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 $VERSION = '1.58';
20 # This would confuse some packagers
21 #$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 # unaryish operators - key maps to handler
33 my @BUILTIN_UNARY_OPS = (
34 # the digits are backcompat stuff
35 { regex => qr/^and (?: \s? \d+ )? $/xi, handler => '_where_op_ANDOR' },
36 { regex => qr/^or (?: \s? \d+ )? $/xi, handler => '_where_op_ANDOR' },
37 { regex => qr/^nest (?: \s? \d+ )? $/xi, handler => '_where_op_NEST' },
38 { regex => qr/^ (?: not \s )? bool $/xi, handler => '_where_op_BOOL' },
41 #======================================================================
42 # DEBUGGING AND ERROR REPORTING
43 #======================================================================
46 return unless $_[0]->{debug}; shift; # a little faster
47 my $func = (caller(1))[3];
48 warn "[$func] ", @_, "\n";
52 my($func) = (caller(1))[3];
53 carp "[$func] Warning: ", @_;
57 my($func) = (caller(1))[3];
58 croak "[$func] Fatal: ", @_;
62 #======================================================================
64 #======================================================================
68 my $class = ref($self) || $self;
69 my %opt = (ref $_[0] eq 'HASH') ? %{$_[0]} : @_;
71 # choose our case by keeping an option around
72 delete $opt{case} if $opt{case} && $opt{case} ne 'lower';
74 # default logic for interpreting arrayrefs
75 $opt{logic} = $opt{logic} ? uc $opt{logic} : 'OR';
77 # how to return bind vars
78 # LDNOTE: changed nwiger code : why this 'delete' ??
79 # $opt{bindtype} ||= delete($opt{bind_type}) || 'normal';
80 $opt{bindtype} ||= 'normal';
82 # default comparison is "=", but can be overridden
85 # try to recognize which are the 'equality' and 'unequality' ops
86 # (temporary quickfix, should go through a more seasoned API)
87 $opt{equality_op} = qr/^(\Q$opt{cmp}\E|is|(is\s+)?like)$/i;
88 $opt{inequality_op} = qr/^(!=|<>|(is\s+)?not(\s+like)?)$/i;
91 $opt{sqltrue} ||= '1=1';
92 $opt{sqlfalse} ||= '0=1';
95 $opt{special_ops} ||= [];
96 push @{$opt{special_ops}}, @BUILTIN_SPECIAL_OPS;
99 $opt{unary_ops} ||= [];
100 push @{$opt{unary_ops}}, @BUILTIN_UNARY_OPS;
102 return bless \%opt, $class;
107 #======================================================================
109 #======================================================================
113 my $table = $self->_table(shift);
114 my $data = shift || return;
116 my $method = $self->_METHOD_FOR_refkind("_insert", $data);
117 my ($sql, @bind) = $self->$method($data);
118 $sql = join " ", $self->_sqlcase('insert into'), $table, $sql;
119 return wantarray ? ($sql, @bind) : $sql;
122 sub _insert_HASHREF { # explicit list of fields and then values
123 my ($self, $data) = @_;
125 my @fields = sort keys %$data;
127 my ($sql, @bind) = $self->_insert_values($data);
130 $_ = $self->_quote($_) foreach @fields;
131 $sql = "( ".join(", ", @fields).") ".$sql;
133 return ($sql, @bind);
136 sub _insert_ARRAYREF { # just generate values(?,?) part (no list of fields)
137 my ($self, $data) = @_;
139 # no names (arrayref) so can't generate bindtype
140 $self->{bindtype} ne 'columns'
141 or belch "can't do 'columns' bindtype when called with arrayref";
143 # fold the list of values into a hash of column name - value pairs
144 # (where the column names are artificially generated, and their
145 # lexicographical ordering keep the ordering of the original list)
146 my $i = "a"; # incremented values will be in lexicographical order
147 my $data_in_hash = { map { ($i++ => $_) } @$data };
149 return $self->_insert_values($data_in_hash);
152 sub _insert_ARRAYREFREF { # literal SQL with bind
153 my ($self, $data) = @_;
155 my ($sql, @bind) = @${$data};
156 $self->_assert_bindval_matches_bindtype(@bind);
158 return ($sql, @bind);
162 sub _insert_SCALARREF { # literal SQL without bind
163 my ($self, $data) = @_;
169 my ($self, $data) = @_;
171 my (@values, @all_bind);
172 foreach my $column (sort keys %$data) {
173 my $v = $data->{$column};
175 $self->_SWITCH_refkind($v, {
178 if ($self->{array_datatypes}) { # if array datatype are activated
180 push @all_bind, $self->_bindtype($column, $v);
182 else { # else literal SQL with bind
183 my ($sql, @bind) = @$v;
184 $self->_assert_bindval_matches_bindtype(@bind);
186 push @all_bind, @bind;
190 ARRAYREFREF => sub { # literal SQL with bind
191 my ($sql, @bind) = @${$v};
192 $self->_assert_bindval_matches_bindtype(@bind);
194 push @all_bind, @bind;
197 # THINK : anything useful to do with a HASHREF ?
198 HASHREF => sub { # (nothing, but old SQLA passed it through)
199 #TODO in SQLA >= 2.0 it will die instead
200 belch "HASH ref as bind value in insert is not supported";
202 push @all_bind, $self->_bindtype($column, $v);
205 SCALARREF => sub { # literal SQL without bind
209 SCALAR_or_UNDEF => sub {
211 push @all_bind, $self->_bindtype($column, $v);
218 my $sql = $self->_sqlcase('values')." ( ".join(", ", @values)." )";
219 return ($sql, @all_bind);
224 #======================================================================
226 #======================================================================
231 my $table = $self->_table(shift);
232 my $data = shift || return;
235 # first build the 'SET' part of the sql statement
236 my (@set, @all_bind);
237 puke "Unsupported data type specified to \$sql->update"
238 unless ref $data eq 'HASH';
240 for my $k (sort keys %$data) {
243 my $label = $self->_quote($k);
245 $self->_SWITCH_refkind($v, {
247 if ($self->{array_datatypes}) { # array datatype
248 push @set, "$label = ?";
249 push @all_bind, $self->_bindtype($k, $v);
251 else { # literal SQL with bind
252 my ($sql, @bind) = @$v;
253 $self->_assert_bindval_matches_bindtype(@bind);
254 push @set, "$label = $sql";
255 push @all_bind, @bind;
258 ARRAYREFREF => sub { # literal SQL with bind
259 my ($sql, @bind) = @${$v};
260 $self->_assert_bindval_matches_bindtype(@bind);
261 push @set, "$label = $sql";
262 push @all_bind, @bind;
264 SCALARREF => sub { # literal SQL without bind
265 push @set, "$label = $$v";
267 SCALAR_or_UNDEF => sub {
268 push @set, "$label = ?";
269 push @all_bind, $self->_bindtype($k, $v);
275 my $sql = $self->_sqlcase('update') . " $table " . $self->_sqlcase('set ')
279 my($where_sql, @where_bind) = $self->where($where);
281 push @all_bind, @where_bind;
284 return wantarray ? ($sql, @all_bind) : $sql;
290 #======================================================================
292 #======================================================================
297 my $table = $self->_table(shift);
298 my $fields = shift || '*';
302 my($where_sql, @bind) = $self->where($where, $order);
304 my $f = (ref $fields eq 'ARRAY') ? join ', ', map { $self->_quote($_) } @$fields
306 my $sql = join(' ', $self->_sqlcase('select'), $f,
307 $self->_sqlcase('from'), $table)
310 return wantarray ? ($sql, @bind) : $sql;
313 #======================================================================
315 #======================================================================
320 my $table = $self->_table(shift);
324 my($where_sql, @bind) = $self->where($where);
325 my $sql = $self->_sqlcase('delete from') . " $table" . $where_sql;
327 return wantarray ? ($sql, @bind) : $sql;
331 #======================================================================
333 #======================================================================
337 # Finally, a separate routine just to handle WHERE clauses
339 my ($self, $where, $order) = @_;
342 my ($sql, @bind) = $self->_recurse_where($where);
343 $sql = $sql ? $self->_sqlcase(' where ') . "( $sql )" : '';
347 $sql .= $self->_order_by($order);
350 return wantarray ? ($sql, @bind) : $sql;
355 my ($self, $where, $logic) = @_;
357 # dispatch on appropriate method according to refkind of $where
358 my $method = $self->_METHOD_FOR_refkind("_where", $where);
361 my ($sql, @bind) = $self->$method($where, $logic);
363 # DBIx::Class directly calls _recurse_where in scalar context, so
364 # we must implement it, even if not in the official API
365 return wantarray ? ($sql, @bind) : $sql;
370 #======================================================================
371 # WHERE: top-level ARRAYREF
372 #======================================================================
375 sub _where_ARRAYREF {
376 my ($self, $where, $logic) = @_;
378 $logic = uc($logic || $self->{logic});
379 $logic eq 'AND' or $logic eq 'OR' or puke "unknown logic: $logic";
381 my @clauses = @$where;
383 my (@sql_clauses, @all_bind);
384 # need to use while() so can shift() for pairs
385 while (my $el = shift @clauses) {
387 # switch according to kind of $el and get corresponding ($sql, @bind)
388 my ($sql, @bind) = $self->_SWITCH_refkind($el, {
390 # skip empty elements, otherwise get invalid trailing AND stuff
391 ARRAYREF => sub {$self->_recurse_where($el) if @$el},
393 ARRAYREFREF => sub { @{${$el}} if @{${$el}}},
395 HASHREF => sub {$self->_recurse_where($el, 'and') if %$el},
396 # LDNOTE : previous SQLA code for hashrefs was creating a dirty
397 # side-effect: the first hashref within an array would change
398 # the global logic to 'AND'. So [ {cond1, cond2}, [cond3, cond4] ]
399 # was interpreted as "(cond1 AND cond2) OR (cond3 AND cond4)",
400 # whereas it should be "(cond1 AND cond2) OR (cond3 OR cond4)".
