1 package SQL::Abstract; # see doc at end of file
10 our @EXPORT_OK = qw(is_plain_value is_literal_value);
20 *SQL::Abstract::_ENV_::DETECT_AUTOGENERATED_STRINGIFICATION = $ENV{SQLA_ISVALUE_IGNORE_AUTOGENERATED_STRINGIFICATION}
26 #======================================================================
28 #======================================================================
30 our $VERSION = '1.87';
32 # This would confuse some packagers
33 $VERSION = eval $VERSION if $VERSION =~ /_/; # numify for warning-free dev releases
37 # special operators (-in, -between). May be extended/overridden by user.
38 # See section WHERE: BUILTIN SPECIAL OPERATORS below for implementation
39 my @BUILTIN_SPECIAL_OPS = (
40 {regex => qr/^ (?: not \s )? between $/ix, handler => '_where_field_BETWEEN'},
41 {regex => qr/^ (?: not \s )? in $/ix, handler => '_where_field_IN'},
42 {regex => qr/^ ident $/ix, handler => '_where_op_IDENT'},
43 {regex => qr/^ value $/ix, handler => '_where_op_VALUE'},
44 {regex => qr/^ is (?: \s+ not )? $/ix, handler => '_where_field_IS'},
47 # unaryish operators - key maps to handler
48 my @BUILTIN_UNARY_OPS = (
49 # the digits are backcompat stuff
50 { regex => qr/^ and (?: [_\s]? \d+ )? $/xi, handler => '_where_op_ANDOR' },
51 { regex => qr/^ or (?: [_\s]? \d+ )? $/xi, handler => '_where_op_ANDOR' },
52 { regex => qr/^ nest (?: [_\s]? \d+ )? $/xi, handler => '_where_op_NEST' },
53 { regex => qr/^ (?: not \s )? bool $/xi, handler => '_where_op_BOOL' },
54 { regex => qr/^ ident $/xi, handler => '_where_op_IDENT' },
55 { regex => qr/^ value $/xi, handler => '_where_op_VALUE' },
56 { regex => qr/^ op $/xi, handler => '_where_op_OP' },
57 { regex => qr/^ bind $/xi, handler => '_where_op_BIND' },
58 { regex => qr/^ literal $/xi, handler => '_where_op_LITERAL' },
61 #======================================================================
62 # DEBUGGING AND ERROR REPORTING
63 #======================================================================
66 return unless $_[0]->{debug}; shift; # a little faster
67 my $func = (caller(1))[3];
68 warn "[$func] ", @_, "\n";
72 my($func) = (caller(1))[3];
73 Carp::carp "[$func] Warning: ", @_;
77 my($func) = (caller(1))[3];
78 Carp::croak "[$func] Fatal: ", @_;
81 sub is_literal_value ($) {
82 ref $_[0] eq 'SCALAR' ? [ ${$_[0]} ]
83 : ( ref $_[0] eq 'REF' and ref ${$_[0]} eq 'ARRAY' ) ? [ @${ $_[0] } ]
87 # FIXME XSify - this can be done so much more efficiently
88 sub is_plain_value ($) {
90 ! length ref $_[0] ? \($_[0])
92 ref $_[0] eq 'HASH' and keys %{$_[0]} == 1
94 exists $_[0]->{-value}
95 ) ? \($_[0]->{-value})
97 # reuse @_ for even moar speedz
98 defined ( $_[1] = Scalar::Util::blessed $_[0] )
100 # deliberately not using Devel::OverloadInfo - the checks we are
101 # intersted in are much more limited than the fullblown thing, and
102 # this is a very hot piece of code
104 # simply using ->can('(""') can leave behind stub methods that
105 # break actually using the overload later (see L<perldiag/Stub
106 # found while resolving method "%s" overloading "%s" in package
107 # "%s"> and the source of overload::mycan())
109 # either has stringification which DBI SHOULD prefer out of the box
110 grep { *{ (qq[${_}::(""]) }{CODE} } @{ $_[2] = mro::get_linear_isa( $_[1] ) }
112 # has nummification or boolification, AND fallback is *not* disabled
114 SQL::Abstract::_ENV_::DETECT_AUTOGENERATED_STRINGIFICATION
117 grep { *{"${_}::(0+"}{CODE} } @{$_[2]}
119 grep { *{"${_}::(bool"}{CODE} } @{$_[2]}
123 # no fallback specified at all
124 ! ( ($_[3]) = grep { *{"${_}::()"}{CODE} } @{$_[2]} )
126 # fallback explicitly undef
127 ! defined ${"$_[3]::()"}
140 #======================================================================
142 #======================================================================
146 my $class = ref($self) || $self;
147 my %opt = (ref $_[0] eq 'HASH') ? %{$_[0]} : @_;
149 # choose our case by keeping an option around
150 delete $opt{case} if $opt{case} && $opt{case} ne 'lower';
152 # default logic for interpreting arrayrefs
153 $opt{logic} = $opt{logic} ? uc $opt{logic} : 'OR';
155 # how to return bind vars
156 $opt{bindtype} ||= 'normal';
158 # default comparison is "=", but can be overridden
161 # try to recognize which are the 'equality' and 'inequality' ops
162 # (temporary quickfix (in 2007), should go through a more seasoned API)
163 $opt{equality_op} = qr/^( \Q$opt{cmp}\E | \= )$/ix;
164 $opt{inequality_op} = qr/^( != | <> )$/ix;
166 $opt{like_op} = qr/^ (is\s+)? r?like $/xi;
167 $opt{not_like_op} = qr/^ (is\s+)? not \s+ r?like $/xi;
170 $opt{sqltrue} ||= '1=1';
171 $opt{sqlfalse} ||= '0=1';
174 $opt{special_ops} ||= [];
175 # regexes are applied in order, thus push after user-defines
176 push @{$opt{special_ops}}, @BUILTIN_SPECIAL_OPS;
179 $opt{unary_ops} ||= [];
180 push @{$opt{unary_ops}}, @BUILTIN_UNARY_OPS;
182 # rudimentary sanity-check for user supplied bits treated as functions/operators
183 # If a purported function matches this regular expression, an exception is thrown.
184 # Literal SQL is *NOT* subject to this check, only functions (and column names
185 # when quoting is not in effect)
188 # need to guard against ()'s in column names too, but this will break tons of
189 # hacks... ideas anyone?
190 $opt{injection_guard} ||= qr/
196 return bless \%opt, $class;
200 sub _assert_pass_injection_guard {
201 if ($_[1] =~ $_[0]->{injection_guard}) {
202 my $class = ref $_[0];
203 puke "Possible SQL injection attempt '$_[1]'. If this is indeed a part of the "
204 . "desired SQL use literal SQL ( \'...' or \[ '...' ] ) or supply your own "
205 . "{injection_guard} attribute to ${class}->new()"
210 #======================================================================
212 #======================================================================
216 my $table = $self->_table(shift);
217 my $data = shift || return;
220 my $method = $self->_METHOD_FOR_refkind("_insert", $data);
221 my ($sql, @bind) = $self->$method($data);
222 $sql = join " ", $self->_sqlcase('insert into'), $table, $sql;
224 if ($options->{returning}) {
225 my ($s, @b) = $self->_insert_returning($options);
230 return wantarray ? ($sql, @bind) : $sql;
233 # So that subclasses can override INSERT ... RETURNING separately from
234 # UPDATE and DELETE (e.g. DBIx::Class::SQLMaker::Oracle does this)
235 sub _insert_returning { shift->_returning(@_) }
238 my ($self, $options) = @_;
240 my $f = $options->{returning};
242 my $fieldlist = $self->_SWITCH_refkind($f, {
243 ARRAYREF => sub {join ', ', map { $self->_quote($_) } @$f;},
244 SCALAR => sub {$self->_quote($f)},
245 SCALARREF => sub {$$f},
247 return $self->_sqlcase(' returning ') . $fieldlist;
250 sub _insert_HASHREF { # explicit list of fields and then values
251 my ($self, $data) = @_;
253 my @fields = sort keys %$data;
255 my ($sql, @bind) = $self->_insert_values($data);
258 $_ = $self->_quote($_) foreach @fields;
259 $sql = "( ".join(", ", @fields).") ".$sql;
261 return ($sql, @bind);
264 sub _insert_ARRAYREF { # just generate values(?,?) part (no list of fields)
265 my ($self, $data) = @_;
267 # no names (arrayref) so can't generate bindtype
268 $self->{bindtype} ne 'columns'
269 or belch "can't do 'columns' bindtype when called with arrayref";
271 my (@values, @all_bind);
272 foreach my $value (@$data) {
273 my ($values, @bind) = $self->_insert_value(undef, $value);
274 push @values, $values;
275 push @all_bind, @bind;
277 my $sql = $self->_sqlcase('values')." ( ".join(", ", @values)." )";
278 return ($sql, @all_bind);
281 sub _insert_ARRAYREFREF { # literal SQL with bind
282 my ($self, $data) = @_;
284 my ($sql, @bind) = @${$data};
285 $self->_assert_bindval_matches_bindtype(@bind);
287 return ($sql, @bind);
291 sub _insert_SCALARREF { # literal SQL without bind
292 my ($self, $data) = @_;
298 my ($self, $data) = @_;
300 my (@values, @all_bind);
301 foreach my $column (sort keys %$data) {
302 my ($values, @bind) = $self->_insert_value($column, $data->{$column});
303 push @values, $values;
304 push @all_bind, @bind;
306 my $sql = $self->_sqlcase('values')." ( ".join(", ", @values)." )";
307 return ($sql, @all_bind);
311 my ($self, $column, $v) = @_;
313 my (@values, @all_bind);
314 $self->_SWITCH_refkind($v, {
317 if ($self->{array_datatypes}) { # if array datatype are activated
319 push @all_bind, $self->_bindtype($column, $v);
321 else { # else literal SQL with bind
322 my ($sql, @bind) = @$v;
323 $self->_assert_bindval_matches_bindtype(@bind);
325 push @all_bind, @bind;
329 ARRAYREFREF => sub { # literal SQL with bind
330 my ($sql, @bind) = @${$v};
331 $self->_assert_bindval_matches_bindtype(@bind);
333 push @all_bind, @bind;
336 # THINK: anything useful to do with a HASHREF ?
