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);
563 if ($k !~ /^-/ and my $literal = is_literal_value($v)) {
565 belch 'Hash-pairs consisting of an empty string with a literal are deprecated, and will be removed in 2.0: use -and => [ $literal ] instead';
568 my ($sql, @bind) = @$literal;
569 return \[ $self->_quote($k).' '.$sql, @bind ];
575 my ($self, $where, $logic) = @_;
577 my $where_exp = $self->_expand_expr($where, $logic);
579 # dispatch on appropriate method according to refkind of $where
580 my $method = $self->_METHOD_FOR_refkind("_where", $where_exp);
582 my ($sql, @bind) = $self->$method($where_exp, $logic);
584 # DBIx::Class used to call _recurse_where in scalar context
585 # something else might too...
587 return ($sql, @bind);
590 belch "Calling _recurse_where in scalar context is deprecated and will go away before 2.0";
597 #======================================================================
598 # WHERE: top-level ARRAYREF
599 #======================================================================
602 sub _where_ARRAYREF {
603 my ($self, $where, $logic) = @_;
605 $logic = uc($logic || $self->{logic});
606 $logic eq 'AND' or $logic eq 'OR' or puke "unknown logic: $logic";
608 my @clauses = @$where;
610 my (@sql_clauses, @all_bind);
611 # need to use while() so can shift() for pairs
613 my $el = shift @clauses;
615 $el = undef if (defined $el and ! length $el);
617 # switch according to kind of $el and get corresponding ($sql, @bind)
618 my ($sql, @bind) = $self->_SWITCH_refkind($el, {
620 # skip empty elements, otherwise get invalid trailing AND stuff
621 ARRAYREF => sub {$self->_recurse_where($el) if @$el},
625 $self->_assert_bindval_matches_bindtype(@b);
629 HASHREF => sub {$self->_recurse_where($el, 'and') if %$el},
631 SCALARREF => sub { ($$el); },
634 # top-level arrayref with scalars, recurse in pairs
635 $self->_recurse_where({$el => shift(@clauses)})
638 UNDEF => sub {puke "Supplying an empty left hand side argument is not supported in array-pairs" },
642 push @sql_clauses, $sql;
643 push @all_bind, @bind;
647 return $self->_join_sql_clauses($logic, \@sql_clauses, \@all_bind);
650 #======================================================================
651 # WHERE: top-level ARRAYREFREF
652 #======================================================================
654 sub _where_ARRAYREFREF {
655 my ($self, $where) = @_;
656 my ($sql, @bind) = @$$where;
657 $self->_assert_bindval_matches_bindtype(@bind);
658 return ($sql, @bind);
661 #======================================================================
662 # WHERE: top-level HASHREF
663 #======================================================================
666 my ($self, $where) = @_;
667 my (@sql_clauses, @all_bind);
669 for my $k (sort keys %$where) {
670 my $v = $where->{$k};
672 # ($k => $v) is either a special unary op or a regular hashpair
673 my ($sql, @bind) = do {
675 # put the operator in canonical form
677 $op = substr $op, 1; # remove initial dash
678 $op =~ s/^\s+|\s+$//g;# remove leading/trailing space
679 $op =~ s/\s+/ /g; # compress whitespace
681 # so that -not_foo works correctly
682 $op =~ s/^not_/NOT /i;
684 $self->_debug("Unary OP(-$op) within hashref, recursing...");
685 my ($s, @b) = $self->_where_unary_op($op, $v);
687 # top level vs nested
688 # we assume that handled unary ops will take care of their ()s
690 List::Util::first {$op =~ $_->{regex}} @{$self->{unary_ops}}
692 ( defined $self->{_nested_func_lhs} and $self->{_nested_func_lhs} eq $k )
698 if (is_literal_value ($v) ) {
699 belch 'Hash-pairs consisting of an empty string with a literal are deprecated, and will be removed in 2.0: use -and => [ $literal ] instead';
702 puke "Supplying an empty left hand side argument is not supported in hash-pairs";
706 my $method = $self->_METHOD_FOR_refkind("_where_hashpair", $v);
707 $self->$method($k, $v);
711 push @sql_clauses, $sql;
712 push @all_bind, @bind;
715 return $self->_join_sql_clauses('and', \@sql_clauses, \@all_bind);
718 sub _where_unary_op {
719 my ($self, $op, $rhs) = @_;
721 # top level special ops are illegal in general
722 # this includes the -ident/-value ops (dual purpose unary and special)
723 puke "Illegal use of top-level '-$op'"
724 if ! defined $self->{_nested_func_lhs} and List::Util::first { $op =~ $_->{regex} } @{$self->{special_ops}};
726 if (my $op_entry = List::Util::first { $op =~ $_->{regex} } @{$self->{unary_ops}}) {
727 my $handler = $op_entry->{handler};
729 if (not ref $handler) {
730 if ($op =~ s/ [_\s]? \d+ $//x ) {
731 belch 'Use of [and|or|nest]_N modifiers is deprecated and will be removed in SQLA v2.0. '
732 . "You probably wanted ...-and => [ -$op => COND1, -$op => COND2 ... ]";
734 return $self->$handler($op, $rhs);
736 elsif (ref $handler eq 'CODE') {
737 return $handler->($self, $op, $rhs);
740 puke "Illegal handler for operator $op - expecting a method name or a coderef";
744 $self->_debug("Generic unary OP: $op - recursing as function");
746 $self->_assert_pass_injection_guard($op);
748 my ($sql, @bind) = $self->_SWITCH_refkind($rhs, {
750 puke "Illegal use of top-level '-$op'"
751 unless defined $self->{_nested_func_lhs};
754 $self->_convert('?'),
755 $self->_bindtype($self->{_nested_func_lhs}, $rhs)
759 $self->_recurse_where($rhs)
763 $sql = sprintf('%s %s',
764 $self->_sqlcase($op),
768 return ($sql, @bind);
771 sub _where_op_ANDOR {
772 my ($self, $op, $v) = @_;
774 $self->_SWITCH_refkind($v, {
776 return $self->_where_ARRAYREF($v, $op);
780 return ($op =~ /^or/i)
781 ? $self->_where_ARRAYREF([ map { $_ => $v->{$_} } (sort keys %$v) ], $op)
782 : $self->_where_HASHREF($v);
786 puke "-$op => \\\$scalar makes little sense, use " .
788 ? '[ \$scalar, \%rest_of_conditions ] instead'
789 : '-and => [ \$scalar, \%rest_of_conditions ] instead'
794 puke "-$op => \\[...] makes little sense, use " .
796 ? '[ \[...], \%rest_of_conditions ] instead'
797 : '-and => [ \[...], \%rest_of_conditions ] instead'
801 SCALAR => sub { # permissively interpreted as SQL
802 puke "-$op => \$value makes little sense, use -bool => \$value instead";
806 puke "-$op => undef not supported";
812 my ($self, $op, $v) = @_;
814 $self->_SWITCH_refkind($v, {
816 SCALAR => sub { # permissively interpreted as SQL
817 belch "literal SQL should be -nest => \\'scalar' "
818 . "instead of -nest => 'scalar' ";
823 puke "-$op => undef not supported";
827 $self->_recurse_where($v);
835 my ($self, $op, $v) = @_;
837 my ($s, @b) = $self->_SWITCH_refkind($v, {
838 SCALAR => sub { # interpreted as SQL column
839 $self->_convert($self->_quote($v));
843 puke "-$op => undef not supported";
847 $self->_recurse_where($v);
851 $s = "(NOT $s)" if $op =~ /^not/i;
856 sub _where_op_IDENT {
858 my ($op, $rhs) = splice @_, -2;
859 if (! defined $rhs or length ref $rhs) {
860 puke "-$op requires a single plain scalar argument (a quotable identifier)";
863 # in case we are called as a top level special op (no '=')
866 $_ = $self->_convert($self->_quote($_)) for ($lhs, $rhs);
874 sub _where_op_VALUE {
876 my ($op, $rhs) = splice @_, -2;
878 # in case we are called as a top level special op (no '=')
882 if (! defined $rhs) {
884 ? $self->_where_hashpair_HASHREF($lhs, { -is => undef })
891 (defined $lhs ? $lhs : $self->{_nested_func_lhs}),
898 $self->_convert($self->_quote($lhs)) . ' = ' . $self->_convert('?'),
902 $self->_convert('?'),
908 sub _where_hashpair_ARRAYREF {
909 my ($self, $k, $v) = @_;
912 my @v = @$v; # need copy because of shift below
913 $self->_debug("ARRAY($k) means distribute over elements");
915 # put apart first element if it is an operator (-and, -or)
917 (defined $v[0] && $v[0] =~ /^ - (?: AND|OR ) $/ix)
921 my @distributed = map { {$k => $_} } @v;
924 $self->_debug("OP($op) reinjected into the distributed array");
925 unshift @distributed, $op;
928 my $logic = $op ? substr($op, 1) : '';
930 return $self->_recurse_where(\@distributed, $logic);
933 $self->_debug("empty ARRAY($k) means 0=1");