402 SCALARREF => sub { ($$el); },
404 SCALAR => sub {# top-level arrayref with scalars, recurse in pairs
405 $self->_recurse_where({$el => shift(@clauses)})},
407 UNDEF => sub {puke "not supported : UNDEF in arrayref" },
411 push @sql_clauses, $sql;
412 push @all_bind, @bind;
416 return $self->_join_sql_clauses($logic, \@sql_clauses, \@all_bind);
419 #======================================================================
420 # WHERE: top-level ARRAYREFREF
421 #======================================================================
423 sub _where_ARRAYREFREF {
424 my ($self, $where) = @_;
425 my ($sql, @bind) = @{${$where}};
427 return ($sql, @bind);
430 #======================================================================
431 # WHERE: top-level HASHREF
432 #======================================================================
435 my ($self, $where) = @_;
436 my (@sql_clauses, @all_bind);
438 for my $k (sort keys %$where) {
439 my $v = $where->{$k};
441 # ($k => $v) is either a special op or a regular hashpair
442 my ($sql, @bind) = ($k =~ /^(-.+)/) ? $self->_where_op_in_hash($1, $v)
444 my $method = $self->_METHOD_FOR_refkind("_where_hashpair", $v);
445 $self->$method($k, $v);
448 push @sql_clauses, $sql;
449 push @all_bind, @bind;
452 return $self->_join_sql_clauses('and', \@sql_clauses, \@all_bind);
456 sub _where_op_in_hash {
457 my ($self, $orig_op, $v) = @_;
459 # put the operator in canonical form
461 $op =~ s/^-//; # remove initial dash
462 $op =~ s/[_\t ]+/ /g; # underscores and whitespace become single spaces
463 $op =~ s/^\s+|\s+$//g;# remove leading/trailing space
465 $self->_debug("OP(-$op) within hashref, recursing...");
467 my $op_entry = first {$op =~ $_->{regex}} @{$self->{unary_ops}};
468 my $handler = $op_entry->{handler};
470 puke "unknown operator: $orig_op";
472 elsif (not ref $handler) {
473 if ($op =~ s/\s?\d+$//) {
474 belch 'Use of [and|or|nest]_N modifiers is deprecated and will be removed in SQLA v2.0. '
475 . "You probably wanted ...-and => [ -$op => COND1, -$op => COND2 ... ]";
477 return $self->$handler ($op, $v);
479 elsif (ref $handler eq 'CODE') {
480 return $handler->($self, $op, $v);
483 puke "Illegal handler for operator $orig_op - expecting a method name or a coderef";
487 sub _where_op_ANDOR {
488 my ($self, $op, $v) = @_;
490 $self->_SWITCH_refkind($v, {
492 return $self->_where_ARRAYREF($v, $op);
496 return ( $op =~ /^or/i )
497 ? $self->_where_ARRAYREF( [ map { $_ => $v->{$_} } ( sort keys %$v ) ], $op )
498 : $self->_where_HASHREF($v);
502 puke "-$op => \\\$scalar not supported, use -nest => ...";
506 puke "-$op => \\[..] not supported, use -nest => ...";
509 SCALAR => sub { # permissively interpreted as SQL
510 puke "-$op => 'scalar' not supported, use -nest => \\'scalar'";
514 puke "-$op => undef not supported";
520 my ($self, $op, $v) = @_;
522 $self->_SWITCH_refkind($v, {
525 return $self->_where_ARRAYREF($v, '');
529 return $self->_where_HASHREF($v);
532 SCALARREF => sub { # literal SQL
536 ARRAYREFREF => sub { # literal SQL
540 SCALAR => sub { # permissively interpreted as SQL
541 belch "literal SQL should be -nest => \\'scalar' "
542 . "instead of -nest => 'scalar' ";
547 puke "-$op => undef not supported";
554 my ($self, $op, $v) = @_;
556 my ( $prefix, $suffix ) = ( $op =~ /\bnot\b/i )
559 $self->_SWITCH_refkind($v, {
561 my ( $sql, @bind ) = $self->_where_ARRAYREF($v, '');
562 return ( ($prefix . $sql . $suffix), @bind );
566 my ( $sql, @bind ) = @{ ${$v} };
567 return ( ($prefix . $sql . $suffix), @bind );
571 my ( $sql, @bind ) = $self->_where_HASHREF($v);
572 return ( ($prefix . $sql . $suffix), @bind );
575 SCALARREF => sub { # literal SQL
576 return ($prefix . $$v . $suffix);
579 SCALAR => sub { # interpreted as SQL column
580 return ($prefix . $self->_convert($self->_quote($v)) . $suffix);
584 puke "-$op => undef not supported";
590 sub _where_hashpair_ARRAYREF {
591 my ($self, $k, $v) = @_;
594 my @v = @$v; # need copy because of shift below
595 $self->_debug("ARRAY($k) means distribute over elements");
597 # put apart first element if it is an operator (-and, -or)
599 (defined $v[0] && $v[0] =~ /^ - (?: AND|OR ) $/ix)
603 my @distributed = map { {$k => $_} } @v;
606 $self->_debug("OP($op) reinjected into the distributed array");
607 unshift @distributed, $op;
610 my $logic = $op ? substr($op, 1) : '';
612 return $self->_recurse_where(\@distributed, $logic);
615 # LDNOTE : not sure of this one. What does "distribute over nothing" mean?
616 $self->_debug("empty ARRAY($k) means 0=1");
617 return ($self->{sqlfalse});
621 sub _where_hashpair_HASHREF {
622 my ($self, $k, $v, $logic) = @_;
625 my ($all_sql, @all_bind);
627 for my $orig_op (sort keys %$v) {
628 my $val = $v->{$orig_op};
630 # put the operator in canonical form
632 $op =~ s/^-//; # remove initial dash
633 $op =~ s/[_\t ]+/ /g; # underscores and whitespace become single spaces
634 $op =~ s/^\s+|\s+$//g;# remove leading/trailing space
638 # CASE: special operators like -in or -between
639 my $special_op = first {$op =~ $_->{regex}} @{$self->{special_ops}};
641 my $handler = $special_op->{handler};
643 puke "No handler supplied for special operator $orig_op";
645 elsif (not ref $handler) {
646 ($sql, @bind) = $self->$handler ($k, $op, $val);
648 elsif (ref $handler eq 'CODE') {
649 ($sql, @bind) = $handler->($self, $k, $op, $val);
652 puke "Illegal handler for special operator $orig_op - expecting a method name or a coderef";
656 $self->_SWITCH_refkind($val, {
658 ARRAYREF => sub { # CASE: col => {op => \@vals}
659 ($sql, @bind) = $self->_where_field_op_ARRAYREF($k, $op, $val);
662 SCALARREF => sub { # CASE: col => {op => \$scalar} (literal SQL without bind)
663 $sql = join ' ', $self->_convert($self->_quote($k)),
664 $self->_sqlcase($op),
668 ARRAYREFREF => sub { # CASE: col => {op => \[$sql, @bind]} (literal SQL with bind)
669 my ($sub_sql, @sub_bind) = @$$val;
670 $self->_assert_bindval_matches_bindtype(@sub_bind);
671 $sql = join ' ', $self->_convert($self->_quote($k)),
672 $self->_sqlcase($op),
678 ($sql, @bind) = $self->_where_hashpair_HASHREF($k, $val, $op);
681 UNDEF => sub { # CASE: col => {op => undef} : sql "IS (NOT)? NULL"
682 my $is = ($op =~ $self->{equality_op}) ? 'is' :
683 ($op =~ $self->{inequality_op}) ? 'is not' :
684 puke "unexpected operator '$orig_op' with undef operand";
685 $sql = $self->_quote($k) . $self->_sqlcase(" $is null");
688 FALLBACK => sub { # CASE: col => {op => $scalar}
689 $sql = join ' ', $self->_convert($self->_quote($k)),
690 $self->_sqlcase($op),
691 $self->_convert('?');
692 @bind = $self->_bindtype($k, $val);
697 ($all_sql) = (defined $all_sql and $all_sql) ? $self->_join_sql_clauses($logic, [$all_sql, $sql], []) : $sql;
698 push @all_bind, @bind;
700 return ($all_sql, @all_bind);
705 sub _where_field_op_ARRAYREF {
706 my ($self, $k, $op, $vals) = @_;
708 my @vals = @$vals; #always work on a copy
711 $self->_debug(sprintf '%s means multiple elements: [ %s ]',
713 join (', ', map { defined $_ ? "'$_'" : 'NULL' } @vals ),
716 # see if the first element is an -and/-or op
718 if (defined $vals[0] && $vals[0] =~ /^ - ( AND|OR ) $/ix) {
723 # distribute $op over each remaining member of @vals, append logic if exists
724 return $self->_recurse_where([map { {$k => {$op, $_}} } @vals], $logic);
726 # LDNOTE : had planned to change the distribution logic when
727 # $op =~ $self->{inequality_op}, because of Morgan laws :
728 # with {field => {'!=' => [22, 33]}}, it would be ridiculous to generate
729 # WHERE field != 22 OR field != 33 : the user probably means
730 # WHERE field != 22 AND field != 33.
