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') {
541 if (keys %$expr > 1) {
543 return +{ "-${logic}" => [
544 map $self->_expand_expr_hashpair($_ => $expr->{$_}, $logic),
548 return $self->_expand_expr_hashpair(%$expr, $logic);
553 sub _expand_expr_hashpair {
554 my ($self, $k, $v, $logic) = @_;
557 return $self->_expand_expr($v);
561 return +{ $k => { $self->{cmp} => $v } };
563 if (my $literal = is_literal_value($v)) {
565 belch 'Hash-pairs consisting of an empty string with a literal are deprecated, and will be removed in 2.0: use -and => [ $literal ] instead';
568 my ($sql, @bind) = @$literal;
569 return \[ $self->_quote($k).' '.$sql, @bind ];
576 my ($self, $where, $logic) = @_;
578 my $where_exp = $self->_expand_expr($where, $logic);
580 # dispatch on appropriate method according to refkind of $where
581 my $method = $self->_METHOD_FOR_refkind("_where", $where_exp);
583 my ($sql, @bind) = $self->$method($where_exp, $logic);
585 # DBIx::Class used to call _recurse_where in scalar context
586 # something else might too...
588 return ($sql, @bind);
591 belch "Calling _recurse_where in scalar context is deprecated and will go away before 2.0";
598 #======================================================================
599 # WHERE: top-level ARRAYREF
600 #======================================================================
603 sub _where_ARRAYREF {
604 my ($self, $where, $logic) = @_;
606 $logic = uc($logic || $self->{logic});
607 $logic eq 'AND' or $logic eq 'OR' or puke "unknown logic: $logic";
609 my @clauses = @$where;
611 my (@sql_clauses, @all_bind);
612 # need to use while() so can shift() for pairs
614 my $el = shift @clauses;
616 $el = undef if (defined $el and ! length $el);
618 # switch according to kind of $el and get corresponding ($sql, @bind)
619 my ($sql, @bind) = $self->_SWITCH_refkind($el, {
621 # skip empty elements, otherwise get invalid trailing AND stuff
622 ARRAYREF => sub {$self->_recurse_where($el) if @$el},
626 $self->_assert_bindval_matches_bindtype(@b);
630 HASHREF => sub {$self->_recurse_where($el, 'and') if %$el},
632 SCALARREF => sub { ($$el); },
635 # top-level arrayref with scalars, recurse in pairs
636 $self->_recurse_where({$el => shift(@clauses)})
639 UNDEF => sub {puke "Supplying an empty left hand side argument is not supported in array-pairs" },
643 push @sql_clauses, $sql;
644 push @all_bind, @bind;
648 return $self->_join_sql_clauses($logic, \@sql_clauses, \@all_bind);
651 #======================================================================
652 # WHERE: top-level ARRAYREFREF
653 #======================================================================
655 sub _where_ARRAYREFREF {
656 my ($self, $where) = @_;
657 my ($sql, @bind) = @$$where;
658 $self->_assert_bindval_matches_bindtype(@bind);
659 return ($sql, @bind);
662 #======================================================================
663 # WHERE: top-level HASHREF
664 #======================================================================
667 my ($self, $where) = @_;
668 my (@sql_clauses, @all_bind);
670 for my $k (sort keys %$where) {
671 my $v = $where->{$k};
673 # ($k => $v) is either a special unary op or a regular hashpair
674 my ($sql, @bind) = do {
676 # put the operator in canonical form
678 $op = substr $op, 1; # remove initial dash
679 $op =~ s/^\s+|\s+$//g;# remove leading/trailing space
680 $op =~ s/\s+/ /g; # compress whitespace
682 # so that -not_foo works correctly
683 $op =~ s/^not_/NOT /i;
685 $self->_debug("Unary OP(-$op) within hashref, recursing...");
686 my ($s, @b) = $self->_where_unary_op($op, $v);
688 # top level vs nested
689 # we assume that handled unary ops will take care of their ()s
691 List::Util::first {$op =~ $_->{regex}} @{$self->{unary_ops}}
693 ( defined $self->{_nested_func_lhs} and $self->{_nested_func_lhs} eq $k )
699 if (is_literal_value ($v) ) {
700 belch 'Hash-pairs consisting of an empty string with a literal are deprecated, and will be removed in 2.0: use -and => [ $literal ] instead';
703 puke "Supplying an empty left hand side argument is not supported in hash-pairs";
707 my $method = $self->_METHOD_FOR_refkind("_where_hashpair", $v);
708 $self->$method($k, $v);
712 push @sql_clauses, $sql;
713 push @all_bind, @bind;
716 return $self->_join_sql_clauses('and', \@sql_clauses, \@all_bind);
719 sub _where_unary_op {
720 my ($self, $op, $rhs) = @_;
722 # top level special ops are illegal in general
723 # this includes the -ident/-value ops (dual purpose unary and special)
724 puke "Illegal use of top-level '-$op'"
725 if ! defined $self->{_nested_func_lhs} and List::Util::first { $op =~ $_->{regex} } @{$self->{special_ops}};
727 if (my $op_entry = List::Util::first { $op =~ $_->{regex} } @{$self->{unary_ops}}) {
728 my $handler = $op_entry->{handler};
730 if (not ref $handler) {
731 if ($op =~ s/ [_\s]? \d+ $//x ) {
732 belch 'Use of [and|or|nest]_N modifiers is deprecated and will be removed in SQLA v2.0. '
733 . "You probably wanted ...-and => [ -$op => COND1, -$op => COND2 ... ]";
735 return $self->$handler($op, $rhs);
737 elsif (ref $handler eq 'CODE') {
738 return $handler->($self, $op, $rhs);
741 puke "Illegal handler for operator $op - expecting a method name or a coderef";
745 $self->_debug("Generic unary OP: $op - recursing as function");
747 $self->_assert_pass_injection_guard($op);
749 my ($sql, @bind) = $self->_SWITCH_refkind($rhs, {
751 puke "Illegal use of top-level '-$op'"
752 unless defined $self->{_nested_func_lhs};
755 $self->_convert('?'),
756 $self->_bindtype($self->{_nested_func_lhs}, $rhs)
760 $self->_recurse_where($rhs)
764 $sql = sprintf('%s %s',
765 $self->_sqlcase($op),
769 return ($sql, @bind);
772 sub _where_op_ANDOR {
773 my ($self, $op, $v) = @_;
775 $self->_SWITCH_refkind($v, {
777 return $self->_where_ARRAYREF($v, $op);
781 return ($op =~ /^or/i)
782 ? $self->_where_ARRAYREF([ map { $_ => $v->{$_} } (sort keys %$v) ], $op)
783 : $self->_where_HASHREF($v);
787 puke "-$op => \\\$scalar makes little sense, use " .
789 ? '[ \$scalar, \%rest_of_conditions ] instead'
790 : '-and => [ \$scalar, \%rest_of_conditions ] instead'
795 puke "-$op => \\[...] makes little sense, use " .
797 ? '[ \[...], \%rest_of_conditions ] instead'
798 : '-and => [ \[...], \%rest_of_conditions ] instead'
802 SCALAR => sub { # permissively interpreted as SQL
803 puke "-$op => \$value makes little sense, use -bool => \$value instead";
807 puke "-$op => undef not supported";
813 my ($self, $op, $v) = @_;
815 $self->_SWITCH_refkind($v, {
817 SCALAR => sub { # permissively interpreted as SQL
818 belch "literal SQL should be -nest => \\'scalar' "
819 . "instead of -nest => 'scalar' ";
824 puke "-$op => undef not supported";
828 $self->_recurse_where($v);
836 my ($self, $op, $v) = @_;
838 my ($s, @b) = $self->_SWITCH_refkind($v, {
839 SCALAR => sub { # interpreted as SQL column
840 $self->_convert($self->_quote($v));
844 puke "-$op => undef not supported";
848 $self->_recurse_where($v);
852 $s = "(NOT $s)" if $op =~ /^not/i;
857 sub _where_op_IDENT {
859 my ($op, $rhs) = splice @_, -2;
860 if (! defined $rhs or length ref $rhs) {
861 puke "-$op requires a single plain scalar argument (a quotable identifier)";
864 # in case we are called as a top level special op (no '=')
867 $_ = $self->_convert($self->_quote($_)) for ($lhs, $rhs);
875 sub _where_op_VALUE {
877 my ($op, $rhs) = splice @_, -2;
879 # in case we are called as a top level special op (no '=')
883 if (! defined $rhs) {
885 ? $self->_where_hashpair_HASHREF($lhs, { -is => undef })
892 (defined $lhs ? $lhs : $self->{_nested_func_lhs}),
899 $self->_convert($self->_quote($lhs)) . ' = ' . $self->_convert('?'),
903 $self->_convert('?'),
909 sub _where_hashpair_ARRAYREF {
910 my ($self, $k, $v) = @_;
913 my @v = @$v; # need copy because of shift below
914 $self->_debug("ARRAY($k) means distribute over elements");
916 # put apart first element if it is an operator (-and, -or)
918 (defined $v[0] && $v[0] =~ /^ - (?: AND|OR ) $/ix)
922 my @distributed = map { {$k => $_} } @v;
925 $self->_debug("OP($op) reinjected into the distributed array");
