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 = '1.86';
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} $v");
1092 return ($self->_where_hashpair_HASHREF($k, { $self->{cmp} => $v }));
1096 sub _where_hashpair_UNDEF {
1097 my ($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 { $self->_quote($_) } @$from;},
1411 SCALAR => sub {$self->_quote($from)},
1412 SCALARREF => sub {$$from},
1417 #======================================================================
1419 #======================================================================
1421 # highly optimized, as it's called way too often
1423 # my ($self, $label) = @_;
1425 return '' unless defined $_[1];
1426 return ${$_[1]} if ref($_[1]) eq 'SCALAR';
1428 $_[0]->{quote_char} or
1429 ($_[0]->_assert_pass_injection_guard($_[1]), return $_[1]);
1431 my $qref = ref $_[0]->{quote_char};
1433 !$qref ? ($_[0]->{quote_char}, $_[0]->{quote_char})
1434 : ($qref eq 'ARRAY') ? @{$_[0]->{quote_char}}
1435 : puke "Unsupported quote_char format: $_[0]->{quote_char}";
1437 my $esc = $_[0]->{escape_char} || $r;
1439 # parts containing * are naturally unquoted
1440 return join($_[0]->{name_sep}||'', map
1441 +( $_ eq '*' ? $_ : do { (my $n = $_) =~ s/(\Q$esc\E|\Q$r\E)/$esc$1/g; $l . $n . $r } ),
1442 ( $_[0]->{name_sep} ? split (/\Q$_[0]->{name_sep}\E/, $_[1] ) : $_[1] )
1447 # Conversion, if applicable
1449 #my ($self, $arg) = @_;
1450 if ($_[0]->{convert}) {
1451 return $_[0]->_sqlcase($_[0]->{convert}) .'(' . $_[1] . ')';
1458 #my ($self, $col, @vals) = @_;
1459 # called often - tighten code
1460 return $_[0]->{bindtype} eq 'columns'
1461 ? map {[$_[1], $_]} @_[2 .. $#_]
1466 # Dies if any element of @bind is not in [colname => value] format
1467 # if bindtype is 'columns'.
1468 sub _assert_bindval_matches_bindtype {
1469 # my ($self, @bind) = @_;
1471 if ($self->{bindtype} eq 'columns') {
1473 if (!defined $_ || ref($_) ne 'ARRAY' || @$_ != 2) {
1474 puke "bindtype 'columns' selected, you need to pass: [column_name => bind_value]"
1480 sub _join_sql_clauses {
1481 my ($self, $logic, $clauses_aref, $bind_aref) = @_;
1483 if (@$clauses_aref > 1) {
1484 my $join = " " . $self->_sqlcase($logic) . " ";
1485 my $sql = '( ' . join($join, @$clauses_aref) . ' )';
1486 return ($sql, @$bind_aref);
1488 elsif (@$clauses_aref) {
1489 return ($clauses_aref->[0], @$bind_aref); # no parentheses
1492 return (); # if no SQL, ignore @$bind_aref
1497 # Fix SQL case, if so requested
1499 # LDNOTE: if $self->{case} is true, then it contains 'lower', so we
1500 # don't touch the argument ... crooked logic, but let's not change it!
1501 return $_[0]->{case} ? $_[1] : uc($_[1]);
1505 #======================================================================
1506 # DISPATCHING FROM REFKIND
1507 #======================================================================
1510 my ($self, $data) = @_;
1512 return 'UNDEF' unless defined $data;
1514 # blessed objects are treated like scalars
1515 my $ref = (Scalar::Util::blessed $data) ? '' : ref $data;
1517 return 'SCALAR' unless $ref;
1520 while ($ref eq 'REF') {
1522 $ref = (Scalar::Util::blessed $data) ? '' : ref $data;
1526 return ($ref||'SCALAR') . ('REF' x $n_steps);
1530 my ($self, $data) = @_;
1531 my @try = ($self->_refkind($data));
1532 push @try, 'SCALAR_or_UNDEF' if $try[0] eq 'SCALAR' || $try[0] eq 'UNDEF';
1533 push @try, 'FALLBACK';
1537 sub _METHOD_FOR_refkind {
1538 my ($self, $meth_prefix, $data) = @_;
1541 for (@{$self->_try_refkind($data)}) {
1542 $method = $self->can($meth_prefix."_".$_)
1546 return $method || puke "cannot dispatch on '$meth_prefix' for ".$self->_refkind($data);
1550 sub _SWITCH_refkind {
1551 my ($self, $data, $dispatch_table) = @_;
1554 for (@{$self->_try_refkind($data)}) {
1555 $coderef = $dispatch_table->{$_}
1559 puke "no dispatch entry for ".$self->_refkind($data)
1568 #======================================================================
1569 # VALUES, GENERATE, AUTOLOAD
1570 #======================================================================
1572 # LDNOTE: original code from nwiger, didn't touch code in that section
1573 # I feel the AUTOLOAD stuff should not be the default, it should
1574 # only be activated on explicit demand by user.
1578 my $data = shift || return;
1579 puke "Argument to ", __PACKAGE__, "->values must be a \\%hash"
1580 unless ref $data eq 'HASH';
1583 foreach my $k (sort keys %$data) {
1584 my $v = $data->{$k};
1585 $self->_SWITCH_refkind($v, {
1587 if ($self->{array_datatypes}) { # array datatype
1588 push @all_bind, $self->_bindtype($k, $v);
1590 else { # literal SQL with bind
1591 my ($sql, @bind) = @$v;
1592 $self->_assert_bindval_matches_bindtype(@bind);
1593 push @all_bind, @bind;
1596 ARRAYREFREF => sub { # literal SQL with bind
1597 my ($sql, @bind) = @${$v};
1598 $self->_assert_bindval_matches_bindtype(@bind);
1599 push @all_bind, @bind;
1601 SCALARREF => sub { # literal SQL without bind
1603 SCALAR_or_UNDEF => sub {
1604 push @all_bind, $self->_bindtype($k, $v);
1615 my(@sql, @sqlq, @sqlv);
1619 if ($ref eq 'HASH') {
1620 for my $k (sort keys %$_) {
1623 my $label = $self->_quote($k);
1624 if ($r eq 'ARRAY') {
1625 # literal SQL with bind
1626 my ($sql, @bind) = @$v;
1627 $self->_assert_bindval_matches_bindtype(@bind);
1628 push @sqlq, "$label = $sql";
1630 } elsif ($r eq 'SCALAR') {
1631 # literal SQL without bind
1632 push @sqlq, "$label = $$v";
1634 push @sqlq, "$label = ?";
1635 push @sqlv, $self->_bindtype($k, $v);
1638 push @sql, $self->_sqlcase('set'), join ', ', @sqlq;
1639 } elsif ($ref eq 'ARRAY') {
1640 # unlike insert(), assume these are ONLY the column names, i.e. for SQL
1643 if ($r eq 'ARRAY') { # literal SQL with bind
1644 my ($sql, @bind) = @$v;
1645 $self->_assert_bindval_matches_bindtype(@bind);
1648 } elsif ($r eq 'SCALAR') { # literal SQL without bind
1649 # embedded literal SQL
1656 push @sql, '(' . join(', ', @sqlq) . ')';
1657 } elsif ($ref eq 'SCALAR') {
1661 # strings get case twiddled
1662 push @sql, $self->_sqlcase($_);
1666 my $sql = join ' ', @sql;
1668 # this is pretty tricky
1669 # if ask for an array, return ($stmt, @bind)
1670 # otherwise, s/?/shift @sqlv/ to put it inline
1672 return ($sql, @sqlv);
1674 1 while $sql =~ s/\?/my $d = shift(@sqlv);
1675 ref $d ? $d->[1] : $d/e;
1684 # This allows us to check for a local, then _form, attr
1686 my($name) = $AUTOLOAD =~ /.*::(.+)/;
1687 return $self->generate($name, @_);
1698 SQL::Abstract - Generate SQL from Perl data structures
1704 my $sql = SQL::Abstract->new;
1706 my($stmt, @bind) = $sql->select($source, \@fields, \%where, $order);
1708 my($stmt, @bind) = $sql->insert($table, \%fieldvals || \@values);
1710 my($stmt, @bind) = $sql->update($table, \%fieldvals, \%where);
1712 my($stmt, @bind) = $sql->delete($table, \%where);
1714 # Then, use these in your DBI statements
1715 my $sth = $dbh->prepare($stmt);
1716 $sth->execute(@bind);
1718 # Just generate the WHERE clause
1719 my($stmt, @bind) = $sql->where(\%where, $order);
1721 # Return values in the same order, for hashed queries
1722 # See PERFORMANCE section for more details
1723 my @bind = $sql->values(\%fieldvals);
