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 = $self->_table(shift);
214 my $data = shift || return;
217 my $method = $self->_METHOD_FOR_refkind("_insert", $data);
218 my ($sql, @bind) = $self->$method($data);
219 $sql = join " ", $self->_sqlcase('insert into'), $table, $sql;
221 if ($options->{returning}) {
222 my ($s, @b) = $self->_insert_returning($options);
227 return wantarray ? ($sql, @bind) : $sql;
230 # So that subclasses can override INSERT ... RETURNING separately from
231 # UPDATE and DELETE (e.g. DBIx::Class::SQLMaker::Oracle does this)
232 sub _insert_returning { shift->_returning(@_) }
235 my ($self, $options) = @_;
237 my $f = $options->{returning};
239 my $fieldlist = $self->_SWITCH_refkind($f, {
240 ARRAYREF => sub {join ', ', map { $self->_quote($_) } @$f;},
241 SCALAR => sub {$self->_quote($f)},
242 SCALARREF => sub {$$f},
244 return $self->_sqlcase(' returning ') . $fieldlist;
247 sub _insert_HASHREF { # explicit list of fields and then values
248 my ($self, $data) = @_;
250 my @fields = sort keys %$data;
252 my ($sql, @bind) = $self->_insert_values($data);
255 $_ = $self->_quote($_) foreach @fields;
256 $sql = "( ".join(", ", @fields).") ".$sql;
258 return ($sql, @bind);
261 sub _insert_ARRAYREF { # just generate values(?,?) part (no list of fields)
262 my ($self, $data) = @_;
264 # no names (arrayref) so can't generate bindtype
265 $self->{bindtype} ne 'columns'
266 or belch "can't do 'columns' bindtype when called with arrayref";
268 my (@values, @all_bind);
269 foreach my $value (@$data) {
270 my ($values, @bind) = $self->_insert_value(undef, $value);
271 push @values, $values;
272 push @all_bind, @bind;
274 my $sql = $self->_sqlcase('values')." ( ".join(", ", @values)." )";
275 return ($sql, @all_bind);
278 sub _insert_ARRAYREFREF { # literal SQL with bind
279 my ($self, $data) = @_;
281 my ($sql, @bind) = @${$data};
282 $self->_assert_bindval_matches_bindtype(@bind);
284 return ($sql, @bind);
288 sub _insert_SCALARREF { # literal SQL without bind
289 my ($self, $data) = @_;
295 my ($self, $data) = @_;
297 my (@values, @all_bind);
298 foreach my $column (sort keys %$data) {
299 my ($values, @bind) = $self->_insert_value($column, $data->{$column});
300 push @values, $values;
301 push @all_bind, @bind;
303 my $sql = $self->_sqlcase('values')." ( ".join(", ", @values)." )";
304 return ($sql, @all_bind);
308 my ($self, $column, $v) = @_;
310 my (@values, @all_bind);
311 $self->_SWITCH_refkind($v, {
314 if ($self->{array_datatypes}) { # if array datatype are activated
316 push @all_bind, $self->_bindtype($column, $v);
318 else { # else literal SQL with bind
319 my ($sql, @bind) = @$v;
320 $self->_assert_bindval_matches_bindtype(@bind);
322 push @all_bind, @bind;
326 ARRAYREFREF => sub { # literal SQL with bind
327 my ($sql, @bind) = @${$v};
328 $self->_assert_bindval_matches_bindtype(@bind);
330 push @all_bind, @bind;
333 # THINK: anything useful to do with a HASHREF ?
334 HASHREF => sub { # (nothing, but old SQLA passed it through)
335 #TODO in SQLA >= 2.0 it will die instead
336 belch "HASH ref as bind value in insert is not supported";
338 push @all_bind, $self->_bindtype($column, $v);
341 SCALARREF => sub { # literal SQL without bind
345 SCALAR_or_UNDEF => sub {
347 push @all_bind, $self->_bindtype($column, $v);
352 my $sql = join(", ", @values);
353 return ($sql, @all_bind);
358 #======================================================================
360 #======================================================================
365 my $table = $self->_table(shift);
366 my $data = shift || return;
370 # first build the 'SET' part of the sql statement
371 puke "Unsupported data type specified to \$sql->update"
372 unless ref $data eq 'HASH';
374 my ($sql, @all_bind) = $self->_update_set_values($data);
375 $sql = $self->_sqlcase('update ') . $table . $self->_sqlcase(' set ')
379 my($where_sql, @where_bind) = $self->where($where);
381 push @all_bind, @where_bind;
384 if ($options->{returning}) {
385 my ($returning_sql, @returning_bind) = $self->_update_returning($options);
386 $sql .= $returning_sql;
387 push @all_bind, @returning_bind;
390 return wantarray ? ($sql, @all_bind) : $sql;
393 sub _update_set_values {
394 my ($self, $data) = @_;
396 my (@set, @all_bind);
397 for my $k (sort keys %$data) {
400 my $label = $self->_quote($k);
402 $self->_SWITCH_refkind($v, {
404 if ($self->{array_datatypes}) { # array datatype
405 push @set, "$label = ?";
406 push @all_bind, $self->_bindtype($k, $v);
408 else { # literal SQL with bind
409 my ($sql, @bind) = @$v;
410 $self->_assert_bindval_matches_bindtype(@bind);
411 push @set, "$label = $sql";
412 push @all_bind, @bind;
415 ARRAYREFREF => sub { # literal SQL with bind
416 my ($sql, @bind) = @${$v};
417 $self->_assert_bindval_matches_bindtype(@bind);
418 push @set, "$label = $sql";
419 push @all_bind, @bind;
421 SCALARREF => sub { # literal SQL without bind
422 push @set, "$label = $$v";
425 my ($op, $arg, @rest) = %$v;
427 puke 'Operator calls in update must be in the form { -op => $arg }'
428 if (@rest or not $op =~ /^\-(.+)/);
430 local $self->{_nested_func_lhs} = $k;
431 my ($sql, @bind) = $self->_where_unary_op($1, $arg);
433 push @set, "$label = $sql";
434 push @all_bind, @bind;
436 SCALAR_or_UNDEF => sub {
437 push @set, "$label = ?";
438 push @all_bind, $self->_bindtype($k, $v);
444 my $sql = join ', ', @set;
446 return ($sql, @all_bind);
449 # So that subclasses can override UPDATE ... RETURNING separately from
451 sub _update_returning { shift->_returning(@_) }
455 #======================================================================
457 #======================================================================
462 my $table = $self->_table(shift);
463 my $fields = shift || '*';
467 my ($fields_sql, @bind) = $self->_select_fields($fields);
469 my ($where_sql, @where_bind) = $self->where($where, $order);
470 push @bind, @where_bind;
472 my $sql = join(' ', $self->_sqlcase('select'), $fields_sql,
473 $self->_sqlcase('from'), $table)
476 return wantarray ? ($sql, @bind) : $sql;
480 my ($self, $fields) = @_;
481 return ref $fields eq 'ARRAY' ? join ', ', map { $self->_quote($_) } @$fields
485 #======================================================================
487 #======================================================================
492 my $table = $self->_table(shift);
496 my($where_sql, @bind) = $self->where($where);
497 my $sql = $self->_sqlcase('delete from ') . $table . $where_sql;
499 if ($options->{returning}) {
500 my ($returning_sql, @returning_bind) = $self->_delete_returning($options);
501 $sql .= $returning_sql;
502 push @bind, @returning_bind;
505 return wantarray ? ($sql, @bind) : $sql;
508 # So that subclasses can override DELETE ... RETURNING separately from
510 sub _delete_returning { shift->_returning(@_) }
514 #======================================================================
516 #======================================================================
520 # Finally, a separate routine just to handle WHERE clauses
522 my ($self, $where, $order) = @_;
525 my ($sql, @bind) = $self->_recurse_where($where);
526 $sql = (defined $sql and length $sql) ? $self->_sqlcase(' where ') . "( $sql )" : '';
530 my ($order_sql, @order_bind) = $self->_order_by($order);
532 push @bind, @order_bind;
535 return wantarray ? ($sql, @bind) : $sql;
540 my ($self, $where, $logic) = @_;
542 # dispatch on appropriate method according to refkind of $where
543 my $method = $self->_METHOD_FOR_refkind("_where", $where);
545 my ($sql, @bind) = $self->$method($where, $logic);
547 # DBIx::Class used to call _recurse_where in scalar context
548 # something else might too...
