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) = @_;
1036 $self->_assert_bindval_matches_bindtype(@{$literal}[1..$#$literal]);
1040 sub _where_hashpair_ARRAYREF {
1041 my ($self, $k, $v) = @_;
1044 my @v = @$v; # need copy because of shift below
1045 $self->_debug("ARRAY($k) means distribute over elements");
1047 # put apart first element if it is an operator (-and, -or)
1049 (defined $v[0] && $v[0] =~ /^ - (?: AND|OR ) $/ix)
1053 my @distributed = map { {$k => $_} } @v;
1056 $self->_debug("OP($op) reinjected into the distributed array");
1057 unshift @distributed, $op;
1060 my $logic = $op ? substr($op, 1) : '';
1062 return $self->_recurse_where(\@distributed, $logic);
1065 $self->_debug("empty ARRAY($k) means 0=1");
1066 return ($self->{sqlfalse});
1070 sub _where_hashpair_HASHREF {
1071 my ($self, $k, $v, $logic) = @_;
1074 local $self->{_nested_func_lhs} = defined $self->{_nested_func_lhs}
1075 ? $self->{_nested_func_lhs}
1079 my ($all_sql, @all_bind);
1081 for my $orig_op (sort keys %$v) {
1082 my $val = $v->{$orig_op};
1084 # put the operator in canonical form
1087 # FIXME - we need to phase out dash-less ops
1088 $op =~ s/^-//; # remove possible initial dash
1089 $op =~ s/^\s+|\s+$//g;# remove leading/trailing space
1090 $op =~ s/\s+/ /g; # compress whitespace
1092 $self->_assert_pass_injection_guard($op);
1095 $op =~ s/^is_not/IS NOT/i;
1097 # so that -not_foo works correctly
1098 $op =~ s/^not_/NOT /i;
1100 # another retarded special case: foo => { $op => { -value => undef } }
1101 if (ref $val eq 'HASH' and keys %$val == 1 and exists $val->{-value} and ! defined $val->{-value} ) {
1107 # CASE: col-value logic modifiers
1108 if ($orig_op =~ /^ \- (and|or) $/xi) {
1109 ($sql, @bind) = $self->_where_hashpair_HASHREF($k, $val, $1);
1111 # CASE: special operators like -in or -between
1112 elsif (my $special_op = List::Util::first { $op =~ $_->{regex} } @{$self->{special_ops}}) {
1113 my $handler = $special_op->{handler};
1115 puke "No handler supplied for special operator $orig_op";
1117 elsif (not ref $handler) {
1118 ($sql, @bind) = $self->$handler($k, $op, $val);
1120 elsif (ref $handler eq 'CODE') {
1121 ($sql, @bind) = $handler->($self, $k, $op, $val);
1124 puke "Illegal handler for special operator $orig_op - expecting a method name or a coderef";
1128 $self->_SWITCH_refkind($val, {
1130 ARRAYREF => sub { # CASE: col => {op => \@vals}
1131 ($sql, @bind) = $self->_where_field_op_ARRAYREF($k, $op, $val);
1134 ARRAYREFREF => sub { # CASE: col => {op => \[$sql, @bind]} (literal SQL with bind)
1135 my ($sub_sql, @sub_bind) = @$$val;
1136 $self->_assert_bindval_matches_bindtype(@sub_bind);
1137 $sql = join ' ', $self->_convert($self->_quote($k)),
1138 $self->_sqlcase($op),
1143 UNDEF => sub { # CASE: col => {op => undef} : sql "IS (NOT)? NULL"
1145 $op =~ /^not$/i ? 'is not' # legacy
1146 : $op =~ $self->{equality_op} ? 'is'
1147 : $op =~ $self->{like_op} ? belch("Supplying an undefined argument to '@{[ uc $op]}' is deprecated") && 'is'
1148 : $op =~ $self->{inequality_op} ? 'is not'
1149 : $op =~ $self->{not_like_op} ? belch("Supplying an undefined argument to '@{[ uc $op]}' is deprecated") && 'is not'
1150 : puke "unexpected operator '$orig_op' with undef operand";
1152 $sql = $self->_quote($k) . $self->_sqlcase(" $is null");
1155 FALLBACK => sub { # CASE: col => {op/func => $stuff}
1156 ($sql, @bind) = $self->_where_unary_op($op, $val);
1159 $self->_convert($self->_quote($k)),
1160 $self->{_nested_func_lhs} eq $k ? $sql : "($sql)", # top level vs nested
1166 ($all_sql) = (defined $all_sql and $all_sql) ? $self->_join_sql_clauses($logic, [$all_sql, $sql], []) : $sql;
1167 push @all_bind, @bind;
1169 return ($all_sql, @all_bind);
1172 sub _where_field_IS {
1173 my ($self, $k, $op, $v) = @_;
1175 my ($s) = $self->_SWITCH_refkind($v, {
1178 $self->_convert($self->_quote($k)),
1179 map { $self->_sqlcase($_)} ($op, 'null')
1182 puke "$op can only take undef as argument";
1189 sub _where_field_op_ARRAYREF {
1190 my ($self, $k, $op, $vals) = @_;
1192 my @vals = @$vals; #always work on a copy
1195 $self->_debug(sprintf '%s means multiple elements: [ %s ]',
1197 join(', ', map { defined $_ ? "'$_'" : 'NULL' } @vals ),
1200 # see if the first element is an -and/-or op
1202 if (defined $vals[0] && $vals[0] =~ /^ - (AND|OR) $/ix) {
1207 # a long standing API wart - an attempt to change this behavior during
1208 # the 1.50 series failed *spectacularly*. Warn instead and leave the
1213 (!$logic or $logic eq 'OR')
1215 ($op =~ $self->{inequality_op} or $op =~ $self->{not_like_op})
1218 belch "A multi-element arrayref as an argument to the inequality op '$o' "
1219 . 'is technically equivalent to an always-true 1=1 (you probably wanted '
1220 . "to say ...{ \$inequality_op => [ -and => \@values ] }... instead)"
1224 # distribute $op over each remaining member of @vals, append logic if exists
1225 return $self->_recurse_where([map { {$k => {$op, $_}} } @vals], $logic);
1229 # try to DWIM on equality operators
1231 $op =~ $self->{equality_op} ? $self->{sqlfalse}
1232 : $op =~ $self->{like_op} ? belch("Supplying an empty arrayref to '@{[ uc $op]}' is deprecated") && $self->{sqlfalse}
1233 : $op =~ $self->{inequality_op} ? $self->{sqltrue}
1234 : $op =~ $self->{not_like_op} ? belch("Supplying an empty arrayref to '@{[ uc $op]}' is deprecated") && $self->{sqltrue}
1235 : puke "operator '$op' applied on an empty array (field '$k')";
1240 sub _where_hashpair_SCALARREF {
1241 my ($self, $k, $v) = @_;
1242 $self->_debug("SCALAR($k) means literal SQL: $$v");
1243 my $sql = $self->_quote($k) . " " . $$v;
1247 # literal SQL with bind
1248 sub _where_hashpair_ARRAYREFREF {
1249 my ($self, $k, $v) = @_;
1250 $self->_debug("REF($k) means literal SQL: @${$v}");
1251 my ($sql, @bind) = @$$v;
1252 $self->_assert_bindval_matches_bindtype(@bind);
1253 $sql = $self->_quote($k) . " " . $sql;
1254 return ($sql, @bind );
1257 # literal SQL without bind
1258 sub _where_hashpair_SCALAR {
1259 my ($self, $k, $v) = @_;
1260 $self->_debug("NOREF($k) means simple key=val: $k $self->{cmp} $v");
1261 return ($self->_where_hashpair_HASHREF($k, { $self->{cmp} => $v }));
1265 sub _where_hashpair_UNDEF {
1266 my ($self, $k, $v) = @_;
1267 $self->_debug("UNDEF($k) means IS NULL");
1268 return $self->_where_hashpair_HASHREF($k, { -is => undef });
1271 #======================================================================
1272 # WHERE: TOP-LEVEL OTHERS (SCALARREF, SCALAR, UNDEF)
1273 #======================================================================
1276 sub _where_SCALARREF {
1277 my ($self, $where) = @_;
1280 $self->_debug("SCALAR(*top) means literal SQL: $$where");
1286 my ($self, $where) = @_;
1289 $self->_debug("NOREF(*top) means literal SQL: $where");
1300 #======================================================================
1301 # WHERE: BUILTIN SPECIAL OPERATORS (-in, -between)
1302 #======================================================================
1305 sub _where_field_BETWEEN {
1306 my ($self, $k, $op, $vals) = @_;
1308 my ($label, $and, $placeholder);
1309 $label = $self->_convert($self->_quote($k));
1310 $and = ' ' . $self->_sqlcase('and') . ' ';
1311 $placeholder = $self->_convert('?');
1312 $op = $self->_sqlcase($op);
1314 my $invalid_args = "Operator '$op' requires either an arrayref with two defined values or expressions, or a single literal scalarref/arrayref-ref";
1316 my ($clause, @bind) = $self->_SWITCH_refkind($vals, {
1317 ARRAYREFREF => sub {
1318 my ($s, @b) = @$$vals;
