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 = '1.86';
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) : $sql;
540 my ($self, $where, $logic) = @_;
542 # dispatch on appropriate method according to refkind of $where
543 my $method = $self->_METHOD_FOR_refkind("_where", $where);
545 my ($sql, @bind) = $self->$method($where, $logic);
547 # DBIx::Class used to call _recurse_where in scalar context
548 # something else might too...
550 return ($sql, @bind);
553 belch "Calling _recurse_where in scalar context is deprecated and will go away before 2.0";
560 #======================================================================
561 # WHERE: top-level ARRAYREF
562 #======================================================================
565 sub _where_ARRAYREF {
566 my ($self, $where, $logic) = @_;
568 $logic = uc($logic || $self->{logic});
569 $logic eq 'AND' or $logic eq 'OR' or puke "unknown logic: $logic";
571 my @clauses = @$where;
573 my (@sql_clauses, @all_bind);
574 # need to use while() so can shift() for pairs
576 my $el = shift @clauses;
578 $el = undef if (defined $el and ! length $el);
580 # switch according to kind of $el and get corresponding ($sql, @bind)
581 my ($sql, @bind) = $self->_SWITCH_refkind($el, {
583 # skip empty elements, otherwise get invalid trailing AND stuff
584 ARRAYREF => sub {$self->_recurse_where($el) if @$el},
588 $self->_assert_bindval_matches_bindtype(@b);
592 HASHREF => sub {$self->_recurse_where($el, 'and') if %$el},
594 SCALARREF => sub { ($$el); },
597 # top-level arrayref with scalars, recurse in pairs
598 $self->_recurse_where({$el => shift(@clauses)})
601 UNDEF => sub {puke "Supplying an empty left hand side argument is not supported in array-pairs" },
605 push @sql_clauses, $sql;
606 push @all_bind, @bind;
610 return $self->_join_sql_clauses($logic, \@sql_clauses, \@all_bind);
613 #======================================================================
614 # WHERE: top-level ARRAYREFREF
615 #======================================================================
617 sub _where_ARRAYREFREF {
618 my ($self, $where) = @_;
619 my ($sql, @bind) = @$$where;
620 $self->_assert_bindval_matches_bindtype(@bind);
621 return ($sql, @bind);
624 #======================================================================
625 # WHERE: top-level HASHREF
626 #======================================================================
629 my ($self, $where) = @_;
630 my (@sql_clauses, @all_bind);
632 for my $k (sort keys %$where) {
633 my $v = $where->{$k};
635 # ($k => $v) is either a special unary op or a regular hashpair
636 my ($sql, @bind) = do {
638 # put the operator in canonical form
640 $op = substr $op, 1; # remove initial dash
641 $op =~ s/^\s+|\s+$//g;# remove leading/trailing space
642 $op =~ s/\s+/ /g; # compress whitespace
644 # so that -not_foo works correctly
645 $op =~ s/^not_/NOT /i;
647 $self->_debug("Unary OP(-$op) within hashref, recursing...");
648 my ($s, @b) = $self->_where_unary_op($op, $v);
650 # top level vs nested
651 # we assume that handled unary ops will take care of their ()s
653 List::Util::first {$op =~ $_->{regex}} @{$self->{unary_ops}}
655 ( defined $self->{_nested_func_lhs} and $self->{_nested_func_lhs} eq $k )
661 if (is_literal_value ($v) ) {
662 belch 'Hash-pairs consisting of an empty string with a literal are deprecated, and will be removed in 2.0: use -and => [ $literal ] instead';
665 puke "Supplying an empty left hand side argument is not supported in hash-pairs";
669 my $method = $self->_METHOD_FOR_refkind("_where_hashpair", $v);
670 $self->$method($k, $v);
674 push @sql_clauses, $sql;
675 push @all_bind, @bind;
678 return $self->_join_sql_clauses('and', \@sql_clauses, \@all_bind);