550 return ($sql, @bind);
553 belch "Calling _recurse_where in scalar context is deprecated and will go away before 2.0";
560 #======================================================================
561 # WHERE: top-level ARRAYREF
562 #======================================================================
565 sub _where_ARRAYREF {
566 my ($self, $where, $logic) = @_;
568 $logic = uc($logic || $self->{logic});
569 $logic eq 'AND' or $logic eq 'OR' or puke "unknown logic: $logic";
571 my @clauses = @$where;
573 my (@sql_clauses, @all_bind);
574 # need to use while() so can shift() for pairs
576 my $el = shift @clauses;
578 $el = undef if (defined $el and ! length $el);
580 # switch according to kind of $el and get corresponding ($sql, @bind)
581 my ($sql, @bind) = $self->_SWITCH_refkind($el, {
583 # skip empty elements, otherwise get invalid trailing AND stuff
584 ARRAYREF => sub {$self->_recurse_where($el) if @$el},
588 $self->_assert_bindval_matches_bindtype(@b);
592 HASHREF => sub {$self->_recurse_where($el, 'and') if %$el},
594 SCALARREF => sub { ($$el); },
597 # top-level arrayref with scalars, recurse in pairs
598 $self->_recurse_where({$el => shift(@clauses)})
601 UNDEF => sub {puke "Supplying an empty left hand side argument is not supported in array-pairs" },
605 push @sql_clauses, $sql;
606 push @all_bind, @bind;
610 return $self->_join_sql_clauses($logic, \@sql_clauses, \@all_bind);
613 #======================================================================
614 # WHERE: top-level ARRAYREFREF
615 #======================================================================
617 sub _where_ARRAYREFREF {
618 my ($self, $where) = @_;
619 my ($sql, @bind) = @$$where;
620 $self->_assert_bindval_matches_bindtype(@bind);
621 return ($sql, @bind);
624 #======================================================================
625 # WHERE: top-level HASHREF
626 #======================================================================
629 my ($self, $where) = @_;
630 my (@sql_clauses, @all_bind);
632 for my $k (sort keys %$where) {
633 my $v = $where->{$k};
635 # ($k => $v) is either a special unary op or a regular hashpair
636 my ($sql, @bind) = do {
638 # put the operator in canonical form
640 $op = substr $op, 1; # remove initial dash
641 $op =~ s/^\s+|\s+$//g;# remove leading/trailing space
642 $op =~ s/\s+/ /g; # compress whitespace
644 # so that -not_foo works correctly
645 $op =~ s/^not_/NOT /i;
647 $self->_debug("Unary OP(-$op) within hashref, recursing...");
648 my ($s, @b) = $self->_where_unary_op($op, $v);
650 # top level vs nested
651 # we assume that handled unary ops will take care of their ()s
653 List::Util::first {$op =~ $_->{regex}} @{$self->{unary_ops}}
655 ( defined $self->{_nested_func_lhs} and $self->{_nested_func_lhs} eq $k )
661 if (is_literal_value ($v) ) {
662 belch 'Hash-pairs consisting of an empty string with a literal are deprecated, and will be removed in 2.0: use -and => [ $literal ] instead';
665 puke "Supplying an empty left hand side argument is not supported in hash-pairs";
669 my $method = $self->_METHOD_FOR_refkind("_where_hashpair", $v);
670 $self->$method($k, $v);
674 push @sql_clauses, $sql;
675 push @all_bind, @bind;
678 return $self->_join_sql_clauses('and', \@sql_clauses, \@all_bind);
681 sub _where_unary_op {
682 my ($self, $op, $rhs) = @_;
684 # top level special ops are illegal in general
685 # this includes the -ident/-value ops (dual purpose unary and special)
686 puke "Illegal use of top-level '-$op'"
687 if ! defined $self->{_nested_func_lhs} and List::Util::first { $op =~ $_->{regex} } @{$self->{special_ops}};
689 if (my $op_entry = List::Util::first { $op =~ $_->{regex} } @{$self->{unary_ops}}) {
690 my $handler = $op_entry->{handler};
692 if (not ref $handler) {
693 if ($op =~ s/ [_\s]? \d+ $//x ) {
694 belch 'Use of [and|or|nest]_N modifiers is deprecated and will be removed in SQLA v2.0. '
695 . "You probably wanted ...-and => [ -$op => COND1, -$op => COND2 ... ]";
697 return $self->$handler($op, $rhs);
699 elsif (ref $handler eq 'CODE') {
700 return $handler->($self, $op, $rhs);
703 puke "Illegal handler for operator $op - expecting a method name or a coderef";
707 $self->_debug("Generic unary OP: $op - recursing as function");
709 $self->_assert_pass_injection_guard($op);
711 my ($sql, @bind) = $self->_SWITCH_refkind($rhs, {
713 puke "Illegal use of top-level '-$op'"
714 unless defined $self->{_nested_func_lhs};
717 $self->_convert('?'),
718 $self->_bindtype($self->{_nested_func_lhs}, $rhs)
722 $self->_recurse_where($rhs)
726 $sql = sprintf('%s %s',
727 $self->_sqlcase($op),
731 return ($sql, @bind);
734 sub _where_op_ANDOR {
735 my ($self, $op, $v) = @_;
737 $self->_SWITCH_refkind($v, {
739 return $self->_where_ARRAYREF($v, $op);
743 return ($op =~ /^or/i)
744 ? $self->_where_ARRAYREF([ map { $_ => $v->{$_} } (sort keys %$v) ], $op)
745 : $self->_where_HASHREF($v);
749 puke "-$op => \\\$scalar makes little sense, use " .
751 ? '[ \$scalar, \%rest_of_conditions ] instead'
752 : '-and => [ \$scalar, \%rest_of_conditions ] instead'
757 puke "-$op => \\[...] makes little sense, use " .
759 ? '[ \[...], \%rest_of_conditions ] instead'
760 : '-and => [ \[...], \%rest_of_conditions ] instead'
764 SCALAR => sub { # permissively interpreted as SQL
765 puke "-$op => \$value makes little sense, use -bool => \$value instead";
769 puke "-$op => undef not supported";
775 my ($self, $op, $v) = @_;
777 $self->_SWITCH_refkind($v, {
779 SCALAR => sub { # permissively interpreted as SQL
780 belch "literal SQL should be -nest => \\'scalar' "
781 . "instead of -nest => 'scalar' ";
786 puke "-$op => undef not supported";
790 $self->_recurse_where($v);
798 my ($self, $op, $v) = @_;
800 my ($s, @b) = $self->_SWITCH_refkind($v, {
801 SCALAR => sub { # interpreted as SQL column
802 $self->_convert($self->_quote($v));
806 puke "-$op => undef not supported";
810 $self->_recurse_where($v);
814 $s = "(NOT $s)" if $op =~ /^not/i;
819 sub _where_op_IDENT {
821 my ($op, $rhs) = splice @_, -2;
822 if (! defined $rhs or length ref $rhs) {
823 puke "-$op requires a single plain scalar argument (a quotable identifier)";
826 # in case we are called as a top level special op (no '=')
829 $_ = $self->_convert($self->_quote($_)) for ($lhs, $rhs);
837 sub _where_op_VALUE {
839 my ($op, $rhs) = splice @_, -2;
841 # in case we are called as a top level special op (no '=')
845 if (! defined $rhs) {
847 ? $self->_convert($self->_quote($lhs)) . ' IS NULL'
854 (defined $lhs ? $lhs : $self->{_nested_func_lhs}),
861 $self->_convert($self->_quote($lhs)) . ' = ' . $self->_convert('?'),
865 $self->_convert('?'),
871 sub _where_hashpair_ARRAYREF {
872 my ($self, $k, $v) = @_;
875 my @v = @$v; # need copy because of shift below
876 $self->_debug("ARRAY($k) means distribute over elements");
878 # put apart first element if it is an operator (-and, -or)
880 (defined $v[0] && $v[0] =~ /^ - (?: AND|OR ) $/ix)
884 my @distributed = map { {$k => $_} } @v;
887 $self->_debug("OP($op) reinjected into the distributed array");
888 unshift @distributed, $op;
891 my $logic = $op ? substr($op, 1) : '';
893 return $self->_recurse_where(\@distributed, $logic);
896 $self->_debug("empty ARRAY($k) means 0=1");
897 return ($self->{sqlfalse});
901 sub _where_hashpair_HASHREF {
902 my ($self, $k, $v, $logic) = @_;
905 local $self->{_nested_func_lhs} = defined $self->{_nested_func_lhs}
906 ? $self->{_nested_func_lhs}
910 my ($all_sql, @all_bind);
912 for my $orig_op (sort keys %$v) {
913 my $val = $v->{$orig_op};
915 # put the operator in canonical form
918 # FIXME - we need to phase out dash-less ops
919 $op =~ s/^-//; # remove possible initial dash
920 $op =~ s/^\s+|\s+$//g;# remove leading/trailing space
921 $op =~ s/\s+/ /g; # compress whitespace
923 $self->_assert_pass_injection_guard($op);
926 $op =~ s/^is_not/IS NOT/i;
928 # so that -not_foo works correctly
929 $op =~ s/^not_/NOT /i;
931 # another retarded special case: foo => { $op => { -value => undef } }
932 if (ref $val eq 'HASH' and keys %$val == 1 and exists $val->{-value} and ! defined $val->{-value} ) {
938 # CASE: col-value logic modifiers
939 if ($orig_op =~ /^ \- (and|or) $/xi) {
940 ($sql, @bind) = $self->_where_hashpair_HASHREF($k, $val, $1);
942 # CASE: special operators like -in or -between
943 elsif (my $special_op = List::Util::first { $op =~ $_->{regex} } @{$self->{special_ops}}) {
944 my $handler = $special_op->{handler};
946 puke "No handler supplied for special operator $orig_op";
948 elsif (not ref $handler) {
949 ($sql, @bind) = $self->$handler($k, $op, $val);
951 elsif (ref $handler eq 'CODE') {
952 ($sql, @bind) = $handler->($self, $k, $op, $val);
955 puke "Illegal handler for special operator $orig_op - expecting a method name or a coderef";
959 $self->_SWITCH_refkind($val, {
961 ARRAYREF => sub { # CASE: col => {op => \@vals}
962 ($sql, @bind) = $self->_where_field_op_ARRAYREF($k, $op, $val);
965 ARRAYREFREF => sub { # CASE: col => {op => \[$sql, @bind]} (literal SQL with bind)
966 my ($sub_sql, @sub_bind) = @$$val;
967 $self->_assert_bindval_matches_bindtype(@sub_bind);
968 $sql = join ' ', $self->_convert($self->_quote($k)),
969 $self->_sqlcase($op),
974 UNDEF => sub { # CASE: col => {op => undef} : sql "IS (NOT)? NULL"
976 $op =~ /^not$/i ? 'is not' # legacy
977 : $op =~ $self->{equality_op} ? 'is'
978 : $op =~ $self->{like_op} ? belch("Supplying an undefined argument to '@{[ uc $op]}' is deprecated") && 'is'
979 : $op =~ $self->{inequality_op} ? 'is not'
980 : $op =~ $self->{not_like_op} ? belch("Supplying an undefined argument to '@{[ uc $op]}' is deprecated") && 'is not'
981 : puke "unexpected operator '$orig_op' with undef operand";
983 $sql = $self->_quote($k) . $self->_sqlcase(" $is null");
986 FALLBACK => sub { # CASE: col => {op/func => $stuff}
987 ($sql, @bind) = $self->_where_unary_op($op, $val);
990 $self->_convert($self->_quote($k)),
991 $self->{_nested_func_lhs} eq $k ? $sql : "($sql)", # top level vs nested
997 ($all_sql) = (defined $all_sql and $all_sql) ? $self->_join_sql_clauses($logic, [$all_sql, $sql], []) : $sql;
998 push @all_bind, @bind;
1000 return ($all_sql, @all_bind);
1003 sub _where_field_IS {
1004 my ($self, $k, $op, $v) = @_;
1006 my ($s) = $self->_SWITCH_refkind($v, {
1009 $self->_convert($self->_quote($k)),
1010 map { $self->_sqlcase($_)} ($op, 'null')
1013 puke "$op can only take undef as argument";
