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 = '1.85';
32 # This would confuse some packagers
33 $VERSION = eval $VERSION if $VERSION =~ /_/; # numify for warning-free dev releases
37 # special operators (-in, -between). May be extended/overridden by user.
38 # See section WHERE: BUILTIN SPECIAL OPERATORS below for implementation
39 my @BUILTIN_SPECIAL_OPS = (
40 {regex => qr/^ (?: not \s )? between $/ix, handler => '_where_field_BETWEEN'},
41 {regex => qr/^ (?: not \s )? in $/ix, handler => '_where_field_IN'},
42 {regex => qr/^ ident $/ix, handler => '_where_op_IDENT'},
43 {regex => qr/^ value $/ix, handler => '_where_op_VALUE'},
44 {regex => qr/^ is (?: \s+ not )? $/ix, handler => '_where_field_IS'},
47 # unaryish operators - key maps to handler
48 my @BUILTIN_UNARY_OPS = (
49 # the digits are backcompat stuff
50 { regex => qr/^ and (?: [_\s]? \d+ )? $/xi, handler => '_where_op_ANDOR' },
51 { regex => qr/^ or (?: [_\s]? \d+ )? $/xi, handler => '_where_op_ANDOR' },
52 { regex => qr/^ nest (?: [_\s]? \d+ )? $/xi, handler => '_where_op_NEST' },
53 { regex => qr/^ (?: not \s )? bool $/xi, handler => '_where_op_BOOL' },
54 { regex => qr/^ ident $/xi, handler => '_where_op_IDENT' },
55 { regex => qr/^ value $/xi, handler => '_where_op_VALUE' },
58 #======================================================================
59 # DEBUGGING AND ERROR REPORTING
60 #======================================================================
63 return unless $_[0]->{debug}; shift; # a little faster
64 my $func = (caller(1))[3];
65 warn "[$func] ", @_, "\n";
69 my($func) = (caller(1))[3];
70 Carp::carp "[$func] Warning: ", @_;
74 my($func) = (caller(1))[3];
75 Carp::croak "[$func] Fatal: ", @_;
78 sub is_literal_value ($) {
79 ref $_[0] eq 'SCALAR' ? [ ${$_[0]} ]
80 : ( ref $_[0] eq 'REF' and ref ${$_[0]} eq 'ARRAY' ) ? [ @${ $_[0] } ]
84 # FIXME XSify - this can be done so much more efficiently
85 sub is_plain_value ($) {
87 ! length ref $_[0] ? \($_[0])
89 ref $_[0] eq 'HASH' and keys %{$_[0]} == 1
91 exists $_[0]->{-value}
92 ) ? \($_[0]->{-value})
94 # reuse @_ for even moar speedz
95 defined ( $_[1] = Scalar::Util::blessed $_[0] )
97 # deliberately not using Devel::OverloadInfo - the checks we are
98 # intersted in are much more limited than the fullblown thing, and
99 # this is a very hot piece of code
101 # simply using ->can('(""') can leave behind stub methods that
102 # break actually using the overload later (see L<perldiag/Stub
103 # found while resolving method "%s" overloading "%s" in package
104 # "%s"> and the source of overload::mycan())
106 # either has stringification which DBI SHOULD prefer out of the box
107 grep { *{ (qq[${_}::(""]) }{CODE} } @{ $_[2] = mro::get_linear_isa( $_[1] ) }
109 # has nummification or boolification, AND fallback is *not* disabled
111 SQL::Abstract::_ENV_::DETECT_AUTOGENERATED_STRINGIFICATION
114 grep { *{"${_}::(0+"}{CODE} } @{$_[2]}
116 grep { *{"${_}::(bool"}{CODE} } @{$_[2]}
120 # no fallback specified at all
121 ! ( ($_[3]) = grep { *{"${_}::()"}{CODE} } @{$_[2]} )
123 # fallback explicitly undef
124 ! defined ${"$_[3]::()"}
137 #======================================================================
139 #======================================================================
143 my $class = ref($self) || $self;
144 my %opt = (ref $_[0] eq 'HASH') ? %{$_[0]} : @_;
146 # choose our case by keeping an option around
147 delete $opt{case} if $opt{case} && $opt{case} ne 'lower';
149 # default logic for interpreting arrayrefs
150 $opt{logic} = $opt{logic} ? uc $opt{logic} : 'OR';
152 # how to return bind vars
153 $opt{bindtype} ||= 'normal';
155 # default comparison is "=", but can be overridden
158 # try to recognize which are the 'equality' and 'inequality' ops
159 # (temporary quickfix (in 2007), should go through a more seasoned API)
160 $opt{equality_op} = qr/^( \Q$opt{cmp}\E | \= )$/ix;
161 $opt{inequality_op} = qr/^( != | <> )$/ix;
163 $opt{like_op} = qr/^ (is\s+)? r?like $/xi;
164 $opt{not_like_op} = qr/^ (is\s+)? not \s+ r?like $/xi;
167 $opt{sqltrue} ||= '1=1';
168 $opt{sqlfalse} ||= '0=1';
171 $opt{special_ops} ||= [];
172 # regexes are applied in order, thus push after user-defines
173 push @{$opt{special_ops}}, @BUILTIN_SPECIAL_OPS;
176 $opt{unary_ops} ||= [];
177 push @{$opt{unary_ops}}, @BUILTIN_UNARY_OPS;
179 # rudimentary sanity-check for user supplied bits treated as functions/operators
180 # If a purported function matches this regular expression, an exception is thrown.
181 # Literal SQL is *NOT* subject to this check, only functions (and column names
182 # when quoting is not in effect)
185 # need to guard against ()'s in column names too, but this will break tons of
186 # hacks... ideas anyone?
187 $opt{injection_guard} ||= qr/
193 return bless \%opt, $class;
197 sub _assert_pass_injection_guard {
198 if ($_[1] =~ $_[0]->{injection_guard}) {
199 my $class = ref $_[0];
200 puke "Possible SQL injection attempt '$_[1]'. If this is indeed a part of the "
201 . "desired SQL use literal SQL ( \'...' or \[ '...' ] ) or supply your own "
202 . "{injection_guard} attribute to ${class}->new()"
207 #======================================================================
209 #======================================================================
213 my ($table_sql, @all_bind) = $self->_table(shift);
214 my $data = shift || return;
217 my $method = $self->_METHOD_FOR_refkind("_insert", $data);
218 my ($values_sql, @values_bind) = $self->$method($data);
219 my $sql = join " ", $self->_sqlcase('insert into'), $table_sql, $values_sql;
220 push @all_bind, @values_bind;
222 if ($options->{returning}) {
223 my ($returning_sql, @returning_bind) = $self->_insert_returning($options);
224 $sql .= $returning_sql;
225 push @all_bind, @returning_bind;
228 return wantarray ? ($sql, @all_bind) : $sql;
231 # So that subclasses can override INSERT ... RETURNING separately from
232 # UPDATE and DELETE (e.g. DBIx::Class::SQLMaker::Oracle does this)
233 sub _insert_returning { shift->_returning(@_) }
236 my ($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 my (@values, @all_bind);
270 foreach my $value (@$data) {
271 my ($values, @bind) = $self->_insert_value(undef, $value);
272 push @values, $values;
273 push @all_bind, @bind;
275 my $sql = $self->_sqlcase('values')." ( ".join(", ", @values)." )";
276 return ($sql, @all_bind);
279 sub _insert_ARRAYREFREF { # literal SQL with bind
280 my ($self, $data) = @_;
282 my ($sql, @bind) = @${$data};
283 $self->_assert_bindval_matches_bindtype(@bind);
285 return ($sql, @bind);
289 sub _insert_SCALARREF { # literal SQL without bind
290 my ($self, $data) = @_;
296 my ($self, $data) = @_;
298 my (@values, @all_bind);
299 foreach my $column (sort keys %$data) {
300 my ($values, @bind) = $self->_insert_value($column, $data->{$column});
301 push @values, $values;
302 push @all_bind, @bind;
304 my $sql = $self->_sqlcase('values')." ( ".join(", ", @values)." )";
305 return ($sql, @all_bind);
309 my ($self, $column, $v) = @_;
311 my (@values, @all_bind);
312 $self->_SWITCH_refkind($v, {
315 if ($self->{array_datatypes}) { # if array datatype are activated
317 push @all_bind, $self->_bindtype($column, $v);
319 else { # else literal SQL with bind
320 my ($sql, @bind) = @$v;
321 $self->_assert_bindval_matches_bindtype(@bind);
323 push @all_bind, @bind;
327 ARRAYREFREF => sub { # literal SQL with bind
328 my ($sql, @bind) = @${$v};
329 $self->_assert_bindval_matches_bindtype(@bind);
331 push @all_bind, @bind;
334 # THINK : anything useful to do with a HASHREF ?
335 HASHREF => sub { # (nothing, but old SQLA passed it through)
336 #TODO in SQLA >= 2.0 it will die instead
337 belch "HASH ref as bind value in insert is not supported";
339 push @all_bind, $self->_bindtype($column, $v);
342 SCALARREF => sub { # literal SQL without bind
346 SCALAR_or_UNDEF => sub {
348 push @all_bind, $self->_bindtype($column, $v);
353 my $sql = join(", ", @values);
354 return ($sql, @all_bind);
359 #======================================================================
361 #======================================================================
366 my ($table_sql, @all_bind) = $self->_table(shift);
367 my $data = shift || return;
371 # first build the 'SET' part of the sql statement
372 puke "Unsupported data type specified to \$sql->update"
373 unless ref $data eq 'HASH';
375 my ($values_sql, @values_bind) = $self->_update_set_values($data);
376 my $sql = $self->_sqlcase('update ') . $table_sql . $self->_sqlcase(' set ')
378 push @all_bind, @values_bind;
381 my($where_sql, @where_bind) = $self->where($where);
383 push @all_bind, @where_bind;
386 if ($options->{returning}) {
387 my ($returning_sql, @returning_bind) = $self->_update_returning($options);
388 $sql .= $returning_sql;
389 push @all_bind, @returning_bind;
392 return wantarray ? ($sql, @all_bind) : $sql;
395 sub _update_set_values {
396 my ($self, $data) = @_;
398 my (@set, @all_bind);
399 for my $k (sort keys %$data) {
402 my $label = $self->_quote($k);
404 $self->_SWITCH_refkind($v, {
406 if ($self->{array_datatypes}) { # array datatype
407 push @set, "$label = ?";
408 push @all_bind, $self->_bindtype($k, $v);
410 else { # literal SQL with bind
411 my ($sql, @bind) = @$v;
412 $self->_assert_bindval_matches_bindtype(@bind);
413 push @set, "$label = $sql";
414 push @all_bind, @bind;
417 ARRAYREFREF => sub { # literal SQL with bind
418 my ($sql, @bind) = @${$v};
419 $self->_assert_bindval_matches_bindtype(@bind);
420 push @set, "$label = $sql";
421 push @all_bind, @bind;
423 SCALARREF => sub { # literal SQL without bind
424 push @set, "$label = $$v";
427 my ($op, $arg, @rest) = %$v;
429 puke 'Operator calls in update must be in the form { -op => $arg }'
430 if (@rest or not $op =~ /^\-(.+)/);
432 local $self->{_nested_func_lhs} = $k;
433 my ($sql, @bind) = $self->_where_unary_op($1, $arg);
435 push @set, "$label = $sql";
436 push @all_bind, @bind;
438 SCALAR_or_UNDEF => sub {
439 push @set, "$label = ?";
440 push @all_bind, $self->_bindtype($k, $v);
446 my $sql = join ', ', @set;
448 return ($sql, @all_bind);
451 # So that subclasses can override UPDATE ... RETURNING separately from
453 sub _update_returning { shift->_returning(@_) }
457 #======================================================================
459 #======================================================================
464 my ($table_sql, @table_bind) = $self->_table(shift);
465 my $fields = shift || '*';
469 my($where_sql, @where_bind) = $self->where($where, $order);
471 my $f = (ref $fields eq 'ARRAY') ? join ', ', map { $self->_quote($_) } @$fields
473 my $sql = join(' ', $self->_sqlcase('select'), $f,
474 $self->_sqlcase('from'), $table_sql)
477 return wantarray ? ($sql, @table_bind, @where_bind) : $sql;
480 #======================================================================
482 #======================================================================
487 my ($table_sql, @all_bind) = $self->_table(shift);
491 my($where_sql, @where_bind) = $self->where($where);
492 my $sql = $self->_sqlcase('delete from ') . $table_sql . $where_sql;
493 push @all_bind, @where_bind;
495 if ($options->{returning}) {
496 my ($returning_sql, @returning_bind) = $self->_delete_returning($options);
497 $sql .= $returning_sql;
498 push @all_bind, @returning_bind;
501 return wantarray ? ($sql, @all_bind) : $sql;
504 # So that subclasses can override DELETE ... RETURNING separately from
506 sub _delete_returning { shift->_returning(@_) }
510 #======================================================================
512 #======================================================================
516 # Finally, a separate routine just to handle WHERE clauses
518 my ($self, $where, $order) = @_;
521 my ($sql, @bind) = $self->_recurse_where($where);
522 $sql = $sql ? $self->_sqlcase(' where ') . "( $sql )" : '';
526 my ($order_sql, @order_bind) = $self->_order_by($order);
528 push @bind, @order_bind;
531 return wantarray ? ($sql, @bind) : $sql;
536 my ($self, $where, $logic) = @_;
538 # dispatch on appropriate method according to refkind of $where
539 my $method = $self->_METHOD_FOR_refkind("_where", $where);
541 my ($sql, @bind) = $self->$method($where, $logic);
543 # DBIx::Class used to call _recurse_where in scalar context
544 # something else might too...
