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);
561 return $self->_expand_expr($v);
563 puke "-bool => undef not supported" unless defined($v);
564 return { -ident => $v };
566 if (my ($rest) = $k =~/^-not[_ ](.*)$/) {
567 return $self->_expand_expr({ -not => { "-${rest}", $v } }, $logic);
571 return +{ $k => { $self->{cmp} => $v } };
573 if (ref($v) eq 'ARRAY') {
574 return $self->{sqlfalse} unless @$v;
575 $self->_debug("ARRAY($k) means distribute over elements");
577 $v->[0] =~ /^-((?:and|or))$/i
578 ? ($v = [ @{$v}[1..$#$v] ], $1)
579 : ($self->{logic} || 'or')
581 return +{ "-${this_logic}" => [ map $self->_expand_expr({ $k => $_ }, $this_logic), @$v ] };
583 if (my $literal = is_literal_value($v)) {
585 belch 'Hash-pairs consisting of an empty string with a literal are deprecated, and will be removed in 2.0: use -and => [ $literal ] instead';
588 my ($sql, @bind) = @$literal;
589 return \[ $self->_quote($k).' '.$sql, @bind ];
596 my ($self, $where, $logic) = @_;
598 my $where_exp = $self->_expand_expr($where, $logic);
600 # dispatch on appropriate method according to refkind of $where
601 my $method = $self->_METHOD_FOR_refkind("_where", $where_exp);
603 my ($sql, @bind) = $self->$method($where_exp, $logic);
605 # DBIx::Class used to call _recurse_where in scalar context
606 # something else might too...
608 return ($sql, @bind);
611 belch "Calling _recurse_where in scalar context is deprecated and will go away before 2.0";
618 #======================================================================
619 # WHERE: top-level ARRAYREF
620 #======================================================================
623 sub _where_ARRAYREF {
624 my ($self, $where, $logic) = @_;
626 $logic = uc($logic || $self->{logic});
627 $logic eq 'AND' or $logic eq 'OR' or puke "unknown logic: $logic";
629 my @clauses = @$where;
631 my (@sql_clauses, @all_bind);
632 # need to use while() so can shift() for pairs
634 my $el = shift @clauses;
636 $el = undef if (defined $el and ! length $el);
638 # switch according to kind of $el and get corresponding ($sql, @bind)
639 my ($sql, @bind) = $self->_SWITCH_refkind($el, {
641 # skip empty elements, otherwise get invalid trailing AND stuff
642 ARRAYREF => sub {$self->_recurse_where($el) if @$el},
646 $self->_assert_bindval_matches_bindtype(@b);
650 HASHREF => sub {$self->_recurse_where($el, 'and') if %$el},
652 SCALARREF => sub { ($$el); },
655 # top-level arrayref with scalars, recurse in pairs
656 $self->_recurse_where({$el => shift(@clauses)})
659 UNDEF => sub {puke "Supplying an empty left hand side argument is not supported in array-pairs" },
663 push @sql_clauses, $sql;
664 push @all_bind, @bind;
668 return $self->_join_sql_clauses($logic, \@sql_clauses, \@all_bind);
671 #======================================================================
672 # WHERE: top-level ARRAYREFREF
673 #======================================================================
675 sub _where_ARRAYREFREF {
676 my ($self, $where) = @_;
677 my ($sql, @bind) = @$$where;
678 $self->_assert_bindval_matches_bindtype(@bind);
679 return ($sql, @bind);
682 #======================================================================
683 # WHERE: top-level HASHREF
684 #======================================================================
687 my ($self, $where) = @_;
688 my (@sql_clauses, @all_bind);
690 for my $k (sort keys %$where) {
691 my $v = $where->{$k};
693 # ($k => $v) is either a special unary op or a regular hashpair
694 my ($sql, @bind) = do {
696 # put the operator in canonical form
698 $op = substr $op, 1; # remove initial dash
699 $op =~ s/^\s+|\s+$//g;# remove leading/trailing space
700 $op =~ s/\s+/ /g; # compress whitespace
702 # so that -not_foo works correctly
703 $op =~ s/^not_/NOT /i;
705 $self->_debug("Unary OP(-$op) within hashref, recursing...");
706 my ($s, @b) = $self->_where_unary_op($op, $v);
708 # top level vs nested
709 # we assume that handled unary ops will take care of their ()s
711 List::Util::first {$op =~ $_->{regex}} @{$self->{unary_ops}}
713 ( defined $self->{_nested_func_lhs} and $self->{_nested_func_lhs} eq $k )
719 if (is_literal_value ($v) ) {
720 belch 'Hash-pairs consisting of an empty string with a literal are deprecated, and will be removed in 2.0: use -and => [ $literal ] instead';
723 puke "Supplying an empty left hand side argument is not supported in hash-pairs";
727 my $method = $self->_METHOD_FOR_refkind("_where_hashpair", $v);
728 $self->$method($k, $v);
732 push @sql_clauses, $sql;
733 push @all_bind, @bind;
736 return $self->_join_sql_clauses('and', \@sql_clauses, \@all_bind);
739 sub _where_unary_op {
740 my ($self, $op, $rhs) = @_;
742 # top level special ops are illegal in general
743 puke "Illegal use of top-level '-$op'"
744 if !(defined $self->{_nested_func_lhs})
745 and List::Util::first { $op =~ $_->{regex} } @{$self->{special_ops}}
746 and not List::Util::first { $op =~ $_->{regex} } @{$self->{unary_ops}};
748 if (my $op_entry = List::Util::first { $op =~ $_->{regex} } @{$self->{unary_ops}}) {
749 my $handler = $op_entry->{handler};
751 if (not ref $handler) {
752 if ($op =~ s/ [_\s]? \d+ $//x ) {
753 belch 'Use of [and|or|nest]_N modifiers is deprecated and will be removed in SQLA v2.0. '
754 . "You probably wanted ...-and => [ -$op => COND1, -$op => COND2 ... ]";
756 return $self->$handler($op, $rhs);
758 elsif (ref $handler eq 'CODE') {
759 return $handler->($self, $op, $rhs);
762 puke "Illegal handler for operator $op - expecting a method name or a coderef";
766 $self->_debug("Generic unary OP: $op - recursing as function");
768 $self->_assert_pass_injection_guard($op);
770 my ($sql, @bind) = $self->_SWITCH_refkind($rhs, {
772 puke "Illegal use of top-level '-$op'"
773 unless defined $self->{_nested_func_lhs};
776 $self->_convert('?'),
777 $self->_bindtype($self->{_nested_func_lhs}, $rhs)
781 $self->_recurse_where($rhs)
785 $sql = sprintf('%s %s',
786 $self->_sqlcase($op),
790 return ($sql, @bind);
793 sub _where_op_ANDOR {
794 my ($self, $op, $v) = @_;
796 $self->_SWITCH_refkind($v, {
798 return $self->_where_ARRAYREF($v, $op);
802 return ($op =~ /^or/i)
803 ? $self->_where_ARRAYREF([ map { $_ => $v->{$_} } (sort keys %$v) ], $op)
804 : $self->_where_HASHREF($v);
808 puke "-$op => \\\$scalar makes little sense, use " .
810 ? '[ \$scalar, \%rest_of_conditions ] instead'
811 : '-and => [ \$scalar, \%rest_of_conditions ] instead'
816 puke "-$op => \\[...] makes little sense, use " .