934 return ($self->{sqlfalse});
938 sub _where_hashpair_HASHREF {
939 my ($self, $k, $v, $logic) = @_;
942 local $self->{_nested_func_lhs} = defined $self->{_nested_func_lhs}
943 ? $self->{_nested_func_lhs}
947 my ($all_sql, @all_bind);
949 for my $orig_op (sort keys %$v) {
950 my $val = $v->{$orig_op};
952 # put the operator in canonical form
955 # FIXME - we need to phase out dash-less ops
956 $op =~ s/^-//; # remove possible initial dash
957 $op =~ s/^\s+|\s+$//g;# remove leading/trailing space
958 $op =~ s/\s+/ /g; # compress whitespace
960 $self->_assert_pass_injection_guard($op);
963 $op =~ s/^is_not/IS NOT/i;
965 # so that -not_foo works correctly
966 $op =~ s/^not_/NOT /i;
968 # another retarded special case: foo => { $op => { -value => undef } }
969 if (ref $val eq 'HASH' and keys %$val == 1 and exists $val->{-value} and ! defined $val->{-value} ) {
975 # CASE: col-value logic modifiers
976 if ($orig_op =~ /^ \- (and|or) $/xi) {
977 ($sql, @bind) = $self->_where_hashpair_HASHREF($k, $val, $1);
979 # CASE: special operators like -in or -between
980 elsif (my $special_op = List::Util::first { $op =~ $_->{regex} } @{$self->{special_ops}}) {
981 my $handler = $special_op->{handler};
983 puke "No handler supplied for special operator $orig_op";
985 elsif (not ref $handler) {
986 ($sql, @bind) = $self->$handler($k, $op, $val);
988 elsif (ref $handler eq 'CODE') {
989 ($sql, @bind) = $handler->($self, $k, $op, $val);
992 puke "Illegal handler for special operator $orig_op - expecting a method name or a coderef";
996 $self->_SWITCH_refkind($val, {
998 ARRAYREF => sub { # CASE: col => {op => \@vals}
999 ($sql, @bind) = $self->_where_field_op_ARRAYREF($k, $op, $val);
1002 ARRAYREFREF => sub { # CASE: col => {op => \[$sql, @bind]} (literal SQL with bind)
1003 my ($sub_sql, @sub_bind) = @$$val;
1004 $self->_assert_bindval_matches_bindtype(@sub_bind);
1005 $sql = join ' ', $self->_convert($self->_quote($k)),
1006 $self->_sqlcase($op),
1011 UNDEF => sub { # CASE: col => {op => undef} : sql "IS (NOT)? NULL"
1013 $op =~ /^not$/i ? 'is not' # legacy
1014 : $op =~ $self->{equality_op} ? 'is'
1015 : $op =~ $self->{like_op} ? belch("Supplying an undefined argument to '@{[ uc $op]}' is deprecated") && 'is'
1016 : $op =~ $self->{inequality_op} ? 'is not'
1017 : $op =~ $self->{not_like_op} ? belch("Supplying an undefined argument to '@{[ uc $op]}' is deprecated") && 'is not'
1018 : puke "unexpected operator '$orig_op' with undef operand";
1020 $sql = $self->_quote($k) . $self->_sqlcase(" $is null");
1023 FALLBACK => sub { # CASE: col => {op/func => $stuff}
1024 ($sql, @bind) = $self->_where_unary_op($op, $val);
1027 $self->_convert($self->_quote($k)),
1028 $self->{_nested_func_lhs} eq $k ? $sql : "($sql)", # top level vs nested
1034 ($all_sql) = (defined $all_sql and $all_sql) ? $self->_join_sql_clauses($logic, [$all_sql, $sql], []) : $sql;
1035 push @all_bind, @bind;
1037 return ($all_sql, @all_bind);
1040 sub _where_field_IS {
1041 my ($self, $k, $op, $v) = @_;
1043 my ($s) = $self->_SWITCH_refkind($v, {
1046 $self->_convert($self->_quote($k)),
1047 map { $self->_sqlcase($_)} ($op, 'null')
1050 puke "$op can only take undef as argument";
1057 sub _where_field_op_ARRAYREF {
1058 my ($self, $k, $op, $vals) = @_;
1060 my @vals = @$vals; #always work on a copy
1063 $self->_debug(sprintf '%s means multiple elements: [ %s ]',
1065 join(', ', map { defined $_ ? "'$_'" : 'NULL' } @vals ),
1068 # see if the first element is an -and/-or op
1070 if (defined $vals[0] && $vals[0] =~ /^ - (AND|OR) $/ix) {
1075 # a long standing API wart - an attempt to change this behavior during
1076 # the 1.50 series failed *spectacularly*. Warn instead and leave the
1081 (!$logic or $logic eq 'OR')
1083 ($op =~ $self->{inequality_op} or $op =~ $self->{not_like_op})
1086 belch "A multi-element arrayref as an argument to the inequality op '$o' "
1087 . 'is technically equivalent to an always-true 1=1 (you probably wanted '
1088 . "to say ...{ \$inequality_op => [ -and => \@values ] }... instead)"
1092 # distribute $op over each remaining member of @vals, append logic if exists
1093 return $self->_recurse_where([map { {$k => {$op, $_}} } @vals], $logic);
1097 # try to DWIM on equality operators
1099 $op =~ $self->{equality_op} ? $self->{sqlfalse}
1100 : $op =~ $self->{like_op} ? belch("Supplying an empty arrayref to '@{[ uc $op]}' is deprecated") && $self->{sqlfalse}
1101 : $op =~ $self->{inequality_op} ? $self->{sqltrue}
1102 : $op =~ $self->{not_like_op} ? belch("Supplying an empty arrayref to '@{[ uc $op]}' is deprecated") && $self->{sqltrue}
1103 : puke "operator '$op' applied on an empty array (field '$k')";
1108 sub _where_hashpair_SCALARREF {
1109 my ($self, $k, $v) = @_;
1110 $self->_debug("SCALAR($k) means literal SQL: $$v");
1111 my $sql = $self->_quote($k) . " " . $$v;
1115 # literal SQL with bind
1116 sub _where_hashpair_ARRAYREFREF {
1117 my ($self, $k, $v) = @_;
1118 $self->_debug("REF($k) means literal SQL: @${$v}");
1119 my ($sql, @bind) = @$$v;
1120 $self->_assert_bindval_matches_bindtype(@bind);
1121 $sql = $self->_quote($k) . " " . $sql;
1122 return ($sql, @bind );
1125 # literal SQL without bind
1126 sub _where_hashpair_SCALAR {
1127 my ($self, $k, $v) = @_;
1128 $self->_debug("NOREF($k) means simple key=val: $k $self->{cmp} $v");
1129 return ($self->_where_hashpair_HASHREF($k, { $self->{cmp} => $v }));
1133 sub _where_hashpair_UNDEF {
1134 my ($self, $k, $v) = @_;
1135 $self->_debug("UNDEF($k) means IS NULL");
1136 return $self->_where_hashpair_HASHREF($k, { -is => undef });
1139 #======================================================================
1140 # WHERE: TOP-LEVEL OTHERS (SCALARREF, SCALAR, UNDEF)
1141 #======================================================================
1144 sub _where_SCALARREF {
1145 my ($self, $where) = @_;
1148 $self->_debug("SCALAR(*top) means literal SQL: $$where");
1154 my ($self, $where) = @_;
1157 $self->_debug("NOREF(*top) means literal SQL: $where");
1168 #======================================================================
1169 # WHERE: BUILTIN SPECIAL OPERATORS (-in, -between)
1170 #======================================================================
1173 sub _where_field_BETWEEN {
1174 my ($self, $k, $op, $vals) = @_;
1176 my ($label, $and, $placeholder);
1177 $label = $self->_convert($self->_quote($k));
1178 $and = ' ' . $self->_sqlcase('and') . ' ';
1179 $placeholder = $self->_convert('?');
1180 $op = $self->_sqlcase($op);
1182 my $invalid_args = "Operator '$op' requires either an arrayref with two defined values or expressions, or a single literal scalarref/arrayref-ref";
1184 my ($clause, @bind) = $self->_SWITCH_refkind($vals, {
1185 ARRAYREFREF => sub {
1186 my ($s, @b) = @$$vals;
1187 $self->_assert_bindval_matches_bindtype(@b);
1194 puke $invalid_args if @$vals != 2;
1196 my (@all_sql, @all_bind);
1197 foreach my $val (@$vals) {
1198 my ($sql, @bind) = $self->_SWITCH_refkind($val, {
1200 return ($placeholder, $self->_bindtype($k, $val) );
1205 ARRAYREFREF => sub {
1206 my ($sql, @bind) = @$$val;
1207 $self->_assert_bindval_matches_bindtype(@bind);
1208 return ($sql, @bind);
1211 my ($func, $arg, @rest) = %$val;
1212 puke "Only simple { -func => arg } functions accepted as sub-arguments to BETWEEN"
1213 if (@rest or $func !~ /^ \- (.+)/x);
1214 $self->_where_unary_op($1 => $arg);
1220 push @all_sql, $sql;
1221 push @all_bind, @bind;
1225 (join $and, @all_sql),
1234 my $sql = "( $label $op $clause )";
1235 return ($sql, @bind)
1239 sub _where_field_IN {
1240 my ($self, $k, $op, $vals) = @_;
1242 # backwards compatibility: if scalar, force into an arrayref
1243 $vals = [$vals] if defined $vals && ! ref $vals;
1245 my ($label) = $self->_convert($self->_quote($k));
1246 my ($placeholder) = $self->_convert('?');
1247 $op = $self->_sqlcase($op);
1249 my ($sql, @bind) = $self->_SWITCH_refkind($vals, {
1250 ARRAYREF => sub { # list of choices
1251 if (@$vals) { # nonempty list