546 return ($sql, @bind);
549 belch "Calling _recurse_where in scalar context is deprecated and will go away before 2.0";
556 #======================================================================
557 # WHERE: top-level ARRAYREF
558 #======================================================================
561 sub _where_ARRAYREF {
562 my ($self, $where, $logic) = @_;
564 $logic = uc($logic || $self->{logic});
565 $logic eq 'AND' or $logic eq 'OR' or puke "unknown logic: $logic";
567 my @clauses = @$where;
569 my (@sql_clauses, @all_bind);
570 # need to use while() so can shift() for pairs
572 my $el = shift @clauses;
574 $el = undef if (defined $el and ! length $el);
576 # switch according to kind of $el and get corresponding ($sql, @bind)
577 my ($sql, @bind) = $self->_SWITCH_refkind($el, {
579 # skip empty elements, otherwise get invalid trailing AND stuff
580 ARRAYREF => sub {$self->_recurse_where($el) if @$el},
584 $self->_assert_bindval_matches_bindtype(@b);
588 HASHREF => sub {$self->_recurse_where($el, 'and') if %$el},
590 SCALARREF => sub { ($$el); },
593 # top-level arrayref with scalars, recurse in pairs
594 $self->_recurse_where({$el => shift(@clauses)})
597 UNDEF => sub {puke "Supplying an empty left hand side argument is not supported in array-pairs" },
601 push @sql_clauses, $sql;
602 push @all_bind, @bind;
606 return $self->_join_sql_clauses($logic, \@sql_clauses, \@all_bind);
609 #======================================================================
610 # WHERE: top-level ARRAYREFREF
611 #======================================================================
613 sub _where_ARRAYREFREF {
614 my ($self, $where) = @_;
615 my ($sql, @bind) = @$$where;
616 $self->_assert_bindval_matches_bindtype(@bind);
617 return ($sql, @bind);
620 #======================================================================
621 # WHERE: top-level HASHREF
622 #======================================================================
625 my ($self, $where) = @_;
626 my (@sql_clauses, @all_bind);
628 for my $k (sort keys %$where) {
629 my $v = $where->{$k};
631 # ($k => $v) is either a special unary op or a regular hashpair
632 my ($sql, @bind) = do {
634 # put the operator in canonical form
636 $op = substr $op, 1; # remove initial dash
637 $op =~ s/^\s+|\s+$//g;# remove leading/trailing space
638 $op =~ s/\s+/ /g; # compress whitespace
640 # so that -not_foo works correctly
641 $op =~ s/^not_/NOT /i;
643 $self->_debug("Unary OP(-$op) within hashref, recursing...");
644 my ($s, @b) = $self->_where_unary_op($op, $v);
646 # top level vs nested
647 # we assume that handled unary ops will take care of their ()s
649 List::Util::first {$op =~ $_->{regex}} @{$self->{unary_ops}}
651 ( defined $self->{_nested_func_lhs} and $self->{_nested_func_lhs} eq $k )
657 if (is_literal_value ($v) ) {
658 belch 'Hash-pairs consisting of an empty string with a literal are deprecated, and will be removed in 2.0: use -and => [ $literal ] instead';
661 puke "Supplying an empty left hand side argument is not supported in hash-pairs";
665 my $method = $self->_METHOD_FOR_refkind("_where_hashpair", $v);
666 $self->$method($k, $v);
670 push @sql_clauses, $sql;
671 push @all_bind, @bind;
674 return $self->_join_sql_clauses('and', \@sql_clauses, \@all_bind);
677 sub _where_unary_op {
678 my ($self, $op, $rhs) = @_;
680 # top level special ops are illegal in general
681 # this includes the -ident/-value ops (dual purpose unary and special)
682 puke "Illegal use of top-level '-$op'"
683 if ! defined $self->{_nested_func_lhs} and List::Util::first { $op =~ $_->{regex} } @{$self->{special_ops}};
685 if (my $op_entry = List::Util::first { $op =~ $_->{regex} } @{$self->{unary_ops}}) {
686 my $handler = $op_entry->{handler};
688 if (not ref $handler) {
689 if ($op =~ s/ [_\s]? \d+ $//x ) {
690 belch 'Use of [and|or|nest]_N modifiers is deprecated and will be removed in SQLA v2.0. '
691 . "You probably wanted ...-and => [ -$op => COND1, -$op => COND2 ... ]";
693 return $self->$handler($op, $rhs);
695 elsif (ref $handler eq 'CODE') {
696 return $handler->($self, $op, $rhs);
699 puke "Illegal handler for operator $op - expecting a method name or a coderef";
703 $self->_debug("Generic unary OP: $op - recursing as function");
705 $self->_assert_pass_injection_guard($op);
707 my ($sql, @bind) = $self->_SWITCH_refkind($rhs, {
709 puke "Illegal use of top-level '-$op'"
710 unless defined $self->{_nested_func_lhs};
713 $self->_convert('?'),
714 $self->_bindtype($self->{_nested_func_lhs}, $rhs)
718 $self->_recurse_where($rhs)
722 $sql = sprintf('%s %s',
723 $self->_sqlcase($op),
727 return ($sql, @bind);
730 sub _where_op_ANDOR {
731 my ($self, $op, $v) = @_;
733 $self->_SWITCH_refkind($v, {
735 return $self->_where_ARRAYREF($v, $op);
739 return ($op =~ /^or/i)
740 ? $self->_where_ARRAYREF([ map { $_ => $v->{$_} } (sort keys %$v) ], $op)
741 : $self->_where_HASHREF($v);
745 puke "-$op => \\\$scalar makes little sense, use " .
747 ? '[ \$scalar, \%rest_of_conditions ] instead'
748 : '-and => [ \$scalar, \%rest_of_conditions ] instead'
753 puke "-$op => \\[...] makes little sense, use " .
755 ? '[ \[...], \%rest_of_conditions ] instead'
756 : '-and => [ \[...], \%rest_of_conditions ] instead'
760 SCALAR => sub { # permissively interpreted as SQL
761 puke "-$op => \$value makes little sense, use -bool => \$value instead";
765 puke "-$op => undef not supported";
771 my ($self, $op, $v) = @_;
773 $self->_SWITCH_refkind($v, {
775 SCALAR => sub { # permissively interpreted as SQL
776 belch "literal SQL should be -nest => \\'scalar' "
777 . "instead of -nest => 'scalar' ";
782 puke "-$op => undef not supported";
786 $self->_recurse_where($v);
794 my ($self, $op, $v) = @_;
796 my ($s, @b) = $self->_SWITCH_refkind($v, {
797 SCALAR => sub { # interpreted as SQL column
798 $self->_convert($self->_quote($v));
802 puke "-$op => undef not supported";
806 $self->_recurse_where($v);
810 $s = "(NOT $s)" if $op =~ /^not/i;
815 sub _where_op_IDENT {
817 my ($op, $rhs) = splice @_, -2;
818 if (! defined $rhs or length ref $rhs) {
819 puke "-$op requires a single plain scalar argument (a quotable identifier)";
822 # in case we are called as a top level special op (no '=')
825 $_ = $self->_convert($self->_quote($_)) for ($lhs, $rhs);
833 sub _where_op_VALUE {
835 my ($op, $rhs) = splice @_, -2;
837 # in case we are called as a top level special op (no '=')
841 if (! defined $rhs) {
843 ? $self->_convert($self->_quote($lhs)) . ' IS NULL'
850 (defined $lhs ? $lhs : $self->{_nested_func_lhs}),
857 $self->_convert($self->_quote($lhs)) . ' = ' . $self->_convert('?'),
861 $self->_convert('?'),
867 sub _where_hashpair_ARRAYREF {
868 my ($self, $k, $v) = @_;
871 my @v = @$v; # need copy because of shift below
872 $self->_debug("ARRAY($k) means distribute over elements");
874 # put apart first element if it is an operator (-and, -or)
876 (defined $v[0] && $v[0] =~ /^ - (?: AND|OR ) $/ix)
880 my @distributed = map { {$k => $_} } @v;
883 $self->_debug("OP($op) reinjected into the distributed array");
884 unshift @distributed, $op;
887 my $logic = $op ? substr($op, 1) : '';
889 return $self->_recurse_where(\@distributed, $logic);
892 $self->_debug("empty ARRAY($k) means 0=1");
893 return ($self->{sqlfalse});
897 sub _where_hashpair_HASHREF {
898 my ($self, $k, $v, $logic) = @_;
901 local $self->{_nested_func_lhs} = defined $self->{_nested_func_lhs}
902 ? $self->{_nested_func_lhs}
906 my ($all_sql, @all_bind);
908 for my $orig_op (sort keys %$v) {
909 my $val = $v->{$orig_op};
911 # put the operator in canonical form
914 # FIXME - we need to phase out dash-less ops
915 $op =~ s/^-//; # remove possible initial dash
916 $op =~ s/^\s+|\s+$//g;# remove leading/trailing space
917 $op =~ s/\s+/ /g; # compress whitespace
919 $self->_assert_pass_injection_guard($op);
922 $op =~ s/^is_not/IS NOT/i;
924 # so that -not_foo works correctly
925 $op =~ s/^not_/NOT /i;
927 # another retarded special case: foo => { $op => { -value => undef } }
928 if (ref $val eq 'HASH' and keys %$val == 1 and exists $val->{-value} and ! defined $val->{-value} ) {
934 # CASE: col-value logic modifiers
935 if ($orig_op =~ /^ \- (and|or) $/xi) {
936 ($sql, @bind) = $self->_where_hashpair_HASHREF($k, $val, $1);
938 # CASE: special operators like -in or -between
939 elsif (my $special_op = List::Util::first { $op =~ $_->{regex} } @{$self->{special_ops}}) {
940 my $handler = $special_op->{handler};
942 puke "No handler supplied for special operator $orig_op";
944 elsif (not ref $handler) {
945 ($sql, @bind) = $self->$handler($k, $op, $val);
947 elsif (ref $handler eq 'CODE') {
948 ($sql, @bind) = $handler->($self, $k, $op, $val);
951 puke "Illegal handler for special operator $orig_op - expecting a method name or a coderef";
955 $self->_SWITCH_refkind($val, {
957 ARRAYREF => sub { # CASE: col => {op => \@vals}
958 ($sql, @bind) = $self->_where_field_op_ARRAYREF($k, $op, $val);
961 ARRAYREFREF => sub { # CASE: col => {op => \[$sql, @bind]} (literal SQL with bind)
962 my ($sub_sql, @sub_bind) = @$$val;
963 $self->_assert_bindval_matches_bindtype(@sub_bind);
964 $sql = join ' ', $self->_convert($self->_quote($k)),
965 $self->_sqlcase($op),
970 UNDEF => sub { # CASE: col => {op => undef} : sql "IS (NOT)? NULL"
972 $op =~ /^not$/i ? 'is not' # legacy
973 : $op =~ $self->{equality_op} ? 'is'
