref $val eq 'ARRAY' ? $val : [$val];
853 return join ', ', map { $self->_quote($_) . ' ' . $self->_sqlcase($order) } @$val;
858 #======================================================================
859 # DATASOURCE (FOR NOW, JUST PLAIN TABLE OR LIST OF TABLES)
860 #======================================================================
865 $self->_SWITCH_refkind($from, {
866 ARRAYREF => sub {join ', ', map { $self->_quote($_) } @$from;},
867 SCALAR => sub {$self->_quote($from)},
868 SCALARREF => sub {$$from},
869 ARRAYREFREF => sub {join ', ', @$from;},
874 #======================================================================
876 #======================================================================
882 $label or puke "can't quote an empty label";
884 # left and right quote characters
885 my ($ql, $qr, @other) = $self->_SWITCH_refkind($self->{quote_char}, {
886 SCALAR => sub {($self->{quote_char}, $self->{quote_char})},
887 ARRAYREF => sub {@{$self->{quote_char}}},
891 or puke "quote_char must be an arrayref of 2 values";
893 # no quoting if no quoting chars
894 $ql or return $label;
896 # no quoting for literal SQL
897 return $$label if ref($label) eq 'SCALAR';
899 # separate table / column (if applicable)
900 my $sep = $self->{name_sep} || '';
901 my @to_quote = $sep ? split /\Q$sep\E/, $label : ($label);
903 # do the quoting, except for "*" or for `table`.*
904 my @quoted = map { $_ eq '*' ? $_: $ql.$_.$qr} @to_quote;
906 # reassemble and return.
907 return join $sep, @quoted;
911 # Conversion, if applicable
913 my ($self, $arg) = @_;
915 # LDNOTE : modified the previous implementation below because
916 # it was not consistent : the first "return" is always an array,
917 # the second "return" is context-dependent. Anyway, _convert
918 # seems always used with just a single argument, so make it a
920 # return @_ unless $self->{convert};
921 # my $conv = $self->_sqlcase($self->{convert});
922 # my @ret = map { $conv.'('.$_.')' } @_;
923 # return wantarray ? @ret : $ret[0];
924 if ($self->{convert}) {
925 my $conv = $self->_sqlcase($self->{convert});
926 $arg = $conv.'('.$arg.')';
934 my($col, @vals) = @_;
936 #LDNOTE : changed original implementation below because it did not make
937 # sense when bindtype eq 'columns' and @vals > 1.
938 # return $self->{bindtype} eq 'columns' ? [ $col, @vals ] : @vals;
940 return $self->{bindtype} eq 'columns' ? map {[$col, $_]} @vals : @vals;
943 # Dies if any element of @bind is not in [colname => value] format
944 # if bindtype is 'columns'.
945 sub _assert_bindval_matches_bindtype {
946 my ($self, @bind) = @_;
948 if ($self->{bindtype} eq 'columns') {
949 foreach my $val (@bind) {
950 if (!defined $val || ref($val) ne 'ARRAY' || @$val != 2) {
951 die "bindtype 'columns' selected, you need to pass: [column_name => bind_value]"
957 sub _join_sql_clauses {
958 my ($self, $logic, $clauses_aref, $bind_aref) = @_;
960 if (@$clauses_aref > 1) {
961 my $join = " " . $self->_sqlcase($logic) . " ";
962 my $sql = '( ' . join($join, @$clauses_aref) . ' )';
963 return ($sql, @$bind_aref);
965 elsif (@$clauses_aref) {
966 return ($clauses_aref->[0], @$bind_aref); # no parentheses
969 return (); # if no SQL, ignore @$bind_aref
974 # Fix SQL case, if so requested
978 # LDNOTE: if $self->{case} is true, then it contains 'lower', so we
979 # don't touch the argument ... crooked logic, but let's not change it!
980 return $self->{case} ? $_[0] : uc($_[0]);
984 #======================================================================
985 # DISPATCHING FROM REFKIND
986 #======================================================================
989 my ($self, $data) = @_;
995 # blessed objects are treated like scalars
996 $ref = (blessed $data) ? '' : ref $data;
997 $n_steps += 1 if $ref;
998 last if $ref ne 'REF';
1002 my $base = $ref || (defined $data ? 'SCALAR' : 'UNDEF');
1004 return $base . ('REF' x $n_steps);
1010 my ($self, $data) = @_;
1011 my @try = ($self->_refkind($data));
1012 push @try, 'SCALAR_or_UNDEF' if $try[0] eq 'SCALAR' || $try[0] eq 'UNDEF';
1013 push @try, 'FALLBACK';
1017 sub _METHOD_FOR_refkind {
1018 my ($self, $meth_prefix, $data) = @_;
1019 my $method = first {$_} map {$self->can($meth_prefix."_".$_)}
1020 $self->_try_refkind($data)
1021 or puke "cannot dispatch on '$meth_prefix' for ".$self->_refkind($data);
1026 sub _SWITCH_refkind {
1027 my ($self, $data, $dispatch_table) = @_;
1029 my $coderef = first {$_} map {$dispatch_table->{$_}}
1030 $self->_try_refkind($data)
1031 or puke "no dispatch entry for ".$self->_refkind($data);
1038 #======================================================================
1039 # VALUES, GENERATE, AUTOLOAD
1040 #======================================================================
1042 # LDNOTE: original code from nwiger, didn't touch code in that section
1043 # I feel the AUTOLOAD stuff should not be the default, it should
1044 # only be activated on explicit demand by user.