1252 my (@all_sql, @all_bind);
1254 for my $val (@$vals) {
1255 my ($sql, @bind) = $self->_SWITCH_refkind($val, {
1257 return ($placeholder, $val);
1262 ARRAYREFREF => sub {
1263 my ($sql, @bind) = @$$val;
1264 $self->_assert_bindval_matches_bindtype(@bind);
1265 return ($sql, @bind);
1268 my ($func, $arg, @rest) = %$val;
1269 puke "Only simple { -func => arg } functions accepted as sub-arguments to IN"
1270 if (@rest or $func !~ /^ \- (.+)/x);
1271 $self->_where_unary_op($1 => $arg);
1275 'SQL::Abstract before v1.75 used to generate incorrect SQL when the '
1276 . "-$op operator was given an undef-containing list: !!!AUDIT YOUR CODE "
1277 . 'AND DATA!!! (the upcoming Data::Query-based version of SQL::Abstract '
1278 . 'will emit the logically correct SQL instead of raising this exception)'
1282 push @all_sql, $sql;
1283 push @all_bind, @bind;
1287 sprintf('%s %s ( %s )',
1290 join(', ', @all_sql)
1292 $self->_bindtype($k, @all_bind),
1295 else { # empty list: some databases won't understand "IN ()", so DWIM
1296 my $sql = ($op =~ /\bnot\b/i) ? $self->{sqltrue} : $self->{sqlfalse};
1301 SCALARREF => sub { # literal SQL
1302 my $sql = $self->_open_outer_paren($$vals);
1303 return ("$label $op ( $sql )");
1305 ARRAYREFREF => sub { # literal SQL with bind
1306 my ($sql, @bind) = @$$vals;
1307 $self->_assert_bindval_matches_bindtype(@bind);
1308 $sql = $self->_open_outer_paren($sql);
1309 return ("$label $op ( $sql )", @bind);
1313 puke "Argument passed to the '$op' operator can not be undefined";
1317 puke "special op $op requires an arrayref (or scalarref/arrayref-ref)";
1321 return ($sql, @bind);
1324 # Some databases (SQLite) treat col IN (1, 2) different from
1325 # col IN ( (1, 2) ). Use this to strip all outer parens while
1326 # adding them back in the corresponding method
1327 sub _open_outer_paren {
1328 my ($self, $sql) = @_;
1330 while (my ($inner) = $sql =~ /^ \s* \( (.*) \) \s* $/xs) {
1332 # there are closing parens inside, need the heavy duty machinery
1333 # to reevaluate the extraction starting from $sql (full reevaluation)
1334 if ($inner =~ /\)/) {
1335 require Text::Balanced;
1337 my (undef, $remainder) = do {
1338 # idiotic design - writes to $@ but *DOES NOT* throw exceptions
1340 Text::Balanced::extract_bracketed($sql, '()', qr/\s*/);
1343 # the entire expression needs to be a balanced bracketed thing
1344 # (after an extract no remainder sans trailing space)
1345 last if defined $remainder and $remainder =~ /\S/;
1355 #======================================================================
1357 #======================================================================
1360 my ($self, $arg) = @_;
1363 for my $c ($self->_order_by_chunks($arg) ) {
1364 $self->_SWITCH_refkind($c, {
1365 SCALAR => sub { push @sql, $c },
1366 ARRAYREF => sub { push @sql, shift @$c; push @bind, @$c },
1372 $self->_sqlcase(' order by'),
1378 return wantarray ? ($sql, @bind) : $sql;
1381 sub _order_by_chunks {
1382 my ($self, $arg) = @_;
1384 return $self->_SWITCH_refkind($arg, {
1387 map { $self->_order_by_chunks($_ ) } @$arg;
1390 ARRAYREFREF => sub {
1391 my ($s, @b) = @$$arg;
1392 $self->_assert_bindval_matches_bindtype(@b);
1396 SCALAR => sub {$self->_quote($arg)},
1398 UNDEF => sub {return () },
1400 SCALARREF => sub {$$arg}, # literal SQL, no quoting
1403 # get first pair in hash
1404 my ($key, $val, @rest) = %$arg;
1406 return () unless $key;
1408 if (@rest or not $key =~ /^-(desc|asc)/i) {
1409 puke "hash passed to _order_by must have exactly one key (-desc or -asc)";
1415 for my $c ($self->_order_by_chunks($val)) {
1418 $self->_SWITCH_refkind($c, {
1423 ($sql, @bind) = @$c;
1427 $sql = $sql . ' ' . $self->_sqlcase($direction);
1429 push @ret, [ $sql, @bind];
1438 #======================================================================
1439 # DATASOURCE (FOR NOW, JUST PLAIN TABLE OR LIST OF TABLES)
1440 #======================================================================
1445 $self->_SWITCH_refkind($from, {
1446 ARRAYREF => sub {join ', ', map { $self->_quote($_) } @$from;},
1447 SCALAR => sub {$self->_quote($from)},
1448 SCALARREF => sub {$$from},
1453 #======================================================================
1455 #======================================================================
1457 # highly optimized, as it's called way too often
1459 # my ($self, $label) = @_;
1461 return '' unless defined $_[1];
1462 return ${$_[1]} if ref($_[1]) eq 'SCALAR';
1464 $_[0]->{quote_char} or
1465 ($_[0]->_assert_pass_injection_guard($_[1]), return $_[1]);
1467 my $qref = ref $_[0]->{quote_char};
1469 !$qref ? ($_[0]->{quote_char}, $_[0]->{quote_char})
1470 : ($qref eq 'ARRAY') ? @{$_[0]->{quote_char}}
1471 : puke "Unsupported quote_char format: $_[0]->{quote_char}";
1473 my $esc = $_[0]->{escape_char} || $r;
1475 # parts containing * are naturally unquoted
1476 return join($_[0]->{name_sep}||'', map
1477 +( $_ eq '*' ? $_ : do { (my $n = $_) =~ s/(\Q$esc\E|\Q$r\E)/$esc$1/g; $l . $n . $r } ),
1478 ( $_[0]->{name_sep} ? split (/\Q$_[0]->{name_sep}\E/, $_[1] ) : $_[1] )
1483 # Conversion, if applicable
1485 #my ($self, $arg) = @_;
1486 if ($_[0]->{convert}) {
1487 return $_[0]->_sqlcase($_[0]->{convert}) .'(' . $_[1] . ')';
1494 #my ($self, $col, @vals) = @_;
1495 # called often - tighten code
1496 return $_[0]->{bindtype} eq 'columns'
1497 ? map {[$_[1], $_]} @_[2 .. $#_]
1502 # Dies if any element of @bind is not in [colname => value] format
1503 # if bindtype is 'columns'.
1504 sub _assert_bindval_matches_bindtype {
1505 # my ($self, @bind) = @_;
1507 if ($self->{bindtype} eq 'columns') {
1509 if (!defined $_ || ref($_) ne 'ARRAY' || @$_ != 2) {
1510 puke "bindtype 'columns' selected, you need to pass: [column_name => bind_value]"
1516 sub _join_sql_clauses {
1517 my ($self, $logic, $clauses_aref, $bind_aref) = @_;
1519 if (@$clauses_aref > 1) {
1520 my $join = " " . $self->_sqlcase($logic) . " ";
1521 my $sql = '( ' . join($join, @$clauses_aref) . ' )';
1522 return ($sql, @$bind_aref);
1524 elsif (@$clauses_aref) {
1525 return ($clauses_aref->[0], @$bind_aref); # no parentheses
1528 return (); # if no SQL, ignore @$bind_aref
1533 # Fix SQL case, if so requested
1535 # LDNOTE: if $self->{case} is true, then it contains 'lower', so we
1536 # don't touch the argument ... crooked logic, but let's not change it!
1537 return $_[0]->{case} ? $_[1] : uc($_[1]);
1541 #======================================================================
1542 # DISPATCHING FROM REFKIND
1543 #======================================================================
1546 my ($self, $data) = @_;
1548 return 'UNDEF' unless defined $data;
1550 # blessed objects are treated like scalars
1551 my $ref = (Scalar::Util::blessed $data) ? '' : ref $data;
1553 return 'SCALAR' unless $ref;
1556 while ($ref eq 'REF') {
1558 $ref = (Scalar::Util::blessed $data) ? '' : ref $data;
1562 return ($ref||'SCALAR') . ('REF' x $n_steps);
1566 my ($self, $data) = @_;
1567 my @try = ($self->_refkind($data));
1568 push @try, 'SCALAR_or_UNDEF' if $try[0] eq 'SCALAR' || $try[0] eq 'UNDEF';
1569 push @try, 'FALLBACK';
1573 sub _METHOD_FOR_refkind {
1574 my ($self, $meth_prefix, $data) = @_;
1577 for (@{$self->_try_refkind($data)}) {
1578 $method = $self->can($meth_prefix."_".$_)
1582 return $method || puke "cannot dispatch on '$meth_prefix' for ".$self->_refkind($data);
1586 sub _SWITCH_refkind {
1587 my ($self, $data, $dispatch_table) = @_;
1590 for (@{$self->_try_refkind($data)}) {
1591 $coderef = $dispatch_table->{$_}
1595 puke "no dispatch entry for ".$self->_refkind($data)
1604 #======================================================================
1605 # VALUES, GENERATE, AUTOLOAD
1606 #======================================================================
1608 # LDNOTE: original code from nwiger, didn't touch code in that section
1609 # I feel the AUTOLOAD stuff should not be the default, it should
1610 # only be activated on explicit demand by user.