974 : $op =~ $self->{like_op} ? belch("Supplying an undefined argument to '@{[ uc $op]}' is deprecated") && 'is'
975 : $op =~ $self->{inequality_op} ? 'is not'
976 : $op =~ $self->{not_like_op} ? belch("Supplying an undefined argument to '@{[ uc $op]}' is deprecated") && 'is not'
977 : puke "unexpected operator '$orig_op' with undef operand";
979 $sql = $self->_quote($k) . $self->_sqlcase(" $is null");
982 FALLBACK => sub { # CASE: col => {op/func => $stuff}
983 ($sql, @bind) = $self->_where_unary_op($op, $val);
986 $self->_convert($self->_quote($k)),
987 $self->{_nested_func_lhs} eq $k ? $sql : "($sql)", # top level vs nested
993 ($all_sql) = (defined $all_sql and $all_sql) ? $self->_join_sql_clauses($logic, [$all_sql, $sql], []) : $sql;
994 push @all_bind, @bind;
996 return ($all_sql, @all_bind);
999 sub _where_field_IS {
1000 my ($self, $k, $op, $v) = @_;
1002 my ($s) = $self->_SWITCH_refkind($v, {
1005 $self->_convert($self->_quote($k)),
1006 map { $self->_sqlcase($_)} ($op, 'null')
1009 puke "$op can only take undef as argument";
1016 sub _where_field_op_ARRAYREF {
1017 my ($self, $k, $op, $vals) = @_;
1019 my @vals = @$vals; #always work on a copy
1022 $self->_debug(sprintf '%s means multiple elements: [ %s ]',
1024 join(', ', map { defined $_ ? "'$_'" : 'NULL' } @vals ),
1027 # see if the first element is an -and/-or op
1029 if (defined $vals[0] && $vals[0] =~ /^ - (AND|OR) $/ix) {
1034 # a long standing API wart - an attempt to change this behavior during
1035 # the 1.50 series failed *spectacularly*. Warn instead and leave the
1040 (!$logic or $logic eq 'OR')
1042 ($op =~ $self->{inequality_op} or $op =~ $self->{not_like_op})
1045 belch "A multi-element arrayref as an argument to the inequality op '$o' "
1046 . 'is technically equivalent to an always-true 1=1 (you probably wanted '
1047 . "to say ...{ \$inequality_op => [ -and => \@values ] }... instead)"
1051 # distribute $op over each remaining member of @vals, append logic if exists
1052 return $self->_recurse_where([map { {$k => {$op, $_}} } @vals], $logic);
1056 # try to DWIM on equality operators
1058 $op =~ $self->{equality_op} ? $self->{sqlfalse}
1059 : $op =~ $self->{like_op} ? belch("Supplying an empty arrayref to '@{[ uc $op]}' is deprecated") && $self->{sqlfalse}
1060 : $op =~ $self->{inequality_op} ? $self->{sqltrue}
1061 : $op =~ $self->{not_like_op} ? belch("Supplying an empty arrayref to '@{[ uc $op]}' is deprecated") && $self->{sqltrue}
1062 : puke "operator '$op' applied on an empty array (field '$k')";
1067 sub _where_hashpair_SCALARREF {
1068 my ($self, $k, $v) = @_;
1069 $self->_debug("SCALAR($k) means literal SQL: $$v");
1070 my $sql = $self->_quote($k) . " " . $$v;
1074 # literal SQL with bind
1075 sub _where_hashpair_ARRAYREFREF {
1076 my ($self, $k, $v) = @_;
1077 $self->_debug("REF($k) means literal SQL: @${$v}");
1078 my ($sql, @bind) = @$$v;
1079 $self->_assert_bindval_matches_bindtype(@bind);
1080 $sql = $self->_quote($k) . " " . $sql;
1081 return ($sql, @bind );
1084 # literal SQL without bind
1085 sub _where_hashpair_SCALAR {
1086 my ($self, $k, $v) = @_;
1087 $self->_debug("NOREF($k) means simple key=val: $k $self->{cmp} $v");
1088 my $sql = join ' ', $self->_convert($self->_quote($k)),
1089 $self->_sqlcase($self->{cmp}),
1090 $self->_convert('?');
1091 my @bind = $self->_bindtype($k, $v);
1092 return ($sql, @bind);
1096 sub _where_hashpair_UNDEF {
1097 my ($self, $k, $v) = @_;
1098 $self->_debug("UNDEF($k) means IS NULL");
1099 my $sql = $self->_quote($k) . $self->_sqlcase(' is null');
1103 #======================================================================
1104 # WHERE: TOP-LEVEL OTHERS (SCALARREF, SCALAR, UNDEF)
1105 #======================================================================
1108 sub _where_SCALARREF {
1109 my ($self, $where) = @_;
1112 $self->_debug("SCALAR(*top) means literal SQL: $$where");
1118 my ($self, $where) = @_;
1121 $self->_debug("NOREF(*top) means literal SQL: $where");
1132 #======================================================================
1133 # WHERE: BUILTIN SPECIAL OPERATORS (-in, -between)
1134 #======================================================================
1137 sub _where_field_BETWEEN {
1138 my ($self, $k, $op, $vals) = @_;
1140 my ($label, $and, $placeholder);
1141 $label = $self->_convert($self->_quote($k));
1142 $and = ' ' . $self->_sqlcase('and') . ' ';
1143 $placeholder = $self->_convert('?');
1144 $op = $self->_sqlcase($op);
1146 my $invalid_args = "Operator '$op' requires either an arrayref with two defined values or expressions, or a single literal scalarref/arrayref-ref";
1148 my ($clause, @bind) = $self->_SWITCH_refkind($vals, {
1149 ARRAYREFREF => sub {
1150 my ($s, @b) = @$$vals;
1151 $self->_assert_bindval_matches_bindtype(@b);
1158 puke $invalid_args if @$vals != 2;
1160 my (@all_sql, @all_bind);
1161 foreach my $val (@$vals) {
1162 my ($sql, @bind) = $self->_SWITCH_refkind($val, {
1164 return ($placeholder, $self->_bindtype($k, $val) );
1169 ARRAYREFREF => sub {
1170 my ($sql, @bind) = @$$val;
1171 $self->_assert_bindval_matches_bindtype(@bind);
1172 return ($sql, @bind);
1175 my ($func, $arg, @rest) = %$val;
1176 puke "Only simple { -func => arg } functions accepted as sub-arguments to BETWEEN"
1177 if (@rest or $func !~ /^ \- (.+)/x);
1178 $self->_where_unary_op($1 => $arg);
1184 push @all_sql, $sql;
1185 push @all_bind, @bind;
1189 (join $and, @all_sql),
1198 my $sql = "( $label $op $clause )";
1199 return ($sql, @bind)
1203 sub _where_field_IN {
1204 my ($self, $k, $op, $vals) = @_;
1206 # backwards compatibility : if scalar, force into an arrayref
1207 $vals = [$vals] if defined $vals && ! ref $vals;
1209 my ($label) = $self->_convert($self->_quote($k));
1210 my ($placeholder) = $self->_convert('?');
1211 $op = $self->_sqlcase($op);
1213 my ($sql, @bind) = $self->_SWITCH_refkind($vals, {
1214 ARRAYREF => sub { # list of choices
1215 if (@$vals) { # nonempty list
1216 my (@all_sql, @all_bind);
1218 for my $val (@$vals) {
1219 my ($sql, @bind) = $self->_SWITCH_refkind($val, {
1221 return ($placeholder, $val);
1226 ARRAYREFREF => sub {
1227 my ($sql, @bind) = @$$val;
1228 $self->_assert_bindval_matches_bindtype(@bind);
1229 return ($sql, @bind);
1232 my ($func, $arg, @rest) = %$val;
1233 puke "Only simple { -func => arg } functions accepted as sub-arguments to IN"
1234 if (@rest or $func !~ /^ \- (.+)/x);
1235 $self->_where_unary_op($1 => $arg);
1239 'SQL::Abstract before v1.75 used to generate incorrect SQL when the '
1240 . "-$op operator was given an undef-containing list: !!!AUDIT YOUR CODE "
1241 . 'AND DATA!!! (the upcoming Data::Query-based version of SQL::Abstract '
1242 . 'will emit the logically correct SQL instead of raising this exception)'
1246 push @all_sql, $sql;
1247 push @all_bind, @bind;
1251 sprintf('%s %s ( %s )',
1254 join(', ', @all_sql)
1256 $self->_bindtype($k, @all_bind),
1259 else { # empty list : some databases won't understand "IN ()", so DWIM
1260 my $sql = ($op =~ /\bnot\b/i) ? $self->{sqltrue} : $self->{sqlfalse};
1265 SCALARREF => sub { # literal SQL
1266 my $sql = $self->_open_outer_paren($$vals);
1267 return ("$label $op ( $sql )");
1269 ARRAYREFREF => sub { # literal SQL with bind
1270 my ($sql, @bind) = @$$vals;
1271 $self->_assert_bindval_matches_bindtype(@bind);
1272 $sql = $self->_open_outer_paren($sql);
1273 return ("$label $op ( $sql )", @bind);
1277 puke "Argument passed to the '$op' operator can not be undefined";
1281 puke "special op $op requires an arrayref (or scalarref/arrayref-ref)";
1285 return ($sql, @bind);
1288 # Some databases (SQLite) treat col IN (1, 2) different from
1289 # col IN ( (1, 2) ). Use this to strip all outer parens while
1290 # adding them back in the corresponding method
1291 sub _open_outer_paren {
1292 my ($self, $sql) = @_;
1294 while (my ($inner) = $sql =~ /^ \s* \( (.*) \) \s* $/xs) {
1296 # there are closing parens inside, need the heavy duty machinery
1297 # to reevaluate the extraction starting from $sql (full reevaluation)
1298 if ($inner =~ /\)/) {
1299 require Text::Balanced;
1301 my (undef, $remainder) = do {
1302 # idiotic design - writes to $@ but *DOES NOT* throw exceptions
1304 Text::Balanced::extract_bracketed($sql, '()', qr/\s*/);
1307 # the entire expression needs to be a balanced bracketed thing
1308 # (after an extract no remainder sans trailing space)
1309 last if defined $remainder and $remainder =~ /\S/;
1319 #======================================================================
1321 #======================================================================
1324 my ($self, $arg) = @_;
1327 for my $c ($self->_order_by_chunks($arg) ) {
1328 $self->_SWITCH_refkind($c, {
1329 SCALAR => sub { push @sql, $c },
1330 ARRAYREF => sub { push @sql, shift @$c; push @bind, @$c },
1336 $self->_sqlcase(' order by'),
1342 return wantarray ? ($sql, @bind) : $sql;
1345 sub _order_by_chunks {
1346 my ($self, $arg) = @_;
1348 return $self->_SWITCH_refkind($arg, {
1351 map { $self->_order_by_chunks($_ ) } @$arg;
1354 ARRAYREFREF => sub {
1355 my ($s, @b) = @$$arg;
1356 $self->_assert_bindval_matches_bindtype(@b);
1360 SCALAR => sub {$self->_quote($arg)},
1362 UNDEF => sub {return () },
1364 SCALARREF => sub {$$arg}, # literal SQL, no quoting
1367 # get first pair in hash
1368 my ($key, $val, @rest) = %$arg;
1370 return () unless $key;
