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' },
58 #======================================================================
59 # DEBUGGING AND ERROR REPORTING
60 #======================================================================
63 return unless $_[0]->{debug}; shift; # a little faster
64 my $func = (caller(1))[3];
65 warn "[$func] ", @_, "\n";
69 my($func) = (caller(1))[3];
70 Carp::carp "[$func] Warning: ", @_;
74 my($func) = (caller(1))[3];
75 Carp::croak "[$func] Fatal: ", @_;
78 sub is_literal_value ($) {
79 ref $_[0] eq 'SCALAR' ? [ ${$_[0]} ]
80 : ( ref $_[0] eq 'REF' and ref ${$_[0]} eq 'ARRAY' ) ? [ @${ $_[0] } ]
84 # FIXME XSify - this can be done so much more efficiently
85 sub is_plain_value ($) {
87 ! length ref $_[0] ? \($_[0])
89 ref $_[0] eq 'HASH' and keys %{$_[0]} == 1
91 exists $_[0]->{-value}
92 ) ? \($_[0]->{-value})
94 # reuse @_ for even moar speedz
95 defined ( $_[1] = Scalar::Util::blessed $_[0] )
97 # deliberately not using Devel::OverloadInfo - the checks we are
98 # intersted in are much more limited than the fullblown thing, and
99 # this is a very hot piece of code
101 # simply using ->can('(""') can leave behind stub methods that
102 # break actually using the overload later (see L<perldiag/Stub
103 # found while resolving method "%s" overloading "%s" in package
104 # "%s"> and the source of overload::mycan())
106 # either has stringification which DBI SHOULD prefer out of the box
107 grep { *{ (qq[${_}::(""]) }{CODE} } @{ $_[2] = mro::get_linear_isa( $_[1] ) }
109 # has nummification or boolification, AND fallback is *not* disabled
111 SQL::Abstract::_ENV_::DETECT_AUTOGENERATED_STRINGIFICATION
114 grep { *{"${_}::(0+"}{CODE} } @{$_[2]}
116 grep { *{"${_}::(bool"}{CODE} } @{$_[2]}
120 # no fallback specified at all
121 ! ( ($_[3]) = grep { *{"${_}::()"}{CODE} } @{$_[2]} )
123 # fallback explicitly undef
124 ! defined ${"$_[3]::()"}
137 #======================================================================
139 #======================================================================
143 my $class = ref($self) || $self;
144 my %opt = (ref $_[0] eq 'HASH') ? %{$_[0]} : @_;
146 # choose our case by keeping an option around
147 delete $opt{case} if $opt{case} && $opt{case} ne 'lower';
149 # default logic for interpreting arrayrefs
150 $opt{logic} = $opt{logic} ? uc $opt{logic} : 'OR';
152 # how to return bind vars
153 $opt{bindtype} ||= 'normal';
155 # default comparison is "=", but can be overridden
158 # try to recognize which are the 'equality' and 'inequality' ops
159 # (temporary quickfix (in 2007), should go through a more seasoned API)
160 $opt{equality_op} = qr/^( \Q$opt{cmp}\E | \= )$/ix;
161 $opt{inequality_op} = qr/^( != | <> )$/ix;
163 $opt{like_op} = qr/^ (is\s+)? r?like $/xi;
164 $opt{not_like_op} = qr/^ (is\s+)? not \s+ r?like $/xi;
167 $opt{sqltrue} ||= '1=1';
168 $opt{sqlfalse} ||= '0=1';
171 $opt{special_ops} ||= [];
172 # regexes are applied in order, thus push after user-defines
173 push @{$opt{special_ops}}, @BUILTIN_SPECIAL_OPS;
176 $opt{unary_ops} ||= [];
177 push @{$opt{unary_ops}}, @BUILTIN_UNARY_OPS;
179 # rudimentary sanity-check for user supplied bits treated as functions/operators
180 # If a purported function matches this regular expression, an exception is thrown.
181 # Literal SQL is *NOT* subject to this check, only functions (and column names
182 # when quoting is not in effect)
185 # need to guard against ()'s in column names too, but this will break tons of
186 # hacks... ideas anyone?
187 $opt{injection_guard} ||= qr/
193 return bless \%opt, $class;
197 sub _assert_pass_injection_guard {
198 if ($_[1] =~ $_[0]->{injection_guard}) {
199 my $class = ref $_[0];
200 puke "Possible SQL injection attempt '$_[1]'. If this is indeed a part of the "
201 . "desired SQL use literal SQL ( \'...' or \[ '...' ] ) or supply your own "
202 . "{injection_guard} attribute to ${class}->new()"
207 #======================================================================
209 #======================================================================
213 my $table = $self->_table(shift);
214 my $data = shift || return;
217 my $method = $self->_METHOD_FOR_refkind("_insert", $data);
218 my ($sql, @bind) = $self->$method($data);
219 $sql = join " ", $self->_sqlcase('insert into'), $table, $sql;
221 if ($options->{returning}) {
222 my ($s, @b) = $self->_insert_returning($options);
227 return wantarray ? ($sql, @bind) : $sql;
230 # So that subclasses can override INSERT ... RETURNING separately from
231 # UPDATE and DELETE (e.g. DBIx::Class::SQLMaker::Oracle does this)
232 sub _insert_returning { shift->_returning(@_) }
235 my ($self, $options) = @_;
237 my $f = $options->{returning};
239 my $fieldlist = $self->_SWITCH_refkind($f, {
240 ARRAYREF => sub {join ', ', map { $self->_quote($_) } @$f;},
241 SCALAR => sub {$self->_quote($f)},
242 SCALARREF => sub {$$f},
244 return $self->_sqlcase(' returning ') . $fieldlist;
247 sub _insert_HASHREF { # explicit list of fields and then values
248 my ($self, $data) = @_;
250 my @fields = sort keys %$data;
252 my ($sql, @bind) = $self->_insert_values($data);
255 $_ = $self->_quote($_) foreach @fields;
256 $sql = "( ".join(", ", @fields).") ".$sql;
258 return ($sql, @bind);
261 sub _insert_ARRAYREF { # just generate values(?,?) part (no list of fields)
262 my ($self, $data) = @_;
264 # no names (arrayref) so can't generate bindtype
265 $self->{bindtype} ne 'columns'
266 or belch "can't do 'columns' bindtype when called with arrayref";
268 my (@values, @all_bind);
269 foreach my $value (@$data) {
270 my ($values, @bind) = $self->_insert_value(undef, $value);
271 push @values, $values;
272 push @all_bind, @bind;
274 my $sql = $self->_sqlcase('values')." ( ".join(", ", @values)." )";
275 return ($sql, @all_bind);
278 sub _insert_ARRAYREFREF { # literal SQL with bind
279 my ($self, $data) = @_;
281 my ($sql, @bind) = @${$data};
282 $self->_assert_bindval_matches_bindtype(@bind);
284 return ($sql, @bind);
288 sub _insert_SCALARREF { # literal SQL without bind
289 my ($self, $data) = @_;
295 my ($self, $data) = @_;
297 my (@values, @all_bind);
298 foreach my $column (sort keys %$data) {
299 my ($values, @bind) = $self->_insert_value($column, $data->{$column});
300 push @values, $values;
301 push @all_bind, @bind;
303 my $sql = $self->_sqlcase('values')." ( ".join(", ", @values)." )";
304 return ($sql, @all_bind);
308 my ($self, $column, $v) = @_;
310 my (@values, @all_bind);
311 $self->_SWITCH_refkind($v, {
314 if ($self->{array_datatypes}) { # if array datatype are activated
316 push @all_bind, $self->_bindtype($column, $v);
318 else { # else literal SQL with bind
319 my ($sql, @bind) = @$v;
320 $self->_assert_bindval_matches_bindtype(@bind);
322 push @all_bind, @bind;
326 ARRAYREFREF => sub { # literal SQL with bind
327 my ($sql, @bind) = @${$v};
328 $self->_assert_bindval_matches_bindtype(@bind);
330 push @all_bind, @bind;
333 # THINK: anything useful to do with a HASHREF ?
334 HASHREF => sub { # (nothing, but old SQLA passed it through)
335 #TODO in SQLA >= 2.0 it will die instead
336 belch "HASH ref as bind value in insert is not supported";
338 push @all_bind, $self->_bindtype($column, $v);
341 SCALARREF => sub { # literal SQL without bind
345 SCALAR_or_UNDEF => sub {
347 push @all_bind, $self->_bindtype($column, $v);
352 my $sql = join(", ", @values);
353 return ($sql, @all_bind);
358 #======================================================================
360 #======================================================================
365 my $table = $self->_table(shift);
366 my $data = shift || return;
370 # first build the 'SET' part of the sql statement
371 puke "Unsupported data type specified to \$sql->update"
372 unless ref $data eq 'HASH';
374 my ($sql, @all_bind) = $self->_update_set_values($data);
375 $sql = $self->_sqlcase('update ') . $table . $self->_sqlcase(' set ')
379 my($where_sql, @where_bind) = $self->where($where);
381 push @all_bind, @where_bind;
384 if ($options->{returning}) {
385 my ($returning_sql, @returning_bind) = $self->_update_returning($options);
386 $sql .= $returning_sql;
387 push @all_bind, @returning_bind;
390 return wantarray ? ($sql, @all_bind) : $sql;
393 sub _update_set_values {
394 my ($self, $data) = @_;
396 my (@set, @all_bind);
397 for my $k (sort keys %$data) {
400 my $label = $self->_quote($k);
402 $self->_SWITCH_refkind($v, {
404 if ($self->{array_datatypes}) { # array datatype
405 push @set, "$label = ?";
406 push @all_bind, $self->_bindtype($k, $v);
408 else { # literal SQL with bind
409 my ($sql, @bind) = @$v;
410 $self->_assert_bindval_matches_bindtype(@bind);
411 push @set, "$label = $sql";
412 push @all_bind, @bind;
415 ARRAYREFREF => sub { # literal SQL with bind
416 my ($sql, @bind) = @${$v};
417 $self->_assert_bindval_matches_bindtype(@bind);
418 push @set, "$label = $sql";
419 push @all_bind, @bind;
421 SCALARREF => sub { # literal SQL without bind
422 push @set, "$label = $$v";
425 my ($op, $arg, @rest) = %$v;
427 puke 'Operator calls in update must be in the form { -op => $arg }'
428 if (@rest or not $op =~ /^\-(.+)/);
430 local $self->{_nested_func_lhs} = $k;
431 my ($sql, @bind) = $self->_where_unary_op($1, $arg);
433 push @set, "$label = $sql";
434 push @all_bind, @bind;
436 SCALAR_or_UNDEF => sub {
437 push @set, "$label = ?";
438 push @all_bind, $self->_bindtype($k, $v);
444 my $sql = join ', ', @set;
446 return ($sql, @all_bind);
449 # So that subclasses can override UPDATE ... RETURNING separately from
451 sub _update_returning { shift->_returning(@_) }
455 #======================================================================
457 #======================================================================
462 my $table = $self->_table(shift);
463 my $fields = shift || '*';
467 my ($fields_sql, @bind) = $self->_select_fields($fields);
469 my ($where_sql, @where_bind) = $self->where($where, $order);
470 push @bind, @where_bind;
472 my $sql = join(' ', $self->_sqlcase('select'), $fields_sql,
473 $self->_sqlcase('from'), $table)
476 return wantarray ? ($sql, @bind) : $sql;
480 my ($self, $fields) = @_;
481 return ref $fields eq 'ARRAY' ? join ', ', map { $self->_quote($_) } @$fields
485 #======================================================================
487 #======================================================================
492 my $table = $self->_table(shift);
496 my($where_sql, @bind) = $self->where($where);
497 my $sql = $self->_sqlcase('delete from ') . $table . $where_sql;
499 if ($options->{returning}) {
500 my ($returning_sql, @returning_bind) = $self->_delete_returning($options);
501 $sql .= $returning_sql;
502 push @bind, @returning_bind;
505 return wantarray ? ($sql, @bind) : $sql;
508 # So that subclasses can override DELETE ... RETURNING separately from
510 sub _delete_returning { shift->_returning(@_) }
514 #======================================================================
516 #======================================================================
520 # Finally, a separate routine just to handle WHERE clauses
522 my ($self, $where, $order) = @_;
525 my ($sql, @bind) = $self->_recurse_where($where);
526 $sql = (defined $sql and length $sql) ? $self->_sqlcase(' where ') . "( $sql )" : '';
530 my ($order_sql, @order_bind) = $self->_order_by($order);
532 push @bind, @order_bind;
535 return wantarray ? ($sql, @bind) : $sql;
539 my ($self, $expr, $logic) = @_;
540 if (ref($expr) eq 'HASH' and keys %$expr > 1) {
542 return +{ "-${logic}" => [
543 map +{ $_ => $expr->{$_} }, sort keys %$expr
550 my ($self, $where, $logic) = @_;
552 my $where_exp = $self->_expand_expr($where, $logic);
554 # dispatch on appropriate method according to refkind of $where
555 my $method = $self->_METHOD_FOR_refkind("_where", $where_exp);
557 my ($sql, @bind) = $self->$method($where_exp, $logic);
559 # DBIx::Class used to call _recurse_where in scalar context
560 # something else might too...