731 # To do this, replace the above to roughly :
732 # my $logic = ($op =~ $self->{inequality_op}) ? 'AND' : 'OR';
733 # return $self->_recurse_where([map { {$k => {$op, $_}} } @vals], $logic);
737 # try to DWIM on equality operators
738 # LDNOTE : not 100% sure this is the correct thing to do ...
739 return ($self->{sqlfalse}) if $op =~ $self->{equality_op};
740 return ($self->{sqltrue}) if $op =~ $self->{inequality_op};
743 puke "operator '$op' applied on an empty array (field '$k')";
748 sub _where_hashpair_SCALARREF {
749 my ($self, $k, $v) = @_;
750 $self->_debug("SCALAR($k) means literal SQL: $$v");
751 my $sql = $self->_quote($k) . " " . $$v;
755 # literal SQL with bind
756 sub _where_hashpair_ARRAYREFREF {
757 my ($self, $k, $v) = @_;
758 $self->_debug("REF($k) means literal SQL: @${$v}");
759 my ($sql, @bind) = @${$v};
760 $self->_assert_bindval_matches_bindtype(@bind);
761 $sql = $self->_quote($k) . " " . $sql;
762 return ($sql, @bind );
765 # literal SQL without bind
766 sub _where_hashpair_SCALAR {
767 my ($self, $k, $v) = @_;
768 $self->_debug("NOREF($k) means simple key=val: $k $self->{cmp} $v");
769 my $sql = join ' ', $self->_convert($self->_quote($k)),
770 $self->_sqlcase($self->{cmp}),
771 $self->_convert('?');
772 my @bind = $self->_bindtype($k, $v);
773 return ( $sql, @bind);
777 sub _where_hashpair_UNDEF {
778 my ($self, $k, $v) = @_;
779 $self->_debug("UNDEF($k) means IS NULL");
780 my $sql = $self->_quote($k) . $self->_sqlcase(' is null');
784 #======================================================================
785 # WHERE: TOP-LEVEL OTHERS (SCALARREF, SCALAR, UNDEF)
786 #======================================================================
789 sub _where_SCALARREF {
790 my ($self, $where) = @_;
793 $self->_debug("SCALAR(*top) means literal SQL: $$where");
799 my ($self, $where) = @_;
802 $self->_debug("NOREF(*top) means literal SQL: $where");
813 #======================================================================
814 # WHERE: BUILTIN SPECIAL OPERATORS (-in, -between)
815 #======================================================================
818 sub _where_field_BETWEEN {
819 my ($self, $k, $op, $vals) = @_;
821 my ($label, $and, $placeholder);
822 $label = $self->_convert($self->_quote($k));
823 $and = ' ' . $self->_sqlcase('and') . ' ';
824 $placeholder = $self->_convert('?');
825 $op = $self->_sqlcase($op);
827 my ($clause, @bind) = $self->_SWITCH_refkind($vals, {
835 puke "special op 'between' accepts an arrayref with exactly two values"
838 my (@all_sql, @all_bind);
839 foreach my $val (@$vals) {
840 my ($sql, @bind) = $self->_SWITCH_refkind($val, {
842 return ($placeholder, ($val));
845 return ($self->_convert($$val), ());
848 my ($sql, @bind) = @$$val;
849 return ($self->_convert($sql), @bind);
853 push @all_bind, @bind;
857 (join $and, @all_sql),
858 $self->_bindtype($k, @all_bind),
862 puke "special op 'between' accepts an arrayref with two values, or a single literal scalarref/arrayref-ref";
866 my $sql = "( $label $op $clause )";
871 sub _where_field_IN {
872 my ($self, $k, $op, $vals) = @_;
874 # backwards compatibility : if scalar, force into an arrayref
875 $vals = [$vals] if defined $vals && ! ref $vals;
877 my ($label) = $self->_convert($self->_quote($k));
878 my ($placeholder) = $self->_convert('?');
879 $op = $self->_sqlcase($op);
881 my ($sql, @bind) = $self->_SWITCH_refkind($vals, {
882 ARRAYREF => sub { # list of choices
883 if (@$vals) { # nonempty list
884 my $placeholders = join ", ", (($placeholder) x @$vals);
885 my $sql = "$label $op ( $placeholders )";
886 my @bind = $self->_bindtype($k, @$vals);
888 return ($sql, @bind);
890 else { # empty list : some databases won't understand "IN ()", so DWIM
891 my $sql = ($op =~ /\bnot\b/i) ? $self->{sqltrue} : $self->{sqlfalse};
896 SCALARREF => sub { # literal SQL
897 my $sql = $self->_open_outer_paren ($$vals);
898 return ("$label $op ( $sql )");
900 ARRAYREFREF => sub { # literal SQL with bind
901 my ($sql, @bind) = @$$vals;
902 $self->_assert_bindval_matches_bindtype(@bind);
903 $sql = $self->_open_outer_paren ($sql);
904 return ("$label $op ( $sql )", @bind);
908 puke "special op 'in' requires an arrayref (or scalarref/arrayref-ref)";
912 return ($sql, @bind);
915 # Some databases (SQLite) treat col IN (1, 2) different from
916 # col IN ( (1, 2) ). Use this to strip all outer parens while
917 # adding them back in the corresponding method
918 sub _open_outer_paren {
919 my ($self, $sql) = @_;
920 $sql = $1 while $sql =~ /^ \s* \( (.*) \) \s* $/x;
925 #======================================================================
927 #======================================================================
930 my ($self, $arg) = @_;
933 for my $c ($self->_order_by_chunks ($arg) ) {
934 $self->_SWITCH_refkind ($c, {
935 SCALAR => sub { push @sql, $c },
936 ARRAYREF => sub { push @sql, shift @$c; push @bind, @$c },
942 $self->_sqlcase(' order by'),
948 return wantarray ? ($sql, @bind) : $sql;
951 sub _order_by_chunks {
952 my ($self, $arg) = @_;
954 return $self->_SWITCH_refkind($arg, {
957 map { $self->_order_by_chunks ($_ ) } @$arg;
960 ARRAYREFREF => sub { [ @$$arg ] },
962 SCALAR => sub {$self->_quote($arg)},
964 UNDEF => sub {return () },
966 SCALARREF => sub {$$arg}, # literal SQL, no quoting
969 # get first pair in hash
970 my ($key, $val) = each %$arg;
972 return () unless $key;
974 if ( (keys %$arg) > 1 or not $key =~ /^-(desc|asc)/i ) {
975 puke "hash passed to _order_by must have exactly one key (-desc or -asc)";
981 for my $c ($self->_order_by_chunks ($val)) {
984 $self->_SWITCH_refkind ($c, {
993 $sql = $sql . ' ' . $self->_sqlcase($direction);
995 push @ret, [ $sql, @bind];
1004 #======================================================================
1005 # DATASOURCE (FOR NOW, JUST PLAIN TABLE OR LIST OF TABLES)
1006 #======================================================================
1011 $self->_SWITCH_refkind($from, {
1012 ARRAYREF => sub {join ', ', map { $self->_quote($_) } @$from;},
1013 SCALAR => sub {$self->_quote($from)},
1014 SCALARREF => sub {$$from},
1015 ARRAYREFREF => sub {join ', ', @$from;},
1020 #======================================================================
1022 #======================================================================
1028 $label or puke "can't quote an empty label";
1030 # left and right quote characters
1031 my ($ql, $qr, @other) = $self->_SWITCH_refkind($self->{quote_char}, {
1032 SCALAR => sub {($self->{quote_char}, $self->{quote_char})},
1033 ARRAYREF => sub {@{$self->{quote_char}}},
1037 or puke "quote_char must be an arrayref of 2 values";
1039 # no quoting if no quoting chars
1040 $ql or return $label;
1042 # no quoting for literal SQL
1043 return $$label if ref($label) eq 'SCALAR';
1045 # separate table / column (if applicable)
1046 my $sep = $self->{name_sep} || '';
1047 my @to_quote = $sep ? split /\Q$sep\E/, $label : ($label);
1049 # do the quoting, except for "*" or for `table`.*
1050 my @quoted = map { $_ eq '*' ? $_: $ql.$_.$qr} @to_quote;
1052 # reassemble and return.
1053 return join $sep, @quoted;
1057 # Conversion, if applicable
1059 my ($self, $arg) = @_;
1061 # LDNOTE : modified the previous implementation below because
1062 # it was not consistent : the first "return" is always an array,
1063 # the second "return" is context-dependent. Anyway, _convert
1064 # seems always used with just a single argument, so make it a
1066 # return @_ unless $self->{convert};
1067 # my $conv = $self->_sqlcase($self->{convert});
1068 # my @ret = map { $conv.'('.$_.')' } @_;
1069 # return wantarray ? @ret : $ret[0];
1070 if ($self->{convert}) {
1071 my $conv = $self->_sqlcase($self->{convert});
1072 $arg = $conv.'('.$arg.')';
1080 my($col, @vals) = @_;
1082 #LDNOTE : changed original implementation below because it did not make
1083 # sense when bindtype eq 'columns' and @vals > 1.