1614 my $data = shift || return;
1615 puke "Argument to ", __PACKAGE__, "->values must be a \\%hash"
1616 unless ref $data eq 'HASH';
1619 foreach my $k (sort keys %$data) {
1620 my $v = $data->{$k};
1621 $self->_SWITCH_refkind($v, {
1623 if ($self->{array_datatypes}) { # array datatype
1624 push @all_bind, $self->_bindtype($k, $v);
1626 else { # literal SQL with bind
1627 my ($sql, @bind) = @$v;
1628 $self->_assert_bindval_matches_bindtype(@bind);
1629 push @all_bind, @bind;
1632 ARRAYREFREF => sub { # literal SQL with bind
1633 my ($sql, @bind) = @${$v};
1634 $self->_assert_bindval_matches_bindtype(@bind);
1635 push @all_bind, @bind;
1637 SCALARREF => sub { # literal SQL without bind
1639 SCALAR_or_UNDEF => sub {
1640 push @all_bind, $self->_bindtype($k, $v);
1651 my(@sql, @sqlq, @sqlv);
1655 if ($ref eq 'HASH') {
1656 for my $k (sort keys %$_) {
1659 my $label = $self->_quote($k);
1660 if ($r eq 'ARRAY') {
1661 # literal SQL with bind
1662 my ($sql, @bind) = @$v;
1663 $self->_assert_bindval_matches_bindtype(@bind);
1664 push @sqlq, "$label = $sql";
1666 } elsif ($r eq 'SCALAR') {
1667 # literal SQL without bind
1668 push @sqlq, "$label = $$v";
1670 push @sqlq, "$label = ?";
1671 push @sqlv, $self->_bindtype($k, $v);
1674 push @sql, $self->_sqlcase('set'), join ', ', @sqlq;
1675 } elsif ($ref eq 'ARRAY') {
1676 # unlike insert(), assume these are ONLY the column names, i.e. for SQL
1679 if ($r eq 'ARRAY') { # literal SQL with bind
1680 my ($sql, @bind) = @$v;
1681 $self->_assert_bindval_matches_bindtype(@bind);
1684 } elsif ($r eq 'SCALAR') { # literal SQL without bind
1685 # embedded literal SQL
1692 push @sql, '(' . join(', ', @sqlq) . ')';
1693 } elsif ($ref eq 'SCALAR') {
1697 # strings get case twiddled
1698 push @sql, $self->_sqlcase($_);
1702 my $sql = join ' ', @sql;
1704 # this is pretty tricky
1705 # if ask for an array, return ($stmt, @bind)
1706 # otherwise, s/?/shift @sqlv/ to put it inline
1708 return ($sql, @sqlv);
1710 1 while $sql =~ s/\?/my $d = shift(@sqlv);
1711 ref $d ? $d->[1] : $d/e;
1720 # This allows us to check for a local, then _form, attr
1722 my($name) = $AUTOLOAD =~ /.*::(.+)/;
1723 return $self->generate($name, @_);
1734 SQL::Abstract - Generate SQL from Perl data structures
1740 my $sql = SQL::Abstract->new;
1742 my($stmt, @bind) = $sql->select($source, \@fields, \%where, $order);
1744 my($stmt, @bind) = $sql->insert($table, \%fieldvals || \@values);
1746 my($stmt, @bind) = $sql->update($table, \%fieldvals, \%where);
1748 my($stmt, @bind) = $sql->delete($table, \%where);
1750 # Then, use these in your DBI statements
1751 my $sth = $dbh->prepare($stmt);
1752 $sth->execute(@bind);
1754 # Just generate the WHERE clause
1755 my($stmt, @bind) = $sql->where(\%where, $order);
1757 # Return values in the same order, for hashed queries
1758 # See PERFORMANCE section for more details
1759 my @bind = $sql->values(\%fieldvals);
1763 This module was inspired by the excellent L<DBIx::Abstract>.
1764 However, in using that module I found that what I really wanted
1765 to do was generate SQL, but still retain complete control over my
1766 statement handles and use the DBI interface. So, I set out to
1767 create an abstract SQL generation module.
1769 While based on the concepts used by L<DBIx::Abstract>, there are
1770 several important differences, especially when it comes to WHERE
1771 clauses. I have modified the concepts used to make the SQL easier
1772 to generate from Perl data structures and, IMO, more intuitive.
1773 The underlying idea is for this module to do what you mean, based
1774 on the data structures you provide it. The big advantage is that
1775 you don't have to modify your code every time your data changes,
1776 as this module figures it out.
1778 To begin with, an SQL INSERT is as easy as just specifying a hash
1779 of C<key=value> pairs:
1782 name => 'Jimbo Bobson',
1783 phone => '123-456-7890',
1784 address => '42 Sister Lane',
1785 city => 'St. Louis',
1786 state => 'Louisiana',
1789 The SQL can then be generated with this:
1791 my($stmt, @bind) = $sql->insert('people', \%data);
1793 Which would give you something like this:
1795 $stmt = "INSERT INTO people
1796 (address, city, name, phone, state)
1797 VALUES (?, ?, ?, ?, ?)";
1798 @bind = ('42 Sister Lane', 'St. Louis', 'Jimbo Bobson',
1799 '123-456-7890', 'Louisiana');
1801 These are then used directly in your DBI code:
1803 my $sth = $dbh->prepare($stmt);
1804 $sth->execute(@bind);
1806 =head2 Inserting and Updating Arrays
1808 If your database has array types (like for example Postgres),
1809 activate the special option C<< array_datatypes => 1 >>
1810 when creating the C<SQL::Abstract> object.
1811 Then you may use an arrayref to insert and update database array types:
1813 my $sql = SQL::Abstract->new(array_datatypes => 1);
1815 planets => [qw/Mercury Venus Earth Mars/]
1818 my($stmt, @bind) = $sql->insert('solar_system', \%data);
1822 $stmt = "INSERT INTO solar_system (planets) VALUES (?)"
1824 @bind = (['Mercury', 'Venus', 'Earth', 'Mars']);
1827 =head2 Inserting and Updating SQL
1829 In order to apply SQL functions to elements of your C<%data> you may
1830 specify a reference to an arrayref for the given hash value. For example,
1831 if you need to execute the Oracle C<to_date> function on a value, you can
1832 say something like this:
1836 date_entered => \[ "to_date(?,'MM/DD/YYYY')", "03/02/2003" ],
1839 The first value in the array is the actual SQL. Any other values are
1840 optional and would be included in the bind values array. This gives
1843 my($stmt, @bind) = $sql->insert('people', \%data);
1845 $stmt = "INSERT INTO people (name, date_entered)
1846 VALUES (?, to_date(?,'MM/DD/YYYY'))";
1847 @bind = ('Bill', '03/02/2003');
1849 An UPDATE is just as easy, all you change is the name of the function:
1851 my($stmt, @bind) = $sql->update('people', \%data);
1853 Notice that your C<%data> isn't touched; the module will generate
1854 the appropriately quirky SQL for you automatically. Usually you'll
1855 want to specify a WHERE clause for your UPDATE, though, which is
1856 where handling C<%where> hashes comes in handy...
1858 =head2 Complex where statements
1860 This module can generate pretty complicated WHERE statements
1861 easily. For example, simple C<key=value> pairs are taken to mean
1862 equality, and if you want to see if a field is within a set
1863 of values, you can use an arrayref. Let's say we wanted to
1864 SELECT some data based on this criteria:
1867 requestor => 'inna',
1868 worker => ['nwiger', 'rcwe', 'sfz'],
1869 status => { '!=', 'completed' }
1872 my($stmt, @bind) = $sql->select('tickets', '*', \%where);
1874 The above would give you something like this:
1876 $stmt = "SELECT * FROM tickets WHERE
1877 ( requestor = ? ) AND ( status != ? )
1878 AND ( worker = ? OR worker = ? OR worker = ? )";
1879 @bind = ('inna', 'completed', 'nwiger', 'rcwe', 'sfz');
1881 Which you could then use in DBI code like so:
1883 my $sth = $dbh->prepare($stmt);
1884 $sth->execute(@bind);
1890 The methods are simple. There's one for every major SQL operation,
1891 and a constructor you use first. The arguments are specified in a
1892 similar order for each method (table, then fields, then a where
1893 clause) to try and simplify things.
1895 =head2 new(option => 'value')
1897 The C<new()> function takes a list of options and values, and returns
1898 a new B<SQL::Abstract> object which can then be used to generate SQL
1899 through the methods below. The options accepted are:
1905 If set to 'lower', then SQL will be generated in all lowercase. By
1906 default SQL is generated in "textbook" case meaning something like:
1908 SELECT a_field FROM a_table WHERE some_field LIKE '%someval%'
1910 Any setting other than 'lower' is ignored.
1914 This determines what the default comparison operator is. By default
1915 it is C<=>, meaning that a hash like this:
1917 %where = (name => 'nwiger', email => 'nate@wiger.org');
1919 Will generate SQL like this:
1921 WHERE name = 'nwiger' AND email = 'nate@wiger.org'
1923 However, you may want loose comparisons by default, so if you set
1924 C<cmp> to C<like> you would get SQL such as:
1926 WHERE name like 'nwiger' AND email like 'nate@wiger.org'
1928 You can also override the comparison on an individual basis - see
1929 the huge section on L</"WHERE CLAUSES"> at the bottom.
1931 =item sqltrue, sqlfalse
1933 Expressions for inserting boolean values within SQL statements.
1934 By default these are C<1=1> and C<1=0>. They are used
1935 by the special operators C<-in> and C<-not_in> for generating
1936 correct SQL even when the argument is an empty array (see below).
1940 This determines the default logical operator for multiple WHERE
1941 statements in arrays or hashes. If absent, the default logic is "or"
1942 for arrays, and "and" for hashes. This means that a WHERE
1946 event_date => {'>=', '2/13/99'},
1947 event_date => {'<=', '4/24/03'},
1950 will generate SQL like this:
1952 WHERE event_date >= '2/13/99' OR event_date <= '4/24/03'
1954 This is probably not what you want given this query, though (look
1955 at the dates). To change the "OR" to an "AND", simply specify:
1957 my $sql = SQL::Abstract->new(logic => 'and');
1959 Which will change the above C<WHERE> to:
1961 WHERE event_date >= '2/13/99' AND event_date <= '4/24/03'
1963 The logic can also be changed locally by inserting
1964 a modifier in front of an arrayref:
1966 @where = (-and => [event_date => {'>=', '2/13/99'},
1967 event_date => {'<=', '4/24/03'} ]);
1969 See the L</"WHERE CLAUSES"> section for explanations.
1973 This will automatically convert comparisons using the specified SQL
1974 function for both column and value. This is mostly used with an argument
1975 of C<upper> or C<lower>, so that the SQL will have the effect of
1976 case-insensitive "searches". For example, this:
1978 $sql = SQL::Abstract->new(convert => 'upper');
1979 %where = (keywords => 'MaKe iT CAse inSeNSItive');
1981 Will turn out the following SQL:
1983 WHERE upper(keywords) like upper('MaKe iT CAse inSeNSItive')
1985 The conversion can be C<upper()>, C<lower()>, or any other SQL function
1986 that can be applied symmetrically to fields (actually B<SQL::Abstract> does
1987 not validate this option; it will just pass through what you specify verbatim).
1991 This is a kludge because many databases suck. For example, you can't
1992 just bind values using DBI's C<execute()> for Oracle C<CLOB> or C<BLOB> fields.
1993 Instead, you have to use C<bind_param()>:
1995 $sth->bind_param(1, 'reg data');
1996 $sth->bind_param(2, $lots, {ora_type => ORA_CLOB});
1998 The problem is, B<SQL::Abstract> will normally just return a C<@bind> array,
1999 which loses track of which field each slot refers to. Fear not.