1319 $self->_assert_bindval_matches_bindtype(@b);
1326 puke $invalid_args if @$vals != 2;
1328 my (@all_sql, @all_bind);
1329 foreach my $val (@$vals) {
1330 my ($sql, @bind) = $self->_SWITCH_refkind($val, {
1332 return ($placeholder, $self->_bindtype($k, $val) );
1337 ARRAYREFREF => sub {
1338 my ($sql, @bind) = @$$val;
1339 $self->_assert_bindval_matches_bindtype(@bind);
1340 return ($sql, @bind);
1343 my ($func, $arg, @rest) = %$val;
1344 puke "Only simple { -func => arg } functions accepted as sub-arguments to BETWEEN"
1345 if (@rest or $func !~ /^ \- (.+)/x);
1346 $self->_where_unary_op($1 => $arg);
1352 push @all_sql, $sql;
1353 push @all_bind, @bind;
1357 (join $and, @all_sql),
1366 my $sql = "( $label $op $clause )";
1367 return ($sql, @bind)
1371 sub _where_field_IN {
1372 my ($self, $k, $op, $vals) = @_;
1374 # backwards compatibility: if scalar, force into an arrayref
1375 $vals = [$vals] if defined $vals && ! ref $vals;
1377 my ($label) = $self->_convert($self->_quote($k));
1378 my ($placeholder) = $self->_convert('?');
1379 $op = $self->_sqlcase($op);
1381 my ($sql, @bind) = $self->_SWITCH_refkind($vals, {
1382 ARRAYREF => sub { # list of choices
1383 if (@$vals) { # nonempty list
1384 my (@all_sql, @all_bind);
1386 for my $val (@$vals) {
1387 my ($sql, @bind) = $self->_SWITCH_refkind($val, {
1389 return ($placeholder, $val);
1394 ARRAYREFREF => sub {
1395 my ($sql, @bind) = @$$val;
1396 $self->_assert_bindval_matches_bindtype(@bind);
1397 return ($sql, @bind);
1400 my ($func, $arg, @rest) = %$val;
1401 puke "Only simple { -func => arg } functions accepted as sub-arguments to IN"
1402 if (@rest or $func !~ /^ \- (.+)/x);
1403 $self->_where_unary_op($1 => $arg);
1407 'SQL::Abstract before v1.75 used to generate incorrect SQL when the '
1408 . "-$op operator was given an undef-containing list: !!!AUDIT YOUR CODE "
1409 . 'AND DATA!!! (the upcoming Data::Query-based version of SQL::Abstract '
1410 . 'will emit the logically correct SQL instead of raising this exception)'
1414 push @all_sql, $sql;
1415 push @all_bind, @bind;
1419 sprintf('%s %s ( %s )',
1422 join(', ', @all_sql)
1424 $self->_bindtype($k, @all_bind),
1427 else { # empty list: some databases won't understand "IN ()", so DWIM
1428 my $sql = ($op =~ /\bnot\b/i) ? $self->{sqltrue} : $self->{sqlfalse};
1433 SCALARREF => sub { # literal SQL
1434 my $sql = $self->_open_outer_paren($$vals);
1435 return ("$label $op ( $sql )");
1437 ARRAYREFREF => sub { # literal SQL with bind
1438 my ($sql, @bind) = @$$vals;
1439 $self->_assert_bindval_matches_bindtype(@bind);
1440 $sql = $self->_open_outer_paren($sql);
1441 return ("$label $op ( $sql )", @bind);
1445 puke "Argument passed to the '$op' operator can not be undefined";
1449 puke "special op $op requires an arrayref (or scalarref/arrayref-ref)";
1453 return ($sql, @bind);
1456 # Some databases (SQLite) treat col IN (1, 2) different from
1457 # col IN ( (1, 2) ). Use this to strip all outer parens while
1458 # adding them back in the corresponding method
1459 sub _open_outer_paren {
1460 my ($self, $sql) = @_;
1462 while (my ($inner) = $sql =~ /^ \s* \( (.*) \) \s* $/xs) {
1464 # there are closing parens inside, need the heavy duty machinery
1465 # to reevaluate the extraction starting from $sql (full reevaluation)
1466 if ($inner =~ /\)/) {
1467 require Text::Balanced;
1469 my (undef, $remainder) = do {
1470 # idiotic design - writes to $@ but *DOES NOT* throw exceptions
1472 Text::Balanced::extract_bracketed($sql, '()', qr/\s*/);
1475 # the entire expression needs to be a balanced bracketed thing
1476 # (after an extract no remainder sans trailing space)
1477 last if defined $remainder and $remainder =~ /\S/;
1487 #======================================================================
1489 #======================================================================
1492 my ($self, $arg) = @_;
1495 for my $c ($self->_order_by_chunks($arg) ) {
1496 $self->_SWITCH_refkind($c, {
1497 SCALAR => sub { push @sql, $c },
1498 ARRAYREF => sub { push @sql, shift @$c; push @bind, @$c },
1504 $self->_sqlcase(' order by'),
1510 return wantarray ? ($sql, @bind) : $sql;
1513 sub _order_by_chunks {
1514 my ($self, $arg) = @_;
1516 return $self->_SWITCH_refkind($arg, {
1519 map { $self->_order_by_chunks($_ ) } @$arg;
1522 ARRAYREFREF => sub {
1523 my ($s, @b) = @$$arg;
1524 $self->_assert_bindval_matches_bindtype(@b);
1528 SCALAR => sub {$self->_quote($arg)},
1530 UNDEF => sub {return () },
1532 SCALARREF => sub {$$arg}, # literal SQL, no quoting
1535 # get first pair in hash
1536 my ($key, $val, @rest) = %$arg;
1538 return () unless $key;
1540 if (@rest or not $key =~ /^-(desc|asc)/i) {
1541 puke "hash passed to _order_by must have exactly one key (-desc or -asc)";
1547 for my $c ($self->_order_by_chunks($val)) {
1550 $self->_SWITCH_refkind($c, {
1555 ($sql, @bind) = @$c;
1559 $sql = $sql . ' ' . $self->_sqlcase($direction);
1561 push @ret, [ $sql, @bind];
1570 #======================================================================
1571 # DATASOURCE (FOR NOW, JUST PLAIN TABLE OR LIST OF TABLES)
1572 #======================================================================
1577 $self->_SWITCH_refkind($from, {
1578 ARRAYREF => sub {join ', ', map { $self->_quote($_) } @$from;},
1579 SCALAR => sub {$self->_quote($from)},
1580 SCALARREF => sub {$$from},
1585 #======================================================================
1587 #======================================================================
1589 # highly optimized, as it's called way too often
1591 # my ($self, $label) = @_;
1593 return '' unless defined $_[1];
1594 return ${$_[1]} if ref($_[1]) eq 'SCALAR';
1596 $_[0]->{quote_char} or
1597 ($_[0]->_assert_pass_injection_guard($_[1]), return $_[1]);
1599 my $qref = ref $_[0]->{quote_char};
1601 !$qref ? ($_[0]->{quote_char}, $_[0]->{quote_char})
1602 : ($qref eq 'ARRAY') ? @{$_[0]->{quote_char}}
1603 : puke "Unsupported quote_char format: $_[0]->{quote_char}";
1605 my $esc = $_[0]->{escape_char} || $r;
1607 # parts containing * are naturally unquoted
1608 return join($_[0]->{name_sep}||'', map
1609 +( $_ eq '*' ? $_ : do { (my $n = $_) =~ s/(\Q$esc\E|\Q$r\E)/$esc$1/g; $l . $n . $r } ),
1610 ( $_[0]->{name_sep} ? split (/\Q$_[0]->{name_sep}\E/, $_[1] ) : $_[1] )
1615 # Conversion, if applicable
1617 #my ($self, $arg) = @_;
1618 if ($_[0]->{convert}) {
1619 return $_[0]->_sqlcase($_[0]->{convert}) .'(' . $_[1] . ')';
1626 #my ($self, $col, @vals) = @_;
1627 # called often - tighten code
1628 return $_[0]->{bindtype} eq 'columns'
1629 ? map {[$_[1], $_]} @_[2 .. $#_]
1634 # Dies if any element of @bind is not in [colname => value] format
1635 # if bindtype is 'columns'.
1636 sub _assert_bindval_matches_bindtype {
1637 # my ($self, @bind) = @_;
1639 if ($self->{bindtype} eq 'columns') {
1641 if (!defined $_ || ref($_) ne 'ARRAY' || @$_ != 2) {
1642 puke "bindtype 'columns' selected, you need to pass: [column_name => bind_value]"
1648 sub _join_sql_clauses {
1649 my ($self, $logic, $clauses_aref, $bind_aref) = @_;
1651 if (@$clauses_aref > 1) {
1652 my $join = " " . $self->_sqlcase($logic) . " ";
1653 my $sql = '( ' . join($join, @$clauses_aref) . ' )';
1654 return ($sql, @$bind_aref);
1656 elsif (@$clauses_aref) {
1657 return ($clauses_aref->[0], @$bind_aref); # no parentheses
1660 return (); # if no SQL, ignore @$bind_aref
1665 # Fix SQL case, if so requested
1667 # LDNOTE: if $self->{case} is true, then it contains 'lower', so we
1668 # don't touch the argument ... crooked logic, but let's not change it!