681 sub _where_unary_op {
682 my ($self, $op, $rhs) = @_;
684 # top level special ops are illegal in general
685 # this includes the -ident/-value ops (dual purpose unary and special)
686 puke "Illegal use of top-level '-$op'"
687 if ! defined $self->{_nested_func_lhs} and List::Util::first { $op =~ $_->{regex} } @{$self->{special_ops}};
689 if (my $op_entry = List::Util::first { $op =~ $_->{regex} } @{$self->{unary_ops}}) {
690 my $handler = $op_entry->{handler};
692 if (not ref $handler) {
693 if ($op =~ s/ [_\s]? \d+ $//x ) {
694 belch 'Use of [and|or|nest]_N modifiers is deprecated and will be removed in SQLA v2.0. '
695 . "You probably wanted ...-and => [ -$op => COND1, -$op => COND2 ... ]";
697 return $self->$handler($op, $rhs);
699 elsif (ref $handler eq 'CODE') {
700 return $handler->($self, $op, $rhs);
703 puke "Illegal handler for operator $op - expecting a method name or a coderef";
707 $self->_debug("Generic unary OP: $op - recursing as function");
709 $self->_assert_pass_injection_guard($op);
711 my ($sql, @bind) = $self->_SWITCH_refkind($rhs, {
713 puke "Illegal use of top-level '-$op'"
714 unless defined $self->{_nested_func_lhs};
717 $self->_convert('?'),
718 $self->_bindtype($self->{_nested_func_lhs}, $rhs)
722 $self->_recurse_where($rhs)
726 $sql = sprintf('%s %s',
727 $self->_sqlcase($op),
731 return ($sql, @bind);
734 sub _where_op_ANDOR {
735 my ($self, $op, $v) = @_;
737 $self->_SWITCH_refkind($v, {
739 return $self->_where_ARRAYREF($v, $op);
743 return ($op =~ /^or/i)
744 ? $self->_where_ARRAYREF([ map { $_ => $v->{$_} } (sort keys %$v) ], $op)
745 : $self->_where_HASHREF($v);
749 puke "-$op => \\\$scalar makes little sense, use " .
751 ? '[ \$scalar, \%rest_of_conditions ] instead'
752 : '-and => [ \$scalar, \%rest_of_conditions ] instead'
757 puke "-$op => \\[...] makes little sense, use " .
759 ? '[ \[...], \%rest_of_conditions ] instead'
760 : '-and => [ \[...], \%rest_of_conditions ] instead'
764 SCALAR => sub { # permissively interpreted as SQL
765 puke "-$op => \$value makes little sense, use -bool => \$value instead";
769 puke "-$op => undef not supported";
775 my ($self, $op, $v) = @_;
777 $self->_SWITCH_refkind($v, {
779 SCALAR => sub { # permissively interpreted as SQL
780 belch "literal SQL should be -nest => \\'scalar' "
781 . "instead of -nest => 'scalar' ";
786 puke "-$op => undef not supported";
790 $self->_recurse_where($v);
798 my ($self, $op, $v) = @_;
800 my ($s, @b) = $self->_SWITCH_refkind($v, {
801 SCALAR => sub { # interpreted as SQL column
802 $self->_convert($self->_quote($v));
806 puke "-$op => undef not supported";
810 $self->_recurse_where($v);
814 $s = "(NOT $s)" if $op =~ /^not/i;
819 sub _where_op_IDENT {
821 my ($op, $rhs) = splice @_, -2;
822 if (! defined $rhs or length ref $rhs) {
823 puke "-$op requires a single plain scalar argument (a quotable identifier)";
826 # in case we are called as a top level special op (no '=')
829 $_ = $self->_convert($self->_quote($_)) for ($lhs, $rhs);
837 sub _where_op_VALUE {
839 my ($op, $rhs) = splice @_, -2;
841 # in case we are called as a top level special op (no '=')
845 if (! defined $rhs) {
847 ? $self->_where_hashpair_HASHREF($lhs, { -is => undef })
854 (defined $lhs ? $lhs : $self->{_nested_func_lhs}),
861 $self->_convert($self->_quote($lhs)) . ' = ' . $self->_convert('?'),
865 $self->_convert('?'),
871 sub _where_hashpair_ARRAYREF {
872 my ($self, $k, $v) = @_;
875 my @v = @$v; # need copy because of shift below
876 $self->_debug("ARRAY($k) means distribute over elements");
878 # put apart first element if it is an operator (-and, -or)
880 (defined $v[0] && $v[0] =~ /^ - (?: AND|OR ) $/ix)