1020 sub _where_field_op_ARRAYREF {
1021 my ($self, $k, $op, $vals) = @_;
1023 my @vals = @$vals; #always work on a copy
1026 $self->_debug(sprintf '%s means multiple elements: [ %s ]',
1028 join(', ', map { defined $_ ? "'$_'" : 'NULL' } @vals ),
1031 # see if the first element is an -and/-or op
1033 if (defined $vals[0] && $vals[0] =~ /^ - (AND|OR) $/ix) {
1038 # a long standing API wart - an attempt to change this behavior during
1039 # the 1.50 series failed *spectacularly*. Warn instead and leave the
1044 (!$logic or $logic eq 'OR')
1046 ($op =~ $self->{inequality_op} or $op =~ $self->{not_like_op})
1049 belch "A multi-element arrayref as an argument to the inequality op '$o' "
1050 . 'is technically equivalent to an always-true 1=1 (you probably wanted '
1051 . "to say ...{ \$inequality_op => [ -and => \@values ] }... instead)"
1055 # distribute $op over each remaining member of @vals, append logic if exists
1056 return $self->_recurse_where([map { {$k => {$op, $_}} } @vals], $logic);
1060 # try to DWIM on equality operators
1062 $op =~ $self->{equality_op} ? $self->{sqlfalse}
1063 : $op =~ $self->{like_op} ? belch("Supplying an empty arrayref to '@{[ uc $op]}' is deprecated") && $self->{sqlfalse}
1064 : $op =~ $self->{inequality_op} ? $self->{sqltrue}
1065 : $op =~ $self->{not_like_op} ? belch("Supplying an empty arrayref to '@{[ uc $op]}' is deprecated") && $self->{sqltrue}
1066 : puke "operator '$op' applied on an empty array (field '$k')";
1071 sub _where_hashpair_SCALARREF {
1072 my ($self, $k, $v) = @_;
1073 $self->_debug("SCALAR($k) means literal SQL: $$v");
1074 my $sql = $self->_quote($k) . " " . $$v;
1078 # literal SQL with bind
1079 sub _where_hashpair_ARRAYREFREF {
1080 my ($self, $k, $v) = @_;
1081 $self->_debug("REF($k) means literal SQL: @${$v}");
1082 my ($sql, @bind) = @$$v;
1083 $self->_assert_bindval_matches_bindtype(@bind);
1084 $sql = $self->_quote($k) . " " . $sql;
1085 return ($sql, @bind );
1088 # literal SQL without bind
1089 sub _where_hashpair_SCALAR {
1090 my ($self, $k, $v) = @_;
1091 $self->_debug("NOREF($k) means simple key=val: $k $self->{cmp} $v");
1092 my $sql = join ' ', $self->_convert($self->_quote($k)),
1093 $self->_sqlcase($self->{cmp}),
1094 $self->_convert('?');
1095 my @bind = $self->_bindtype($k, $v);
1096 return ($sql, @bind);
1100 sub _where_hashpair_UNDEF {
1101 my ($self, $k, $v) = @_;
1102 $self->_debug("UNDEF($k) means IS NULL");
1103 my $sql = $self->_quote($k) . $self->_sqlcase(' is null');
1107 #======================================================================
1108 # WHERE: TOP-LEVEL OTHERS (SCALARREF, SCALAR, UNDEF)
1109 #======================================================================
1112 sub _where_SCALARREF {
1113 my ($self, $where) = @_;
1116 $self->_debug("SCALAR(*top) means literal SQL: $$where");
1122 my ($self, $where) = @_;
1125 $self->_debug("NOREF(*top) means literal SQL: $where");
1136 #======================================================================
1137 # WHERE: BUILTIN SPECIAL OPERATORS (-in, -between)
1138 #======================================================================
1141 sub _where_field_BETWEEN {
1142 my ($self, $k, $op, $vals) = @_;
1144 my ($label, $and, $placeholder);
1145 $label = $self->_convert($self->_quote($k));
1146 $and = ' ' . $self->_sqlcase('and') . ' ';
1147 $placeholder = $self->_convert('?');
1148 $op = $self->_sqlcase($op);
1150 my $invalid_args = "Operator '$op' requires either an arrayref with two defined values or expressions, or a single literal scalarref/arrayref-ref";
1152 my ($clause, @bind) = $self->_SWITCH_refkind($vals, {
1153 ARRAYREFREF => sub {
1154 my ($s, @b) = @$$vals;
1155 $self->_assert_bindval_matches_bindtype(@b);
1162 puke $invalid_args if @$vals != 2;
1164 my (@all_sql, @all_bind);
1165 foreach my $val (@$vals) {
1166 my ($sql, @bind) = $self->_SWITCH_refkind($val, {
1168 return ($placeholder, $self->_bindtype($k, $val) );
1173 ARRAYREFREF => sub {
1174 my ($sql, @bind) = @$$val;
1175 $self->_assert_bindval_matches_bindtype(@bind);
1176 return ($sql, @bind);
1179 my ($func, $arg, @rest) = %$val;
1180 puke "Only simple { -func => arg } functions accepted as sub-arguments to BETWEEN"
1181 if (@rest or $func !~ /^ \- (.+)/x);
1182 $self->_where_unary_op($1 => $arg);
1188 push @all_sql, $sql;
1189 push @all_bind, @bind;
1193 (join $and, @all_sql),
1202 my $sql = "( $label $op $clause )";
1203 return ($sql, @bind)
1207 sub _where_field_IN {
1208 my ($self, $k, $op, $vals) = @_;
1210 # backwards compatibility: if scalar, force into an arrayref
1211 $vals = [$vals] if defined $vals && ! ref $vals;
1213 my ($label) = $self->_convert($self->_quote($k));
1214 my ($placeholder) = $self->_convert('?');
1215 $op = $self->_sqlcase($op);
1217 my ($sql, @bind) = $self->_SWITCH_refkind($vals, {
1218 ARRAYREF => sub { # list of choices
1219 if (@$vals) { # nonempty list
1220 my (@all_sql, @all_bind);
1222 for my $val (@$vals) {
1223 my ($sql, @bind) = $self->_SWITCH_refkind($val, {
1225 return ($placeholder, $val);
1230 ARRAYREFREF => sub {
1231 my ($sql, @bind) = @$$val;
1232 $self->_assert_bindval_matches_bindtype(@bind);
1233 return ($sql, @bind);
1236 my ($func, $arg, @rest) = %$val;
1237 puke "Only simple { -func => arg } functions accepted as sub-arguments to IN"
1238 if (@rest or $func !~ /^ \- (.+)/x);
1239 $self->_where_unary_op($1 => $arg);
1243 'SQL::Abstract before v1.75 used to generate incorrect SQL when the '
1244 . "-$op operator was given an undef-containing list: !!!AUDIT YOUR CODE "
1245 . 'AND DATA!!! (the upcoming Data::Query-based version of SQL::Abstract '
1246 . 'will emit the logically correct SQL instead of raising this exception)'
1250 push @all_sql, $sql;
1251 push @all_bind, @bind;
1255 sprintf('%s %s ( %s )',
1258 join(', ', @all_sql)
1260 $self->_bindtype($k, @all_bind),
1263 else { # empty list: some databases won't understand "IN ()", so DWIM
1264 my $sql = ($op =~ /\bnot\b/i) ? $self->{sqltrue} : $self->{sqlfalse};
1269 SCALARREF => sub { # literal SQL
1270 my $sql = $self->_open_outer_paren($$vals);
1271 return ("$label $op ( $sql )");
1273 ARRAYREFREF => sub { # literal SQL with bind
1274 my ($sql, @bind) = @$$vals;
1275 $self->_assert_bindval_matches_bindtype(@bind);
1276 $sql = $self->_open_outer_paren($sql);
1277 return ("$label $op ( $sql )", @bind);
1281 puke "Argument passed to the '$op' operator can not be undefined";
1285 puke "special op $op requires an arrayref (or scalarref/arrayref-ref)";
1289 return ($sql, @bind);
1292 # Some databases (SQLite) treat col IN (1, 2) different from
1293 # col IN ( (1, 2) ). Use this to strip all outer parens while
1294 # adding them back in the corresponding method
1295 sub _open_outer_paren {
1296 my ($self, $sql) = @_;
1298 while (my ($inner) = $sql =~ /^ \s* \( (.*) \) \s* $/xs) {
1300 # there are closing parens inside, need the heavy duty machinery
1301 # to reevaluate the extraction starting from $sql (full reevaluation)
1302 if ($inner =~ /\)/) {
1303 require Text::Balanced;
1305 my (undef, $remainder) = do {
1306 # idiotic design - writes to $@ but *DOES NOT* throw exceptions
1308 Text::Balanced::extract_bracketed($sql, '()', qr/\s*/);
1311 # the entire expression needs to be a balanced bracketed thing
1312 # (after an extract no remainder sans trailing space)
1313 last if defined $remainder and $remainder =~ /\S/;
1323 #======================================================================
1325 #======================================================================
1328 my ($self, $arg) = @_;
1331 for my $c ($self->_order_by_chunks($arg) ) {
1332 $self->_SWITCH_refkind($c, {
1333 SCALAR => sub { push @sql, $c },
1334 ARRAYREF => sub { push @sql, shift @$c; push @bind, @$c },
1340 $self->_sqlcase(' order by'),
1346 return wantarray ? ($sql, @bind) : $sql;
1349 sub _order_by_chunks {
1350 my ($self, $arg) = @_;
1352 return $self->_SWITCH_refkind($arg, {
1355 map { $self->_order_by_chunks($_ ) } @$arg;
1358 ARRAYREFREF => sub {
1359 my ($s, @b) = @$$arg;
1360 $self->_assert_bindval_matches_bindtype(@b);
1364 SCALAR => sub {$self->_quote($arg)},
1366 UNDEF => sub {return () },
1368 SCALARREF => sub {$$arg}, # literal SQL, no quoting
1371 # get first pair in hash
1372 my ($key, $val, @rest) = %$arg;
1374 return () unless $key;
1376 if (@rest or not $key =~ /^-(desc|asc)/i) {
1377 puke "hash passed to _order_by must have exactly one key (-desc or -asc)";
1383 for my $c ($self->_order_by_chunks($val)) {
1386 $self->_SWITCH_refkind($c, {
1391 ($sql, @bind) = @$c;
1395 $sql = $sql . ' ' . $self->_sqlcase($direction);
1397 push @ret, [ $sql, @bind];
1406 #======================================================================
1407 # DATASOURCE (FOR NOW, JUST PLAIN TABLE OR LIST OF TABLES)
1408 #======================================================================
1413 $self->_SWITCH_refkind($from, {
1414 ARRAYREF => sub {join ', ', map { $self->_quote($_) } @$from;},
1415 SCALAR => sub {$self->_quote($from)},
1416 SCALARREF => sub {$$from},
1421 #======================================================================
1423 #======================================================================
1425 # highly optimized, as it's called way too often
1427 # my ($self, $label) = @_;
1429 return '' unless defined $_[1];
1430 return ${$_[1]} if ref($_[1]) eq 'SCALAR';
1432 $_[0]->{quote_char} or
1433 ($_[0]->_assert_pass_injection_guard($_[1]), return $_[1]);
1435 my $qref = ref $_[0]->{quote_char};
1437 !$qref ? ($_[0]->{quote_char}, $_[0]->{quote_char})
1438 : ($qref eq 'ARRAY') ? @{$_[0]->{quote_char}}
1439 : puke "Unsupported quote_char format: $_[0]->{quote_char}";
1441 my $esc = $_[0]->{escape_char} || $r;
1443 # parts containing * are naturally unquoted
1444 return join($_[0]->{name_sep}||'', map
1445 +( $_ eq '*' ? $_ : do { (my $n = $_) =~ s/(\Q$esc\E|\Q$r\E)/$esc$1/g; $l . $n . $r } ),
1446 ( $_[0]->{name_sep} ? split (/\Q$_[0]->{name_sep}\E/, $_[1] ) : $_[1] )
1451 # Conversion, if applicable
1453 #my ($self, $arg) = @_;
1454 if ($_[0]->{convert}) {
1455 return $_[0]->_sqlcase($_[0]->{convert}) .'(' . $_[1] . ')';
1462 #my ($self, $col, @vals) = @_;
1463 # called often - tighten code
1464 return $_[0]->{bindtype} eq 'columns'
1465 ? map {[$_[1], $_]} @_[2 .. $#_]
1470 # Dies if any element of @bind is not in [colname => value] format
1471 # if bindtype is 'columns'.
