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 = '1.87';
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) : $sql;
539 my ($self, $expr, $logic) = @_;
540 if (ref($expr) eq 'HASH') {
541 if (keys %$expr > 1) {
543 return +{ "-${logic}" => [
544 map $self->_expand_expr_hashpair($_ => $expr->{$_}, $logic),
548 return $self->_expand_expr_hashpair(%$expr, $logic);
553 sub _expand_expr_hashpair {
554 my ($self, $k, $v, $logic) = @_;
557 return +{ $k => { $self->{cmp} => $v } };
561 return $self->_expand_expr($v);
567 my ($self, $where, $logic) = @_;
569 my $where_exp = $self->_expand_expr($where, $logic);
571 # dispatch on appropriate method according to refkind of $where
572 my $method = $self->_METHOD_FOR_refkind("_where", $where_exp);
574 my ($sql, @bind) = $self->$method($where_exp, $logic);
576 # DBIx::Class used to call _recurse_where in scalar context
577 # something else might too...
579 return ($sql, @bind);
582 belch "Calling _recurse_where in scalar context is deprecated and will go away before 2.0";
589 #======================================================================
590 # WHERE: top-level ARRAYREF
591 #======================================================================
594 sub _where_ARRAYREF {
595 my ($self, $where, $logic) = @_;
597 $logic = uc($logic || $self->{logic});
598 $logic eq 'AND' or $logic eq 'OR' or puke "unknown logic: $logic";
600 my @clauses = @$where;
602 my (@sql_clauses, @all_bind);
603 # need to use while() so can shift() for pairs
605 my $el = shift @clauses;
607 $el = undef if (defined $el and ! length $el);
609 # switch according to kind of $el and get corresponding ($sql, @bind)
610 my ($sql, @bind) = $self->_SWITCH_refkind($el, {
612 # skip empty elements, otherwise get invalid trailing AND stuff
613 ARRAYREF => sub {$self->_recurse_where($el) if @$el},
617 $self->_assert_bindval_matches_bindtype(@b);
621 HASHREF => sub {$self->_recurse_where($el, 'and') if %$el},
623 SCALARREF => sub { ($$el); },
626 # top-level arrayref with scalars, recurse in pairs
627 $self->_recurse_where({$el => shift(@clauses)})
630 UNDEF => sub {puke "Supplying an empty left hand side argument is not supported in array-pairs" },
634 push @sql_clauses, $sql;
635 push @all_bind, @bind;
639 return $self->_join_sql_clauses($logic, \@sql_clauses, \@all_bind);
642 #======================================================================
643 # WHERE: top-level ARRAYREFREF
644 #======================================================================
646 sub _where_ARRAYREFREF {
647 my ($self, $where) = @_;
648 my ($sql, @bind) = @$$where;
649 $self->_assert_bindval_matches_bindtype(@bind);
650 return ($sql, @bind);
653 #======================================================================
654 # WHERE: top-level HASHREF
655 #======================================================================
658 my ($self, $where) = @_;
659 my (@sql_clauses, @all_bind);
661 for my $k (sort keys %$where) {
662 my $v = $where->{$k};
664 # ($k => $v) is either a special unary op or a regular hashpair
665 my ($sql, @bind) = do {
667 # put the operator in canonical form
669 $op = substr $op, 1; # remove initial dash
670 $op =~ s/^\s+|\s+$//g;# remove leading/trailing space
671 $op =~ s/\s+/ /g; # compress whitespace
673 # so that -not_foo works correctly
674 $op =~ s/^not_/NOT /i;
676 $self->_debug("Unary OP(-$op) within hashref, recursing...");
677 my ($s, @b) = $self->_where_unary_op($op, $v);
679 # top level vs nested
680 # we assume that handled unary ops will take care of their ()s
682 List::Util::first {$op =~ $_->{regex}} @{$self->{unary_ops}}
684 ( defined $self->{_nested_func_lhs} and $self->{_nested_func_lhs} eq $k )
690 if (is_literal_value ($v) ) {
691 belch 'Hash-pairs consisting of an empty string with a literal are deprecated, and will be removed in 2.0: use -and => [ $literal ] instead';
694 puke "Supplying an empty left hand side argument is not supported in hash-pairs";
698 my $method = $self->_METHOD_FOR_refkind("_where_hashpair", $v);
699 $self->$method($k, $v);
703 push @sql_clauses, $sql;
704 push @all_bind, @bind;
707 return $self->_join_sql_clauses('and', \@sql_clauses, \@all_bind);
710 sub _where_unary_op {
711 my ($self, $op, $rhs) = @_;
713 # top level special ops are illegal in general
714 # this includes the -ident/-value ops (dual purpose unary and special)
715 puke "Illegal use of top-level '-$op'"
716 if ! defined $self->{_nested_func_lhs} and List::Util::first { $op =~ $_->{regex} } @{$self->{special_ops}};
718 if (my $op_entry = List::Util::first { $op =~ $_->{regex} } @{$self->{unary_ops}}) {
719 my $handler = $op_entry->{handler};
721 if (not ref $handler) {
722 if ($op =~ s/ [_\s]? \d+ $//x ) {
723 belch 'Use of [and|or|nest]_N modifiers is deprecated and will be removed in SQLA v2.0. '
724 . "You probably wanted ...-and => [ -$op => COND1, -$op => COND2 ... ]";
726 return $self->$handler($op, $rhs);
728 elsif (ref $handler eq 'CODE') {
729 return $handler->($self, $op, $rhs);
732 puke "Illegal handler for operator $op - expecting a method name or a coderef";
736 $self->_debug("Generic unary OP: $op - recursing as function");
738 $self->_assert_pass_injection_guard($op);
740 my ($sql, @bind) = $self->_SWITCH_refkind($rhs, {
742 puke "Illegal use of top-level '-$op'"
743 unless defined $self->{_nested_func_lhs};
746 $self->_convert('?'),
747 $self->_bindtype($self->{_nested_func_lhs}, $rhs)
751 $self->_recurse_where($rhs)
755 $sql = sprintf('%s %s',
756 $self->_sqlcase($op),
760 return ($sql, @bind);
763 sub _where_op_ANDOR {
764 my ($self, $op, $v) = @_;
766 $self->_SWITCH_refkind($v, {
768 return $self->_where_ARRAYREF($v, $op);
772 return ($op =~ /^or/i)
773 ? $self->_where_ARRAYREF([ map { $_ => $v->{$_} } (sort keys %$v) ], $op)
774 : $self->_where_HASHREF($v);
778 puke "-$op => \\\$scalar makes little sense, use " .
780 ? '[ \$scalar, \%rest_of_conditions ] instead'
781 : '-and => [ \$scalar, \%rest_of_conditions ] instead'
786 puke "-$op => \\[...] makes little sense, use " .
788 ? '[ \[...], \%rest_of_conditions ] instead'
789 : '-and => [ \[...], \%rest_of_conditions ] instead'
793 SCALAR => sub { # permissively interpreted as SQL
794 puke "-$op => \$value makes little sense, use -bool => \$value instead";
798 puke "-$op => undef not supported";
804 my ($self, $op, $v) = @_;
806 $self->_SWITCH_refkind($v, {
808 SCALAR => sub { # permissively interpreted as SQL
809 belch "literal SQL should be -nest => \\'scalar' "
810 . "instead of -nest => 'scalar' ";
815 puke "-$op => undef not supported";
819 $self->_recurse_where($v);
827 my ($self, $op, $v) = @_;
829 my ($s, @b) = $self->_SWITCH_refkind($v, {
830 SCALAR => sub { # interpreted as SQL column
831 $self->_convert($self->_quote($v));
835 puke "-$op => undef not supported";
839 $self->_recurse_where($v);
843 $s = "(NOT $s)" if $op =~ /^not/i;
848 sub _where_op_IDENT {
850 my ($op, $rhs) = splice @_, -2;
851 if (! defined $rhs or length ref $rhs) {
852 puke "-$op requires a single plain scalar argument (a quotable identifier)";
855 # in case we are called as a top level special op (no '=')
858 $_ = $self->_convert($self->_quote($_)) for ($lhs, $rhs);
866 sub _where_op_VALUE {
868 my ($op, $rhs) = splice @_, -2;
870 # in case we are called as a top level special op (no '=')
874 if (! defined $rhs) {
876 ? $self->_where_hashpair_HASHREF($lhs, { -is => undef })
883 (defined $lhs ? $lhs : $self->{_nested_func_lhs}),
890 $self->_convert($self->_quote($lhs)) . ' = ' . $self->_convert('?'),
894 $self->_convert('?'),
900 sub _where_hashpair_ARRAYREF {
901 my ($self, $k, $v) = @_;
904 my @v = @$v; # need copy because of shift below
905 $self->_debug("ARRAY($k) means distribute over elements");
907 # put apart first element if it is an operator (-and, -or)
909 (defined $v[0] && $v[0] =~ /^ - (?: AND|OR ) $/ix)
913 my @distributed = map { {$k => $_} } @v;
916 $self->_debug("OP($op) reinjected into the distributed array");
917 unshift @distributed, $op;
920 my $logic = $op ? substr($op, 1) : '';