926 unshift @distributed, $op;
929 my $logic = $op ? substr($op, 1) : '';
931 return $self->_recurse_where(\@distributed, $logic);
934 $self->_debug("empty ARRAY($k) means 0=1");
935 return ($self->{sqlfalse});
939 sub _where_hashpair_HASHREF {
940 my ($self, $k, $v, $logic) = @_;
943 local $self->{_nested_func_lhs} = defined $self->{_nested_func_lhs}
944 ? $self->{_nested_func_lhs}
948 my ($all_sql, @all_bind);
950 for my $orig_op (sort keys %$v) {
951 my $val = $v->{$orig_op};
953 # put the operator in canonical form
956 # FIXME - we need to phase out dash-less ops
957 $op =~ s/^-//; # remove possible initial dash
958 $op =~ s/^\s+|\s+$//g;# remove leading/trailing space
959 $op =~ s/\s+/ /g; # compress whitespace
961 $self->_assert_pass_injection_guard($op);
964 $op =~ s/^is_not/IS NOT/i;
966 # so that -not_foo works correctly
967 $op =~ s/^not_/NOT /i;
969 # another retarded special case: foo => { $op => { -value => undef } }
970 if (ref $val eq 'HASH' and keys %$val == 1 and exists $val->{-value} and ! defined $val->{-value} ) {
976 # CASE: col-value logic modifiers
977 if ($orig_op =~ /^ \- (and|or) $/xi) {
978 ($sql, @bind) = $self->_where_hashpair_HASHREF($k, $val, $1);
980 # CASE: special operators like -in or -between
981 elsif (my $special_op = List::Util::first { $op =~ $_->{regex} } @{$self->{special_ops}}) {
982 my $handler = $special_op->{handler};
984 puke "No handler supplied for special operator $orig_op";
986 elsif (not ref $handler) {
987 ($sql, @bind) = $self->$handler($k, $op, $val);
989 elsif (ref $handler eq 'CODE') {
990 ($sql, @bind) = $handler->($self, $k, $op, $val);
993 puke "Illegal handler for special operator $orig_op - expecting a method name or a coderef";
997 $self->_SWITCH_refkind($val, {
999 ARRAYREF => sub { # CASE: col => {op => \@vals}
1000 ($sql, @bind) = $self->_where_field_op_ARRAYREF($k, $op, $val);
1003 ARRAYREFREF => sub { # CASE: col => {op => \[$sql, @bind]} (literal SQL with bind)
1004 my ($sub_sql, @sub_bind) = @$$val;
1005 $self->_assert_bindval_matches_bindtype(@sub_bind);
1006 $sql = join ' ', $self->_convert($self->_quote($k)),
1007 $self->_sqlcase($op),
1012 UNDEF => sub { # CASE: col => {op => undef} : sql "IS (NOT)? NULL"
1014 $op =~ /^not$/i ? 'is not' # legacy
1015 : $op =~ $self->{equality_op} ? 'is'
1016 : $op =~ $self->{like_op} ? belch("Supplying an undefined argument to '@{[ uc $op]}' is deprecated") && 'is'
1017 : $op =~ $self->{inequality_op} ? 'is not'
1018 : $op =~ $self->{not_like_op} ? belch("Supplying an undefined argument to '@{[ uc $op]}' is deprecated") && 'is not'
1019 : puke "unexpected operator '$orig_op' with undef operand";
1021 $sql = $self->_quote($k) . $self->_sqlcase(" $is null");
1024 FALLBACK => sub { # CASE: col => {op/func => $stuff}
1025 ($sql, @bind) = $self->_where_unary_op($op, $val);
1028 $self->_convert($self->_quote($k)),
1029 $self->{_nested_func_lhs} eq $k ? $sql : "($sql)", # top level vs nested
1035 ($all_sql) = (defined $all_sql and $all_sql) ? $self->_join_sql_clauses($logic, [$all_sql, $sql], []) : $sql;
1036 push @all_bind, @bind;
1038 return ($all_sql, @all_bind);
1041 sub _where_field_IS {
1042 my ($self, $k, $op, $v) = @_;
1044 my ($s) = $self->_SWITCH_refkind($v, {
1047 $self->_convert($self->_quote($k)),
1048 map { $self->_sqlcase($_)} ($op, 'null')
1051 puke "$op can only take undef as argument";
1058 sub _where_field_op_ARRAYREF {
1059 my ($self, $k, $op, $vals) = @_;
1061 my @vals = @$vals; #always work on a copy
1064 $self->_debug(sprintf '%s means multiple elements: [ %s ]',
1066 join(', ', map { defined $_ ? "'$_'" : 'NULL' } @vals ),
1069 # see if the first element is an -and/-or op
1071 if (defined $vals[0] && $vals[0] =~ /^ - (AND|OR) $/ix) {
1076 # a long standing API wart - an attempt to change this behavior during
1077 # the 1.50 series failed *spectacularly*. Warn instead and leave the
1082 (!$logic or $logic eq 'OR')
1084 ($op =~ $self->{inequality_op} or $op =~ $self->{not_like_op})
1087 belch "A multi-element arrayref as an argument to the inequality op '$o' "
1088 . 'is technically equivalent to an always-true 1=1 (you probably wanted '
1089 . "to say ...{ \$inequality_op => [ -and => \@values ] }... instead)"
1093 # distribute $op over each remaining member of @vals, append logic if exists
1094 return $self->_recurse_where([map { {$k => {$op, $_}} } @vals], $logic);
1098 # try to DWIM on equality operators
1100 $op =~ $self->{equality_op} ? $self->{sqlfalse}
1101 : $op =~ $self->{like_op} ? belch("Supplying an empty arrayref to '@{[ uc $op]}' is deprecated") && $self->{sqlfalse}
1102 : $op =~ $self->{inequality_op} ? $self->{sqltrue}
1103 : $op =~ $self->{not_like_op} ? belch("Supplying an empty arrayref to '@{[ uc $op]}' is deprecated") && $self->{sqltrue}
1104 : puke "operator '$op' applied on an empty array (field '$k')";
1109 sub _where_hashpair_SCALARREF {
1110 my ($self, $k, $v) = @_;
1111 $self->_debug("SCALAR($k) means literal SQL: $$v");
1112 my $sql = $self->_quote($k) . " " . $$v;
1116 # literal SQL with bind
1117 sub _where_hashpair_ARRAYREFREF {
1118 my ($self, $k, $v) = @_;
1119 $self->_debug("REF($k) means literal SQL: @${$v}");
1120 my ($sql, @bind) = @$$v;
1121 $self->_assert_bindval_matches_bindtype(@bind);
1122 $sql = $self->_quote($k) . " " . $sql;
1123 return ($sql, @bind );
1126 # literal SQL without bind
1127 sub _where_hashpair_SCALAR {
1128 my ($self, $k, $v) = @_;
1129 $self->_debug("NOREF($k) means simple key=val: $k $self->{cmp} $v");
1130 return ($self->_where_hashpair_HASHREF($k, { $self->{cmp} => $v }));
1134 sub _where_hashpair_UNDEF {
1135 my ($self, $k, $v) = @_;
1136 $self->_debug("UNDEF($k) means IS NULL");
1137 return $self->_where_hashpair_HASHREF($k, { -is => undef });
1140 #======================================================================
1141 # WHERE: TOP-LEVEL OTHERS (SCALARREF, SCALAR, UNDEF)
1142 #======================================================================
1145 sub _where_SCALARREF {
1146 my ($self, $where) = @_;
1149 $self->_debug("SCALAR(*top) means literal SQL: $$where");
1155 my ($self, $where) = @_;
1158 $self->_debug("NOREF(*top) means literal SQL: $where");
1169 #======================================================================
1170 # WHERE: BUILTIN SPECIAL OPERATORS (-in, -between)
1171 #======================================================================
1174 sub _where_field_BETWEEN {
1175 my ($self, $k, $op, $vals) = @_;
1177 my ($label, $and, $placeholder);
1178 $label = $self->_convert($self->_quote($k));
1179 $and = ' ' . $self->_sqlcase('and') . ' ';
1180 $placeholder = $self->_convert('?');
1181 $op = $self->_sqlcase($op);
1183 my $invalid_args = "Operator '$op' requires either an arrayref with two defined values or expressions, or a single literal scalarref/arrayref-ref";
1185 my ($clause, @bind) = $self->_SWITCH_refkind($vals, {
1186 ARRAYREFREF => sub {
1187 my ($s, @b) = @$$vals;
1188 $self->_assert_bindval_matches_bindtype(@b);
1195 puke $invalid_args if @$vals != 2;
1197 my (@all_sql, @all_bind);
1198 foreach my $val (@$vals) {
1199 my ($sql, @bind) = $self->_SWITCH_refkind($val, {
1201 return ($placeholder, $self->_bindtype($k, $val) );
1206 ARRAYREFREF => sub {
1207 my ($sql, @bind) = @$$val;
1208 $self->_assert_bindval_matches_bindtype(@bind);
1209 return ($sql, @bind);
1212 my ($func, $arg, @rest) = %$val;
1213 puke "Only simple { -func => arg } functions accepted as sub-arguments to BETWEEN"
1214 if (@rest or $func !~ /^ \- (.+)/x);
1215 $self->_where_unary_op($1 => $arg);
1221 push @all_sql, $sql;
1222 push @all_bind, @bind;
1226 (join $and, @all_sql),
1235 my $sql = "( $label $op $clause )";
1236 return ($sql, @bind)
1240 sub _where_field_IN {
1241 my ($self, $k, $op, $vals) = @_;
1243 # backwards compatibility: if scalar, force into an arrayref
1244 $vals = [$vals] if defined $vals && ! ref $vals;
1246 my ($label) = $self->_convert($self->_quote($k));
1247 my ($placeholder) = $self->_convert('?');
1248 $op = $self->_sqlcase($op);