1669 return $_[0]->{case} ? $_[1] : uc($_[1]);
1673 #======================================================================
1674 # DISPATCHING FROM REFKIND
1675 #======================================================================
1678 my ($self, $data) = @_;
1680 return 'UNDEF' unless defined $data;
1682 # blessed objects are treated like scalars
1683 my $ref = (Scalar::Util::blessed $data) ? '' : ref $data;
1685 return 'SCALAR' unless $ref;
1688 while ($ref eq 'REF') {
1690 $ref = (Scalar::Util::blessed $data) ? '' : ref $data;
1694 return ($ref||'SCALAR') . ('REF' x $n_steps);
1698 my ($self, $data) = @_;
1699 my @try = ($self->_refkind($data));
1700 push @try, 'SCALAR_or_UNDEF' if $try[0] eq 'SCALAR' || $try[0] eq 'UNDEF';
1701 push @try, 'FALLBACK';
1705 sub _METHOD_FOR_refkind {
1706 my ($self, $meth_prefix, $data) = @_;
1709 for (@{$self->_try_refkind($data)}) {
1710 $method = $self->can($meth_prefix."_".$_)
1714 return $method || puke "cannot dispatch on '$meth_prefix' for ".$self->_refkind($data);
1718 sub _SWITCH_refkind {
1719 my ($self, $data, $dispatch_table) = @_;
1722 for (@{$self->_try_refkind($data)}) {
1723 $coderef = $dispatch_table->{$_}
1727 puke "no dispatch entry for ".$self->_refkind($data)
1736 #======================================================================
1737 # VALUES, GENERATE, AUTOLOAD
1738 #======================================================================
1740 # LDNOTE: original code from nwiger, didn't touch code in that section
1741 # I feel the AUTOLOAD stuff should not be the default, it should
1742 # only be activated on explicit demand by user.
1746 my $data = shift || return;
1747 puke "Argument to ", __PACKAGE__, "->values must be a \\%hash"
1748 unless ref $data eq 'HASH';
1751 foreach my $k (sort keys %$data) {
1752 my $v = $data->{$k};
1753 $self->_SWITCH_refkind($v, {
1755 if ($self->{array_datatypes}) { # array datatype
1756 push @all_bind, $self->_bindtype($k, $v);
1758 else { # literal SQL with bind
1759 my ($sql, @bind) = @$v;
1760 $self->_assert_bindval_matches_bindtype(@bind);
1761 push @all_bind, @bind;
1764 ARRAYREFREF => sub { # literal SQL with bind
1765 my ($sql, @bind) = @${$v};
1766 $self->_assert_bindval_matches_bindtype(@bind);
1767 push @all_bind, @bind;
1769 SCALARREF => sub { # literal SQL without bind
1771 SCALAR_or_UNDEF => sub {
1772 push @all_bind, $self->_bindtype($k, $v);
1783 my(@sql, @sqlq, @sqlv);
1787 if ($ref eq 'HASH') {
1788 for my $k (sort keys %$_) {
1791 my $label = $self->_quote($k);
1792 if ($r eq 'ARRAY') {
1793 # literal SQL with bind
1794 my ($sql, @bind) = @$v;
1795 $self->_assert_bindval_matches_bindtype(@bind);
1796 push @sqlq, "$label = $sql";
1798 } elsif ($r eq 'SCALAR') {
1799 # literal SQL without bind
1800 push @sqlq, "$label = $$v";
1802 push @sqlq, "$label = ?";
1803 push @sqlv, $self->_bindtype($k, $v);
1806 push @sql, $self->_sqlcase('set'), join ', ', @sqlq;
1807 } elsif ($ref eq 'ARRAY') {
1808 # unlike insert(), assume these are ONLY the column names, i.e. for SQL
1811 if ($r eq 'ARRAY') { # literal SQL with bind
1812 my ($sql, @bind) = @$v;
1813 $self->_assert_bindval_matches_bindtype(@bind);
1816 } elsif ($r eq 'SCALAR') { # literal SQL without bind
1817 # embedded literal SQL
1824 push @sql, '(' . join(', ', @sqlq) . ')';
1825 } elsif ($ref eq 'SCALAR') {
1829 # strings get case twiddled
1830 push @sql, $self->_sqlcase($_);
1834 my $sql = join ' ', @sql;
1836 # this is pretty tricky
1837 # if ask for an array, return ($stmt, @bind)
1838 # otherwise, s/?/shift @sqlv/ to put it inline
1840 return ($sql, @sqlv);
1842 1 while $sql =~ s/\?/my $d = shift(@sqlv);
1843 ref $d ? $d->[1] : $d/e;
1852 # This allows us to check for a local, then _form, attr
1854 my($name) = $AUTOLOAD =~ /.*::(.+)/;
1855 return $self->generate($name, @_);
1866 SQL::Abstract - Generate SQL from Perl data structures
1872 my $sql = SQL::Abstract->new;
1874 my($stmt, @bind) = $sql->select($source, \@fields, \%where, $order);
1876 my($stmt, @bind) = $sql->insert($table, \%fieldvals || \@values);
1878 my($stmt, @bind) = $sql->update($table, \%fieldvals, \%where);
1880 my($stmt, @bind) = $sql->delete($table, \%where);
1882 # Then, use these in your DBI statements
1883 my $sth = $dbh->prepare($stmt);
1884 $sth->execute(@bind);
1886 # Just generate the WHERE clause
1887 my($stmt, @bind) = $sql->where(\%where, $order);
1889 # Return values in the same order, for hashed queries
1890 # See PERFORMANCE section for more details
1891 my @bind = $sql->values(\%fieldvals);
1895 This module was inspired by the excellent L<DBIx::Abstract>.
1896 However, in using that module I found that what I really wanted
1897 to do was generate SQL, but still retain complete control over my
1898 statement handles and use the DBI interface. So, I set out to
1899 create an abstract SQL generation module.
1901 While based on the concepts used by L<DBIx::Abstract>, there are
1902 several important differences, especially when it comes to WHERE
1903 clauses. I have modified the concepts used to make the SQL easier
1904 to generate from Perl data structures and, IMO, more intuitive.
1905 The underlying idea is for this module to do what you mean, based
1906 on the data structures you provide it. The big advantage is that
1907 you don't have to modify your code every time your data changes,
1908 as this module figures it out.
1910 To begin with, an SQL INSERT is as easy as just specifying a hash
1911 of C<key=value> pairs:
1914 name => 'Jimbo Bobson',
1915 phone => '123-456-7890',
1916 address => '42 Sister Lane',
1917 city => 'St. Louis',
1918 state => 'Louisiana',
1921 The SQL can then be generated with this:
1923 my($stmt, @bind) = $sql->insert('people', \%data);
1925 Which would give you something like this:
1927 $stmt = "INSERT INTO people
1928 (address, city, name, phone, state)
1929 VALUES (?, ?, ?, ?, ?)";
1930 @bind = ('42 Sister Lane', 'St. Louis', 'Jimbo Bobson',
1931 '123-456-7890', 'Louisiana');
1933 These are then used directly in your DBI code:
1935 my $sth = $dbh->prepare($stmt);
1936 $sth->execute(@bind);
1938 =head2 Inserting and Updating Arrays
1940 If your database has array types (like for example Postgres),
1941 activate the special option C<< array_datatypes => 1 >>
1942 when creating the C<SQL::Abstract> object.
1943 Then you may use an arrayref to insert and update database array types:
1945 my $sql = SQL::Abstract->new(array_datatypes => 1);
1947 planets => [qw/Mercury Venus Earth Mars/]
1950 my($stmt, @bind) = $sql->insert('solar_system', \%data);
1954 $stmt = "INSERT INTO solar_system (planets) VALUES (?)"
1956 @bind = (['Mercury', 'Venus', 'Earth', 'Mars']);
1959 =head2 Inserting and Updating SQL
1961 In order to apply SQL functions to elements of your C<%data> you may
1962 specify a reference to an arrayref for the given hash value. For example,
1963 if you need to execute the Oracle C<to_date> function on a value, you can
1964 say something like this:
1968 date_entered => \[ "to_date(?,'MM/DD/YYYY')", "03/02/2003" ],
1971 The first value in the array is the actual SQL. Any other values are
1972 optional and would be included in the bind values array. This gives
1975 my($stmt, @bind) = $sql->insert('people', \%data);
1977 $stmt = "INSERT INTO people (name, date_entered)
1978 VALUES (?, to_date(?,'MM/DD/YYYY'))";
1979 @bind = ('Bill', '03/02/2003');
1981 An UPDATE is just as easy, all you change is the name of the function:
1983 my($stmt, @bind) = $sql->update('people', \%data);
1985 Notice that your C<%data> isn't touched; the module will generate
1986 the appropriately quirky SQL for you automatically. Usually you'll
1987 want to specify a WHERE clause for your UPDATE, though, which is
1988 where handling C<%where> hashes comes in handy...