337 HASHREF => sub { # (nothing, but old SQLA passed it through)
338 #TODO in SQLA >= 2.0 it will die instead
339 belch "HASH ref as bind value in insert is not supported";
341 push @all_bind, $self->_bindtype($column, $v);
344 SCALARREF => sub { # literal SQL without bind
348 SCALAR_or_UNDEF => sub {
350 push @all_bind, $self->_bindtype($column, $v);
355 my $sql = join(", ", @values);
356 return ($sql, @all_bind);
361 #======================================================================
363 #======================================================================
368 my $table = $self->_table(shift);
369 my $data = shift || return;
373 # first build the 'SET' part of the sql statement
374 puke "Unsupported data type specified to \$sql->update"
375 unless ref $data eq 'HASH';
377 my ($sql, @all_bind) = $self->_update_set_values($data);
378 $sql = $self->_sqlcase('update ') . $table . $self->_sqlcase(' set ')
382 my($where_sql, @where_bind) = $self->where($where);
384 push @all_bind, @where_bind;
387 if ($options->{returning}) {
388 my ($returning_sql, @returning_bind) = $self->_update_returning($options);
389 $sql .= $returning_sql;
390 push @all_bind, @returning_bind;
393 return wantarray ? ($sql, @all_bind) : $sql;
396 sub _update_set_values {
397 my ($self, $data) = @_;
399 my (@set, @all_bind);
400 for my $k (sort keys %$data) {
403 my $label = $self->_quote($k);
405 $self->_SWITCH_refkind($v, {
407 if ($self->{array_datatypes}) { # array datatype
408 push @set, "$label = ?";
409 push @all_bind, $self->_bindtype($k, $v);
411 else { # literal SQL with bind
412 my ($sql, @bind) = @$v;
413 $self->_assert_bindval_matches_bindtype(@bind);
414 push @set, "$label = $sql";
415 push @all_bind, @bind;
418 ARRAYREFREF => sub { # literal SQL with bind
419 my ($sql, @bind) = @${$v};
420 $self->_assert_bindval_matches_bindtype(@bind);
421 push @set, "$label = $sql";
422 push @all_bind, @bind;
424 SCALARREF => sub { # literal SQL without bind
425 push @set, "$label = $$v";
428 my ($op, $arg, @rest) = %$v;
430 puke 'Operator calls in update must be in the form { -op => $arg }'
431 if (@rest or not $op =~ /^\-(.+)/);
433 local $self->{_nested_func_lhs} = $k;
434 my ($sql, @bind) = $self->_where_unary_op($1, $arg);
436 push @set, "$label = $sql";
437 push @all_bind, @bind;
439 SCALAR_or_UNDEF => sub {
440 push @set, "$label = ?";
441 push @all_bind, $self->_bindtype($k, $v);
447 my $sql = join ', ', @set;
449 return ($sql, @all_bind);
452 # So that subclasses can override UPDATE ... RETURNING separately from
454 sub _update_returning { shift->_returning(@_) }
458 #======================================================================
460 #======================================================================
465 my $table = $self->_table(shift);
466 my $fields = shift || '*';
470 my ($fields_sql, @bind) = $self->_select_fields($fields);
472 my ($where_sql, @where_bind) = $self->where($where, $order);
473 push @bind, @where_bind;
475 my $sql = join(' ', $self->_sqlcase('select'), $fields_sql,
476 $self->_sqlcase('from'), $table)
479 return wantarray ? ($sql, @bind) : $sql;
483 my ($self, $fields) = @_;
484 return ref $fields eq 'ARRAY' ? join ', ', map { $self->_quote($_) } @$fields
488 #======================================================================
490 #======================================================================
495 my $table = $self->_table(shift);
499 my($where_sql, @bind) = $self->where($where);
500 my $sql = $self->_sqlcase('delete from ') . $table . $where_sql;
502 if ($options->{returning}) {
503 my ($returning_sql, @returning_bind) = $self->_delete_returning($options);
504 $sql .= $returning_sql;
505 push @bind, @returning_bind;
508 return wantarray ? ($sql, @bind) : $sql;
511 # So that subclasses can override DELETE ... RETURNING separately from
513 sub _delete_returning { shift->_returning(@_) }
517 #======================================================================
519 #======================================================================
523 # Finally, a separate routine just to handle WHERE clauses
525 my ($self, $where, $order) = @_;
528 my ($sql, @bind) = $self->_recurse_where($where);
529 $sql = (defined $sql and length $sql) ? $self->_sqlcase(' where ') . "( $sql )" : '';
533 my ($order_sql, @order_bind) = $self->_order_by($order);
535 push @bind, @order_bind;
538 return wantarray ? ($sql, @bind) : $sql;
542 my ($self, $expr, $logic) = @_;
543 if (ref($expr) eq 'HASH') {
544 if (keys %$expr > 1) {
546 return +{ "-${logic}" => [
547 map $self->_expand_expr_hashpair($_ => $expr->{$_}, $logic),
551 return $self->_expand_expr_hashpair(%$expr, $logic);
553 if (ref($expr) eq 'ARRAY') {
554 $logic = lc($logic || $self->{logic});
555 $logic eq 'and' or $logic eq 'or' or puke "unknown logic: $logic";
561 while (my ($el) = splice @expr, 0, 1) {
562 puke "Supplying an empty left hand side argument is not supported in array-pairs"
563 unless defined($el) and length($el);
564 my $elref = ref($el);
566 push(@res, $self->_expand_expr({ $el, shift(@expr) }));
567 } elsif ($elref eq 'ARRAY') {
568 push(@res, $self->_expand_expr($el)) if @$el;
569 } elsif (is_literal_value($el)) {
571 } elsif ($elref eq 'HASH') {
572 push @res, $self->_expand_expr($el);
577 return { '-'.$logic => \@res };
579 if (my $literal = is_literal_value($expr)) {
580 return +{ -literal => $literal };
585 sub _expand_expr_hashpair {
586 my ($self, $k, $v, $logic) = @_;
587 unless (defined($k) and length($k)) {
588 if (defined($k) and my $literal = is_literal_value($v)) {
589 belch 'Hash-pairs consisting of an empty string with a literal are deprecated, and will be removed in 2.0: use -and => [ $literal ] instead';
590 return { -literal => $literal };
592 puke "Supplying an empty left hand side argument is not supported";
596 return $self->_expand_expr($v);
600 return $self->_expand_expr($v);
602 puke "-bool => undef not supported" unless defined($v);
603 return { -ident => $v };
605 if (my ($rest) = $k =~/^-not[_ ](.*)$/) {
606 return $self->_expand_expr({ -not => { "-${rest}", $v } }, $logic);
608 if (my ($logic) = $k =~ /^-(and|or)$/) {
609 if (ref($v) eq 'HASH') {
610 return $self->_expand_expr($v, $logic);
614 unless (defined($v)) {
615 my $orig_op = my $op = $self->{cmp};
617 $op =~ /^not$/i ? 'is not' # legacy
618 : $op =~ $self->{equality_op} ? 'is'
619 : $op =~ $self->{like_op} ? belch("Supplying an undefined argument to '@{[ uc $op]}' is deprecated") && 'is'
620 : $op =~ $self->{inequality_op} ? 'is not'
621 : $op =~ $self->{not_like_op} ? belch("Supplying an undefined argument to '@{[ uc $op]}' is deprecated") && 'is not'
622 : puke "unexpected operator '$orig_op' with undef operand";
623 return +{ -op => [ $is.' null', { -ident => $k } ] };
630 { -bind => [ $k, $v ] }
634 if (ref($v) eq 'HASH' and keys %$v > 1) {
636 map $self->_expand_expr_hashpair($k => { $_ => $v->{$_} }),
640 if (ref($v) eq 'ARRAY') {
641 return $self->{sqlfalse} unless @$v;
642 $self->_debug("ARRAY($k) means distribute over elements");
644 $v->[0] =~ /^-((?:and|or))$/i
645 ? ($v = [ @{$v}[1..$#$v] ], $1)
646 : ($self->{logic} || 'or')
648 return +{ "-${this_logic}" => [ map $self->_expand_expr({ $k => $_ }, $this_logic), @$v ] };
650 if (my $literal = is_literal_value($v)) {
652 belch 'Hash-pairs consisting of an empty string with a literal are deprecated, and will be removed in 2.0: use -and => [ $literal ] instead';
655 my ($sql, @bind) = @$literal;
656 if ($self->{bindtype} eq 'columns') {
658 if (!defined $_ || ref($_) ne 'ARRAY' || @$_ != 2) {
659 puke "bindtype 'columns' selected, you need to pass: [column_name => bind_value]"
663 return +{ -literal => [ $self->_quote($k).' '.$sql, @bind ] };
670 my ($self, $where, $logic) = @_;
672 my $where_exp = $self->_expand_expr($where, $logic);
674 # dispatch on appropriate method according to refkind of $where
675 my $method = $self->_METHOD_FOR_refkind("_where", $where_exp);
677 my ($sql, @bind) = $self->$method($where_exp, $logic);
679 # DBIx::Class used to call _recurse_where in scalar context
680 # something else might too...
682 return ($sql, @bind);
685 belch "Calling _recurse_where in scalar context is deprecated and will go away before 2.0";
692 #======================================================================
693 # WHERE: top-level ARRAYREF
694 #======================================================================
697 sub _where_ARRAYREF {
698 my ($self, $where, $logic) = @_;
700 $logic = uc($logic || $self->{logic});
701 $logic eq 'AND' or $logic eq 'OR' or puke "unknown logic: $logic";
703 my @clauses = @$where;
705 my (@sql_clauses, @all_bind);
706 # need to use while() so can shift() for pairs
708 my $el = shift @clauses;
710 $el = undef if (defined $el and ! length $el);
712 # switch according to kind of $el and get corresponding ($sql, @bind)
713 my ($sql, @bind) = $self->_SWITCH_refkind($el, {
715 # skip empty elements, otherwise get invalid trailing AND stuff
716 ARRAYREF => sub {$self->_recurse_where($el) if @$el},
720 $self->_assert_bindval_matches_bindtype(@b);
724 HASHREF => sub {$self->_recurse_where($el, 'and') if %$el},
726 SCALARREF => sub { ($$el); },
729 # top-level arrayref with scalars, recurse in pairs
730 $self->_recurse_where({$el => shift(@clauses)})
733 UNDEF => sub {puke "Supplying an empty left hand side argument is not supported in array-pairs" },
737 push @sql_clauses, $sql;
738 push @all_bind, @bind;
742 return $self->_join_sql_clauses($logic, \@sql_clauses, \@all_bind);
745 #======================================================================
746 # WHERE: top-level ARRAYREFREF
747 #======================================================================
749 sub _where_ARRAYREFREF {
750 my ($self, $where) = @_;
751 my ($sql, @bind) = @$$where;
752 $self->_assert_bindval_matches_bindtype(@bind);
753 return ($sql, @bind);
756 #======================================================================
757 # WHERE: top-level HASHREF
758 #======================================================================
761 my ($self, $where) = @_;
762 my (@sql_clauses, @all_bind);
764 for my $k (sort keys %$where) {
765 my $v = $where->{$k};
767 # ($k => $v) is either a special unary op or a regular hashpair
768 my ($sql, @bind) = do {
770 # put the operator in canonical form
772 $op = substr $op, 1; # remove initial dash
773 $op =~ s/^\s+|\s+$//g;# remove leading/trailing space
774 $op =~ s/\s+/ /g; # compress whitespace
776 # so that -not_foo works correctly
777 $op =~ s/^not_/NOT /i;
779 $self->_debug("Unary OP(-$op) within hashref, recursing...");
780 my ($s, @b) = $self->_where_unary_op($op, $v);
782 # top level vs nested
783 # we assume that handled unary ops will take care of their ()s
785 List::Util::first {$op =~ $_->{regex}} @{$self->{unary_ops}}
787 ( defined $self->{_nested_func_lhs} and $self->{_nested_func_lhs} eq $k )
793 if (is_literal_value ($v) ) {
794 belch 'Hash-pairs consisting of an empty string with a literal are deprecated, and will be removed in 2.0: use -and => [ $literal ] instead';
797 puke "Supplying an empty left hand side argument is not supported in hash-pairs";
801 my $method = $self->_METHOD_FOR_refkind("_where_hashpair", $v);
802 $self->$method($k, $v);
806 push @sql_clauses, $sql;
807 push @all_bind, @bind;
810 return $self->_join_sql_clauses('and', \@sql_clauses, \@all_bind);
813 sub _where_unary_op {
814 my ($self, $op, $rhs) = @_;
816 # top level special ops are illegal in general
817 puke "Illegal use of top-level '-$op'"
818 if !(defined $self->{_nested_func_lhs})
819 and List::Util::first { $op =~ $_->{regex} } @{$self->{special_ops}}
820 and not List::Util::first { $op =~ $_->{regex} } @{$self->{unary_ops}};
822 if (my $op_entry = List::Util::first { $op =~ $_->{regex} } @{$self->{unary_ops}}) {
823 my $handler = $op_entry->{handler};
825 if (not ref $handler) {
826 if ($op =~ s/ [_\s]? \d+ $//x ) {
827 belch 'Use of [and|or|nest]_N modifiers is deprecated and will be removed in SQLA v2.0. '
828 . "You probably wanted ...-and => [ -$op => COND1, -$op => COND2 ... ]";
830 return $self->$handler($op, $rhs);
832 elsif (ref $handler eq 'CODE') {
833 return $handler->($self, $op, $rhs);
836 puke "Illegal handler for operator $op - expecting a method name or a coderef";
840 $self->_debug("Generic unary OP: $op - recursing as function");
842 $self->_assert_pass_injection_guard($op);
844 my ($sql, @bind) = $self->_SWITCH_refkind($rhs, {
846 puke "Illegal use of top-level '-$op'"
847 unless defined $self->{_nested_func_lhs};
850 $self->_convert('?'),
851 $self->_bindtype($self->{_nested_func_lhs}, $rhs)
855 $self->_recurse_where($rhs)
859 $sql = sprintf('%s %s',
860 $self->_sqlcase($op),
864 return ($sql, @bind);
867 sub _where_op_ANDOR {
868 my ($self, $op, $v) = @_;
870 $self->_SWITCH_refkind($v, {
872 return $self->_where_ARRAYREF($v, $op);
876 return ($op =~ /^or/i)
877 ? $self->_where_ARRAYREF([ map { $_ => $v->{$_} } (sort keys %$v) ], $op)
878 : $self->_where_HASHREF($v);
882 puke "-$op => \\\$scalar makes little sense, use " .