1250 my ($sql, @bind) = $self->_SWITCH_refkind($vals, {
1251 ARRAYREF => sub { # list of choices
1252 if (@$vals) { # nonempty list
1253 my (@all_sql, @all_bind);
1255 for my $val (@$vals) {
1256 my ($sql, @bind) = $self->_SWITCH_refkind($val, {
1258 return ($placeholder, $val);
1263 ARRAYREFREF => sub {
1264 my ($sql, @bind) = @$$val;
1265 $self->_assert_bindval_matches_bindtype(@bind);
1266 return ($sql, @bind);
1269 my ($func, $arg, @rest) = %$val;
1270 puke "Only simple { -func => arg } functions accepted as sub-arguments to IN"
1271 if (@rest or $func !~ /^ \- (.+)/x);
1272 $self->_where_unary_op($1 => $arg);
1276 'SQL::Abstract before v1.75 used to generate incorrect SQL when the '
1277 . "-$op operator was given an undef-containing list: !!!AUDIT YOUR CODE "
1278 . 'AND DATA!!! (the upcoming Data::Query-based version of SQL::Abstract '
1279 . 'will emit the logically correct SQL instead of raising this exception)'
1283 push @all_sql, $sql;
1284 push @all_bind, @bind;
1288 sprintf('%s %s ( %s )',
1291 join(', ', @all_sql)
1293 $self->_bindtype($k, @all_bind),
1296 else { # empty list: some databases won't understand "IN ()", so DWIM
1297 my $sql = ($op =~ /\bnot\b/i) ? $self->{sqltrue} : $self->{sqlfalse};
1302 SCALARREF => sub { # literal SQL
1303 my $sql = $self->_open_outer_paren($$vals);
1304 return ("$label $op ( $sql )");
1306 ARRAYREFREF => sub { # literal SQL with bind
1307 my ($sql, @bind) = @$$vals;
1308 $self->_assert_bindval_matches_bindtype(@bind);
1309 $sql = $self->_open_outer_paren($sql);
1310 return ("$label $op ( $sql )", @bind);
1314 puke "Argument passed to the '$op' operator can not be undefined";
1318 puke "special op $op requires an arrayref (or scalarref/arrayref-ref)";
1322 return ($sql, @bind);
1325 # Some databases (SQLite) treat col IN (1, 2) different from
1326 # col IN ( (1, 2) ). Use this to strip all outer parens while
1327 # adding them back in the corresponding method
1328 sub _open_outer_paren {
1329 my ($self, $sql) = @_;
1331 while (my ($inner) = $sql =~ /^ \s* \( (.*) \) \s* $/xs) {
1333 # there are closing parens inside, need the heavy duty machinery
1334 # to reevaluate the extraction starting from $sql (full reevaluation)
1335 if ($inner =~ /\)/) {
1336 require Text::Balanced;
1338 my (undef, $remainder) = do {
1339 # idiotic design - writes to $@ but *DOES NOT* throw exceptions
1341 Text::Balanced::extract_bracketed($sql, '()', qr/\s*/);
1344 # the entire expression needs to be a balanced bracketed thing
1345 # (after an extract no remainder sans trailing space)
1346 last if defined $remainder and $remainder =~ /\S/;
1356 #======================================================================
1358 #======================================================================
1361 my ($self, $arg) = @_;
1364 for my $c ($self->_order_by_chunks($arg) ) {
1365 $self->_SWITCH_refkind($c, {
1366 SCALAR => sub { push @sql, $c },
1367 ARRAYREF => sub { push @sql, shift @$c; push @bind, @$c },
1373 $self->_sqlcase(' order by'),
1379 return wantarray ? ($sql, @bind) : $sql;
1382 sub _order_by_chunks {
1383 my ($self, $arg) = @_;
1385 return $self->_SWITCH_refkind($arg, {
1388 map { $self->_order_by_chunks($_ ) } @$arg;
1391 ARRAYREFREF => sub {
1392 my ($s, @b) = @$$arg;
1393 $self->_assert_bindval_matches_bindtype(@b);
1397 SCALAR => sub {$self->_quote($arg)},
1399 UNDEF => sub {return () },
1401 SCALARREF => sub {$$arg}, # literal SQL, no quoting
1404 # get first pair in hash
1405 my ($key, $val, @rest) = %$arg;
1407 return () unless $key;
1409 if (@rest or not $key =~ /^-(desc|asc)/i) {
1410 puke "hash passed to _order_by must have exactly one key (-desc or -asc)";
1416 for my $c ($self->_order_by_chunks($val)) {
1419 $self->_SWITCH_refkind($c, {
1424 ($sql, @bind) = @$c;
1428 $sql = $sql . ' ' . $self->_sqlcase($direction);
1430 push @ret, [ $sql, @bind];
1439 #======================================================================
1440 # DATASOURCE (FOR NOW, JUST PLAIN TABLE OR LIST OF TABLES)
1441 #======================================================================
1446 $self->_SWITCH_refkind($from, {
1447 ARRAYREF => sub {join ', ', map { $self->_quote($_) } @$from;},
1448 SCALAR => sub {$self->_quote($from)},
1449 SCALARREF => sub {$$from},
1454 #======================================================================
1456 #======================================================================
1458 # highly optimized, as it's called way too often
1460 # my ($self, $label) = @_;
1462 return '' unless defined $_[1];
1463 return ${$_[1]} if ref($_[1]) eq 'SCALAR';
1465 $_[0]->{quote_char} or
1466 ($_[0]->_assert_pass_injection_guard($_[1]), return $_[1]);
1468 my $qref = ref $_[0]->{quote_char};
1470 !$qref ? ($_[0]->{quote_char}, $_[0]->{quote_char})
1471 : ($qref eq 'ARRAY') ? @{$_[0]->{quote_char}}
1472 : puke "Unsupported quote_char format: $_[0]->{quote_char}";
1474 my $esc = $_[0]->{escape_char} || $r;
1476 # parts containing * are naturally unquoted
1477 return join($_[0]->{name_sep}||'', map
1478 +( $_ eq '*' ? $_ : do { (my $n = $_) =~ s/(\Q$esc\E|\Q$r\E)/$esc$1/g; $l . $n . $r } ),
1479 ( $_[0]->{name_sep} ? split (/\Q$_[0]->{name_sep}\E/, $_[1] ) : $_[1] )
1484 # Conversion, if applicable
1486 #my ($self, $arg) = @_;
1487 if ($_[0]->{convert}) {
1488 return $_[0]->_sqlcase($_[0]->{convert}) .'(' . $_[1] . ')';
1495 #my ($self, $col, @vals) = @_;
1496 # called often - tighten code
1497 return $_[0]->{bindtype} eq 'columns'
1498 ? map {[$_[1], $_]} @_[2 .. $#_]
1503 # Dies if any element of @bind is not in [colname => value] format
1504 # if bindtype is 'columns'.
1505 sub _assert_bindval_matches_bindtype {
1506 # my ($self, @bind) = @_;
1508 if ($self->{bindtype} eq 'columns') {
1510 if (!defined $_ || ref($_) ne 'ARRAY' || @$_ != 2) {
1511 puke "bindtype 'columns' selected, you need to pass: [column_name => bind_value]"
1517 sub _join_sql_clauses {
1518 my ($self, $logic, $clauses_aref, $bind_aref) = @_;
1520 if (@$clauses_aref > 1) {
1521 my $join = " " . $self->_sqlcase($logic) . " ";
1522 my $sql = '( ' . join($join, @$clauses_aref) . ' )';
1523 return ($sql, @$bind_aref);
1525 elsif (@$clauses_aref) {
1526 return ($clauses_aref->[0], @$bind_aref); # no parentheses
1529 return (); # if no SQL, ignore @$bind_aref
1534 # Fix SQL case, if so requested
1536 # LDNOTE: if $self->{case} is true, then it contains 'lower', so we
1537 # don't touch the argument ... crooked logic, but let's not change it!
1538 return $_[0]->{case} ? $_[1] : uc($_[1]);
1542 #======================================================================
1543 # DISPATCHING FROM REFKIND
1544 #======================================================================
1547 my ($self, $data) = @_;
1549 return 'UNDEF' unless defined $data;
1551 # blessed objects are treated like scalars
1552 my $ref = (Scalar::Util::blessed $data) ? '' : ref $data;
1554 return 'SCALAR' unless $ref;
1557 while ($ref eq 'REF') {
1559 $ref = (Scalar::Util::blessed $data) ? '' : ref $data;
1563 return ($ref||'SCALAR') . ('REF' x $n_steps);
1567 my ($self, $data) = @_;
1568 my @try = ($self->_refkind($data));
1569 push @try, 'SCALAR_or_UNDEF' if $try[0] eq 'SCALAR' || $try[0] eq 'UNDEF';
1570 push @try, 'FALLBACK';
1574 sub _METHOD_FOR_refkind {
1575 my ($self, $meth_prefix, $data) = @_;
1578 for (@{$self->_try_refkind($data)}) {
1579 $method = $self->can($meth_prefix."_".$_)
1583 return $method || puke "cannot dispatch on '$meth_prefix' for ".$self->_refkind($data);
1587 sub _SWITCH_refkind {
1588 my ($self, $data, $dispatch_table) = @_;
1591 for (@{$self->_try_refkind($data)}) {
1592 $coderef = $dispatch_table->{$_}
1596 puke "no dispatch entry for ".$self->_refkind($data)
1605 #======================================================================
1606 # VALUES, GENERATE, AUTOLOAD
1607 #======================================================================
1609 # LDNOTE: original code from nwiger, didn't touch code in that section
1610 # I feel the AUTOLOAD stuff should not be the default, it should
1611 # only be activated on explicit demand by user.