1472 sub _assert_bindval_matches_bindtype {
1473 # my ($self, @bind) = @_;
1475 if ($self->{bindtype} eq 'columns') {
1477 if (!defined $_ || ref($_) ne 'ARRAY' || @$_ != 2) {
1478 puke "bindtype 'columns' selected, you need to pass: [column_name => bind_value]"
1484 sub _join_sql_clauses {
1485 my ($self, $logic, $clauses_aref, $bind_aref) = @_;
1487 if (@$clauses_aref > 1) {
1488 my $join = " " . $self->_sqlcase($logic) . " ";
1489 my $sql = '( ' . join($join, @$clauses_aref) . ' )';
1490 return ($sql, @$bind_aref);
1492 elsif (@$clauses_aref) {
1493 return ($clauses_aref->[0], @$bind_aref); # no parentheses
1496 return (); # if no SQL, ignore @$bind_aref
1501 # Fix SQL case, if so requested
1503 # LDNOTE: if $self->{case} is true, then it contains 'lower', so we
1504 # don't touch the argument ... crooked logic, but let's not change it!
1505 return $_[0]->{case} ? $_[1] : uc($_[1]);
1509 #======================================================================
1510 # DISPATCHING FROM REFKIND
1511 #======================================================================
1514 my ($self, $data) = @_;
1516 return 'UNDEF' unless defined $data;
1518 # blessed objects are treated like scalars
1519 my $ref = (Scalar::Util::blessed $data) ? '' : ref $data;
1521 return 'SCALAR' unless $ref;
1524 while ($ref eq 'REF') {
1526 $ref = (Scalar::Util::blessed $data) ? '' : ref $data;
1530 return ($ref||'SCALAR') . ('REF' x $n_steps);
1534 my ($self, $data) = @_;
1535 my @try = ($self->_refkind($data));
1536 push @try, 'SCALAR_or_UNDEF' if $try[0] eq 'SCALAR' || $try[0] eq 'UNDEF';
1537 push @try, 'FALLBACK';
1541 sub _METHOD_FOR_refkind {
1542 my ($self, $meth_prefix, $data) = @_;
1545 for (@{$self->_try_refkind($data)}) {
1546 $method = $self->can($meth_prefix."_".$_)
1550 return $method || puke "cannot dispatch on '$meth_prefix' for ".$self->_refkind($data);
1554 sub _SWITCH_refkind {
1555 my ($self, $data, $dispatch_table) = @_;
1558 for (@{$self->_try_refkind($data)}) {
1559 $coderef = $dispatch_table->{$_}
1563 puke "no dispatch entry for ".$self->_refkind($data)
1572 #======================================================================
1573 # VALUES, GENERATE, AUTOLOAD
1574 #======================================================================
1576 # LDNOTE: original code from nwiger, didn't touch code in that section
1577 # I feel the AUTOLOAD stuff should not be the default, it should
1578 # only be activated on explicit demand by user.
1582 my $data = shift || return;
1583 puke "Argument to ", __PACKAGE__, "->values must be a \\%hash"
1584 unless ref $data eq 'HASH';
1587 foreach my $k (sort keys %$data) {
1588 my $v = $data->{$k};
1589 $self->_SWITCH_refkind($v, {
1591 if ($self->{array_datatypes}) { # array datatype
1592 push @all_bind, $self->_bindtype($k, $v);
1594 else { # literal SQL with bind
1595 my ($sql, @bind) = @$v;
1596 $self->_assert_bindval_matches_bindtype(@bind);
1597 push @all_bind, @bind;
1600 ARRAYREFREF => sub { # literal SQL with bind
1601 my ($sql, @bind) = @${$v};
1602 $self->_assert_bindval_matches_bindtype(@bind);
1603 push @all_bind, @bind;
1605 SCALARREF => sub { # literal SQL without bind
1607 SCALAR_or_UNDEF => sub {
1608 push @all_bind, $self->_bindtype($k, $v);
1619 my(@sql, @sqlq, @sqlv);
1623 if ($ref eq 'HASH') {
1624 for my $k (sort keys %$_) {
1627 my $label = $self->_quote($k);
1628 if ($r eq 'ARRAY') {
1629 # literal SQL with bind
1630 my ($sql, @bind) = @$v;
1631 $self->_assert_bindval_matches_bindtype(@bind);
1632 push @sqlq, "$label = $sql";
1634 } elsif ($r eq 'SCALAR') {
1635 # literal SQL without bind
1636 push @sqlq, "$label = $$v";
1638 push @sqlq, "$label = ?";
1639 push @sqlv, $self->_bindtype($k, $v);
1642 push @sql, $self->_sqlcase('set'), join ', ', @sqlq;
1643 } elsif ($ref eq 'ARRAY') {
1644 # unlike insert(), assume these are ONLY the column names, i.e. for SQL
1647 if ($r eq 'ARRAY') { # literal SQL with bind
1648 my ($sql, @bind) = @$v;
1649 $self->_assert_bindval_matches_bindtype(@bind);
1652 } elsif ($r eq 'SCALAR') { # literal SQL without bind
1653 # embedded literal SQL
1660 push @sql, '(' . join(', ', @sqlq) . ')';
1661 } elsif ($ref eq 'SCALAR') {
1665 # strings get case twiddled
1666 push @sql, $self->_sqlcase($_);
1670 my $sql = join ' ', @sql;
1672 # this is pretty tricky
1673 # if ask for an array, return ($stmt, @bind)
1674 # otherwise, s/?/shift @sqlv/ to put it inline
1676 return ($sql, @sqlv);
1678 1 while $sql =~ s/\?/my $d = shift(@sqlv);
1679 ref $d ? $d->[1] : $d/e;
1688 # This allows us to check for a local, then _form, attr
1690 my($name) = $AUTOLOAD =~ /.*::(.+)/;
1691 return $self->generate($name, @_);
1702 SQL::Abstract - Generate SQL from Perl data structures
1708 my $sql = SQL::Abstract->new;
1710 my($stmt, @bind) = $sql->select($source, \@fields, \%where, $order);
1712 my($stmt, @bind) = $sql->insert($table, \%fieldvals || \@values);
1714 my($stmt, @bind) = $sql->update($table, \%fieldvals, \%where);
1716 my($stmt, @bind) = $sql->delete($table, \%where);
1718 # Then, use these in your DBI statements
1719 my $sth = $dbh->prepare($stmt);
1720 $sth->execute(@bind);
1722 # Just generate the WHERE clause
1723 my($stmt, @bind) = $sql->where(\%where, $order);
1725 # Return values in the same order, for hashed queries
1726 # See PERFORMANCE section for more details
1727 my @bind = $sql->values(\%fieldvals);
1731 This module was inspired by the excellent L<DBIx::Abstract>.
1732 However, in using that module I found that what I really wanted
1733 to do was generate SQL, but still retain complete control over my
1734 statement handles and use the DBI interface. So, I set out to
1735 create an abstract SQL generation module.
1737 While based on the concepts used by L<DBIx::Abstract>, there are
1738 several important differences, especially when it comes to WHERE
1739 clauses. I have modified the concepts used to make the SQL easier
1740 to generate from Perl data structures and, IMO, more intuitive.
1741 The underlying idea is for this module to do what you mean, based
1742 on the data structures you provide it. The big advantage is that
1743 you don't have to modify your code every time your data changes,
1744 as this module figures it out.
1746 To begin with, an SQL INSERT is as easy as just specifying a hash
1747 of C<key=value> pairs:
1750 name => 'Jimbo Bobson',
1751 phone => '123-456-7890',
1752 address => '42 Sister Lane',
1753 city => 'St. Louis',
1754 state => 'Louisiana',
1757 The SQL can then be generated with this:
1759 my($stmt, @bind) = $sql->insert('people', \%data);
1761 Which would give you something like this:
1763 $stmt = "INSERT INTO people
1764 (address, city, name, phone, state)
1765 VALUES (?, ?, ?, ?, ?)";
1766 @bind = ('42 Sister Lane', 'St. Louis', 'Jimbo Bobson',
1767 '123-456-7890', 'Louisiana');
1769 These are then used directly in your DBI code:
1771 my $sth = $dbh->prepare($stmt);
1772 $sth->execute(@bind);
1774 =head2 Inserting and Updating Arrays
1776 If your database has array types (like for example Postgres),
1777 activate the special option C<< array_datatypes => 1 >>
1778 when creating the C<SQL::Abstract> object.
1779 Then you may use an arrayref to insert and update database array types:
1781 my $sql = SQL::Abstract->new(array_datatypes => 1);
1783 planets => [qw/Mercury Venus Earth Mars/]
1786 my($stmt, @bind) = $sql->insert('solar_system', \%data);
1790 $stmt = "INSERT INTO solar_system (planets) VALUES (?)"
1792 @bind = (['Mercury', 'Venus', 'Earth', 'Mars']);
1795 =head2 Inserting and Updating SQL
1797 In order to apply SQL functions to elements of your C<%data> you may
1798 specify a reference to an arrayref for the given hash value. For example,
1799 if you need to execute the Oracle C<to_date> function on a value, you can
1800 say something like this:
1804 date_entered => \[ "to_date(?,'MM/DD/YYYY')", "03/02/2003" ],
1807 The first value in the array is the actual SQL. Any other values are
1808 optional and would be included in the bind values array. This gives
1811 my($stmt, @bind) = $sql->insert('people', \%data);
1813 $stmt = "INSERT INTO people (name, date_entered)
1814 VALUES (?, to_date(?,'MM/DD/YYYY'))";
1815 @bind = ('Bill', '03/02/2003');
1817 An UPDATE is just as easy, all you change is the name of the function:
1819 my($stmt, @bind) = $sql->update('people', \%data);
1821 Notice that your C<%data> isn't touched; the module will generate
1822 the appropriately quirky SQL for you automatically. Usually you'll
1823 want to specify a WHERE clause for your UPDATE, though, which is
1824 where handling C<%where> hashes comes in handy...
1826 =head2 Complex where statements
1828 This module can generate pretty complicated WHERE statements
1829 easily. For example, simple C<key=value> pairs are taken to mean
1830 equality, and if you want to see if a field is within a set
1831 of values, you can use an arrayref. Let's say we wanted to
1832 SELECT some data based on this criteria:
1835 requestor => 'inna',
1836 worker => ['nwiger', 'rcwe', 'sfz'],
1837 status => { '!=', 'completed' }
1840 my($stmt, @bind) = $sql->select('tickets', '*', \%where);
1842 The above would give you something like this:
1844 $stmt = "SELECT * FROM tickets WHERE
1845 ( requestor = ? ) AND ( status != ? )
1846 AND ( worker = ? OR worker = ? OR worker = ? )";
1847 @bind = ('inna', 'completed', 'nwiger', 'rcwe', 'sfz');
1849 Which you could then use in DBI code like so:
1851 my $sth = $dbh->prepare($stmt);
1852 $sth->execute(@bind);
1858 The methods are simple. There's one for every major SQL operation,
1859 and a constructor you use first. The arguments are specified in a
1860 similar order for each method (table, then fields, then a where
1861 clause) to try and simplify things.