884 ? '[ \$scalar, \%rest_of_conditions ] instead'
885 : '-and => [ \$scalar, \%rest_of_conditions ] instead'
890 puke "-$op => \\[...] makes little sense, use " .
892 ? '[ \[...], \%rest_of_conditions ] instead'
893 : '-and => [ \[...], \%rest_of_conditions ] instead'
897 SCALAR => sub { # permissively interpreted as SQL
898 puke "-$op => \$value makes little sense, use -bool => \$value instead";
902 puke "-$op => undef not supported";
908 my ($self, $op, $v) = @_;
910 $self->_SWITCH_refkind($v, {
912 SCALAR => sub { # permissively interpreted as SQL
913 belch "literal SQL should be -nest => \\'scalar' "
914 . "instead of -nest => 'scalar' ";
919 puke "-$op => undef not supported";
923 $self->_recurse_where($v);
931 my ($self, $op, $v) = @_;
933 my ($s, @b) = $self->_SWITCH_refkind($v, {
934 SCALAR => sub { # interpreted as SQL column
935 $self->_convert($self->_quote($v));
939 puke "-$op => undef not supported";
943 $self->_recurse_where($v);
947 $s = "(NOT $s)" if $op =~ /^not/i;
952 sub _where_op_IDENT {
954 my ($op, $rhs) = splice @_, -2;
955 if (! defined $rhs or length ref $rhs) {
956 puke "-$op requires a single plain scalar argument (a quotable identifier)";
959 # in case we are called as a top level special op (no '=')
960 my $has_lhs = my $lhs = shift;
962 $_ = $self->_convert($self->_quote($_)) for ($lhs, $rhs);
970 sub _where_op_VALUE {
972 my ($op, $rhs) = splice @_, -2;
974 # in case we are called as a top level special op (no '=')
978 if (! defined $rhs) {
980 ? $self->_where_hashpair_HASHREF($lhs, { -is => undef })
987 (defined $lhs ? $lhs : $self->{_nested_func_lhs}),
994 $self->_convert($self->_quote($lhs)) . ' = ' . $self->_convert('?'),
998 $self->_convert('?'),
1005 my %unop_postfix = map +($_ => 1), 'is null', 'is not null';
1008 my ($self, undef, $v) = @_;
1009 my ($op, @args) = @$v;
1010 $op =~ s/^-// if length($op) > 1;
1011 local $self->{_nested_func_lhs};
1013 my ($expr_sql, @bind) = $self->_recurse_where($args[0]);
1014 my $final_op = join ' ', split '_', $op;
1015 my $op_sql = $self->_sqlcase($final_op);
1017 $unop_postfix{lc($final_op)}
1018 ? "${expr_sql} ${op_sql}"
1019 : "${op_sql} ${expr_sql}"
1021 return ($final_sql, @bind);
1022 } elsif (@args == 2) {
1023 my ($l, $r) = map [ $self->_recurse_where($_) ], @args;
1024 return ( $l->[0].' '.$self->_sqlcase(join ' ', split '_', $op).' '.$r->[0], @{$l}[1..$#$l], @{$r}[1..$#$r] );
1029 sub _where_op_BIND {
1030 my ($self, undef, $bind) = @_;
1031 return ($self->_convert('?'), $self->_bindtype(@$bind));
1034 sub _where_op_LITERAL {
1035 my ($self, undef, $literal) = @_;
1039 sub _where_hashpair_ARRAYREF {
1040 my ($self, $k, $v) = @_;
1043 my @v = @$v; # need copy because of shift below
1044 $self->_debug("ARRAY($k) means distribute over elements");
1046 # put apart first element if it is an operator (-and, -or)
1048 (defined $v[0] && $v[0] =~ /^ - (?: AND|OR ) $/ix)
1052 my @distributed = map { {$k => $_} } @v;
1055 $self->_debug("OP($op) reinjected into the distributed array");
1056 unshift @distributed, $op;
1059 my $logic = $op ? substr($op, 1) : '';
1061 return $self->_recurse_where(\@distributed, $logic);
1064 $self->_debug("empty ARRAY($k) means 0=1");
1065 return ($self->{sqlfalse});
1069 sub _where_hashpair_HASHREF {
1070 my ($self, $k, $v, $logic) = @_;
1073 local $self->{_nested_func_lhs} = defined $self->{_nested_func_lhs}
1074 ? $self->{_nested_func_lhs}
1078 my ($all_sql, @all_bind);
1080 for my $orig_op (sort keys %$v) {
1081 my $val = $v->{$orig_op};
1083 # put the operator in canonical form
1086 # FIXME - we need to phase out dash-less ops
1087 $op =~ s/^-//; # remove possible initial dash
1088 $op =~ s/^\s+|\s+$//g;# remove leading/trailing space
1089 $op =~ s/\s+/ /g; # compress whitespace
1091 $self->_assert_pass_injection_guard($op);
1094 $op =~ s/^is_not/IS NOT/i;
1096 # so that -not_foo works correctly
1097 $op =~ s/^not_/NOT /i;
1099 # another retarded special case: foo => { $op => { -value => undef } }
1100 if (ref $val eq 'HASH' and keys %$val == 1 and exists $val->{-value} and ! defined $val->{-value} ) {
1106 # CASE: col-value logic modifiers
1107 if ($orig_op =~ /^ \- (and|or) $/xi) {
1108 ($sql, @bind) = $self->_where_hashpair_HASHREF($k, $val, $1);
1110 # CASE: special operators like -in or -between
1111 elsif (my $special_op = List::Util::first { $op =~ $_->{regex} } @{$self->{special_ops}}) {
1112 my $handler = $special_op->{handler};
1114 puke "No handler supplied for special operator $orig_op";
1116 elsif (not ref $handler) {
1117 ($sql, @bind) = $self->$handler($k, $op, $val);
1119 elsif (ref $handler eq 'CODE') {
1120 ($sql, @bind) = $handler->($self, $k, $op, $val);
1123 puke "Illegal handler for special operator $orig_op - expecting a method name or a coderef";
1127 $self->_SWITCH_refkind($val, {
1129 ARRAYREF => sub { # CASE: col => {op => \@vals}
1130 ($sql, @bind) = $self->_where_field_op_ARRAYREF($k, $op, $val);
1133 ARRAYREFREF => sub { # CASE: col => {op => \[$sql, @bind]} (literal SQL with bind)
1134 my ($sub_sql, @sub_bind) = @$$val;
1135 $self->_assert_bindval_matches_bindtype(@sub_bind);
1136 $sql = join ' ', $self->_convert($self->_quote($k)),
1137 $self->_sqlcase($op),
1142 UNDEF => sub { # CASE: col => {op => undef} : sql "IS (NOT)? NULL"
1144 $op =~ /^not$/i ? 'is not' # legacy
1145 : $op =~ $self->{equality_op} ? 'is'
1146 : $op =~ $self->{like_op} ? belch("Supplying an undefined argument to '@{[ uc $op]}' is deprecated") && 'is'
1147 : $op =~ $self->{inequality_op} ? 'is not'
1148 : $op =~ $self->{not_like_op} ? belch("Supplying an undefined argument to '@{[ uc $op]}' is deprecated") && 'is not'
1149 : puke "unexpected operator '$orig_op' with undef operand";
1151 $sql = $self->_quote($k) . $self->_sqlcase(" $is null");
1154 FALLBACK => sub { # CASE: col => {op/func => $stuff}
1155 ($sql, @bind) = $self->_where_unary_op($op, $val);
1158 $self->_convert($self->_quote($k)),
1159 $self->{_nested_func_lhs} eq $k ? $sql : "($sql)", # top level vs nested
1165 ($all_sql) = (defined $all_sql and $all_sql) ? $self->_join_sql_clauses($logic, [$all_sql, $sql], []) : $sql;
1166 push @all_bind, @bind;
1168 return ($all_sql, @all_bind);
1171 sub _where_field_IS {
1172 my ($self, $k, $op, $v) = @_;
1174 my ($s) = $self->_SWITCH_refkind($v, {
1177 $self->_convert($self->_quote($k)),
1178 map { $self->_sqlcase($_)} ($op, 'null')
1181 puke "$op can only take undef as argument";
1188 sub _where_field_op_ARRAYREF {
1189 my ($self, $k, $op, $vals) = @_;
1191 my @vals = @$vals; #always work on a copy
1194 $self->_debug(sprintf '%s means multiple elements: [ %s ]',
1196 join(', ', map { defined $_ ? "'$_'" : 'NULL' } @vals ),
1199 # see if the first element is an -and/-or op
1201 if (defined $vals[0] && $vals[0] =~ /^ - (AND|OR) $/ix) {
1206 # a long standing API wart - an attempt to change this behavior during
1207 # the 1.50 series failed *spectacularly*. Warn instead and leave the
1212 (!$logic or $logic eq 'OR')
1214 ($op =~ $self->{inequality_op} or $op =~ $self->{not_like_op})
1217 belch "A multi-element arrayref as an argument to the inequality op '$o' "
1218 . 'is technically equivalent to an always-true 1=1 (you probably wanted '
1219 . "to say ...{ \$inequality_op => [ -and => \@values ] }... instead)"
1223 # distribute $op over each remaining member of @vals, append logic if exists
1224 return $self->_recurse_where([map { {$k => {$op, $_}} } @vals], $logic);
1228 # try to DWIM on equality operators
1230 $op =~ $self->{equality_op} ? $self->{sqlfalse}
1231 : $op =~ $self->{like_op} ? belch("Supplying an empty arrayref to '@{[ uc $op]}' is deprecated") && $self->{sqlfalse}
1232 : $op =~ $self->{inequality_op} ? $self->{sqltrue}
1233 : $op =~ $self->{not_like_op} ? belch("Supplying an empty arrayref to '@{[ uc $op]}' is deprecated") && $self->{sqltrue}
1234 : puke "operator '$op' applied on an empty array (field '$k')";
1239 sub _where_hashpair_SCALARREF {
1240 my ($self, $k, $v) = @_;
1241 $self->_debug("SCALAR($k) means literal SQL: $$v");
1242 my $sql = $self->_quote($k) . " " . $$v;
1246 # literal SQL with bind
1247 sub _where_hashpair_ARRAYREFREF {
1248 my ($self, $k, $v) = @_;
1249 $self->_debug("REF($k) means literal SQL: @${$v}");
1250 my ($sql, @bind) = @$$v;
1251 $self->_assert_bindval_matches_bindtype(@bind);
1252 $sql = $self->_quote($k) . " " . $sql;
1253 return ($sql, @bind );
1256 # literal SQL without bind
1257 sub _where_hashpair_SCALAR {
1258 my ($self, $k, $v) = @_;
1259 $self->_debug("NOREF($k) means simple key=val: $k $self->{cmp} $v");
1260 return ($self->_where_hashpair_HASHREF($k, { $self->{cmp} => $v }));
1264 sub _where_hashpair_UNDEF {
1265 my ($self, $k, $v) = @_;
1266 $self->_debug("UNDEF($k) means IS NULL");
1267 return $self->_where_hashpair_HASHREF($k, { -is => undef });
1270 #======================================================================
1271 # WHERE: TOP-LEVEL OTHERS (SCALARREF, SCALAR, UNDEF)
1272 #======================================================================
1275 sub _where_SCALARREF {
1276 my ($self, $where) = @_;
1279 $self->_debug("SCALAR(*top) means literal SQL: $$where");
1285 my ($self, $where) = @_;
1288 $self->_debug("NOREF(*top) means literal SQL: $where");
1299 #======================================================================
1300 # WHERE: BUILTIN SPECIAL OPERATORS (-in, -between)
1301 #======================================================================
1304 sub _where_field_BETWEEN {
1305 my ($self, $k, $op, $vals) = @_;
1307 my ($label, $and, $placeholder);
1308 $label = $self->_convert($self->_quote($k));