1084 # return $self->{bindtype} eq 'columns' ? [ $col, @vals ] : @vals;
1086 return $self->{bindtype} eq 'columns' ? map {[$col, $_]} @vals : @vals;
1089 # Dies if any element of @bind is not in [colname => value] format
1090 # if bindtype is 'columns'.
1091 sub _assert_bindval_matches_bindtype {
1092 my ($self, @bind) = @_;
1094 if ($self->{bindtype} eq 'columns') {
1095 foreach my $val (@bind) {
1096 if (!defined $val || ref($val) ne 'ARRAY' || @$val != 2) {
1097 die "bindtype 'columns' selected, you need to pass: [column_name => bind_value]"
1103 sub _join_sql_clauses {
1104 my ($self, $logic, $clauses_aref, $bind_aref) = @_;
1106 if (@$clauses_aref > 1) {
1107 my $join = " " . $self->_sqlcase($logic) . " ";
1108 my $sql = '( ' . join($join, @$clauses_aref) . ' )';
1109 return ($sql, @$bind_aref);
1111 elsif (@$clauses_aref) {
1112 return ($clauses_aref->[0], @$bind_aref); # no parentheses
1115 return (); # if no SQL, ignore @$bind_aref
1120 # Fix SQL case, if so requested
1124 # LDNOTE: if $self->{case} is true, then it contains 'lower', so we
1125 # don't touch the argument ... crooked logic, but let's not change it!
1126 return $self->{case} ? $_[0] : uc($_[0]);
1130 #======================================================================
1131 # DISPATCHING FROM REFKIND
1132 #======================================================================
1135 my ($self, $data) = @_;
1141 # blessed objects are treated like scalars
1142 $ref = (blessed $data) ? '' : ref $data;
1143 $n_steps += 1 if $ref;
1144 last if $ref ne 'REF';
1148 my $base = $ref || (defined $data ? 'SCALAR' : 'UNDEF');
1150 return $base . ('REF' x $n_steps);
1156 my ($self, $data) = @_;
1157 my @try = ($self->_refkind($data));
1158 push @try, 'SCALAR_or_UNDEF' if $try[0] eq 'SCALAR' || $try[0] eq 'UNDEF';
1159 push @try, 'FALLBACK';
1163 sub _METHOD_FOR_refkind {
1164 my ($self, $meth_prefix, $data) = @_;
1165 my $method = first {$_} map {$self->can($meth_prefix."_".$_)}
1166 $self->_try_refkind($data)
1167 or puke "cannot dispatch on '$meth_prefix' for ".$self->_refkind($data);
1172 sub _SWITCH_refkind {
1173 my ($self, $data, $dispatch_table) = @_;
1175 my $coderef = first {$_} map {$dispatch_table->{$_}}
1176 $self->_try_refkind($data)
1177 or puke "no dispatch entry for ".$self->_refkind($data);
1184 #======================================================================
1185 # VALUES, GENERATE, AUTOLOAD
1186 #======================================================================
1188 # LDNOTE: original code from nwiger, didn't touch code in that section
1189 # I feel the AUTOLOAD stuff should not be the default, it should
1190 # only be activated on explicit demand by user.
1194 my $data = shift || return;
1195 puke "Argument to ", __PACKAGE__, "->values must be a \\%hash"
1196 unless ref $data eq 'HASH';
1199 foreach my $k ( sort keys %$data ) {
1200 my $v = $data->{$k};
1201 $self->_SWITCH_refkind($v, {
1203 if ($self->{array_datatypes}) { # array datatype
1204 push @all_bind, $self->_bindtype($k, $v);
1206 else { # literal SQL with bind
1207 my ($sql, @bind) = @$v;
1208 $self->_assert_bindval_matches_bindtype(@bind);
1209 push @all_bind, @bind;
1212 ARRAYREFREF => sub { # literal SQL with bind
1213 my ($sql, @bind) = @${$v};
1214 $self->_assert_bindval_matches_bindtype(@bind);
1215 push @all_bind, @bind;
1217 SCALARREF => sub { # literal SQL without bind
1219 SCALAR_or_UNDEF => sub {
1220 push @all_bind, $self->_bindtype($k, $v);
1231 my(@sql, @sqlq, @sqlv);
1235 if ($ref eq 'HASH') {
1236 for my $k (sort keys %$_) {
1239 my $label = $self->_quote($k);
1240 if ($r eq 'ARRAY') {
1241 # literal SQL with bind
1242 my ($sql, @bind) = @$v;
1243 $self->_assert_bindval_matches_bindtype(@bind);
1244 push @sqlq, "$label = $sql";
1246 } elsif ($r eq 'SCALAR') {
1247 # literal SQL without bind
1248 push @sqlq, "$label = $$v";
1250 push @sqlq, "$label = ?";
1251 push @sqlv, $self->_bindtype($k, $v);
1254 push @sql, $self->_sqlcase('set'), join ', ', @sqlq;
1255 } elsif ($ref eq 'ARRAY') {
1256 # unlike insert(), assume these are ONLY the column names, i.e. for SQL
1259 if ($r eq 'ARRAY') { # literal SQL with bind
1260 my ($sql, @bind) = @$v;
1261 $self->_assert_bindval_matches_bindtype(@bind);
1264 } elsif ($r eq 'SCALAR') { # literal SQL without bind
1265 # embedded literal SQL
1272 push @sql, '(' . join(', ', @sqlq) . ')';
1273 } elsif ($ref eq 'SCALAR') {
1277 # strings get case twiddled
1278 push @sql, $self->_sqlcase($_);
1282 my $sql = join ' ', @sql;
1284 # this is pretty tricky
1285 # if ask for an array, return ($stmt, @bind)
1286 # otherwise, s/?/shift @sqlv/ to put it inline
1288 return ($sql, @sqlv);
1290 1 while $sql =~ s/\?/my $d = shift(@sqlv);
1291 ref $d ? $d->[1] : $d/e;
1300 # This allows us to check for a local, then _form, attr
1302 my($name) = $AUTOLOAD =~ /.*::(.+)/;
1303 return $self->generate($name, @_);
1314 SQL::Abstract - Generate SQL from Perl data structures
1320 my $sql = SQL::Abstract->new;
1322 my($stmt, @bind) = $sql->select($table, \@fields, \%where, \@order);
1324 my($stmt, @bind) = $sql->insert($table, \%fieldvals || \@values);
1326 my($stmt, @bind) = $sql->update($table, \%fieldvals, \%where);
1328 my($stmt, @bind) = $sql->delete($table, \%where);
1330 # Then, use these in your DBI statements
1331 my $sth = $dbh->prepare($stmt);
1332 $sth->execute(@bind);
1334 # Just generate the WHERE clause
1335 my($stmt, @bind) = $sql->where(\%where, \@order);
1337 # Return values in the same order, for hashed queries
1338 # See PERFORMANCE section for more details
1339 my @bind = $sql->values(\%fieldvals);
1343 This module was inspired by the excellent L<DBIx::Abstract>.
1344 However, in using that module I found that what I really wanted
1345 to do was generate SQL, but still retain complete control over my
1346 statement handles and use the DBI interface. So, I set out to
1347 create an abstract SQL generation module.
1349 While based on the concepts used by L<DBIx::Abstract>, there are
1350 several important differences, especially when it comes to WHERE
1351 clauses. I have modified the concepts used to make the SQL easier
1352 to generate from Perl data structures and, IMO, more intuitive.
1353 The underlying idea is for this module to do what you mean, based
1354 on the data structures you provide it. The big advantage is that
1355 you don't have to modify your code every time your data changes,
1356 as this module figures it out.
1358 To begin with, an SQL INSERT is as easy as just specifying a hash
1359 of C<key=value> pairs:
1362 name => 'Jimbo Bobson',
1363 phone => '123-456-7890',
1364 address => '42 Sister Lane',
1365 city => 'St. Louis',
1366 state => 'Louisiana',
1369 The SQL can then be generated with this:
1371 my($stmt, @bind) = $sql->insert('people', \%data);
1373 Which would give you something like this:
1375 $stmt = "INSERT INTO people
1376 (address, city, name, phone, state)
1377 VALUES (?, ?, ?, ?, ?)";
1378 @bind = ('42 Sister Lane', 'St. Louis', 'Jimbo Bobson',
1379 '123-456-7890', 'Louisiana');
1381 These are then used directly in your DBI code:
1383 my $sth = $dbh->prepare($stmt);
1384 $sth->execute(@bind);
1386 =head2 Inserting and Updating Arrays
1388 If your database has array types (like for example Postgres),
1389 activate the special option C<< array_datatypes => 1 >>
1390 when creating the C<SQL::Abstract> object.
1391 Then you may use an arrayref to insert and update database array types:
1393 my $sql = SQL::Abstract->new(array_datatypes => 1);
1395 planets => [qw/Mercury Venus Earth Mars/]
1398 my($stmt, @bind) = $sql->insert('solar_system', \%data);
1402 $stmt = "INSERT INTO solar_system (planets) VALUES (?)"
