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 $VERSION = '1.54';
20 # This would confuse some packagers
21 #$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) = @_;
618 my @vals = @$vals; #always work on a copy
621 $self->_debug("ARRAY($vals) means multiple elements: [ @vals ]");
623 # see if the first element is an -and/-or op
625 if ($vals[0] =~ /^ - ( AND|OR ) $/ix) {
630 # distribute $op over each remaining member of @vals, append logic if exists
631 return $self->_recurse_where([map { {$k => {$op, $_}} } @vals], $logic);
633 # LDNOTE : had planned to change the distribution logic when
634 # $op =~ $self->{inequality_op}, because of Morgan laws :
635 # with {field => {'!=' => [22, 33]}}, it would be ridiculous to generate
636 # WHERE field != 22 OR field != 33 : the user probably means
637 # WHERE field != 22 AND field != 33.
638 # To do this, replace the above to roughly :
639 # my $logic = ($op =~ $self->{inequality_op}) ? 'AND' : 'OR';
640 # return $self->_recurse_where([map { {$k => {$op, $_}} } @vals], $logic);
644 # try to DWIM on equality operators
645 # LDNOTE : not 100% sure this is the correct thing to do ...
646 return ($self->{sqlfalse}) if $op =~ $self->{equality_op};
647 return ($self->{sqltrue}) if $op =~ $self->{inequality_op};
650 puke "operator '$op' applied on an empty array (field '$k')";
655 sub _where_hashpair_SCALARREF {
656 my ($self, $k, $v) = @_;
657 $self->_debug("SCALAR($k) means literal SQL: $$v");
658 my $sql = $self->_quote($k) . " " . $$v;
662 # literal SQL with bind
663 sub _where_hashpair_ARRAYREFREF {
664 my ($self, $k, $v) = @_;
665 $self->_debug("REF($k) means literal SQL: @${$v}");
666 my ($sql, @bind) = @${$v};
667 $self->_assert_bindval_matches_bindtype(@bind);
668 $sql = $self->_quote($k) . " " . $sql;
669 return ($sql, @bind );
672 # literal SQL without bind
673 sub _where_hashpair_SCALAR {
674 my ($self, $k, $v) = @_;
675 $self->_debug("NOREF($k) means simple key=val: $k $self->{cmp} $v");
676 my $sql = join ' ', $self->_convert($self->_quote($k)),
677 $self->_sqlcase($self->{cmp}),
678 $self->_convert('?');
679 my @bind = $self->_bindtype($k, $v);
680 return ( $sql, @bind);
684 sub _where_hashpair_UNDEF {
685 my ($self, $k, $v) = @_;
686 $self->_debug("UNDEF($k) means IS NULL");
687 my $sql = $self->_quote($k) . $self->_sqlcase(' is null');
691 #======================================================================
692 # WHERE: TOP-LEVEL OTHERS (SCALARREF, SCALAR, UNDEF)
693 #======================================================================
696 sub _where_SCALARREF {
697 my ($self, $where) = @_;
700 $self->_debug("SCALAR(*top) means literal SQL: $$where");
706 my ($self, $where) = @_;
709 $self->_debug("NOREF(*top) means literal SQL: $where");
720 #======================================================================
721 # WHERE: BUILTIN SPECIAL OPERATORS (-in, -between)
722 #======================================================================
725 sub _where_field_BETWEEN {
726 my ($self, $k, $op, $vals) = @_;
728 (ref $vals eq 'ARRAY' && @$vals == 2) or
729 (ref $vals eq 'REF' && (@$$vals == 1 || @$$vals == 2 || @$$vals == 3))
730 or puke "special op 'between' requires an arrayref of two values (or a scalarref or arrayrefref for literal SQL)";
732 my ($clause, @bind, $label, $and, $placeholder);
733 $label = $self->_convert($self->_quote($k));
734 $and = ' ' . $self->_sqlcase('and') . ' ';
735 $placeholder = $self->_convert('?');
736 $op = $self->_sqlcase($op);
738 if (ref $vals eq 'REF') {
739 ($clause, @bind) = @$$vals;
742 my (@all_sql, @all_bind);
744 foreach my $val (@$vals) {
745 my ($sql, @bind) = $self->_SWITCH_refkind($val, {
747 return ($placeholder, ($val));
750 return ($self->_convert($$val), ());
754 push @all_bind, @bind;
757 $clause = (join $and, @all_sql);
758 @bind = $self->_bindtype($k, @all_bind);
760 my $sql = "( $label $op $clause )";
765 sub _where_field_IN {
766 my ($self, $k, $op, $vals) = @_;
768 # backwards compatibility : if scalar, force into an arrayref
769 $vals = [$vals] if defined $vals && ! ref $vals;
771 my ($label) = $self->_convert($self->_quote($k));
772 my ($placeholder) = $self->_convert('?');
773 $op = $self->_sqlcase($op);
775 my ($sql, @bind) = $self->_SWITCH_refkind($vals, {
776 ARRAYREF => sub { # list of choices
777 if (@$vals) { # nonempty list
778 my $placeholders = join ", ", (($placeholder) x @$vals);
779 my $sql = "$label $op ( $placeholders )";
780 my @bind = $self->_bindtype($k, @$vals);
782 return ($sql, @bind);
784 else { # empty list : some databases won't understand "IN ()", so DWIM
785 my $sql = ($op =~ /\bnot\b/i) ? $self->{sqltrue} : $self->{sqlfalse};
790 ARRAYREFREF => sub { # literal SQL with bind
791 my ($sql, @bind) = @$$vals;
792 $self->_assert_bindval_matches_bindtype(@bind);
793 return ("$label $op ( $sql )", @bind);
797 puke "special op 'in' requires an arrayref (or arrayref-ref)";
801 return ($sql, @bind);
809 #======================================================================
811 #======================================================================
814 my ($self, $arg) = @_;
816 # construct list of ordering instructions
817 my @order = $self->_SWITCH_refkind($arg, {
820 map {$self->_SWITCH_refkind($_, {
821 SCALAR => sub {$self->_quote($_)},
823 SCALARREF => sub {$$_}, # literal SQL, no quoting
824 HASHREF => sub {$self->_order_by_hash($_)}
828 SCALAR => sub {$self->_quote($arg)},
830 SCALARREF => sub {$$arg}, # literal SQL, no quoting
831 HASHREF => sub {$self->_order_by_hash($arg)},
836 my $order = join ', ', @order;
837 return $order ? $self->_sqlcase(' order by')." $order" : '';
842 my ($self, $hash) = @_;
844 # get first pair in hash
845 my ($key, $val) = each %$hash;
847 # check if one pair was found and no other pair in hash
848 $key && !(each %$hash)
849 or puke "hash passed to _order_by must have exactly one key (-desc or -asc)";
851 my ($order) = ($key =~ /^-(desc|asc)/i)
852 or puke "invalid key in _order_by hash : $key";
854 $val = ref $val eq 'ARRAY' ? $val : [$val];
855 return join ', ', map { $self->_quote($_) . ' ' . $self->_sqlcase($order) } @$val;
860 #======================================================================
861 # DATASOURCE (FOR NOW, JUST PLAIN TABLE OR LIST OF TABLES)
862 #======================================================================
867 $self->_SWITCH_refkind($from, {
868 ARRAYREF => sub {join ', ', map { $self->_quote($_) } @$from;},
869 SCALAR => sub {$self->_quote($from)},
870 SCALARREF => sub {$$from},
871 ARRAYREFREF => sub {join ', ', @$from;},
876 #======================================================================
878 #======================================================================
884 $label or puke "can't quote an empty label";
886 # left and right quote characters
887 my ($ql, $qr, @other) = $self->_SWITCH_refkind($self->{quote_char}, {
888 SCALAR => sub {($self->{quote_char}, $self->{quote_char})},
889 ARRAYREF => sub {@{$self->{quote_char}}},
893 or puke "quote_char must be an arrayref of 2 values";
895 # no quoting if no quoting chars
896 $ql or return $label;
898 # no quoting for literal SQL
899 return $$label if ref($label) eq 'SCALAR';
901 # separate table / column (if applicable)
902 my $sep = $self->{name_sep} || '';
903 my @to_quote = $sep ? split /\Q$sep\E/, $label : ($label);
905 # do the quoting, except for "*" or for `table`.*
906 my @quoted = map { $_ eq '*' ? $_: $ql.$_.$qr} @to_quote;
908 # reassemble and return.
