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 = '1.78';
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] } ]
82 ref $_[0] eq 'HASH' and keys %{$_[0]} == 1
84 defined $_[0]->{-ident} and ! length ref $_[0]->{-ident}
85 ) ? [ $_[0]->{-ident} ]
89 # FIXME XSify - this can be done so much more efficiently
90 sub is_plain_value ($) {
92 ! length ref $_[0] ? \($_[0])
94 ref $_[0] eq 'HASH' and keys %{$_[0]} == 1
96 exists $_[0]->{-value}
97 ) ? \($_[0]->{-value})
99 # reuse @_ for even moar speedz
100 defined ( $_[1] = Scalar::Util::blessed $_[0] )
102 # deliberately not using Devel::OverloadInfo - the checks we are
103 # intersted in are much more limited than the fullblown thing, and
104 # this is a very hot piece of code
106 # simply using ->can('(""') can leave behind stub methods that
107 # break actually using the overload later (see L<perldiag/Stub
108 # found while resolving method "%s" overloading "%s" in package
109 # "%s"> and the source of overload::mycan())
111 # either has stringification which DBI SHOULD prefer out of the box
112 grep { *{ (qq[${_}::(""]) }{CODE} } @{ $_[2] = mro::get_linear_isa( $_[1] ) }
114 # has nummification or boolification, AND fallback is *not* disabled
116 SQL::Abstract::_ENV_::DETECT_AUTOGENERATED_STRINGIFICATION
119 grep { *{"${_}::(0+"}{CODE} } @{$_[2]}
121 grep { *{"${_}::(bool"}{CODE} } @{$_[2]}
125 # no fallback specified at all
126 ! ( ($_[3]) = grep { *{"${_}::()"}{CODE} } @{$_[2]} )
128 # fallback explicitly undef
129 ! defined ${"$_[3]::()"}
142 #======================================================================
144 #======================================================================
148 my $class = ref($self) || $self;
149 my %opt = (ref $_[0] eq 'HASH') ? %{$_[0]} : @_;
151 # choose our case by keeping an option around
152 delete $opt{case} if $opt{case} && $opt{case} ne 'lower';
154 # default logic for interpreting arrayrefs
155 $opt{logic} = $opt{logic} ? uc $opt{logic} : 'OR';
157 # how to return bind vars
158 $opt{bindtype} ||= 'normal';
160 # default comparison is "=", but can be overridden
163 # try to recognize which are the 'equality' and 'inequality' ops
164 # (temporary quickfix (in 2007), should go through a more seasoned API)
165 $opt{equality_op} = qr/^( \Q$opt{cmp}\E | \= )$/ix;
166 $opt{inequality_op} = qr/^( != | <> )$/ix;
168 $opt{like_op} = qr/^ (is\s+)? r?like $/xi;
169 $opt{not_like_op} = qr/^ (is\s+)? not \s+ r?like $/xi;
172 $opt{sqltrue} ||= '1=1';
173 $opt{sqlfalse} ||= '0=1';
176 $opt{special_ops} ||= [];
177 # regexes are applied in order, thus push after user-defines
178 push @{$opt{special_ops}}, @BUILTIN_SPECIAL_OPS;
181 $opt{unary_ops} ||= [];
182 push @{$opt{unary_ops}}, @BUILTIN_UNARY_OPS;
184 # rudimentary sanity-check for user supplied bits treated as functions/operators
185 # If a purported function matches this regular expression, an exception is thrown.
186 # Literal SQL is *NOT* subject to this check, only functions (and column names
187 # when quoting is not in effect)
190 # need to guard against ()'s in column names too, but this will break tons of
191 # hacks... ideas anyone?
192 $opt{injection_guard} ||= qr/
198 return bless \%opt, $class;
202 sub _assert_pass_injection_guard {
203 if ($_[1] =~ $_[0]->{injection_guard}) {
204 my $class = ref $_[0];
205 puke "Possible SQL injection attempt '$_[1]'. If this is indeed a part of the "
206 . "desired SQL use literal SQL ( \'...' or \[ '...' ] ) or supply your own "
207 . "{injection_guard} attribute to ${class}->new()"
212 #======================================================================
214 #======================================================================
218 my $table = $self->_table(shift);
219 my $data = shift || return;
222 my $method = $self->_METHOD_FOR_refkind("_insert", $data);
223 my ($sql, @bind) = $self->$method($data);
224 $sql = join " ", $self->_sqlcase('insert into'), $table, $sql;
226 if ($options->{returning}) {
227 my ($s, @b) = $self->_insert_returning ($options);
232 return wantarray ? ($sql, @bind) : $sql;
235 sub _insert_returning {
236 my ($self, $options) = @_;
238 my $f = $options->{returning};
240 my $fieldlist = $self->_SWITCH_refkind($f, {
241 ARRAYREF => sub {join ', ', map { $self->_quote($_) } @$f;},
242 SCALAR => sub {$self->_quote($f)},
243 SCALARREF => sub {$$f},
245 return $self->_sqlcase(' returning ') . $fieldlist;
248 sub _insert_HASHREF { # explicit list of fields and then values
249 my ($self, $data) = @_;
251 my @fields = sort keys %$data;
253 my ($sql, @bind) = $self->_insert_values($data);
256 $_ = $self->_quote($_) foreach @fields;
257 $sql = "( ".join(", ", @fields).") ".$sql;
259 return ($sql, @bind);
262 sub _insert_ARRAYREF { # just generate values(?,?) part (no list of fields)
263 my ($self, $data) = @_;
265 # no names (arrayref) so can't generate bindtype
266 $self->{bindtype} ne 'columns'
267 or belch "can't do 'columns' bindtype when called with arrayref";
269 # fold the list of values into a hash of column name - value pairs
270 # (where the column names are artificially generated, and their
271 # lexicographical ordering keep the ordering of the original list)
272 my $i = "a"; # incremented values will be in lexicographical order
273 my $data_in_hash = { map { ($i++ => $_) } @$data };
275 return $self->_insert_values($data_in_hash);
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 $v = $data->{$column};
301 $self->_SWITCH_refkind($v, {
304 if ($self->{array_datatypes}) { # if array datatype are activated
306 push @all_bind, $self->_bindtype($column, $v);
308 else { # else literal SQL with bind
309 my ($sql, @bind) = @$v;
310 $self->_assert_bindval_matches_bindtype(@bind);
312 push @all_bind, @bind;
316 ARRAYREFREF => sub { # literal SQL with bind
317 my ($sql, @bind) = @${$v};
318 $self->_assert_bindval_matches_bindtype(@bind);
320 push @all_bind, @bind;
323 # THINK : anything useful to do with a HASHREF ?
324 HASHREF => sub { # (nothing, but old SQLA passed it through)
325 #TODO in SQLA >= 2.0 it will die instead
326 belch "HASH ref as bind value in insert is not supported";
328 push @all_bind, $self->_bindtype($column, $v);
331 SCALARREF => sub { # literal SQL without bind
335 SCALAR_or_UNDEF => sub {
337 push @all_bind, $self->_bindtype($column, $v);
344 my $sql = $self->_sqlcase('values')." ( ".join(", ", @values)." )";
345 return ($sql, @all_bind);
350 #======================================================================
352 #======================================================================
357 my $table = $self->_table(shift);
358 my $data = shift || return;
361 # first build the 'SET' part of the sql statement
362 my (@set, @all_bind);
363 puke "Unsupported data type specified to \$sql->update"
364 unless ref $data eq 'HASH';
366 for my $k (sort keys %$data) {
369 my $label = $self->_quote($k);
371 $self->_SWITCH_refkind($v, {
373 if ($self->{array_datatypes}) { # array datatype
374 push @set, "$label = ?";
375 push @all_bind, $self->_bindtype($k, $v);
377 else { # literal SQL with bind
378 my ($sql, @bind) = @$v;
379 $self->_assert_bindval_matches_bindtype(@bind);
380 push @set, "$label = $sql";
381 push @all_bind, @bind;
384 ARRAYREFREF => sub { # literal SQL with bind
385 my ($sql, @bind) = @${$v};
386 $self->_assert_bindval_matches_bindtype(@bind);
387 push @set, "$label = $sql";
388 push @all_bind, @bind;
390 SCALARREF => sub { # literal SQL without bind
391 push @set, "$label = $$v";
394 my ($op, $arg, @rest) = %$v;
396 puke 'Operator calls in update must be in the form { -op => $arg }'
397 if (@rest or not $op =~ /^\-(.+)/);
399 local $self->{_nested_func_lhs} = $k;
400 my ($sql, @bind) = $self->_where_unary_op ($1, $arg);
402 push @set, "$label = $sql";
403 push @all_bind, @bind;
405 SCALAR_or_UNDEF => sub {
406 push @set, "$label = ?";
407 push @all_bind, $self->_bindtype($k, $v);
413 my $sql = $self->_sqlcase('update') . " $table " . $self->_sqlcase('set ')
417 my($where_sql, @where_bind) = $self->where($where);
419 push @all_bind, @where_bind;
422 return wantarray ? ($sql, @all_bind) : $sql;
428 #======================================================================
430 #======================================================================
435 my $table = $self->_table(shift);
436 my $fields = shift || '*';
440 my($where_sql, @bind) = $self->where($where, $order);
442 my $f = (ref $fields eq 'ARRAY') ? join ', ', map { $self->_quote($_) } @$fields
444 my $sql = join(' ', $self->_sqlcase('select'), $f,
445 $self->_sqlcase('from'), $table)
448 return wantarray ? ($sql, @bind) : $sql;
451 #======================================================================
453 #======================================================================
458 my $table = $self->_table(shift);
462 my($where_sql, @bind) = $self->where($where);
463 my $sql = $self->_sqlcase('delete from') . " $table" . $where_sql;
465 return wantarray ? ($sql, @bind) : $sql;
469 #======================================================================
471 #======================================================================
475 # Finally, a separate routine just to handle WHERE clauses
477 my ($self, $where, $order) = @_;
480 my ($sql, @bind) = $self->_recurse_where($where);
481 $sql = $sql ? $self->_sqlcase(' where ') . "( $sql )" : '';
485 $sql .= $self->_order_by($order);
488 return wantarray ? ($sql, @bind) : $sql;
493 my ($self, $where, $logic) = @_;
495 # dispatch on appropriate method according to refkind of $where
496 my $method = $self->_METHOD_FOR_refkind("_where", $where);
498 my ($sql, @bind) = $self->$method($where, $logic);
500 # DBIx::Class directly calls _recurse_where in scalar context, so
501 # we must implement it, even if not in the official API
502 return wantarray ? ($sql, @bind) : $sql;
