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, $logic);
500 # DBIx::Class used to call _recurse_where in scalar context
501 # something else might too...
503 return ($sql, @bind);
506 belch "Calling _recurse_where in scalar context is deprecated and will go away before 2.0";
513 #======================================================================
514 # WHERE: top-level ARRAYREF
515 #======================================================================
518 sub _where_ARRAYREF {
519 my ($self, $where, $logic) = @_;
521 $logic = uc($logic || $self->{logic});
522 $logic eq 'AND' or $logic eq 'OR' or puke "unknown logic: $logic";
524 my @clauses = @$where;
526 my (@sql_clauses, @all_bind);
527 # need to use while() so can shift() for pairs
528 while (my $el = shift @clauses) {
530 # switch according to kind of $el and get corresponding ($sql, @bind)
531 my ($sql, @bind) = $self->_SWITCH_refkind($el, {
533 # skip empty elements, otherwise get invalid trailing AND stuff
534 ARRAYREF => sub {$self->_recurse_where($el) if @$el},
538 $self->_assert_bindval_matches_bindtype(@b);
542 HASHREF => sub {$self->_recurse_where($el, 'and') if %$el},
544 SCALARREF => sub { ($$el); },
546 SCALAR => sub {# top-level arrayref with scalars, recurse in pairs
547 $self->_recurse_where({$el => shift(@clauses)})},
549 UNDEF => sub {puke "not supported : UNDEF in arrayref" },
553 push @sql_clauses, $sql;
554 push @all_bind, @bind;
558 return $self->_join_sql_clauses($logic, \@sql_clauses, \@all_bind);
561 #======================================================================
562 # WHERE: top-level ARRAYREFREF
563 #======================================================================
565 sub _where_ARRAYREFREF {
566 my ($self, $where) = @_;
567 my ($sql, @bind) = @$$where;
568 $self->_assert_bindval_matches_bindtype(@bind);
569 return ($sql, @bind);
572 #======================================================================
573 # WHERE: top-level HASHREF
574 #======================================================================
577 my ($self, $where) = @_;
578 my (@sql_clauses, @all_bind);
580 for my $k (sort keys %$where) {
581 my $v = $where->{$k};
583 # ($k => $v) is either a special unary op or a regular hashpair
584 my ($sql, @bind) = do {
586 # put the operator in canonical form
588 $op = substr $op, 1; # remove initial dash
589 $op =~ s/^\s+|\s+$//g;# remove leading/trailing space
590 $op =~ s/\s+/ /g; # compress whitespace
592 # so that -not_foo works correctly
593 $op =~ s/^not_/NOT /i;
595 $self->_debug("Unary OP(-$op) within hashref, recursing...");
596 my ($s, @b) = $self->_where_unary_op ($op, $v);
598 # top level vs nested
599 # we assume that handled unary ops will take care of their ()s
601 List::Util::first {$op =~ $_->{regex}} @{$self->{unary_ops}}
603 defined($self->{_nested_func_lhs}) && ($self->{_nested_func_lhs} eq $k)
608 my $method = $self->_METHOD_FOR_refkind("_where_hashpair", $v);
609 $self->$method($k, $v);
613 push @sql_clauses, $sql;
614 push @all_bind, @bind;
617 return $self->_join_sql_clauses('and', \@sql_clauses, \@all_bind);
620 sub _where_unary_op {
621 my ($self, $op, $rhs) = @_;
623 if (my $op_entry = List::Util::first {$op =~ $_->{regex}} @{$self->{unary_ops}}) {
624 my $handler = $op_entry->{handler};
626 if (not ref $handler) {
627 if ($op =~ s/ [_\s]? \d+ $//x ) {
628 belch 'Use of [and|or|nest]_N modifiers is deprecated and will be removed in SQLA v2.0. '
629 . "You probably wanted ...-and => [ -$op => COND1, -$op => COND2 ... ]";
631 return $self->$handler ($op, $rhs);
633 elsif (ref $handler eq 'CODE') {
634 return $handler->($self, $op, $rhs);
637 puke "Illegal handler for operator $op - expecting a method name or a coderef";
641 $self->_debug("Generic unary OP: $op - recursing as function");
643 $self->_assert_pass_injection_guard($op);
645 my ($sql, @bind) = $self->_SWITCH_refkind ($rhs, {
647 puke "Illegal use of top-level '$op'"
648 unless $self->{_nested_func_lhs};
651 $self->_convert('?'),
652 $self->_bindtype($self->{_nested_func_lhs}, $rhs)
656 $self->_recurse_where ($rhs)
660 $sql = sprintf ('%s %s',
661 $self->_sqlcase($op),
665 return ($sql, @bind);
668 sub _where_op_ANDOR {
669 my ($self, $op, $v) = @_;
671 $self->_SWITCH_refkind($v, {
673 return $self->_where_ARRAYREF($v, $op);
677 return ( $op =~ /^or/i )
678 ? $self->_where_ARRAYREF( [ map { $_ => $v->{$_} } ( sort keys %$v ) ], $op )
679 : $self->_where_HASHREF($v);
683 puke "-$op => \\\$scalar makes little sense, use " .
685 ? '[ \$scalar, \%rest_of_conditions ] instead'
686 : '-and => [ \$scalar, \%rest_of_conditions ] instead'
691 puke "-$op => \\[...] makes little sense, use " .
693 ? '[ \[...], \%rest_of_conditions ] instead'
694 : '-and => [ \[...], \%rest_of_conditions ] instead'
698 SCALAR => sub { # permissively interpreted as SQL
699 puke "-$op => \$value makes little sense, use -bool => \$value instead";
703 puke "-$op => undef not supported";
709 my ($self, $op, $v) = @_;
711 $self->_SWITCH_refkind($v, {
713 SCALAR => sub { # permissively interpreted as SQL
714 belch "literal SQL should be -nest => \\'scalar' "
715 . "instead of -nest => 'scalar' ";
720 puke "-$op => undef not supported";
724 $self->_recurse_where ($v);
732 my ($self, $op, $v) = @_;
734 my ($s, @b) = $self->_SWITCH_refkind($v, {
735 SCALAR => sub { # interpreted as SQL column
736 $self->_convert($self->_quote($v));
740 puke "-$op => undef not supported";
744 $self->_recurse_where ($v);
748 $s = "(NOT $s)" if $op =~ /^not/i;
753 sub _where_op_IDENT {
755 my ($op, $rhs) = splice @_, -2;
756 if (! defined $rhs or length ref $rhs) {
757 puke "-$op requires a single plain scalar argument (a quotable identifier)";
760 # in case we are called as a top level special op (no '=')
763 $_ = $self->_convert($self->_quote($_)) for ($lhs, $rhs);
771 sub _where_op_VALUE {
773 my ($op, $rhs) = splice @_, -2;
775 # in case we are called as a top level special op (no '=')
779 if (! defined $rhs) {
781 ? $self->_convert($self->_quote($lhs)) . ' IS NULL'
788 ($lhs || $self->{_nested_func_lhs}),
795 $self->_convert($self->_quote($lhs)) . ' = ' . $self->_convert('?'),
799 $self->_convert('?'),
805 sub _where_hashpair_ARRAYREF {
806 my ($self, $k, $v) = @_;
809 my @v = @$v; # need copy because of shift below
810 $self->_debug("ARRAY($k) means distribute over elements");
812 # put apart first element if it is an operator (-and, -or)
814 (defined $v[0] && $v[0] =~ /^ - (?: AND|OR ) $/ix)
818 my @distributed = map { {$k => $_} } @v;
821 $self->_debug("OP($op) reinjected into the distributed array");
822 unshift @distributed, $op;
825 my $logic = $op ? substr($op, 1) : '';
827 return $self->_recurse_where(\@distributed, $logic);
830 $self->_debug("empty ARRAY($k) means 0=1");
831 return ($self->{sqlfalse});
835 sub _where_hashpair_HASHREF {
836 my ($self, $k, $v, $logic) = @_;
839 local $self->{_nested_func_lhs} = $self->{_nested_func_lhs};
841 my ($all_sql, @all_bind);
843 for my $orig_op (sort keys %$v) {
844 my $val = $v->{$orig_op};
846 # put the operator in canonical form
849 # FIXME - we need to phase out dash-less ops
850 $op =~ s/^-//; # remove possible initial dash
851 $op =~ s/^\s+|\s+$//g;# remove leading/trailing space
852 $op =~ s/\s+/ /g; # compress whitespace
854 $self->_assert_pass_injection_guard($op);
857 $op =~ s/^is_not/IS NOT/i;
859 # so that -not_foo works correctly
860 $op =~ s/^not_/NOT /i;
862 # another retarded special case: foo => { $op => { -value => undef } }
863 if (ref $val eq 'HASH' and keys %$val == 1 and exists $val->{-value} and ! defined $val->{-value} ) {
869 # CASE: col-value logic modifiers
870 if ( $orig_op =~ /^ \- (and|or) $/xi ) {
871 ($sql, @bind) = $self->_where_hashpair_HASHREF($k, $val, $1);
873 # CASE: special operators like -in or -between
874 elsif ( my $special_op = List::Util::first {$op =~ $_->{regex}} @{$self->{special_ops}} ) {
875 my $handler = $special_op->{handler};
877 puke "No handler supplied for special operator $orig_op";
879 elsif (not ref $handler) {
880 ($sql, @bind) = $self->$handler ($k, $op, $val);
882 elsif (ref $handler eq 'CODE') {
883 ($sql, @bind) = $handler->($self, $k, $op, $val);
886 puke "Illegal handler for special operator $orig_op - expecting a method name or a coderef";
890 $self->_SWITCH_refkind($val, {
892 ARRAYREF => sub { # CASE: col => {op => \@vals}
893 ($sql, @bind) = $self->_where_field_op_ARRAYREF($k, $op, $val);