1990 =head2 Complex where statements
1992 This module can generate pretty complicated WHERE statements
1993 easily. For example, simple C<key=value> pairs are taken to mean
1994 equality, and if you want to see if a field is within a set
1995 of values, you can use an arrayref. Let's say we wanted to
1996 SELECT some data based on this criteria:
1999 requestor => 'inna',
2000 worker => ['nwiger', 'rcwe', 'sfz'],
2001 status => { '!=', 'completed' }
2004 my($stmt, @bind) = $sql->select('tickets', '*', \%where);
2006 The above would give you something like this:
2008 $stmt = "SELECT * FROM tickets WHERE
2009 ( requestor = ? ) AND ( status != ? )
2010 AND ( worker = ? OR worker = ? OR worker = ? )";
2011 @bind = ('inna', 'completed', 'nwiger', 'rcwe', 'sfz');
2013 Which you could then use in DBI code like so:
2015 my $sth = $dbh->prepare($stmt);
2016 $sth->execute(@bind);
2022 The methods are simple. There's one for every major SQL operation,
2023 and a constructor you use first. The arguments are specified in a
2024 similar order for each method (table, then fields, then a where
2025 clause) to try and simplify things.
2027 =head2 new(option => 'value')
2029 The C<new()> function takes a list of options and values, and returns
2030 a new B<SQL::Abstract> object which can then be used to generate SQL
2031 through the methods below. The options accepted are:
2037 If set to 'lower', then SQL will be generated in all lowercase. By
2038 default SQL is generated in "textbook" case meaning something like:
2040 SELECT a_field FROM a_table WHERE some_field LIKE '%someval%'
2042 Any setting other than 'lower' is ignored.
2046 This determines what the default comparison operator is. By default
2047 it is C<=>, meaning that a hash like this:
2049 %where = (name => 'nwiger', email => 'nate@wiger.org');
2051 Will generate SQL like this:
2053 WHERE name = 'nwiger' AND email = 'nate@wiger.org'
2055 However, you may want loose comparisons by default, so if you set
2056 C<cmp> to C<like> you would get SQL such as:
2058 WHERE name like 'nwiger' AND email like 'nate@wiger.org'
2060 You can also override the comparison on an individual basis - see
2061 the huge section on L</"WHERE CLAUSES"> at the bottom.
2063 =item sqltrue, sqlfalse
2065 Expressions for inserting boolean values within SQL statements.
2066 By default these are C<1=1> and C<1=0>. They are used
2067 by the special operators C<-in> and C<-not_in> for generating
2068 correct SQL even when the argument is an empty array (see below).
2072 This determines the default logical operator for multiple WHERE
2073 statements in arrays or hashes. If absent, the default logic is "or"
2074 for arrays, and "and" for hashes. This means that a WHERE
2078 event_date => {'>=', '2/13/99'},
2079 event_date => {'<=', '4/24/03'},
2082 will generate SQL like this:
2084 WHERE event_date >= '2/13/99' OR event_date <= '4/24/03'
2086 This is probably not what you want given this query, though (look
2087 at the dates). To change the "OR" to an "AND", simply specify:
2089 my $sql = SQL::Abstract->new(logic => 'and');
2091 Which will change the above C<WHERE> to:
2093 WHERE event_date >= '2/13/99' AND event_date <= '4/24/03'
2095 The logic can also be changed locally by inserting
2096 a modifier in front of an arrayref:
2098 @where = (-and => [event_date => {'>=', '2/13/99'},
2099 event_date => {'<=', '4/24/03'} ]);
2101 See the L</"WHERE CLAUSES"> section for explanations.
2105 This will automatically convert comparisons using the specified SQL
2106 function for both column and value. This is mostly used with an argument
2107 of C<upper> or C<lower>, so that the SQL will have the effect of
2108 case-insensitive "searches". For example, this:
2110 $sql = SQL::Abstract->new(convert => 'upper');
2111 %where = (keywords => 'MaKe iT CAse inSeNSItive');
2113 Will turn out the following SQL:
2115 WHERE upper(keywords) like upper('MaKe iT CAse inSeNSItive')
2117 The conversion can be C<upper()>, C<lower()>, or any other SQL function
2118 that can be applied symmetrically to fields (actually B<SQL::Abstract> does
2119 not validate this option; it will just pass through what you specify verbatim).
2123 This is a kludge because many databases suck. For example, you can't
2124 just bind values using DBI's C<execute()> for Oracle C<CLOB> or C<BLOB> fields.
2125 Instead, you have to use C<bind_param()>:
2127 $sth->bind_param(1, 'reg data');
2128 $sth->bind_param(2, $lots, {ora_type => ORA_CLOB});
2130 The problem is, B<SQL::Abstract> will normally just return a C<@bind> array,
2131 which loses track of which field each slot refers to. Fear not.
2133 If you specify C<bindtype> in new, you can determine how C<@bind> is returned.
2134 Currently, you can specify either C<normal> (default) or C<columns>. If you
2135 specify C<columns>, you will get an array that looks like this:
2137 my $sql = SQL::Abstract->new(bindtype => 'columns');
2138 my($stmt, @bind) = $sql->insert(...);
2141 [ 'column1', 'value1' ],
2142 [ 'column2', 'value2' ],
2143 [ 'column3', 'value3' ],
2146 You can then iterate through this manually, using DBI's C<bind_param()>.
2148 $sth->prepare($stmt);
2151 my($col, $data) = @$_;
2152 if ($col eq 'details' || $col eq 'comments') {
2153 $sth->bind_param($i, $data, {ora_type => ORA_CLOB});
2154 } elsif ($col eq 'image') {
2155 $sth->bind_param($i, $data, {ora_type => ORA_BLOB});
2157 $sth->bind_param($i, $data);
2161 $sth->execute; # execute without @bind now
2163 Now, why would you still use B<SQL::Abstract> if you have to do this crap?
2164 Basically, the advantage is still that you don't have to care which fields
2165 are or are not included. You could wrap that above C<for> loop in a simple
2166 sub called C<bind_fields()> or something and reuse it repeatedly. You still
2167 get a layer of abstraction over manual SQL specification.
2169 Note that if you set L</bindtype> to C<columns>, the C<\[ $sql, @bind ]>
2170 construct (see L</Literal SQL with placeholders and bind values (subqueries)>)
2171 will expect the bind values in this format.
2175 This is the character that a table or column name will be quoted
2176 with. By default this is an empty string, but you could set it to
2177 the character C<`>, to generate SQL like this:
2179 SELECT `a_field` FROM `a_table` WHERE `some_field` LIKE '%someval%'
2181 Alternatively, you can supply an array ref of two items, the first being the left
2182 hand quote character, and the second the right hand quote character. For
2183 example, you could supply C<['[',']']> for SQL Server 2000 compliant quotes
2184 that generates SQL like this:
2186 SELECT [a_field] FROM [a_table] WHERE [some_field] LIKE '%someval%'
2188 Quoting is useful if you have tables or columns names that are reserved
2189 words in your database's SQL dialect.
2193 This is the character that will be used to escape L</quote_char>s appearing
2194 in an identifier before it has been quoted.
2196 The parameter default in case of a single L</quote_char> character is the quote
2199 When opening-closing-style quoting is used (L</quote_char> is an arrayref)
2200 this parameter defaults to the B<closing (right)> L</quote_char>. Occurrences
2201 of the B<opening (left)> L</quote_char> within the identifier are currently left
2202 untouched. The default for opening-closing-style quotes may change in future
2203 versions, thus you are B<strongly encouraged> to specify the escape character
2208 This is the character that separates a table and column name. It is
2209 necessary to specify this when the C<quote_char> option is selected,
2210 so that tables and column names can be individually quoted like this:
2212 SELECT `table`.`one_field` FROM `table` WHERE `table`.`other_field` = 1
2214 =item injection_guard
2216 A regular expression C<qr/.../> that is applied to any C<-function> and unquoted
2217 column name specified in a query structure. This is a safety mechanism to avoid
2218 injection attacks when mishandling user input e.g.:
2220 my %condition_as_column_value_pairs = get_values_from_user();
2221 $sqla->select( ... , \%condition_as_column_value_pairs );
2223 If the expression matches an exception is thrown. Note that literal SQL
2224 supplied via C<\'...'> or C<\['...']> is B<not> checked in any way.
2226 Defaults to checking for C<;> and the C<GO> keyword (TransactSQL)
2228 =item array_datatypes
2230 When this option is true, arrayrefs in INSERT or UPDATE are
2231 interpreted as array datatypes and are passed directly
2233 When this option is false, arrayrefs are interpreted
2234 as literal SQL, just like refs to arrayrefs
2235 (but this behavior is for backwards compatibility; when writing
2236 new queries, use the "reference to arrayref" syntax
2242 Takes a reference to a list of "special operators"
2243 to extend the syntax understood by L<SQL::Abstract>.