1309 $and = ' ' . $self->_sqlcase('and') . ' ';
1310 $placeholder = $self->_convert('?');
1311 $op = $self->_sqlcase($op);
1313 my $invalid_args = "Operator '$op' requires either an arrayref with two defined values or expressions, or a single literal scalarref/arrayref-ref";
1315 my ($clause, @bind) = $self->_SWITCH_refkind($vals, {
1316 ARRAYREFREF => sub {
1317 my ($s, @b) = @$$vals;
1318 $self->_assert_bindval_matches_bindtype(@b);
1325 puke $invalid_args if @$vals != 2;
1327 my (@all_sql, @all_bind);
1328 foreach my $val (@$vals) {
1329 my ($sql, @bind) = $self->_SWITCH_refkind($val, {
1331 return ($placeholder, $self->_bindtype($k, $val) );
1336 ARRAYREFREF => sub {
1337 my ($sql, @bind) = @$$val;
1338 $self->_assert_bindval_matches_bindtype(@bind);
1339 return ($sql, @bind);
1342 my ($func, $arg, @rest) = %$val;
1343 puke "Only simple { -func => arg } functions accepted as sub-arguments to BETWEEN"
1344 if (@rest or $func !~ /^ \- (.+)/x);
1345 $self->_where_unary_op($1 => $arg);
1351 push @all_sql, $sql;
1352 push @all_bind, @bind;
1356 (join $and, @all_sql),
1365 my $sql = "( $label $op $clause )";
1366 return ($sql, @bind)
1370 sub _where_field_IN {
1371 my ($self, $k, $op, $vals) = @_;
1373 # backwards compatibility: if scalar, force into an arrayref
1374 $vals = [$vals] if defined $vals && ! ref $vals;
1376 my ($label) = $self->_convert($self->_quote($k));
1377 my ($placeholder) = $self->_convert('?');
1378 $op = $self->_sqlcase($op);
1380 my ($sql, @bind) = $self->_SWITCH_refkind($vals, {
1381 ARRAYREF => sub { # list of choices
1382 if (@$vals) { # nonempty list
1383 my (@all_sql, @all_bind);
1385 for my $val (@$vals) {
1386 my ($sql, @bind) = $self->_SWITCH_refkind($val, {
1388 return ($placeholder, $val);
1393 ARRAYREFREF => sub {
1394 my ($sql, @bind) = @$$val;
1395 $self->_assert_bindval_matches_bindtype(@bind);
1396 return ($sql, @bind);
1399 my ($func, $arg, @rest) = %$val;
1400 puke "Only simple { -func => arg } functions accepted as sub-arguments to IN"
1401 if (@rest or $func !~ /^ \- (.+)/x);
1402 $self->_where_unary_op($1 => $arg);
1406 'SQL::Abstract before v1.75 used to generate incorrect SQL when the '
1407 . "-$op operator was given an undef-containing list: !!!AUDIT YOUR CODE "
1408 . 'AND DATA!!! (the upcoming Data::Query-based version of SQL::Abstract '
1409 . 'will emit the logically correct SQL instead of raising this exception)'
1413 push @all_sql, $sql;
1414 push @all_bind, @bind;
1418 sprintf('%s %s ( %s )',
1421 join(', ', @all_sql)
1423 $self->_bindtype($k, @all_bind),
1426 else { # empty list: some databases won't understand "IN ()", so DWIM
1427 my $sql = ($op =~ /\bnot\b/i) ? $self->{sqltrue} : $self->{sqlfalse};
1432 SCALARREF => sub { # literal SQL
1433 my $sql = $self->_open_outer_paren($$vals);
1434 return ("$label $op ( $sql )");
1436 ARRAYREFREF => sub { # literal SQL with bind
1437 my ($sql, @bind) = @$$vals;
1438 $self->_assert_bindval_matches_bindtype(@bind);
1439 $sql = $self->_open_outer_paren($sql);
1440 return ("$label $op ( $sql )", @bind);
1444 puke "Argument passed to the '$op' operator can not be undefined";
1448 puke "special op $op requires an arrayref (or scalarref/arrayref-ref)";
1452 return ($sql, @bind);
1455 # Some databases (SQLite) treat col IN (1, 2) different from
1456 # col IN ( (1, 2) ). Use this to strip all outer parens while
1457 # adding them back in the corresponding method
1458 sub _open_outer_paren {
1459 my ($self, $sql) = @_;
1461 while (my ($inner) = $sql =~ /^ \s* \( (.*) \) \s* $/xs) {
1463 # there are closing parens inside, need the heavy duty machinery
1464 # to reevaluate the extraction starting from $sql (full reevaluation)
1465 if ($inner =~ /\)/) {
1466 require Text::Balanced;
1468 my (undef, $remainder) = do {
1469 # idiotic design - writes to $@ but *DOES NOT* throw exceptions
1471 Text::Balanced::extract_bracketed($sql, '()', qr/\s*/);
1474 # the entire expression needs to be a balanced bracketed thing
1475 # (after an extract no remainder sans trailing space)
1476 last if defined $remainder and $remainder =~ /\S/;
1486 #======================================================================
1488 #======================================================================
1491 my ($self, $arg) = @_;
1494 for my $c ($self->_order_by_chunks($arg) ) {
1495 $self->_SWITCH_refkind($c, {
1496 SCALAR => sub { push @sql, $c },
1497 ARRAYREF => sub { push @sql, shift @$c; push @bind, @$c },
1503 $self->_sqlcase(' order by'),
1509 return wantarray ? ($sql, @bind) : $sql;
1512 sub _order_by_chunks {
1513 my ($self, $arg) = @_;
1515 return $self->_SWITCH_refkind($arg, {
1518 map { $self->_order_by_chunks($_ ) } @$arg;
1521 ARRAYREFREF => sub {
1522 my ($s, @b) = @$$arg;
1523 $self->_assert_bindval_matches_bindtype(@b);
1527 SCALAR => sub {$self->_quote($arg)},
1529 UNDEF => sub {return () },
1531 SCALARREF => sub {$$arg}, # literal SQL, no quoting
1534 # get first pair in hash
1535 my ($key, $val, @rest) = %$arg;
1537 return () unless $key;
1539 if (@rest or not $key =~ /^-(desc|asc)/i) {
1540 puke "hash passed to _order_by must have exactly one key (-desc or -asc)";
1546 for my $c ($self->_order_by_chunks($val)) {
1549 $self->_SWITCH_refkind($c, {
1554 ($sql, @bind) = @$c;
1558 $sql = $sql . ' ' . $self->_sqlcase($direction);
1560 push @ret, [ $sql, @bind];
1569 #======================================================================
1570 # DATASOURCE (FOR NOW, JUST PLAIN TABLE OR LIST OF TABLES)
1571 #======================================================================
1576 $self->_SWITCH_refkind($from, {
1577 ARRAYREF => sub {join ', ', map { $self->_quote($_) } @$from;},
1578 SCALAR => sub {$self->_quote($from)},
1579 SCALARREF => sub {$$from},
1584 #======================================================================
1586 #======================================================================
1588 # highly optimized, as it's called way too often
1590 # my ($self, $label) = @_;
1592 return '' unless defined $_[1];
1593 return ${$_[1]} if ref($_[1]) eq 'SCALAR';
1595 $_[0]->{quote_char} or
1596 ($_[0]->_assert_pass_injection_guard($_[1]), return $_[1]);
1598 my $qref = ref $_[0]->{quote_char};
1600 !$qref ? ($_[0]->{quote_char}, $_[0]->{quote_char})
1601 : ($qref eq 'ARRAY') ? @{$_[0]->{quote_char}}
1602 : puke "Unsupported quote_char format: $_[0]->{quote_char}";
1604 my $esc = $_[0]->{escape_char} || $r;
1606 # parts containing * are naturally unquoted
1607 return join($_[0]->{name_sep}||'', map
1608 +( $_ eq '*' ? $_ : do { (my $n = $_) =~ s/(\Q$esc\E|\Q$r\E)/$esc$1/g; $l . $n . $r } ),
1609 ( $_[0]->{name_sep} ? split (/\Q$_[0]->{name_sep}\E/, $_[1] ) : $_[1] )
1614 # Conversion, if applicable
1616 #my ($self, $arg) = @_;
1617 if ($_[0]->{convert}) {
1618 return $_[0]->_sqlcase($_[0]->{convert}) .'(' . $_[1] . ')';
1625 #my ($self, $col, @vals) = @_;
1626 # called often - tighten code
1627 return $_[0]->{bindtype} eq 'columns'
1628 ? map {[$_[1], $_]} @_[2 .. $#_]
1633 # Dies if any element of @bind is not in [colname => value] format
1634 # if bindtype is 'columns'.
1635 sub _assert_bindval_matches_bindtype {
1636 # my ($self, @bind) = @_;
1638 if ($self->{bindtype} eq 'columns') {
1640 if (!defined $_ || ref($_) ne 'ARRAY' || @$_ != 2) {
1641 puke "bindtype 'columns' selected, you need to pass: [column_name => bind_value]"
1647 sub _join_sql_clauses {
1648 my ($self, $logic, $clauses_aref, $bind_aref) = @_;
1650 if (@$clauses_aref > 1) {
1651 my $join = " " . $self->_sqlcase($logic) . " ";
1652 my $sql = '( ' . join($join, @$clauses_aref) . ' )';
1653 return ($sql, @$bind_aref);
1655 elsif (@$clauses_aref) {
1656 return ($clauses_aref->[0], @$bind_aref); # no parentheses
1659 return (); # if no SQL, ignore @$bind_aref
1664 # Fix SQL case, if so requested
1666 # LDNOTE: if $self->{case} is true, then it contains 'lower', so we
1667 # don't touch the argument ... crooked logic, but let's not change it!
1668 return $_[0]->{case} ? $_[1] : uc($_[1]);
1672 #======================================================================
1673 # DISPATCHING FROM REFKIND
1674 #======================================================================
1677 my ($self, $data) = @_;
1679 return 'UNDEF' unless defined $data;
1681 # blessed objects are treated like scalars
1682 my $ref = (Scalar::Util::blessed $data) ? '' : ref $data;
1684 return 'SCALAR' unless $ref;
1687 while ($ref eq 'REF') {
1689 $ref = (Scalar::Util::blessed $data) ? '' : ref $data;
1693 return ($ref||'SCALAR') . ('REF' x $n_steps);
1697 my ($self, $data) = @_;
1698 my @try = ($self->_refkind($data));
1699 push @try, 'SCALAR_or_UNDEF' if $try[0] eq 'SCALAR' || $try[0] eq 'UNDEF';
1700 push @try, 'FALLBACK';
1704 sub _METHOD_FOR_refkind {
1705 my ($self, $meth_prefix, $data) = @_;
1708 for (@{$self->_try_refkind($data)}) {
1709 $method = $self->can($meth_prefix."_".$_)
1713 return $method || puke "cannot dispatch on '$meth_prefix' for ".$self->_refkind($data);
1717 sub _SWITCH_refkind {
1718 my ($self, $data, $dispatch_table) = @_;
1721 for (@{$self->_try_refkind($data)}) {
1722 $coderef = $dispatch_table->{$_}
1726 puke "no dispatch entry for ".$self->_refkind($data)
1735 #======================================================================
1736 # VALUES, GENERATE, AUTOLOAD
1737 #======================================================================
1739 # LDNOTE: original code from nwiger, didn't touch code in that section
1740 # I feel the AUTOLOAD stuff should not be the default, it should
1741 # only be activated on explicit demand by user.