2001 If you specify C<bindtype> in new, you can determine how C<@bind> is returned.
2002 Currently, you can specify either C<normal> (default) or C<columns>. If you
2003 specify C<columns>, you will get an array that looks like this:
2005 my $sql = SQL::Abstract->new(bindtype => 'columns');
2006 my($stmt, @bind) = $sql->insert(...);
2009 [ 'column1', 'value1' ],
2010 [ 'column2', 'value2' ],
2011 [ 'column3', 'value3' ],
2014 You can then iterate through this manually, using DBI's C<bind_param()>.
2016 $sth->prepare($stmt);
2019 my($col, $data) = @$_;
2020 if ($col eq 'details' || $col eq 'comments') {
2021 $sth->bind_param($i, $data, {ora_type => ORA_CLOB});
2022 } elsif ($col eq 'image') {
2023 $sth->bind_param($i, $data, {ora_type => ORA_BLOB});
2025 $sth->bind_param($i, $data);
2029 $sth->execute; # execute without @bind now
2031 Now, why would you still use B<SQL::Abstract> if you have to do this crap?
2032 Basically, the advantage is still that you don't have to care which fields
2033 are or are not included. You could wrap that above C<for> loop in a simple
2034 sub called C<bind_fields()> or something and reuse it repeatedly. You still
2035 get a layer of abstraction over manual SQL specification.
2037 Note that if you set L</bindtype> to C<columns>, the C<\[ $sql, @bind ]>
2038 construct (see L</Literal SQL with placeholders and bind values (subqueries)>)
2039 will expect the bind values in this format.
2043 This is the character that a table or column name will be quoted
2044 with. By default this is an empty string, but you could set it to
2045 the character C<`>, to generate SQL like this:
2047 SELECT `a_field` FROM `a_table` WHERE `some_field` LIKE '%someval%'
2049 Alternatively, you can supply an array ref of two items, the first being the left
2050 hand quote character, and the second the right hand quote character. For
2051 example, you could supply C<['[',']']> for SQL Server 2000 compliant quotes
2052 that generates SQL like this:
2054 SELECT [a_field] FROM [a_table] WHERE [some_field] LIKE '%someval%'
2056 Quoting is useful if you have tables or columns names that are reserved
2057 words in your database's SQL dialect.
2061 This is the character that will be used to escape L</quote_char>s appearing
2062 in an identifier before it has been quoted.
2064 The parameter default in case of a single L</quote_char> character is the quote
2067 When opening-closing-style quoting is used (L</quote_char> is an arrayref)
2068 this parameter defaults to the B<closing (right)> L</quote_char>. Occurrences
2069 of the B<opening (left)> L</quote_char> within the identifier are currently left
2070 untouched. The default for opening-closing-style quotes may change in future
2071 versions, thus you are B<strongly encouraged> to specify the escape character
2076 This is the character that separates a table and column name. It is
2077 necessary to specify this when the C<quote_char> option is selected,
2078 so that tables and column names can be individually quoted like this:
2080 SELECT `table`.`one_field` FROM `table` WHERE `table`.`other_field` = 1
2082 =item injection_guard
2084 A regular expression C<qr/.../> that is applied to any C<-function> and unquoted
2085 column name specified in a query structure. This is a safety mechanism to avoid
2086 injection attacks when mishandling user input e.g.:
2088 my %condition_as_column_value_pairs = get_values_from_user();
2089 $sqla->select( ... , \%condition_as_column_value_pairs );
2091 If the expression matches an exception is thrown. Note that literal SQL
2092 supplied via C<\'...'> or C<\['...']> is B<not> checked in any way.
2094 Defaults to checking for C<;> and the C<GO> keyword (TransactSQL)
2096 =item array_datatypes
2098 When this option is true, arrayrefs in INSERT or UPDATE are
2099 interpreted as array datatypes and are passed directly
2101 When this option is false, arrayrefs are interpreted
2102 as literal SQL, just like refs to arrayrefs
2103 (but this behavior is for backwards compatibility; when writing
2104 new queries, use the "reference to arrayref" syntax
2110 Takes a reference to a list of "special operators"
2111 to extend the syntax understood by L<SQL::Abstract>.
2112 See section L</"SPECIAL OPERATORS"> for details.
2116 Takes a reference to a list of "unary operators"
2117 to extend the syntax understood by L<SQL::Abstract>.
2118 See section L</"UNARY OPERATORS"> for details.
2124 =head2 insert($table, \@values || \%fieldvals, \%options)
2126 This is the simplest function. You simply give it a table name
2127 and either an arrayref of values or hashref of field/value pairs.
2128 It returns an SQL INSERT statement and a list of bind values.
2129 See the sections on L</"Inserting and Updating Arrays"> and
2130 L</"Inserting and Updating SQL"> for information on how to insert
2131 with those data types.
2133 The optional C<\%options> hash reference may contain additional
2134 options to generate the insert SQL. Currently supported options
2141 Takes either a scalar of raw SQL fields, or an array reference of
2142 field names, and adds on an SQL C<RETURNING> statement at the end.
2143 This allows you to return data generated by the insert statement
2144 (such as row IDs) without performing another C<SELECT> statement.
2145 Note, however, this is not part of the SQL standard and may not
2146 be supported by all database engines.
2150 =head2 update($table, \%fieldvals, \%where, \%options)
2152 This takes a table, hashref of field/value pairs, and an optional
2153 hashref L<WHERE clause|/WHERE CLAUSES>. It returns an SQL UPDATE function and a list
2155 See the sections on L</"Inserting and Updating Arrays"> and
2156 L</"Inserting and Updating SQL"> for information on how to insert
2157 with those data types.
2159 The optional C<\%options> hash reference may contain additional
2160 options to generate the update SQL. Currently supported options
2167 See the C<returning> option to
2168 L<insert|/insert($table, \@values || \%fieldvals, \%options)>.
2172 =head2 select($source, $fields, $where, $order)
2174 This returns a SQL SELECT statement and associated list of bind values, as
2175 specified by the arguments:
2181 Specification of the 'FROM' part of the statement.
2182 The argument can be either a plain scalar (interpreted as a table
2183 name, will be quoted), or an arrayref (interpreted as a list
2184 of table names, joined by commas, quoted), or a scalarref
2185 (literal SQL, not quoted).
2189 Specification of the list of fields to retrieve from
2191 The argument can be either an arrayref (interpreted as a list
2192 of field names, will be joined by commas and quoted), or a
2193 plain scalar (literal SQL, not quoted).
2194 Please observe that this API is not as flexible as that of
2195 the first argument C<$source>, for backwards compatibility reasons.
2199 Optional argument to specify the WHERE part of the query.
2200 The argument is most often a hashref, but can also be
2201 an arrayref or plain scalar --
2202 see section L<WHERE clause|/"WHERE CLAUSES"> for details.
2206 Optional argument to specify the ORDER BY part of the query.
2207 The argument can be a scalar, a hashref or an arrayref
2208 -- see section L<ORDER BY clause|/"ORDER BY CLAUSES">
2214 =head2 delete($table, \%where, \%options)
2216 This takes a table name and optional hashref L<WHERE clause|/WHERE CLAUSES>.
2217 It returns an SQL DELETE statement and list of bind values.
2219 The optional C<\%options> hash reference may contain additional
2220 options to generate the delete SQL. Currently supported options
2227 See the C<returning> option to
2228 L<insert|/insert($table, \@values || \%fieldvals, \%options)>.
2232 =head2 where(\%where, $order)
2234 This is used to generate just the WHERE clause. For example,
2235 if you have an arbitrary data structure and know what the
2236 rest of your SQL is going to look like, but want an easy way
2237 to produce a WHERE clause, use this. It returns an SQL WHERE
2238 clause and list of bind values.
2241 =head2 values(\%data)
2243 This just returns the values from the hash C<%data>, in the same
2244 order that would be returned from any of the other above queries.
2245 Using this allows you to markedly speed up your queries if you
2246 are affecting lots of rows. See below under the L</"PERFORMANCE"> section.
2248 =head2 generate($any, 'number', $of, \@data, $struct, \%types)
2250 Warning: This is an experimental method and subject to change.
2252 This returns arbitrarily generated SQL. It's a really basic shortcut.
2253 It will return two different things, depending on return context:
2255 my($stmt, @bind) = $sql->generate('create table', \$table, \@fields);
2256 my $stmt_and_val = $sql->generate('create table', \$table, \@fields);
2258 These would return the following:
2260 # First calling form
2261 $stmt = "CREATE TABLE test (?, ?)";
2262 @bind = (field1, field2);
2264 # Second calling form
2265 $stmt_and_val = "CREATE TABLE test (field1, field2)";
2267 Depending on what you're trying to do, it's up to you to choose the correct
2268 format. In this example, the second form is what you would want.
2272 $sql->generate('alter session', { nls_date_format => 'MM/YY' });
2276 ALTER SESSION SET nls_date_format = 'MM/YY'
2278 You get the idea. Strings get their case twiddled, but everything
2279 else remains verbatim.
2281 =head1 EXPORTABLE FUNCTIONS
2283 =head2 is_plain_value
2285 Determines if the supplied argument is a plain value as understood by this
2290 =item * The value is C<undef>
2292 =item * The value is a non-reference
2294 =item * The value is an object with stringification overloading
2296 =item * The value is of the form C<< { -value => $anything } >>
2300 On failure returns C<undef>, on success returns a B<scalar> reference
2301 to the original supplied argument.
2307 The stringification overloading detection is rather advanced: it takes
2308 into consideration not only the presence of a C<""> overload, but if that
2309 fails also checks for enabled
2310 L<autogenerated versions of C<"">|overload/Magic Autogeneration>, based
2311 on either C<0+> or C<bool>.
2313 Unfortunately testing in the field indicates that this
2314 detection B<< may tickle a latent bug in perl versions before 5.018 >>,
2315 but only when very large numbers of stringifying objects are involved.