507 #======================================================================
508 # WHERE: top-level ARRAYREF
509 #======================================================================
512 sub _where_ARRAYREF {
513 my ($self, $where, $logic) = @_;
515 $logic = uc($logic || $self->{logic});
516 $logic eq 'AND' or $logic eq 'OR' or puke "unknown logic: $logic";
518 my @clauses = @$where;
520 my (@sql_clauses, @all_bind);
521 # need to use while() so can shift() for pairs
522 while (my $el = shift @clauses) {
524 # switch according to kind of $el and get corresponding ($sql, @bind)
525 my ($sql, @bind) = $self->_SWITCH_refkind($el, {
527 # skip empty elements, otherwise get invalid trailing AND stuff
528 ARRAYREF => sub {$self->_recurse_where($el) if @$el},
532 $self->_assert_bindval_matches_bindtype(@b);
536 HASHREF => sub {$self->_recurse_where($el, 'and') if %$el},
538 SCALARREF => sub { ($$el); },
540 SCALAR => sub {# top-level arrayref with scalars, recurse in pairs
541 $self->_recurse_where({$el => shift(@clauses)})},
543 UNDEF => sub {puke "not supported : UNDEF in arrayref" },
547 push @sql_clauses, $sql;
548 push @all_bind, @bind;
552 return $self->_join_sql_clauses($logic, \@sql_clauses, \@all_bind);
555 #======================================================================
556 # WHERE: top-level ARRAYREFREF
557 #======================================================================
559 sub _where_ARRAYREFREF {
560 my ($self, $where) = @_;
561 my ($sql, @bind) = @$$where;
562 $self->_assert_bindval_matches_bindtype(@bind);
563 return ($sql, @bind);
566 #======================================================================
567 # WHERE: top-level HASHREF
568 #======================================================================
571 my ($self, $where) = @_;
572 my (@sql_clauses, @all_bind);
574 for my $k (sort keys %$where) {
575 my $v = $where->{$k};
577 # ($k => $v) is either a special unary op or a regular hashpair
578 my ($sql, @bind) = do {
580 # put the operator in canonical form
582 $op = substr $op, 1; # remove initial dash
583 $op =~ s/^\s+|\s+$//g;# remove leading/trailing space
584 $op =~ s/\s+/ /g; # compress whitespace
586 # so that -not_foo works correctly
587 $op =~ s/^not_/NOT /i;
589 $self->_debug("Unary OP(-$op) within hashref, recursing...");
590 my ($s, @b) = $self->_where_unary_op ($op, $v);
592 # top level vs nested
593 # we assume that handled unary ops will take care of their ()s
595 List::Util::first {$op =~ $_->{regex}} @{$self->{unary_ops}}
597 defined($self->{_nested_func_lhs}) && ($self->{_nested_func_lhs} eq $k)
602 my $method = $self->_METHOD_FOR_refkind("_where_hashpair", $v);
603 $self->$method($k, $v);
607 push @sql_clauses, $sql;
608 push @all_bind, @bind;
611 return $self->_join_sql_clauses('and', \@sql_clauses, \@all_bind);
614 sub _where_unary_op {
615 my ($self, $op, $rhs) = @_;
617 if (my $op_entry = List::Util::first {$op =~ $_->{regex}} @{$self->{unary_ops}}) {
618 my $handler = $op_entry->{handler};
620 if (not ref $handler) {
621 if ($op =~ s/ [_\s]? \d+ $//x ) {
622 belch 'Use of [and|or|nest]_N modifiers is deprecated and will be removed in SQLA v2.0. '
623 . "You probably wanted ...-and => [ -$op => COND1, -$op => COND2 ... ]";
625 return $self->$handler ($op, $rhs);
627 elsif (ref $handler eq 'CODE') {
628 return $handler->($self, $op, $rhs);
631 puke "Illegal handler for operator $op - expecting a method name or a coderef";
635 $self->_debug("Generic unary OP: $op - recursing as function");
637 $self->_assert_pass_injection_guard($op);
639 my ($sql, @bind) = $self->_SWITCH_refkind ($rhs, {
641 puke "Illegal use of top-level '$op'"
642 unless $self->{_nested_func_lhs};
645 $self->_convert('?'),
646 $self->_bindtype($self->{_nested_func_lhs}, $rhs)
650 $self->_recurse_where ($rhs)
654 $sql = sprintf ('%s %s',
655 $self->_sqlcase($op),
659 return ($sql, @bind);
662 sub _where_op_ANDOR {
663 my ($self, $op, $v) = @_;
665 $self->_SWITCH_refkind($v, {
667 return $self->_where_ARRAYREF($v, $op);
671 return ( $op =~ /^or/i )
672 ? $self->_where_ARRAYREF( [ map { $_ => $v->{$_} } ( sort keys %$v ) ], $op )
673 : $self->_where_HASHREF($v);
677 puke "-$op => \\\$scalar makes little sense, use " .
679 ? '[ \$scalar, \%rest_of_conditions ] instead'
680 : '-and => [ \$scalar, \%rest_of_conditions ] instead'
685 puke "-$op => \\[...] makes little sense, use " .
687 ? '[ \[...], \%rest_of_conditions ] instead'
688 : '-and => [ \[...], \%rest_of_conditions ] instead'
692 SCALAR => sub { # permissively interpreted as SQL
693 puke "-$op => \$value makes little sense, use -bool => \$value instead";
697 puke "-$op => undef not supported";
703 my ($self, $op, $v) = @_;
705 $self->_SWITCH_refkind($v, {
707 SCALAR => sub { # permissively interpreted as SQL
708 belch "literal SQL should be -nest => \\'scalar' "
709 . "instead of -nest => 'scalar' ";
714 puke "-$op => undef not supported";
718 $self->_recurse_where ($v);
726 my ($self, $op, $v) = @_;
728 my ($s, @b) = $self->_SWITCH_refkind($v, {
729 SCALAR => sub { # interpreted as SQL column
730 $self->_convert($self->_quote($v));
734 puke "-$op => undef not supported";
738 $self->_recurse_where ($v);
742 $s = "(NOT $s)" if $op =~ /^not/i;
747 sub _where_op_IDENT {
749 my ($op, $rhs) = splice @_, -2;
750 if (! defined $rhs or length ref $rhs) {
751 puke "-$op requires a single plain scalar argument (a quotable identifier)";
754 # in case we are called as a top level special op (no '=')
757 $_ = $self->_convert($self->_quote($_)) for ($lhs, $rhs);
765 sub _where_op_VALUE {
767 my ($op, $rhs) = splice @_, -2;
769 # in case we are called as a top level special op (no '=')
773 if (! defined $rhs) {
775 ? $self->_convert($self->_quote($lhs)) . ' IS NULL'
782 ($lhs || $self->{_nested_func_lhs}),
789 $self->_convert($self->_quote($lhs)) . ' = ' . $self->_convert('?'),
793 $self->_convert('?'),
799 sub _where_hashpair_ARRAYREF {
800 my ($self, $k, $v) = @_;
803 my @v = @$v; # need copy because of shift below
804 $self->_debug("ARRAY($k) means distribute over elements");
806 # put apart first element if it is an operator (-and, -or)
808 (defined $v[0] && $v[0] =~ /^ - (?: AND|OR ) $/ix)
812 my @distributed = map { {$k => $_} } @v;
815 $self->_debug("OP($op) reinjected into the distributed array");
816 unshift @distributed, $op;
819 my $logic = $op ? substr($op, 1) : '';
821 return $self->_recurse_where(\@distributed, $logic);
824 $self->_debug("empty ARRAY($k) means 0=1");
825 return ($self->{sqlfalse});
829 sub _where_hashpair_HASHREF {
830 my ($self, $k, $v, $logic) = @_;
833 local $self->{_nested_func_lhs} = $self->{_nested_func_lhs};
835 my ($all_sql, @all_bind);
837 for my $orig_op (sort keys %$v) {
838 my $val = $v->{$orig_op};
840 # put the operator in canonical form
843 # FIXME - we need to phase out dash-less ops
844 $op =~ s/^-//; # remove possible initial dash
845 $op =~ s/^\s+|\s+$//g;# remove leading/trailing space
846 $op =~ s/\s+/ /g; # compress whitespace
848 $self->_assert_pass_injection_guard($op);
851 $op =~ s/^is_not/IS NOT/i;
853 # so that -not_foo works correctly
854 $op =~ s/^not_/NOT /i;
856 # another retarded special case: foo => { $op => { -value => undef } }
857 if (ref $val eq 'HASH' and keys %$val == 1 and exists $val->{-value} and ! defined $val->{-value} ) {
863 # CASE: col-value logic modifiers
864 if ( $orig_op =~ /^ \- (and|or) $/xi ) {
865 ($sql, @bind) = $self->_where_hashpair_HASHREF($k, $val, $1);
867 # CASE: special operators like -in or -between
868 elsif ( my $special_op = List::Util::first {$op =~ $_->{regex}} @{$self->{special_ops}} ) {
869 my $handler = $special_op->{handler};
871 puke "No handler supplied for special operator $orig_op";
873 elsif (not ref $handler) {
874 ($sql, @bind) = $self->$handler ($k, $op, $val);
876 elsif (ref $handler eq 'CODE') {
877 ($sql, @bind) = $handler->($self, $k, $op, $val);
880 puke "Illegal handler for special operator $orig_op - expecting a method name or a coderef";
884 $self->_SWITCH_refkind($val, {
886 ARRAYREF => sub { # CASE: col => {op => \@vals}
887 ($sql, @bind) = $self->_where_field_op_ARRAYREF($k, $op, $val);
890 ARRAYREFREF => sub { # CASE: col => {op => \[$sql, @bind]} (literal SQL with bind)
891 my ($sub_sql, @sub_bind) = @$$val;
892 $self->_assert_bindval_matches_bindtype(@sub_bind);
893 $sql = join ' ', $self->_convert($self->_quote($k)),
894 $self->_sqlcase($op),
899 UNDEF => sub { # CASE: col => {op => undef} : sql "IS (NOT)? NULL"
901 $op =~ /^not$/i ? 'is not' # legacy
902 : $op =~ $self->{equality_op} ? 'is'
903 : $op =~ $self->{like_op} ? belch("Supplying an undefined argument to '@{[ uc $op]}' is deprecated") && 'is'
904 : $op =~ $self->{inequality_op} ? 'is not'
905 : $op =~ $self->{not_like_op} ? belch("Supplying an undefined argument to '@{[ uc $op]}' is deprecated") && 'is not'
906 : puke "unexpected operator '$orig_op' with undef operand";
908 $sql = $self->_quote($k) . $self->_sqlcase(" $is null");
911 FALLBACK => sub { # CASE: col => {op/func => $stuff}
913 # retain for proper column type bind
914 $self->{_nested_func_lhs} ||= $k;
916 ($sql, @bind) = $self->_where_unary_op ($op, $val);
919 $self->_convert($self->_quote($k)),
920 $self->{_nested_func_lhs} eq $k ? $sql : "($sql)", # top level vs nested
926 ($all_sql) = (defined $all_sql and $all_sql) ? $self->_join_sql_clauses($logic, [$all_sql, $sql], []) : $sql;
927 push @all_bind, @bind;
929 return ($all_sql, @all_bind);
932 sub _where_field_IS {
933 my ($self, $k, $op, $v) = @_;
935 my ($s) = $self->_SWITCH_refkind($v, {
938 $self->_convert($self->_quote($k)),
939 map { $self->_sqlcase($_)} ($op, 'null')
942 puke "$op can only take undef as argument";
949 sub _where_field_op_ARRAYREF {
950 my ($self, $k, $op, $vals) = @_;
952 my @vals = @$vals; #always work on a copy