1404 @bind = (['Mercury', 'Venus', 'Earth', 'Mars']);
1407 =head2 Inserting and Updating SQL
1409 In order to apply SQL functions to elements of your C<%data> you may
1410 specify a reference to an arrayref for the given hash value. For example,
1411 if you need to execute the Oracle C<to_date> function on a value, you can
1412 say something like this:
1416 date_entered => \["to_date(?,'MM/DD/YYYY')", "03/02/2003"],
1419 The first value in the array is the actual SQL. Any other values are
1420 optional and would be included in the bind values array. This gives
1423 my($stmt, @bind) = $sql->insert('people', \%data);
1425 $stmt = "INSERT INTO people (name, date_entered)
1426 VALUES (?, to_date(?,'MM/DD/YYYY'))";
1427 @bind = ('Bill', '03/02/2003');
1429 An UPDATE is just as easy, all you change is the name of the function:
1431 my($stmt, @bind) = $sql->update('people', \%data);
1433 Notice that your C<%data> isn't touched; the module will generate
1434 the appropriately quirky SQL for you automatically. Usually you'll
1435 want to specify a WHERE clause for your UPDATE, though, which is
1436 where handling C<%where> hashes comes in handy...
1438 =head2 Complex where statements
1440 This module can generate pretty complicated WHERE statements
1441 easily. For example, simple C<key=value> pairs are taken to mean
1442 equality, and if you want to see if a field is within a set
1443 of values, you can use an arrayref. Let's say we wanted to
1444 SELECT some data based on this criteria:
1447 requestor => 'inna',
1448 worker => ['nwiger', 'rcwe', 'sfz'],
1449 status => { '!=', 'completed' }
1452 my($stmt, @bind) = $sql->select('tickets', '*', \%where);
1454 The above would give you something like this:
1456 $stmt = "SELECT * FROM tickets WHERE
1457 ( requestor = ? ) AND ( status != ? )
1458 AND ( worker = ? OR worker = ? OR worker = ? )";
1459 @bind = ('inna', 'completed', 'nwiger', 'rcwe', 'sfz');
1461 Which you could then use in DBI code like so:
1463 my $sth = $dbh->prepare($stmt);
1464 $sth->execute(@bind);
1470 The functions are simple. There's one for each major SQL operation,
1471 and a constructor you use first. The arguments are specified in a
1472 similar order to each function (table, then fields, then a where
1473 clause) to try and simplify things.
1478 =head2 new(option => 'value')
1480 The C<new()> function takes a list of options and values, and returns
1481 a new B<SQL::Abstract> object which can then be used to generate SQL
1482 through the methods below. The options accepted are:
1488 If set to 'lower', then SQL will be generated in all lowercase. By
1489 default SQL is generated in "textbook" case meaning something like:
1491 SELECT a_field FROM a_table WHERE some_field LIKE '%someval%'
1493 Any setting other than 'lower' is ignored.
1497 This determines what the default comparison operator is. By default
1498 it is C<=>, meaning that a hash like this:
1500 %where = (name => 'nwiger', email => 'nate@wiger.org');
1502 Will generate SQL like this:
1504 WHERE name = 'nwiger' AND email = 'nate@wiger.org'
1506 However, you may want loose comparisons by default, so if you set
1507 C<cmp> to C<like> you would get SQL such as:
1509 WHERE name like 'nwiger' AND email like 'nate@wiger.org'
1511 You can also override the comparsion on an individual basis - see
1512 the huge section on L</"WHERE CLAUSES"> at the bottom.
1514 =item sqltrue, sqlfalse
1516 Expressions for inserting boolean values within SQL statements.
1517 By default these are C<1=1> and C<1=0>. They are used
1518 by the special operators C<-in> and C<-not_in> for generating
1519 correct SQL even when the argument is an empty array (see below).
1523 This determines the default logical operator for multiple WHERE
1524 statements in arrays or hashes. If absent, the default logic is "or"
1525 for arrays, and "and" for hashes. This means that a WHERE
1529 event_date => {'>=', '2/13/99'},
1530 event_date => {'<=', '4/24/03'},
1533 will generate SQL like this:
1535 WHERE event_date >= '2/13/99' OR event_date <= '4/24/03'
1537 This is probably not what you want given this query, though (look
1538 at the dates). To change the "OR" to an "AND", simply specify:
1540 my $sql = SQL::Abstract->new(logic => 'and');
1542 Which will change the above C<WHERE> to:
1544 WHERE event_date >= '2/13/99' AND event_date <= '4/24/03'
1546 The logic can also be changed locally by inserting
1547 a modifier in front of an arrayref :
1549 @where = (-and => [event_date => {'>=', '2/13/99'},
1550 event_date => {'<=', '4/24/03'} ]);
1552 See the L</"WHERE CLAUSES"> section for explanations.
1556 This will automatically convert comparisons using the specified SQL
1557 function for both column and value. This is mostly used with an argument
1558 of C<upper> or C<lower>, so that the SQL will have the effect of
1559 case-insensitive "searches". For example, this:
1561 $sql = SQL::Abstract->new(convert => 'upper');
1562 %where = (keywords => 'MaKe iT CAse inSeNSItive');
1564 Will turn out the following SQL:
1566 WHERE upper(keywords) like upper('MaKe iT CAse inSeNSItive')
1568 The conversion can be C<upper()>, C<lower()>, or any other SQL function
1569 that can be applied symmetrically to fields (actually B<SQL::Abstract> does
1570 not validate this option; it will just pass through what you specify verbatim).
1574 This is a kludge because many databases suck. For example, you can't
1575 just bind values using DBI's C<execute()> for Oracle C<CLOB> or C<BLOB> fields.
1576 Instead, you have to use C<bind_param()>:
1578 $sth->bind_param(1, 'reg data');
1579 $sth->bind_param(2, $lots, {ora_type => ORA_CLOB});
1581 The problem is, B<SQL::Abstract> will normally just return a C<@bind> array,
1582 which loses track of which field each slot refers to. Fear not.
1584 If you specify C<bindtype> in new, you can determine how C<@bind> is returned.
1585 Currently, you can specify either C<normal> (default) or C<columns>. If you
1586 specify C<columns>, you will get an array that looks like this:
1588 my $sql = SQL::Abstract->new(bindtype => 'columns');
1589 my($stmt, @bind) = $sql->insert(...);
1592 [ 'column1', 'value1' ],
1593 [ 'column2', 'value2' ],
1594 [ 'column3', 'value3' ],
1597 You can then iterate through this manually, using DBI's C<bind_param()>.
1599 $sth->prepare($stmt);
1602 my($col, $data) = @$_;
1603 if ($col eq 'details' || $col eq 'comments') {
1604 $sth->bind_param($i, $data, {ora_type => ORA_CLOB});
1605 } elsif ($col eq 'image') {
1606 $sth->bind_param($i, $data, {ora_type => ORA_BLOB});
1608 $sth->bind_param($i, $data);
1612 $sth->execute; # execute without @bind now
1614 Now, why would you still use B<SQL::Abstract> if you have to do this crap?
1615 Basically, the advantage is still that you don't have to care which fields
1616 are or are not included. You could wrap that above C<for> loop in a simple
1617 sub called C<bind_fields()> or something and reuse it repeatedly. You still
1618 get a layer of abstraction over manual SQL specification.
1620 Note that if you set L</bindtype> to C<columns>, the C<\[$sql, @bind]>
1621 construct (see L</Literal SQL with placeholders and bind values (subqueries)>)
1622 will expect the bind values in this format.
1626 This is the character that a table or column name will be quoted
1627 with. By default this is an empty string, but you could set it to
1628 the character C<`>, to generate SQL like this:
1630 SELECT `a_field` FROM `a_table` WHERE `some_field` LIKE '%someval%'
1632 Alternatively, you can supply an array ref of two items, the first being the left
1633 hand quote character, and the second the right hand quote character. For
1634 example, you could supply C<['[',']']> for SQL Server 2000 compliant quotes
1635 that generates SQL like this:
1637 SELECT [a_field] FROM [a_table] WHERE [some_field] LIKE '%someval%'
1639 Quoting is useful if you have tables or columns names that are reserved
1640 words in your database's SQL dialect.
1644 This is the character that separates a table and column name. It is
1645 necessary to specify this when the C<quote_char> option is selected,
1646 so that tables and column names can be individually quoted like this:
1648 SELECT `table`.`one_field` FROM `table` WHERE `table`.`other_field` = 1
1650 =item array_datatypes
1652 When this option is true, arrayrefs in INSERT or UPDATE are
1653 interpreted as array datatypes and are passed directly
1655 When this option is false, arrayrefs are interpreted
1656 as literal SQL, just like refs to arrayrefs
1657 (but this behavior is for backwards compatibility; when writing
1658 new queries, use the "reference to arrayref" syntax
1664 Takes a reference to a list of "special operators"
1665 to extend the syntax understood by L<SQL::Abstract>.
1666 See section L</"SPECIAL OPERATORS"> for details.
1670 Takes a reference to a list of "unary operators"
1671 to extend the syntax understood by L<SQL::Abstract>.
1672 See section L</"UNARY OPERATORS"> for details.
1678 =head2 insert($table, \@values || \%fieldvals)
1680 This is the simplest function. You simply give it a table name
1681 and either an arrayref of values or hashref of field/value pairs.
1682 It returns an SQL INSERT statement and a list of bind values.
1683 See the sections on L</"Inserting and Updating Arrays"> and
1684 L</"Inserting and Updating SQL"> for information on how to insert
1685 with those data types.