562 return ($sql, @bind);
565 belch "Calling _recurse_where in scalar context is deprecated and will go away before 2.0";
572 #======================================================================
573 # WHERE: top-level ARRAYREF
574 #======================================================================
577 sub _where_ARRAYREF {
578 my ($self, $where, $logic) = @_;
580 $logic = uc($logic || $self->{logic});
581 $logic eq 'AND' or $logic eq 'OR' or puke "unknown logic: $logic";
583 my @clauses = @$where;
585 my (@sql_clauses, @all_bind);
586 # need to use while() so can shift() for pairs
588 my $el = shift @clauses;
590 $el = undef if (defined $el and ! length $el);
592 # switch according to kind of $el and get corresponding ($sql, @bind)
593 my ($sql, @bind) = $self->_SWITCH_refkind($el, {
595 # skip empty elements, otherwise get invalid trailing AND stuff
596 ARRAYREF => sub {$self->_recurse_where($el) if @$el},
600 $self->_assert_bindval_matches_bindtype(@b);
604 HASHREF => sub {$self->_recurse_where($el, 'and') if %$el},
606 SCALARREF => sub { ($$el); },
609 # top-level arrayref with scalars, recurse in pairs
610 $self->_recurse_where({$el => shift(@clauses)})
613 UNDEF => sub {puke "Supplying an empty left hand side argument is not supported in array-pairs" },
617 push @sql_clauses, $sql;
618 push @all_bind, @bind;
622 return $self->_join_sql_clauses($logic, \@sql_clauses, \@all_bind);
625 #======================================================================
626 # WHERE: top-level ARRAYREFREF
627 #======================================================================
629 sub _where_ARRAYREFREF {
630 my ($self, $where) = @_;
631 my ($sql, @bind) = @$$where;
632 $self->_assert_bindval_matches_bindtype(@bind);
633 return ($sql, @bind);
636 #======================================================================
637 # WHERE: top-level HASHREF
638 #======================================================================
641 my ($self, $where) = @_;
642 my (@sql_clauses, @all_bind);
644 for my $k (sort keys %$where) {
645 my $v = $where->{$k};
647 # ($k => $v) is either a special unary op or a regular hashpair
648 my ($sql, @bind) = do {
650 # put the operator in canonical form
652 $op = substr $op, 1; # remove initial dash
653 $op =~ s/^\s+|\s+$//g;# remove leading/trailing space
654 $op =~ s/\s+/ /g; # compress whitespace
656 # so that -not_foo works correctly
657 $op =~ s/^not_/NOT /i;
659 $self->_debug("Unary OP(-$op) within hashref, recursing...");
660 my ($s, @b) = $self->_where_unary_op($op, $v);
662 # top level vs nested
663 # we assume that handled unary ops will take care of their ()s
665 List::Util::first {$op =~ $_->{regex}} @{$self->{unary_ops}}
667 ( defined $self->{_nested_func_lhs} and $self->{_nested_func_lhs} eq $k )
673 if (is_literal_value ($v) ) {
674 belch 'Hash-pairs consisting of an empty string with a literal are deprecated, and will be removed in 2.0: use -and => [ $literal ] instead';
677 puke "Supplying an empty left hand side argument is not supported in hash-pairs";
681 my $method = $self->_METHOD_FOR_refkind("_where_hashpair", $v);
682 $self->$method($k, $v);
686 push @sql_clauses, $sql;
687 push @all_bind, @bind;
690 return $self->_join_sql_clauses('and', \@sql_clauses, \@all_bind);
693 sub _where_unary_op {
694 my ($self, $op, $rhs) = @_;
696 # top level special ops are illegal in general
697 # this includes the -ident/-value ops (dual purpose unary and special)
698 puke "Illegal use of top-level '-$op'"
699 if ! defined $self->{_nested_func_lhs} and List::Util::first { $op =~ $_->{regex} } @{$self->{special_ops}};
701 if (my $op_entry = List::Util::first { $op =~ $_->{regex} } @{$self->{unary_ops}}) {
702 my $handler = $op_entry->{handler};
704 if (not ref $handler) {
705 if ($op =~ s/ [_\s]? \d+ $//x ) {
706 belch 'Use of [and|or|nest]_N modifiers is deprecated and will be removed in SQLA v2.0. '
707 . "You probably wanted ...-and => [ -$op => COND1, -$op => COND2 ... ]";
709 return $self->$handler($op, $rhs);
711 elsif (ref $handler eq 'CODE') {
712 return $handler->($self, $op, $rhs);
715 puke "Illegal handler for operator $op - expecting a method name or a coderef";
719 $self->_debug("Generic unary OP: $op - recursing as function");
721 $self->_assert_pass_injection_guard($op);
723 my ($sql, @bind) = $self->_SWITCH_refkind($rhs, {
725 puke "Illegal use of top-level '-$op'"
726 unless defined $self->{_nested_func_lhs};
729 $self->_convert('?'),
730 $self->_bindtype($self->{_nested_func_lhs}, $rhs)
734 $self->_recurse_where($rhs)
738 $sql = sprintf('%s %s',
739 $self->_sqlcase($op),
743 return ($sql, @bind);
746 sub _where_op_ANDOR {
747 my ($self, $op, $v) = @_;
749 $self->_SWITCH_refkind($v, {
751 return $self->_where_ARRAYREF($v, $op);
755 return ($op =~ /^or/i)
756 ? $self->_where_ARRAYREF([ map { $_ => $v->{$_} } (sort keys %$v) ], $op)
757 : $self->_where_HASHREF($v);
761 puke "-$op => \\\$scalar makes little sense, use " .
763 ? '[ \$scalar, \%rest_of_conditions ] instead'
764 : '-and => [ \$scalar, \%rest_of_conditions ] instead'
769 puke "-$op => \\[...] makes little sense, use " .
771 ? '[ \[...], \%rest_of_conditions ] instead'
772 : '-and => [ \[...], \%rest_of_conditions ] instead'
776 SCALAR => sub { # permissively interpreted as SQL
777 puke "-$op => \$value makes little sense, use -bool => \$value instead";
781 puke "-$op => undef not supported";
787 my ($self, $op, $v) = @_;
789 $self->_SWITCH_refkind($v, {
791 SCALAR => sub { # permissively interpreted as SQL
792 belch "literal SQL should be -nest => \\'scalar' "
793 . "instead of -nest => 'scalar' ";
798 puke "-$op => undef not supported";
802 $self->_recurse_where($v);
810 my ($self, $op, $v) = @_;
812 my ($s, @b) = $self->_SWITCH_refkind($v, {
813 SCALAR => sub { # interpreted as SQL column
814 $self->_convert($self->_quote($v));
818 puke "-$op => undef not supported";
822 $self->_recurse_where($v);
826 $s = "(NOT $s)" if $op =~ /^not/i;
831 sub _where_op_IDENT {
833 my ($op, $rhs) = splice @_, -2;
834 if (! defined $rhs or length ref $rhs) {
835 puke "-$op requires a single plain scalar argument (a quotable identifier)";
838 # in case we are called as a top level special op (no '=')
841 $_ = $self->_convert($self->_quote($_)) for ($lhs, $rhs);
849 sub _where_op_VALUE {
851 my ($op, $rhs) = splice @_, -2;
853 # in case we are called as a top level special op (no '=')
857 if (! defined $rhs) {
859 ? $self->_where_hashpair_HASHREF($lhs, { -is => undef })
866 (defined $lhs ? $lhs : $self->{_nested_func_lhs}),
873 $self->_convert($self->_quote($lhs)) . ' = ' . $self->_convert('?'),
877 $self->_convert('?'),
883 sub _where_hashpair_ARRAYREF {
884 my ($self, $k, $v) = @_;
887 my @v = @$v; # need copy because of shift below
888 $self->_debug("ARRAY($k) means distribute over elements");
890 # put apart first element if it is an operator (-and, -or)
892 (defined $v[0] && $v[0] =~ /^ - (?: AND|OR ) $/ix)
896 my @distributed = map { {$k => $_} } @v;
899 $self->_debug("OP($op) reinjected into the distributed array");
900 unshift @distributed, $op;
903 my $logic = $op ? substr($op, 1) : '';
905 return $self->_recurse_where(\@distributed, $logic);
908 $self->_debug("empty ARRAY($k) means 0=1");
909 return ($self->{sqlfalse});
913 sub _where_hashpair_HASHREF {
914 my ($self, $k, $v, $logic) = @_;
917 local $self->{_nested_func_lhs} = defined $self->{_nested_func_lhs}
918 ? $self->{_nested_func_lhs}
922 my ($all_sql, @all_bind);
924 for my $orig_op (sort keys %$v) {
925 my $val = $v->{$orig_op};
927 # put the operator in canonical form
930 # FIXME - we need to phase out dash-less ops
931 $op =~ s/^-//; # remove possible initial dash
932 $op =~ s/^\s+|\s+$//g;# remove leading/trailing space
933 $op =~ s/\s+/ /g; # compress whitespace
935 $self->_assert_pass_injection_guard($op);
938 $op =~ s/^is_not/IS NOT/i;
940 # so that -not_foo works correctly
941 $op =~ s/^not_/NOT /i;
943 # another retarded special case: foo => { $op => { -value => undef } }
944 if (ref $val eq 'HASH' and keys %$val == 1 and exists $val->{-value} and ! defined $val->{-value} ) {
950 # CASE: col-value logic modifiers
951 if ($orig_op =~ /^ \- (and|or) $/xi) {
952 ($sql, @bind) = $self->_where_hashpair_HASHREF($k, $val, $1);
954 # CASE: special operators like -in or -between
955 elsif (my $special_op = List::Util::first { $op =~ $_->{regex} } @{$self->{special_ops}}) {
956 my $handler = $special_op->{handler};
958 puke "No handler supplied for special operator $orig_op";
960 elsif (not ref $handler) {
961 ($sql, @bind) = $self->$handler($k, $op, $val);
963 elsif (ref $handler eq 'CODE') {
964 ($sql, @bind) = $handler->($self, $k, $op, $val);
967 puke "Illegal handler for special operator $orig_op - expecting a method name or a coderef";
971 $self->_SWITCH_refkind($val, {
973 ARRAYREF => sub { # CASE: col => {op => \@vals}
974 ($sql, @bind) = $self->_where_field_op_ARRAYREF($k, $op, $val);
977 ARRAYREFREF => sub { # CASE: col => {op => \[$sql, @bind]} (literal SQL with bind)
978 my ($sub_sql, @sub_bind) = @$$val;
979 $self->_assert_bindval_matches_bindtype(@sub_bind);
980 $sql = join ' ', $self->_convert($self->_quote($k)),
981 $self->_sqlcase($op),
986 UNDEF => sub { # CASE: col => {op => undef} : sql "IS (NOT)? NULL"
988 $op =~ /^not$/i ? 'is not' # legacy
989 : $op =~ $self->{equality_op} ? 'is'
990 : $op =~ $self->{like_op} ? belch("Supplying an undefined argument to '@{[ uc $op]}' is deprecated") && 'is'
991 : $op =~ $self->{inequality_op} ? 'is not'
992 : $op =~ $self->{not_like_op} ? belch("Supplying an undefined argument to '@{[ uc $op]}' is deprecated") && 'is not'
993 : puke "unexpected operator '$orig_op' with undef operand";
995 $sql = $self->_quote($k) . $self->_sqlcase(" $is null");
998 FALLBACK => sub { # CASE: col => {op/func => $stuff}
999 ($sql, @bind) = $self->_where_unary_op($op, $val);
1002 $self->_convert($self->_quote($k)),
1003 $self->{_nested_func_lhs} eq $k ? $sql : "($sql)", # top level vs nested
1009 ($all_sql) = (defined $all_sql and $all_sql) ? $self->_join_sql_clauses($logic, [$all_sql, $sql], []) : $sql;
1010 push @all_bind, @bind;
1012 return ($all_sql, @all_bind);
1015 sub _where_field_IS {
1016 my ($self, $k, $op, $v) = @_;
1018 my ($s) = $self->_SWITCH_refkind($v, {
1021 $self->_convert($self->_quote($k)),
1022 map { $self->_sqlcase($_)} ($op, 'null')