955 $self->_debug(sprintf '%s means multiple elements: [ %s ]',
957 join (', ', map { defined $_ ? "'$_'" : 'NULL' } @vals ),
960 # see if the first element is an -and/-or op
962 if (defined $vals[0] && $vals[0] =~ /^ - ( AND|OR ) $/ix) {
967 # a long standing API wart - an attempt to change this behavior during
968 # the 1.50 series failed *spectacularly*. Warn instead and leave the
973 (!$logic or $logic eq 'OR')
975 ( $op =~ $self->{inequality_op} or $op =~ $self->{not_like_op} )
978 belch "A multi-element arrayref as an argument to the inequality op '$o' "
979 . 'is technically equivalent to an always-true 1=1 (you probably wanted '
980 . "to say ...{ \$inequality_op => [ -and => \@values ] }... instead)"
984 # distribute $op over each remaining member of @vals, append logic if exists
985 return $self->_recurse_where([map { {$k => {$op, $_}} } @vals], $logic);
989 # try to DWIM on equality operators
991 $op =~ $self->{equality_op} ? $self->{sqlfalse}
992 : $op =~ $self->{like_op} ? belch("Supplying an empty arrayref to '@{[ uc $op]}' is deprecated") && $self->{sqlfalse}
993 : $op =~ $self->{inequality_op} ? $self->{sqltrue}
994 : $op =~ $self->{not_like_op} ? belch("Supplying an empty arrayref to '@{[ uc $op]}' is deprecated") && $self->{sqltrue}
995 : puke "operator '$op' applied on an empty array (field '$k')";
1000 sub _where_hashpair_SCALARREF {
1001 my ($self, $k, $v) = @_;
1002 $self->_debug("SCALAR($k) means literal SQL: $$v");
1003 my $sql = $self->_quote($k) . " " . $$v;
1007 # literal SQL with bind
1008 sub _where_hashpair_ARRAYREFREF {
1009 my ($self, $k, $v) = @_;
1010 $self->_debug("REF($k) means literal SQL: @${$v}");
1011 my ($sql, @bind) = @$$v;
1012 $self->_assert_bindval_matches_bindtype(@bind);
1013 $sql = $self->_quote($k) . " " . $sql;
1014 return ($sql, @bind );
1017 # literal SQL without bind
1018 sub _where_hashpair_SCALAR {
1019 my ($self, $k, $v) = @_;
1020 $self->_debug("NOREF($k) means simple key=val: $k $self->{cmp} $v");
1021 my $sql = join ' ', $self->_convert($self->_quote($k)),
1022 $self->_sqlcase($self->{cmp}),
1023 $self->_convert('?');
1024 my @bind = $self->_bindtype($k, $v);
1025 return ( $sql, @bind);
1029 sub _where_hashpair_UNDEF {
1030 my ($self, $k, $v) = @_;
1031 $self->_debug("UNDEF($k) means IS NULL");
1032 my $sql = $self->_quote($k) . $self->_sqlcase(' is null');
1036 #======================================================================
1037 # WHERE: TOP-LEVEL OTHERS (SCALARREF, SCALAR, UNDEF)
1038 #======================================================================
1041 sub _where_SCALARREF {
1042 my ($self, $where) = @_;
1045 $self->_debug("SCALAR(*top) means literal SQL: $$where");
1051 my ($self, $where) = @_;
1054 $self->_debug("NOREF(*top) means literal SQL: $where");
1065 #======================================================================
1066 # WHERE: BUILTIN SPECIAL OPERATORS (-in, -between)
1067 #======================================================================
1070 sub _where_field_BETWEEN {
1071 my ($self, $k, $op, $vals) = @_;
1073 my ($label, $and, $placeholder);
1074 $label = $self->_convert($self->_quote($k));
1075 $and = ' ' . $self->_sqlcase('and') . ' ';
1076 $placeholder = $self->_convert('?');
1077 $op = $self->_sqlcase($op);
1079 my $invalid_args = "Operator '$op' requires either an arrayref with two defined values or expressions, or a single literal scalarref/arrayref-ref";
1081 my ($clause, @bind) = $self->_SWITCH_refkind($vals, {
1082 ARRAYREFREF => sub {
1083 my ($s, @b) = @$$vals;
1084 $self->_assert_bindval_matches_bindtype(@b);
1091 puke $invalid_args if @$vals != 2;
1093 my (@all_sql, @all_bind);
1094 foreach my $val (@$vals) {
1095 my ($sql, @bind) = $self->_SWITCH_refkind($val, {
1097 return ($placeholder, $self->_bindtype($k, $val) );
1102 ARRAYREFREF => sub {
1103 my ($sql, @bind) = @$$val;
1104 $self->_assert_bindval_matches_bindtype(@bind);
1105 return ($sql, @bind);
1108 my ($func, $arg, @rest) = %$val;
1109 puke ("Only simple { -func => arg } functions accepted as sub-arguments to BETWEEN")
1110 if (@rest or $func !~ /^ \- (.+)/x);
1111 local $self->{_nested_func_lhs} = $k;
1112 $self->_where_unary_op ($1 => $arg);
1118 push @all_sql, $sql;
1119 push @all_bind, @bind;
1123 (join $and, @all_sql),
1132 my $sql = "( $label $op $clause )";
1133 return ($sql, @bind)
1137 sub _where_field_IN {
1138 my ($self, $k, $op, $vals) = @_;
1140 # backwards compatibility : if scalar, force into an arrayref
1141 $vals = [$vals] if defined $vals && ! ref $vals;
1143 my ($label) = $self->_convert($self->_quote($k));
1144 my ($placeholder) = $self->_convert('?');
1145 $op = $self->_sqlcase($op);
1147 my ($sql, @bind) = $self->_SWITCH_refkind($vals, {
1148 ARRAYREF => sub { # list of choices
1149 if (@$vals) { # nonempty list
1150 my (@all_sql, @all_bind);
1152 for my $val (@$vals) {
1153 my ($sql, @bind) = $self->_SWITCH_refkind($val, {
1155 return ($placeholder, $val);
1160 ARRAYREFREF => sub {
1161 my ($sql, @bind) = @$$val;
1162 $self->_assert_bindval_matches_bindtype(@bind);
1163 return ($sql, @bind);
1166 my ($func, $arg, @rest) = %$val;
1167 puke ("Only simple { -func => arg } functions accepted as sub-arguments to IN")
1168 if (@rest or $func !~ /^ \- (.+)/x);
1169 local $self->{_nested_func_lhs} = $k;
1170 $self->_where_unary_op ($1 => $arg);
1174 'SQL::Abstract before v1.75 used to generate incorrect SQL when the '
1175 . "-$op operator was given an undef-containing list: !!!AUDIT YOUR CODE "
1176 . 'AND DATA!!! (the upcoming Data::Query-based version of SQL::Abstract '
1177 . 'will emit the logically correct SQL instead of raising this exception)'
1181 push @all_sql, $sql;
1182 push @all_bind, @bind;
1186 sprintf ('%s %s ( %s )',
1189 join (', ', @all_sql)
1191 $self->_bindtype($k, @all_bind),
1194 else { # empty list : some databases won't understand "IN ()", so DWIM
1195 my $sql = ($op =~ /\bnot\b/i) ? $self->{sqltrue} : $self->{sqlfalse};
1200 SCALARREF => sub { # literal SQL
1201 my $sql = $self->_open_outer_paren ($$vals);
1202 return ("$label $op ( $sql )");
1204 ARRAYREFREF => sub { # literal SQL with bind
1205 my ($sql, @bind) = @$$vals;
1206 $self->_assert_bindval_matches_bindtype(@bind);
1207 $sql = $self->_open_outer_paren ($sql);
1208 return ("$label $op ( $sql )", @bind);
1212 puke "Argument passed to the '$op' operator can not be undefined";
1216 puke "special op $op requires an arrayref (or scalarref/arrayref-ref)";
1220 return ($sql, @bind);
1223 # Some databases (SQLite) treat col IN (1, 2) different from
1224 # col IN ( (1, 2) ). Use this to strip all outer parens while
1225 # adding them back in the corresponding method
1226 sub _open_outer_paren {
1227 my ($self, $sql) = @_;
1228 $sql = $1 while $sql =~ /^ \s* \( (.*) \) \s* $/xs;
1233 #======================================================================
1235 #======================================================================
1238 my ($self, $arg) = @_;
1241 for my $c ($self->_order_by_chunks ($arg) ) {
1242 $self->_SWITCH_refkind ($c, {
1243 SCALAR => sub { push @sql, $c },
1244 ARRAYREF => sub { push @sql, shift @$c; push @bind, @$c },
1250 $self->_sqlcase(' order by'),
1256 return wantarray ? ($sql, @bind) : $sql;
1259 sub _order_by_chunks {
1260 my ($self, $arg) = @_;
1262 return $self->_SWITCH_refkind($arg, {
1265 map { $self->_order_by_chunks ($_ ) } @$arg;
1268 ARRAYREFREF => sub {
1269 my ($s, @b) = @$$arg;
1270 $self->_assert_bindval_matches_bindtype(@b);
1274 SCALAR => sub {$self->_quote($arg)},
1276 UNDEF => sub {return () },
1278 SCALARREF => sub {$$arg}, # literal SQL, no quoting
1281 # get first pair in hash
1282 my ($key, $val, @rest) = %$arg;
1284 return () unless $key;
1286 if ( @rest or not $key =~ /^-(desc|asc)/i ) {
1287 puke "hash passed to _order_by must have exactly one key (-desc or -asc)";
1293 for my $c ($self->_order_by_chunks ($val)) {
1296 $self->_SWITCH_refkind ($c, {
1301 ($sql, @bind) = @$c;
1305 $sql = $sql . ' ' . $self->_sqlcase($direction);
1307 push @ret, [ $sql, @bind];
1316 #======================================================================
1317 # DATASOURCE (FOR NOW, JUST PLAIN TABLE OR LIST OF TABLES)
1318 #======================================================================
1323 $self->_SWITCH_refkind($from, {
1324 ARRAYREF => sub {join ', ', map { $self->_quote($_) } @$from;},
1325 SCALAR => sub {$self->_quote($from)},
1326 SCALARREF => sub {$$from},
1331 #======================================================================
1333 #======================================================================
1335 # highly optimized, as it's called way too often
1337 # my ($self, $label) = @_;
1339 return '' unless defined $_[1];
1340 return ${$_[1]} if ref($_[1]) eq 'SCALAR';
1342 unless ($_[0]->{quote_char}) {
1343 $_[0]->_assert_pass_injection_guard($_[1]);
1347 my $qref = ref $_[0]->{quote_char};
1350 ($l, $r) = ( $_[0]->{quote_char}, $_[0]->{quote_char} );
1352 elsif ($qref eq 'ARRAY') {
1353 ($l, $r) = @{$_[0]->{quote_char}};
1356 puke "Unsupported quote_char format: $_[0]->{quote_char}";
1358 my $esc = $_[0]->{escape_char} || $r;
1360 # parts containing * are naturally unquoted
1361 return join( $_[0]->{name_sep}||'', map
1362 { $_ eq '*' ? $_ : do { (my $n = $_) =~ s/(\Q$esc\E|\Q$r\E)/$esc$1/g; $l . $n . $r } }
1363 ( $_[0]->{name_sep} ? split (/\Q$_[0]->{name_sep}\E/, $_[1] ) : $_[1] )
1368 # Conversion, if applicable
1370 #my ($self, $arg) = @_;
1371 if ($_[0]->{convert}) {
1372 return $_[0]->_sqlcase($_[0]->{convert}) .'(' . $_[1] . ')';
1379 #my ($self, $col, @vals) = @_;
1380 # called often - tighten code
1381 return $_[0]->{bindtype} eq 'columns'
1382 ? map {[$_[1], $_]} @_[2 .. $#_]
1387 # Dies if any element of @bind is not in [colname => value] format
1388 # if bindtype is 'columns'.