1863 =head2 new(option => 'value')
1865 The C<new()> function takes a list of options and values, and returns
1866 a new B<SQL::Abstract> object which can then be used to generate SQL
1867 through the methods below. The options accepted are:
1873 If set to 'lower', then SQL will be generated in all lowercase. By
1874 default SQL is generated in "textbook" case meaning something like:
1876 SELECT a_field FROM a_table WHERE some_field LIKE '%someval%'
1878 Any setting other than 'lower' is ignored.
1882 This determines what the default comparison operator is. By default
1883 it is C<=>, meaning that a hash like this:
1885 %where = (name => 'nwiger', email => 'nate@wiger.org');
1887 Will generate SQL like this:
1889 WHERE name = 'nwiger' AND email = 'nate@wiger.org'
1891 However, you may want loose comparisons by default, so if you set
1892 C<cmp> to C<like> you would get SQL such as:
1894 WHERE name like 'nwiger' AND email like 'nate@wiger.org'
1896 You can also override the comparison on an individual basis - see
1897 the huge section on L</"WHERE CLAUSES"> at the bottom.
1899 =item sqltrue, sqlfalse
1901 Expressions for inserting boolean values within SQL statements.
1902 By default these are C<1=1> and C<1=0>. They are used
1903 by the special operators C<-in> and C<-not_in> for generating
1904 correct SQL even when the argument is an empty array (see below).
1908 This determines the default logical operator for multiple WHERE
1909 statements in arrays or hashes. If absent, the default logic is "or"
1910 for arrays, and "and" for hashes. This means that a WHERE
1914 event_date => {'>=', '2/13/99'},
1915 event_date => {'<=', '4/24/03'},
1918 will generate SQL like this:
1920 WHERE event_date >= '2/13/99' OR event_date <= '4/24/03'
1922 This is probably not what you want given this query, though (look
1923 at the dates). To change the "OR" to an "AND", simply specify:
1925 my $sql = SQL::Abstract->new(logic => 'and');
1927 Which will change the above C<WHERE> to:
1929 WHERE event_date >= '2/13/99' AND event_date <= '4/24/03'
1931 The logic can also be changed locally by inserting
1932 a modifier in front of an arrayref:
1934 @where = (-and => [event_date => {'>=', '2/13/99'},
1935 event_date => {'<=', '4/24/03'} ]);
1937 See the L</"WHERE CLAUSES"> section for explanations.
1941 This will automatically convert comparisons using the specified SQL
1942 function for both column and value. This is mostly used with an argument
1943 of C<upper> or C<lower>, so that the SQL will have the effect of
1944 case-insensitive "searches". For example, this:
1946 $sql = SQL::Abstract->new(convert => 'upper');
1947 %where = (keywords => 'MaKe iT CAse inSeNSItive');
1949 Will turn out the following SQL:
1951 WHERE upper(keywords) like upper('MaKe iT CAse inSeNSItive')
1953 The conversion can be C<upper()>, C<lower()>, or any other SQL function
1954 that can be applied symmetrically to fields (actually B<SQL::Abstract> does
1955 not validate this option; it will just pass through what you specify verbatim).
1959 This is a kludge because many databases suck. For example, you can't
1960 just bind values using DBI's C<execute()> for Oracle C<CLOB> or C<BLOB> fields.
1961 Instead, you have to use C<bind_param()>:
1963 $sth->bind_param(1, 'reg data');
1964 $sth->bind_param(2, $lots, {ora_type => ORA_CLOB});
1966 The problem is, B<SQL::Abstract> will normally just return a C<@bind> array,
1967 which loses track of which field each slot refers to. Fear not.
1969 If you specify C<bindtype> in new, you can determine how C<@bind> is returned.
1970 Currently, you can specify either C<normal> (default) or C<columns>. If you
1971 specify C<columns>, you will get an array that looks like this:
1973 my $sql = SQL::Abstract->new(bindtype => 'columns');
1974 my($stmt, @bind) = $sql->insert(...);
1977 [ 'column1', 'value1' ],
1978 [ 'column2', 'value2' ],
1979 [ 'column3', 'value3' ],
1982 You can then iterate through this manually, using DBI's C<bind_param()>.
1984 $sth->prepare($stmt);
1987 my($col, $data) = @$_;
1988 if ($col eq 'details' || $col eq 'comments') {
1989 $sth->bind_param($i, $data, {ora_type => ORA_CLOB});
1990 } elsif ($col eq 'image') {
1991 $sth->bind_param($i, $data, {ora_type => ORA_BLOB});
1993 $sth->bind_param($i, $data);
1997 $sth->execute; # execute without @bind now
1999 Now, why would you still use B<SQL::Abstract> if you have to do this crap?
2000 Basically, the advantage is still that you don't have to care which fields
2001 are or are not included. You could wrap that above C<for> loop in a simple
2002 sub called C<bind_fields()> or something and reuse it repeatedly. You still
2003 get a layer of abstraction over manual SQL specification.
2005 Note that if you set L</bindtype> to C<columns>, the C<\[ $sql, @bind ]>
2006 construct (see L</Literal SQL with placeholders and bind values (subqueries)>)
2007 will expect the bind values in this format.
2011 This is the character that a table or column name will be quoted
2012 with. By default this is an empty string, but you could set it to
2013 the character C<`>, to generate SQL like this:
2015 SELECT `a_field` FROM `a_table` WHERE `some_field` LIKE '%someval%'
2017 Alternatively, you can supply an array ref of two items, the first being the left
2018 hand quote character, and the second the right hand quote character. For
2019 example, you could supply C<['[',']']> for SQL Server 2000 compliant quotes
2020 that generates SQL like this:
2022 SELECT [a_field] FROM [a_table] WHERE [some_field] LIKE '%someval%'
2024 Quoting is useful if you have tables or columns names that are reserved
2025 words in your database's SQL dialect.
2029 This is the character that will be used to escape L</quote_char>s appearing
2030 in an identifier before it has been quoted.
2032 The parameter default in case of a single L</quote_char> character is the quote
2035 When opening-closing-style quoting is used (L</quote_char> is an arrayref)
2036 this parameter defaults to the B<closing (right)> L</quote_char>. Occurrences
2037 of the B<opening (left)> L</quote_char> within the identifier are currently left
2038 untouched. The default for opening-closing-style quotes may change in future
2039 versions, thus you are B<strongly encouraged> to specify the escape character
2044 This is the character that separates a table and column name. It is
2045 necessary to specify this when the C<quote_char> option is selected,
2046 so that tables and column names can be individually quoted like this:
2048 SELECT `table`.`one_field` FROM `table` WHERE `table`.`other_field` = 1
2050 =item injection_guard
2052 A regular expression C<qr/.../> that is applied to any C<-function> and unquoted
2053 column name specified in a query structure. This is a safety mechanism to avoid
2054 injection attacks when mishandling user input e.g.:
2056 my %condition_as_column_value_pairs = get_values_from_user();
2057 $sqla->select( ... , \%condition_as_column_value_pairs );
2059 If the expression matches an exception is thrown. Note that literal SQL
2060 supplied via C<\'...'> or C<\['...']> is B<not> checked in any way.
2062 Defaults to checking for C<;> and the C<GO> keyword (TransactSQL)
2064 =item array_datatypes
2066 When this option is true, arrayrefs in INSERT or UPDATE are
2067 interpreted as array datatypes and are passed directly
2069 When this option is false, arrayrefs are interpreted
2070 as literal SQL, just like refs to arrayrefs
2071 (but this behavior is for backwards compatibility; when writing
2072 new queries, use the "reference to arrayref" syntax
2078 Takes a reference to a list of "special operators"
2079 to extend the syntax understood by L<SQL::Abstract>.
2080 See section L</"SPECIAL OPERATORS"> for details.
2084 Takes a reference to a list of "unary operators"
2085 to extend the syntax understood by L<SQL::Abstract>.
2086 See section L</"UNARY OPERATORS"> for details.
2092 =head2 insert($table, \@values || \%fieldvals, \%options)
2094 This is the simplest function. You simply give it a table name
2095 and either an arrayref of values or hashref of field/value pairs.
2096 It returns an SQL INSERT statement and a list of bind values.
2097 See the sections on L</"Inserting and Updating Arrays"> and
2098 L</"Inserting and Updating SQL"> for information on how to insert
2099 with those data types.
2101 The optional C<\%options> hash reference may contain additional
2102 options to generate the insert SQL. Currently supported options
2109 Takes either a scalar of raw SQL fields, or an array reference of
2110 field names, and adds on an SQL C<RETURNING> statement at the end.
2111 This allows you to return data generated by the insert statement
2112 (such as row IDs) without performing another C<SELECT> statement.
2113 Note, however, this is not part of the SQL standard and may not
2114 be supported by all database engines.
2118 =head2 update($table, \%fieldvals, \%where, \%options)
2120 This takes a table, hashref of field/value pairs, and an optional
2121 hashref L<WHERE clause|/WHERE CLAUSES>. It returns an SQL UPDATE function and a list
2123 See the sections on L</"Inserting and Updating Arrays"> and
2124 L</"Inserting and Updating SQL"> for information on how to insert
2125 with those data types.
2127 The optional C<\%options> hash reference may contain additional
2128 options to generate the update SQL. Currently supported options
2135 See the C<returning> option to
2136 L<insert|/insert($table, \@values || \%fieldvals, \%options)>.
2140 =head2 select($source, $fields, $where, $order)
2142 This returns a SQL SELECT statement and associated list of bind values, as
2143 specified by the arguments:
2149 Specification of the 'FROM' part of the statement.
2150 The argument can be either a plain scalar (interpreted as a table
2151 name, will be quoted), or an arrayref (interpreted as a list
2152 of table names, joined by commas, quoted), or a scalarref
2153 (literal SQL, not quoted).
2157 Specification of the list of fields to retrieve from
2159 The argument can be either an arrayref (interpreted as a list
2160 of field names, will be joined by commas and quoted), or a
2161 plain scalar (literal SQL, not quoted).
2162 Please observe that this API is not as flexible as that of
2163 the first argument C<$source>, for backwards compatibility reasons.
2167 Optional argument to specify the WHERE part of the query.
2168 The argument is most often a hashref, but can also be
2169 an arrayref or plain scalar --
2170 see section L<WHERE clause|/"WHERE CLAUSES"> for details.