922 return $self->_recurse_where(\@distributed, $logic);
925 $self->_debug("empty ARRAY($k) means 0=1");
926 return ($self->{sqlfalse});
930 sub _where_hashpair_HASHREF {
931 my ($self, $k, $v, $logic) = @_;
934 local $self->{_nested_func_lhs} = defined $self->{_nested_func_lhs}
935 ? $self->{_nested_func_lhs}
939 my ($all_sql, @all_bind);
941 for my $orig_op (sort keys %$v) {
942 my $val = $v->{$orig_op};
944 # put the operator in canonical form
947 # FIXME - we need to phase out dash-less ops
948 $op =~ s/^-//; # remove possible initial dash
949 $op =~ s/^\s+|\s+$//g;# remove leading/trailing space
950 $op =~ s/\s+/ /g; # compress whitespace
952 $self->_assert_pass_injection_guard($op);
955 $op =~ s/^is_not/IS NOT/i;
957 # so that -not_foo works correctly
958 $op =~ s/^not_/NOT /i;
960 # another retarded special case: foo => { $op => { -value => undef } }
961 if (ref $val eq 'HASH' and keys %$val == 1 and exists $val->{-value} and ! defined $val->{-value} ) {
967 # CASE: col-value logic modifiers
968 if ($orig_op =~ /^ \- (and|or) $/xi) {
969 ($sql, @bind) = $self->_where_hashpair_HASHREF($k, $val, $1);
971 # CASE: special operators like -in or -between
972 elsif (my $special_op = List::Util::first { $op =~ $_->{regex} } @{$self->{special_ops}}) {
973 my $handler = $special_op->{handler};
975 puke "No handler supplied for special operator $orig_op";
977 elsif (not ref $handler) {
978 ($sql, @bind) = $self->$handler($k, $op, $val);
980 elsif (ref $handler eq 'CODE') {
981 ($sql, @bind) = $handler->($self, $k, $op, $val);
984 puke "Illegal handler for special operator $orig_op - expecting a method name or a coderef";
988 $self->_SWITCH_refkind($val, {
990 ARRAYREF => sub { # CASE: col => {op => \@vals}
991 ($sql, @bind) = $self->_where_field_op_ARRAYREF($k, $op, $val);
994 ARRAYREFREF => sub { # CASE: col => {op => \[$sql, @bind]} (literal SQL with bind)
995 my ($sub_sql, @sub_bind) = @$$val;
996 $self->_assert_bindval_matches_bindtype(@sub_bind);
997 $sql = join ' ', $self->_convert($self->_quote($k)),
998 $self->_sqlcase($op),
1003 UNDEF => sub { # CASE: col => {op => undef} : sql "IS (NOT)? NULL"
1005 $op =~ /^not$/i ? 'is not' # legacy
1006 : $op =~ $self->{equality_op} ? 'is'
1007 : $op =~ $self->{like_op} ? belch("Supplying an undefined argument to '@{[ uc $op]}' is deprecated") && 'is'
1008 : $op =~ $self->{inequality_op} ? 'is not'
1009 : $op =~ $self->{not_like_op} ? belch("Supplying an undefined argument to '@{[ uc $op]}' is deprecated") && 'is not'
1010 : puke "unexpected operator '$orig_op' with undef operand";
1012 $sql = $self->_quote($k) . $self->_sqlcase(" $is null");
1015 FALLBACK => sub { # CASE: col => {op/func => $stuff}
1016 ($sql, @bind) = $self->_where_unary_op($op, $val);
1019 $self->_convert($self->_quote($k)),
1020 $self->{_nested_func_lhs} eq $k ? $sql : "($sql)", # top level vs nested
1026 ($all_sql) = (defined $all_sql and $all_sql) ? $self->_join_sql_clauses($logic, [$all_sql, $sql], []) : $sql;
1027 push @all_bind, @bind;
1029 return ($all_sql, @all_bind);
1032 sub _where_field_IS {
1033 my ($self, $k, $op, $v) = @_;
1035 my ($s) = $self->_SWITCH_refkind($v, {
1038 $self->_convert($self->_quote($k)),
1039 map { $self->_sqlcase($_)} ($op, 'null')
1042 puke "$op can only take undef as argument";
1049 sub _where_field_op_ARRAYREF {
1050 my ($self, $k, $op, $vals) = @_;
1052 my @vals = @$vals; #always work on a copy
1055 $self->_debug(sprintf '%s means multiple elements: [ %s ]',
1057 join(', ', map { defined $_ ? "'$_'" : 'NULL' } @vals ),
1060 # see if the first element is an -and/-or op
1062 if (defined $vals[0] && $vals[0] =~ /^ - (AND|OR) $/ix) {
1067 # a long standing API wart - an attempt to change this behavior during
1068 # the 1.50 series failed *spectacularly*. Warn instead and leave the
1073 (!$logic or $logic eq 'OR')
1075 ($op =~ $self->{inequality_op} or $op =~ $self->{not_like_op})
1078 belch "A multi-element arrayref as an argument to the inequality op '$o' "
1079 . 'is technically equivalent to an always-true 1=1 (you probably wanted '
1080 . "to say ...{ \$inequality_op => [ -and => \@values ] }... instead)"
1084 # distribute $op over each remaining member of @vals, append logic if exists
1085 return $self->_recurse_where([map { {$k => {$op, $_}} } @vals], $logic);
1089 # try to DWIM on equality operators
1091 $op =~ $self->{equality_op} ? $self->{sqlfalse}
1092 : $op =~ $self->{like_op} ? belch("Supplying an empty arrayref to '@{[ uc $op]}' is deprecated") && $self->{sqlfalse}
1093 : $op =~ $self->{inequality_op} ? $self->{sqltrue}
1094 : $op =~ $self->{not_like_op} ? belch("Supplying an empty arrayref to '@{[ uc $op]}' is deprecated") && $self->{sqltrue}
1095 : puke "operator '$op' applied on an empty array (field '$k')";
1100 sub _where_hashpair_SCALARREF {
1101 my ($self, $k, $v) = @_;
1102 $self->_debug("SCALAR($k) means literal SQL: $$v");
1103 my $sql = $self->_quote($k) . " " . $$v;
1107 # literal SQL with bind
1108 sub _where_hashpair_ARRAYREFREF {
1109 my ($self, $k, $v) = @_;
1110 $self->_debug("REF($k) means literal SQL: @${$v}");
1111 my ($sql, @bind) = @$$v;
1112 $self->_assert_bindval_matches_bindtype(@bind);
1113 $sql = $self->_quote($k) . " " . $sql;
1114 return ($sql, @bind );
1117 # literal SQL without bind
1118 sub _where_hashpair_SCALAR {
1119 my ($self, $k, $v) = @_;
1120 $self->_debug("NOREF($k) means simple key=val: $k $self->{cmp} $v");
1121 return ($self->_where_hashpair_HASHREF($k, { $self->{cmp} => $v }));
1125 sub _where_hashpair_UNDEF {
1126 my ($self, $k, $v) = @_;
1127 $self->_debug("UNDEF($k) means IS NULL");
1128 return $self->_where_hashpair_HASHREF($k, { -is => undef });
1131 #======================================================================
1132 # WHERE: TOP-LEVEL OTHERS (SCALARREF, SCALAR, UNDEF)
1133 #======================================================================
1136 sub _where_SCALARREF {
1137 my ($self, $where) = @_;
1140 $self->_debug("SCALAR(*top) means literal SQL: $$where");
1146 my ($self, $where) = @_;
1149 $self->_debug("NOREF(*top) means literal SQL: $where");
1160 #======================================================================
1161 # WHERE: BUILTIN SPECIAL OPERATORS (-in, -between)
1162 #======================================================================
1165 sub _where_field_BETWEEN {
1166 my ($self, $k, $op, $vals) = @_;
1168 my ($label, $and, $placeholder);
1169 $label = $self->_convert($self->_quote($k));
1170 $and = ' ' . $self->_sqlcase('and') . ' ';
1171 $placeholder = $self->_convert('?');
1172 $op = $self->_sqlcase($op);
1174 my $invalid_args = "Operator '$op' requires either an arrayref with two defined values or expressions, or a single literal scalarref/arrayref-ref";
1176 my ($clause, @bind) = $self->_SWITCH_refkind($vals, {
1177 ARRAYREFREF => sub {
1178 my ($s, @b) = @$$vals;
1179 $self->_assert_bindval_matches_bindtype(@b);
1186 puke $invalid_args if @$vals != 2;
1188 my (@all_sql, @all_bind);
1189 foreach my $val (@$vals) {
1190 my ($sql, @bind) = $self->_SWITCH_refkind($val, {
1192 return ($placeholder, $self->_bindtype($k, $val) );
1197 ARRAYREFREF => sub {
1198 my ($sql, @bind) = @$$val;
1199 $self->_assert_bindval_matches_bindtype(@bind);
1200 return ($sql, @bind);
1203 my ($func, $arg, @rest) = %$val;
1204 puke "Only simple { -func => arg } functions accepted as sub-arguments to BETWEEN"
1205 if (@rest or $func !~ /^ \- (.+)/x);
1206 $self->_where_unary_op($1 => $arg);
1212 push @all_sql, $sql;
1213 push @all_bind, @bind;
1217 (join $and, @all_sql),
1226 my $sql = "( $label $op $clause )";
1227 return ($sql, @bind)
1231 sub _where_field_IN {
1232 my ($self, $k, $op, $vals) = @_;
1234 # backwards compatibility: if scalar, force into an arrayref
1235 $vals = [$vals] if defined $vals && ! ref $vals;
1237 my ($label) = $self->_convert($self->_quote($k));
1238 my ($placeholder) = $self->_convert('?');
1239 $op = $self->_sqlcase($op);
1241 my ($sql, @bind) = $self->_SWITCH_refkind($vals, {