909 return join $sep, @quoted;
913 # Conversion, if applicable
915 my ($self, $arg) = @_;
917 # LDNOTE : modified the previous implementation below because
918 # it was not consistent : the first "return" is always an array,
919 # the second "return" is context-dependent. Anyway, _convert
920 # seems always used with just a single argument, so make it a
922 # return @_ unless $self->{convert};
923 # my $conv = $self->_sqlcase($self->{convert});
924 # my @ret = map { $conv.'('.$_.')' } @_;
925 # return wantarray ? @ret : $ret[0];
926 if ($self->{convert}) {
927 my $conv = $self->_sqlcase($self->{convert});
928 $arg = $conv.'('.$arg.')';
936 my($col, @vals) = @_;
938 #LDNOTE : changed original implementation below because it did not make
939 # sense when bindtype eq 'columns' and @vals > 1.
940 # return $self->{bindtype} eq 'columns' ? [ $col, @vals ] : @vals;
942 return $self->{bindtype} eq 'columns' ? map {[$col, $_]} @vals : @vals;
945 # Dies if any element of @bind is not in [colname => value] format
946 # if bindtype is 'columns'.
947 sub _assert_bindval_matches_bindtype {
948 my ($self, @bind) = @_;
950 if ($self->{bindtype} eq 'columns') {
951 foreach my $val (@bind) {
952 if (!defined $val || ref($val) ne 'ARRAY' || @$val != 2) {
953 die "bindtype 'columns' selected, you need to pass: [column_name => bind_value]"
959 sub _join_sql_clauses {
960 my ($self, $logic, $clauses_aref, $bind_aref) = @_;
962 if (@$clauses_aref > 1) {
963 my $join = " " . $self->_sqlcase($logic) . " ";
964 my $sql = '( ' . join($join, @$clauses_aref) . ' )';
965 return ($sql, @$bind_aref);
967 elsif (@$clauses_aref) {
968 return ($clauses_aref->[0], @$bind_aref); # no parentheses
971 return (); # if no SQL, ignore @$bind_aref
976 # Fix SQL case, if so requested
980 # LDNOTE: if $self->{case} is true, then it contains 'lower', so we
981 # don't touch the argument ... crooked logic, but let's not change it!
982 return $self->{case} ? $_[0] : uc($_[0]);
986 #======================================================================
987 # DISPATCHING FROM REFKIND
988 #======================================================================
991 my ($self, $data) = @_;
997 # blessed objects are treated like scalars
998 $ref = (blessed $data) ? '' : ref $data;
999 $n_steps += 1 if $ref;
1000 last if $ref ne 'REF';
1004 my $base = $ref || (defined $data ? 'SCALAR' : 'UNDEF');
1006 return $base . ('REF' x $n_steps);
1012 my ($self, $data) = @_;
1013 my @try = ($self->_refkind($data));
1014 push @try, 'SCALAR_or_UNDEF' if $try[0] eq 'SCALAR' || $try[0] eq 'UNDEF';
1015 push @try, 'FALLBACK';
1019 sub _METHOD_FOR_refkind {
1020 my ($self, $meth_prefix, $data) = @_;
1021 my $method = first {$_} map {$self->can($meth_prefix."_".$_)}
1022 $self->_try_refkind($data)
1023 or puke "cannot dispatch on '$meth_prefix' for ".$self->_refkind($data);
1028 sub _SWITCH_refkind {
1029 my ($self, $data, $dispatch_table) = @_;
1031 my $coderef = first {$_} map {$dispatch_table->{$_}}
1032 $self->_try_refkind($data)
1033 or puke "no dispatch entry for ".$self->_refkind($data);
1040 #======================================================================
1041 # VALUES, GENERATE, AUTOLOAD
1042 #======================================================================
1044 # LDNOTE: original code from nwiger, didn't touch code in that section
1045 # I feel the AUTOLOAD stuff should not be the default, it should
1046 # only be activated on explicit demand by user.