1727 This module was inspired by the excellent L<DBIx::Abstract>.
1728 However, in using that module I found that what I really wanted
1729 to do was generate SQL, but still retain complete control over my
1730 statement handles and use the DBI interface. So, I set out to
1731 create an abstract SQL generation module.
1733 While based on the concepts used by L<DBIx::Abstract>, there are
1734 several important differences, especially when it comes to WHERE
1735 clauses. I have modified the concepts used to make the SQL easier
1736 to generate from Perl data structures and, IMO, more intuitive.
1737 The underlying idea is for this module to do what you mean, based
1738 on the data structures you provide it. The big advantage is that
1739 you don't have to modify your code every time your data changes,
1740 as this module figures it out.
1742 To begin with, an SQL INSERT is as easy as just specifying a hash
1743 of C<key=value> pairs:
1746 name => 'Jimbo Bobson',
1747 phone => '123-456-7890',
1748 address => '42 Sister Lane',
1749 city => 'St. Louis',
1750 state => 'Louisiana',
1753 The SQL can then be generated with this:
1755 my($stmt, @bind) = $sql->insert('people', \%data);
1757 Which would give you something like this:
1759 $stmt = "INSERT INTO people
1760 (address, city, name, phone, state)
1761 VALUES (?, ?, ?, ?, ?)";
1762 @bind = ('42 Sister Lane', 'St. Louis', 'Jimbo Bobson',
1763 '123-456-7890', 'Louisiana');
1765 These are then used directly in your DBI code:
1767 my $sth = $dbh->prepare($stmt);
1768 $sth->execute(@bind);
1770 =head2 Inserting and Updating Arrays
1772 If your database has array types (like for example Postgres),
1773 activate the special option C<< array_datatypes => 1 >>
1774 when creating the C<SQL::Abstract> object.
1775 Then you may use an arrayref to insert and update database array types:
1777 my $sql = SQL::Abstract->new(array_datatypes => 1);
1779 planets => [qw/Mercury Venus Earth Mars/]
1782 my($stmt, @bind) = $sql->insert('solar_system', \%data);
1786 $stmt = "INSERT INTO solar_system (planets) VALUES (?)"
1788 @bind = (['Mercury', 'Venus', 'Earth', 'Mars']);
1791 =head2 Inserting and Updating SQL
1793 In order to apply SQL functions to elements of your C<%data> you may
1794 specify a reference to an arrayref for the given hash value. For example,
1795 if you need to execute the Oracle C<to_date> function on a value, you can
1796 say something like this:
1800 date_entered => \[ "to_date(?,'MM/DD/YYYY')", "03/02/2003" ],
1803 The first value in the array is the actual SQL. Any other values are
1804 optional and would be included in the bind values array. This gives
1807 my($stmt, @bind) = $sql->insert('people', \%data);
1809 $stmt = "INSERT INTO people (name, date_entered)
1810 VALUES (?, to_date(?,'MM/DD/YYYY'))";
1811 @bind = ('Bill', '03/02/2003');
1813 An UPDATE is just as easy, all you change is the name of the function:
1815 my($stmt, @bind) = $sql->update('people', \%data);
1817 Notice that your C<%data> isn't touched; the module will generate
1818 the appropriately quirky SQL for you automatically. Usually you'll
1819 want to specify a WHERE clause for your UPDATE, though, which is
1820 where handling C<%where> hashes comes in handy...
1822 =head2 Complex where statements
1824 This module can generate pretty complicated WHERE statements
1825 easily. For example, simple C<key=value> pairs are taken to mean
1826 equality, and if you want to see if a field is within a set
1827 of values, you can use an arrayref. Let's say we wanted to
1828 SELECT some data based on this criteria:
1831 requestor => 'inna',
1832 worker => ['nwiger', 'rcwe', 'sfz'],
1833 status => { '!=', 'completed' }
1836 my($stmt, @bind) = $sql->select('tickets', '*', \%where);
1838 The above would give you something like this:
1840 $stmt = "SELECT * FROM tickets WHERE
1841 ( requestor = ? ) AND ( status != ? )
1842 AND ( worker = ? OR worker = ? OR worker = ? )";
1843 @bind = ('inna', 'completed', 'nwiger', 'rcwe', 'sfz');
1845 Which you could then use in DBI code like so:
1847 my $sth = $dbh->prepare($stmt);
1848 $sth->execute(@bind);
1854 The methods are simple. There's one for every major SQL operation,
1855 and a constructor you use first. The arguments are specified in a
1856 similar order for each method (table, then fields, then a where
1857 clause) to try and simplify things.
1859 =head2 new(option => 'value')
1861 The C<new()> function takes a list of options and values, and returns
1862 a new B<SQL::Abstract> object which can then be used to generate SQL
1863 through the methods below. The options accepted are:
1869 If set to 'lower', then SQL will be generated in all lowercase. By
1870 default SQL is generated in "textbook" case meaning something like:
1872 SELECT a_field FROM a_table WHERE some_field LIKE '%someval%'
1874 Any setting other than 'lower' is ignored.
1878 This determines what the default comparison operator is. By default
1879 it is C<=>, meaning that a hash like this:
1881 %where = (name => 'nwiger', email => 'nate@wiger.org');
1883 Will generate SQL like this:
1885 WHERE name = 'nwiger' AND email = 'nate@wiger.org'
1887 However, you may want loose comparisons by default, so if you set
1888 C<cmp> to C<like> you would get SQL such as:
1890 WHERE name like 'nwiger' AND email like 'nate@wiger.org'
1892 You can also override the comparison on an individual basis - see
1893 the huge section on L</"WHERE CLAUSES"> at the bottom.
