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 $self->_expand_expr($v);
559 if (my ($rest) = $k =~/^-not[_ ](.*)$/) {
560 return $self->_expand_expr({ -not => { "-${rest}", $v } }, $logic);
564 return +{ $k => { $self->{cmp} => $v } };
566 if (ref($v) eq 'ARRAY') {
567 return $self->{sqlfalse} unless @$v;
568 $self->_debug("ARRAY($k) means distribute over elements");
570 $v->[0] =~ /^-((?:and|or))$/i
571 ? ($v = [ @{$v}[1..$#$v] ], $1)
572 : ($self->{logic} || 'or')
574 return +{ "-${this_logic}" => [ map $self->_expand_expr({ $k => $_ }, $this_logic), @$v ] };
576 if (my $literal = is_literal_value($v)) {
578 belch 'Hash-pairs consisting of an empty string with a literal are deprecated, and will be removed in 2.0: use -and => [ $literal ] instead';
581 my ($sql, @bind) = @$literal;
582 return \[ $self->_quote($k).' '.$sql, @bind ];
589 my ($self, $where, $logic) = @_;
591 my $where_exp = $self->_expand_expr($where, $logic);
593 # dispatch on appropriate method according to refkind of $where
594 my $method = $self->_METHOD_FOR_refkind("_where", $where_exp);
596 my ($sql, @bind) = $self->$method($where_exp, $logic);
598 # DBIx::Class used to call _recurse_where in scalar context
599 # something else might too...
601 return ($sql, @bind);
604 belch "Calling _recurse_where in scalar context is deprecated and will go away before 2.0";
611 #======================================================================
612 # WHERE: top-level ARRAYREF
613 #======================================================================
616 sub _where_ARRAYREF {
617 my ($self, $where, $logic) = @_;
619 $logic = uc($logic || $self->{logic});
620 $logic eq 'AND' or $logic eq 'OR' or puke "unknown logic: $logic";
622 my @clauses = @$where;
624 my (@sql_clauses, @all_bind);
625 # need to use while() so can shift() for pairs
627 my $el = shift @clauses;
629 $el = undef if (defined $el and ! length $el);
631 # switch according to kind of $el and get corresponding ($sql, @bind)
632 my ($sql, @bind) = $self->_SWITCH_refkind($el, {
634 # skip empty elements, otherwise get invalid trailing AND stuff
635 ARRAYREF => sub {$self->_recurse_where($el) if @$el},
639 $self->_assert_bindval_matches_bindtype(@b);
643 HASHREF => sub {$self->_recurse_where($el, 'and') if %$el},
645 SCALARREF => sub { ($$el); },
648 # top-level arrayref with scalars, recurse in pairs
649 $self->_recurse_where({$el => shift(@clauses)})
652 UNDEF => sub {puke "Supplying an empty left hand side argument is not supported in array-pairs" },
656 push @sql_clauses, $sql;
657 push @all_bind, @bind;
661 return $self->_join_sql_clauses($logic, \@sql_clauses, \@all_bind);
664 #======================================================================
665 # WHERE: top-level ARRAYREFREF
666 #======================================================================
668 sub _where_ARRAYREFREF {
669 my ($self, $where) = @_;
670 my ($sql, @bind) = @$$where;
671 $self->_assert_bindval_matches_bindtype(@bind);
672 return ($sql, @bind);
675 #======================================================================
676 # WHERE: top-level HASHREF
677 #======================================================================
680 my ($self, $where) = @_;
681 my (@sql_clauses, @all_bind);
683 for my $k (sort keys %$where) {
684 my $v = $where->{$k};
686 # ($k => $v) is either a special unary op or a regular hashpair
687 my ($sql, @bind) = do {
689 # put the operator in canonical form
691 $op = substr $op, 1; # remove initial dash
692 $op =~ s/^\s+|\s+$//g;# remove leading/trailing space
693 $op =~ s/\s+/ /g; # compress whitespace
695 # so that -not_foo works correctly
696 $op =~ s/^not_/NOT /i;
698 $self->_debug("Unary OP(-$op) within hashref, recursing...");
699 my ($s, @b) = $self->_where_unary_op($op, $v);
701 # top level vs nested
702 # we assume that handled unary ops will take care of their ()s
704 List::Util::first {$op =~ $_->{regex}} @{$self->{unary_ops}}
706 ( defined $self->{_nested_func_lhs} and $self->{_nested_func_lhs} eq $k )
712 if (is_literal_value ($v) ) {
713 belch 'Hash-pairs consisting of an empty string with a literal are deprecated, and will be removed in 2.0: use -and => [ $literal ] instead';
716 puke "Supplying an empty left hand side argument is not supported in hash-pairs";
720 my $method = $self->_METHOD_FOR_refkind("_where_hashpair", $v);
721 $self->$method($k, $v);
725 push @sql_clauses, $sql;
726 push @all_bind, @bind;
729 return $self->_join_sql_clauses('and', \@sql_clauses, \@all_bind);
732 sub _where_unary_op {
733 my ($self, $op, $rhs) = @_;
735 # top level special ops are illegal in general
736 # this includes the -ident/-value ops (dual purpose unary and special)
737 puke "Illegal use of top-level '-$op'"
738 if ! defined $self->{_nested_func_lhs} and List::Util::first { $op =~ $_->{regex} } @{$self->{special_ops}};
740 if (my $op_entry = List::Util::first { $op =~ $_->{regex} } @{$self->{unary_ops}}) {
741 my $handler = $op_entry->{handler};
743 if (not ref $handler) {
744 if ($op =~ s/ [_\s]? \d+ $//x ) {
745 belch 'Use of [and|or|nest]_N modifiers is deprecated and will be removed in SQLA v2.0. '
746 . "You probably wanted ...-and => [ -$op => COND1, -$op => COND2 ... ]";
748 return $self->$handler($op, $rhs);
750 elsif (ref $handler eq 'CODE') {
751 return $handler->($self, $op, $rhs);
754 puke "Illegal handler for operator $op - expecting a method name or a coderef";
758 $self->_debug("Generic unary OP: $op - recursing as function");
760 $self->_assert_pass_injection_guard($op);
762 my ($sql, @bind) = $self->_SWITCH_refkind($rhs, {
764 puke "Illegal use of top-level '-$op'"
765 unless defined $self->{_nested_func_lhs};
768 $self->_convert('?'),
769 $self->_bindtype($self->{_nested_func_lhs}, $rhs)
773 $self->_recurse_where($rhs)
777 $sql = sprintf('%s %s',
778 $self->_sqlcase($op),
782 return ($sql, @bind);
785 sub _where_op_ANDOR {
786 my ($self, $op, $v) = @_;
788 $self->_SWITCH_refkind($v, {
790 return $self->_where_ARRAYREF($v, $op);
794 return ($op =~ /^or/i)
795 ? $self->_where_ARRAYREF([ map { $_ => $v->{$_} } (sort keys %$v) ], $op)
796 : $self->_where_HASHREF($v);
800 puke "-$op => \\\$scalar makes little sense, use " .
802 ? '[ \$scalar, \%rest_of_conditions ] instead'
803 : '-and => [ \$scalar, \%rest_of_conditions ] instead'
808 puke "-$op => \\[...] makes little sense, use " .