1615 my $data = shift || return;
1616 puke "Argument to ", __PACKAGE__, "->values must be a \\%hash"
1617 unless ref $data eq 'HASH';
1620 foreach my $k (sort keys %$data) {
1621 my $v = $data->{$k};
1622 $self->_SWITCH_refkind($v, {
1624 if ($self->{array_datatypes}) { # array datatype
1625 push @all_bind, $self->_bindtype($k, $v);
1627 else { # literal SQL with bind
1628 my ($sql, @bind) = @$v;
1629 $self->_assert_bindval_matches_bindtype(@bind);
1630 push @all_bind, @bind;
1633 ARRAYREFREF => sub { # literal SQL with bind
1634 my ($sql, @bind) = @${$v};
1635 $self->_assert_bindval_matches_bindtype(@bind);
1636 push @all_bind, @bind;
1638 SCALARREF => sub { # literal SQL without bind
1640 SCALAR_or_UNDEF => sub {
1641 push @all_bind, $self->_bindtype($k, $v);
1652 my(@sql, @sqlq, @sqlv);
1656 if ($ref eq 'HASH') {
1657 for my $k (sort keys %$_) {
1660 my $label = $self->_quote($k);
1661 if ($r eq 'ARRAY') {
1662 # literal SQL with bind
1663 my ($sql, @bind) = @$v;
1664 $self->_assert_bindval_matches_bindtype(@bind);
1665 push @sqlq, "$label = $sql";
1667 } elsif ($r eq 'SCALAR') {
1668 # literal SQL without bind
1669 push @sqlq, "$label = $$v";
1671 push @sqlq, "$label = ?";
1672 push @sqlv, $self->_bindtype($k, $v);
1675 push @sql, $self->_sqlcase('set'), join ', ', @sqlq;
1676 } elsif ($ref eq 'ARRAY') {
1677 # unlike insert(), assume these are ONLY the column names, i.e. for SQL
1680 if ($r eq 'ARRAY') { # literal SQL with bind
1681 my ($sql, @bind) = @$v;
1682 $self->_assert_bindval_matches_bindtype(@bind);
1685 } elsif ($r eq 'SCALAR') { # literal SQL without bind
1686 # embedded literal SQL
1693 push @sql, '(' . join(', ', @sqlq) . ')';
1694 } elsif ($ref eq 'SCALAR') {
1698 # strings get case twiddled
1699 push @sql, $self->_sqlcase($_);
1703 my $sql = join ' ', @sql;
1705 # this is pretty tricky
1706 # if ask for an array, return ($stmt, @bind)
1707 # otherwise, s/?/shift @sqlv/ to put it inline
1709 return ($sql, @sqlv);
1711 1 while $sql =~ s/\?/my $d = shift(@sqlv);
1712 ref $d ? $d->[1] : $d/e;
1721 # This allows us to check for a local, then _form, attr
1723 my($name) = $AUTOLOAD =~ /.*::(.+)/;
1724 return $self->generate($name, @_);
1735 SQL::Abstract - Generate SQL from Perl data structures
1741 my $sql = SQL::Abstract->new;
1743 my($stmt, @bind) = $sql->select($source, \@fields, \%where, $order);
1745 my($stmt, @bind) = $sql->insert($table, \%fieldvals || \@values);
1747 my($stmt, @bind) = $sql->update($table, \%fieldvals, \%where);
1749 my($stmt, @bind) = $sql->delete($table, \%where);
1751 # Then, use these in your DBI statements
1752 my $sth = $dbh->prepare($stmt);
1753 $sth->execute(@bind);
1755 # Just generate the WHERE clause
1756 my($stmt, @bind) = $sql->where(\%where, $order);
1758 # Return values in the same order, for hashed queries
1759 # See PERFORMANCE section for more details
1760 my @bind = $sql->values(\%fieldvals);
1764 This module was inspired by the excellent L<DBIx::Abstract>.
1765 However, in using that module I found that what I really wanted
1766 to do was generate SQL, but still retain complete control over my
1767 statement handles and use the DBI interface. So, I set out to
1768 create an abstract SQL generation module.
1770 While based on the concepts used by L<DBIx::Abstract>, there are
1771 several important differences, especially when it comes to WHERE
1772 clauses. I have modified the concepts used to make the SQL easier
1773 to generate from Perl data structures and, IMO, more intuitive.
1774 The underlying idea is for this module to do what you mean, based
1775 on the data structures you provide it. The big advantage is that
1776 you don't have to modify your code every time your data changes,
1777 as this module figures it out.
1779 To begin with, an SQL INSERT is as easy as just specifying a hash
1780 of C<key=value> pairs:
1783 name => 'Jimbo Bobson',
1784 phone => '123-456-7890',
1785 address => '42 Sister Lane',
1786 city => 'St. Louis',
1787 state => 'Louisiana',
1790 The SQL can then be generated with this:
1792 my($stmt, @bind) = $sql->insert('people', \%data);
1794 Which would give you something like this:
1796 $stmt = "INSERT INTO people
1797 (address, city, name, phone, state)
1798 VALUES (?, ?, ?, ?, ?)";
1799 @bind = ('42 Sister Lane', 'St. Louis', 'Jimbo Bobson',
1800 '123-456-7890', 'Louisiana');
1802 These are then used directly in your DBI code:
1804 my $sth = $dbh->prepare($stmt);
1805 $sth->execute(@bind);
1807 =head2 Inserting and Updating Arrays
1809 If your database has array types (like for example Postgres),
1810 activate the special option C<< array_datatypes => 1 >>
1811 when creating the C<SQL::Abstract> object.
1812 Then you may use an arrayref to insert and update database array types:
1814 my $sql = SQL::Abstract->new(array_datatypes => 1);
1816 planets => [qw/Mercury Venus Earth Mars/]
1819 my($stmt, @bind) = $sql->insert('solar_system', \%data);
1823 $stmt = "INSERT INTO solar_system (planets) VALUES (?)"
1825 @bind = (['Mercury', 'Venus', 'Earth', 'Mars']);
1828 =head2 Inserting and Updating SQL
1830 In order to apply SQL functions to elements of your C<%data> you may
1831 specify a reference to an arrayref for the given hash value. For example,
1832 if you need to execute the Oracle C<to_date> function on a value, you can
1833 say something like this:
1837 date_entered => \[ "to_date(?,'MM/DD/YYYY')", "03/02/2003" ],
1840 The first value in the array is the actual SQL. Any other values are
1841 optional and would be included in the bind values array. This gives
1844 my($stmt, @bind) = $sql->insert('people', \%data);
1846 $stmt = "INSERT INTO people (name, date_entered)
1847 VALUES (?, to_date(?,'MM/DD/YYYY'))";
1848 @bind = ('Bill', '03/02/2003');
1850 An UPDATE is just as easy, all you change is the name of the function:
1852 my($stmt, @bind) = $sql->update('people', \%data);
1854 Notice that your C<%data> isn't touched; the module will generate
1855 the appropriately quirky SQL for you automatically. Usually you'll
1856 want to specify a WHERE clause for your UPDATE, though, which is
1857 where handling C<%where> hashes comes in handy...
1859 =head2 Complex where statements
1861 This module can generate pretty complicated WHERE statements
1862 easily. For example, simple C<key=value> pairs are taken to mean
1863 equality, and if you want to see if a field is within a set
1864 of values, you can use an arrayref. Let's say we wanted to
1865 SELECT some data based on this criteria:
1868 requestor => 'inna',
1869 worker => ['nwiger', 'rcwe', 'sfz'],
1870 status => { '!=', 'completed' }
1873 my($stmt, @bind) = $sql->select('tickets', '*', \%where);
1875 The above would give you something like this:
1877 $stmt = "SELECT * FROM tickets WHERE
1878 ( requestor = ? ) AND ( status != ? )
1879 AND ( worker = ? OR worker = ? OR worker = ? )";
1880 @bind = ('inna', 'completed', 'nwiger', 'rcwe', 'sfz');
1882 Which you could then use in DBI code like so:
1884 my $sth = $dbh->prepare($stmt);
1885 $sth->execute(@bind);
1891 The methods are simple. There's one for every major SQL operation,
1892 and a constructor you use first. The arguments are specified in a
1893 similar order for each method (table, then fields, then a where
1894 clause) to try and simplify things.
1896 =head2 new(option => 'value')
1898 The C<new()> function takes a list of options and values, and returns
1899 a new B<SQL::Abstract> object which can then be used to generate SQL
1900 through the methods below. The options accepted are:
1906 If set to 'lower', then SQL will be generated in all lowercase. By
1907 default SQL is generated in "textbook" case meaning something like:
1909 SELECT a_field FROM a_table WHERE some_field LIKE '%someval%'
1911 Any setting other than 'lower' is ignored.
1915 This determines what the default comparison operator is. By default
1916 it is C<=>, meaning that a hash like this:
1918 %where = (name => 'nwiger', email => 'nate@wiger.org');
1920 Will generate SQL like this:
1922 WHERE name = 'nwiger' AND email = 'nate@wiger.org'
1924 However, you may want loose comparisons by default, so if you set
1925 C<cmp> to C<like> you would get SQL such as:
1927 WHERE name like 'nwiger' AND email like 'nate@wiger.org'
1929 You can also override the comparison on an individual basis - see
1930 the huge section on L</"WHERE CLAUSES"> at the bottom.
1932 =item sqltrue, sqlfalse
1934 Expressions for inserting boolean values within SQL statements.
1935 By default these are C<1=1> and C<1=0>. They are used
1936 by the special operators C<-in> and C<-not_in> for generating
1937 correct SQL even when the argument is an empty array (see below).
1941 This determines the default logical operator for multiple WHERE
1942 statements in arrays or hashes. If absent, the default logic is "or"
1943 for arrays, and "and" for hashes. This means that a WHERE
1947 event_date => {'>=', '2/13/99'},
1948 event_date => {'<=', '4/24/03'},
1951 will generate SQL like this:
1953 WHERE event_date >= '2/13/99' OR event_date <= '4/24/03'
1955 This is probably not what you want given this query, though (look
1956 at the dates). To change the "OR" to an "AND", simply specify:
1958 my $sql = SQL::Abstract->new(logic => 'and');
1960 Which will change the above C<WHERE> to:
1962 WHERE event_date >= '2/13/99' AND event_date <= '4/24/03'
1964 The logic can also be changed locally by inserting
1965 a modifier in front of an arrayref:
1967 @where = (-and => [event_date => {'>=', '2/13/99'},
1968 event_date => {'<=', '4/24/03'} ]);
1970 See the L</"WHERE CLAUSES"> section for explanations.
1974 This will automatically convert comparisons using the specified SQL
1975 function for both column and value. This is mostly used with an argument
1976 of C<upper> or C<lower>, so that the SQL will have the effect of
1977 case-insensitive "searches". For example, this:
1979 $sql = SQL::Abstract->new(convert => 'upper');
1980 %where = (keywords => 'MaKe iT CAse inSeNSItive');
1982 Will turn out the following SQL:
1984 WHERE upper(keywords) like upper('MaKe iT CAse inSeNSItive')
1986 The conversion can be C<upper()>, C<lower()>, or any other SQL function
1987 that can be applied symmetrically to fields (actually B<SQL::Abstract> does
1988 not validate this option; it will just pass through what you specify verbatim).
1992 This is a kludge because many databases suck. For example, you can't
1993 just bind values using DBI's C<execute()> for Oracle C<CLOB> or C<BLOB> fields.