1687 =head2 update($table, \%fieldvals, \%where)
1689 This takes a table, hashref of field/value pairs, and an optional
1690 hashref L<WHERE clause|/WHERE CLAUSES>. It returns an SQL UPDATE function and a list
1692 See the sections on L</"Inserting and Updating Arrays"> and
1693 L</"Inserting and Updating SQL"> for information on how to insert
1694 with those data types.
1696 =head2 select($source, $fields, $where, $order)
1698 This returns a SQL SELECT statement and associated list of bind values, as
1699 specified by the arguments :
1705 Specification of the 'FROM' part of the statement.
1706 The argument can be either a plain scalar (interpreted as a table
1707 name, will be quoted), or an arrayref (interpreted as a list
1708 of table names, joined by commas, quoted), or a scalarref
1709 (literal table name, not quoted), or a ref to an arrayref
1710 (list of literal table names, joined by commas, not quoted).
1714 Specification of the list of fields to retrieve from
1716 The argument can be either an arrayref (interpreted as a list
1717 of field names, will be joined by commas and quoted), or a
1718 plain scalar (literal SQL, not quoted).
1719 Please observe that this API is not as flexible as for
1720 the first argument C<$table>, for backwards compatibility reasons.
1724 Optional argument to specify the WHERE part of the query.
1725 The argument is most often a hashref, but can also be
1726 an arrayref or plain scalar --
1727 see section L<WHERE clause|/"WHERE CLAUSES"> for details.
1731 Optional argument to specify the ORDER BY part of the query.
1732 The argument can be a scalar, a hashref or an arrayref
1733 -- see section L<ORDER BY clause|/"ORDER BY CLAUSES">
1739 =head2 delete($table, \%where)
1741 This takes a table name and optional hashref L<WHERE clause|/WHERE CLAUSES>.
1742 It returns an SQL DELETE statement and list of bind values.
1744 =head2 where(\%where, \@order)
1746 This is used to generate just the WHERE clause. For example,
1747 if you have an arbitrary data structure and know what the
1748 rest of your SQL is going to look like, but want an easy way
1749 to produce a WHERE clause, use this. It returns an SQL WHERE
1750 clause and list of bind values.
1753 =head2 values(\%data)
1755 This just returns the values from the hash C<%data>, in the same
1756 order that would be returned from any of the other above queries.
1757 Using this allows you to markedly speed up your queries if you
1758 are affecting lots of rows. See below under the L</"PERFORMANCE"> section.
1760 =head2 generate($any, 'number', $of, \@data, $struct, \%types)
1762 Warning: This is an experimental method and subject to change.
1764 This returns arbitrarily generated SQL. It's a really basic shortcut.
1765 It will return two different things, depending on return context:
1767 my($stmt, @bind) = $sql->generate('create table', \$table, \@fields);
1768 my $stmt_and_val = $sql->generate('create table', \$table, \@fields);
1770 These would return the following:
1772 # First calling form
1773 $stmt = "CREATE TABLE test (?, ?)";
1774 @bind = (field1, field2);
1776 # Second calling form
1777 $stmt_and_val = "CREATE TABLE test (field1, field2)";
1779 Depending on what you're trying to do, it's up to you to choose the correct
1780 format. In this example, the second form is what you would want.
1784 $sql->generate('alter session', { nls_date_format => 'MM/YY' });
1788 ALTER SESSION SET nls_date_format = 'MM/YY'
1790 You get the idea. Strings get their case twiddled, but everything
1791 else remains verbatim.
1796 =head1 WHERE CLAUSES
1800 This module uses a variation on the idea from L<DBIx::Abstract>. It
1801 is B<NOT>, repeat I<not> 100% compatible. B<The main logic of this
1802 module is that things in arrays are OR'ed, and things in hashes
1805 The easiest way to explain is to show lots of examples. After
1806 each C<%where> hash shown, it is assumed you used:
1808 my($stmt, @bind) = $sql->where(\%where);
1810 However, note that the C<%where> hash can be used directly in any
1811 of the other functions as well, as described above.
1813 =head2 Key-value pairs
1815 So, let's get started. To begin, a simple hash:
1819 status => 'completed'
1822 Is converted to SQL C<key = val> statements:
1824 $stmt = "WHERE user = ? AND status = ?";
1825 @bind = ('nwiger', 'completed');
1827 One common thing I end up doing is having a list of values that
1828 a field can be in. To do this, simply specify a list inside of
1833 status => ['assigned', 'in-progress', 'pending'];
1836 This simple code will create the following:
1838 $stmt = "WHERE user = ? AND ( status = ? OR status = ? OR status = ? )";
1839 @bind = ('nwiger', 'assigned', 'in-progress', 'pending');
1841 A field associated to an empty arrayref will be considered a
1842 logical false and will generate 0=1.
1844 =head2 Specific comparison operators
1846 If you want to specify a different type of operator for your comparison,
1847 you can use a hashref for a given column:
1851 status => { '!=', 'completed' }
1854 Which would generate:
1856 $stmt = "WHERE user = ? AND status != ?";
1857 @bind = ('nwiger', 'completed');
1859 To test against multiple values, just enclose the values in an arrayref:
1861 status => { '=', ['assigned', 'in-progress', 'pending'] };
1863 Which would give you:
1865 "WHERE status = ? OR status = ? OR status = ?"
1868 The hashref can also contain multiple pairs, in which case it is expanded
1869 into an C<AND> of its elements:
1873 status => { '!=', 'completed', -not_like => 'pending%' }
1876 # Or more dynamically, like from a form
1877 $where{user} = 'nwiger';
1878 $where{status}{'!='} = 'completed';
1879 $where{status}{'-not_like'} = 'pending%';
1881 # Both generate this
1882 $stmt = "WHERE user = ? AND status != ? AND status NOT LIKE ?";
1883 @bind = ('nwiger', 'completed', 'pending%');
1886 To get an OR instead, you can combine it with the arrayref idea:
1890 priority => [ {'=', 2}, {'!=', 1} ]
1893 Which would generate:
1895 $stmt = "WHERE user = ? AND priority = ? OR priority != ?";
1896 @bind = ('nwiger', '2', '1');
1898 If you want to include literal SQL (with or without bind values), just use a
1899 scalar reference or array reference as the value:
1902 date_entered => { '>' => \["to_date(?, 'MM/DD/YYYY')", "11/26/2008"] },
1903 date_expires => { '<' => \"now()" }
1906 Which would generate:
1908 $stmt = "WHERE date_entered > "to_date(?, 'MM/DD/YYYY') AND date_expires < now()";
1909 @bind = ('11/26/2008');
1912 =head2 Logic and nesting operators
1914 In the example above,
1915 there is a subtle trap if you want to say something like
1916 this (notice the C<AND>):
1918 WHERE priority != ? AND priority != ?
1920 Because, in Perl you I<can't> do this:
1922 priority => { '!=', 2, '!=', 1 }
1924 As the second C<!=> key will obliterate the first. The solution
1925 is to use the special C<-modifier> form inside an arrayref:
1927 priority => [ -and => {'!=', 2},
1931 Normally, these would be joined by C<OR>, but the modifier tells it
1932 to use C<AND> instead. (Hint: You can use this in conjunction with the
1933 C<logic> option to C<new()> in order to change the way your queries
1934 work by default.) B<Important:> Note that the C<-modifier> goes
1935 B<INSIDE> the arrayref, as an extra first element. This will
1936 B<NOT> do what you think it might:
1938 priority => -and => [{'!=', 2}, {'!=', 1}] # WRONG!
1940 Here is a quick list of equivalencies, since there is some overlap:
1943 status => {'!=', 'completed', 'not like', 'pending%' }
1944 status => [ -and => {'!=', 'completed'}, {'not like', 'pending%'}]
1947 status => {'=', ['assigned', 'in-progress']}
1948 status => [ -or => {'=', 'assigned'}, {'=', 'in-progress'}]
1949 status => [ {'=', 'assigned'}, {'=', 'in-progress'} ]
1953 =head2 Special operators : IN, BETWEEN, etc.
1955 You can also use the hashref format to compare a list of fields using the
1956 C<IN> comparison operator, by specifying the list as an arrayref:
1959 status => 'completed',
1960 reportid => { -in => [567, 2335, 2] }
1963 Which would generate:
1965 $stmt = "WHERE status = ? AND reportid IN (?,?,?)";
1966 @bind = ('completed', '567', '2335', '2');
1968 The reverse operator C<-not_in> generates SQL C<NOT IN> and is used in
1971 If the argument to C<-in> is an empty array, 'sqlfalse' is generated
1972 (by default : C<1=0>). Similarly, C<< -not_in => [] >> generates
1973 'sqltrue' (by default : C<1=1>).