810 ? '[ \[...], \%rest_of_conditions ] instead'
811 : '-and => [ \[...], \%rest_of_conditions ] instead'
815 SCALAR => sub { # permissively interpreted as SQL
816 puke "-$op => \$value makes little sense, use -bool => \$value instead";
820 puke "-$op => undef not supported";
826 my ($self, $op, $v) = @_;
828 $self->_SWITCH_refkind($v, {
830 SCALAR => sub { # permissively interpreted as SQL
831 belch "literal SQL should be -nest => \\'scalar' "
832 . "instead of -nest => 'scalar' ";
837 puke "-$op => undef not supported";
841 $self->_recurse_where($v);
849 my ($self, $op, $v) = @_;
851 my ($s, @b) = $self->_SWITCH_refkind($v, {
852 SCALAR => sub { # interpreted as SQL column
853 $self->_convert($self->_quote($v));
857 puke "-$op => undef not supported";
861 $self->_recurse_where($v);
865 $s = "(NOT $s)" if $op =~ /^not/i;
870 sub _where_op_IDENT {
872 my ($op, $rhs) = splice @_, -2;
873 if (! defined $rhs or length ref $rhs) {
874 puke "-$op requires a single plain scalar argument (a quotable identifier)";
877 # in case we are called as a top level special op (no '=')
880 $_ = $self->_convert($self->_quote($_)) for ($lhs, $rhs);
888 sub _where_op_VALUE {
890 my ($op, $rhs) = splice @_, -2;
892 # in case we are called as a top level special op (no '=')
896 if (! defined $rhs) {
898 ? $self->_where_hashpair_HASHREF($lhs, { -is => undef })
905 (defined $lhs ? $lhs : $self->{_nested_func_lhs}),
912 $self->_convert($self->_quote($lhs)) . ' = ' . $self->_convert('?'),
916 $self->_convert('?'),
922 sub _where_hashpair_ARRAYREF {
923 my ($self, $k, $v) = @_;
926 my @v = @$v; # need copy because of shift below
927 $self->_debug("ARRAY($k) means distribute over elements");
929 # put apart first element if it is an operator (-and, -or)
931 (defined $v[0] && $v[0] =~ /^ - (?: AND|OR ) $/ix)
935 my @distributed = map { {$k => $_} } @v;
938 $self->_debug("OP($op) reinjected into the distributed array");
939 unshift @distributed, $op;
942 my $logic = $op ? substr($op, 1) : '';
944 return $self->_recurse_where(\@distributed, $logic);
947 $self->_debug("empty ARRAY($k) means 0=1");
948 return ($self->{sqlfalse});
952 sub _where_hashpair_HASHREF {
953 my ($self, $k, $v, $logic) = @_;
956 local $self->{_nested_func_lhs} = defined $self->{_nested_func_lhs}
957 ? $self->{_nested_func_lhs}
961 my ($all_sql, @all_bind);
963 for my $orig_op (sort keys %$v) {
964 my $val = $v->{$orig_op};
966 # put the operator in canonical form
969 # FIXME - we need to phase out dash-less ops
970 $op =~ s/^-//; # remove possible initial dash
971 $op =~ s/^\s+|\s+$//g;# remove leading/trailing space
972 $op =~ s/\s+/ /g; # compress whitespace
974 $self->_assert_pass_injection_guard($op);
977 $op =~ s/^is_not/IS NOT/i;
979 # so that -not_foo works correctly
980 $op =~ s/^not_/NOT /i;
982 # another retarded special case: foo => { $op => { -value => undef } }
983 if (ref $val eq 'HASH' and keys %$val == 1 and exists $val->{-value} and ! defined $val->{-value} ) {
989 # CASE: col-value logic modifiers
990 if ($orig_op =~ /^ \- (and|or) $/xi) {
991 ($sql, @bind) = $self->_where_hashpair_HASHREF($k, $val, $1);
993 # CASE: special operators like -in or -between
994 elsif (my $special_op = List::Util::first { $op =~ $_->{regex} } @{$self->{special_ops}}) {
995 my $handler = $special_op->{handler};
997 puke "No handler supplied for special operator $orig_op";
999 elsif (not ref $handler) {
1000 ($sql, @bind) = $self->$handler($k, $op, $val);
1002 elsif (ref $handler eq 'CODE') {
1003 ($sql, @bind) = $handler->($self, $k, $op, $val);
1006 puke "Illegal handler for special operator $orig_op - expecting a method name or a coderef";
1010 $self->_SWITCH_refkind($val, {
1012 ARRAYREF => sub { # CASE: col => {op => \@vals}
1013 ($sql, @bind) = $self->_where_field_op_ARRAYREF($k, $op, $val);
1016 ARRAYREFREF => sub { # CASE: col => {op => \[$sql, @bind]} (literal SQL with bind)
1017 my ($sub_sql, @sub_bind) = @$$val;
1018 $self->_assert_bindval_matches_bindtype(@sub_bind);
1019 $sql = join ' ', $self->_convert($self->_quote($k)),
1020 $self->_sqlcase($op),
1025 UNDEF => sub { # CASE: col => {op => undef} : sql "IS (NOT)? NULL"
1027 $op =~ /^not$/i ? 'is not' # legacy
1028 : $op =~ $self->{equality_op} ? 'is'
1029 : $op =~ $self->{like_op} ? belch("Supplying an undefined argument to '@{[ uc $op]}' is deprecated") && 'is'
1030 : $op =~ $self->{inequality_op} ? 'is not'
1031 : $op =~ $self->{not_like_op} ? belch("Supplying an undefined argument to '@{[ uc $op]}' is deprecated") && 'is not'
1032 : puke "unexpected operator '$orig_op' with undef operand";
1034 $sql = $self->_quote($k) . $self->_sqlcase(" $is null");
1037 FALLBACK => sub { # CASE: col => {op/func => $stuff}
1038 ($sql, @bind) = $self->_where_unary_op($op, $val);
1041 $self->_convert($self->_quote($k)),
1042 $self->{_nested_func_lhs} eq $k ? $sql : "($sql)", # top level vs nested
1048 ($all_sql) = (defined $all_sql and $all_sql) ? $self->_join_sql_clauses($logic, [$all_sql, $sql], []) : $sql;
1049 push @all_bind, @bind;
1051 return ($all_sql, @all_bind);
1054 sub _where_field_IS {
1055 my ($self, $k, $op, $v) = @_;
1057 my ($s) = $self->_SWITCH_refkind($v, {
1060 $self->_convert($self->_quote($k)),
1061 map { $self->_sqlcase($_)} ($op, 'null')
1064 puke "$op can only take undef as argument";
1071 sub _where_field_op_ARRAYREF {
1072 my ($self, $k, $op, $vals) = @_;
1074 my @vals = @$vals; #always work on a copy
1077 $self->_debug(sprintf '%s means multiple elements: [ %s ]',
1079 join(', ', map { defined $_ ? "'$_'" : 'NULL' } @vals ),
1082 # see if the first element is an -and/-or op
1084 if (defined $vals[0] && $vals[0] =~ /^ - (AND|OR) $/ix) {
1089 # a long standing API wart - an attempt to change this behavior during
1090 # the 1.50 series failed *spectacularly*. Warn instead and leave the
1095 (!$logic or $logic eq 'OR')
1097 ($op =~ $self->{inequality_op} or $op =~ $self->{not_like_op})
1100 belch "A multi-element arrayref as an argument to the inequality op '$o' "
1101 . 'is technically equivalent to an always-true 1=1 (you probably wanted '
1102 . "to say ...{ \$inequality_op => [ -and => \@values ] }... instead)"
1106 # distribute $op over each remaining member of @vals, append logic if exists
1107 return $self->_recurse_where([map { {$k => {$op, $_}} } @vals], $logic);
1111 # try to DWIM on equality operators
1113 $op =~ $self->{equality_op} ? $self->{sqlfalse}
1114 : $op =~ $self->{like_op} ? belch("Supplying an empty arrayref to '@{[ uc $op]}' is deprecated") && $self->{sqlfalse}
1115 : $op =~ $self->{inequality_op} ? $self->{sqltrue}
1116 : $op =~ $self->{not_like_op} ? belch("Supplying an empty arrayref to '@{[ uc $op]}' is deprecated") && $self->{sqltrue}
1117 : puke "operator '$op' applied on an empty array (field '$k')";
1122 sub _where_hashpair_SCALARREF {
1123 my ($self, $k, $v) = @_;
1124 $self->_debug("SCALAR($k) means literal SQL: $$v");
1125 my $sql = $self->_quote($k) . " " . $$v;
1129 # literal SQL with bind
1130 sub _where_hashpair_ARRAYREFREF {
1131 my ($self, $k, $v) = @_;
1132 $self->_debug("REF($k) means literal SQL: @${$v}");
1133 my ($sql, @bind) = @$$v;
1134 $self->_assert_bindval_matches_bindtype(@bind);
1135 $sql = $self->_quote($k) . " " . $sql;
1136 return ($sql, @bind );
1139 # literal SQL without bind
1140 sub _where_hashpair_SCALAR {
1141 my ($self, $k, $v) = @_;
1142 $self->_debug("NOREF($k) means simple key=val: $k $self->{cmp} $v");
1143 return ($self->_where_hashpair_HASHREF($k, { $self->{cmp} => $v }));
1147 sub _where_hashpair_UNDEF {
1148 my ($self, $k, $v) = @_;
1149 $self->_debug("UNDEF($k) means IS NULL");
1150 return $self->_where_hashpair_HASHREF($k, { -is => undef });
1153 #======================================================================
1154 # WHERE: TOP-LEVEL OTHERS (SCALARREF, SCALAR, UNDEF)
1155 #======================================================================
1158 sub _where_SCALARREF {
1159 my ($self, $where) = @_;
1162 $self->_debug("SCALAR(*top) means literal SQL: $$where");
1168 my ($self, $where) = @_;
1171 $self->_debug("NOREF(*top) means literal SQL: $where");
1182 #======================================================================
1183 # WHERE: BUILTIN SPECIAL OPERATORS (-in, -between)
1184 #======================================================================
1187 sub _where_field_BETWEEN {
1188 my ($self, $k, $op, $vals) = @_;
1190 my ($label, $and, $placeholder);
1191 $label = $self->_convert($self->_quote($k));
1192 $and = ' ' . $self->_sqlcase('and') . ' ';
1193 $placeholder = $self->_convert('?');
1194 $op = $self->_sqlcase($op);
1196 my $invalid_args = "Operator '$op' requires either an arrayref with two defined values or expressions, or a single literal scalarref/arrayref-ref";
1198 my ($clause, @bind) = $self->_SWITCH_refkind($vals, {
1199 ARRAYREFREF => sub {
1200 my ($s, @b) = @$$vals;
1201 $self->_assert_bindval_matches_bindtype(@b);
1208 puke $invalid_args if @$vals != 2;
1210 my (@all_sql, @all_bind);
1211 foreach my $val (@$vals) {