1372 if (@rest or not $key =~ /^-(desc|asc)/i) {
1373 puke "hash passed to _order_by must have exactly one key (-desc or -asc)";
1379 for my $c ($self->_order_by_chunks($val)) {
1382 $self->_SWITCH_refkind($c, {
1387 ($sql, @bind) = @$c;
1391 $sql = $sql . ' ' . $self->_sqlcase($direction);
1393 push @ret, [ $sql, @bind];
1402 #======================================================================
1403 # DATASOURCE (FOR NOW, JUST PLAIN TABLE OR LIST OF TABLES)
1404 #======================================================================
1409 $self->_SWITCH_refkind($from, {
1410 ARRAYREF => sub {join ', ', map { $self->_quote($_) } @$from;},
1411 ARRAYREFREF => sub {
1412 my ($sql, @bind) = @$$from;
1413 $self->_assert_bindval_matches_bindtype(@bind);
1414 return ($sql, @bind);
1416 SCALAR => sub {$self->_quote($from)},
1417 SCALARREF => sub {$$from},
1422 #======================================================================
1424 #======================================================================
1426 # highly optimized, as it's called way too often
1428 # my ($self, $label) = @_;
1430 return '' unless defined $_[1];
1431 return ${$_[1]} if ref($_[1]) eq 'SCALAR';
1433 $_[0]->{quote_char} or
1434 ($_[0]->_assert_pass_injection_guard($_[1]), return $_[1]);
1436 my $qref = ref $_[0]->{quote_char};
1438 !$qref ? ($_[0]->{quote_char}, $_[0]->{quote_char})
1439 : ($qref eq 'ARRAY') ? @{$_[0]->{quote_char}}
1440 : puke "Unsupported quote_char format: $_[0]->{quote_char}";
1442 my $esc = $_[0]->{escape_char} || $r;
1444 # parts containing * are naturally unquoted
1445 return join($_[0]->{name_sep}||'', map
1446 +( $_ eq '*' ? $_ : do { (my $n = $_) =~ s/(\Q$esc\E|\Q$r\E)/$esc$1/g; $l . $n . $r } ),
1447 ( $_[0]->{name_sep} ? split (/\Q$_[0]->{name_sep}\E/, $_[1] ) : $_[1] )
1452 # Conversion, if applicable
1454 #my ($self, $arg) = @_;
1455 if ($_[0]->{convert}) {
1456 return $_[0]->_sqlcase($_[0]->{convert}) .'(' . $_[1] . ')';
1463 #my ($self, $col, @vals) = @_;
1464 # called often - tighten code
1465 return $_[0]->{bindtype} eq 'columns'
1466 ? map {[$_[1], $_]} @_[2 .. $#_]
1471 # Dies if any element of @bind is not in [colname => value] format
1472 # if bindtype is 'columns'.
1473 sub _assert_bindval_matches_bindtype {
1474 # my ($self, @bind) = @_;
1476 if ($self->{bindtype} eq 'columns') {
1478 if (!defined $_ || ref($_) ne 'ARRAY' || @$_ != 2) {
1479 puke "bindtype 'columns' selected, you need to pass: [column_name => bind_value]"
1485 sub _join_sql_clauses {
1486 my ($self, $logic, $clauses_aref, $bind_aref) = @_;
1488 if (@$clauses_aref > 1) {
1489 my $join = " " . $self->_sqlcase($logic) . " ";
1490 my $sql = '( ' . join($join, @$clauses_aref) . ' )';
1491 return ($sql, @$bind_aref);
1493 elsif (@$clauses_aref) {
1494 return ($clauses_aref->[0], @$bind_aref); # no parentheses
1497 return (); # if no SQL, ignore @$bind_aref
1502 # Fix SQL case, if so requested
1504 # LDNOTE: if $self->{case} is true, then it contains 'lower', so we
1505 # don't touch the argument ... crooked logic, but let's not change it!
1506 return $_[0]->{case} ? $_[1] : uc($_[1]);
1510 #======================================================================
1511 # DISPATCHING FROM REFKIND
1512 #======================================================================
1515 my ($self, $data) = @_;
1517 return 'UNDEF' unless defined $data;
1519 # blessed objects are treated like scalars
1520 my $ref = (Scalar::Util::blessed $data) ? '' : ref $data;
1522 return 'SCALAR' unless $ref;
1525 while ($ref eq 'REF') {
1527 $ref = (Scalar::Util::blessed $data) ? '' : ref $data;
1531 return ($ref||'SCALAR') . ('REF' x $n_steps);
1535 my ($self, $data) = @_;
1536 my @try = ($self->_refkind($data));
1537 push @try, 'SCALAR_or_UNDEF' if $try[0] eq 'SCALAR' || $try[0] eq 'UNDEF';
1538 push @try, 'FALLBACK';
1542 sub _METHOD_FOR_refkind {
1543 my ($self, $meth_prefix, $data) = @_;
1546 for (@{$self->_try_refkind($data)}) {
1547 $method = $self->can($meth_prefix."_".$_)
1551 return $method || puke "cannot dispatch on '$meth_prefix' for ".$self->_refkind($data);
1555 sub _SWITCH_refkind {
1556 my ($self, $data, $dispatch_table) = @_;
1559 for (@{$self->_try_refkind($data)}) {
1560 $coderef = $dispatch_table->{$_}
1564 puke "no dispatch entry for ".$self->_refkind($data)
1573 #======================================================================
1574 # VALUES, GENERATE, AUTOLOAD
1575 #======================================================================
1577 # LDNOTE: original code from nwiger, didn't touch code in that section
1578 # I feel the AUTOLOAD stuff should not be the default, it should
1579 # only be activated on explicit demand by user.
1583 my $data = shift || return;
1584 puke "Argument to ", __PACKAGE__, "->values must be a \\%hash"
1585 unless ref $data eq 'HASH';
1588 foreach my $k (sort keys %$data) {
1589 my $v = $data->{$k};
1590 $self->_SWITCH_refkind($v, {
1592 if ($self->{array_datatypes}) { # array datatype
1593 push @all_bind, $self->_bindtype($k, $v);
1595 else { # literal SQL with bind
1596 my ($sql, @bind) = @$v;
1597 $self->_assert_bindval_matches_bindtype(@bind);
1598 push @all_bind, @bind;
1601 ARRAYREFREF => sub { # literal SQL with bind
1602 my ($sql, @bind) = @${$v};
1603 $self->_assert_bindval_matches_bindtype(@bind);
1604 push @all_bind, @bind;
1606 SCALARREF => sub { # literal SQL without bind
1608 SCALAR_or_UNDEF => sub {
1609 push @all_bind, $self->_bindtype($k, $v);
1620 my(@sql, @sqlq, @sqlv);
1624 if ($ref eq 'HASH') {
1625 for my $k (sort keys %$_) {
1628 my $label = $self->_quote($k);
1629 if ($r eq 'ARRAY') {
1630 # literal SQL with bind
1631 my ($sql, @bind) = @$v;
1632 $self->_assert_bindval_matches_bindtype(@bind);
1633 push @sqlq, "$label = $sql";
1635 } elsif ($r eq 'SCALAR') {
1636 # literal SQL without bind
1637 push @sqlq, "$label = $$v";
1639 push @sqlq, "$label = ?";
1640 push @sqlv, $self->_bindtype($k, $v);
1643 push @sql, $self->_sqlcase('set'), join ', ', @sqlq;
1644 } elsif ($ref eq 'ARRAY') {
1645 # unlike insert(), assume these are ONLY the column names, i.e. for SQL
1648 if ($r eq 'ARRAY') { # literal SQL with bind
1649 my ($sql, @bind) = @$v;
1650 $self->_assert_bindval_matches_bindtype(@bind);
1653 } elsif ($r eq 'SCALAR') { # literal SQL without bind
1654 # embedded literal SQL
1661 push @sql, '(' . join(', ', @sqlq) . ')';
1662 } elsif ($ref eq 'SCALAR') {
1666 # strings get case twiddled
1667 push @sql, $self->_sqlcase($_);
1671 my $sql = join ' ', @sql;
1673 # this is pretty tricky
1674 # if ask for an array, return ($stmt, @bind)
1675 # otherwise, s/?/shift @sqlv/ to put it inline
1677 return ($sql, @sqlv);
1679 1 while $sql =~ s/\?/my $d = shift(@sqlv);
1680 ref $d ? $d->[1] : $d/e;
1689 # This allows us to check for a local, then _form, attr
1691 my($name) = $AUTOLOAD =~ /.*::(.+)/;
1692 return $self->generate($name, @_);
1703 SQL::Abstract - Generate SQL from Perl data structures
1709 my $sql = SQL::Abstract->new;
1711 my($stmt, @bind) = $sql->select($source, \@fields, \%where, $order);
1713 my($stmt, @bind) = $sql->insert($table, \%fieldvals || \@values);
1715 my($stmt, @bind) = $sql->update($table, \%fieldvals, \%where);
1717 my($stmt, @bind) = $sql->delete($table, \%where);
1719 # Then, use these in your DBI statements
1720 my $sth = $dbh->prepare($stmt);
1721 $sth->execute(@bind);
1723 # Just generate the WHERE clause
1724 my($stmt, @bind) = $sql->where(\%where, $order);
1726 # Return values in the same order, for hashed queries
1727 # See PERFORMANCE section for more details
1728 my @bind = $sql->values(\%fieldvals);
1732 This module was inspired by the excellent L<DBIx::Abstract>.
1733 However, in using that module I found that what I really wanted
1734 to do was generate SQL, but still retain complete control over my
1735 statement handles and use the DBI interface. So, I set out to
1736 create an abstract SQL generation module.
1738 While based on the concepts used by L<DBIx::Abstract>, there are
1739 several important differences, especially when it comes to WHERE
1740 clauses. I have modified the concepts used to make the SQL easier
1741 to generate from Perl data structures and, IMO, more intuitive.
1742 The underlying idea is for this module to do what you mean, based
1743 on the data structures you provide it. The big advantage is that
1744 you don't have to modify your code every time your data changes,
1745 as this module figures it out.
1747 To begin with, an SQL INSERT is as easy as just specifying a hash
1748 of C<key=value> pairs:
1751 name => 'Jimbo Bobson',
1752 phone => '123-456-7890',
1753 address => '42 Sister Lane',
1754 city => 'St. Louis',
1755 state => 'Louisiana',
1758 The SQL can then be generated with this:
1760 my($stmt, @bind) = $sql->insert('people', \%data);
1762 Which would give you something like this:
1764 $stmt = "INSERT INTO people
1765 (address, city, name, phone, state)
1766 VALUES (?, ?, ?, ?, ?)";
1767 @bind = ('42 Sister Lane', 'St. Louis', 'Jimbo Bobson',
1768 '123-456-7890', 'Louisiana');
1770 These are then used directly in your DBI code:
1772 my $sth = $dbh->prepare($stmt);
1773 $sth->execute(@bind);
1775 =head2 Inserting and Updating Arrays
1777 If your database has array types (like for example Postgres),
1778 activate the special option C<< array_datatypes => 1 >>
1779 when creating the C<SQL::Abstract> object.