1895 =item sqltrue, sqlfalse
1897 Expressions for inserting boolean values within SQL statements.
1898 By default these are C<1=1> and C<1=0>. They are used
1899 by the special operators C<-in> and C<-not_in> for generating
1900 correct SQL even when the argument is an empty array (see below).
1904 This determines the default logical operator for multiple WHERE
1905 statements in arrays or hashes. If absent, the default logic is "or"
1906 for arrays, and "and" for hashes. This means that a WHERE
1910 event_date => {'>=', '2/13/99'},
1911 event_date => {'<=', '4/24/03'},
1914 will generate SQL like this:
1916 WHERE event_date >= '2/13/99' OR event_date <= '4/24/03'
1918 This is probably not what you want given this query, though (look
1919 at the dates). To change the "OR" to an "AND", simply specify:
1921 my $sql = SQL::Abstract->new(logic => 'and');
1923 Which will change the above C<WHERE> to:
1925 WHERE event_date >= '2/13/99' AND event_date <= '4/24/03'
1927 The logic can also be changed locally by inserting
1928 a modifier in front of an arrayref:
1930 @where = (-and => [event_date => {'>=', '2/13/99'},
1931 event_date => {'<=', '4/24/03'} ]);
1933 See the L</"WHERE CLAUSES"> section for explanations.
1937 This will automatically convert comparisons using the specified SQL
1938 function for both column and value. This is mostly used with an argument
1939 of C<upper> or C<lower>, so that the SQL will have the effect of
1940 case-insensitive "searches". For example, this:
1942 $sql = SQL::Abstract->new(convert => 'upper');
1943 %where = (keywords => 'MaKe iT CAse inSeNSItive');
1945 Will turn out the following SQL:
1947 WHERE upper(keywords) like upper('MaKe iT CAse inSeNSItive')
1949 The conversion can be C<upper()>, C<lower()>, or any other SQL function
1950 that can be applied symmetrically to fields (actually B<SQL::Abstract> does
1951 not validate this option; it will just pass through what you specify verbatim).
1955 This is a kludge because many databases suck. For example, you can't
1956 just bind values using DBI's C<execute()> for Oracle C<CLOB> or C<BLOB> fields.
1957 Instead, you have to use C<bind_param()>:
1959 $sth->bind_param(1, 'reg data');
1960 $sth->bind_param(2, $lots, {ora_type => ORA_CLOB});
1962 The problem is, B<SQL::Abstract> will normally just return a C<@bind> array,
1963 which loses track of which field each slot refers to. Fear not.
1965 If you specify C<bindtype> in new, you can determine how C<@bind> is returned.
1966 Currently, you can specify either C<normal> (default) or C<columns>. If you
1967 specify C<columns>, you will get an array that looks like this:
1969 my $sql = SQL::Abstract->new(bindtype => 'columns');
1970 my($stmt, @bind) = $sql->insert(...);
1973 [ 'column1', 'value1' ],
1974 [ 'column2', 'value2' ],
1975 [ 'column3', 'value3' ],
1978 You can then iterate through this manually, using DBI's C<bind_param()>.
1980 $sth->prepare($stmt);
1983 my($col, $data) = @$_;
1984 if ($col eq 'details' || $col eq 'comments') {
1985 $sth->bind_param($i, $data, {ora_type => ORA_CLOB});
1986 } elsif ($col eq 'image') {
1987 $sth->bind_param($i, $data, {ora_type => ORA_BLOB});
1989 $sth->bind_param($i, $data);
1993 $sth->execute; # execute without @bind now
1995 Now, why would you still use B<SQL::Abstract> if you have to do this crap?
1996 Basically, the advantage is still that you don't have to care which fields
1997 are or are not included. You could wrap that above C<for> loop in a simple
1998 sub called C<bind_fields()> or something and reuse it repeatedly. You still
1999 get a layer of abstraction over manual SQL specification.
2001 Note that if you set L</bindtype> to C<columns>, the C<\[ $sql, @bind ]>
2002 construct (see L</Literal SQL with placeholders and bind values (subqueries)>)
2003 will expect the bind values in this format.
2007 This is the character that a table or column name will be quoted
2008 with. By default this is an empty string, but you could set it to
2009 the character C<`>, to generate SQL like this:
2011 SELECT `a_field` FROM `a_table` WHERE `some_field` LIKE '%someval%'
2013 Alternatively, you can supply an array ref of two items, the first being the left
2014 hand quote character, and the second the right hand quote character. For
2015 example, you could supply C<['[',']']> for SQL Server 2000 compliant quotes
2016 that generates SQL like this:
2018 SELECT [a_field] FROM [a_table] WHERE [some_field] LIKE '%someval%'
2020 Quoting is useful if you have tables or columns names that are reserved
2021 words in your database's SQL dialect.
2025 This is the character that will be used to escape L</quote_char>s appearing
2026 in an identifier before it has been quoted.
2028 The parameter default in case of a single L</quote_char> character is the quote
2031 When opening-closing-style quoting is used (L</quote_char> is an arrayref)
2032 this parameter defaults to the B<closing (right)> L</quote_char>. Occurrences
2033 of the B<opening (left)> L</quote_char> within the identifier are currently left
2034 untouched. The default for opening-closing-style quotes may change in future
2035 versions, thus you are B<strongly encouraged> to specify the escape character
2040 This is the character that separates a table and column name. It is
2041 necessary to specify this when the C<quote_char> option is selected,
2042 so that tables and column names can be individually quoted like this:
2044 SELECT `table`.`one_field` FROM `table` WHERE `table`.`other_field` = 1
2046 =item injection_guard
2048 A regular expression C<qr/.../> that is applied to any C<-function> and unquoted
2049 column name specified in a query structure. This is a safety mechanism to avoid
2050 injection attacks when mishandling user input e.g.:
2052 my %condition_as_column_value_pairs = get_values_from_user();
2053 $sqla->select( ... , \%condition_as_column_value_pairs );
2055 If the expression matches an exception is thrown. Note that literal SQL
2056 supplied via C<\'...'> or C<\['...']> is B<not> checked in any way.
2058 Defaults to checking for C<;> and the C<GO> keyword (TransactSQL)
2060 =item array_datatypes
2062 When this option is true, arrayrefs in INSERT or UPDATE are
2063 interpreted as array datatypes and are passed directly
2065 When this option is false, arrayrefs are interpreted
2066 as literal SQL, just like refs to arrayrefs
2067 (but this behavior is for backwards compatibility; when writing
2068 new queries, use the "reference to arrayref" syntax
2074 Takes a reference to a list of "special operators"
2075 to extend the syntax understood by L<SQL::Abstract>.
2076 See section L</"SPECIAL OPERATORS"> for details.
2080 Takes a reference to a list of "unary operators"
2081 to extend the syntax understood by L<SQL::Abstract>.
2082 See section L</"UNARY OPERATORS"> for details.
2088 =head2 insert($table, \@values || \%fieldvals, \%options)
2090 This is the simplest function. You simply give it a table name
2091 and either an arrayref of values or hashref of field/value pairs.