818 ? '[ \[...], \%rest_of_conditions ] instead'
819 : '-and => [ \[...], \%rest_of_conditions ] instead'
823 SCALAR => sub { # permissively interpreted as SQL
824 puke "-$op => \$value makes little sense, use -bool => \$value instead";
828 puke "-$op => undef not supported";
834 my ($self, $op, $v) = @_;
836 $self->_SWITCH_refkind($v, {
838 SCALAR => sub { # permissively interpreted as SQL
839 belch "literal SQL should be -nest => \\'scalar' "
840 . "instead of -nest => 'scalar' ";
845 puke "-$op => undef not supported";
849 $self->_recurse_where($v);
857 my ($self, $op, $v) = @_;
859 my ($s, @b) = $self->_SWITCH_refkind($v, {
860 SCALAR => sub { # interpreted as SQL column
861 $self->_convert($self->_quote($v));
865 puke "-$op => undef not supported";
869 $self->_recurse_where($v);
873 $s = "(NOT $s)" if $op =~ /^not/i;
878 sub _where_op_IDENT {
880 my ($op, $rhs) = splice @_, -2;
881 if (! defined $rhs or length ref $rhs) {
882 puke "-$op requires a single plain scalar argument (a quotable identifier)";
885 # in case we are called as a top level special op (no '=')
888 $_ = $self->_convert($self->_quote($_)) for ($lhs, $rhs);
896 sub _where_op_VALUE {
898 my ($op, $rhs) = splice @_, -2;
900 # in case we are called as a top level special op (no '=')
904 if (! defined $rhs) {
906 ? $self->_where_hashpair_HASHREF($lhs, { -is => undef })
913 (defined $lhs ? $lhs : $self->{_nested_func_lhs}),
920 $self->_convert($self->_quote($lhs)) . ' = ' . $self->_convert('?'),
924 $self->_convert('?'),
930 sub _where_hashpair_ARRAYREF {
931 my ($self, $k, $v) = @_;
934 my @v = @$v; # need copy because of shift below
935 $self->_debug("ARRAY($k) means distribute over elements");
937 # put apart first element if it is an operator (-and, -or)
939 (defined $v[0] && $v[0] =~ /^ - (?: AND|OR ) $/ix)
943 my @distributed = map { {$k => $_} } @v;
946 $self->_debug("OP($op) reinjected into the distributed array");
947 unshift @distributed, $op;
950 my $logic = $op ? substr($op, 1) : '';
952 return $self->_recurse_where(\@distributed, $logic);
955 $self->_debug("empty ARRAY($k) means 0=1");
956 return ($self->{sqlfalse});
960 sub _where_hashpair_HASHREF {
961 my ($self, $k, $v, $logic) = @_;
964 local $self->{_nested_func_lhs} = defined $self->{_nested_func_lhs}
965 ? $self->{_nested_func_lhs}
969 my ($all_sql, @all_bind);
971 for my $orig_op (sort keys %$v) {
972 my $val = $v->{$orig_op};
974 # put the operator in canonical form
977 # FIXME - we need to phase out dash-less ops
978 $op =~ s/^-//; # remove possible initial dash
979 $op =~ s/^\s+|\s+$//g;# remove leading/trailing space
980 $op =~ s/\s+/ /g; # compress whitespace
982 $self->_assert_pass_injection_guard($op);
985 $op =~ s/^is_not/IS NOT/i;
987 # so that -not_foo works correctly
988 $op =~ s/^not_/NOT /i;
990 # another retarded special case: foo => { $op => { -value => undef } }
991 if (ref $val eq 'HASH' and keys %$val == 1 and exists $val->{-value} and ! defined $val->{-value} ) {
997 # CASE: col-value logic modifiers
998 if ($orig_op =~ /^ \- (and|or) $/xi) {
999 ($sql, @bind) = $self->_where_hashpair_HASHREF($k, $val, $1);
1001 # CASE: special operators like -in or -between
1002 elsif (my $special_op = List::Util::first { $op =~ $_->{regex} } @{$self->{special_ops}}) {
1003 my $handler = $special_op->{handler};
1005 puke "No handler supplied for special operator $orig_op";
1007 elsif (not ref $handler) {
1008 ($sql, @bind) = $self->$handler($k, $op, $val);
1010 elsif (ref $handler eq 'CODE') {
1011 ($sql, @bind) = $handler->($self, $k, $op, $val);
1014 puke "Illegal handler for special operator $orig_op - expecting a method name or a coderef";
1018 $self->_SWITCH_refkind($val, {
1020 ARRAYREF => sub { # CASE: col => {op => \@vals}
1021 ($sql, @bind) = $self->_where_field_op_ARRAYREF($k, $op, $val);
1024 ARRAYREFREF => sub { # CASE: col => {op => \[$sql, @bind]} (literal SQL with bind)
1025 my ($sub_sql, @sub_bind) = @$$val;
1026 $self->_assert_bindval_matches_bindtype(@sub_bind);
1027 $sql = join ' ', $self->_convert($self->_quote($k)),
1028 $self->_sqlcase($op),
1033 UNDEF => sub { # CASE: col => {op => undef} : sql "IS (NOT)? NULL"
1035 $op =~ /^not$/i ? 'is not' # legacy
1036 : $op =~ $self->{equality_op} ? 'is'
1037 : $op =~ $self->{like_op} ? belch("Supplying an undefined argument to '@{[ uc $op]}' is deprecated") && 'is'
1038 : $op =~ $self->{inequality_op} ? 'is not'
1039 : $op =~ $self->{not_like_op} ? belch("Supplying an undefined argument to '@{[ uc $op]}' is deprecated") && 'is not'
1040 : puke "unexpected operator '$orig_op' with undef operand";
1042 $sql = $self->_quote($k) . $self->_sqlcase(" $is null");
1045 FALLBACK => sub { # CASE: col => {op/func => $stuff}
1046 ($sql, @bind) = $self->_where_unary_op($op, $val);
1049 $self->_convert($self->_quote($k)),
1050 $self->{_nested_func_lhs} eq $k ? $sql : "($sql)", # top level vs nested
1056 ($all_sql) = (defined $all_sql and $all_sql) ? $self->_join_sql_clauses($logic, [$all_sql, $sql], []) : $sql;
1057 push @all_bind, @bind;
1059 return ($all_sql, @all_bind);
1062 sub _where_field_IS {
1063 my ($self, $k, $op, $v) = @_;
1065 my ($s) = $self->_SWITCH_refkind($v, {
1068 $self->_convert($self->_quote($k)),
1069 map { $self->_sqlcase($_)} ($op, 'null')
1072 puke "$op can only take undef as argument";
1079 sub _where_field_op_ARRAYREF {
1080 my ($self, $k, $op, $vals) = @_;
1082 my @vals = @$vals; #always work on a copy
1085 $self->_debug(sprintf '%s means multiple elements: [ %s ]',
1087 join(', ', map { defined $_ ? "'$_'" : 'NULL' } @vals ),
1090 # see if the first element is an -and/-or op
1092 if (defined $vals[0] && $vals[0] =~ /^ - (AND|OR) $/ix) {
1097 # a long standing API wart - an attempt to change this behavior during
1098 # the 1.50 series failed *spectacularly*. Warn instead and leave the
1103 (!$logic or $logic eq 'OR')
1105 ($op =~ $self->{inequality_op} or $op =~ $self->{not_like_op})
1108 belch "A multi-element arrayref as an argument to the inequality op '$o' "
1109 . 'is technically equivalent to an always-true 1=1 (you probably wanted '
1110 . "to say ...{ \$inequality_op => [ -and => \@values ] }... instead)"
1114 # distribute $op over each remaining member of @vals, append logic if exists
1115 return $self->_recurse_where([map { {$k => {$op, $_}} } @vals], $logic);
1119 # try to DWIM on equality operators
1121 $op =~ $self->{equality_op} ? $self->{sqlfalse}
1122 : $op =~ $self->{like_op} ? belch("Supplying an empty arrayref to '@{[ uc $op]}' is deprecated") && $self->{sqlfalse}
1123 : $op =~ $self->{inequality_op} ? $self->{sqltrue}
1124 : $op =~ $self->{not_like_op} ? belch("Supplying an empty arrayref to '@{[ uc $op]}' is deprecated") && $self->{sqltrue}
1125 : puke "operator '$op' applied on an empty array (field '$k')";
1130 sub _where_hashpair_SCALARREF {
1131 my ($self, $k, $v) = @_;
1132 $self->_debug("SCALAR($k) means literal SQL: $$v");
1133 my $sql = $self->_quote($k) . " " . $$v;
1137 # literal SQL with bind
1138 sub _where_hashpair_ARRAYREFREF {
1139 my ($self, $k, $v) = @_;
1140 $self->_debug("REF($k) means literal SQL: @${$v}");
1141 my ($sql, @bind) = @$$v;
1142 $self->_assert_bindval_matches_bindtype(@bind);
1143 $sql = $self->_quote($k) . " " . $sql;
1144 return ($sql, @bind );
1147 # literal SQL without bind
1148 sub _where_hashpair_SCALAR {
1149 my ($self, $k, $v) = @_;
1150 $self->_debug("NOREF($k) means simple key=val: $k $self->{cmp} $v");
1151 return ($self->_where_hashpair_HASHREF($k, { $self->{cmp} => $v }));
1155 sub _where_hashpair_UNDEF {
1156 my ($self, $k, $v) = @_;
1157 $self->_debug("UNDEF($k) means IS NULL");
1158 return $self->_where_hashpair_HASHREF($k, { -is => undef });
1161 #======================================================================
1162 # WHERE: TOP-LEVEL OTHERS (SCALARREF, SCALAR, UNDEF)
1163 #======================================================================
1166 sub _where_SCALARREF {
1167 my ($self, $where) = @_;
1170 $self->_debug("SCALAR(*top) means literal SQL: $$where");
1176 my ($self, $where) = @_;
1179 $self->_debug("NOREF(*top) means literal SQL: $where");
1190 #======================================================================
1191 # WHERE: BUILTIN SPECIAL OPERATORS (-in, -between)
1192 #======================================================================
1195 sub _where_field_BETWEEN {
1196 my ($self, $k, $op, $vals) = @_;
1198 my ($label, $and, $placeholder);
1199 $label = $self->_convert($self->_quote($k));
1200 $and = ' ' . $self->_sqlcase('and') . ' ';
1201 $placeholder = $self->_convert('?');
1202 $op = $self->_sqlcase($op);
1204 my $invalid_args = "Operator '$op' requires either an arrayref with two defined values or expressions, or a single literal scalarref/arrayref-ref";
1206 my ($clause, @bind) = $self->_SWITCH_refkind($vals, {
1207 ARRAYREFREF => sub {
1208 my ($s, @b) = @$$vals;
1209 $self->_assert_bindval_matches_bindtype(@b);
1216 puke $invalid_args if @$vals != 2;
1218 my (@all_sql, @all_bind);
1219 foreach my $val (@$vals) {