1780 Then you may use an arrayref to insert and update database array types:
1782 my $sql = SQL::Abstract->new(array_datatypes => 1);
1784 planets => [qw/Mercury Venus Earth Mars/]
1787 my($stmt, @bind) = $sql->insert('solar_system', \%data);
1791 $stmt = "INSERT INTO solar_system (planets) VALUES (?)"
1793 @bind = (['Mercury', 'Venus', 'Earth', 'Mars']);
1796 =head2 Inserting and Updating SQL
1798 In order to apply SQL functions to elements of your C<%data> you may
1799 specify a reference to an arrayref for the given hash value. For example,
1800 if you need to execute the Oracle C<to_date> function on a value, you can
1801 say something like this:
1805 date_entered => \[ "to_date(?,'MM/DD/YYYY')", "03/02/2003" ],
1808 The first value in the array is the actual SQL. Any other values are
1809 optional and would be included in the bind values array. This gives
1812 my($stmt, @bind) = $sql->insert('people', \%data);
1814 $stmt = "INSERT INTO people (name, date_entered)
1815 VALUES (?, to_date(?,'MM/DD/YYYY'))";
1816 @bind = ('Bill', '03/02/2003');
1818 An UPDATE is just as easy, all you change is the name of the function:
1820 my($stmt, @bind) = $sql->update('people', \%data);
1822 Notice that your C<%data> isn't touched; the module will generate
1823 the appropriately quirky SQL for you automatically. Usually you'll
1824 want to specify a WHERE clause for your UPDATE, though, which is
1825 where handling C<%where> hashes comes in handy...
1827 =head2 Complex where statements
1829 This module can generate pretty complicated WHERE statements
1830 easily. For example, simple C<key=value> pairs are taken to mean
1831 equality, and if you want to see if a field is within a set
1832 of values, you can use an arrayref. Let's say we wanted to
1833 SELECT some data based on this criteria:
1836 requestor => 'inna',
1837 worker => ['nwiger', 'rcwe', 'sfz'],
1838 status => { '!=', 'completed' }
1841 my($stmt, @bind) = $sql->select('tickets', '*', \%where);
1843 The above would give you something like this:
1845 $stmt = "SELECT * FROM tickets WHERE
1846 ( requestor = ? ) AND ( status != ? )
1847 AND ( worker = ? OR worker = ? OR worker = ? )";
1848 @bind = ('inna', 'completed', 'nwiger', 'rcwe', 'sfz');
1850 Which you could then use in DBI code like so:
1852 my $sth = $dbh->prepare($stmt);
1853 $sth->execute(@bind);
1859 The methods are simple. There's one for every major SQL operation,
1860 and a constructor you use first. The arguments are specified in a
1861 similar order for each method (table, then fields, then a where
1862 clause) to try and simplify things.
1864 =head2 new(option => 'value')
1866 The C<new()> function takes a list of options and values, and returns
1867 a new B<SQL::Abstract> object which can then be used to generate SQL
1868 through the methods below. The options accepted are:
1874 If set to 'lower', then SQL will be generated in all lowercase. By
1875 default SQL is generated in "textbook" case meaning something like:
1877 SELECT a_field FROM a_table WHERE some_field LIKE '%someval%'
1879 Any setting other than 'lower' is ignored.
1883 This determines what the default comparison operator is. By default
1884 it is C<=>, meaning that a hash like this:
1886 %where = (name => 'nwiger', email => 'nate@wiger.org');
1888 Will generate SQL like this:
1890 WHERE name = 'nwiger' AND email = 'nate@wiger.org'
1892 However, you may want loose comparisons by default, so if you set
1893 C<cmp> to C<like> you would get SQL such as:
1895 WHERE name like 'nwiger' AND email like 'nate@wiger.org'
1897 You can also override the comparison on an individual basis - see
1898 the huge section on L</"WHERE CLAUSES"> at the bottom.
1900 =item sqltrue, sqlfalse
1902 Expressions for inserting boolean values within SQL statements.
1903 By default these are C<1=1> and C<1=0>. They are used
1904 by the special operators C<-in> and C<-not_in> for generating
1905 correct SQL even when the argument is an empty array (see below).
1909 This determines the default logical operator for multiple WHERE
1910 statements in arrays or hashes. If absent, the default logic is "or"
1911 for arrays, and "and" for hashes. This means that a WHERE
1915 event_date => {'>=', '2/13/99'},
1916 event_date => {'<=', '4/24/03'},
1919 will generate SQL like this:
1921 WHERE event_date >= '2/13/99' OR event_date <= '4/24/03'
1923 This is probably not what you want given this query, though (look
1924 at the dates). To change the "OR" to an "AND", simply specify:
1926 my $sql = SQL::Abstract->new(logic => 'and');
1928 Which will change the above C<WHERE> to:
1930 WHERE event_date >= '2/13/99' AND event_date <= '4/24/03'
1932 The logic can also be changed locally by inserting
1933 a modifier in front of an arrayref :
1935 @where = (-and => [event_date => {'>=', '2/13/99'},
1936 event_date => {'<=', '4/24/03'} ]);
1938 See the L</"WHERE CLAUSES"> section for explanations.
1942 This will automatically convert comparisons using the specified SQL
1943 function for both column and value. This is mostly used with an argument
1944 of C<upper> or C<lower>, so that the SQL will have the effect of
1945 case-insensitive "searches". For example, this:
1947 $sql = SQL::Abstract->new(convert => 'upper');
1948 %where = (keywords => 'MaKe iT CAse inSeNSItive');
1950 Will turn out the following SQL:
1952 WHERE upper(keywords) like upper('MaKe iT CAse inSeNSItive')
1954 The conversion can be C<upper()>, C<lower()>, or any other SQL function
1955 that can be applied symmetrically to fields (actually B<SQL::Abstract> does
1956 not validate this option; it will just pass through what you specify verbatim).
1960 This is a kludge because many databases suck. For example, you can't
1961 just bind values using DBI's C<execute()> for Oracle C<CLOB> or C<BLOB> fields.
1962 Instead, you have to use C<bind_param()>:
1964 $sth->bind_param(1, 'reg data');
1965 $sth->bind_param(2, $lots, {ora_type => ORA_CLOB});
1967 The problem is, B<SQL::Abstract> will normally just return a C<@bind> array,
1968 which loses track of which field each slot refers to. Fear not.
1970 If you specify C<bindtype> in new, you can determine how C<@bind> is returned.
1971 Currently, you can specify either C<normal> (default) or C<columns>. If you
1972 specify C<columns>, you will get an array that looks like this:
1974 my $sql = SQL::Abstract->new(bindtype => 'columns');
1975 my($stmt, @bind) = $sql->insert(...);
1978 [ 'column1', 'value1' ],
1979 [ 'column2', 'value2' ],
1980 [ 'column3', 'value3' ],
1983 You can then iterate through this manually, using DBI's C<bind_param()>.
1985 $sth->prepare($stmt);
1988 my($col, $data) = @$_;
1989 if ($col eq 'details' || $col eq 'comments') {
1990 $sth->bind_param($i, $data, {ora_type => ORA_CLOB});
1991 } elsif ($col eq 'image') {
1992 $sth->bind_param($i, $data, {ora_type => ORA_BLOB});
1994 $sth->bind_param($i, $data);
1998 $sth->execute; # execute without @bind now
2000 Now, why would you still use B<SQL::Abstract> if you have to do this crap?
2001 Basically, the advantage is still that you don't have to care which fields
2002 are or are not included. You could wrap that above C<for> loop in a simple
2003 sub called C<bind_fields()> or something and reuse it repeatedly. You still
2004 get a layer of abstraction over manual SQL specification.
2006 Note that if you set L</bindtype> to C<columns>, the C<\[ $sql, @bind ]>
2007 construct (see L</Literal SQL with placeholders and bind values (subqueries)>)
2008 will expect the bind values in this format.
2012 This is the character that a table or column name will be quoted
2013 with. By default this is an empty string, but you could set it to
2014 the character C<`>, to generate SQL like this:
2016 SELECT `a_field` FROM `a_table` WHERE `some_field` LIKE '%someval%'
2018 Alternatively, you can supply an array ref of two items, the first being the left
2019 hand quote character, and the second the right hand quote character. For
2020 example, you could supply C<['[',']']> for SQL Server 2000 compliant quotes
2021 that generates SQL like this:
2023 SELECT [a_field] FROM [a_table] WHERE [some_field] LIKE '%someval%'
2025 Quoting is useful if you have tables or columns names that are reserved
2026 words in your database's SQL dialect.
2030 This is the character that will be used to escape L</quote_char>s appearing
2031 in an identifier before it has been quoted.
2033 The parameter default in case of a single L</quote_char> character is the quote
2036 When opening-closing-style quoting is used (L</quote_char> is an arrayref)
2037 this parameter defaults to the B<closing (right)> L</quote_char>. Occurrences
2038 of the B<opening (left)> L</quote_char> within the identifier are currently left
2039 untouched. The default for opening-closing-style quotes may change in future
2040 versions, thus you are B<strongly encouraged> to specify the escape character
2045 This is the character that separates a table and column name. It is
2046 necessary to specify this when the C<quote_char> option is selected,
2047 so that tables and column names can be individually quoted like this:
2049 SELECT `table`.`one_field` FROM `table` WHERE `table`.`other_field` = 1
2051 =item injection_guard
2053 A regular expression C<qr/.../> that is applied to any C<-function> and unquoted
2054 column name specified in a query structure. This is a safety mechanism to avoid
2055 injection attacks when mishandling user input e.g.:
2057 my %condition_as_column_value_pairs = get_values_from_user();
2058 $sqla->select( ... , \%condition_as_column_value_pairs );
2060 If the expression matches an exception is thrown. Note that literal SQL
2061 supplied via C<\'...'> or C<\['...']> is B<not> checked in any way.
2063 Defaults to checking for C<;> and the C<GO> keyword (TransactSQL)
2065 =item array_datatypes
2067 When this option is true, arrayrefs in INSERT or UPDATE are
2068 interpreted as array datatypes and are passed directly
2070 When this option is false, arrayrefs are interpreted
2071 as literal SQL, just like refs to arrayrefs
2072 (but this behavior is for backwards compatibility; when writing
2073 new queries, use the "reference to arrayref" syntax
2079 Takes a reference to a list of "special operators"
2080 to extend the syntax understood by L<SQL::Abstract>.
2081 See section L</"SPECIAL OPERATORS"> for details.
2085 Takes a reference to a list of "unary operators"
2086 to extend the syntax understood by L<SQL::Abstract>.
2087 See section L</"UNARY OPERATORS"> for details.
2093 =head2 insert($table, \@values || \%fieldvals, \%options)
2095 This is the simplest function. You simply give it a table name
2096 and either an arrayref of values or hashref of field/value pairs.
2097 It returns an SQL INSERT statement and a list of bind values.
2098 See the sections on L</"Inserting and Updating Arrays"> and
2099 L</"Inserting and Updating SQL"> for information on how to insert
2100 with those data types.
2102 The optional C<\%options> hash reference may contain additional
2103 options to generate the insert SQL. Currently supported options
2110 Takes either a scalar of raw SQL fields, or an array reference of
2111 field names, and adds on an SQL C<RETURNING> statement at the end.