1212 my ($sql, @bind) = $self->_SWITCH_refkind($val, {
1214 return ($placeholder, $self->_bindtype($k, $val) );
1219 ARRAYREFREF => sub {
1220 my ($sql, @bind) = @$$val;
1221 $self->_assert_bindval_matches_bindtype(@bind);
1222 return ($sql, @bind);
1225 my ($func, $arg, @rest) = %$val;
1226 puke "Only simple { -func => arg } functions accepted as sub-arguments to BETWEEN"
1227 if (@rest or $func !~ /^ \- (.+)/x);
1228 $self->_where_unary_op($1 => $arg);
1234 push @all_sql, $sql;
1235 push @all_bind, @bind;
1239 (join $and, @all_sql),
1248 my $sql = "( $label $op $clause )";
1249 return ($sql, @bind)
1253 sub _where_field_IN {
1254 my ($self, $k, $op, $vals) = @_;
1256 # backwards compatibility: if scalar, force into an arrayref
1257 $vals = [$vals] if defined $vals && ! ref $vals;
1259 my ($label) = $self->_convert($self->_quote($k));
1260 my ($placeholder) = $self->_convert('?');
1261 $op = $self->_sqlcase($op);
1263 my ($sql, @bind) = $self->_SWITCH_refkind($vals, {
1264 ARRAYREF => sub { # list of choices
1265 if (@$vals) { # nonempty list
1266 my (@all_sql, @all_bind);
1268 for my $val (@$vals) {
1269 my ($sql, @bind) = $self->_SWITCH_refkind($val, {
1271 return ($placeholder, $val);
1276 ARRAYREFREF => sub {
1277 my ($sql, @bind) = @$$val;
1278 $self->_assert_bindval_matches_bindtype(@bind);
1279 return ($sql, @bind);
1282 my ($func, $arg, @rest) = %$val;
1283 puke "Only simple { -func => arg } functions accepted as sub-arguments to IN"
1284 if (@rest or $func !~ /^ \- (.+)/x);
1285 $self->_where_unary_op($1 => $arg);
1289 'SQL::Abstract before v1.75 used to generate incorrect SQL when the '
1290 . "-$op operator was given an undef-containing list: !!!AUDIT YOUR CODE "
1291 . 'AND DATA!!! (the upcoming Data::Query-based version of SQL::Abstract '
1292 . 'will emit the logically correct SQL instead of raising this exception)'
1296 push @all_sql, $sql;
1297 push @all_bind, @bind;
1301 sprintf('%s %s ( %s )',
1304 join(', ', @all_sql)
1306 $self->_bindtype($k, @all_bind),
1309 else { # empty list: some databases won't understand "IN ()", so DWIM
1310 my $sql = ($op =~ /\bnot\b/i) ? $self->{sqltrue} : $self->{sqlfalse};
1315 SCALARREF => sub { # literal SQL
1316 my $sql = $self->_open_outer_paren($$vals);
1317 return ("$label $op ( $sql )");
1319 ARRAYREFREF => sub { # literal SQL with bind
1320 my ($sql, @bind) = @$$vals;
1321 $self->_assert_bindval_matches_bindtype(@bind);
1322 $sql = $self->_open_outer_paren($sql);
1323 return ("$label $op ( $sql )", @bind);
1327 puke "Argument passed to the '$op' operator can not be undefined";
1331 puke "special op $op requires an arrayref (or scalarref/arrayref-ref)";
1335 return ($sql, @bind);
1338 # Some databases (SQLite) treat col IN (1, 2) different from
1339 # col IN ( (1, 2) ). Use this to strip all outer parens while
1340 # adding them back in the corresponding method
1341 sub _open_outer_paren {
1342 my ($self, $sql) = @_;
1344 while (my ($inner) = $sql =~ /^ \s* \( (.*) \) \s* $/xs) {
1346 # there are closing parens inside, need the heavy duty machinery
1347 # to reevaluate the extraction starting from $sql (full reevaluation)
1348 if ($inner =~ /\)/) {
1349 require Text::Balanced;
1351 my (undef, $remainder) = do {
1352 # idiotic design - writes to $@ but *DOES NOT* throw exceptions
1354 Text::Balanced::extract_bracketed($sql, '()', qr/\s*/);
1357 # the entire expression needs to be a balanced bracketed thing
1358 # (after an extract no remainder sans trailing space)
1359 last if defined $remainder and $remainder =~ /\S/;
1369 #======================================================================
1371 #======================================================================
1374 my ($self, $arg) = @_;
1377 for my $c ($self->_order_by_chunks($arg) ) {
1378 $self->_SWITCH_refkind($c, {
1379 SCALAR => sub { push @sql, $c },
1380 ARRAYREF => sub { push @sql, shift @$c; push @bind, @$c },
1386 $self->_sqlcase(' order by'),
1392 return wantarray ? ($sql, @bind) : $sql;
1395 sub _order_by_chunks {
1396 my ($self, $arg) = @_;
1398 return $self->_SWITCH_refkind($arg, {
1401 map { $self->_order_by_chunks($_ ) } @$arg;
1404 ARRAYREFREF => sub {
1405 my ($s, @b) = @$$arg;
1406 $self->_assert_bindval_matches_bindtype(@b);
1410 SCALAR => sub {$self->_quote($arg)},
1412 UNDEF => sub {return () },
1414 SCALARREF => sub {$$arg}, # literal SQL, no quoting
1417 # get first pair in hash
1418 my ($key, $val, @rest) = %$arg;
1420 return () unless $key;
1422 if (@rest or not $key =~ /^-(desc|asc)/i) {
1423 puke "hash passed to _order_by must have exactly one key (-desc or -asc)";
1429 for my $c ($self->_order_by_chunks($val)) {
1432 $self->_SWITCH_refkind($c, {
1437 ($sql, @bind) = @$c;
1441 $sql = $sql . ' ' . $self->_sqlcase($direction);
1443 push @ret, [ $sql, @bind];
1452 #======================================================================
1453 # DATASOURCE (FOR NOW, JUST PLAIN TABLE OR LIST OF TABLES)
1454 #======================================================================
1459 $self->_SWITCH_refkind($from, {
1460 ARRAYREF => sub {join ', ', map { $self->_quote($_) } @$from;},
1461 SCALAR => sub {$self->_quote($from)},
1462 SCALARREF => sub {$$from},
1467 #======================================================================
1469 #======================================================================
1471 # highly optimized, as it's called way too often
1473 # my ($self, $label) = @_;
1475 return '' unless defined $_[1];
1476 return ${$_[1]} if ref($_[1]) eq 'SCALAR';
1478 $_[0]->{quote_char} or
1479 ($_[0]->_assert_pass_injection_guard($_[1]), return $_[1]);
1481 my $qref = ref $_[0]->{quote_char};
1483 !$qref ? ($_[0]->{quote_char}, $_[0]->{quote_char})
1484 : ($qref eq 'ARRAY') ? @{$_[0]->{quote_char}}
1485 : puke "Unsupported quote_char format: $_[0]->{quote_char}";
1487 my $esc = $_[0]->{escape_char} || $r;
1489 # parts containing * are naturally unquoted
1490 return join($_[0]->{name_sep}||'', map
1491 +( $_ eq '*' ? $_ : do { (my $n = $_) =~ s/(\Q$esc\E|\Q$r\E)/$esc$1/g; $l . $n . $r } ),
1492 ( $_[0]->{name_sep} ? split (/\Q$_[0]->{name_sep}\E/, $_[1] ) : $_[1] )
1497 # Conversion, if applicable
1499 #my ($self, $arg) = @_;
1500 if ($_[0]->{convert}) {
1501 return $_[0]->_sqlcase($_[0]->{convert}) .'(' . $_[1] . ')';
1508 #my ($self, $col, @vals) = @_;
1509 # called often - tighten code
1510 return $_[0]->{bindtype} eq 'columns'
1511 ? map {[$_[1], $_]} @_[2 .. $#_]
1516 # Dies if any element of @bind is not in [colname => value] format
1517 # if bindtype is 'columns'.
1518 sub _assert_bindval_matches_bindtype {
1519 # my ($self, @bind) = @_;
1521 if ($self->{bindtype} eq 'columns') {
1523 if (!defined $_ || ref($_) ne 'ARRAY' || @$_ != 2) {
1524 puke "bindtype 'columns' selected, you need to pass: [column_name => bind_value]"
1530 sub _join_sql_clauses {
1531 my ($self, $logic, $clauses_aref, $bind_aref) = @_;
1533 if (@$clauses_aref > 1) {
1534 my $join = " " . $self->_sqlcase($logic) . " ";
1535 my $sql = '( ' . join($join, @$clauses_aref) . ' )';
1536 return ($sql, @$bind_aref);
1538 elsif (@$clauses_aref) {
1539 return ($clauses_aref->[0], @$bind_aref); # no parentheses
1542 return (); # if no SQL, ignore @$bind_aref
1547 # Fix SQL case, if so requested
1549 # LDNOTE: if $self->{case} is true, then it contains 'lower', so we
1550 # don't touch the argument ... crooked logic, but let's not change it!
1551 return $_[0]->{case} ? $_[1] : uc($_[1]);
1555 #======================================================================
1556 # DISPATCHING FROM REFKIND
1557 #======================================================================
1560 my ($self, $data) = @_;
1562 return 'UNDEF' unless defined $data;
1564 # blessed objects are treated like scalars
1565 my $ref = (Scalar::Util::blessed $data) ? '' : ref $data;
1567 return 'SCALAR' unless $ref;
1570 while ($ref eq 'REF') {
1572 $ref = (Scalar::Util::blessed $data) ? '' : ref $data;
1576 return ($ref||'SCALAR') . ('REF' x $n_steps);
1580 my ($self, $data) = @_;
1581 my @try = ($self->_refkind($data));
1582 push @try, 'SCALAR_or_UNDEF' if $try[0] eq 'SCALAR' || $try[0] eq 'UNDEF';
1583 push @try, 'FALLBACK';
1587 sub _METHOD_FOR_refkind {
1588 my ($self, $meth_prefix, $data) = @_;
1591 for (@{$self->_try_refkind($data)}) {
1592 $method = $self->can($meth_prefix."_".$_)
1596 return $method || puke "cannot dispatch on '$meth_prefix' for ".$self->_refkind($data);
1600 sub _SWITCH_refkind {
1601 my ($self, $data, $dispatch_table) = @_;
1604 for (@{$self->_try_refkind($data)}) {
1605 $coderef = $dispatch_table->{$_}
1609 puke "no dispatch entry for ".$self->_refkind($data)
1618 #======================================================================
1619 # VALUES, GENERATE, AUTOLOAD
1620 #======================================================================
1622 # LDNOTE: original code from nwiger, didn't touch code in that section
1623 # I feel the AUTOLOAD stuff should not be the default, it should
1624 # only be activated on explicit demand by user.
