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 = '1.81';
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) : $sql;
230 # Used by DBIx::Class::SQLMaker->insert
231 sub _insert_returning { shift->_returning(@_) }
234 my ($self, $options) = @_;
236 my $f = $options->{returning};
238 my $fieldlist = $self->_SWITCH_refkind($f, {
239 ARRAYREF => sub {join ', ', map { $self->_quote($_) } @$f;},
240 SCALAR => sub {$self->_quote($f)},
241 SCALARREF => sub {$$f},
243 return $self->_sqlcase(' returning ') . $fieldlist;
246 sub _insert_HASHREF { # explicit list of fields and then values
247 my ($self, $data) = @_;
249 my @fields = sort keys %$data;
251 my ($sql, @bind) = $self->_insert_values($data);
254 $_ = $self->_quote($_) foreach @fields;
255 $sql = "( ".join(", ", @fields).") ".$sql;
257 return ($sql, @bind);
260 sub _insert_ARRAYREF { # just generate values(?,?) part (no list of fields)
261 my ($self, $data) = @_;
263 # no names (arrayref) so can't generate bindtype
264 $self->{bindtype} ne 'columns'
265 or belch "can't do 'columns' bindtype when called with arrayref";
267 # fold the list of values into a hash of column name - value pairs
268 # (where the column names are artificially generated, and their
269 # lexicographical ordering keep the ordering of the original list)
270 my $i = "a"; # incremented values will be in lexicographical order
271 my $data_in_hash = { map { ($i++ => $_) } @$data };
273 return $self->_insert_values($data_in_hash);
276 sub _insert_ARRAYREFREF { # literal SQL with bind
277 my ($self, $data) = @_;
279 my ($sql, @bind) = @${$data};
280 $self->_assert_bindval_matches_bindtype(@bind);
282 return ($sql, @bind);
286 sub _insert_SCALARREF { # literal SQL without bind
287 my ($self, $data) = @_;
293 my ($self, $data) = @_;
295 my (@values, @all_bind);
296 foreach my $column (sort keys %$data) {
297 my $v = $data->{$column};
299 $self->_SWITCH_refkind($v, {
302 if ($self->{array_datatypes}) { # if array datatype are activated
304 push @all_bind, $self->_bindtype($column, $v);
306 else { # else literal SQL with bind
307 my ($sql, @bind) = @$v;
308 $self->_assert_bindval_matches_bindtype(@bind);
310 push @all_bind, @bind;
314 ARRAYREFREF => sub { # literal SQL with bind
315 my ($sql, @bind) = @${$v};
316 $self->_assert_bindval_matches_bindtype(@bind);
318 push @all_bind, @bind;
321 # THINK : anything useful to do with a HASHREF ?
322 HASHREF => sub { # (nothing, but old SQLA passed it through)
323 #TODO in SQLA >= 2.0 it will die instead
324 belch "HASH ref as bind value in insert is not supported";
326 push @all_bind, $self->_bindtype($column, $v);
329 SCALARREF => sub { # literal SQL without bind
333 SCALAR_or_UNDEF => sub {
335 push @all_bind, $self->_bindtype($column, $v);
342 my $sql = $self->_sqlcase('values')." ( ".join(", ", @values)." )";
343 return ($sql, @all_bind);
348 #======================================================================
350 #======================================================================
355 my $table = $self->_table(shift);
356 my $data = shift || return;
360 # first build the 'SET' part of the sql statement
361 my (@set, @all_bind);
362 puke "Unsupported data type specified to \$sql->update"
363 unless ref $data eq 'HASH';
365 for my $k (sort keys %$data) {
368 my $label = $self->_quote($k);
370 $self->_SWITCH_refkind($v, {
372 if ($self->{array_datatypes}) { # array datatype
373 push @set, "$label = ?";
374 push @all_bind, $self->_bindtype($k, $v);
376 else { # literal SQL with bind
377 my ($sql, @bind) = @$v;
378 $self->_assert_bindval_matches_bindtype(@bind);
379 push @set, "$label = $sql";
380 push @all_bind, @bind;
383 ARRAYREFREF => sub { # literal SQL with bind
384 my ($sql, @bind) = @${$v};
385 $self->_assert_bindval_matches_bindtype(@bind);
386 push @set, "$label = $sql";
387 push @all_bind, @bind;
389 SCALARREF => sub { # literal SQL without bind
390 push @set, "$label = $$v";
393 my ($op, $arg, @rest) = %$v;
395 puke 'Operator calls in update must be in the form { -op => $arg }'
396 if (@rest or not $op =~ /^\-(.+)/);
398 local $self->{_nested_func_lhs} = $k;
399 my ($sql, @bind) = $self->_where_unary_op ($1, $arg);
401 push @set, "$label = $sql";
402 push @all_bind, @bind;
404 SCALAR_or_UNDEF => sub {
405 push @set, "$label = ?";
406 push @all_bind, $self->_bindtype($k, $v);
412 my $sql = $self->_sqlcase('update') . " $table " . $self->_sqlcase('set ')
416 my($where_sql, @where_bind) = $self->where($where);
418 push @all_bind, @where_bind;
421 if ($options->{returning}) {
422 my ($returning_sql, @returning_bind) = $self->_update_returning ($options);
423 $sql .= $returning_sql;
424 push @all_bind, @returning_bind;
427 return wantarray ? ($sql, @all_bind) : $sql;
430 sub _update_returning { shift->_returning(@_) }
434 #======================================================================
436 #======================================================================
441 my $table = $self->_table(shift);
442 my $fields = shift || '*';
446 my($where_sql, @bind) = $self->where($where, $order);
448 my $f = (ref $fields eq 'ARRAY') ? join ', ', map { $self->_quote($_) } @$fields
450 my $sql = join(' ', $self->_sqlcase('select'), $f,
451 $self->_sqlcase('from'), $table)
454 return wantarray ? ($sql, @bind) : $sql;
457 #======================================================================
459 #======================================================================
464 my $table = $self->_table(shift);
468 my($where_sql, @bind) = $self->where($where);
469 my $sql = $self->_sqlcase('delete from') . " $table" . $where_sql;
471 return wantarray ? ($sql, @bind) : $sql;
475 #======================================================================
477 #======================================================================
481 # Finally, a separate routine just to handle WHERE clauses
483 my ($self, $where, $order) = @_;
486 my ($sql, @bind) = $self->_recurse_where($where);
487 $sql = $sql ? $self->_sqlcase(' where ') . "( $sql )" : '';
491 my ($order_sql, @order_bind) = $self->_order_by($order);
493 push @bind, @order_bind;
496 return wantarray ? ($sql, @bind) : $sql;
501 my ($self, $where, $logic) = @_;
503 # dispatch on appropriate method according to refkind of $where
504 my $method = $self->_METHOD_FOR_refkind("_where", $where);
506 my ($sql, @bind) = $self->$method($where, $logic);
508 # DBIx::Class used to call _recurse_where in scalar context
509 # something else might too...
511 return ($sql, @bind);
514 belch "Calling _recurse_where in scalar context is deprecated and will go away before 2.0";
521 #======================================================================
522 # WHERE: top-level ARRAYREF
523 #======================================================================
526 sub _where_ARRAYREF {
527 my ($self, $where, $logic) = @_;
529 $logic = uc($logic || $self->{logic});
530 $logic eq 'AND' or $logic eq 'OR' or puke "unknown logic: $logic";
532 my @clauses = @$where;
534 my (@sql_clauses, @all_bind);
535 # need to use while() so can shift() for pairs
537 my $el = shift @clauses;
539 $el = undef if (defined $el and ! length $el);
541 # switch according to kind of $el and get corresponding ($sql, @bind)
542 my ($sql, @bind) = $self->_SWITCH_refkind($el, {
544 # skip empty elements, otherwise get invalid trailing AND stuff
545 ARRAYREF => sub {$self->_recurse_where($el) if @$el},
549 $self->_assert_bindval_matches_bindtype(@b);
553 HASHREF => sub {$self->_recurse_where($el, 'and') if %$el},
555 SCALARREF => sub { ($$el); },
558 # top-level arrayref with scalars, recurse in pairs
559 $self->_recurse_where({$el => shift(@clauses)})
562 UNDEF => sub {puke "Supplying an empty left hand side argument is not supported in array-pairs" },
566 push @sql_clauses, $sql;
567 push @all_bind, @bind;
571 return $self->_join_sql_clauses($logic, \@sql_clauses, \@all_bind);
574 #======================================================================
575 # WHERE: top-level ARRAYREFREF
576 #======================================================================
578 sub _where_ARRAYREFREF {
579 my ($self, $where) = @_;
580 my ($sql, @bind) = @$$where;
581 $self->_assert_bindval_matches_bindtype(@bind);
582 return ($sql, @bind);
585 #======================================================================
586 # WHERE: top-level HASHREF
587 #======================================================================
590 my ($self, $where) = @_;
591 my (@sql_clauses, @all_bind);
593 for my $k (sort keys %$where) {
594 my $v = $where->{$k};
596 # ($k => $v) is either a special unary op or a regular hashpair
597 my ($sql, @bind) = do {
599 # put the operator in canonical form
601 $op = substr $op, 1; # remove initial dash
602 $op =~ s/^\s+|\s+$//g;# remove leading/trailing space
603 $op =~ s/\s+/ /g; # compress whitespace
605 # so that -not_foo works correctly
606 $op =~ s/^not_/NOT /i;
608 $self->_debug("Unary OP(-$op) within hashref, recursing...");
609 my ($s, @b) = $self->_where_unary_op ($op, $v);
611 # top level vs nested
612 # we assume that handled unary ops will take care of their ()s
614 List::Util::first {$op =~ $_->{regex}} @{$self->{unary_ops}}
616 ( defined $self->{_nested_func_lhs} and $self->{_nested_func_lhs} eq $k )
622 if (is_literal_value ($v) ) {
623 belch 'Hash-pairs consisting of an empty string with a literal are deprecated, and will be removed in 2.0: use -and => [ $literal ] instead';
626 puke "Supplying an empty left hand side argument is not supported in hash-pairs";
630 my $method = $self->_METHOD_FOR_refkind("_where_hashpair", $v);
631 $self->$method($k, $v);
635 push @sql_clauses, $sql;
636 push @all_bind, @bind;
639 return $self->_join_sql_clauses('and', \@sql_clauses, \@all_bind);
642 sub _where_unary_op {
643 my ($self, $op, $rhs) = @_;
645 # top level special ops are illegal in general
646 # this includes the -ident/-value ops (dual purpose unary and special)
647 puke "Illegal use of top-level '-$op'"
648 if ! defined $self->{_nested_func_lhs} and List::Util::first {$op =~ $_->{regex}} @{$self->{special_ops}};
650 if (my $op_entry = List::Util::first {$op =~ $_->{regex}} @{$self->{unary_ops}}) {
651 my $handler = $op_entry->{handler};
653 if (not ref $handler) {
654 if ($op =~ s/ [_\s]? \d+ $//x ) {
655 belch 'Use of [and|or|nest]_N modifiers is deprecated and will be removed in SQLA v2.0. '
656 . "You probably wanted ...-and => [ -$op => COND1, -$op => COND2 ... ]";
658 return $self->$handler ($op, $rhs);
660 elsif (ref $handler eq 'CODE') {
661 return $handler->($self, $op, $rhs);
664 puke "Illegal handler for operator $op - expecting a method name or a coderef";
668 $self->_debug("Generic unary OP: $op - recursing as function");
670 $self->_assert_pass_injection_guard($op);
672 my ($sql, @bind) = $self->_SWITCH_refkind ($rhs, {
674 puke "Illegal use of top-level '-$op'"
675 unless defined $self->{_nested_func_lhs};
678 $self->_convert('?'),
679 $self->_bindtype($self->{_nested_func_lhs}, $rhs)
683 $self->_recurse_where ($rhs)
687 $sql = sprintf ('%s %s',
688 $self->_sqlcase($op),
692 return ($sql, @bind);
695 sub _where_op_ANDOR {
696 my ($self, $op, $v) = @_;
698 $self->_SWITCH_refkind($v, {
700 return $self->_where_ARRAYREF($v, $op);
704 return ( $op =~ /^or/i )
705 ? $self->_where_ARRAYREF( [ map { $_ => $v->{$_} } ( sort keys %$v ) ], $op )
706 : $self->_where_HASHREF($v);
710 puke "-$op => \\\$scalar makes little sense, use " .