1994 Instead, you have to use C<bind_param()>:
1996 $sth->bind_param(1, 'reg data');
1997 $sth->bind_param(2, $lots, {ora_type => ORA_CLOB});
1999 The problem is, B<SQL::Abstract> will normally just return a C<@bind> array,
2000 which loses track of which field each slot refers to. Fear not.
2002 If you specify C<bindtype> in new, you can determine how C<@bind> is returned.
2003 Currently, you can specify either C<normal> (default) or C<columns>. If you
2004 specify C<columns>, you will get an array that looks like this:
2006 my $sql = SQL::Abstract->new(bindtype => 'columns');
2007 my($stmt, @bind) = $sql->insert(...);
2010 [ 'column1', 'value1' ],
2011 [ 'column2', 'value2' ],
2012 [ 'column3', 'value3' ],
2015 You can then iterate through this manually, using DBI's C<bind_param()>.
2017 $sth->prepare($stmt);
2020 my($col, $data) = @$_;
2021 if ($col eq 'details' || $col eq 'comments') {
2022 $sth->bind_param($i, $data, {ora_type => ORA_CLOB});
2023 } elsif ($col eq 'image') {
2024 $sth->bind_param($i, $data, {ora_type => ORA_BLOB});
2026 $sth->bind_param($i, $data);
2030 $sth->execute; # execute without @bind now
2032 Now, why would you still use B<SQL::Abstract> if you have to do this crap?
2033 Basically, the advantage is still that you don't have to care which fields
2034 are or are not included. You could wrap that above C<for> loop in a simple
2035 sub called C<bind_fields()> or something and reuse it repeatedly. You still
2036 get a layer of abstraction over manual SQL specification.
2038 Note that if you set L</bindtype> to C<columns>, the C<\[ $sql, @bind ]>
2039 construct (see L</Literal SQL with placeholders and bind values (subqueries)>)
2040 will expect the bind values in this format.
2044 This is the character that a table or column name will be quoted
2045 with. By default this is an empty string, but you could set it to
2046 the character C<`>, to generate SQL like this:
2048 SELECT `a_field` FROM `a_table` WHERE `some_field` LIKE '%someval%'
2050 Alternatively, you can supply an array ref of two items, the first being the left
2051 hand quote character, and the second the right hand quote character. For
2052 example, you could supply C<['[',']']> for SQL Server 2000 compliant quotes
2053 that generates SQL like this:
2055 SELECT [a_field] FROM [a_table] WHERE [some_field] LIKE '%someval%'
2057 Quoting is useful if you have tables or columns names that are reserved
2058 words in your database's SQL dialect.
2062 This is the character that will be used to escape L</quote_char>s appearing
2063 in an identifier before it has been quoted.
2065 The parameter default in case of a single L</quote_char> character is the quote
2068 When opening-closing-style quoting is used (L</quote_char> is an arrayref)
2069 this parameter defaults to the B<closing (right)> L</quote_char>. Occurrences
2070 of the B<opening (left)> L</quote_char> within the identifier are currently left
2071 untouched. The default for opening-closing-style quotes may change in future
2072 versions, thus you are B<strongly encouraged> to specify the escape character
2077 This is the character that separates a table and column name. It is
2078 necessary to specify this when the C<quote_char> option is selected,
2079 so that tables and column names can be individually quoted like this:
2081 SELECT `table`.`one_field` FROM `table` WHERE `table`.`other_field` = 1
2083 =item injection_guard
2085 A regular expression C<qr/.../> that is applied to any C<-function> and unquoted
2086 column name specified in a query structure. This is a safety mechanism to avoid
2087 injection attacks when mishandling user input e.g.:
2089 my %condition_as_column_value_pairs = get_values_from_user();
2090 $sqla->select( ... , \%condition_as_column_value_pairs );
2092 If the expression matches an exception is thrown. Note that literal SQL
2093 supplied via C<\'...'> or C<\['...']> is B<not> checked in any way.
2095 Defaults to checking for C<;> and the C<GO> keyword (TransactSQL)
2097 =item array_datatypes
2099 When this option is true, arrayrefs in INSERT or UPDATE are
2100 interpreted as array datatypes and are passed directly
2102 When this option is false, arrayrefs are interpreted
2103 as literal SQL, just like refs to arrayrefs
2104 (but this behavior is for backwards compatibility; when writing
2105 new queries, use the "reference to arrayref" syntax
2111 Takes a reference to a list of "special operators"
2112 to extend the syntax understood by L<SQL::Abstract>.
2113 See section L</"SPECIAL OPERATORS"> for details.
2117 Takes a reference to a list of "unary operators"
2118 to extend the syntax understood by L<SQL::Abstract>.
2119 See section L</"UNARY OPERATORS"> for details.
2125 =head2 insert($table, \@values || \%fieldvals, \%options)
2127 This is the simplest function. You simply give it a table name
2128 and either an arrayref of values or hashref of field/value pairs.
2129 It returns an SQL INSERT statement and a list of bind values.
2130 See the sections on L</"Inserting and Updating Arrays"> and
2131 L</"Inserting and Updating SQL"> for information on how to insert
2132 with those data types.
2134 The optional C<\%options> hash reference may contain additional
2135 options to generate the insert SQL. Currently supported options
2142 Takes either a scalar of raw SQL fields, or an array reference of
2143 field names, and adds on an SQL C<RETURNING> statement at the end.
2144 This allows you to return data generated by the insert statement
2145 (such as row IDs) without performing another C<SELECT> statement.
2146 Note, however, this is not part of the SQL standard and may not
2147 be supported by all database engines.
2151 =head2 update($table, \%fieldvals, \%where, \%options)
2153 This takes a table, hashref of field/value pairs, and an optional
2154 hashref L<WHERE clause|/WHERE CLAUSES>. It returns an SQL UPDATE function and a list
2156 See the sections on L</"Inserting and Updating Arrays"> and
2157 L</"Inserting and Updating SQL"> for information on how to insert
2158 with those data types.
2160 The optional C<\%options> hash reference may contain additional
2161 options to generate the update SQL. Currently supported options
2168 See the C<returning> option to
2169 L<insert|/insert($table, \@values || \%fieldvals, \%options)>.
2173 =head2 select($source, $fields, $where, $order)
2175 This returns a SQL SELECT statement and associated list of bind values, as
2176 specified by the arguments:
2182 Specification of the 'FROM' part of the statement.
2183 The argument can be either a plain scalar (interpreted as a table
2184 name, will be quoted), or an arrayref (interpreted as a list
2185 of table names, joined by commas, quoted), or a scalarref
2186 (literal SQL, not quoted).
2190 Specification of the list of fields to retrieve from
2192 The argument can be either an arrayref (interpreted as a list
2193 of field names, will be joined by commas and quoted), or a
2194 plain scalar (literal SQL, not quoted).
2195 Please observe that this API is not as flexible as that of
2196 the first argument C<$source>, for backwards compatibility reasons.
2200 Optional argument to specify the WHERE part of the query.
2201 The argument is most often a hashref, but can also be
2202 an arrayref or plain scalar --
2203 see section L<WHERE clause|/"WHERE CLAUSES"> for details.
2207 Optional argument to specify the ORDER BY part of the query.
2208 The argument can be a scalar, a hashref or an arrayref
2209 -- see section L<ORDER BY clause|/"ORDER BY CLAUSES">
2215 =head2 delete($table, \%where, \%options)
2217 This takes a table name and optional hashref L<WHERE clause|/WHERE CLAUSES>.
2218 It returns an SQL DELETE statement and list of bind values.
2220 The optional C<\%options> hash reference may contain additional
2221 options to generate the delete SQL. Currently supported options
2228 See the C<returning> option to
2229 L<insert|/insert($table, \@values || \%fieldvals, \%options)>.
2233 =head2 where(\%where, $order)
2235 This is used to generate just the WHERE clause. For example,
2236 if you have an arbitrary data structure and know what the
2237 rest of your SQL is going to look like, but want an easy way
2238 to produce a WHERE clause, use this. It returns an SQL WHERE
2239 clause and list of bind values.
2242 =head2 values(\%data)
2244 This just returns the values from the hash C<%data>, in the same
2245 order that would be returned from any of the other above queries.
2246 Using this allows you to markedly speed up your queries if you
2247 are affecting lots of rows. See below under the L</"PERFORMANCE"> section.
2249 =head2 generate($any, 'number', $of, \@data, $struct, \%types)
2251 Warning: This is an experimental method and subject to change.
2253 This returns arbitrarily generated SQL. It's a really basic shortcut.
2254 It will return two different things, depending on return context:
2256 my($stmt, @bind) = $sql->generate('create table', \$table, \@fields);
2257 my $stmt_and_val = $sql->generate('create table', \$table, \@fields);
2259 These would return the following:
2261 # First calling form
2262 $stmt = "CREATE TABLE test (?, ?)";
2263 @bind = (field1, field2);
2265 # Second calling form
2266 $stmt_and_val = "CREATE TABLE test (field1, field2)";
2268 Depending on what you're trying to do, it's up to you to choose the correct
2269 format. In this example, the second form is what you would want.
2273 $sql->generate('alter session', { nls_date_format => 'MM/YY' });
2277 ALTER SESSION SET nls_date_format = 'MM/YY'
2279 You get the idea. Strings get their case twiddled, but everything
2280 else remains verbatim.
2282 =head1 EXPORTABLE FUNCTIONS
2284 =head2 is_plain_value
2286 Determines if the supplied argument is a plain value as understood by this
2291 =item * The value is C<undef>
2293 =item * The value is a non-reference
2295 =item * The value is an object with stringification overloading
2297 =item * The value is of the form C<< { -value => $anything } >>
2301 On failure returns C<undef>, on success returns a B<scalar> reference
2302 to the original supplied argument.
2308 The stringification overloading detection is rather advanced: it takes
2309 into consideration not only the presence of a C<""> overload, but if that
2310 fails also checks for enabled
2311 L<autogenerated versions of C<"">|overload/Magic Autogeneration>, based
2312 on either C<0+> or C<bool>.