1628 my $data = shift || return;
1629 puke "Argument to ", __PACKAGE__, "->values must be a \\%hash"
1630 unless ref $data eq 'HASH';
1633 foreach my $k (sort keys %$data) {
1634 my $v = $data->{$k};
1635 $self->_SWITCH_refkind($v, {
1637 if ($self->{array_datatypes}) { # array datatype
1638 push @all_bind, $self->_bindtype($k, $v);
1640 else { # literal SQL with bind
1641 my ($sql, @bind) = @$v;
1642 $self->_assert_bindval_matches_bindtype(@bind);
1643 push @all_bind, @bind;
1646 ARRAYREFREF => sub { # literal SQL with bind
1647 my ($sql, @bind) = @${$v};
1648 $self->_assert_bindval_matches_bindtype(@bind);
1649 push @all_bind, @bind;
1651 SCALARREF => sub { # literal SQL without bind
1653 SCALAR_or_UNDEF => sub {
1654 push @all_bind, $self->_bindtype($k, $v);
1665 my(@sql, @sqlq, @sqlv);
1669 if ($ref eq 'HASH') {
1670 for my $k (sort keys %$_) {
1673 my $label = $self->_quote($k);
1674 if ($r eq 'ARRAY') {
1675 # literal SQL with bind
1676 my ($sql, @bind) = @$v;
1677 $self->_assert_bindval_matches_bindtype(@bind);
1678 push @sqlq, "$label = $sql";
1680 } elsif ($r eq 'SCALAR') {
1681 # literal SQL without bind
1682 push @sqlq, "$label = $$v";
1684 push @sqlq, "$label = ?";
1685 push @sqlv, $self->_bindtype($k, $v);
1688 push @sql, $self->_sqlcase('set'), join ', ', @sqlq;
1689 } elsif ($ref eq 'ARRAY') {
1690 # unlike insert(), assume these are ONLY the column names, i.e. for SQL
1693 if ($r eq 'ARRAY') { # literal SQL with bind
1694 my ($sql, @bind) = @$v;
1695 $self->_assert_bindval_matches_bindtype(@bind);
1698 } elsif ($r eq 'SCALAR') { # literal SQL without bind
1699 # embedded literal SQL
1706 push @sql, '(' . join(', ', @sqlq) . ')';
1707 } elsif ($ref eq 'SCALAR') {
1711 # strings get case twiddled
1712 push @sql, $self->_sqlcase($_);
1716 my $sql = join ' ', @sql;
1718 # this is pretty tricky
1719 # if ask for an array, return ($stmt, @bind)
1720 # otherwise, s/?/shift @sqlv/ to put it inline
1722 return ($sql, @sqlv);
1724 1 while $sql =~ s/\?/my $d = shift(@sqlv);
1725 ref $d ? $d->[1] : $d/e;
1734 # This allows us to check for a local, then _form, attr
1736 my($name) = $AUTOLOAD =~ /.*::(.+)/;
1737 return $self->generate($name, @_);
1748 SQL::Abstract - Generate SQL from Perl data structures
1754 my $sql = SQL::Abstract->new;
1756 my($stmt, @bind) = $sql->select($source, \@fields, \%where, $order);
1758 my($stmt, @bind) = $sql->insert($table, \%fieldvals || \@values);
1760 my($stmt, @bind) = $sql->update($table, \%fieldvals, \%where);
1762 my($stmt, @bind) = $sql->delete($table, \%where);
1764 # Then, use these in your DBI statements
1765 my $sth = $dbh->prepare($stmt);
1766 $sth->execute(@bind);
1768 # Just generate the WHERE clause
1769 my($stmt, @bind) = $sql->where(\%where, $order);
1771 # Return values in the same order, for hashed queries
1772 # See PERFORMANCE section for more details
1773 my @bind = $sql->values(\%fieldvals);
1777 This module was inspired by the excellent L<DBIx::Abstract>.
1778 However, in using that module I found that what I really wanted
1779 to do was generate SQL, but still retain complete control over my
1780 statement handles and use the DBI interface. So, I set out to
1781 create an abstract SQL generation module.
1783 While based on the concepts used by L<DBIx::Abstract>, there are
1784 several important differences, especially when it comes to WHERE
1785 clauses. I have modified the concepts used to make the SQL easier
1786 to generate from Perl data structures and, IMO, more intuitive.
1787 The underlying idea is for this module to do what you mean, based
1788 on the data structures you provide it. The big advantage is that
1789 you don't have to modify your code every time your data changes,
1790 as this module figures it out.
1792 To begin with, an SQL INSERT is as easy as just specifying a hash
1793 of C<key=value> pairs:
1796 name => 'Jimbo Bobson',
1797 phone => '123-456-7890',
1798 address => '42 Sister Lane',
1799 city => 'St. Louis',
1800 state => 'Louisiana',
1803 The SQL can then be generated with this:
1805 my($stmt, @bind) = $sql->insert('people', \%data);
1807 Which would give you something like this:
1809 $stmt = "INSERT INTO people
1810 (address, city, name, phone, state)
1811 VALUES (?, ?, ?, ?, ?)";
1812 @bind = ('42 Sister Lane', 'St. Louis', 'Jimbo Bobson',
1813 '123-456-7890', 'Louisiana');
1815 These are then used directly in your DBI code:
1817 my $sth = $dbh->prepare($stmt);
1818 $sth->execute(@bind);
1820 =head2 Inserting and Updating Arrays
1822 If your database has array types (like for example Postgres),
1823 activate the special option C<< array_datatypes => 1 >>
1824 when creating the C<SQL::Abstract> object.
1825 Then you may use an arrayref to insert and update database array types:
1827 my $sql = SQL::Abstract->new(array_datatypes => 1);
1829 planets => [qw/Mercury Venus Earth Mars/]
1832 my($stmt, @bind) = $sql->insert('solar_system', \%data);
1836 $stmt = "INSERT INTO solar_system (planets) VALUES (?)"
1838 @bind = (['Mercury', 'Venus', 'Earth', 'Mars']);
1841 =head2 Inserting and Updating SQL
1843 In order to apply SQL functions to elements of your C<%data> you may
1844 specify a reference to an arrayref for the given hash value. For example,
1845 if you need to execute the Oracle C<to_date> function on a value, you can
1846 say something like this:
1850 date_entered => \[ "to_date(?,'MM/DD/YYYY')", "03/02/2003" ],
1853 The first value in the array is the actual SQL. Any other values are
1854 optional and would be included in the bind values array. This gives
1857 my($stmt, @bind) = $sql->insert('people', \%data);
1859 $stmt = "INSERT INTO people (name, date_entered)
1860 VALUES (?, to_date(?,'MM/DD/YYYY'))";
1861 @bind = ('Bill', '03/02/2003');
1863 An UPDATE is just as easy, all you change is the name of the function:
1865 my($stmt, @bind) = $sql->update('people', \%data);
1867 Notice that your C<%data> isn't touched; the module will generate
1868 the appropriately quirky SQL for you automatically. Usually you'll
1869 want to specify a WHERE clause for your UPDATE, though, which is
1870 where handling C<%where> hashes comes in handy...
1872 =head2 Complex where statements
1874 This module can generate pretty complicated WHERE statements
1875 easily. For example, simple C<key=value> pairs are taken to mean
1876 equality, and if you want to see if a field is within a set
1877 of values, you can use an arrayref. Let's say we wanted to
1878 SELECT some data based on this criteria:
1881 requestor => 'inna',
1882 worker => ['nwiger', 'rcwe', 'sfz'],
1883 status => { '!=', 'completed' }
1886 my($stmt, @bind) = $sql->select('tickets', '*', \%where);
1888 The above would give you something like this:
1890 $stmt = "SELECT * FROM tickets WHERE
1891 ( requestor = ? ) AND ( status != ? )
1892 AND ( worker = ? OR worker = ? OR worker = ? )";
1893 @bind = ('inna', 'completed', 'nwiger', 'rcwe', 'sfz');
1895 Which you could then use in DBI code like so:
1897 my $sth = $dbh->prepare($stmt);
1898 $sth->execute(@bind);
1904 The methods are simple. There's one for every major SQL operation,
1905 and a constructor you use first. The arguments are specified in a
1906 similar order for each method (table, then fields, then a where
1907 clause) to try and simplify things.
1909 =head2 new(option => 'value')
1911 The C<new()> function takes a list of options and values, and returns
1912 a new B<SQL::Abstract> object which can then be used to generate SQL
1913 through the methods below. The options accepted are:
1919 If set to 'lower', then SQL will be generated in all lowercase. By
1920 default SQL is generated in "textbook" case meaning something like:
1922 SELECT a_field FROM a_table WHERE some_field LIKE '%someval%'
1924 Any setting other than 'lower' is ignored.
1928 This determines what the default comparison operator is. By default
1929 it is C<=>, meaning that a hash like this:
1931 %where = (name => 'nwiger', email => 'nate@wiger.org');
1933 Will generate SQL like this:
1935 WHERE name = 'nwiger' AND email = 'nate@wiger.org'
1937 However, you may want loose comparisons by default, so if you set
1938 C<cmp> to C<like> you would get SQL such as:
1940 WHERE name like 'nwiger' AND email like 'nate@wiger.org'
1942 You can also override the comparison on an individual basis - see
1943 the huge section on L</"WHERE CLAUSES"> at the bottom.
1945 =item sqltrue, sqlfalse
1947 Expressions for inserting boolean values within SQL statements.
1948 By default these are C<1=1> and C<1=0>. They are used
1949 by the special operators C<-in> and C<-not_in> for generating
1950 correct SQL even when the argument is an empty array (see below).
1954 This determines the default logical operator for multiple WHERE
1955 statements in arrays or hashes. If absent, the default logic is "or"
1956 for arrays, and "and" for hashes. This means that a WHERE
1960 event_date => {'>=', '2/13/99'},
1961 event_date => {'<=', '4/24/03'},
1964 will generate SQL like this:
1966 WHERE event_date >= '2/13/99' OR event_date <= '4/24/03'
1968 This is probably not what you want given this query, though (look
1969 at the dates). To change the "OR" to an "AND", simply specify:
1971 my $sql = SQL::Abstract->new(logic => 'and');
1973 Which will change the above C<WHERE> to:
1975 WHERE event_date >= '2/13/99' AND event_date <= '4/24/03'
1977 The logic can also be changed locally by inserting
1978 a modifier in front of an arrayref:
1980 @where = (-and => [event_date => {'>=', '2/13/99'},
1981 event_date => {'<=', '4/24/03'} ]);
1983 See the L</"WHERE CLAUSES"> section for explanations.
1987 This will automatically convert comparisons using the specified SQL
1988 function for both column and value. This is mostly used with an argument
1989 of C<upper> or C<lower>, so that the SQL will have the effect of
1990 case-insensitive "searches". For example, this:
1992 $sql = SQL::Abstract->new(convert => 'upper');
1993 %where = (keywords => 'MaKe iT CAse inSeNSItive');
1995 Will turn out the following SQL:
1997 WHERE upper(keywords) like upper('MaKe iT CAse inSeNSItive')
1999 The conversion can be C<upper()>, C<lower()>, or any other SQL function
2000 that can be applied symmetrically to fields (actually B<SQL::Abstract> does
2001 not validate this option; it will just pass through what you specify verbatim).