712 ? '[ \$scalar, \%rest_of_conditions ] instead'
713 : '-and => [ \$scalar, \%rest_of_conditions ] instead'
718 puke "-$op => \\[...] makes little sense, use " .
720 ? '[ \[...], \%rest_of_conditions ] instead'
721 : '-and => [ \[...], \%rest_of_conditions ] instead'
725 SCALAR => sub { # permissively interpreted as SQL
726 puke "-$op => \$value makes little sense, use -bool => \$value instead";
730 puke "-$op => undef not supported";
736 my ($self, $op, $v) = @_;
738 $self->_SWITCH_refkind($v, {
740 SCALAR => sub { # permissively interpreted as SQL
741 belch "literal SQL should be -nest => \\'scalar' "
742 . "instead of -nest => 'scalar' ";
747 puke "-$op => undef not supported";
751 $self->_recurse_where ($v);
759 my ($self, $op, $v) = @_;
761 my ($s, @b) = $self->_SWITCH_refkind($v, {
762 SCALAR => sub { # interpreted as SQL column
763 $self->_convert($self->_quote($v));
767 puke "-$op => undef not supported";
771 $self->_recurse_where ($v);
775 $s = "(NOT $s)" if $op =~ /^not/i;
780 sub _where_op_IDENT {
782 my ($op, $rhs) = splice @_, -2;
783 if (! defined $rhs or length ref $rhs) {
784 puke "-$op requires a single plain scalar argument (a quotable identifier)";
787 # in case we are called as a top level special op (no '=')
790 $_ = $self->_convert($self->_quote($_)) for ($lhs, $rhs);
798 sub _where_op_VALUE {
800 my ($op, $rhs) = splice @_, -2;
802 # in case we are called as a top level special op (no '=')
806 if (! defined $rhs) {
808 ? $self->_convert($self->_quote($lhs)) . ' IS NULL'
815 ( defined $lhs ? $lhs : $self->{_nested_func_lhs} ),
822 $self->_convert($self->_quote($lhs)) . ' = ' . $self->_convert('?'),
826 $self->_convert('?'),
832 sub _where_hashpair_ARRAYREF {
833 my ($self, $k, $v) = @_;
836 my @v = @$v; # need copy because of shift below
837 $self->_debug("ARRAY($k) means distribute over elements");
839 # put apart first element if it is an operator (-and, -or)
841 (defined $v[0] && $v[0] =~ /^ - (?: AND|OR ) $/ix)
845 my @distributed = map { {$k => $_} } @v;
848 $self->_debug("OP($op) reinjected into the distributed array");
849 unshift @distributed, $op;
852 my $logic = $op ? substr($op, 1) : '';
854 return $self->_recurse_where(\@distributed, $logic);
857 $self->_debug("empty ARRAY($k) means 0=1");
858 return ($self->{sqlfalse});
862 sub _where_hashpair_HASHREF {
863 my ($self, $k, $v, $logic) = @_;
866 local $self->{_nested_func_lhs} = defined $self->{_nested_func_lhs}
867 ? $self->{_nested_func_lhs}
871 my ($all_sql, @all_bind);
873 for my $orig_op (sort keys %$v) {
874 my $val = $v->{$orig_op};
876 # put the operator in canonical form
879 # FIXME - we need to phase out dash-less ops
880 $op =~ s/^-//; # remove possible initial dash
881 $op =~ s/^\s+|\s+$//g;# remove leading/trailing space
882 $op =~ s/\s+/ /g; # compress whitespace
884 $self->_assert_pass_injection_guard($op);
887 $op =~ s/^is_not/IS NOT/i;
889 # so that -not_foo works correctly
890 $op =~ s/^not_/NOT /i;
892 # another retarded special case: foo => { $op => { -value => undef } }
893 if (ref $val eq 'HASH' and keys %$val == 1 and exists $val->{-value} and ! defined $val->{-value} ) {
899 # CASE: col-value logic modifiers
900 if ( $orig_op =~ /^ \- (and|or) $/xi ) {
901 ($sql, @bind) = $self->_where_hashpair_HASHREF($k, $val, $1);
903 # CASE: special operators like -in or -between
904 elsif ( my $special_op = List::Util::first {$op =~ $_->{regex}} @{$self->{special_ops}} ) {
905 my $handler = $special_op->{handler};
907 puke "No handler supplied for special operator $orig_op";
909 elsif (not ref $handler) {
910 ($sql, @bind) = $self->$handler ($k, $op, $val);
912 elsif (ref $handler eq 'CODE') {
913 ($sql, @bind) = $handler->($self, $k, $op, $val);
916 puke "Illegal handler for special operator $orig_op - expecting a method name or a coderef";
920 $self->_SWITCH_refkind($val, {
922 ARRAYREF => sub { # CASE: col => {op => \@vals}
923 ($sql, @bind) = $self->_where_field_op_ARRAYREF($k, $op, $val);
926 ARRAYREFREF => sub { # CASE: col => {op => \[$sql, @bind]} (literal SQL with bind)
927 my ($sub_sql, @sub_bind) = @$$val;
928 $self->_assert_bindval_matches_bindtype(@sub_bind);
929 $sql = join ' ', $self->_convert($self->_quote($k)),
930 $self->_sqlcase($op),
935 UNDEF => sub { # CASE: col => {op => undef} : sql "IS (NOT)? NULL"
937 $op =~ /^not$/i ? 'is not' # legacy
938 : $op =~ $self->{equality_op} ? 'is'
939 : $op =~ $self->{like_op} ? belch("Supplying an undefined argument to '@{[ uc $op]}' is deprecated") && 'is'
940 : $op =~ $self->{inequality_op} ? 'is not'
941 : $op =~ $self->{not_like_op} ? belch("Supplying an undefined argument to '@{[ uc $op]}' is deprecated") && 'is not'
942 : puke "unexpected operator '$orig_op' with undef operand";
944 $sql = $self->_quote($k) . $self->_sqlcase(" $is null");
947 FALLBACK => sub { # CASE: col => {op/func => $stuff}
948 ($sql, @bind) = $self->_where_unary_op ($op, $val);
951 $self->_convert($self->_quote($k)),
952 $self->{_nested_func_lhs} eq $k ? $sql : "($sql)", # top level vs nested
958 ($all_sql) = (defined $all_sql and $all_sql) ? $self->_join_sql_clauses($logic, [$all_sql, $sql], []) : $sql;
959 push @all_bind, @bind;
961 return ($all_sql, @all_bind);
964 sub _where_field_IS {
965 my ($self, $k, $op, $v) = @_;
967 my ($s) = $self->_SWITCH_refkind($v, {
970 $self->_convert($self->_quote($k)),
971 map { $self->_sqlcase($_)} ($op, 'null')
974 puke "$op can only take undef as argument";
981 sub _where_field_op_ARRAYREF {
982 my ($self, $k, $op, $vals) = @_;
984 my @vals = @$vals; #always work on a copy
987 $self->_debug(sprintf '%s means multiple elements: [ %s ]',
989 join (', ', map { defined $_ ? "'$_'" : 'NULL' } @vals ),
992 # see if the first element is an -and/-or op
994 if (defined $vals[0] && $vals[0] =~ /^ - ( AND|OR ) $/ix) {
999 # a long standing API wart - an attempt to change this behavior during
1000 # the 1.50 series failed *spectacularly*. Warn instead and leave the
1005 (!$logic or $logic eq 'OR')
1007 ( $op =~ $self->{inequality_op} or $op =~ $self->{not_like_op} )
1010 belch "A multi-element arrayref as an argument to the inequality op '$o' "
1011 . 'is technically equivalent to an always-true 1=1 (you probably wanted '
1012 . "to say ...{ \$inequality_op => [ -and => \@values ] }... instead)"
1016 # distribute $op over each remaining member of @vals, append logic if exists
1017 return $self->_recurse_where([map { {$k => {$op, $_}} } @vals], $logic);
1021 # try to DWIM on equality operators
1023 $op =~ $self->{equality_op} ? $self->{sqlfalse}
1024 : $op =~ $self->{like_op} ? belch("Supplying an empty arrayref to '@{[ uc $op]}' is deprecated") && $self->{sqlfalse}
1025 : $op =~ $self->{inequality_op} ? $self->{sqltrue}
1026 : $op =~ $self->{not_like_op} ? belch("Supplying an empty arrayref to '@{[ uc $op]}' is deprecated") && $self->{sqltrue}
1027 : puke "operator '$op' applied on an empty array (field '$k')";
1032 sub _where_hashpair_SCALARREF {
1033 my ($self, $k, $v) = @_;
1034 $self->_debug("SCALAR($k) means literal SQL: $$v");
1035 my $sql = $self->_quote($k) . " " . $$v;
1039 # literal SQL with bind
1040 sub _where_hashpair_ARRAYREFREF {
1041 my ($self, $k, $v) = @_;
1042 $self->_debug("REF($k) means literal SQL: @${$v}");
1043 my ($sql, @bind) = @$$v;
1044 $self->_assert_bindval_matches_bindtype(@bind);
1045 $sql = $self->_quote($k) . " " . $sql;
1046 return ($sql, @bind );
1049 # literal SQL without bind
1050 sub _where_hashpair_SCALAR {
1051 my ($self, $k, $v) = @_;
1052 $self->_debug("NOREF($k) means simple key=val: $k $self->{cmp} $v");
1053 my $sql = join ' ', $self->_convert($self->_quote($k)),
1054 $self->_sqlcase($self->{cmp}),
1055 $self->_convert('?');
1056 my @bind = $self->_bindtype($k, $v);
1057 return ( $sql, @bind);
1061 sub _where_hashpair_UNDEF {
1062 my ($self, $k, $v) = @_;
1063 $self->_debug("UNDEF($k) means IS NULL");
1064 my $sql = $self->_quote($k) . $self->_sqlcase(' is null');