1048 my $data = shift || return;
1049 puke "Argument to ", __PACKAGE__, "->values must be a \\%hash"
1050 unless ref $data eq 'HASH';
1053 foreach my $k ( sort keys %$data ) {
1054 my $v = $data->{$k};
1055 $self->_SWITCH_refkind($v, {
1057 if ($self->{array_datatypes}) { # array datatype
1058 push @all_bind, $self->_bindtype($k, $v);
1060 else { # literal SQL with bind
1061 my ($sql, @bind) = @$v;
1062 $self->_assert_bindval_matches_bindtype(@bind);
1063 push @all_bind, @bind;
1066 ARRAYREFREF => sub { # literal SQL with bind
1067 my ($sql, @bind) = @${$v};
1068 $self->_assert_bindval_matches_bindtype(@bind);
1069 push @all_bind, @bind;
1071 SCALARREF => sub { # literal SQL without bind
1073 SCALAR_or_UNDEF => sub {
1074 push @all_bind, $self->_bindtype($k, $v);
1085 my(@sql, @sqlq, @sqlv);
1089 if ($ref eq 'HASH') {
1090 for my $k (sort keys %$_) {
1093 my $label = $self->_quote($k);
1094 if ($r eq 'ARRAY') {
1095 # literal SQL with bind
1096 my ($sql, @bind) = @$v;
1097 $self->_assert_bindval_matches_bindtype(@bind);
1098 push @sqlq, "$label = $sql";
1100 } elsif ($r eq 'SCALAR') {
1101 # literal SQL without bind
1102 push @sqlq, "$label = $$v";
1104 push @sqlq, "$label = ?";
1105 push @sqlv, $self->_bindtype($k, $v);
1108 push @sql, $self->_sqlcase('set'), join ', ', @sqlq;
1109 } elsif ($ref eq 'ARRAY') {
1110 # unlike insert(), assume these are ONLY the column names, i.e. for SQL
1113 if ($r eq 'ARRAY') { # literal SQL with bind
1114 my ($sql, @bind) = @$v;
1115 $self->_assert_bindval_matches_bindtype(@bind);
1118 } elsif ($r eq 'SCALAR') { # literal SQL without bind
1119 # embedded literal SQL
1126 push @sql, '(' . join(', ', @sqlq) . ')';
1127 } elsif ($ref eq 'SCALAR') {
1131 # strings get case twiddled
1132 push @sql, $self->_sqlcase($_);
1136 my $sql = join ' ', @sql;
1138 # this is pretty tricky
1139 # if ask for an array, return ($stmt, @bind)
1140 # otherwise, s/?/shift @sqlv/ to put it inline
1142 return ($sql, @sqlv);
1144 1 while $sql =~ s/\?/my $d = shift(@sqlv);
1145 ref $d ? $d->[1] : $d/e;
1154 # This allows us to check for a local, then _form, attr
1156 my($name) = $AUTOLOAD =~ /.*::(.+)/;
1157 return $self->generate($name, @_);
1168 SQL::Abstract - Generate SQL from Perl data structures
1174 my $sql = SQL::Abstract->new;
1176 my($stmt, @bind) = $sql->select($table, \@fields, \%where, \@order);
1178 my($stmt, @bind) = $sql->insert($table, \%fieldvals || \@values);
1180 my($stmt, @bind) = $sql->update($table, \%fieldvals, \%where);
1182 my($stmt, @bind) = $sql->delete($table, \%where);
1184 # Then, use these in your DBI statements
1185 my $sth = $dbh->prepare($stmt);
1186 $sth->execute(@bind);
1188 # Just generate the WHERE clause
1189 my($stmt, @bind) = $sql->where(\%where, \@order);
1191 # Return values in the same order, for hashed queries
1192 # See PERFORMANCE section for more details
1193 my @bind = $sql->values(\%fieldvals);
1197 This module was inspired by the excellent L<DBIx::Abstract>.
1198 However, in using that module I found that what I really wanted
1199 to do was generate SQL, but still retain complete control over my
1200 statement handles and use the DBI interface. So, I set out to
1201 create an abstract SQL generation module.
1203 While based on the concepts used by L<DBIx::Abstract>, there are
1204 several important differences, especially when it comes to WHERE
1205 clauses. I have modified the concepts used to make the SQL easier
1206 to generate from Perl data structures and, IMO, more intuitive.
1207 The underlying idea is for this module to do what you mean, based
1208 on the data structures you provide it. The big advantage is that
1209 you don't have to modify your code every time your data changes,
1210 as this module figures it out.
1212 To begin with, an SQL INSERT is as easy as just specifying a hash
1213 of C<key=value> pairs:
1216 name => 'Jimbo Bobson',
1217 phone => '123-456-7890',
1218 address => '42 Sister Lane',
1219 city => 'St. Louis',
1220 state => 'Louisiana',
1223 The SQL can then be generated with this:
1225 my($stmt, @bind) = $sql->insert('people', \%data);
1227 Which would give you something like this:
1229 $stmt = "INSERT INTO people
1230 (address, city, name, phone, state)
1231 VALUES (?, ?, ?, ?, ?)";
1232 @bind = ('42 Sister Lane', 'St. Louis', 'Jimbo Bobson',
1233 '123-456-7890', 'Louisiana');
1235 These are then used directly in your DBI code:
1237 my $sth = $dbh->prepare($stmt);
1238 $sth->execute(@bind);
1240 =head2 Inserting and Updating Arrays
1242 If your database has array types (like for example Postgres),
1243 activate the special option C<< array_datatypes => 1 >>
1244 when creating the C<SQL::Abstract> object.
1245 Then you may use an arrayref to insert and update database array types:
1247 my $sql = SQL::Abstract->new(array_datatypes => 1);
1249 planets => [qw/Mercury Venus Earth Mars/]
1252 my($stmt, @bind) = $sql->insert('solar_system', \%data);
1256 $stmt = "INSERT INTO solar_system (planets) VALUES (?)"