1220 my ($sql, @bind) = $self->_SWITCH_refkind($val, {
1222 return ($placeholder, $self->_bindtype($k, $val) );
1227 ARRAYREFREF => sub {
1228 my ($sql, @bind) = @$$val;
1229 $self->_assert_bindval_matches_bindtype(@bind);
1230 return ($sql, @bind);
1233 my ($func, $arg, @rest) = %$val;
1234 puke "Only simple { -func => arg } functions accepted as sub-arguments to BETWEEN"
1235 if (@rest or $func !~ /^ \- (.+)/x);
1236 $self->_where_unary_op($1 => $arg);
1242 push @all_sql, $sql;
1243 push @all_bind, @bind;
1247 (join $and, @all_sql),
1256 my $sql = "( $label $op $clause )";
1257 return ($sql, @bind)
1261 sub _where_field_IN {
1262 my ($self, $k, $op, $vals) = @_;
1264 # backwards compatibility: if scalar, force into an arrayref
1265 $vals = [$vals] if defined $vals && ! ref $vals;
1267 my ($label) = $self->_convert($self->_quote($k));
1268 my ($placeholder) = $self->_convert('?');
1269 $op = $self->_sqlcase($op);
1271 my ($sql, @bind) = $self->_SWITCH_refkind($vals, {
1272 ARRAYREF => sub { # list of choices
1273 if (@$vals) { # nonempty list
1274 my (@all_sql, @all_bind);
1276 for my $val (@$vals) {
1277 my ($sql, @bind) = $self->_SWITCH_refkind($val, {
1279 return ($placeholder, $val);
1284 ARRAYREFREF => sub {
1285 my ($sql, @bind) = @$$val;
1286 $self->_assert_bindval_matches_bindtype(@bind);
1287 return ($sql, @bind);
1290 my ($func, $arg, @rest) = %$val;
1291 puke "Only simple { -func => arg } functions accepted as sub-arguments to IN"
1292 if (@rest or $func !~ /^ \- (.+)/x);
1293 $self->_where_unary_op($1 => $arg);
1297 'SQL::Abstract before v1.75 used to generate incorrect SQL when the '
1298 . "-$op operator was given an undef-containing list: !!!AUDIT YOUR CODE "
1299 . 'AND DATA!!! (the upcoming Data::Query-based version of SQL::Abstract '
1300 . 'will emit the logically correct SQL instead of raising this exception)'
1304 push @all_sql, $sql;
1305 push @all_bind, @bind;
1309 sprintf('%s %s ( %s )',
1312 join(', ', @all_sql)
1314 $self->_bindtype($k, @all_bind),
1317 else { # empty list: some databases won't understand "IN ()", so DWIM
1318 my $sql = ($op =~ /\bnot\b/i) ? $self->{sqltrue} : $self->{sqlfalse};
1323 SCALARREF => sub { # literal SQL
1324 my $sql = $self->_open_outer_paren($$vals);
1325 return ("$label $op ( $sql )");
1327 ARRAYREFREF => sub { # literal SQL with bind
1328 my ($sql, @bind) = @$$vals;
1329 $self->_assert_bindval_matches_bindtype(@bind);
1330 $sql = $self->_open_outer_paren($sql);
1331 return ("$label $op ( $sql )", @bind);
1335 puke "Argument passed to the '$op' operator can not be undefined";
1339 puke "special op $op requires an arrayref (or scalarref/arrayref-ref)";
1343 return ($sql, @bind);
1346 # Some databases (SQLite) treat col IN (1, 2) different from
1347 # col IN ( (1, 2) ). Use this to strip all outer parens while
1348 # adding them back in the corresponding method
1349 sub _open_outer_paren {
1350 my ($self, $sql) = @_;
1352 while (my ($inner) = $sql =~ /^ \s* \( (.*) \) \s* $/xs) {
1354 # there are closing parens inside, need the heavy duty machinery
1355 # to reevaluate the extraction starting from $sql (full reevaluation)
1356 if ($inner =~ /\)/) {
1357 require Text::Balanced;
1359 my (undef, $remainder) = do {
1360 # idiotic design - writes to $@ but *DOES NOT* throw exceptions
1362 Text::Balanced::extract_bracketed($sql, '()', qr/\s*/);
1365 # the entire expression needs to be a balanced bracketed thing
1366 # (after an extract no remainder sans trailing space)
1367 last if defined $remainder and $remainder =~ /\S/;
1377 #======================================================================
1379 #======================================================================
1382 my ($self, $arg) = @_;
1385 for my $c ($self->_order_by_chunks($arg) ) {
1386 $self->_SWITCH_refkind($c, {
1387 SCALAR => sub { push @sql, $c },
1388 ARRAYREF => sub { push @sql, shift @$c; push @bind, @$c },
1394 $self->_sqlcase(' order by'),
1400 return wantarray ? ($sql, @bind) : $sql;
1403 sub _order_by_chunks {
1404 my ($self, $arg) = @_;
1406 return $self->_SWITCH_refkind($arg, {
1409 map { $self->_order_by_chunks($_ ) } @$arg;
1412 ARRAYREFREF => sub {
1413 my ($s, @b) = @$$arg;
1414 $self->_assert_bindval_matches_bindtype(@b);
1418 SCALAR => sub {$self->_quote($arg)},
1420 UNDEF => sub {return () },
1422 SCALARREF => sub {$$arg}, # literal SQL, no quoting
1425 # get first pair in hash
1426 my ($key, $val, @rest) = %$arg;
1428 return () unless $key;
1430 if (@rest or not $key =~ /^-(desc|asc)/i) {
1431 puke "hash passed to _order_by must have exactly one key (-desc or -asc)";
1437 for my $c ($self->_order_by_chunks($val)) {
1440 $self->_SWITCH_refkind($c, {
1445 ($sql, @bind) = @$c;
1449 $sql = $sql . ' ' . $self->_sqlcase($direction);
1451 push @ret, [ $sql, @bind];
1460 #======================================================================
1461 # DATASOURCE (FOR NOW, JUST PLAIN TABLE OR LIST OF TABLES)
1462 #======================================================================
1467 $self->_SWITCH_refkind($from, {
1468 ARRAYREF => sub {join ', ', map { $self->_quote($_) } @$from;},
1469 SCALAR => sub {$self->_quote($from)},
1470 SCALARREF => sub {$$from},
1475 #======================================================================
1477 #======================================================================
1479 # highly optimized, as it's called way too often
1481 # my ($self, $label) = @_;
1483 return '' unless defined $_[1];
1484 return ${$_[1]} if ref($_[1]) eq 'SCALAR';
1486 $_[0]->{quote_char} or
1487 ($_[0]->_assert_pass_injection_guard($_[1]), return $_[1]);
1489 my $qref = ref $_[0]->{quote_char};
1491 !$qref ? ($_[0]->{quote_char}, $_[0]->{quote_char})
1492 : ($qref eq 'ARRAY') ? @{$_[0]->{quote_char}}
1493 : puke "Unsupported quote_char format: $_[0]->{quote_char}";
1495 my $esc = $_[0]->{escape_char} || $r;
1497 # parts containing * are naturally unquoted
1498 return join($_[0]->{name_sep}||'', map
1499 +( $_ eq '*' ? $_ : do { (my $n = $_) =~ s/(\Q$esc\E|\Q$r\E)/$esc$1/g; $l . $n . $r } ),
1500 ( $_[0]->{name_sep} ? split (/\Q$_[0]->{name_sep}\E/, $_[1] ) : $_[1] )
1505 # Conversion, if applicable
1507 #my ($self, $arg) = @_;
1508 if ($_[0]->{convert}) {
1509 return $_[0]->_sqlcase($_[0]->{convert}) .'(' . $_[1] . ')';
1516 #my ($self, $col, @vals) = @_;
1517 # called often - tighten code
1518 return $_[0]->{bindtype} eq 'columns'
1519 ? map {[$_[1], $_]} @_[2 .. $#_]
1524 # Dies if any element of @bind is not in [colname => value] format
1525 # if bindtype is 'columns'.
1526 sub _assert_bindval_matches_bindtype {
1527 # my ($self, @bind) = @_;
1529 if ($self->{bindtype} eq 'columns') {
1531 if (!defined $_ || ref($_) ne 'ARRAY' || @$_ != 2) {
1532 puke "bindtype 'columns' selected, you need to pass: [column_name => bind_value]"
1538 sub _join_sql_clauses {
1539 my ($self, $logic, $clauses_aref, $bind_aref) = @_;
1541 if (@$clauses_aref > 1) {
1542 my $join = " " . $self->_sqlcase($logic) . " ";
1543 my $sql = '( ' . join($join, @$clauses_aref) . ' )';
1544 return ($sql, @$bind_aref);
1546 elsif (@$clauses_aref) {
1547 return ($clauses_aref->[0], @$bind_aref); # no parentheses
1550 return (); # if no SQL, ignore @$bind_aref
1555 # Fix SQL case, if so requested
1557 # LDNOTE: if $self->{case} is true, then it contains 'lower', so we
1558 # don't touch the argument ... crooked logic, but let's not change it!
1559 return $_[0]->{case} ? $_[1] : uc($_[1]);
1563 #======================================================================
1564 # DISPATCHING FROM REFKIND
1565 #======================================================================
1568 my ($self, $data) = @_;
1570 return 'UNDEF' unless defined $data;
1572 # blessed objects are treated like scalars
1573 my $ref = (Scalar::Util::blessed $data) ? '' : ref $data;
1575 return 'SCALAR' unless $ref;
1578 while ($ref eq 'REF') {
1580 $ref = (Scalar::Util::blessed $data) ? '' : ref $data;
1584 return ($ref||'SCALAR') . ('REF' x $n_steps);
1588 my ($self, $data) = @_;
1589 my @try = ($self->_refkind($data));
1590 push @try, 'SCALAR_or_UNDEF' if $try[0] eq 'SCALAR' || $try[0] eq 'UNDEF';
1591 push @try, 'FALLBACK';
1595 sub _METHOD_FOR_refkind {
1596 my ($self, $meth_prefix, $data) = @_;
1599 for (@{$self->_try_refkind($data)}) {
1600 $method = $self->can($meth_prefix."_".$_)
1604 return $method || puke "cannot dispatch on '$meth_prefix' for ".$self->_refkind($data);
1608 sub _SWITCH_refkind {
1609 my ($self, $data, $dispatch_table) = @_;
1612 for (@{$self->_try_refkind($data)}) {
1613 $coderef = $dispatch_table->{$_}
1617 puke "no dispatch entry for ".$self->_refkind($data)
1626 #======================================================================
1627 # VALUES, GENERATE, AUTOLOAD
1628 #======================================================================
1630 # LDNOTE: original code from nwiger, didn't touch code in that section
1631 # I feel the AUTOLOAD stuff should not be the default, it should
1632 # only be activated on explicit demand by user.
