1 package SQL::Abstract; # see doc at end of file
10 our @EXPORT_OK = qw(is_plain_value is_literal_value);
20 *SQL::Abstract::_ENV_::DETECT_AUTOGENERATED_STRINGIFICATION = $ENV{SQLA_ISVALUE_IGNORE_AUTOGENERATED_STRINGIFICATION}
26 #======================================================================
28 #======================================================================
30 our $VERSION = '1.87';
32 # This would confuse some packagers
33 $VERSION = eval $VERSION if $VERSION =~ /_/; # numify for warning-free dev releases
37 # special operators (-in, -between). May be extended/overridden by user.
38 # See section WHERE: BUILTIN SPECIAL OPERATORS below for implementation
39 my @BUILTIN_SPECIAL_OPS = (
40 {regex => qr/^ (?: not \s )? between $/ix, handler => '_where_field_BETWEEN'},
41 {regex => qr/^ (?: not \s )? in $/ix, handler => '_where_field_IN'},
42 {regex => qr/^ ident $/ix, handler => '_where_op_IDENT'},
43 {regex => qr/^ value $/ix, handler => '_where_op_VALUE'},
44 {regex => qr/^ is (?: \s+ not )? $/ix, handler => '_where_field_IS'},
47 # unaryish operators - key maps to handler
48 my @BUILTIN_UNARY_OPS = (
49 # the digits are backcompat stuff
50 { regex => qr/^ and (?: [_\s]? \d+ )? $/xi, handler => '_where_op_ANDOR' },
51 { regex => qr/^ or (?: [_\s]? \d+ )? $/xi, handler => '_where_op_ANDOR' },
52 { regex => qr/^ nest (?: [_\s]? \d+ )? $/xi, handler => '_where_op_NEST' },
53 { regex => qr/^ (?: not \s )? bool $/xi, handler => '_where_op_BOOL' },
54 { regex => qr/^ ident $/xi, handler => '_where_op_IDENT' },
55 { regex => qr/^ value $/xi, handler => '_where_op_VALUE' },
58 #======================================================================
59 # DEBUGGING AND ERROR REPORTING
60 #======================================================================
63 return unless $_[0]->{debug}; shift; # a little faster
64 my $func = (caller(1))[3];
65 warn "[$func] ", @_, "\n";
69 my($func) = (caller(1))[3];
70 Carp::carp "[$func] Warning: ", @_;
74 my($func) = (caller(1))[3];
75 Carp::croak "[$func] Fatal: ", @_;
78 sub is_literal_value ($) {
79 ref $_[0] eq 'SCALAR' ? [ ${$_[0]} ]
80 : ( ref $_[0] eq 'REF' and ref ${$_[0]} eq 'ARRAY' ) ? [ @${ $_[0] } ]
84 # FIXME XSify - this can be done so much more efficiently
85 sub is_plain_value ($) {
87 ! length ref $_[0] ? \($_[0])
89 ref $_[0] eq 'HASH' and keys %{$_[0]} == 1
91 exists $_[0]->{-value}
92 ) ? \($_[0]->{-value})
94 # reuse @_ for even moar speedz
95 defined ( $_[1] = Scalar::Util::blessed $_[0] )
97 # deliberately not using Devel::OverloadInfo - the checks we are
98 # intersted in are much more limited than the fullblown thing, and
99 # this is a very hot piece of code
101 # simply using ->can('(""') can leave behind stub methods that
102 # break actually using the overload later (see L<perldiag/Stub
103 # found while resolving method "%s" overloading "%s" in package
104 # "%s"> and the source of overload::mycan())
106 # either has stringification which DBI SHOULD prefer out of the box
107 grep { *{ (qq[${_}::(""]) }{CODE} } @{ $_[2] = mro::get_linear_isa( $_[1] ) }
109 # has nummification or boolification, AND fallback is *not* disabled
111 SQL::Abstract::_ENV_::DETECT_AUTOGENERATED_STRINGIFICATION
114 grep { *{"${_}::(0+"}{CODE} } @{$_[2]}
116 grep { *{"${_}::(bool"}{CODE} } @{$_[2]}
120 # no fallback specified at all
121 ! ( ($_[3]) = grep { *{"${_}::()"}{CODE} } @{$_[2]} )
123 # fallback explicitly undef
124 ! defined ${"$_[3]::()"}
137 #======================================================================
139 #======================================================================
143 my $class = ref($self) || $self;
144 my %opt = (ref $_[0] eq 'HASH') ? %{$_[0]} : @_;
146 # choose our case by keeping an option around
147 delete $opt{case} if $opt{case} && $opt{case} ne 'lower';
149 # default logic for interpreting arrayrefs
150 $opt{logic} = $opt{logic} ? uc $opt{logic} : 'OR';
152 # how to return bind vars
153 $opt{bindtype} ||= 'normal';
155 # default comparison is "=", but can be overridden
158 # try to recognize which are the 'equality' and 'inequality' ops
159 # (temporary quickfix (in 2007), should go through a more seasoned API)
160 $opt{equality_op} = qr/^( \Q$opt{cmp}\E | \= )$/ix;
161 $opt{inequality_op} = qr/^( != | <> )$/ix;
163 $opt{like_op} = qr/^ (is\s+)? r?like $/xi;
164 $opt{not_like_op} = qr/^ (is\s+)? not \s+ r?like $/xi;
167 $opt{sqltrue} ||= '1=1';
168 $opt{sqlfalse} ||= '0=1';
171 $opt{special_ops} ||= [];
172 # regexes are applied in order, thus push after user-defines
173 push @{$opt{special_ops}}, @BUILTIN_SPECIAL_OPS;
176 $opt{unary_ops} ||= [];
177 push @{$opt{unary_ops}}, @BUILTIN_UNARY_OPS;
179 # rudimentary sanity-check for user supplied bits treated as functions/operators
180 # If a purported function matches this regular expression, an exception is thrown.
181 # Literal SQL is *NOT* subject to this check, only functions (and column names
182 # when quoting is not in effect)
185 # need to guard against ()'s in column names too, but this will break tons of
186 # hacks... ideas anyone?
187 $opt{injection_guard} ||= qr/
193 return bless \%opt, $class;
197 sub _assert_pass_injection_guard {
198 if ($_[1] =~ $_[0]->{injection_guard}) {
199 my $class = ref $_[0];
200 puke "Possible SQL injection attempt '$_[1]'. If this is indeed a part of the "
201 . "desired SQL use literal SQL ( \'...' or \[ '...' ] ) or supply your own "
202 . "{injection_guard} attribute to ${class}->new()"
207 #======================================================================
209 #======================================================================
213 my $table = $self->_table(shift);
214 my $data = shift || return;
217 my $method = $self->_METHOD_FOR_refkind("_insert", $data);
218 my ($sql, @bind) = $self->$method($data);
219 $sql = join " ", $self->_sqlcase('insert into'), $table, $sql;
221 if ($options->{returning}) {
222 my ($s, @b) = $self->_insert_returning($options);
227 return wantarray ? ($sql, @bind) : $sql;
230 # So that subclasses can override INSERT ... RETURNING separately from
231 # UPDATE and DELETE (e.g. DBIx::Class::SQLMaker::Oracle does this)
232 sub _insert_returning { shift->_returning(@_) }
235 my ($self, $options) = @_;
237 my $f = $options->{returning};
239 my $fieldlist = $self->_SWITCH_refkind($f, {
240 ARRAYREF => sub {join ', ', map { $self->_quote($_) } @$f;},
241 SCALAR => sub {$self->_quote($f)},
242 SCALARREF => sub {$$f},
244 return $self->_sqlcase(' returning ') . $fieldlist;
247 sub _insert_HASHREF { # explicit list of fields and then values
248 my ($self, $data) = @_;
250 my @fields = sort keys %$data;
252 my ($sql, @bind) = $self->_insert_values($data);
255 $_ = $self->_quote($_) foreach @fields;
256 $sql = "( ".join(", ", @fields).") ".$sql;
258 return ($sql, @bind);
261 sub _insert_ARRAYREF { # just generate values(?,?) part (no list of fields)
262 my ($self, $data) = @_;
264 # no names (arrayref) so can't generate bindtype
265 $self->{bindtype} ne 'columns'
266 or belch "can't do 'columns' bindtype when called with arrayref";
268 my (@values, @all_bind);
269 foreach my $value (@$data) {
270 my ($values, @bind) = $self->_insert_value(undef, $value);
271 push @values, $values;
272 push @all_bind, @bind;
274 my $sql = $self->_sqlcase('values')." ( ".join(", ", @values)." )";
275 return ($sql, @all_bind);
278 sub _insert_ARRAYREFREF { # literal SQL with bind
279 my ($self, $data) = @_;
281 my ($sql, @bind) = @${$data};
282 $self->_assert_bindval_matches_bindtype(@bind);
284 return ($sql, @bind);
288 sub _insert_SCALARREF { # literal SQL without bind
289 my ($self, $data) = @_;
295 my ($self, $data) = @_;
297 my (@values, @all_bind);
298 foreach my $column (sort keys %$data) {
299 my ($values, @bind) = $self->_insert_value($column, $data->{$column});
300 push @values, $values;
301 push @all_bind, @bind;
303 my $sql = $self->_sqlcase('values')." ( ".join(", ", @values)." )";
304 return ($sql, @all_bind);
308 my ($self, $column, $v) = @_;
310 my (@values, @all_bind);
311 $self->_SWITCH_refkind($v, {
314 if ($self->{array_datatypes}) { # if array datatype are activated
316 push @all_bind, $self->_bindtype($column, $v);
318 else { # else literal SQL with bind
319 my ($sql, @bind) = @$v;
320 $self->_assert_bindval_matches_bindtype(@bind);
322 push @all_bind, @bind;
326 ARRAYREFREF => sub { # literal SQL with bind
327 my ($sql, @bind) = @${$v};
328 $self->_assert_bindval_matches_bindtype(@bind);
330 push @all_bind, @bind;
333 # THINK: anything useful to do with a HASHREF ?