1258 @bind = (['Mercury', 'Venus', 'Earth', 'Mars']);
1261 =head2 Inserting and Updating SQL
1263 In order to apply SQL functions to elements of your C<%data> you may
1264 specify a reference to an arrayref for the given hash value. For example,
1265 if you need to execute the Oracle C<to_date> function on a value, you can
1266 say something like this:
1270 date_entered => \["to_date(?,'MM/DD/YYYY')", "03/02/2003"],
1273 The first value in the array is the actual SQL. Any other values are
1274 optional and would be included in the bind values array. This gives
1277 my($stmt, @bind) = $sql->insert('people', \%data);
1279 $stmt = "INSERT INTO people (name, date_entered)
1280 VALUES (?, to_date(?,'MM/DD/YYYY'))";
1281 @bind = ('Bill', '03/02/2003');
1283 An UPDATE is just as easy, all you change is the name of the function:
1285 my($stmt, @bind) = $sql->update('people', \%data);
1287 Notice that your C<%data> isn't touched; the module will generate
1288 the appropriately quirky SQL for you automatically. Usually you'll
1289 want to specify a WHERE clause for your UPDATE, though, which is
1290 where handling C<%where> hashes comes in handy...
1292 =head2 Complex where statements
1294 This module can generate pretty complicated WHERE statements
1295 easily. For example, simple C<key=value> pairs are taken to mean
1296 equality, and if you want to see if a field is within a set
1297 of values, you can use an arrayref. Let's say we wanted to
1298 SELECT some data based on this criteria:
1301 requestor => 'inna',
1302 worker => ['nwiger', 'rcwe', 'sfz'],
1303 status => { '!=', 'completed' }
1306 my($stmt, @bind) = $sql->select('tickets', '*', \%where);
1308 The above would give you something like this:
1310 $stmt = "SELECT * FROM tickets WHERE
1311 ( requestor = ? ) AND ( status != ? )
1312 AND ( worker = ? OR worker = ? OR worker = ? )";
1313 @bind = ('inna', 'completed', 'nwiger', 'rcwe', 'sfz');
1315 Which you could then use in DBI code like so:
1317 my $sth = $dbh->prepare($stmt);
1318 $sth->execute(@bind);
1324 The functions are simple. There's one for each major SQL operation,
1325 and a constructor you use first. The arguments are specified in a
1326 similar order to each function (table, then fields, then a where
1327 clause) to try and simplify things.
1332 =head2 new(option => 'value')
1334 The C<new()> function takes a list of options and values, and returns
1335 a new B<SQL::Abstract> object which can then be used to generate SQL
1336 through the methods below. The options accepted are:
1342 If set to 'lower', then SQL will be generated in all lowercase. By
1343 default SQL is generated in "textbook" case meaning something like:
1345 SELECT a_field FROM a_table WHERE some_field LIKE '%someval%'
1347 Any setting other than 'lower' is ignored.
1351 This determines what the default comparison operator is. By default
1352 it is C<=>, meaning that a hash like this:
1354 %where = (name => 'nwiger', email => 'nate@wiger.org');
1356 Will generate SQL like this:
1358 WHERE name = 'nwiger' AND email = 'nate@wiger.org'
1360 However, you may want loose comparisons by default, so if you set
1361 C<cmp> to C<like> you would get SQL such as:
1363 WHERE name like 'nwiger' AND email like 'nate@wiger.org'
1365 You can also override the comparsion on an individual basis - see
1366 the huge section on L</"WHERE CLAUSES"> at the bottom.
1368 =item sqltrue, sqlfalse
1370 Expressions for inserting boolean values within SQL statements.
1371 By default these are C<1=1> and C<1=0>. They are used
1372 by the special operators C<-in> and C<-not_in> for generating
1373 correct SQL even when the argument is an empty array (see below).
1377 This determines the default logical operator for multiple WHERE
1378 statements in arrays or hashes. If absent, the default logic is "or"
1379 for arrays, and "and" for hashes. This means that a WHERE
1383 event_date => {'>=', '2/13/99'},
1384 event_date => {'<=', '4/24/03'},
1387 will generate SQL like this:
1389 WHERE event_date >= '2/13/99' OR event_date <= '4/24/03'
1391 This is probably not what you want given this query, though (look
1392 at the dates). To change the "OR" to an "AND", simply specify:
1394 my $sql = SQL::Abstract->new(logic => 'and');
1396 Which will change the above C<WHERE> to:
1398 WHERE event_date >= '2/13/99' AND event_date <= '4/24/03'
1400 The logic can also be changed locally by inserting
1401 a modifier in front of an arrayref :
1403 @where = (-and => [event_date => {'>=', '2/13/99'},
1404 event_date => {'<=', '4/24/03'} ]);
1406 See the L</"WHERE CLAUSES"> section for explanations.
1410 This will automatically convert comparisons using the specified SQL
1411 function for both column and value. This is mostly used with an argument
1412 of C<upper> or C<lower>, so that the SQL will have the effect of
1413 case-insensitive "searches". For example, this:
1415 $sql = SQL::Abstract->new(convert => 'upper');
1416 %where = (keywords => 'MaKe iT CAse inSeNSItive');
1418 Will turn out the following SQL:
1420 WHERE upper(keywords) like upper('MaKe iT CAse inSeNSItive')
1422 The conversion can be C<upper()>, C<lower()>, or any other SQL function
1423 that can be applied symmetrically to fields (actually B<SQL::Abstract> does
1424 not validate this option; it will just pass through what you specify verbatim).
1428 This is a kludge because many databases suck. For example, you can't
1429 just bind values using DBI's C<execute()> for Oracle C<CLOB> or C<BLOB> fields.
1430 Instead, you have to use C<bind_param()>:
1432 $sth->bind_param(1, 'reg data');
1433 $sth->bind_param(2, $lots, {ora_type => ORA_CLOB});
1435 The problem is, B<SQL::Abstract> will normally just return a C<@bind> array,
1436 which loses track of which field each slot refers to. Fear not.