2316 At the time of writing ( Sep 2014 ) there is no clear explanation of
2317 the direct cause, nor is there a manageably small test case that reliably
2318 reproduces the problem.
2320 If you encounter any of the following exceptions in B<random places within
2321 your application stack> - this module may be to blame:
2323 Operation "ne": no method found,
2324 left argument in overloaded package <something>,
2325 right argument in overloaded package <something>
2329 Stub found while resolving method "???" overloading """" in package <something>
2331 If you fall victim to the above - please attempt to reduce the problem
2332 to something that could be sent to the L<SQL::Abstract developers
2333 |DBIx::Class/GETTING HELP/SUPPORT>
2334 (either publicly or privately). As a workaround in the meantime you can
2335 set C<$ENV{SQLA_ISVALUE_IGNORE_AUTOGENERATED_STRINGIFICATION}> to a true
2336 value, which will most likely eliminate your problem (at the expense of
2337 not being able to properly detect exotic forms of stringification).
2339 This notice and environment variable will be removed in a future version,
2340 as soon as the underlying problem is found and a reliable workaround is
2345 =head2 is_literal_value
2347 Determines if the supplied argument is a literal value as understood by this
2352 =item * C<\$sql_string>
2354 =item * C<\[ $sql_string, @bind_values ]>
2358 On failure returns C<undef>, on success returns an B<array> reference
2359 containing the unpacked version of the supplied literal SQL and bind values.
2361 =head1 WHERE CLAUSES
2365 This module uses a variation on the idea from L<DBIx::Abstract>. It
2366 is B<NOT>, repeat I<not> 100% compatible. B<The main logic of this
2367 module is that things in arrays are OR'ed, and things in hashes
2370 The easiest way to explain is to show lots of examples. After
2371 each C<%where> hash shown, it is assumed you used:
2373 my($stmt, @bind) = $sql->where(\%where);
2375 However, note that the C<%where> hash can be used directly in any
2376 of the other functions as well, as described above.
2378 =head2 Key-value pairs
2380 So, let's get started. To begin, a simple hash:
2384 status => 'completed'
2387 Is converted to SQL C<key = val> statements:
2389 $stmt = "WHERE user = ? AND status = ?";
2390 @bind = ('nwiger', 'completed');
2392 One common thing I end up doing is having a list of values that
2393 a field can be in. To do this, simply specify a list inside of
2398 status => ['assigned', 'in-progress', 'pending'];
2401 This simple code will create the following:
2403 $stmt = "WHERE user = ? AND ( status = ? OR status = ? OR status = ? )";
2404 @bind = ('nwiger', 'assigned', 'in-progress', 'pending');
2406 A field associated to an empty arrayref will be considered a
2407 logical false and will generate 0=1.
2409 =head2 Tests for NULL values
2411 If the value part is C<undef> then this is converted to SQL <IS NULL>
2420 $stmt = "WHERE user = ? AND status IS NULL";
2423 To test if a column IS NOT NULL:
2427 status => { '!=', undef },
2430 =head2 Specific comparison operators
2432 If you want to specify a different type of operator for your comparison,
2433 you can use a hashref for a given column:
2437 status => { '!=', 'completed' }
2440 Which would generate:
2442 $stmt = "WHERE user = ? AND status != ?";
2443 @bind = ('nwiger', 'completed');
2445 To test against multiple values, just enclose the values in an arrayref:
2447 status => { '=', ['assigned', 'in-progress', 'pending'] };
2449 Which would give you:
2451 "WHERE status = ? OR status = ? OR status = ?"
2454 The hashref can also contain multiple pairs, in which case it is expanded
2455 into an C<AND> of its elements:
2459 status => { '!=', 'completed', -not_like => 'pending%' }
2462 # Or more dynamically, like from a form
2463 $where{user} = 'nwiger';
2464 $where{status}{'!='} = 'completed';
2465 $where{status}{'-not_like'} = 'pending%';
2467 # Both generate this
2468 $stmt = "WHERE user = ? AND status != ? AND status NOT LIKE ?";
2469 @bind = ('nwiger', 'completed', 'pending%');
2472 To get an OR instead, you can combine it with the arrayref idea:
2476 priority => [ { '=', 2 }, { '>', 5 } ]
2479 Which would generate:
2481 $stmt = "WHERE ( priority = ? OR priority > ? ) AND user = ?";
2482 @bind = ('2', '5', 'nwiger');
2484 If you want to include literal SQL (with or without bind values), just use a
2485 scalar reference or reference to an arrayref as the value:
2488 date_entered => { '>' => \["to_date(?, 'MM/DD/YYYY')", "11/26/2008"] },
2489 date_expires => { '<' => \"now()" }
2492 Which would generate:
2494 $stmt = "WHERE date_entered > to_date(?, 'MM/DD/YYYY') AND date_expires < now()";
2495 @bind = ('11/26/2008');
2498 =head2 Logic and nesting operators
2500 In the example above,
2501 there is a subtle trap if you want to say something like
2502 this (notice the C<AND>):
2504 WHERE priority != ? AND priority != ?
2506 Because, in Perl you I<can't> do this:
2508 priority => { '!=' => 2, '!=' => 1 }
2510 As the second C<!=> key will obliterate the first. The solution
2511 is to use the special C<-modifier> form inside an arrayref:
2513 priority => [ -and => {'!=', 2},
2517 Normally, these would be joined by C<OR>, but the modifier tells it
2518 to use C<AND> instead. (Hint: You can use this in conjunction with the
2519 C<logic> option to C<new()> in order to change the way your queries
2520 work by default.) B<Important:> Note that the C<-modifier> goes
2521 B<INSIDE> the arrayref, as an extra first element. This will
2522 B<NOT> do what you think it might:
2524 priority => -and => [{'!=', 2}, {'!=', 1}] # WRONG!
2526 Here is a quick list of equivalencies, since there is some overlap:
2529 status => {'!=', 'completed', 'not like', 'pending%' }
2530 status => [ -and => {'!=', 'completed'}, {'not like', 'pending%'}]
2533 status => {'=', ['assigned', 'in-progress']}
2534 status => [ -or => {'=', 'assigned'}, {'=', 'in-progress'}]
2535 status => [ {'=', 'assigned'}, {'=', 'in-progress'} ]
2539 =head2 Special operators: IN, BETWEEN, etc.
2541 You can also use the hashref format to compare a list of fields using the
2542 C<IN> comparison operator, by specifying the list as an arrayref:
2545 status => 'completed',
2546 reportid => { -in => [567, 2335, 2] }
2549 Which would generate:
2551 $stmt = "WHERE status = ? AND reportid IN (?,?,?)";
2552 @bind = ('completed', '567', '2335', '2');
2554 The reverse operator C<-not_in> generates SQL C<NOT IN> and is used in
2557 If the argument to C<-in> is an empty array, 'sqlfalse' is generated
2558 (by default: C<1=0>). Similarly, C<< -not_in => [] >> generates
2559 'sqltrue' (by default: C<1=1>).
2561 In addition to the array you can supply a chunk of literal sql or
2562 literal sql with bind:
2565 customer => { -in => \[
2566 'SELECT cust_id FROM cust WHERE balance > ?',
2569 status => { -in => \'SELECT status_codes FROM states' },
2575 customer IN ( SELECT cust_id FROM cust WHERE balance > ? )
2576 AND status IN ( SELECT status_codes FROM states )
2580 Finally, if the argument to C<-in> is not a reference, it will be
2581 treated as a single-element array.
2583 Another pair of operators is C<-between> and C<-not_between>,
2584 used with an arrayref of two values:
2588 completion_date => {
2589 -not_between => ['2002-10-01', '2003-02-06']
2595 WHERE user = ? AND completion_date NOT BETWEEN ( ? AND ? )
2597 Just like with C<-in> all plausible combinations of literal SQL
2601 start0 => { -between => [ 1, 2 ] },
2602 start1 => { -between => \["? AND ?", 1, 2] },
2603 start2 => { -between => \"lower(x) AND upper(y)" },
2604 start3 => { -between => [
2606 \["upper(?)", 'stuff' ],
2613 ( start0 BETWEEN ? AND ? )
2614 AND ( start1 BETWEEN ? AND ? )
2615 AND ( start2 BETWEEN lower(x) AND upper(y) )
2616 AND ( start3 BETWEEN lower(x) AND upper(?) )
2618 @bind = (1, 2, 1, 2, 'stuff');
2621 These are the two builtin "special operators"; but the
2622 list can be expanded: see section L</"SPECIAL OPERATORS"> below.
2624 =head2 Unary operators: bool
2626 If you wish to test against boolean columns or functions within your
2627 database you can use the C<-bool> and C<-not_bool> operators. For
2628 example to test the column C<is_user> being true and the column
2629 C<is_enabled> being false you would use:-
2633 -not_bool => 'is_enabled',
2638 WHERE is_user AND NOT is_enabled
2640 If a more complex combination is required, testing more conditions,
2641 then you should use the and/or operators:-
2646 -not_bool => { two=> { -rlike => 'bar' } },
2647 -not_bool => { three => [ { '=', 2 }, { '>', 5 } ] },
2658 (NOT ( three = ? OR three > ? ))
2661 =head2 Nested conditions, -and/-or prefixes
2663 So far, we've seen how multiple conditions are joined with a top-level
2664 C<AND>. We can change this by putting the different conditions we want in
2665 hashes and then putting those hashes in an array. For example:
2670 status => { -like => ['pending%', 'dispatched'] },
2674 status => 'unassigned',
2678 This data structure would create the following:
2680 $stmt = "WHERE ( user = ? AND ( status LIKE ? OR status LIKE ? ) )
2681 OR ( user = ? AND status = ? ) )";
2682 @bind = ('nwiger', 'pending', 'dispatched', 'robot', 'unassigned');
2685 Clauses in hashrefs or arrayrefs can be prefixed with an C<-and> or C<-or>
2686 to change the logic inside:
2692 -and => [ workhrs => {'>', 20}, geo => 'ASIA' ],
2693 -or => { workhrs => {'<', 50}, geo => 'EURO' },
2700 $stmt = "WHERE ( user = ?