334 HASHREF => sub { # (nothing, but old SQLA passed it through)
335 #TODO in SQLA >= 2.0 it will die instead
336 belch "HASH ref as bind value in insert is not supported";
338 push @all_bind, $self->_bindtype($column, $v);
341 SCALARREF => sub { # literal SQL without bind
345 SCALAR_or_UNDEF => sub {
347 push @all_bind, $self->_bindtype($column, $v);
352 my $sql = join(", ", @values);
353 return ($sql, @all_bind);
358 #======================================================================
360 #======================================================================
365 my $table = $self->_table(shift);
366 my $data = shift || return;
370 # first build the 'SET' part of the sql statement
371 puke "Unsupported data type specified to \$sql->update"
372 unless ref $data eq 'HASH';
374 my ($sql, @all_bind) = $self->_update_set_values($data);
375 $sql = $self->_sqlcase('update ') . $table . $self->_sqlcase(' set ')
379 my($where_sql, @where_bind) = $self->where($where);
381 push @all_bind, @where_bind;
384 if ($options->{returning}) {
385 my ($returning_sql, @returning_bind) = $self->_update_returning($options);
386 $sql .= $returning_sql;
387 push @all_bind, @returning_bind;
390 return wantarray ? ($sql, @all_bind) : $sql;
393 sub _update_set_values {
394 my ($self, $data) = @_;
396 my (@set, @all_bind);
397 for my $k (sort keys %$data) {
400 my $label = $self->_quote($k);
402 $self->_SWITCH_refkind($v, {
404 if ($self->{array_datatypes}) { # array datatype
405 push @set, "$label = ?";
406 push @all_bind, $self->_bindtype($k, $v);
408 else { # literal SQL with bind
409 my ($sql, @bind) = @$v;
410 $self->_assert_bindval_matches_bindtype(@bind);
411 push @set, "$label = $sql";
412 push @all_bind, @bind;
415 ARRAYREFREF => sub { # literal SQL with bind
416 my ($sql, @bind) = @${$v};
417 $self->_assert_bindval_matches_bindtype(@bind);
418 push @set, "$label = $sql";
419 push @all_bind, @bind;
421 SCALARREF => sub { # literal SQL without bind
422 push @set, "$label = $$v";
425 my ($op, $arg, @rest) = %$v;
427 puke 'Operator calls in update must be in the form { -op => $arg }'
428 if (@rest or not $op =~ /^\-(.+)/);
430 local $self->{_nested_func_lhs} = $k;
431 my ($sql, @bind) = $self->_where_unary_op($1, $arg);
433 push @set, "$label = $sql";
434 push @all_bind, @bind;
436 SCALAR_or_UNDEF => sub {
437 push @set, "$label = ?";
438 push @all_bind, $self->_bindtype($k, $v);
444 my $sql = join ', ', @set;
446 return ($sql, @all_bind);
449 # So that subclasses can override UPDATE ... RETURNING separately from
451 sub _update_returning { shift->_returning(@_) }
455 #======================================================================
457 #======================================================================
462 my $table = $self->_table(shift);
463 my $fields = shift || '*';
467 my ($fields_sql, @bind) = $self->_select_fields($fields);
469 my ($where_sql, @where_bind) = $self->where($where, $order);
470 push @bind, @where_bind;
472 my $sql = join(' ', $self->_sqlcase('select'), $fields_sql,
473 $self->_sqlcase('from'), $table)
476 return wantarray ? ($sql, @bind) : $sql;
480 my ($self, $fields) = @_;
481 return ref $fields eq 'ARRAY' ? join ', ', map { $self->_quote($_) } @$fields
485 #======================================================================
487 #======================================================================
492 my $table = $self->_table(shift);
496 my($where_sql, @bind) = $self->where($where);
497 my $sql = $self->_sqlcase('delete from ') . $table . $where_sql;
499 if ($options->{returning}) {
500 my ($returning_sql, @returning_bind) = $self->_delete_returning($options);
501 $sql .= $returning_sql;
502 push @bind, @returning_bind;
505 return wantarray ? ($sql, @bind) : $sql;
508 # So that subclasses can override DELETE ... RETURNING separately from
510 sub _delete_returning { shift->_returning(@_) }
514 #======================================================================
516 #======================================================================
520 # Finally, a separate routine just to handle WHERE clauses
522 my ($self, $where, $order) = @_;
525 my ($sql, @bind) = $self->_recurse_where($where);
526 $sql = (defined $sql and length $sql) ? $self->_sqlcase(' where ') . "( $sql )" : '';
530 my ($order_sql, @order_bind) = $self->_order_by($order);
532 push @bind, @order_bind;
535 return wantarray ? ($sql, @bind) : $sql;
539 my ($self, $expr, $logic) = @_;
540 if (ref($expr) eq 'HASH') {
541 if (keys %$expr > 1) {
543 return +{ "-${logic}" => [
544 map $self->_expand_expr_hashpair($_ => $expr->{$_}, $logic),
548 return $self->_expand_expr_hashpair(%$expr, $logic);
553 sub _expand_expr_hashpair {
554 my ($self, $k, $v, $logic) = @_;
557 return +{ $k => { $self->{cmp} => $v } };
564 my ($self, $where, $logic) = @_;
566 my $where_exp = $self->_expand_expr($where, $logic);
568 # dispatch on appropriate method according to refkind of $where
569 my $method = $self->_METHOD_FOR_refkind("_where", $where_exp);
571 my ($sql, @bind) = $self->$method($where_exp, $logic);
573 # DBIx::Class used to call _recurse_where in scalar context
574 # something else might too...
576 return ($sql, @bind);
579 belch "Calling _recurse_where in scalar context is deprecated and will go away before 2.0";
586 #======================================================================
587 # WHERE: top-level ARRAYREF
588 #======================================================================
591 sub _where_ARRAYREF {
592 my ($self, $where, $logic) = @_;
594 $logic = uc($logic || $self->{logic});
595 $logic eq 'AND' or $logic eq 'OR' or puke "unknown logic: $logic";
597 my @clauses = @$where;
599 my (@sql_clauses, @all_bind);
600 # need to use while() so can shift() for pairs
602 my $el = shift @clauses;
604 $el = undef if (defined $el and ! length $el);
606 # switch according to kind of $el and get corresponding ($sql, @bind)
607 my ($sql, @bind) = $self->_SWITCH_refkind($el, {
609 # skip empty elements, otherwise get invalid trailing AND stuff
610 ARRAYREF => sub {$self->_recurse_where($el) if @$el},
614 $self->_assert_bindval_matches_bindtype(@b);
618 HASHREF => sub {$self->_recurse_where($el, 'and') if %$el},
620 SCALARREF => sub { ($$el); },
623 # top-level arrayref with scalars, recurse in pairs
624 $self->_recurse_where({$el => shift(@clauses)})
627 UNDEF => sub {puke "Supplying an empty left hand side argument is not supported in array-pairs" },
631 push @sql_clauses, $sql;
632 push @all_bind, @bind;
636 return $self->_join_sql_clauses($logic, \@sql_clauses, \@all_bind);
639 #======================================================================
640 # WHERE: top-level ARRAYREFREF
641 #======================================================================
643 sub _where_ARRAYREFREF {
644 my ($self, $where) = @_;
645 my ($sql, @bind) = @$$where;
646 $self->_assert_bindval_matches_bindtype(@bind);
647 return ($sql, @bind);
650 #======================================================================
651 # WHERE: top-level HASHREF
652 #======================================================================
655 my ($self, $where) = @_;
656 my (@sql_clauses, @all_bind);
658 for my $k (sort keys %$where) {
659 my $v = $where->{$k};
661 # ($k => $v) is either a special unary op or a regular hashpair
662 my ($sql, @bind) = do {
664 # put the operator in canonical form
666 $op = substr $op, 1; # remove initial dash
667 $op =~ s/^\s+|\s+$//g;# remove leading/trailing space
668 $op =~ s/\s+/ /g; # compress whitespace
670 # so that -not_foo works correctly
671 $op =~ s/^not_/NOT /i;
673 $self->_debug("Unary OP(-$op) within hashref, recursing...");
674 my ($s, @b) = $self->_where_unary_op($op, $v);
676 # top level vs nested
677 # we assume that handled unary ops will take care of their ()s
679 List::Util::first {$op =~ $_->{regex}} @{$self->{unary_ops}}
681 ( defined $self->{_nested_func_lhs} and $self->{_nested_func_lhs} eq $k )
687 if (is_literal_value ($v) ) {
688 belch 'Hash-pairs consisting of an empty string with a literal are deprecated, and will be removed in 2.0: use -and => [ $literal ] instead';
691 puke "Supplying an empty left hand side argument is not supported in hash-pairs";
695 my $method = $self->_METHOD_FOR_refkind("_where_hashpair", $v);
696 $self->$method($k, $v);
700 push @sql_clauses, $sql;
701 push @all_bind, @bind;
704 return $self->_join_sql_clauses('and', \@sql_clauses, \@all_bind);
707 sub _where_unary_op {
708 my ($self, $op, $rhs) = @_;
710 # top level special ops are illegal in general
711 # this includes the -ident/-value ops (dual purpose unary and special)
712 puke "Illegal use of top-level '-$op'"
713 if ! defined $self->{_nested_func_lhs} and List::Util::first { $op =~ $_->{regex} } @{$self->{special_ops}};
715 if (my $op_entry = List::Util::first { $op =~ $_->{regex} } @{$self->{unary_ops}}) {
716 my $handler = $op_entry->{handler};
718 if (not ref $handler) {
719 if ($op =~ s/ [_\s]? \d+ $//x ) {
720 belch 'Use of [and|or|nest]_N modifiers is deprecated and will be removed in SQLA v2.0. '
721 . "You probably wanted ...-and => [ -$op => COND1, -$op => COND2 ... ]";
723 return $self->$handler($op, $rhs);
725 elsif (ref $handler eq 'CODE') {
726 return $handler->($self, $op, $rhs);
729 puke "Illegal handler for operator $op - expecting a method name or a coderef";
733 $self->_debug("Generic unary OP: $op - recursing as function");
735 $self->_assert_pass_injection_guard($op);
737 my ($sql, @bind) = $self->_SWITCH_refkind($rhs, {
739 puke "Illegal use of top-level '-$op'"
740 unless defined $self->{_nested_func_lhs};
743 $self->_convert('?'),
744 $self->_bindtype($self->{_nested_func_lhs}, $rhs)
748 $self->_recurse_where($rhs)
752 $sql = sprintf('%s %s',
753 $self->_sqlcase($op),
757 return ($sql, @bind);
760 sub _where_op_ANDOR {
761 my ($self, $op, $v) = @_;
763 $self->_SWITCH_refkind($v, {
765 return $self->_where_ARRAYREF($v, $op);
769 return ($op =~ /^or/i)
770 ? $self->_where_ARRAYREF([ map { $_ => $v->{$_} } (sort keys %$v) ], $op)
771 : $self->_where_HASHREF($v);
775 puke "-$op => \\\$scalar makes little sense, use " .
777 ? '[ \$scalar, \%rest_of_conditions ] instead'
778 : '-and => [ \$scalar, \%rest_of_conditions ] instead'
783 puke "-$op => \\[...] makes little sense, use " .