1050 my $data = shift || return;
1051 puke "Argument to ", __PACKAGE__, "->values must be a \\%hash"
1052 unless ref $data eq 'HASH';
1055 foreach my $k ( sort keys %$data ) {
1056 my $v = $data->{$k};
1057 $self->_SWITCH_refkind($v, {
1059 if ($self->{array_datatypes}) { # array datatype
1060 push @all_bind, $self->_bindtype($k, $v);
1062 else { # literal SQL with bind
1063 my ($sql, @bind) = @$v;
1064 $self->_assert_bindval_matches_bindtype(@bind);
1065 push @all_bind, @bind;
1068 ARRAYREFREF => sub { # literal SQL with bind
1069 my ($sql, @bind) = @${$v};
1070 $self->_assert_bindval_matches_bindtype(@bind);
1071 push @all_bind, @bind;
1073 SCALARREF => sub { # literal SQL without bind
1075 SCALAR_or_UNDEF => sub {
1076 push @all_bind, $self->_bindtype($k, $v);
1087 my(@sql, @sqlq, @sqlv);
1091 if ($ref eq 'HASH') {
1092 for my $k (sort keys %$_) {
1095 my $label = $self->_quote($k);
1096 if ($r eq 'ARRAY') {
1097 # literal SQL with bind
1098 my ($sql, @bind) = @$v;
1099 $self->_assert_bindval_matches_bindtype(@bind);
1100 push @sqlq, "$label = $sql";
1102 } elsif ($r eq 'SCALAR') {
1103 # literal SQL without bind
1104 push @sqlq, "$label = $$v";
1106 push @sqlq, "$label = ?";
1107 push @sqlv, $self->_bindtype($k, $v);
1110 push @sql, $self->_sqlcase('set'), join ', ', @sqlq;
1111 } elsif ($ref eq 'ARRAY') {
1112 # unlike insert(), assume these are ONLY the column names, i.e. for SQL
1115 if ($r eq 'ARRAY') { # literal SQL with bind
1116 my ($sql, @bind) = @$v;
1117 $self->_assert_bindval_matches_bindtype(@bind);
1120 } elsif ($r eq 'SCALAR') { # literal SQL without bind
1121 # embedded literal SQL
1128 push @sql, '(' . join(', ', @sqlq) . ')';
1129 } elsif ($ref eq 'SCALAR') {
1133 # strings get case twiddled
1134 push @sql, $self->_sqlcase($_);
1138 my $sql = join ' ', @sql;
1140 # this is pretty tricky
1141 # if ask for an array, return ($stmt, @bind)
1142 # otherwise, s/?/shift @sqlv/ to put it inline
1144 return ($sql, @sqlv);
1146 1 while $sql =~ s/\?/my $d = shift(@sqlv);
1147 ref $d ? $d->[1] : $d/e;
1156 # This allows us to check for a local, then _form, attr
1158 my($name) = $AUTOLOAD =~ /.*::(.+)/;
1159 return $self->generate($name, @_);
1170 SQL::Abstract - Generate SQL from Perl data structures
1176 my $sql = SQL::Abstract->new;
1178 my($stmt, @bind) = $sql->select($table, \@fields, \%where, \@order);
1180 my($stmt, @bind) = $sql->insert($table, \%fieldvals || \@values);
1182 my($stmt, @bind) = $sql->update($table, \%fieldvals, \%where);
1184 my($stmt, @bind) = $sql->delete($table, \%where);
1186 # Then, use these in your DBI statements
1187 my $sth = $dbh->prepare($stmt);
1188 $sth->execute(@bind);
1190 # Just generate the WHERE clause
1191 my($stmt, @bind) = $sql->where(\%where, \@order);
1193 # Return values in the same order, for hashed queries
1194 # See PERFORMANCE section for more details
1195 my @bind = $sql->values(\%fieldvals);
1199 This module was inspired by the excellent L<DBIx::Abstract>.
1200 However, in using that module I found that what I really wanted
1201 to do was generate SQL, but still retain complete control over my
1202 statement handles and use the DBI interface. So, I set out to
1203 create an abstract SQL generation module.
1205 While based on the concepts used by L<DBIx::Abstract>, there are
1206 several important differences, especially when it comes to WHERE
1207 clauses. I have modified the concepts used to make the SQL easier
1208 to generate from Perl data structures and, IMO, more intuitive.
1209 The underlying idea is for this module to do what you mean, based
1210 on the data structures you provide it. The big advantage is that
1211 you don't have to modify your code every time your data changes,
1212 as this module figures it out.
1214 To begin with, an SQL INSERT is as easy as just specifying a hash
1215 of C<key=value> pairs:
1218 name => 'Jimbo Bobson',
1219 phone => '123-456-7890',
1220 address => '42 Sister Lane',
1221 city => 'St. Louis',
1222 state => 'Louisiana',
1225 The SQL can then be generated with this:
1227 my($stmt, @bind) = $sql->insert('people', \%data);
1229 Which would give you something like this:
1231 $stmt = "INSERT INTO people
1232 (address, city, name, phone, state)
1233 VALUES (?, ?, ?, ?, ?)";
1234 @bind = ('42 Sister Lane', 'St. Louis', 'Jimbo Bobson',
1235 '123-456-7890', 'Louisiana');
1237 These are then used directly in your DBI code:
1239 my $sth = $dbh->prepare($stmt);
1240 $sth->execute(@bind);
1242 =head2 Inserting and Updating Arrays
1244 If your database has array types (like for example Postgres),
1245 activate the special option C<< array_datatypes => 1 >>
1246 when creating the C<SQL::Abstract> object.
1247 Then you may use an arrayref to insert and update database array types:
1249 my $sql = SQL::Abstract->new(array_datatypes => 1);
1251 planets => [qw/Mercury Venus Earth Mars/]
1254 my($stmt, @bind) = $sql->insert('solar_system', \%data);
1258 $stmt = "INSERT INTO solar_system (planets) VALUES (?)"
1260 @bind = (['Mercury', 'Venus', 'Earth', 'Mars']);
1263 =head2 Inserting and Updating SQL
1265 In order to apply SQL functions to elements of your C<%data> you may
1266 specify a reference to an arrayref for the given hash value. For example,
1267 if you need to execute the Oracle C<to_date> function on a value, you can
1268 say something like this:
1272 date_entered => \["to_date(?,'MM/DD/YYYY')", "03/02/2003"],
1275 The first value in the array is the actual SQL. Any other values are
1276 optional and would be included in the bind values array. This gives
1279 my($stmt, @bind) = $sql->insert('people', \%data);
1281 $stmt = "INSERT INTO people (name, date_entered)
1282 VALUES (?, to_date(?,'MM/DD/YYYY'))";
1283 @bind = ('Bill', '03/02/2003');
1285 An UPDATE is just as easy, all you change is the name of the function:
1287 my($stmt, @bind) = $sql->update('people', \%data);
1289 Notice that your C<%data> isn't touched; the module will generate
1290 the appropriately quirky SQL for you automatically. Usually you'll
1291 want to specify a WHERE clause for your UPDATE, though, which is
1292 where handling C<%where> hashes comes in handy...