1068 #======================================================================
1069 # WHERE: TOP-LEVEL OTHERS (SCALARREF, SCALAR, UNDEF)
1070 #======================================================================
1073 sub _where_SCALARREF {
1074 my ($self, $where) = @_;
1077 $self->_debug("SCALAR(*top) means literal SQL: $$where");
1083 my ($self, $where) = @_;
1086 $self->_debug("NOREF(*top) means literal SQL: $where");
1097 #======================================================================
1098 # WHERE: BUILTIN SPECIAL OPERATORS (-in, -between)
1099 #======================================================================
1102 sub _where_field_BETWEEN {
1103 my ($self, $k, $op, $vals) = @_;
1105 my ($label, $and, $placeholder);
1106 $label = $self->_convert($self->_quote($k));
1107 $and = ' ' . $self->_sqlcase('and') . ' ';
1108 $placeholder = $self->_convert('?');
1109 $op = $self->_sqlcase($op);
1111 my $invalid_args = "Operator '$op' requires either an arrayref with two defined values or expressions, or a single literal scalarref/arrayref-ref";
1113 my ($clause, @bind) = $self->_SWITCH_refkind($vals, {
1114 ARRAYREFREF => sub {
1115 my ($s, @b) = @$$vals;
1116 $self->_assert_bindval_matches_bindtype(@b);
1123 puke $invalid_args if @$vals != 2;
1125 my (@all_sql, @all_bind);
1126 foreach my $val (@$vals) {
1127 my ($sql, @bind) = $self->_SWITCH_refkind($val, {
1129 return ($placeholder, $self->_bindtype($k, $val) );
1134 ARRAYREFREF => sub {
1135 my ($sql, @bind) = @$$val;
1136 $self->_assert_bindval_matches_bindtype(@bind);
1137 return ($sql, @bind);
1140 my ($func, $arg, @rest) = %$val;
1141 puke ("Only simple { -func => arg } functions accepted as sub-arguments to BETWEEN")
1142 if (@rest or $func !~ /^ \- (.+)/x);
1143 $self->_where_unary_op ($1 => $arg);
1149 push @all_sql, $sql;
1150 push @all_bind, @bind;
1154 (join $and, @all_sql),
1163 my $sql = "( $label $op $clause )";
1164 return ($sql, @bind)
1168 sub _where_field_IN {
1169 my ($self, $k, $op, $vals) = @_;
1171 # backwards compatibility : if scalar, force into an arrayref
1172 $vals = [$vals] if defined $vals && ! ref $vals;
1174 my ($label) = $self->_convert($self->_quote($k));
1175 my ($placeholder) = $self->_convert('?');
1176 $op = $self->_sqlcase($op);
1178 my ($sql, @bind) = $self->_SWITCH_refkind($vals, {
1179 ARRAYREF => sub { # list of choices
1180 if (@$vals) { # nonempty list
1181 my (@all_sql, @all_bind);
1183 for my $val (@$vals) {
1184 my ($sql, @bind) = $self->_SWITCH_refkind($val, {
1186 return ($placeholder, $val);
1191 ARRAYREFREF => sub {
1192 my ($sql, @bind) = @$$val;
1193 $self->_assert_bindval_matches_bindtype(@bind);
1194 return ($sql, @bind);
1197 my ($func, $arg, @rest) = %$val;
1198 puke ("Only simple { -func => arg } functions accepted as sub-arguments to IN")
1199 if (@rest or $func !~ /^ \- (.+)/x);
1200 $self->_where_unary_op ($1 => $arg);
1204 'SQL::Abstract before v1.75 used to generate incorrect SQL when the '
1205 . "-$op operator was given an undef-containing list: !!!AUDIT YOUR CODE "
1206 . 'AND DATA!!! (the upcoming Data::Query-based version of SQL::Abstract '
1207 . 'will emit the logically correct SQL instead of raising this exception)'
1211 push @all_sql, $sql;
1212 push @all_bind, @bind;
1216 sprintf ('%s %s ( %s )',
1219 join (', ', @all_sql)
1221 $self->_bindtype($k, @all_bind),
1224 else { # empty list : some databases won't understand "IN ()", so DWIM
1225 my $sql = ($op =~ /\bnot\b/i) ? $self->{sqltrue} : $self->{sqlfalse};
1230 SCALARREF => sub { # literal SQL
1231 my $sql = $self->_open_outer_paren ($$vals);
1232 return ("$label $op ( $sql )");
1234 ARRAYREFREF => sub { # literal SQL with bind
1235 my ($sql, @bind) = @$$vals;
1236 $self->_assert_bindval_matches_bindtype(@bind);
1237 $sql = $self->_open_outer_paren ($sql);
1238 return ("$label $op ( $sql )", @bind);
1242 puke "Argument passed to the '$op' operator can not be undefined";
1246 puke "special op $op requires an arrayref (or scalarref/arrayref-ref)";
1250 return ($sql, @bind);
1253 # Some databases (SQLite) treat col IN (1, 2) different from
1254 # col IN ( (1, 2) ). Use this to strip all outer parens while
1255 # adding them back in the corresponding method
1256 sub _open_outer_paren {
1257 my ($self, $sql) = @_;
1259 while ( my ($inner) = $sql =~ /^ \s* \( (.*) \) \s* $/xs ) {
1261 # there are closing parens inside, need the heavy duty machinery
1262 # to reevaluate the extraction starting from $sql (full reevaluation)
1263 if ( $inner =~ /\)/ ) {
1264 require Text::Balanced;
1266 my (undef, $remainder) = do {
1267 # idiotic design - writes to $@ but *DOES NOT* throw exceptions
1269 Text::Balanced::extract_bracketed( $sql, '()', qr/\s*/ );
1272 # the entire expression needs to be a balanced bracketed thing
1273 # (after an extract no remainder sans trailing space)
1274 last if defined $remainder and $remainder =~ /\S/;
1284 #======================================================================
1286 #======================================================================
1289 my ($self, $arg) = @_;
1292 for my $c ($self->_order_by_chunks ($arg) ) {
1293 $self->_SWITCH_refkind ($c, {
1294 SCALAR => sub { push @sql, $c },
1295 ARRAYREF => sub { push @sql, shift @$c; push @bind, @$c },
1301 $self->_sqlcase(' order by'),
1307 return wantarray ? ($sql, @bind) : $sql;
1310 sub _order_by_chunks {
1311 my ($self, $arg) = @_;
1313 return $self->_SWITCH_refkind($arg, {
1316 map { $self->_order_by_chunks ($_ ) } @$arg;
1319 ARRAYREFREF => sub {
1320 my ($s, @b) = @$$arg;
1321 $self->_assert_bindval_matches_bindtype(@b);
1325 SCALAR => sub {$self->_quote($arg)},
1327 UNDEF => sub {return () },
1329 SCALARREF => sub {$$arg}, # literal SQL, no quoting
1332 # get first pair in hash
1333 my ($key, $val, @rest) = %$arg;
1335 return () unless $key;
1337 if ( @rest or not $key =~ /^-(desc|asc)/i ) {
1338 puke "hash passed to _order_by must have exactly one key (-desc or -asc)";
1344 for my $c ($self->_order_by_chunks ($val)) {
1347 $self->_SWITCH_refkind ($c, {
1352 ($sql, @bind) = @$c;
1356 $sql = $sql . ' ' . $self->_sqlcase($direction);
1358 push @ret, [ $sql, @bind];
1367 #======================================================================
1368 # DATASOURCE (FOR NOW, JUST PLAIN TABLE OR LIST OF TABLES)
1369 #======================================================================
1374 $self->_SWITCH_refkind($from, {
1375 ARRAYREF => sub {join ', ', map { $self->_quote($_) } @$from;},
1376 SCALAR => sub {$self->_quote($from)},
1377 SCALARREF => sub {$$from},
1382 #======================================================================
1384 #======================================================================
1386 # highly optimized, as it's called way too often
1388 # my ($self, $label) = @_;
1390 return '' unless defined $_[1];
1391 return ${$_[1]} if ref($_[1]) eq 'SCALAR';
1393 unless ($_[0]->{quote_char}) {
1394 $_[0]->_assert_pass_injection_guard($_[1]);
1398 my $qref = ref $_[0]->{quote_char};
1401 ($l, $r) = ( $_[0]->{quote_char}, $_[0]->{quote_char} );
1403 elsif ($qref eq 'ARRAY') {
1404 ($l, $r) = @{$_[0]->{quote_char}};
1407 puke "Unsupported quote_char format: $_[0]->{quote_char}";
1409 my $esc = $_[0]->{escape_char} || $r;
1411 # parts containing * are naturally unquoted
1412 return join( $_[0]->{name_sep}||'', map
1413 { $_ eq '*' ? $_ : do { (my $n = $_) =~ s/(\Q$esc\E|\Q$r\E)/$esc$1/g; $l . $n . $r } }
1414 ( $_[0]->{name_sep} ? split (/\Q$_[0]->{name_sep}\E/, $_[1] ) : $_[1] )
1419 # Conversion, if applicable
1421 #my ($self, $arg) = @_;
1422 if ($_[0]->{convert}) {
1423 return $_[0]->_sqlcase($_[0]->{convert}) .'(' . $_[1] . ')';
1430 #my ($self, $col, @vals) = @_;
1431 # called often - tighten code
1432 return $_[0]->{bindtype} eq 'columns'
1433 ? map {[$_[1], $_]} @_[2 .. $#_]
1438 # Dies if any element of @bind is not in [colname => value] format
1439 # if bindtype is 'columns'.