785 ? '[ \[...], \%rest_of_conditions ] instead'
786 : '-and => [ \[...], \%rest_of_conditions ] instead'
790 SCALAR => sub { # permissively interpreted as SQL
791 puke "-$op => \$value makes little sense, use -bool => \$value instead";
795 puke "-$op => undef not supported";
801 my ($self, $op, $v) = @_;
803 $self->_SWITCH_refkind($v, {
805 SCALAR => sub { # permissively interpreted as SQL
806 belch "literal SQL should be -nest => \\'scalar' "
807 . "instead of -nest => 'scalar' ";
812 puke "-$op => undef not supported";
816 $self->_recurse_where($v);
824 my ($self, $op, $v) = @_;
826 my ($s, @b) = $self->_SWITCH_refkind($v, {
827 SCALAR => sub { # interpreted as SQL column
828 $self->_convert($self->_quote($v));
832 puke "-$op => undef not supported";
836 $self->_recurse_where($v);
840 $s = "(NOT $s)" if $op =~ /^not/i;
845 sub _where_op_IDENT {
847 my ($op, $rhs) = splice @_, -2;
848 if (! defined $rhs or length ref $rhs) {
849 puke "-$op requires a single plain scalar argument (a quotable identifier)";
852 # in case we are called as a top level special op (no '=')
855 $_ = $self->_convert($self->_quote($_)) for ($lhs, $rhs);
863 sub _where_op_VALUE {
865 my ($op, $rhs) = splice @_, -2;
867 # in case we are called as a top level special op (no '=')
871 if (! defined $rhs) {
873 ? $self->_where_hashpair_HASHREF($lhs, { -is => undef })
880 (defined $lhs ? $lhs : $self->{_nested_func_lhs}),
887 $self->_convert($self->_quote($lhs)) . ' = ' . $self->_convert('?'),
891 $self->_convert('?'),
897 sub _where_hashpair_ARRAYREF {
898 my ($self, $k, $v) = @_;
901 my @v = @$v; # need copy because of shift below
902 $self->_debug("ARRAY($k) means distribute over elements");
904 # put apart first element if it is an operator (-and, -or)
906 (defined $v[0] && $v[0] =~ /^ - (?: AND|OR ) $/ix)
910 my @distributed = map { {$k => $_} } @v;
913 $self->_debug("OP($op) reinjected into the distributed array");
914 unshift @distributed, $op;
917 my $logic = $op ? substr($op, 1) : '';
919 return $self->_recurse_where(\@distributed, $logic);
922 $self->_debug("empty ARRAY($k) means 0=1");
923 return ($self->{sqlfalse});
927 sub _where_hashpair_HASHREF {
928 my ($self, $k, $v, $logic) = @_;
931 local $self->{_nested_func_lhs} = defined $self->{_nested_func_lhs}
932 ? $self->{_nested_func_lhs}
936 my ($all_sql, @all_bind);
938 for my $orig_op (sort keys %$v) {
939 my $val = $v->{$orig_op};
941 # put the operator in canonical form
944 # FIXME - we need to phase out dash-less ops
945 $op =~ s/^-//; # remove possible initial dash
946 $op =~ s/^\s+|\s+$//g;# remove leading/trailing space
947 $op =~ s/\s+/ /g; # compress whitespace
949 $self->_assert_pass_injection_guard($op);
952 $op =~ s/^is_not/IS NOT/i;
954 # so that -not_foo works correctly
955 $op =~ s/^not_/NOT /i;
957 # another retarded special case: foo => { $op => { -value => undef } }
958 if (ref $val eq 'HASH' and keys %$val == 1 and exists $val->{-value} and ! defined $val->{-value} ) {
964 # CASE: col-value logic modifiers
965 if ($orig_op =~ /^ \- (and|or) $/xi) {
966 ($sql, @bind) = $self->_where_hashpair_HASHREF($k, $val, $1);
968 # CASE: special operators like -in or -between
969 elsif (my $special_op = List::Util::first { $op =~ $_->{regex} } @{$self->{special_ops}}) {
970 my $handler = $special_op->{handler};
972 puke "No handler supplied for special operator $orig_op";
974 elsif (not ref $handler) {
975 ($sql, @bind) = $self->$handler($k, $op, $val);
977 elsif (ref $handler eq 'CODE') {
978 ($sql, @bind) = $handler->($self, $k, $op, $val);
981 puke "Illegal handler for special operator $orig_op - expecting a method name or a coderef";
985 $self->_SWITCH_refkind($val, {
987 ARRAYREF => sub { # CASE: col => {op => \@vals}
988 ($sql, @bind) = $self->_where_field_op_ARRAYREF($k, $op, $val);
991 ARRAYREFREF => sub { # CASE: col => {op => \[$sql, @bind]} (literal SQL with bind)
992 my ($sub_sql, @sub_bind) = @$$val;
993 $self->_assert_bindval_matches_bindtype(@sub_bind);
994 $sql = join ' ', $self->_convert($self->_quote($k)),
995 $self->_sqlcase($op),
1000 UNDEF => sub { # CASE: col => {op => undef} : sql "IS (NOT)? NULL"
1002 $op =~ /^not$/i ? 'is not' # legacy
1003 : $op =~ $self->{equality_op} ? 'is'
1004 : $op =~ $self->{like_op} ? belch("Supplying an undefined argument to '@{[ uc $op]}' is deprecated") && 'is'
1005 : $op =~ $self->{inequality_op} ? 'is not'
1006 : $op =~ $self->{not_like_op} ? belch("Supplying an undefined argument to '@{[ uc $op]}' is deprecated") && 'is not'
1007 : puke "unexpected operator '$orig_op' with undef operand";
1009 $sql = $self->_quote($k) . $self->_sqlcase(" $is null");
1012 FALLBACK => sub { # CASE: col => {op/func => $stuff}
1013 ($sql, @bind) = $self->_where_unary_op($op, $val);
1016 $self->_convert($self->_quote($k)),
1017 $self->{_nested_func_lhs} eq $k ? $sql : "($sql)", # top level vs nested
1023 ($all_sql) = (defined $all_sql and $all_sql) ? $self->_join_sql_clauses($logic, [$all_sql, $sql], []) : $sql;
1024 push @all_bind, @bind;
1026 return ($all_sql, @all_bind);
1029 sub _where_field_IS {
1030 my ($self, $k, $op, $v) = @_;
1032 my ($s) = $self->_SWITCH_refkind($v, {
1035 $self->_convert($self->_quote($k)),
1036 map { $self->_sqlcase($_)} ($op, 'null')
1039 puke "$op can only take undef as argument";
1046 sub _where_field_op_ARRAYREF {
1047 my ($self, $k, $op, $vals) = @_;
1049 my @vals = @$vals; #always work on a copy
1052 $self->_debug(sprintf '%s means multiple elements: [ %s ]',
1054 join(', ', map { defined $_ ? "'$_'" : 'NULL' } @vals ),
1057 # see if the first element is an -and/-or op
1059 if (defined $vals[0] && $vals[0] =~ /^ - (AND|OR) $/ix) {
1064 # a long standing API wart - an attempt to change this behavior during
1065 # the 1.50 series failed *spectacularly*. Warn instead and leave the
1070 (!$logic or $logic eq 'OR')
1072 ($op =~ $self->{inequality_op} or $op =~ $self->{not_like_op})
1075 belch "A multi-element arrayref as an argument to the inequality op '$o' "
1076 . 'is technically equivalent to an always-true 1=1 (you probably wanted '
1077 . "to say ...{ \$inequality_op => [ -and => \@values ] }... instead)"
1081 # distribute $op over each remaining member of @vals, append logic if exists
1082 return $self->_recurse_where([map { {$k => {$op, $_}} } @vals], $logic);
1086 # try to DWIM on equality operators
1088 $op =~ $self->{equality_op} ? $self->{sqlfalse}
1089 : $op =~ $self->{like_op} ? belch("Supplying an empty arrayref to '@{[ uc $op]}' is deprecated") && $self->{sqlfalse}
1090 : $op =~ $self->{inequality_op} ? $self->{sqltrue}
1091 : $op =~ $self->{not_like_op} ? belch("Supplying an empty arrayref to '@{[ uc $op]}' is deprecated") && $self->{sqltrue}
1092 : puke "operator '$op' applied on an empty array (field '$k')";
1097 sub _where_hashpair_SCALARREF {
1098 my ($self, $k, $v) = @_;
1099 $self->_debug("SCALAR($k) means literal SQL: $$v");
1100 my $sql = $self->_quote($k) . " " . $$v;
1104 # literal SQL with bind
1105 sub _where_hashpair_ARRAYREFREF {
1106 my ($self, $k, $v) = @_;
1107 $self->_debug("REF($k) means literal SQL: @${$v}");
1108 my ($sql, @bind) = @$$v;
1109 $self->_assert_bindval_matches_bindtype(@bind);
1110 $sql = $self->_quote($k) . " " . $sql;
1111 return ($sql, @bind );
1114 # literal SQL without bind
1115 sub _where_hashpair_SCALAR {
1116 my ($self, $k, $v) = @_;
1117 $self->_debug("NOREF($k) means simple key=val: $k $self->{cmp} $v");
1118 return ($self->_where_hashpair_HASHREF($k, { $self->{cmp} => $v }));
1122 sub _where_hashpair_UNDEF {
1123 my ($self, $k, $v) = @_;
1124 $self->_debug("UNDEF($k) means IS NULL");
1125 return $self->_where_hashpair_HASHREF($k, { -is => undef });
1128 #======================================================================
1129 # WHERE: TOP-LEVEL OTHERS (SCALARREF, SCALAR, UNDEF)
1130 #======================================================================
1133 sub _where_SCALARREF {
1134 my ($self, $where) = @_;
1137 $self->_debug("SCALAR(*top) means literal SQL: $$where");
1143 my ($self, $where) = @_;
1146 $self->_debug("NOREF(*top) means literal SQL: $where");
1157 #======================================================================
1158 # WHERE: BUILTIN SPECIAL OPERATORS (-in, -between)
1159 #======================================================================
1162 sub _where_field_BETWEEN {
1163 my ($self, $k, $op, $vals) = @_;
1165 my ($label, $and, $placeholder);
1166 $label = $self->_convert($self->_quote($k));
1167 $and = ' ' . $self->_sqlcase('and') . ' ';
1168 $placeholder = $self->_convert('?');
1169 $op = $self->_sqlcase($op);
1171 my $invalid_args = "Operator '$op' requires either an arrayref with two defined values or expressions, or a single literal scalarref/arrayref-ref";
1173 my ($clause, @bind) = $self->_SWITCH_refkind($vals, {
1174 ARRAYREFREF => sub {
1175 my ($s, @b) = @$$vals;
1176 $self->_assert_bindval_matches_bindtype(@b);
1183 puke $invalid_args if @$vals != 2;
1185 my (@all_sql, @all_bind);
1186 foreach my $val (@$vals) {
1187 my ($sql, @bind) = $self->_SWITCH_refkind($val, {
1189 return ($placeholder, $self->_bindtype($k, $val) );
1194 ARRAYREFREF => sub {
1195 my ($sql, @bind) = @$$val;
1196 $self->_assert_bindval_matches_bindtype(@bind);
1197 return ($sql, @bind);
1200 my ($func, $arg, @rest) = %$val;
1201 puke "Only simple { -func => arg } functions accepted as sub-arguments to BETWEEN"
1202 if (@rest or $func !~ /^ \- (.+)/x);
1203 $self->_where_unary_op($1 => $arg);
1209 push @all_sql, $sql;
1210 push @all_bind, @bind;
1214 (join $and, @all_sql),
1223 my $sql = "( $label $op $clause )";
1224 return ($sql, @bind)
1228 sub _where_field_IN {
1229 my ($self, $k, $op, $vals) = @_;
1231 # backwards compatibility: if scalar, force into an arrayref
1232 $vals = [$vals] if defined $vals && ! ref $vals;
1234 my ($label) = $self->_convert($self->_quote($k));
1235 my ($placeholder) = $self->_convert('?');
1236 $op = $self->_sqlcase($op);
1238 my ($sql, @bind) = $self->_SWITCH_refkind($vals, {
1239 ARRAYREF => sub { # list of choices
1240 if (@$vals) { # nonempty list
1241 my (@all_sql, @all_bind);
1243 for my $val (@$vals) {
1244 my ($sql, @bind) = $self->_SWITCH_refkind($val, {
1246 return ($placeholder, $val);
1251 ARRAYREFREF => sub {
1252 my ($sql, @bind) = @$$val;
1253 $self->_assert_bindval_matches_bindtype(@bind);
1254 return ($sql, @bind);
1257 my ($func, $arg, @rest) = %$val;
1258 puke "Only simple { -func => arg } functions accepted as sub-arguments to IN"
1259 if (@rest or $func !~ /^ \- (.+)/x);
1260 $self->_where_unary_op($1 => $arg);
1264 'SQL::Abstract before v1.75 used to generate incorrect SQL when the '
1265 . "-$op operator was given an undef-containing list: !!!AUDIT YOUR CODE "
1266 . 'AND DATA!!! (the upcoming Data::Query-based version of SQL::Abstract '