1294 =head2 Complex where statements
1296 This module can generate pretty complicated WHERE statements
1297 easily. For example, simple C<key=value> pairs are taken to mean
1298 equality, and if you want to see if a field is within a set
1299 of values, you can use an arrayref. Let's say we wanted to
1300 SELECT some data based on this criteria:
1303 requestor => 'inna',
1304 worker => ['nwiger', 'rcwe', 'sfz'],
1305 status => { '!=', 'completed' }
1308 my($stmt, @bind) = $sql->select('tickets', '*', \%where);
1310 The above would give you something like this:
1312 $stmt = "SELECT * FROM tickets WHERE
1313 ( requestor = ? ) AND ( status != ? )
1314 AND ( worker = ? OR worker = ? OR worker = ? )";
1315 @bind = ('inna', 'completed', 'nwiger', 'rcwe', 'sfz');
1317 Which you could then use in DBI code like so:
1319 my $sth = $dbh->prepare($stmt);
1320 $sth->execute(@bind);
1326 The functions are simple. There's one for each major SQL operation,
1327 and a constructor you use first. The arguments are specified in a
1328 similar order to each function (table, then fields, then a where
1329 clause) to try and simplify things.
1334 =head2 new(option => 'value')
1336 The C<new()> function takes a list of options and values, and returns
1337 a new B<SQL::Abstract> object which can then be used to generate SQL
1338 through the methods below. The options accepted are:
1344 If set to 'lower', then SQL will be generated in all lowercase. By
1345 default SQL is generated in "textbook" case meaning something like:
1347 SELECT a_field FROM a_table WHERE some_field LIKE '%someval%'
1349 Any setting other than 'lower' is ignored.
1353 This determines what the default comparison operator is. By default
1354 it is C<=>, meaning that a hash like this:
1356 %where = (name => 'nwiger', email => 'nate@wiger.org');
1358 Will generate SQL like this:
1360 WHERE name = 'nwiger' AND email = 'nate@wiger.org'
1362 However, you may want loose comparisons by default, so if you set
1363 C<cmp> to C<like> you would get SQL such as:
1365 WHERE name like 'nwiger' AND email like 'nate@wiger.org'
1367 You can also override the comparsion on an individual basis - see
1368 the huge section on L</"WHERE CLAUSES"> at the bottom.
1370 =item sqltrue, sqlfalse
1372 Expressions for inserting boolean values within SQL statements.
1373 By default these are C<1=1> and C<1=0>. They are used
1374 by the special operators C<-in> and C<-not_in> for generating
1375 correct SQL even when the argument is an empty array (see below).
1379 This determines the default logical operator for multiple WHERE
1380 statements in arrays or hashes. If absent, the default logic is "or"
1381 for arrays, and "and" for hashes. This means that a WHERE
1385 event_date => {'>=', '2/13/99'},
1386 event_date => {'<=', '4/24/03'},
1389 will generate SQL like this:
1391 WHERE event_date >= '2/13/99' OR event_date <= '4/24/03'
1393 This is probably not what you want given this query, though (look
1394 at the dates). To change the "OR" to an "AND", simply specify:
1396 my $sql = SQL::Abstract->new(logic => 'and');
1398 Which will change the above C<WHERE> to:
1400 WHERE event_date >= '2/13/99' AND event_date <= '4/24/03'
1402 The logic can also be changed locally by inserting
1403 a modifier in front of an arrayref :
1405 @where = (-and => [event_date => {'>=', '2/13/99'},
1406 event_date => {'<=', '4/24/03'} ]);
1408 See the L</"WHERE CLAUSES"> section for explanations.
1412 This will automatically convert comparisons using the specified SQL
1413 function for both column and value. This is mostly used with an argument
1414 of C<upper> or C<lower>, so that the SQL will have the effect of
1415 case-insensitive "searches". For example, this:
1417 $sql = SQL::Abstract->new(convert => 'upper');
1418 %where = (keywords => 'MaKe iT CAse inSeNSItive');
1420 Will turn out the following SQL:
1422 WHERE upper(keywords) like upper('MaKe iT CAse inSeNSItive')
1424 The conversion can be C<upper()>, C<lower()>, or any other SQL function
1425 that can be applied symmetrically to fields (actually B<SQL::Abstract> does
1426 not validate this option; it will just pass through what you specify verbatim).
1430 This is a kludge because many databases suck. For example, you can't
1431 just bind values using DBI's C<execute()> for Oracle C<CLOB> or C<BLOB> fields.
1432 Instead, you have to use C<bind_param()>:
1434 $sth->bind_param(1, 'reg data');
1435 $sth->bind_param(2, $lots, {ora_type => ORA_CLOB});
1437 The problem is, B<SQL::Abstract> will normally just return a C<@bind> array,
1438 which loses track of which field each slot refers to. Fear not.
1440 If you specify C<bindtype> in new, you can determine how C<@bind> is returned.
1441 Currently, you can specify either C<normal> (default) or C<columns>. If you
1442 specify C<columns>, you will get an array that looks like this:
1444 my $sql = SQL::Abstract->new(bindtype => 'columns');
1445 my($stmt, @bind) = $sql->insert(...);
1448 [ 'column1', 'value1' ],
1449 [ 'column2', 'value2' ],
1450 [ 'column3', 'value3' ],
1453 You can then iterate through this manually, using DBI's C<bind_param()>.
1455 $sth->prepare($stmt);
1458 my($col, $data) = @$_;
1459 if ($col eq 'details' || $col eq 'comments') {
1460 $sth->bind_param($i, $data, {ora_type => ORA_CLOB});
1461 } elsif ($col eq 'image') {
1462 $sth->bind_param($i, $data, {ora_type => ORA_BLOB});
1464 $sth->bind_param($i, $data);
1468 $sth->execute; # execute without @bind now
1470 Now, why would you still use B<SQL::Abstract> if you have to do this crap?
1471 Basically, the advantage is still that you don't have to care which fields
1472 are or are not included. You could wrap that above C<for> loop in a simple
1473 sub called C<bind_fields()> or something and reuse it repeatedly. You still
1474 get a layer of abstraction over manual SQL specification.
1476 Note that if you set L</bindtype> to C<columns>, the C<\[$sql, @bind]>
1477 construct (see L</Literal SQL with placeholders and bind values (subqueries)>)
1478 will expect the bind values in this format.
1482 This is the character that a table or column name will be quoted
1483 with. By default this is an empty string, but you could set it to
1484 the character C<`>, to generate SQL like this:
1486 SELECT `a_field` FROM `a_table` WHERE `some_field` LIKE '%someval%'
1488 Alternatively, you can supply an array ref of two items, the first being the left
1489 hand quote character, and the second the right hand quote character. For
1490 example, you could supply C<['[',']']> for SQL Server 2000 compliant quotes
1491 that generates SQL like this:
1493 SELECT [a_field] FROM [a_table] WHERE [some_field] LIKE '%someval%'
1495 Quoting is useful if you have tables or columns names that are reserved
1496 words in your database's SQL dialect.