1745 my $data = shift || return;
1746 puke "Argument to ", __PACKAGE__, "->values must be a \\%hash"
1747 unless ref $data eq 'HASH';
1750 foreach my $k (sort keys %$data) {
1751 my $v = $data->{$k};
1752 $self->_SWITCH_refkind($v, {
1754 if ($self->{array_datatypes}) { # array datatype
1755 push @all_bind, $self->_bindtype($k, $v);
1757 else { # literal SQL with bind
1758 my ($sql, @bind) = @$v;
1759 $self->_assert_bindval_matches_bindtype(@bind);
1760 push @all_bind, @bind;
1763 ARRAYREFREF => sub { # literal SQL with bind
1764 my ($sql, @bind) = @${$v};
1765 $self->_assert_bindval_matches_bindtype(@bind);
1766 push @all_bind, @bind;
1768 SCALARREF => sub { # literal SQL without bind
1770 SCALAR_or_UNDEF => sub {
1771 push @all_bind, $self->_bindtype($k, $v);
1782 my(@sql, @sqlq, @sqlv);
1786 if ($ref eq 'HASH') {
1787 for my $k (sort keys %$_) {
1790 my $label = $self->_quote($k);
1791 if ($r eq 'ARRAY') {
1792 # literal SQL with bind
1793 my ($sql, @bind) = @$v;
1794 $self->_assert_bindval_matches_bindtype(@bind);
1795 push @sqlq, "$label = $sql";
1797 } elsif ($r eq 'SCALAR') {
1798 # literal SQL without bind
1799 push @sqlq, "$label = $$v";
1801 push @sqlq, "$label = ?";
1802 push @sqlv, $self->_bindtype($k, $v);
1805 push @sql, $self->_sqlcase('set'), join ', ', @sqlq;
1806 } elsif ($ref eq 'ARRAY') {
1807 # unlike insert(), assume these are ONLY the column names, i.e. for SQL
1810 if ($r eq 'ARRAY') { # literal SQL with bind
1811 my ($sql, @bind) = @$v;
1812 $self->_assert_bindval_matches_bindtype(@bind);
1815 } elsif ($r eq 'SCALAR') { # literal SQL without bind
1816 # embedded literal SQL
1823 push @sql, '(' . join(', ', @sqlq) . ')';
1824 } elsif ($ref eq 'SCALAR') {
1828 # strings get case twiddled
1829 push @sql, $self->_sqlcase($_);
1833 my $sql = join ' ', @sql;
1835 # this is pretty tricky
1836 # if ask for an array, return ($stmt, @bind)
1837 # otherwise, s/?/shift @sqlv/ to put it inline
1839 return ($sql, @sqlv);
1841 1 while $sql =~ s/\?/my $d = shift(@sqlv);
1842 ref $d ? $d->[1] : $d/e;
1851 # This allows us to check for a local, then _form, attr
1853 my($name) = $AUTOLOAD =~ /.*::(.+)/;
1854 return $self->generate($name, @_);
1865 SQL::Abstract - Generate SQL from Perl data structures
1871 my $sql = SQL::Abstract->new;
1873 my($stmt, @bind) = $sql->select($source, \@fields, \%where, $order);
1875 my($stmt, @bind) = $sql->insert($table, \%fieldvals || \@values);
1877 my($stmt, @bind) = $sql->update($table, \%fieldvals, \%where);
1879 my($stmt, @bind) = $sql->delete($table, \%where);
1881 # Then, use these in your DBI statements
1882 my $sth = $dbh->prepare($stmt);
1883 $sth->execute(@bind);
1885 # Just generate the WHERE clause
1886 my($stmt, @bind) = $sql->where(\%where, $order);
1888 # Return values in the same order, for hashed queries
1889 # See PERFORMANCE section for more details
1890 my @bind = $sql->values(\%fieldvals);
1894 This module was inspired by the excellent L<DBIx::Abstract>.
1895 However, in using that module I found that what I really wanted
1896 to do was generate SQL, but still retain complete control over my
1897 statement handles and use the DBI interface. So, I set out to
1898 create an abstract SQL generation module.
1900 While based on the concepts used by L<DBIx::Abstract>, there are
1901 several important differences, especially when it comes to WHERE
1902 clauses. I have modified the concepts used to make the SQL easier
1903 to generate from Perl data structures and, IMO, more intuitive.
1904 The underlying idea is for this module to do what you mean, based
1905 on the data structures you provide it. The big advantage is that
1906 you don't have to modify your code every time your data changes,
1907 as this module figures it out.
1909 To begin with, an SQL INSERT is as easy as just specifying a hash
1910 of C<key=value> pairs:
1913 name => 'Jimbo Bobson',
1914 phone => '123-456-7890',
1915 address => '42 Sister Lane',
1916 city => 'St. Louis',
1917 state => 'Louisiana',
1920 The SQL can then be generated with this:
1922 my($stmt, @bind) = $sql->insert('people', \%data);
1924 Which would give you something like this:
1926 $stmt = "INSERT INTO people
1927 (address, city, name, phone, state)
1928 VALUES (?, ?, ?, ?, ?)";
1929 @bind = ('42 Sister Lane', 'St. Louis', 'Jimbo Bobson',
1930 '123-456-7890', 'Louisiana');
1932 These are then used directly in your DBI code:
1934 my $sth = $dbh->prepare($stmt);
1935 $sth->execute(@bind);
1937 =head2 Inserting and Updating Arrays
1939 If your database has array types (like for example Postgres),
1940 activate the special option C<< array_datatypes => 1 >>
1941 when creating the C<SQL::Abstract> object.
1942 Then you may use an arrayref to insert and update database array types:
1944 my $sql = SQL::Abstract->new(array_datatypes => 1);
1946 planets => [qw/Mercury Venus Earth Mars/]
1949 my($stmt, @bind) = $sql->insert('solar_system', \%data);
1953 $stmt = "INSERT INTO solar_system (planets) VALUES (?)"
1955 @bind = (['Mercury', 'Venus', 'Earth', 'Mars']);
1958 =head2 Inserting and Updating SQL
1960 In order to apply SQL functions to elements of your C<%data> you may
1961 specify a reference to an arrayref for the given hash value. For example,
1962 if you need to execute the Oracle C<to_date> function on a value, you can
1963 say something like this:
1967 date_entered => \[ "to_date(?,'MM/DD/YYYY')", "03/02/2003" ],
1970 The first value in the array is the actual SQL. Any other values are
1971 optional and would be included in the bind values array. This gives
1974 my($stmt, @bind) = $sql->insert('people', \%data);
1976 $stmt = "INSERT INTO people (name, date_entered)
1977 VALUES (?, to_date(?,'MM/DD/YYYY'))";
1978 @bind = ('Bill', '03/02/2003');
1980 An UPDATE is just as easy, all you change is the name of the function:
1982 my($stmt, @bind) = $sql->update('people', \%data);
1984 Notice that your C<%data> isn't touched; the module will generate
1985 the appropriately quirky SQL for you automatically. Usually you'll
1986 want to specify a WHERE clause for your UPDATE, though, which is
1987 where handling C<%where> hashes comes in handy...
1989 =head2 Complex where statements
1991 This module can generate pretty complicated WHERE statements
1992 easily. For example, simple C<key=value> pairs are taken to mean
1993 equality, and if you want to see if a field is within a set
1994 of values, you can use an arrayref. Let's say we wanted to
1995 SELECT some data based on this criteria:
1998 requestor => 'inna',
1999 worker => ['nwiger', 'rcwe', 'sfz'],
2000 status => { '!=', 'completed' }
2003 my($stmt, @bind) = $sql->select('tickets', '*', \%where);
2005 The above would give you something like this:
2007 $stmt = "SELECT * FROM tickets WHERE
2008 ( requestor = ? ) AND ( status != ? )
2009 AND ( worker = ? OR worker = ? OR worker = ? )";
2010 @bind = ('inna', 'completed', 'nwiger', 'rcwe', 'sfz');
2012 Which you could then use in DBI code like so:
2014 my $sth = $dbh->prepare($stmt);
2015 $sth->execute(@bind);
2021 The methods are simple. There's one for every major SQL operation,
2022 and a constructor you use first. The arguments are specified in a
2023 similar order for each method (table, then fields, then a where
2024 clause) to try and simplify things.
2026 =head2 new(option => 'value')
2028 The C<new()> function takes a list of options and values, and returns
2029 a new B<SQL::Abstract> object which can then be used to generate SQL
2030 through the methods below. The options accepted are:
2036 If set to 'lower', then SQL will be generated in all lowercase. By
2037 default SQL is generated in "textbook" case meaning something like:
2039 SELECT a_field FROM a_table WHERE some_field LIKE '%someval%'
2041 Any setting other than 'lower' is ignored.
2045 This determines what the default comparison operator is. By default
2046 it is C<=>, meaning that a hash like this:
2048 %where = (name => 'nwiger', email => 'nate@wiger.org');
2050 Will generate SQL like this:
2052 WHERE name = 'nwiger' AND email = 'nate@wiger.org'
2054 However, you may want loose comparisons by default, so if you set
2055 C<cmp> to C<like> you would get SQL such as:
2057 WHERE name like 'nwiger' AND email like 'nate@wiger.org'
2059 You can also override the comparison on an individual basis - see
2060 the huge section on L</"WHERE CLAUSES"> at the bottom.
2062 =item sqltrue, sqlfalse
2064 Expressions for inserting boolean values within SQL statements.
2065 By default these are C<1=1> and C<1=0>. They are used
2066 by the special operators C<-in> and C<-not_in> for generating
2067 correct SQL even when the argument is an empty array (see below).
2071 This determines the default logical operator for multiple WHERE
2072 statements in arrays or hashes. If absent, the default logic is "or"
2073 for arrays, and "and" for hashes. This means that a WHERE
2077 event_date => {'>=', '2/13/99'},
2078 event_date => {'<=', '4/24/03'},
2081 will generate SQL like this:
2083 WHERE event_date >= '2/13/99' OR event_date <= '4/24/03'
2085 This is probably not what you want given this query, though (look
2086 at the dates). To change the "OR" to an "AND", simply specify:
2088 my $sql = SQL::Abstract->new(logic => 'and');
2090 Which will change the above C<WHERE> to:
2092 WHERE event_date >= '2/13/99' AND event_date <= '4/24/03'
2094 The logic can also be changed locally by inserting
2095 a modifier in front of an arrayref:
2097 @where = (-and => [event_date => {'>=', '2/13/99'},
2098 event_date => {'<=', '4/24/03'} ]);
2100 See the L</"WHERE CLAUSES"> section for explanations.
2104 This will automatically convert comparisons using the specified SQL
2105 function for both column and value. This is mostly used with an argument
2106 of C<upper> or C<lower>, so that the SQL will have the effect of
2107 case-insensitive "searches". For example, this:
2109 $sql = SQL::Abstract->new(convert => 'upper');
2110 %where = (keywords => 'MaKe iT CAse inSeNSItive');
2112 Will turn out the following SQL:
2114 WHERE upper(keywords) like upper('MaKe iT CAse inSeNSItive')
2116 The conversion can be C<upper()>, C<lower()>, or any other SQL function
2117 that can be applied symmetrically to fields (actually B<SQL::Abstract> does
2118 not validate this option; it will just pass through what you specify verbatim).
2122 This is a kludge because many databases suck. For example, you can't
2123 just bind values using DBI's C<execute()> for Oracle C<CLOB> or C<BLOB> fields.
2124 Instead, you have to use C<bind_param()>:
2126 $sth->bind_param(1, 'reg data');
2127 $sth->bind_param(2, $lots, {ora_type => ORA_CLOB});
2129 The problem is, B<SQL::Abstract> will normally just return a C<@bind> array,
2130 which loses track of which field each slot refers to. Fear not.
2132 If you specify C<bindtype> in new, you can determine how C<@bind> is returned.
2133 Currently, you can specify either C<normal> (default) or C<columns>. If you
2134 specify C<columns>, you will get an array that looks like this:
2136 my $sql = SQL::Abstract->new(bindtype => 'columns');
2137 my($stmt, @bind) = $sql->insert(...);
2140 [ 'column1', 'value1' ],
2141 [ 'column2', 'value2' ],
2142 [ 'column3', 'value3' ],
2145 You can then iterate through this manually, using DBI's C<bind_param()>.
2147 $sth->prepare($stmt);
2150 my($col, $data) = @$_;
2151 if ($col eq 'details' || $col eq 'comments') {
2152 $sth->bind_param($i, $data, {ora_type => ORA_CLOB});
2153 } elsif ($col eq 'image') {
2154 $sth->bind_param($i, $data, {ora_type => ORA_BLOB});
2156 $sth->bind_param($i, $data);
2160 $sth->execute; # execute without @bind now
2162 Now, why would you still use B<SQL::Abstract> if you have to do this crap?
2163 Basically, the advantage is still that you don't have to care which fields
2164 are or are not included. You could wrap that above C<for> loop in a simple
2165 sub called C<bind_fields()> or something and reuse it repeatedly. You still
2166 get a layer of abstraction over manual SQL specification.
2168 Note that if you set L</bindtype> to C<columns>, the C<\[ $sql, @bind ]>
2169 construct (see L</Literal SQL with placeholders and bind values (subqueries)>)
2170 will expect the bind values in this format.
2174 This is the character that a table or column name will be quoted
2175 with. By default this is an empty string, but you could set it to
2176 the character C<`>, to generate SQL like this:
2178 SELECT `a_field` FROM `a_table` WHERE `some_field` LIKE '%someval%'
2180 Alternatively, you can supply an array ref of two items, the first being the left
2181 hand quote character, and the second the right hand quote character. For
2182 example, you could supply C<['[',']']> for SQL Server 2000 compliant quotes
2183 that generates SQL like this:
2185 SELECT [a_field] FROM [a_table] WHERE [some_field] LIKE '%someval%'
2187 Quoting is useful if you have tables or columns names that are reserved
2188 words in your database's SQL dialect.
2192 This is the character that will be used to escape L</quote_char>s appearing
2193 in an identifier before it has been quoted.