1267 . 'will emit the logically correct SQL instead of raising this exception)'
1271 push @all_sql, $sql;
1272 push @all_bind, @bind;
1276 sprintf('%s %s ( %s )',
1279 join(', ', @all_sql)
1281 $self->_bindtype($k, @all_bind),
1284 else { # empty list: some databases won't understand "IN ()", so DWIM
1285 my $sql = ($op =~ /\bnot\b/i) ? $self->{sqltrue} : $self->{sqlfalse};
1290 SCALARREF => sub { # literal SQL
1291 my $sql = $self->_open_outer_paren($$vals);
1292 return ("$label $op ( $sql )");
1294 ARRAYREFREF => sub { # literal SQL with bind
1295 my ($sql, @bind) = @$$vals;
1296 $self->_assert_bindval_matches_bindtype(@bind);
1297 $sql = $self->_open_outer_paren($sql);
1298 return ("$label $op ( $sql )", @bind);
1302 puke "Argument passed to the '$op' operator can not be undefined";
1306 puke "special op $op requires an arrayref (or scalarref/arrayref-ref)";
1310 return ($sql, @bind);
1313 # Some databases (SQLite) treat col IN (1, 2) different from
1314 # col IN ( (1, 2) ). Use this to strip all outer parens while
1315 # adding them back in the corresponding method
1316 sub _open_outer_paren {
1317 my ($self, $sql) = @_;
1319 while (my ($inner) = $sql =~ /^ \s* \( (.*) \) \s* $/xs) {
1321 # there are closing parens inside, need the heavy duty machinery
1322 # to reevaluate the extraction starting from $sql (full reevaluation)
1323 if ($inner =~ /\)/) {
1324 require Text::Balanced;
1326 my (undef, $remainder) = do {
1327 # idiotic design - writes to $@ but *DOES NOT* throw exceptions
1329 Text::Balanced::extract_bracketed($sql, '()', qr/\s*/);
1332 # the entire expression needs to be a balanced bracketed thing
1333 # (after an extract no remainder sans trailing space)
1334 last if defined $remainder and $remainder =~ /\S/;
1344 #======================================================================
1346 #======================================================================
1349 my ($self, $arg) = @_;
1352 for my $c ($self->_order_by_chunks($arg) ) {
1353 $self->_SWITCH_refkind($c, {
1354 SCALAR => sub { push @sql, $c },
1355 ARRAYREF => sub { push @sql, shift @$c; push @bind, @$c },
1361 $self->_sqlcase(' order by'),
1367 return wantarray ? ($sql, @bind) : $sql;
1370 sub _order_by_chunks {
1371 my ($self, $arg) = @_;
1373 return $self->_SWITCH_refkind($arg, {
1376 map { $self->_order_by_chunks($_ ) } @$arg;
1379 ARRAYREFREF => sub {
1380 my ($s, @b) = @$$arg;
1381 $self->_assert_bindval_matches_bindtype(@b);
1385 SCALAR => sub {$self->_quote($arg)},
1387 UNDEF => sub {return () },
1389 SCALARREF => sub {$$arg}, # literal SQL, no quoting
1392 # get first pair in hash
1393 my ($key, $val, @rest) = %$arg;
1395 return () unless $key;
1397 if (@rest or not $key =~ /^-(desc|asc)/i) {
1398 puke "hash passed to _order_by must have exactly one key (-desc or -asc)";
1404 for my $c ($self->_order_by_chunks($val)) {
1407 $self->_SWITCH_refkind($c, {
1412 ($sql, @bind) = @$c;
1416 $sql = $sql . ' ' . $self->_sqlcase($direction);
1418 push @ret, [ $sql, @bind];
1427 #======================================================================
1428 # DATASOURCE (FOR NOW, JUST PLAIN TABLE OR LIST OF TABLES)
1429 #======================================================================
1434 $self->_SWITCH_refkind($from, {
1435 ARRAYREF => sub {join ', ', map { $self->_quote($_) } @$from;},
1436 SCALAR => sub {$self->_quote($from)},
1437 SCALARREF => sub {$$from},
1442 #======================================================================
1444 #======================================================================
1446 # highly optimized, as it's called way too often
1448 # my ($self, $label) = @_;
1450 return '' unless defined $_[1];
1451 return ${$_[1]} if ref($_[1]) eq 'SCALAR';
1453 $_[0]->{quote_char} or
1454 ($_[0]->_assert_pass_injection_guard($_[1]), return $_[1]);
1456 my $qref = ref $_[0]->{quote_char};
1458 !$qref ? ($_[0]->{quote_char}, $_[0]->{quote_char})
1459 : ($qref eq 'ARRAY') ? @{$_[0]->{quote_char}}
1460 : puke "Unsupported quote_char format: $_[0]->{quote_char}";
1462 my $esc = $_[0]->{escape_char} || $r;
1464 # parts containing * are naturally unquoted
1465 return join($_[0]->{name_sep}||'', map
1466 +( $_ eq '*' ? $_ : do { (my $n = $_) =~ s/(\Q$esc\E|\Q$r\E)/$esc$1/g; $l . $n . $r } ),
1467 ( $_[0]->{name_sep} ? split (/\Q$_[0]->{name_sep}\E/, $_[1] ) : $_[1] )
1472 # Conversion, if applicable
1474 #my ($self, $arg) = @_;
1475 if ($_[0]->{convert}) {
1476 return $_[0]->_sqlcase($_[0]->{convert}) .'(' . $_[1] . ')';
1483 #my ($self, $col, @vals) = @_;
1484 # called often - tighten code
1485 return $_[0]->{bindtype} eq 'columns'
1486 ? map {[$_[1], $_]} @_[2 .. $#_]
1491 # Dies if any element of @bind is not in [colname => value] format
1492 # if bindtype is 'columns'.
1493 sub _assert_bindval_matches_bindtype {
1494 # my ($self, @bind) = @_;
1496 if ($self->{bindtype} eq 'columns') {
1498 if (!defined $_ || ref($_) ne 'ARRAY' || @$_ != 2) {
1499 puke "bindtype 'columns' selected, you need to pass: [column_name => bind_value]"
1505 sub _join_sql_clauses {
1506 my ($self, $logic, $clauses_aref, $bind_aref) = @_;
1508 if (@$clauses_aref > 1) {
1509 my $join = " " . $self->_sqlcase($logic) . " ";
1510 my $sql = '( ' . join($join, @$clauses_aref) . ' )';
1511 return ($sql, @$bind_aref);
1513 elsif (@$clauses_aref) {
1514 return ($clauses_aref->[0], @$bind_aref); # no parentheses
1517 return (); # if no SQL, ignore @$bind_aref
1522 # Fix SQL case, if so requested
1524 # LDNOTE: if $self->{case} is true, then it contains 'lower', so we
1525 # don't touch the argument ... crooked logic, but let's not change it!
1526 return $_[0]->{case} ? $_[1] : uc($_[1]);
1530 #======================================================================
1531 # DISPATCHING FROM REFKIND
1532 #======================================================================
1535 my ($self, $data) = @_;
1537 return 'UNDEF' unless defined $data;
1539 # blessed objects are treated like scalars
1540 my $ref = (Scalar::Util::blessed $data) ? '' : ref $data;
1542 return 'SCALAR' unless $ref;
1545 while ($ref eq 'REF') {
1547 $ref = (Scalar::Util::blessed $data) ? '' : ref $data;
1551 return ($ref||'SCALAR') . ('REF' x $n_steps);
1555 my ($self, $data) = @_;
1556 my @try = ($self->_refkind($data));
1557 push @try, 'SCALAR_or_UNDEF' if $try[0] eq 'SCALAR' || $try[0] eq 'UNDEF';
1558 push @try, 'FALLBACK';
1562 sub _METHOD_FOR_refkind {
1563 my ($self, $meth_prefix, $data) = @_;
1566 for (@{$self->_try_refkind($data)}) {
1567 $method = $self->can($meth_prefix."_".$_)
1571 return $method || puke "cannot dispatch on '$meth_prefix' for ".$self->_refkind($data);
1575 sub _SWITCH_refkind {
1576 my ($self, $data, $dispatch_table) = @_;
1579 for (@{$self->_try_refkind($data)}) {
1580 $coderef = $dispatch_table->{$_}
1584 puke "no dispatch entry for ".$self->_refkind($data)
1593 #======================================================================
1594 # VALUES, GENERATE, AUTOLOAD
1595 #======================================================================
1597 # LDNOTE: original code from nwiger, didn't touch code in that section
1598 # I feel the AUTOLOAD stuff should not be the default, it should
1599 # only be activated on explicit demand by user.
1603 my $data = shift || return;
1604 puke "Argument to ", __PACKAGE__, "->values must be a \\%hash"
1605 unless ref $data eq 'HASH';
1608 foreach my $k (sort keys %$data) {
1609 my $v = $data->{$k};
1610 $self->_SWITCH_refkind($v, {
1612 if ($self->{array_datatypes}) { # array datatype
1613 push @all_bind, $self->_bindtype($k, $v);
1615 else { # literal SQL with bind
1616 my ($sql, @bind) = @$v;
1617 $self->_assert_bindval_matches_bindtype(@bind);
1618 push @all_bind, @bind;
1621 ARRAYREFREF => sub { # literal SQL with bind
1622 my ($sql, @bind) = @${$v};
1623 $self->_assert_bindval_matches_bindtype(@bind);
1624 push @all_bind, @bind;
1626 SCALARREF => sub { # literal SQL without bind
1628 SCALAR_or_UNDEF => sub {
1629 push @all_bind, $self->_bindtype($k, $v);
1640 my(@sql, @sqlq, @sqlv);
1644 if ($ref eq 'HASH') {
1645 for my $k (sort keys %$_) {
1648 my $label = $self->_quote($k);
1649 if ($r eq 'ARRAY') {
1650 # literal SQL with bind
1651 my ($sql, @bind) = @$v;
1652 $self->_assert_bindval_matches_bindtype(@bind);
1653 push @sqlq, "$label = $sql";
1655 } elsif ($r eq 'SCALAR') {
1656 # literal SQL without bind
1657 push @sqlq, "$label = $$v";
1659 push @sqlq, "$label = ?";
1660 push @sqlv, $self->_bindtype($k, $v);
1663 push @sql, $self->_sqlcase('set'), join ', ', @sqlq;
1664 } elsif ($ref eq 'ARRAY') {
1665 # unlike insert(), assume these are ONLY the column names, i.e. for SQL
1668 if ($r eq 'ARRAY') { # literal SQL with bind
1669 my ($sql, @bind) = @$v;
1670 $self->_assert_bindval_matches_bindtype(@bind);
1673 } elsif ($r eq 'SCALAR') { # literal SQL without bind
1674 # embedded literal SQL
1681 push @sql, '(' . join(', ', @sqlq) . ')';
1682 } elsif ($ref eq 'SCALAR') {
1686 # strings get case twiddled
1687 push @sql, $self->_sqlcase($_);
1691 my $sql = join ' ', @sql;
1693 # this is pretty tricky
1694 # if ask for an array, return ($stmt, @bind)
1695 # otherwise, s/?/shift @sqlv/ to put it inline
1697 return ($sql, @sqlv);
1699 1 while $sql =~ s/\?/my $d = shift(@sqlv);
1700 ref $d ? $d->[1] : $d/e;
1709 # This allows us to check for a local, then _form, attr
1711 my($name) = $AUTOLOAD =~ /.*::(.+)/;
1712 return $self->generate($name, @_);
1723 SQL::Abstract - Generate SQL from Perl data structures
1729 my $sql = SQL::Abstract->new;
1731 my($stmt, @bind) = $sql->select($source, \@fields, \%where, $order);
1733 my($stmt, @bind) = $sql->insert($table, \%fieldvals || \@values);
1735 my($stmt, @bind) = $sql->update($table, \%fieldvals, \%where);
1737 my($stmt, @bind) = $sql->delete($table, \%where);
1739 # Then, use these in your DBI statements
1740 my $sth = $dbh->prepare($stmt);
1741 $sth->execute(@bind);
1743 # Just generate the WHERE clause
1744 my($stmt, @bind) = $sql->where(\%where, $order);
1746 # Return values in the same order, for hashed queries
1747 # See PERFORMANCE section for more details
1748 my @bind = $sql->values(\%fieldvals);
1752 This module was inspired by the excellent L<DBIx::Abstract>.
1753 However, in using that module I found that what I really wanted
1754 to do was generate SQL, but still retain complete control over my
1755 statement handles and use the DBI interface. So, I set out to
1756 create an abstract SQL generation module.
1758 While based on the concepts used by L<DBIx::Abstract>, there are
1759 several important differences, especially when it comes to WHERE
1760 clauses. I have modified the concepts used to make the SQL easier
1761 to generate from Perl data structures and, IMO, more intuitive.