1389 sub _assert_bindval_matches_bindtype {
1390 # my ($self, @bind) = @_;
1392 if ($self->{bindtype} eq 'columns') {
1394 if (!defined $_ || ref($_) ne 'ARRAY' || @$_ != 2) {
1395 puke "bindtype 'columns' selected, you need to pass: [column_name => bind_value]"
1401 sub _join_sql_clauses {
1402 my ($self, $logic, $clauses_aref, $bind_aref) = @_;
1404 if (@$clauses_aref > 1) {
1405 my $join = " " . $self->_sqlcase($logic) . " ";
1406 my $sql = '( ' . join($join, @$clauses_aref) . ' )';
1407 return ($sql, @$bind_aref);
1409 elsif (@$clauses_aref) {
1410 return ($clauses_aref->[0], @$bind_aref); # no parentheses
1413 return (); # if no SQL, ignore @$bind_aref
1418 # Fix SQL case, if so requested
1420 # LDNOTE: if $self->{case} is true, then it contains 'lower', so we
1421 # don't touch the argument ... crooked logic, but let's not change it!
1422 return $_[0]->{case} ? $_[1] : uc($_[1]);
1426 #======================================================================
1427 # DISPATCHING FROM REFKIND
1428 #======================================================================
1431 my ($self, $data) = @_;
1433 return 'UNDEF' unless defined $data;
1435 # blessed objects are treated like scalars
1436 my $ref = (Scalar::Util::blessed $data) ? '' : ref $data;
1438 return 'SCALAR' unless $ref;
1441 while ($ref eq 'REF') {
1443 $ref = (Scalar::Util::blessed $data) ? '' : ref $data;
1447 return ($ref||'SCALAR') . ('REF' x $n_steps);
1451 my ($self, $data) = @_;
1452 my @try = ($self->_refkind($data));
1453 push @try, 'SCALAR_or_UNDEF' if $try[0] eq 'SCALAR' || $try[0] eq 'UNDEF';
1454 push @try, 'FALLBACK';
1458 sub _METHOD_FOR_refkind {
1459 my ($self, $meth_prefix, $data) = @_;
1462 for (@{$self->_try_refkind($data)}) {
1463 $method = $self->can($meth_prefix."_".$_)
1467 return $method || puke "cannot dispatch on '$meth_prefix' for ".$self->_refkind($data);
1471 sub _SWITCH_refkind {
1472 my ($self, $data, $dispatch_table) = @_;
1475 for (@{$self->_try_refkind($data)}) {
1476 $coderef = $dispatch_table->{$_}
1480 puke "no dispatch entry for ".$self->_refkind($data)
1489 #======================================================================
1490 # VALUES, GENERATE, AUTOLOAD
1491 #======================================================================
1493 # LDNOTE: original code from nwiger, didn't touch code in that section
1494 # I feel the AUTOLOAD stuff should not be the default, it should
1495 # only be activated on explicit demand by user.
1499 my $data = shift || return;
1500 puke "Argument to ", __PACKAGE__, "->values must be a \\%hash"
1501 unless ref $data eq 'HASH';
1504 foreach my $k ( sort keys %$data ) {
1505 my $v = $data->{$k};
1506 $self->_SWITCH_refkind($v, {
1508 if ($self->{array_datatypes}) { # array datatype
1509 push @all_bind, $self->_bindtype($k, $v);
1511 else { # literal SQL with bind
1512 my ($sql, @bind) = @$v;
1513 $self->_assert_bindval_matches_bindtype(@bind);
1514 push @all_bind, @bind;
1517 ARRAYREFREF => sub { # literal SQL with bind
1518 my ($sql, @bind) = @${$v};
1519 $self->_assert_bindval_matches_bindtype(@bind);
1520 push @all_bind, @bind;
1522 SCALARREF => sub { # literal SQL without bind
1524 SCALAR_or_UNDEF => sub {
1525 push @all_bind, $self->_bindtype($k, $v);
1536 my(@sql, @sqlq, @sqlv);
1540 if ($ref eq 'HASH') {
1541 for my $k (sort keys %$_) {
1544 my $label = $self->_quote($k);
1545 if ($r eq 'ARRAY') {
1546 # literal SQL with bind
1547 my ($sql, @bind) = @$v;
1548 $self->_assert_bindval_matches_bindtype(@bind);
1549 push @sqlq, "$label = $sql";
1551 } elsif ($r eq 'SCALAR') {
1552 # literal SQL without bind
1553 push @sqlq, "$label = $$v";
1555 push @sqlq, "$label = ?";
1556 push @sqlv, $self->_bindtype($k, $v);
1559 push @sql, $self->_sqlcase('set'), join ', ', @sqlq;
1560 } elsif ($ref eq 'ARRAY') {
1561 # unlike insert(), assume these are ONLY the column names, i.e. for SQL
1564 if ($r eq 'ARRAY') { # literal SQL with bind
1565 my ($sql, @bind) = @$v;
1566 $self->_assert_bindval_matches_bindtype(@bind);
1569 } elsif ($r eq 'SCALAR') { # literal SQL without bind
1570 # embedded literal SQL
1577 push @sql, '(' . join(', ', @sqlq) . ')';
1578 } elsif ($ref eq 'SCALAR') {
1582 # strings get case twiddled
1583 push @sql, $self->_sqlcase($_);
1587 my $sql = join ' ', @sql;
1589 # this is pretty tricky
1590 # if ask for an array, return ($stmt, @bind)
1591 # otherwise, s/?/shift @sqlv/ to put it inline
1593 return ($sql, @sqlv);
1595 1 while $sql =~ s/\?/my $d = shift(@sqlv);
1596 ref $d ? $d->[1] : $d/e;
1605 # This allows us to check for a local, then _form, attr
1607 my($name) = $AUTOLOAD =~ /.*::(.+)/;
1608 return $self->generate($name, @_);
1619 SQL::Abstract - Generate SQL from Perl data structures
1625 my $sql = SQL::Abstract->new;
1627 my($stmt, @bind) = $sql->select($source, \@fields, \%where, \@order);
1629 my($stmt, @bind) = $sql->insert($table, \%fieldvals || \@values);
1631 my($stmt, @bind) = $sql->update($table, \%fieldvals, \%where);
1633 my($stmt, @bind) = $sql->delete($table, \%where);
1635 # Then, use these in your DBI statements
1636 my $sth = $dbh->prepare($stmt);
1637 $sth->execute(@bind);
1639 # Just generate the WHERE clause
1640 my($stmt, @bind) = $sql->where(\%where, \@order);
1642 # Return values in the same order, for hashed queries
1643 # See PERFORMANCE section for more details
1644 my @bind = $sql->values(\%fieldvals);
1648 This module was inspired by the excellent L<DBIx::Abstract>.
1649 However, in using that module I found that what I really wanted
1650 to do was generate SQL, but still retain complete control over my
1651 statement handles and use the DBI interface. So, I set out to
1652 create an abstract SQL generation module.
1654 While based on the concepts used by L<DBIx::Abstract>, there are
1655 several important differences, especially when it comes to WHERE
1656 clauses. I have modified the concepts used to make the SQL easier
1657 to generate from Perl data structures and, IMO, more intuitive.
1658 The underlying idea is for this module to do what you mean, based
1659 on the data structures you provide it. The big advantage is that
1660 you don't have to modify your code every time your data changes,
1661 as this module figures it out.
1663 To begin with, an SQL INSERT is as easy as just specifying a hash
1664 of C<key=value> pairs:
1667 name => 'Jimbo Bobson',
1668 phone => '123-456-7890',
1669 address => '42 Sister Lane',
1670 city => 'St. Louis',
1671 state => 'Louisiana',
1674 The SQL can then be generated with this:
1676 my($stmt, @bind) = $sql->insert('people', \%data);
1678 Which would give you something like this:
1680 $stmt = "INSERT INTO people
1681 (address, city, name, phone, state)
1682 VALUES (?, ?, ?, ?, ?)";
1683 @bind = ('42 Sister Lane', 'St. Louis', 'Jimbo Bobson',
1684 '123-456-7890', 'Louisiana');
1686 These are then used directly in your DBI code:
1688 my $sth = $dbh->prepare($stmt);
1689 $sth->execute(@bind);
1691 =head2 Inserting and Updating Arrays
1693 If your database has array types (like for example Postgres),
1694 activate the special option C<< array_datatypes => 1 >>
1695 when creating the C<SQL::Abstract> object.
1696 Then you may use an arrayref to insert and update database array types:
1698 my $sql = SQL::Abstract->new(array_datatypes => 1);
1700 planets => [qw/Mercury Venus Earth Mars/]
1703 my($stmt, @bind) = $sql->insert('solar_system', \%data);
1707 $stmt = "INSERT INTO solar_system (planets) VALUES (?)"
1709 @bind = (['Mercury', 'Venus', 'Earth', 'Mars']);
1712 =head2 Inserting and Updating SQL
1714 In order to apply SQL functions to elements of your C<%data> you may
1715 specify a reference to an arrayref for the given hash value. For example,
1716 if you need to execute the Oracle C<to_date> function on a value, you can
1717 say something like this:
1721 date_entered => \[ "to_date(?,'MM/DD/YYYY')", "03/02/2003" ],
1724 The first value in the array is the actual SQL. Any other values are
1725 optional and would be included in the bind values array. This gives
1728 my($stmt, @bind) = $sql->insert('people', \%data);
1730 $stmt = "INSERT INTO people (name, date_entered)
1731 VALUES (?, to_date(?,'MM/DD/YYYY'))";
1732 @bind = ('Bill', '03/02/2003');
1734 An UPDATE is just as easy, all you change is the name of the function:
1736 my($stmt, @bind) = $sql->update('people', \%data);
1738 Notice that your C<%data> isn't touched; the module will generate
1739 the appropriately quirky SQL for you automatically. Usually you'll
1740 want to specify a WHERE clause for your UPDATE, though, which is
1741 where handling C<%where> hashes comes in handy...