2244 See section L</"SPECIAL OPERATORS"> for details.
2248 Takes a reference to a list of "unary operators"
2249 to extend the syntax understood by L<SQL::Abstract>.
2250 See section L</"UNARY OPERATORS"> for details.
2256 =head2 insert($table, \@values || \%fieldvals, \%options)
2258 This is the simplest function. You simply give it a table name
2259 and either an arrayref of values or hashref of field/value pairs.
2260 It returns an SQL INSERT statement and a list of bind values.
2261 See the sections on L</"Inserting and Updating Arrays"> and
2262 L</"Inserting and Updating SQL"> for information on how to insert
2263 with those data types.
2265 The optional C<\%options> hash reference may contain additional
2266 options to generate the insert SQL. Currently supported options
2273 Takes either a scalar of raw SQL fields, or an array reference of
2274 field names, and adds on an SQL C<RETURNING> statement at the end.
2275 This allows you to return data generated by the insert statement
2276 (such as row IDs) without performing another C<SELECT> statement.
2277 Note, however, this is not part of the SQL standard and may not
2278 be supported by all database engines.
2282 =head2 update($table, \%fieldvals, \%where, \%options)
2284 This takes a table, hashref of field/value pairs, and an optional
2285 hashref L<WHERE clause|/WHERE CLAUSES>. It returns an SQL UPDATE function and a list
2287 See the sections on L</"Inserting and Updating Arrays"> and
2288 L</"Inserting and Updating SQL"> for information on how to insert
2289 with those data types.
2291 The optional C<\%options> hash reference may contain additional
2292 options to generate the update SQL. Currently supported options
2299 See the C<returning> option to
2300 L<insert|/insert($table, \@values || \%fieldvals, \%options)>.
2304 =head2 select($source, $fields, $where, $order)
2306 This returns a SQL SELECT statement and associated list of bind values, as
2307 specified by the arguments:
2313 Specification of the 'FROM' part of the statement.
2314 The argument can be either a plain scalar (interpreted as a table
2315 name, will be quoted), or an arrayref (interpreted as a list
2316 of table names, joined by commas, quoted), or a scalarref
2317 (literal SQL, not quoted).
2321 Specification of the list of fields to retrieve from
2323 The argument can be either an arrayref (interpreted as a list
2324 of field names, will be joined by commas and quoted), or a
2325 plain scalar (literal SQL, not quoted).
2326 Please observe that this API is not as flexible as that of
2327 the first argument C<$source>, for backwards compatibility reasons.
2331 Optional argument to specify the WHERE part of the query.
2332 The argument is most often a hashref, but can also be
2333 an arrayref or plain scalar --
2334 see section L<WHERE clause|/"WHERE CLAUSES"> for details.
2338 Optional argument to specify the ORDER BY part of the query.
2339 The argument can be a scalar, a hashref or an arrayref
2340 -- see section L<ORDER BY clause|/"ORDER BY CLAUSES">
2346 =head2 delete($table, \%where, \%options)
2348 This takes a table name and optional hashref L<WHERE clause|/WHERE CLAUSES>.
2349 It returns an SQL DELETE statement and list of bind values.
2351 The optional C<\%options> hash reference may contain additional
2352 options to generate the delete SQL. Currently supported options
2359 See the C<returning> option to
2360 L<insert|/insert($table, \@values || \%fieldvals, \%options)>.
2364 =head2 where(\%where, $order)
2366 This is used to generate just the WHERE clause. For example,
2367 if you have an arbitrary data structure and know what the
2368 rest of your SQL is going to look like, but want an easy way
2369 to produce a WHERE clause, use this. It returns an SQL WHERE
2370 clause and list of bind values.
2373 =head2 values(\%data)
2375 This just returns the values from the hash C<%data>, in the same
2376 order that would be returned from any of the other above queries.
2377 Using this allows you to markedly speed up your queries if you
2378 are affecting lots of rows. See below under the L</"PERFORMANCE"> section.
2380 =head2 generate($any, 'number', $of, \@data, $struct, \%types)
2382 Warning: This is an experimental method and subject to change.
2384 This returns arbitrarily generated SQL. It's a really basic shortcut.
2385 It will return two different things, depending on return context:
2387 my($stmt, @bind) = $sql->generate('create table', \$table, \@fields);
2388 my $stmt_and_val = $sql->generate('create table', \$table, \@fields);
2390 These would return the following:
2392 # First calling form
2393 $stmt = "CREATE TABLE test (?, ?)";
2394 @bind = (field1, field2);
2396 # Second calling form
2397 $stmt_and_val = "CREATE TABLE test (field1, field2)";
2399 Depending on what you're trying to do, it's up to you to choose the correct
2400 format. In this example, the second form is what you would want.
2404 $sql->generate('alter session', { nls_date_format => 'MM/YY' });
2408 ALTER SESSION SET nls_date_format = 'MM/YY'
2410 You get the idea. Strings get their case twiddled, but everything
2411 else remains verbatim.
2413 =head1 EXPORTABLE FUNCTIONS
2415 =head2 is_plain_value
2417 Determines if the supplied argument is a plain value as understood by this
2422 =item * The value is C<undef>
2424 =item * The value is a non-reference
2426 =item * The value is an object with stringification overloading
2428 =item * The value is of the form C<< { -value => $anything } >>
2432 On failure returns C<undef>, on success returns a B<scalar> reference
2433 to the original supplied argument.
2439 The stringification overloading detection is rather advanced: it takes
2440 into consideration not only the presence of a C<""> overload, but if that
2441 fails also checks for enabled
2442 L<autogenerated versions of C<"">|overload/Magic Autogeneration>, based
2443 on either C<0+> or C<bool>.
2445 Unfortunately testing in the field indicates that this
2446 detection B<< may tickle a latent bug in perl versions before 5.018 >>,
2447 but only when very large numbers of stringifying objects are involved.
2448 At the time of writing ( Sep 2014 ) there is no clear explanation of
2449 the direct cause, nor is there a manageably small test case that reliably
2450 reproduces the problem.
2452 If you encounter any of the following exceptions in B<random places within
2453 your application stack> - this module may be to blame:
2455 Operation "ne": no method found,
2456 left argument in overloaded package <something>,
2457 right argument in overloaded package <something>
2461 Stub found while resolving method "???" overloading """" in package <something>
2463 If you fall victim to the above - please attempt to reduce the problem
2464 to something that could be sent to the L<SQL::Abstract developers
2465 |DBIx::Class/GETTING HELP/SUPPORT>
2466 (either publicly or privately). As a workaround in the meantime you can
2467 set C<$ENV{SQLA_ISVALUE_IGNORE_AUTOGENERATED_STRINGIFICATION}> to a true
2468 value, which will most likely eliminate your problem (at the expense of
2469 not being able to properly detect exotic forms of stringification).
2471 This notice and environment variable will be removed in a future version,
2472 as soon as the underlying problem is found and a reliable workaround is
2477 =head2 is_literal_value
2479 Determines if the supplied argument is a literal value as understood by this
2484 =item * C<\$sql_string>
2486 =item * C<\[ $sql_string, @bind_values ]>
2490 On failure returns C<undef>, on success returns an B<array> reference
2491 containing the unpacked version of the supplied literal SQL and bind values.
2493 =head1 WHERE CLAUSES
2497 This module uses a variation on the idea from L<DBIx::Abstract>. It
2498 is B<NOT>, repeat I<not> 100% compatible. B<The main logic of this
2499 module is that things in arrays are OR'ed, and things in hashes
2502 The easiest way to explain is to show lots of examples. After
2503 each C<%where> hash shown, it is assumed you used:
2505 my($stmt, @bind) = $sql->where(\%where);
2507 However, note that the C<%where> hash can be used directly in any
2508 of the other functions as well, as described above.
2510 =head2 Key-value pairs
2512 So, let's get started. To begin, a simple hash:
2516 status => 'completed'
2519 Is converted to SQL C<key = val> statements:
2521 $stmt = "WHERE user = ? AND status = ?";
2522 @bind = ('nwiger', 'completed');
2524 One common thing I end up doing is having a list of values that
2525 a field can be in. To do this, simply specify a list inside of
2530 status => ['assigned', 'in-progress', 'pending'];
2533 This simple code will create the following:
2535 $stmt = "WHERE user = ? AND ( status = ? OR status = ? OR status = ? )";
2536 @bind = ('nwiger', 'assigned', 'in-progress', 'pending');
2538 A field associated to an empty arrayref will be considered a
2539 logical false and will generate 0=1.