1762 The underlying idea is for this module to do what you mean, based
1763 on the data structures you provide it. The big advantage is that
1764 you don't have to modify your code every time your data changes,
1765 as this module figures it out.
1767 To begin with, an SQL INSERT is as easy as just specifying a hash
1768 of C<key=value> pairs:
1771 name => 'Jimbo Bobson',
1772 phone => '123-456-7890',
1773 address => '42 Sister Lane',
1774 city => 'St. Louis',
1775 state => 'Louisiana',
1778 The SQL can then be generated with this:
1780 my($stmt, @bind) = $sql->insert('people', \%data);
1782 Which would give you something like this:
1784 $stmt = "INSERT INTO people
1785 (address, city, name, phone, state)
1786 VALUES (?, ?, ?, ?, ?)";
1787 @bind = ('42 Sister Lane', 'St. Louis', 'Jimbo Bobson',
1788 '123-456-7890', 'Louisiana');
1790 These are then used directly in your DBI code:
1792 my $sth = $dbh->prepare($stmt);
1793 $sth->execute(@bind);
1795 =head2 Inserting and Updating Arrays
1797 If your database has array types (like for example Postgres),
1798 activate the special option C<< array_datatypes => 1 >>
1799 when creating the C<SQL::Abstract> object.
1800 Then you may use an arrayref to insert and update database array types:
1802 my $sql = SQL::Abstract->new(array_datatypes => 1);
1804 planets => [qw/Mercury Venus Earth Mars/]
1807 my($stmt, @bind) = $sql->insert('solar_system', \%data);
1811 $stmt = "INSERT INTO solar_system (planets) VALUES (?)"
1813 @bind = (['Mercury', 'Venus', 'Earth', 'Mars']);
1816 =head2 Inserting and Updating SQL
1818 In order to apply SQL functions to elements of your C<%data> you may
1819 specify a reference to an arrayref for the given hash value. For example,
1820 if you need to execute the Oracle C<to_date> function on a value, you can
1821 say something like this:
1825 date_entered => \[ "to_date(?,'MM/DD/YYYY')", "03/02/2003" ],
1828 The first value in the array is the actual SQL. Any other values are
1829 optional and would be included in the bind values array. This gives
1832 my($stmt, @bind) = $sql->insert('people', \%data);
1834 $stmt = "INSERT INTO people (name, date_entered)
1835 VALUES (?, to_date(?,'MM/DD/YYYY'))";
1836 @bind = ('Bill', '03/02/2003');
1838 An UPDATE is just as easy, all you change is the name of the function:
1840 my($stmt, @bind) = $sql->update('people', \%data);
1842 Notice that your C<%data> isn't touched; the module will generate
1843 the appropriately quirky SQL for you automatically. Usually you'll
1844 want to specify a WHERE clause for your UPDATE, though, which is
1845 where handling C<%where> hashes comes in handy...
1847 =head2 Complex where statements
1849 This module can generate pretty complicated WHERE statements
1850 easily. For example, simple C<key=value> pairs are taken to mean
1851 equality, and if you want to see if a field is within a set
1852 of values, you can use an arrayref. Let's say we wanted to
1853 SELECT some data based on this criteria:
1856 requestor => 'inna',
1857 worker => ['nwiger', 'rcwe', 'sfz'],
1858 status => { '!=', 'completed' }
1861 my($stmt, @bind) = $sql->select('tickets', '*', \%where);
1863 The above would give you something like this:
1865 $stmt = "SELECT * FROM tickets WHERE
1866 ( requestor = ? ) AND ( status != ? )
1867 AND ( worker = ? OR worker = ? OR worker = ? )";
1868 @bind = ('inna', 'completed', 'nwiger', 'rcwe', 'sfz');
1870 Which you could then use in DBI code like so:
1872 my $sth = $dbh->prepare($stmt);
1873 $sth->execute(@bind);
1879 The methods are simple. There's one for every major SQL operation,
1880 and a constructor you use first. The arguments are specified in a
1881 similar order for each method (table, then fields, then a where
1882 clause) to try and simplify things.
1884 =head2 new(option => 'value')
1886 The C<new()> function takes a list of options and values, and returns
1887 a new B<SQL::Abstract> object which can then be used to generate SQL
1888 through the methods below. The options accepted are:
1894 If set to 'lower', then SQL will be generated in all lowercase. By
1895 default SQL is generated in "textbook" case meaning something like:
1897 SELECT a_field FROM a_table WHERE some_field LIKE '%someval%'
1899 Any setting other than 'lower' is ignored.
1903 This determines what the default comparison operator is. By default
1904 it is C<=>, meaning that a hash like this:
1906 %where = (name => 'nwiger', email => 'nate@wiger.org');
1908 Will generate SQL like this:
1910 WHERE name = 'nwiger' AND email = 'nate@wiger.org'
1912 However, you may want loose comparisons by default, so if you set
1913 C<cmp> to C<like> you would get SQL such as:
1915 WHERE name like 'nwiger' AND email like 'nate@wiger.org'
1917 You can also override the comparison on an individual basis - see
1918 the huge section on L</"WHERE CLAUSES"> at the bottom.
1920 =item sqltrue, sqlfalse
1922 Expressions for inserting boolean values within SQL statements.
1923 By default these are C<1=1> and C<1=0>. They are used
1924 by the special operators C<-in> and C<-not_in> for generating
1925 correct SQL even when the argument is an empty array (see below).
1929 This determines the default logical operator for multiple WHERE
1930 statements in arrays or hashes. If absent, the default logic is "or"
1931 for arrays, and "and" for hashes. This means that a WHERE
1935 event_date => {'>=', '2/13/99'},
1936 event_date => {'<=', '4/24/03'},
1939 will generate SQL like this:
1941 WHERE event_date >= '2/13/99' OR event_date <= '4/24/03'
1943 This is probably not what you want given this query, though (look
1944 at the dates). To change the "OR" to an "AND", simply specify:
1946 my $sql = SQL::Abstract->new(logic => 'and');
1948 Which will change the above C<WHERE> to:
1950 WHERE event_date >= '2/13/99' AND event_date <= '4/24/03'
1952 The logic can also be changed locally by inserting
1953 a modifier in front of an arrayref:
1955 @where = (-and => [event_date => {'>=', '2/13/99'},
1956 event_date => {'<=', '4/24/03'} ]);
1958 See the L</"WHERE CLAUSES"> section for explanations.
1962 This will automatically convert comparisons using the specified SQL
1963 function for both column and value. This is mostly used with an argument
1964 of C<upper> or C<lower>, so that the SQL will have the effect of
1965 case-insensitive "searches". For example, this:
1967 $sql = SQL::Abstract->new(convert => 'upper');
1968 %where = (keywords => 'MaKe iT CAse inSeNSItive');
1970 Will turn out the following SQL:
1972 WHERE upper(keywords) like upper('MaKe iT CAse inSeNSItive')
1974 The conversion can be C<upper()>, C<lower()>, or any other SQL function
1975 that can be applied symmetrically to fields (actually B<SQL::Abstract> does
1976 not validate this option; it will just pass through what you specify verbatim).
1980 This is a kludge because many databases suck. For example, you can't
1981 just bind values using DBI's C<execute()> for Oracle C<CLOB> or C<BLOB> fields.
1982 Instead, you have to use C<bind_param()>:
1984 $sth->bind_param(1, 'reg data');
1985 $sth->bind_param(2, $lots, {ora_type => ORA_CLOB});
1987 The problem is, B<SQL::Abstract> will normally just return a C<@bind> array,
1988 which loses track of which field each slot refers to. Fear not.
1990 If you specify C<bindtype> in new, you can determine how C<@bind> is returned.
1991 Currently, you can specify either C<normal> (default) or C<columns>. If you
1992 specify C<columns>, you will get an array that looks like this:
1994 my $sql = SQL::Abstract->new(bindtype => 'columns');
1995 my($stmt, @bind) = $sql->insert(...);
1998 [ 'column1', 'value1' ],
1999 [ 'column2', 'value2' ],
2000 [ 'column3', 'value3' ],
2003 You can then iterate through this manually, using DBI's C<bind_param()>.
2005 $sth->prepare($stmt);
2008 my($col, $data) = @$_;
2009 if ($col eq 'details' || $col eq 'comments') {
2010 $sth->bind_param($i, $data, {ora_type => ORA_CLOB});
2011 } elsif ($col eq 'image') {
2012 $sth->bind_param($i, $data, {ora_type => ORA_BLOB});
2014 $sth->bind_param($i, $data);
2018 $sth->execute; # execute without @bind now
2020 Now, why would you still use B<SQL::Abstract> if you have to do this crap?
2021 Basically, the advantage is still that you don't have to care which fields
2022 are or are not included. You could wrap that above C<for> loop in a simple
2023 sub called C<bind_fields()> or something and reuse it repeatedly. You still
2024 get a layer of abstraction over manual SQL specification.
2026 Note that if you set L</bindtype> to C<columns>, the C<\[ $sql, @bind ]>
2027 construct (see L</Literal SQL with placeholders and bind values (subqueries)>)
2028 will expect the bind values in this format.
2032 This is the character that a table or column name will be quoted
2033 with. By default this is an empty string, but you could set it to
2034 the character C<`>, to generate SQL like this:
2036 SELECT `a_field` FROM `a_table` WHERE `some_field` LIKE '%someval%'
2038 Alternatively, you can supply an array ref of two items, the first being the left
2039 hand quote character, and the second the right hand quote character. For
2040 example, you could supply C<['[',']']> for SQL Server 2000 compliant quotes
2041 that generates SQL like this:
2043 SELECT [a_field] FROM [a_table] WHERE [some_field] LIKE '%someval%'
2045 Quoting is useful if you have tables or columns names that are reserved
2046 words in your database's SQL dialect.
2050 This is the character that will be used to escape L</quote_char>s appearing
2051 in an identifier before it has been quoted.
2053 The parameter default in case of a single L</quote_char> character is the quote
2056 When opening-closing-style quoting is used (L</quote_char> is an arrayref)
2057 this parameter defaults to the B<closing (right)> L</quote_char>. Occurrences
2058 of the B<opening (left)> L</quote_char> within the identifier are currently left
2059 untouched. The default for opening-closing-style quotes may change in future
2060 versions, thus you are B<strongly encouraged> to specify the escape character
2065 This is the character that separates a table and column name. It is
2066 necessary to specify this when the C<quote_char> option is selected,
2067 so that tables and column names can be individually quoted like this:
2069 SELECT `table`.`one_field` FROM `table` WHERE `table`.`other_field` = 1
2071 =item injection_guard
2073 A regular expression C<qr/.../> that is applied to any C<-function> and unquoted
2074 column name specified in a query structure. This is a safety mechanism to avoid
2075 injection attacks when mishandling user input e.g.:
2077 my %condition_as_column_value_pairs = get_values_from_user();
2078 $sqla->select( ... , \%condition_as_column_value_pairs );
2080 If the expression matches an exception is thrown. Note that literal SQL
2081 supplied via C<\'...'> or C<\['...']> is B<not> checked in any way.
2083 Defaults to checking for C<;> and the C<GO> keyword (TransactSQL)
2085 =item array_datatypes
2087 When this option is true, arrayrefs in INSERT or UPDATE are
2088 interpreted as array datatypes and are passed directly
2090 When this option is false, arrayrefs are interpreted
2091 as literal SQL, just like refs to arrayrefs
2092 (but this behavior is for backwards compatibility; when writing
2093 new queries, use the "reference to arrayref" syntax
2099 Takes a reference to a list of "special operators"
2100 to extend the syntax understood by L<SQL::Abstract>.
2101 See section L</"SPECIAL OPERATORS"> for details.
2105 Takes a reference to a list of "unary operators"
2106 to extend the syntax understood by L<SQL::Abstract>.