1636 my $data = shift || return;
1637 puke "Argument to ", __PACKAGE__, "->values must be a \\%hash"
1638 unless ref $data eq 'HASH';
1641 foreach my $k (sort keys %$data) {
1642 my $v = $data->{$k};
1643 $self->_SWITCH_refkind($v, {
1645 if ($self->{array_datatypes}) { # array datatype
1646 push @all_bind, $self->_bindtype($k, $v);
1648 else { # literal SQL with bind
1649 my ($sql, @bind) = @$v;
1650 $self->_assert_bindval_matches_bindtype(@bind);
1651 push @all_bind, @bind;
1654 ARRAYREFREF => sub { # literal SQL with bind
1655 my ($sql, @bind) = @${$v};
1656 $self->_assert_bindval_matches_bindtype(@bind);
1657 push @all_bind, @bind;
1659 SCALARREF => sub { # literal SQL without bind
1661 SCALAR_or_UNDEF => sub {
1662 push @all_bind, $self->_bindtype($k, $v);
1673 my(@sql, @sqlq, @sqlv);
1677 if ($ref eq 'HASH') {
1678 for my $k (sort keys %$_) {
1681 my $label = $self->_quote($k);
1682 if ($r eq 'ARRAY') {
1683 # literal SQL with bind
1684 my ($sql, @bind) = @$v;
1685 $self->_assert_bindval_matches_bindtype(@bind);
1686 push @sqlq, "$label = $sql";
1688 } elsif ($r eq 'SCALAR') {
1689 # literal SQL without bind
1690 push @sqlq, "$label = $$v";
1692 push @sqlq, "$label = ?";
1693 push @sqlv, $self->_bindtype($k, $v);
1696 push @sql, $self->_sqlcase('set'), join ', ', @sqlq;
1697 } elsif ($ref eq 'ARRAY') {
1698 # unlike insert(), assume these are ONLY the column names, i.e. for SQL
1701 if ($r eq 'ARRAY') { # literal SQL with bind
1702 my ($sql, @bind) = @$v;
1703 $self->_assert_bindval_matches_bindtype(@bind);
1706 } elsif ($r eq 'SCALAR') { # literal SQL without bind
1707 # embedded literal SQL
1714 push @sql, '(' . join(', ', @sqlq) . ')';
1715 } elsif ($ref eq 'SCALAR') {
1719 # strings get case twiddled
1720 push @sql, $self->_sqlcase($_);
1724 my $sql = join ' ', @sql;
1726 # this is pretty tricky
1727 # if ask for an array, return ($stmt, @bind)
1728 # otherwise, s/?/shift @sqlv/ to put it inline
1730 return ($sql, @sqlv);
1732 1 while $sql =~ s/\?/my $d = shift(@sqlv);
1733 ref $d ? $d->[1] : $d/e;
1742 # This allows us to check for a local, then _form, attr
1744 my($name) = $AUTOLOAD =~ /.*::(.+)/;
1745 return $self->generate($name, @_);
1756 SQL::Abstract - Generate SQL from Perl data structures
1762 my $sql = SQL::Abstract->new;
1764 my($stmt, @bind) = $sql->select($source, \@fields, \%where, $order);
1766 my($stmt, @bind) = $sql->insert($table, \%fieldvals || \@values);
1768 my($stmt, @bind) = $sql->update($table, \%fieldvals, \%where);
1770 my($stmt, @bind) = $sql->delete($table, \%where);
1772 # Then, use these in your DBI statements
1773 my $sth = $dbh->prepare($stmt);
1774 $sth->execute(@bind);
1776 # Just generate the WHERE clause
1777 my($stmt, @bind) = $sql->where(\%where, $order);
1779 # Return values in the same order, for hashed queries
1780 # See PERFORMANCE section for more details
1781 my @bind = $sql->values(\%fieldvals);
1785 This module was inspired by the excellent L<DBIx::Abstract>.
1786 However, in using that module I found that what I really wanted
1787 to do was generate SQL, but still retain complete control over my
1788 statement handles and use the DBI interface. So, I set out to
1789 create an abstract SQL generation module.
1791 While based on the concepts used by L<DBIx::Abstract>, there are
1792 several important differences, especially when it comes to WHERE
1793 clauses. I have modified the concepts used to make the SQL easier
1794 to generate from Perl data structures and, IMO, more intuitive.
1795 The underlying idea is for this module to do what you mean, based
1796 on the data structures you provide it. The big advantage is that
1797 you don't have to modify your code every time your data changes,
1798 as this module figures it out.
1800 To begin with, an SQL INSERT is as easy as just specifying a hash
1801 of C<key=value> pairs:
1804 name => 'Jimbo Bobson',
1805 phone => '123-456-7890',
1806 address => '42 Sister Lane',
1807 city => 'St. Louis',
1808 state => 'Louisiana',
1811 The SQL can then be generated with this:
1813 my($stmt, @bind) = $sql->insert('people', \%data);
1815 Which would give you something like this:
1817 $stmt = "INSERT INTO people
1818 (address, city, name, phone, state)
1819 VALUES (?, ?, ?, ?, ?)";
1820 @bind = ('42 Sister Lane', 'St. Louis', 'Jimbo Bobson',
1821 '123-456-7890', 'Louisiana');
1823 These are then used directly in your DBI code:
1825 my $sth = $dbh->prepare($stmt);
1826 $sth->execute(@bind);
1828 =head2 Inserting and Updating Arrays
1830 If your database has array types (like for example Postgres),
1831 activate the special option C<< array_datatypes => 1 >>
1832 when creating the C<SQL::Abstract> object.
1833 Then you may use an arrayref to insert and update database array types:
1835 my $sql = SQL::Abstract->new(array_datatypes => 1);
1837 planets => [qw/Mercury Venus Earth Mars/]
1840 my($stmt, @bind) = $sql->insert('solar_system', \%data);
1844 $stmt = "INSERT INTO solar_system (planets) VALUES (?)"
1846 @bind = (['Mercury', 'Venus', 'Earth', 'Mars']);
1849 =head2 Inserting and Updating SQL
1851 In order to apply SQL functions to elements of your C<%data> you may
1852 specify a reference to an arrayref for the given hash value. For example,
1853 if you need to execute the Oracle C<to_date> function on a value, you can
1854 say something like this:
1858 date_entered => \[ "to_date(?,'MM/DD/YYYY')", "03/02/2003" ],
1861 The first value in the array is the actual SQL. Any other values are
1862 optional and would be included in the bind values array. This gives
1865 my($stmt, @bind) = $sql->insert('people', \%data);
1867 $stmt = "INSERT INTO people (name, date_entered)
1868 VALUES (?, to_date(?,'MM/DD/YYYY'))";
1869 @bind = ('Bill', '03/02/2003');
1871 An UPDATE is just as easy, all you change is the name of the function:
1873 my($stmt, @bind) = $sql->update('people', \%data);
1875 Notice that your C<%data> isn't touched; the module will generate
1876 the appropriately quirky SQL for you automatically. Usually you'll
1877 want to specify a WHERE clause for your UPDATE, though, which is
1878 where handling C<%where> hashes comes in handy...
1880 =head2 Complex where statements
1882 This module can generate pretty complicated WHERE statements
1883 easily. For example, simple C<key=value> pairs are taken to mean
1884 equality, and if you want to see if a field is within a set
1885 of values, you can use an arrayref. Let's say we wanted to
1886 SELECT some data based on this criteria:
1889 requestor => 'inna',
1890 worker => ['nwiger', 'rcwe', 'sfz'],
1891 status => { '!=', 'completed' }
1894 my($stmt, @bind) = $sql->select('tickets', '*', \%where);
1896 The above would give you something like this:
1898 $stmt = "SELECT * FROM tickets WHERE
1899 ( requestor = ? ) AND ( status != ? )
1900 AND ( worker = ? OR worker = ? OR worker = ? )";
1901 @bind = ('inna', 'completed', 'nwiger', 'rcwe', 'sfz');
1903 Which you could then use in DBI code like so:
1905 my $sth = $dbh->prepare($stmt);
1906 $sth->execute(@bind);
1912 The methods are simple. There's one for every major SQL operation,
1913 and a constructor you use first. The arguments are specified in a
1914 similar order for each method (table, then fields, then a where
1915 clause) to try and simplify things.
1917 =head2 new(option => 'value')
1919 The C<new()> function takes a list of options and values, and returns
1920 a new B<SQL::Abstract> object which can then be used to generate SQL
1921 through the methods below. The options accepted are:
1927 If set to 'lower', then SQL will be generated in all lowercase. By
1928 default SQL is generated in "textbook" case meaning something like:
1930 SELECT a_field FROM a_table WHERE some_field LIKE '%someval%'
1932 Any setting other than 'lower' is ignored.
1936 This determines what the default comparison operator is. By default
1937 it is C<=>, meaning that a hash like this:
1939 %where = (name => 'nwiger', email => 'nate@wiger.org');
1941 Will generate SQL like this:
1943 WHERE name = 'nwiger' AND email = 'nate@wiger.org'
1945 However, you may want loose comparisons by default, so if you set
1946 C<cmp> to C<like> you would get SQL such as:
1948 WHERE name like 'nwiger' AND email like 'nate@wiger.org'
1950 You can also override the comparison on an individual basis - see
1951 the huge section on L</"WHERE CLAUSES"> at the bottom.
1953 =item sqltrue, sqlfalse
1955 Expressions for inserting boolean values within SQL statements.
1956 By default these are C<1=1> and C<1=0>. They are used
1957 by the special operators C<-in> and C<-not_in> for generating
1958 correct SQL even when the argument is an empty array (see below).
1962 This determines the default logical operator for multiple WHERE
1963 statements in arrays or hashes. If absent, the default logic is "or"
1964 for arrays, and "and" for hashes. This means that a WHERE
1968 event_date => {'>=', '2/13/99'},
1969 event_date => {'<=', '4/24/03'},
1972 will generate SQL like this:
1974 WHERE event_date >= '2/13/99' OR event_date <= '4/24/03'
1976 This is probably not what you want given this query, though (look
1977 at the dates). To change the "OR" to an "AND", simply specify:
1979 my $sql = SQL::Abstract->new(logic => 'and');
1981 Which will change the above C<WHERE> to:
1983 WHERE event_date >= '2/13/99' AND event_date <= '4/24/03'
1985 The logic can also be changed locally by inserting
1986 a modifier in front of an arrayref:
1988 @where = (-and => [event_date => {'>=', '2/13/99'},
1989 event_date => {'<=', '4/24/03'} ]);
1991 See the L</"WHERE CLAUSES"> section for explanations.
1995 This will automatically convert comparisons using the specified SQL
1996 function for both column and value. This is mostly used with an argument
1997 of C<upper> or C<lower>, so that the SQL will have the effect of
1998 case-insensitive "searches". For example, this:
2000 $sql = SQL::Abstract->new(convert => 'upper');
2001 %where = (keywords => 'MaKe iT CAse inSeNSItive');
2003 Will turn out the following SQL:
2005 WHERE upper(keywords) like upper('MaKe iT CAse inSeNSItive')
2007 The conversion can be C<upper()>, C<lower()>, or any other SQL function
2008 that can be applied symmetrically to fields (actually B<SQL::Abstract> does
2009 not validate this option; it will just pass through what you specify verbatim).