1743 =head2 Complex where statements
1745 This module can generate pretty complicated WHERE statements
1746 easily. For example, simple C<key=value> pairs are taken to mean
1747 equality, and if you want to see if a field is within a set
1748 of values, you can use an arrayref. Let's say we wanted to
1749 SELECT some data based on this criteria:
1752 requestor => 'inna',
1753 worker => ['nwiger', 'rcwe', 'sfz'],
1754 status => { '!=', 'completed' }
1757 my($stmt, @bind) = $sql->select('tickets', '*', \%where);
1759 The above would give you something like this:
1761 $stmt = "SELECT * FROM tickets WHERE
1762 ( requestor = ? ) AND ( status != ? )
1763 AND ( worker = ? OR worker = ? OR worker = ? )";
1764 @bind = ('inna', 'completed', 'nwiger', 'rcwe', 'sfz');
1766 Which you could then use in DBI code like so:
1768 my $sth = $dbh->prepare($stmt);
1769 $sth->execute(@bind);
1775 The methods are simple. There's one for each major SQL operation,
1776 and a constructor you use first. The arguments are specified in a
1777 similar order to each method (table, then fields, then a where
1778 clause) to try and simplify things.
1780 =head2 new(option => 'value')
1782 The C<new()> function takes a list of options and values, and returns
1783 a new B<SQL::Abstract> object which can then be used to generate SQL
1784 through the methods below. The options accepted are:
1790 If set to 'lower', then SQL will be generated in all lowercase. By
1791 default SQL is generated in "textbook" case meaning something like:
1793 SELECT a_field FROM a_table WHERE some_field LIKE '%someval%'
1795 Any setting other than 'lower' is ignored.
1799 This determines what the default comparison operator is. By default
1800 it is C<=>, meaning that a hash like this:
1802 %where = (name => 'nwiger', email => 'nate@wiger.org');
1804 Will generate SQL like this:
1806 WHERE name = 'nwiger' AND email = 'nate@wiger.org'
1808 However, you may want loose comparisons by default, so if you set
1809 C<cmp> to C<like> you would get SQL such as:
1811 WHERE name like 'nwiger' AND email like 'nate@wiger.org'
1813 You can also override the comparison on an individual basis - see
1814 the huge section on L</"WHERE CLAUSES"> at the bottom.
1816 =item sqltrue, sqlfalse
1818 Expressions for inserting boolean values within SQL statements.
1819 By default these are C<1=1> and C<1=0>. They are used
1820 by the special operators C<-in> and C<-not_in> for generating
1821 correct SQL even when the argument is an empty array (see below).
1825 This determines the default logical operator for multiple WHERE
1826 statements in arrays or hashes. If absent, the default logic is "or"
1827 for arrays, and "and" for hashes. This means that a WHERE
1831 event_date => {'>=', '2/13/99'},
1832 event_date => {'<=', '4/24/03'},
1835 will generate SQL like this:
1837 WHERE event_date >= '2/13/99' OR event_date <= '4/24/03'
1839 This is probably not what you want given this query, though (look
1840 at the dates). To change the "OR" to an "AND", simply specify:
1842 my $sql = SQL::Abstract->new(logic => 'and');
1844 Which will change the above C<WHERE> to:
1846 WHERE event_date >= '2/13/99' AND event_date <= '4/24/03'
1848 The logic can also be changed locally by inserting
1849 a modifier in front of an arrayref :
1851 @where = (-and => [event_date => {'>=', '2/13/99'},
1852 event_date => {'<=', '4/24/03'} ]);
1854 See the L</"WHERE CLAUSES"> section for explanations.
1858 This will automatically convert comparisons using the specified SQL
1859 function for both column and value. This is mostly used with an argument
1860 of C<upper> or C<lower>, so that the SQL will have the effect of
1861 case-insensitive "searches". For example, this:
1863 $sql = SQL::Abstract->new(convert => 'upper');
1864 %where = (keywords => 'MaKe iT CAse inSeNSItive');
1866 Will turn out the following SQL:
1868 WHERE upper(keywords) like upper('MaKe iT CAse inSeNSItive')
1870 The conversion can be C<upper()>, C<lower()>, or any other SQL function
1871 that can be applied symmetrically to fields (actually B<SQL::Abstract> does
1872 not validate this option; it will just pass through what you specify verbatim).
1876 This is a kludge because many databases suck. For example, you can't
1877 just bind values using DBI's C<execute()> for Oracle C<CLOB> or C<BLOB> fields.
1878 Instead, you have to use C<bind_param()>:
1880 $sth->bind_param(1, 'reg data');
1881 $sth->bind_param(2, $lots, {ora_type => ORA_CLOB});
1883 The problem is, B<SQL::Abstract> will normally just return a C<@bind> array,
1884 which loses track of which field each slot refers to. Fear not.
1886 If you specify C<bindtype> in new, you can determine how C<@bind> is returned.
1887 Currently, you can specify either C<normal> (default) or C<columns>. If you
1888 specify C<columns>, you will get an array that looks like this:
1890 my $sql = SQL::Abstract->new(bindtype => 'columns');
1891 my($stmt, @bind) = $sql->insert(...);
1894 [ 'column1', 'value1' ],
1895 [ 'column2', 'value2' ],
1896 [ 'column3', 'value3' ],
1899 You can then iterate through this manually, using DBI's C<bind_param()>.
1901 $sth->prepare($stmt);
1904 my($col, $data) = @$_;
1905 if ($col eq 'details' || $col eq 'comments') {
1906 $sth->bind_param($i, $data, {ora_type => ORA_CLOB});
1907 } elsif ($col eq 'image') {
1908 $sth->bind_param($i, $data, {ora_type => ORA_BLOB});
1910 $sth->bind_param($i, $data);
1914 $sth->execute; # execute without @bind now
1916 Now, why would you still use B<SQL::Abstract> if you have to do this crap?
1917 Basically, the advantage is still that you don't have to care which fields
1918 are or are not included. You could wrap that above C<for> loop in a simple
1919 sub called C<bind_fields()> or something and reuse it repeatedly. You still
1920 get a layer of abstraction over manual SQL specification.
1922 Note that if you set L</bindtype> to C<columns>, the C<\[ $sql, @bind ]>
1923 construct (see L</Literal SQL with placeholders and bind values (subqueries)>)
1924 will expect the bind values in this format.
1928 This is the character that a table or column name will be quoted
1929 with. By default this is an empty string, but you could set it to
1930 the character C<`>, to generate SQL like this:
1932 SELECT `a_field` FROM `a_table` WHERE `some_field` LIKE '%someval%'
1934 Alternatively, you can supply an array ref of two items, the first being the left
1935 hand quote character, and the second the right hand quote character. For
1936 example, you could supply C<['[',']']> for SQL Server 2000 compliant quotes
1937 that generates SQL like this:
1939 SELECT [a_field] FROM [a_table] WHERE [some_field] LIKE '%someval%'
1941 Quoting is useful if you have tables or columns names that are reserved
1942 words in your database's SQL dialect.
1946 This is the character that will be used to escape L</quote_char>s appearing
1947 in an identifier before it has been quoted.
1949 The paramter default in case of a single L</quote_char> character is the quote
1952 When opening-closing-style quoting is used (L</quote_char> is an arrayref)
1953 this parameter defaults to the B<closing (right)> L</quote_char>. Occurences
1954 of the B<opening (left)> L</quote_char> within the identifier are currently left
1955 untouched. The default for opening-closing-style quotes may change in future
1956 versions, thus you are B<strongly encouraged> to specify the escape character
1961 This is the character that separates a table and column name. It is
1962 necessary to specify this when the C<quote_char> option is selected,
1963 so that tables and column names can be individually quoted like this:
1965 SELECT `table`.`one_field` FROM `table` WHERE `table`.`other_field` = 1
1967 =item injection_guard
1969 A regular expression C<qr/.../> that is applied to any C<-function> and unquoted
1970 column name specified in a query structure. This is a safety mechanism to avoid
1971 injection attacks when mishandling user input e.g.:
1973 my %condition_as_column_value_pairs = get_values_from_user();
1974 $sqla->select( ... , \%condition_as_column_value_pairs );
1976 If the expression matches an exception is thrown. Note that literal SQL
1977 supplied via C<\'...'> or C<\['...']> is B<not> checked in any way.
1979 Defaults to checking for C<;> and the C<GO> keyword (TransactSQL)
1981 =item array_datatypes
1983 When this option is true, arrayrefs in INSERT or UPDATE are
1984 interpreted as array datatypes and are passed directly
1986 When this option is false, arrayrefs are interpreted
1987 as literal SQL, just like refs to arrayrefs
1988 (but this behavior is for backwards compatibility; when writing
1989 new queries, use the "reference to arrayref" syntax
1995 Takes a reference to a list of "special operators"
1996 to extend the syntax understood by L<SQL::Abstract>.
1997 See section L</"SPECIAL OPERATORS"> for details.
2001 Takes a reference to a list of "unary operators"
2002 to extend the syntax understood by L<SQL::Abstract>.
2003 See section L</"UNARY OPERATORS"> for details.
2009 =head2 insert($table, \@values || \%fieldvals, \%options)
2011 This is the simplest function. You simply give it a table name
2012 and either an arrayref of values or hashref of field/value pairs.
2013 It returns an SQL INSERT statement and a list of bind values.
2014 See the sections on L</"Inserting and Updating Arrays"> and
2015 L</"Inserting and Updating SQL"> for information on how to insert
2016 with those data types.
2018 The optional C<\%options> hash reference may contain additional
2019 options to generate the insert SQL. Currently supported options
2026 Takes either a scalar of raw SQL fields, or an array reference of
2027 field names, and adds on an SQL C<RETURNING> statement at the end.
2028 This allows you to return data generated by the insert statement
2029 (such as row IDs) without performing another C<SELECT> statement.
2030 Note, however, this is not part of the SQL standard and may not
2031 be supported by all database engines.
2035 =head2 update($table, \%fieldvals, \%where)
2037 This takes a table, hashref of field/value pairs, and an optional
2038 hashref L<WHERE clause|/WHERE CLAUSES>. It returns an SQL UPDATE function and a list
2040 See the sections on L</"Inserting and Updating Arrays"> and
2041 L</"Inserting and Updating SQL"> for information on how to insert
2042 with those data types.