2107 See section L</"UNARY OPERATORS"> for details.
2113 =head2 insert($table, \@values || \%fieldvals, \%options)
2115 This is the simplest function. You simply give it a table name
2116 and either an arrayref of values or hashref of field/value pairs.
2117 It returns an SQL INSERT statement and a list of bind values.
2118 See the sections on L</"Inserting and Updating Arrays"> and
2119 L</"Inserting and Updating SQL"> for information on how to insert
2120 with those data types.
2122 The optional C<\%options> hash reference may contain additional
2123 options to generate the insert SQL. Currently supported options
2130 Takes either a scalar of raw SQL fields, or an array reference of
2131 field names, and adds on an SQL C<RETURNING> statement at the end.
2132 This allows you to return data generated by the insert statement
2133 (such as row IDs) without performing another C<SELECT> statement.
2134 Note, however, this is not part of the SQL standard and may not
2135 be supported by all database engines.
2139 =head2 update($table, \%fieldvals, \%where, \%options)
2141 This takes a table, hashref of field/value pairs, and an optional
2142 hashref L<WHERE clause|/WHERE CLAUSES>. It returns an SQL UPDATE function and a list
2144 See the sections on L</"Inserting and Updating Arrays"> and
2145 L</"Inserting and Updating SQL"> for information on how to insert
2146 with those data types.
2148 The optional C<\%options> hash reference may contain additional
2149 options to generate the update SQL. Currently supported options
2156 See the C<returning> option to
2157 L<insert|/insert($table, \@values || \%fieldvals, \%options)>.
2161 =head2 select($source, $fields, $where, $order)
2163 This returns a SQL SELECT statement and associated list of bind values, as
2164 specified by the arguments:
2170 Specification of the 'FROM' part of the statement.
2171 The argument can be either a plain scalar (interpreted as a table
2172 name, will be quoted), or an arrayref (interpreted as a list
2173 of table names, joined by commas, quoted), or a scalarref
2174 (literal SQL, not quoted).
2178 Specification of the list of fields to retrieve from
2180 The argument can be either an arrayref (interpreted as a list
2181 of field names, will be joined by commas and quoted), or a
2182 plain scalar (literal SQL, not quoted).
2183 Please observe that this API is not as flexible as that of
2184 the first argument C<$source>, for backwards compatibility reasons.
2188 Optional argument to specify the WHERE part of the query.
2189 The argument is most often a hashref, but can also be
2190 an arrayref or plain scalar --
2191 see section L<WHERE clause|/"WHERE CLAUSES"> for details.
2195 Optional argument to specify the ORDER BY part of the query.
2196 The argument can be a scalar, a hashref or an arrayref
2197 -- see section L<ORDER BY clause|/"ORDER BY CLAUSES">
2203 =head2 delete($table, \%where, \%options)
2205 This takes a table name and optional hashref L<WHERE clause|/WHERE CLAUSES>.
2206 It returns an SQL DELETE statement and list of bind values.
2208 The optional C<\%options> hash reference may contain additional
2209 options to generate the delete SQL. Currently supported options
2216 See the C<returning> option to
2217 L<insert|/insert($table, \@values || \%fieldvals, \%options)>.
2221 =head2 where(\%where, $order)
2223 This is used to generate just the WHERE clause. For example,
2224 if you have an arbitrary data structure and know what the
2225 rest of your SQL is going to look like, but want an easy way
2226 to produce a WHERE clause, use this. It returns an SQL WHERE
2227 clause and list of bind values.
2230 =head2 values(\%data)
2232 This just returns the values from the hash C<%data>, in the same
2233 order that would be returned from any of the other above queries.
2234 Using this allows you to markedly speed up your queries if you
2235 are affecting lots of rows. See below under the L</"PERFORMANCE"> section.
2237 =head2 generate($any, 'number', $of, \@data, $struct, \%types)
2239 Warning: This is an experimental method and subject to change.
2241 This returns arbitrarily generated SQL. It's a really basic shortcut.
2242 It will return two different things, depending on return context:
2244 my($stmt, @bind) = $sql->generate('create table', \$table, \@fields);
2245 my $stmt_and_val = $sql->generate('create table', \$table, \@fields);
2247 These would return the following:
2249 # First calling form
2250 $stmt = "CREATE TABLE test (?, ?)";
2251 @bind = (field1, field2);
2253 # Second calling form
2254 $stmt_and_val = "CREATE TABLE test (field1, field2)";
2256 Depending on what you're trying to do, it's up to you to choose the correct
2257 format. In this example, the second form is what you would want.
2261 $sql->generate('alter session', { nls_date_format => 'MM/YY' });
2265 ALTER SESSION SET nls_date_format = 'MM/YY'
2267 You get the idea. Strings get their case twiddled, but everything
2268 else remains verbatim.
2270 =head1 EXPORTABLE FUNCTIONS
2272 =head2 is_plain_value
2274 Determines if the supplied argument is a plain value as understood by this
2279 =item * The value is C<undef>
2281 =item * The value is a non-reference
2283 =item * The value is an object with stringification overloading
2285 =item * The value is of the form C<< { -value => $anything } >>
2289 On failure returns C<undef>, on success returns a B<scalar> reference
2290 to the original supplied argument.
2296 The stringification overloading detection is rather advanced: it takes
2297 into consideration not only the presence of a C<""> overload, but if that
2298 fails also checks for enabled
2299 L<autogenerated versions of C<"">|overload/Magic Autogeneration>, based
2300 on either C<0+> or C<bool>.
2302 Unfortunately testing in the field indicates that this
2303 detection B<< may tickle a latent bug in perl versions before 5.018 >>,
2304 but only when very large numbers of stringifying objects are involved.
2305 At the time of writing ( Sep 2014 ) there is no clear explanation of
2306 the direct cause, nor is there a manageably small test case that reliably
2307 reproduces the problem.
2309 If you encounter any of the following exceptions in B<random places within
2310 your application stack> - this module may be to blame:
2312 Operation "ne": no method found,
2313 left argument in overloaded package <something>,
2314 right argument in overloaded package <something>
2318 Stub found while resolving method "???" overloading """" in package <something>
2320 If you fall victim to the above - please attempt to reduce the problem
2321 to something that could be sent to the L<SQL::Abstract developers
2322 |DBIx::Class/GETTING HELP/SUPPORT>
2323 (either publicly or privately). As a workaround in the meantime you can
2324 set C<$ENV{SQLA_ISVALUE_IGNORE_AUTOGENERATED_STRINGIFICATION}> to a true
2325 value, which will most likely eliminate your problem (at the expense of
2326 not being able to properly detect exotic forms of stringification).
2328 This notice and environment variable will be removed in a future version,
2329 as soon as the underlying problem is found and a reliable workaround is
2334 =head2 is_literal_value
2336 Determines if the supplied argument is a literal value as understood by this
2341 =item * C<\$sql_string>
2343 =item * C<\[ $sql_string, @bind_values ]>
2347 On failure returns C<undef>, on success returns an B<array> reference
2348 containing the unpacked version of the supplied literal SQL and bind values.
2350 =head1 WHERE CLAUSES
2354 This module uses a variation on the idea from L<DBIx::Abstract>. It
2355 is B<NOT>, repeat I<not> 100% compatible. B<The main logic of this
2356 module is that things in arrays are OR'ed, and things in hashes
2359 The easiest way to explain is to show lots of examples. After
2360 each C<%where> hash shown, it is assumed you used:
2362 my($stmt, @bind) = $sql->where(\%where);
2364 However, note that the C<%where> hash can be used directly in any
2365 of the other functions as well, as described above.
2367 =head2 Key-value pairs
2369 So, let's get started. To begin, a simple hash:
2373 status => 'completed'
2376 Is converted to SQL C<key = val> statements:
2378 $stmt = "WHERE user = ? AND status = ?";
2379 @bind = ('nwiger', 'completed');
2381 One common thing I end up doing is having a list of values that
2382 a field can be in. To do this, simply specify a list inside of
2387 status => ['assigned', 'in-progress', 'pending'];
2390 This simple code will create the following:
2392 $stmt = "WHERE user = ? AND ( status = ? OR status = ? OR status = ? )";
2393 @bind = ('nwiger', 'assigned', 'in-progress', 'pending');
2395 A field associated to an empty arrayref will be considered a
2396 logical false and will generate 0=1.
2398 =head2 Tests for NULL values
2400 If the value part is C<undef> then this is converted to SQL <IS NULL>
2409 $stmt = "WHERE user = ? AND status IS NULL";
2412 To test if a column IS NOT NULL:
2416 status => { '!=', undef },
2419 =head2 Specific comparison operators
2421 If you want to specify a different type of operator for your comparison,
2422 you can use a hashref for a given column:
2426 status => { '!=', 'completed' }
2429 Which would generate:
2431 $stmt = "WHERE user = ? AND status != ?";
2432 @bind = ('nwiger', 'completed');
2434 To test against multiple values, just enclose the values in an arrayref:
2436 status => { '=', ['assigned', 'in-progress', 'pending'] };
2438 Which would give you:
2440 "WHERE status = ? OR status = ? OR status = ?"
2443 The hashref can also contain multiple pairs, in which case it is expanded
2444 into an C<AND> of its elements:
2448 status => { '!=', 'completed', -not_like => 'pending%' }
2451 # Or more dynamically, like from a form
2452 $where{user} = 'nwiger';
2453 $where{status}{'!='} = 'completed';
2454 $where{status}{'-not_like'} = 'pending%';
2456 # Both generate this
2457 $stmt = "WHERE user = ? AND status != ? AND status NOT LIKE ?";
2458 @bind = ('nwiger', 'completed', 'pending%');
2461 To get an OR instead, you can combine it with the arrayref idea:
2465 priority => [ { '=', 2 }, { '>', 5 } ]
2468 Which would generate:
2470 $stmt = "WHERE ( priority = ? OR priority > ? ) AND user = ?";
2471 @bind = ('2', '5', 'nwiger');
2473 If you want to include literal SQL (with or without bind values), just use a
2474 scalar reference or reference to an arrayref as the value:
2477 date_entered => { '>' => \["to_date(?, 'MM/DD/YYYY')", "11/26/2008"] },
2478 date_expires => { '<' => \"now()" }
2481 Which would generate:
2483 $stmt = "WHERE date_entered > to_date(?, 'MM/DD/YYYY') AND date_expires < now()";
2484 @bind = ('11/26/2008');
2487 =head2 Logic and nesting operators
2489 In the example above,
2490 there is a subtle trap if you want to say something like
2491 this (notice the C<AND>):
2493 WHERE priority != ? AND priority != ?
2495 Because, in Perl you I<can't> do this:
2497 priority => { '!=' => 2, '!=' => 1 }
2499 As the second C<!=> key will obliterate the first. The solution
2500 is to use the special C<-modifier> form inside an arrayref:
2502 priority => [ -and => {'!=', 2},
2506 Normally, these would be joined by C<OR>, but the modifier tells it
2507 to use C<AND> instead. (Hint: You can use this in conjunction with the
2508 C<logic> option to C<new()> in order to change the way your queries
2509 work by default.) B<Important:> Note that the C<-modifier> goes
2510 B<INSIDE> the arrayref, as an extra first element. This will
2511 B<NOT> do what you think it might:
2513 priority => -and => [{'!=', 2}, {'!=', 1}] # WRONG!
2515 Here is a quick list of equivalencies, since there is some overlap:
2518 status => {'!=', 'completed', 'not like', 'pending%' }
2519 status => [ -and => {'!=', 'completed'}, {'not like', 'pending%'}]
2522 status => {'=', ['assigned', 'in-progress']}
2523 status => [ -or => {'=', 'assigned'}, {'=', 'in-progress'}]
2524 status => [ {'=', 'assigned'}, {'=', 'in-progress'} ]
2528 =head2 Special operators: IN, BETWEEN, etc.