2092 It returns an SQL INSERT statement and a list of bind values.
2093 See the sections on L</"Inserting and Updating Arrays"> and
2094 L</"Inserting and Updating SQL"> for information on how to insert
2095 with those data types.
2097 The optional C<\%options> hash reference may contain additional
2098 options to generate the insert SQL. Currently supported options
2105 Takes either a scalar of raw SQL fields, or an array reference of
2106 field names, and adds on an SQL C<RETURNING> statement at the end.
2107 This allows you to return data generated by the insert statement
2108 (such as row IDs) without performing another C<SELECT> statement.
2109 Note, however, this is not part of the SQL standard and may not
2110 be supported by all database engines.
2114 =head2 update($table, \%fieldvals, \%where, \%options)
2116 This takes a table, hashref of field/value pairs, and an optional
2117 hashref L<WHERE clause|/WHERE CLAUSES>. It returns an SQL UPDATE function and a list
2119 See the sections on L</"Inserting and Updating Arrays"> and
2120 L</"Inserting and Updating SQL"> for information on how to insert
2121 with those data types.
2123 The optional C<\%options> hash reference may contain additional
2124 options to generate the update SQL. Currently supported options
2131 See the C<returning> option to
2132 L<insert|/insert($table, \@values || \%fieldvals, \%options)>.
2136 =head2 select($source, $fields, $where, $order)
2138 This returns a SQL SELECT statement and associated list of bind values, as
2139 specified by the arguments:
2145 Specification of the 'FROM' part of the statement.
2146 The argument can be either a plain scalar (interpreted as a table
2147 name, will be quoted), or an arrayref (interpreted as a list
2148 of table names, joined by commas, quoted), or a scalarref
2149 (literal SQL, not quoted).
2153 Specification of the list of fields to retrieve from
2155 The argument can be either an arrayref (interpreted as a list
2156 of field names, will be joined by commas and quoted), or a
2157 plain scalar (literal SQL, not quoted).
2158 Please observe that this API is not as flexible as that of
2159 the first argument C<$source>, for backwards compatibility reasons.
2163 Optional argument to specify the WHERE part of the query.
2164 The argument is most often a hashref, but can also be
2165 an arrayref or plain scalar --
2166 see section L<WHERE clause|/"WHERE CLAUSES"> for details.
2170 Optional argument to specify the ORDER BY part of the query.
2171 The argument can be a scalar, a hashref or an arrayref
2172 -- see section L<ORDER BY clause|/"ORDER BY CLAUSES">
2178 =head2 delete($table, \%where, \%options)
2180 This takes a table name and optional hashref L<WHERE clause|/WHERE CLAUSES>.
2181 It returns an SQL DELETE statement and list of bind values.
2183 The optional C<\%options> hash reference may contain additional
2184 options to generate the delete SQL. Currently supported options
2191 See the C<returning> option to
2192 L<insert|/insert($table, \@values || \%fieldvals, \%options)>.
2196 =head2 where(\%where, $order)
2198 This is used to generate just the WHERE clause. For example,
2199 if you have an arbitrary data structure and know what the
2200 rest of your SQL is going to look like, but want an easy way
2201 to produce a WHERE clause, use this. It returns an SQL WHERE
2202 clause and list of bind values.
2205 =head2 values(\%data)
2207 This just returns the values from the hash C<%data>, in the same
2208 order that would be returned from any of the other above queries.
2209 Using this allows you to markedly speed up your queries if you
2210 are affecting lots of rows. See below under the L</"PERFORMANCE"> section.
2212 =head2 generate($any, 'number', $of, \@data, $struct, \%types)
2214 Warning: This is an experimental method and subject to change.
2216 This returns arbitrarily generated SQL. It's a really basic shortcut.
2217 It will return two different things, depending on return context:
2219 my($stmt, @bind) = $sql->generate('create table', \$table, \@fields);
2220 my $stmt_and_val = $sql->generate('create table', \$table, \@fields);
2222 These would return the following:
2224 # First calling form
2225 $stmt = "CREATE TABLE test (?, ?)";
2226 @bind = (field1, field2);
2228 # Second calling form
2229 $stmt_and_val = "CREATE TABLE test (field1, field2)";
2231 Depending on what you're trying to do, it's up to you to choose the correct
2232 format. In this example, the second form is what you would want.
2236 $sql->generate('alter session', { nls_date_format => 'MM/YY' });
2240 ALTER SESSION SET nls_date_format = 'MM/YY'
2242 You get the idea. Strings get their case twiddled, but everything
2243 else remains verbatim.
2245 =head1 EXPORTABLE FUNCTIONS
2247 =head2 is_plain_value
2249 Determines if the supplied argument is a plain value as understood by this
2254 =item * The value is C<undef>
2256 =item * The value is a non-reference
2258 =item * The value is an object with stringification overloading
2260 =item * The value is of the form C<< { -value => $anything } >>
2264 On failure returns C<undef>, on success returns a B<scalar> reference
2265 to the original supplied argument.
2271 The stringification overloading detection is rather advanced: it takes
2272 into consideration not only the presence of a C<""> overload, but if that
2273 fails also checks for enabled
2274 L<autogenerated versions of C<"">|overload/Magic Autogeneration>, based
2275 on either C<0+> or C<bool>.
2277 Unfortunately testing in the field indicates that this
2278 detection B<< may tickle a latent bug in perl versions before 5.018 >>,
2279 but only when very large numbers of stringifying objects are involved.
2280 At the time of writing ( Sep 2014 ) there is no clear explanation of
2281 the direct cause, nor is there a manageably small test case that reliably
2282 reproduces the problem.
2284 If you encounter any of the following exceptions in B<random places within
2285 your application stack> - this module may be to blame:
2287 Operation "ne": no method found,
2288 left argument in overloaded package <something>,
2289 right argument in overloaded package <something>
2293 Stub found while resolving method "???" overloading """" in package <something>
2295 If you fall victim to the above - please attempt to reduce the problem
2296 to something that could be sent to the L<SQL::Abstract developers
2297 |DBIx::Class/GETTING HELP/SUPPORT>
2298 (either publicly or privately). As a workaround in the meantime you can
2299 set C<$ENV{SQLA_ISVALUE_IGNORE_AUTOGENERATED_STRINGIFICATION}> to a true
2300 value, which will most likely eliminate your problem (at the expense of
2301 not being able to properly detect exotic forms of stringification).
2303 This notice and environment variable will be removed in a future version,
2304 as soon as the underlying problem is found and a reliable workaround is
2309 =head2 is_literal_value
2311 Determines if the supplied argument is a literal value as understood by this
2316 =item * C<\$sql_string>
2318 =item * C<\[ $sql_string, @bind_values ]>
2322 On failure returns C<undef>, on success returns an B<array> reference
2323 containing the unpacked version of the supplied literal SQL and bind values.