1438 If you specify C<bindtype> in new, you can determine how C<@bind> is returned.
1439 Currently, you can specify either C<normal> (default) or C<columns>. If you
1440 specify C<columns>, you will get an array that looks like this:
1442 my $sql = SQL::Abstract->new(bindtype => 'columns');
1443 my($stmt, @bind) = $sql->insert(...);
1446 [ 'column1', 'value1' ],
1447 [ 'column2', 'value2' ],
1448 [ 'column3', 'value3' ],
1451 You can then iterate through this manually, using DBI's C<bind_param()>.
1453 $sth->prepare($stmt);
1456 my($col, $data) = @$_;
1457 if ($col eq 'details' || $col eq 'comments') {
1458 $sth->bind_param($i, $data, {ora_type => ORA_CLOB});
1459 } elsif ($col eq 'image') {
1460 $sth->bind_param($i, $data, {ora_type => ORA_BLOB});
1462 $sth->bind_param($i, $data);
1466 $sth->execute; # execute without @bind now
1468 Now, why would you still use B<SQL::Abstract> if you have to do this crap?
1469 Basically, the advantage is still that you don't have to care which fields
1470 are or are not included. You could wrap that above C<for> loop in a simple
1471 sub called C<bind_fields()> or something and reuse it repeatedly. You still
1472 get a layer of abstraction over manual SQL specification.
1474 Note that if you set L</bindtype> to C<columns>, the C<\[$sql, @bind]>
1475 construct (see L</Literal SQL with placeholders and bind values (subqueries)>)
1476 will expect the bind values in this format.
1480 This is the character that a table or column name will be quoted
1481 with. By default this is an empty string, but you could set it to
1482 the character C<`>, to generate SQL like this:
1484 SELECT `a_field` FROM `a_table` WHERE `some_field` LIKE '%someval%'
1486 Alternatively, you can supply an array ref of two items, the first being the left
1487 hand quote character, and the second the right hand quote character. For
1488 example, you could supply C<['[',']']> for SQL Server 2000 compliant quotes
1489 that generates SQL like this:
1491 SELECT [a_field] FROM [a_table] WHERE [some_field] LIKE '%someval%'
1493 Quoting is useful if you have tables or columns names that are reserved
1494 words in your database's SQL dialect.
1498 This is the character that separates a table and column name. It is
1499 necessary to specify this when the C<quote_char> option is selected,
1500 so that tables and column names can be individually quoted like this:
1502 SELECT `table`.`one_field` FROM `table` WHERE `table`.`other_field` = 1
1504 =item array_datatypes
1506 When this option is true, arrayrefs in INSERT or UPDATE are
1507 interpreted as array datatypes and are passed directly
1509 When this option is false, arrayrefs are interpreted
1510 as literal SQL, just like refs to arrayrefs
1511 (but this behavior is for backwards compatibility; when writing
1512 new queries, use the "reference to arrayref" syntax
1518 Takes a reference to a list of "special operators"
1519 to extend the syntax understood by L<SQL::Abstract>.
1520 See section L</"SPECIAL OPERATORS"> for details.
1526 =head2 insert($table, \@values || \%fieldvals)
1528 This is the simplest function. You simply give it a table name
1529 and either an arrayref of values or hashref of field/value pairs.
1530 It returns an SQL INSERT statement and a list of bind values.
1531 See the sections on L</"Inserting and Updating Arrays"> and
1532 L</"Inserting and Updating SQL"> for information on how to insert
1533 with those data types.
1535 =head2 update($table, \%fieldvals, \%where)
1537 This takes a table, hashref of field/value pairs, and an optional
1538 hashref L<WHERE clause|/WHERE CLAUSES>. It returns an SQL UPDATE function and a list
1540 See the sections on L</"Inserting and Updating Arrays"> and
1541 L</"Inserting and Updating SQL"> for information on how to insert
1542 with those data types.
1544 =head2 select($source, $fields, $where, $order)
1546 This returns a SQL SELECT statement and associated list of bind values, as
1547 specified by the arguments :
1553 Specification of the 'FROM' part of the statement.
1554 The argument can be either a plain scalar (interpreted as a table
1555 name, will be quoted), or an arrayref (interpreted as a list
1556 of table names, joined by commas, quoted), or a scalarref
1557 (literal table name, not quoted), or a ref to an arrayref
1558 (list of literal table names, joined by commas, not quoted).
1562 Specification of the list of fields to retrieve from
1564 The argument can be either an arrayref (interpreted as a list
1565 of field names, will be joined by commas and quoted), or a
1566 plain scalar (literal SQL, not quoted).
1567 Please observe that this API is not as flexible as for
1568 the first argument C<$table>, for backwards compatibility reasons.
1572 Optional argument to specify the WHERE part of the query.
1573 The argument is most often a hashref, but can also be
1574 an arrayref or plain scalar --
1575 see section L<WHERE clause|/"WHERE CLAUSES"> for details.
1579 Optional argument to specify the ORDER BY part of the query.
1580 The argument can be a scalar, a hashref or an arrayref
1581 -- see section L<ORDER BY clause|/"ORDER BY CLAUSES">
1587 =head2 delete($table, \%where)
1589 This takes a table name and optional hashref L<WHERE clause|/WHERE CLAUSES>.
1590 It returns an SQL DELETE statement and list of bind values.
1592 =head2 where(\%where, \@order)
1594 This is used to generate just the WHERE clause. For example,
1595 if you have an arbitrary data structure and know what the
1596 rest of your SQL is going to look like, but want an easy way
1597 to produce a WHERE clause, use this. It returns an SQL WHERE
1598 clause and list of bind values.
1601 =head2 values(\%data)
1603 This just returns the values from the hash C<%data>, in the same
1604 order that would be returned from any of the other above queries.