2314 Unfortunately testing in the field indicates that this
2315 detection B<< may tickle a latent bug in perl versions before 5.018 >>,
2316 but only when very large numbers of stringifying objects are involved.
2317 At the time of writing ( Sep 2014 ) there is no clear explanation of
2318 the direct cause, nor is there a manageably small test case that reliably
2319 reproduces the problem.
2321 If you encounter any of the following exceptions in B<random places within
2322 your application stack> - this module may be to blame:
2324 Operation "ne": no method found,
2325 left argument in overloaded package <something>,
2326 right argument in overloaded package <something>
2330 Stub found while resolving method "???" overloading """" in package <something>
2332 If you fall victim to the above - please attempt to reduce the problem
2333 to something that could be sent to the L<SQL::Abstract developers
2334 |DBIx::Class/GETTING HELP/SUPPORT>
2335 (either publicly or privately). As a workaround in the meantime you can
2336 set C<$ENV{SQLA_ISVALUE_IGNORE_AUTOGENERATED_STRINGIFICATION}> to a true
2337 value, which will most likely eliminate your problem (at the expense of
2338 not being able to properly detect exotic forms of stringification).
2340 This notice and environment variable will be removed in a future version,
2341 as soon as the underlying problem is found and a reliable workaround is
2346 =head2 is_literal_value
2348 Determines if the supplied argument is a literal value as understood by this
2353 =item * C<\$sql_string>
2355 =item * C<\[ $sql_string, @bind_values ]>
2359 On failure returns C<undef>, on success returns an B<array> reference
2360 containing the unpacked version of the supplied literal SQL and bind values.
2362 =head1 WHERE CLAUSES
2366 This module uses a variation on the idea from L<DBIx::Abstract>. It
2367 is B<NOT>, repeat I<not> 100% compatible. B<The main logic of this
2368 module is that things in arrays are OR'ed, and things in hashes
2371 The easiest way to explain is to show lots of examples. After
2372 each C<%where> hash shown, it is assumed you used:
2374 my($stmt, @bind) = $sql->where(\%where);
2376 However, note that the C<%where> hash can be used directly in any
2377 of the other functions as well, as described above.
2379 =head2 Key-value pairs
2381 So, let's get started. To begin, a simple hash:
2385 status => 'completed'
2388 Is converted to SQL C<key = val> statements:
2390 $stmt = "WHERE user = ? AND status = ?";
2391 @bind = ('nwiger', 'completed');
2393 One common thing I end up doing is having a list of values that
2394 a field can be in. To do this, simply specify a list inside of
2399 status => ['assigned', 'in-progress', 'pending'];
2402 This simple code will create the following:
2404 $stmt = "WHERE user = ? AND ( status = ? OR status = ? OR status = ? )";
2405 @bind = ('nwiger', 'assigned', 'in-progress', 'pending');
2407 A field associated to an empty arrayref will be considered a
2408 logical false and will generate 0=1.
2410 =head2 Tests for NULL values
2412 If the value part is C<undef> then this is converted to SQL <IS NULL>
2421 $stmt = "WHERE user = ? AND status IS NULL";
2424 To test if a column IS NOT NULL:
2428 status => { '!=', undef },
2431 =head2 Specific comparison operators
2433 If you want to specify a different type of operator for your comparison,
2434 you can use a hashref for a given column:
2438 status => { '!=', 'completed' }
2441 Which would generate:
2443 $stmt = "WHERE user = ? AND status != ?";
2444 @bind = ('nwiger', 'completed');
2446 To test against multiple values, just enclose the values in an arrayref:
2448 status => { '=', ['assigned', 'in-progress', 'pending'] };
2450 Which would give you:
2452 "WHERE status = ? OR status = ? OR status = ?"
2455 The hashref can also contain multiple pairs, in which case it is expanded
2456 into an C<AND> of its elements:
2460 status => { '!=', 'completed', -not_like => 'pending%' }
2463 # Or more dynamically, like from a form
2464 $where{user} = 'nwiger';
2465 $where{status}{'!='} = 'completed';
2466 $where{status}{'-not_like'} = 'pending%';
2468 # Both generate this
2469 $stmt = "WHERE user = ? AND status != ? AND status NOT LIKE ?";
2470 @bind = ('nwiger', 'completed', 'pending%');
2473 To get an OR instead, you can combine it with the arrayref idea:
2477 priority => [ { '=', 2 }, { '>', 5 } ]
2480 Which would generate:
2482 $stmt = "WHERE ( priority = ? OR priority > ? ) AND user = ?";
2483 @bind = ('2', '5', 'nwiger');
2485 If you want to include literal SQL (with or without bind values), just use a
2486 scalar reference or reference to an arrayref as the value:
2489 date_entered => { '>' => \["to_date(?, 'MM/DD/YYYY')", "11/26/2008"] },
2490 date_expires => { '<' => \"now()" }
2493 Which would generate:
2495 $stmt = "WHERE date_entered > to_date(?, 'MM/DD/YYYY') AND date_expires < now()";
2496 @bind = ('11/26/2008');
2499 =head2 Logic and nesting operators
2501 In the example above,
2502 there is a subtle trap if you want to say something like
2503 this (notice the C<AND>):
2505 WHERE priority != ? AND priority != ?
2507 Because, in Perl you I<can't> do this:
2509 priority => { '!=' => 2, '!=' => 1 }
2511 As the second C<!=> key will obliterate the first. The solution
2512 is to use the special C<-modifier> form inside an arrayref:
2514 priority => [ -and => {'!=', 2},
2518 Normally, these would be joined by C<OR>, but the modifier tells it
2519 to use C<AND> instead. (Hint: You can use this in conjunction with the
2520 C<logic> option to C<new()> in order to change the way your queries
2521 work by default.) B<Important:> Note that the C<-modifier> goes
2522 B<INSIDE> the arrayref, as an extra first element. This will
2523 B<NOT> do what you think it might:
2525 priority => -and => [{'!=', 2}, {'!=', 1}] # WRONG!
2527 Here is a quick list of equivalencies, since there is some overlap:
2530 status => {'!=', 'completed', 'not like', 'pending%' }
2531 status => [ -and => {'!=', 'completed'}, {'not like', 'pending%'}]
2534 status => {'=', ['assigned', 'in-progress']}
2535 status => [ -or => {'=', 'assigned'}, {'=', 'in-progress'}]
2536 status => [ {'=', 'assigned'}, {'=', 'in-progress'} ]
2540 =head2 Special operators: IN, BETWEEN, etc.
2542 You can also use the hashref format to compare a list of fields using the
2543 C<IN> comparison operator, by specifying the list as an arrayref:
2546 status => 'completed',
2547 reportid => { -in => [567, 2335, 2] }
2550 Which would generate:
2552 $stmt = "WHERE status = ? AND reportid IN (?,?,?)";
2553 @bind = ('completed', '567', '2335', '2');
2555 The reverse operator C<-not_in> generates SQL C<NOT IN> and is used in
2558 If the argument to C<-in> is an empty array, 'sqlfalse' is generated
2559 (by default: C<1=0>). Similarly, C<< -not_in => [] >> generates
2560 'sqltrue' (by default: C<1=1>).
2562 In addition to the array you can supply a chunk of literal sql or
2563 literal sql with bind:
2566 customer => { -in => \[
2567 'SELECT cust_id FROM cust WHERE balance > ?',
2570 status => { -in => \'SELECT status_codes FROM states' },
2576 customer IN ( SELECT cust_id FROM cust WHERE balance > ? )
2577 AND status IN ( SELECT status_codes FROM states )
2581 Finally, if the argument to C<-in> is not a reference, it will be
2582 treated as a single-element array.
2584 Another pair of operators is C<-between> and C<-not_between>,
2585 used with an arrayref of two values:
2589 completion_date => {
2590 -not_between => ['2002-10-01', '2003-02-06']
2596 WHERE user = ? AND completion_date NOT BETWEEN ( ? AND ? )
2598 Just like with C<-in> all plausible combinations of literal SQL
2602 start0 => { -between => [ 1, 2 ] },
2603 start1 => { -between => \["? AND ?", 1, 2] },
2604 start2 => { -between => \"lower(x) AND upper(y)" },
2605 start3 => { -between => [
2607 \["upper(?)", 'stuff' ],
2614 ( start0 BETWEEN ? AND ? )
2615 AND ( start1 BETWEEN ? AND ? )
2616 AND ( start2 BETWEEN lower(x) AND upper(y) )
2617 AND ( start3 BETWEEN lower(x) AND upper(?) )
2619 @bind = (1, 2, 1, 2, 'stuff');
2622 These are the two builtin "special operators"; but the
2623 list can be expanded: see section L</"SPECIAL OPERATORS"> below.
2625 =head2 Unary operators: bool
2627 If you wish to test against boolean columns or functions within your
2628 database you can use the C<-bool> and C<-not_bool> operators. For
2629 example to test the column C<is_user> being true and the column
2630 C<is_enabled> being false you would use:-
2634 -not_bool => 'is_enabled',
2639 WHERE is_user AND NOT is_enabled
2641 If a more complex combination is required, testing more conditions,
2642 then you should use the and/or operators:-
2647 -not_bool => { two=> { -rlike => 'bar' } },
2648 -not_bool => { three => [ { '=', 2 }, { '>', 5 } ] },
2659 (NOT ( three = ? OR three > ? ))
2662 =head2 Nested conditions, -and/-or prefixes
2664 So far, we've seen how multiple conditions are joined with a top-level
2665 C<AND>. We can change this by putting the different conditions we want in
2666 hashes and then putting those hashes in an array. For example:
2671 status => { -like => ['pending%', 'dispatched'] },
2675 status => 'unassigned',
2679 This data structure would create the following:
2681 $stmt = "WHERE ( user = ? AND ( status LIKE ? OR status LIKE ? ) )
2682 OR ( user = ? AND status = ? ) )";
2683 @bind = ('nwiger', 'pending', 'dispatched', 'robot', 'unassigned');
2686 Clauses in hashrefs or arrayrefs can be prefixed with an C<-and> or C<-or>
2687 to change the logic inside:
2693 -and => [ workhrs => {'>', 20}, geo => 'ASIA' ],
2694 -or => { workhrs => {'<', 50}, geo => 'EURO' },
2701 $stmt = "WHERE ( user = ?