1025 puke "$op can only take undef as argument";
1032 sub _where_field_op_ARRAYREF {
1033 my ($self, $k, $op, $vals) = @_;
1035 my @vals = @$vals; #always work on a copy
1038 $self->_debug(sprintf '%s means multiple elements: [ %s ]',
1040 join(', ', map { defined $_ ? "'$_'" : 'NULL' } @vals ),
1043 # see if the first element is an -and/-or op
1045 if (defined $vals[0] && $vals[0] =~ /^ - (AND|OR) $/ix) {
1050 # a long standing API wart - an attempt to change this behavior during
1051 # the 1.50 series failed *spectacularly*. Warn instead and leave the
1056 (!$logic or $logic eq 'OR')
1058 ($op =~ $self->{inequality_op} or $op =~ $self->{not_like_op})
1061 belch "A multi-element arrayref as an argument to the inequality op '$o' "
1062 . 'is technically equivalent to an always-true 1=1 (you probably wanted '
1063 . "to say ...{ \$inequality_op => [ -and => \@values ] }... instead)"
1067 # distribute $op over each remaining member of @vals, append logic if exists
1068 return $self->_recurse_where([map { {$k => {$op, $_}} } @vals], $logic);
1072 # try to DWIM on equality operators
1074 $op =~ $self->{equality_op} ? $self->{sqlfalse}
1075 : $op =~ $self->{like_op} ? belch("Supplying an empty arrayref to '@{[ uc $op]}' is deprecated") && $self->{sqlfalse}
1076 : $op =~ $self->{inequality_op} ? $self->{sqltrue}
1077 : $op =~ $self->{not_like_op} ? belch("Supplying an empty arrayref to '@{[ uc $op]}' is deprecated") && $self->{sqltrue}
1078 : puke "operator '$op' applied on an empty array (field '$k')";
1083 sub _where_hashpair_SCALARREF {
1084 my ($self, $k, $v) = @_;
1085 $self->_debug("SCALAR($k) means literal SQL: $$v");
1086 my $sql = $self->_quote($k) . " " . $$v;
1090 # literal SQL with bind
1091 sub _where_hashpair_ARRAYREFREF {
1092 my ($self, $k, $v) = @_;
1093 $self->_debug("REF($k) means literal SQL: @${$v}");
1094 my ($sql, @bind) = @$$v;
1095 $self->_assert_bindval_matches_bindtype(@bind);
1096 $sql = $self->_quote($k) . " " . $sql;
1097 return ($sql, @bind );
1100 # literal SQL without bind
1101 sub _where_hashpair_SCALAR {
1102 my ($self, $k, $v) = @_;
1103 $self->_debug("NOREF($k) means simple key=val: $k $self->{cmp} $v");
1104 return ($self->_where_hashpair_HASHREF($k, { $self->{cmp} => $v }));
1108 sub _where_hashpair_UNDEF {
1109 my ($self, $k, $v) = @_;
1110 $self->_debug("UNDEF($k) means IS NULL");
1111 return $self->_where_hashpair_HASHREF($k, { -is => undef });
1114 #======================================================================
1115 # WHERE: TOP-LEVEL OTHERS (SCALARREF, SCALAR, UNDEF)
1116 #======================================================================
1119 sub _where_SCALARREF {
1120 my ($self, $where) = @_;
1123 $self->_debug("SCALAR(*top) means literal SQL: $$where");
1129 my ($self, $where) = @_;
1132 $self->_debug("NOREF(*top) means literal SQL: $where");
1143 #======================================================================
1144 # WHERE: BUILTIN SPECIAL OPERATORS (-in, -between)
1145 #======================================================================
1148 sub _where_field_BETWEEN {
1149 my ($self, $k, $op, $vals) = @_;
1151 my ($label, $and, $placeholder);
1152 $label = $self->_convert($self->_quote($k));
1153 $and = ' ' . $self->_sqlcase('and') . ' ';
1154 $placeholder = $self->_convert('?');
1155 $op = $self->_sqlcase($op);
1157 my $invalid_args = "Operator '$op' requires either an arrayref with two defined values or expressions, or a single literal scalarref/arrayref-ref";
1159 my ($clause, @bind) = $self->_SWITCH_refkind($vals, {
1160 ARRAYREFREF => sub {
1161 my ($s, @b) = @$$vals;
1162 $self->_assert_bindval_matches_bindtype(@b);
1169 puke $invalid_args if @$vals != 2;
1171 my (@all_sql, @all_bind);
1172 foreach my $val (@$vals) {
1173 my ($sql, @bind) = $self->_SWITCH_refkind($val, {
1175 return ($placeholder, $self->_bindtype($k, $val) );
1180 ARRAYREFREF => sub {
1181 my ($sql, @bind) = @$$val;
1182 $self->_assert_bindval_matches_bindtype(@bind);
1183 return ($sql, @bind);
1186 my ($func, $arg, @rest) = %$val;
1187 puke "Only simple { -func => arg } functions accepted as sub-arguments to BETWEEN"
1188 if (@rest or $func !~ /^ \- (.+)/x);
1189 $self->_where_unary_op($1 => $arg);
1195 push @all_sql, $sql;
1196 push @all_bind, @bind;
1200 (join $and, @all_sql),
1209 my $sql = "( $label $op $clause )";
1210 return ($sql, @bind)
1214 sub _where_field_IN {
1215 my ($self, $k, $op, $vals) = @_;
1217 # backwards compatibility: if scalar, force into an arrayref
1218 $vals = [$vals] if defined $vals && ! ref $vals;
1220 my ($label) = $self->_convert($self->_quote($k));
1221 my ($placeholder) = $self->_convert('?');
1222 $op = $self->_sqlcase($op);
1224 my ($sql, @bind) = $self->_SWITCH_refkind($vals, {
1225 ARRAYREF => sub { # list of choices
1226 if (@$vals) { # nonempty list
1227 my (@all_sql, @all_bind);
1229 for my $val (@$vals) {
1230 my ($sql, @bind) = $self->_SWITCH_refkind($val, {
1232 return ($placeholder, $val);
1237 ARRAYREFREF => sub {
1238 my ($sql, @bind) = @$$val;
1239 $self->_assert_bindval_matches_bindtype(@bind);
1240 return ($sql, @bind);
1243 my ($func, $arg, @rest) = %$val;
1244 puke "Only simple { -func => arg } functions accepted as sub-arguments to IN"
1245 if (@rest or $func !~ /^ \- (.+)/x);
1246 $self->_where_unary_op($1 => $arg);
1250 'SQL::Abstract before v1.75 used to generate incorrect SQL when the '
1251 . "-$op operator was given an undef-containing list: !!!AUDIT YOUR CODE "
1252 . 'AND DATA!!! (the upcoming Data::Query-based version of SQL::Abstract '
1253 . 'will emit the logically correct SQL instead of raising this exception)'
1257 push @all_sql, $sql;
1258 push @all_bind, @bind;
1262 sprintf('%s %s ( %s )',
1265 join(', ', @all_sql)
1267 $self->_bindtype($k, @all_bind),
1270 else { # empty list: some databases won't understand "IN ()", so DWIM
1271 my $sql = ($op =~ /\bnot\b/i) ? $self->{sqltrue} : $self->{sqlfalse};
1276 SCALARREF => sub { # literal SQL
1277 my $sql = $self->_open_outer_paren($$vals);
1278 return ("$label $op ( $sql )");
1280 ARRAYREFREF => sub { # literal SQL with bind
1281 my ($sql, @bind) = @$$vals;
1282 $self->_assert_bindval_matches_bindtype(@bind);
1283 $sql = $self->_open_outer_paren($sql);
1284 return ("$label $op ( $sql )", @bind);
1288 puke "Argument passed to the '$op' operator can not be undefined";
1292 puke "special op $op requires an arrayref (or scalarref/arrayref-ref)";
1296 return ($sql, @bind);
1299 # Some databases (SQLite) treat col IN (1, 2) different from
1300 # col IN ( (1, 2) ). Use this to strip all outer parens while
1301 # adding them back in the corresponding method
1302 sub _open_outer_paren {
1303 my ($self, $sql) = @_;
1305 while (my ($inner) = $sql =~ /^ \s* \( (.*) \) \s* $/xs) {
1307 # there are closing parens inside, need the heavy duty machinery
1308 # to reevaluate the extraction starting from $sql (full reevaluation)
1309 if ($inner =~ /\)/) {
1310 require Text::Balanced;
1312 my (undef, $remainder) = do {
1313 # idiotic design - writes to $@ but *DOES NOT* throw exceptions
1315 Text::Balanced::extract_bracketed($sql, '()', qr/\s*/);
1318 # the entire expression needs to be a balanced bracketed thing
1319 # (after an extract no remainder sans trailing space)
1320 last if defined $remainder and $remainder =~ /\S/;
1330 #======================================================================
1332 #======================================================================
1335 my ($self, $arg) = @_;
1338 for my $c ($self->_order_by_chunks($arg) ) {
1339 $self->_SWITCH_refkind($c, {
1340 SCALAR => sub { push @sql, $c },
1341 ARRAYREF => sub { push @sql, shift @$c; push @bind, @$c },
1347 $self->_sqlcase(' order by'),
1353 return wantarray ? ($sql, @bind) : $sql;
1356 sub _order_by_chunks {
1357 my ($self, $arg) = @_;
1359 return $self->_SWITCH_refkind($arg, {
1362 map { $self->_order_by_chunks($_ ) } @$arg;
1365 ARRAYREFREF => sub {
1366 my ($s, @b) = @$$arg;
1367 $self->_assert_bindval_matches_bindtype(@b);
1371 SCALAR => sub {$self->_quote($arg)},
1373 UNDEF => sub {return () },
1375 SCALARREF => sub {$$arg}, # literal SQL, no quoting
1378 # get first pair in hash
1379 my ($key, $val, @rest) = %$arg;
1381 return () unless $key;
1383 if (@rest or not $key =~ /^-(desc|asc)/i) {
1384 puke "hash passed to _order_by must have exactly one key (-desc or -asc)";
1390 for my $c ($self->_order_by_chunks($val)) {
1393 $self->_SWITCH_refkind($c, {
1398 ($sql, @bind) = @$c;
1402 $sql = $sql . ' ' . $self->_sqlcase($direction);
1404 push @ret, [ $sql, @bind];
1413 #======================================================================
1414 # DATASOURCE (FOR NOW, JUST PLAIN TABLE OR LIST OF TABLES)
1415 #======================================================================
1420 $self->_SWITCH_refkind($from, {
1421 ARRAYREF => sub {join ', ', map { $self->_quote($_) } @$from;},
1422 SCALAR => sub {$self->_quote($from)},
1423 SCALARREF => sub {$$from},
1428 #======================================================================
1430 #======================================================================
1432 # highly optimized, as it's called way too often
1434 # my ($self, $label) = @_;
1436 return '' unless defined $_[1];
1437 return ${$_[1]} if ref($_[1]) eq 'SCALAR';
1439 $_[0]->{quote_char} or
1440 ($_[0]->_assert_pass_injection_guard($_[1]), return $_[1]);
1442 my $qref = ref $_[0]->{quote_char};
1444 !$qref ? ($_[0]->{quote_char}, $_[0]->{quote_char})
1445 : ($qref eq 'ARRAY') ? @{$_[0]->{quote_char}}
1446 : puke "Unsupported quote_char format: $_[0]->{quote_char}";
1448 my $esc = $_[0]->{escape_char} || $r;
1450 # parts containing * are naturally unquoted
1451 return join($_[0]->{name_sep}||'', map
1452 +( $_ eq '*' ? $_ : do { (my $n = $_) =~ s/(\Q$esc\E|\Q$r\E)/$esc$1/g; $l . $n . $r } ),
1453 ( $_[0]->{name_sep} ? split (/\Q$_[0]->{name_sep}\E/, $_[1] ) : $_[1] )
1458 # Conversion, if applicable
1460 #my ($self, $arg) = @_;
1461 if ($_[0]->{convert}) {
1462 return $_[0]->_sqlcase($_[0]->{convert}) .'(' . $_[1] . ')';
1469 #my ($self, $col, @vals) = @_;
1470 # called often - tighten code
1471 return $_[0]->{bindtype} eq 'columns'
1472 ? map {[$_[1], $_]} @_[2 .. $#_]
1477 # Dies if any element of @bind is not in [colname => value] format
1478 # if bindtype is 'columns'.