1500 This is the character that separates a table and column name. It is
1501 necessary to specify this when the C<quote_char> option is selected,
1502 so that tables and column names can be individually quoted like this:
1504 SELECT `table`.`one_field` FROM `table` WHERE `table`.`other_field` = 1
1506 =item array_datatypes
1508 When this option is true, arrayrefs in INSERT or UPDATE are
1509 interpreted as array datatypes and are passed directly
1511 When this option is false, arrayrefs are interpreted
1512 as literal SQL, just like refs to arrayrefs
1513 (but this behavior is for backwards compatibility; when writing
1514 new queries, use the "reference to arrayref" syntax
1520 Takes a reference to a list of "special operators"
1521 to extend the syntax understood by L<SQL::Abstract>.
1522 See section L</"SPECIAL OPERATORS"> for details.
1528 =head2 insert($table, \@values || \%fieldvals)
1530 This is the simplest function. You simply give it a table name
1531 and either an arrayref of values or hashref of field/value pairs.
1532 It returns an SQL INSERT statement and a list of bind values.
1533 See the sections on L</"Inserting and Updating Arrays"> and
1534 L</"Inserting and Updating SQL"> for information on how to insert
1535 with those data types.
1537 =head2 update($table, \%fieldvals, \%where)
1539 This takes a table, hashref of field/value pairs, and an optional
1540 hashref L<WHERE clause|/WHERE CLAUSES>. It returns an SQL UPDATE function and a list
1542 See the sections on L</"Inserting and Updating Arrays"> and
1543 L</"Inserting and Updating SQL"> for information on how to insert
1544 with those data types.
1546 =head2 select($source, $fields, $where, $order)
1548 This returns a SQL SELECT statement and associated list of bind values, as
1549 specified by the arguments :
1555 Specification of the 'FROM' part of the statement.
1556 The argument can be either a plain scalar (interpreted as a table
1557 name, will be quoted), or an arrayref (interpreted as a list
1558 of table names, joined by commas, quoted), or a scalarref
1559 (literal table name, not quoted), or a ref to an arrayref
1560 (list of literal table names, joined by commas, not quoted).
1564 Specification of the list of fields to retrieve from
1566 The argument can be either an arrayref (interpreted as a list
1567 of field names, will be joined by commas and quoted), or a
1568 plain scalar (literal SQL, not quoted).
1569 Please observe that this API is not as flexible as for
1570 the first argument C<$table>, for backwards compatibility reasons.
1574 Optional argument to specify the WHERE part of the query.
1575 The argument is most often a hashref, but can also be
1576 an arrayref or plain scalar --
1577 see section L<WHERE clause|/"WHERE CLAUSES"> for details.
1581 Optional argument to specify the ORDER BY part of the query.
1582 The argument can be a scalar, a hashref or an arrayref
1583 -- see section L<ORDER BY clause|/"ORDER BY CLAUSES">
1589 =head2 delete($table, \%where)
1591 This takes a table name and optional hashref L<WHERE clause|/WHERE CLAUSES>.
1592 It returns an SQL DELETE statement and list of bind values.
1594 =head2 where(\%where, \@order)
1596 This is used to generate just the WHERE clause. For example,
1597 if you have an arbitrary data structure and know what the
1598 rest of your SQL is going to look like, but want an easy way
1599 to produce a WHERE clause, use this. It returns an SQL WHERE
1600 clause and list of bind values.
1603 =head2 values(\%data)
1605 This just returns the values from the hash C<%data>, in the same
1606 order that would be returned from any of the other above queries.
1607 Using this allows you to markedly speed up your queries if you
1608 are affecting lots of rows. See below under the L</"PERFORMANCE"> section.
1610 =head2 generate($any, 'number', $of, \@data, $struct, \%types)
1612 Warning: This is an experimental method and subject to change.
1614 This returns arbitrarily generated SQL. It's a really basic shortcut.
1615 It will return two different things, depending on return context:
1617 my($stmt, @bind) = $sql->generate('create table', \$table, \@fields);
1618 my $stmt_and_val = $sql->generate('create table', \$table, \@fields);
1620 These would return the following:
1622 # First calling form
1623 $stmt = "CREATE TABLE test (?, ?)";
1624 @bind = (field1, field2);
1626 # Second calling form
1627 $stmt_and_val = "CREATE TABLE test (field1, field2)";
1629 Depending on what you're trying to do, it's up to you to choose the correct
1630 format. In this example, the second form is what you would want.
1634 $sql->generate('alter session', { nls_date_format => 'MM/YY' });
1638 ALTER SESSION SET nls_date_format = 'MM/YY'
1640 You get the idea. Strings get their case twiddled, but everything
1641 else remains verbatim.
1646 =head1 WHERE CLAUSES
1650 This module uses a variation on the idea from L<DBIx::Abstract>. It
1651 is B<NOT>, repeat I<not> 100% compatible. B<The main logic of this
1652 module is that things in arrays are OR'ed, and things in hashes
1655 The easiest way to explain is to show lots of examples. After
1656 each C<%where> hash shown, it is assumed you used:
1658 my($stmt, @bind) = $sql->where(\%where);
1660 However, note that the C<%where> hash can be used directly in any
1661 of the other functions as well, as described above.
1663 =head2 Key-value pairs
1665 So, let's get started. To begin, a simple hash:
1669 status => 'completed'
1672 Is converted to SQL C<key = val> statements:
1674 $stmt = "WHERE user = ? AND status = ?";
1675 @bind = ('nwiger', 'completed');
1677 One common thing I end up doing is having a list of values that
1678 a field can be in. To do this, simply specify a list inside of
1683 status => ['assigned', 'in-progress', 'pending'];
1686 This simple code will create the following:
1688 $stmt = "WHERE user = ? AND ( status = ? OR status = ? OR status = ? )";
1689 @bind = ('nwiger', 'assigned', 'in-progress', 'pending');
1691 A field associated to an empty arrayref will be considered a
1692 logical false and will generate 0=1.
1694 =head2 Specific comparison operators
1696 If you want to specify a different type of operator for your comparison,
1697 you can use a hashref for a given column:
1701 status => { '!=', 'completed' }
1704 Which would generate:
1706 $stmt = "WHERE user = ? AND status != ?";
1707 @bind = ('nwiger', 'completed');
1709 To test against multiple values, just enclose the values in an arrayref:
1711 status => { '=', ['assigned', 'in-progress', 'pending'] };
1713 Which would give you:
1715 "WHERE status = ? OR status = ? OR status = ?"