2013 This is a kludge because many databases suck. For example, you can't
2014 just bind values using DBI's C<execute()> for Oracle C<CLOB> or C<BLOB> fields.
2015 Instead, you have to use C<bind_param()>:
2017 $sth->bind_param(1, 'reg data');
2018 $sth->bind_param(2, $lots, {ora_type => ORA_CLOB});
2020 The problem is, B<SQL::Abstract> will normally just return a C<@bind> array,
2021 which loses track of which field each slot refers to. Fear not.
2023 If you specify C<bindtype> in new, you can determine how C<@bind> is returned.
2024 Currently, you can specify either C<normal> (default) or C<columns>. If you
2025 specify C<columns>, you will get an array that looks like this:
2027 my $sql = SQL::Abstract->new(bindtype => 'columns');
2028 my($stmt, @bind) = $sql->insert(...);
2031 [ 'column1', 'value1' ],
2032 [ 'column2', 'value2' ],
2033 [ 'column3', 'value3' ],
2036 You can then iterate through this manually, using DBI's C<bind_param()>.
2038 $sth->prepare($stmt);
2041 my($col, $data) = @$_;
2042 if ($col eq 'details' || $col eq 'comments') {
2043 $sth->bind_param($i, $data, {ora_type => ORA_CLOB});
2044 } elsif ($col eq 'image') {
2045 $sth->bind_param($i, $data, {ora_type => ORA_BLOB});
2047 $sth->bind_param($i, $data);
2051 $sth->execute; # execute without @bind now
2053 Now, why would you still use B<SQL::Abstract> if you have to do this crap?
2054 Basically, the advantage is still that you don't have to care which fields
2055 are or are not included. You could wrap that above C<for> loop in a simple
2056 sub called C<bind_fields()> or something and reuse it repeatedly. You still
2057 get a layer of abstraction over manual SQL specification.
2059 Note that if you set L</bindtype> to C<columns>, the C<\[ $sql, @bind ]>
2060 construct (see L</Literal SQL with placeholders and bind values (subqueries)>)
2061 will expect the bind values in this format.
2065 This is the character that a table or column name will be quoted
2066 with. By default this is an empty string, but you could set it to
2067 the character C<`>, to generate SQL like this:
2069 SELECT `a_field` FROM `a_table` WHERE `some_field` LIKE '%someval%'
2071 Alternatively, you can supply an array ref of two items, the first being the left
2072 hand quote character, and the second the right hand quote character. For
2073 example, you could supply C<['[',']']> for SQL Server 2000 compliant quotes
2074 that generates SQL like this:
2076 SELECT [a_field] FROM [a_table] WHERE [some_field] LIKE '%someval%'
2078 Quoting is useful if you have tables or columns names that are reserved
2079 words in your database's SQL dialect.
2083 This is the character that will be used to escape L</quote_char>s appearing
2084 in an identifier before it has been quoted.
2086 The parameter default in case of a single L</quote_char> character is the quote
2089 When opening-closing-style quoting is used (L</quote_char> is an arrayref)
2090 this parameter defaults to the B<closing (right)> L</quote_char>. Occurrences
2091 of the B<opening (left)> L</quote_char> within the identifier are currently left
2092 untouched. The default for opening-closing-style quotes may change in future
2093 versions, thus you are B<strongly encouraged> to specify the escape character
2098 This is the character that separates a table and column name. It is
2099 necessary to specify this when the C<quote_char> option is selected,
2100 so that tables and column names can be individually quoted like this:
2102 SELECT `table`.`one_field` FROM `table` WHERE `table`.`other_field` = 1
2104 =item injection_guard
2106 A regular expression C<qr/.../> that is applied to any C<-function> and unquoted
2107 column name specified in a query structure. This is a safety mechanism to avoid
2108 injection attacks when mishandling user input e.g.:
2110 my %condition_as_column_value_pairs = get_values_from_user();
2111 $sqla->select( ... , \%condition_as_column_value_pairs );
2113 If the expression matches an exception is thrown. Note that literal SQL
2114 supplied via C<\'...'> or C<\['...']> is B<not> checked in any way.
2116 Defaults to checking for C<;> and the C<GO> keyword (TransactSQL)
2118 =item array_datatypes
2120 When this option is true, arrayrefs in INSERT or UPDATE are
2121 interpreted as array datatypes and are passed directly
2123 When this option is false, arrayrefs are interpreted
2124 as literal SQL, just like refs to arrayrefs
2125 (but this behavior is for backwards compatibility; when writing
2126 new queries, use the "reference to arrayref" syntax
2132 Takes a reference to a list of "special operators"
2133 to extend the syntax understood by L<SQL::Abstract>.
2134 See section L</"SPECIAL OPERATORS"> for details.
2138 Takes a reference to a list of "unary operators"
2139 to extend the syntax understood by L<SQL::Abstract>.
2140 See section L</"UNARY OPERATORS"> for details.
2146 =head2 insert($table, \@values || \%fieldvals, \%options)
2148 This is the simplest function. You simply give it a table name
2149 and either an arrayref of values or hashref of field/value pairs.
2150 It returns an SQL INSERT statement and a list of bind values.
2151 See the sections on L</"Inserting and Updating Arrays"> and
2152 L</"Inserting and Updating SQL"> for information on how to insert
2153 with those data types.
2155 The optional C<\%options> hash reference may contain additional
2156 options to generate the insert SQL. Currently supported options
2163 Takes either a scalar of raw SQL fields, or an array reference of
2164 field names, and adds on an SQL C<RETURNING> statement at the end.
2165 This allows you to return data generated by the insert statement
2166 (such as row IDs) without performing another C<SELECT> statement.
2167 Note, however, this is not part of the SQL standard and may not
2168 be supported by all database engines.
2172 =head2 update($table, \%fieldvals, \%where, \%options)
2174 This takes a table, hashref of field/value pairs, and an optional
2175 hashref L<WHERE clause|/WHERE CLAUSES>. It returns an SQL UPDATE function and a list
2177 See the sections on L</"Inserting and Updating Arrays"> and
2178 L</"Inserting and Updating SQL"> for information on how to insert
2179 with those data types.
2181 The optional C<\%options> hash reference may contain additional
2182 options to generate the update SQL. Currently supported options
2189 See the C<returning> option to
2190 L<insert|/insert($table, \@values || \%fieldvals, \%options)>.
2194 =head2 select($source, $fields, $where, $order)
2196 This returns a SQL SELECT statement and associated list of bind values, as
2197 specified by the arguments:
2203 Specification of the 'FROM' part of the statement.
2204 The argument can be either a plain scalar (interpreted as a table
2205 name, will be quoted), or an arrayref (interpreted as a list
2206 of table names, joined by commas, quoted), or a scalarref
2207 (literal SQL, not quoted).
2211 Specification of the list of fields to retrieve from
2213 The argument can be either an arrayref (interpreted as a list
2214 of field names, will be joined by commas and quoted), or a
2215 plain scalar (literal SQL, not quoted).
2216 Please observe that this API is not as flexible as that of
2217 the first argument C<$source>, for backwards compatibility reasons.
2221 Optional argument to specify the WHERE part of the query.
2222 The argument is most often a hashref, but can also be
2223 an arrayref or plain scalar --
2224 see section L<WHERE clause|/"WHERE CLAUSES"> for details.
2228 Optional argument to specify the ORDER BY part of the query.
2229 The argument can be a scalar, a hashref or an arrayref
2230 -- see section L<ORDER BY clause|/"ORDER BY CLAUSES">
2236 =head2 delete($table, \%where, \%options)
2238 This takes a table name and optional hashref L<WHERE clause|/WHERE CLAUSES>.
2239 It returns an SQL DELETE statement and list of bind values.
2241 The optional C<\%options> hash reference may contain additional
2242 options to generate the delete SQL. Currently supported options
2249 See the C<returning> option to
2250 L<insert|/insert($table, \@values || \%fieldvals, \%options)>.
2254 =head2 where(\%where, $order)
2256 This is used to generate just the WHERE clause. For example,
2257 if you have an arbitrary data structure and know what the
2258 rest of your SQL is going to look like, but want an easy way
2259 to produce a WHERE clause, use this. It returns an SQL WHERE
2260 clause and list of bind values.
2263 =head2 values(\%data)
2265 This just returns the values from the hash C<%data>, in the same
2266 order that would be returned from any of the other above queries.
2267 Using this allows you to markedly speed up your queries if you
2268 are affecting lots of rows. See below under the L</"PERFORMANCE"> section.
2270 =head2 generate($any, 'number', $of, \@data, $struct, \%types)
2272 Warning: This is an experimental method and subject to change.
2274 This returns arbitrarily generated SQL. It's a really basic shortcut.
2275 It will return two different things, depending on return context:
2277 my($stmt, @bind) = $sql->generate('create table', \$table, \@fields);
2278 my $stmt_and_val = $sql->generate('create table', \$table, \@fields);
2280 These would return the following:
2282 # First calling form
2283 $stmt = "CREATE TABLE test (?, ?)";
2284 @bind = (field1, field2);
2286 # Second calling form
2287 $stmt_and_val = "CREATE TABLE test (field1, field2)";
2289 Depending on what you're trying to do, it's up to you to choose the correct
2290 format. In this example, the second form is what you would want.
2294 $sql->generate('alter session', { nls_date_format => 'MM/YY' });
2298 ALTER SESSION SET nls_date_format = 'MM/YY'
2300 You get the idea. Strings get their case twiddled, but everything
2301 else remains verbatim.
2303 =head1 EXPORTABLE FUNCTIONS
2305 =head2 is_plain_value
2307 Determines if the supplied argument is a plain value as understood by this
2312 =item * The value is C<undef>
2314 =item * The value is a non-reference
2316 =item * The value is an object with stringification overloading
2318 =item * The value is of the form C<< { -value => $anything } >>
2322 On failure returns C<undef>, on success returns a B<scalar> reference
2323 to the original supplied argument.