1242 ARRAYREF => sub { # list of choices
1243 if (@$vals) { # nonempty list
1244 my (@all_sql, @all_bind);
1246 for my $val (@$vals) {
1247 my ($sql, @bind) = $self->_SWITCH_refkind($val, {
1249 return ($placeholder, $val);
1254 ARRAYREFREF => sub {
1255 my ($sql, @bind) = @$$val;
1256 $self->_assert_bindval_matches_bindtype(@bind);
1257 return ($sql, @bind);
1260 my ($func, $arg, @rest) = %$val;
1261 puke "Only simple { -func => arg } functions accepted as sub-arguments to IN"
1262 if (@rest or $func !~ /^ \- (.+)/x);
1263 $self->_where_unary_op($1 => $arg);
1267 'SQL::Abstract before v1.75 used to generate incorrect SQL when the '
1268 . "-$op operator was given an undef-containing list: !!!AUDIT YOUR CODE "
1269 . 'AND DATA!!! (the upcoming Data::Query-based version of SQL::Abstract '
1270 . 'will emit the logically correct SQL instead of raising this exception)'
1274 push @all_sql, $sql;
1275 push @all_bind, @bind;
1279 sprintf('%s %s ( %s )',
1282 join(', ', @all_sql)
1284 $self->_bindtype($k, @all_bind),
1287 else { # empty list: some databases won't understand "IN ()", so DWIM
1288 my $sql = ($op =~ /\bnot\b/i) ? $self->{sqltrue} : $self->{sqlfalse};
1293 SCALARREF => sub { # literal SQL
1294 my $sql = $self->_open_outer_paren($$vals);
1295 return ("$label $op ( $sql )");
1297 ARRAYREFREF => sub { # literal SQL with bind
1298 my ($sql, @bind) = @$$vals;
1299 $self->_assert_bindval_matches_bindtype(@bind);
1300 $sql = $self->_open_outer_paren($sql);
1301 return ("$label $op ( $sql )", @bind);
1305 puke "Argument passed to the '$op' operator can not be undefined";
1309 puke "special op $op requires an arrayref (or scalarref/arrayref-ref)";
1313 return ($sql, @bind);
1316 # Some databases (SQLite) treat col IN (1, 2) different from
1317 # col IN ( (1, 2) ). Use this to strip all outer parens while
1318 # adding them back in the corresponding method
1319 sub _open_outer_paren {
1320 my ($self, $sql) = @_;
1322 while (my ($inner) = $sql =~ /^ \s* \( (.*) \) \s* $/xs) {
1324 # there are closing parens inside, need the heavy duty machinery
1325 # to reevaluate the extraction starting from $sql (full reevaluation)
1326 if ($inner =~ /\)/) {
1327 require Text::Balanced;
1329 my (undef, $remainder) = do {
1330 # idiotic design - writes to $@ but *DOES NOT* throw exceptions
1332 Text::Balanced::extract_bracketed($sql, '()', qr/\s*/);
1335 # the entire expression needs to be a balanced bracketed thing
1336 # (after an extract no remainder sans trailing space)
1337 last if defined $remainder and $remainder =~ /\S/;
1347 #======================================================================
1349 #======================================================================
1352 my ($self, $arg) = @_;
1355 for my $c ($self->_order_by_chunks($arg) ) {
1356 $self->_SWITCH_refkind($c, {
1357 SCALAR => sub { push @sql, $c },
1358 ARRAYREF => sub { push @sql, shift @$c; push @bind, @$c },
1364 $self->_sqlcase(' order by'),
1370 return wantarray ? ($sql, @bind) : $sql;
1373 sub _order_by_chunks {
1374 my ($self, $arg) = @_;
1376 return $self->_SWITCH_refkind($arg, {
1379 map { $self->_order_by_chunks($_ ) } @$arg;
1382 ARRAYREFREF => sub {
1383 my ($s, @b) = @$$arg;
1384 $self->_assert_bindval_matches_bindtype(@b);
1388 SCALAR => sub {$self->_quote($arg)},
1390 UNDEF => sub {return () },
1392 SCALARREF => sub {$$arg}, # literal SQL, no quoting
1395 # get first pair in hash
1396 my ($key, $val, @rest) = %$arg;
1398 return () unless $key;
1400 if (@rest or not $key =~ /^-(desc|asc)/i) {
1401 puke "hash passed to _order_by must have exactly one key (-desc or -asc)";
1407 for my $c ($self->_order_by_chunks($val)) {
1410 $self->_SWITCH_refkind($c, {
1415 ($sql, @bind) = @$c;
1419 $sql = $sql . ' ' . $self->_sqlcase($direction);
1421 push @ret, [ $sql, @bind];
1430 #======================================================================
1431 # DATASOURCE (FOR NOW, JUST PLAIN TABLE OR LIST OF TABLES)
1432 #======================================================================
1437 $self->_SWITCH_refkind($from, {
1438 ARRAYREF => sub {join ', ', map { $self->_quote($_) } @$from;},
1439 SCALAR => sub {$self->_quote($from)},
1440 SCALARREF => sub {$$from},
1445 #======================================================================
1447 #======================================================================
1449 # highly optimized, as it's called way too often
1451 # my ($self, $label) = @_;
1453 return '' unless defined $_[1];
1454 return ${$_[1]} if ref($_[1]) eq 'SCALAR';
1456 $_[0]->{quote_char} or
1457 ($_[0]->_assert_pass_injection_guard($_[1]), return $_[1]);
1459 my $qref = ref $_[0]->{quote_char};
1461 !$qref ? ($_[0]->{quote_char}, $_[0]->{quote_char})
1462 : ($qref eq 'ARRAY') ? @{$_[0]->{quote_char}}
1463 : puke "Unsupported quote_char format: $_[0]->{quote_char}";
1465 my $esc = $_[0]->{escape_char} || $r;
1467 # parts containing * are naturally unquoted
1468 return join($_[0]->{name_sep}||'', map
1469 +( $_ eq '*' ? $_ : do { (my $n = $_) =~ s/(\Q$esc\E|\Q$r\E)/$esc$1/g; $l . $n . $r } ),
1470 ( $_[0]->{name_sep} ? split (/\Q$_[0]->{name_sep}\E/, $_[1] ) : $_[1] )
1475 # Conversion, if applicable
1477 #my ($self, $arg) = @_;
1478 if ($_[0]->{convert}) {
1479 return $_[0]->_sqlcase($_[0]->{convert}) .'(' . $_[1] . ')';
1486 #my ($self, $col, @vals) = @_;
1487 # called often - tighten code
1488 return $_[0]->{bindtype} eq 'columns'
1489 ? map {[$_[1], $_]} @_[2 .. $#_]
1494 # Dies if any element of @bind is not in [colname => value] format
1495 # if bindtype is 'columns'.
1496 sub _assert_bindval_matches_bindtype {
1497 # my ($self, @bind) = @_;
1499 if ($self->{bindtype} eq 'columns') {
1501 if (!defined $_ || ref($_) ne 'ARRAY' || @$_ != 2) {
1502 puke "bindtype 'columns' selected, you need to pass: [column_name => bind_value]"
1508 sub _join_sql_clauses {
1509 my ($self, $logic, $clauses_aref, $bind_aref) = @_;
1511 if (@$clauses_aref > 1) {
1512 my $join = " " . $self->_sqlcase($logic) . " ";
1513 my $sql = '( ' . join($join, @$clauses_aref) . ' )';
1514 return ($sql, @$bind_aref);
1516 elsif (@$clauses_aref) {
1517 return ($clauses_aref->[0], @$bind_aref); # no parentheses
1520 return (); # if no SQL, ignore @$bind_aref
1525 # Fix SQL case, if so requested
1527 # LDNOTE: if $self->{case} is true, then it contains 'lower', so we
1528 # don't touch the argument ... crooked logic, but let's not change it!
1529 return $_[0]->{case} ? $_[1] : uc($_[1]);
1533 #======================================================================
1534 # DISPATCHING FROM REFKIND
1535 #======================================================================
1538 my ($self, $data) = @_;
1540 return 'UNDEF' unless defined $data;
1542 # blessed objects are treated like scalars
1543 my $ref = (Scalar::Util::blessed $data) ? '' : ref $data;
1545 return 'SCALAR' unless $ref;
1548 while ($ref eq 'REF') {
1550 $ref = (Scalar::Util::blessed $data) ? '' : ref $data;
1554 return ($ref||'SCALAR') . ('REF' x $n_steps);
1558 my ($self, $data) = @_;
1559 my @try = ($self->_refkind($data));
1560 push @try, 'SCALAR_or_UNDEF' if $try[0] eq 'SCALAR' || $try[0] eq 'UNDEF';
1561 push @try, 'FALLBACK';
1565 sub _METHOD_FOR_refkind {
1566 my ($self, $meth_prefix, $data) = @_;
1569 for (@{$self->_try_refkind($data)}) {
1570 $method = $self->can($meth_prefix."_".$_)
1574 return $method || puke "cannot dispatch on '$meth_prefix' for ".$self->_refkind($data);
1578 sub _SWITCH_refkind {
1579 my ($self, $data, $dispatch_table) = @_;
1582 for (@{$self->_try_refkind($data)}) {
1583 $coderef = $dispatch_table->{$_}
1587 puke "no dispatch entry for ".$self->_refkind($data)
1596 #======================================================================
1597 # VALUES, GENERATE, AUTOLOAD
1598 #======================================================================
1600 # LDNOTE: original code from nwiger, didn't touch code in that section
1601 # I feel the AUTOLOAD stuff should not be the default, it should
1602 # only be activated on explicit demand by user.