2195 The parameter default in case of a single L</quote_char> character is the quote
2198 When opening-closing-style quoting is used (L</quote_char> is an arrayref)
2199 this parameter defaults to the B<closing (right)> L</quote_char>. Occurrences
2200 of the B<opening (left)> L</quote_char> within the identifier are currently left
2201 untouched. The default for opening-closing-style quotes may change in future
2202 versions, thus you are B<strongly encouraged> to specify the escape character
2207 This is the character that separates a table and column name. It is
2208 necessary to specify this when the C<quote_char> option is selected,
2209 so that tables and column names can be individually quoted like this:
2211 SELECT `table`.`one_field` FROM `table` WHERE `table`.`other_field` = 1
2213 =item injection_guard
2215 A regular expression C<qr/.../> that is applied to any C<-function> and unquoted
2216 column name specified in a query structure. This is a safety mechanism to avoid
2217 injection attacks when mishandling user input e.g.:
2219 my %condition_as_column_value_pairs = get_values_from_user();
2220 $sqla->select( ... , \%condition_as_column_value_pairs );
2222 If the expression matches an exception is thrown. Note that literal SQL
2223 supplied via C<\'...'> or C<\['...']> is B<not> checked in any way.
2225 Defaults to checking for C<;> and the C<GO> keyword (TransactSQL)
2227 =item array_datatypes
2229 When this option is true, arrayrefs in INSERT or UPDATE are
2230 interpreted as array datatypes and are passed directly
2232 When this option is false, arrayrefs are interpreted
2233 as literal SQL, just like refs to arrayrefs
2234 (but this behavior is for backwards compatibility; when writing
2235 new queries, use the "reference to arrayref" syntax
2241 Takes a reference to a list of "special operators"
2242 to extend the syntax understood by L<SQL::Abstract>.
2243 See section L</"SPECIAL OPERATORS"> for details.
2247 Takes a reference to a list of "unary operators"
2248 to extend the syntax understood by L<SQL::Abstract>.
2249 See section L</"UNARY OPERATORS"> for details.
2255 =head2 insert($table, \@values || \%fieldvals, \%options)
2257 This is the simplest function. You simply give it a table name
2258 and either an arrayref of values or hashref of field/value pairs.
2259 It returns an SQL INSERT statement and a list of bind values.
2260 See the sections on L</"Inserting and Updating Arrays"> and
2261 L</"Inserting and Updating SQL"> for information on how to insert
2262 with those data types.
2264 The optional C<\%options> hash reference may contain additional
2265 options to generate the insert SQL. Currently supported options
2272 Takes either a scalar of raw SQL fields, or an array reference of
2273 field names, and adds on an SQL C<RETURNING> statement at the end.
2274 This allows you to return data generated by the insert statement
2275 (such as row IDs) without performing another C<SELECT> statement.
2276 Note, however, this is not part of the SQL standard and may not
2277 be supported by all database engines.
2281 =head2 update($table, \%fieldvals, \%where, \%options)
2283 This takes a table, hashref of field/value pairs, and an optional
2284 hashref L<WHERE clause|/WHERE CLAUSES>. It returns an SQL UPDATE function and a list
2286 See the sections on L</"Inserting and Updating Arrays"> and
2287 L</"Inserting and Updating SQL"> for information on how to insert
2288 with those data types.
2290 The optional C<\%options> hash reference may contain additional
2291 options to generate the update SQL. Currently supported options
2298 See the C<returning> option to
2299 L<insert|/insert($table, \@values || \%fieldvals, \%options)>.
2303 =head2 select($source, $fields, $where, $order)
2305 This returns a SQL SELECT statement and associated list of bind values, as
2306 specified by the arguments:
2312 Specification of the 'FROM' part of the statement.
2313 The argument can be either a plain scalar (interpreted as a table
2314 name, will be quoted), or an arrayref (interpreted as a list
2315 of table names, joined by commas, quoted), or a scalarref
2316 (literal SQL, not quoted).
2320 Specification of the list of fields to retrieve from
2322 The argument can be either an arrayref (interpreted as a list
2323 of field names, will be joined by commas and quoted), or a
2324 plain scalar (literal SQL, not quoted).
2325 Please observe that this API is not as flexible as that of
2326 the first argument C<$source>, for backwards compatibility reasons.
2330 Optional argument to specify the WHERE part of the query.
2331 The argument is most often a hashref, but can also be
2332 an arrayref or plain scalar --
2333 see section L<WHERE clause|/"WHERE CLAUSES"> for details.
2337 Optional argument to specify the ORDER BY part of the query.
2338 The argument can be a scalar, a hashref or an arrayref
2339 -- see section L<ORDER BY clause|/"ORDER BY CLAUSES">
2345 =head2 delete($table, \%where, \%options)
2347 This takes a table name and optional hashref L<WHERE clause|/WHERE CLAUSES>.
2348 It returns an SQL DELETE statement and list of bind values.
2350 The optional C<\%options> hash reference may contain additional
2351 options to generate the delete SQL. Currently supported options
2358 See the C<returning> option to
2359 L<insert|/insert($table, \@values || \%fieldvals, \%options)>.
2363 =head2 where(\%where, $order)
2365 This is used to generate just the WHERE clause. For example,
2366 if you have an arbitrary data structure and know what the
2367 rest of your SQL is going to look like, but want an easy way
2368 to produce a WHERE clause, use this. It returns an SQL WHERE
2369 clause and list of bind values.
2372 =head2 values(\%data)
2374 This just returns the values from the hash C<%data>, in the same
2375 order that would be returned from any of the other above queries.
2376 Using this allows you to markedly speed up your queries if you
2377 are affecting lots of rows. See below under the L</"PERFORMANCE"> section.
2379 =head2 generate($any, 'number', $of, \@data, $struct, \%types)
2381 Warning: This is an experimental method and subject to change.
2383 This returns arbitrarily generated SQL. It's a really basic shortcut.
2384 It will return two different things, depending on return context:
2386 my($stmt, @bind) = $sql->generate('create table', \$table, \@fields);
2387 my $stmt_and_val = $sql->generate('create table', \$table, \@fields);
2389 These would return the following:
2391 # First calling form
2392 $stmt = "CREATE TABLE test (?, ?)";
2393 @bind = (field1, field2);
2395 # Second calling form
2396 $stmt_and_val = "CREATE TABLE test (field1, field2)";
2398 Depending on what you're trying to do, it's up to you to choose the correct
2399 format. In this example, the second form is what you would want.
2403 $sql->generate('alter session', { nls_date_format => 'MM/YY' });
2407 ALTER SESSION SET nls_date_format = 'MM/YY'
2409 You get the idea. Strings get their case twiddled, but everything
2410 else remains verbatim.
2412 =head1 EXPORTABLE FUNCTIONS
2414 =head2 is_plain_value
2416 Determines if the supplied argument is a plain value as understood by this
2421 =item * The value is C<undef>
2423 =item * The value is a non-reference
2425 =item * The value is an object with stringification overloading
2427 =item * The value is of the form C<< { -value => $anything } >>
2431 On failure returns C<undef>, on success returns a B<scalar> reference
2432 to the original supplied argument.
2438 The stringification overloading detection is rather advanced: it takes
2439 into consideration not only the presence of a C<""> overload, but if that
2440 fails also checks for enabled
2441 L<autogenerated versions of C<"">|overload/Magic Autogeneration>, based
2442 on either C<0+> or C<bool>.
2444 Unfortunately testing in the field indicates that this
2445 detection B<< may tickle a latent bug in perl versions before 5.018 >>,
2446 but only when very large numbers of stringifying objects are involved.
2447 At the time of writing ( Sep 2014 ) there is no clear explanation of
2448 the direct cause, nor is there a manageably small test case that reliably
2449 reproduces the problem.
2451 If you encounter any of the following exceptions in B<random places within
2452 your application stack> - this module may be to blame:
2454 Operation "ne": no method found,
2455 left argument in overloaded package <something>,
2456 right argument in overloaded package <something>
2460 Stub found while resolving method "???" overloading """" in package <something>
2462 If you fall victim to the above - please attempt to reduce the problem
2463 to something that could be sent to the L<SQL::Abstract developers
2464 |DBIx::Class/GETTING HELP/SUPPORT>
2465 (either publicly or privately). As a workaround in the meantime you can
2466 set C<$ENV{SQLA_ISVALUE_IGNORE_AUTOGENERATED_STRINGIFICATION}> to a true
2467 value, which will most likely eliminate your problem (at the expense of
2468 not being able to properly detect exotic forms of stringification).
2470 This notice and environment variable will be removed in a future version,
2471 as soon as the underlying problem is found and a reliable workaround is
2476 =head2 is_literal_value
2478 Determines if the supplied argument is a literal value as understood by this
2483 =item * C<\$sql_string>
2485 =item * C<\[ $sql_string, @bind_values ]>
2489 On failure returns C<undef>, on success returns an B<array> reference
2490 containing the unpacked version of the supplied literal SQL and bind values.
2492 =head1 WHERE CLAUSES
2496 This module uses a variation on the idea from L<DBIx::Abstract>. It
2497 is B<NOT>, repeat I<not> 100% compatible. B<The main logic of this
2498 module is that things in arrays are OR'ed, and things in hashes
2501 The easiest way to explain is to show lots of examples. After
2502 each C<%where> hash shown, it is assumed you used:
2504 my($stmt, @bind) = $sql->where(\%where);
2506 However, note that the C<%where> hash can be used directly in any
2507 of the other functions as well, as described above.
2509 =head2 Key-value pairs
2511 So, let's get started. To begin, a simple hash:
2515 status => 'completed'
2518 Is converted to SQL C<key = val> statements:
2520 $stmt = "WHERE user = ? AND status = ?";
2521 @bind = ('nwiger', 'completed');
2523 One common thing I end up doing is having a list of values that
2524 a field can be in. To do this, simply specify a list inside of
2529 status => ['assigned', 'in-progress', 'pending'];
2532 This simple code will create the following:
2534 $stmt = "WHERE user = ? AND ( status = ? OR status = ? OR status = ? )";
2535 @bind = ('nwiger', 'assigned', 'in-progress', 'pending');
2537 A field associated to an empty arrayref will be considered a
2538 logical false and will generate 0=1.
2540 =head2 Tests for NULL values
2542 If the value part is C<undef> then this is converted to SQL <IS NULL>
2551 $stmt = "WHERE user = ? AND status IS NULL";
2554 To test if a column IS NOT NULL:
2558 status => { '!=', undef },
2561 =head2 Specific comparison operators
2563 If you want to specify a different type of operator for your comparison,
2564 you can use a hashref for a given column:
2568 status => { '!=', 'completed' }
2571 Which would generate:
2573 $stmt = "WHERE user = ? AND status != ?";
2574 @bind = ('nwiger', 'completed');
2576 To test against multiple values, just enclose the values in an arrayref:
2578 status => { '=', ['assigned', 'in-progress', 'pending'] };
2580 Which would give you:
2582 "WHERE status = ? OR status = ? OR status = ?"
2585 The hashref can also contain multiple pairs, in which case it is expanded
2586 into an C<AND> of its elements:
2590 status => { '!=', 'completed', -not_like => 'pending%' }
2593 # Or more dynamically, like from a form
2594 $where{user} = 'nwiger';
2595 $where{status}{'!='} = 'completed';
2596 $where{status}{'-not_like'} = 'pending%';
2598 # Both generate this
2599 $stmt = "WHERE user = ? AND status != ? AND status NOT LIKE ?";
2600 @bind = ('nwiger', 'completed', 'pending%');
2603 To get an OR instead, you can combine it with the arrayref idea:
2607 priority => [ { '=', 2 }, { '>', 5 } ]
2610 Which would generate:
2612 $stmt = "WHERE ( priority = ? OR priority > ? ) AND user = ?";
2613 @bind = ('2', '5', 'nwiger');
2615 If you want to include literal SQL (with or without bind values), just use a
2616 scalar reference or reference to an arrayref as the value:
2619 date_entered => { '>' => \["to_date(?, 'MM/DD/YYYY')", "11/26/2008"] },
2620 date_expires => { '<' => \"now()" }
2623 Which would generate:
2625 $stmt = "WHERE date_entered > to_date(?, 'MM/DD/YYYY') AND date_expires < now()";
2626 @bind = ('11/26/2008');
2629 =head2 Logic and nesting operators
2631 In the example above,
2632 there is a subtle trap if you want to say something like
2633 this (notice the C<AND>):
2635 WHERE priority != ? AND priority != ?
2637 Because, in Perl you I<can't> do this:
2639 priority => { '!=' => 2, '!=' => 1 }
2641 As the second C<!=> key will obliterate the first. The solution
2642 is to use the special C<-modifier> form inside an arrayref:
2644 priority => [ -and => {'!=', 2},
2648 Normally, these would be joined by C<OR>, but the modifier tells it
2649 to use C<AND> instead. (Hint: You can use this in conjunction with the
2650 C<logic> option to C<new()> in order to change the way your queries
2651 work by default.) B<Important:> Note that the C<-modifier> goes
2652 B<INSIDE> the arrayref, as an extra first element. This will
2653 B<NOT> do what you think it might:
2655 priority => -and => [{'!=', 2}, {'!=', 1}] # WRONG!