2530 You can also use the hashref format to compare a list of fields using the
2531 C<IN> comparison operator, by specifying the list as an arrayref:
2534 status => 'completed',
2535 reportid => { -in => [567, 2335, 2] }
2538 Which would generate:
2540 $stmt = "WHERE status = ? AND reportid IN (?,?,?)";
2541 @bind = ('completed', '567', '2335', '2');
2543 The reverse operator C<-not_in> generates SQL C<NOT IN> and is used in
2546 If the argument to C<-in> is an empty array, 'sqlfalse' is generated
2547 (by default: C<1=0>). Similarly, C<< -not_in => [] >> generates
2548 'sqltrue' (by default: C<1=1>).
2550 In addition to the array you can supply a chunk of literal sql or
2551 literal sql with bind:
2554 customer => { -in => \[
2555 'SELECT cust_id FROM cust WHERE balance > ?',
2558 status => { -in => \'SELECT status_codes FROM states' },
2564 customer IN ( SELECT cust_id FROM cust WHERE balance > ? )
2565 AND status IN ( SELECT status_codes FROM states )
2569 Finally, if the argument to C<-in> is not a reference, it will be
2570 treated as a single-element array.
2572 Another pair of operators is C<-between> and C<-not_between>,
2573 used with an arrayref of two values:
2577 completion_date => {
2578 -not_between => ['2002-10-01', '2003-02-06']
2584 WHERE user = ? AND completion_date NOT BETWEEN ( ? AND ? )
2586 Just like with C<-in> all plausible combinations of literal SQL
2590 start0 => { -between => [ 1, 2 ] },
2591 start1 => { -between => \["? AND ?", 1, 2] },
2592 start2 => { -between => \"lower(x) AND upper(y)" },
2593 start3 => { -between => [
2595 \["upper(?)", 'stuff' ],
2602 ( start0 BETWEEN ? AND ? )
2603 AND ( start1 BETWEEN ? AND ? )
2604 AND ( start2 BETWEEN lower(x) AND upper(y) )
2605 AND ( start3 BETWEEN lower(x) AND upper(?) )
2607 @bind = (1, 2, 1, 2, 'stuff');
2610 These are the two builtin "special operators"; but the
2611 list can be expanded: see section L</"SPECIAL OPERATORS"> below.
2613 =head2 Unary operators: bool
2615 If you wish to test against boolean columns or functions within your
2616 database you can use the C<-bool> and C<-not_bool> operators. For
2617 example to test the column C<is_user> being true and the column
2618 C<is_enabled> being false you would use:-
2622 -not_bool => 'is_enabled',
2627 WHERE is_user AND NOT is_enabled
2629 If a more complex combination is required, testing more conditions,
2630 then you should use the and/or operators:-
2635 -not_bool => { two=> { -rlike => 'bar' } },
2636 -not_bool => { three => [ { '=', 2 }, { '>', 5 } ] },
2647 (NOT ( three = ? OR three > ? ))
2650 =head2 Nested conditions, -and/-or prefixes
2652 So far, we've seen how multiple conditions are joined with a top-level
2653 C<AND>. We can change this by putting the different conditions we want in
2654 hashes and then putting those hashes in an array. For example:
2659 status => { -like => ['pending%', 'dispatched'] },
2663 status => 'unassigned',
2667 This data structure would create the following:
2669 $stmt = "WHERE ( user = ? AND ( status LIKE ? OR status LIKE ? ) )
2670 OR ( user = ? AND status = ? ) )";
2671 @bind = ('nwiger', 'pending', 'dispatched', 'robot', 'unassigned');
2674 Clauses in hashrefs or arrayrefs can be prefixed with an C<-and> or C<-or>
2675 to change the logic inside:
2681 -and => [ workhrs => {'>', 20}, geo => 'ASIA' ],
2682 -or => { workhrs => {'<', 50}, geo => 'EURO' },
2689 $stmt = "WHERE ( user = ?
2690 AND ( ( workhrs > ? AND geo = ? )
2691 OR ( workhrs < ? OR geo = ? ) ) )";
2692 @bind = ('nwiger', '20', 'ASIA', '50', 'EURO');
2694 =head3 Algebraic inconsistency, for historical reasons
2696 C<Important note>: when connecting several conditions, the C<-and->|C<-or>
2697 operator goes C<outside> of the nested structure; whereas when connecting
2698 several constraints on one column, the C<-and> operator goes
2699 C<inside> the arrayref. Here is an example combining both features:
2702 -and => [a => 1, b => 2],
2703 -or => [c => 3, d => 4],
2704 e => [-and => {-like => 'foo%'}, {-like => '%bar'} ]
2709 WHERE ( ( ( a = ? AND b = ? )
2710 OR ( c = ? OR d = ? )
2711 OR ( e LIKE ? AND e LIKE ? ) ) )
2713 This difference in syntax is unfortunate but must be preserved for
2714 historical reasons. So be careful: the two examples below would
2715 seem algebraically equivalent, but they are not
2718 { -like => 'foo%' },
2719 { -like => '%bar' },
2721 # yields: WHERE ( ( col LIKE ? AND col LIKE ? ) )
2724 { col => { -like => 'foo%' } },
2725 { col => { -like => '%bar' } },
2727 # yields: WHERE ( ( col LIKE ? OR col LIKE ? ) )
2730 =head2 Literal SQL and value type operators
2732 The basic premise of SQL::Abstract is that in WHERE specifications the "left
2733 side" is a column name and the "right side" is a value (normally rendered as
2734 a placeholder). This holds true for both hashrefs and arrayref pairs as you
2735 see in the L</WHERE CLAUSES> examples above. Sometimes it is necessary to
2736 alter this behavior. There are several ways of doing so.
2740 This is a virtual operator that signals the string to its right side is an
2741 identifier (a column name) and not a value. For example to compare two
2742 columns you would write:
2745 priority => { '<', 2 },
2746 requestor => { -ident => 'submitter' },
2751 $stmt = "WHERE priority < ? AND requestor = submitter";
2754 If you are maintaining legacy code you may see a different construct as
2755 described in L</Deprecated usage of Literal SQL>, please use C<-ident> in new
2760 This is a virtual operator that signals that the construct to its right side
2761 is a value to be passed to DBI. This is for example necessary when you want
2762 to write a where clause against an array (for RDBMS that support such
2763 datatypes). For example:
2766 array => { -value => [1, 2, 3] }
2771 $stmt = 'WHERE array = ?';
2772 @bind = ([1, 2, 3]);
2774 Note that if you were to simply say:
2780 the result would probably not be what you wanted:
2782 $stmt = 'WHERE array = ? OR array = ? OR array = ?';
2787 Finally, sometimes only literal SQL will do. To include a random snippet
2788 of SQL verbatim, you specify it as a scalar reference. Consider this only
2789 as a last resort. Usually there is a better way. For example:
2792 priority => { '<', 2 },
2793 requestor => { -in => \'(SELECT name FROM hitmen)' },
2798 $stmt = "WHERE priority < ? AND requestor IN (SELECT name FROM hitmen)"
2801 Note that in this example, you only get one bind parameter back, since
2802 the verbatim SQL is passed as part of the statement.
2806 Never use untrusted input as a literal SQL argument - this is a massive
2807 security risk (there is no way to check literal snippets for SQL
2808 injections and other nastyness). If you need to deal with untrusted input
2809 use literal SQL with placeholders as described next.
2811 =head3 Literal SQL with placeholders and bind values (subqueries)
2813 If the literal SQL to be inserted has placeholders and bind values,
2814 use a reference to an arrayref (yes this is a double reference --
2815 not so common, but perfectly legal Perl). For example, to find a date
2816 in Postgres you can use something like this:
2819 date_column => \[ "= date '2008-09-30' - ?::integer", 10 ]
2824 $stmt = "WHERE ( date_column = date '2008-09-30' - ?::integer )"
2827 Note that you must pass the bind values in the same format as they are returned
2828 by L<where|/where(\%where, $order)>. This means that if you set L</bindtype>
2829 to C<columns>, you must provide the bind values in the
2830 C<< [ column_meta => value ] >> format, where C<column_meta> is an opaque
2831 scalar value; most commonly the column name, but you can use any scalar value
2832 (including references and blessed references), L<SQL::Abstract> will simply
2833 pass it through intact. So if C<bindtype> is set to C<columns> the above
2834 example will look like:
2837 date_column => \[ "= date '2008-09-30' - ?::integer", [ {} => 10 ] ]
2840 Literal SQL is especially useful for nesting parenthesized clauses in the
2841 main SQL query. Here is a first example:
2843 my ($sub_stmt, @sub_bind) = ("SELECT c1 FROM t1 WHERE c2 < ? AND c3 LIKE ?",
2847 bar => \["IN ($sub_stmt)" => @sub_bind],
2852 $stmt = "WHERE (foo = ? AND bar IN (SELECT c1 FROM t1
2853 WHERE c2 < ? AND c3 LIKE ?))";
2854 @bind = (1234, 100, "foo%");
2856 Other subquery operators, like for example C<"E<gt> ALL"> or C<"NOT IN">,
2857 are expressed in the same way. Of course the C<$sub_stmt> and
2858 its associated bind values can be generated through a former call
2861 my ($sub_stmt, @sub_bind)
2862 = $sql->select("t1", "c1", {c2 => {"<" => 100},
2863 c3 => {-like => "foo%"}});
2866 bar => \["> ALL ($sub_stmt)" => @sub_bind],
2869 In the examples above, the subquery was used as an operator on a column;
2870 but the same principle also applies for a clause within the main C<%where>
2871 hash, like an EXISTS subquery:
2873 my ($sub_stmt, @sub_bind)
2874 = $sql->select("t1", "*", {c1 => 1, c2 => \"> t0.c0"});
2875 my %where = ( -and => [
2877 \["EXISTS ($sub_stmt)" => @sub_bind],
2882 $stmt = "WHERE (foo = ? AND EXISTS (SELECT * FROM t1
2883 WHERE c1 = ? AND c2 > t0.c0))";
2887 Observe that the condition on C<c2> in the subquery refers to
2888 column C<t0.c0> of the main query: this is I<not> a bind
2889 value, so we have to express it through a scalar ref.
2890 Writing C<< c2 => {">" => "t0.c0"} >> would have generated
2891 C<< c2 > ? >> with bind value C<"t0.c0"> ... not exactly
2892 what we wanted here.
2894 Finally, here is an example where a subquery is used
2895 for expressing unary negation:
2897 my ($sub_stmt, @sub_bind)
2898 = $sql->where({age => [{"<" => 10}, {">" => 20}]});
2899 $sub_stmt =~ s/^ where //i; # don't want "WHERE" in the subclause
2901 lname => {like => '%son%'},
2902 \["NOT ($sub_stmt)" => @sub_bind],
2907 $stmt = "lname LIKE ? AND NOT ( age < ? OR age > ? )"
2908 @bind = ('%son%', 10, 20)
2910 =head3 Deprecated usage of Literal SQL
2912 Below are some examples of archaic use of literal SQL. It is shown only as
2913 reference for those who deal with legacy code. Each example has a much
2914 better, cleaner and safer alternative that users should opt for in new code.
2920 my %where = ( requestor => \'IS NOT NULL' )
2922 $stmt = "WHERE requestor IS NOT NULL"
2924 This used to be the way of generating NULL comparisons, before the handling
2925 of C<undef> got formalized. For new code please use the superior syntax as
2926 described in L</Tests for NULL values>.
2930 my %where = ( requestor => \'= submitter' )
2932 $stmt = "WHERE requestor = submitter"
2934 This used to be the only way to compare columns. Use the superior L</-ident>
2935 method for all new code. For example an identifier declared in such a way
2936 will be properly quoted if L</quote_char> is properly set, while the legacy
2937 form will remain as supplied.