1479 sub _assert_bindval_matches_bindtype {
1480 # my ($self, @bind) = @_;
1482 if ($self->{bindtype} eq 'columns') {
1484 if (!defined $_ || ref($_) ne 'ARRAY' || @$_ != 2) {
1485 puke "bindtype 'columns' selected, you need to pass: [column_name => bind_value]"
1491 sub _join_sql_clauses {
1492 my ($self, $logic, $clauses_aref, $bind_aref) = @_;
1494 if (@$clauses_aref > 1) {
1495 my $join = " " . $self->_sqlcase($logic) . " ";
1496 my $sql = '( ' . join($join, @$clauses_aref) . ' )';
1497 return ($sql, @$bind_aref);
1499 elsif (@$clauses_aref) {
1500 return ($clauses_aref->[0], @$bind_aref); # no parentheses
1503 return (); # if no SQL, ignore @$bind_aref
1508 # Fix SQL case, if so requested
1510 # LDNOTE: if $self->{case} is true, then it contains 'lower', so we
1511 # don't touch the argument ... crooked logic, but let's not change it!
1512 return $_[0]->{case} ? $_[1] : uc($_[1]);
1516 #======================================================================
1517 # DISPATCHING FROM REFKIND
1518 #======================================================================
1521 my ($self, $data) = @_;
1523 return 'UNDEF' unless defined $data;
1525 # blessed objects are treated like scalars
1526 my $ref = (Scalar::Util::blessed $data) ? '' : ref $data;
1528 return 'SCALAR' unless $ref;
1531 while ($ref eq 'REF') {
1533 $ref = (Scalar::Util::blessed $data) ? '' : ref $data;
1537 return ($ref||'SCALAR') . ('REF' x $n_steps);
1541 my ($self, $data) = @_;
1542 my @try = ($self->_refkind($data));
1543 push @try, 'SCALAR_or_UNDEF' if $try[0] eq 'SCALAR' || $try[0] eq 'UNDEF';
1544 push @try, 'FALLBACK';
1548 sub _METHOD_FOR_refkind {
1549 my ($self, $meth_prefix, $data) = @_;
1552 for (@{$self->_try_refkind($data)}) {
1553 $method = $self->can($meth_prefix."_".$_)
1557 return $method || puke "cannot dispatch on '$meth_prefix' for ".$self->_refkind($data);
1561 sub _SWITCH_refkind {
1562 my ($self, $data, $dispatch_table) = @_;
1565 for (@{$self->_try_refkind($data)}) {
1566 $coderef = $dispatch_table->{$_}
1570 puke "no dispatch entry for ".$self->_refkind($data)
1579 #======================================================================
1580 # VALUES, GENERATE, AUTOLOAD
1581 #======================================================================
1583 # LDNOTE: original code from nwiger, didn't touch code in that section
1584 # I feel the AUTOLOAD stuff should not be the default, it should
1585 # only be activated on explicit demand by user.
1589 my $data = shift || return;
1590 puke "Argument to ", __PACKAGE__, "->values must be a \\%hash"
1591 unless ref $data eq 'HASH';
1594 foreach my $k (sort keys %$data) {
1595 my $v = $data->{$k};
1596 $self->_SWITCH_refkind($v, {
1598 if ($self->{array_datatypes}) { # array datatype
1599 push @all_bind, $self->_bindtype($k, $v);
1601 else { # literal SQL with bind
1602 my ($sql, @bind) = @$v;
1603 $self->_assert_bindval_matches_bindtype(@bind);
1604 push @all_bind, @bind;
1607 ARRAYREFREF => sub { # literal SQL with bind
1608 my ($sql, @bind) = @${$v};
1609 $self->_assert_bindval_matches_bindtype(@bind);
1610 push @all_bind, @bind;
1612 SCALARREF => sub { # literal SQL without bind
1614 SCALAR_or_UNDEF => sub {
1615 push @all_bind, $self->_bindtype($k, $v);
1626 my(@sql, @sqlq, @sqlv);
1630 if ($ref eq 'HASH') {
1631 for my $k (sort keys %$_) {
1634 my $label = $self->_quote($k);
1635 if ($r eq 'ARRAY') {
1636 # literal SQL with bind
1637 my ($sql, @bind) = @$v;
1638 $self->_assert_bindval_matches_bindtype(@bind);
1639 push @sqlq, "$label = $sql";
1641 } elsif ($r eq 'SCALAR') {
1642 # literal SQL without bind
1643 push @sqlq, "$label = $$v";
1645 push @sqlq, "$label = ?";
1646 push @sqlv, $self->_bindtype($k, $v);
1649 push @sql, $self->_sqlcase('set'), join ', ', @sqlq;
1650 } elsif ($ref eq 'ARRAY') {
1651 # unlike insert(), assume these are ONLY the column names, i.e. for SQL
1654 if ($r eq 'ARRAY') { # literal SQL with bind
1655 my ($sql, @bind) = @$v;
1656 $self->_assert_bindval_matches_bindtype(@bind);
1659 } elsif ($r eq 'SCALAR') { # literal SQL without bind
1660 # embedded literal SQL
1667 push @sql, '(' . join(', ', @sqlq) . ')';
1668 } elsif ($ref eq 'SCALAR') {
1672 # strings get case twiddled
1673 push @sql, $self->_sqlcase($_);
1677 my $sql = join ' ', @sql;
1679 # this is pretty tricky
1680 # if ask for an array, return ($stmt, @bind)
1681 # otherwise, s/?/shift @sqlv/ to put it inline
1683 return ($sql, @sqlv);
1685 1 while $sql =~ s/\?/my $d = shift(@sqlv);
1686 ref $d ? $d->[1] : $d/e;
1695 # This allows us to check for a local, then _form, attr
1697 my($name) = $AUTOLOAD =~ /.*::(.+)/;
1698 return $self->generate($name, @_);
1709 SQL::Abstract - Generate SQL from Perl data structures
1715 my $sql = SQL::Abstract->new;
1717 my($stmt, @bind) = $sql->select($source, \@fields, \%where, $order);
1719 my($stmt, @bind) = $sql->insert($table, \%fieldvals || \@values);
1721 my($stmt, @bind) = $sql->update($table, \%fieldvals, \%where);
1723 my($stmt, @bind) = $sql->delete($table, \%where);
1725 # Then, use these in your DBI statements
1726 my $sth = $dbh->prepare($stmt);
1727 $sth->execute(@bind);
1729 # Just generate the WHERE clause
1730 my($stmt, @bind) = $sql->where(\%where, $order);
1732 # Return values in the same order, for hashed queries
1733 # See PERFORMANCE section for more details
1734 my @bind = $sql->values(\%fieldvals);
1738 This module was inspired by the excellent L<DBIx::Abstract>.
1739 However, in using that module I found that what I really wanted
1740 to do was generate SQL, but still retain complete control over my
1741 statement handles and use the DBI interface. So, I set out to
1742 create an abstract SQL generation module.
1744 While based on the concepts used by L<DBIx::Abstract>, there are
1745 several important differences, especially when it comes to WHERE
1746 clauses. I have modified the concepts used to make the SQL easier
1747 to generate from Perl data structures and, IMO, more intuitive.
1748 The underlying idea is for this module to do what you mean, based
1749 on the data structures you provide it. The big advantage is that
1750 you don't have to modify your code every time your data changes,
1751 as this module figures it out.
1753 To begin with, an SQL INSERT is as easy as just specifying a hash
1754 of C<key=value> pairs:
1757 name => 'Jimbo Bobson',
1758 phone => '123-456-7890',
1759 address => '42 Sister Lane',
1760 city => 'St. Louis',
1761 state => 'Louisiana',
1764 The SQL can then be generated with this:
1766 my($stmt, @bind) = $sql->insert('people', \%data);
1768 Which would give you something like this:
1770 $stmt = "INSERT INTO people
1771 (address, city, name, phone, state)
1772 VALUES (?, ?, ?, ?, ?)";
1773 @bind = ('42 Sister Lane', 'St. Louis', 'Jimbo Bobson',
1774 '123-456-7890', 'Louisiana');
1776 These are then used directly in your DBI code:
1778 my $sth = $dbh->prepare($stmt);
1779 $sth->execute(@bind);
1781 =head2 Inserting and Updating Arrays
1783 If your database has array types (like for example Postgres),
1784 activate the special option C<< array_datatypes => 1 >>
1785 when creating the C<SQL::Abstract> object.