2325 =head1 WHERE CLAUSES
2329 This module uses a variation on the idea from L<DBIx::Abstract>. It
2330 is B<NOT>, repeat I<not> 100% compatible. B<The main logic of this
2331 module is that things in arrays are OR'ed, and things in hashes
2334 The easiest way to explain is to show lots of examples. After
2335 each C<%where> hash shown, it is assumed you used:
2337 my($stmt, @bind) = $sql->where(\%where);
2339 However, note that the C<%where> hash can be used directly in any
2340 of the other functions as well, as described above.
2342 =head2 Key-value pairs
2344 So, let's get started. To begin, a simple hash:
2348 status => 'completed'
2351 Is converted to SQL C<key = val> statements:
2353 $stmt = "WHERE user = ? AND status = ?";
2354 @bind = ('nwiger', 'completed');
2356 One common thing I end up doing is having a list of values that
2357 a field can be in. To do this, simply specify a list inside of
2362 status => ['assigned', 'in-progress', 'pending'];
2365 This simple code will create the following:
2367 $stmt = "WHERE user = ? AND ( status = ? OR status = ? OR status = ? )";
2368 @bind = ('nwiger', 'assigned', 'in-progress', 'pending');
2370 A field associated to an empty arrayref will be considered a
2371 logical false and will generate 0=1.
2373 =head2 Tests for NULL values
2375 If the value part is C<undef> then this is converted to SQL <IS NULL>
2384 $stmt = "WHERE user = ? AND status IS NULL";
2387 To test if a column IS NOT NULL:
2391 status => { '!=', undef },
2394 =head2 Specific comparison operators
2396 If you want to specify a different type of operator for your comparison,
2397 you can use a hashref for a given column:
2401 status => { '!=', 'completed' }
2404 Which would generate:
2406 $stmt = "WHERE user = ? AND status != ?";
2407 @bind = ('nwiger', 'completed');
2409 To test against multiple values, just enclose the values in an arrayref:
2411 status => { '=', ['assigned', 'in-progress', 'pending'] };
2413 Which would give you:
2415 "WHERE status = ? OR status = ? OR status = ?"
2418 The hashref can also contain multiple pairs, in which case it is expanded
2419 into an C<AND> of its elements:
2423 status => { '!=', 'completed', -not_like => 'pending%' }
2426 # Or more dynamically, like from a form
2427 $where{user} = 'nwiger';
2428 $where{status}{'!='} = 'completed';
2429 $where{status}{'-not_like'} = 'pending%';
2431 # Both generate this
2432 $stmt = "WHERE user = ? AND status != ? AND status NOT LIKE ?";
2433 @bind = ('nwiger', 'completed', 'pending%');
2436 To get an OR instead, you can combine it with the arrayref idea:
2440 priority => [ { '=', 2 }, { '>', 5 } ]
2443 Which would generate:
2445 $stmt = "WHERE ( priority = ? OR priority > ? ) AND user = ?";
2446 @bind = ('2', '5', 'nwiger');
2448 If you want to include literal SQL (with or without bind values), just use a
2449 scalar reference or reference to an arrayref as the value:
2452 date_entered => { '>' => \["to_date(?, 'MM/DD/YYYY')", "11/26/2008"] },
2453 date_expires => { '<' => \"now()" }
2456 Which would generate:
2458 $stmt = "WHERE date_entered > to_date(?, 'MM/DD/YYYY') AND date_expires < now()";
2459 @bind = ('11/26/2008');
2462 =head2 Logic and nesting operators
2464 In the example above,
2465 there is a subtle trap if you want to say something like
2466 this (notice the C<AND>):
2468 WHERE priority != ? AND priority != ?
2470 Because, in Perl you I<can't> do this:
2472 priority => { '!=' => 2, '!=' => 1 }
2474 As the second C<!=> key will obliterate the first. The solution
2475 is to use the special C<-modifier> form inside an arrayref:
2477 priority => [ -and => {'!=', 2},
2481 Normally, these would be joined by C<OR>, but the modifier tells it
2482 to use C<AND> instead. (Hint: You can use this in conjunction with the
2483 C<logic> option to C<new()> in order to change the way your queries
2484 work by default.) B<Important:> Note that the C<-modifier> goes
2485 B<INSIDE> the arrayref, as an extra first element. This will
2486 B<NOT> do what you think it might:
2488 priority => -and => [{'!=', 2}, {'!=', 1}] # WRONG!
2490 Here is a quick list of equivalencies, since there is some overlap:
2493 status => {'!=', 'completed', 'not like', 'pending%' }
2494 status => [ -and => {'!=', 'completed'}, {'not like', 'pending%'}]
2497 status => {'=', ['assigned', 'in-progress']}
2498 status => [ -or => {'=', 'assigned'}, {'=', 'in-progress'}]
2499 status => [ {'=', 'assigned'}, {'=', 'in-progress'} ]
2503 =head2 Special operators: IN, BETWEEN, etc.
2505 You can also use the hashref format to compare a list of fields using the
2506 C<IN> comparison operator, by specifying the list as an arrayref:
2509 status => 'completed',
2510 reportid => { -in => [567, 2335, 2] }
2513 Which would generate:
2515 $stmt = "WHERE status = ? AND reportid IN (?,?,?)";
2516 @bind = ('completed', '567', '2335', '2');
2518 The reverse operator C<-not_in> generates SQL C<NOT IN> and is used in
2521 If the argument to C<-in> is an empty array, 'sqlfalse' is generated
2522 (by default: C<1=0>). Similarly, C<< -not_in => [] >> generates
2523 'sqltrue' (by default: C<1=1>).
2525 In addition to the array you can supply a chunk of literal sql or
2526 literal sql with bind:
2529 customer => { -in => \[
2530 'SELECT cust_id FROM cust WHERE balance > ?',
2533 status => { -in => \'SELECT status_codes FROM states' },
2539 customer IN ( SELECT cust_id FROM cust WHERE balance > ? )
2540 AND status IN ( SELECT status_codes FROM states )
2544 Finally, if the argument to C<-in> is not a reference, it will be
2545 treated as a single-element array.
2547 Another pair of operators is C<-between> and C<-not_between>,
2548 used with an arrayref of two values:
2552 completion_date => {
2553 -not_between => ['2002-10-01', '2003-02-06']
2559 WHERE user = ? AND completion_date NOT BETWEEN ( ? AND ? )
2561 Just like with C<-in> all plausible combinations of literal SQL
2565 start0 => { -between => [ 1, 2 ] },
2566 start1 => { -between => \["? AND ?", 1, 2] },
2567 start2 => { -between => \"lower(x) AND upper(y)" },
2568 start3 => { -between => [
2570 \["upper(?)", 'stuff' ],
2577 ( start0 BETWEEN ? AND ? )
2578 AND ( start1 BETWEEN ? AND ? )
2579 AND ( start2 BETWEEN lower(x) AND upper(y) )
2580 AND ( start3 BETWEEN lower(x) AND upper(?) )
2582 @bind = (1, 2, 1, 2, 'stuff');
2585 These are the two builtin "special operators"; but the
2586 list can be expanded: see section L</"SPECIAL OPERATORS"> below.