2329 The stringification overloading detection is rather advanced: it takes
2330 into consideration not only the presence of a C<""> overload, but if that
2331 fails also checks for enabled
2332 L<autogenerated versions of C<"">|overload/Magic Autogeneration>, based
2333 on either C<0+> or C<bool>.
2335 Unfortunately testing in the field indicates that this
2336 detection B<< may tickle a latent bug in perl versions before 5.018 >>,
2337 but only when very large numbers of stringifying objects are involved.
2338 At the time of writing ( Sep 2014 ) there is no clear explanation of
2339 the direct cause, nor is there a manageably small test case that reliably
2340 reproduces the problem.
2342 If you encounter any of the following exceptions in B<random places within
2343 your application stack> - this module may be to blame:
2345 Operation "ne": no method found,
2346 left argument in overloaded package <something>,
2347 right argument in overloaded package <something>
2351 Stub found while resolving method "???" overloading """" in package <something>
2353 If you fall victim to the above - please attempt to reduce the problem
2354 to something that could be sent to the L<SQL::Abstract developers
2355 |DBIx::Class/GETTING HELP/SUPPORT>
2356 (either publicly or privately). As a workaround in the meantime you can
2357 set C<$ENV{SQLA_ISVALUE_IGNORE_AUTOGENERATED_STRINGIFICATION}> to a true
2358 value, which will most likely eliminate your problem (at the expense of
2359 not being able to properly detect exotic forms of stringification).
2361 This notice and environment variable will be removed in a future version,
2362 as soon as the underlying problem is found and a reliable workaround is
2367 =head2 is_literal_value
2369 Determines if the supplied argument is a literal value as understood by this
2374 =item * C<\$sql_string>
2376 =item * C<\[ $sql_string, @bind_values ]>
2380 On failure returns C<undef>, on success returns an B<array> reference
2381 containing the unpacked version of the supplied literal SQL and bind values.
2383 =head1 WHERE CLAUSES
2387 This module uses a variation on the idea from L<DBIx::Abstract>. It
2388 is B<NOT>, repeat I<not> 100% compatible. B<The main logic of this
2389 module is that things in arrays are OR'ed, and things in hashes
2392 The easiest way to explain is to show lots of examples. After
2393 each C<%where> hash shown, it is assumed you used:
2395 my($stmt, @bind) = $sql->where(\%where);
2397 However, note that the C<%where> hash can be used directly in any
2398 of the other functions as well, as described above.
2400 =head2 Key-value pairs
2402 So, let's get started. To begin, a simple hash:
2406 status => 'completed'
2409 Is converted to SQL C<key = val> statements:
2411 $stmt = "WHERE user = ? AND status = ?";
2412 @bind = ('nwiger', 'completed');
2414 One common thing I end up doing is having a list of values that
2415 a field can be in. To do this, simply specify a list inside of
2420 status => ['assigned', 'in-progress', 'pending'];
2423 This simple code will create the following:
2425 $stmt = "WHERE user = ? AND ( status = ? OR status = ? OR status = ? )";
2426 @bind = ('nwiger', 'assigned', 'in-progress', 'pending');
2428 A field associated to an empty arrayref will be considered a
2429 logical false and will generate 0=1.
2431 =head2 Tests for NULL values
2433 If the value part is C<undef> then this is converted to SQL <IS NULL>
2442 $stmt = "WHERE user = ? AND status IS NULL";
2445 To test if a column IS NOT NULL:
2449 status => { '!=', undef },
2452 =head2 Specific comparison operators
2454 If you want to specify a different type of operator for your comparison,
2455 you can use a hashref for a given column:
2459 status => { '!=', 'completed' }
2462 Which would generate:
2464 $stmt = "WHERE user = ? AND status != ?";
2465 @bind = ('nwiger', 'completed');
2467 To test against multiple values, just enclose the values in an arrayref:
2469 status => { '=', ['assigned', 'in-progress', 'pending'] };
2471 Which would give you:
2473 "WHERE status = ? OR status = ? OR status = ?"
2476 The hashref can also contain multiple pairs, in which case it is expanded
2477 into an C<AND> of its elements:
2481 status => { '!=', 'completed', -not_like => 'pending%' }
2484 # Or more dynamically, like from a form
2485 $where{user} = 'nwiger';
2486 $where{status}{'!='} = 'completed';
2487 $where{status}{'-not_like'} = 'pending%';
2489 # Both generate this
2490 $stmt = "WHERE user = ? AND status != ? AND status NOT LIKE ?";
2491 @bind = ('nwiger', 'completed', 'pending%');
2494 To get an OR instead, you can combine it with the arrayref idea:
2498 priority => [ { '=', 2 }, { '>', 5 } ]
2501 Which would generate:
2503 $stmt = "WHERE ( priority = ? OR priority > ? ) AND user = ?";
2504 @bind = ('2', '5', 'nwiger');
2506 If you want to include literal SQL (with or without bind values), just use a
2507 scalar reference or reference to an arrayref as the value:
2510 date_entered => { '>' => \["to_date(?, 'MM/DD/YYYY')", "11/26/2008"] },
2511 date_expires => { '<' => \"now()" }
2514 Which would generate:
2516 $stmt = "WHERE date_entered > to_date(?, 'MM/DD/YYYY') AND date_expires < now()";
2517 @bind = ('11/26/2008');
2520 =head2 Logic and nesting operators
2522 In the example above,
2523 there is a subtle trap if you want to say something like
2524 this (notice the C<AND>):
2526 WHERE priority != ? AND priority != ?
2528 Because, in Perl you I<can't> do this:
2530 priority => { '!=' => 2, '!=' => 1 }
2532 As the second C<!=> key will obliterate the first. The solution
2533 is to use the special C<-modifier> form inside an arrayref:
2535 priority => [ -and => {'!=', 2},
2539 Normally, these would be joined by C<OR>, but the modifier tells it
2540 to use C<AND> instead. (Hint: You can use this in conjunction with the
2541 C<logic> option to C<new()> in order to change the way your queries
2542 work by default.) B<Important:> Note that the C<-modifier> goes
2543 B<INSIDE> the arrayref, as an extra first element. This will
2544 B<NOT> do what you think it might:
2546 priority => -and => [{'!=', 2}, {'!=', 1}] # WRONG!
2548 Here is a quick list of equivalencies, since there is some overlap:
2551 status => {'!=', 'completed', 'not like', 'pending%' }
2552 status => [ -and => {'!=', 'completed'}, {'not like', 'pending%'}]
2555 status => {'=', ['assigned', 'in-progress']}
2556 status => [ -or => {'=', 'assigned'}, {'=', 'in-progress'}]
2557 status => [ {'=', 'assigned'}, {'=', 'in-progress'} ]
2561 =head2 Special operators: IN, BETWEEN, etc.
2563 You can also use the hashref format to compare a list of fields using the
2564 C<IN> comparison operator, by specifying the list as an arrayref:
2567 status => 'completed',
2568 reportid => { -in => [567, 2335, 2] }
2571 Which would generate:
2573 $stmt = "WHERE status = ? AND reportid IN (?,?,?)";
2574 @bind = ('completed', '567', '2335', '2');
2576 The reverse operator C<-not_in> generates SQL C<NOT IN> and is used in
2579 If the argument to C<-in> is an empty array, 'sqlfalse' is generated
2580 (by default: C<1=0>). Similarly, C<< -not_in => [] >> generates
2581 'sqltrue' (by default: C<1=1>).
2583 In addition to the array you can supply a chunk of literal sql or
2584 literal sql with bind:
2587 customer => { -in => \[
2588 'SELECT cust_id FROM cust WHERE balance > ?',
2591 status => { -in => \'SELECT status_codes FROM states' },
2597 customer IN ( SELECT cust_id FROM cust WHERE balance > ? )
2598 AND status IN ( SELECT status_codes FROM states )
2602 Finally, if the argument to C<-in> is not a reference, it will be
2603 treated as a single-element array.
2605 Another pair of operators is C<-between> and C<-not_between>,
2606 used with an arrayref of two values:
2610 completion_date => {
2611 -not_between => ['2002-10-01', '2003-02-06']
2617 WHERE user = ? AND completion_date NOT BETWEEN ( ? AND ? )
2619 Just like with C<-in> all plausible combinations of literal SQL
2623 start0 => { -between => [ 1, 2 ] },
2624 start1 => { -between => \["? AND ?", 1, 2] },
2625 start2 => { -between => \"lower(x) AND upper(y)" },
2626 start3 => { -between => [
2628 \["upper(?)", 'stuff' ],
2635 ( start0 BETWEEN ? AND ? )
2636 AND ( start1 BETWEEN ? AND ? )
2637 AND ( start2 BETWEEN lower(x) AND upper(y) )
2638 AND ( start3 BETWEEN lower(x) AND upper(?) )
2640 @bind = (1, 2, 1, 2, 'stuff');
2643 These are the two builtin "special operators"; but the
2644 list can be expanded: see section L</"SPECIAL OPERATORS"> below.
2646 =head2 Unary operators: bool
2648 If you wish to test against boolean columns or functions within your
2649 database you can use the C<-bool> and C<-not_bool> operators. For
2650 example to test the column C<is_user> being true and the column
2651 C<is_enabled> being false you would use:-
2655 -not_bool => 'is_enabled',
2660 WHERE is_user AND NOT is_enabled
2662 If a more complex combination is required, testing more conditions,
2663 then you should use the and/or operators:-
2668 -not_bool => { two=> { -rlike => 'bar' } },
2669 -not_bool => { three => [ { '=', 2 }, { '>', 5 } ] },
2680 (NOT ( three = ? OR three > ? ))
2683 =head2 Nested conditions, -and/-or prefixes
2685 So far, we've seen how multiple conditions are joined with a top-level
2686 C<AND>. We can change this by putting the different conditions we want in
2687 hashes and then putting those hashes in an array. For example:
2692 status => { -like => ['pending%', 'dispatched'] },
2696 status => 'unassigned',
2700 This data structure would create the following:
2702 $stmt = "WHERE ( user = ? AND ( status LIKE ? OR status LIKE ? ) )
2703 OR ( user = ? AND status = ? ) )";
2704 @bind = ('nwiger', 'pending', 'dispatched', 'robot', 'unassigned');
2707 Clauses in hashrefs or arrayrefs can be prefixed with an C<-and> or C<-or>
2708 to change the logic inside:
2714 -and => [ workhrs => {'>', 20}, geo => 'ASIA' ],
2715 -or => { workhrs => {'<', 50}, geo => 'EURO' },
2722 $stmt = "WHERE ( user = ?