2702 AND ( ( workhrs > ? AND geo = ? )
2703 OR ( workhrs < ? OR geo = ? ) ) )";
2704 @bind = ('nwiger', '20', 'ASIA', '50', 'EURO');
2706 =head3 Algebraic inconsistency, for historical reasons
2708 C<Important note>: when connecting several conditions, the C<-and->|C<-or>
2709 operator goes C<outside> of the nested structure; whereas when connecting
2710 several constraints on one column, the C<-and> operator goes
2711 C<inside> the arrayref. Here is an example combining both features:
2714 -and => [a => 1, b => 2],
2715 -or => [c => 3, d => 4],
2716 e => [-and => {-like => 'foo%'}, {-like => '%bar'} ]
2721 WHERE ( ( ( a = ? AND b = ? )
2722 OR ( c = ? OR d = ? )
2723 OR ( e LIKE ? AND e LIKE ? ) ) )
2725 This difference in syntax is unfortunate but must be preserved for
2726 historical reasons. So be careful: the two examples below would
2727 seem algebraically equivalent, but they are not
2730 { -like => 'foo%' },
2731 { -like => '%bar' },
2733 # yields: WHERE ( ( col LIKE ? AND col LIKE ? ) )
2736 { col => { -like => 'foo%' } },
2737 { col => { -like => '%bar' } },
2739 # yields: WHERE ( ( col LIKE ? OR col LIKE ? ) )
2742 =head2 Literal SQL and value type operators
2744 The basic premise of SQL::Abstract is that in WHERE specifications the "left
2745 side" is a column name and the "right side" is a value (normally rendered as
2746 a placeholder). This holds true for both hashrefs and arrayref pairs as you
2747 see in the L</WHERE CLAUSES> examples above. Sometimes it is necessary to
2748 alter this behavior. There are several ways of doing so.
2752 This is a virtual operator that signals the string to its right side is an
2753 identifier (a column name) and not a value. For example to compare two
2754 columns you would write:
2757 priority => { '<', 2 },
2758 requestor => { -ident => 'submitter' },
2763 $stmt = "WHERE priority < ? AND requestor = submitter";
2766 If you are maintaining legacy code you may see a different construct as
2767 described in L</Deprecated usage of Literal SQL>, please use C<-ident> in new
2772 This is a virtual operator that signals that the construct to its right side
2773 is a value to be passed to DBI. This is for example necessary when you want
2774 to write a where clause against an array (for RDBMS that support such
2775 datatypes). For example:
2778 array => { -value => [1, 2, 3] }
2783 $stmt = 'WHERE array = ?';
2784 @bind = ([1, 2, 3]);
2786 Note that if you were to simply say:
2792 the result would probably not be what you wanted:
2794 $stmt = 'WHERE array = ? OR array = ? OR array = ?';
2799 Finally, sometimes only literal SQL will do. To include a random snippet
2800 of SQL verbatim, you specify it as a scalar reference. Consider this only
2801 as a last resort. Usually there is a better way. For example:
2804 priority => { '<', 2 },
2805 requestor => { -in => \'(SELECT name FROM hitmen)' },
2810 $stmt = "WHERE priority < ? AND requestor IN (SELECT name FROM hitmen)"
2813 Note that in this example, you only get one bind parameter back, since
2814 the verbatim SQL is passed as part of the statement.
2818 Never use untrusted input as a literal SQL argument - this is a massive
2819 security risk (there is no way to check literal snippets for SQL
2820 injections and other nastyness). If you need to deal with untrusted input
2821 use literal SQL with placeholders as described next.
2823 =head3 Literal SQL with placeholders and bind values (subqueries)
2825 If the literal SQL to be inserted has placeholders and bind values,
2826 use a reference to an arrayref (yes this is a double reference --
2827 not so common, but perfectly legal Perl). For example, to find a date
2828 in Postgres you can use something like this:
2831 date_column => \[ "= date '2008-09-30' - ?::integer", 10 ]
2836 $stmt = "WHERE ( date_column = date '2008-09-30' - ?::integer )"
2839 Note that you must pass the bind values in the same format as they are returned
2840 by L<where|/where(\%where, $order)>. This means that if you set L</bindtype>
2841 to C<columns>, you must provide the bind values in the
2842 C<< [ column_meta => value ] >> format, where C<column_meta> is an opaque
2843 scalar value; most commonly the column name, but you can use any scalar value
2844 (including references and blessed references), L<SQL::Abstract> will simply
2845 pass it through intact. So if C<bindtype> is set to C<columns> the above
2846 example will look like:
2849 date_column => \[ "= date '2008-09-30' - ?::integer", [ {} => 10 ] ]
2852 Literal SQL is especially useful for nesting parenthesized clauses in the
2853 main SQL query. Here is a first example:
2855 my ($sub_stmt, @sub_bind) = ("SELECT c1 FROM t1 WHERE c2 < ? AND c3 LIKE ?",
2859 bar => \["IN ($sub_stmt)" => @sub_bind],
2864 $stmt = "WHERE (foo = ? AND bar IN (SELECT c1 FROM t1
2865 WHERE c2 < ? AND c3 LIKE ?))";
2866 @bind = (1234, 100, "foo%");
2868 Other subquery operators, like for example C<"E<gt> ALL"> or C<"NOT IN">,
2869 are expressed in the same way. Of course the C<$sub_stmt> and
2870 its associated bind values can be generated through a former call
2873 my ($sub_stmt, @sub_bind)
2874 = $sql->select("t1", "c1", {c2 => {"<" => 100},
2875 c3 => {-like => "foo%"}});
2878 bar => \["> ALL ($sub_stmt)" => @sub_bind],
2881 In the examples above, the subquery was used as an operator on a column;
2882 but the same principle also applies for a clause within the main C<%where>
2883 hash, like an EXISTS subquery:
2885 my ($sub_stmt, @sub_bind)
2886 = $sql->select("t1", "*", {c1 => 1, c2 => \"> t0.c0"});
2887 my %where = ( -and => [
2889 \["EXISTS ($sub_stmt)" => @sub_bind],
2894 $stmt = "WHERE (foo = ? AND EXISTS (SELECT * FROM t1
2895 WHERE c1 = ? AND c2 > t0.c0))";
2899 Observe that the condition on C<c2> in the subquery refers to
2900 column C<t0.c0> of the main query: this is I<not> a bind
2901 value, so we have to express it through a scalar ref.
2902 Writing C<< c2 => {">" => "t0.c0"} >> would have generated
2903 C<< c2 > ? >> with bind value C<"t0.c0"> ... not exactly
2904 what we wanted here.
2906 Finally, here is an example where a subquery is used
2907 for expressing unary negation:
2909 my ($sub_stmt, @sub_bind)
2910 = $sql->where({age => [{"<" => 10}, {">" => 20}]});
2911 $sub_stmt =~ s/^ where //i; # don't want "WHERE" in the subclause
2913 lname => {like => '%son%'},
2914 \["NOT ($sub_stmt)" => @sub_bind],
2919 $stmt = "lname LIKE ? AND NOT ( age < ? OR age > ? )"
2920 @bind = ('%son%', 10, 20)
2922 =head3 Deprecated usage of Literal SQL
2924 Below are some examples of archaic use of literal SQL. It is shown only as
2925 reference for those who deal with legacy code. Each example has a much
2926 better, cleaner and safer alternative that users should opt for in new code.
2932 my %where = ( requestor => \'IS NOT NULL' )
2934 $stmt = "WHERE requestor IS NOT NULL"
2936 This used to be the way of generating NULL comparisons, before the handling
2937 of C<undef> got formalized. For new code please use the superior syntax as
2938 described in L</Tests for NULL values>.
2942 my %where = ( requestor => \'= submitter' )
2944 $stmt = "WHERE requestor = submitter"
2946 This used to be the only way to compare columns. Use the superior L</-ident>
2947 method for all new code. For example an identifier declared in such a way
2948 will be properly quoted if L</quote_char> is properly set, while the legacy
2949 form will remain as supplied.
2953 my %where = ( is_ready => \"", completed => { '>', '2012-12-21' } )
2955 $stmt = "WHERE completed > ? AND is_ready"
2956 @bind = ('2012-12-21')
2958 Using an empty string literal used to be the only way to express a boolean.
2959 For all new code please use the much more readable
2960 L<-bool|/Unary operators: bool> operator.
2966 These pages could go on for a while, since the nesting of the data
2967 structures this module can handle are pretty much unlimited (the
2968 module implements the C<WHERE> expansion as a recursive function
2969 internally). Your best bet is to "play around" with the module a
2970 little to see how the data structures behave, and choose the best
2971 format for your data based on that.
2973 And of course, all the values above will probably be replaced with
2974 variables gotten from forms or the command line. After all, if you
2975 knew everything ahead of time, you wouldn't have to worry about
2976 dynamically-generating SQL and could just hardwire it into your
2979 =head1 ORDER BY CLAUSES
2981 Some functions take an order by clause. This can either be a scalar (just a
2982 column name), a hashref of C<< { -desc => 'col' } >> or C<< { -asc => 'col' }
2983 >>, a scalarref, an arrayref-ref, or an arrayref of any of the previous
2986 Given | Will Generate
2987 ---------------------------------------------------------------
2989 'colA' | ORDER BY colA
2991 [qw/colA colB/] | ORDER BY colA, colB
2993 {-asc => 'colA'} | ORDER BY colA ASC
2995 {-desc => 'colB'} | ORDER BY colB DESC
2997 ['colA', {-asc => 'colB'}] | ORDER BY colA, colB ASC
2999 { -asc => [qw/colA colB/] } | ORDER BY colA ASC, colB ASC
3001 \'colA DESC' | ORDER BY colA DESC
3003 \[ 'FUNC(colA, ?)', $x ] | ORDER BY FUNC(colA, ?)