1718 The hashref can also contain multiple pairs, in which case it is expanded
1719 into an C<AND> of its elements:
1723 status => { '!=', 'completed', -not_like => 'pending%' }
1726 # Or more dynamically, like from a form
1727 $where{user} = 'nwiger';
1728 $where{status}{'!='} = 'completed';
1729 $where{status}{'-not_like'} = 'pending%';
1731 # Both generate this
1732 $stmt = "WHERE user = ? AND status != ? AND status NOT LIKE ?";
1733 @bind = ('nwiger', 'completed', 'pending%');
1736 To get an OR instead, you can combine it with the arrayref idea:
1740 priority => [ {'=', 2}, {'!=', 1} ]
1743 Which would generate:
1745 $stmt = "WHERE user = ? AND priority = ? OR priority != ?";
1746 @bind = ('nwiger', '2', '1');
1748 If you want to include literal SQL (with or without bind values), just use a
1749 scalar reference or array reference as the value:
1752 date_entered => { '>' => \["to_date(?, 'MM/DD/YYYY')", "11/26/2008"] },
1753 date_expires => { '<' => \"now()" }
1756 Which would generate:
1758 $stmt = "WHERE date_entered > "to_date(?, 'MM/DD/YYYY') AND date_expires < now()";
1759 @bind = ('11/26/2008');
1762 =head2 Logic and nesting operators
1764 In the example above,
1765 there is a subtle trap if you want to say something like
1766 this (notice the C<AND>):
1768 WHERE priority != ? AND priority != ?
1770 Because, in Perl you I<can't> do this:
1772 priority => { '!=', 2, '!=', 1 }
1774 As the second C<!=> key will obliterate the first. The solution
1775 is to use the special C<-modifier> form inside an arrayref:
1777 priority => [ -and => {'!=', 2},
1781 Normally, these would be joined by C<OR>, but the modifier tells it
1782 to use C<AND> instead. (Hint: You can use this in conjunction with the
1783 C<logic> option to C<new()> in order to change the way your queries
1784 work by default.) B<Important:> Note that the C<-modifier> goes
1785 B<INSIDE> the arrayref, as an extra first element. This will
1786 B<NOT> do what you think it might:
1788 priority => -and => [{'!=', 2}, {'!=', 1}] # WRONG!
1790 Here is a quick list of equivalencies, since there is some overlap:
1793 status => {'!=', 'completed', 'not like', 'pending%' }
1794 status => [ -and => {'!=', 'completed'}, {'not like', 'pending%'}]
1797 status => {'=', ['assigned', 'in-progress']}
1798 status => [ -or => {'=', 'assigned'}, {'=', 'in-progress'}]
1799 status => [ {'=', 'assigned'}, {'=', 'in-progress'} ]
1803 =head2 Special operators : IN, BETWEEN, etc.
1805 You can also use the hashref format to compare a list of fields using the
1806 C<IN> comparison operator, by specifying the list as an arrayref:
1809 status => 'completed',
1810 reportid => { -in => [567, 2335, 2] }
1813 Which would generate:
1815 $stmt = "WHERE status = ? AND reportid IN (?,?,?)";
1816 @bind = ('completed', '567', '2335', '2');
1818 The reverse operator C<-not_in> generates SQL C<NOT IN> and is used in
1821 If the argument to C<-in> is an empty array, 'sqlfalse' is generated
1822 (by default : C<1=0>). Similarly, C<< -not_in => [] >> generates
1823 'sqltrue' (by default : C<1=1>).
1827 Another pair of operators is C<-between> and C<-not_between>,
1828 used with an arrayref of two values:
1832 completion_date => {
1833 -not_between => ['2002-10-01', '2003-02-06']
1839 WHERE user = ? AND completion_date NOT BETWEEN ( ? AND ? )
1841 These are the two builtin "special operators"; but the
1842 list can be expanded : see section L</"SPECIAL OPERATORS"> below.
1844 =head2 Nested conditions, -and/-or prefixes
1846 So far, we've seen how multiple conditions are joined with a top-level
1847 C<AND>. We can change this by putting the different conditions we want in
1848 hashes and then putting those hashes in an array. For example:
1853 status => { -like => ['pending%', 'dispatched'] },
1857 status => 'unassigned',
1861 This data structure would create the following:
1863 $stmt = "WHERE ( user = ? AND ( status LIKE ? OR status LIKE ? ) )
1864 OR ( user = ? AND status = ? ) )";
1865 @bind = ('nwiger', 'pending', 'dispatched', 'robot', 'unassigned');
1868 There is also a special C<-nest>
1869 operator which adds an additional set of parens, to create a subquery.
1870 For example, to get something like this:
1872 $stmt = "WHERE user = ? AND ( workhrs > ? OR geo = ? )";
1873 @bind = ('nwiger', '20', 'ASIA');
1879 -nest => [ workhrs => {'>', 20}, geo => 'ASIA' ],
1883 Finally, clauses in hashrefs or arrayrefs can be
1884 prefixed with an C<-and> or C<-or> to change the logic
1891 -and => [workhrs => {'>', 20}, geo => 'ASIA' ],
1892 -and => [workhrs => {'<', 50}, geo => 'EURO' ]
1899 WHERE ( user = ? AND
1900 ( ( workhrs > ? AND geo = ? )
1901 OR ( workhrs < ? AND geo = ? ) ) )
1904 =head2 Algebraic inconsistency, for historical reasons
1906 C<Important note>: when connecting several conditions, the C<-and->|C<-or>
1907 operator goes C<outside> of the nested structure; whereas when connecting
1908 several constraints on one column, the C<-and> operator goes
1909 C<inside> the arrayref. Here is an example combining both features :
1912 -and => [a => 1, b => 2],
1913 -or => [c => 3, d => 4],
1914 e => [-and => {-like => 'foo%'}, {-like => '%bar'} ]
1919 WHERE ( ( ( a = ? AND b = ? )
1920 OR ( c = ? OR d = ? )
1921 OR ( e LIKE ? AND e LIKE ? ) ) )
1923 This difference in syntax is unfortunate but must be preserved for
1924 historical reasons. So be careful : the two examples below would
1925 seem algebraically equivalent, but they are not
1927 {col => [-and => {-like => 'foo%'}, {-like => '%bar'}]}
1928 # yields : WHERE ( ( col LIKE ? AND col LIKE ? ) )
1930 [-and => {col => {-like => 'foo%'}, {col => {-like => '%bar'}}]]
1931 # yields : WHERE ( ( col LIKE ? OR col LIKE ? ) )
1936 Finally, sometimes only literal SQL will do. If you want to include
1937 literal SQL verbatim, you can specify it as a scalar reference, namely:
1939 my $inn = 'is Not Null';
1941 priority => { '<', 2 },
1947 $stmt = "WHERE priority < ? AND requestor is Not Null";
1950 Note that in this example, you only get one bind parameter back, since
1951 the verbatim SQL is passed as part of the statement.