2657 Here is a quick list of equivalencies, since there is some overlap:
2660 status => {'!=', 'completed', 'not like', 'pending%' }
2661 status => [ -and => {'!=', 'completed'}, {'not like', 'pending%'}]
2664 status => {'=', ['assigned', 'in-progress']}
2665 status => [ -or => {'=', 'assigned'}, {'=', 'in-progress'}]
2666 status => [ {'=', 'assigned'}, {'=', 'in-progress'} ]
2670 =head2 Special operators: IN, BETWEEN, etc.
2672 You can also use the hashref format to compare a list of fields using the
2673 C<IN> comparison operator, by specifying the list as an arrayref:
2676 status => 'completed',
2677 reportid => { -in => [567, 2335, 2] }
2680 Which would generate:
2682 $stmt = "WHERE status = ? AND reportid IN (?,?,?)";
2683 @bind = ('completed', '567', '2335', '2');
2685 The reverse operator C<-not_in> generates SQL C<NOT IN> and is used in
2688 If the argument to C<-in> is an empty array, 'sqlfalse' is generated
2689 (by default: C<1=0>). Similarly, C<< -not_in => [] >> generates
2690 'sqltrue' (by default: C<1=1>).
2692 In addition to the array you can supply a chunk of literal sql or
2693 literal sql with bind:
2696 customer => { -in => \[
2697 'SELECT cust_id FROM cust WHERE balance > ?',
2700 status => { -in => \'SELECT status_codes FROM states' },
2706 customer IN ( SELECT cust_id FROM cust WHERE balance > ? )
2707 AND status IN ( SELECT status_codes FROM states )
2711 Finally, if the argument to C<-in> is not a reference, it will be
2712 treated as a single-element array.
2714 Another pair of operators is C<-between> and C<-not_between>,
2715 used with an arrayref of two values:
2719 completion_date => {
2720 -not_between => ['2002-10-01', '2003-02-06']
2726 WHERE user = ? AND completion_date NOT BETWEEN ( ? AND ? )
2728 Just like with C<-in> all plausible combinations of literal SQL
2732 start0 => { -between => [ 1, 2 ] },
2733 start1 => { -between => \["? AND ?", 1, 2] },
2734 start2 => { -between => \"lower(x) AND upper(y)" },
2735 start3 => { -between => [
2737 \["upper(?)", 'stuff' ],
2744 ( start0 BETWEEN ? AND ? )
2745 AND ( start1 BETWEEN ? AND ? )
2746 AND ( start2 BETWEEN lower(x) AND upper(y) )
2747 AND ( start3 BETWEEN lower(x) AND upper(?) )
2749 @bind = (1, 2, 1, 2, 'stuff');
2752 These are the two builtin "special operators"; but the
2753 list can be expanded: see section L</"SPECIAL OPERATORS"> below.
2755 =head2 Unary operators: bool
2757 If you wish to test against boolean columns or functions within your
2758 database you can use the C<-bool> and C<-not_bool> operators. For
2759 example to test the column C<is_user> being true and the column
2760 C<is_enabled> being false you would use:-
2764 -not_bool => 'is_enabled',
2769 WHERE is_user AND NOT is_enabled
2771 If a more complex combination is required, testing more conditions,
2772 then you should use the and/or operators:-
2777 -not_bool => { two=> { -rlike => 'bar' } },
2778 -not_bool => { three => [ { '=', 2 }, { '>', 5 } ] },
2789 (NOT ( three = ? OR three > ? ))
2792 =head2 Nested conditions, -and/-or prefixes
2794 So far, we've seen how multiple conditions are joined with a top-level
2795 C<AND>. We can change this by putting the different conditions we want in
2796 hashes and then putting those hashes in an array. For example:
2801 status => { -like => ['pending%', 'dispatched'] },
2805 status => 'unassigned',
2809 This data structure would create the following:
2811 $stmt = "WHERE ( user = ? AND ( status LIKE ? OR status LIKE ? ) )
2812 OR ( user = ? AND status = ? ) )";
2813 @bind = ('nwiger', 'pending', 'dispatched', 'robot', 'unassigned');
2816 Clauses in hashrefs or arrayrefs can be prefixed with an C<-and> or C<-or>
2817 to change the logic inside:
2823 -and => [ workhrs => {'>', 20}, geo => 'ASIA' ],
2824 -or => { workhrs => {'<', 50}, geo => 'EURO' },
2831 $stmt = "WHERE ( user = ?
2832 AND ( ( workhrs > ? AND geo = ? )
2833 OR ( workhrs < ? OR geo = ? ) ) )";
2834 @bind = ('nwiger', '20', 'ASIA', '50', 'EURO');
2836 =head3 Algebraic inconsistency, for historical reasons
2838 C<Important note>: when connecting several conditions, the C<-and->|C<-or>
2839 operator goes C<outside> of the nested structure; whereas when connecting
2840 several constraints on one column, the C<-and> operator goes
2841 C<inside> the arrayref. Here is an example combining both features:
2844 -and => [a => 1, b => 2],
2845 -or => [c => 3, d => 4],
2846 e => [-and => {-like => 'foo%'}, {-like => '%bar'} ]
2851 WHERE ( ( ( a = ? AND b = ? )
2852 OR ( c = ? OR d = ? )
2853 OR ( e LIKE ? AND e LIKE ? ) ) )
2855 This difference in syntax is unfortunate but must be preserved for
2856 historical reasons. So be careful: the two examples below would
2857 seem algebraically equivalent, but they are not
2860 { -like => 'foo%' },
2861 { -like => '%bar' },
2863 # yields: WHERE ( ( col LIKE ? AND col LIKE ? ) )
2866 { col => { -like => 'foo%' } },
2867 { col => { -like => '%bar' } },
2869 # yields: WHERE ( ( col LIKE ? OR col LIKE ? ) )
2872 =head2 Literal SQL and value type operators
2874 The basic premise of SQL::Abstract is that in WHERE specifications the "left
2875 side" is a column name and the "right side" is a value (normally rendered as
2876 a placeholder). This holds true for both hashrefs and arrayref pairs as you
2877 see in the L</WHERE CLAUSES> examples above. Sometimes it is necessary to
2878 alter this behavior. There are several ways of doing so.
2882 This is a virtual operator that signals the string to its right side is an
2883 identifier (a column name) and not a value. For example to compare two
2884 columns you would write:
2887 priority => { '<', 2 },
2888 requestor => { -ident => 'submitter' },
2893 $stmt = "WHERE priority < ? AND requestor = submitter";
2896 If you are maintaining legacy code you may see a different construct as
2897 described in L</Deprecated usage of Literal SQL>, please use C<-ident> in new
2902 This is a virtual operator that signals that the construct to its right side
2903 is a value to be passed to DBI. This is for example necessary when you want
2904 to write a where clause against an array (for RDBMS that support such
2905 datatypes). For example:
2908 array => { -value => [1, 2, 3] }
2913 $stmt = 'WHERE array = ?';
2914 @bind = ([1, 2, 3]);
2916 Note that if you were to simply say:
2922 the result would probably not be what you wanted:
2924 $stmt = 'WHERE array = ? OR array = ? OR array = ?';
2929 Finally, sometimes only literal SQL will do. To include a random snippet
2930 of SQL verbatim, you specify it as a scalar reference. Consider this only
2931 as a last resort. Usually there is a better way. For example:
2934 priority => { '<', 2 },
2935 requestor => { -in => \'(SELECT name FROM hitmen)' },
2940 $stmt = "WHERE priority < ? AND requestor IN (SELECT name FROM hitmen)"
2943 Note that in this example, you only get one bind parameter back, since
2944 the verbatim SQL is passed as part of the statement.
2948 Never use untrusted input as a literal SQL argument - this is a massive
2949 security risk (there is no way to check literal snippets for SQL
2950 injections and other nastyness). If you need to deal with untrusted input
2951 use literal SQL with placeholders as described next.
2953 =head3 Literal SQL with placeholders and bind values (subqueries)
2955 If the literal SQL to be inserted has placeholders and bind values,
2956 use a reference to an arrayref (yes this is a double reference --
2957 not so common, but perfectly legal Perl). For example, to find a date
2958 in Postgres you can use something like this:
2961 date_column => \[ "= date '2008-09-30' - ?::integer", 10 ]
2966 $stmt = "WHERE ( date_column = date '2008-09-30' - ?::integer )"
2969 Note that you must pass the bind values in the same format as they are returned
2970 by L<where|/where(\%where, $order)>. This means that if you set L</bindtype>
2971 to C<columns>, you must provide the bind values in the
2972 C<< [ column_meta => value ] >> format, where C<column_meta> is an opaque
2973 scalar value; most commonly the column name, but you can use any scalar value
2974 (including references and blessed references), L<SQL::Abstract> will simply
2975 pass it through intact. So if C<bindtype> is set to C<columns> the above
2976 example will look like:
2979 date_column => \[ "= date '2008-09-30' - ?::integer", [ {} => 10 ] ]
2982 Literal SQL is especially useful for nesting parenthesized clauses in the
2983 main SQL query. Here is a first example:
2985 my ($sub_stmt, @sub_bind) = ("SELECT c1 FROM t1 WHERE c2 < ? AND c3 LIKE ?",
2989 bar => \["IN ($sub_stmt)" => @sub_bind],
2994 $stmt = "WHERE (foo = ? AND bar IN (SELECT c1 FROM t1
2995 WHERE c2 < ? AND c3 LIKE ?))";
2996 @bind = (1234, 100, "foo%");
2998 Other subquery operators, like for example C<"E<gt> ALL"> or C<"NOT IN">,
2999 are expressed in the same way. Of course the C<$sub_stmt> and
3000 its associated bind values can be generated through a former call
3003 my ($sub_stmt, @sub_bind)
3004 = $sql->select("t1", "c1", {c2 => {"<" => 100},
3005 c3 => {-like => "foo%"}});
3008 bar => \["> ALL ($sub_stmt)" => @sub_bind],
3011 In the examples above, the subquery was used as an operator on a column;
3012 but the same principle also applies for a clause within the main C<%where>
3013 hash, like an EXISTS subquery:
3015 my ($sub_stmt, @sub_bind)
3016 = $sql->select("t1", "*", {c1 => 1, c2 => \"> t0.c0"});
3017 my %where = ( -and => [
3019 \["EXISTS ($sub_stmt)" => @sub_bind],
3024 $stmt = "WHERE (foo = ? AND EXISTS (SELECT * FROM t1
3025 WHERE c1 = ? AND c2 > t0.c0))";
3029 Observe that the condition on C<c2> in the subquery refers to
3030 column C<t0.c0> of the main query: this is I<not> a bind
3031 value, so we have to express it through a scalar ref.
3032 Writing C<< c2 => {">" => "t0.c0"} >> would have generated
3033 C<< c2 > ? >> with bind value C<"t0.c0"> ... not exactly
3034 what we wanted here.
3036 Finally, here is an example where a subquery is used
3037 for expressing unary negation:
3039 my ($sub_stmt, @sub_bind)
3040 = $sql->where({age => [{"<" => 10}, {">" => 20}]});
3041 $sub_stmt =~ s/^ where //i; # don't want "WHERE" in the subclause
3043 lname => {like => '%son%'},
3044 \["NOT ($sub_stmt)" => @sub_bind],
3049 $stmt = "lname LIKE ? AND NOT ( age < ? OR age > ? )"
3050 @bind = ('%son%', 10, 20)
3052 =head3 Deprecated usage of Literal SQL
3054 Below are some examples of archaic use of literal SQL. It is shown only as
3055 reference for those who deal with legacy code. Each example has a much
3056 better, cleaner and safer alternative that users should opt for in new code.