896 ARRAYREFREF => sub { # CASE: col => {op => \[$sql, @bind]} (literal SQL with bind)
897 my ($sub_sql, @sub_bind) = @$$val;
898 $self->_assert_bindval_matches_bindtype(@sub_bind);
899 $sql = join ' ', $self->_convert($self->_quote($k)),
900 $self->_sqlcase($op),
905 UNDEF => sub { # CASE: col => {op => undef} : sql "IS (NOT)? NULL"
907 $op =~ /^not$/i ? 'is not' # legacy
908 : $op =~ $self->{equality_op} ? 'is'
909 : $op =~ $self->{like_op} ? belch("Supplying an undefined argument to '@{[ uc $op]}' is deprecated") && 'is'
910 : $op =~ $self->{inequality_op} ? 'is not'
911 : $op =~ $self->{not_like_op} ? belch("Supplying an undefined argument to '@{[ uc $op]}' is deprecated") && 'is not'
912 : puke "unexpected operator '$orig_op' with undef operand";
914 $sql = $self->_quote($k) . $self->_sqlcase(" $is null");
917 FALLBACK => sub { # CASE: col => {op/func => $stuff}
919 # retain for proper column type bind
920 $self->{_nested_func_lhs} ||= $k;
922 ($sql, @bind) = $self->_where_unary_op ($op, $val);
925 $self->_convert($self->_quote($k)),
926 $self->{_nested_func_lhs} eq $k ? $sql : "($sql)", # top level vs nested
932 ($all_sql) = (defined $all_sql and $all_sql) ? $self->_join_sql_clauses($logic, [$all_sql, $sql], []) : $sql;
933 push @all_bind, @bind;
935 return ($all_sql, @all_bind);
938 sub _where_field_IS {
939 my ($self, $k, $op, $v) = @_;
941 my ($s) = $self->_SWITCH_refkind($v, {
944 $self->_convert($self->_quote($k)),
945 map { $self->_sqlcase($_)} ($op, 'null')
948 puke "$op can only take undef as argument";
955 sub _where_field_op_ARRAYREF {
956 my ($self, $k, $op, $vals) = @_;
958 my @vals = @$vals; #always work on a copy
961 $self->_debug(sprintf '%s means multiple elements: [ %s ]',
963 join (', ', map { defined $_ ? "'$_'" : 'NULL' } @vals ),
966 # see if the first element is an -and/-or op
968 if (defined $vals[0] && $vals[0] =~ /^ - ( AND|OR ) $/ix) {
973 # a long standing API wart - an attempt to change this behavior during
974 # the 1.50 series failed *spectacularly*. Warn instead and leave the
979 (!$logic or $logic eq 'OR')
981 ( $op =~ $self->{inequality_op} or $op =~ $self->{not_like_op} )
984 belch "A multi-element arrayref as an argument to the inequality op '$o' "
985 . 'is technically equivalent to an always-true 1=1 (you probably wanted '
986 . "to say ...{ \$inequality_op => [ -and => \@values ] }... instead)"
990 # distribute $op over each remaining member of @vals, append logic if exists
991 return $self->_recurse_where([map { {$k => {$op, $_}} } @vals], $logic);
995 # try to DWIM on equality operators
997 $op =~ $self->{equality_op} ? $self->{sqlfalse}
998 : $op =~ $self->{like_op} ? belch("Supplying an empty arrayref to '@{[ uc $op]}' is deprecated") && $self->{sqlfalse}
999 : $op =~ $self->{inequality_op} ? $self->{sqltrue}
1000 : $op =~ $self->{not_like_op} ? belch("Supplying an empty arrayref to '@{[ uc $op]}' is deprecated") && $self->{sqltrue}
1001 : puke "operator '$op' applied on an empty array (field '$k')";
1006 sub _where_hashpair_SCALARREF {
1007 my ($self, $k, $v) = @_;
1008 $self->_debug("SCALAR($k) means literal SQL: $$v");
1009 my $sql = $self->_quote($k) . " " . $$v;
1013 # literal SQL with bind
1014 sub _where_hashpair_ARRAYREFREF {
1015 my ($self, $k, $v) = @_;
1016 $self->_debug("REF($k) means literal SQL: @${$v}");
1017 my ($sql, @bind) = @$$v;
1018 $self->_assert_bindval_matches_bindtype(@bind);
1019 $sql = $self->_quote($k) . " " . $sql;
1020 return ($sql, @bind );
1023 # literal SQL without bind
1024 sub _where_hashpair_SCALAR {
1025 my ($self, $k, $v) = @_;
1026 $self->_debug("NOREF($k) means simple key=val: $k $self->{cmp} $v");
1027 my $sql = join ' ', $self->_convert($self->_quote($k)),
1028 $self->_sqlcase($self->{cmp}),
1029 $self->_convert('?');
1030 my @bind = $self->_bindtype($k, $v);
1031 return ( $sql, @bind);
1035 sub _where_hashpair_UNDEF {
1036 my ($self, $k, $v) = @_;
1037 $self->_debug("UNDEF($k) means IS NULL");
1038 my $sql = $self->_quote($k) . $self->_sqlcase(' is null');
1042 #======================================================================
1043 # WHERE: TOP-LEVEL OTHERS (SCALARREF, SCALAR, UNDEF)
1044 #======================================================================
1047 sub _where_SCALARREF {
1048 my ($self, $where) = @_;
1051 $self->_debug("SCALAR(*top) means literal SQL: $$where");
1057 my ($self, $where) = @_;
1060 $self->_debug("NOREF(*top) means literal SQL: $where");
1071 #======================================================================
1072 # WHERE: BUILTIN SPECIAL OPERATORS (-in, -between)
1073 #======================================================================
1076 sub _where_field_BETWEEN {
1077 my ($self, $k, $op, $vals) = @_;
1079 my ($label, $and, $placeholder);
1080 $label = $self->_convert($self->_quote($k));
1081 $and = ' ' . $self->_sqlcase('and') . ' ';
1082 $placeholder = $self->_convert('?');
1083 $op = $self->_sqlcase($op);
1085 my $invalid_args = "Operator '$op' requires either an arrayref with two defined values or expressions, or a single literal scalarref/arrayref-ref";
1087 my ($clause, @bind) = $self->_SWITCH_refkind($vals, {
1088 ARRAYREFREF => sub {
1089 my ($s, @b) = @$$vals;
1090 $self->_assert_bindval_matches_bindtype(@b);
1097 puke $invalid_args if @$vals != 2;
1099 my (@all_sql, @all_bind);
1100 foreach my $val (@$vals) {
1101 my ($sql, @bind) = $self->_SWITCH_refkind($val, {
1103 return ($placeholder, $self->_bindtype($k, $val) );
1108 ARRAYREFREF => sub {
1109 my ($sql, @bind) = @$$val;
1110 $self->_assert_bindval_matches_bindtype(@bind);
1111 return ($sql, @bind);
1114 my ($func, $arg, @rest) = %$val;
1115 puke ("Only simple { -func => arg } functions accepted as sub-arguments to BETWEEN")
1116 if (@rest or $func !~ /^ \- (.+)/x);
1117 local $self->{_nested_func_lhs} = $k;
1118 $self->_where_unary_op ($1 => $arg);
1124 push @all_sql, $sql;
1125 push @all_bind, @bind;
1129 (join $and, @all_sql),
1138 my $sql = "( $label $op $clause )";
1139 return ($sql, @bind)
1143 sub _where_field_IN {
1144 my ($self, $k, $op, $vals) = @_;
1146 # backwards compatibility : if scalar, force into an arrayref
1147 $vals = [$vals] if defined $vals && ! ref $vals;
1149 my ($label) = $self->_convert($self->_quote($k));
1150 my ($placeholder) = $self->_convert('?');
1151 $op = $self->_sqlcase($op);
1153 my ($sql, @bind) = $self->_SWITCH_refkind($vals, {
1154 ARRAYREF => sub { # list of choices
1155 if (@$vals) { # nonempty list
1156 my (@all_sql, @all_bind);
1158 for my $val (@$vals) {
1159 my ($sql, @bind) = $self->_SWITCH_refkind($val, {
1161 return ($placeholder, $val);
1166 ARRAYREFREF => sub {
1167 my ($sql, @bind) = @$$val;
1168 $self->_assert_bindval_matches_bindtype(@bind);
1169 return ($sql, @bind);
1172 my ($func, $arg, @rest) = %$val;
1173 puke ("Only simple { -func => arg } functions accepted as sub-arguments to IN")
1174 if (@rest or $func !~ /^ \- (.+)/x);
1175 local $self->{_nested_func_lhs} = $k;
1176 $self->_where_unary_op ($1 => $arg);
1180 'SQL::Abstract before v1.75 used to generate incorrect SQL when the '
1181 . "-$op operator was given an undef-containing list: !!!AUDIT YOUR CODE "
1182 . 'AND DATA!!! (the upcoming Data::Query-based version of SQL::Abstract '
1183 . 'will emit the logically correct SQL instead of raising this exception)'
1187 push @all_sql, $sql;
1188 push @all_bind, @bind;
1192 sprintf ('%s %s ( %s )',
1195 join (', ', @all_sql)
1197 $self->_bindtype($k, @all_bind),
1200 else { # empty list : some databases won't understand "IN ()", so DWIM
1201 my $sql = ($op =~ /\bnot\b/i) ? $self->{sqltrue} : $self->{sqlfalse};
1206 SCALARREF => sub { # literal SQL
1207 my $sql = $self->_open_outer_paren ($$vals);
1208 return ("$label $op ( $sql )");
1210 ARRAYREFREF => sub { # literal SQL with bind
1211 my ($sql, @bind) = @$$vals;
1212 $self->_assert_bindval_matches_bindtype(@bind);
1213 $sql = $self->_open_outer_paren ($sql);
1214 return ("$label $op ( $sql )", @bind);
1218 puke "Argument passed to the '$op' operator can not be undefined";
1222 puke "special op $op requires an arrayref (or scalarref/arrayref-ref)";
1226 return ($sql, @bind);
1229 # Some databases (SQLite) treat col IN (1, 2) different from
1230 # col IN ( (1, 2) ). Use this to strip all outer parens while
1231 # adding them back in the corresponding method
1232 sub _open_outer_paren {
1233 my ($self, $sql) = @_;
1234 $sql = $1 while $sql =~ /^ \s* \( (.*) \) \s* $/xs;