1606 my $data = shift || return;
1607 puke "Argument to ", __PACKAGE__, "->values must be a \\%hash"
1608 unless ref $data eq 'HASH';
1611 foreach my $k (sort keys %$data) {
1612 my $v = $data->{$k};
1613 $self->_SWITCH_refkind($v, {
1615 if ($self->{array_datatypes}) { # array datatype
1616 push @all_bind, $self->_bindtype($k, $v);
1618 else { # literal SQL with bind
1619 my ($sql, @bind) = @$v;
1620 $self->_assert_bindval_matches_bindtype(@bind);
1621 push @all_bind, @bind;
1624 ARRAYREFREF => sub { # literal SQL with bind
1625 my ($sql, @bind) = @${$v};
1626 $self->_assert_bindval_matches_bindtype(@bind);
1627 push @all_bind, @bind;
1629 SCALARREF => sub { # literal SQL without bind
1631 SCALAR_or_UNDEF => sub {
1632 push @all_bind, $self->_bindtype($k, $v);
1643 my(@sql, @sqlq, @sqlv);
1647 if ($ref eq 'HASH') {
1648 for my $k (sort keys %$_) {
1651 my $label = $self->_quote($k);
1652 if ($r eq 'ARRAY') {
1653 # literal SQL with bind
1654 my ($sql, @bind) = @$v;
1655 $self->_assert_bindval_matches_bindtype(@bind);
1656 push @sqlq, "$label = $sql";
1658 } elsif ($r eq 'SCALAR') {
1659 # literal SQL without bind
1660 push @sqlq, "$label = $$v";
1662 push @sqlq, "$label = ?";
1663 push @sqlv, $self->_bindtype($k, $v);
1666 push @sql, $self->_sqlcase('set'), join ', ', @sqlq;
1667 } elsif ($ref eq 'ARRAY') {
1668 # unlike insert(), assume these are ONLY the column names, i.e. for SQL
1671 if ($r eq 'ARRAY') { # literal SQL with bind
1672 my ($sql, @bind) = @$v;
1673 $self->_assert_bindval_matches_bindtype(@bind);
1676 } elsif ($r eq 'SCALAR') { # literal SQL without bind
1677 # embedded literal SQL
1684 push @sql, '(' . join(', ', @sqlq) . ')';
1685 } elsif ($ref eq 'SCALAR') {
1689 # strings get case twiddled
1690 push @sql, $self->_sqlcase($_);
1694 my $sql = join ' ', @sql;
1696 # this is pretty tricky
1697 # if ask for an array, return ($stmt, @bind)
1698 # otherwise, s/?/shift @sqlv/ to put it inline
1700 return ($sql, @sqlv);
1702 1 while $sql =~ s/\?/my $d = shift(@sqlv);
1703 ref $d ? $d->[1] : $d/e;
1712 # This allows us to check for a local, then _form, attr
1714 my($name) = $AUTOLOAD =~ /.*::(.+)/;
1715 return $self->generate($name, @_);
1726 SQL::Abstract - Generate SQL from Perl data structures
1732 my $sql = SQL::Abstract->new;
1734 my($stmt, @bind) = $sql->select($source, \@fields, \%where, $order);
1736 my($stmt, @bind) = $sql->insert($table, \%fieldvals || \@values);
1738 my($stmt, @bind) = $sql->update($table, \%fieldvals, \%where);
1740 my($stmt, @bind) = $sql->delete($table, \%where);
1742 # Then, use these in your DBI statements
1743 my $sth = $dbh->prepare($stmt);
1744 $sth->execute(@bind);
1746 # Just generate the WHERE clause
1747 my($stmt, @bind) = $sql->where(\%where, $order);
1749 # Return values in the same order, for hashed queries
1750 # See PERFORMANCE section for more details
1751 my @bind = $sql->values(\%fieldvals);
1755 This module was inspired by the excellent L<DBIx::Abstract>.
1756 However, in using that module I found that what I really wanted
1757 to do was generate SQL, but still retain complete control over my
1758 statement handles and use the DBI interface. So, I set out to
1759 create an abstract SQL generation module.
1761 While based on the concepts used by L<DBIx::Abstract>, there are
1762 several important differences, especially when it comes to WHERE
1763 clauses. I have modified the concepts used to make the SQL easier
1764 to generate from Perl data structures and, IMO, more intuitive.
1765 The underlying idea is for this module to do what you mean, based
1766 on the data structures you provide it. The big advantage is that
1767 you don't have to modify your code every time your data changes,
1768 as this module figures it out.
1770 To begin with, an SQL INSERT is as easy as just specifying a hash
1771 of C<key=value> pairs:
1774 name => 'Jimbo Bobson',
1775 phone => '123-456-7890',
1776 address => '42 Sister Lane',
1777 city => 'St. Louis',
1778 state => 'Louisiana',
1781 The SQL can then be generated with this:
1783 my($stmt, @bind) = $sql->insert('people', \%data);
1785 Which would give you something like this:
1787 $stmt = "INSERT INTO people
1788 (address, city, name, phone, state)
1789 VALUES (?, ?, ?, ?, ?)";
1790 @bind = ('42 Sister Lane', 'St. Louis', 'Jimbo Bobson',
1791 '123-456-7890', 'Louisiana');
1793 These are then used directly in your DBI code:
1795 my $sth = $dbh->prepare($stmt);
1796 $sth->execute(@bind);
1798 =head2 Inserting and Updating Arrays
1800 If your database has array types (like for example Postgres),
1801 activate the special option C<< array_datatypes => 1 >>
1802 when creating the C<SQL::Abstract> object.
1803 Then you may use an arrayref to insert and update database array types:
1805 my $sql = SQL::Abstract->new(array_datatypes => 1);
1807 planets => [qw/Mercury Venus Earth Mars/]
1810 my($stmt, @bind) = $sql->insert('solar_system', \%data);
1814 $stmt = "INSERT INTO solar_system (planets) VALUES (?)"
1816 @bind = (['Mercury', 'Venus', 'Earth', 'Mars']);
1819 =head2 Inserting and Updating SQL
1821 In order to apply SQL functions to elements of your C<%data> you may
1822 specify a reference to an arrayref for the given hash value. For example,
1823 if you need to execute the Oracle C<to_date> function on a value, you can
1824 say something like this:
1828 date_entered => \[ "to_date(?,'MM/DD/YYYY')", "03/02/2003" ],
1831 The first value in the array is the actual SQL. Any other values are
1832 optional and would be included in the bind values array. This gives
1835 my($stmt, @bind) = $sql->insert('people', \%data);
1837 $stmt = "INSERT INTO people (name, date_entered)
1838 VALUES (?, to_date(?,'MM/DD/YYYY'))";
1839 @bind = ('Bill', '03/02/2003');
1841 An UPDATE is just as easy, all you change is the name of the function:
1843 my($stmt, @bind) = $sql->update('people', \%data);
1845 Notice that your C<%data> isn't touched; the module will generate
1846 the appropriately quirky SQL for you automatically. Usually you'll
1847 want to specify a WHERE clause for your UPDATE, though, which is
1848 where handling C<%where> hashes comes in handy...
1850 =head2 Complex where statements
1852 This module can generate pretty complicated WHERE statements
1853 easily. For example, simple C<key=value> pairs are taken to mean
1854 equality, and if you want to see if a field is within a set
1855 of values, you can use an arrayref. Let's say we wanted to
1856 SELECT some data based on this criteria:
1859 requestor => 'inna',
1860 worker => ['nwiger', 'rcwe', 'sfz'],
1861 status => { '!=', 'completed' }
1864 my($stmt, @bind) = $sql->select('tickets', '*', \%where);
1866 The above would give you something like this:
1868 $stmt = "SELECT * FROM tickets WHERE
1869 ( requestor = ? ) AND ( status != ? )
1870 AND ( worker = ? OR worker = ? OR worker = ? )";
1871 @bind = ('inna', 'completed', 'nwiger', 'rcwe', 'sfz');
1873 Which you could then use in DBI code like so:
1875 my $sth = $dbh->prepare($stmt);
1876 $sth->execute(@bind);
1882 The methods are simple. There's one for every major SQL operation,
1883 and a constructor you use first. The arguments are specified in a
1884 similar order for each method (table, then fields, then a where
1885 clause) to try and simplify things.
1887 =head2 new(option => 'value')
1889 The C<new()> function takes a list of options and values, and returns
1890 a new B<SQL::Abstract> object which can then be used to generate SQL
1891 through the methods below. The options accepted are:
1897 If set to 'lower', then SQL will be generated in all lowercase. By
1898 default SQL is generated in "textbook" case meaning something like:
1900 SELECT a_field FROM a_table WHERE some_field LIKE '%someval%'
1902 Any setting other than 'lower' is ignored.
1906 This determines what the default comparison operator is. By default
1907 it is C<=>, meaning that a hash like this:
1909 %where = (name => 'nwiger', email => 'nate@wiger.org');
1911 Will generate SQL like this:
1913 WHERE name = 'nwiger' AND email = 'nate@wiger.org'
1915 However, you may want loose comparisons by default, so if you set
1916 C<cmp> to C<like> you would get SQL such as:
1918 WHERE name like 'nwiger' AND email like 'nate@wiger.org'
1920 You can also override the comparison on an individual basis - see
1921 the huge section on L</"WHERE CLAUSES"> at the bottom.
1923 =item sqltrue, sqlfalse
1925 Expressions for inserting boolean values within SQL statements.
1926 By default these are C<1=1> and C<1=0>. They are used
1927 by the special operators C<-in> and C<-not_in> for generating
1928 correct SQL even when the argument is an empty array (see below).
1932 This determines the default logical operator for multiple WHERE
1933 statements in arrays or hashes. If absent, the default logic is "or"
1934 for arrays, and "and" for hashes. This means that a WHERE
1938 event_date => {'>=', '2/13/99'},
1939 event_date => {'<=', '4/24/03'},
1942 will generate SQL like this:
1944 WHERE event_date >= '2/13/99' OR event_date <= '4/24/03'
1946 This is probably not what you want given this query, though (look
1947 at the dates). To change the "OR" to an "AND", simply specify:
1949 my $sql = SQL::Abstract->new(logic => 'and');
1951 Which will change the above C<WHERE> to:
1953 WHERE event_date >= '2/13/99' AND event_date <= '4/24/03'
1955 The logic can also be changed locally by inserting
1956 a modifier in front of an arrayref:
1958 @where = (-and => [event_date => {'>=', '2/13/99'},
1959 event_date => {'<=', '4/24/03'} ]);
1961 See the L</"WHERE CLAUSES"> section for explanations.
1965 This will automatically convert comparisons using the specified SQL
1966 function for both column and value. This is mostly used with an argument
1967 of C<upper> or C<lower>, so that the SQL will have the effect of
1968 case-insensitive "searches". For example, this:
1970 $sql = SQL::Abstract->new(convert => 'upper');
1971 %where = (keywords => 'MaKe iT CAse inSeNSItive');
1973 Will turn out the following SQL:
1975 WHERE upper(keywords) like upper('MaKe iT CAse inSeNSItive')
1977 The conversion can be C<upper()>, C<lower()>, or any other SQL function
1978 that can be applied symmetrically to fields (actually B<SQL::Abstract> does
1979 not validate this option; it will just pass through what you specify verbatim).
1983 This is a kludge because many databases suck. For example, you can't
1984 just bind values using DBI's C<execute()> for Oracle C<CLOB> or C<BLOB> fields.