2723 AND ( ( workhrs > ? AND geo = ? )
2724 OR ( workhrs < ? OR geo = ? ) ) )";
2725 @bind = ('nwiger', '20', 'ASIA', '50', 'EURO');
2727 =head3 Algebraic inconsistency, for historical reasons
2729 C<Important note>: when connecting several conditions, the C<-and->|C<-or>
2730 operator goes C<outside> of the nested structure; whereas when connecting
2731 several constraints on one column, the C<-and> operator goes
2732 C<inside> the arrayref. Here is an example combining both features:
2735 -and => [a => 1, b => 2],
2736 -or => [c => 3, d => 4],
2737 e => [-and => {-like => 'foo%'}, {-like => '%bar'} ]
2742 WHERE ( ( ( a = ? AND b = ? )
2743 OR ( c = ? OR d = ? )
2744 OR ( e LIKE ? AND e LIKE ? ) ) )
2746 This difference in syntax is unfortunate but must be preserved for
2747 historical reasons. So be careful: the two examples below would
2748 seem algebraically equivalent, but they are not
2751 { -like => 'foo%' },
2752 { -like => '%bar' },
2754 # yields: WHERE ( ( col LIKE ? AND col LIKE ? ) )
2757 { col => { -like => 'foo%' } },
2758 { col => { -like => '%bar' } },
2760 # yields: WHERE ( ( col LIKE ? OR col LIKE ? ) )
2763 =head2 Literal SQL and value type operators
2765 The basic premise of SQL::Abstract is that in WHERE specifications the "left
2766 side" is a column name and the "right side" is a value (normally rendered as
2767 a placeholder). This holds true for both hashrefs and arrayref pairs as you
2768 see in the L</WHERE CLAUSES> examples above. Sometimes it is necessary to
2769 alter this behavior. There are several ways of doing so.
2773 This is a virtual operator that signals the string to its right side is an
2774 identifier (a column name) and not a value. For example to compare two
2775 columns you would write:
2778 priority => { '<', 2 },
2779 requestor => { -ident => 'submitter' },
2784 $stmt = "WHERE priority < ? AND requestor = submitter";
2787 If you are maintaining legacy code you may see a different construct as
2788 described in L</Deprecated usage of Literal SQL>, please use C<-ident> in new
2793 This is a virtual operator that signals that the construct to its right side
2794 is a value to be passed to DBI. This is for example necessary when you want
2795 to write a where clause against an array (for RDBMS that support such
2796 datatypes). For example:
2799 array => { -value => [1, 2, 3] }
2804 $stmt = 'WHERE array = ?';
2805 @bind = ([1, 2, 3]);
2807 Note that if you were to simply say:
2813 the result would probably not be what you wanted:
2815 $stmt = 'WHERE array = ? OR array = ? OR array = ?';
2820 Finally, sometimes only literal SQL will do. To include a random snippet
2821 of SQL verbatim, you specify it as a scalar reference. Consider this only
2822 as a last resort. Usually there is a better way. For example:
2825 priority => { '<', 2 },
2826 requestor => { -in => \'(SELECT name FROM hitmen)' },
2831 $stmt = "WHERE priority < ? AND requestor IN (SELECT name FROM hitmen)"
2834 Note that in this example, you only get one bind parameter back, since
2835 the verbatim SQL is passed as part of the statement.
2839 Never use untrusted input as a literal SQL argument - this is a massive
2840 security risk (there is no way to check literal snippets for SQL
2841 injections and other nastyness). If you need to deal with untrusted input
2842 use literal SQL with placeholders as described next.
2844 =head3 Literal SQL with placeholders and bind values (subqueries)
2846 If the literal SQL to be inserted has placeholders and bind values,
2847 use a reference to an arrayref (yes this is a double reference --
2848 not so common, but perfectly legal Perl). For example, to find a date
2849 in Postgres you can use something like this:
2852 date_column => \[ "= date '2008-09-30' - ?::integer", 10 ]
2857 $stmt = "WHERE ( date_column = date '2008-09-30' - ?::integer )"
2860 Note that you must pass the bind values in the same format as they are returned
2861 by L<where|/where(\%where, $order)>. This means that if you set L</bindtype>
2862 to C<columns>, you must provide the bind values in the
2863 C<< [ column_meta => value ] >> format, where C<column_meta> is an opaque
2864 scalar value; most commonly the column name, but you can use any scalar value
2865 (including references and blessed references), L<SQL::Abstract> will simply
2866 pass it through intact. So if C<bindtype> is set to C<columns> the above
2867 example will look like:
2870 date_column => \[ "= date '2008-09-30' - ?::integer", [ {} => 10 ] ]
2873 Literal SQL is especially useful for nesting parenthesized clauses in the
2874 main SQL query. Here is a first example:
2876 my ($sub_stmt, @sub_bind) = ("SELECT c1 FROM t1 WHERE c2 < ? AND c3 LIKE ?",
2880 bar => \["IN ($sub_stmt)" => @sub_bind],
2885 $stmt = "WHERE (foo = ? AND bar IN (SELECT c1 FROM t1
2886 WHERE c2 < ? AND c3 LIKE ?))";
2887 @bind = (1234, 100, "foo%");
2889 Other subquery operators, like for example C<"E<gt> ALL"> or C<"NOT IN">,
2890 are expressed in the same way. Of course the C<$sub_stmt> and
2891 its associated bind values can be generated through a former call
2894 my ($sub_stmt, @sub_bind)
2895 = $sql->select("t1", "c1", {c2 => {"<" => 100},
2896 c3 => {-like => "foo%"}});
2899 bar => \["> ALL ($sub_stmt)" => @sub_bind],
2902 In the examples above, the subquery was used as an operator on a column;
2903 but the same principle also applies for a clause within the main C<%where>
2904 hash, like an EXISTS subquery:
2906 my ($sub_stmt, @sub_bind)
2907 = $sql->select("t1", "*", {c1 => 1, c2 => \"> t0.c0"});
2908 my %where = ( -and => [
2910 \["EXISTS ($sub_stmt)" => @sub_bind],
2915 $stmt = "WHERE (foo = ? AND EXISTS (SELECT * FROM t1
2916 WHERE c1 = ? AND c2 > t0.c0))";
2920 Observe that the condition on C<c2> in the subquery refers to
2921 column C<t0.c0> of the main query: this is I<not> a bind
2922 value, so we have to express it through a scalar ref.
2923 Writing C<< c2 => {">" => "t0.c0"} >> would have generated
2924 C<< c2 > ? >> with bind value C<"t0.c0"> ... not exactly
2925 what we wanted here.
2927 Finally, here is an example where a subquery is used
2928 for expressing unary negation:
2930 my ($sub_stmt, @sub_bind)
2931 = $sql->where({age => [{"<" => 10}, {">" => 20}]});
2932 $sub_stmt =~ s/^ where //i; # don't want "WHERE" in the subclause
2934 lname => {like => '%son%'},
2935 \["NOT ($sub_stmt)" => @sub_bind],
2940 $stmt = "lname LIKE ? AND NOT ( age < ? OR age > ? )"
2941 @bind = ('%son%', 10, 20)
2943 =head3 Deprecated usage of Literal SQL
2945 Below are some examples of archaic use of literal SQL. It is shown only as
2946 reference for those who deal with legacy code. Each example has a much
2947 better, cleaner and safer alternative that users should opt for in new code.
2953 my %where = ( requestor => \'IS NOT NULL' )
2955 $stmt = "WHERE requestor IS NOT NULL"
2957 This used to be the way of generating NULL comparisons, before the handling
2958 of C<undef> got formalized. For new code please use the superior syntax as
2959 described in L</Tests for NULL values>.
2963 my %where = ( requestor => \'= submitter' )
2965 $stmt = "WHERE requestor = submitter"
2967 This used to be the only way to compare columns. Use the superior L</-ident>
2968 method for all new code. For example an identifier declared in such a way
2969 will be properly quoted if L</quote_char> is properly set, while the legacy
2970 form will remain as supplied.
2974 my %where = ( is_ready => \"", completed => { '>', '2012-12-21' } )
2976 $stmt = "WHERE completed > ? AND is_ready"
2977 @bind = ('2012-12-21')
2979 Using an empty string literal used to be the only way to express a boolean.
2980 For all new code please use the much more readable
2981 L<-bool|/Unary operators: bool> operator.
2987 These pages could go on for a while, since the nesting of the data
2988 structures this module can handle are pretty much unlimited (the
2989 module implements the C<WHERE> expansion as a recursive function
2990 internally). Your best bet is to "play around" with the module a
2991 little to see how the data structures behave, and choose the best
2992 format for your data based on that.
2994 And of course, all the values above will probably be replaced with
2995 variables gotten from forms or the command line. After all, if you
2996 knew everything ahead of time, you wouldn't have to worry about
2997 dynamically-generating SQL and could just hardwire it into your
3000 =head1 ORDER BY CLAUSES
3002 Some functions take an order by clause. This can either be a scalar (just a
3003 column name), a hashref of C<< { -desc => 'col' } >> or C<< { -asc => 'col' }
3004 >>, a scalarref, an arrayref-ref, or an arrayref of any of the previous
3007 Given | Will Generate
3008 ---------------------------------------------------------------
3010 'colA' | ORDER BY colA
3012 [qw/colA colB/] | ORDER BY colA, colB
3014 {-asc => 'colA'} | ORDER BY colA ASC
3016 {-desc => 'colB'} | ORDER BY colB DESC
3018 ['colA', {-asc => 'colB'}] | ORDER BY colA, colB ASC
3020 { -asc => [qw/colA colB/] } | ORDER BY colA ASC, colB ASC
3022 \'colA DESC' | ORDER BY colA DESC
3024 \[ 'FUNC(colA, ?)', $x ] | ORDER BY FUNC(colA, ?)