2941 my %where = ( is_ready => \"", completed => { '>', '2012-12-21' } )
2943 $stmt = "WHERE completed > ? AND is_ready"
2944 @bind = ('2012-12-21')
2946 Using an empty string literal used to be the only way to express a boolean.
2947 For all new code please use the much more readable
2948 L<-bool|/Unary operators: bool> operator.
2954 These pages could go on for a while, since the nesting of the data
2955 structures this module can handle are pretty much unlimited (the
2956 module implements the C<WHERE> expansion as a recursive function
2957 internally). Your best bet is to "play around" with the module a
2958 little to see how the data structures behave, and choose the best
2959 format for your data based on that.
2961 And of course, all the values above will probably be replaced with
2962 variables gotten from forms or the command line. After all, if you
2963 knew everything ahead of time, you wouldn't have to worry about
2964 dynamically-generating SQL and could just hardwire it into your
2967 =head1 ORDER BY CLAUSES
2969 Some functions take an order by clause. This can either be a scalar (just a
2970 column name), a hashref of C<< { -desc => 'col' } >> or C<< { -asc => 'col' }
2971 >>, a scalarref, an arrayref-ref, or an arrayref of any of the previous
2974 Given | Will Generate
2975 ---------------------------------------------------------------
2977 'colA' | ORDER BY colA
2979 [qw/colA colB/] | ORDER BY colA, colB
2981 {-asc => 'colA'} | ORDER BY colA ASC
2983 {-desc => 'colB'} | ORDER BY colB DESC
2985 ['colA', {-asc => 'colB'}] | ORDER BY colA, colB ASC
2987 { -asc => [qw/colA colB/] } | ORDER BY colA ASC, colB ASC
2989 \'colA DESC' | ORDER BY colA DESC
2991 \[ 'FUNC(colA, ?)', $x ] | ORDER BY FUNC(colA, ?)
2992 | /* ...with $x bound to ? */
2995 { -asc => 'colA' }, | colA ASC,
2996 { -desc => [qw/colB/] }, | colB DESC,
2997 { -asc => [qw/colC colD/] },| colC ASC, colD ASC,
2998 \'colE DESC', | colE DESC,
2999 \[ 'FUNC(colF, ?)', $x ], | FUNC(colF, ?)
3000 ] | /* ...with $x bound to ? */
3001 ===============================================================
3005 =head1 SPECIAL OPERATORS
3007 my $sqlmaker = SQL::Abstract->new(special_ops => [
3011 my ($self, $field, $op, $arg) = @_;
3017 handler => 'method_name',
3021 A "special operator" is a SQL syntactic clause that can be
3022 applied to a field, instead of a usual binary operator.
3025 WHERE field IN (?, ?, ?)
3026 WHERE field BETWEEN ? AND ?
3027 WHERE MATCH(field) AGAINST (?, ?)
3029 Special operators IN and BETWEEN are fairly standard and therefore
3030 are builtin within C<SQL::Abstract> (as the overridable methods
3031 C<_where_field_IN> and C<_where_field_BETWEEN>). For other operators,
3032 like the MATCH .. AGAINST example above which is specific to MySQL,
3033 you can write your own operator handlers - supply a C<special_ops>
3034 argument to the C<new> method. That argument takes an arrayref of
3035 operator definitions; each operator definition is a hashref with two
3042 the regular expression to match the operator
3046 Either a coderef or a plain scalar method name. In both cases
3047 the expected return is C<< ($sql, @bind) >>.
3049 When supplied with a method name, it is simply called on the
3050 L<SQL::Abstract> object as:
3052 $self->$method_name($field, $op, $arg)
3056 $field is the LHS of the operator
3057 $op is the part that matched the handler regex
3060 When supplied with a coderef, it is called as:
3062 $coderef->($self, $field, $op, $arg)
3067 For example, here is an implementation
3068 of the MATCH .. AGAINST syntax for MySQL
3070 my $sqlmaker = SQL::Abstract->new(special_ops => [
3072 # special op for MySql MATCH (field) AGAINST(word1, word2, ...)
3073 {regex => qr/^match$/i,
3075 my ($self, $field, $op, $arg) = @_;
3076 $arg = [$arg] if not ref $arg;
3077 my $label = $self->_quote($field);
3078 my ($placeholder) = $self->_convert('?');
3079 my $placeholders = join ", ", (($placeholder) x @$arg);
3080 my $sql = $self->_sqlcase('match') . " ($label) "
3081 . $self->_sqlcase('against') . " ($placeholders) ";
3082 my @bind = $self->_bindtype($field, @$arg);
3083 return ($sql, @bind);
3090 =head1 UNARY OPERATORS
3092 my $sqlmaker = SQL::Abstract->new(unary_ops => [
3096 my ($self, $op, $arg) = @_;
3102 handler => 'method_name',
3106 A "unary operator" is a SQL syntactic clause that can be
3107 applied to a field - the operator goes before the field
3109 You can write your own operator handlers - supply a C<unary_ops>
3110 argument to the C<new> method. That argument takes an arrayref of
3111 operator definitions; each operator definition is a hashref with two
3118 the regular expression to match the operator
3122 Either a coderef or a plain scalar method name. In both cases
3123 the expected return is C<< $sql >>.
3125 When supplied with a method name, it is simply called on the
3126 L<SQL::Abstract> object as:
3128 $self->$method_name($op, $arg)
3132 $op is the part that matched the handler regex
3133 $arg is the RHS or argument of the operator
3135 When supplied with a coderef, it is called as:
3137 $coderef->($self, $op, $arg)
3145 Thanks to some benchmarking by Mark Stosberg, it turns out that
3146 this module is many orders of magnitude faster than using C<DBIx::Abstract>.
3147 I must admit this wasn't an intentional design issue, but it's a
3148 byproduct of the fact that you get to control your C<DBI> handles
3151 To maximize performance, use a code snippet like the following:
3153 # prepare a statement handle using the first row
3154 # and then reuse it for the rest of the rows
3156 for my $href (@array_of_hashrefs) {
3157 $stmt ||= $sql->insert('table', $href);
3158 $sth ||= $dbh->prepare($stmt);
3159 $sth->execute($sql->values($href));
3162 The reason this works is because the keys in your C<$href> are sorted
3163 internally by B<SQL::Abstract>. Thus, as long as your data retains
3164 the same structure, you only have to generate the SQL the first time
3165 around. On subsequent queries, simply use the C<values> function provided
3166 by this module to return your values in the correct order.
3168 However this depends on the values having the same type - if, for
3169 example, the values of a where clause may either have values
3170 (resulting in sql of the form C<column = ?> with a single bind
3171 value), or alternatively the values might be C<undef> (resulting in
3172 sql of the form C<column IS NULL> with no bind value) then the
3173 caching technique suggested will not work.
3177 If you use my C<CGI::FormBuilder> module at all, you'll hopefully
3178 really like this part (I do, at least). Building up a complex query
3179 can be as simple as the following:
3186 use CGI::FormBuilder;
3189 my $form = CGI::FormBuilder->new(...);
3190 my $sql = SQL::Abstract->new;
3192 if ($form->submitted) {
3193 my $field = $form->field;
3194 my $id = delete $field->{id};
3195 my($stmt, @bind) = $sql->update('table', $field, {id => $id});
3198 Of course, you would still have to connect using C<DBI> to run the
3199 query, but the point is that if you make your form look like your
3200 table, the actual query script can be extremely simplistic.
3202 If you're B<REALLY> lazy (I am), check out C<HTML::QuickTable> for
3203 a fast interface to returning and formatting data. I frequently
3204 use these three modules together to write complex database query
3205 apps in under 50 lines.
3207 =head1 HOW TO CONTRIBUTE
3209 Contributions are always welcome, in all usable forms (we especially
3210 welcome documentation improvements). The delivery methods include git-
3211 or unified-diff formatted patches, GitHub pull requests, or plain bug
3212 reports either via RT or the Mailing list. Contributors are generally
3213 granted full access to the official repository after their first several
3214 patches pass successful review.
3216 This project is maintained in a git repository. The code and related tools are
3217 accessible at the following locations:
3221 =item * Official repo: L<git://git.shadowcat.co.uk/dbsrgits/SQL-Abstract.git>
3223 =item * Official gitweb: L<http://git.shadowcat.co.uk/gitweb/gitweb.cgi?p=dbsrgits/SQL-Abstract.git>
3225 =item * GitHub mirror: L<https://github.com/dbsrgits/sql-abstract>
3227 =item * Authorized committers: L<ssh://dbsrgits@git.shadowcat.co.uk/SQL-Abstract.git>
3233 Version 1.50 was a major internal refactoring of C<SQL::Abstract>.
3234 Great care has been taken to preserve the I<published> behavior
3235 documented in previous versions in the 1.* family; however,
3236 some features that were previously undocumented, or behaved
3237 differently from the documentation, had to be changed in order
3238 to clarify the semantics. Hence, client code that was relying
3239 on some dark areas of C<SQL::Abstract> v1.*
3240 B<might behave differently> in v1.50.
3242 The main changes are:
3248 support for literal SQL through the C<< \ [ $sql, @bind ] >> syntax.
3252 support for the { operator => \"..." } construct (to embed literal SQL)
3256 support for the { operator => \["...", @bind] } construct (to embed literal SQL with bind values)
3260 optional support for L<array datatypes|/"Inserting and Updating Arrays">
3264 defensive programming: check arguments
3268 fixed bug with global logic, which was previously implemented
3269 through global variables yielding side-effects. Prior versions would
3270 interpret C<< [ {cond1, cond2}, [cond3, cond4] ] >>
3271 as C<< "(cond1 AND cond2) OR (cond3 AND cond4)" >>.
3272 Now this is interpreted
3273 as C<< "(cond1 AND cond2) OR (cond3 OR cond4)" >>.
3278 fixed semantics of _bindtype on array args
3282 dropped the C<_anoncopy> of the %where tree. No longer necessary,
3283 we just avoid shifting arrays within that tree.
3287 dropped the C<_modlogic> function
3291 =head1 ACKNOWLEDGEMENTS
3293 There are a number of individuals that have really helped out with
3294 this module. Unfortunately, most of them submitted bugs via CPAN
3295 so I have no idea who they are! But the people I do know are:
3297 Ash Berlin (order_by hash term support)
3298 Matt Trout (DBIx::Class support)
3299 Mark Stosberg (benchmarking)
3300 Chas Owens (initial "IN" operator support)
3301 Philip Collins (per-field SQL functions)
3302 Eric Kolve (hashref "AND" support)
3303 Mike Fragassi (enhancements to "BETWEEN" and "LIKE")
3304 Dan Kubb (support for "quote_char" and "name_sep")
3305 Guillermo Roditi (patch to cleanup "IN" and "BETWEEN", fix and tests for _order_by)
3306 Laurent Dami (internal refactoring, extensible list of special operators, literal SQL)
3307 Norbert Buchmuller (support for literal SQL in hashpair, misc. fixes & tests)
3308 Peter Rabbitson (rewrite of SQLA::Test, misc. fixes & tests)
3309 Oliver Charles (support for "RETURNING" after "INSERT")
3315 L<DBIx::Class>, L<DBIx::Abstract>, L<CGI::FormBuilder>, L<HTML::QuickTable>.
3319 Copyright (c) 2001-2007 Nathan Wiger <nwiger@cpan.org>. All Rights Reserved.
3321 This module is actively maintained by Matt Trout <mst@shadowcatsystems.co.uk>
3323 For support, your best bet is to try the C<DBIx::Class> users mailing list.
3324 While not an official support venue, C<DBIx::Class> makes heavy use of
3325 C<SQL::Abstract>, and as such list members there are very familiar with
3326 how to create queries.
3330 This module is free software; you may copy this under the same
3331 terms as perl itself (either the GNU General Public License or
3332 the Artistic License)