2541 =head2 Tests for NULL values
2543 If the value part is C<undef> then this is converted to SQL <IS NULL>
2552 $stmt = "WHERE user = ? AND status IS NULL";
2555 To test if a column IS NOT NULL:
2559 status => { '!=', undef },
2562 =head2 Specific comparison operators
2564 If you want to specify a different type of operator for your comparison,
2565 you can use a hashref for a given column:
2569 status => { '!=', 'completed' }
2572 Which would generate:
2574 $stmt = "WHERE user = ? AND status != ?";
2575 @bind = ('nwiger', 'completed');
2577 To test against multiple values, just enclose the values in an arrayref:
2579 status => { '=', ['assigned', 'in-progress', 'pending'] };
2581 Which would give you:
2583 "WHERE status = ? OR status = ? OR status = ?"
2586 The hashref can also contain multiple pairs, in which case it is expanded
2587 into an C<AND> of its elements:
2591 status => { '!=', 'completed', -not_like => 'pending%' }
2594 # Or more dynamically, like from a form
2595 $where{user} = 'nwiger';
2596 $where{status}{'!='} = 'completed';
2597 $where{status}{'-not_like'} = 'pending%';
2599 # Both generate this
2600 $stmt = "WHERE user = ? AND status != ? AND status NOT LIKE ?";
2601 @bind = ('nwiger', 'completed', 'pending%');
2604 To get an OR instead, you can combine it with the arrayref idea:
2608 priority => [ { '=', 2 }, { '>', 5 } ]
2611 Which would generate:
2613 $stmt = "WHERE ( priority = ? OR priority > ? ) AND user = ?";
2614 @bind = ('2', '5', 'nwiger');
2616 If you want to include literal SQL (with or without bind values), just use a
2617 scalar reference or reference to an arrayref as the value:
2620 date_entered => { '>' => \["to_date(?, 'MM/DD/YYYY')", "11/26/2008"] },
2621 date_expires => { '<' => \"now()" }
2624 Which would generate:
2626 $stmt = "WHERE date_entered > to_date(?, 'MM/DD/YYYY') AND date_expires < now()";
2627 @bind = ('11/26/2008');
2630 =head2 Logic and nesting operators
2632 In the example above,
2633 there is a subtle trap if you want to say something like
2634 this (notice the C<AND>):
2636 WHERE priority != ? AND priority != ?
2638 Because, in Perl you I<can't> do this:
2640 priority => { '!=' => 2, '!=' => 1 }
2642 As the second C<!=> key will obliterate the first. The solution
2643 is to use the special C<-modifier> form inside an arrayref:
2645 priority => [ -and => {'!=', 2},
2649 Normally, these would be joined by C<OR>, but the modifier tells it
2650 to use C<AND> instead. (Hint: You can use this in conjunction with the
2651 C<logic> option to C<new()> in order to change the way your queries
2652 work by default.) B<Important:> Note that the C<-modifier> goes
2653 B<INSIDE> the arrayref, as an extra first element. This will
2654 B<NOT> do what you think it might:
2656 priority => -and => [{'!=', 2}, {'!=', 1}] # WRONG!
2658 Here is a quick list of equivalencies, since there is some overlap:
2661 status => {'!=', 'completed', 'not like', 'pending%' }
2662 status => [ -and => {'!=', 'completed'}, {'not like', 'pending%'}]
2665 status => {'=', ['assigned', 'in-progress']}
2666 status => [ -or => {'=', 'assigned'}, {'=', 'in-progress'}]
2667 status => [ {'=', 'assigned'}, {'=', 'in-progress'} ]
2671 =head2 Special operators: IN, BETWEEN, etc.
2673 You can also use the hashref format to compare a list of fields using the
2674 C<IN> comparison operator, by specifying the list as an arrayref:
2677 status => 'completed',
2678 reportid => { -in => [567, 2335, 2] }
2681 Which would generate:
2683 $stmt = "WHERE status = ? AND reportid IN (?,?,?)";
2684 @bind = ('completed', '567', '2335', '2');
2686 The reverse operator C<-not_in> generates SQL C<NOT IN> and is used in
2689 If the argument to C<-in> is an empty array, 'sqlfalse' is generated
2690 (by default: C<1=0>). Similarly, C<< -not_in => [] >> generates
2691 'sqltrue' (by default: C<1=1>).
2693 In addition to the array you can supply a chunk of literal sql or
2694 literal sql with bind:
2697 customer => { -in => \[
2698 'SELECT cust_id FROM cust WHERE balance > ?',
2701 status => { -in => \'SELECT status_codes FROM states' },
2707 customer IN ( SELECT cust_id FROM cust WHERE balance > ? )
2708 AND status IN ( SELECT status_codes FROM states )
2712 Finally, if the argument to C<-in> is not a reference, it will be
2713 treated as a single-element array.
2715 Another pair of operators is C<-between> and C<-not_between>,
2716 used with an arrayref of two values:
2720 completion_date => {
2721 -not_between => ['2002-10-01', '2003-02-06']
2727 WHERE user = ? AND completion_date NOT BETWEEN ( ? AND ? )
2729 Just like with C<-in> all plausible combinations of literal SQL
2733 start0 => { -between => [ 1, 2 ] },
2734 start1 => { -between => \["? AND ?", 1, 2] },
2735 start2 => { -between => \"lower(x) AND upper(y)" },
2736 start3 => { -between => [
2738 \["upper(?)", 'stuff' ],
2745 ( start0 BETWEEN ? AND ? )
2746 AND ( start1 BETWEEN ? AND ? )
2747 AND ( start2 BETWEEN lower(x) AND upper(y) )
2748 AND ( start3 BETWEEN lower(x) AND upper(?) )
2750 @bind = (1, 2, 1, 2, 'stuff');
2753 These are the two builtin "special operators"; but the
2754 list can be expanded: see section L</"SPECIAL OPERATORS"> below.
2756 =head2 Unary operators: bool
2758 If you wish to test against boolean columns or functions within your
2759 database you can use the C<-bool> and C<-not_bool> operators. For
2760 example to test the column C<is_user> being true and the column
2761 C<is_enabled> being false you would use:-
2765 -not_bool => 'is_enabled',
2770 WHERE is_user AND NOT is_enabled
2772 If a more complex combination is required, testing more conditions,
2773 then you should use the and/or operators:-
2778 -not_bool => { two=> { -rlike => 'bar' } },
2779 -not_bool => { three => [ { '=', 2 }, { '>', 5 } ] },
2790 (NOT ( three = ? OR three > ? ))
2793 =head2 Nested conditions, -and/-or prefixes
2795 So far, we've seen how multiple conditions are joined with a top-level
2796 C<AND>. We can change this by putting the different conditions we want in
2797 hashes and then putting those hashes in an array. For example:
2802 status => { -like => ['pending%', 'dispatched'] },
2806 status => 'unassigned',
2810 This data structure would create the following:
2812 $stmt = "WHERE ( user = ? AND ( status LIKE ? OR status LIKE ? ) )
2813 OR ( user = ? AND status = ? ) )";
2814 @bind = ('nwiger', 'pending', 'dispatched', 'robot', 'unassigned');
2817 Clauses in hashrefs or arrayrefs can be prefixed with an C<-and> or C<-or>
2818 to change the logic inside:
2824 -and => [ workhrs => {'>', 20}, geo => 'ASIA' ],
2825 -or => { workhrs => {'<', 50}, geo => 'EURO' },
2832 $stmt = "WHERE ( user = ?
2833 AND ( ( workhrs > ? AND geo = ? )
2834 OR ( workhrs < ? OR geo = ? ) ) )";
2835 @bind = ('nwiger', '20', 'ASIA', '50', 'EURO');
2837 =head3 Algebraic inconsistency, for historical reasons
2839 C<Important note>: when connecting several conditions, the C<-and->|C<-or>
2840 operator goes C<outside> of the nested structure; whereas when connecting
2841 several constraints on one column, the C<-and> operator goes
2842 C<inside> the arrayref. Here is an example combining both features:
2845 -and => [a => 1, b => 2],
2846 -or => [c => 3, d => 4],
2847 e => [-and => {-like => 'foo%'}, {-like => '%bar'} ]
2852 WHERE ( ( ( a = ? AND b = ? )
2853 OR ( c = ? OR d = ? )
2854 OR ( e LIKE ? AND e LIKE ? ) ) )
2856 This difference in syntax is unfortunate but must be preserved for
2857 historical reasons. So be careful: the two examples below would
2858 seem algebraically equivalent, but they are not
2861 { -like => 'foo%' },
2862 { -like => '%bar' },
2864 # yields: WHERE ( ( col LIKE ? AND col LIKE ? ) )
2867 { col => { -like => 'foo%' } },
2868 { col => { -like => '%bar' } },
2870 # yields: WHERE ( ( col LIKE ? OR col LIKE ? ) )
2873 =head2 Literal SQL and value type operators
2875 The basic premise of SQL::Abstract is that in WHERE specifications the "left
2876 side" is a column name and the "right side" is a value (normally rendered as
2877 a placeholder). This holds true for both hashrefs and arrayref pairs as you
2878 see in the L</WHERE CLAUSES> examples above. Sometimes it is necessary to
2879 alter this behavior. There are several ways of doing so.