3004 | /* ...with $x bound to ? */
3007 { -asc => 'colA' }, | colA ASC,
3008 { -desc => [qw/colB/] }, | colB DESC,
3009 { -asc => [qw/colC colD/] },| colC ASC, colD ASC,
3010 \'colE DESC', | colE DESC,
3011 \[ 'FUNC(colF, ?)', $x ], | FUNC(colF, ?)
3012 ] | /* ...with $x bound to ? */
3013 ===============================================================
3017 =head1 SPECIAL OPERATORS
3019 my $sqlmaker = SQL::Abstract->new(special_ops => [
3023 my ($self, $field, $op, $arg) = @_;
3029 handler => 'method_name',
3033 A "special operator" is a SQL syntactic clause that can be
3034 applied to a field, instead of a usual binary operator.
3037 WHERE field IN (?, ?, ?)
3038 WHERE field BETWEEN ? AND ?
3039 WHERE MATCH(field) AGAINST (?, ?)
3041 Special operators IN and BETWEEN are fairly standard and therefore
3042 are builtin within C<SQL::Abstract> (as the overridable methods
3043 C<_where_field_IN> and C<_where_field_BETWEEN>). For other operators,
3044 like the MATCH .. AGAINST example above which is specific to MySQL,
3045 you can write your own operator handlers - supply a C<special_ops>
3046 argument to the C<new> method. That argument takes an arrayref of
3047 operator definitions; each operator definition is a hashref with two
3054 the regular expression to match the operator
3058 Either a coderef or a plain scalar method name. In both cases
3059 the expected return is C<< ($sql, @bind) >>.
3061 When supplied with a method name, it is simply called on the
3062 L<SQL::Abstract> object as:
3064 $self->$method_name($field, $op, $arg)
3068 $field is the LHS of the operator
3069 $op is the part that matched the handler regex
3072 When supplied with a coderef, it is called as:
3074 $coderef->($self, $field, $op, $arg)
3079 For example, here is an implementation
3080 of the MATCH .. AGAINST syntax for MySQL
3082 my $sqlmaker = SQL::Abstract->new(special_ops => [
3084 # special op for MySql MATCH (field) AGAINST(word1, word2, ...)
3085 {regex => qr/^match$/i,
3087 my ($self, $field, $op, $arg) = @_;
3088 $arg = [$arg] if not ref $arg;
3089 my $label = $self->_quote($field);
3090 my ($placeholder) = $self->_convert('?');
3091 my $placeholders = join ", ", (($placeholder) x @$arg);
3092 my $sql = $self->_sqlcase('match') . " ($label) "
3093 . $self->_sqlcase('against') . " ($placeholders) ";
3094 my @bind = $self->_bindtype($field, @$arg);
3095 return ($sql, @bind);
3102 =head1 UNARY OPERATORS
3104 my $sqlmaker = SQL::Abstract->new(unary_ops => [
3108 my ($self, $op, $arg) = @_;
3114 handler => 'method_name',
3118 A "unary operator" is a SQL syntactic clause that can be
3119 applied to a field - the operator goes before the field
3121 You can write your own operator handlers - supply a C<unary_ops>
3122 argument to the C<new> method. That argument takes an arrayref of
3123 operator definitions; each operator definition is a hashref with two
3130 the regular expression to match the operator
3134 Either a coderef or a plain scalar method name. In both cases
3135 the expected return is C<< $sql >>.
3137 When supplied with a method name, it is simply called on the
3138 L<SQL::Abstract> object as:
3140 $self->$method_name($op, $arg)
3144 $op is the part that matched the handler regex
3145 $arg is the RHS or argument of the operator
3147 When supplied with a coderef, it is called as:
3149 $coderef->($self, $op, $arg)
3157 Thanks to some benchmarking by Mark Stosberg, it turns out that
3158 this module is many orders of magnitude faster than using C<DBIx::Abstract>.
3159 I must admit this wasn't an intentional design issue, but it's a
3160 byproduct of the fact that you get to control your C<DBI> handles
3163 To maximize performance, use a code snippet like the following:
3165 # prepare a statement handle using the first row
3166 # and then reuse it for the rest of the rows
3168 for my $href (@array_of_hashrefs) {
3169 $stmt ||= $sql->insert('table', $href);
3170 $sth ||= $dbh->prepare($stmt);
3171 $sth->execute($sql->values($href));
3174 The reason this works is because the keys in your C<$href> are sorted
3175 internally by B<SQL::Abstract>. Thus, as long as your data retains
3176 the same structure, you only have to generate the SQL the first time
3177 around. On subsequent queries, simply use the C<values> function provided
3178 by this module to return your values in the correct order.
3180 However this depends on the values having the same type - if, for
3181 example, the values of a where clause may either have values
3182 (resulting in sql of the form C<column = ?> with a single bind
3183 value), or alternatively the values might be C<undef> (resulting in
3184 sql of the form C<column IS NULL> with no bind value) then the
3185 caching technique suggested will not work.
3189 If you use my C<CGI::FormBuilder> module at all, you'll hopefully
3190 really like this part (I do, at least). Building up a complex query
3191 can be as simple as the following:
3198 use CGI::FormBuilder;
3201 my $form = CGI::FormBuilder->new(...);
3202 my $sql = SQL::Abstract->new;
3204 if ($form->submitted) {
3205 my $field = $form->field;
3206 my $id = delete $field->{id};
3207 my($stmt, @bind) = $sql->update('table', $field, {id => $id});
3210 Of course, you would still have to connect using C<DBI> to run the
3211 query, but the point is that if you make your form look like your
3212 table, the actual query script can be extremely simplistic.
3214 If you're B<REALLY> lazy (I am), check out C<HTML::QuickTable> for
3215 a fast interface to returning and formatting data. I frequently
3216 use these three modules together to write complex database query
3217 apps in under 50 lines.
3219 =head1 HOW TO CONTRIBUTE
3221 Contributions are always welcome, in all usable forms (we especially
3222 welcome documentation improvements). The delivery methods include git-
3223 or unified-diff formatted patches, GitHub pull requests, or plain bug
3224 reports either via RT or the Mailing list. Contributors are generally
3225 granted full access to the official repository after their first several
3226 patches pass successful review.
3228 This project is maintained in a git repository. The code and related tools are
3229 accessible at the following locations:
3233 =item * Official repo: L<git://git.shadowcat.co.uk/dbsrgits/SQL-Abstract.git>
3235 =item * Official gitweb: L<http://git.shadowcat.co.uk/gitweb/gitweb.cgi?p=dbsrgits/SQL-Abstract.git>
3237 =item * GitHub mirror: L<https://github.com/dbsrgits/sql-abstract>
3239 =item * Authorized committers: L<ssh://dbsrgits@git.shadowcat.co.uk/SQL-Abstract.git>
3245 Version 1.50 was a major internal refactoring of C<SQL::Abstract>.
3246 Great care has been taken to preserve the I<published> behavior
3247 documented in previous versions in the 1.* family; however,
3248 some features that were previously undocumented, or behaved
3249 differently from the documentation, had to be changed in order
3250 to clarify the semantics. Hence, client code that was relying
3251 on some dark areas of C<SQL::Abstract> v1.*
3252 B<might behave differently> in v1.50.
3254 The main changes are:
3260 support for literal SQL through the C<< \ [ $sql, @bind ] >> syntax.
3264 support for the { operator => \"..." } construct (to embed literal SQL)
3268 support for the { operator => \["...", @bind] } construct (to embed literal SQL with bind values)
3272 optional support for L<array datatypes|/"Inserting and Updating Arrays">
3276 defensive programming: check arguments
3280 fixed bug with global logic, which was previously implemented
3281 through global variables yielding side-effects. Prior versions would
3282 interpret C<< [ {cond1, cond2}, [cond3, cond4] ] >>
3283 as C<< "(cond1 AND cond2) OR (cond3 AND cond4)" >>.
3284 Now this is interpreted
3285 as C<< "(cond1 AND cond2) OR (cond3 OR cond4)" >>.
3290 fixed semantics of _bindtype on array args
3294 dropped the C<_anoncopy> of the %where tree. No longer necessary,
3295 we just avoid shifting arrays within that tree.
3299 dropped the C<_modlogic> function
3303 =head1 ACKNOWLEDGEMENTS
3305 There are a number of individuals that have really helped out with
3306 this module. Unfortunately, most of them submitted bugs via CPAN
3307 so I have no idea who they are! But the people I do know are:
3309 Ash Berlin (order_by hash term support)
3310 Matt Trout (DBIx::Class support)
3311 Mark Stosberg (benchmarking)
3312 Chas Owens (initial "IN" operator support)
3313 Philip Collins (per-field SQL functions)
3314 Eric Kolve (hashref "AND" support)
3315 Mike Fragassi (enhancements to "BETWEEN" and "LIKE")
3316 Dan Kubb (support for "quote_char" and "name_sep")
3317 Guillermo Roditi (patch to cleanup "IN" and "BETWEEN", fix and tests for _order_by)
3318 Laurent Dami (internal refactoring, extensible list of special operators, literal SQL)
3319 Norbert Buchmuller (support for literal SQL in hashpair, misc. fixes & tests)
3320 Peter Rabbitson (rewrite of SQLA::Test, misc. fixes & tests)
3321 Oliver Charles (support for "RETURNING" after "INSERT")
3327 L<DBIx::Class>, L<DBIx::Abstract>, L<CGI::FormBuilder>, L<HTML::QuickTable>.
3331 Copyright (c) 2001-2007 Nathan Wiger <nwiger@cpan.org>. All Rights Reserved.
3333 This module is actively maintained by Matt Trout <mst@shadowcatsystems.co.uk>
3335 For support, your best bet is to try the C<DBIx::Class> users mailing list.
3336 While not an official support venue, C<DBIx::Class> makes heavy use of
3337 C<SQL::Abstract>, and as such list members there are very familiar with
3338 how to create queries.
3342 This module is free software; you may copy this under the same
3343 terms as perl itself (either the GNU General Public License or
3344 the Artistic License)