1786 Then you may use an arrayref to insert and update database array types:
1788 my $sql = SQL::Abstract->new(array_datatypes => 1);
1790 planets => [qw/Mercury Venus Earth Mars/]
1793 my($stmt, @bind) = $sql->insert('solar_system', \%data);
1797 $stmt = "INSERT INTO solar_system (planets) VALUES (?)"
1799 @bind = (['Mercury', 'Venus', 'Earth', 'Mars']);
1802 =head2 Inserting and Updating SQL
1804 In order to apply SQL functions to elements of your C<%data> you may
1805 specify a reference to an arrayref for the given hash value. For example,
1806 if you need to execute the Oracle C<to_date> function on a value, you can
1807 say something like this:
1811 date_entered => \[ "to_date(?,'MM/DD/YYYY')", "03/02/2003" ],
1814 The first value in the array is the actual SQL. Any other values are
1815 optional and would be included in the bind values array. This gives
1818 my($stmt, @bind) = $sql->insert('people', \%data);
1820 $stmt = "INSERT INTO people (name, date_entered)
1821 VALUES (?, to_date(?,'MM/DD/YYYY'))";
1822 @bind = ('Bill', '03/02/2003');
1824 An UPDATE is just as easy, all you change is the name of the function:
1826 my($stmt, @bind) = $sql->update('people', \%data);
1828 Notice that your C<%data> isn't touched; the module will generate
1829 the appropriately quirky SQL for you automatically. Usually you'll
1830 want to specify a WHERE clause for your UPDATE, though, which is
1831 where handling C<%where> hashes comes in handy...
1833 =head2 Complex where statements
1835 This module can generate pretty complicated WHERE statements
1836 easily. For example, simple C<key=value> pairs are taken to mean
1837 equality, and if you want to see if a field is within a set
1838 of values, you can use an arrayref. Let's say we wanted to
1839 SELECT some data based on this criteria:
1842 requestor => 'inna',
1843 worker => ['nwiger', 'rcwe', 'sfz'],
1844 status => { '!=', 'completed' }
1847 my($stmt, @bind) = $sql->select('tickets', '*', \%where);
1849 The above would give you something like this:
1851 $stmt = "SELECT * FROM tickets WHERE
1852 ( requestor = ? ) AND ( status != ? )
1853 AND ( worker = ? OR worker = ? OR worker = ? )";
1854 @bind = ('inna', 'completed', 'nwiger', 'rcwe', 'sfz');
1856 Which you could then use in DBI code like so:
1858 my $sth = $dbh->prepare($stmt);
1859 $sth->execute(@bind);
1865 The methods are simple. There's one for every major SQL operation,
1866 and a constructor you use first. The arguments are specified in a
1867 similar order for each method (table, then fields, then a where
1868 clause) to try and simplify things.
1870 =head2 new(option => 'value')
1872 The C<new()> function takes a list of options and values, and returns
1873 a new B<SQL::Abstract> object which can then be used to generate SQL
1874 through the methods below. The options accepted are:
1880 If set to 'lower', then SQL will be generated in all lowercase. By
1881 default SQL is generated in "textbook" case meaning something like:
1883 SELECT a_field FROM a_table WHERE some_field LIKE '%someval%'
1885 Any setting other than 'lower' is ignored.
1889 This determines what the default comparison operator is. By default
1890 it is C<=>, meaning that a hash like this:
1892 %where = (name => 'nwiger', email => 'nate@wiger.org');
1894 Will generate SQL like this:
1896 WHERE name = 'nwiger' AND email = 'nate@wiger.org'
1898 However, you may want loose comparisons by default, so if you set
1899 C<cmp> to C<like> you would get SQL such as:
1901 WHERE name like 'nwiger' AND email like 'nate@wiger.org'
1903 You can also override the comparison on an individual basis - see
1904 the huge section on L</"WHERE CLAUSES"> at the bottom.
1906 =item sqltrue, sqlfalse
1908 Expressions for inserting boolean values within SQL statements.
1909 By default these are C<1=1> and C<1=0>. They are used
1910 by the special operators C<-in> and C<-not_in> for generating
1911 correct SQL even when the argument is an empty array (see below).
1915 This determines the default logical operator for multiple WHERE
1916 statements in arrays or hashes. If absent, the default logic is "or"
1917 for arrays, and "and" for hashes. This means that a WHERE
1921 event_date => {'>=', '2/13/99'},
1922 event_date => {'<=', '4/24/03'},
1925 will generate SQL like this:
1927 WHERE event_date >= '2/13/99' OR event_date <= '4/24/03'
1929 This is probably not what you want given this query, though (look
1930 at the dates). To change the "OR" to an "AND", simply specify:
1932 my $sql = SQL::Abstract->new(logic => 'and');
1934 Which will change the above C<WHERE> to:
1936 WHERE event_date >= '2/13/99' AND event_date <= '4/24/03'
1938 The logic can also be changed locally by inserting
1939 a modifier in front of an arrayref:
1941 @where = (-and => [event_date => {'>=', '2/13/99'},
1942 event_date => {'<=', '4/24/03'} ]);
1944 See the L</"WHERE CLAUSES"> section for explanations.
1948 This will automatically convert comparisons using the specified SQL
1949 function for both column and value. This is mostly used with an argument
1950 of C<upper> or C<lower>, so that the SQL will have the effect of
1951 case-insensitive "searches". For example, this:
1953 $sql = SQL::Abstract->new(convert => 'upper');
1954 %where = (keywords => 'MaKe iT CAse inSeNSItive');
1956 Will turn out the following SQL:
1958 WHERE upper(keywords) like upper('MaKe iT CAse inSeNSItive')
1960 The conversion can be C<upper()>, C<lower()>, or any other SQL function
1961 that can be applied symmetrically to fields (actually B<SQL::Abstract> does
1962 not validate this option; it will just pass through what you specify verbatim).
1966 This is a kludge because many databases suck. For example, you can't
1967 just bind values using DBI's C<execute()> for Oracle C<CLOB> or C<BLOB> fields.
1968 Instead, you have to use C<bind_param()>:
1970 $sth->bind_param(1, 'reg data');
1971 $sth->bind_param(2, $lots, {ora_type => ORA_CLOB});
1973 The problem is, B<SQL::Abstract> will normally just return a C<@bind> array,
1974 which loses track of which field each slot refers to. Fear not.
1976 If you specify C<bindtype> in new, you can determine how C<@bind> is returned.
1977 Currently, you can specify either C<normal> (default) or C<columns>. If you
1978 specify C<columns>, you will get an array that looks like this:
1980 my $sql = SQL::Abstract->new(bindtype => 'columns');
1981 my($stmt, @bind) = $sql->insert(...);
1984 [ 'column1', 'value1' ],
1985 [ 'column2', 'value2' ],
1986 [ 'column3', 'value3' ],
1989 You can then iterate through this manually, using DBI's C<bind_param()>.
1991 $sth->prepare($stmt);
1994 my($col, $data) = @$_;
1995 if ($col eq 'details' || $col eq 'comments') {
1996 $sth->bind_param($i, $data, {ora_type => ORA_CLOB});
1997 } elsif ($col eq 'image') {
1998 $sth->bind_param($i, $data, {ora_type => ORA_BLOB});
2000 $sth->bind_param($i, $data);
2004 $sth->execute; # execute without @bind now
2006 Now, why would you still use B<SQL::Abstract> if you have to do this crap?
2007 Basically, the advantage is still that you don't have to care which fields
2008 are or are not included. You could wrap that above C<for> loop in a simple
2009 sub called C<bind_fields()> or something and reuse it repeatedly. You still
2010 get a layer of abstraction over manual SQL specification.
2012 Note that if you set L</bindtype> to C<columns>, the C<\[ $sql, @bind ]>
2013 construct (see L</Literal SQL with placeholders and bind values (subqueries)>)
2014 will expect the bind values in this format.
2018 This is the character that a table or column name will be quoted
2019 with. By default this is an empty string, but you could set it to
2020 the character C<`>, to generate SQL like this:
2022 SELECT `a_field` FROM `a_table` WHERE `some_field` LIKE '%someval%'
2024 Alternatively, you can supply an array ref of two items, the first being the left
2025 hand quote character, and the second the right hand quote character. For
2026 example, you could supply C<['[',']']> for SQL Server 2000 compliant quotes
2027 that generates SQL like this:
2029 SELECT [a_field] FROM [a_table] WHERE [some_field] LIKE '%someval%'
2031 Quoting is useful if you have tables or columns names that are reserved
2032 words in your database's SQL dialect.
2036 This is the character that will be used to escape L</quote_char>s appearing
2037 in an identifier before it has been quoted.
2039 The parameter default in case of a single L</quote_char> character is the quote
2042 When opening-closing-style quoting is used (L</quote_char> is an arrayref)
2043 this parameter defaults to the B<closing (right)> L</quote_char>. Occurrences
2044 of the B<opening (left)> L</quote_char> within the identifier are currently left
2045 untouched. The default for opening-closing-style quotes may change in future
2046 versions, thus you are B<strongly encouraged> to specify the escape character
2051 This is the character that separates a table and column name. It is
2052 necessary to specify this when the C<quote_char> option is selected,
2053 so that tables and column names can be individually quoted like this:
2055 SELECT `table`.`one_field` FROM `table` WHERE `table`.`other_field` = 1
2057 =item injection_guard
2059 A regular expression C<qr/.../> that is applied to any C<-function> and unquoted
2060 column name specified in a query structure. This is a safety mechanism to avoid
2061 injection attacks when mishandling user input e.g.:
2063 my %condition_as_column_value_pairs = get_values_from_user();
2064 $sqla->select( ... , \%condition_as_column_value_pairs );
2066 If the expression matches an exception is thrown. Note that literal SQL
2067 supplied via C<\'...'> or C<\['...']> is B<not> checked in any way.
2069 Defaults to checking for C<;> and the C<GO> keyword (TransactSQL)
2071 =item array_datatypes
2073 When this option is true, arrayrefs in INSERT or UPDATE are
2074 interpreted as array datatypes and are passed directly
2076 When this option is false, arrayrefs are interpreted
2077 as literal SQL, just like refs to arrayrefs
2078 (but this behavior is for backwards compatibility; when writing
2079 new queries, use the "reference to arrayref" syntax
2085 Takes a reference to a list of "special operators"
2086 to extend the syntax understood by L<SQL::Abstract>.
2087 See section L</"SPECIAL OPERATORS"> for details.
2091 Takes a reference to a list of "unary operators"
2092 to extend the syntax understood by L<SQL::Abstract>.
2093 See section L</"UNARY OPERATORS"> for details.
2099 =head2 insert($table, \@values || \%fieldvals, \%options)
2101 This is the simplest function. You simply give it a table name
2102 and either an arrayref of values or hashref of field/value pairs.
2103 It returns an SQL INSERT statement and a list of bind values.
2104 See the sections on L</"Inserting and Updating Arrays"> and
2105 L</"Inserting and Updating SQL"> for information on how to insert
2106 with those data types.
2108 The optional C<\%options> hash reference may contain additional
2109 options to generate the insert SQL. Currently supported options
2116 Takes either a scalar of raw SQL fields, or an array reference of
2117 field names, and adds on an SQL C<RETURNING> statement at the end.
2118 This allows you to return data generated by the insert statement
2119 (such as row IDs) without performing another C<SELECT> statement.
2120 Note, however, this is not part of the SQL standard and may not
2121 be supported by all database engines.