3025 | /* ...with $x bound to ? */
3028 { -asc => 'colA' }, | colA ASC,
3029 { -desc => [qw/colB/] }, | colB DESC,
3030 { -asc => [qw/colC colD/] },| colC ASC, colD ASC,
3031 \'colE DESC', | colE DESC,
3032 \[ 'FUNC(colF, ?)', $x ], | FUNC(colF, ?)
3033 ] | /* ...with $x bound to ? */
3034 ===============================================================
3038 =head1 SPECIAL OPERATORS
3040 my $sqlmaker = SQL::Abstract->new(special_ops => [
3044 my ($self, $field, $op, $arg) = @_;
3050 handler => 'method_name',
3054 A "special operator" is a SQL syntactic clause that can be
3055 applied to a field, instead of a usual binary operator.
3058 WHERE field IN (?, ?, ?)
3059 WHERE field BETWEEN ? AND ?
3060 WHERE MATCH(field) AGAINST (?, ?)
3062 Special operators IN and BETWEEN are fairly standard and therefore
3063 are builtin within C<SQL::Abstract> (as the overridable methods
3064 C<_where_field_IN> and C<_where_field_BETWEEN>). For other operators,
3065 like the MATCH .. AGAINST example above which is specific to MySQL,
3066 you can write your own operator handlers - supply a C<special_ops>
3067 argument to the C<new> method. That argument takes an arrayref of
3068 operator definitions; each operator definition is a hashref with two
3075 the regular expression to match the operator
3079 Either a coderef or a plain scalar method name. In both cases
3080 the expected return is C<< ($sql, @bind) >>.
3082 When supplied with a method name, it is simply called on the
3083 L<SQL::Abstract> object as:
3085 $self->$method_name($field, $op, $arg)
3089 $field is the LHS of the operator
3090 $op is the part that matched the handler regex
3093 When supplied with a coderef, it is called as:
3095 $coderef->($self, $field, $op, $arg)
3100 For example, here is an implementation
3101 of the MATCH .. AGAINST syntax for MySQL
3103 my $sqlmaker = SQL::Abstract->new(special_ops => [
3105 # special op for MySql MATCH (field) AGAINST(word1, word2, ...)
3106 {regex => qr/^match$/i,
3108 my ($self, $field, $op, $arg) = @_;
3109 $arg = [$arg] if not ref $arg;
3110 my $label = $self->_quote($field);
3111 my ($placeholder) = $self->_convert('?');
3112 my $placeholders = join ", ", (($placeholder) x @$arg);
3113 my $sql = $self->_sqlcase('match') . " ($label) "
3114 . $self->_sqlcase('against') . " ($placeholders) ";
3115 my @bind = $self->_bindtype($field, @$arg);
3116 return ($sql, @bind);
3123 =head1 UNARY OPERATORS
3125 my $sqlmaker = SQL::Abstract->new(unary_ops => [
3129 my ($self, $op, $arg) = @_;
3135 handler => 'method_name',
3139 A "unary operator" is a SQL syntactic clause that can be
3140 applied to a field - the operator goes before the field
3142 You can write your own operator handlers - supply a C<unary_ops>
3143 argument to the C<new> method. That argument takes an arrayref of
3144 operator definitions; each operator definition is a hashref with two
3151 the regular expression to match the operator
3155 Either a coderef or a plain scalar method name. In both cases
3156 the expected return is C<< $sql >>.
3158 When supplied with a method name, it is simply called on the
3159 L<SQL::Abstract> object as:
3161 $self->$method_name($op, $arg)
3165 $op is the part that matched the handler regex
3166 $arg is the RHS or argument of the operator
3168 When supplied with a coderef, it is called as:
3170 $coderef->($self, $op, $arg)
3178 Thanks to some benchmarking by Mark Stosberg, it turns out that
3179 this module is many orders of magnitude faster than using C<DBIx::Abstract>.
3180 I must admit this wasn't an intentional design issue, but it's a
3181 byproduct of the fact that you get to control your C<DBI> handles
3184 To maximize performance, use a code snippet like the following:
3186 # prepare a statement handle using the first row
3187 # and then reuse it for the rest of the rows
3189 for my $href (@array_of_hashrefs) {
3190 $stmt ||= $sql->insert('table', $href);
3191 $sth ||= $dbh->prepare($stmt);
3192 $sth->execute($sql->values($href));
3195 The reason this works is because the keys in your C<$href> are sorted
3196 internally by B<SQL::Abstract>. Thus, as long as your data retains
3197 the same structure, you only have to generate the SQL the first time
3198 around. On subsequent queries, simply use the C<values> function provided
3199 by this module to return your values in the correct order.
3201 However this depends on the values having the same type - if, for
3202 example, the values of a where clause may either have values
3203 (resulting in sql of the form C<column = ?> with a single bind
3204 value), or alternatively the values might be C<undef> (resulting in
3205 sql of the form C<column IS NULL> with no bind value) then the
3206 caching technique suggested will not work.
3210 If you use my C<CGI::FormBuilder> module at all, you'll hopefully
3211 really like this part (I do, at least). Building up a complex query
3212 can be as simple as the following:
3219 use CGI::FormBuilder;
3222 my $form = CGI::FormBuilder->new(...);
3223 my $sql = SQL::Abstract->new;
3225 if ($form->submitted) {
3226 my $field = $form->field;
3227 my $id = delete $field->{id};
3228 my($stmt, @bind) = $sql->update('table', $field, {id => $id});
3231 Of course, you would still have to connect using C<DBI> to run the
3232 query, but the point is that if you make your form look like your
3233 table, the actual query script can be extremely simplistic.
3235 If you're B<REALLY> lazy (I am), check out C<HTML::QuickTable> for
3236 a fast interface to returning and formatting data. I frequently
3237 use these three modules together to write complex database query
3238 apps in under 50 lines.
3240 =head1 HOW TO CONTRIBUTE
3242 Contributions are always welcome, in all usable forms (we especially
3243 welcome documentation improvements). The delivery methods include git-
3244 or unified-diff formatted patches, GitHub pull requests, or plain bug
3245 reports either via RT or the Mailing list. Contributors are generally
3246 granted full access to the official repository after their first several
3247 patches pass successful review.
3249 This project is maintained in a git repository. The code and related tools are
3250 accessible at the following locations:
3254 =item * Official repo: L<git://git.shadowcat.co.uk/dbsrgits/SQL-Abstract.git>
3256 =item * Official gitweb: L<http://git.shadowcat.co.uk/gitweb/gitweb.cgi?p=dbsrgits/SQL-Abstract.git>
3258 =item * GitHub mirror: L<https://github.com/dbsrgits/sql-abstract>
3260 =item * Authorized committers: L<ssh://dbsrgits@git.shadowcat.co.uk/SQL-Abstract.git>
3266 Version 1.50 was a major internal refactoring of C<SQL::Abstract>.
3267 Great care has been taken to preserve the I<published> behavior
3268 documented in previous versions in the 1.* family; however,
3269 some features that were previously undocumented, or behaved
3270 differently from the documentation, had to be changed in order
3271 to clarify the semantics. Hence, client code that was relying
3272 on some dark areas of C<SQL::Abstract> v1.*
3273 B<might behave differently> in v1.50.
3275 The main changes are:
3281 support for literal SQL through the C<< \ [ $sql, @bind ] >> syntax.
3285 support for the { operator => \"..." } construct (to embed literal SQL)
3289 support for the { operator => \["...", @bind] } construct (to embed literal SQL with bind values)
3293 optional support for L<array datatypes|/"Inserting and Updating Arrays">
3297 defensive programming: check arguments
3301 fixed bug with global logic, which was previously implemented
3302 through global variables yielding side-effects. Prior versions would
3303 interpret C<< [ {cond1, cond2}, [cond3, cond4] ] >>
3304 as C<< "(cond1 AND cond2) OR (cond3 AND cond4)" >>.
3305 Now this is interpreted
3306 as C<< "(cond1 AND cond2) OR (cond3 OR cond4)" >>.
3311 fixed semantics of _bindtype on array args
3315 dropped the C<_anoncopy> of the %where tree. No longer necessary,
3316 we just avoid shifting arrays within that tree.
3320 dropped the C<_modlogic> function
3324 =head1 ACKNOWLEDGEMENTS
3326 There are a number of individuals that have really helped out with
3327 this module. Unfortunately, most of them submitted bugs via CPAN
3328 so I have no idea who they are! But the people I do know are:
3330 Ash Berlin (order_by hash term support)
3331 Matt Trout (DBIx::Class support)
3332 Mark Stosberg (benchmarking)
3333 Chas Owens (initial "IN" operator support)
3334 Philip Collins (per-field SQL functions)
3335 Eric Kolve (hashref "AND" support)
3336 Mike Fragassi (enhancements to "BETWEEN" and "LIKE")
3337 Dan Kubb (support for "quote_char" and "name_sep")
3338 Guillermo Roditi (patch to cleanup "IN" and "BETWEEN", fix and tests for _order_by)
3339 Laurent Dami (internal refactoring, extensible list of special operators, literal SQL)
3340 Norbert Buchmuller (support for literal SQL in hashpair, misc. fixes & tests)
3341 Peter Rabbitson (rewrite of SQLA::Test, misc. fixes & tests)
3342 Oliver Charles (support for "RETURNING" after "INSERT")
3348 L<DBIx::Class>, L<DBIx::Abstract>, L<CGI::FormBuilder>, L<HTML::QuickTable>.
3352 Copyright (c) 2001-2007 Nathan Wiger <nwiger@cpan.org>. All Rights Reserved.
3354 This module is actively maintained by Matt Trout <mst@shadowcatsystems.co.uk>
3356 For support, your best bet is to try the C<DBIx::Class> users mailing list.
3357 While not an official support venue, C<DBIx::Class> makes heavy use of
3358 C<SQL::Abstract>, and as such list members there are very familiar with
3359 how to create queries.
3363 This module is free software; you may copy this under the same
3364 terms as perl itself (either the GNU General Public License or
3365 the Artistic License)