2005 This is a kludge because many databases suck. For example, you can't
2006 just bind values using DBI's C<execute()> for Oracle C<CLOB> or C<BLOB> fields.
2007 Instead, you have to use C<bind_param()>:
2009 $sth->bind_param(1, 'reg data');
2010 $sth->bind_param(2, $lots, {ora_type => ORA_CLOB});
2012 The problem is, B<SQL::Abstract> will normally just return a C<@bind> array,
2013 which loses track of which field each slot refers to. Fear not.
2015 If you specify C<bindtype> in new, you can determine how C<@bind> is returned.
2016 Currently, you can specify either C<normal> (default) or C<columns>. If you
2017 specify C<columns>, you will get an array that looks like this:
2019 my $sql = SQL::Abstract->new(bindtype => 'columns');
2020 my($stmt, @bind) = $sql->insert(...);
2023 [ 'column1', 'value1' ],
2024 [ 'column2', 'value2' ],
2025 [ 'column3', 'value3' ],
2028 You can then iterate through this manually, using DBI's C<bind_param()>.
2030 $sth->prepare($stmt);
2033 my($col, $data) = @$_;
2034 if ($col eq 'details' || $col eq 'comments') {
2035 $sth->bind_param($i, $data, {ora_type => ORA_CLOB});
2036 } elsif ($col eq 'image') {
2037 $sth->bind_param($i, $data, {ora_type => ORA_BLOB});
2039 $sth->bind_param($i, $data);
2043 $sth->execute; # execute without @bind now
2045 Now, why would you still use B<SQL::Abstract> if you have to do this crap?
2046 Basically, the advantage is still that you don't have to care which fields
2047 are or are not included. You could wrap that above C<for> loop in a simple
2048 sub called C<bind_fields()> or something and reuse it repeatedly. You still
2049 get a layer of abstraction over manual SQL specification.
2051 Note that if you set L</bindtype> to C<columns>, the C<\[ $sql, @bind ]>
2052 construct (see L</Literal SQL with placeholders and bind values (subqueries)>)
2053 will expect the bind values in this format.
2057 This is the character that a table or column name will be quoted
2058 with. By default this is an empty string, but you could set it to
2059 the character C<`>, to generate SQL like this:
2061 SELECT `a_field` FROM `a_table` WHERE `some_field` LIKE '%someval%'
2063 Alternatively, you can supply an array ref of two items, the first being the left
2064 hand quote character, and the second the right hand quote character. For
2065 example, you could supply C<['[',']']> for SQL Server 2000 compliant quotes
2066 that generates SQL like this:
2068 SELECT [a_field] FROM [a_table] WHERE [some_field] LIKE '%someval%'
2070 Quoting is useful if you have tables or columns names that are reserved
2071 words in your database's SQL dialect.
2075 This is the character that will be used to escape L</quote_char>s appearing
2076 in an identifier before it has been quoted.
2078 The parameter default in case of a single L</quote_char> character is the quote
2081 When opening-closing-style quoting is used (L</quote_char> is an arrayref)
2082 this parameter defaults to the B<closing (right)> L</quote_char>. Occurrences
2083 of the B<opening (left)> L</quote_char> within the identifier are currently left
2084 untouched. The default for opening-closing-style quotes may change in future
2085 versions, thus you are B<strongly encouraged> to specify the escape character
2090 This is the character that separates a table and column name. It is
2091 necessary to specify this when the C<quote_char> option is selected,
2092 so that tables and column names can be individually quoted like this:
2094 SELECT `table`.`one_field` FROM `table` WHERE `table`.`other_field` = 1
2096 =item injection_guard
2098 A regular expression C<qr/.../> that is applied to any C<-function> and unquoted
2099 column name specified in a query structure. This is a safety mechanism to avoid
2100 injection attacks when mishandling user input e.g.:
2102 my %condition_as_column_value_pairs = get_values_from_user();
2103 $sqla->select( ... , \%condition_as_column_value_pairs );
2105 If the expression matches an exception is thrown. Note that literal SQL
2106 supplied via C<\'...'> or C<\['...']> is B<not> checked in any way.
2108 Defaults to checking for C<;> and the C<GO> keyword (TransactSQL)
2110 =item array_datatypes
2112 When this option is true, arrayrefs in INSERT or UPDATE are
2113 interpreted as array datatypes and are passed directly
2115 When this option is false, arrayrefs are interpreted
2116 as literal SQL, just like refs to arrayrefs
2117 (but this behavior is for backwards compatibility; when writing
2118 new queries, use the "reference to arrayref" syntax
2124 Takes a reference to a list of "special operators"
2125 to extend the syntax understood by L<SQL::Abstract>.
2126 See section L</"SPECIAL OPERATORS"> for details.
2130 Takes a reference to a list of "unary operators"
2131 to extend the syntax understood by L<SQL::Abstract>.
2132 See section L</"UNARY OPERATORS"> for details.
2138 =head2 insert($table, \@values || \%fieldvals, \%options)
2140 This is the simplest function. You simply give it a table name
2141 and either an arrayref of values or hashref of field/value pairs.
2142 It returns an SQL INSERT statement and a list of bind values.
2143 See the sections on L</"Inserting and Updating Arrays"> and
2144 L</"Inserting and Updating SQL"> for information on how to insert
2145 with those data types.
2147 The optional C<\%options> hash reference may contain additional
2148 options to generate the insert SQL. Currently supported options
2155 Takes either a scalar of raw SQL fields, or an array reference of
2156 field names, and adds on an SQL C<RETURNING> statement at the end.
2157 This allows you to return data generated by the insert statement
2158 (such as row IDs) without performing another C<SELECT> statement.
2159 Note, however, this is not part of the SQL standard and may not
2160 be supported by all database engines.
2164 =head2 update($table, \%fieldvals, \%where, \%options)
2166 This takes a table, hashref of field/value pairs, and an optional
2167 hashref L<WHERE clause|/WHERE CLAUSES>. It returns an SQL UPDATE function and a list
2169 See the sections on L</"Inserting and Updating Arrays"> and
2170 L</"Inserting and Updating SQL"> for information on how to insert
2171 with those data types.
2173 The optional C<\%options> hash reference may contain additional
2174 options to generate the update SQL. Currently supported options
2181 See the C<returning> option to
2182 L<insert|/insert($table, \@values || \%fieldvals, \%options)>.
2186 =head2 select($source, $fields, $where, $order)
2188 This returns a SQL SELECT statement and associated list of bind values, as
2189 specified by the arguments:
2195 Specification of the 'FROM' part of the statement.
2196 The argument can be either a plain scalar (interpreted as a table
2197 name, will be quoted), or an arrayref (interpreted as a list
2198 of table names, joined by commas, quoted), or a scalarref
2199 (literal SQL, not quoted).
2203 Specification of the list of fields to retrieve from
2205 The argument can be either an arrayref (interpreted as a list
2206 of field names, will be joined by commas and quoted), or a
2207 plain scalar (literal SQL, not quoted).
2208 Please observe that this API is not as flexible as that of
2209 the first argument C<$source>, for backwards compatibility reasons.
2213 Optional argument to specify the WHERE part of the query.
2214 The argument is most often a hashref, but can also be
2215 an arrayref or plain scalar --
2216 see section L<WHERE clause|/"WHERE CLAUSES"> for details.
2220 Optional argument to specify the ORDER BY part of the query.
2221 The argument can be a scalar, a hashref or an arrayref
2222 -- see section L<ORDER BY clause|/"ORDER BY CLAUSES">
2228 =head2 delete($table, \%where, \%options)
2230 This takes a table name and optional hashref L<WHERE clause|/WHERE CLAUSES>.
2231 It returns an SQL DELETE statement and list of bind values.
2233 The optional C<\%options> hash reference may contain additional
2234 options to generate the delete SQL. Currently supported options
2241 See the C<returning> option to
2242 L<insert|/insert($table, \@values || \%fieldvals, \%options)>.
2246 =head2 where(\%where, $order)
2248 This is used to generate just the WHERE clause. For example,
2249 if you have an arbitrary data structure and know what the
2250 rest of your SQL is going to look like, but want an easy way
2251 to produce a WHERE clause, use this. It returns an SQL WHERE
2252 clause and list of bind values.
2255 =head2 values(\%data)
2257 This just returns the values from the hash C<%data>, in the same
2258 order that would be returned from any of the other above queries.
2259 Using this allows you to markedly speed up your queries if you
2260 are affecting lots of rows. See below under the L</"PERFORMANCE"> section.
2262 =head2 generate($any, 'number', $of, \@data, $struct, \%types)
2264 Warning: This is an experimental method and subject to change.
2266 This returns arbitrarily generated SQL. It's a really basic shortcut.
2267 It will return two different things, depending on return context:
2269 my($stmt, @bind) = $sql->generate('create table', \$table, \@fields);
2270 my $stmt_and_val = $sql->generate('create table', \$table, \@fields);
2272 These would return the following:
2274 # First calling form
2275 $stmt = "CREATE TABLE test (?, ?)";
2276 @bind = (field1, field2);
2278 # Second calling form
2279 $stmt_and_val = "CREATE TABLE test (field1, field2)";
2281 Depending on what you're trying to do, it's up to you to choose the correct
2282 format. In this example, the second form is what you would want.
2286 $sql->generate('alter session', { nls_date_format => 'MM/YY' });
2290 ALTER SESSION SET nls_date_format = 'MM/YY'
2292 You get the idea. Strings get their case twiddled, but everything
2293 else remains verbatim.
2295 =head1 EXPORTABLE FUNCTIONS
2297 =head2 is_plain_value
2299 Determines if the supplied argument is a plain value as understood by this
2304 =item * The value is C<undef>
2306 =item * The value is a non-reference
2308 =item * The value is an object with stringification overloading
2310 =item * The value is of the form C<< { -value => $anything } >>
2314 On failure returns C<undef>, on success returns a B<scalar> reference
2315 to the original supplied argument.
2321 The stringification overloading detection is rather advanced: it takes
2322 into consideration not only the presence of a C<""> overload, but if that
2323 fails also checks for enabled
2324 L<autogenerated versions of C<"">|overload/Magic Autogeneration>, based
2325 on either C<0+> or C<bool>.