1605 Using this allows you to markedly speed up your queries if you
1606 are affecting lots of rows. See below under the L</"PERFORMANCE"> section.
1608 =head2 generate($any, 'number', $of, \@data, $struct, \%types)
1610 Warning: This is an experimental method and subject to change.
1612 This returns arbitrarily generated SQL. It's a really basic shortcut.
1613 It will return two different things, depending on return context:
1615 my($stmt, @bind) = $sql->generate('create table', \$table, \@fields);
1616 my $stmt_and_val = $sql->generate('create table', \$table, \@fields);
1618 These would return the following:
1620 # First calling form
1621 $stmt = "CREATE TABLE test (?, ?)";
1622 @bind = (field1, field2);
1624 # Second calling form
1625 $stmt_and_val = "CREATE TABLE test (field1, field2)";
1627 Depending on what you're trying to do, it's up to you to choose the correct
1628 format. In this example, the second form is what you would want.
1632 $sql->generate('alter session', { nls_date_format => 'MM/YY' });
1636 ALTER SESSION SET nls_date_format = 'MM/YY'
1638 You get the idea. Strings get their case twiddled, but everything
1639 else remains verbatim.
1644 =head1 WHERE CLAUSES
1648 This module uses a variation on the idea from L<DBIx::Abstract>. It
1649 is B<NOT>, repeat I<not> 100% compatible. B<The main logic of this
1650 module is that things in arrays are OR'ed, and things in hashes
1653 The easiest way to explain is to show lots of examples. After
1654 each C<%where> hash shown, it is assumed you used:
1656 my($stmt, @bind) = $sql->where(\%where);
1658 However, note that the C<%where> hash can be used directly in any
1659 of the other functions as well, as described above.
1661 =head2 Key-value pairs
1663 So, let's get started. To begin, a simple hash:
1667 status => 'completed'
1670 Is converted to SQL C<key = val> statements:
1672 $stmt = "WHERE user = ? AND status = ?";
1673 @bind = ('nwiger', 'completed');
1675 One common thing I end up doing is having a list of values that
1676 a field can be in. To do this, simply specify a list inside of
1681 status => ['assigned', 'in-progress', 'pending'];
1684 This simple code will create the following:
1686 $stmt = "WHERE user = ? AND ( status = ? OR status = ? OR status = ? )";
1687 @bind = ('nwiger', 'assigned', 'in-progress', 'pending');
1689 A field associated to an empty arrayref will be considered a
1690 logical false and will generate 0=1.
1692 =head2 Specific comparison operators
1694 If you want to specify a different type of operator for your comparison,
1695 you can use a hashref for a given column:
1699 status => { '!=', 'completed' }
1702 Which would generate:
1704 $stmt = "WHERE user = ? AND status != ?";
1705 @bind = ('nwiger', 'completed');
1707 To test against multiple values, just enclose the values in an arrayref:
1709 status => { '=', ['assigned', 'in-progress', 'pending'] };
1711 Which would give you:
1713 "WHERE status = ? OR status = ? OR status = ?"
1716 The hashref can also contain multiple pairs, in which case it is expanded
1717 into an C<AND> of its elements:
1721 status => { '!=', 'completed', -not_like => 'pending%' }
1724 # Or more dynamically, like from a form
1725 $where{user} = 'nwiger';
1726 $where{status}{'!='} = 'completed';
1727 $where{status}{'-not_like'} = 'pending%';
1729 # Both generate this
1730 $stmt = "WHERE user = ? AND status != ? AND status NOT LIKE ?";
1731 @bind = ('nwiger', 'completed', 'pending%');
1734 To get an OR instead, you can combine it with the arrayref idea:
1738 priority => [ {'=', 2}, {'!=', 1} ]
1741 Which would generate:
1743 $stmt = "WHERE user = ? AND priority = ? OR priority != ?";
1744 @bind = ('nwiger', '2', '1');
1746 If you want to include literal SQL (with or without bind values), just use a
1747 scalar reference or array reference as the value:
1750 date_entered => { '>' => \["to_date(?, 'MM/DD/YYYY')", "11/26/2008"] },
1751 date_expires => { '<' => \"now()" }
1754 Which would generate:
1756 $stmt = "WHERE date_entered > "to_date(?, 'MM/DD/YYYY') AND date_expires < now()";
1757 @bind = ('11/26/2008');
1760 =head2 Logic and nesting operators
1762 In the example above,
1763 there is a subtle trap if you want to say something like
1764 this (notice the C<AND>):
1766 WHERE priority != ? AND priority != ?
1768 Because, in Perl you I<can't> do this:
1770 priority => { '!=', 2, '!=', 1 }
1772 As the second C<!=> key will obliterate the first. The solution
1773 is to use the special C<-modifier> form inside an arrayref:
1775 priority => [ -and => {'!=', 2},
1779 Normally, these would be joined by C<OR>, but the modifier tells it
1780 to use C<AND> instead. (Hint: You can use this in conjunction with the
1781 C<logic> option to C<new()> in order to change the way your queries
1782 work by default.) B<Important:> Note that the C<-modifier> goes
1783 B<INSIDE> the arrayref, as an extra first element. This will
1784 B<NOT> do what you think it might:
1786 priority => -and => [{'!=', 2}, {'!=', 1}] # WRONG!
1788 Here is a quick list of equivalencies, since there is some overlap:
1791 status => {'!=', 'completed', 'not like', 'pending%' }
1792 status => [ -and => {'!=', 'completed'}, {'not like', 'pending%'}]
1795 status => {'=', ['assigned', 'in-progress']}
1796 status => [ -or => {'=', 'assigned'}, {'=', 'in-progress'}]
1797 status => [ {'=', 'assigned'}, {'=', 'in-progress'} ]
1801 =head2 Special operators : IN, BETWEEN, etc.