1440 sub _assert_bindval_matches_bindtype {
1441 # my ($self, @bind) = @_;
1443 if ($self->{bindtype} eq 'columns') {
1445 if (!defined $_ || ref($_) ne 'ARRAY' || @$_ != 2) {
1446 puke "bindtype 'columns' selected, you need to pass: [column_name => bind_value]"
1452 sub _join_sql_clauses {
1453 my ($self, $logic, $clauses_aref, $bind_aref) = @_;
1455 if (@$clauses_aref > 1) {
1456 my $join = " " . $self->_sqlcase($logic) . " ";
1457 my $sql = '( ' . join($join, @$clauses_aref) . ' )';
1458 return ($sql, @$bind_aref);
1460 elsif (@$clauses_aref) {
1461 return ($clauses_aref->[0], @$bind_aref); # no parentheses
1464 return (); # if no SQL, ignore @$bind_aref
1469 # Fix SQL case, if so requested
1471 # LDNOTE: if $self->{case} is true, then it contains 'lower', so we
1472 # don't touch the argument ... crooked logic, but let's not change it!
1473 return $_[0]->{case} ? $_[1] : uc($_[1]);
1477 #======================================================================
1478 # DISPATCHING FROM REFKIND
1479 #======================================================================
1482 my ($self, $data) = @_;
1484 return 'UNDEF' unless defined $data;
1486 # blessed objects are treated like scalars
1487 my $ref = (Scalar::Util::blessed $data) ? '' : ref $data;
1489 return 'SCALAR' unless $ref;
1492 while ($ref eq 'REF') {
1494 $ref = (Scalar::Util::blessed $data) ? '' : ref $data;
1498 return ($ref||'SCALAR') . ('REF' x $n_steps);
1502 my ($self, $data) = @_;
1503 my @try = ($self->_refkind($data));
1504 push @try, 'SCALAR_or_UNDEF' if $try[0] eq 'SCALAR' || $try[0] eq 'UNDEF';
1505 push @try, 'FALLBACK';
1509 sub _METHOD_FOR_refkind {
1510 my ($self, $meth_prefix, $data) = @_;
1513 for (@{$self->_try_refkind($data)}) {
1514 $method = $self->can($meth_prefix."_".$_)
1518 return $method || puke "cannot dispatch on '$meth_prefix' for ".$self->_refkind($data);
1522 sub _SWITCH_refkind {
1523 my ($self, $data, $dispatch_table) = @_;
1526 for (@{$self->_try_refkind($data)}) {
1527 $coderef = $dispatch_table->{$_}
1531 puke "no dispatch entry for ".$self->_refkind($data)
1540 #======================================================================
1541 # VALUES, GENERATE, AUTOLOAD
1542 #======================================================================
1544 # LDNOTE: original code from nwiger, didn't touch code in that section
1545 # I feel the AUTOLOAD stuff should not be the default, it should
1546 # only be activated on explicit demand by user.
1550 my $data = shift || return;
1551 puke "Argument to ", __PACKAGE__, "->values must be a \\%hash"
1552 unless ref $data eq 'HASH';
1555 foreach my $k ( sort keys %$data ) {
1556 my $v = $data->{$k};
1557 $self->_SWITCH_refkind($v, {
1559 if ($self->{array_datatypes}) { # array datatype
1560 push @all_bind, $self->_bindtype($k, $v);
1562 else { # literal SQL with bind
1563 my ($sql, @bind) = @$v;
1564 $self->_assert_bindval_matches_bindtype(@bind);
1565 push @all_bind, @bind;
1568 ARRAYREFREF => sub { # literal SQL with bind
1569 my ($sql, @bind) = @${$v};
1570 $self->_assert_bindval_matches_bindtype(@bind);
1571 push @all_bind, @bind;
1573 SCALARREF => sub { # literal SQL without bind
1575 SCALAR_or_UNDEF => sub {
1576 push @all_bind, $self->_bindtype($k, $v);
1587 my(@sql, @sqlq, @sqlv);
1591 if ($ref eq 'HASH') {
1592 for my $k (sort keys %$_) {
1595 my $label = $self->_quote($k);
1596 if ($r eq 'ARRAY') {
1597 # literal SQL with bind
1598 my ($sql, @bind) = @$v;
1599 $self->_assert_bindval_matches_bindtype(@bind);
1600 push @sqlq, "$label = $sql";
1602 } elsif ($r eq 'SCALAR') {
1603 # literal SQL without bind
1604 push @sqlq, "$label = $$v";
1606 push @sqlq, "$label = ?";
1607 push @sqlv, $self->_bindtype($k, $v);
1610 push @sql, $self->_sqlcase('set'), join ', ', @sqlq;
1611 } elsif ($ref eq 'ARRAY') {
1612 # unlike insert(), assume these are ONLY the column names, i.e. for SQL
1615 if ($r eq 'ARRAY') { # literal SQL with bind
1616 my ($sql, @bind) = @$v;
1617 $self->_assert_bindval_matches_bindtype(@bind);
1620 } elsif ($r eq 'SCALAR') { # literal SQL without bind
1621 # embedded literal SQL
1628 push @sql, '(' . join(', ', @sqlq) . ')';
1629 } elsif ($ref eq 'SCALAR') {
1633 # strings get case twiddled
1634 push @sql, $self->_sqlcase($_);
1638 my $sql = join ' ', @sql;
1640 # this is pretty tricky
1641 # if ask for an array, return ($stmt, @bind)
1642 # otherwise, s/?/shift @sqlv/ to put it inline
1644 return ($sql, @sqlv);
1646 1 while $sql =~ s/\?/my $d = shift(@sqlv);
1647 ref $d ? $d->[1] : $d/e;
1656 # This allows us to check for a local, then _form, attr
1658 my($name) = $AUTOLOAD =~ /.*::(.+)/;
1659 return $self->generate($name, @_);
1670 SQL::Abstract - Generate SQL from Perl data structures
1676 my $sql = SQL::Abstract->new;
1678 my($stmt, @bind) = $sql->select($source, \@fields, \%where, $order);
1680 my($stmt, @bind) = $sql->insert($table, \%fieldvals || \@values);
1682 my($stmt, @bind) = $sql->update($table, \%fieldvals, \%where);
1684 my($stmt, @bind) = $sql->delete($table, \%where);
1686 # Then, use these in your DBI statements
1687 my $sth = $dbh->prepare($stmt);
1688 $sth->execute(@bind);
1690 # Just generate the WHERE clause
1691 my($stmt, @bind) = $sql->where(\%where, $order);
1693 # Return values in the same order, for hashed queries
1694 # See PERFORMANCE section for more details
1695 my @bind = $sql->values(\%fieldvals);
1699 This module was inspired by the excellent L<DBIx::Abstract>.
1700 However, in using that module I found that what I really wanted
1701 to do was generate SQL, but still retain complete control over my
1702 statement handles and use the DBI interface. So, I set out to
1703 create an abstract SQL generation module.
1705 While based on the concepts used by L<DBIx::Abstract>, there are
1706 several important differences, especially when it comes to WHERE
1707 clauses. I have modified the concepts used to make the SQL easier
1708 to generate from Perl data structures and, IMO, more intuitive.
1709 The underlying idea is for this module to do what you mean, based
1710 on the data structures you provide it. The big advantage is that
1711 you don't have to modify your code every time your data changes,
1712 as this module figures it out.
1714 To begin with, an SQL INSERT is as easy as just specifying a hash
1715 of C<key=value> pairs:
1718 name => 'Jimbo Bobson',
1719 phone => '123-456-7890',
1720 address => '42 Sister Lane',
1721 city => 'St. Louis',
1722 state => 'Louisiana',
1725 The SQL can then be generated with this:
1727 my($stmt, @bind) = $sql->insert('people', \%data);
1729 Which would give you something like this:
1731 $stmt = "INSERT INTO people
1732 (address, city, name, phone, state)
1733 VALUES (?, ?, ?, ?, ?)";
1734 @bind = ('42 Sister Lane', 'St. Louis', 'Jimbo Bobson',
1735 '123-456-7890', 'Louisiana');
1737 These are then used directly in your DBI code:
1739 my $sth = $dbh->prepare($stmt);
1740 $sth->execute(@bind);
1742 =head2 Inserting and Updating Arrays
1744 If your database has array types (like for example Postgres),
1745 activate the special option C<< array_datatypes => 1 >>
1746 when creating the C<SQL::Abstract> object.
1747 Then you may use an arrayref to insert and update database array types:
1749 my $sql = SQL::Abstract->new(array_datatypes => 1);
1751 planets => [qw/Mercury Venus Earth Mars/]
1754 my($stmt, @bind) = $sql->insert('solar_system', \%data);
1758 $stmt = "INSERT INTO solar_system (planets) VALUES (?)"
1760 @bind = (['Mercury', 'Venus', 'Earth', 'Mars']);
1763 =head2 Inserting and Updating SQL
1765 In order to apply SQL functions to elements of your C<%data> you may
1766 specify a reference to an arrayref for the given hash value. For example,
1767 if you need to execute the Oracle C<to_date> function on a value, you can
1768 say something like this:
1772 date_entered => \[ "to_date(?,'MM/DD/YYYY')", "03/02/2003" ],
1775 The first value in the array is the actual SQL. Any other values are
1776 optional and would be included in the bind values array. This gives
1779 my($stmt, @bind) = $sql->insert('people', \%data);
1781 $stmt = "INSERT INTO people (name, date_entered)
1782 VALUES (?, to_date(?,'MM/DD/YYYY'))";
1783 @bind = ('Bill', '03/02/2003');
1785 An UPDATE is just as easy, all you change is the name of the function:
1787 my($stmt, @bind) = $sql->update('people', \%data);
1789 Notice that your C<%data> isn't touched; the module will generate
1790 the appropriately quirky SQL for you automatically. Usually you'll
1791 want to specify a WHERE clause for your UPDATE, though, which is
1792 where handling C<%where> hashes comes in handy...
1794 =head2 Complex where statements
1796 This module can generate pretty complicated WHERE statements
1797 easily. For example, simple C<key=value> pairs are taken to mean
1798 equality, and if you want to see if a field is within a set
1799 of values, you can use an arrayref. Let's say we wanted to
1800 SELECT some data based on this criteria:
1803 requestor => 'inna',
1804 worker => ['nwiger', 'rcwe', 'sfz'],
1805 status => { '!=', 'completed' }
1808 my($stmt, @bind) = $sql->select('tickets', '*', \%where);
1810 The above would give you something like this:
1812 $stmt = "SELECT * FROM tickets WHERE
1813 ( requestor = ? ) AND ( status != ? )
1814 AND ( worker = ? OR worker = ? OR worker = ? )";
1815 @bind = ('inna', 'completed', 'nwiger', 'rcwe', 'sfz');
1817 Which you could then use in DBI code like so:
1819 my $sth = $dbh->prepare($stmt);
1820 $sth->execute(@bind);
1826 The methods are simple. There's one for every major SQL operation,
1827 and a constructor you use first. The arguments are specified in a
1828 similar order for each method (table, then fields, then a where
1829 clause) to try and simplify things.