2588 =head2 Unary operators: bool
2590 If you wish to test against boolean columns or functions within your
2591 database you can use the C<-bool> and C<-not_bool> operators. For
2592 example to test the column C<is_user> being true and the column
2593 C<is_enabled> being false you would use:-
2597 -not_bool => 'is_enabled',
2602 WHERE is_user AND NOT is_enabled
2604 If a more complex combination is required, testing more conditions,
2605 then you should use the and/or operators:-
2610 -not_bool => { two=> { -rlike => 'bar' } },
2611 -not_bool => { three => [ { '=', 2 }, { '>', 5 } ] },
2622 (NOT ( three = ? OR three > ? ))
2625 =head2 Nested conditions, -and/-or prefixes
2627 So far, we've seen how multiple conditions are joined with a top-level
2628 C<AND>. We can change this by putting the different conditions we want in
2629 hashes and then putting those hashes in an array. For example:
2634 status => { -like => ['pending%', 'dispatched'] },
2638 status => 'unassigned',
2642 This data structure would create the following:
2644 $stmt = "WHERE ( user = ? AND ( status LIKE ? OR status LIKE ? ) )
2645 OR ( user = ? AND status = ? ) )";
2646 @bind = ('nwiger', 'pending', 'dispatched', 'robot', 'unassigned');
2649 Clauses in hashrefs or arrayrefs can be prefixed with an C<-and> or C<-or>
2650 to change the logic inside:
2656 -and => [ workhrs => {'>', 20}, geo => 'ASIA' ],
2657 -or => { workhrs => {'<', 50}, geo => 'EURO' },
2664 $stmt = "WHERE ( user = ?
2665 AND ( ( workhrs > ? AND geo = ? )
2666 OR ( workhrs < ? OR geo = ? ) ) )";
2667 @bind = ('nwiger', '20', 'ASIA', '50', 'EURO');
2669 =head3 Algebraic inconsistency, for historical reasons
2671 C<Important note>: when connecting several conditions, the C<-and->|C<-or>
2672 operator goes C<outside> of the nested structure; whereas when connecting
2673 several constraints on one column, the C<-and> operator goes
2674 C<inside> the arrayref. Here is an example combining both features:
2677 -and => [a => 1, b => 2],
2678 -or => [c => 3, d => 4],
2679 e => [-and => {-like => 'foo%'}, {-like => '%bar'} ]
2684 WHERE ( ( ( a = ? AND b = ? )
2685 OR ( c = ? OR d = ? )
2686 OR ( e LIKE ? AND e LIKE ? ) ) )
2688 This difference in syntax is unfortunate but must be preserved for
2689 historical reasons. So be careful: the two examples below would
2690 seem algebraically equivalent, but they are not
2693 { -like => 'foo%' },
2694 { -like => '%bar' },
2696 # yields: WHERE ( ( col LIKE ? AND col LIKE ? ) )
2699 { col => { -like => 'foo%' } },
2700 { col => { -like => '%bar' } },
2702 # yields: WHERE ( ( col LIKE ? OR col LIKE ? ) )
2705 =head2 Literal SQL and value type operators
2707 The basic premise of SQL::Abstract is that in WHERE specifications the "left
2708 side" is a column name and the "right side" is a value (normally rendered as
2709 a placeholder). This holds true for both hashrefs and arrayref pairs as you
2710 see in the L</WHERE CLAUSES> examples above. Sometimes it is necessary to
2711 alter this behavior. There are several ways of doing so.
2715 This is a virtual operator that signals the string to its right side is an
2716 identifier (a column name) and not a value. For example to compare two
2717 columns you would write:
2720 priority => { '<', 2 },
2721 requestor => { -ident => 'submitter' },
2726 $stmt = "WHERE priority < ? AND requestor = submitter";
2729 If you are maintaining legacy code you may see a different construct as
2730 described in L</Deprecated usage of Literal SQL>, please use C<-ident> in new
2735 This is a virtual operator that signals that the construct to its right side
2736 is a value to be passed to DBI. This is for example necessary when you want
2737 to write a where clause against an array (for RDBMS that support such
2738 datatypes). For example:
2741 array => { -value => [1, 2, 3] }
2746 $stmt = 'WHERE array = ?';
2747 @bind = ([1, 2, 3]);
2749 Note that if you were to simply say:
2755 the result would probably not be what you wanted:
2757 $stmt = 'WHERE array = ? OR array = ? OR array = ?';
2762 Finally, sometimes only literal SQL will do. To include a random snippet
2763 of SQL verbatim, you specify it as a scalar reference. Consider this only
2764 as a last resort. Usually there is a better way. For example:
2767 priority => { '<', 2 },
2768 requestor => { -in => \'(SELECT name FROM hitmen)' },
2773 $stmt = "WHERE priority < ? AND requestor IN (SELECT name FROM hitmen)"
2776 Note that in this example, you only get one bind parameter back, since
2777 the verbatim SQL is passed as part of the statement.
2781 Never use untrusted input as a literal SQL argument - this is a massive
2782 security risk (there is no way to check literal snippets for SQL
2783 injections and other nastyness). If you need to deal with untrusted input
2784 use literal SQL with placeholders as described next.
2786 =head3 Literal SQL with placeholders and bind values (subqueries)
2788 If the literal SQL to be inserted has placeholders and bind values,
2789 use a reference to an arrayref (yes this is a double reference --
2790 not so common, but perfectly legal Perl). For example, to find a date
2791 in Postgres you can use something like this:
2794 date_column => \[ "= date '2008-09-30' - ?::integer", 10 ]
2799 $stmt = "WHERE ( date_column = date '2008-09-30' - ?::integer )"
2802 Note that you must pass the bind values in the same format as they are returned
2803 by L<where|/where(\%where, $order)>. This means that if you set L</bindtype>
2804 to C<columns>, you must provide the bind values in the
2805 C<< [ column_meta => value ] >> format, where C<column_meta> is an opaque
2806 scalar value; most commonly the column name, but you can use any scalar value
2807 (including references and blessed references), L<SQL::Abstract> will simply
2808 pass it through intact. So if C<bindtype> is set to C<columns> the above
2809 example will look like:
2812 date_column => \[ "= date '2008-09-30' - ?::integer", [ {} => 10 ] ]
2815 Literal SQL is especially useful for nesting parenthesized clauses in the
2816 main SQL query. Here is a first example:
2818 my ($sub_stmt, @sub_bind) = ("SELECT c1 FROM t1 WHERE c2 < ? AND c3 LIKE ?",
2822 bar => \["IN ($sub_stmt)" => @sub_bind],
2827 $stmt = "WHERE (foo = ? AND bar IN (SELECT c1 FROM t1
2828 WHERE c2 < ? AND c3 LIKE ?))";
2829 @bind = (1234, 100, "foo%");
2831 Other subquery operators, like for example C<"E<gt> ALL"> or C<"NOT IN">,
2832 are expressed in the same way. Of course the C<$sub_stmt> and
2833 its associated bind values can be generated through a former call
2836 my ($sub_stmt, @sub_bind)
2837 = $sql->select("t1", "c1", {c2 => {"<" => 100},
2838 c3 => {-like => "foo%"}});
2841 bar => \["> ALL ($sub_stmt)" => @sub_bind],
2844 In the examples above, the subquery was used as an operator on a column;
2845 but the same principle also applies for a clause within the main C<%where>
2846 hash, like an EXISTS subquery:
2848 my ($sub_stmt, @sub_bind)
2849 = $sql->select("t1", "*", {c1 => 1, c2 => \"> t0.c0"});
2850 my %where = ( -and => [
2852 \["EXISTS ($sub_stmt)" => @sub_bind],
2857 $stmt = "WHERE (foo = ? AND EXISTS (SELECT * FROM t1
2858 WHERE c1 = ? AND c2 > t0.c0))";
2862 Observe that the condition on C<c2> in the subquery refers to
2863 column C<t0.c0> of the main query: this is I<not> a bind
2864 value, so we have to express it through a scalar ref.