2327 Unfortunately testing in the field indicates that this
2328 detection B<< may tickle a latent bug in perl versions before 5.018 >>,
2329 but only when very large numbers of stringifying objects are involved.
2330 At the time of writing ( Sep 2014 ) there is no clear explanation of
2331 the direct cause, nor is there a manageably small test case that reliably
2332 reproduces the problem.
2334 If you encounter any of the following exceptions in B<random places within
2335 your application stack> - this module may be to blame:
2337 Operation "ne": no method found,
2338 left argument in overloaded package <something>,
2339 right argument in overloaded package <something>
2343 Stub found while resolving method "???" overloading """" in package <something>
2345 If you fall victim to the above - please attempt to reduce the problem
2346 to something that could be sent to the L<SQL::Abstract developers
2347 |DBIx::Class/GETTING HELP/SUPPORT>
2348 (either publicly or privately). As a workaround in the meantime you can
2349 set C<$ENV{SQLA_ISVALUE_IGNORE_AUTOGENERATED_STRINGIFICATION}> to a true
2350 value, which will most likely eliminate your problem (at the expense of
2351 not being able to properly detect exotic forms of stringification).
2353 This notice and environment variable will be removed in a future version,
2354 as soon as the underlying problem is found and a reliable workaround is
2359 =head2 is_literal_value
2361 Determines if the supplied argument is a literal value as understood by this
2366 =item * C<\$sql_string>
2368 =item * C<\[ $sql_string, @bind_values ]>
2372 On failure returns C<undef>, on success returns an B<array> reference
2373 containing the unpacked version of the supplied literal SQL and bind values.
2375 =head1 WHERE CLAUSES
2379 This module uses a variation on the idea from L<DBIx::Abstract>. It
2380 is B<NOT>, repeat I<not> 100% compatible. B<The main logic of this
2381 module is that things in arrays are OR'ed, and things in hashes
2384 The easiest way to explain is to show lots of examples. After
2385 each C<%where> hash shown, it is assumed you used:
2387 my($stmt, @bind) = $sql->where(\%where);
2389 However, note that the C<%where> hash can be used directly in any
2390 of the other functions as well, as described above.
2392 =head2 Key-value pairs
2394 So, let's get started. To begin, a simple hash:
2398 status => 'completed'
2401 Is converted to SQL C<key = val> statements:
2403 $stmt = "WHERE user = ? AND status = ?";
2404 @bind = ('nwiger', 'completed');
2406 One common thing I end up doing is having a list of values that
2407 a field can be in. To do this, simply specify a list inside of
2412 status => ['assigned', 'in-progress', 'pending'];
2415 This simple code will create the following:
2417 $stmt = "WHERE user = ? AND ( status = ? OR status = ? OR status = ? )";
2418 @bind = ('nwiger', 'assigned', 'in-progress', 'pending');
2420 A field associated to an empty arrayref will be considered a
2421 logical false and will generate 0=1.
2423 =head2 Tests for NULL values
2425 If the value part is C<undef> then this is converted to SQL <IS NULL>
2434 $stmt = "WHERE user = ? AND status IS NULL";
2437 To test if a column IS NOT NULL:
2441 status => { '!=', undef },
2444 =head2 Specific comparison operators
2446 If you want to specify a different type of operator for your comparison,
2447 you can use a hashref for a given column:
2451 status => { '!=', 'completed' }
2454 Which would generate:
2456 $stmt = "WHERE user = ? AND status != ?";
2457 @bind = ('nwiger', 'completed');
2459 To test against multiple values, just enclose the values in an arrayref:
2461 status => { '=', ['assigned', 'in-progress', 'pending'] };
2463 Which would give you:
2465 "WHERE status = ? OR status = ? OR status = ?"
2468 The hashref can also contain multiple pairs, in which case it is expanded
2469 into an C<AND> of its elements:
2473 status => { '!=', 'completed', -not_like => 'pending%' }
2476 # Or more dynamically, like from a form
2477 $where{user} = 'nwiger';
2478 $where{status}{'!='} = 'completed';
2479 $where{status}{'-not_like'} = 'pending%';
2481 # Both generate this
2482 $stmt = "WHERE user = ? AND status != ? AND status NOT LIKE ?";
2483 @bind = ('nwiger', 'completed', 'pending%');
2486 To get an OR instead, you can combine it with the arrayref idea:
2490 priority => [ { '=', 2 }, { '>', 5 } ]
2493 Which would generate:
2495 $stmt = "WHERE ( priority = ? OR priority > ? ) AND user = ?";
2496 @bind = ('2', '5', 'nwiger');
2498 If you want to include literal SQL (with or without bind values), just use a
2499 scalar reference or reference to an arrayref as the value:
2502 date_entered => { '>' => \["to_date(?, 'MM/DD/YYYY')", "11/26/2008"] },
2503 date_expires => { '<' => \"now()" }
2506 Which would generate:
2508 $stmt = "WHERE date_entered > to_date(?, 'MM/DD/YYYY') AND date_expires < now()";
2509 @bind = ('11/26/2008');
2512 =head2 Logic and nesting operators
2514 In the example above,
2515 there is a subtle trap if you want to say something like
2516 this (notice the C<AND>):
2518 WHERE priority != ? AND priority != ?
2520 Because, in Perl you I<can't> do this:
2522 priority => { '!=' => 2, '!=' => 1 }
2524 As the second C<!=> key will obliterate the first. The solution
2525 is to use the special C<-modifier> form inside an arrayref:
2527 priority => [ -and => {'!=', 2},
2531 Normally, these would be joined by C<OR>, but the modifier tells it
2532 to use C<AND> instead. (Hint: You can use this in conjunction with the
2533 C<logic> option to C<new()> in order to change the way your queries
2534 work by default.) B<Important:> Note that the C<-modifier> goes
2535 B<INSIDE> the arrayref, as an extra first element. This will
2536 B<NOT> do what you think it might:
2538 priority => -and => [{'!=', 2}, {'!=', 1}] # WRONG!
2540 Here is a quick list of equivalencies, since there is some overlap:
2543 status => {'!=', 'completed', 'not like', 'pending%' }
2544 status => [ -and => {'!=', 'completed'}, {'not like', 'pending%'}]
2547 status => {'=', ['assigned', 'in-progress']}
2548 status => [ -or => {'=', 'assigned'}, {'=', 'in-progress'}]
2549 status => [ {'=', 'assigned'}, {'=', 'in-progress'} ]
2553 =head2 Special operators: IN, BETWEEN, etc.
2555 You can also use the hashref format to compare a list of fields using the
2556 C<IN> comparison operator, by specifying the list as an arrayref:
2559 status => 'completed',
2560 reportid => { -in => [567, 2335, 2] }
2563 Which would generate:
2565 $stmt = "WHERE status = ? AND reportid IN (?,?,?)";
2566 @bind = ('completed', '567', '2335', '2');
2568 The reverse operator C<-not_in> generates SQL C<NOT IN> and is used in
2571 If the argument to C<-in> is an empty array, 'sqlfalse' is generated
2572 (by default: C<1=0>). Similarly, C<< -not_in => [] >> generates
2573 'sqltrue' (by default: C<1=1>).
2575 In addition to the array you can supply a chunk of literal sql or
2576 literal sql with bind:
2579 customer => { -in => \[
2580 'SELECT cust_id FROM cust WHERE balance > ?',
2583 status => { -in => \'SELECT status_codes FROM states' },
2589 customer IN ( SELECT cust_id FROM cust WHERE balance > ? )
2590 AND status IN ( SELECT status_codes FROM states )
2594 Finally, if the argument to C<-in> is not a reference, it will be
2595 treated as a single-element array.
2597 Another pair of operators is C<-between> and C<-not_between>,
2598 used with an arrayref of two values:
2602 completion_date => {
2603 -not_between => ['2002-10-01', '2003-02-06']
2609 WHERE user = ? AND completion_date NOT BETWEEN ( ? AND ? )
2611 Just like with C<-in> all plausible combinations of literal SQL
2615 start0 => { -between => [ 1, 2 ] },
2616 start1 => { -between => \["? AND ?", 1, 2] },
2617 start2 => { -between => \"lower(x) AND upper(y)" },
2618 start3 => { -between => [
2620 \["upper(?)", 'stuff' ],
2627 ( start0 BETWEEN ? AND ? )
2628 AND ( start1 BETWEEN ? AND ? )
2629 AND ( start2 BETWEEN lower(x) AND upper(y) )
2630 AND ( start3 BETWEEN lower(x) AND upper(?) )
2632 @bind = (1, 2, 1, 2, 'stuff');
2635 These are the two builtin "special operators"; but the
2636 list can be expanded: see section L</"SPECIAL OPERATORS"> below.
2638 =head2 Unary operators: bool
2640 If you wish to test against boolean columns or functions within your
2641 database you can use the C<-bool> and C<-not_bool> operators. For
2642 example to test the column C<is_user> being true and the column
2643 C<is_enabled> being false you would use:-
2647 -not_bool => 'is_enabled',
2652 WHERE is_user AND NOT is_enabled
2654 If a more complex combination is required, testing more conditions,
2655 then you should use the and/or operators:-
2660 -not_bool => { two=> { -rlike => 'bar' } },
2661 -not_bool => { three => [ { '=', 2 }, { '>', 5 } ] },
2672 (NOT ( three = ? OR three > ? ))
2675 =head2 Nested conditions, -and/-or prefixes
2677 So far, we've seen how multiple conditions are joined with a top-level
2678 C<AND>. We can change this by putting the different conditions we want in
2679 hashes and then putting those hashes in an array. For example:
2684 status => { -like => ['pending%', 'dispatched'] },
2688 status => 'unassigned',
2692 This data structure would create the following:
2694 $stmt = "WHERE ( user = ? AND ( status LIKE ? OR status LIKE ? ) )
2695 OR ( user = ? AND status = ? ) )";
2696 @bind = ('nwiger', 'pending', 'dispatched', 'robot', 'unassigned');
2699 Clauses in hashrefs or arrayrefs can be prefixed with an C<-and> or C<-or>
2700 to change the logic inside:
2706 -and => [ workhrs => {'>', 20}, geo => 'ASIA' ],
2707 -or => { workhrs => {'<', 50}, geo => 'EURO' },
2714 $stmt = "WHERE ( user = ?