1831 =head2 new(option => 'value')
1833 The C<new()> function takes a list of options and values, and returns
1834 a new B<SQL::Abstract> object which can then be used to generate SQL
1835 through the methods below. The options accepted are:
1841 If set to 'lower', then SQL will be generated in all lowercase. By
1842 default SQL is generated in "textbook" case meaning something like:
1844 SELECT a_field FROM a_table WHERE some_field LIKE '%someval%'
1846 Any setting other than 'lower' is ignored.
1850 This determines what the default comparison operator is. By default
1851 it is C<=>, meaning that a hash like this:
1853 %where = (name => 'nwiger', email => 'nate@wiger.org');
1855 Will generate SQL like this:
1857 WHERE name = 'nwiger' AND email = 'nate@wiger.org'
1859 However, you may want loose comparisons by default, so if you set
1860 C<cmp> to C<like> you would get SQL such as:
1862 WHERE name like 'nwiger' AND email like 'nate@wiger.org'
1864 You can also override the comparison on an individual basis - see
1865 the huge section on L</"WHERE CLAUSES"> at the bottom.
1867 =item sqltrue, sqlfalse
1869 Expressions for inserting boolean values within SQL statements.
1870 By default these are C<1=1> and C<1=0>. They are used
1871 by the special operators C<-in> and C<-not_in> for generating
1872 correct SQL even when the argument is an empty array (see below).
1876 This determines the default logical operator for multiple WHERE
1877 statements in arrays or hashes. If absent, the default logic is "or"
1878 for arrays, and "and" for hashes. This means that a WHERE
1882 event_date => {'>=', '2/13/99'},
1883 event_date => {'<=', '4/24/03'},
1886 will generate SQL like this:
1888 WHERE event_date >= '2/13/99' OR event_date <= '4/24/03'
1890 This is probably not what you want given this query, though (look
1891 at the dates). To change the "OR" to an "AND", simply specify:
1893 my $sql = SQL::Abstract->new(logic => 'and');
1895 Which will change the above C<WHERE> to:
1897 WHERE event_date >= '2/13/99' AND event_date <= '4/24/03'
1899 The logic can also be changed locally by inserting
1900 a modifier in front of an arrayref :
1902 @where = (-and => [event_date => {'>=', '2/13/99'},
1903 event_date => {'<=', '4/24/03'} ]);
1905 See the L</"WHERE CLAUSES"> section for explanations.
1909 This will automatically convert comparisons using the specified SQL
1910 function for both column and value. This is mostly used with an argument
1911 of C<upper> or C<lower>, so that the SQL will have the effect of
1912 case-insensitive "searches". For example, this:
1914 $sql = SQL::Abstract->new(convert => 'upper');
1915 %where = (keywords => 'MaKe iT CAse inSeNSItive');
1917 Will turn out the following SQL:
1919 WHERE upper(keywords) like upper('MaKe iT CAse inSeNSItive')
1921 The conversion can be C<upper()>, C<lower()>, or any other SQL function
1922 that can be applied symmetrically to fields (actually B<SQL::Abstract> does
1923 not validate this option; it will just pass through what you specify verbatim).
1927 This is a kludge because many databases suck. For example, you can't
1928 just bind values using DBI's C<execute()> for Oracle C<CLOB> or C<BLOB> fields.
1929 Instead, you have to use C<bind_param()>:
1931 $sth->bind_param(1, 'reg data');
1932 $sth->bind_param(2, $lots, {ora_type => ORA_CLOB});
1934 The problem is, B<SQL::Abstract> will normally just return a C<@bind> array,
1935 which loses track of which field each slot refers to. Fear not.
1937 If you specify C<bindtype> in new, you can determine how C<@bind> is returned.
1938 Currently, you can specify either C<normal> (default) or C<columns>. If you
1939 specify C<columns>, you will get an array that looks like this:
1941 my $sql = SQL::Abstract->new(bindtype => 'columns');
1942 my($stmt, @bind) = $sql->insert(...);
1945 [ 'column1', 'value1' ],
1946 [ 'column2', 'value2' ],
1947 [ 'column3', 'value3' ],
1950 You can then iterate through this manually, using DBI's C<bind_param()>.
1952 $sth->prepare($stmt);
1955 my($col, $data) = @$_;
1956 if ($col eq 'details' || $col eq 'comments') {
1957 $sth->bind_param($i, $data, {ora_type => ORA_CLOB});
1958 } elsif ($col eq 'image') {
1959 $sth->bind_param($i, $data, {ora_type => ORA_BLOB});
1961 $sth->bind_param($i, $data);
1965 $sth->execute; # execute without @bind now
1967 Now, why would you still use B<SQL::Abstract> if you have to do this crap?
1968 Basically, the advantage is still that you don't have to care which fields
1969 are or are not included. You could wrap that above C<for> loop in a simple
1970 sub called C<bind_fields()> or something and reuse it repeatedly. You still
1971 get a layer of abstraction over manual SQL specification.
1973 Note that if you set L</bindtype> to C<columns>, the C<\[ $sql, @bind ]>
1974 construct (see L</Literal SQL with placeholders and bind values (subqueries)>)
1975 will expect the bind values in this format.
1979 This is the character that a table or column name will be quoted
1980 with. By default this is an empty string, but you could set it to
1981 the character C<`>, to generate SQL like this:
1983 SELECT `a_field` FROM `a_table` WHERE `some_field` LIKE '%someval%'
1985 Alternatively, you can supply an array ref of two items, the first being the left
1986 hand quote character, and the second the right hand quote character. For
1987 example, you could supply C<['[',']']> for SQL Server 2000 compliant quotes
1988 that generates SQL like this:
1990 SELECT [a_field] FROM [a_table] WHERE [some_field] LIKE '%someval%'
1992 Quoting is useful if you have tables or columns names that are reserved
1993 words in your database's SQL dialect.
1997 This is the character that will be used to escape L</quote_char>s appearing
1998 in an identifier before it has been quoted.
2000 The parameter default in case of a single L</quote_char> character is the quote
2003 When opening-closing-style quoting is used (L</quote_char> is an arrayref)
2004 this parameter defaults to the B<closing (right)> L</quote_char>. Occurences
2005 of the B<opening (left)> L</quote_char> within the identifier are currently left
2006 untouched. The default for opening-closing-style quotes may change in future
2007 versions, thus you are B<strongly encouraged> to specify the escape character
2012 This is the character that separates a table and column name. It is
2013 necessary to specify this when the C<quote_char> option is selected,
2014 so that tables and column names can be individually quoted like this:
2016 SELECT `table`.`one_field` FROM `table` WHERE `table`.`other_field` = 1
2018 =item injection_guard
2020 A regular expression C<qr/.../> that is applied to any C<-function> and unquoted
2021 column name specified in a query structure. This is a safety mechanism to avoid
2022 injection attacks when mishandling user input e.g.:
2024 my %condition_as_column_value_pairs = get_values_from_user();
2025 $sqla->select( ... , \%condition_as_column_value_pairs );
2027 If the expression matches an exception is thrown. Note that literal SQL
2028 supplied via C<\'...'> or C<\['...']> is B<not> checked in any way.
2030 Defaults to checking for C<;> and the C<GO> keyword (TransactSQL)
2032 =item array_datatypes
2034 When this option is true, arrayrefs in INSERT or UPDATE are
2035 interpreted as array datatypes and are passed directly
2037 When this option is false, arrayrefs are interpreted
2038 as literal SQL, just like refs to arrayrefs
2039 (but this behavior is for backwards compatibility; when writing
2040 new queries, use the "reference to arrayref" syntax
2046 Takes a reference to a list of "special operators"
2047 to extend the syntax understood by L<SQL::Abstract>.
2048 See section L</"SPECIAL OPERATORS"> for details.
2052 Takes a reference to a list of "unary operators"
2053 to extend the syntax understood by L<SQL::Abstract>.
2054 See section L</"UNARY OPERATORS"> for details.
2060 =head2 insert($table, \@values || \%fieldvals, \%options)
2062 This is the simplest function. You simply give it a table name
2063 and either an arrayref of values or hashref of field/value pairs.
2064 It returns an SQL INSERT statement and a list of bind values.
2065 See the sections on L</"Inserting and Updating Arrays"> and
2066 L</"Inserting and Updating SQL"> for information on how to insert
2067 with those data types.
2069 The optional C<\%options> hash reference may contain additional
2070 options to generate the insert SQL. Currently supported options
2077 Takes either a scalar of raw SQL fields, or an array reference of
2078 field names, and adds on an SQL C<RETURNING> statement at the end.
2079 This allows you to return data generated by the insert statement
2080 (such as row IDs) without performing another C<SELECT> statement.
2081 Note, however, this is not part of the SQL standard and may not
2082 be supported by all database engines.
2086 =head2 update($table, \%fieldvals, \%where, \%options)
2088 This takes a table, hashref of field/value pairs, and an optional
2089 hashref L<WHERE clause|/WHERE CLAUSES>. It returns an SQL UPDATE function and a list
2091 See the sections on L</"Inserting and Updating Arrays"> and
2092 L</"Inserting and Updating SQL"> for information on how to insert
2093 with those data types.
2095 The optional C<\%options> hash reference may contain additional
2096 options to generate the update SQL. Currently supported options
2103 See the C<returning> option to
2104 L<insert|/insert($table, \@values || \%fieldvals, \%options)>.
2108 =head2 select($source, $fields, $where, $order)
2110 This returns a SQL SELECT statement and associated list of bind values, as
2111 specified by the arguments :
2117 Specification of the 'FROM' part of the statement.
2118 The argument can be either a plain scalar (interpreted as a table
2119 name, will be quoted), or an arrayref (interpreted as a list
2120 of table names, joined by commas, quoted), or a scalarref
2121 (literal table name, not quoted), or a ref to an arrayref
2122 (list of literal table names, joined by commas, not quoted).
2126 Specification of the list of fields to retrieve from
2128 The argument can be either an arrayref (interpreted as a list
2129 of field names, will be joined by commas and quoted), or a
2130 plain scalar (literal SQL, not quoted).
2131 Please observe that this API is not as flexible as that of
2132 the first argument C<$source>, for backwards compatibility reasons.
2136 Optional argument to specify the WHERE part of the query.
2137 The argument is most often a hashref, but can also be
2138 an arrayref or plain scalar --
2139 see section L<WHERE clause|/"WHERE CLAUSES"> for details.