2125 =head2 update($table, \%fieldvals, \%where, \%options)
2127 This takes a table, hashref of field/value pairs, and an optional
2128 hashref L<WHERE clause|/WHERE CLAUSES>. It returns an SQL UPDATE function and a list
2130 See the sections on L</"Inserting and Updating Arrays"> and
2131 L</"Inserting and Updating SQL"> for information on how to insert
2132 with those data types.
2134 The optional C<\%options> hash reference may contain additional
2135 options to generate the update SQL. Currently supported options
2142 See the C<returning> option to
2143 L<insert|/insert($table, \@values || \%fieldvals, \%options)>.
2147 =head2 select($source, $fields, $where, $order)
2149 This returns a SQL SELECT statement and associated list of bind values, as
2150 specified by the arguments:
2156 Specification of the 'FROM' part of the statement.
2157 The argument can be either a plain scalar (interpreted as a table
2158 name, will be quoted), or an arrayref (interpreted as a list
2159 of table names, joined by commas, quoted), or a scalarref
2160 (literal SQL, not quoted).
2164 Specification of the list of fields to retrieve from
2166 The argument can be either an arrayref (interpreted as a list
2167 of field names, will be joined by commas and quoted), or a
2168 plain scalar (literal SQL, not quoted).
2169 Please observe that this API is not as flexible as that of
2170 the first argument C<$source>, for backwards compatibility reasons.
2174 Optional argument to specify the WHERE part of the query.
2175 The argument is most often a hashref, but can also be
2176 an arrayref or plain scalar --
2177 see section L<WHERE clause|/"WHERE CLAUSES"> for details.
2181 Optional argument to specify the ORDER BY part of the query.
2182 The argument can be a scalar, a hashref or an arrayref
2183 -- see section L<ORDER BY clause|/"ORDER BY CLAUSES">
2189 =head2 delete($table, \%where, \%options)
2191 This takes a table name and optional hashref L<WHERE clause|/WHERE CLAUSES>.
2192 It returns an SQL DELETE statement and list of bind values.
2194 The optional C<\%options> hash reference may contain additional
2195 options to generate the delete SQL. Currently supported options
2202 See the C<returning> option to
2203 L<insert|/insert($table, \@values || \%fieldvals, \%options)>.
2207 =head2 where(\%where, $order)
2209 This is used to generate just the WHERE clause. For example,
2210 if you have an arbitrary data structure and know what the
2211 rest of your SQL is going to look like, but want an easy way
2212 to produce a WHERE clause, use this. It returns an SQL WHERE
2213 clause and list of bind values.
2216 =head2 values(\%data)
2218 This just returns the values from the hash C<%data>, in the same
2219 order that would be returned from any of the other above queries.
2220 Using this allows you to markedly speed up your queries if you
2221 are affecting lots of rows. See below under the L</"PERFORMANCE"> section.
2223 =head2 generate($any, 'number', $of, \@data, $struct, \%types)
2225 Warning: This is an experimental method and subject to change.
2227 This returns arbitrarily generated SQL. It's a really basic shortcut.
2228 It will return two different things, depending on return context:
2230 my($stmt, @bind) = $sql->generate('create table', \$table, \@fields);
2231 my $stmt_and_val = $sql->generate('create table', \$table, \@fields);
2233 These would return the following:
2235 # First calling form
2236 $stmt = "CREATE TABLE test (?, ?)";
2237 @bind = (field1, field2);
2239 # Second calling form
2240 $stmt_and_val = "CREATE TABLE test (field1, field2)";
2242 Depending on what you're trying to do, it's up to you to choose the correct
2243 format. In this example, the second form is what you would want.
2247 $sql->generate('alter session', { nls_date_format => 'MM/YY' });
2251 ALTER SESSION SET nls_date_format = 'MM/YY'
2253 You get the idea. Strings get their case twiddled, but everything
2254 else remains verbatim.
2256 =head1 EXPORTABLE FUNCTIONS
2258 =head2 is_plain_value
2260 Determines if the supplied argument is a plain value as understood by this
2265 =item * The value is C<undef>
2267 =item * The value is a non-reference
2269 =item * The value is an object with stringification overloading
2271 =item * The value is of the form C<< { -value => $anything } >>
2275 On failure returns C<undef>, on success returns a B<scalar> reference
2276 to the original supplied argument.
2282 The stringification overloading detection is rather advanced: it takes
2283 into consideration not only the presence of a C<""> overload, but if that
2284 fails also checks for enabled
2285 L<autogenerated versions of C<"">|overload/Magic Autogeneration>, based
2286 on either C<0+> or C<bool>.
2288 Unfortunately testing in the field indicates that this
2289 detection B<< may tickle a latent bug in perl versions before 5.018 >>,
2290 but only when very large numbers of stringifying objects are involved.
2291 At the time of writing ( Sep 2014 ) there is no clear explanation of
2292 the direct cause, nor is there a manageably small test case that reliably
2293 reproduces the problem.
2295 If you encounter any of the following exceptions in B<random places within
2296 your application stack> - this module may be to blame:
2298 Operation "ne": no method found,
2299 left argument in overloaded package <something>,
2300 right argument in overloaded package <something>
2304 Stub found while resolving method "???" overloading """" in package <something>
2306 If you fall victim to the above - please attempt to reduce the problem
2307 to something that could be sent to the L<SQL::Abstract developers
2308 |DBIx::Class/GETTING HELP/SUPPORT>
2309 (either publicly or privately). As a workaround in the meantime you can
2310 set C<$ENV{SQLA_ISVALUE_IGNORE_AUTOGENERATED_STRINGIFICATION}> to a true
2311 value, which will most likely eliminate your problem (at the expense of
2312 not being able to properly detect exotic forms of stringification).
2314 This notice and environment variable will be removed in a future version,
2315 as soon as the underlying problem is found and a reliable workaround is
2320 =head2 is_literal_value
2322 Determines if the supplied argument is a literal value as understood by this
2327 =item * C<\$sql_string>
2329 =item * C<\[ $sql_string, @bind_values ]>
2333 On failure returns C<undef>, on success returns an B<array> reference
2334 containing the unpacked version of the supplied literal SQL and bind values.
2336 =head1 WHERE CLAUSES
2340 This module uses a variation on the idea from L<DBIx::Abstract>. It
2341 is B<NOT>, repeat I<not> 100% compatible. B<The main logic of this
2342 module is that things in arrays are OR'ed, and things in hashes
2345 The easiest way to explain is to show lots of examples. After
2346 each C<%where> hash shown, it is assumed you used:
2348 my($stmt, @bind) = $sql->where(\%where);
2350 However, note that the C<%where> hash can be used directly in any
2351 of the other functions as well, as described above.
2353 =head2 Key-value pairs
2355 So, let's get started. To begin, a simple hash:
2359 status => 'completed'
2362 Is converted to SQL C<key = val> statements:
2364 $stmt = "WHERE user = ? AND status = ?";
2365 @bind = ('nwiger', 'completed');
2367 One common thing I end up doing is having a list of values that
2368 a field can be in. To do this, simply specify a list inside of
2373 status => ['assigned', 'in-progress', 'pending'];
2376 This simple code will create the following:
2378 $stmt = "WHERE user = ? AND ( status = ? OR status = ? OR status = ? )";
2379 @bind = ('nwiger', 'assigned', 'in-progress', 'pending');
2381 A field associated to an empty arrayref will be considered a
2382 logical false and will generate 0=1.
2384 =head2 Tests for NULL values
2386 If the value part is C<undef> then this is converted to SQL <IS NULL>
2395 $stmt = "WHERE user = ? AND status IS NULL";
2398 To test if a column IS NOT NULL:
2402 status => { '!=', undef },
2405 =head2 Specific comparison operators
2407 If you want to specify a different type of operator for your comparison,
2408 you can use a hashref for a given column:
2412 status => { '!=', 'completed' }
2415 Which would generate:
2417 $stmt = "WHERE user = ? AND status != ?";
2418 @bind = ('nwiger', 'completed');
2420 To test against multiple values, just enclose the values in an arrayref:
2422 status => { '=', ['assigned', 'in-progress', 'pending'] };
2424 Which would give you:
2426 "WHERE status = ? OR status = ? OR status = ?"
2429 The hashref can also contain multiple pairs, in which case it is expanded
2430 into an C<AND> of its elements:
2434 status => { '!=', 'completed', -not_like => 'pending%' }
2437 # Or more dynamically, like from a form
2438 $where{user} = 'nwiger';
2439 $where{status}{'!='} = 'completed';
2440 $where{status}{'-not_like'} = 'pending%';
2442 # Both generate this
2443 $stmt = "WHERE user = ? AND status != ? AND status NOT LIKE ?";
2444 @bind = ('nwiger', 'completed', 'pending%');
2447 To get an OR instead, you can combine it with the arrayref idea:
2451 priority => [ { '=', 2 }, { '>', 5 } ]
2454 Which would generate:
2456 $stmt = "WHERE ( priority = ? OR priority > ? ) AND user = ?";
2457 @bind = ('2', '5', 'nwiger');
2459 If you want to include literal SQL (with or without bind values), just use a
2460 scalar reference or reference to an arrayref as the value:
2463 date_entered => { '>' => \["to_date(?, 'MM/DD/YYYY')", "11/26/2008"] },
2464 date_expires => { '<' => \"now()" }
2467 Which would generate:
2469 $stmt = "WHERE date_entered > to_date(?, 'MM/DD/YYYY') AND date_expires < now()";
2470 @bind = ('11/26/2008');
2473 =head2 Logic and nesting operators
2475 In the example above,
2476 there is a subtle trap if you want to say something like
2477 this (notice the C<AND>):
2479 WHERE priority != ? AND priority != ?
2481 Because, in Perl you I<can't> do this:
2483 priority => { '!=' => 2, '!=' => 1 }
2485 As the second C<!=> key will obliterate the first. The solution
2486 is to use the special C<-modifier> form inside an arrayref:
2488 priority => [ -and => {'!=', 2},
2492 Normally, these would be joined by C<OR>, but the modifier tells it
2493 to use C<AND> instead. (Hint: You can use this in conjunction with the
2494 C<logic> option to C<new()> in order to change the way your queries
2495 work by default.) B<Important:> Note that the C<-modifier> goes
2496 B<INSIDE> the arrayref, as an extra first element. This will
2497 B<NOT> do what you think it might:
2499 priority => -and => [{'!=', 2}, {'!=', 1}] # WRONG!
2501 Here is a quick list of equivalencies, since there is some overlap:
2504 status => {'!=', 'completed', 'not like', 'pending%' }
2505 status => [ -and => {'!=', 'completed'}, {'not like', 'pending%'}]
2508 status => {'=', ['assigned', 'in-progress']}
2509 status => [ -or => {'=', 'assigned'}, {'=', 'in-progress'}]
2510 status => [ {'=', 'assigned'}, {'=', 'in-progress'} ]
2514 =head2 Special operators: IN, BETWEEN, etc.
2516 You can also use the hashref format to compare a list of fields using the
2517 C<IN> comparison operator, by specifying the list as an arrayref:
2520 status => 'completed',
2521 reportid => { -in => [567, 2335, 2] }
2524 Which would generate:
2526 $stmt = "WHERE status = ? AND reportid IN (?,?,?)";
2527 @bind = ('completed', '567', '2335', '2');
2529 The reverse operator C<-not_in> generates SQL C<NOT IN> and is used in
2532 If the argument to C<-in> is an empty array, 'sqlfalse' is generated
2533 (by default: C<1=0>). Similarly, C<< -not_in => [] >> generates
2534 'sqltrue' (by default: C<1=1>).