2883 This is a virtual operator that signals the string to its right side is an
2884 identifier (a column name) and not a value. For example to compare two
2885 columns you would write:
2888 priority => { '<', 2 },
2889 requestor => { -ident => 'submitter' },
2894 $stmt = "WHERE priority < ? AND requestor = submitter";
2897 If you are maintaining legacy code you may see a different construct as
2898 described in L</Deprecated usage of Literal SQL>, please use C<-ident> in new
2903 This is a virtual operator that signals that the construct to its right side
2904 is a value to be passed to DBI. This is for example necessary when you want
2905 to write a where clause against an array (for RDBMS that support such
2906 datatypes). For example:
2909 array => { -value => [1, 2, 3] }
2914 $stmt = 'WHERE array = ?';
2915 @bind = ([1, 2, 3]);
2917 Note that if you were to simply say:
2923 the result would probably not be what you wanted:
2925 $stmt = 'WHERE array = ? OR array = ? OR array = ?';
2930 Finally, sometimes only literal SQL will do. To include a random snippet
2931 of SQL verbatim, you specify it as a scalar reference. Consider this only
2932 as a last resort. Usually there is a better way. For example:
2935 priority => { '<', 2 },
2936 requestor => { -in => \'(SELECT name FROM hitmen)' },
2941 $stmt = "WHERE priority < ? AND requestor IN (SELECT name FROM hitmen)"
2944 Note that in this example, you only get one bind parameter back, since
2945 the verbatim SQL is passed as part of the statement.
2949 Never use untrusted input as a literal SQL argument - this is a massive
2950 security risk (there is no way to check literal snippets for SQL
2951 injections and other nastyness). If you need to deal with untrusted input
2952 use literal SQL with placeholders as described next.
2954 =head3 Literal SQL with placeholders and bind values (subqueries)
2956 If the literal SQL to be inserted has placeholders and bind values,
2957 use a reference to an arrayref (yes this is a double reference --
2958 not so common, but perfectly legal Perl). For example, to find a date
2959 in Postgres you can use something like this:
2962 date_column => \[ "= date '2008-09-30' - ?::integer", 10 ]
2967 $stmt = "WHERE ( date_column = date '2008-09-30' - ?::integer )"
2970 Note that you must pass the bind values in the same format as they are returned
2971 by L<where|/where(\%where, $order)>. This means that if you set L</bindtype>
2972 to C<columns>, you must provide the bind values in the
2973 C<< [ column_meta => value ] >> format, where C<column_meta> is an opaque
2974 scalar value; most commonly the column name, but you can use any scalar value
2975 (including references and blessed references), L<SQL::Abstract> will simply
2976 pass it through intact. So if C<bindtype> is set to C<columns> the above
2977 example will look like:
2980 date_column => \[ "= date '2008-09-30' - ?::integer", [ {} => 10 ] ]
2983 Literal SQL is especially useful for nesting parenthesized clauses in the
2984 main SQL query. Here is a first example:
2986 my ($sub_stmt, @sub_bind) = ("SELECT c1 FROM t1 WHERE c2 < ? AND c3 LIKE ?",
2990 bar => \["IN ($sub_stmt)" => @sub_bind],
2995 $stmt = "WHERE (foo = ? AND bar IN (SELECT c1 FROM t1
2996 WHERE c2 < ? AND c3 LIKE ?))";
2997 @bind = (1234, 100, "foo%");
2999 Other subquery operators, like for example C<"E<gt> ALL"> or C<"NOT IN">,
3000 are expressed in the same way. Of course the C<$sub_stmt> and
3001 its associated bind values can be generated through a former call
3004 my ($sub_stmt, @sub_bind)
3005 = $sql->select("t1", "c1", {c2 => {"<" => 100},
3006 c3 => {-like => "foo%"}});
3009 bar => \["> ALL ($sub_stmt)" => @sub_bind],
3012 In the examples above, the subquery was used as an operator on a column;
3013 but the same principle also applies for a clause within the main C<%where>
3014 hash, like an EXISTS subquery:
3016 my ($sub_stmt, @sub_bind)
3017 = $sql->select("t1", "*", {c1 => 1, c2 => \"> t0.c0"});
3018 my %where = ( -and => [
3020 \["EXISTS ($sub_stmt)" => @sub_bind],
3025 $stmt = "WHERE (foo = ? AND EXISTS (SELECT * FROM t1
3026 WHERE c1 = ? AND c2 > t0.c0))";
3030 Observe that the condition on C<c2> in the subquery refers to
3031 column C<t0.c0> of the main query: this is I<not> a bind
3032 value, so we have to express it through a scalar ref.
3033 Writing C<< c2 => {">" => "t0.c0"} >> would have generated
3034 C<< c2 > ? >> with bind value C<"t0.c0"> ... not exactly
3035 what we wanted here.
3037 Finally, here is an example where a subquery is used
3038 for expressing unary negation:
3040 my ($sub_stmt, @sub_bind)
3041 = $sql->where({age => [{"<" => 10}, {">" => 20}]});
3042 $sub_stmt =~ s/^ where //i; # don't want "WHERE" in the subclause
3044 lname => {like => '%son%'},
3045 \["NOT ($sub_stmt)" => @sub_bind],
3050 $stmt = "lname LIKE ? AND NOT ( age < ? OR age > ? )"
3051 @bind = ('%son%', 10, 20)
3053 =head3 Deprecated usage of Literal SQL
3055 Below are some examples of archaic use of literal SQL. It is shown only as
3056 reference for those who deal with legacy code. Each example has a much
3057 better, cleaner and safer alternative that users should opt for in new code.
3063 my %where = ( requestor => \'IS NOT NULL' )
3065 $stmt = "WHERE requestor IS NOT NULL"
3067 This used to be the way of generating NULL comparisons, before the handling
3068 of C<undef> got formalized. For new code please use the superior syntax as
3069 described in L</Tests for NULL values>.
3073 my %where = ( requestor => \'= submitter' )
3075 $stmt = "WHERE requestor = submitter"
3077 This used to be the only way to compare columns. Use the superior L</-ident>
3078 method for all new code. For example an identifier declared in such a way
3079 will be properly quoted if L</quote_char> is properly set, while the legacy
3080 form will remain as supplied.
3084 my %where = ( is_ready => \"", completed => { '>', '2012-12-21' } )
3086 $stmt = "WHERE completed > ? AND is_ready"
3087 @bind = ('2012-12-21')
3089 Using an empty string literal used to be the only way to express a boolean.
3090 For all new code please use the much more readable
3091 L<-bool|/Unary operators: bool> operator.
3097 These pages could go on for a while, since the nesting of the data
3098 structures this module can handle are pretty much unlimited (the
3099 module implements the C<WHERE> expansion as a recursive function
3100 internally). Your best bet is to "play around" with the module a
3101 little to see how the data structures behave, and choose the best
3102 format for your data based on that.
3104 And of course, all the values above will probably be replaced with
3105 variables gotten from forms or the command line. After all, if you
3106 knew everything ahead of time, you wouldn't have to worry about
3107 dynamically-generating SQL and could just hardwire it into your
3110 =head1 ORDER BY CLAUSES
3112 Some functions take an order by clause. This can either be a scalar (just a
3113 column name), a hashref of C<< { -desc => 'col' } >> or C<< { -asc => 'col' }
3114 >>, a scalarref, an arrayref-ref, or an arrayref of any of the previous
3117 Given | Will Generate
3118 ---------------------------------------------------------------
3120 'colA' | ORDER BY colA
3122 [qw/colA colB/] | ORDER BY colA, colB
3124 {-asc => 'colA'} | ORDER BY colA ASC
3126 {-desc => 'colB'} | ORDER BY colB DESC
3128 ['colA', {-asc => 'colB'}] | ORDER BY colA, colB ASC
3130 { -asc => [qw/colA colB/] } | ORDER BY colA ASC, colB ASC
3132 \'colA DESC' | ORDER BY colA DESC
3134 \[ 'FUNC(colA, ?)', $x ] | ORDER BY FUNC(colA, ?)
3135 | /* ...with $x bound to ? */
3138 { -asc => 'colA' }, | colA ASC,
3139 { -desc => [qw/colB/] }, | colB DESC,
3140 { -asc => [qw/colC colD/] },| colC ASC, colD ASC,
3141 \'colE DESC', | colE DESC,
3142 \[ 'FUNC(colF, ?)', $x ], | FUNC(colF, ?)
3143 ] | /* ...with $x bound to ? */
3144 ===============================================================
3148 =head1 SPECIAL OPERATORS
3150 my $sqlmaker = SQL::Abstract->new(special_ops => [
3154 my ($self, $field, $op, $arg) = @_;
3160 handler => 'method_name',
3164 A "special operator" is a SQL syntactic clause that can be
3165 applied to a field, instead of a usual binary operator.