1239 #======================================================================
1241 #======================================================================
1244 my ($self, $arg) = @_;
1247 for my $c ($self->_order_by_chunks ($arg) ) {
1248 $self->_SWITCH_refkind ($c, {
1249 SCALAR => sub { push @sql, $c },
1250 ARRAYREF => sub { push @sql, shift @$c; push @bind, @$c },
1256 $self->_sqlcase(' order by'),
1262 return wantarray ? ($sql, @bind) : $sql;
1265 sub _order_by_chunks {
1266 my ($self, $arg) = @_;
1268 return $self->_SWITCH_refkind($arg, {
1271 map { $self->_order_by_chunks ($_ ) } @$arg;
1274 ARRAYREFREF => sub {
1275 my ($s, @b) = @$$arg;
1276 $self->_assert_bindval_matches_bindtype(@b);
1280 SCALAR => sub {$self->_quote($arg)},
1282 UNDEF => sub {return () },
1284 SCALARREF => sub {$$arg}, # literal SQL, no quoting
1287 # get first pair in hash
1288 my ($key, $val, @rest) = %$arg;
1290 return () unless $key;
1292 if ( @rest or not $key =~ /^-(desc|asc)/i ) {
1293 puke "hash passed to _order_by must have exactly one key (-desc or -asc)";
1299 for my $c ($self->_order_by_chunks ($val)) {
1302 $self->_SWITCH_refkind ($c, {
1307 ($sql, @bind) = @$c;
1311 $sql = $sql . ' ' . $self->_sqlcase($direction);
1313 push @ret, [ $sql, @bind];
1322 #======================================================================
1323 # DATASOURCE (FOR NOW, JUST PLAIN TABLE OR LIST OF TABLES)
1324 #======================================================================
1329 $self->_SWITCH_refkind($from, {
1330 ARRAYREF => sub {join ', ', map { $self->_quote($_) } @$from;},
1331 SCALAR => sub {$self->_quote($from)},
1332 SCALARREF => sub {$$from},
1337 #======================================================================
1339 #======================================================================
1341 # highly optimized, as it's called way too often
1343 # my ($self, $label) = @_;
1345 return '' unless defined $_[1];
1346 return ${$_[1]} if ref($_[1]) eq 'SCALAR';
1348 unless ($_[0]->{quote_char}) {
1349 $_[0]->_assert_pass_injection_guard($_[1]);
1353 my $qref = ref $_[0]->{quote_char};
1356 ($l, $r) = ( $_[0]->{quote_char}, $_[0]->{quote_char} );
1358 elsif ($qref eq 'ARRAY') {
1359 ($l, $r) = @{$_[0]->{quote_char}};
1362 puke "Unsupported quote_char format: $_[0]->{quote_char}";
1364 my $esc = $_[0]->{escape_char} || $r;
1366 # parts containing * are naturally unquoted
1367 return join( $_[0]->{name_sep}||'', map
1368 { $_ eq '*' ? $_ : do { (my $n = $_) =~ s/(\Q$esc\E|\Q$r\E)/$esc$1/g; $l . $n . $r } }
1369 ( $_[0]->{name_sep} ? split (/\Q$_[0]->{name_sep}\E/, $_[1] ) : $_[1] )
1374 # Conversion, if applicable
1376 #my ($self, $arg) = @_;
1377 if ($_[0]->{convert}) {
1378 return $_[0]->_sqlcase($_[0]->{convert}) .'(' . $_[1] . ')';
1385 #my ($self, $col, @vals) = @_;
1386 # called often - tighten code
1387 return $_[0]->{bindtype} eq 'columns'
1388 ? map {[$_[1], $_]} @_[2 .. $#_]
1393 # Dies if any element of @bind is not in [colname => value] format
1394 # if bindtype is 'columns'.
1395 sub _assert_bindval_matches_bindtype {
1396 # my ($self, @bind) = @_;
1398 if ($self->{bindtype} eq 'columns') {
1400 if (!defined $_ || ref($_) ne 'ARRAY' || @$_ != 2) {
1401 puke "bindtype 'columns' selected, you need to pass: [column_name => bind_value]"
1407 sub _join_sql_clauses {
1408 my ($self, $logic, $clauses_aref, $bind_aref) = @_;
1410 if (@$clauses_aref > 1) {
1411 my $join = " " . $self->_sqlcase($logic) . " ";
1412 my $sql = '( ' . join($join, @$clauses_aref) . ' )';
1413 return ($sql, @$bind_aref);
1415 elsif (@$clauses_aref) {
1416 return ($clauses_aref->[0], @$bind_aref); # no parentheses
1419 return (); # if no SQL, ignore @$bind_aref
1424 # Fix SQL case, if so requested
1426 # LDNOTE: if $self->{case} is true, then it contains 'lower', so we
1427 # don't touch the argument ... crooked logic, but let's not change it!
1428 return $_[0]->{case} ? $_[1] : uc($_[1]);
1432 #======================================================================
1433 # DISPATCHING FROM REFKIND
1434 #======================================================================
1437 my ($self, $data) = @_;
1439 return 'UNDEF' unless defined $data;
1441 # blessed objects are treated like scalars
1442 my $ref = (Scalar::Util::blessed $data) ? '' : ref $data;
1444 return 'SCALAR' unless $ref;
1447 while ($ref eq 'REF') {
1449 $ref = (Scalar::Util::blessed $data) ? '' : ref $data;
1453 return ($ref||'SCALAR') . ('REF' x $n_steps);
1457 my ($self, $data) = @_;
1458 my @try = ($self->_refkind($data));
1459 push @try, 'SCALAR_or_UNDEF' if $try[0] eq 'SCALAR' || $try[0] eq 'UNDEF';
1460 push @try, 'FALLBACK';
1464 sub _METHOD_FOR_refkind {
1465 my ($self, $meth_prefix, $data) = @_;
1468 for (@{$self->_try_refkind($data)}) {
1469 $method = $self->can($meth_prefix."_".$_)
1473 return $method || puke "cannot dispatch on '$meth_prefix' for ".$self->_refkind($data);
1477 sub _SWITCH_refkind {
1478 my ($self, $data, $dispatch_table) = @_;
1481 for (@{$self->_try_refkind($data)}) {
1482 $coderef = $dispatch_table->{$_}
1486 puke "no dispatch entry for ".$self->_refkind($data)
1495 #======================================================================
1496 # VALUES, GENERATE, AUTOLOAD
1497 #======================================================================
1499 # LDNOTE: original code from nwiger, didn't touch code in that section
1500 # I feel the AUTOLOAD stuff should not be the default, it should
1501 # only be activated on explicit demand by user.
1505 my $data = shift || return;
1506 puke "Argument to ", __PACKAGE__, "->values must be a \\%hash"
1507 unless ref $data eq 'HASH';
1510 foreach my $k ( sort keys %$data ) {
1511 my $v = $data->{$k};
1512 $self->_SWITCH_refkind($v, {
1514 if ($self->{array_datatypes}) { # array datatype
1515 push @all_bind, $self->_bindtype($k, $v);
1517 else { # literal SQL with bind
1518 my ($sql, @bind) = @$v;
1519 $self->_assert_bindval_matches_bindtype(@bind);
1520 push @all_bind, @bind;
1523 ARRAYREFREF => sub { # literal SQL with bind
1524 my ($sql, @bind) = @${$v};
1525 $self->_assert_bindval_matches_bindtype(@bind);
1526 push @all_bind, @bind;
1528 SCALARREF => sub { # literal SQL without bind
1530 SCALAR_or_UNDEF => sub {
1531 push @all_bind, $self->_bindtype($k, $v);
1542 my(@sql, @sqlq, @sqlv);
1546 if ($ref eq 'HASH') {
1547 for my $k (sort keys %$_) {
1550 my $label = $self->_quote($k);
1551 if ($r eq 'ARRAY') {
1552 # literal SQL with bind
1553 my ($sql, @bind) = @$v;
1554 $self->_assert_bindval_matches_bindtype(@bind);
1555 push @sqlq, "$label = $sql";
1557 } elsif ($r eq 'SCALAR') {
1558 # literal SQL without bind
1559 push @sqlq, "$label = $$v";
1561 push @sqlq, "$label = ?";
1562 push @sqlv, $self->_bindtype($k, $v);
1565 push @sql, $self->_sqlcase('set'), join ', ', @sqlq;
1566 } elsif ($ref eq 'ARRAY') {
1567 # unlike insert(), assume these are ONLY the column names, i.e. for SQL
1570 if ($r eq 'ARRAY') { # literal SQL with bind
1571 my ($sql, @bind) = @$v;
1572 $self->_assert_bindval_matches_bindtype(@bind);
1575 } elsif ($r eq 'SCALAR') { # literal SQL without bind
1576 # embedded literal SQL
1583 push @sql, '(' . join(', ', @sqlq) . ')';
1584 } elsif ($ref eq 'SCALAR') {
1588 # strings get case twiddled
1589 push @sql, $self->_sqlcase($_);
1593 my $sql = join ' ', @sql;
1595 # this is pretty tricky
1596 # if ask for an array, return ($stmt, @bind)
1597 # otherwise, s/?/shift @sqlv/ to put it inline
1599 return ($sql, @sqlv);
1601 1 while $sql =~ s/\?/my $d = shift(@sqlv);
1602 ref $d ? $d->[1] : $d/e;
1611 # This allows us to check for a local, then _form, attr
1613 my($name) = $AUTOLOAD =~ /.*::(.+)/;
1614 return $self->generate($name, @_);
1625 SQL::Abstract - Generate SQL from Perl data structures
1631 my $sql = SQL::Abstract->new;
1633 my($stmt, @bind) = $sql->select($source, \@fields, \%where, \@order);
1635 my($stmt, @bind) = $sql->insert($table, \%fieldvals || \@values);
1637 my($stmt, @bind) = $sql->update($table, \%fieldvals, \%where);
1639 my($stmt, @bind) = $sql->delete($table, \%where);
1641 # Then, use these in your DBI statements
1642 my $sth = $dbh->prepare($stmt);
1643 $sth->execute(@bind);
1645 # Just generate the WHERE clause
1646 my($stmt, @bind) = $sql->where(\%where, \@order);
1648 # Return values in the same order, for hashed queries
1649 # See PERFORMANCE section for more details
1650 my @bind = $sql->values(\%fieldvals);