1953 Of course, just to prove a point, the above can also be accomplished
1957 priority => { '<', 2 },
1958 requestor => { '!=', undef },
1964 Conditions on boolean columns can be expressed in the
1965 same way, passing a reference to an empty string :
1968 priority => { '<', 2 },
1974 $stmt = "WHERE priority < ? AND is_ready";
1978 =head2 Literal SQL with placeholders and bind values (subqueries)
1980 If the literal SQL to be inserted has placeholders and bind values,
1981 use a reference to an arrayref (yes this is a double reference --
1982 not so common, but perfectly legal Perl). For example, to find a date
1983 in Postgres you can use something like this:
1986 date_column => \[q/= date '2008-09-30' - ?::integer/, 10/]
1991 $stmt = "WHERE ( date_column = date '2008-09-30' - ?::integer )"
1994 Note that you must pass the bind values in the same format as they are returned
1995 by L</where>. That means that if you set L</bindtype> to C<columns>, you must
1996 provide the bind values in the C<< [ column_meta => value ] >> format, where
1997 C<column_meta> is an opaque scalar value; most commonly the column name, but
1998 you can use any scalar value (including references and blessed references),
1999 L<SQL::Abstract> will simply pass it through intact. So if C<bindtype> is set
2000 to C<columns> the above example will look like:
2003 date_column => \[q/= date '2008-09-30' - ?::integer/, [ dummy => 10 ]/]
2006 Literal SQL is especially useful for nesting parenthesized clauses in the
2007 main SQL query. Here is a first example :
2009 my ($sub_stmt, @sub_bind) = ("SELECT c1 FROM t1 WHERE c2 < ? AND c3 LIKE ?",
2013 bar => \["IN ($sub_stmt)" => @sub_bind],
2018 $stmt = "WHERE (foo = ? AND bar IN (SELECT c1 FROM t1
2019 WHERE c2 < ? AND c3 LIKE ?))";
2020 @bind = (1234, 100, "foo%");
2022 Other subquery operators, like for example C<"E<gt> ALL"> or C<"NOT IN">,
2023 are expressed in the same way. Of course the C<$sub_stmt> and
2024 its associated bind values can be generated through a former call
2027 my ($sub_stmt, @sub_bind)
2028 = $sql->select("t1", "c1", {c2 => {"<" => 100},
2029 c3 => {-like => "foo%"}});
2032 bar => \["> ALL ($sub_stmt)" => @sub_bind],
2035 In the examples above, the subquery was used as an operator on a column;
2036 but the same principle also applies for a clause within the main C<%where>
2037 hash, like an EXISTS subquery :
2039 my ($sub_stmt, @sub_bind)
2040 = $sql->select("t1", "*", {c1 => 1, c2 => \"> t0.c0"});
2043 -nest => \["EXISTS ($sub_stmt)" => @sub_bind],
2048 $stmt = "WHERE (foo = ? AND EXISTS (SELECT * FROM t1
2049 WHERE c1 = ? AND c2 > t0.c0))";
2053 Observe that the condition on C<c2> in the subquery refers to
2054 column C<t0.c0> of the main query : this is I<not> a bind
2055 value, so we have to express it through a scalar ref.
2056 Writing C<< c2 => {">" => "t0.c0"} >> would have generated
2057 C<< c2 > ? >> with bind value C<"t0.c0"> ... not exactly
2058 what we wanted here.
2060 Another use of the subquery technique is when some SQL clauses need
2061 parentheses, as it often occurs with some proprietary SQL extensions
2062 like for example fulltext expressions, geospatial expressions,
2063 NATIVE clauses, etc. Here is an example of a fulltext query in MySQL :
2066 -nest => \["MATCH (col1, col2) AGAINST (?)" => qw/apples/]
2069 Finally, here is an example where a subquery is used
2070 for expressing unary negation:
2072 my ($sub_stmt, @sub_bind)
2073 = $sql->where({age => [{"<" => 10}, {">" => 20}]});
2074 $sub_stmt =~ s/^ where //i; # don't want "WHERE" in the subclause
2076 lname => {like => '%son%'},
2077 -nest => \["NOT ($sub_stmt)" => @sub_bind],
2082 $stmt = "lname LIKE ? AND NOT ( age < ? OR age > ? )"
2083 @bind = ('%son%', 10, 20)
2089 These pages could go on for a while, since the nesting of the data
2090 structures this module can handle are pretty much unlimited (the
2091 module implements the C<WHERE> expansion as a recursive function
2092 internally). Your best bet is to "play around" with the module a
2093 little to see how the data structures behave, and choose the best
2094 format for your data based on that.
2096 And of course, all the values above will probably be replaced with
2097 variables gotten from forms or the command line. After all, if you
2098 knew everything ahead of time, you wouldn't have to worry about
2099 dynamically-generating SQL and could just hardwire it into your
2105 =head1 ORDER BY CLAUSES
2107 Some functions take an order by clause. This can either be a scalar (just a
2108 column name,) a hash of C<< { -desc => 'col' } >> or C<< { -asc => 'col' } >>,
2109 or an array of either of the two previous forms. Examples:
2111 Given | Will Generate
2112 ----------------------------------------------------------
2114 \'colA DESC' | ORDER BY colA DESC
2116 'colA' | ORDER BY colA
2118 [qw/colA colB/] | ORDER BY colA, colB
2120 {-asc => 'colA'} | ORDER BY colA ASC
2122 {-desc => 'colB'} | ORDER BY colB DESC
2124 ['colA', {-asc => 'colB'}] | ORDER BY colA, colB ASC
2126 { -asc => [qw/colA colB] } | ORDER BY colA ASC, colB ASC
2129 { -asc => 'colA' }, | ORDER BY colA ASC, colB DESC,
2130 { -desc => [qw/colB/], | colC ASC, colD ASC
2131 { -asc => [qw/colC colD/],|
2133 ===========================================================
2137 =head1 SPECIAL OPERATORS
2139 my $sqlmaker = SQL::Abstract->new(special_ops => [
2143 my ($self, $field, $op, $arg) = @_;
2149 handler => 'method_name',
2153 A "special operator" is a SQL syntactic clause that can be
2154 applied to a field, instead of a usual binary operator.