2044 =head2 select($source, $fields, $where, $order)
2046 This returns a SQL SELECT statement and associated list of bind values, as
2047 specified by the arguments :
2053 Specification of the 'FROM' part of the statement.
2054 The argument can be either a plain scalar (interpreted as a table
2055 name, will be quoted), or an arrayref (interpreted as a list
2056 of table names, joined by commas, quoted), or a scalarref
2057 (literal table name, not quoted), or a ref to an arrayref
2058 (list of literal table names, joined by commas, not quoted).
2062 Specification of the list of fields to retrieve from
2064 The argument can be either an arrayref (interpreted as a list
2065 of field names, will be joined by commas and quoted), or a
2066 plain scalar (literal SQL, not quoted).
2067 Please observe that this API is not as flexible as that of
2068 the first argument C<$source>, for backwards compatibility reasons.
2072 Optional argument to specify the WHERE part of the query.
2073 The argument is most often a hashref, but can also be
2074 an arrayref or plain scalar --
2075 see section L<WHERE clause|/"WHERE CLAUSES"> for details.
2079 Optional argument to specify the ORDER BY part of the query.
2080 The argument can be a scalar, a hashref or an arrayref
2081 -- see section L<ORDER BY clause|/"ORDER BY CLAUSES">
2087 =head2 delete($table, \%where)
2089 This takes a table name and optional hashref L<WHERE clause|/WHERE CLAUSES>.
2090 It returns an SQL DELETE statement and list of bind values.
2092 =head2 where(\%where, \@order)
2094 This is used to generate just the WHERE clause. For example,
2095 if you have an arbitrary data structure and know what the
2096 rest of your SQL is going to look like, but want an easy way
2097 to produce a WHERE clause, use this. It returns an SQL WHERE
2098 clause and list of bind values.
2101 =head2 values(\%data)
2103 This just returns the values from the hash C<%data>, in the same
2104 order that would be returned from any of the other above queries.
2105 Using this allows you to markedly speed up your queries if you
2106 are affecting lots of rows. See below under the L</"PERFORMANCE"> section.
2108 =head2 generate($any, 'number', $of, \@data, $struct, \%types)
2110 Warning: This is an experimental method and subject to change.
2112 This returns arbitrarily generated SQL. It's a really basic shortcut.
2113 It will return two different things, depending on return context:
2115 my($stmt, @bind) = $sql->generate('create table', \$table, \@fields);
2116 my $stmt_and_val = $sql->generate('create table', \$table, \@fields);
2118 These would return the following:
2120 # First calling form
2121 $stmt = "CREATE TABLE test (?, ?)";
2122 @bind = (field1, field2);
2124 # Second calling form
2125 $stmt_and_val = "CREATE TABLE test (field1, field2)";
2127 Depending on what you're trying to do, it's up to you to choose the correct
2128 format. In this example, the second form is what you would want.
2132 $sql->generate('alter session', { nls_date_format => 'MM/YY' });
2136 ALTER SESSION SET nls_date_format = 'MM/YY'
2138 You get the idea. Strings get their case twiddled, but everything
2139 else remains verbatim.
2141 =head1 EXPORTABLE FUNCTIONS
2143 =head2 is_plain_value
2145 Determines if the supplied argument is a plain value as understood by this
2150 =item * The value is C<undef>
2152 =item * The value is a non-reference
2154 =item * The value is an object with stringification overloading
2156 =item * The value is of the form C<< { -value => $anything } >>
2160 On failure returns C<undef>, on sucess returns a B<scalar> reference
2161 to the original supplied argument.
2167 The stringification overloading detection is rather advanced: it takes
2168 into consideration not only the presence of a C<""> overload, but if that
2169 fails also checks for enabled
2170 L<autogenerated versions of C<"">|overload/Magic Autogeneration>, based
2171 on either C<0+> or C<bool>.
2173 Unfortunately testing in the field indicates that this
2174 detection B<< may tickle a latent bug in perl versions before 5.018 >>,
2175 but only when very large numbers of stringifying objects are involved.
2176 At the time of writing ( Sep 2014 ) there is no clear explanation of
2177 the direct cause, nor is there a manageably small test case that reliably
2178 reproduces the problem.
2180 If you encounter any of the following exceptions in B<random places within
2181 your application stack> - this module may be to blame:
2183 Operation "ne": no method found,
2184 left argument in overloaded package <something>,
2185 right argument in overloaded package <something>
2189 Stub found while resolving method "???" overloading """" in package <something>
2191 If you fall victim to the above - please attempt to reduce the problem
2192 to something that could be sent to the L<SQL::Abstract developers
2193 |DBIx::Class/GETTING_HELP/SUPPORT>
2194 (either publicly or privately). As a workaround in the meantime you can
2195 set C<$ENV{SQLA_ISVALUE_IGNORE_AUTOGENERATED_STRINGIFICATION}> to a true
2196 value, which will most likely eliminate your problem (at the expense of
2197 not being able to properly detect exotic forms of stringification).
2199 This notice and environment variable will be removed in a future version,
2200 as soon as the underlying problem is found and a reliable workaround is
2205 =head2 is_literal_value
2207 Determines if the supplied argument is a literal value as understood by this
2212 =item * C<\$sql_string>
2214 =item * C<\[ $sql_string, @bind_values ]>
2216 =item * C<< { -ident => $plain_defined_string } >>
2220 On failure returns C<undef>, on sucess returns an B<array> reference
2221 containing the unpacked version of the supplied literal SQL and bind values.
2223 =head1 WHERE CLAUSES
2227 This module uses a variation on the idea from L<DBIx::Abstract>. It
2228 is B<NOT>, repeat I<not> 100% compatible. B<The main logic of this
2229 module is that things in arrays are OR'ed, and things in hashes
2232 The easiest way to explain is to show lots of examples. After
2233 each C<%where> hash shown, it is assumed you used:
2235 my($stmt, @bind) = $sql->where(\%where);
2237 However, note that the C<%where> hash can be used directly in any
2238 of the other functions as well, as described above.
2240 =head2 Key-value pairs
2242 So, let's get started. To begin, a simple hash:
2246 status => 'completed'
2249 Is converted to SQL C<key = val> statements:
2251 $stmt = "WHERE user = ? AND status = ?";
2252 @bind = ('nwiger', 'completed');
2254 One common thing I end up doing is having a list of values that
2255 a field can be in. To do this, simply specify a list inside of
2260 status => ['assigned', 'in-progress', 'pending'];
2263 This simple code will create the following:
2265 $stmt = "WHERE user = ? AND ( status = ? OR status = ? OR status = ? )";
2266 @bind = ('nwiger', 'assigned', 'in-progress', 'pending');
2268 A field associated to an empty arrayref will be considered a
2269 logical false and will generate 0=1.
2271 =head2 Tests for NULL values
2273 If the value part is C<undef> then this is converted to SQL <IS NULL>
2282 $stmt = "WHERE user = ? AND status IS NULL";
2285 To test if a column IS NOT NULL:
2289 status => { '!=', undef },
2292 =head2 Specific comparison operators
2294 If you want to specify a different type of operator for your comparison,
2295 you can use a hashref for a given column:
2299 status => { '!=', 'completed' }
2302 Which would generate:
2304 $stmt = "WHERE user = ? AND status != ?";
2305 @bind = ('nwiger', 'completed');
2307 To test against multiple values, just enclose the values in an arrayref:
2309 status => { '=', ['assigned', 'in-progress', 'pending'] };
2311 Which would give you:
2313 "WHERE status = ? OR status = ? OR status = ?"
2316 The hashref can also contain multiple pairs, in which case it is expanded
2317 into an C<AND> of its elements:
2321 status => { '!=', 'completed', -not_like => 'pending%' }
2324 # Or more dynamically, like from a form
2325 $where{user} = 'nwiger';
2326 $where{status}{'!='} = 'completed';
2327 $where{status}{'-not_like'} = 'pending%';
2329 # Both generate this
2330 $stmt = "WHERE user = ? AND status != ? AND status NOT LIKE ?";
2331 @bind = ('nwiger', 'completed', 'pending%');
2334 To get an OR instead, you can combine it with the arrayref idea:
2338 priority => [ { '=', 2 }, { '>', 5 } ]
2341 Which would generate:
2343 $stmt = "WHERE ( priority = ? OR priority > ? ) AND user = ?";
2344 @bind = ('2', '5', 'nwiger');
2346 If you want to include literal SQL (with or without bind values), just use a
2347 scalar reference or array reference as the value:
2350 date_entered => { '>' => \["to_date(?, 'MM/DD/YYYY')", "11/26/2008"] },
2351 date_expires => { '<' => \"now()" }
2354 Which would generate:
2356 $stmt = "WHERE date_entered > "to_date(?, 'MM/DD/YYYY') AND date_expires < now()";
2357 @bind = ('11/26/2008');
2360 =head2 Logic and nesting operators
2362 In the example above,
2363 there is a subtle trap if you want to say something like
2364 this (notice the C<AND>):
2366 WHERE priority != ? AND priority != ?
2368 Because, in Perl you I<can't> do this:
2370 priority => { '!=', 2, '!=', 1 }
2372 As the second C<!=> key will obliterate the first. The solution
2373 is to use the special C<-modifier> form inside an arrayref:
2375 priority => [ -and => {'!=', 2},
2379 Normally, these would be joined by C<OR>, but the modifier tells it
2380 to use C<AND> instead. (Hint: You can use this in conjunction with the
2381 C<logic> option to C<new()> in order to change the way your queries
2382 work by default.) B<Important:> Note that the C<-modifier> goes
2383 B<INSIDE> the arrayref, as an extra first element. This will
2384 B<NOT> do what you think it might:
2386 priority => -and => [{'!=', 2}, {'!=', 1}] # WRONG!
2388 Here is a quick list of equivalencies, since there is some overlap:
2391 status => {'!=', 'completed', 'not like', 'pending%' }
2392 status => [ -and => {'!=', 'completed'}, {'not like', 'pending%'}]
2395 status => {'=', ['assigned', 'in-progress']}
2396 status => [ -or => {'=', 'assigned'}, {'=', 'in-progress'}]
2397 status => [ {'=', 'assigned'}, {'=', 'in-progress'} ]
2401 =head2 Special operators : IN, BETWEEN, etc.
2403 You can also use the hashref format to compare a list of fields using the
2404 C<IN> comparison operator, by specifying the list as an arrayref:
2407 status => 'completed',
2408 reportid => { -in => [567, 2335, 2] }
2411 Which would generate:
2413 $stmt = "WHERE status = ? AND reportid IN (?,?,?)";
2414 @bind = ('completed', '567', '2335', '2');
2416 The reverse operator C<-not_in> generates SQL C<NOT IN> and is used in
2419 If the argument to C<-in> is an empty array, 'sqlfalse' is generated
2420 (by default : C<1=0>). Similarly, C<< -not_in => [] >> generates
2421 'sqltrue' (by default : C<1=1>).