1985 Instead, you have to use C<bind_param()>:
1987 $sth->bind_param(1, 'reg data');
1988 $sth->bind_param(2, $lots, {ora_type => ORA_CLOB});
1990 The problem is, B<SQL::Abstract> will normally just return a C<@bind> array,
1991 which loses track of which field each slot refers to. Fear not.
1993 If you specify C<bindtype> in new, you can determine how C<@bind> is returned.
1994 Currently, you can specify either C<normal> (default) or C<columns>. If you
1995 specify C<columns>, you will get an array that looks like this:
1997 my $sql = SQL::Abstract->new(bindtype => 'columns');
1998 my($stmt, @bind) = $sql->insert(...);
2001 [ 'column1', 'value1' ],
2002 [ 'column2', 'value2' ],
2003 [ 'column3', 'value3' ],
2006 You can then iterate through this manually, using DBI's C<bind_param()>.
2008 $sth->prepare($stmt);
2011 my($col, $data) = @$_;
2012 if ($col eq 'details' || $col eq 'comments') {
2013 $sth->bind_param($i, $data, {ora_type => ORA_CLOB});
2014 } elsif ($col eq 'image') {
2015 $sth->bind_param($i, $data, {ora_type => ORA_BLOB});
2017 $sth->bind_param($i, $data);
2021 $sth->execute; # execute without @bind now
2023 Now, why would you still use B<SQL::Abstract> if you have to do this crap?
2024 Basically, the advantage is still that you don't have to care which fields
2025 are or are not included. You could wrap that above C<for> loop in a simple
2026 sub called C<bind_fields()> or something and reuse it repeatedly. You still
2027 get a layer of abstraction over manual SQL specification.
2029 Note that if you set L</bindtype> to C<columns>, the C<\[ $sql, @bind ]>
2030 construct (see L</Literal SQL with placeholders and bind values (subqueries)>)
2031 will expect the bind values in this format.
2035 This is the character that a table or column name will be quoted
2036 with. By default this is an empty string, but you could set it to
2037 the character C<`>, to generate SQL like this:
2039 SELECT `a_field` FROM `a_table` WHERE `some_field` LIKE '%someval%'
2041 Alternatively, you can supply an array ref of two items, the first being the left
2042 hand quote character, and the second the right hand quote character. For
2043 example, you could supply C<['[',']']> for SQL Server 2000 compliant quotes
2044 that generates SQL like this:
2046 SELECT [a_field] FROM [a_table] WHERE [some_field] LIKE '%someval%'
2048 Quoting is useful if you have tables or columns names that are reserved
2049 words in your database's SQL dialect.
2053 This is the character that will be used to escape L</quote_char>s appearing
2054 in an identifier before it has been quoted.
2056 The parameter default in case of a single L</quote_char> character is the quote
2059 When opening-closing-style quoting is used (L</quote_char> is an arrayref)
2060 this parameter defaults to the B<closing (right)> L</quote_char>. Occurrences
2061 of the B<opening (left)> L</quote_char> within the identifier are currently left
2062 untouched. The default for opening-closing-style quotes may change in future
2063 versions, thus you are B<strongly encouraged> to specify the escape character
2068 This is the character that separates a table and column name. It is
2069 necessary to specify this when the C<quote_char> option is selected,
2070 so that tables and column names can be individually quoted like this:
2072 SELECT `table`.`one_field` FROM `table` WHERE `table`.`other_field` = 1
2074 =item injection_guard
2076 A regular expression C<qr/.../> that is applied to any C<-function> and unquoted
2077 column name specified in a query structure. This is a safety mechanism to avoid
2078 injection attacks when mishandling user input e.g.:
2080 my %condition_as_column_value_pairs = get_values_from_user();
2081 $sqla->select( ... , \%condition_as_column_value_pairs );
2083 If the expression matches an exception is thrown. Note that literal SQL
2084 supplied via C<\'...'> or C<\['...']> is B<not> checked in any way.
2086 Defaults to checking for C<;> and the C<GO> keyword (TransactSQL)
2088 =item array_datatypes
2090 When this option is true, arrayrefs in INSERT or UPDATE are
2091 interpreted as array datatypes and are passed directly
2093 When this option is false, arrayrefs are interpreted
2094 as literal SQL, just like refs to arrayrefs
2095 (but this behavior is for backwards compatibility; when writing
2096 new queries, use the "reference to arrayref" syntax
2102 Takes a reference to a list of "special operators"
2103 to extend the syntax understood by L<SQL::Abstract>.
2104 See section L</"SPECIAL OPERATORS"> for details.
2108 Takes a reference to a list of "unary operators"
2109 to extend the syntax understood by L<SQL::Abstract>.
2110 See section L</"UNARY OPERATORS"> for details.
2116 =head2 insert($table, \@values || \%fieldvals, \%options)
2118 This is the simplest function. You simply give it a table name
2119 and either an arrayref of values or hashref of field/value pairs.
2120 It returns an SQL INSERT statement and a list of bind values.
2121 See the sections on L</"Inserting and Updating Arrays"> and
2122 L</"Inserting and Updating SQL"> for information on how to insert
2123 with those data types.
2125 The optional C<\%options> hash reference may contain additional
2126 options to generate the insert SQL. Currently supported options
2133 Takes either a scalar of raw SQL fields, or an array reference of
2134 field names, and adds on an SQL C<RETURNING> statement at the end.
2135 This allows you to return data generated by the insert statement
2136 (such as row IDs) without performing another C<SELECT> statement.
2137 Note, however, this is not part of the SQL standard and may not
2138 be supported by all database engines.
2142 =head2 update($table, \%fieldvals, \%where, \%options)
2144 This takes a table, hashref of field/value pairs, and an optional
2145 hashref L<WHERE clause|/WHERE CLAUSES>. It returns an SQL UPDATE function and a list
2147 See the sections on L</"Inserting and Updating Arrays"> and
2148 L</"Inserting and Updating SQL"> for information on how to insert
2149 with those data types.
2151 The optional C<\%options> hash reference may contain additional
2152 options to generate the update SQL. Currently supported options
2159 See the C<returning> option to
2160 L<insert|/insert($table, \@values || \%fieldvals, \%options)>.
2164 =head2 select($source, $fields, $where, $order)
2166 This returns a SQL SELECT statement and associated list of bind values, as
2167 specified by the arguments:
2173 Specification of the 'FROM' part of the statement.
2174 The argument can be either a plain scalar (interpreted as a table
2175 name, will be quoted), or an arrayref (interpreted as a list
2176 of table names, joined by commas, quoted), or a scalarref
2177 (literal SQL, not quoted).
2181 Specification of the list of fields to retrieve from
2183 The argument can be either an arrayref (interpreted as a list
2184 of field names, will be joined by commas and quoted), or a
2185 plain scalar (literal SQL, not quoted).
2186 Please observe that this API is not as flexible as that of
2187 the first argument C<$source>, for backwards compatibility reasons.
2191 Optional argument to specify the WHERE part of the query.
2192 The argument is most often a hashref, but can also be
2193 an arrayref or plain scalar --
2194 see section L<WHERE clause|/"WHERE CLAUSES"> for details.
2198 Optional argument to specify the ORDER BY part of the query.
2199 The argument can be a scalar, a hashref or an arrayref
2200 -- see section L<ORDER BY clause|/"ORDER BY CLAUSES">
2206 =head2 delete($table, \%where, \%options)
2208 This takes a table name and optional hashref L<WHERE clause|/WHERE CLAUSES>.
2209 It returns an SQL DELETE statement and list of bind values.
2211 The optional C<\%options> hash reference may contain additional
2212 options to generate the delete SQL. Currently supported options
2219 See the C<returning> option to
2220 L<insert|/insert($table, \@values || \%fieldvals, \%options)>.
2224 =head2 where(\%where, $order)
2226 This is used to generate just the WHERE clause. For example,
2227 if you have an arbitrary data structure and know what the
2228 rest of your SQL is going to look like, but want an easy way
2229 to produce a WHERE clause, use this. It returns an SQL WHERE
2230 clause and list of bind values.
2233 =head2 values(\%data)
2235 This just returns the values from the hash C<%data>, in the same
2236 order that would be returned from any of the other above queries.
2237 Using this allows you to markedly speed up your queries if you
2238 are affecting lots of rows. See below under the L</"PERFORMANCE"> section.
2240 =head2 generate($any, 'number', $of, \@data, $struct, \%types)
2242 Warning: This is an experimental method and subject to change.
2244 This returns arbitrarily generated SQL. It's a really basic shortcut.
2245 It will return two different things, depending on return context:
2247 my($stmt, @bind) = $sql->generate('create table', \$table, \@fields);
2248 my $stmt_and_val = $sql->generate('create table', \$table, \@fields);
2250 These would return the following:
2252 # First calling form
2253 $stmt = "CREATE TABLE test (?, ?)";
2254 @bind = (field1, field2);
2256 # Second calling form
2257 $stmt_and_val = "CREATE TABLE test (field1, field2)";
2259 Depending on what you're trying to do, it's up to you to choose the correct
2260 format. In this example, the second form is what you would want.
2264 $sql->generate('alter session', { nls_date_format => 'MM/YY' });
2268 ALTER SESSION SET nls_date_format = 'MM/YY'
2270 You get the idea. Strings get their case twiddled, but everything
2271 else remains verbatim.
2273 =head1 EXPORTABLE FUNCTIONS
2275 =head2 is_plain_value
2277 Determines if the supplied argument is a plain value as understood by this
2282 =item * The value is C<undef>
2284 =item * The value is a non-reference
2286 =item * The value is an object with stringification overloading
2288 =item * The value is of the form C<< { -value => $anything } >>
2292 On failure returns C<undef>, on success returns a B<scalar> reference
2293 to the original supplied argument.
2299 The stringification overloading detection is rather advanced: it takes
2300 into consideration not only the presence of a C<""> overload, but if that
2301 fails also checks for enabled
2302 L<autogenerated versions of C<"">|overload/Magic Autogeneration>, based
2303 on either C<0+> or C<bool>.