2112 This allows you to return data generated by the insert statement
2113 (such as row IDs) without performing another C<SELECT> statement.
2114 Note, however, this is not part of the SQL standard and may not
2115 be supported by all database engines.
2119 =head2 update($table, \%fieldvals, \%where, \%options)
2121 This takes a table, hashref of field/value pairs, and an optional
2122 hashref L<WHERE clause|/WHERE CLAUSES>. It returns an SQL UPDATE function and a list
2124 See the sections on L</"Inserting and Updating Arrays"> and
2125 L</"Inserting and Updating SQL"> for information on how to insert
2126 with those data types.
2128 The optional C<\%options> hash reference may contain additional
2129 options to generate the update SQL. Currently supported options
2136 See the C<returning> option to
2137 L<insert|/insert($table, \@values || \%fieldvals, \%options)>.
2141 =head2 select($source, $fields, $where, $order)
2143 This returns a SQL SELECT statement and associated list of bind values, as
2144 specified by the arguments :
2150 Specification of the 'FROM' part of the statement.
2151 The argument can be either a plain scalar (interpreted as a table
2152 name, will be quoted), or an arrayref (interpreted as a list
2153 of table names, joined by commas, quoted), or a scalarref
2154 (literal SQL, not quoted), or a reference to an arrayref
2155 (literal SQL and bind values).
2159 Specification of the list of fields to retrieve from
2161 The argument can be either an arrayref (interpreted as a list
2162 of field names, will be joined by commas and quoted), or a
2163 plain scalar (literal SQL, not quoted).
2164 Please observe that this API is not as flexible as that of
2165 the first argument C<$source>, for backwards compatibility reasons.
2169 Optional argument to specify the WHERE part of the query.
2170 The argument is most often a hashref, but can also be
2171 an arrayref or plain scalar --
2172 see section L<WHERE clause|/"WHERE CLAUSES"> for details.
2176 Optional argument to specify the ORDER BY part of the query.
2177 The argument can be a scalar, a hashref or an arrayref
2178 -- see section L<ORDER BY clause|/"ORDER BY CLAUSES">
2184 =head2 delete($table, \%where, \%options)
2186 This takes a table name and optional hashref L<WHERE clause|/WHERE CLAUSES>.
2187 It returns an SQL DELETE statement and list of bind values.
2189 The optional C<\%options> hash reference may contain additional
2190 options to generate the delete SQL. Currently supported options
2197 See the C<returning> option to
2198 L<insert|/insert($table, \@values || \%fieldvals, \%options)>.
2202 =head2 where(\%where, $order)
2204 This is used to generate just the WHERE clause. For example,
2205 if you have an arbitrary data structure and know what the
2206 rest of your SQL is going to look like, but want an easy way
2207 to produce a WHERE clause, use this. It returns an SQL WHERE
2208 clause and list of bind values.
2211 =head2 values(\%data)
2213 This just returns the values from the hash C<%data>, in the same
2214 order that would be returned from any of the other above queries.
2215 Using this allows you to markedly speed up your queries if you
2216 are affecting lots of rows. See below under the L</"PERFORMANCE"> section.
2218 =head2 generate($any, 'number', $of, \@data, $struct, \%types)
2220 Warning: This is an experimental method and subject to change.
2222 This returns arbitrarily generated SQL. It's a really basic shortcut.
2223 It will return two different things, depending on return context:
2225 my($stmt, @bind) = $sql->generate('create table', \$table, \@fields);
2226 my $stmt_and_val = $sql->generate('create table', \$table, \@fields);
2228 These would return the following:
2230 # First calling form
2231 $stmt = "CREATE TABLE test (?, ?)";
2232 @bind = (field1, field2);
2234 # Second calling form
2235 $stmt_and_val = "CREATE TABLE test (field1, field2)";
2237 Depending on what you're trying to do, it's up to you to choose the correct
2238 format. In this example, the second form is what you would want.
2242 $sql->generate('alter session', { nls_date_format => 'MM/YY' });
2246 ALTER SESSION SET nls_date_format = 'MM/YY'
2248 You get the idea. Strings get their case twiddled, but everything
2249 else remains verbatim.
2251 =head1 EXPORTABLE FUNCTIONS
2253 =head2 is_plain_value
2255 Determines if the supplied argument is a plain value as understood by this
2260 =item * The value is C<undef>
2262 =item * The value is a non-reference
2264 =item * The value is an object with stringification overloading
2266 =item * The value is of the form C<< { -value => $anything } >>
2270 On failure returns C<undef>, on success returns a B<scalar> reference
2271 to the original supplied argument.
2277 The stringification overloading detection is rather advanced: it takes
2278 into consideration not only the presence of a C<""> overload, but if that
2279 fails also checks for enabled
2280 L<autogenerated versions of C<"">|overload/Magic Autogeneration>, based
2281 on either C<0+> or C<bool>.
2283 Unfortunately testing in the field indicates that this
2284 detection B<< may tickle a latent bug in perl versions before 5.018 >>,
2285 but only when very large numbers of stringifying objects are involved.
2286 At the time of writing ( Sep 2014 ) there is no clear explanation of
2287 the direct cause, nor is there a manageably small test case that reliably
2288 reproduces the problem.
2290 If you encounter any of the following exceptions in B<random places within
2291 your application stack> - this module may be to blame:
2293 Operation "ne": no method found,
2294 left argument in overloaded package <something>,
2295 right argument in overloaded package <something>
2299 Stub found while resolving method "???" overloading """" in package <something>
2301 If you fall victim to the above - please attempt to reduce the problem
2302 to something that could be sent to the L<SQL::Abstract developers
2303 |DBIx::Class/GETTING HELP/SUPPORT>
2304 (either publicly or privately). As a workaround in the meantime you can
2305 set C<$ENV{SQLA_ISVALUE_IGNORE_AUTOGENERATED_STRINGIFICATION}> to a true
2306 value, which will most likely eliminate your problem (at the expense of
2307 not being able to properly detect exotic forms of stringification).
2309 This notice and environment variable will be removed in a future version,
2310 as soon as the underlying problem is found and a reliable workaround is
2315 =head2 is_literal_value
2317 Determines if the supplied argument is a literal value as understood by this
2322 =item * C<\$sql_string>
2324 =item * C<\[ $sql_string, @bind_values ]>
2328 On failure returns C<undef>, on success returns an B<array> reference
2329 containing the unpacked version of the supplied literal SQL and bind values.
2331 =head1 WHERE CLAUSES
2335 This module uses a variation on the idea from L<DBIx::Abstract>. It
2336 is B<NOT>, repeat I<not> 100% compatible. B<The main logic of this
2337 module is that things in arrays are OR'ed, and things in hashes
2340 The easiest way to explain is to show lots of examples. After
2341 each C<%where> hash shown, it is assumed you used:
2343 my($stmt, @bind) = $sql->where(\%where);
2345 However, note that the C<%where> hash can be used directly in any
2346 of the other functions as well, as described above.
2348 =head2 Key-value pairs
2350 So, let's get started. To begin, a simple hash:
2354 status => 'completed'
2357 Is converted to SQL C<key = val> statements:
2359 $stmt = "WHERE user = ? AND status = ?";
2360 @bind = ('nwiger', 'completed');
2362 One common thing I end up doing is having a list of values that
2363 a field can be in. To do this, simply specify a list inside of
2368 status => ['assigned', 'in-progress', 'pending'];
2371 This simple code will create the following:
2373 $stmt = "WHERE user = ? AND ( status = ? OR status = ? OR status = ? )";
2374 @bind = ('nwiger', 'assigned', 'in-progress', 'pending');
2376 A field associated to an empty arrayref will be considered a
2377 logical false and will generate 0=1.
2379 =head2 Tests for NULL values
2381 If the value part is C<undef> then this is converted to SQL <IS NULL>
2390 $stmt = "WHERE user = ? AND status IS NULL";
2393 To test if a column IS NOT NULL:
2397 status => { '!=', undef },
2400 =head2 Specific comparison operators
2402 If you want to specify a different type of operator for your comparison,
2403 you can use a hashref for a given column:
2407 status => { '!=', 'completed' }
2410 Which would generate:
2412 $stmt = "WHERE user = ? AND status != ?";
2413 @bind = ('nwiger', 'completed');
2415 To test against multiple values, just enclose the values in an arrayref:
2417 status => { '=', ['assigned', 'in-progress', 'pending'] };
2419 Which would give you:
2421 "WHERE status = ? OR status = ? OR status = ?"
2424 The hashref can also contain multiple pairs, in which case it is expanded
2425 into an C<AND> of its elements:
2429 status => { '!=', 'completed', -not_like => 'pending%' }
2432 # Or more dynamically, like from a form
2433 $where{user} = 'nwiger';
2434 $where{status}{'!='} = 'completed';
2435 $where{status}{'-not_like'} = 'pending%';
2437 # Both generate this
2438 $stmt = "WHERE user = ? AND status != ? AND status NOT LIKE ?";
2439 @bind = ('nwiger', 'completed', 'pending%');
2442 To get an OR instead, you can combine it with the arrayref idea:
2446 priority => [ { '=', 2 }, { '>', 5 } ]
2449 Which would generate:
2451 $stmt = "WHERE ( priority = ? OR priority > ? ) AND user = ?";
2452 @bind = ('2', '5', 'nwiger');
2454 If you want to include literal SQL (with or without bind values), just use a
2455 scalar reference or reference to an arrayref as the value:
2458 date_entered => { '>' => \["to_date(?, 'MM/DD/YYYY')", "11/26/2008"] },
2459 date_expires => { '<' => \"now()" }
2462 Which would generate:
2464 $stmt = "WHERE date_entered > to_date(?, 'MM/DD/YYYY') AND date_expires < now()";
2465 @bind = ('11/26/2008');
2468 =head2 Logic and nesting operators
2470 In the example above,
2471 there is a subtle trap if you want to say something like
2472 this (notice the C<AND>):
2474 WHERE priority != ? AND priority != ?
2476 Because, in Perl you I<can't> do this:
2478 priority => { '!=' => 2, '!=' => 1 }
2480 As the second C<!=> key will obliterate the first. The solution
2481 is to use the special C<-modifier> form inside an arrayref:
2483 priority => [ -and => {'!=', 2},
2487 Normally, these would be joined by C<OR>, but the modifier tells it
2488 to use C<AND> instead. (Hint: You can use this in conjunction with the
2489 C<logic> option to C<new()> in order to change the way your queries
2490 work by default.) B<Important:> Note that the C<-modifier> goes
2491 B<INSIDE> the arrayref, as an extra first element. This will
2492 B<NOT> do what you think it might:
2494 priority => -and => [{'!=', 2}, {'!=', 1}] # WRONG!
2496 Here is a quick list of equivalencies, since there is some overlap:
2499 status => {'!=', 'completed', 'not like', 'pending%' }
2500 status => [ -and => {'!=', 'completed'}, {'not like', 'pending%'}]
2503 status => {'=', ['assigned', 'in-progress']}
2504 status => [ -or => {'=', 'assigned'}, {'=', 'in-progress'}]
2505 status => [ {'=', 'assigned'}, {'=', 'in-progress'} ]
2509 =head2 Special operators : IN, BETWEEN, etc.