1975 In addition to the array you can supply a chunk of literal sql or
1976 literal sql with bind:
1979 customer => { -in => \[
1980 'SELECT cust_id FROM cust WHERE balance > ?',
1983 status => { -in => \'SELECT status_codes FROM states' },
1989 customer IN ( SELECT cust_id FROM cust WHERE balance > ? )
1990 AND status IN ( SELECT status_codes FROM states )
1996 Another pair of operators is C<-between> and C<-not_between>,
1997 used with an arrayref of two values:
2001 completion_date => {
2002 -not_between => ['2002-10-01', '2003-02-06']
2008 WHERE user = ? AND completion_date NOT BETWEEN ( ? AND ? )
2010 Just like with C<-in> all plausible combinations of literal SQL
2014 start0 => { -between => [ 1, 2 ] },
2015 start1 => { -between => \["? AND ?", 1, 2] },
2016 start2 => { -between => \"lower(x) AND upper(y)" },
2017 start3 => { -between => [
2019 \["upper(?)", 'stuff' ],
2026 ( start0 BETWEEN ? AND ? )
2027 AND ( start1 BETWEEN ? AND ? )
2028 AND ( start2 BETWEEN lower(x) AND upper(y) )
2029 AND ( start3 BETWEEN lower(x) AND upper(?) )
2031 @bind = (1, 2, 1, 2, 'stuff');
2034 These are the two builtin "special operators"; but the
2035 list can be expanded : see section L</"SPECIAL OPERATORS"> below.
2037 =head2 Unary operators: bool
2039 If you wish to test against boolean columns or functions within your
2040 database you can use the C<-bool> and C<-not_bool> operators. For
2041 example to test the column C<is_user> being true and the column
2042 <is_enabled> being false you would use:-
2046 -not_bool => 'is_enabled',
2051 WHERE is_user AND NOT is_enabled
2053 If a more complex combination is required, testing more conditions,
2054 then you should use the and/or operators:-
2061 -not_bool => 'four',
2067 WHERE one AND two AND three AND NOT four
2070 =head2 Nested conditions, -and/-or prefixes
2072 So far, we've seen how multiple conditions are joined with a top-level
2073 C<AND>. We can change this by putting the different conditions we want in
2074 hashes and then putting those hashes in an array. For example:
2079 status => { -like => ['pending%', 'dispatched'] },
2083 status => 'unassigned',
2087 This data structure would create the following:
2089 $stmt = "WHERE ( user = ? AND ( status LIKE ? OR status LIKE ? ) )
2090 OR ( user = ? AND status = ? ) )";
2091 @bind = ('nwiger', 'pending', 'dispatched', 'robot', 'unassigned');
2094 There is also a special C<-nest>
2095 operator which adds an additional set of parens, to create a subquery.
2096 For example, to get something like this:
2098 $stmt = "WHERE user = ? AND ( workhrs > ? OR geo = ? )";
2099 @bind = ('nwiger', '20', 'ASIA');
2105 -nest => [ workhrs => {'>', 20}, geo => 'ASIA' ],
2109 Finally, clauses in hashrefs or arrayrefs can be
2110 prefixed with an C<-and> or C<-or> to change the logic
2117 -and => [workhrs => {'>', 20}, geo => 'ASIA' ],
2118 -and => [workhrs => {'<', 50}, geo => 'EURO' ]
2125 WHERE ( user = ? AND
2126 ( ( workhrs > ? AND geo = ? )
2127 OR ( workhrs < ? AND geo = ? ) ) )
2130 =head2 Algebraic inconsistency, for historical reasons
2132 C<Important note>: when connecting several conditions, the C<-and->|C<-or>
2133 operator goes C<outside> of the nested structure; whereas when connecting
2134 several constraints on one column, the C<-and> operator goes
2135 C<inside> the arrayref. Here is an example combining both features :
2138 -and => [a => 1, b => 2],
2139 -or => [c => 3, d => 4],
2140 e => [-and => {-like => 'foo%'}, {-like => '%bar'} ]
2145 WHERE ( ( ( a = ? AND b = ? )
2146 OR ( c = ? OR d = ? )
2147 OR ( e LIKE ? AND e LIKE ? ) ) )
2149 This difference in syntax is unfortunate but must be preserved for
2150 historical reasons. So be careful : the two examples below would
2151 seem algebraically equivalent, but they are not
2153 {col => [-and => {-like => 'foo%'}, {-like => '%bar'}]}
2154 # yields : WHERE ( ( col LIKE ? AND col LIKE ? ) )
2156 [-and => {col => {-like => 'foo%'}, {col => {-like => '%bar'}}]]
2157 # yields : WHERE ( ( col LIKE ? OR col LIKE ? ) )
2162 Finally, sometimes only literal SQL will do. If you want to include
2163 literal SQL verbatim, you can specify it as a scalar reference, namely:
2165 my $inn = 'is Not Null';
2167 priority => { '<', 2 },
2173 $stmt = "WHERE priority < ? AND requestor is Not Null";
2176 Note that in this example, you only get one bind parameter back, since
2177 the verbatim SQL is passed as part of the statement.
2179 Of course, just to prove a point, the above can also be accomplished
2183 priority => { '<', 2 },
2184 requestor => { '!=', undef },
2190 Conditions on boolean columns can be expressed in the same way, passing
2191 a reference to an empty string, however using liternal SQL in this way
2192 is deprecated - the preferred method is to use the boolean operators -
2193 see L</"Unary operators: bool"> :
2196 priority => { '<', 2 },
2202 $stmt = "WHERE priority < ? AND is_ready";
2206 =head2 Literal SQL with placeholders and bind values (subqueries)
2208 If the literal SQL to be inserted has placeholders and bind values,
2209 use a reference to an arrayref (yes this is a double reference --
2210 not so common, but perfectly legal Perl). For example, to find a date
2211 in Postgres you can use something like this:
2214 date_column => \[q/= date '2008-09-30' - ?::integer/, 10/]
2219 $stmt = "WHERE ( date_column = date '2008-09-30' - ?::integer )"
2222 Note that you must pass the bind values in the same format as they are returned
2223 by L</where>. That means that if you set L</bindtype> to C<columns>, you must
2224 provide the bind values in the C<< [ column_meta => value ] >> format, where
2225 C<column_meta> is an opaque scalar value; most commonly the column name, but
2226 you can use any scalar value (including references and blessed references),
2227 L<SQL::Abstract> will simply pass it through intact. So if C<bindtype> is set
2228 to C<columns> the above example will look like:
2231 date_column => \[q/= date '2008-09-30' - ?::integer/, [ dummy => 10 ]/]
2234 Literal SQL is especially useful for nesting parenthesized clauses in the
2235 main SQL query. Here is a first example :
2237 my ($sub_stmt, @sub_bind) = ("SELECT c1 FROM t1 WHERE c2 < ? AND c3 LIKE ?",
2241 bar => \["IN ($sub_stmt)" => @sub_bind],
2246 $stmt = "WHERE (foo = ? AND bar IN (SELECT c1 FROM t1
2247 WHERE c2 < ? AND c3 LIKE ?))";
2248 @bind = (1234, 100, "foo%");
2250 Other subquery operators, like for example C<"E<gt> ALL"> or C<"NOT IN">,
2251 are expressed in the same way. Of course the C<$sub_stmt> and
2252 its associated bind values can be generated through a former call
2255 my ($sub_stmt, @sub_bind)
2256 = $sql->select("t1", "c1", {c2 => {"<" => 100},
2257 c3 => {-like => "foo%"}});
2260 bar => \["> ALL ($sub_stmt)" => @sub_bind],
2263 In the examples above, the subquery was used as an operator on a column;
2264 but the same principle also applies for a clause within the main C<%where>
2265 hash, like an EXISTS subquery :
2267 my ($sub_stmt, @sub_bind)
2268 = $sql->select("t1", "*", {c1 => 1, c2 => \"> t0.c0"});
2271 -nest => \["EXISTS ($sub_stmt)" => @sub_bind],
2276 $stmt = "WHERE (foo = ? AND EXISTS (SELECT * FROM t1
2277 WHERE c1 = ? AND c2 > t0.c0))";
2281 Observe that the condition on C<c2> in the subquery refers to
2282 column C<t0.c0> of the main query : this is I<not> a bind
2283 value, so we have to express it through a scalar ref.
2284 Writing C<< c2 => {">" => "t0.c0"} >> would have generated
2285 C<< c2 > ? >> with bind value C<"t0.c0"> ... not exactly
2286 what we wanted here.
2288 Another use of the subquery technique is when some SQL clauses need
2289 parentheses, as it often occurs with some proprietary SQL extensions
2290 like for example fulltext expressions, geospatial expressions,
2291 NATIVE clauses, etc. Here is an example of a fulltext query in MySQL :
2294 -nest => \["MATCH (col1, col2) AGAINST (?)" => qw/apples/]
2297 Finally, here is an example where a subquery is used
2298 for expressing unary negation:
2300 my ($sub_stmt, @sub_bind)
2301 = $sql->where({age => [{"<" => 10}, {">" => 20}]});
2302 $sub_stmt =~ s/^ where //i; # don't want "WHERE" in the subclause
2304 lname => {like => '%son%'},
2305 -nest => \["NOT ($sub_stmt)" => @sub_bind],
2310 $stmt = "lname LIKE ? AND NOT ( age < ? OR age > ? )"
2311 @bind = ('%son%', 10, 20)
2317 These pages could go on for a while, since the nesting of the data
2318 structures this module can handle are pretty much unlimited (the
2319 module implements the C<WHERE> expansion as a recursive function
2320 internally). Your best bet is to "play around" with the module a
2321 little to see how the data structures behave, and choose the best
2322 format for your data based on that.