2143 Optional argument to specify the ORDER BY part of the query.
2144 The argument can be a scalar, a hashref or an arrayref
2145 -- see section L<ORDER BY clause|/"ORDER BY CLAUSES">
2151 =head2 delete($table, \%where)
2153 This takes a table name and optional hashref L<WHERE clause|/WHERE CLAUSES>.
2154 It returns an SQL DELETE statement and list of bind values.
2156 =head2 where(\%where, $order)
2158 This is used to generate just the WHERE clause. For example,
2159 if you have an arbitrary data structure and know what the
2160 rest of your SQL is going to look like, but want an easy way
2161 to produce a WHERE clause, use this. It returns an SQL WHERE
2162 clause and list of bind values.
2165 =head2 values(\%data)
2167 This just returns the values from the hash C<%data>, in the same
2168 order that would be returned from any of the other above queries.
2169 Using this allows you to markedly speed up your queries if you
2170 are affecting lots of rows. See below under the L</"PERFORMANCE"> section.
2172 =head2 generate($any, 'number', $of, \@data, $struct, \%types)
2174 Warning: This is an experimental method and subject to change.
2176 This returns arbitrarily generated SQL. It's a really basic shortcut.
2177 It will return two different things, depending on return context:
2179 my($stmt, @bind) = $sql->generate('create table', \$table, \@fields);
2180 my $stmt_and_val = $sql->generate('create table', \$table, \@fields);
2182 These would return the following:
2184 # First calling form
2185 $stmt = "CREATE TABLE test (?, ?)";
2186 @bind = (field1, field2);
2188 # Second calling form
2189 $stmt_and_val = "CREATE TABLE test (field1, field2)";
2191 Depending on what you're trying to do, it's up to you to choose the correct
2192 format. In this example, the second form is what you would want.
2196 $sql->generate('alter session', { nls_date_format => 'MM/YY' });
2200 ALTER SESSION SET nls_date_format = 'MM/YY'
2202 You get the idea. Strings get their case twiddled, but everything
2203 else remains verbatim.
2205 =head1 EXPORTABLE FUNCTIONS
2207 =head2 is_plain_value
2209 Determines if the supplied argument is a plain value as understood by this
2214 =item * The value is C<undef>
2216 =item * The value is a non-reference
2218 =item * The value is an object with stringification overloading
2220 =item * The value is of the form C<< { -value => $anything } >>
2224 On failure returns C<undef>, on sucess returns a B<scalar> reference
2225 to the original supplied argument.
2231 The stringification overloading detection is rather advanced: it takes
2232 into consideration not only the presence of a C<""> overload, but if that
2233 fails also checks for enabled
2234 L<autogenerated versions of C<"">|overload/Magic Autogeneration>, based
2235 on either C<0+> or C<bool>.
2237 Unfortunately testing in the field indicates that this
2238 detection B<< may tickle a latent bug in perl versions before 5.018 >>,
2239 but only when very large numbers of stringifying objects are involved.
2240 At the time of writing ( Sep 2014 ) there is no clear explanation of
2241 the direct cause, nor is there a manageably small test case that reliably
2242 reproduces the problem.
2244 If you encounter any of the following exceptions in B<random places within
2245 your application stack> - this module may be to blame:
2247 Operation "ne": no method found,
2248 left argument in overloaded package <something>,
2249 right argument in overloaded package <something>
2253 Stub found while resolving method "???" overloading """" in package <something>
2255 If you fall victim to the above - please attempt to reduce the problem
2256 to something that could be sent to the L<SQL::Abstract developers
2257 |DBIx::Class/GETTING HELP/SUPPORT>
2258 (either publicly or privately). As a workaround in the meantime you can
2259 set C<$ENV{SQLA_ISVALUE_IGNORE_AUTOGENERATED_STRINGIFICATION}> to a true
2260 value, which will most likely eliminate your problem (at the expense of
2261 not being able to properly detect exotic forms of stringification).
2263 This notice and environment variable will be removed in a future version,
2264 as soon as the underlying problem is found and a reliable workaround is
2269 =head2 is_literal_value
2271 Determines if the supplied argument is a literal value as understood by this
2276 =item * C<\$sql_string>
2278 =item * C<\[ $sql_string, @bind_values ]>
2282 On failure returns C<undef>, on sucess returns an B<array> reference
2283 containing the unpacked version of the supplied literal SQL and bind values.
2285 =head1 WHERE CLAUSES
2289 This module uses a variation on the idea from L<DBIx::Abstract>. It
2290 is B<NOT>, repeat I<not> 100% compatible. B<The main logic of this
2291 module is that things in arrays are OR'ed, and things in hashes
2294 The easiest way to explain is to show lots of examples. After
2295 each C<%where> hash shown, it is assumed you used:
2297 my($stmt, @bind) = $sql->where(\%where);
2299 However, note that the C<%where> hash can be used directly in any
2300 of the other functions as well, as described above.
2302 =head2 Key-value pairs
2304 So, let's get started. To begin, a simple hash:
2308 status => 'completed'
2311 Is converted to SQL C<key = val> statements:
2313 $stmt = "WHERE user = ? AND status = ?";
2314 @bind = ('nwiger', 'completed');
2316 One common thing I end up doing is having a list of values that
2317 a field can be in. To do this, simply specify a list inside of
2322 status => ['assigned', 'in-progress', 'pending'];
2325 This simple code will create the following:
2327 $stmt = "WHERE user = ? AND ( status = ? OR status = ? OR status = ? )";
2328 @bind = ('nwiger', 'assigned', 'in-progress', 'pending');
2330 A field associated to an empty arrayref will be considered a
2331 logical false and will generate 0=1.
2333 =head2 Tests for NULL values
2335 If the value part is C<undef> then this is converted to SQL <IS NULL>
2344 $stmt = "WHERE user = ? AND status IS NULL";
2347 To test if a column IS NOT NULL:
2351 status => { '!=', undef },
2354 =head2 Specific comparison operators
2356 If you want to specify a different type of operator for your comparison,
2357 you can use a hashref for a given column:
2361 status => { '!=', 'completed' }
2364 Which would generate:
2366 $stmt = "WHERE user = ? AND status != ?";
2367 @bind = ('nwiger', 'completed');
2369 To test against multiple values, just enclose the values in an arrayref:
2371 status => { '=', ['assigned', 'in-progress', 'pending'] };
2373 Which would give you:
2375 "WHERE status = ? OR status = ? OR status = ?"
2378 The hashref can also contain multiple pairs, in which case it is expanded
2379 into an C<AND> of its elements:
2383 status => { '!=', 'completed', -not_like => 'pending%' }
2386 # Or more dynamically, like from a form
2387 $where{user} = 'nwiger';
2388 $where{status}{'!='} = 'completed';
2389 $where{status}{'-not_like'} = 'pending%';
2391 # Both generate this
2392 $stmt = "WHERE user = ? AND status != ? AND status NOT LIKE ?";
2393 @bind = ('nwiger', 'completed', 'pending%');
2396 To get an OR instead, you can combine it with the arrayref idea:
2400 priority => [ { '=', 2 }, { '>', 5 } ]
2403 Which would generate:
2405 $stmt = "WHERE ( priority = ? OR priority > ? ) AND user = ?";
2406 @bind = ('2', '5', 'nwiger');
2408 If you want to include literal SQL (with or without bind values), just use a
2409 scalar reference or reference to an arrayref as the value:
2412 date_entered => { '>' => \["to_date(?, 'MM/DD/YYYY')", "11/26/2008"] },
2413 date_expires => { '<' => \"now()" }
2416 Which would generate:
2418 $stmt = "WHERE date_entered > to_date(?, 'MM/DD/YYYY') AND date_expires < now()";
2419 @bind = ('11/26/2008');
2422 =head2 Logic and nesting operators
2424 In the example above,
2425 there is a subtle trap if you want to say something like
2426 this (notice the C<AND>):
2428 WHERE priority != ? AND priority != ?
2430 Because, in Perl you I<can't> do this:
2432 priority => { '!=' => 2, '!=' => 1 }
2434 As the second C<!=> key will obliterate the first. The solution
2435 is to use the special C<-modifier> form inside an arrayref:
2437 priority => [ -and => {'!=', 2},
2441 Normally, these would be joined by C<OR>, but the modifier tells it
2442 to use C<AND> instead. (Hint: You can use this in conjunction with the
2443 C<logic> option to C<new()> in order to change the way your queries
2444 work by default.) B<Important:> Note that the C<-modifier> goes
2445 B<INSIDE> the arrayref, as an extra first element. This will
2446 B<NOT> do what you think it might:
2448 priority => -and => [{'!=', 2}, {'!=', 1}] # WRONG!
2450 Here is a quick list of equivalencies, since there is some overlap:
2453 status => {'!=', 'completed', 'not like', 'pending%' }
2454 status => [ -and => {'!=', 'completed'}, {'not like', 'pending%'}]
2457 status => {'=', ['assigned', 'in-progress']}
2458 status => [ -or => {'=', 'assigned'}, {'=', 'in-progress'}]
2459 status => [ {'=', 'assigned'}, {'=', 'in-progress'} ]
2463 =head2 Special operators : IN, BETWEEN, etc.
2465 You can also use the hashref format to compare a list of fields using the
2466 C<IN> comparison operator, by specifying the list as an arrayref:
2469 status => 'completed',
2470 reportid => { -in => [567, 2335, 2] }
2473 Which would generate:
2475 $stmt = "WHERE status = ? AND reportid IN (?,?,?)";
2476 @bind = ('completed', '567', '2335', '2');
2478 The reverse operator C<-not_in> generates SQL C<NOT IN> and is used in
2481 If the argument to C<-in> is an empty array, 'sqlfalse' is generated
2482 (by default : C<1=0>). Similarly, C<< -not_in => [] >> generates
2483 'sqltrue' (by default : C<1=1>).
2485 In addition to the array you can supply a chunk of literal sql or
2486 literal sql with bind:
2489 customer => { -in => \[
2490 'SELECT cust_id FROM cust WHERE balance > ?',
2493 status => { -in => \'SELECT status_codes FROM states' },
2499 customer IN ( SELECT cust_id FROM cust WHERE balance > ? )
2500 AND status IN ( SELECT status_codes FROM states )
2504 Finally, if the argument to C<-in> is not a reference, it will be
2505 treated as a single-element array.
2507 Another pair of operators is C<-between> and C<-not_between>,
2508 used with an arrayref of two values:
2512 completion_date => {
2513 -not_between => ['2002-10-01', '2003-02-06']
2519 WHERE user = ? AND completion_date NOT BETWEEN ( ? AND ? )
2521 Just like with C<-in> all plausible combinations of literal SQL
2525 start0 => { -between => [ 1, 2 ] },
2526 start1 => { -between => \["? AND ?", 1, 2] },
2527 start2 => { -between => \"lower(x) AND upper(y)" },
2528 start3 => { -between => [
2530 \["upper(?)", 'stuff' ],
2537 ( start0 BETWEEN ? AND ? )
2538 AND ( start1 BETWEEN ? AND ? )
2539 AND ( start2 BETWEEN lower(x) AND upper(y) )
2540 AND ( start3 BETWEEN lower(x) AND upper(?) )
2542 @bind = (1, 2, 1, 2, 'stuff');
2545 These are the two builtin "special operators"; but the
2546 list can be expanded : see section L</"SPECIAL OPERATORS"> below.