2174 Optional argument to specify the ORDER BY part of the query.
2175 The argument can be a scalar, a hashref or an arrayref
2176 -- see section L<ORDER BY clause|/"ORDER BY CLAUSES">
2182 =head2 delete($table, \%where, \%options)
2184 This takes a table name and optional hashref L<WHERE clause|/WHERE CLAUSES>.
2185 It returns an SQL DELETE statement and list of bind values.
2187 The optional C<\%options> hash reference may contain additional
2188 options to generate the delete SQL. Currently supported options
2195 See the C<returning> option to
2196 L<insert|/insert($table, \@values || \%fieldvals, \%options)>.
2200 =head2 where(\%where, $order)
2202 This is used to generate just the WHERE clause. For example,
2203 if you have an arbitrary data structure and know what the
2204 rest of your SQL is going to look like, but want an easy way
2205 to produce a WHERE clause, use this. It returns an SQL WHERE
2206 clause and list of bind values.
2209 =head2 values(\%data)
2211 This just returns the values from the hash C<%data>, in the same
2212 order that would be returned from any of the other above queries.
2213 Using this allows you to markedly speed up your queries if you
2214 are affecting lots of rows. See below under the L</"PERFORMANCE"> section.
2216 =head2 generate($any, 'number', $of, \@data, $struct, \%types)
2218 Warning: This is an experimental method and subject to change.
2220 This returns arbitrarily generated SQL. It's a really basic shortcut.
2221 It will return two different things, depending on return context:
2223 my($stmt, @bind) = $sql->generate('create table', \$table, \@fields);
2224 my $stmt_and_val = $sql->generate('create table', \$table, \@fields);
2226 These would return the following:
2228 # First calling form
2229 $stmt = "CREATE TABLE test (?, ?)";
2230 @bind = (field1, field2);
2232 # Second calling form
2233 $stmt_and_val = "CREATE TABLE test (field1, field2)";
2235 Depending on what you're trying to do, it's up to you to choose the correct
2236 format. In this example, the second form is what you would want.
2240 $sql->generate('alter session', { nls_date_format => 'MM/YY' });
2244 ALTER SESSION SET nls_date_format = 'MM/YY'
2246 You get the idea. Strings get their case twiddled, but everything
2247 else remains verbatim.
2249 =head1 EXPORTABLE FUNCTIONS
2251 =head2 is_plain_value
2253 Determines if the supplied argument is a plain value as understood by this
2258 =item * The value is C<undef>
2260 =item * The value is a non-reference
2262 =item * The value is an object with stringification overloading
2264 =item * The value is of the form C<< { -value => $anything } >>
2268 On failure returns C<undef>, on success returns a B<scalar> reference
2269 to the original supplied argument.
2275 The stringification overloading detection is rather advanced: it takes
2276 into consideration not only the presence of a C<""> overload, but if that
2277 fails also checks for enabled
2278 L<autogenerated versions of C<"">|overload/Magic Autogeneration>, based
2279 on either C<0+> or C<bool>.
2281 Unfortunately testing in the field indicates that this
2282 detection B<< may tickle a latent bug in perl versions before 5.018 >>,
2283 but only when very large numbers of stringifying objects are involved.
2284 At the time of writing ( Sep 2014 ) there is no clear explanation of
2285 the direct cause, nor is there a manageably small test case that reliably
2286 reproduces the problem.
2288 If you encounter any of the following exceptions in B<random places within
2289 your application stack> - this module may be to blame:
2291 Operation "ne": no method found,
2292 left argument in overloaded package <something>,
2293 right argument in overloaded package <something>
2297 Stub found while resolving method "???" overloading """" in package <something>
2299 If you fall victim to the above - please attempt to reduce the problem
2300 to something that could be sent to the L<SQL::Abstract developers
2301 |DBIx::Class/GETTING HELP/SUPPORT>
2302 (either publicly or privately). As a workaround in the meantime you can
2303 set C<$ENV{SQLA_ISVALUE_IGNORE_AUTOGENERATED_STRINGIFICATION}> to a true
2304 value, which will most likely eliminate your problem (at the expense of
2305 not being able to properly detect exotic forms of stringification).
2307 This notice and environment variable will be removed in a future version,
2308 as soon as the underlying problem is found and a reliable workaround is
2313 =head2 is_literal_value
2315 Determines if the supplied argument is a literal value as understood by this
2320 =item * C<\$sql_string>
2322 =item * C<\[ $sql_string, @bind_values ]>
2326 On failure returns C<undef>, on success returns an B<array> reference
2327 containing the unpacked version of the supplied literal SQL and bind values.
2329 =head1 WHERE CLAUSES
2333 This module uses a variation on the idea from L<DBIx::Abstract>. It
2334 is B<NOT>, repeat I<not> 100% compatible. B<The main logic of this
2335 module is that things in arrays are OR'ed, and things in hashes
2338 The easiest way to explain is to show lots of examples. After
2339 each C<%where> hash shown, it is assumed you used:
2341 my($stmt, @bind) = $sql->where(\%where);
2343 However, note that the C<%where> hash can be used directly in any
2344 of the other functions as well, as described above.
2346 =head2 Key-value pairs
2348 So, let's get started. To begin, a simple hash:
2352 status => 'completed'
2355 Is converted to SQL C<key = val> statements:
2357 $stmt = "WHERE user = ? AND status = ?";
2358 @bind = ('nwiger', 'completed');
2360 One common thing I end up doing is having a list of values that
2361 a field can be in. To do this, simply specify a list inside of
2366 status => ['assigned', 'in-progress', 'pending'];
2369 This simple code will create the following:
2371 $stmt = "WHERE user = ? AND ( status = ? OR status = ? OR status = ? )";
2372 @bind = ('nwiger', 'assigned', 'in-progress', 'pending');
2374 A field associated to an empty arrayref will be considered a
2375 logical false and will generate 0=1.
2377 =head2 Tests for NULL values
2379 If the value part is C<undef> then this is converted to SQL <IS NULL>
2388 $stmt = "WHERE user = ? AND status IS NULL";
2391 To test if a column IS NOT NULL:
2395 status => { '!=', undef },
2398 =head2 Specific comparison operators
2400 If you want to specify a different type of operator for your comparison,
2401 you can use a hashref for a given column:
2405 status => { '!=', 'completed' }
2408 Which would generate:
2410 $stmt = "WHERE user = ? AND status != ?";
2411 @bind = ('nwiger', 'completed');
2413 To test against multiple values, just enclose the values in an arrayref:
2415 status => { '=', ['assigned', 'in-progress', 'pending'] };
2417 Which would give you:
2419 "WHERE status = ? OR status = ? OR status = ?"
2422 The hashref can also contain multiple pairs, in which case it is expanded
2423 into an C<AND> of its elements:
2427 status => { '!=', 'completed', -not_like => 'pending%' }
2430 # Or more dynamically, like from a form
2431 $where{user} = 'nwiger';
2432 $where{status}{'!='} = 'completed';
2433 $where{status}{'-not_like'} = 'pending%';
2435 # Both generate this
2436 $stmt = "WHERE user = ? AND status != ? AND status NOT LIKE ?";
2437 @bind = ('nwiger', 'completed', 'pending%');
2440 To get an OR instead, you can combine it with the arrayref idea:
2444 priority => [ { '=', 2 }, { '>', 5 } ]
2447 Which would generate:
2449 $stmt = "WHERE ( priority = ? OR priority > ? ) AND user = ?";
2450 @bind = ('2', '5', 'nwiger');
2452 If you want to include literal SQL (with or without bind values), just use a
2453 scalar reference or reference to an arrayref as the value:
2456 date_entered => { '>' => \["to_date(?, 'MM/DD/YYYY')", "11/26/2008"] },
2457 date_expires => { '<' => \"now()" }
2460 Which would generate:
2462 $stmt = "WHERE date_entered > to_date(?, 'MM/DD/YYYY') AND date_expires < now()";
2463 @bind = ('11/26/2008');
2466 =head2 Logic and nesting operators
2468 In the example above,
2469 there is a subtle trap if you want to say something like
2470 this (notice the C<AND>):
2472 WHERE priority != ? AND priority != ?
2474 Because, in Perl you I<can't> do this:
2476 priority => { '!=' => 2, '!=' => 1 }
2478 As the second C<!=> key will obliterate the first. The solution
2479 is to use the special C<-modifier> form inside an arrayref:
2481 priority => [ -and => {'!=', 2},
2485 Normally, these would be joined by C<OR>, but the modifier tells it
2486 to use C<AND> instead. (Hint: You can use this in conjunction with the
2487 C<logic> option to C<new()> in order to change the way your queries
2488 work by default.) B<Important:> Note that the C<-modifier> goes
2489 B<INSIDE> the arrayref, as an extra first element. This will
2490 B<NOT> do what you think it might:
2492 priority => -and => [{'!=', 2}, {'!=', 1}] # WRONG!
2494 Here is a quick list of equivalencies, since there is some overlap:
2497 status => {'!=', 'completed', 'not like', 'pending%' }
2498 status => [ -and => {'!=', 'completed'}, {'not like', 'pending%'}]
2501 status => {'=', ['assigned', 'in-progress']}
2502 status => [ -or => {'=', 'assigned'}, {'=', 'in-progress'}]
2503 status => [ {'=', 'assigned'}, {'=', 'in-progress'} ]
2507 =head2 Special operators: IN, BETWEEN, etc.
2509 You can also use the hashref format to compare a list of fields using the
2510 C<IN> comparison operator, by specifying the list as an arrayref:
2513 status => 'completed',
2514 reportid => { -in => [567, 2335, 2] }
2517 Which would generate:
2519 $stmt = "WHERE status = ? AND reportid IN (?,?,?)";
2520 @bind = ('completed', '567', '2335', '2');
2522 The reverse operator C<-not_in> generates SQL C<NOT IN> and is used in
2525 If the argument to C<-in> is an empty array, 'sqlfalse' is generated
2526 (by default: C<1=0>). Similarly, C<< -not_in => [] >> generates
2527 'sqltrue' (by default: C<1=1>).
2529 In addition to the array you can supply a chunk of literal sql or
2530 literal sql with bind:
2533 customer => { -in => \[
2534 'SELECT cust_id FROM cust WHERE balance > ?',
2537 status => { -in => \'SELECT status_codes FROM states' },
2543 customer IN ( SELECT cust_id FROM cust WHERE balance > ? )
2544 AND status IN ( SELECT status_codes FROM states )
2548 Finally, if the argument to C<-in> is not a reference, it will be
2549 treated as a single-element array.
2551 Another pair of operators is C<-between> and C<-not_between>,
2552 used with an arrayref of two values:
2556 completion_date => {
2557 -not_between => ['2002-10-01', '2003-02-06']
2563 WHERE user = ? AND completion_date NOT BETWEEN ( ? AND ? )
2565 Just like with C<-in> all plausible combinations of literal SQL
2569 start0 => { -between => [ 1, 2 ] },
2570 start1 => { -between => \["? AND ?", 1, 2] },
2571 start2 => { -between => \"lower(x) AND upper(y)" },
2572 start3 => { -between => [
2574 \["upper(?)", 'stuff' ],
2581 ( start0 BETWEEN ? AND ? )
2582 AND ( start1 BETWEEN ? AND ? )
2583 AND ( start2 BETWEEN lower(x) AND upper(y) )
2584 AND ( start3 BETWEEN lower(x) AND upper(?) )
2586 @bind = (1, 2, 1, 2, 'stuff');
2589 These are the two builtin "special operators"; but the
2590 list can be expanded: see section L</"SPECIAL OPERATORS"> below.