2305 Unfortunately testing in the field indicates that this
2306 detection B<< may tickle a latent bug in perl versions before 5.018 >>,
2307 but only when very large numbers of stringifying objects are involved.
2308 At the time of writing ( Sep 2014 ) there is no clear explanation of
2309 the direct cause, nor is there a manageably small test case that reliably
2310 reproduces the problem.
2312 If you encounter any of the following exceptions in B<random places within
2313 your application stack> - this module may be to blame:
2315 Operation "ne": no method found,
2316 left argument in overloaded package <something>,
2317 right argument in overloaded package <something>
2321 Stub found while resolving method "???" overloading """" in package <something>
2323 If you fall victim to the above - please attempt to reduce the problem
2324 to something that could be sent to the L<SQL::Abstract developers
2325 |DBIx::Class/GETTING HELP/SUPPORT>
2326 (either publicly or privately). As a workaround in the meantime you can
2327 set C<$ENV{SQLA_ISVALUE_IGNORE_AUTOGENERATED_STRINGIFICATION}> to a true
2328 value, which will most likely eliminate your problem (at the expense of
2329 not being able to properly detect exotic forms of stringification).
2331 This notice and environment variable will be removed in a future version,
2332 as soon as the underlying problem is found and a reliable workaround is
2337 =head2 is_literal_value
2339 Determines if the supplied argument is a literal value as understood by this
2344 =item * C<\$sql_string>
2346 =item * C<\[ $sql_string, @bind_values ]>
2350 On failure returns C<undef>, on success returns an B<array> reference
2351 containing the unpacked version of the supplied literal SQL and bind values.
2353 =head1 WHERE CLAUSES
2357 This module uses a variation on the idea from L<DBIx::Abstract>. It
2358 is B<NOT>, repeat I<not> 100% compatible. B<The main logic of this
2359 module is that things in arrays are OR'ed, and things in hashes
2362 The easiest way to explain is to show lots of examples. After
2363 each C<%where> hash shown, it is assumed you used:
2365 my($stmt, @bind) = $sql->where(\%where);
2367 However, note that the C<%where> hash can be used directly in any
2368 of the other functions as well, as described above.
2370 =head2 Key-value pairs
2372 So, let's get started. To begin, a simple hash:
2376 status => 'completed'
2379 Is converted to SQL C<key = val> statements:
2381 $stmt = "WHERE user = ? AND status = ?";
2382 @bind = ('nwiger', 'completed');
2384 One common thing I end up doing is having a list of values that
2385 a field can be in. To do this, simply specify a list inside of
2390 status => ['assigned', 'in-progress', 'pending'];
2393 This simple code will create the following:
2395 $stmt = "WHERE user = ? AND ( status = ? OR status = ? OR status = ? )";
2396 @bind = ('nwiger', 'assigned', 'in-progress', 'pending');
2398 A field associated to an empty arrayref will be considered a
2399 logical false and will generate 0=1.
2401 =head2 Tests for NULL values
2403 If the value part is C<undef> then this is converted to SQL <IS NULL>
2412 $stmt = "WHERE user = ? AND status IS NULL";
2415 To test if a column IS NOT NULL:
2419 status => { '!=', undef },
2422 =head2 Specific comparison operators
2424 If you want to specify a different type of operator for your comparison,
2425 you can use a hashref for a given column:
2429 status => { '!=', 'completed' }
2432 Which would generate:
2434 $stmt = "WHERE user = ? AND status != ?";
2435 @bind = ('nwiger', 'completed');
2437 To test against multiple values, just enclose the values in an arrayref:
2439 status => { '=', ['assigned', 'in-progress', 'pending'] };
2441 Which would give you:
2443 "WHERE status = ? OR status = ? OR status = ?"
2446 The hashref can also contain multiple pairs, in which case it is expanded
2447 into an C<AND> of its elements:
2451 status => { '!=', 'completed', -not_like => 'pending%' }
2454 # Or more dynamically, like from a form
2455 $where{user} = 'nwiger';
2456 $where{status}{'!='} = 'completed';
2457 $where{status}{'-not_like'} = 'pending%';
2459 # Both generate this
2460 $stmt = "WHERE user = ? AND status != ? AND status NOT LIKE ?";
2461 @bind = ('nwiger', 'completed', 'pending%');
2464 To get an OR instead, you can combine it with the arrayref idea:
2468 priority => [ { '=', 2 }, { '>', 5 } ]
2471 Which would generate:
2473 $stmt = "WHERE ( priority = ? OR priority > ? ) AND user = ?";
2474 @bind = ('2', '5', 'nwiger');
2476 If you want to include literal SQL (with or without bind values), just use a
2477 scalar reference or reference to an arrayref as the value:
2480 date_entered => { '>' => \["to_date(?, 'MM/DD/YYYY')", "11/26/2008"] },
2481 date_expires => { '<' => \"now()" }
2484 Which would generate:
2486 $stmt = "WHERE date_entered > to_date(?, 'MM/DD/YYYY') AND date_expires < now()";
2487 @bind = ('11/26/2008');
2490 =head2 Logic and nesting operators
2492 In the example above,
2493 there is a subtle trap if you want to say something like
2494 this (notice the C<AND>):
2496 WHERE priority != ? AND priority != ?
2498 Because, in Perl you I<can't> do this:
2500 priority => { '!=' => 2, '!=' => 1 }
2502 As the second C<!=> key will obliterate the first. The solution
2503 is to use the special C<-modifier> form inside an arrayref:
2505 priority => [ -and => {'!=', 2},
2509 Normally, these would be joined by C<OR>, but the modifier tells it
2510 to use C<AND> instead. (Hint: You can use this in conjunction with the
2511 C<logic> option to C<new()> in order to change the way your queries
2512 work by default.) B<Important:> Note that the C<-modifier> goes
2513 B<INSIDE> the arrayref, as an extra first element. This will
2514 B<NOT> do what you think it might:
2516 priority => -and => [{'!=', 2}, {'!=', 1}] # WRONG!
2518 Here is a quick list of equivalencies, since there is some overlap:
2521 status => {'!=', 'completed', 'not like', 'pending%' }
2522 status => [ -and => {'!=', 'completed'}, {'not like', 'pending%'}]
2525 status => {'=', ['assigned', 'in-progress']}
2526 status => [ -or => {'=', 'assigned'}, {'=', 'in-progress'}]
2527 status => [ {'=', 'assigned'}, {'=', 'in-progress'} ]
2531 =head2 Special operators: IN, BETWEEN, etc.
2533 You can also use the hashref format to compare a list of fields using the
2534 C<IN> comparison operator, by specifying the list as an arrayref:
2537 status => 'completed',
2538 reportid => { -in => [567, 2335, 2] }
2541 Which would generate:
2543 $stmt = "WHERE status = ? AND reportid IN (?,?,?)";
2544 @bind = ('completed', '567', '2335', '2');
2546 The reverse operator C<-not_in> generates SQL C<NOT IN> and is used in
2549 If the argument to C<-in> is an empty array, 'sqlfalse' is generated
2550 (by default: C<1=0>). Similarly, C<< -not_in => [] >> generates
2551 'sqltrue' (by default: C<1=1>).
2553 In addition to the array you can supply a chunk of literal sql or
2554 literal sql with bind:
2557 customer => { -in => \[
2558 'SELECT cust_id FROM cust WHERE balance > ?',
2561 status => { -in => \'SELECT status_codes FROM states' },
2567 customer IN ( SELECT cust_id FROM cust WHERE balance > ? )
2568 AND status IN ( SELECT status_codes FROM states )
2572 Finally, if the argument to C<-in> is not a reference, it will be
2573 treated as a single-element array.
2575 Another pair of operators is C<-between> and C<-not_between>,
2576 used with an arrayref of two values:
2580 completion_date => {
2581 -not_between => ['2002-10-01', '2003-02-06']
2587 WHERE user = ? AND completion_date NOT BETWEEN ( ? AND ? )
2589 Just like with C<-in> all plausible combinations of literal SQL
2593 start0 => { -between => [ 1, 2 ] },
2594 start1 => { -between => \["? AND ?", 1, 2] },
2595 start2 => { -between => \"lower(x) AND upper(y)" },
2596 start3 => { -between => [
2598 \["upper(?)", 'stuff' ],
2605 ( start0 BETWEEN ? AND ? )
2606 AND ( start1 BETWEEN ? AND ? )
2607 AND ( start2 BETWEEN lower(x) AND upper(y) )
2608 AND ( start3 BETWEEN lower(x) AND upper(?) )
2610 @bind = (1, 2, 1, 2, 'stuff');
2613 These are the two builtin "special operators"; but the
2614 list can be expanded: see section L</"SPECIAL OPERATORS"> below.
2616 =head2 Unary operators: bool
2618 If you wish to test against boolean columns or functions within your
2619 database you can use the C<-bool> and C<-not_bool> operators. For
2620 example to test the column C<is_user> being true and the column
2621 C<is_enabled> being false you would use:-
2625 -not_bool => 'is_enabled',
2630 WHERE is_user AND NOT is_enabled
2632 If a more complex combination is required, testing more conditions,
2633 then you should use the and/or operators:-
2638 -not_bool => { two=> { -rlike => 'bar' } },
2639 -not_bool => { three => [ { '=', 2 }, { '>', 5 } ] },
2650 (NOT ( three = ? OR three > ? ))
2653 =head2 Nested conditions, -and/-or prefixes
2655 So far, we've seen how multiple conditions are joined with a top-level
2656 C<AND>. We can change this by putting the different conditions we want in
2657 hashes and then putting those hashes in an array. For example:
2662 status => { -like => ['pending%', 'dispatched'] },
2666 status => 'unassigned',
2670 This data structure would create the following:
2672 $stmt = "WHERE ( user = ? AND ( status LIKE ? OR status LIKE ? ) )
2673 OR ( user = ? AND status = ? ) )";
2674 @bind = ('nwiger', 'pending', 'dispatched', 'robot', 'unassigned');
2677 Clauses in hashrefs or arrayrefs can be prefixed with an C<-and> or C<-or>
2678 to change the logic inside:
2684 -and => [ workhrs => {'>', 20}, geo => 'ASIA' ],
2685 -or => { workhrs => {'<', 50}, geo => 'EURO' },
2692 $stmt = "WHERE ( user = ?