1654 This module was inspired by the excellent L<DBIx::Abstract>.
1655 However, in using that module I found that what I really wanted
1656 to do was generate SQL, but still retain complete control over my
1657 statement handles and use the DBI interface. So, I set out to
1658 create an abstract SQL generation module.
1660 While based on the concepts used by L<DBIx::Abstract>, there are
1661 several important differences, especially when it comes to WHERE
1662 clauses. I have modified the concepts used to make the SQL easier
1663 to generate from Perl data structures and, IMO, more intuitive.
1664 The underlying idea is for this module to do what you mean, based
1665 on the data structures you provide it. The big advantage is that
1666 you don't have to modify your code every time your data changes,
1667 as this module figures it out.
1669 To begin with, an SQL INSERT is as easy as just specifying a hash
1670 of C<key=value> pairs:
1673 name => 'Jimbo Bobson',
1674 phone => '123-456-7890',
1675 address => '42 Sister Lane',
1676 city => 'St. Louis',
1677 state => 'Louisiana',
1680 The SQL can then be generated with this:
1682 my($stmt, @bind) = $sql->insert('people', \%data);
1684 Which would give you something like this:
1686 $stmt = "INSERT INTO people
1687 (address, city, name, phone, state)
1688 VALUES (?, ?, ?, ?, ?)";
1689 @bind = ('42 Sister Lane', 'St. Louis', 'Jimbo Bobson',
1690 '123-456-7890', 'Louisiana');
1692 These are then used directly in your DBI code:
1694 my $sth = $dbh->prepare($stmt);
1695 $sth->execute(@bind);
1697 =head2 Inserting and Updating Arrays
1699 If your database has array types (like for example Postgres),
1700 activate the special option C<< array_datatypes => 1 >>
1701 when creating the C<SQL::Abstract> object.
1702 Then you may use an arrayref to insert and update database array types:
1704 my $sql = SQL::Abstract->new(array_datatypes => 1);
1706 planets => [qw/Mercury Venus Earth Mars/]
1709 my($stmt, @bind) = $sql->insert('solar_system', \%data);
1713 $stmt = "INSERT INTO solar_system (planets) VALUES (?)"
1715 @bind = (['Mercury', 'Venus', 'Earth', 'Mars']);
1718 =head2 Inserting and Updating SQL
1720 In order to apply SQL functions to elements of your C<%data> you may
1721 specify a reference to an arrayref for the given hash value. For example,
1722 if you need to execute the Oracle C<to_date> function on a value, you can
1723 say something like this:
1727 date_entered => \[ "to_date(?,'MM/DD/YYYY')", "03/02/2003" ],
1730 The first value in the array is the actual SQL. Any other values are
1731 optional and would be included in the bind values array. This gives
1734 my($stmt, @bind) = $sql->insert('people', \%data);
1736 $stmt = "INSERT INTO people (name, date_entered)
1737 VALUES (?, to_date(?,'MM/DD/YYYY'))";
1738 @bind = ('Bill', '03/02/2003');
1740 An UPDATE is just as easy, all you change is the name of the function:
1742 my($stmt, @bind) = $sql->update('people', \%data);
1744 Notice that your C<%data> isn't touched; the module will generate
1745 the appropriately quirky SQL for you automatically. Usually you'll
1746 want to specify a WHERE clause for your UPDATE, though, which is
1747 where handling C<%where> hashes comes in handy...
1749 =head2 Complex where statements
1751 This module can generate pretty complicated WHERE statements
1752 easily. For example, simple C<key=value> pairs are taken to mean
1753 equality, and if you want to see if a field is within a set
1754 of values, you can use an arrayref. Let's say we wanted to
1755 SELECT some data based on this criteria:
1758 requestor => 'inna',
1759 worker => ['nwiger', 'rcwe', 'sfz'],
1760 status => { '!=', 'completed' }
1763 my($stmt, @bind) = $sql->select('tickets', '*', \%where);
1765 The above would give you something like this:
1767 $stmt = "SELECT * FROM tickets WHERE
1768 ( requestor = ? ) AND ( status != ? )
1769 AND ( worker = ? OR worker = ? OR worker = ? )";
1770 @bind = ('inna', 'completed', 'nwiger', 'rcwe', 'sfz');
1772 Which you could then use in DBI code like so:
1774 my $sth = $dbh->prepare($stmt);
1775 $sth->execute(@bind);
1781 The methods are simple. There's one for every major SQL operation,
1782 and a constructor you use first. The arguments are specified in a
1783 similar order for each method (table, then fields, then a where
1784 clause) to try and simplify things.
1786 =head2 new(option => 'value')
1788 The C<new()> function takes a list of options and values, and returns
1789 a new B<SQL::Abstract> object which can then be used to generate SQL
1790 through the methods below. The options accepted are:
1796 If set to 'lower', then SQL will be generated in all lowercase. By
1797 default SQL is generated in "textbook" case meaning something like:
1799 SELECT a_field FROM a_table WHERE some_field LIKE '%someval%'
1801 Any setting other than 'lower' is ignored.
1805 This determines what the default comparison operator is. By default
1806 it is C<=>, meaning that a hash like this:
1808 %where = (name => 'nwiger', email => 'nate@wiger.org');
1810 Will generate SQL like this:
1812 WHERE name = 'nwiger' AND email = 'nate@wiger.org'
1814 However, you may want loose comparisons by default, so if you set
1815 C<cmp> to C<like> you would get SQL such as:
1817 WHERE name like 'nwiger' AND email like 'nate@wiger.org'
1819 You can also override the comparison on an individual basis - see
1820 the huge section on L</"WHERE CLAUSES"> at the bottom.
1822 =item sqltrue, sqlfalse
1824 Expressions for inserting boolean values within SQL statements.
1825 By default these are C<1=1> and C<1=0>. They are used
1826 by the special operators C<-in> and C<-not_in> for generating
1827 correct SQL even when the argument is an empty array (see below).
1831 This determines the default logical operator for multiple WHERE
1832 statements in arrays or hashes. If absent, the default logic is "or"
1833 for arrays, and "and" for hashes. This means that a WHERE
1837 event_date => {'>=', '2/13/99'},
1838 event_date => {'<=', '4/24/03'},
1841 will generate SQL like this:
1843 WHERE event_date >= '2/13/99' OR event_date <= '4/24/03'
1845 This is probably not what you want given this query, though (look
1846 at the dates). To change the "OR" to an "AND", simply specify:
1848 my $sql = SQL::Abstract->new(logic => 'and');
1850 Which will change the above C<WHERE> to:
1852 WHERE event_date >= '2/13/99' AND event_date <= '4/24/03'
1854 The logic can also be changed locally by inserting
1855 a modifier in front of an arrayref :
1857 @where = (-and => [event_date => {'>=', '2/13/99'},
1858 event_date => {'<=', '4/24/03'} ]);
1860 See the L</"WHERE CLAUSES"> section for explanations.
1864 This will automatically convert comparisons using the specified SQL
1865 function for both column and value. This is mostly used with an argument
1866 of C<upper> or C<lower>, so that the SQL will have the effect of
1867 case-insensitive "searches". For example, this:
1869 $sql = SQL::Abstract->new(convert => 'upper');
1870 %where = (keywords => 'MaKe iT CAse inSeNSItive');
1872 Will turn out the following SQL:
1874 WHERE upper(keywords) like upper('MaKe iT CAse inSeNSItive')
1876 The conversion can be C<upper()>, C<lower()>, or any other SQL function
1877 that can be applied symmetrically to fields (actually B<SQL::Abstract> does
1878 not validate this option; it will just pass through what you specify verbatim).
1882 This is a kludge because many databases suck. For example, you can't
1883 just bind values using DBI's C<execute()> for Oracle C<CLOB> or C<BLOB> fields.
1884 Instead, you have to use C<bind_param()>:
1886 $sth->bind_param(1, 'reg data');
1887 $sth->bind_param(2, $lots, {ora_type => ORA_CLOB});
1889 The problem is, B<SQL::Abstract> will normally just return a C<@bind> array,
1890 which loses track of which field each slot refers to. Fear not.