2511 You can also use the hashref format to compare a list of fields using the
2512 C<IN> comparison operator, by specifying the list as an arrayref:
2515 status => 'completed',
2516 reportid => { -in => [567, 2335, 2] }
2519 Which would generate:
2521 $stmt = "WHERE status = ? AND reportid IN (?,?,?)";
2522 @bind = ('completed', '567', '2335', '2');
2524 The reverse operator C<-not_in> generates SQL C<NOT IN> and is used in
2527 If the argument to C<-in> is an empty array, 'sqlfalse' is generated
2528 (by default : C<1=0>). Similarly, C<< -not_in => [] >> generates
2529 'sqltrue' (by default : C<1=1>).
2531 In addition to the array you can supply a chunk of literal sql or
2532 literal sql with bind:
2535 customer => { -in => \[
2536 'SELECT cust_id FROM cust WHERE balance > ?',
2539 status => { -in => \'SELECT status_codes FROM states' },
2545 customer IN ( SELECT cust_id FROM cust WHERE balance > ? )
2546 AND status IN ( SELECT status_codes FROM states )
2550 Finally, if the argument to C<-in> is not a reference, it will be
2551 treated as a single-element array.
2553 Another pair of operators is C<-between> and C<-not_between>,
2554 used with an arrayref of two values:
2558 completion_date => {
2559 -not_between => ['2002-10-01', '2003-02-06']
2565 WHERE user = ? AND completion_date NOT BETWEEN ( ? AND ? )
2567 Just like with C<-in> all plausible combinations of literal SQL
2571 start0 => { -between => [ 1, 2 ] },
2572 start1 => { -between => \["? AND ?", 1, 2] },
2573 start2 => { -between => \"lower(x) AND upper(y)" },
2574 start3 => { -between => [
2576 \["upper(?)", 'stuff' ],
2583 ( start0 BETWEEN ? AND ? )
2584 AND ( start1 BETWEEN ? AND ? )
2585 AND ( start2 BETWEEN lower(x) AND upper(y) )
2586 AND ( start3 BETWEEN lower(x) AND upper(?) )
2588 @bind = (1, 2, 1, 2, 'stuff');
2591 These are the two builtin "special operators"; but the
2592 list can be expanded : see section L</"SPECIAL OPERATORS"> below.
2594 =head2 Unary operators: bool
2596 If you wish to test against boolean columns or functions within your
2597 database you can use the C<-bool> and C<-not_bool> operators. For
2598 example to test the column C<is_user> being true and the column
2599 C<is_enabled> being false you would use:-
2603 -not_bool => 'is_enabled',
2608 WHERE is_user AND NOT is_enabled
2610 If a more complex combination is required, testing more conditions,
2611 then you should use the and/or operators:-
2616 -not_bool => { two=> { -rlike => 'bar' } },
2617 -not_bool => { three => [ { '=', 2 }, { '>', 5 } ] },
2628 (NOT ( three = ? OR three > ? ))
2631 =head2 Nested conditions, -and/-or prefixes
2633 So far, we've seen how multiple conditions are joined with a top-level
2634 C<AND>. We can change this by putting the different conditions we want in
2635 hashes and then putting those hashes in an array. For example:
2640 status => { -like => ['pending%', 'dispatched'] },
2644 status => 'unassigned',
2648 This data structure would create the following:
2650 $stmt = "WHERE ( user = ? AND ( status LIKE ? OR status LIKE ? ) )
2651 OR ( user = ? AND status = ? ) )";
2652 @bind = ('nwiger', 'pending', 'dispatched', 'robot', 'unassigned');
2655 Clauses in hashrefs or arrayrefs can be prefixed with an C<-and> or C<-or>
2656 to change the logic inside :
2662 -and => [ workhrs => {'>', 20}, geo => 'ASIA' ],
2663 -or => { workhrs => {'<', 50}, geo => 'EURO' },
2670 $stmt = "WHERE ( user = ?
2671 AND ( ( workhrs > ? AND geo = ? )
2672 OR ( workhrs < ? OR geo = ? ) ) )";
2673 @bind = ('nwiger', '20', 'ASIA', '50', 'EURO');
2675 =head3 Algebraic inconsistency, for historical reasons
2677 C<Important note>: when connecting several conditions, the C<-and->|C<-or>
2678 operator goes C<outside> of the nested structure; whereas when connecting
2679 several constraints on one column, the C<-and> operator goes
2680 C<inside> the arrayref. Here is an example combining both features :
2683 -and => [a => 1, b => 2],
2684 -or => [c => 3, d => 4],
2685 e => [-and => {-like => 'foo%'}, {-like => '%bar'} ]
2690 WHERE ( ( ( a = ? AND b = ? )
2691 OR ( c = ? OR d = ? )
2692 OR ( e LIKE ? AND e LIKE ? ) ) )
2694 This difference in syntax is unfortunate but must be preserved for
2695 historical reasons. So be careful : the two examples below would
2696 seem algebraically equivalent, but they are not
2699 { -like => 'foo%' },
2700 { -like => '%bar' },
2702 # yields : WHERE ( ( col LIKE ? AND col LIKE ? ) )
2705 { col => { -like => 'foo%' } },
2706 { col => { -like => '%bar' } },
2708 # yields : WHERE ( ( col LIKE ? OR col LIKE ? ) )
2711 =head2 Literal SQL and value type operators
2713 The basic premise of SQL::Abstract is that in WHERE specifications the "left
2714 side" is a column name and the "right side" is a value (normally rendered as
2715 a placeholder). This holds true for both hashrefs and arrayref pairs as you
2716 see in the L</WHERE CLAUSES> examples above. Sometimes it is necessary to
2717 alter this behavior. There are several ways of doing so.
2721 This is a virtual operator that signals the string to its right side is an
2722 identifier (a column name) and not a value. For example to compare two
2723 columns you would write:
2726 priority => { '<', 2 },
2727 requestor => { -ident => 'submitter' },
2732 $stmt = "WHERE priority < ? AND requestor = submitter";
2735 If you are maintaining legacy code you may see a different construct as
2736 described in L</Deprecated usage of Literal SQL>, please use C<-ident> in new
2741 This is a virtual operator that signals that the construct to its right side
2742 is a value to be passed to DBI. This is for example necessary when you want
2743 to write a where clause against an array (for RDBMS that support such
2744 datatypes). For example:
2747 array => { -value => [1, 2, 3] }
2752 $stmt = 'WHERE array = ?';
2753 @bind = ([1, 2, 3]);
2755 Note that if you were to simply say:
2761 the result would probably not be what you wanted:
2763 $stmt = 'WHERE array = ? OR array = ? OR array = ?';
2768 Finally, sometimes only literal SQL will do. To include a random snippet
2769 of SQL verbatim, you specify it as a scalar reference. Consider this only
2770 as a last resort. Usually there is a better way. For example:
2773 priority => { '<', 2 },
2774 requestor => { -in => \'(SELECT name FROM hitmen)' },
2779 $stmt = "WHERE priority < ? AND requestor IN (SELECT name FROM hitmen)"
2782 Note that in this example, you only get one bind parameter back, since
2783 the verbatim SQL is passed as part of the statement.
2787 Never use untrusted input as a literal SQL argument - this is a massive
2788 security risk (there is no way to check literal snippets for SQL
2789 injections and other nastyness). If you need to deal with untrusted input
2790 use literal SQL with placeholders as described next.
2792 =head3 Literal SQL with placeholders and bind values (subqueries)
2794 If the literal SQL to be inserted has placeholders and bind values,
2795 use a reference to an arrayref (yes this is a double reference --
2796 not so common, but perfectly legal Perl). For example, to find a date
2797 in Postgres you can use something like this:
2800 date_column => \[ "= date '2008-09-30' - ?::integer", 10 ]
2805 $stmt = "WHERE ( date_column = date '2008-09-30' - ?::integer )"
2808 Note that you must pass the bind values in the same format as they are returned
2809 by L<where|/where(\%where, $order)>. This means that if you set L</bindtype>
2810 to C<columns>, you must provide the bind values in the
2811 C<< [ column_meta => value ] >> format, where C<column_meta> is an opaque
2812 scalar value; most commonly the column name, but you can use any scalar value
2813 (including references and blessed references), L<SQL::Abstract> will simply
2814 pass it through intact. So if C<bindtype> is set to C<columns> the above
2815 example will look like:
2818 date_column => \[ "= date '2008-09-30' - ?::integer", [ {} => 10 ] ]
2821 Literal SQL is especially useful for nesting parenthesized clauses in the
2822 main SQL query. Here is a first example :
2824 my ($sub_stmt, @sub_bind) = ("SELECT c1 FROM t1 WHERE c2 < ? AND c3 LIKE ?",
2828 bar => \["IN ($sub_stmt)" => @sub_bind],
2833 $stmt = "WHERE (foo = ? AND bar IN (SELECT c1 FROM t1
2834 WHERE c2 < ? AND c3 LIKE ?))";
2835 @bind = (1234, 100, "foo%");
2837 Other subquery operators, like for example C<"E<gt> ALL"> or C<"NOT IN">,
2838 are expressed in the same way. Of course the C<$sub_stmt> and
2839 its associated bind values can be generated through a former call
2842 my ($sub_stmt, @sub_bind)
2843 = $sql->select("t1", "c1", {c2 => {"<" => 100},
2844 c3 => {-like => "foo%"}});
2847 bar => \["> ALL ($sub_stmt)" => @sub_bind],
2850 In the examples above, the subquery was used as an operator on a column;
2851 but the same principle also applies for a clause within the main C<%where>
2852 hash, like an EXISTS subquery :
2854 my ($sub_stmt, @sub_bind)
2855 = $sql->select("t1", "*", {c1 => 1, c2 => \"> t0.c0"});
2856 my %where = ( -and => [
2858 \["EXISTS ($sub_stmt)" => @sub_bind],
2863 $stmt = "WHERE (foo = ? AND EXISTS (SELECT * FROM t1
2864 WHERE c1 = ? AND c2 > t0.c0))";
2868 Observe that the condition on C<c2> in the subquery refers to
2869 column C<t0.c0> of the main query : this is I<not> a bind
2870 value, so we have to express it through a scalar ref.
2871 Writing C<< c2 => {">" => "t0.c0"} >> would have generated
2872 C<< c2 > ? >> with bind value C<"t0.c0"> ... not exactly
2873 what we wanted here.
2875 Finally, here is an example where a subquery is used
2876 for expressing unary negation:
2878 my ($sub_stmt, @sub_bind)
2879 = $sql->where({age => [{"<" => 10}, {">" => 20}]});
2880 $sub_stmt =~ s/^ where //i; # don't want "WHERE" in the subclause
2882 lname => {like => '%son%'},
2883 \["NOT ($sub_stmt)" => @sub_bind],
2888 $stmt = "lname LIKE ? AND NOT ( age < ? OR age > ? )"
2889 @bind = ('%son%', 10, 20)
2891 =head3 Deprecated usage of Literal SQL
2893 Below are some examples of archaic use of literal SQL. It is shown only as
2894 reference for those who deal with legacy code. Each example has a much
2895 better, cleaner and safer alternative that users should opt for in new code.
2901 my %where = ( requestor => \'IS NOT NULL' )
2903 $stmt = "WHERE requestor IS NOT NULL"
2905 This used to be the way of generating NULL comparisons, before the handling
2906 of C<undef> got formalized. For new code please use the superior syntax as
2907 described in L</Tests for NULL values>.
2911 my %where = ( requestor => \'= submitter' )
2913 $stmt = "WHERE requestor = submitter"
2915 This used to be the only way to compare columns. Use the superior L</-ident>
2916 method for all new code. For example an identifier declared in such a way
2917 will be properly quoted if L</quote_char> is properly set, while the legacy
2918 form will remain as supplied.