2701 AND ( ( workhrs > ? AND geo = ? )
2702 OR ( workhrs < ? OR geo = ? ) ) )";
2703 @bind = ('nwiger', '20', 'ASIA', '50', 'EURO');
2705 =head3 Algebraic inconsistency, for historical reasons
2707 C<Important note>: when connecting several conditions, the C<-and->|C<-or>
2708 operator goes C<outside> of the nested structure; whereas when connecting
2709 several constraints on one column, the C<-and> operator goes
2710 C<inside> the arrayref. Here is an example combining both features:
2713 -and => [a => 1, b => 2],
2714 -or => [c => 3, d => 4],
2715 e => [-and => {-like => 'foo%'}, {-like => '%bar'} ]
2720 WHERE ( ( ( a = ? AND b = ? )
2721 OR ( c = ? OR d = ? )
2722 OR ( e LIKE ? AND e LIKE ? ) ) )
2724 This difference in syntax is unfortunate but must be preserved for
2725 historical reasons. So be careful: the two examples below would
2726 seem algebraically equivalent, but they are not
2729 { -like => 'foo%' },
2730 { -like => '%bar' },
2732 # yields: WHERE ( ( col LIKE ? AND col LIKE ? ) )
2735 { col => { -like => 'foo%' } },
2736 { col => { -like => '%bar' } },
2738 # yields: WHERE ( ( col LIKE ? OR col LIKE ? ) )
2741 =head2 Literal SQL and value type operators
2743 The basic premise of SQL::Abstract is that in WHERE specifications the "left
2744 side" is a column name and the "right side" is a value (normally rendered as
2745 a placeholder). This holds true for both hashrefs and arrayref pairs as you
2746 see in the L</WHERE CLAUSES> examples above. Sometimes it is necessary to
2747 alter this behavior. There are several ways of doing so.
2751 This is a virtual operator that signals the string to its right side is an
2752 identifier (a column name) and not a value. For example to compare two
2753 columns you would write:
2756 priority => { '<', 2 },
2757 requestor => { -ident => 'submitter' },
2762 $stmt = "WHERE priority < ? AND requestor = submitter";
2765 If you are maintaining legacy code you may see a different construct as
2766 described in L</Deprecated usage of Literal SQL>, please use C<-ident> in new
2771 This is a virtual operator that signals that the construct to its right side
2772 is a value to be passed to DBI. This is for example necessary when you want
2773 to write a where clause against an array (for RDBMS that support such
2774 datatypes). For example:
2777 array => { -value => [1, 2, 3] }
2782 $stmt = 'WHERE array = ?';
2783 @bind = ([1, 2, 3]);
2785 Note that if you were to simply say:
2791 the result would probably not be what you wanted:
2793 $stmt = 'WHERE array = ? OR array = ? OR array = ?';
2798 Finally, sometimes only literal SQL will do. To include a random snippet
2799 of SQL verbatim, you specify it as a scalar reference. Consider this only
2800 as a last resort. Usually there is a better way. For example:
2803 priority => { '<', 2 },
2804 requestor => { -in => \'(SELECT name FROM hitmen)' },
2809 $stmt = "WHERE priority < ? AND requestor IN (SELECT name FROM hitmen)"
2812 Note that in this example, you only get one bind parameter back, since
2813 the verbatim SQL is passed as part of the statement.
2817 Never use untrusted input as a literal SQL argument - this is a massive
2818 security risk (there is no way to check literal snippets for SQL
2819 injections and other nastyness). If you need to deal with untrusted input
2820 use literal SQL with placeholders as described next.
2822 =head3 Literal SQL with placeholders and bind values (subqueries)
2824 If the literal SQL to be inserted has placeholders and bind values,
2825 use a reference to an arrayref (yes this is a double reference --
2826 not so common, but perfectly legal Perl). For example, to find a date
2827 in Postgres you can use something like this:
2830 date_column => \[ "= date '2008-09-30' - ?::integer", 10 ]
2835 $stmt = "WHERE ( date_column = date '2008-09-30' - ?::integer )"
2838 Note that you must pass the bind values in the same format as they are returned
2839 by L<where|/where(\%where, $order)>. This means that if you set L</bindtype>
2840 to C<columns>, you must provide the bind values in the
2841 C<< [ column_meta => value ] >> format, where C<column_meta> is an opaque
2842 scalar value; most commonly the column name, but you can use any scalar value
2843 (including references and blessed references), L<SQL::Abstract> will simply
2844 pass it through intact. So if C<bindtype> is set to C<columns> the above
2845 example will look like:
2848 date_column => \[ "= date '2008-09-30' - ?::integer", [ {} => 10 ] ]
2851 Literal SQL is especially useful for nesting parenthesized clauses in the
2852 main SQL query. Here is a first example:
2854 my ($sub_stmt, @sub_bind) = ("SELECT c1 FROM t1 WHERE c2 < ? AND c3 LIKE ?",
2858 bar => \["IN ($sub_stmt)" => @sub_bind],
2863 $stmt = "WHERE (foo = ? AND bar IN (SELECT c1 FROM t1
2864 WHERE c2 < ? AND c3 LIKE ?))";
2865 @bind = (1234, 100, "foo%");
2867 Other subquery operators, like for example C<"E<gt> ALL"> or C<"NOT IN">,
2868 are expressed in the same way. Of course the C<$sub_stmt> and
2869 its associated bind values can be generated through a former call
2872 my ($sub_stmt, @sub_bind)
2873 = $sql->select("t1", "c1", {c2 => {"<" => 100},
2874 c3 => {-like => "foo%"}});
2877 bar => \["> ALL ($sub_stmt)" => @sub_bind],
2880 In the examples above, the subquery was used as an operator on a column;
2881 but the same principle also applies for a clause within the main C<%where>
2882 hash, like an EXISTS subquery:
2884 my ($sub_stmt, @sub_bind)
2885 = $sql->select("t1", "*", {c1 => 1, c2 => \"> t0.c0"});
2886 my %where = ( -and => [
2888 \["EXISTS ($sub_stmt)" => @sub_bind],
2893 $stmt = "WHERE (foo = ? AND EXISTS (SELECT * FROM t1
2894 WHERE c1 = ? AND c2 > t0.c0))";
2898 Observe that the condition on C<c2> in the subquery refers to
2899 column C<t0.c0> of the main query: this is I<not> a bind
2900 value, so we have to express it through a scalar ref.
2901 Writing C<< c2 => {">" => "t0.c0"} >> would have generated
2902 C<< c2 > ? >> with bind value C<"t0.c0"> ... not exactly
2903 what we wanted here.
2905 Finally, here is an example where a subquery is used
2906 for expressing unary negation:
2908 my ($sub_stmt, @sub_bind)
2909 = $sql->where({age => [{"<" => 10}, {">" => 20}]});
2910 $sub_stmt =~ s/^ where //i; # don't want "WHERE" in the subclause
2912 lname => {like => '%son%'},
2913 \["NOT ($sub_stmt)" => @sub_bind],
2918 $stmt = "lname LIKE ? AND NOT ( age < ? OR age > ? )"
2919 @bind = ('%son%', 10, 20)
2921 =head3 Deprecated usage of Literal SQL
2923 Below are some examples of archaic use of literal SQL. It is shown only as
2924 reference for those who deal with legacy code. Each example has a much
2925 better, cleaner and safer alternative that users should opt for in new code.
2931 my %where = ( requestor => \'IS NOT NULL' )
2933 $stmt = "WHERE requestor IS NOT NULL"
2935 This used to be the way of generating NULL comparisons, before the handling
2936 of C<undef> got formalized. For new code please use the superior syntax as
2937 described in L</Tests for NULL values>.
2941 my %where = ( requestor => \'= submitter' )
2943 $stmt = "WHERE requestor = submitter"
2945 This used to be the only way to compare columns. Use the superior L</-ident>
2946 method for all new code. For example an identifier declared in such a way
2947 will be properly quoted if L</quote_char> is properly set, while the legacy
2948 form will remain as supplied.
2952 my %where = ( is_ready => \"", completed => { '>', '2012-12-21' } )
2954 $stmt = "WHERE completed > ? AND is_ready"
2955 @bind = ('2012-12-21')
2957 Using an empty string literal used to be the only way to express a boolean.
2958 For all new code please use the much more readable
2959 L<-bool|/Unary operators: bool> operator.
2965 These pages could go on for a while, since the nesting of the data
2966 structures this module can handle are pretty much unlimited (the
2967 module implements the C<WHERE> expansion as a recursive function
2968 internally). Your best bet is to "play around" with the module a
2969 little to see how the data structures behave, and choose the best
2970 format for your data based on that.
2972 And of course, all the values above will probably be replaced with
2973 variables gotten from forms or the command line. After all, if you
2974 knew everything ahead of time, you wouldn't have to worry about
2975 dynamically-generating SQL and could just hardwire it into your
2978 =head1 ORDER BY CLAUSES
2980 Some functions take an order by clause. This can either be a scalar (just a
2981 column name), a hashref of C<< { -desc => 'col' } >> or C<< { -asc => 'col' }
2982 >>, a scalarref, an arrayref-ref, or an arrayref of any of the previous
2985 Given | Will Generate
2986 ---------------------------------------------------------------
2988 'colA' | ORDER BY colA
2990 [qw/colA colB/] | ORDER BY colA, colB
2992 {-asc => 'colA'} | ORDER BY colA ASC
2994 {-desc => 'colB'} | ORDER BY colB DESC
2996 ['colA', {-asc => 'colB'}] | ORDER BY colA, colB ASC
2998 { -asc => [qw/colA colB/] } | ORDER BY colA ASC, colB ASC
3000 \'colA DESC' | ORDER BY colA DESC
3002 \[ 'FUNC(colA, ?)', $x ] | ORDER BY FUNC(colA, ?)