1892 If you specify C<bindtype> in new, you can determine how C<@bind> is returned.
1893 Currently, you can specify either C<normal> (default) or C<columns>. If you
1894 specify C<columns>, you will get an array that looks like this:
1896 my $sql = SQL::Abstract->new(bindtype => 'columns');
1897 my($stmt, @bind) = $sql->insert(...);
1900 [ 'column1', 'value1' ],
1901 [ 'column2', 'value2' ],
1902 [ 'column3', 'value3' ],
1905 You can then iterate through this manually, using DBI's C<bind_param()>.
1907 $sth->prepare($stmt);
1910 my($col, $data) = @$_;
1911 if ($col eq 'details' || $col eq 'comments') {
1912 $sth->bind_param($i, $data, {ora_type => ORA_CLOB});
1913 } elsif ($col eq 'image') {
1914 $sth->bind_param($i, $data, {ora_type => ORA_BLOB});
1916 $sth->bind_param($i, $data);
1920 $sth->execute; # execute without @bind now
1922 Now, why would you still use B<SQL::Abstract> if you have to do this crap?
1923 Basically, the advantage is still that you don't have to care which fields
1924 are or are not included. You could wrap that above C<for> loop in a simple
1925 sub called C<bind_fields()> or something and reuse it repeatedly. You still
1926 get a layer of abstraction over manual SQL specification.
1928 Note that if you set L</bindtype> to C<columns>, the C<\[ $sql, @bind ]>
1929 construct (see L</Literal SQL with placeholders and bind values (subqueries)>)
1930 will expect the bind values in this format.
1934 This is the character that a table or column name will be quoted
1935 with. By default this is an empty string, but you could set it to
1936 the character C<`>, to generate SQL like this:
1938 SELECT `a_field` FROM `a_table` WHERE `some_field` LIKE '%someval%'
1940 Alternatively, you can supply an array ref of two items, the first being the left
1941 hand quote character, and the second the right hand quote character. For
1942 example, you could supply C<['[',']']> for SQL Server 2000 compliant quotes
1943 that generates SQL like this:
1945 SELECT [a_field] FROM [a_table] WHERE [some_field] LIKE '%someval%'
1947 Quoting is useful if you have tables or columns names that are reserved
1948 words in your database's SQL dialect.
1952 This is the character that will be used to escape L</quote_char>s appearing
1953 in an identifier before it has been quoted.
1955 The paramter default in case of a single L</quote_char> character is the quote
1958 When opening-closing-style quoting is used (L</quote_char> is an arrayref)
1959 this parameter defaults to the B<closing (right)> L</quote_char>. Occurences
1960 of the B<opening (left)> L</quote_char> within the identifier are currently left
1961 untouched. The default for opening-closing-style quotes may change in future
1962 versions, thus you are B<strongly encouraged> to specify the escape character
1967 This is the character that separates a table and column name. It is
1968 necessary to specify this when the C<quote_char> option is selected,
1969 so that tables and column names can be individually quoted like this:
1971 SELECT `table`.`one_field` FROM `table` WHERE `table`.`other_field` = 1
1973 =item injection_guard
1975 A regular expression C<qr/.../> that is applied to any C<-function> and unquoted
1976 column name specified in a query structure. This is a safety mechanism to avoid
1977 injection attacks when mishandling user input e.g.:
1979 my %condition_as_column_value_pairs = get_values_from_user();
1980 $sqla->select( ... , \%condition_as_column_value_pairs );
1982 If the expression matches an exception is thrown. Note that literal SQL
1983 supplied via C<\'...'> or C<\['...']> is B<not> checked in any way.
1985 Defaults to checking for C<;> and the C<GO> keyword (TransactSQL)
1987 =item array_datatypes
1989 When this option is true, arrayrefs in INSERT or UPDATE are
1990 interpreted as array datatypes and are passed directly
1992 When this option is false, arrayrefs are interpreted
1993 as literal SQL, just like refs to arrayrefs
1994 (but this behavior is for backwards compatibility; when writing
1995 new queries, use the "reference to arrayref" syntax
2001 Takes a reference to a list of "special operators"
2002 to extend the syntax understood by L<SQL::Abstract>.
2003 See section L</"SPECIAL OPERATORS"> for details.
2007 Takes a reference to a list of "unary operators"
2008 to extend the syntax understood by L<SQL::Abstract>.
2009 See section L</"UNARY OPERATORS"> for details.
2015 =head2 insert($table, \@values || \%fieldvals, \%options)
2017 This is the simplest function. You simply give it a table name
2018 and either an arrayref of values or hashref of field/value pairs.
2019 It returns an SQL INSERT statement and a list of bind values.
2020 See the sections on L</"Inserting and Updating Arrays"> and
2021 L</"Inserting and Updating SQL"> for information on how to insert
2022 with those data types.
2024 The optional C<\%options> hash reference may contain additional
2025 options to generate the insert SQL. Currently supported options
2032 Takes either a scalar of raw SQL fields, or an array reference of
2033 field names, and adds on an SQL C<RETURNING> statement at the end.
2034 This allows you to return data generated by the insert statement
2035 (such as row IDs) without performing another C<SELECT> statement.
2036 Note, however, this is not part of the SQL standard and may not
2037 be supported by all database engines.
2041 =head2 update($table, \%fieldvals, \%where)
2043 This takes a table, hashref of field/value pairs, and an optional
2044 hashref L<WHERE clause|/WHERE CLAUSES>. It returns an SQL UPDATE function and a list
2046 See the sections on L</"Inserting and Updating Arrays"> and
2047 L</"Inserting and Updating SQL"> for information on how to insert
2048 with those data types.
2050 =head2 select($source, $fields, $where, $order)
2052 This returns a SQL SELECT statement and associated list of bind values, as
2053 specified by the arguments :
2059 Specification of the 'FROM' part of the statement.
2060 The argument can be either a plain scalar (interpreted as a table
2061 name, will be quoted), or an arrayref (interpreted as a list
2062 of table names, joined by commas, quoted), or a scalarref
2063 (literal table name, not quoted), or a ref to an arrayref
2064 (list of literal table names, joined by commas, not quoted).
2068 Specification of the list of fields to retrieve from
2070 The argument can be either an arrayref (interpreted as a list
2071 of field names, will be joined by commas and quoted), or a
2072 plain scalar (literal SQL, not quoted).
2073 Please observe that this API is not as flexible as that of
2074 the first argument C<$source>, for backwards compatibility reasons.
2078 Optional argument to specify the WHERE part of the query.
2079 The argument is most often a hashref, but can also be
2080 an arrayref or plain scalar --
2081 see section L<WHERE clause|/"WHERE CLAUSES"> for details.
2085 Optional argument to specify the ORDER BY part of the query.
2086 The argument can be a scalar, a hashref or an arrayref
2087 -- see section L<ORDER BY clause|/"ORDER BY CLAUSES">
2093 =head2 delete($table, \%where)
2095 This takes a table name and optional hashref L<WHERE clause|/WHERE CLAUSES>.
2096 It returns an SQL DELETE statement and list of bind values.
2098 =head2 where(\%where, \@order)
2100 This is used to generate just the WHERE clause. For example,
2101 if you have an arbitrary data structure and know what the
2102 rest of your SQL is going to look like, but want an easy way
2103 to produce a WHERE clause, use this. It returns an SQL WHERE
2104 clause and list of bind values.
2107 =head2 values(\%data)
2109 This just returns the values from the hash C<%data>, in the same
2110 order that would be returned from any of the other above queries.
2111 Using this allows you to markedly speed up your queries if you
2112 are affecting lots of rows. See below under the L</"PERFORMANCE"> section.
2114 =head2 generate($any, 'number', $of, \@data, $struct, \%types)
2116 Warning: This is an experimental method and subject to change.
2118 This returns arbitrarily generated SQL. It's a really basic shortcut.
2119 It will return two different things, depending on return context:
2121 my($stmt, @bind) = $sql->generate('create table', \$table, \@fields);
2122 my $stmt_and_val = $sql->generate('create table', \$table, \@fields);
2124 These would return the following:
2126 # First calling form
2127 $stmt = "CREATE TABLE test (?, ?)";
2128 @bind = (field1, field2);
2130 # Second calling form
2131 $stmt_and_val = "CREATE TABLE test (field1, field2)";
2133 Depending on what you're trying to do, it's up to you to choose the correct
2134 format. In this example, the second form is what you would want.
2138 $sql->generate('alter session', { nls_date_format => 'MM/YY' });
2142 ALTER SESSION SET nls_date_format = 'MM/YY'
2144 You get the idea. Strings get their case twiddled, but everything
2145 else remains verbatim.
2147 =head1 EXPORTABLE FUNCTIONS
2149 =head2 is_plain_value
2151 Determines if the supplied argument is a plain value as understood by this
2156 =item * The value is C<undef>
2158 =item * The value is a non-reference
2160 =item * The value is an object with stringification overloading
2162 =item * The value is of the form C<< { -value => $anything } >>
2166 On failure returns C<undef>, on sucess returns a B<scalar> reference
2167 to the original supplied argument.
2173 The stringification overloading detection is rather advanced: it takes
2174 into consideration not only the presence of a C<""> overload, but if that
2175 fails also checks for enabled
2176 L<autogenerated versions of C<"">|overload/Magic Autogeneration>, based
2177 on either C<0+> or C<bool>.
2179 Unfortunately testing in the field indicates that this
2180 detection B<< may tickle a latent bug in perl versions before 5.018 >>,
2181 but only when very large numbers of stringifying objects are involved.