2592 =head2 Unary operators: bool
2594 If you wish to test against boolean columns or functions within your
2595 database you can use the C<-bool> and C<-not_bool> operators. For
2596 example to test the column C<is_user> being true and the column
2597 C<is_enabled> being false you would use:-
2601 -not_bool => 'is_enabled',
2606 WHERE is_user AND NOT is_enabled
2608 If a more complex combination is required, testing more conditions,
2609 then you should use the and/or operators:-
2614 -not_bool => { two=> { -rlike => 'bar' } },
2615 -not_bool => { three => [ { '=', 2 }, { '>', 5 } ] },
2626 (NOT ( three = ? OR three > ? ))
2629 =head2 Nested conditions, -and/-or prefixes
2631 So far, we've seen how multiple conditions are joined with a top-level
2632 C<AND>. We can change this by putting the different conditions we want in
2633 hashes and then putting those hashes in an array. For example:
2638 status => { -like => ['pending%', 'dispatched'] },
2642 status => 'unassigned',
2646 This data structure would create the following:
2648 $stmt = "WHERE ( user = ? AND ( status LIKE ? OR status LIKE ? ) )
2649 OR ( user = ? AND status = ? ) )";
2650 @bind = ('nwiger', 'pending', 'dispatched', 'robot', 'unassigned');
2653 Clauses in hashrefs or arrayrefs can be prefixed with an C<-and> or C<-or>
2654 to change the logic inside:
2660 -and => [ workhrs => {'>', 20}, geo => 'ASIA' ],
2661 -or => { workhrs => {'<', 50}, geo => 'EURO' },
2668 $stmt = "WHERE ( user = ?
2669 AND ( ( workhrs > ? AND geo = ? )
2670 OR ( workhrs < ? OR geo = ? ) ) )";
2671 @bind = ('nwiger', '20', 'ASIA', '50', 'EURO');
2673 =head3 Algebraic inconsistency, for historical reasons
2675 C<Important note>: when connecting several conditions, the C<-and->|C<-or>
2676 operator goes C<outside> of the nested structure; whereas when connecting
2677 several constraints on one column, the C<-and> operator goes
2678 C<inside> the arrayref. Here is an example combining both features:
2681 -and => [a => 1, b => 2],
2682 -or => [c => 3, d => 4],
2683 e => [-and => {-like => 'foo%'}, {-like => '%bar'} ]
2688 WHERE ( ( ( a = ? AND b = ? )
2689 OR ( c = ? OR d = ? )
2690 OR ( e LIKE ? AND e LIKE ? ) ) )
2692 This difference in syntax is unfortunate but must be preserved for
2693 historical reasons. So be careful: the two examples below would
2694 seem algebraically equivalent, but they are not
2697 { -like => 'foo%' },
2698 { -like => '%bar' },
2700 # yields: WHERE ( ( col LIKE ? AND col LIKE ? ) )
2703 { col => { -like => 'foo%' } },
2704 { col => { -like => '%bar' } },
2706 # yields: WHERE ( ( col LIKE ? OR col LIKE ? ) )
2709 =head2 Literal SQL and value type operators
2711 The basic premise of SQL::Abstract is that in WHERE specifications the "left
2712 side" is a column name and the "right side" is a value (normally rendered as
2713 a placeholder). This holds true for both hashrefs and arrayref pairs as you
2714 see in the L</WHERE CLAUSES> examples above. Sometimes it is necessary to
2715 alter this behavior. There are several ways of doing so.
2719 This is a virtual operator that signals the string to its right side is an
2720 identifier (a column name) and not a value. For example to compare two
2721 columns you would write:
2724 priority => { '<', 2 },
2725 requestor => { -ident => 'submitter' },
2730 $stmt = "WHERE priority < ? AND requestor = submitter";
2733 If you are maintaining legacy code you may see a different construct as
2734 described in L</Deprecated usage of Literal SQL>, please use C<-ident> in new
2739 This is a virtual operator that signals that the construct to its right side
2740 is a value to be passed to DBI. This is for example necessary when you want
2741 to write a where clause against an array (for RDBMS that support such
2742 datatypes). For example:
2745 array => { -value => [1, 2, 3] }
2750 $stmt = 'WHERE array = ?';
2751 @bind = ([1, 2, 3]);
2753 Note that if you were to simply say:
2759 the result would probably not be what you wanted:
2761 $stmt = 'WHERE array = ? OR array = ? OR array = ?';
2766 Finally, sometimes only literal SQL will do. To include a random snippet
2767 of SQL verbatim, you specify it as a scalar reference. Consider this only
2768 as a last resort. Usually there is a better way. For example:
2771 priority => { '<', 2 },
2772 requestor => { -in => \'(SELECT name FROM hitmen)' },
2777 $stmt = "WHERE priority < ? AND requestor IN (SELECT name FROM hitmen)"
2780 Note that in this example, you only get one bind parameter back, since
2781 the verbatim SQL is passed as part of the statement.
2785 Never use untrusted input as a literal SQL argument - this is a massive
2786 security risk (there is no way to check literal snippets for SQL
2787 injections and other nastyness). If you need to deal with untrusted input
2788 use literal SQL with placeholders as described next.
2790 =head3 Literal SQL with placeholders and bind values (subqueries)
2792 If the literal SQL to be inserted has placeholders and bind values,
2793 use a reference to an arrayref (yes this is a double reference --
2794 not so common, but perfectly legal Perl). For example, to find a date
2795 in Postgres you can use something like this:
2798 date_column => \[ "= date '2008-09-30' - ?::integer", 10 ]
2803 $stmt = "WHERE ( date_column = date '2008-09-30' - ?::integer )"
2806 Note that you must pass the bind values in the same format as they are returned
2807 by L<where|/where(\%where, $order)>. This means that if you set L</bindtype>
2808 to C<columns>, you must provide the bind values in the
2809 C<< [ column_meta => value ] >> format, where C<column_meta> is an opaque
2810 scalar value; most commonly the column name, but you can use any scalar value
2811 (including references and blessed references), L<SQL::Abstract> will simply
2812 pass it through intact. So if C<bindtype> is set to C<columns> the above
2813 example will look like:
2816 date_column => \[ "= date '2008-09-30' - ?::integer", [ {} => 10 ] ]
2819 Literal SQL is especially useful for nesting parenthesized clauses in the
2820 main SQL query. Here is a first example:
2822 my ($sub_stmt, @sub_bind) = ("SELECT c1 FROM t1 WHERE c2 < ? AND c3 LIKE ?",
2826 bar => \["IN ($sub_stmt)" => @sub_bind],
2831 $stmt = "WHERE (foo = ? AND bar IN (SELECT c1 FROM t1
2832 WHERE c2 < ? AND c3 LIKE ?))";
2833 @bind = (1234, 100, "foo%");
2835 Other subquery operators, like for example C<"E<gt> ALL"> or C<"NOT IN">,
2836 are expressed in the same way. Of course the C<$sub_stmt> and
2837 its associated bind values can be generated through a former call
2840 my ($sub_stmt, @sub_bind)
2841 = $sql->select("t1", "c1", {c2 => {"<" => 100},
2842 c3 => {-like => "foo%"}});
2845 bar => \["> ALL ($sub_stmt)" => @sub_bind],
2848 In the examples above, the subquery was used as an operator on a column;
2849 but the same principle also applies for a clause within the main C<%where>
2850 hash, like an EXISTS subquery:
2852 my ($sub_stmt, @sub_bind)
2853 = $sql->select("t1", "*", {c1 => 1, c2 => \"> t0.c0"});
2854 my %where = ( -and => [
2856 \["EXISTS ($sub_stmt)" => @sub_bind],
2861 $stmt = "WHERE (foo = ? AND EXISTS (SELECT * FROM t1
2862 WHERE c1 = ? AND c2 > t0.c0))";
2866 Observe that the condition on C<c2> in the subquery refers to
2867 column C<t0.c0> of the main query: this is I<not> a bind
2868 value, so we have to express it through a scalar ref.
2869 Writing C<< c2 => {">" => "t0.c0"} >> would have generated
2870 C<< c2 > ? >> with bind value C<"t0.c0"> ... not exactly
2871 what we wanted here.
2873 Finally, here is an example where a subquery is used
2874 for expressing unary negation:
2876 my ($sub_stmt, @sub_bind)
2877 = $sql->where({age => [{"<" => 10}, {">" => 20}]});
2878 $sub_stmt =~ s/^ where //i; # don't want "WHERE" in the subclause
2880 lname => {like => '%son%'},
2881 \["NOT ($sub_stmt)" => @sub_bind],
2886 $stmt = "lname LIKE ? AND NOT ( age < ? OR age > ? )"
2887 @bind = ('%son%', 10, 20)
2889 =head3 Deprecated usage of Literal SQL
2891 Below are some examples of archaic use of literal SQL. It is shown only as
2892 reference for those who deal with legacy code. Each example has a much
2893 better, cleaner and safer alternative that users should opt for in new code.
2899 my %where = ( requestor => \'IS NOT NULL' )
2901 $stmt = "WHERE requestor IS NOT NULL"
2903 This used to be the way of generating NULL comparisons, before the handling
2904 of C<undef> got formalized. For new code please use the superior syntax as
2905 described in L</Tests for NULL values>.
2909 my %where = ( requestor => \'= submitter' )
2911 $stmt = "WHERE requestor = submitter"
2913 This used to be the only way to compare columns. Use the superior L</-ident>
2914 method for all new code. For example an identifier declared in such a way
2915 will be properly quoted if L</quote_char> is properly set, while the legacy
2916 form will remain as supplied.
2920 my %where = ( is_ready => \"", completed => { '>', '2012-12-21' } )
2922 $stmt = "WHERE completed > ? AND is_ready"
2923 @bind = ('2012-12-21')
2925 Using an empty string literal used to be the only way to express a boolean.
2926 For all new code please use the much more readable
2927 L<-bool|/Unary operators: bool> operator.
2933 These pages could go on for a while, since the nesting of the data
2934 structures this module can handle are pretty much unlimited (the
2935 module implements the C<WHERE> expansion as a recursive function
2936 internally). Your best bet is to "play around" with the module a
2937 little to see how the data structures behave, and choose the best
2938 format for your data based on that.