3062 my %where = ( requestor => \'IS NOT NULL' )
3064 $stmt = "WHERE requestor IS NOT NULL"
3066 This used to be the way of generating NULL comparisons, before the handling
3067 of C<undef> got formalized. For new code please use the superior syntax as
3068 described in L</Tests for NULL values>.
3072 my %where = ( requestor => \'= submitter' )
3074 $stmt = "WHERE requestor = submitter"
3076 This used to be the only way to compare columns. Use the superior L</-ident>
3077 method for all new code. For example an identifier declared in such a way
3078 will be properly quoted if L</quote_char> is properly set, while the legacy
3079 form will remain as supplied.
3083 my %where = ( is_ready => \"", completed => { '>', '2012-12-21' } )
3085 $stmt = "WHERE completed > ? AND is_ready"
3086 @bind = ('2012-12-21')
3088 Using an empty string literal used to be the only way to express a boolean.
3089 For all new code please use the much more readable
3090 L<-bool|/Unary operators: bool> operator.
3096 These pages could go on for a while, since the nesting of the data
3097 structures this module can handle are pretty much unlimited (the
3098 module implements the C<WHERE> expansion as a recursive function
3099 internally). Your best bet is to "play around" with the module a
3100 little to see how the data structures behave, and choose the best
3101 format for your data based on that.
3103 And of course, all the values above will probably be replaced with
3104 variables gotten from forms or the command line. After all, if you
3105 knew everything ahead of time, you wouldn't have to worry about
3106 dynamically-generating SQL and could just hardwire it into your
3109 =head1 ORDER BY CLAUSES
3111 Some functions take an order by clause. This can either be a scalar (just a
3112 column name), a hashref of C<< { -desc => 'col' } >> or C<< { -asc => 'col' }
3113 >>, a scalarref, an arrayref-ref, or an arrayref of any of the previous
3116 Given | Will Generate
3117 ---------------------------------------------------------------
3119 'colA' | ORDER BY colA
3121 [qw/colA colB/] | ORDER BY colA, colB
3123 {-asc => 'colA'} | ORDER BY colA ASC
3125 {-desc => 'colB'} | ORDER BY colB DESC
3127 ['colA', {-asc => 'colB'}] | ORDER BY colA, colB ASC
3129 { -asc => [qw/colA colB/] } | ORDER BY colA ASC, colB ASC
3131 \'colA DESC' | ORDER BY colA DESC
3133 \[ 'FUNC(colA, ?)', $x ] | ORDER BY FUNC(colA, ?)
3134 | /* ...with $x bound to ? */
3137 { -asc => 'colA' }, | colA ASC,
3138 { -desc => [qw/colB/] }, | colB DESC,
3139 { -asc => [qw/colC colD/] },| colC ASC, colD ASC,
3140 \'colE DESC', | colE DESC,
3141 \[ 'FUNC(colF, ?)', $x ], | FUNC(colF, ?)
3142 ] | /* ...with $x bound to ? */
3143 ===============================================================
3147 =head1 SPECIAL OPERATORS
3149 my $sqlmaker = SQL::Abstract->new(special_ops => [
3153 my ($self, $field, $op, $arg) = @_;
3159 handler => 'method_name',
3163 A "special operator" is a SQL syntactic clause that can be
3164 applied to a field, instead of a usual binary operator.
3167 WHERE field IN (?, ?, ?)
3168 WHERE field BETWEEN ? AND ?
3169 WHERE MATCH(field) AGAINST (?, ?)
3171 Special operators IN and BETWEEN are fairly standard and therefore
3172 are builtin within C<SQL::Abstract> (as the overridable methods
3173 C<_where_field_IN> and C<_where_field_BETWEEN>). For other operators,
3174 like the MATCH .. AGAINST example above which is specific to MySQL,
3175 you can write your own operator handlers - supply a C<special_ops>
3176 argument to the C<new> method. That argument takes an arrayref of
3177 operator definitions; each operator definition is a hashref with two
3184 the regular expression to match the operator
3188 Either a coderef or a plain scalar method name. In both cases
3189 the expected return is C<< ($sql, @bind) >>.
3191 When supplied with a method name, it is simply called on the
3192 L<SQL::Abstract> object as:
3194 $self->$method_name($field, $op, $arg)
3198 $field is the LHS of the operator
3199 $op is the part that matched the handler regex
3202 When supplied with a coderef, it is called as:
3204 $coderef->($self, $field, $op, $arg)
3209 For example, here is an implementation
3210 of the MATCH .. AGAINST syntax for MySQL
3212 my $sqlmaker = SQL::Abstract->new(special_ops => [
3214 # special op for MySql MATCH (field) AGAINST(word1, word2, ...)
3215 {regex => qr/^match$/i,
3217 my ($self, $field, $op, $arg) = @_;
3218 $arg = [$arg] if not ref $arg;
3219 my $label = $self->_quote($field);
3220 my ($placeholder) = $self->_convert('?');
3221 my $placeholders = join ", ", (($placeholder) x @$arg);
3222 my $sql = $self->_sqlcase('match') . " ($label) "
3223 . $self->_sqlcase('against') . " ($placeholders) ";
3224 my @bind = $self->_bindtype($field, @$arg);
3225 return ($sql, @bind);
3232 =head1 UNARY OPERATORS
3234 my $sqlmaker = SQL::Abstract->new(unary_ops => [
3238 my ($self, $op, $arg) = @_;
3244 handler => 'method_name',
3248 A "unary operator" is a SQL syntactic clause that can be
3249 applied to a field - the operator goes before the field
3251 You can write your own operator handlers - supply a C<unary_ops>
3252 argument to the C<new> method. That argument takes an arrayref of
3253 operator definitions; each operator definition is a hashref with two
3260 the regular expression to match the operator
3264 Either a coderef or a plain scalar method name. In both cases
3265 the expected return is C<< $sql >>.
3267 When supplied with a method name, it is simply called on the
3268 L<SQL::Abstract> object as:
3270 $self->$method_name($op, $arg)
3274 $op is the part that matched the handler regex
3275 $arg is the RHS or argument of the operator
3277 When supplied with a coderef, it is called as:
3279 $coderef->($self, $op, $arg)
3287 Thanks to some benchmarking by Mark Stosberg, it turns out that
3288 this module is many orders of magnitude faster than using C<DBIx::Abstract>.
3289 I must admit this wasn't an intentional design issue, but it's a
3290 byproduct of the fact that you get to control your C<DBI> handles
3293 To maximize performance, use a code snippet like the following:
3295 # prepare a statement handle using the first row
3296 # and then reuse it for the rest of the rows
3298 for my $href (@array_of_hashrefs) {
3299 $stmt ||= $sql->insert('table', $href);
3300 $sth ||= $dbh->prepare($stmt);
3301 $sth->execute($sql->values($href));
3304 The reason this works is because the keys in your C<$href> are sorted
3305 internally by B<SQL::Abstract>. Thus, as long as your data retains
3306 the same structure, you only have to generate the SQL the first time
3307 around. On subsequent queries, simply use the C<values> function provided
3308 by this module to return your values in the correct order.
3310 However this depends on the values having the same type - if, for
3311 example, the values of a where clause may either have values
3312 (resulting in sql of the form C<column = ?> with a single bind
3313 value), or alternatively the values might be C<undef> (resulting in
3314 sql of the form C<column IS NULL> with no bind value) then the
3315 caching technique suggested will not work.
3319 If you use my C<CGI::FormBuilder> module at all, you'll hopefully
3320 really like this part (I do, at least). Building up a complex query
3321 can be as simple as the following:
3328 use CGI::FormBuilder;
3331 my $form = CGI::FormBuilder->new(...);
3332 my $sql = SQL::Abstract->new;
3334 if ($form->submitted) {
3335 my $field = $form->field;
3336 my $id = delete $field->{id};
3337 my($stmt, @bind) = $sql->update('table', $field, {id => $id});
3340 Of course, you would still have to connect using C<DBI> to run the
3341 query, but the point is that if you make your form look like your
3342 table, the actual query script can be extremely simplistic.
3344 If you're B<REALLY> lazy (I am), check out C<HTML::QuickTable> for
3345 a fast interface to returning and formatting data. I frequently
3346 use these three modules together to write complex database query
3347 apps in under 50 lines.
3349 =head1 HOW TO CONTRIBUTE
3351 Contributions are always welcome, in all usable forms (we especially
3352 welcome documentation improvements). The delivery methods include git-
3353 or unified-diff formatted patches, GitHub pull requests, or plain bug
3354 reports either via RT or the Mailing list. Contributors are generally
3355 granted full access to the official repository after their first several
3356 patches pass successful review.
3358 This project is maintained in a git repository. The code and related tools are
3359 accessible at the following locations:
3363 =item * Official repo: L<git://git.shadowcat.co.uk/dbsrgits/SQL-Abstract.git>
3365 =item * Official gitweb: L<http://git.shadowcat.co.uk/gitweb/gitweb.cgi?p=dbsrgits/SQL-Abstract.git>
3367 =item * GitHub mirror: L<https://github.com/dbsrgits/sql-abstract>
3369 =item * Authorized committers: L<ssh://dbsrgits@git.shadowcat.co.uk/SQL-Abstract.git>
3375 Version 1.50 was a major internal refactoring of C<SQL::Abstract>.
3376 Great care has been taken to preserve the I<published> behavior
3377 documented in previous versions in the 1.* family; however,
3378 some features that were previously undocumented, or behaved
3379 differently from the documentation, had to be changed in order
3380 to clarify the semantics. Hence, client code that was relying
3381 on some dark areas of C<SQL::Abstract> v1.*
3382 B<might behave differently> in v1.50.
3384 The main changes are:
3390 support for literal SQL through the C<< \ [ $sql, @bind ] >> syntax.
3394 support for the { operator => \"..." } construct (to embed literal SQL)
3398 support for the { operator => \["...", @bind] } construct (to embed literal SQL with bind values)
3402 optional support for L<array datatypes|/"Inserting and Updating Arrays">
3406 defensive programming: check arguments
3410 fixed bug with global logic, which was previously implemented
3411 through global variables yielding side-effects. Prior versions would
3412 interpret C<< [ {cond1, cond2}, [cond3, cond4] ] >>
3413 as C<< "(cond1 AND cond2) OR (cond3 AND cond4)" >>.
3414 Now this is interpreted
3415 as C<< "(cond1 AND cond2) OR (cond3 OR cond4)" >>.
3420 fixed semantics of _bindtype on array args
3424 dropped the C<_anoncopy> of the %where tree. No longer necessary,
3425 we just avoid shifting arrays within that tree.
3429 dropped the C<_modlogic> function
3433 =head1 ACKNOWLEDGEMENTS
3435 There are a number of individuals that have really helped out with
3436 this module. Unfortunately, most of them submitted bugs via CPAN
3437 so I have no idea who they are! But the people I do know are:
3439 Ash Berlin (order_by hash term support)
3440 Matt Trout (DBIx::Class support)
3441 Mark Stosberg (benchmarking)
3442 Chas Owens (initial "IN" operator support)
3443 Philip Collins (per-field SQL functions)
3444 Eric Kolve (hashref "AND" support)
3445 Mike Fragassi (enhancements to "BETWEEN" and "LIKE")
3446 Dan Kubb (support for "quote_char" and "name_sep")
3447 Guillermo Roditi (patch to cleanup "IN" and "BETWEEN", fix and tests for _order_by)
3448 Laurent Dami (internal refactoring, extensible list of special operators, literal SQL)
3449 Norbert Buchmuller (support for literal SQL in hashpair, misc. fixes & tests)
3450 Peter Rabbitson (rewrite of SQLA::Test, misc. fixes & tests)
3451 Oliver Charles (support for "RETURNING" after "INSERT")
3457 L<DBIx::Class>, L<DBIx::Abstract>, L<CGI::FormBuilder>, L<HTML::QuickTable>.
3461 Copyright (c) 2001-2007 Nathan Wiger <nwiger@cpan.org>. All Rights Reserved.
3463 This module is actively maintained by Matt Trout <mst@shadowcatsystems.co.uk>
3465 For support, your best bet is to try the C<DBIx::Class> users mailing list.
3466 While not an official support venue, C<DBIx::Class> makes heavy use of
3467 C<SQL::Abstract>, and as such list members there are very familiar with
3468 how to create queries.
3472 This module is free software; you may copy this under the same
3473 terms as perl itself (either the GNU General Public License or
3474 the Artistic License)