2182 At the time of writing ( Sep 2014 ) there is no clear explanation of
2183 the direct cause, nor is there a manageably small test case that reliably
2184 reproduces the problem.
2186 If you encounter any of the following exceptions in B<random places within
2187 your application stack> - this module may be to blame:
2189 Operation "ne": no method found,
2190 left argument in overloaded package <something>,
2191 right argument in overloaded package <something>
2195 Stub found while resolving method "???" overloading """" in package <something>
2197 If you fall victim to the above - please attempt to reduce the problem
2198 to something that could be sent to the L<SQL::Abstract developers
2199 |DBIx::Class/GETTING HELP/SUPPORT>
2200 (either publicly or privately). As a workaround in the meantime you can
2201 set C<$ENV{SQLA_ISVALUE_IGNORE_AUTOGENERATED_STRINGIFICATION}> to a true
2202 value, which will most likely eliminate your problem (at the expense of
2203 not being able to properly detect exotic forms of stringification).
2205 This notice and environment variable will be removed in a future version,
2206 as soon as the underlying problem is found and a reliable workaround is
2211 =head2 is_literal_value
2213 Determines if the supplied argument is a literal value as understood by this
2218 =item * C<\$sql_string>
2220 =item * C<\[ $sql_string, @bind_values ]>
2222 =item * C<< { -ident => $plain_defined_string } >>
2226 On failure returns C<undef>, on sucess returns an B<array> reference
2227 containing the unpacked version of the supplied literal SQL and bind values.
2229 =head1 WHERE CLAUSES
2233 This module uses a variation on the idea from L<DBIx::Abstract>. It
2234 is B<NOT>, repeat I<not> 100% compatible. B<The main logic of this
2235 module is that things in arrays are OR'ed, and things in hashes
2238 The easiest way to explain is to show lots of examples. After
2239 each C<%where> hash shown, it is assumed you used:
2241 my($stmt, @bind) = $sql->where(\%where);
2243 However, note that the C<%where> hash can be used directly in any
2244 of the other functions as well, as described above.
2246 =head2 Key-value pairs
2248 So, let's get started. To begin, a simple hash:
2252 status => 'completed'
2255 Is converted to SQL C<key = val> statements:
2257 $stmt = "WHERE user = ? AND status = ?";
2258 @bind = ('nwiger', 'completed');
2260 One common thing I end up doing is having a list of values that
2261 a field can be in. To do this, simply specify a list inside of
2266 status => ['assigned', 'in-progress', 'pending'];
2269 This simple code will create the following:
2271 $stmt = "WHERE user = ? AND ( status = ? OR status = ? OR status = ? )";
2272 @bind = ('nwiger', 'assigned', 'in-progress', 'pending');
2274 A field associated to an empty arrayref will be considered a
2275 logical false and will generate 0=1.
2277 =head2 Tests for NULL values
2279 If the value part is C<undef> then this is converted to SQL <IS NULL>
2288 $stmt = "WHERE user = ? AND status IS NULL";
2291 To test if a column IS NOT NULL:
2295 status => { '!=', undef },
2298 =head2 Specific comparison operators
2300 If you want to specify a different type of operator for your comparison,
2301 you can use a hashref for a given column:
2305 status => { '!=', 'completed' }
2308 Which would generate:
2310 $stmt = "WHERE user = ? AND status != ?";
2311 @bind = ('nwiger', 'completed');
2313 To test against multiple values, just enclose the values in an arrayref:
2315 status => { '=', ['assigned', 'in-progress', 'pending'] };
2317 Which would give you:
2319 "WHERE status = ? OR status = ? OR status = ?"
2322 The hashref can also contain multiple pairs, in which case it is expanded
2323 into an C<AND> of its elements:
2327 status => { '!=', 'completed', -not_like => 'pending%' }
2330 # Or more dynamically, like from a form
2331 $where{user} = 'nwiger';
2332 $where{status}{'!='} = 'completed';
2333 $where{status}{'-not_like'} = 'pending%';
2335 # Both generate this
2336 $stmt = "WHERE user = ? AND status != ? AND status NOT LIKE ?";
2337 @bind = ('nwiger', 'completed', 'pending%');
2340 To get an OR instead, you can combine it with the arrayref idea:
2344 priority => [ { '=', 2 }, { '>', 5 } ]
2347 Which would generate:
2349 $stmt = "WHERE ( priority = ? OR priority > ? ) AND user = ?";
2350 @bind = ('2', '5', 'nwiger');
2352 If you want to include literal SQL (with or without bind values), just use a
2353 scalar reference or reference to an arrayref as the value:
2356 date_entered => { '>' => \["to_date(?, 'MM/DD/YYYY')", "11/26/2008"] },
2357 date_expires => { '<' => \"now()" }
2360 Which would generate:
2362 $stmt = "WHERE date_entered > to_date(?, 'MM/DD/YYYY') AND date_expires < now()";
2363 @bind = ('11/26/2008');
2366 =head2 Logic and nesting operators
2368 In the example above,
2369 there is a subtle trap if you want to say something like
2370 this (notice the C<AND>):
2372 WHERE priority != ? AND priority != ?
2374 Because, in Perl you I<can't> do this:
2376 priority => { '!=' => 2, '!=' => 1 }
2378 As the second C<!=> key will obliterate the first. The solution
2379 is to use the special C<-modifier> form inside an arrayref:
2381 priority => [ -and => {'!=', 2},
2385 Normally, these would be joined by C<OR>, but the modifier tells it
2386 to use C<AND> instead. (Hint: You can use this in conjunction with the
2387 C<logic> option to C<new()> in order to change the way your queries
2388 work by default.) B<Important:> Note that the C<-modifier> goes
2389 B<INSIDE> the arrayref, as an extra first element. This will
2390 B<NOT> do what you think it might:
2392 priority => -and => [{'!=', 2}, {'!=', 1}] # WRONG!
2394 Here is a quick list of equivalencies, since there is some overlap:
2397 status => {'!=', 'completed', 'not like', 'pending%' }
2398 status => [ -and => {'!=', 'completed'}, {'not like', 'pending%'}]
2401 status => {'=', ['assigned', 'in-progress']}
2402 status => [ -or => {'=', 'assigned'}, {'=', 'in-progress'}]
2403 status => [ {'=', 'assigned'}, {'=', 'in-progress'} ]
2407 =head2 Special operators : IN, BETWEEN, etc.
2409 You can also use the hashref format to compare a list of fields using the
2410 C<IN> comparison operator, by specifying the list as an arrayref:
2413 status => 'completed',
2414 reportid => { -in => [567, 2335, 2] }
2417 Which would generate:
2419 $stmt = "WHERE status = ? AND reportid IN (?,?,?)";
2420 @bind = ('completed', '567', '2335', '2');
2422 The reverse operator C<-not_in> generates SQL C<NOT IN> and is used in
2425 If the argument to C<-in> is an empty array, 'sqlfalse' is generated
2426 (by default : C<1=0>). Similarly, C<< -not_in => [] >> generates
2427 'sqltrue' (by default : C<1=1>).
2429 In addition to the array you can supply a chunk of literal sql or
2430 literal sql with bind:
2433 customer => { -in => \[
2434 'SELECT cust_id FROM cust WHERE balance > ?',
2437 status => { -in => \'SELECT status_codes FROM states' },
2443 customer IN ( SELECT cust_id FROM cust WHERE balance > ? )
2444 AND status IN ( SELECT status_codes FROM states )
2448 Finally, if the argument to C<-in> is not a reference, it will be
2449 treated as a single-element array.
2451 Another pair of operators is C<-between> and C<-not_between>,
2452 used with an arrayref of two values:
2456 completion_date => {
2457 -not_between => ['2002-10-01', '2003-02-06']
2463 WHERE user = ? AND completion_date NOT BETWEEN ( ? AND ? )
2465 Just like with C<-in> all plausible combinations of literal SQL
2469 start0 => { -between => [ 1, 2 ] },
2470 start1 => { -between => \["? AND ?", 1, 2] },
2471 start2 => { -between => \"lower(x) AND upper(y)" },
2472 start3 => { -between => [
2474 \["upper(?)", 'stuff' ],
2481 ( start0 BETWEEN ? AND ? )
2482 AND ( start1 BETWEEN ? AND ? )
2483 AND ( start2 BETWEEN lower(x) AND upper(y) )
2484 AND ( start3 BETWEEN lower(x) AND upper(?) )
2486 @bind = (1, 2, 1, 2, 'stuff');
2489 These are the two builtin "special operators"; but the
2490 list can be expanded : see section L</"SPECIAL OPERATORS"> below.
2492 =head2 Unary operators: bool
2494 If you wish to test against boolean columns or functions within your
2495 database you can use the C<-bool> and C<-not_bool> operators. For
2496 example to test the column C<is_user> being true and the column
2497 C<is_enabled> being false you would use:-
2501 -not_bool => 'is_enabled',
2506 WHERE is_user AND NOT is_enabled
2508 If a more complex combination is required, testing more conditions,
2509 then you should use the and/or operators:-
2514 -not_bool => { two=> { -rlike => 'bar' } },
2515 -not_bool => { three => [ { '=', 2 }, { '>', 5 } ] },
2526 (NOT ( three = ? OR three > ? ))
2529 =head2 Nested conditions, -and/-or prefixes
2531 So far, we've seen how multiple conditions are joined with a top-level
2532 C<AND>. We can change this by putting the different conditions we want in
2533 hashes and then putting those hashes in an array. For example:
2538 status => { -like => ['pending%', 'dispatched'] },
2542 status => 'unassigned',
2546 This data structure would create the following:
2548 $stmt = "WHERE ( user = ? AND ( status LIKE ? OR status LIKE ? ) )
2549 OR ( user = ? AND status = ? ) )";
2550 @bind = ('nwiger', 'pending', 'dispatched', 'robot', 'unassigned');
2553 Clauses in hashrefs or arrayrefs can be prefixed with an C<-and> or C<-or>
2554 to change the logic inside :
2560 -and => [ workhrs => {'>', 20}, geo => 'ASIA' ],
2561 -or => { workhrs => {'<', 50}, geo => 'EURO' },
2568 $stmt = "WHERE ( user = ?