2693 AND ( ( workhrs > ? AND geo = ? )
2694 OR ( workhrs < ? OR geo = ? ) ) )";
2695 @bind = ('nwiger', '20', 'ASIA', '50', 'EURO');
2697 =head3 Algebraic inconsistency, for historical reasons
2699 C<Important note>: when connecting several conditions, the C<-and->|C<-or>
2700 operator goes C<outside> of the nested structure; whereas when connecting
2701 several constraints on one column, the C<-and> operator goes
2702 C<inside> the arrayref. Here is an example combining both features:
2705 -and => [a => 1, b => 2],
2706 -or => [c => 3, d => 4],
2707 e => [-and => {-like => 'foo%'}, {-like => '%bar'} ]
2712 WHERE ( ( ( a = ? AND b = ? )
2713 OR ( c = ? OR d = ? )
2714 OR ( e LIKE ? AND e LIKE ? ) ) )
2716 This difference in syntax is unfortunate but must be preserved for
2717 historical reasons. So be careful: the two examples below would
2718 seem algebraically equivalent, but they are not
2721 { -like => 'foo%' },
2722 { -like => '%bar' },
2724 # yields: WHERE ( ( col LIKE ? AND col LIKE ? ) )
2727 { col => { -like => 'foo%' } },
2728 { col => { -like => '%bar' } },
2730 # yields: WHERE ( ( col LIKE ? OR col LIKE ? ) )
2733 =head2 Literal SQL and value type operators
2735 The basic premise of SQL::Abstract is that in WHERE specifications the "left
2736 side" is a column name and the "right side" is a value (normally rendered as
2737 a placeholder). This holds true for both hashrefs and arrayref pairs as you
2738 see in the L</WHERE CLAUSES> examples above. Sometimes it is necessary to
2739 alter this behavior. There are several ways of doing so.
2743 This is a virtual operator that signals the string to its right side is an
2744 identifier (a column name) and not a value. For example to compare two
2745 columns you would write:
2748 priority => { '<', 2 },
2749 requestor => { -ident => 'submitter' },
2754 $stmt = "WHERE priority < ? AND requestor = submitter";
2757 If you are maintaining legacy code you may see a different construct as
2758 described in L</Deprecated usage of Literal SQL>, please use C<-ident> in new
2763 This is a virtual operator that signals that the construct to its right side
2764 is a value to be passed to DBI. This is for example necessary when you want
2765 to write a where clause against an array (for RDBMS that support such
2766 datatypes). For example:
2769 array => { -value => [1, 2, 3] }
2774 $stmt = 'WHERE array = ?';
2775 @bind = ([1, 2, 3]);
2777 Note that if you were to simply say:
2783 the result would probably not be what you wanted:
2785 $stmt = 'WHERE array = ? OR array = ? OR array = ?';
2790 Finally, sometimes only literal SQL will do. To include a random snippet
2791 of SQL verbatim, you specify it as a scalar reference. Consider this only
2792 as a last resort. Usually there is a better way. For example:
2795 priority => { '<', 2 },
2796 requestor => { -in => \'(SELECT name FROM hitmen)' },
2801 $stmt = "WHERE priority < ? AND requestor IN (SELECT name FROM hitmen)"
2804 Note that in this example, you only get one bind parameter back, since
2805 the verbatim SQL is passed as part of the statement.
2809 Never use untrusted input as a literal SQL argument - this is a massive
2810 security risk (there is no way to check literal snippets for SQL
2811 injections and other nastyness). If you need to deal with untrusted input
2812 use literal SQL with placeholders as described next.
2814 =head3 Literal SQL with placeholders and bind values (subqueries)
2816 If the literal SQL to be inserted has placeholders and bind values,
2817 use a reference to an arrayref (yes this is a double reference --
2818 not so common, but perfectly legal Perl). For example, to find a date
2819 in Postgres you can use something like this:
2822 date_column => \[ "= date '2008-09-30' - ?::integer", 10 ]
2827 $stmt = "WHERE ( date_column = date '2008-09-30' - ?::integer )"
2830 Note that you must pass the bind values in the same format as they are returned
2831 by L<where|/where(\%where, $order)>. This means that if you set L</bindtype>
2832 to C<columns>, you must provide the bind values in the
2833 C<< [ column_meta => value ] >> format, where C<column_meta> is an opaque
2834 scalar value; most commonly the column name, but you can use any scalar value
2835 (including references and blessed references), L<SQL::Abstract> will simply
2836 pass it through intact. So if C<bindtype> is set to C<columns> the above
2837 example will look like:
2840 date_column => \[ "= date '2008-09-30' - ?::integer", [ {} => 10 ] ]
2843 Literal SQL is especially useful for nesting parenthesized clauses in the
2844 main SQL query. Here is a first example:
2846 my ($sub_stmt, @sub_bind) = ("SELECT c1 FROM t1 WHERE c2 < ? AND c3 LIKE ?",
2850 bar => \["IN ($sub_stmt)" => @sub_bind],
2855 $stmt = "WHERE (foo = ? AND bar IN (SELECT c1 FROM t1
2856 WHERE c2 < ? AND c3 LIKE ?))";
2857 @bind = (1234, 100, "foo%");
2859 Other subquery operators, like for example C<"E<gt> ALL"> or C<"NOT IN">,
2860 are expressed in the same way. Of course the C<$sub_stmt> and
2861 its associated bind values can be generated through a former call
2864 my ($sub_stmt, @sub_bind)
2865 = $sql->select("t1", "c1", {c2 => {"<" => 100},
2866 c3 => {-like => "foo%"}});
2869 bar => \["> ALL ($sub_stmt)" => @sub_bind],
2872 In the examples above, the subquery was used as an operator on a column;
2873 but the same principle also applies for a clause within the main C<%where>
2874 hash, like an EXISTS subquery:
2876 my ($sub_stmt, @sub_bind)
2877 = $sql->select("t1", "*", {c1 => 1, c2 => \"> t0.c0"});
2878 my %where = ( -and => [
2880 \["EXISTS ($sub_stmt)" => @sub_bind],
2885 $stmt = "WHERE (foo = ? AND EXISTS (SELECT * FROM t1
2886 WHERE c1 = ? AND c2 > t0.c0))";
2890 Observe that the condition on C<c2> in the subquery refers to
2891 column C<t0.c0> of the main query: this is I<not> a bind
2892 value, so we have to express it through a scalar ref.
2893 Writing C<< c2 => {">" => "t0.c0"} >> would have generated
2894 C<< c2 > ? >> with bind value C<"t0.c0"> ... not exactly
2895 what we wanted here.
2897 Finally, here is an example where a subquery is used
2898 for expressing unary negation:
2900 my ($sub_stmt, @sub_bind)
2901 = $sql->where({age => [{"<" => 10}, {">" => 20}]});
2902 $sub_stmt =~ s/^ where //i; # don't want "WHERE" in the subclause
2904 lname => {like => '%son%'},
2905 \["NOT ($sub_stmt)" => @sub_bind],
2910 $stmt = "lname LIKE ? AND NOT ( age < ? OR age > ? )"
2911 @bind = ('%son%', 10, 20)
2913 =head3 Deprecated usage of Literal SQL
2915 Below are some examples of archaic use of literal SQL. It is shown only as
2916 reference for those who deal with legacy code. Each example has a much
2917 better, cleaner and safer alternative that users should opt for in new code.
2923 my %where = ( requestor => \'IS NOT NULL' )
2925 $stmt = "WHERE requestor IS NOT NULL"
2927 This used to be the way of generating NULL comparisons, before the handling
2928 of C<undef> got formalized. For new code please use the superior syntax as
2929 described in L</Tests for NULL values>.
2933 my %where = ( requestor => \'= submitter' )
2935 $stmt = "WHERE requestor = submitter"
2937 This used to be the only way to compare columns. Use the superior L</-ident>
2938 method for all new code. For example an identifier declared in such a way
2939 will be properly quoted if L</quote_char> is properly set, while the legacy
2940 form will remain as supplied.
2944 my %where = ( is_ready => \"", completed => { '>', '2012-12-21' } )
2946 $stmt = "WHERE completed > ? AND is_ready"
2947 @bind = ('2012-12-21')
2949 Using an empty string literal used to be the only way to express a boolean.
2950 For all new code please use the much more readable
2951 L<-bool|/Unary operators: bool> operator.
2957 These pages could go on for a while, since the nesting of the data
2958 structures this module can handle are pretty much unlimited (the
2959 module implements the C<WHERE> expansion as a recursive function
2960 internally). Your best bet is to "play around" with the module a
2961 little to see how the data structures behave, and choose the best
2962 format for your data based on that.
2964 And of course, all the values above will probably be replaced with
2965 variables gotten from forms or the command line. After all, if you
2966 knew everything ahead of time, you wouldn't have to worry about
2967 dynamically-generating SQL and could just hardwire it into your
2970 =head1 ORDER BY CLAUSES
2972 Some functions take an order by clause. This can either be a scalar (just a
2973 column name), a hashref of C<< { -desc => 'col' } >> or C<< { -asc => 'col' }
2974 >>, a scalarref, an arrayref-ref, or an arrayref of any of the previous
2977 Given | Will Generate
2978 ---------------------------------------------------------------
2980 'colA' | ORDER BY colA
2982 [qw/colA colB/] | ORDER BY colA, colB
2984 {-asc => 'colA'} | ORDER BY colA ASC
2986 {-desc => 'colB'} | ORDER BY colB DESC
2988 ['colA', {-asc => 'colB'}] | ORDER BY colA, colB ASC
2990 { -asc => [qw/colA colB/] } | ORDER BY colA ASC, colB ASC
2992 \'colA DESC' | ORDER BY colA DESC
2994 \[ 'FUNC(colA, ?)', $x ] | ORDER BY FUNC(colA, ?)