2536 In addition to the array you can supply a chunk of literal sql or
2537 literal sql with bind:
2540 customer => { -in => \[
2541 'SELECT cust_id FROM cust WHERE balance > ?',
2544 status => { -in => \'SELECT status_codes FROM states' },
2550 customer IN ( SELECT cust_id FROM cust WHERE balance > ? )
2551 AND status IN ( SELECT status_codes FROM states )
2555 Finally, if the argument to C<-in> is not a reference, it will be
2556 treated as a single-element array.
2558 Another pair of operators is C<-between> and C<-not_between>,
2559 used with an arrayref of two values:
2563 completion_date => {
2564 -not_between => ['2002-10-01', '2003-02-06']
2570 WHERE user = ? AND completion_date NOT BETWEEN ( ? AND ? )
2572 Just like with C<-in> all plausible combinations of literal SQL
2576 start0 => { -between => [ 1, 2 ] },
2577 start1 => { -between => \["? AND ?", 1, 2] },
2578 start2 => { -between => \"lower(x) AND upper(y)" },
2579 start3 => { -between => [
2581 \["upper(?)", 'stuff' ],
2588 ( start0 BETWEEN ? AND ? )
2589 AND ( start1 BETWEEN ? AND ? )
2590 AND ( start2 BETWEEN lower(x) AND upper(y) )
2591 AND ( start3 BETWEEN lower(x) AND upper(?) )
2593 @bind = (1, 2, 1, 2, 'stuff');
2596 These are the two builtin "special operators"; but the
2597 list can be expanded: see section L</"SPECIAL OPERATORS"> below.
2599 =head2 Unary operators: bool
2601 If you wish to test against boolean columns or functions within your
2602 database you can use the C<-bool> and C<-not_bool> operators. For
2603 example to test the column C<is_user> being true and the column
2604 C<is_enabled> being false you would use:-
2608 -not_bool => 'is_enabled',
2613 WHERE is_user AND NOT is_enabled
2615 If a more complex combination is required, testing more conditions,
2616 then you should use the and/or operators:-
2621 -not_bool => { two=> { -rlike => 'bar' } },
2622 -not_bool => { three => [ { '=', 2 }, { '>', 5 } ] },
2633 (NOT ( three = ? OR three > ? ))
2636 =head2 Nested conditions, -and/-or prefixes
2638 So far, we've seen how multiple conditions are joined with a top-level
2639 C<AND>. We can change this by putting the different conditions we want in
2640 hashes and then putting those hashes in an array. For example:
2645 status => { -like => ['pending%', 'dispatched'] },
2649 status => 'unassigned',
2653 This data structure would create the following:
2655 $stmt = "WHERE ( user = ? AND ( status LIKE ? OR status LIKE ? ) )
2656 OR ( user = ? AND status = ? ) )";
2657 @bind = ('nwiger', 'pending', 'dispatched', 'robot', 'unassigned');
2660 Clauses in hashrefs or arrayrefs can be prefixed with an C<-and> or C<-or>
2661 to change the logic inside:
2667 -and => [ workhrs => {'>', 20}, geo => 'ASIA' ],
2668 -or => { workhrs => {'<', 50}, geo => 'EURO' },
2675 $stmt = "WHERE ( user = ?
2676 AND ( ( workhrs > ? AND geo = ? )
2677 OR ( workhrs < ? OR geo = ? ) ) )";
2678 @bind = ('nwiger', '20', 'ASIA', '50', 'EURO');
2680 =head3 Algebraic inconsistency, for historical reasons
2682 C<Important note>: when connecting several conditions, the C<-and->|C<-or>
2683 operator goes C<outside> of the nested structure; whereas when connecting
2684 several constraints on one column, the C<-and> operator goes
2685 C<inside> the arrayref. Here is an example combining both features:
2688 -and => [a => 1, b => 2],
2689 -or => [c => 3, d => 4],
2690 e => [-and => {-like => 'foo%'}, {-like => '%bar'} ]
2695 WHERE ( ( ( a = ? AND b = ? )
2696 OR ( c = ? OR d = ? )
2697 OR ( e LIKE ? AND e LIKE ? ) ) )
2699 This difference in syntax is unfortunate but must be preserved for
2700 historical reasons. So be careful: the two examples below would
2701 seem algebraically equivalent, but they are not
2704 { -like => 'foo%' },
2705 { -like => '%bar' },
2707 # yields: WHERE ( ( col LIKE ? AND col LIKE ? ) )
2710 { col => { -like => 'foo%' } },
2711 { col => { -like => '%bar' } },
2713 # yields: WHERE ( ( col LIKE ? OR col LIKE ? ) )
2716 =head2 Literal SQL and value type operators
2718 The basic premise of SQL::Abstract is that in WHERE specifications the "left
2719 side" is a column name and the "right side" is a value (normally rendered as
2720 a placeholder). This holds true for both hashrefs and arrayref pairs as you
2721 see in the L</WHERE CLAUSES> examples above. Sometimes it is necessary to
2722 alter this behavior. There are several ways of doing so.
2726 This is a virtual operator that signals the string to its right side is an
2727 identifier (a column name) and not a value. For example to compare two
2728 columns you would write:
2731 priority => { '<', 2 },
2732 requestor => { -ident => 'submitter' },
2737 $stmt = "WHERE priority < ? AND requestor = submitter";
2740 If you are maintaining legacy code you may see a different construct as
2741 described in L</Deprecated usage of Literal SQL>, please use C<-ident> in new
2746 This is a virtual operator that signals that the construct to its right side
2747 is a value to be passed to DBI. This is for example necessary when you want
2748 to write a where clause against an array (for RDBMS that support such
2749 datatypes). For example:
2752 array => { -value => [1, 2, 3] }
2757 $stmt = 'WHERE array = ?';
2758 @bind = ([1, 2, 3]);
2760 Note that if you were to simply say:
2766 the result would probably not be what you wanted:
2768 $stmt = 'WHERE array = ? OR array = ? OR array = ?';
2773 Finally, sometimes only literal SQL will do. To include a random snippet
2774 of SQL verbatim, you specify it as a scalar reference. Consider this only
2775 as a last resort. Usually there is a better way. For example:
2778 priority => { '<', 2 },
2779 requestor => { -in => \'(SELECT name FROM hitmen)' },
2784 $stmt = "WHERE priority < ? AND requestor IN (SELECT name FROM hitmen)"
2787 Note that in this example, you only get one bind parameter back, since
2788 the verbatim SQL is passed as part of the statement.
2792 Never use untrusted input as a literal SQL argument - this is a massive
2793 security risk (there is no way to check literal snippets for SQL
2794 injections and other nastyness). If you need to deal with untrusted input
2795 use literal SQL with placeholders as described next.
2797 =head3 Literal SQL with placeholders and bind values (subqueries)
2799 If the literal SQL to be inserted has placeholders and bind values,
2800 use a reference to an arrayref (yes this is a double reference --
2801 not so common, but perfectly legal Perl). For example, to find a date
2802 in Postgres you can use something like this:
2805 date_column => \[ "= date '2008-09-30' - ?::integer", 10 ]
2810 $stmt = "WHERE ( date_column = date '2008-09-30' - ?::integer )"
2813 Note that you must pass the bind values in the same format as they are returned
2814 by L<where|/where(\%where, $order)>. This means that if you set L</bindtype>
2815 to C<columns>, you must provide the bind values in the
2816 C<< [ column_meta => value ] >> format, where C<column_meta> is an opaque
2817 scalar value; most commonly the column name, but you can use any scalar value
2818 (including references and blessed references), L<SQL::Abstract> will simply
2819 pass it through intact. So if C<bindtype> is set to C<columns> the above
2820 example will look like:
2823 date_column => \[ "= date '2008-09-30' - ?::integer", [ {} => 10 ] ]
2826 Literal SQL is especially useful for nesting parenthesized clauses in the
2827 main SQL query. Here is a first example:
2829 my ($sub_stmt, @sub_bind) = ("SELECT c1 FROM t1 WHERE c2 < ? AND c3 LIKE ?",
2833 bar => \["IN ($sub_stmt)" => @sub_bind],
2838 $stmt = "WHERE (foo = ? AND bar IN (SELECT c1 FROM t1
2839 WHERE c2 < ? AND c3 LIKE ?))";
2840 @bind = (1234, 100, "foo%");
2842 Other subquery operators, like for example C<"E<gt> ALL"> or C<"NOT IN">,
2843 are expressed in the same way. Of course the C<$sub_stmt> and
2844 its associated bind values can be generated through a former call
2847 my ($sub_stmt, @sub_bind)
2848 = $sql->select("t1", "c1", {c2 => {"<" => 100},
2849 c3 => {-like => "foo%"}});
2852 bar => \["> ALL ($sub_stmt)" => @sub_bind],
2855 In the examples above, the subquery was used as an operator on a column;
2856 but the same principle also applies for a clause within the main C<%where>
2857 hash, like an EXISTS subquery:
2859 my ($sub_stmt, @sub_bind)
2860 = $sql->select("t1", "*", {c1 => 1, c2 => \"> t0.c0"});
2861 my %where = ( -and => [
2863 \["EXISTS ($sub_stmt)" => @sub_bind],
2868 $stmt = "WHERE (foo = ? AND EXISTS (SELECT * FROM t1
2869 WHERE c1 = ? AND c2 > t0.c0))";
2873 Observe that the condition on C<c2> in the subquery refers to
2874 column C<t0.c0> of the main query: this is I<not> a bind
2875 value, so we have to express it through a scalar ref.
2876 Writing C<< c2 => {">" => "t0.c0"} >> would have generated
2877 C<< c2 > ? >> with bind value C<"t0.c0"> ... not exactly
2878 what we wanted here.
2880 Finally, here is an example where a subquery is used
2881 for expressing unary negation:
2883 my ($sub_stmt, @sub_bind)
2884 = $sql->where({age => [{"<" => 10}, {">" => 20}]});
2885 $sub_stmt =~ s/^ where //i; # don't want "WHERE" in the subclause
2887 lname => {like => '%son%'},
2888 \["NOT ($sub_stmt)" => @sub_bind],
2893 $stmt = "lname LIKE ? AND NOT ( age < ? OR age > ? )"
2894 @bind = ('%son%', 10, 20)
2896 =head3 Deprecated usage of Literal SQL
2898 Below are some examples of archaic use of literal SQL. It is shown only as
2899 reference for those who deal with legacy code. Each example has a much
2900 better, cleaner and safer alternative that users should opt for in new code.
2906 my %where = ( requestor => \'IS NOT NULL' )
2908 $stmt = "WHERE requestor IS NOT NULL"
2910 This used to be the way of generating NULL comparisons, before the handling
2911 of C<undef> got formalized. For new code please use the superior syntax as
2912 described in L</Tests for NULL values>.
2916 my %where = ( requestor => \'= submitter' )
2918 $stmt = "WHERE requestor = submitter"
2920 This used to be the only way to compare columns. Use the superior L</-ident>
2921 method for all new code. For example an identifier declared in such a way
2922 will be properly quoted if L</quote_char> is properly set, while the legacy
2923 form will remain as supplied.