2569 AND ( ( workhrs > ? AND geo = ? )
2570 OR ( workhrs < ? OR geo = ? ) ) )";
2571 @bind = ('nwiger', '20', 'ASIA', '50', 'EURO');
2573 =head3 Algebraic inconsistency, for historical reasons
2575 C<Important note>: when connecting several conditions, the C<-and->|C<-or>
2576 operator goes C<outside> of the nested structure; whereas when connecting
2577 several constraints on one column, the C<-and> operator goes
2578 C<inside> the arrayref. Here is an example combining both features :
2581 -and => [a => 1, b => 2],
2582 -or => [c => 3, d => 4],
2583 e => [-and => {-like => 'foo%'}, {-like => '%bar'} ]
2588 WHERE ( ( ( a = ? AND b = ? )
2589 OR ( c = ? OR d = ? )
2590 OR ( e LIKE ? AND e LIKE ? ) ) )
2592 This difference in syntax is unfortunate but must be preserved for
2593 historical reasons. So be careful : the two examples below would
2594 seem algebraically equivalent, but they are not
2596 {col => [-and => {-like => 'foo%'}, {-like => '%bar'}]}
2597 # yields : WHERE ( ( col LIKE ? AND col LIKE ? ) )
2599 [-and => {col => {-like => 'foo%'}, {col => {-like => '%bar'}}]]
2600 # yields : WHERE ( ( col LIKE ? OR col LIKE ? ) )
2603 =head2 Literal SQL and value type operators
2605 The basic premise of SQL::Abstract is that in WHERE specifications the "left
2606 side" is a column name and the "right side" is a value (normally rendered as
2607 a placeholder). This holds true for both hashrefs and arrayref pairs as you
2608 see in the L</WHERE CLAUSES> examples above. Sometimes it is necessary to
2609 alter this behavior. There are several ways of doing so.
2613 This is a virtual operator that signals the string to its right side is an
2614 identifier (a column name) and not a value. For example to compare two
2615 columns you would write:
2618 priority => { '<', 2 },
2619 requestor => { -ident => 'submitter' },
2624 $stmt = "WHERE priority < ? AND requestor = submitter";
2627 If you are maintaining legacy code you may see a different construct as
2628 described in L</Deprecated usage of Literal SQL>, please use C<-ident> in new
2633 This is a virtual operator that signals that the construct to its right side
2634 is a value to be passed to DBI. This is for example necessary when you want
2635 to write a where clause against an array (for RDBMS that support such
2636 datatypes). For example:
2639 array => { -value => [1, 2, 3] }
2644 $stmt = 'WHERE array = ?';
2645 @bind = ([1, 2, 3]);
2647 Note that if you were to simply say:
2653 the result would probably not be what you wanted:
2655 $stmt = 'WHERE array = ? OR array = ? OR array = ?';
2660 Finally, sometimes only literal SQL will do. To include a random snippet
2661 of SQL verbatim, you specify it as a scalar reference. Consider this only
2662 as a last resort. Usually there is a better way. For example:
2665 priority => { '<', 2 },
2666 requestor => { -in => \'(SELECT name FROM hitmen)' },
2671 $stmt = "WHERE priority < ? AND requestor IN (SELECT name FROM hitmen)"
2674 Note that in this example, you only get one bind parameter back, since
2675 the verbatim SQL is passed as part of the statement.
2679 Never use untrusted input as a literal SQL argument - this is a massive
2680 security risk (there is no way to check literal snippets for SQL
2681 injections and other nastyness). If you need to deal with untrusted input
2682 use literal SQL with placeholders as described next.
2684 =head3 Literal SQL with placeholders and bind values (subqueries)
2686 If the literal SQL to be inserted has placeholders and bind values,
2687 use a reference to an arrayref (yes this is a double reference --
2688 not so common, but perfectly legal Perl). For example, to find a date
2689 in Postgres you can use something like this:
2692 date_column => \[ "= date '2008-09-30' - ?::integer", 10 ]
2697 $stmt = "WHERE ( date_column = date '2008-09-30' - ?::integer )"
2700 Note that you must pass the bind values in the same format as they are returned
2701 by L<where|/where(\%where, \@order)>. This means that if you set L</bindtype>
2702 to C<columns>, you must provide the bind values in the
2703 C<< [ column_meta => value ] >> format, where C<column_meta> is an opaque
2704 scalar value; most commonly the column name, but you can use any scalar value
2705 (including references and blessed references), L<SQL::Abstract> will simply
2706 pass it through intact. So if C<bindtype> is set to C<columns> the above
2707 example will look like:
2710 date_column => \[ "= date '2008-09-30' - ?::integer", [ {} => 10 ] ]
2713 Literal SQL is especially useful for nesting parenthesized clauses in the
2714 main SQL query. Here is a first example :
2716 my ($sub_stmt, @sub_bind) = ("SELECT c1 FROM t1 WHERE c2 < ? AND c3 LIKE ?",
2720 bar => \["IN ($sub_stmt)" => @sub_bind],
2725 $stmt = "WHERE (foo = ? AND bar IN (SELECT c1 FROM t1
2726 WHERE c2 < ? AND c3 LIKE ?))";
2727 @bind = (1234, 100, "foo%");
2729 Other subquery operators, like for example C<"E<gt> ALL"> or C<"NOT IN">,
2730 are expressed in the same way. Of course the C<$sub_stmt> and
2731 its associated bind values can be generated through a former call
2734 my ($sub_stmt, @sub_bind)
2735 = $sql->select("t1", "c1", {c2 => {"<" => 100},
2736 c3 => {-like => "foo%"}});
2739 bar => \["> ALL ($sub_stmt)" => @sub_bind],
2742 In the examples above, the subquery was used as an operator on a column;
2743 but the same principle also applies for a clause within the main C<%where>
2744 hash, like an EXISTS subquery :
2746 my ($sub_stmt, @sub_bind)
2747 = $sql->select("t1", "*", {c1 => 1, c2 => \"> t0.c0"});
2748 my %where = ( -and => [
2750 \["EXISTS ($sub_stmt)" => @sub_bind],
2755 $stmt = "WHERE (foo = ? AND EXISTS (SELECT * FROM t1
2756 WHERE c1 = ? AND c2 > t0.c0))";
2760 Observe that the condition on C<c2> in the subquery refers to
2761 column C<t0.c0> of the main query : this is I<not> a bind
2762 value, so we have to express it through a scalar ref.
2763 Writing C<< c2 => {">" => "t0.c0"} >> would have generated
2764 C<< c2 > ? >> with bind value C<"t0.c0"> ... not exactly
2765 what we wanted here.
2767 Finally, here is an example where a subquery is used
2768 for expressing unary negation:
2770 my ($sub_stmt, @sub_bind)
2771 = $sql->where({age => [{"<" => 10}, {">" => 20}]});
2772 $sub_stmt =~ s/^ where //i; # don't want "WHERE" in the subclause
2774 lname => {like => '%son%'},
2775 \["NOT ($sub_stmt)" => @sub_bind],
2780 $stmt = "lname LIKE ? AND NOT ( age < ? OR age > ? )"
2781 @bind = ('%son%', 10, 20)
2783 =head3 Deprecated usage of Literal SQL
2785 Below are some examples of archaic use of literal SQL. It is shown only as
2786 reference for those who deal with legacy code. Each example has a much
2787 better, cleaner and safer alternative that users should opt for in new code.
2793 my %where = ( requestor => \'IS NOT NULL' )
2795 $stmt = "WHERE requestor IS NOT NULL"
2797 This used to be the way of generating NULL comparisons, before the handling
2798 of C<undef> got formalized. For new code please use the superior syntax as
2799 described in L</Tests for NULL values>.
2803 my %where = ( requestor => \'= submitter' )
2805 $stmt = "WHERE requestor = submitter"
2807 This used to be the only way to compare columns. Use the superior L</-ident>
2808 method for all new code. For example an identifier declared in such a way
2809 will be properly quoted if L</quote_char> is properly set, while the legacy
2810 form will remain as supplied.
2814 my %where = ( is_ready => \"", completed => { '>', '2012-12-21' } )
2816 $stmt = "WHERE completed > ? AND is_ready"
2817 @bind = ('2012-12-21')
2819 Using an empty string literal used to be the only way to express a boolean.
2820 For all new code please use the much more readable
2821 L<-bool|/Unary operators: bool> operator.
2827 These pages could go on for a while, since the nesting of the data
2828 structures this module can handle are pretty much unlimited (the
2829 module implements the C<WHERE> expansion as a recursive function
2830 internally). Your best bet is to "play around" with the module a
2831 little to see how the data structures behave, and choose the best
2832 format for your data based on that.