2423 In addition to the array you can supply a chunk of literal sql or
2424 literal sql with bind:
2427 customer => { -in => \[
2428 'SELECT cust_id FROM cust WHERE balance > ?',
2431 status => { -in => \'SELECT status_codes FROM states' },
2437 customer IN ( SELECT cust_id FROM cust WHERE balance > ? )
2438 AND status IN ( SELECT status_codes FROM states )
2442 Finally, if the argument to C<-in> is not a reference, it will be
2443 treated as a single-element array.
2445 Another pair of operators is C<-between> and C<-not_between>,
2446 used with an arrayref of two values:
2450 completion_date => {
2451 -not_between => ['2002-10-01', '2003-02-06']
2457 WHERE user = ? AND completion_date NOT BETWEEN ( ? AND ? )
2459 Just like with C<-in> all plausible combinations of literal SQL
2463 start0 => { -between => [ 1, 2 ] },
2464 start1 => { -between => \["? AND ?", 1, 2] },
2465 start2 => { -between => \"lower(x) AND upper(y)" },
2466 start3 => { -between => [
2468 \["upper(?)", 'stuff' ],
2475 ( start0 BETWEEN ? AND ? )
2476 AND ( start1 BETWEEN ? AND ? )
2477 AND ( start2 BETWEEN lower(x) AND upper(y) )
2478 AND ( start3 BETWEEN lower(x) AND upper(?) )
2480 @bind = (1, 2, 1, 2, 'stuff');
2483 These are the two builtin "special operators"; but the
2484 list can be expanded : see section L</"SPECIAL OPERATORS"> below.
2486 =head2 Unary operators: bool
2488 If you wish to test against boolean columns or functions within your
2489 database you can use the C<-bool> and C<-not_bool> operators. For
2490 example to test the column C<is_user> being true and the column
2491 C<is_enabled> being false you would use:-
2495 -not_bool => 'is_enabled',
2500 WHERE is_user AND NOT is_enabled
2502 If a more complex combination is required, testing more conditions,
2503 then you should use the and/or operators:-
2508 -not_bool => { two=> { -rlike => 'bar' } },
2509 -not_bool => { three => [ { '=', 2 }, { '>', 5 } ] },
2520 (NOT ( three = ? OR three > ? ))
2523 =head2 Nested conditions, -and/-or prefixes
2525 So far, we've seen how multiple conditions are joined with a top-level
2526 C<AND>. We can change this by putting the different conditions we want in
2527 hashes and then putting those hashes in an array. For example:
2532 status => { -like => ['pending%', 'dispatched'] },
2536 status => 'unassigned',
2540 This data structure would create the following:
2542 $stmt = "WHERE ( user = ? AND ( status LIKE ? OR status LIKE ? ) )
2543 OR ( user = ? AND status = ? ) )";
2544 @bind = ('nwiger', 'pending', 'dispatched', 'robot', 'unassigned');
2547 Clauses in hashrefs or arrayrefs can be prefixed with an C<-and> or C<-or>
2548 to change the logic inside :
2554 -and => [ workhrs => {'>', 20}, geo => 'ASIA' ],
2555 -or => { workhrs => {'<', 50}, geo => 'EURO' },
2562 WHERE ( user = ? AND (
2563 ( workhrs > ? AND geo = ? )
2564 OR ( workhrs < ? OR geo = ? )
2567 =head3 Algebraic inconsistency, for historical reasons
2569 C<Important note>: when connecting several conditions, the C<-and->|C<-or>
2570 operator goes C<outside> of the nested structure; whereas when connecting
2571 several constraints on one column, the C<-and> operator goes
2572 C<inside> the arrayref. Here is an example combining both features :
2575 -and => [a => 1, b => 2],
2576 -or => [c => 3, d => 4],
2577 e => [-and => {-like => 'foo%'}, {-like => '%bar'} ]
2582 WHERE ( ( ( a = ? AND b = ? )
2583 OR ( c = ? OR d = ? )
2584 OR ( e LIKE ? AND e LIKE ? ) ) )
2586 This difference in syntax is unfortunate but must be preserved for
2587 historical reasons. So be careful : the two examples below would
2588 seem algebraically equivalent, but they are not
2590 {col => [-and => {-like => 'foo%'}, {-like => '%bar'}]}
2591 # yields : WHERE ( ( col LIKE ? AND col LIKE ? ) )
2593 [-and => {col => {-like => 'foo%'}, {col => {-like => '%bar'}}]]
2594 # yields : WHERE ( ( col LIKE ? OR col LIKE ? ) )
2597 =head2 Literal SQL and value type operators
2599 The basic premise of SQL::Abstract is that in WHERE specifications the "left
2600 side" is a column name and the "right side" is a value (normally rendered as
2601 a placeholder). This holds true for both hashrefs and arrayref pairs as you
2602 see in the L</WHERE CLAUSES> examples above. Sometimes it is necessary to
2603 alter this behavior. There are several ways of doing so.
2607 This is a virtual operator that signals the string to its right side is an
2608 identifier (a column name) and not a value. For example to compare two
2609 columns you would write:
2612 priority => { '<', 2 },
2613 requestor => { -ident => 'submitter' },
2618 $stmt = "WHERE priority < ? AND requestor = submitter";
2621 If you are maintaining legacy code you may see a different construct as
2622 described in L</Deprecated usage of Literal SQL>, please use C<-ident> in new
2627 This is a virtual operator that signals that the construct to its right side
2628 is a value to be passed to DBI. This is for example necessary when you want
2629 to write a where clause against an array (for RDBMS that support such
2630 datatypes). For example:
2633 array => { -value => [1, 2, 3] }
2638 $stmt = 'WHERE array = ?';
2639 @bind = ([1, 2, 3]);
2641 Note that if you were to simply say:
2647 the result would probably not be what you wanted:
2649 $stmt = 'WHERE array = ? OR array = ? OR array = ?';
2654 Finally, sometimes only literal SQL will do. To include a random snippet
2655 of SQL verbatim, you specify it as a scalar reference. Consider this only
2656 as a last resort. Usually there is a better way. For example:
2659 priority => { '<', 2 },
2660 requestor => { -in => \'(SELECT name FROM hitmen)' },
2665 $stmt = "WHERE priority < ? AND requestor IN (SELECT name FROM hitmen)"
2668 Note that in this example, you only get one bind parameter back, since
2669 the verbatim SQL is passed as part of the statement.
2673 Never use untrusted input as a literal SQL argument - this is a massive
2674 security risk (there is no way to check literal snippets for SQL
2675 injections and other nastyness). If you need to deal with untrusted input
2676 use literal SQL with placeholders as described next.
2678 =head3 Literal SQL with placeholders and bind values (subqueries)
2680 If the literal SQL to be inserted has placeholders and bind values,
2681 use a reference to an arrayref (yes this is a double reference --
2682 not so common, but perfectly legal Perl). For example, to find a date
2683 in Postgres you can use something like this:
2686 date_column => \[ "= date '2008-09-30' - ?::integer", 10 ]
2691 $stmt = "WHERE ( date_column = date '2008-09-30' - ?::integer )"
2694 Note that you must pass the bind values in the same format as they are returned
2695 by L</where>. That means that if you set L</bindtype> to C<columns>, you must
2696 provide the bind values in the C<< [ column_meta => value ] >> format, where
2697 C<column_meta> is an opaque scalar value; most commonly the column name, but
2698 you can use any scalar value (including references and blessed references),
2699 L<SQL::Abstract> will simply pass it through intact. So if C<bindtype> is set
2700 to C<columns> the above example will look like:
2703 date_column => \[ "= date '2008-09-30' - ?::integer", [ {} => 10 ] ]
2706 Literal SQL is especially useful for nesting parenthesized clauses in the
2707 main SQL query. Here is a first example :
2709 my ($sub_stmt, @sub_bind) = ("SELECT c1 FROM t1 WHERE c2 < ? AND c3 LIKE ?",
2713 bar => \["IN ($sub_stmt)" => @sub_bind],
2718 $stmt = "WHERE (foo = ? AND bar IN (SELECT c1 FROM t1
2719 WHERE c2 < ? AND c3 LIKE ?))";
2720 @bind = (1234, 100, "foo%");
2722 Other subquery operators, like for example C<"E<gt> ALL"> or C<"NOT IN">,
2723 are expressed in the same way. Of course the C<$sub_stmt> and
2724 its associated bind values can be generated through a former call
2727 my ($sub_stmt, @sub_bind)
2728 = $sql->select("t1", "c1", {c2 => {"<" => 100},
2729 c3 => {-like => "foo%"}});
2732 bar => \["> ALL ($sub_stmt)" => @sub_bind],
2735 In the examples above, the subquery was used as an operator on a column;
2736 but the same principle also applies for a clause within the main C<%where>
2737 hash, like an EXISTS subquery :
2739 my ($sub_stmt, @sub_bind)
2740 = $sql->select("t1", "*", {c1 => 1, c2 => \"> t0.c0"});
2741 my %where = ( -and => [
2743 \["EXISTS ($sub_stmt)" => @sub_bind],
2748 $stmt = "WHERE (foo = ? AND EXISTS (SELECT * FROM t1
2749 WHERE c1 = ? AND c2 > t0.c0))";
2753 Observe that the condition on C<c2> in the subquery refers to
2754 column C<t0.c0> of the main query : this is I<not> a bind
2755 value, so we have to express it through a scalar ref.
2756 Writing C<< c2 => {">" => "t0.c0"} >> would have generated
2757 C<< c2 > ? >> with bind value C<"t0.c0"> ... not exactly
2758 what we wanted here.
2760 Finally, here is an example where a subquery is used
2761 for expressing unary negation:
2763 my ($sub_stmt, @sub_bind)
2764 = $sql->where({age => [{"<" => 10}, {">" => 20}]});
2765 $sub_stmt =~ s/^ where //i; # don't want "WHERE" in the subclause
2767 lname => {like => '%son%'},
2768 \["NOT ($sub_stmt)" => @sub_bind],
2773 $stmt = "lname LIKE ? AND NOT ( age < ? OR age > ? )"
2774 @bind = ('%son%', 10, 20)
2776 =head3 Deprecated usage of Literal SQL
2778 Below are some examples of archaic use of literal SQL. It is shown only as
2779 reference for those who deal with legacy code. Each example has a much
2780 better, cleaner and safer alternative that users should opt for in new code.
2786 my %where = ( requestor => \'IS NOT NULL' )
2788 $stmt = "WHERE requestor IS NOT NULL"
2790 This used to be the way of generating NULL comparisons, before the handling
2791 of C<undef> got formalized. For new code please use the superior syntax as
2792 described in L</Tests for NULL values>.
2796 my %where = ( requestor => \'= submitter' )
2798 $stmt = "WHERE requestor = submitter"
2800 This used to be the only way to compare columns. Use the superior L</-ident>
2801 method for all new code. For example an identifier declared in such a way
2802 will be properly quoted if L</quote_char> is properly set, while the legacy
2803 form will remain as supplied.