3003 | /* ...with $x bound to ? */
3006 { -asc => 'colA' }, | colA ASC,
3007 { -desc => [qw/colB/] }, | colB DESC,
3008 { -asc => [qw/colC colD/] },| colC ASC, colD ASC,
3009 \'colE DESC', | colE DESC,
3010 \[ 'FUNC(colF, ?)', $x ], | FUNC(colF, ?)
3011 ] | /* ...with $x bound to ? */
3012 ===============================================================
3016 =head1 SPECIAL OPERATORS
3018 my $sqlmaker = SQL::Abstract->new(special_ops => [
3022 my ($self, $field, $op, $arg) = @_;
3028 handler => 'method_name',
3032 A "special operator" is a SQL syntactic clause that can be
3033 applied to a field, instead of a usual binary operator.
3036 WHERE field IN (?, ?, ?)
3037 WHERE field BETWEEN ? AND ?
3038 WHERE MATCH(field) AGAINST (?, ?)
3040 Special operators IN and BETWEEN are fairly standard and therefore
3041 are builtin within C<SQL::Abstract> (as the overridable methods
3042 C<_where_field_IN> and C<_where_field_BETWEEN>). For other operators,
3043 like the MATCH .. AGAINST example above which is specific to MySQL,
3044 you can write your own operator handlers - supply a C<special_ops>
3045 argument to the C<new> method. That argument takes an arrayref of
3046 operator definitions; each operator definition is a hashref with two
3053 the regular expression to match the operator
3057 Either a coderef or a plain scalar method name. In both cases
3058 the expected return is C<< ($sql, @bind) >>.
3060 When supplied with a method name, it is simply called on the
3061 L<SQL::Abstract> object as:
3063 $self->$method_name($field, $op, $arg)
3067 $field is the LHS of the operator
3068 $op is the part that matched the handler regex
3071 When supplied with a coderef, it is called as:
3073 $coderef->($self, $field, $op, $arg)
3078 For example, here is an implementation
3079 of the MATCH .. AGAINST syntax for MySQL
3081 my $sqlmaker = SQL::Abstract->new(special_ops => [
3083 # special op for MySql MATCH (field) AGAINST(word1, word2, ...)
3084 {regex => qr/^match$/i,
3086 my ($self, $field, $op, $arg) = @_;
3087 $arg = [$arg] if not ref $arg;
3088 my $label = $self->_quote($field);
3089 my ($placeholder) = $self->_convert('?');
3090 my $placeholders = join ", ", (($placeholder) x @$arg);
3091 my $sql = $self->_sqlcase('match') . " ($label) "
3092 . $self->_sqlcase('against') . " ($placeholders) ";
3093 my @bind = $self->_bindtype($field, @$arg);
3094 return ($sql, @bind);
3101 =head1 UNARY OPERATORS
3103 my $sqlmaker = SQL::Abstract->new(unary_ops => [
3107 my ($self, $op, $arg) = @_;
3113 handler => 'method_name',
3117 A "unary operator" is a SQL syntactic clause that can be
3118 applied to a field - the operator goes before the field
3120 You can write your own operator handlers - supply a C<unary_ops>
3121 argument to the C<new> method. That argument takes an arrayref of
3122 operator definitions; each operator definition is a hashref with two
3129 the regular expression to match the operator
3133 Either a coderef or a plain scalar method name. In both cases
3134 the expected return is C<< $sql >>.
3136 When supplied with a method name, it is simply called on the
3137 L<SQL::Abstract> object as:
3139 $self->$method_name($op, $arg)
3143 $op is the part that matched the handler regex
3144 $arg is the RHS or argument of the operator
3146 When supplied with a coderef, it is called as:
3148 $coderef->($self, $op, $arg)
3156 Thanks to some benchmarking by Mark Stosberg, it turns out that
3157 this module is many orders of magnitude faster than using C<DBIx::Abstract>.
3158 I must admit this wasn't an intentional design issue, but it's a
3159 byproduct of the fact that you get to control your C<DBI> handles
3162 To maximize performance, use a code snippet like the following:
3164 # prepare a statement handle using the first row
3165 # and then reuse it for the rest of the rows
3167 for my $href (@array_of_hashrefs) {
3168 $stmt ||= $sql->insert('table', $href);
3169 $sth ||= $dbh->prepare($stmt);
3170 $sth->execute($sql->values($href));
3173 The reason this works is because the keys in your C<$href> are sorted
3174 internally by B<SQL::Abstract>. Thus, as long as your data retains
3175 the same structure, you only have to generate the SQL the first time
3176 around. On subsequent queries, simply use the C<values> function provided
3177 by this module to return your values in the correct order.
3179 However this depends on the values having the same type - if, for
3180 example, the values of a where clause may either have values
3181 (resulting in sql of the form C<column = ?> with a single bind
3182 value), or alternatively the values might be C<undef> (resulting in
3183 sql of the form C<column IS NULL> with no bind value) then the
3184 caching technique suggested will not work.
3188 If you use my C<CGI::FormBuilder> module at all, you'll hopefully
3189 really like this part (I do, at least). Building up a complex query
3190 can be as simple as the following:
3197 use CGI::FormBuilder;
3200 my $form = CGI::FormBuilder->new(...);
3201 my $sql = SQL::Abstract->new;
3203 if ($form->submitted) {
3204 my $field = $form->field;
3205 my $id = delete $field->{id};
3206 my($stmt, @bind) = $sql->update('table', $field, {id => $id});
3209 Of course, you would still have to connect using C<DBI> to run the
3210 query, but the point is that if you make your form look like your
3211 table, the actual query script can be extremely simplistic.
3213 If you're B<REALLY> lazy (I am), check out C<HTML::QuickTable> for
3214 a fast interface to returning and formatting data. I frequently
3215 use these three modules together to write complex database query
3216 apps in under 50 lines.
3218 =head1 HOW TO CONTRIBUTE
3220 Contributions are always welcome, in all usable forms (we especially
3221 welcome documentation improvements). The delivery methods include git-
3222 or unified-diff formatted patches, GitHub pull requests, or plain bug
3223 reports either via RT or the Mailing list. Contributors are generally
3224 granted full access to the official repository after their first several
3225 patches pass successful review.
3227 This project is maintained in a git repository. The code and related tools are
3228 accessible at the following locations:
3232 =item * Official repo: L<git://git.shadowcat.co.uk/dbsrgits/SQL-Abstract.git>
3234 =item * Official gitweb: L<http://git.shadowcat.co.uk/gitweb/gitweb.cgi?p=dbsrgits/SQL-Abstract.git>
3236 =item * GitHub mirror: L<https://github.com/dbsrgits/sql-abstract>
3238 =item * Authorized committers: L<ssh://dbsrgits@git.shadowcat.co.uk/SQL-Abstract.git>
3244 Version 1.50 was a major internal refactoring of C<SQL::Abstract>.
3245 Great care has been taken to preserve the I<published> behavior
3246 documented in previous versions in the 1.* family; however,
3247 some features that were previously undocumented, or behaved
3248 differently from the documentation, had to be changed in order
3249 to clarify the semantics. Hence, client code that was relying
3250 on some dark areas of C<SQL::Abstract> v1.*
3251 B<might behave differently> in v1.50.
3253 The main changes are:
3259 support for literal SQL through the C<< \ [ $sql, @bind ] >> syntax.
3263 support for the { operator => \"..." } construct (to embed literal SQL)
3267 support for the { operator => \["...", @bind] } construct (to embed literal SQL with bind values)
3271 optional support for L<array datatypes|/"Inserting and Updating Arrays">
3275 defensive programming: check arguments
3279 fixed bug with global logic, which was previously implemented
3280 through global variables yielding side-effects. Prior versions would
3281 interpret C<< [ {cond1, cond2}, [cond3, cond4] ] >>
3282 as C<< "(cond1 AND cond2) OR (cond3 AND cond4)" >>.
3283 Now this is interpreted
3284 as C<< "(cond1 AND cond2) OR (cond3 OR cond4)" >>.
3289 fixed semantics of _bindtype on array args
3293 dropped the C<_anoncopy> of the %where tree. No longer necessary,
3294 we just avoid shifting arrays within that tree.
3298 dropped the C<_modlogic> function
3302 =head1 ACKNOWLEDGEMENTS
3304 There are a number of individuals that have really helped out with
3305 this module. Unfortunately, most of them submitted bugs via CPAN
3306 so I have no idea who they are! But the people I do know are:
3308 Ash Berlin (order_by hash term support)
3309 Matt Trout (DBIx::Class support)
3310 Mark Stosberg (benchmarking)
3311 Chas Owens (initial "IN" operator support)
3312 Philip Collins (per-field SQL functions)
3313 Eric Kolve (hashref "AND" support)
3314 Mike Fragassi (enhancements to "BETWEEN" and "LIKE")
3315 Dan Kubb (support for "quote_char" and "name_sep")
3316 Guillermo Roditi (patch to cleanup "IN" and "BETWEEN", fix and tests for _order_by)
3317 Laurent Dami (internal refactoring, extensible list of special operators, literal SQL)
3318 Norbert Buchmuller (support for literal SQL in hashpair, misc. fixes & tests)
3319 Peter Rabbitson (rewrite of SQLA::Test, misc. fixes & tests)
3320 Oliver Charles (support for "RETURNING" after "INSERT")
3326 L<DBIx::Class>, L<DBIx::Abstract>, L<CGI::FormBuilder>, L<HTML::QuickTable>.
3330 Copyright (c) 2001-2007 Nathan Wiger <nwiger@cpan.org>. All Rights Reserved.
3332 This module is actively maintained by Matt Trout <mst@shadowcatsystems.co.uk>
3334 For support, your best bet is to try the C<DBIx::Class> users mailing list.
3335 While not an official support venue, C<DBIx::Class> makes heavy use of
3336 C<SQL::Abstract>, and as such list members there are very familiar with
3337 how to create queries.
3341 This module is free software; you may copy this under the same
3342 terms as perl itself (either the GNU General Public License or
3343 the Artistic License)