2548 =head2 Unary operators: bool
2550 If you wish to test against boolean columns or functions within your
2551 database you can use the C<-bool> and C<-not_bool> operators. For
2552 example to test the column C<is_user> being true and the column
2553 C<is_enabled> being false you would use:-
2557 -not_bool => 'is_enabled',
2562 WHERE is_user AND NOT is_enabled
2564 If a more complex combination is required, testing more conditions,
2565 then you should use the and/or operators:-
2570 -not_bool => { two=> { -rlike => 'bar' } },
2571 -not_bool => { three => [ { '=', 2 }, { '>', 5 } ] },
2582 (NOT ( three = ? OR three > ? ))
2585 =head2 Nested conditions, -and/-or prefixes
2587 So far, we've seen how multiple conditions are joined with a top-level
2588 C<AND>. We can change this by putting the different conditions we want in
2589 hashes and then putting those hashes in an array. For example:
2594 status => { -like => ['pending%', 'dispatched'] },
2598 status => 'unassigned',
2602 This data structure would create the following:
2604 $stmt = "WHERE ( user = ? AND ( status LIKE ? OR status LIKE ? ) )
2605 OR ( user = ? AND status = ? ) )";
2606 @bind = ('nwiger', 'pending', 'dispatched', 'robot', 'unassigned');
2609 Clauses in hashrefs or arrayrefs can be prefixed with an C<-and> or C<-or>
2610 to change the logic inside :
2616 -and => [ workhrs => {'>', 20}, geo => 'ASIA' ],
2617 -or => { workhrs => {'<', 50}, geo => 'EURO' },
2624 $stmt = "WHERE ( user = ?
2625 AND ( ( workhrs > ? AND geo = ? )
2626 OR ( workhrs < ? OR geo = ? ) ) )";
2627 @bind = ('nwiger', '20', 'ASIA', '50', 'EURO');
2629 =head3 Algebraic inconsistency, for historical reasons
2631 C<Important note>: when connecting several conditions, the C<-and->|C<-or>
2632 operator goes C<outside> of the nested structure; whereas when connecting
2633 several constraints on one column, the C<-and> operator goes
2634 C<inside> the arrayref. Here is an example combining both features :
2637 -and => [a => 1, b => 2],
2638 -or => [c => 3, d => 4],
2639 e => [-and => {-like => 'foo%'}, {-like => '%bar'} ]
2644 WHERE ( ( ( a = ? AND b = ? )
2645 OR ( c = ? OR d = ? )
2646 OR ( e LIKE ? AND e LIKE ? ) ) )
2648 This difference in syntax is unfortunate but must be preserved for
2649 historical reasons. So be careful : the two examples below would
2650 seem algebraically equivalent, but they are not
2653 { -like => 'foo%' },
2654 { -like => '%bar' },
2656 # yields : WHERE ( ( col LIKE ? AND col LIKE ? ) )
2659 { col => { -like => 'foo%' } },
2660 { col => { -like => '%bar' } },
2662 # yields : WHERE ( ( col LIKE ? OR col LIKE ? ) )
2665 =head2 Literal SQL and value type operators
2667 The basic premise of SQL::Abstract is that in WHERE specifications the "left
2668 side" is a column name and the "right side" is a value (normally rendered as
2669 a placeholder). This holds true for both hashrefs and arrayref pairs as you
2670 see in the L</WHERE CLAUSES> examples above. Sometimes it is necessary to
2671 alter this behavior. There are several ways of doing so.
2675 This is a virtual operator that signals the string to its right side is an
2676 identifier (a column name) and not a value. For example to compare two
2677 columns you would write:
2680 priority => { '<', 2 },
2681 requestor => { -ident => 'submitter' },
2686 $stmt = "WHERE priority < ? AND requestor = submitter";
2689 If you are maintaining legacy code you may see a different construct as
2690 described in L</Deprecated usage of Literal SQL>, please use C<-ident> in new
2695 This is a virtual operator that signals that the construct to its right side
2696 is a value to be passed to DBI. This is for example necessary when you want
2697 to write a where clause against an array (for RDBMS that support such
2698 datatypes). For example:
2701 array => { -value => [1, 2, 3] }
2706 $stmt = 'WHERE array = ?';
2707 @bind = ([1, 2, 3]);
2709 Note that if you were to simply say:
2715 the result would probably not be what you wanted:
2717 $stmt = 'WHERE array = ? OR array = ? OR array = ?';
2722 Finally, sometimes only literal SQL will do. To include a random snippet
2723 of SQL verbatim, you specify it as a scalar reference. Consider this only
2724 as a last resort. Usually there is a better way. For example:
2727 priority => { '<', 2 },
2728 requestor => { -in => \'(SELECT name FROM hitmen)' },
2733 $stmt = "WHERE priority < ? AND requestor IN (SELECT name FROM hitmen)"
2736 Note that in this example, you only get one bind parameter back, since
2737 the verbatim SQL is passed as part of the statement.
2741 Never use untrusted input as a literal SQL argument - this is a massive
2742 security risk (there is no way to check literal snippets for SQL
2743 injections and other nastyness). If you need to deal with untrusted input
2744 use literal SQL with placeholders as described next.
2746 =head3 Literal SQL with placeholders and bind values (subqueries)
2748 If the literal SQL to be inserted has placeholders and bind values,
2749 use a reference to an arrayref (yes this is a double reference --
2750 not so common, but perfectly legal Perl). For example, to find a date
2751 in Postgres you can use something like this:
2754 date_column => \[ "= date '2008-09-30' - ?::integer", 10 ]
2759 $stmt = "WHERE ( date_column = date '2008-09-30' - ?::integer )"
2762 Note that you must pass the bind values in the same format as they are returned
2763 by L<where|/where(\%where, $order)>. This means that if you set L</bindtype>
2764 to C<columns>, you must provide the bind values in the
2765 C<< [ column_meta => value ] >> format, where C<column_meta> is an opaque
2766 scalar value; most commonly the column name, but you can use any scalar value
2767 (including references and blessed references), L<SQL::Abstract> will simply
2768 pass it through intact. So if C<bindtype> is set to C<columns> the above
2769 example will look like:
2772 date_column => \[ "= date '2008-09-30' - ?::integer", [ {} => 10 ] ]
2775 Literal SQL is especially useful for nesting parenthesized clauses in the
2776 main SQL query. Here is a first example :
2778 my ($sub_stmt, @sub_bind) = ("SELECT c1 FROM t1 WHERE c2 < ? AND c3 LIKE ?",
2782 bar => \["IN ($sub_stmt)" => @sub_bind],
2787 $stmt = "WHERE (foo = ? AND bar IN (SELECT c1 FROM t1
2788 WHERE c2 < ? AND c3 LIKE ?))";
2789 @bind = (1234, 100, "foo%");
2791 Other subquery operators, like for example C<"E<gt> ALL"> or C<"NOT IN">,
2792 are expressed in the same way. Of course the C<$sub_stmt> and
2793 its associated bind values can be generated through a former call
2796 my ($sub_stmt, @sub_bind)
2797 = $sql->select("t1", "c1", {c2 => {"<" => 100},
2798 c3 => {-like => "foo%"}});
2801 bar => \["> ALL ($sub_stmt)" => @sub_bind],
2804 In the examples above, the subquery was used as an operator on a column;
2805 but the same principle also applies for a clause within the main C<%where>
2806 hash, like an EXISTS subquery :
2808 my ($sub_stmt, @sub_bind)
2809 = $sql->select("t1", "*", {c1 => 1, c2 => \"> t0.c0"});
2810 my %where = ( -and => [
2812 \["EXISTS ($sub_stmt)" => @sub_bind],
2817 $stmt = "WHERE (foo = ? AND EXISTS (SELECT * FROM t1
2818 WHERE c1 = ? AND c2 > t0.c0))";
2822 Observe that the condition on C<c2> in the subquery refers to
2823 column C<t0.c0> of the main query : this is I<not> a bind
2824 value, so we have to express it through a scalar ref.
2825 Writing C<< c2 => {">" => "t0.c0"} >> would have generated
2826 C<< c2 > ? >> with bind value C<"t0.c0"> ... not exactly
2827 what we wanted here.
2829 Finally, here is an example where a subquery is used
2830 for expressing unary negation:
2832 my ($sub_stmt, @sub_bind)
2833 = $sql->where({age => [{"<" => 10}, {">" => 20}]});
2834 $sub_stmt =~ s/^ where //i; # don't want "WHERE" in the subclause
2836 lname => {like => '%son%'},
2837 \["NOT ($sub_stmt)" => @sub_bind],
2842 $stmt = "lname LIKE ? AND NOT ( age < ? OR age > ? )"
2843 @bind = ('%son%', 10, 20)
2845 =head3 Deprecated usage of Literal SQL
2847 Below are some examples of archaic use of literal SQL. It is shown only as
2848 reference for those who deal with legacy code. Each example has a much
2849 better, cleaner and safer alternative that users should opt for in new code.
2855 my %where = ( requestor => \'IS NOT NULL' )
2857 $stmt = "WHERE requestor IS NOT NULL"
2859 This used to be the way of generating NULL comparisons, before the handling
2860 of C<undef> got formalized. For new code please use the superior syntax as
2861 described in L</Tests for NULL values>.
2865 my %where = ( requestor => \'= submitter' )
2867 $stmt = "WHERE requestor = submitter"
2869 This used to be the only way to compare columns. Use the superior L</-ident>
2870 method for all new code. For example an identifier declared in such a way
2871 will be properly quoted if L</quote_char> is properly set, while the legacy
2872 form will remain as supplied.
2876 my %where = ( is_ready => \"", completed => { '>', '2012-12-21' } )
2878 $stmt = "WHERE completed > ? AND is_ready"
2879 @bind = ('2012-12-21')
2881 Using an empty string literal used to be the only way to express a boolean.
2882 For all new code please use the much more readable
2883 L<-bool|/Unary operators: bool> operator.
2889 These pages could go on for a while, since the nesting of the data
2890 structures this module can handle are pretty much unlimited (the
2891 module implements the C<WHERE> expansion as a recursive function
2892 internally). Your best bet is to "play around" with the module a
2893 little to see how the data structures behave, and choose the best
2894 format for your data based on that.