2715 AND ( ( workhrs > ? AND geo = ? )
2716 OR ( workhrs < ? OR geo = ? ) ) )";
2717 @bind = ('nwiger', '20', 'ASIA', '50', 'EURO');
2719 =head3 Algebraic inconsistency, for historical reasons
2721 C<Important note>: when connecting several conditions, the C<-and->|C<-or>
2722 operator goes C<outside> of the nested structure; whereas when connecting
2723 several constraints on one column, the C<-and> operator goes
2724 C<inside> the arrayref. Here is an example combining both features:
2727 -and => [a => 1, b => 2],
2728 -or => [c => 3, d => 4],
2729 e => [-and => {-like => 'foo%'}, {-like => '%bar'} ]
2734 WHERE ( ( ( a = ? AND b = ? )
2735 OR ( c = ? OR d = ? )
2736 OR ( e LIKE ? AND e LIKE ? ) ) )
2738 This difference in syntax is unfortunate but must be preserved for
2739 historical reasons. So be careful: the two examples below would
2740 seem algebraically equivalent, but they are not
2743 { -like => 'foo%' },
2744 { -like => '%bar' },
2746 # yields: WHERE ( ( col LIKE ? AND col LIKE ? ) )
2749 { col => { -like => 'foo%' } },
2750 { col => { -like => '%bar' } },
2752 # yields: WHERE ( ( col LIKE ? OR col LIKE ? ) )
2755 =head2 Literal SQL and value type operators
2757 The basic premise of SQL::Abstract is that in WHERE specifications the "left
2758 side" is a column name and the "right side" is a value (normally rendered as
2759 a placeholder). This holds true for both hashrefs and arrayref pairs as you
2760 see in the L</WHERE CLAUSES> examples above. Sometimes it is necessary to
2761 alter this behavior. There are several ways of doing so.
2765 This is a virtual operator that signals the string to its right side is an
2766 identifier (a column name) and not a value. For example to compare two
2767 columns you would write:
2770 priority => { '<', 2 },
2771 requestor => { -ident => 'submitter' },
2776 $stmt = "WHERE priority < ? AND requestor = submitter";
2779 If you are maintaining legacy code you may see a different construct as
2780 described in L</Deprecated usage of Literal SQL>, please use C<-ident> in new
2785 This is a virtual operator that signals that the construct to its right side
2786 is a value to be passed to DBI. This is for example necessary when you want
2787 to write a where clause against an array (for RDBMS that support such
2788 datatypes). For example:
2791 array => { -value => [1, 2, 3] }
2796 $stmt = 'WHERE array = ?';
2797 @bind = ([1, 2, 3]);
2799 Note that if you were to simply say:
2805 the result would probably not be what you wanted:
2807 $stmt = 'WHERE array = ? OR array = ? OR array = ?';
2812 Finally, sometimes only literal SQL will do. To include a random snippet
2813 of SQL verbatim, you specify it as a scalar reference. Consider this only
2814 as a last resort. Usually there is a better way. For example:
2817 priority => { '<', 2 },
2818 requestor => { -in => \'(SELECT name FROM hitmen)' },
2823 $stmt = "WHERE priority < ? AND requestor IN (SELECT name FROM hitmen)"
2826 Note that in this example, you only get one bind parameter back, since
2827 the verbatim SQL is passed as part of the statement.
2831 Never use untrusted input as a literal SQL argument - this is a massive
2832 security risk (there is no way to check literal snippets for SQL
2833 injections and other nastyness). If you need to deal with untrusted input
2834 use literal SQL with placeholders as described next.
2836 =head3 Literal SQL with placeholders and bind values (subqueries)
2838 If the literal SQL to be inserted has placeholders and bind values,
2839 use a reference to an arrayref (yes this is a double reference --
2840 not so common, but perfectly legal Perl). For example, to find a date
2841 in Postgres you can use something like this:
2844 date_column => \[ "= date '2008-09-30' - ?::integer", 10 ]
2849 $stmt = "WHERE ( date_column = date '2008-09-30' - ?::integer )"
2852 Note that you must pass the bind values in the same format as they are returned
2853 by L<where|/where(\%where, $order)>. This means that if you set L</bindtype>
2854 to C<columns>, you must provide the bind values in the
2855 C<< [ column_meta => value ] >> format, where C<column_meta> is an opaque
2856 scalar value; most commonly the column name, but you can use any scalar value
2857 (including references and blessed references), L<SQL::Abstract> will simply
2858 pass it through intact. So if C<bindtype> is set to C<columns> the above
2859 example will look like:
2862 date_column => \[ "= date '2008-09-30' - ?::integer", [ {} => 10 ] ]
2865 Literal SQL is especially useful for nesting parenthesized clauses in the
2866 main SQL query. Here is a first example:
2868 my ($sub_stmt, @sub_bind) = ("SELECT c1 FROM t1 WHERE c2 < ? AND c3 LIKE ?",
2872 bar => \["IN ($sub_stmt)" => @sub_bind],
2877 $stmt = "WHERE (foo = ? AND bar IN (SELECT c1 FROM t1
2878 WHERE c2 < ? AND c3 LIKE ?))";
2879 @bind = (1234, 100, "foo%");
2881 Other subquery operators, like for example C<"E<gt> ALL"> or C<"NOT IN">,
2882 are expressed in the same way. Of course the C<$sub_stmt> and
2883 its associated bind values can be generated through a former call
2886 my ($sub_stmt, @sub_bind)
2887 = $sql->select("t1", "c1", {c2 => {"<" => 100},
2888 c3 => {-like => "foo%"}});
2891 bar => \["> ALL ($sub_stmt)" => @sub_bind],
2894 In the examples above, the subquery was used as an operator on a column;
2895 but the same principle also applies for a clause within the main C<%where>
2896 hash, like an EXISTS subquery:
2898 my ($sub_stmt, @sub_bind)
2899 = $sql->select("t1", "*", {c1 => 1, c2 => \"> t0.c0"});
2900 my %where = ( -and => [
2902 \["EXISTS ($sub_stmt)" => @sub_bind],
2907 $stmt = "WHERE (foo = ? AND EXISTS (SELECT * FROM t1
2908 WHERE c1 = ? AND c2 > t0.c0))";
2912 Observe that the condition on C<c2> in the subquery refers to
2913 column C<t0.c0> of the main query: this is I<not> a bind
2914 value, so we have to express it through a scalar ref.
2915 Writing C<< c2 => {">" => "t0.c0"} >> would have generated
2916 C<< c2 > ? >> with bind value C<"t0.c0"> ... not exactly
2917 what we wanted here.
2919 Finally, here is an example where a subquery is used
2920 for expressing unary negation:
2922 my ($sub_stmt, @sub_bind)
2923 = $sql->where({age => [{"<" => 10}, {">" => 20}]});
2924 $sub_stmt =~ s/^ where //i; # don't want "WHERE" in the subclause
2926 lname => {like => '%son%'},
2927 \["NOT ($sub_stmt)" => @sub_bind],
2932 $stmt = "lname LIKE ? AND NOT ( age < ? OR age > ? )"
2933 @bind = ('%son%', 10, 20)
2935 =head3 Deprecated usage of Literal SQL
2937 Below are some examples of archaic use of literal SQL. It is shown only as
2938 reference for those who deal with legacy code. Each example has a much
2939 better, cleaner and safer alternative that users should opt for in new code.
2945 my %where = ( requestor => \'IS NOT NULL' )
2947 $stmt = "WHERE requestor IS NOT NULL"
2949 This used to be the way of generating NULL comparisons, before the handling
2950 of C<undef> got formalized. For new code please use the superior syntax as
2951 described in L</Tests for NULL values>.
2955 my %where = ( requestor => \'= submitter' )
2957 $stmt = "WHERE requestor = submitter"
2959 This used to be the only way to compare columns. Use the superior L</-ident>
2960 method for all new code. For example an identifier declared in such a way
2961 will be properly quoted if L</quote_char> is properly set, while the legacy
2962 form will remain as supplied.
2966 my %where = ( is_ready => \"", completed => { '>', '2012-12-21' } )
2968 $stmt = "WHERE completed > ? AND is_ready"
2969 @bind = ('2012-12-21')
2971 Using an empty string literal used to be the only way to express a boolean.
2972 For all new code please use the much more readable
2973 L<-bool|/Unary operators: bool> operator.
2979 These pages could go on for a while, since the nesting of the data
2980 structures this module can handle are pretty much unlimited (the
2981 module implements the C<WHERE> expansion as a recursive function
2982 internally). Your best bet is to "play around" with the module a
2983 little to see how the data structures behave, and choose the best
2984 format for your data based on that.
2986 And of course, all the values above will probably be replaced with
2987 variables gotten from forms or the command line. After all, if you
2988 knew everything ahead of time, you wouldn't have to worry about
2989 dynamically-generating SQL and could just hardwire it into your
2992 =head1 ORDER BY CLAUSES
2994 Some functions take an order by clause. This can either be a scalar (just a
2995 column name), a hashref of C<< { -desc => 'col' } >> or C<< { -asc => 'col' }
2996 >>, a scalarref, an arrayref-ref, or an arrayref of any of the previous
2999 Given | Will Generate
3000 ---------------------------------------------------------------
3002 'colA' | ORDER BY colA
3004 [qw/colA colB/] | ORDER BY colA, colB
3006 {-asc => 'colA'} | ORDER BY colA ASC
3008 {-desc => 'colB'} | ORDER BY colB DESC
3010 ['colA', {-asc => 'colB'}] | ORDER BY colA, colB ASC
3012 { -asc => [qw/colA colB/] } | ORDER BY colA ASC, colB ASC
3014 \'colA DESC' | ORDER BY colA DESC
3016 \[ 'FUNC(colA, ?)', $x ] | ORDER BY FUNC(colA, ?)