2865 Writing C<< c2 => {">" => "t0.c0"} >> would have generated
2866 C<< c2 > ? >> with bind value C<"t0.c0"> ... not exactly
2867 what we wanted here.
2869 Finally, here is an example where a subquery is used
2870 for expressing unary negation:
2872 my ($sub_stmt, @sub_bind)
2873 = $sql->where({age => [{"<" => 10}, {">" => 20}]});
2874 $sub_stmt =~ s/^ where //i; # don't want "WHERE" in the subclause
2876 lname => {like => '%son%'},
2877 \["NOT ($sub_stmt)" => @sub_bind],
2882 $stmt = "lname LIKE ? AND NOT ( age < ? OR age > ? )"
2883 @bind = ('%son%', 10, 20)
2885 =head3 Deprecated usage of Literal SQL
2887 Below are some examples of archaic use of literal SQL. It is shown only as
2888 reference for those who deal with legacy code. Each example has a much
2889 better, cleaner and safer alternative that users should opt for in new code.
2895 my %where = ( requestor => \'IS NOT NULL' )
2897 $stmt = "WHERE requestor IS NOT NULL"
2899 This used to be the way of generating NULL comparisons, before the handling
2900 of C<undef> got formalized. For new code please use the superior syntax as
2901 described in L</Tests for NULL values>.
2905 my %where = ( requestor => \'= submitter' )
2907 $stmt = "WHERE requestor = submitter"
2909 This used to be the only way to compare columns. Use the superior L</-ident>
2910 method for all new code. For example an identifier declared in such a way
2911 will be properly quoted if L</quote_char> is properly set, while the legacy
2912 form will remain as supplied.
2916 my %where = ( is_ready => \"", completed => { '>', '2012-12-21' } )
2918 $stmt = "WHERE completed > ? AND is_ready"
2919 @bind = ('2012-12-21')
2921 Using an empty string literal used to be the only way to express a boolean.
2922 For all new code please use the much more readable
2923 L<-bool|/Unary operators: bool> operator.
2929 These pages could go on for a while, since the nesting of the data
2930 structures this module can handle are pretty much unlimited (the
2931 module implements the C<WHERE> expansion as a recursive function
2932 internally). Your best bet is to "play around" with the module a
2933 little to see how the data structures behave, and choose the best
2934 format for your data based on that.
2936 And of course, all the values above will probably be replaced with
2937 variables gotten from forms or the command line. After all, if you
2938 knew everything ahead of time, you wouldn't have to worry about
2939 dynamically-generating SQL and could just hardwire it into your
2942 =head1 ORDER BY CLAUSES
2944 Some functions take an order by clause. This can either be a scalar (just a
2945 column name), a hashref of C<< { -desc => 'col' } >> or C<< { -asc => 'col' }
2946 >>, a scalarref, an arrayref-ref, or an arrayref of any of the previous
2949 Given | Will Generate
2950 ---------------------------------------------------------------
2952 'colA' | ORDER BY colA
2954 [qw/colA colB/] | ORDER BY colA, colB
2956 {-asc => 'colA'} | ORDER BY colA ASC
2958 {-desc => 'colB'} | ORDER BY colB DESC
2960 ['colA', {-asc => 'colB'}] | ORDER BY colA, colB ASC
2962 { -asc => [qw/colA colB/] } | ORDER BY colA ASC, colB ASC
2964 \'colA DESC' | ORDER BY colA DESC
2966 \[ 'FUNC(colA, ?)', $x ] | ORDER BY FUNC(colA, ?)
2967 | /* ...with $x bound to ? */
2970 { -asc => 'colA' }, | colA ASC,
2971 { -desc => [qw/colB/] }, | colB DESC,
2972 { -asc => [qw/colC colD/] },| colC ASC, colD ASC,
2973 \'colE DESC', | colE DESC,
2974 \[ 'FUNC(colF, ?)', $x ], | FUNC(colF, ?)
2975 ] | /* ...with $x bound to ? */
2976 ===============================================================
2980 =head1 SPECIAL OPERATORS
2982 my $sqlmaker = SQL::Abstract->new(special_ops => [
2986 my ($self, $field, $op, $arg) = @_;
2992 handler => 'method_name',
2996 A "special operator" is a SQL syntactic clause that can be
2997 applied to a field, instead of a usual binary operator.
3000 WHERE field IN (?, ?, ?)
3001 WHERE field BETWEEN ? AND ?
3002 WHERE MATCH(field) AGAINST (?, ?)
3004 Special operators IN and BETWEEN are fairly standard and therefore
3005 are builtin within C<SQL::Abstract> (as the overridable methods
3006 C<_where_field_IN> and C<_where_field_BETWEEN>). For other operators,
3007 like the MATCH .. AGAINST example above which is specific to MySQL,
3008 you can write your own operator handlers - supply a C<special_ops>
3009 argument to the C<new> method. That argument takes an arrayref of
3010 operator definitions; each operator definition is a hashref with two
3017 the regular expression to match the operator
3021 Either a coderef or a plain scalar method name. In both cases
3022 the expected return is C<< ($sql, @bind) >>.
3024 When supplied with a method name, it is simply called on the
3025 L<SQL::Abstract> object as:
3027 $self->$method_name($field, $op, $arg)
3031 $field is the LHS of the operator
3032 $op is the part that matched the handler regex
3035 When supplied with a coderef, it is called as:
3037 $coderef->($self, $field, $op, $arg)
3042 For example, here is an implementation
3043 of the MATCH .. AGAINST syntax for MySQL
3045 my $sqlmaker = SQL::Abstract->new(special_ops => [
3047 # special op for MySql MATCH (field) AGAINST(word1, word2, ...)
3048 {regex => qr/^match$/i,
3050 my ($self, $field, $op, $arg) = @_;
3051 $arg = [$arg] if not ref $arg;
3052 my $label = $self->_quote($field);
3053 my ($placeholder) = $self->_convert('?');
3054 my $placeholders = join ", ", (($placeholder) x @$arg);
3055 my $sql = $self->_sqlcase('match') . " ($label) "
3056 . $self->_sqlcase('against') . " ($placeholders) ";
3057 my @bind = $self->_bindtype($field, @$arg);
3058 return ($sql, @bind);
3065 =head1 UNARY OPERATORS
3067 my $sqlmaker = SQL::Abstract->new(unary_ops => [
3071 my ($self, $op, $arg) = @_;
3077 handler => 'method_name',
3081 A "unary operator" is a SQL syntactic clause that can be
3082 applied to a field - the operator goes before the field
3084 You can write your own operator handlers - supply a C<unary_ops>
3085 argument to the C<new> method. That argument takes an arrayref of
3086 operator definitions; each operator definition is a hashref with two
3093 the regular expression to match the operator
3097 Either a coderef or a plain scalar method name. In both cases
3098 the expected return is C<< $sql >>.