2807 my %where = ( is_ready => \"", completed => { '>', '2012-12-21' } )
2809 $stmt = "WHERE completed > ? AND is_ready"
2810 @bind = ('2012-12-21')
2812 Using an empty string literal used to be the only way to express a boolean.
2813 For all new code please use the much more readable
2814 L<-bool|/Unary operators: bool> operator.
2820 These pages could go on for a while, since the nesting of the data
2821 structures this module can handle are pretty much unlimited (the
2822 module implements the C<WHERE> expansion as a recursive function
2823 internally). Your best bet is to "play around" with the module a
2824 little to see how the data structures behave, and choose the best
2825 format for your data based on that.
2827 And of course, all the values above will probably be replaced with
2828 variables gotten from forms or the command line. After all, if you
2829 knew everything ahead of time, you wouldn't have to worry about
2830 dynamically-generating SQL and could just hardwire it into your
2833 =head1 ORDER BY CLAUSES
2835 Some functions take an order by clause. This can either be a scalar (just a
2836 column name,) a hash of C<< { -desc => 'col' } >> or C<< { -asc => 'col' } >>,
2837 or an array of either of the two previous forms. Examples:
2839 Given | Will Generate
2840 ----------------------------------------------------------
2842 \'colA DESC' | ORDER BY colA DESC
2844 'colA' | ORDER BY colA
2846 [qw/colA colB/] | ORDER BY colA, colB
2848 {-asc => 'colA'} | ORDER BY colA ASC
2850 {-desc => 'colB'} | ORDER BY colB DESC
2852 ['colA', {-asc => 'colB'}] | ORDER BY colA, colB ASC
2854 { -asc => [qw/colA colB/] } | ORDER BY colA ASC, colB ASC
2857 { -asc => 'colA' }, | ORDER BY colA ASC, colB DESC,
2858 { -desc => [qw/colB/], | colC ASC, colD ASC
2859 { -asc => [qw/colC colD/],|
2861 ===========================================================
2865 =head1 SPECIAL OPERATORS
2867 my $sqlmaker = SQL::Abstract->new(special_ops => [
2871 my ($self, $field, $op, $arg) = @_;
2877 handler => 'method_name',
2881 A "special operator" is a SQL syntactic clause that can be
2882 applied to a field, instead of a usual binary operator.
2885 WHERE field IN (?, ?, ?)
2886 WHERE field BETWEEN ? AND ?
2887 WHERE MATCH(field) AGAINST (?, ?)
2889 Special operators IN and BETWEEN are fairly standard and therefore
2890 are builtin within C<SQL::Abstract> (as the overridable methods
2891 C<_where_field_IN> and C<_where_field_BETWEEN>). For other operators,
2892 like the MATCH .. AGAINST example above which is specific to MySQL,
2893 you can write your own operator handlers - supply a C<special_ops>
2894 argument to the C<new> method. That argument takes an arrayref of
2895 operator definitions; each operator definition is a hashref with two
2902 the regular expression to match the operator
2906 Either a coderef or a plain scalar method name. In both cases
2907 the expected return is C<< ($sql, @bind) >>.
2909 When supplied with a method name, it is simply called on the
2910 L<SQL::Abstract/> object as:
2912 $self->$method_name ($field, $op, $arg)
2916 $op is the part that matched the handler regex
2917 $field is the LHS of the operator
2920 When supplied with a coderef, it is called as:
2922 $coderef->($self, $field, $op, $arg)
2927 For example, here is an implementation
2928 of the MATCH .. AGAINST syntax for MySQL
2930 my $sqlmaker = SQL::Abstract->new(special_ops => [
2932 # special op for MySql MATCH (field) AGAINST(word1, word2, ...)
2933 {regex => qr/^match$/i,
2935 my ($self, $field, $op, $arg) = @_;
2936 $arg = [$arg] if not ref $arg;
2937 my $label = $self->_quote($field);
2938 my ($placeholder) = $self->_convert('?');
2939 my $placeholders = join ", ", (($placeholder) x @$arg);
2940 my $sql = $self->_sqlcase('match') . " ($label) "
2941 . $self->_sqlcase('against') . " ($placeholders) ";
2942 my @bind = $self->_bindtype($field, @$arg);
2943 return ($sql, @bind);
2950 =head1 UNARY OPERATORS
2952 my $sqlmaker = SQL::Abstract->new(unary_ops => [
2956 my ($self, $op, $arg) = @_;
2962 handler => 'method_name',
2966 A "unary operator" is a SQL syntactic clause that can be
2967 applied to a field - the operator goes before the field
2969 You can write your own operator handlers - supply a C<unary_ops>
2970 argument to the C<new> method. That argument takes an arrayref of
2971 operator definitions; each operator definition is a hashref with two
2978 the regular expression to match the operator
2982 Either a coderef or a plain scalar method name. In both cases
2983 the expected return is C<< $sql >>.
2985 When supplied with a method name, it is simply called on the
2986 L<SQL::Abstract/> object as:
2988 $self->$method_name ($op, $arg)
2992 $op is the part that matched the handler regex
2993 $arg is the RHS or argument of the operator
2995 When supplied with a coderef, it is called as:
2997 $coderef->($self, $op, $arg)
3005 Thanks to some benchmarking by Mark Stosberg, it turns out that
3006 this module is many orders of magnitude faster than using C<DBIx::Abstract>.
3007 I must admit this wasn't an intentional design issue, but it's a
3008 byproduct of the fact that you get to control your C<DBI> handles
3011 To maximize performance, use a code snippet like the following:
3013 # prepare a statement handle using the first row
3014 # and then reuse it for the rest of the rows
3016 for my $href (@array_of_hashrefs) {
3017 $stmt ||= $sql->insert('table', $href);
3018 $sth ||= $dbh->prepare($stmt);
3019 $sth->execute($sql->values($href));
3022 The reason this works is because the keys in your C<$href> are sorted
3023 internally by B<SQL::Abstract>. Thus, as long as your data retains
3024 the same structure, you only have to generate the SQL the first time
3025 around. On subsequent queries, simply use the C<values> function provided
3026 by this module to return your values in the correct order.
3028 However this depends on the values having the same type - if, for
3029 example, the values of a where clause may either have values
3030 (resulting in sql of the form C<column = ?> with a single bind
3031 value), or alternatively the values might be C<undef> (resulting in
3032 sql of the form C<column IS NULL> with no bind value) then the
3033 caching technique suggested will not work.
3037 If you use my C<CGI::FormBuilder> module at all, you'll hopefully
3038 really like this part (I do, at least). Building up a complex query
3039 can be as simple as the following:
3046 use CGI::FormBuilder;
3049 my $form = CGI::FormBuilder->new(...);
3050 my $sql = SQL::Abstract->new;
3052 if ($form->submitted) {
3053 my $field = $form->field;
3054 my $id = delete $field->{id};
3055 my($stmt, @bind) = $sql->update('table', $field, {id => $id});
3058 Of course, you would still have to connect using C<DBI> to run the
3059 query, but the point is that if you make your form look like your
3060 table, the actual query script can be extremely simplistic.
3062 If you're B<REALLY> lazy (I am), check out C<HTML::QuickTable> for
3063 a fast interface to returning and formatting data. I frequently
3064 use these three modules together to write complex database query
3065 apps in under 50 lines.
3071 =item * gitweb: L<http://git.shadowcat.co.uk/gitweb/gitweb.cgi?p=dbsrgits/SQL-Abstract.git>
3073 =item * git: L<git://git.shadowcat.co.uk/dbsrgits/SQL-Abstract.git>
3079 Version 1.50 was a major internal refactoring of C<SQL::Abstract>.
3080 Great care has been taken to preserve the I<published> behavior
3081 documented in previous versions in the 1.* family; however,
3082 some features that were previously undocumented, or behaved
3083 differently from the documentation, had to be changed in order
3084 to clarify the semantics. Hence, client code that was relying
3085 on some dark areas of C<SQL::Abstract> v1.*
3086 B<might behave differently> in v1.50.
3088 The main changes are :
3094 support for literal SQL through the C<< \ [ $sql, @bind ] >> syntax.
3098 support for the { operator => \"..." } construct (to embed literal SQL)
3102 support for the { operator => \["...", @bind] } construct (to embed literal SQL with bind values)
3106 optional support for L<array datatypes|/"Inserting and Updating Arrays">
3110 defensive programming : check arguments
3114 fixed bug with global logic, which was previously implemented
3115 through global variables yielding side-effects. Prior versions would
3116 interpret C<< [ {cond1, cond2}, [cond3, cond4] ] >>
3117 as C<< "(cond1 AND cond2) OR (cond3 AND cond4)" >>.
3118 Now this is interpreted
3119 as C<< "(cond1 AND cond2) OR (cond3 OR cond4)" >>.
3124 fixed semantics of _bindtype on array args
3128 dropped the C<_anoncopy> of the %where tree. No longer necessary,
3129 we just avoid shifting arrays within that tree.
3133 dropped the C<_modlogic> function
3137 =head1 ACKNOWLEDGEMENTS
3139 There are a number of individuals that have really helped out with
3140 this module. Unfortunately, most of them submitted bugs via CPAN
3141 so I have no idea who they are! But the people I do know are:
3143 Ash Berlin (order_by hash term support)
3144 Matt Trout (DBIx::Class support)
3145 Mark Stosberg (benchmarking)
3146 Chas Owens (initial "IN" operator support)
3147 Philip Collins (per-field SQL functions)
3148 Eric Kolve (hashref "AND" support)
3149 Mike Fragassi (enhancements to "BETWEEN" and "LIKE")
3150 Dan Kubb (support for "quote_char" and "name_sep")
3151 Guillermo Roditi (patch to cleanup "IN" and "BETWEEN", fix and tests for _order_by)
3152 Laurent Dami (internal refactoring, extensible list of special operators, literal SQL)
3153 Norbert Buchmuller (support for literal SQL in hashpair, misc. fixes & tests)
3154 Peter Rabbitson (rewrite of SQLA::Test, misc. fixes & tests)
3155 Oliver Charles (support for "RETURNING" after "INSERT")
3161 L<DBIx::Class>, L<DBIx::Abstract>, L<CGI::FormBuilder>, L<HTML::QuickTable>.
3165 Copyright (c) 2001-2007 Nathan Wiger <nwiger@cpan.org>. All Rights Reserved.
3167 This module is actively maintained by Matt Trout <mst@shadowcatsystems.co.uk>
3169 For support, your best bet is to try the C<DBIx::Class> users mailing list.
3170 While not an official support venue, C<DBIx::Class> makes heavy use of
3171 C<SQL::Abstract>, and as such list members there are very familiar with
3172 how to create queries.
3176 This module is free software; you may copy this under the same
3177 terms as perl itself (either the GNU General Public License or
3178 the Artistic License)