2834 And of course, all the values above will probably be replaced with
2835 variables gotten from forms or the command line. After all, if you
2836 knew everything ahead of time, you wouldn't have to worry about
2837 dynamically-generating SQL and could just hardwire it into your
2840 =head1 ORDER BY CLAUSES
2842 Some functions take an order by clause. This can either be a scalar (just a
2843 column name,) a hash of C<< { -desc => 'col' } >> or C<< { -asc => 'col' } >>,
2844 or an array of either of the two previous forms. Examples:
2846 Given | Will Generate
2847 ----------------------------------------------------------
2849 \'colA DESC' | ORDER BY colA DESC
2851 'colA' | ORDER BY colA
2853 [qw/colA colB/] | ORDER BY colA, colB
2855 {-asc => 'colA'} | ORDER BY colA ASC
2857 {-desc => 'colB'} | ORDER BY colB DESC
2859 ['colA', {-asc => 'colB'}] | ORDER BY colA, colB ASC
2861 { -asc => [qw/colA colB/] } | ORDER BY colA ASC, colB ASC
2864 { -asc => 'colA' }, | ORDER BY colA ASC, colB DESC,
2865 { -desc => [qw/colB/], | colC ASC, colD ASC
2866 { -asc => [qw/colC colD/],|
2868 ===========================================================
2872 =head1 SPECIAL OPERATORS
2874 my $sqlmaker = SQL::Abstract->new(special_ops => [
2878 my ($self, $field, $op, $arg) = @_;
2884 handler => 'method_name',
2888 A "special operator" is a SQL syntactic clause that can be
2889 applied to a field, instead of a usual binary operator.
2892 WHERE field IN (?, ?, ?)
2893 WHERE field BETWEEN ? AND ?
2894 WHERE MATCH(field) AGAINST (?, ?)
2896 Special operators IN and BETWEEN are fairly standard and therefore
2897 are builtin within C<SQL::Abstract> (as the overridable methods
2898 C<_where_field_IN> and C<_where_field_BETWEEN>). For other operators,
2899 like the MATCH .. AGAINST example above which is specific to MySQL,
2900 you can write your own operator handlers - supply a C<special_ops>
2901 argument to the C<new> method. That argument takes an arrayref of
2902 operator definitions; each operator definition is a hashref with two
2909 the regular expression to match the operator
2913 Either a coderef or a plain scalar method name. In both cases
2914 the expected return is C<< ($sql, @bind) >>.
2916 When supplied with a method name, it is simply called on the
2917 L<SQL::Abstract> object as:
2919 $self->$method_name ($field, $op, $arg)
2923 $field is the LHS of the operator
2924 $op is the part that matched the handler regex
2927 When supplied with a coderef, it is called as:
2929 $coderef->($self, $field, $op, $arg)
2934 For example, here is an implementation
2935 of the MATCH .. AGAINST syntax for MySQL
2937 my $sqlmaker = SQL::Abstract->new(special_ops => [
2939 # special op for MySql MATCH (field) AGAINST(word1, word2, ...)
2940 {regex => qr/^match$/i,
2942 my ($self, $field, $op, $arg) = @_;
2943 $arg = [$arg] if not ref $arg;
2944 my $label = $self->_quote($field);
2945 my ($placeholder) = $self->_convert('?');
2946 my $placeholders = join ", ", (($placeholder) x @$arg);
2947 my $sql = $self->_sqlcase('match') . " ($label) "
2948 . $self->_sqlcase('against') . " ($placeholders) ";
2949 my @bind = $self->_bindtype($field, @$arg);
2950 return ($sql, @bind);
2957 =head1 UNARY OPERATORS
2959 my $sqlmaker = SQL::Abstract->new(unary_ops => [
2963 my ($self, $op, $arg) = @_;
2969 handler => 'method_name',
2973 A "unary operator" is a SQL syntactic clause that can be
2974 applied to a field - the operator goes before the field
2976 You can write your own operator handlers - supply a C<unary_ops>
2977 argument to the C<new> method. That argument takes an arrayref of
2978 operator definitions; each operator definition is a hashref with two
2985 the regular expression to match the operator
2989 Either a coderef or a plain scalar method name. In both cases
2990 the expected return is C<< $sql >>.
2992 When supplied with a method name, it is simply called on the
2993 L<SQL::Abstract> object as:
2995 $self->$method_name ($op, $arg)
2999 $op is the part that matched the handler regex
3000 $arg is the RHS or argument of the operator
3002 When supplied with a coderef, it is called as:
3004 $coderef->($self, $op, $arg)
3012 Thanks to some benchmarking by Mark Stosberg, it turns out that
3013 this module is many orders of magnitude faster than using C<DBIx::Abstract>.
3014 I must admit this wasn't an intentional design issue, but it's a
3015 byproduct of the fact that you get to control your C<DBI> handles
3018 To maximize performance, use a code snippet like the following:
3020 # prepare a statement handle using the first row
3021 # and then reuse it for the rest of the rows
3023 for my $href (@array_of_hashrefs) {
3024 $stmt ||= $sql->insert('table', $href);
3025 $sth ||= $dbh->prepare($stmt);
3026 $sth->execute($sql->values($href));
3029 The reason this works is because the keys in your C<$href> are sorted
3030 internally by B<SQL::Abstract>. Thus, as long as your data retains
3031 the same structure, you only have to generate the SQL the first time
3032 around. On subsequent queries, simply use the C<values> function provided
3033 by this module to return your values in the correct order.
3035 However this depends on the values having the same type - if, for
3036 example, the values of a where clause may either have values
3037 (resulting in sql of the form C<column = ?> with a single bind
3038 value), or alternatively the values might be C<undef> (resulting in
3039 sql of the form C<column IS NULL> with no bind value) then the
3040 caching technique suggested will not work.
3044 If you use my C<CGI::FormBuilder> module at all, you'll hopefully
3045 really like this part (I do, at least). Building up a complex query
3046 can be as simple as the following:
3053 use CGI::FormBuilder;
3056 my $form = CGI::FormBuilder->new(...);
3057 my $sql = SQL::Abstract->new;
3059 if ($form->submitted) {
3060 my $field = $form->field;
3061 my $id = delete $field->{id};
3062 my($stmt, @bind) = $sql->update('table', $field, {id => $id});
3065 Of course, you would still have to connect using C<DBI> to run the
3066 query, but the point is that if you make your form look like your
3067 table, the actual query script can be extremely simplistic.
3069 If you're B<REALLY> lazy (I am), check out C<HTML::QuickTable> for
3070 a fast interface to returning and formatting data. I frequently
3071 use these three modules together to write complex database query
3072 apps in under 50 lines.
3078 =item * gitweb: L<http://git.shadowcat.co.uk/gitweb/gitweb.cgi?p=dbsrgits/SQL-Abstract.git>
3080 =item * git: L<git://git.shadowcat.co.uk/dbsrgits/SQL-Abstract.git>
3086 Version 1.50 was a major internal refactoring of C<SQL::Abstract>.
3087 Great care has been taken to preserve the I<published> behavior
3088 documented in previous versions in the 1.* family; however,
3089 some features that were previously undocumented, or behaved
3090 differently from the documentation, had to be changed in order
3091 to clarify the semantics. Hence, client code that was relying
3092 on some dark areas of C<SQL::Abstract> v1.*
3093 B<might behave differently> in v1.50.
3095 The main changes are :
3101 support for literal SQL through the C<< \ [ $sql, @bind ] >> syntax.
3105 support for the { operator => \"..." } construct (to embed literal SQL)
3109 support for the { operator => \["...", @bind] } construct (to embed literal SQL with bind values)
3113 optional support for L<array datatypes|/"Inserting and Updating Arrays">
3117 defensive programming : check arguments
3121 fixed bug with global logic, which was previously implemented
3122 through global variables yielding side-effects. Prior versions would
3123 interpret C<< [ {cond1, cond2}, [cond3, cond4] ] >>
3124 as C<< "(cond1 AND cond2) OR (cond3 AND cond4)" >>.
3125 Now this is interpreted
3126 as C<< "(cond1 AND cond2) OR (cond3 OR cond4)" >>.
3131 fixed semantics of _bindtype on array args
3135 dropped the C<_anoncopy> of the %where tree. No longer necessary,
3136 we just avoid shifting arrays within that tree.
3140 dropped the C<_modlogic> function
3144 =head1 ACKNOWLEDGEMENTS
3146 There are a number of individuals that have really helped out with
3147 this module. Unfortunately, most of them submitted bugs via CPAN
3148 so I have no idea who they are! But the people I do know are:
3150 Ash Berlin (order_by hash term support)
3151 Matt Trout (DBIx::Class support)
3152 Mark Stosberg (benchmarking)
3153 Chas Owens (initial "IN" operator support)
3154 Philip Collins (per-field SQL functions)
3155 Eric Kolve (hashref "AND" support)
3156 Mike Fragassi (enhancements to "BETWEEN" and "LIKE")
3157 Dan Kubb (support for "quote_char" and "name_sep")
3158 Guillermo Roditi (patch to cleanup "IN" and "BETWEEN", fix and tests for _order_by)
3159 Laurent Dami (internal refactoring, extensible list of special operators, literal SQL)
3160 Norbert Buchmuller (support for literal SQL in hashpair, misc. fixes & tests)
3161 Peter Rabbitson (rewrite of SQLA::Test, misc. fixes & tests)
3162 Oliver Charles (support for "RETURNING" after "INSERT")
3168 L<DBIx::Class>, L<DBIx::Abstract>, L<CGI::FormBuilder>, L<HTML::QuickTable>.
3172 Copyright (c) 2001-2007 Nathan Wiger <nwiger@cpan.org>. All Rights Reserved.
3174 This module is actively maintained by Matt Trout <mst@shadowcatsystems.co.uk>
3176 For support, your best bet is to try the C<DBIx::Class> users mailing list.
3177 While not an official support venue, C<DBIx::Class> makes heavy use of
3178 C<SQL::Abstract>, and as such list members there are very familiar with
3179 how to create queries.
3183 This module is free software; you may copy this under the same
3184 terms as perl itself (either the GNU General Public License or
3185 the Artistic License)