3017 | /* ...with $x bound to ? */
3020 { -asc => 'colA' }, | colA ASC,
3021 { -desc => [qw/colB/] }, | colB DESC,
3022 { -asc => [qw/colC colD/] },| colC ASC, colD ASC,
3023 \'colE DESC', | colE DESC,
3024 \[ 'FUNC(colF, ?)', $x ], | FUNC(colF, ?)
3025 ] | /* ...with $x bound to ? */
3026 ===============================================================
3030 =head1 SPECIAL OPERATORS
3032 my $sqlmaker = SQL::Abstract->new(special_ops => [
3036 my ($self, $field, $op, $arg) = @_;
3042 handler => 'method_name',
3046 A "special operator" is a SQL syntactic clause that can be
3047 applied to a field, instead of a usual binary operator.
3050 WHERE field IN (?, ?, ?)
3051 WHERE field BETWEEN ? AND ?
3052 WHERE MATCH(field) AGAINST (?, ?)
3054 Special operators IN and BETWEEN are fairly standard and therefore
3055 are builtin within C<SQL::Abstract> (as the overridable methods
3056 C<_where_field_IN> and C<_where_field_BETWEEN>). For other operators,
3057 like the MATCH .. AGAINST example above which is specific to MySQL,
3058 you can write your own operator handlers - supply a C<special_ops>
3059 argument to the C<new> method. That argument takes an arrayref of
3060 operator definitions; each operator definition is a hashref with two
3067 the regular expression to match the operator
3071 Either a coderef or a plain scalar method name. In both cases
3072 the expected return is C<< ($sql, @bind) >>.
3074 When supplied with a method name, it is simply called on the
3075 L<SQL::Abstract> object as:
3077 $self->$method_name($field, $op, $arg)
3081 $field is the LHS of the operator
3082 $op is the part that matched the handler regex
3085 When supplied with a coderef, it is called as:
3087 $coderef->($self, $field, $op, $arg)
3092 For example, here is an implementation
3093 of the MATCH .. AGAINST syntax for MySQL
3095 my $sqlmaker = SQL::Abstract->new(special_ops => [
3097 # special op for MySql MATCH (field) AGAINST(word1, word2, ...)
3098 {regex => qr/^match$/i,
3100 my ($self, $field, $op, $arg) = @_;
3101 $arg = [$arg] if not ref $arg;
3102 my $label = $self->_quote($field);
3103 my ($placeholder) = $self->_convert('?');
3104 my $placeholders = join ", ", (($placeholder) x @$arg);
3105 my $sql = $self->_sqlcase('match') . " ($label) "
3106 . $self->_sqlcase('against') . " ($placeholders) ";
3107 my @bind = $self->_bindtype($field, @$arg);
3108 return ($sql, @bind);
3115 =head1 UNARY OPERATORS
3117 my $sqlmaker = SQL::Abstract->new(unary_ops => [
3121 my ($self, $op, $arg) = @_;
3127 handler => 'method_name',
3131 A "unary operator" is a SQL syntactic clause that can be
3132 applied to a field - the operator goes before the field
3134 You can write your own operator handlers - supply a C<unary_ops>
3135 argument to the C<new> method. That argument takes an arrayref of
3136 operator definitions; each operator definition is a hashref with two
3143 the regular expression to match the operator
3147 Either a coderef or a plain scalar method name. In both cases
3148 the expected return is C<< $sql >>.
3150 When supplied with a method name, it is simply called on the
3151 L<SQL::Abstract> object as:
3153 $self->$method_name($op, $arg)
3157 $op is the part that matched the handler regex
3158 $arg is the RHS or argument of the operator
3160 When supplied with a coderef, it is called as:
3162 $coderef->($self, $op, $arg)
3170 Thanks to some benchmarking by Mark Stosberg, it turns out that
3171 this module is many orders of magnitude faster than using C<DBIx::Abstract>.
3172 I must admit this wasn't an intentional design issue, but it's a
3173 byproduct of the fact that you get to control your C<DBI> handles
3176 To maximize performance, use a code snippet like the following:
3178 # prepare a statement handle using the first row
3179 # and then reuse it for the rest of the rows
3181 for my $href (@array_of_hashrefs) {
3182 $stmt ||= $sql->insert('table', $href);
3183 $sth ||= $dbh->prepare($stmt);
3184 $sth->execute($sql->values($href));
3187 The reason this works is because the keys in your C<$href> are sorted
3188 internally by B<SQL::Abstract>. Thus, as long as your data retains
3189 the same structure, you only have to generate the SQL the first time
3190 around. On subsequent queries, simply use the C<values> function provided
3191 by this module to return your values in the correct order.
3193 However this depends on the values having the same type - if, for
3194 example, the values of a where clause may either have values
3195 (resulting in sql of the form C<column = ?> with a single bind
3196 value), or alternatively the values might be C<undef> (resulting in
3197 sql of the form C<column IS NULL> with no bind value) then the
3198 caching technique suggested will not work.
3202 If you use my C<CGI::FormBuilder> module at all, you'll hopefully
3203 really like this part (I do, at least). Building up a complex query
3204 can be as simple as the following:
3211 use CGI::FormBuilder;
3214 my $form = CGI::FormBuilder->new(...);
3215 my $sql = SQL::Abstract->new;
3217 if ($form->submitted) {
3218 my $field = $form->field;
3219 my $id = delete $field->{id};
3220 my($stmt, @bind) = $sql->update('table', $field, {id => $id});
3223 Of course, you would still have to connect using C<DBI> to run the
3224 query, but the point is that if you make your form look like your
3225 table, the actual query script can be extremely simplistic.
3227 If you're B<REALLY> lazy (I am), check out C<HTML::QuickTable> for
3228 a fast interface to returning and formatting data. I frequently
3229 use these three modules together to write complex database query
3230 apps in under 50 lines.
3232 =head1 HOW TO CONTRIBUTE
3234 Contributions are always welcome, in all usable forms (we especially
3235 welcome documentation improvements). The delivery methods include git-
3236 or unified-diff formatted patches, GitHub pull requests, or plain bug
3237 reports either via RT or the Mailing list. Contributors are generally
3238 granted full access to the official repository after their first several
3239 patches pass successful review.
3241 This project is maintained in a git repository. The code and related tools are
3242 accessible at the following locations:
3246 =item * Official repo: L<git://git.shadowcat.co.uk/dbsrgits/SQL-Abstract.git>
3248 =item * Official gitweb: L<http://git.shadowcat.co.uk/gitweb/gitweb.cgi?p=dbsrgits/SQL-Abstract.git>
3250 =item * GitHub mirror: L<https://github.com/dbsrgits/sql-abstract>
3252 =item * Authorized committers: L<ssh://dbsrgits@git.shadowcat.co.uk/SQL-Abstract.git>
3258 Version 1.50 was a major internal refactoring of C<SQL::Abstract>.
3259 Great care has been taken to preserve the I<published> behavior
3260 documented in previous versions in the 1.* family; however,
3261 some features that were previously undocumented, or behaved
3262 differently from the documentation, had to be changed in order
3263 to clarify the semantics. Hence, client code that was relying
3264 on some dark areas of C<SQL::Abstract> v1.*
3265 B<might behave differently> in v1.50.
3267 The main changes are:
3273 support for literal SQL through the C<< \ [ $sql, @bind ] >> syntax.
3277 support for the { operator => \"..." } construct (to embed literal SQL)
3281 support for the { operator => \["...", @bind] } construct (to embed literal SQL with bind values)
3285 optional support for L<array datatypes|/"Inserting and Updating Arrays">
3289 defensive programming: check arguments
3293 fixed bug with global logic, which was previously implemented
3294 through global variables yielding side-effects. Prior versions would
3295 interpret C<< [ {cond1, cond2}, [cond3, cond4] ] >>
3296 as C<< "(cond1 AND cond2) OR (cond3 AND cond4)" >>.
3297 Now this is interpreted
3298 as C<< "(cond1 AND cond2) OR (cond3 OR cond4)" >>.
3303 fixed semantics of _bindtype on array args
3307 dropped the C<_anoncopy> of the %where tree. No longer necessary,
3308 we just avoid shifting arrays within that tree.
3312 dropped the C<_modlogic> function
3316 =head1 ACKNOWLEDGEMENTS
3318 There are a number of individuals that have really helped out with
3319 this module. Unfortunately, most of them submitted bugs via CPAN
3320 so I have no idea who they are! But the people I do know are:
3322 Ash Berlin (order_by hash term support)
3323 Matt Trout (DBIx::Class support)
3324 Mark Stosberg (benchmarking)
3325 Chas Owens (initial "IN" operator support)
3326 Philip Collins (per-field SQL functions)
3327 Eric Kolve (hashref "AND" support)
3328 Mike Fragassi (enhancements to "BETWEEN" and "LIKE")
3329 Dan Kubb (support for "quote_char" and "name_sep")
3330 Guillermo Roditi (patch to cleanup "IN" and "BETWEEN", fix and tests for _order_by)
3331 Laurent Dami (internal refactoring, extensible list of special operators, literal SQL)
3332 Norbert Buchmuller (support for literal SQL in hashpair, misc. fixes & tests)
3333 Peter Rabbitson (rewrite of SQLA::Test, misc. fixes & tests)
3334 Oliver Charles (support for "RETURNING" after "INSERT")
3340 L<DBIx::Class>, L<DBIx::Abstract>, L<CGI::FormBuilder>, L<HTML::QuickTable>.
3344 Copyright (c) 2001-2007 Nathan Wiger <nwiger@cpan.org>. All Rights Reserved.
3346 This module is actively maintained by Matt Trout <mst@shadowcatsystems.co.uk>
3348 For support, your best bet is to try the C<DBIx::Class> users mailing list.
3349 While not an official support venue, C<DBIx::Class> makes heavy use of
3350 C<SQL::Abstract>, and as such list members there are very familiar with
3351 how to create queries.
3355 This module is free software; you may copy this under the same
3356 terms as perl itself (either the GNU General Public License or
3357 the Artistic License)