2940 And of course, all the values above will probably be replaced with
2941 variables gotten from forms or the command line. After all, if you
2942 knew everything ahead of time, you wouldn't have to worry about
2943 dynamically-generating SQL and could just hardwire it into your
2946 =head1 ORDER BY CLAUSES
2948 Some functions take an order by clause. This can either be a scalar (just a
2949 column name), a hashref of C<< { -desc => 'col' } >> or C<< { -asc => 'col' }
2950 >>, a scalarref, an arrayref-ref, or an arrayref of any of the previous
2953 Given | Will Generate
2954 ---------------------------------------------------------------
2956 'colA' | ORDER BY colA
2958 [qw/colA colB/] | ORDER BY colA, colB
2960 {-asc => 'colA'} | ORDER BY colA ASC
2962 {-desc => 'colB'} | ORDER BY colB DESC
2964 ['colA', {-asc => 'colB'}] | ORDER BY colA, colB ASC
2966 { -asc => [qw/colA colB/] } | ORDER BY colA ASC, colB ASC
2968 \'colA DESC' | ORDER BY colA DESC
2970 \[ 'FUNC(colA, ?)', $x ] | ORDER BY FUNC(colA, ?)
2971 | /* ...with $x bound to ? */
2974 { -asc => 'colA' }, | colA ASC,
2975 { -desc => [qw/colB/] }, | colB DESC,
2976 { -asc => [qw/colC colD/] },| colC ASC, colD ASC,
2977 \'colE DESC', | colE DESC,
2978 \[ 'FUNC(colF, ?)', $x ], | FUNC(colF, ?)
2979 ] | /* ...with $x bound to ? */
2980 ===============================================================
2984 =head1 SPECIAL OPERATORS
2986 my $sqlmaker = SQL::Abstract->new(special_ops => [
2990 my ($self, $field, $op, $arg) = @_;
2996 handler => 'method_name',
3000 A "special operator" is a SQL syntactic clause that can be
3001 applied to a field, instead of a usual binary operator.
3004 WHERE field IN (?, ?, ?)
3005 WHERE field BETWEEN ? AND ?
3006 WHERE MATCH(field) AGAINST (?, ?)
3008 Special operators IN and BETWEEN are fairly standard and therefore
3009 are builtin within C<SQL::Abstract> (as the overridable methods
3010 C<_where_field_IN> and C<_where_field_BETWEEN>). For other operators,
3011 like the MATCH .. AGAINST example above which is specific to MySQL,
3012 you can write your own operator handlers - supply a C<special_ops>
3013 argument to the C<new> method. That argument takes an arrayref of
3014 operator definitions; each operator definition is a hashref with two
3021 the regular expression to match the operator
3025 Either a coderef or a plain scalar method name. In both cases
3026 the expected return is C<< ($sql, @bind) >>.
3028 When supplied with a method name, it is simply called on the
3029 L<SQL::Abstract> object as:
3031 $self->$method_name($field, $op, $arg)
3035 $field is the LHS of the operator
3036 $op is the part that matched the handler regex
3039 When supplied with a coderef, it is called as:
3041 $coderef->($self, $field, $op, $arg)
3046 For example, here is an implementation
3047 of the MATCH .. AGAINST syntax for MySQL
3049 my $sqlmaker = SQL::Abstract->new(special_ops => [
3051 # special op for MySql MATCH (field) AGAINST(word1, word2, ...)
3052 {regex => qr/^match$/i,
3054 my ($self, $field, $op, $arg) = @_;
3055 $arg = [$arg] if not ref $arg;
3056 my $label = $self->_quote($field);
3057 my ($placeholder) = $self->_convert('?');
3058 my $placeholders = join ", ", (($placeholder) x @$arg);
3059 my $sql = $self->_sqlcase('match') . " ($label) "
3060 . $self->_sqlcase('against') . " ($placeholders) ";
3061 my @bind = $self->_bindtype($field, @$arg);
3062 return ($sql, @bind);
3069 =head1 UNARY OPERATORS
3071 my $sqlmaker = SQL::Abstract->new(unary_ops => [
3075 my ($self, $op, $arg) = @_;
3081 handler => 'method_name',
3085 A "unary operator" is a SQL syntactic clause that can be
3086 applied to a field - the operator goes before the field
3088 You can write your own operator handlers - supply a C<unary_ops>
3089 argument to the C<new> method. That argument takes an arrayref of
3090 operator definitions; each operator definition is a hashref with two
3097 the regular expression to match the operator
3101 Either a coderef or a plain scalar method name. In both cases
3102 the expected return is C<< $sql >>.
3104 When supplied with a method name, it is simply called on the
3105 L<SQL::Abstract> object as:
3107 $self->$method_name($op, $arg)
3111 $op is the part that matched the handler regex
3112 $arg is the RHS or argument of the operator
3114 When supplied with a coderef, it is called as:
3116 $coderef->($self, $op, $arg)
3124 Thanks to some benchmarking by Mark Stosberg, it turns out that
3125 this module is many orders of magnitude faster than using C<DBIx::Abstract>.
3126 I must admit this wasn't an intentional design issue, but it's a
3127 byproduct of the fact that you get to control your C<DBI> handles
3130 To maximize performance, use a code snippet like the following:
3132 # prepare a statement handle using the first row
3133 # and then reuse it for the rest of the rows
3135 for my $href (@array_of_hashrefs) {
3136 $stmt ||= $sql->insert('table', $href);
3137 $sth ||= $dbh->prepare($stmt);
3138 $sth->execute($sql->values($href));
3141 The reason this works is because the keys in your C<$href> are sorted
3142 internally by B<SQL::Abstract>. Thus, as long as your data retains
3143 the same structure, you only have to generate the SQL the first time
3144 around. On subsequent queries, simply use the C<values> function provided
3145 by this module to return your values in the correct order.
3147 However this depends on the values having the same type - if, for
3148 example, the values of a where clause may either have values
3149 (resulting in sql of the form C<column = ?> with a single bind
3150 value), or alternatively the values might be C<undef> (resulting in
3151 sql of the form C<column IS NULL> with no bind value) then the
3152 caching technique suggested will not work.
3156 If you use my C<CGI::FormBuilder> module at all, you'll hopefully
3157 really like this part (I do, at least). Building up a complex query
3158 can be as simple as the following:
3165 use CGI::FormBuilder;
3168 my $form = CGI::FormBuilder->new(...);
3169 my $sql = SQL::Abstract->new;
3171 if ($form->submitted) {
3172 my $field = $form->field;
3173 my $id = delete $field->{id};
3174 my($stmt, @bind) = $sql->update('table', $field, {id => $id});
3177 Of course, you would still have to connect using C<DBI> to run the
3178 query, but the point is that if you make your form look like your
3179 table, the actual query script can be extremely simplistic.
3181 If you're B<REALLY> lazy (I am), check out C<HTML::QuickTable> for
3182 a fast interface to returning and formatting data. I frequently
3183 use these three modules together to write complex database query
3184 apps in under 50 lines.
3186 =head1 HOW TO CONTRIBUTE
3188 Contributions are always welcome, in all usable forms (we especially
3189 welcome documentation improvements). The delivery methods include git-
3190 or unified-diff formatted patches, GitHub pull requests, or plain bug
3191 reports either via RT or the Mailing list. Contributors are generally
3192 granted full access to the official repository after their first several
3193 patches pass successful review.
3195 This project is maintained in a git repository. The code and related tools are
3196 accessible at the following locations:
3200 =item * Official repo: L<git://git.shadowcat.co.uk/dbsrgits/SQL-Abstract.git>
3202 =item * Official gitweb: L<http://git.shadowcat.co.uk/gitweb/gitweb.cgi?p=dbsrgits/SQL-Abstract.git>
3204 =item * GitHub mirror: L<https://github.com/dbsrgits/sql-abstract>
3206 =item * Authorized committers: L<ssh://dbsrgits@git.shadowcat.co.uk/SQL-Abstract.git>
3212 Version 1.50 was a major internal refactoring of C<SQL::Abstract>.
3213 Great care has been taken to preserve the I<published> behavior
3214 documented in previous versions in the 1.* family; however,
3215 some features that were previously undocumented, or behaved
3216 differently from the documentation, had to be changed in order
3217 to clarify the semantics. Hence, client code that was relying
3218 on some dark areas of C<SQL::Abstract> v1.*
3219 B<might behave differently> in v1.50.
3221 The main changes are:
3227 support for literal SQL through the C<< \ [ $sql, @bind ] >> syntax.
3231 support for the { operator => \"..." } construct (to embed literal SQL)
3235 support for the { operator => \["...", @bind] } construct (to embed literal SQL with bind values)
3239 optional support for L<array datatypes|/"Inserting and Updating Arrays">
3243 defensive programming: check arguments
3247 fixed bug with global logic, which was previously implemented
3248 through global variables yielding side-effects. Prior versions would
3249 interpret C<< [ {cond1, cond2}, [cond3, cond4] ] >>
3250 as C<< "(cond1 AND cond2) OR (cond3 AND cond4)" >>.
3251 Now this is interpreted
3252 as C<< "(cond1 AND cond2) OR (cond3 OR cond4)" >>.
3257 fixed semantics of _bindtype on array args
3261 dropped the C<_anoncopy> of the %where tree. No longer necessary,
3262 we just avoid shifting arrays within that tree.
3266 dropped the C<_modlogic> function
3270 =head1 ACKNOWLEDGEMENTS
3272 There are a number of individuals that have really helped out with
3273 this module. Unfortunately, most of them submitted bugs via CPAN
3274 so I have no idea who they are! But the people I do know are:
3276 Ash Berlin (order_by hash term support)
3277 Matt Trout (DBIx::Class support)
3278 Mark Stosberg (benchmarking)
3279 Chas Owens (initial "IN" operator support)
3280 Philip Collins (per-field SQL functions)
3281 Eric Kolve (hashref "AND" support)
3282 Mike Fragassi (enhancements to "BETWEEN" and "LIKE")
3283 Dan Kubb (support for "quote_char" and "name_sep")
3284 Guillermo Roditi (patch to cleanup "IN" and "BETWEEN", fix and tests for _order_by)
3285 Laurent Dami (internal refactoring, extensible list of special operators, literal SQL)
3286 Norbert Buchmuller (support for literal SQL in hashpair, misc. fixes & tests)
3287 Peter Rabbitson (rewrite of SQLA::Test, misc. fixes & tests)
3288 Oliver Charles (support for "RETURNING" after "INSERT")
3294 L<DBIx::Class>, L<DBIx::Abstract>, L<CGI::FormBuilder>, L<HTML::QuickTable>.
3298 Copyright (c) 2001-2007 Nathan Wiger <nwiger@cpan.org>. All Rights Reserved.
3300 This module is actively maintained by Matt Trout <mst@shadowcatsystems.co.uk>
3302 For support, your best bet is to try the C<DBIx::Class> users mailing list.
3303 While not an official support venue, C<DBIx::Class> makes heavy use of
3304 C<SQL::Abstract>, and as such list members there are very familiar with
3305 how to create queries.
3309 This module is free software; you may copy this under the same
3310 terms as perl itself (either the GNU General Public License or
3311 the Artistic License)