2896 And of course, all the values above will probably be replaced with
2897 variables gotten from forms or the command line. After all, if you
2898 knew everything ahead of time, you wouldn't have to worry about
2899 dynamically-generating SQL and could just hardwire it into your
2902 =head1 ORDER BY CLAUSES
2904 Some functions take an order by clause. This can either be a scalar (just a
2905 column name,) a hash of C<< { -desc => 'col' } >> or C<< { -asc => 'col' } >>,
2906 or an array of either of the two previous forms. Examples:
2908 Given | Will Generate
2909 ----------------------------------------------------------
2911 \'colA DESC' | ORDER BY colA DESC
2913 'colA' | ORDER BY colA
2915 [qw/colA colB/] | ORDER BY colA, colB
2917 {-asc => 'colA'} | ORDER BY colA ASC
2919 {-desc => 'colB'} | ORDER BY colB DESC
2921 ['colA', {-asc => 'colB'}] | ORDER BY colA, colB ASC
2923 { -asc => [qw/colA colB/] } | ORDER BY colA ASC, colB ASC
2926 { -asc => 'colA' }, | ORDER BY colA ASC, colB DESC,
2927 { -desc => [qw/colB/], | colC ASC, colD ASC
2928 { -asc => [qw/colC colD/],|
2930 ===========================================================
2934 =head1 SPECIAL OPERATORS
2936 my $sqlmaker = SQL::Abstract->new(special_ops => [
2940 my ($self, $field, $op, $arg) = @_;
2946 handler => 'method_name',
2950 A "special operator" is a SQL syntactic clause that can be
2951 applied to a field, instead of a usual binary operator.
2954 WHERE field IN (?, ?, ?)
2955 WHERE field BETWEEN ? AND ?
2956 WHERE MATCH(field) AGAINST (?, ?)
2958 Special operators IN and BETWEEN are fairly standard and therefore
2959 are builtin within C<SQL::Abstract> (as the overridable methods
2960 C<_where_field_IN> and C<_where_field_BETWEEN>). For other operators,
2961 like the MATCH .. AGAINST example above which is specific to MySQL,
2962 you can write your own operator handlers - supply a C<special_ops>
2963 argument to the C<new> method. That argument takes an arrayref of
2964 operator definitions; each operator definition is a hashref with two
2971 the regular expression to match the operator
2975 Either a coderef or a plain scalar method name. In both cases
2976 the expected return is C<< ($sql, @bind) >>.
2978 When supplied with a method name, it is simply called on the
2979 L<SQL::Abstract> object as:
2981 $self->$method_name ($field, $op, $arg)
2985 $field is the LHS of the operator
2986 $op is the part that matched the handler regex
2989 When supplied with a coderef, it is called as:
2991 $coderef->($self, $field, $op, $arg)
2996 For example, here is an implementation
2997 of the MATCH .. AGAINST syntax for MySQL
2999 my $sqlmaker = SQL::Abstract->new(special_ops => [
3001 # special op for MySql MATCH (field) AGAINST(word1, word2, ...)
3002 {regex => qr/^match$/i,
3004 my ($self, $field, $op, $arg) = @_;
3005 $arg = [$arg] if not ref $arg;
3006 my $label = $self->_quote($field);
3007 my ($placeholder) = $self->_convert('?');
3008 my $placeholders = join ", ", (($placeholder) x @$arg);
3009 my $sql = $self->_sqlcase('match') . " ($label) "
3010 . $self->_sqlcase('against') . " ($placeholders) ";
3011 my @bind = $self->_bindtype($field, @$arg);
3012 return ($sql, @bind);
3019 =head1 UNARY OPERATORS
3021 my $sqlmaker = SQL::Abstract->new(unary_ops => [
3025 my ($self, $op, $arg) = @_;
3031 handler => 'method_name',
3035 A "unary operator" is a SQL syntactic clause that can be
3036 applied to a field - the operator goes before the field
3038 You can write your own operator handlers - supply a C<unary_ops>
3039 argument to the C<new> method. That argument takes an arrayref of
3040 operator definitions; each operator definition is a hashref with two
3047 the regular expression to match the operator
3051 Either a coderef or a plain scalar method name. In both cases
3052 the expected return is C<< $sql >>.
3054 When supplied with a method name, it is simply called on the
3055 L<SQL::Abstract> object as:
3057 $self->$method_name ($op, $arg)
3061 $op is the part that matched the handler regex
3062 $arg is the RHS or argument of the operator
3064 When supplied with a coderef, it is called as:
3066 $coderef->($self, $op, $arg)
3074 Thanks to some benchmarking by Mark Stosberg, it turns out that
3075 this module is many orders of magnitude faster than using C<DBIx::Abstract>.
3076 I must admit this wasn't an intentional design issue, but it's a
3077 byproduct of the fact that you get to control your C<DBI> handles
3080 To maximize performance, use a code snippet like the following:
3082 # prepare a statement handle using the first row
3083 # and then reuse it for the rest of the rows
3085 for my $href (@array_of_hashrefs) {
3086 $stmt ||= $sql->insert('table', $href);
3087 $sth ||= $dbh->prepare($stmt);
3088 $sth->execute($sql->values($href));
3091 The reason this works is because the keys in your C<$href> are sorted
3092 internally by B<SQL::Abstract>. Thus, as long as your data retains
3093 the same structure, you only have to generate the SQL the first time
3094 around. On subsequent queries, simply use the C<values> function provided
3095 by this module to return your values in the correct order.
3097 However this depends on the values having the same type - if, for
3098 example, the values of a where clause may either have values
3099 (resulting in sql of the form C<column = ?> with a single bind
3100 value), or alternatively the values might be C<undef> (resulting in
3101 sql of the form C<column IS NULL> with no bind value) then the
3102 caching technique suggested will not work.
3106 If you use my C<CGI::FormBuilder> module at all, you'll hopefully
3107 really like this part (I do, at least). Building up a complex query
3108 can be as simple as the following:
3115 use CGI::FormBuilder;
3118 my $form = CGI::FormBuilder->new(...);
3119 my $sql = SQL::Abstract->new;
3121 if ($form->submitted) {
3122 my $field = $form->field;
3123 my $id = delete $field->{id};
3124 my($stmt, @bind) = $sql->update('table', $field, {id => $id});
3127 Of course, you would still have to connect using C<DBI> to run the
3128 query, but the point is that if you make your form look like your
3129 table, the actual query script can be extremely simplistic.
3131 If you're B<REALLY> lazy (I am), check out C<HTML::QuickTable> for
3132 a fast interface to returning and formatting data. I frequently
3133 use these three modules together to write complex database query
3134 apps in under 50 lines.
3136 =head1 HOW TO CONTRIBUTE
3138 Contributions are always welcome, in all usable forms (we especially
3139 welcome documentation improvements). The delivery methods include git-
3140 or unified-diff formatted patches, GitHub pull requests, or plain bug
3141 reports either via RT or the Mailing list. Contributors are generally
3142 granted full access to the official repository after their first several
3143 patches pass successful review.
3145 This project is maintained in a git repository. The code and related tools are
3146 accessible at the following locations:
3150 =item * Official repo: L<git://git.shadowcat.co.uk/dbsrgits/SQL-Abstract.git>
3152 =item * Official gitweb: L<http://git.shadowcat.co.uk/gitweb/gitweb.cgi?p=dbsrgits/SQL-Abstract.git>
3154 =item * GitHub mirror: L<https://github.com/dbsrgits/sql-abstract>
3156 =item * Authorized committers: L<ssh://dbsrgits@git.shadowcat.co.uk/SQL-Abstract.git>
3162 Version 1.50 was a major internal refactoring of C<SQL::Abstract>.
3163 Great care has been taken to preserve the I<published> behavior
3164 documented in previous versions in the 1.* family; however,
3165 some features that were previously undocumented, or behaved
3166 differently from the documentation, had to be changed in order
3167 to clarify the semantics. Hence, client code that was relying
3168 on some dark areas of C<SQL::Abstract> v1.*
3169 B<might behave differently> in v1.50.
3171 The main changes are :
3177 support for literal SQL through the C<< \ [ $sql, @bind ] >> syntax.
3181 support for the { operator => \"..." } construct (to embed literal SQL)
3185 support for the { operator => \["...", @bind] } construct (to embed literal SQL with bind values)
3189 optional support for L<array datatypes|/"Inserting and Updating Arrays">
3193 defensive programming : check arguments
3197 fixed bug with global logic, which was previously implemented
3198 through global variables yielding side-effects. Prior versions would
3199 interpret C<< [ {cond1, cond2}, [cond3, cond4] ] >>
3200 as C<< "(cond1 AND cond2) OR (cond3 AND cond4)" >>.
3201 Now this is interpreted
3202 as C<< "(cond1 AND cond2) OR (cond3 OR cond4)" >>.
3207 fixed semantics of _bindtype on array args
3211 dropped the C<_anoncopy> of the %where tree. No longer necessary,
3212 we just avoid shifting arrays within that tree.
3216 dropped the C<_modlogic> function
3220 =head1 ACKNOWLEDGEMENTS
3222 There are a number of individuals that have really helped out with
3223 this module. Unfortunately, most of them submitted bugs via CPAN
3224 so I have no idea who they are! But the people I do know are:
3226 Ash Berlin (order_by hash term support)
3227 Matt Trout (DBIx::Class support)
3228 Mark Stosberg (benchmarking)
3229 Chas Owens (initial "IN" operator support)
3230 Philip Collins (per-field SQL functions)
3231 Eric Kolve (hashref "AND" support)
3232 Mike Fragassi (enhancements to "BETWEEN" and "LIKE")
3233 Dan Kubb (support for "quote_char" and "name_sep")
3234 Guillermo Roditi (patch to cleanup "IN" and "BETWEEN", fix and tests for _order_by)
3235 Laurent Dami (internal refactoring, extensible list of special operators, literal SQL)
3236 Norbert Buchmuller (support for literal SQL in hashpair, misc. fixes & tests)
3237 Peter Rabbitson (rewrite of SQLA::Test, misc. fixes & tests)
3238 Oliver Charles (support for "RETURNING" after "INSERT")
3244 L<DBIx::Class>, L<DBIx::Abstract>, L<CGI::FormBuilder>, L<HTML::QuickTable>.
3248 Copyright (c) 2001-2007 Nathan Wiger <nwiger@cpan.org>. All Rights Reserved.
3250 This module is actively maintained by Matt Trout <mst@shadowcatsystems.co.uk>
3252 For support, your best bet is to try the C<DBIx::Class> users mailing list.
3253 While not an official support venue, C<DBIx::Class> makes heavy use of
3254 C<SQL::Abstract>, and as such list members there are very familiar with
3255 how to create queries.
3259 This module is free software; you may copy this under the same
3260 terms as perl itself (either the GNU General Public License or
3261 the Artistic License)