1803 You can also use the hashref format to compare a list of fields using the
1804 C<IN> comparison operator, by specifying the list as an arrayref:
1807 status => 'completed',
1808 reportid => { -in => [567, 2335, 2] }
1811 Which would generate:
1813 $stmt = "WHERE status = ? AND reportid IN (?,?,?)";
1814 @bind = ('completed', '567', '2335', '2');
1816 The reverse operator C<-not_in> generates SQL C<NOT IN> and is used in
1819 If the argument to C<-in> is an empty array, 'sqlfalse' is generated
1820 (by default : C<1=0>). Similarly, C<< -not_in => [] >> generates
1821 'sqltrue' (by default : C<1=1>).
1825 Another pair of operators is C<-between> and C<-not_between>,
1826 used with an arrayref of two values:
1830 completion_date => {
1831 -not_between => ['2002-10-01', '2003-02-06']
1837 WHERE user = ? AND completion_date NOT BETWEEN ( ? AND ? )
1839 These are the two builtin "special operators"; but the
1840 list can be expanded : see section L</"SPECIAL OPERATORS"> below.
1842 =head2 Nested conditions, -and/-or prefixes
1844 So far, we've seen how multiple conditions are joined with a top-level
1845 C<AND>. We can change this by putting the different conditions we want in
1846 hashes and then putting those hashes in an array. For example:
1851 status => { -like => ['pending%', 'dispatched'] },
1855 status => 'unassigned',
1859 This data structure would create the following:
1861 $stmt = "WHERE ( user = ? AND ( status LIKE ? OR status LIKE ? ) )
1862 OR ( user = ? AND status = ? ) )";
1863 @bind = ('nwiger', 'pending', 'dispatched', 'robot', 'unassigned');
1866 There is also a special C<-nest>
1867 operator which adds an additional set of parens, to create a subquery.
1868 For example, to get something like this:
1870 $stmt = "WHERE user = ? AND ( workhrs > ? OR geo = ? )";
1871 @bind = ('nwiger', '20', 'ASIA');
1877 -nest => [ workhrs => {'>', 20}, geo => 'ASIA' ],
1881 Finally, clauses in hashrefs or arrayrefs can be
1882 prefixed with an C<-and> or C<-or> to change the logic
1889 -and => [workhrs => {'>', 20}, geo => 'ASIA' ],
1890 -and => [workhrs => {'<', 50}, geo => 'EURO' ]
1897 WHERE ( user = ? AND
1898 ( ( workhrs > ? AND geo = ? )
1899 OR ( workhrs < ? AND geo = ? ) ) )
1902 =head2 Algebraic inconsistency, for historical reasons
1904 C<Important note>: when connecting several conditions, the C<-and->|C<-or>
1905 operator goes C<outside> of the nested structure; whereas when connecting
1906 several constraints on one column, the C<-and> operator goes
1907 C<inside> the arrayref. Here is an example combining both features :
1910 -and => [a => 1, b => 2],
1911 -or => [c => 3, d => 4],
1912 e => [-and => {-like => 'foo%'}, {-like => '%bar'} ]
1917 WHERE ( ( ( a = ? AND b = ? )
1918 OR ( c = ? OR d = ? )
1919 OR ( e LIKE ? AND e LIKE ? ) ) )
1921 This difference in syntax is unfortunate but must be preserved for
1922 historical reasons. So be careful : the two examples below would
1923 seem algebraically equivalent, but they are not
1925 {col => [-and => {-like => 'foo%'}, {-like => '%bar'}]}
1926 # yields : WHERE ( ( col LIKE ? AND col LIKE ? ) )
1928 [-and => {col => {-like => 'foo%'}, {col => {-like => '%bar'}}]]
1929 # yields : WHERE ( ( col LIKE ? OR col LIKE ? ) )
1934 Finally, sometimes only literal SQL will do. If you want to include
1935 literal SQL verbatim, you can specify it as a scalar reference, namely:
1937 my $inn = 'is Not Null';
1939 priority => { '<', 2 },
1945 $stmt = "WHERE priority < ? AND requestor is Not Null";
1948 Note that in this example, you only get one bind parameter back, since
1949 the verbatim SQL is passed as part of the statement.
1951 Of course, just to prove a point, the above can also be accomplished
1955 priority => { '<', 2 },
1956 requestor => { '!=', undef },
1962 Conditions on boolean columns can be expressed in the
1963 same way, passing a reference to an empty string :
1966 priority => { '<', 2 },
1972 $stmt = "WHERE priority < ? AND is_ready";
1976 =head2 Literal SQL with placeholders and bind values (subqueries)
1978 If the literal SQL to be inserted has placeholders and bind values,
1979 use a reference to an arrayref (yes this is a double reference --
1980 not so common, but perfectly legal Perl). For example, to find a date
1981 in Postgres you can use something like this:
1984 date_column => \[q/= date '2008-09-30' - ?::integer/, 10/]
1989 $stmt = "WHERE ( date_column = date '2008-09-30' - ?::integer )"
1992 Note that you must pass the bind values in the same format as they are returned
1993 by L</where>. That means that if you set L</bindtype> to C<columns>, you must
1994 provide the bind values in the C<< [ column_meta => value ] >> format, where
1995 C<column_meta> is an opaque scalar value; most commonly the column name, but
1996 you can use any scalar value (including references and blessed references),
1997 L<SQL::Abstract> will simply pass it through intact. So if C<bindtype> is set
1998 to C<columns> the above example will look like:
2001 date_column => \[q/= date '2008-09-30' - ?::integer/, [ dummy => 10 ]/]
2004 Literal SQL is especially useful for nesting parenthesized clauses in the
2005 main SQL query. Here is a first example :
2007 my ($sub_stmt, @sub_bind) = ("SELECT c1 FROM t1 WHERE c2 < ? AND c3 LIKE ?",
2011 bar => \["IN ($sub_stmt)" => @sub_bind],
2016 $stmt = "WHERE (foo = ? AND bar IN (SELECT c1 FROM t1
2017 WHERE c2 < ? AND c3 LIKE ?))";
2018 @bind = (1234, 100, "foo%");
2020 Other subquery operators, like for example C<"E<gt> ALL"> or C<"NOT IN">,
2021 are expressed in the same way. Of course the C<$sub_stmt> and
2022 its associated bind values can be generated through a former call
2025 my ($sub_stmt, @sub_bind)
2026 = $sql->select("t1", "c1", {c2 => {"<" => 100},
2027 c3 => {-like => "foo%"}});
2030 bar => \["> ALL ($sub_stmt)" => @sub_bind],
2033 In the examples above, the subquery was used as an operator on a column;
2034 but the same principle also applies for a clause within the main C<%where>
2035 hash, like an EXISTS subquery :
2037 my ($sub_stmt, @sub_bind)
2038 = $sql->select("t1", "*", {c1 => 1, c2 => \"> t0.c0"});
2041 -nest => \["EXISTS ($sub_stmt)" => @sub_bind],
2046 $stmt = "WHERE (foo = ? AND EXISTS (SELECT * FROM t1
2047 WHERE c1 = ? AND c2 > t0.c0))";
2051 Observe that the condition on C<c2> in the subquery refers to
2052 column C<t0.c0> of the main query : this is I<not> a bind
2053 value, so we have to express it through a scalar ref.