3168 WHERE field IN (?, ?, ?)
3169 WHERE field BETWEEN ? AND ?
3170 WHERE MATCH(field) AGAINST (?, ?)
3172 Special operators IN and BETWEEN are fairly standard and therefore
3173 are builtin within C<SQL::Abstract> (as the overridable methods
3174 C<_where_field_IN> and C<_where_field_BETWEEN>). For other operators,
3175 like the MATCH .. AGAINST example above which is specific to MySQL,
3176 you can write your own operator handlers - supply a C<special_ops>
3177 argument to the C<new> method. That argument takes an arrayref of
3178 operator definitions; each operator definition is a hashref with two
3185 the regular expression to match the operator
3189 Either a coderef or a plain scalar method name. In both cases
3190 the expected return is C<< ($sql, @bind) >>.
3192 When supplied with a method name, it is simply called on the
3193 L<SQL::Abstract> object as:
3195 $self->$method_name($field, $op, $arg)
3199 $field is the LHS of the operator
3200 $op is the part that matched the handler regex
3203 When supplied with a coderef, it is called as:
3205 $coderef->($self, $field, $op, $arg)
3210 For example, here is an implementation
3211 of the MATCH .. AGAINST syntax for MySQL
3213 my $sqlmaker = SQL::Abstract->new(special_ops => [
3215 # special op for MySql MATCH (field) AGAINST(word1, word2, ...)
3216 {regex => qr/^match$/i,
3218 my ($self, $field, $op, $arg) = @_;
3219 $arg = [$arg] if not ref $arg;
3220 my $label = $self->_quote($field);
3221 my ($placeholder) = $self->_convert('?');
3222 my $placeholders = join ", ", (($placeholder) x @$arg);
3223 my $sql = $self->_sqlcase('match') . " ($label) "
3224 . $self->_sqlcase('against') . " ($placeholders) ";
3225 my @bind = $self->_bindtype($field, @$arg);
3226 return ($sql, @bind);
3233 =head1 UNARY OPERATORS
3235 my $sqlmaker = SQL::Abstract->new(unary_ops => [
3239 my ($self, $op, $arg) = @_;
3245 handler => 'method_name',
3249 A "unary operator" is a SQL syntactic clause that can be
3250 applied to a field - the operator goes before the field
3252 You can write your own operator handlers - supply a C<unary_ops>
3253 argument to the C<new> method. That argument takes an arrayref of
3254 operator definitions; each operator definition is a hashref with two
3261 the regular expression to match the operator
3265 Either a coderef or a plain scalar method name. In both cases
3266 the expected return is C<< $sql >>.
3268 When supplied with a method name, it is simply called on the
3269 L<SQL::Abstract> object as:
3271 $self->$method_name($op, $arg)
3275 $op is the part that matched the handler regex
3276 $arg is the RHS or argument of the operator
3278 When supplied with a coderef, it is called as:
3280 $coderef->($self, $op, $arg)
3288 Thanks to some benchmarking by Mark Stosberg, it turns out that
3289 this module is many orders of magnitude faster than using C<DBIx::Abstract>.
3290 I must admit this wasn't an intentional design issue, but it's a
3291 byproduct of the fact that you get to control your C<DBI> handles
3294 To maximize performance, use a code snippet like the following:
3296 # prepare a statement handle using the first row
3297 # and then reuse it for the rest of the rows
3299 for my $href (@array_of_hashrefs) {
3300 $stmt ||= $sql->insert('table', $href);
3301 $sth ||= $dbh->prepare($stmt);
3302 $sth->execute($sql->values($href));
3305 The reason this works is because the keys in your C<$href> are sorted
3306 internally by B<SQL::Abstract>. Thus, as long as your data retains
3307 the same structure, you only have to generate the SQL the first time
3308 around. On subsequent queries, simply use the C<values> function provided
3309 by this module to return your values in the correct order.
3311 However this depends on the values having the same type - if, for
3312 example, the values of a where clause may either have values
3313 (resulting in sql of the form C<column = ?> with a single bind
3314 value), or alternatively the values might be C<undef> (resulting in
3315 sql of the form C<column IS NULL> with no bind value) then the
3316 caching technique suggested will not work.
3320 If you use my C<CGI::FormBuilder> module at all, you'll hopefully
3321 really like this part (I do, at least). Building up a complex query
3322 can be as simple as the following:
3329 use CGI::FormBuilder;
3332 my $form = CGI::FormBuilder->new(...);
3333 my $sql = SQL::Abstract->new;
3335 if ($form->submitted) {
3336 my $field = $form->field;
3337 my $id = delete $field->{id};
3338 my($stmt, @bind) = $sql->update('table', $field, {id => $id});
3341 Of course, you would still have to connect using C<DBI> to run the
3342 query, but the point is that if you make your form look like your
3343 table, the actual query script can be extremely simplistic.
3345 If you're B<REALLY> lazy (I am), check out C<HTML::QuickTable> for
3346 a fast interface to returning and formatting data. I frequently
3347 use these three modules together to write complex database query
3348 apps in under 50 lines.
3350 =head1 HOW TO CONTRIBUTE
3352 Contributions are always welcome, in all usable forms (we especially
3353 welcome documentation improvements). The delivery methods include git-
3354 or unified-diff formatted patches, GitHub pull requests, or plain bug
3355 reports either via RT or the Mailing list. Contributors are generally
3356 granted full access to the official repository after their first several
3357 patches pass successful review.
3359 This project is maintained in a git repository. The code and related tools are
3360 accessible at the following locations:
3364 =item * Official repo: L<git://git.shadowcat.co.uk/dbsrgits/SQL-Abstract.git>
3366 =item * Official gitweb: L<http://git.shadowcat.co.uk/gitweb/gitweb.cgi?p=dbsrgits/SQL-Abstract.git>
3368 =item * GitHub mirror: L<https://github.com/dbsrgits/sql-abstract>
3370 =item * Authorized committers: L<ssh://dbsrgits@git.shadowcat.co.uk/SQL-Abstract.git>
3376 Version 1.50 was a major internal refactoring of C<SQL::Abstract>.
3377 Great care has been taken to preserve the I<published> behavior
3378 documented in previous versions in the 1.* family; however,
3379 some features that were previously undocumented, or behaved
3380 differently from the documentation, had to be changed in order
3381 to clarify the semantics. Hence, client code that was relying
3382 on some dark areas of C<SQL::Abstract> v1.*
3383 B<might behave differently> in v1.50.
3385 The main changes are:
3391 support for literal SQL through the C<< \ [ $sql, @bind ] >> syntax.
3395 support for the { operator => \"..." } construct (to embed literal SQL)
3399 support for the { operator => \["...", @bind] } construct (to embed literal SQL with bind values)
3403 optional support for L<array datatypes|/"Inserting and Updating Arrays">
3407 defensive programming: check arguments
3411 fixed bug with global logic, which was previously implemented
3412 through global variables yielding side-effects. Prior versions would
3413 interpret C<< [ {cond1, cond2}, [cond3, cond4] ] >>
3414 as C<< "(cond1 AND cond2) OR (cond3 AND cond4)" >>.
3415 Now this is interpreted
3416 as C<< "(cond1 AND cond2) OR (cond3 OR cond4)" >>.
3421 fixed semantics of _bindtype on array args
3425 dropped the C<_anoncopy> of the %where tree. No longer necessary,
3426 we just avoid shifting arrays within that tree.
3430 dropped the C<_modlogic> function
3434 =head1 ACKNOWLEDGEMENTS
3436 There are a number of individuals that have really helped out with
3437 this module. Unfortunately, most of them submitted bugs via CPAN
3438 so I have no idea who they are! But the people I do know are:
3440 Ash Berlin (order_by hash term support)
3441 Matt Trout (DBIx::Class support)
3442 Mark Stosberg (benchmarking)
3443 Chas Owens (initial "IN" operator support)
3444 Philip Collins (per-field SQL functions)
3445 Eric Kolve (hashref "AND" support)
3446 Mike Fragassi (enhancements to "BETWEEN" and "LIKE")
3447 Dan Kubb (support for "quote_char" and "name_sep")
3448 Guillermo Roditi (patch to cleanup "IN" and "BETWEEN", fix and tests for _order_by)
3449 Laurent Dami (internal refactoring, extensible list of special operators, literal SQL)
3450 Norbert Buchmuller (support for literal SQL in hashpair, misc. fixes & tests)
3451 Peter Rabbitson (rewrite of SQLA::Test, misc. fixes & tests)
3452 Oliver Charles (support for "RETURNING" after "INSERT")
3458 L<DBIx::Class>, L<DBIx::Abstract>, L<CGI::FormBuilder>, L<HTML::QuickTable>.
3462 Copyright (c) 2001-2007 Nathan Wiger <nwiger@cpan.org>. All Rights Reserved.
3464 This module is actively maintained by Matt Trout <mst@shadowcatsystems.co.uk>
3466 For support, your best bet is to try the C<DBIx::Class> users mailing list.
3467 While not an official support venue, C<DBIx::Class> makes heavy use of
3468 C<SQL::Abstract>, and as such list members there are very familiar with
3469 how to create queries.
3473 This module is free software; you may copy this under the same
3474 terms as perl itself (either the GNU General Public License or
3475 the Artistic License)