2324 And of course, all the values above will probably be replaced with
2325 variables gotten from forms or the command line. After all, if you
2326 knew everything ahead of time, you wouldn't have to worry about
2327 dynamically-generating SQL and could just hardwire it into your
2333 =head1 ORDER BY CLAUSES
2335 Some functions take an order by clause. This can either be a scalar (just a
2336 column name,) a hash of C<< { -desc => 'col' } >> or C<< { -asc => 'col' } >>,
2337 or an array of either of the two previous forms. Examples:
2339 Given | Will Generate
2340 ----------------------------------------------------------
2342 \'colA DESC' | ORDER BY colA DESC
2344 'colA' | ORDER BY colA
2346 [qw/colA colB/] | ORDER BY colA, colB
2348 {-asc => 'colA'} | ORDER BY colA ASC
2350 {-desc => 'colB'} | ORDER BY colB DESC
2352 ['colA', {-asc => 'colB'}] | ORDER BY colA, colB ASC
2354 { -asc => [qw/colA colB] } | ORDER BY colA ASC, colB ASC
2357 { -asc => 'colA' }, | ORDER BY colA ASC, colB DESC,
2358 { -desc => [qw/colB/], | colC ASC, colD ASC
2359 { -asc => [qw/colC colD/],|
2361 ===========================================================
2365 =head1 SPECIAL OPERATORS
2367 my $sqlmaker = SQL::Abstract->new(special_ops => [
2371 my ($self, $field, $op, $arg) = @_;
2377 handler => 'method_name',
2381 A "special operator" is a SQL syntactic clause that can be
2382 applied to a field, instead of a usual binary operator.
2385 WHERE field IN (?, ?, ?)
2386 WHERE field BETWEEN ? AND ?
2387 WHERE MATCH(field) AGAINST (?, ?)
2389 Special operators IN and BETWEEN are fairly standard and therefore
2390 are builtin within C<SQL::Abstract> (as the overridable methods
2391 C<_where_field_IN> and C<_where_field_BETWEEN>). For other operators,
2392 like the MATCH .. AGAINST example above which is specific to MySQL,
2393 you can write your own operator handlers - supply a C<special_ops>
2394 argument to the C<new> method. That argument takes an arrayref of
2395 operator definitions; each operator definition is a hashref with two
2402 the regular expression to match the operator
2406 Either a coderef or a plain scalar method name. In both cases
2407 the expected return is C<< ($sql, @bind) >>.
2409 When supplied with a method name, it is simply called on the
2410 L<SQL::Abstract/> object as:
2412 $self->$method_name ($field, $op, $arg)
2416 $op is the part that matched the handler regex
2417 $field is the LHS of the operator
2420 When supplied with a coderef, it is called as:
2422 $coderef->($self, $field, $op, $arg)
2427 For example, here is an implementation
2428 of the MATCH .. AGAINST syntax for MySQL
2430 my $sqlmaker = SQL::Abstract->new(special_ops => [
2432 # special op for MySql MATCH (field) AGAINST(word1, word2, ...)
2433 {regex => qr/^match$/i,
2435 my ($self, $field, $op, $arg) = @_;
2436 $arg = [$arg] if not ref $arg;
2437 my $label = $self->_quote($field);
2438 my ($placeholder) = $self->_convert('?');
2439 my $placeholders = join ", ", (($placeholder) x @$arg);
2440 my $sql = $self->_sqlcase('match') . " ($label) "
2441 . $self->_sqlcase('against') . " ($placeholders) ";
2442 my @bind = $self->_bindtype($field, @$arg);
2443 return ($sql, @bind);
2450 =head1 UNARY OPERATORS
2452 my $sqlmaker = SQL::Abstract->new(unary_ops => [
2456 my ($self, $op, $arg) = @_;
2462 handler => 'method_name',
2466 A "unary operator" is a SQL syntactic clause that can be
2467 applied to a field - the operator goes before the field
2469 You can write your own operator handlers - supply a C<unary_ops>
2470 argument to the C<new> method. That argument takes an arrayref of
2471 operator definitions; each operator definition is a hashref with two
2478 the regular expression to match the operator
2482 Either a coderef or a plain scalar method name. In both cases
2483 the expected return is C<< $sql >>.
2485 When supplied with a method name, it is simply called on the
2486 L<SQL::Abstract/> object as:
2488 $self->$method_name ($op, $arg)
2492 $op is the part that matched the handler regex
2493 $arg is the RHS or argument of the operator
2495 When supplied with a coderef, it is called as:
2497 $coderef->($self, $op, $arg)
2505 Thanks to some benchmarking by Mark Stosberg, it turns out that
2506 this module is many orders of magnitude faster than using C<DBIx::Abstract>.
2507 I must admit this wasn't an intentional design issue, but it's a
2508 byproduct of the fact that you get to control your C<DBI> handles
2511 To maximize performance, use a code snippet like the following:
2513 # prepare a statement handle using the first row
2514 # and then reuse it for the rest of the rows
2516 for my $href (@array_of_hashrefs) {
2517 $stmt ||= $sql->insert('table', $href);
2518 $sth ||= $dbh->prepare($stmt);
2519 $sth->execute($sql->values($href));
2522 The reason this works is because the keys in your C<$href> are sorted
2523 internally by B<SQL::Abstract>. Thus, as long as your data retains
2524 the same structure, you only have to generate the SQL the first time
2525 around. On subsequent queries, simply use the C<values> function provided
2526 by this module to return your values in the correct order.
2531 If you use my C<CGI::FormBuilder> module at all, you'll hopefully
2532 really like this part (I do, at least). Building up a complex query
2533 can be as simple as the following:
2537 use CGI::FormBuilder;
2540 my $form = CGI::FormBuilder->new(...);
2541 my $sql = SQL::Abstract->new;
2543 if ($form->submitted) {
2544 my $field = $form->field;
2545 my $id = delete $field->{id};
2546 my($stmt, @bind) = $sql->update('table', $field, {id => $id});
2549 Of course, you would still have to connect using C<DBI> to run the
2550 query, but the point is that if you make your form look like your
2551 table, the actual query script can be extremely simplistic.
2553 If you're B<REALLY> lazy (I am), check out C<HTML::QuickTable> for
2554 a fast interface to returning and formatting data. I frequently
2555 use these three modules together to write complex database query
2556 apps in under 50 lines.
2561 Version 1.50 was a major internal refactoring of C<SQL::Abstract>.
2562 Great care has been taken to preserve the I<published> behavior
2563 documented in previous versions in the 1.* family; however,
2564 some features that were previously undocumented, or behaved
2565 differently from the documentation, had to be changed in order
2566 to clarify the semantics. Hence, client code that was relying
2567 on some dark areas of C<SQL::Abstract> v1.*
2568 B<might behave differently> in v1.50.
2570 The main changes are :
2576 support for literal SQL through the C<< \ [$sql, bind] >> syntax.
2580 support for the { operator => \"..." } construct (to embed literal SQL)
2584 support for the { operator => \["...", @bind] } construct (to embed literal SQL with bind values)
2588 optional support for L<array datatypes|/"Inserting and Updating Arrays">
2592 defensive programming : check arguments
2596 fixed bug with global logic, which was previously implemented
2597 through global variables yielding side-effects. Prior versions would
2598 interpret C<< [ {cond1, cond2}, [cond3, cond4] ] >>
2599 as C<< "(cond1 AND cond2) OR (cond3 AND cond4)" >>.
2600 Now this is interpreted
2601 as C<< "(cond1 AND cond2) OR (cond3 OR cond4)" >>.
2606 fixed semantics of _bindtype on array args
2610 dropped the C<_anoncopy> of the %where tree. No longer necessary,
2611 we just avoid shifting arrays within that tree.
2615 dropped the C<_modlogic> function
2621 =head1 ACKNOWLEDGEMENTS
2623 There are a number of individuals that have really helped out with
2624 this module. Unfortunately, most of them submitted bugs via CPAN
2625 so I have no idea who they are! But the people I do know are:
2627 Ash Berlin (order_by hash term support)
2628 Matt Trout (DBIx::Class support)
2629 Mark Stosberg (benchmarking)
2630 Chas Owens (initial "IN" operator support)
2631 Philip Collins (per-field SQL functions)
2632 Eric Kolve (hashref "AND" support)
2633 Mike Fragassi (enhancements to "BETWEEN" and "LIKE")
2634 Dan Kubb (support for "quote_char" and "name_sep")
2635 Guillermo Roditi (patch to cleanup "IN" and "BETWEEN", fix and tests for _order_by)
2636 Laurent Dami (internal refactoring, multiple -nest, extensible list of special operators, literal SQL)
2637 Norbert Buchmuller (support for literal SQL in hashpair, misc. fixes & tests)
2638 Peter Rabbitson (rewrite of SQLA::Test, misc. fixes & tests)
2644 L<DBIx::Class>, L<DBIx::Abstract>, L<CGI::FormBuilder>, L<HTML::QuickTable>.
2648 Copyright (c) 2001-2007 Nathan Wiger <nwiger@cpan.org>. All Rights Reserved.
2650 This module is actively maintained by Matt Trout <mst@shadowcatsystems.co.uk>
2652 For support, your best bet is to try the C<DBIx::Class> users mailing list.
2653 While not an official support venue, C<DBIx::Class> makes heavy use of
2654 C<SQL::Abstract>, and as such list members there are very familiar with
2655 how to create queries.
2659 This module is free software; you may copy this under the terms of
2660 the GNU General Public License, or the Artistic License, copies of
2661 which should have accompanied your Perl kit.