2054 Writing C<< c2 => {">" => "t0.c0"} >> would have generated
2055 C<< c2 > ? >> with bind value C<"t0.c0"> ... not exactly
2056 what we wanted here.
2058 Another use of the subquery technique is when some SQL clauses need
2059 parentheses, as it often occurs with some proprietary SQL extensions
2060 like for example fulltext expressions, geospatial expressions,
2061 NATIVE clauses, etc. Here is an example of a fulltext query in MySQL :
2064 -nest => \["MATCH (col1, col2) AGAINST (?)" => qw/apples/]
2067 Finally, here is an example where a subquery is used
2068 for expressing unary negation:
2070 my ($sub_stmt, @sub_bind)
2071 = $sql->where({age => [{"<" => 10}, {">" => 20}]});
2072 $sub_stmt =~ s/^ where //i; # don't want "WHERE" in the subclause
2074 lname => {like => '%son%'},
2075 -nest => \["NOT ($sub_stmt)" => @sub_bind],
2080 $stmt = "lname LIKE ? AND NOT ( age < ? OR age > ? )"
2081 @bind = ('%son%', 10, 20)
2087 These pages could go on for a while, since the nesting of the data
2088 structures this module can handle are pretty much unlimited (the
2089 module implements the C<WHERE> expansion as a recursive function
2090 internally). Your best bet is to "play around" with the module a
2091 little to see how the data structures behave, and choose the best
2092 format for your data based on that.
2094 And of course, all the values above will probably be replaced with
2095 variables gotten from forms or the command line. After all, if you
2096 knew everything ahead of time, you wouldn't have to worry about
2097 dynamically-generating SQL and could just hardwire it into your
2103 =head1 ORDER BY CLAUSES
2105 Some functions take an order by clause. This can either be a scalar (just a
2106 column name,) a hash of C<< { -desc => 'col' } >> or C<< { -asc => 'col' } >>,
2107 or an array of either of the two previous forms. Examples:
2109 Given | Will Generate
2110 ----------------------------------------------------------
2112 \'colA DESC' | ORDER BY colA DESC
2114 'colA' | ORDER BY colA
2116 [qw/colA colB/] | ORDER BY colA, colB
2118 {-asc => 'colA'} | ORDER BY colA ASC
2120 {-desc => 'colB'} | ORDER BY colB DESC
2122 ['colA', {-asc => 'colB'}] | ORDER BY colA, colB ASC
2124 { -asc => [qw/colA colB] } | ORDER BY colA ASC, colB ASC
2127 { -asc => 'colA' }, | ORDER BY colA ASC, colB DESC,
2128 { -desc => [qw/colB/], | colC ASC, colD ASC
2129 { -asc => [qw/colC colD/],|
2131 ===========================================================
2135 =head1 SPECIAL OPERATORS
2137 my $sqlmaker = SQL::Abstract->new(special_ops => [
2141 my ($self, $field, $op, $arg) = @_;
2147 handler => 'method_name',
2151 A "special operator" is a SQL syntactic clause that can be
2152 applied to a field, instead of a usual binary operator.
2155 WHERE field IN (?, ?, ?)
2156 WHERE field BETWEEN ? AND ?
2157 WHERE MATCH(field) AGAINST (?, ?)
2159 Special operators IN and BETWEEN are fairly standard and therefore
2160 are builtin within C<SQL::Abstract> (as the overridable methods
2161 C<_where_field_IN> and C<_where_field_BETWEEN>). For other operators,
2162 like the MATCH .. AGAINST example above which is specific to MySQL,
2163 you can write your own operator handlers - supply a C<special_ops>
2164 argument to the C<new> method. That argument takes an arrayref of
2165 operator definitions; each operator definition is a hashref with two
2172 the regular expression to match the operator
2176 Either a coderef or a plain scalar method name. In both cases
2177 the expected return is C<< ($sql, @bind) >>.
2179 When supplied with a method name, it is simply called on the
2180 L<SQL::Abstract/> object as:
2182 $self->$method_name ($field, $op, $arg)
2186 $op is the part that matched the handler regex
2187 $field is the LHS of the operator
2190 When supplied with a coderef, it is called as:
2192 $coderef->($self, $field, $op, $arg)
2197 For example, here is an implementation
2198 of the MATCH .. AGAINST syntax for MySQL
2200 my $sqlmaker = SQL::Abstract->new(special_ops => [
2202 # special op for MySql MATCH (field) AGAINST(word1, word2, ...)