2922 my %where = ( is_ready => \"", completed => { '>', '2012-12-21' } )
2924 $stmt = "WHERE completed > ? AND is_ready"
2925 @bind = ('2012-12-21')
2927 Using an empty string literal used to be the only way to express a boolean.
2928 For all new code please use the much more readable
2929 L<-bool|/Unary operators: bool> operator.
2935 These pages could go on for a while, since the nesting of the data
2936 structures this module can handle are pretty much unlimited (the
2937 module implements the C<WHERE> expansion as a recursive function
2938 internally). Your best bet is to "play around" with the module a
2939 little to see how the data structures behave, and choose the best
2940 format for your data based on that.
2942 And of course, all the values above will probably be replaced with
2943 variables gotten from forms or the command line. After all, if you
2944 knew everything ahead of time, you wouldn't have to worry about
2945 dynamically-generating SQL and could just hardwire it into your
2948 =head1 ORDER BY CLAUSES
2950 Some functions take an order by clause. This can either be a scalar (just a
2951 column name), a hashref of C<< { -desc => 'col' } >> or C<< { -asc => 'col' }
2952 >>, a scalarref, an arrayref-ref, or an arrayref of any of the previous
2955 Given | Will Generate
2956 ---------------------------------------------------------------
2958 'colA' | ORDER BY colA
2960 [qw/colA colB/] | ORDER BY colA, colB
2962 {-asc => 'colA'} | ORDER BY colA ASC
2964 {-desc => 'colB'} | ORDER BY colB DESC
2966 ['colA', {-asc => 'colB'}] | ORDER BY colA, colB ASC
2968 { -asc => [qw/colA colB/] } | ORDER BY colA ASC, colB ASC
2970 \'colA DESC' | ORDER BY colA DESC
2972 \[ 'FUNC(colA, ?)', $x ] | ORDER BY FUNC(colA, ?)
2973 | /* ...with $x bound to ? */
2976 { -asc => 'colA' }, | colA ASC,
2977 { -desc => [qw/colB/] }, | colB DESC,
2978 { -asc => [qw/colC colD/] },| colC ASC, colD ASC,
2979 \'colE DESC', | colE DESC,
2980 \[ 'FUNC(colF, ?)', $x ], | FUNC(colF, ?)
2981 ] | /* ...with $x bound to ? */
2982 ===============================================================
2986 =head1 SPECIAL OPERATORS
2988 my $sqlmaker = SQL::Abstract->new(special_ops => [
2992 my ($self, $field, $op, $arg) = @_;
2998 handler => 'method_name',
3002 A "special operator" is a SQL syntactic clause that can be
3003 applied to a field, instead of a usual binary operator.
3006 WHERE field IN (?, ?, ?)
3007 WHERE field BETWEEN ? AND ?
3008 WHERE MATCH(field) AGAINST (?, ?)
3010 Special operators IN and BETWEEN are fairly standard and therefore
3011 are builtin within C<SQL::Abstract> (as the overridable methods
3012 C<_where_field_IN> and C<_where_field_BETWEEN>). For other operators,
3013 like the MATCH .. AGAINST example above which is specific to MySQL,
3014 you can write your own operator handlers - supply a C<special_ops>
3015 argument to the C<new> method. That argument takes an arrayref of
3016 operator definitions; each operator definition is a hashref with two
3023 the regular expression to match the operator
3027 Either a coderef or a plain scalar method name. In both cases
3028 the expected return is C<< ($sql, @bind) >>.
3030 When supplied with a method name, it is simply called on the
3031 L<SQL::Abstract> object as:
3033 $self->$method_name($field, $op, $arg)
3037 $field is the LHS of the operator
3038 $op is the part that matched the handler regex
3041 When supplied with a coderef, it is called as:
3043 $coderef->($self, $field, $op, $arg)
3048 For example, here is an implementation
3049 of the MATCH .. AGAINST syntax for MySQL
3051 my $sqlmaker = SQL::Abstract->new(special_ops => [
3053 # special op for MySql MATCH (field) AGAINST(word1, word2, ...)
3054 {regex => qr/^match$/i,
3056 my ($self, $field, $op, $arg) = @_;
3057 $arg = [$arg] if not ref $arg;
3058 my $label = $self->_quote($field);
3059 my ($placeholder) = $self->_convert('?');
3060 my $placeholders = join ", ", (($placeholder) x @$arg);
3061 my $sql = $self->_sqlcase('match') . " ($label) "
3062 . $self->_sqlcase('against') . " ($placeholders) ";
3063 my @bind = $self->_bindtype($field, @$arg);
3064 return ($sql, @bind);
3071 =head1 UNARY OPERATORS
3073 my $sqlmaker = SQL::Abstract->new(unary_ops => [
3077 my ($self, $op, $arg) = @_;
3083 handler => 'method_name',
3087 A "unary operator" is a SQL syntactic clause that can be
3088 applied to a field - the operator goes before the field
3090 You can write your own operator handlers - supply a C<unary_ops>
3091 argument to the C<new> method. That argument takes an arrayref of
3092 operator definitions; each operator definition is a hashref with two
3099 the regular expression to match the operator
3103 Either a coderef or a plain scalar method name. In both cases
3104 the expected return is C<< $sql >>.
3106 When supplied with a method name, it is simply called on the
3107 L<SQL::Abstract> object as:
3109 $self->$method_name($op, $arg)
3113 $op is the part that matched the handler regex
3114 $arg is the RHS or argument of the operator
3116 When supplied with a coderef, it is called as:
3118 $coderef->($self, $op, $arg)
3126 Thanks to some benchmarking by Mark Stosberg, it turns out that
3127 this module is many orders of magnitude faster than using C<DBIx::Abstract>.
3128 I must admit this wasn't an intentional design issue, but it's a
3129 byproduct of the fact that you get to control your C<DBI> handles
3132 To maximize performance, use a code snippet like the following:
3134 # prepare a statement handle using the first row
3135 # and then reuse it for the rest of the rows
3137 for my $href (@array_of_hashrefs) {
3138 $stmt ||= $sql->insert('table', $href);
3139 $sth ||= $dbh->prepare($stmt);
3140 $sth->execute($sql->values($href));
3143 The reason this works is because the keys in your C<$href> are sorted
3144 internally by B<SQL::Abstract>. Thus, as long as your data retains
3145 the same structure, you only have to generate the SQL the first time
3146 around. On subsequent queries, simply use the C<values> function provided
3147 by this module to return your values in the correct order.
3149 However this depends on the values having the same type - if, for
3150 example, the values of a where clause may either have values
3151 (resulting in sql of the form C<column = ?> with a single bind
3152 value), or alternatively the values might be C<undef> (resulting in
3153 sql of the form C<column IS NULL> with no bind value) then the
3154 caching technique suggested will not work.
3158 If you use my C<CGI::FormBuilder> module at all, you'll hopefully
3159 really like this part (I do, at least). Building up a complex query
3160 can be as simple as the following:
3167 use CGI::FormBuilder;
3170 my $form = CGI::FormBuilder->new(...);
3171 my $sql = SQL::Abstract->new;
3173 if ($form->submitted) {
3174 my $field = $form->field;
3175 my $id = delete $field->{id};
3176 my($stmt, @bind) = $sql->update('table', $field, {id => $id});
3179 Of course, you would still have to connect using C<DBI> to run the
3180 query, but the point is that if you make your form look like your
3181 table, the actual query script can be extremely simplistic.
3183 If you're B<REALLY> lazy (I am), check out C<HTML::QuickTable> for
3184 a fast interface to returning and formatting data. I frequently
3185 use these three modules together to write complex database query
3186 apps in under 50 lines.
3188 =head1 HOW TO CONTRIBUTE
3190 Contributions are always welcome, in all usable forms (we especially
3191 welcome documentation improvements). The delivery methods include git-
3192 or unified-diff formatted patches, GitHub pull requests, or plain bug
3193 reports either via RT or the Mailing list. Contributors are generally
3194 granted full access to the official repository after their first several
3195 patches pass successful review.
3197 This project is maintained in a git repository. The code and related tools are
3198 accessible at the following locations:
3202 =item * Official repo: L<git://git.shadowcat.co.uk/dbsrgits/SQL-Abstract.git>
3204 =item * Official gitweb: L<http://git.shadowcat.co.uk/gitweb/gitweb.cgi?p=dbsrgits/SQL-Abstract.git>
3206 =item * GitHub mirror: L<https://github.com/dbsrgits/sql-abstract>
3208 =item * Authorized committers: L<ssh://dbsrgits@git.shadowcat.co.uk/SQL-Abstract.git>
3214 Version 1.50 was a major internal refactoring of C<SQL::Abstract>.
3215 Great care has been taken to preserve the I<published> behavior
3216 documented in previous versions in the 1.* family; however,
3217 some features that were previously undocumented, or behaved
3218 differently from the documentation, had to be changed in order
3219 to clarify the semantics. Hence, client code that was relying
3220 on some dark areas of C<SQL::Abstract> v1.*
3221 B<might behave differently> in v1.50.
3223 The main changes are :
3229 support for literal SQL through the C<< \ [ $sql, @bind ] >> syntax.
3233 support for the { operator => \"..." } construct (to embed literal SQL)
3237 support for the { operator => \["...", @bind] } construct (to embed literal SQL with bind values)
3241 optional support for L<array datatypes|/"Inserting and Updating Arrays">
3245 defensive programming : check arguments
3249 fixed bug with global logic, which was previously implemented
3250 through global variables yielding side-effects. Prior versions would
3251 interpret C<< [ {cond1, cond2}, [cond3, cond4] ] >>
3252 as C<< "(cond1 AND cond2) OR (cond3 AND cond4)" >>.
3253 Now this is interpreted
3254 as C<< "(cond1 AND cond2) OR (cond3 OR cond4)" >>.
3259 fixed semantics of _bindtype on array args
3263 dropped the C<_anoncopy> of the %where tree. No longer necessary,
3264 we just avoid shifting arrays within that tree.
3268 dropped the C<_modlogic> function
3272 =head1 ACKNOWLEDGEMENTS
3274 There are a number of individuals that have really helped out with
3275 this module. Unfortunately, most of them submitted bugs via CPAN
3276 so I have no idea who they are! But the people I do know are:
3278 Ash Berlin (order_by hash term support)
3279 Matt Trout (DBIx::Class support)
3280 Mark Stosberg (benchmarking)
3281 Chas Owens (initial "IN" operator support)
3282 Philip Collins (per-field SQL functions)
3283 Eric Kolve (hashref "AND" support)
3284 Mike Fragassi (enhancements to "BETWEEN" and "LIKE")
3285 Dan Kubb (support for "quote_char" and "name_sep")
3286 Guillermo Roditi (patch to cleanup "IN" and "BETWEEN", fix and tests for _order_by)
3287 Laurent Dami (internal refactoring, extensible list of special operators, literal SQL)
3288 Norbert Buchmuller (support for literal SQL in hashpair, misc. fixes & tests)
3289 Peter Rabbitson (rewrite of SQLA::Test, misc. fixes & tests)
3290 Oliver Charles (support for "RETURNING" after "INSERT")
3296 L<DBIx::Class>, L<DBIx::Abstract>, L<CGI::FormBuilder>, L<HTML::QuickTable>.
3300 Copyright (c) 2001-2007 Nathan Wiger <nwiger@cpan.org>. All Rights Reserved.
3302 This module is actively maintained by Matt Trout <mst@shadowcatsystems.co.uk>
3304 For support, your best bet is to try the C<DBIx::Class> users mailing list.
3305 While not an official support venue, C<DBIx::Class> makes heavy use of
3306 C<SQL::Abstract>, and as such list members there are very familiar with
3307 how to create queries.
3311 This module is free software; you may copy this under the same
3312 terms as perl itself (either the GNU General Public License or
3313 the Artistic License)