2927 my %where = ( is_ready => \"", completed => { '>', '2012-12-21' } )
2929 $stmt = "WHERE completed > ? AND is_ready"
2930 @bind = ('2012-12-21')
2932 Using an empty string literal used to be the only way to express a boolean.
2933 For all new code please use the much more readable
2934 L<-bool|/Unary operators: bool> operator.
2940 These pages could go on for a while, since the nesting of the data
2941 structures this module can handle are pretty much unlimited (the
2942 module implements the C<WHERE> expansion as a recursive function
2943 internally). Your best bet is to "play around" with the module a
2944 little to see how the data structures behave, and choose the best
2945 format for your data based on that.
2947 And of course, all the values above will probably be replaced with
2948 variables gotten from forms or the command line. After all, if you
2949 knew everything ahead of time, you wouldn't have to worry about
2950 dynamically-generating SQL and could just hardwire it into your
2953 =head1 ORDER BY CLAUSES
2955 Some functions take an order by clause. This can either be a scalar (just a
2956 column name), a hashref of C<< { -desc => 'col' } >> or C<< { -asc => 'col' }
2957 >>, a scalarref, an arrayref-ref, or an arrayref of any of the previous
2960 Given | Will Generate
2961 ---------------------------------------------------------------
2963 'colA' | ORDER BY colA
2965 [qw/colA colB/] | ORDER BY colA, colB
2967 {-asc => 'colA'} | ORDER BY colA ASC
2969 {-desc => 'colB'} | ORDER BY colB DESC
2971 ['colA', {-asc => 'colB'}] | ORDER BY colA, colB ASC
2973 { -asc => [qw/colA colB/] } | ORDER BY colA ASC, colB ASC
2975 \'colA DESC' | ORDER BY colA DESC
2977 \[ 'FUNC(colA, ?)', $x ] | ORDER BY FUNC(colA, ?)
2978 | /* ...with $x bound to ? */
2981 { -asc => 'colA' }, | colA ASC,
2982 { -desc => [qw/colB/] }, | colB DESC,
2983 { -asc => [qw/colC colD/] },| colC ASC, colD ASC,
2984 \'colE DESC', | colE DESC,
2985 \[ 'FUNC(colF, ?)', $x ], | FUNC(colF, ?)
2986 ] | /* ...with $x bound to ? */
2987 ===============================================================
2991 =head1 SPECIAL OPERATORS
2993 my $sqlmaker = SQL::Abstract->new(special_ops => [
2997 my ($self, $field, $op, $arg) = @_;
3003 handler => 'method_name',
3007 A "special operator" is a SQL syntactic clause that can be
3008 applied to a field, instead of a usual binary operator.
3011 WHERE field IN (?, ?, ?)
3012 WHERE field BETWEEN ? AND ?
3013 WHERE MATCH(field) AGAINST (?, ?)
3015 Special operators IN and BETWEEN are fairly standard and therefore
3016 are builtin within C<SQL::Abstract> (as the overridable methods
3017 C<_where_field_IN> and C<_where_field_BETWEEN>). For other operators,
3018 like the MATCH .. AGAINST example above which is specific to MySQL,
3019 you can write your own operator handlers - supply a C<special_ops>
3020 argument to the C<new> method. That argument takes an arrayref of
3021 operator definitions; each operator definition is a hashref with two
3028 the regular expression to match the operator
3032 Either a coderef or a plain scalar method name. In both cases
3033 the expected return is C<< ($sql, @bind) >>.
3035 When supplied with a method name, it is simply called on the
3036 L<SQL::Abstract> object as:
3038 $self->$method_name($field, $op, $arg)
3042 $field is the LHS of the operator
3043 $op is the part that matched the handler regex
3046 When supplied with a coderef, it is called as:
3048 $coderef->($self, $field, $op, $arg)
3053 For example, here is an implementation
3054 of the MATCH .. AGAINST syntax for MySQL
3056 my $sqlmaker = SQL::Abstract->new(special_ops => [
3058 # special op for MySql MATCH (field) AGAINST(word1, word2, ...)
3059 {regex => qr/^match$/i,
3061 my ($self, $field, $op, $arg) = @_;
3062 $arg = [$arg] if not ref $arg;
3063 my $label = $self->_quote($field);
3064 my ($placeholder) = $self->_convert('?');
3065 my $placeholders = join ", ", (($placeholder) x @$arg);
3066 my $sql = $self->_sqlcase('match') . " ($label) "
3067 . $self->_sqlcase('against') . " ($placeholders) ";
3068 my @bind = $self->_bindtype($field, @$arg);
3069 return ($sql, @bind);
3076 =head1 UNARY OPERATORS
3078 my $sqlmaker = SQL::Abstract->new(unary_ops => [
3082 my ($self, $op, $arg) = @_;
3088 handler => 'method_name',
3092 A "unary operator" is a SQL syntactic clause that can be
3093 applied to a field - the operator goes before the field
3095 You can write your own operator handlers - supply a C<unary_ops>
3096 argument to the C<new> method. That argument takes an arrayref of
3097 operator definitions; each operator definition is a hashref with two
3104 the regular expression to match the operator
3108 Either a coderef or a plain scalar method name. In both cases
3109 the expected return is C<< $sql >>.
3111 When supplied with a method name, it is simply called on the
3112 L<SQL::Abstract> object as:
3114 $self->$method_name($op, $arg)
3118 $op is the part that matched the handler regex
3119 $arg is the RHS or argument of the operator
3121 When supplied with a coderef, it is called as:
3123 $coderef->($self, $op, $arg)
3131 Thanks to some benchmarking by Mark Stosberg, it turns out that
3132 this module is many orders of magnitude faster than using C<DBIx::Abstract>.
3133 I must admit this wasn't an intentional design issue, but it's a
3134 byproduct of the fact that you get to control your C<DBI> handles
3137 To maximize performance, use a code snippet like the following:
3139 # prepare a statement handle using the first row
3140 # and then reuse it for the rest of the rows
3142 for my $href (@array_of_hashrefs) {
3143 $stmt ||= $sql->insert('table', $href);
3144 $sth ||= $dbh->prepare($stmt);
3145 $sth->execute($sql->values($href));
3148 The reason this works is because the keys in your C<$href> are sorted
3149 internally by B<SQL::Abstract>. Thus, as long as your data retains
3150 the same structure, you only have to generate the SQL the first time
3151 around. On subsequent queries, simply use the C<values> function provided
3152 by this module to return your values in the correct order.
3154 However this depends on the values having the same type - if, for
3155 example, the values of a where clause may either have values
3156 (resulting in sql of the form C<column = ?> with a single bind
3157 value), or alternatively the values might be C<undef> (resulting in
3158 sql of the form C<column IS NULL> with no bind value) then the
3159 caching technique suggested will not work.
3163 If you use my C<CGI::FormBuilder> module at all, you'll hopefully
3164 really like this part (I do, at least). Building up a complex query
3165 can be as simple as the following:
3172 use CGI::FormBuilder;
3175 my $form = CGI::FormBuilder->new(...);
3176 my $sql = SQL::Abstract->new;
3178 if ($form->submitted) {
3179 my $field = $form->field;
3180 my $id = delete $field->{id};
3181 my($stmt, @bind) = $sql->update('table', $field, {id => $id});
3184 Of course, you would still have to connect using C<DBI> to run the
3185 query, but the point is that if you make your form look like your
3186 table, the actual query script can be extremely simplistic.
3188 If you're B<REALLY> lazy (I am), check out C<HTML::QuickTable> for
3189 a fast interface to returning and formatting data. I frequently
3190 use these three modules together to write complex database query
3191 apps in under 50 lines.
3193 =head1 HOW TO CONTRIBUTE
3195 Contributions are always welcome, in all usable forms (we especially
3196 welcome documentation improvements). The delivery methods include git-
3197 or unified-diff formatted patches, GitHub pull requests, or plain bug
3198 reports either via RT or the Mailing list. Contributors are generally
3199 granted full access to the official repository after their first several
3200 patches pass successful review.
3202 This project is maintained in a git repository. The code and related tools are
3203 accessible at the following locations:
3207 =item * Official repo: L<git://git.shadowcat.co.uk/dbsrgits/SQL-Abstract.git>
3209 =item * Official gitweb: L<http://git.shadowcat.co.uk/gitweb/gitweb.cgi?p=dbsrgits/SQL-Abstract.git>
3211 =item * GitHub mirror: L<https://github.com/dbsrgits/sql-abstract>
3213 =item * Authorized committers: L<ssh://dbsrgits@git.shadowcat.co.uk/SQL-Abstract.git>
3219 Version 1.50 was a major internal refactoring of C<SQL::Abstract>.
3220 Great care has been taken to preserve the I<published> behavior
3221 documented in previous versions in the 1.* family; however,
3222 some features that were previously undocumented, or behaved
3223 differently from the documentation, had to be changed in order
3224 to clarify the semantics. Hence, client code that was relying
3225 on some dark areas of C<SQL::Abstract> v1.*
3226 B<might behave differently> in v1.50.
3228 The main changes are:
3234 support for literal SQL through the C<< \ [ $sql, @bind ] >> syntax.
3238 support for the { operator => \"..." } construct (to embed literal SQL)
3242 support for the { operator => \["...", @bind] } construct (to embed literal SQL with bind values)
3246 optional support for L<array datatypes|/"Inserting and Updating Arrays">
3250 defensive programming: check arguments
3254 fixed bug with global logic, which was previously implemented
3255 through global variables yielding side-effects. Prior versions would
3256 interpret C<< [ {cond1, cond2}, [cond3, cond4] ] >>
3257 as C<< "(cond1 AND cond2) OR (cond3 AND cond4)" >>.
3258 Now this is interpreted
3259 as C<< "(cond1 AND cond2) OR (cond3 OR cond4)" >>.
3264 fixed semantics of _bindtype on array args
3268 dropped the C<_anoncopy> of the %where tree. No longer necessary,
3269 we just avoid shifting arrays within that tree.
3273 dropped the C<_modlogic> function
3277 =head1 ACKNOWLEDGEMENTS
3279 There are a number of individuals that have really helped out with
3280 this module. Unfortunately, most of them submitted bugs via CPAN
3281 so I have no idea who they are! But the people I do know are:
3283 Ash Berlin (order_by hash term support)
3284 Matt Trout (DBIx::Class support)
3285 Mark Stosberg (benchmarking)
3286 Chas Owens (initial "IN" operator support)
3287 Philip Collins (per-field SQL functions)
3288 Eric Kolve (hashref "AND" support)
3289 Mike Fragassi (enhancements to "BETWEEN" and "LIKE")
3290 Dan Kubb (support for "quote_char" and "name_sep")
3291 Guillermo Roditi (patch to cleanup "IN" and "BETWEEN", fix and tests for _order_by)
3292 Laurent Dami (internal refactoring, extensible list of special operators, literal SQL)
3293 Norbert Buchmuller (support for literal SQL in hashpair, misc. fixes & tests)
3294 Peter Rabbitson (rewrite of SQLA::Test, misc. fixes & tests)
3295 Oliver Charles (support for "RETURNING" after "INSERT")
3301 L<DBIx::Class>, L<DBIx::Abstract>, L<CGI::FormBuilder>, L<HTML::QuickTable>.
3305 Copyright (c) 2001-2007 Nathan Wiger <nwiger@cpan.org>. All Rights Reserved.
3307 This module is actively maintained by Matt Trout <mst@shadowcatsystems.co.uk>
3309 For support, your best bet is to try the C<DBIx::Class> users mailing list.
3310 While not an official support venue, C<DBIx::Class> makes heavy use of
3311 C<SQL::Abstract>, and as such list members there are very familiar with
3312 how to create queries.
3316 This module is free software; you may copy this under the same
3317 terms as perl itself (either the GNU General Public License or
3318 the Artistic License)