2203 {regex => qr/^match$/i,
2205 my ($self, $field, $op, $arg) = @_;
2206 $arg = [$arg] if not ref $arg;
2207 my $label = $self->_quote($field);
2208 my ($placeholder) = $self->_convert('?');
2209 my $placeholders = join ", ", (($placeholder) x @$arg);
2210 my $sql = $self->_sqlcase('match') . " ($label) "
2211 . $self->_sqlcase('against') . " ($placeholders) ";
2212 my @bind = $self->_bindtype($field, @$arg);
2213 return ($sql, @bind);
2222 Thanks to some benchmarking by Mark Stosberg, it turns out that
2223 this module is many orders of magnitude faster than using C<DBIx::Abstract>.
2224 I must admit this wasn't an intentional design issue, but it's a
2225 byproduct of the fact that you get to control your C<DBI> handles
2228 To maximize performance, use a code snippet like the following:
2230 # prepare a statement handle using the first row
2231 # and then reuse it for the rest of the rows
2233 for my $href (@array_of_hashrefs) {
2234 $stmt ||= $sql->insert('table', $href);
2235 $sth ||= $dbh->prepare($stmt);
2236 $sth->execute($sql->values($href));
2239 The reason this works is because the keys in your C<$href> are sorted
2240 internally by B<SQL::Abstract>. Thus, as long as your data retains
2241 the same structure, you only have to generate the SQL the first time
2242 around. On subsequent queries, simply use the C<values> function provided
2243 by this module to return your values in the correct order.
2248 If you use my C<CGI::FormBuilder> module at all, you'll hopefully
2249 really like this part (I do, at least). Building up a complex query
2250 can be as simple as the following:
2254 use CGI::FormBuilder;
2257 my $form = CGI::FormBuilder->new(...);
2258 my $sql = SQL::Abstract->new;
2260 if ($form->submitted) {
2261 my $field = $form->field;
2262 my $id = delete $field->{id};
2263 my($stmt, @bind) = $sql->update('table', $field, {id => $id});
2266 Of course, you would still have to connect using C<DBI> to run the
2267 query, but the point is that if you make your form look like your
2268 table, the actual query script can be extremely simplistic.
2270 If you're B<REALLY> lazy (I am), check out C<HTML::QuickTable> for
2271 a fast interface to returning and formatting data. I frequently
2272 use these three modules together to write complex database query
2273 apps in under 50 lines.
2278 Version 1.50 was a major internal refactoring of C<SQL::Abstract>.
2279 Great care has been taken to preserve the I<published> behavior
2280 documented in previous versions in the 1.* family; however,
2281 some features that were previously undocumented, or behaved
2282 differently from the documentation, had to be changed in order
2283 to clarify the semantics. Hence, client code that was relying
2284 on some dark areas of C<SQL::Abstract> v1.*
2285 B<might behave differently> in v1.50.
2287 The main changes are :
2293 support for literal SQL through the C<< \ [$sql, bind] >> syntax.
2297 support for the { operator => \"..." } construct (to embed literal SQL)
2301 support for the { operator => \["...", @bind] } construct (to embed literal SQL with bind values)
2305 optional support for L<array datatypes|/"Inserting and Updating Arrays">
2309 defensive programming : check arguments
2313 fixed bug with global logic, which was previously implemented
2314 through global variables yielding side-effects. Prior versions would
2315 interpret C<< [ {cond1, cond2}, [cond3, cond4] ] >>
2316 as C<< "(cond1 AND cond2) OR (cond3 AND cond4)" >>.
2317 Now this is interpreted
2318 as C<< "(cond1 AND cond2) OR (cond3 OR cond4)" >>.
2323 fixed semantics of _bindtype on array args
2327 dropped the C<_anoncopy> of the %where tree. No longer necessary,
2328 we just avoid shifting arrays within that tree.
2332 dropped the C<_modlogic> function
2338 =head1 ACKNOWLEDGEMENTS
2340 There are a number of individuals that have really helped out with
2341 this module. Unfortunately, most of them submitted bugs via CPAN
2342 so I have no idea who they are! But the people I do know are:
2344 Ash Berlin (order_by hash term support)
2345 Matt Trout (DBIx::Class support)
2346 Mark Stosberg (benchmarking)
2347 Chas Owens (initial "IN" operator support)
2348 Philip Collins (per-field SQL functions)
2349 Eric Kolve (hashref "AND" support)
2350 Mike Fragassi (enhancements to "BETWEEN" and "LIKE")
2351 Dan Kubb (support for "quote_char" and "name_sep")
2352 Guillermo Roditi (patch to cleanup "IN" and "BETWEEN", fix and tests for _order_by)
2353 Laurent Dami (internal refactoring, multiple -nest, extensible list of special operators, literal SQL)
2354 Norbert Buchmuller (support for literal SQL in hashpair, misc. fixes & tests)
2355 Peter Rabbitson (rewrite of SQLA::Test, misc. fixes & tests)
2361 L<DBIx::Class>, L<DBIx::Abstract>, L<CGI::FormBuilder>, L<HTML::QuickTable>.
2365 Copyright (c) 2001-2007 Nathan Wiger <nwiger@cpan.org>. All Rights Reserved.
2367 This module is actively maintained by Matt Trout <mst@shadowcatsystems.co.uk>
2369 For support, your best bet is to try the C<DBIx::Class> users mailing list.
2370 While not an official support venue, C<DBIx::Class> makes heavy use of
2371 C<SQL::Abstract>, and as such list members there are very familiar with
2372 how to create queries.
2376 This module is free software; you may copy this under the terms of
2377 the GNU General Public License, or the Artistic License, copies of
2378 which should have accompanied your Perl kit.