2157 WHERE field IN (?, ?, ?)
2158 WHERE field BETWEEN ? AND ?
2159 WHERE MATCH(field) AGAINST (?, ?)
2161 Special operators IN and BETWEEN are fairly standard and therefore
2162 are builtin within C<SQL::Abstract> (as the overridable methods
2163 C<_where_field_IN> and C<_where_field_BETWEEN>). For other operators,
2164 like the MATCH .. AGAINST example above which is specific to MySQL,
2165 you can write your own operator handlers - supply a C<special_ops>
2166 argument to the C<new> method. That argument takes an arrayref of
2167 operator definitions; each operator definition is a hashref with two
2174 the regular expression to match the operator
2178 Either a coderef or a plain scalar method name. In both cases
2179 the expected return is C<< ($sql, @bind) >>.
2181 When supplied with a method name, it is simply called on the
2182 L<SQL::Abstract/> object as:
2184 $self->$method_name ($field, $op, $arg)
2188 $op is the part that matched the handler regex
2189 $field is the LHS of the operator
2192 When supplied with a coderef, it is called as:
2194 $coderef->($self, $field, $op, $arg)
2199 For example, here is an implementation
2200 of the MATCH .. AGAINST syntax for MySQL
2202 my $sqlmaker = SQL::Abstract->new(special_ops => [
2204 # special op for MySql MATCH (field) AGAINST(word1, word2, ...)
2205 {regex => qr/^match$/i,
2207 my ($self, $field, $op, $arg) = @_;
2208 $arg = [$arg] if not ref $arg;
2209 my $label = $self->_quote($field);
2210 my ($placeholder) = $self->_convert('?');
2211 my $placeholders = join ", ", (($placeholder) x @$arg);
2212 my $sql = $self->_sqlcase('match') . " ($label) "
2213 . $self->_sqlcase('against') . " ($placeholders) ";
2214 my @bind = $self->_bindtype($field, @$arg);
2215 return ($sql, @bind);
2224 Thanks to some benchmarking by Mark Stosberg, it turns out that
2225 this module is many orders of magnitude faster than using C<DBIx::Abstract>.
2226 I must admit this wasn't an intentional design issue, but it's a
2227 byproduct of the fact that you get to control your C<DBI> handles
2230 To maximize performance, use a code snippet like the following:
2232 # prepare a statement handle using the first row
2233 # and then reuse it for the rest of the rows
2235 for my $href (@array_of_hashrefs) {
2236 $stmt ||= $sql->insert('table', $href);
2237 $sth ||= $dbh->prepare($stmt);
2238 $sth->execute($sql->values($href));
2241 The reason this works is because the keys in your C<$href> are sorted
2242 internally by B<SQL::Abstract>. Thus, as long as your data retains
2243 the same structure, you only have to generate the SQL the first time
2244 around. On subsequent queries, simply use the C<values> function provided
2245 by this module to return your values in the correct order.
2250 If you use my C<CGI::FormBuilder> module at all, you'll hopefully
2251 really like this part (I do, at least). Building up a complex query
2252 can be as simple as the following:
2256 use CGI::FormBuilder;
2259 my $form = CGI::FormBuilder->new(...);
2260 my $sql = SQL::Abstract->new;
2262 if ($form->submitted) {
2263 my $field = $form->field;
2264 my $id = delete $field->{id};
2265 my($stmt, @bind) = $sql->update('table', $field, {id => $id});
2268 Of course, you would still have to connect using C<DBI> to run the
2269 query, but the point is that if you make your form look like your
2270 table, the actual query script can be extremely simplistic.
2272 If you're B<REALLY> lazy (I am), check out C<HTML::QuickTable> for
2273 a fast interface to returning and formatting data. I frequently
2274 use these three modules together to write complex database query
2275 apps in under 50 lines.
2280 Version 1.50 was a major internal refactoring of C<SQL::Abstract>.
2281 Great care has been taken to preserve the I<published> behavior
2282 documented in previous versions in the 1.* family; however,
2283 some features that were previously undocumented, or behaved
2284 differently from the documentation, had to be changed in order
2285 to clarify the semantics. Hence, client code that was relying
2286 on some dark areas of C<SQL::Abstract> v1.*
2287 B<might behave differently> in v1.50.
2289 The main changes are :
2295 support for literal SQL through the C<< \ [$sql, bind] >> syntax.
2299 support for the { operator => \"..." } construct (to embed literal SQL)
2303 support for the { operator => \["...", @bind] } construct (to embed literal SQL with bind values)
2307 optional support for L<array datatypes|/"Inserting and Updating Arrays">
2311 defensive programming : check arguments
2315 fixed bug with global logic, which was previously implemented
2316 through global variables yielding side-effects. Prior versions would
2317 interpret C<< [ {cond1, cond2}, [cond3, cond4] ] >>
2318 as C<< "(cond1 AND cond2) OR (cond3 AND cond4)" >>.
2319 Now this is interpreted
2320 as C<< "(cond1 AND cond2) OR (cond3 OR cond4)" >>.
2325 fixed semantics of _bindtype on array args
2329 dropped the C<_anoncopy> of the %where tree. No longer necessary,
2330 we just avoid shifting arrays within that tree.
2334 dropped the C<_modlogic> function
2340 =head1 ACKNOWLEDGEMENTS
2342 There are a number of individuals that have really helped out with
2343 this module. Unfortunately, most of them submitted bugs via CPAN
2344 so I have no idea who they are! But the people I do know are:
2346 Ash Berlin (order_by hash term support)
2347 Matt Trout (DBIx::Class support)
2348 Mark Stosberg (benchmarking)
2349 Chas Owens (initial "IN" operator support)
2350 Philip Collins (per-field SQL functions)
2351 Eric Kolve (hashref "AND" support)
2352 Mike Fragassi (enhancements to "BETWEEN" and "LIKE")
2353 Dan Kubb (support for "quote_char" and "name_sep")
2354 Guillermo Roditi (patch to cleanup "IN" and "BETWEEN", fix and tests for _order_by)
2355 Laurent Dami (internal refactoring, multiple -nest, extensible list of special operators, literal SQL)
2356 Norbert Buchmuller (support for literal SQL in hashpair, misc. fixes & tests)
2357 Peter Rabbitson (rewrite of SQLA::Test, misc. fixes & tests)
2363 L<DBIx::Class>, L<DBIx::Abstract>, L<CGI::FormBuilder>, L<HTML::QuickTable>.
2367 Copyright (c) 2001-2007 Nathan Wiger <nwiger@cpan.org>. All Rights Reserved.
2369 This module is actively maintained by Matt Trout <mst@shadowcatsystems.co.uk>
2371 For support, your best bet is to try the C<DBIx::Class> users mailing list.
2372 While not an official support venue, C<DBIx::Class> makes heavy use of
2373 C<SQL::Abstract>, and as such list members there are very familiar with
2374 how to create queries.
2378 This module is free software; you may copy this under the terms of
2379 the GNU General Public License, or the Artistic License, copies of
2380 which should have accompanied your Perl kit.