3100 When supplied with a method name, it is simply called on the
3101 L<SQL::Abstract> object as:
3103 $self->$method_name($op, $arg)
3107 $op is the part that matched the handler regex
3108 $arg is the RHS or argument of the operator
3110 When supplied with a coderef, it is called as:
3112 $coderef->($self, $op, $arg)
3120 Thanks to some benchmarking by Mark Stosberg, it turns out that
3121 this module is many orders of magnitude faster than using C<DBIx::Abstract>.
3122 I must admit this wasn't an intentional design issue, but it's a
3123 byproduct of the fact that you get to control your C<DBI> handles
3126 To maximize performance, use a code snippet like the following:
3128 # prepare a statement handle using the first row
3129 # and then reuse it for the rest of the rows
3131 for my $href (@array_of_hashrefs) {
3132 $stmt ||= $sql->insert('table', $href);
3133 $sth ||= $dbh->prepare($stmt);
3134 $sth->execute($sql->values($href));
3137 The reason this works is because the keys in your C<$href> are sorted
3138 internally by B<SQL::Abstract>. Thus, as long as your data retains
3139 the same structure, you only have to generate the SQL the first time
3140 around. On subsequent queries, simply use the C<values> function provided
3141 by this module to return your values in the correct order.
3143 However this depends on the values having the same type - if, for
3144 example, the values of a where clause may either have values
3145 (resulting in sql of the form C<column = ?> with a single bind
3146 value), or alternatively the values might be C<undef> (resulting in
3147 sql of the form C<column IS NULL> with no bind value) then the
3148 caching technique suggested will not work.
3152 If you use my C<CGI::FormBuilder> module at all, you'll hopefully
3153 really like this part (I do, at least). Building up a complex query
3154 can be as simple as the following:
3161 use CGI::FormBuilder;
3164 my $form = CGI::FormBuilder->new(...);
3165 my $sql = SQL::Abstract->new;
3167 if ($form->submitted) {
3168 my $field = $form->field;
3169 my $id = delete $field->{id};
3170 my($stmt, @bind) = $sql->update('table', $field, {id => $id});
3173 Of course, you would still have to connect using C<DBI> to run the
3174 query, but the point is that if you make your form look like your
3175 table, the actual query script can be extremely simplistic.
3177 If you're B<REALLY> lazy (I am), check out C<HTML::QuickTable> for
3178 a fast interface to returning and formatting data. I frequently
3179 use these three modules together to write complex database query
3180 apps in under 50 lines.
3182 =head1 HOW TO CONTRIBUTE
3184 Contributions are always welcome, in all usable forms (we especially
3185 welcome documentation improvements). The delivery methods include git-
3186 or unified-diff formatted patches, GitHub pull requests, or plain bug
3187 reports either via RT or the Mailing list. Contributors are generally
3188 granted full access to the official repository after their first several
3189 patches pass successful review.
3191 This project is maintained in a git repository. The code and related tools are
3192 accessible at the following locations:
3196 =item * Official repo: L<git://git.shadowcat.co.uk/dbsrgits/SQL-Abstract.git>
3198 =item * Official gitweb: L<http://git.shadowcat.co.uk/gitweb/gitweb.cgi?p=dbsrgits/SQL-Abstract.git>
3200 =item * GitHub mirror: L<https://github.com/dbsrgits/sql-abstract>
3202 =item * Authorized committers: L<ssh://dbsrgits@git.shadowcat.co.uk/SQL-Abstract.git>
3208 Version 1.50 was a major internal refactoring of C<SQL::Abstract>.
3209 Great care has been taken to preserve the I<published> behavior
3210 documented in previous versions in the 1.* family; however,
3211 some features that were previously undocumented, or behaved
3212 differently from the documentation, had to be changed in order
3213 to clarify the semantics. Hence, client code that was relying
3214 on some dark areas of C<SQL::Abstract> v1.*
3215 B<might behave differently> in v1.50.
3217 The main changes are:
3223 support for literal SQL through the C<< \ [ $sql, @bind ] >> syntax.
3227 support for the { operator => \"..." } construct (to embed literal SQL)
3231 support for the { operator => \["...", @bind] } construct (to embed literal SQL with bind values)
3235 optional support for L<array datatypes|/"Inserting and Updating Arrays">
3239 defensive programming: check arguments
3243 fixed bug with global logic, which was previously implemented
3244 through global variables yielding side-effects. Prior versions would
3245 interpret C<< [ {cond1, cond2}, [cond3, cond4] ] >>
3246 as C<< "(cond1 AND cond2) OR (cond3 AND cond4)" >>.
3247 Now this is interpreted
3248 as C<< "(cond1 AND cond2) OR (cond3 OR cond4)" >>.
3253 fixed semantics of _bindtype on array args
3257 dropped the C<_anoncopy> of the %where tree. No longer necessary,
3258 we just avoid shifting arrays within that tree.
3262 dropped the C<_modlogic> function
3266 =head1 ACKNOWLEDGEMENTS
3268 There are a number of individuals that have really helped out with
3269 this module. Unfortunately, most of them submitted bugs via CPAN
3270 so I have no idea who they are! But the people I do know are:
3272 Ash Berlin (order_by hash term support)
3273 Matt Trout (DBIx::Class support)
3274 Mark Stosberg (benchmarking)
3275 Chas Owens (initial "IN" operator support)
3276 Philip Collins (per-field SQL functions)
3277 Eric Kolve (hashref "AND" support)
3278 Mike Fragassi (enhancements to "BETWEEN" and "LIKE")
3279 Dan Kubb (support for "quote_char" and "name_sep")
3280 Guillermo Roditi (patch to cleanup "IN" and "BETWEEN", fix and tests for _order_by)
3281 Laurent Dami (internal refactoring, extensible list of special operators, literal SQL)
3282 Norbert Buchmuller (support for literal SQL in hashpair, misc. fixes & tests)
3283 Peter Rabbitson (rewrite of SQLA::Test, misc. fixes & tests)
3284 Oliver Charles (support for "RETURNING" after "INSERT")
3290 L<DBIx::Class>, L<DBIx::Abstract>, L<CGI::FormBuilder>, L<HTML::QuickTable>.
3294 Copyright (c) 2001-2007 Nathan Wiger <nwiger@cpan.org>. All Rights Reserved.
3296 This module is actively maintained by Matt Trout <mst@shadowcatsystems.co.uk>
3298 For support, your best bet is to try the C<DBIx::Class> users mailing list.
3299 While not an official support venue, C<DBIx::Class> makes heavy use of
3300 C<SQL::Abstract>, and as such list members there are very familiar with
3301 how to create queries.
3305 This module is free software; you may copy this under the same
3306 terms as perl itself (either the GNU General Public License or
3307 the Artistic License)