2995 | /* ...with $x bound to ? */
2998 { -asc => 'colA' }, | colA ASC,
2999 { -desc => [qw/colB/] }, | colB DESC,
3000 { -asc => [qw/colC colD/] },| colC ASC, colD ASC,
3001 \'colE DESC', | colE DESC,
3002 \[ 'FUNC(colF, ?)', $x ], | FUNC(colF, ?)
3003 ] | /* ...with $x bound to ? */
3004 ===============================================================
3008 =head1 SPECIAL OPERATORS
3010 my $sqlmaker = SQL::Abstract->new(special_ops => [
3014 my ($self, $field, $op, $arg) = @_;
3020 handler => 'method_name',
3024 A "special operator" is a SQL syntactic clause that can be
3025 applied to a field, instead of a usual binary operator.
3028 WHERE field IN (?, ?, ?)
3029 WHERE field BETWEEN ? AND ?
3030 WHERE MATCH(field) AGAINST (?, ?)
3032 Special operators IN and BETWEEN are fairly standard and therefore
3033 are builtin within C<SQL::Abstract> (as the overridable methods
3034 C<_where_field_IN> and C<_where_field_BETWEEN>). For other operators,
3035 like the MATCH .. AGAINST example above which is specific to MySQL,
3036 you can write your own operator handlers - supply a C<special_ops>
3037 argument to the C<new> method. That argument takes an arrayref of
3038 operator definitions; each operator definition is a hashref with two
3045 the regular expression to match the operator
3049 Either a coderef or a plain scalar method name. In both cases
3050 the expected return is C<< ($sql, @bind) >>.
3052 When supplied with a method name, it is simply called on the
3053 L<SQL::Abstract> object as:
3055 $self->$method_name($field, $op, $arg)
3059 $field is the LHS of the operator
3060 $op is the part that matched the handler regex
3063 When supplied with a coderef, it is called as:
3065 $coderef->($self, $field, $op, $arg)
3070 For example, here is an implementation
3071 of the MATCH .. AGAINST syntax for MySQL
3073 my $sqlmaker = SQL::Abstract->new(special_ops => [
3075 # special op for MySql MATCH (field) AGAINST(word1, word2, ...)
3076 {regex => qr/^match$/i,
3078 my ($self, $field, $op, $arg) = @_;
3079 $arg = [$arg] if not ref $arg;
3080 my $label = $self->_quote($field);
3081 my ($placeholder) = $self->_convert('?');
3082 my $placeholders = join ", ", (($placeholder) x @$arg);
3083 my $sql = $self->_sqlcase('match') . " ($label) "
3084 . $self->_sqlcase('against') . " ($placeholders) ";
3085 my @bind = $self->_bindtype($field, @$arg);
3086 return ($sql, @bind);
3093 =head1 UNARY OPERATORS
3095 my $sqlmaker = SQL::Abstract->new(unary_ops => [
3099 my ($self, $op, $arg) = @_;
3105 handler => 'method_name',
3109 A "unary operator" is a SQL syntactic clause that can be
3110 applied to a field - the operator goes before the field
3112 You can write your own operator handlers - supply a C<unary_ops>
3113 argument to the C<new> method. That argument takes an arrayref of
3114 operator definitions; each operator definition is a hashref with two
3121 the regular expression to match the operator
3125 Either a coderef or a plain scalar method name. In both cases
3126 the expected return is C<< $sql >>.
3128 When supplied with a method name, it is simply called on the
3129 L<SQL::Abstract> object as:
3131 $self->$method_name($op, $arg)
3135 $op is the part that matched the handler regex
3136 $arg is the RHS or argument of the operator
3138 When supplied with a coderef, it is called as:
3140 $coderef->($self, $op, $arg)
3148 Thanks to some benchmarking by Mark Stosberg, it turns out that
3149 this module is many orders of magnitude faster than using C<DBIx::Abstract>.
3150 I must admit this wasn't an intentional design issue, but it's a
3151 byproduct of the fact that you get to control your C<DBI> handles
3154 To maximize performance, use a code snippet like the following:
3156 # prepare a statement handle using the first row
3157 # and then reuse it for the rest of the rows
3159 for my $href (@array_of_hashrefs) {
3160 $stmt ||= $sql->insert('table', $href);
3161 $sth ||= $dbh->prepare($stmt);
3162 $sth->execute($sql->values($href));
3165 The reason this works is because the keys in your C<$href> are sorted
3166 internally by B<SQL::Abstract>. Thus, as long as your data retains
3167 the same structure, you only have to generate the SQL the first time
3168 around. On subsequent queries, simply use the C<values> function provided
3169 by this module to return your values in the correct order.
3171 However this depends on the values having the same type - if, for
3172 example, the values of a where clause may either have values
3173 (resulting in sql of the form C<column = ?> with a single bind
3174 value), or alternatively the values might be C<undef> (resulting in
3175 sql of the form C<column IS NULL> with no bind value) then the
3176 caching technique suggested will not work.
3180 If you use my C<CGI::FormBuilder> module at all, you'll hopefully
3181 really like this part (I do, at least). Building up a complex query
3182 can be as simple as the following:
3189 use CGI::FormBuilder;
3192 my $form = CGI::FormBuilder->new(...);
3193 my $sql = SQL::Abstract->new;
3195 if ($form->submitted) {
3196 my $field = $form->field;
3197 my $id = delete $field->{id};
3198 my($stmt, @bind) = $sql->update('table', $field, {id => $id});
3201 Of course, you would still have to connect using C<DBI> to run the
3202 query, but the point is that if you make your form look like your
3203 table, the actual query script can be extremely simplistic.
3205 If you're B<REALLY> lazy (I am), check out C<HTML::QuickTable> for
3206 a fast interface to returning and formatting data. I frequently
3207 use these three modules together to write complex database query
3208 apps in under 50 lines.
3210 =head1 HOW TO CONTRIBUTE
3212 Contributions are always welcome, in all usable forms (we especially
3213 welcome documentation improvements). The delivery methods include git-
3214 or unified-diff formatted patches, GitHub pull requests, or plain bug
3215 reports either via RT or the Mailing list. Contributors are generally
3216 granted full access to the official repository after their first several
3217 patches pass successful review.
3219 This project is maintained in a git repository. The code and related tools are
3220 accessible at the following locations:
3224 =item * Official repo: L<git://git.shadowcat.co.uk/dbsrgits/SQL-Abstract.git>
3226 =item * Official gitweb: L<http://git.shadowcat.co.uk/gitweb/gitweb.cgi?p=dbsrgits/SQL-Abstract.git>
3228 =item * GitHub mirror: L<https://github.com/dbsrgits/sql-abstract>
3230 =item * Authorized committers: L<ssh://dbsrgits@git.shadowcat.co.uk/SQL-Abstract.git>
3236 Version 1.50 was a major internal refactoring of C<SQL::Abstract>.
3237 Great care has been taken to preserve the I<published> behavior
3238 documented in previous versions in the 1.* family; however,
3239 some features that were previously undocumented, or behaved
3240 differently from the documentation, had to be changed in order
3241 to clarify the semantics. Hence, client code that was relying
3242 on some dark areas of C<SQL::Abstract> v1.*
3243 B<might behave differently> in v1.50.
3245 The main changes are:
3251 support for literal SQL through the C<< \ [ $sql, @bind ] >> syntax.
3255 support for the { operator => \"..." } construct (to embed literal SQL)
3259 support for the { operator => \["...", @bind] } construct (to embed literal SQL with bind values)
3263 optional support for L<array datatypes|/"Inserting and Updating Arrays">
3267 defensive programming: check arguments
3271 fixed bug with global logic, which was previously implemented
3272 through global variables yielding side-effects. Prior versions would
3273 interpret C<< [ {cond1, cond2}, [cond3, cond4] ] >>
3274 as C<< "(cond1 AND cond2) OR (cond3 AND cond4)" >>.
3275 Now this is interpreted
3276 as C<< "(cond1 AND cond2) OR (cond3 OR cond4)" >>.
3281 fixed semantics of _bindtype on array args
3285 dropped the C<_anoncopy> of the %where tree. No longer necessary,
3286 we just avoid shifting arrays within that tree.
3290 dropped the C<_modlogic> function
3294 =head1 ACKNOWLEDGEMENTS
3296 There are a number of individuals that have really helped out with
3297 this module. Unfortunately, most of them submitted bugs via CPAN
3298 so I have no idea who they are! But the people I do know are:
3300 Ash Berlin (order_by hash term support)
3301 Matt Trout (DBIx::Class support)
3302 Mark Stosberg (benchmarking)
3303 Chas Owens (initial "IN" operator support)
3304 Philip Collins (per-field SQL functions)
3305 Eric Kolve (hashref "AND" support)
3306 Mike Fragassi (enhancements to "BETWEEN" and "LIKE")
3307 Dan Kubb (support for "quote_char" and "name_sep")
3308 Guillermo Roditi (patch to cleanup "IN" and "BETWEEN", fix and tests for _order_by)
3309 Laurent Dami (internal refactoring, extensible list of special operators, literal SQL)
3310 Norbert Buchmuller (support for literal SQL in hashpair, misc. fixes & tests)
3311 Peter Rabbitson (rewrite of SQLA::Test, misc. fixes & tests)
3312 Oliver Charles (support for "RETURNING" after "INSERT")
3318 L<DBIx::Class>, L<DBIx::Abstract>, L<CGI::FormBuilder>, L<HTML::QuickTable>.
3322 Copyright (c) 2001-2007 Nathan Wiger <nwiger@cpan.org>. All Rights Reserved.
3324 This module is actively maintained by Matt Trout <mst@shadowcatsystems.co.uk>
3326 For support, your best bet is to try the C<DBIx::Class> users mailing list.
3327 While not an official support venue, C<DBIx::Class> makes heavy use of
3328 C<SQL::Abstract>, and as such list members there are very familiar with
3329 how to create queries.
3333 This module is free software; you may copy this under the same
3334 terms as perl itself (either the GNU General Public License or
3335 the Artistic License)