884 my @distributed = map { {$k => $_} } @v;
887 $self->_debug("OP($op) reinjected into the distributed array");
888 unshift @distributed, $op;
891 my $logic = $op ? substr($op, 1) : '';
893 return $self->_recurse_where(\@distributed, $logic);
896 $self->_debug("empty ARRAY($k) means 0=1");
897 return ($self->{sqlfalse});
901 sub _where_hashpair_HASHREF {
902 my ($self, $k, $v, $logic) = @_;
905 local $self->{_nested_func_lhs} = defined $self->{_nested_func_lhs}
906 ? $self->{_nested_func_lhs}
910 my ($all_sql, @all_bind);
912 for my $orig_op (sort keys %$v) {
913 my $val = $v->{$orig_op};
915 # put the operator in canonical form
918 # FIXME - we need to phase out dash-less ops
919 $op =~ s/^-//; # remove possible initial dash
920 $op =~ s/^\s+|\s+$//g;# remove leading/trailing space
921 $op =~ s/\s+/ /g; # compress whitespace
923 $self->_assert_pass_injection_guard($op);
926 $op =~ s/^is_not/IS NOT/i;
928 # so that -not_foo works correctly
929 $op =~ s/^not_/NOT /i;
931 # another retarded special case: foo => { $op => { -value => undef } }
932 if (ref $val eq 'HASH' and keys %$val == 1 and exists $val->{-value} and ! defined $val->{-value} ) {
938 # CASE: col-value logic modifiers
939 if ($orig_op =~ /^ \- (and|or) $/xi) {
940 ($sql, @bind) = $self->_where_hashpair_HASHREF($k, $val, $1);
942 # CASE: special operators like -in or -between
943 elsif (my $special_op = List::Util::first { $op =~ $_->{regex} } @{$self->{special_ops}}) {
944 my $handler = $special_op->{handler};
946 puke "No handler supplied for special operator $orig_op";
948 elsif (not ref $handler) {
949 ($sql, @bind) = $self->$handler($k, $op, $val);
951 elsif (ref $handler eq 'CODE') {
952 ($sql, @bind) = $handler->($self, $k, $op, $val);
955 puke "Illegal handler for special operator $orig_op - expecting a method name or a coderef";
959 $self->_SWITCH_refkind($val, {
961 ARRAYREF => sub { # CASE: col => {op => \@vals}
962 ($sql, @bind) = $self->_where_field_op_ARRAYREF($k, $op, $val);
965 ARRAYREFREF => sub { # CASE: col => {op => \[$sql, @bind]} (literal SQL with bind)
966 my ($sub_sql, @sub_bind) = @$$val;
967 $self->_assert_bindval_matches_bindtype(@sub_bind);
968 $sql = join ' ', $self->_convert($self->_quote($k)),
969 $self->_sqlcase($op),
974 UNDEF => sub { # CASE: col => {op => undef} : sql "IS (NOT)? NULL"
976 $op =~ /^not$/i ? 'is not' # legacy
977 : $op =~ $self->{equality_op} ? 'is'
978 : $op =~ $self->{like_op} ? belch("Supplying an undefined argument to '@{[ uc $op]}' is deprecated") && 'is'
979 : $op =~ $self->{inequality_op} ? 'is not'
980 : $op =~ $self->{not_like_op} ? belch("Supplying an undefined argument to '@{[ uc $op]}' is deprecated") && 'is not'
981 : puke "unexpected operator '$orig_op' with undef operand";
983 $sql = $self->_quote($k) . $self->_sqlcase(" $is null");
986 FALLBACK => sub { # CASE: col => {op/func => $stuff}
987 ($sql, @bind) = $self->_where_unary_op($op, $val);
990 $self->_convert($self->_quote($k)),
991 $self->{_nested_func_lhs} eq $k ? $sql : "($sql)", # top level vs nested
997 ($all_sql) = (defined $all_sql and $all_sql) ? $self->_join_sql_clauses($logic, [$all_sql, $sql], []) : $sql;
998 push @all_bind, @bind;
1000 return ($all_sql, @all_bind);
1003 sub _where_field_IS {
1004 my ($self, $k, $op, $v) = @_;
1006 my ($s) = $self->_SWITCH_refkind($v, {
1009 $self->_convert($self->_quote($k)),
1010 map { $self->_sqlcase($_)} ($op, 'null')
1013 puke "$op can only take undef as argument";
1020 sub _where_field_op_ARRAYREF {
1021 my ($self, $k, $op, $vals) = @_;
1023 my @vals = @$vals; #always work on a copy
1026 $self->_debug(sprintf '%s means multiple elements: [ %s ]',
1028 join(', ', map { defined $_ ? "'$_'" : 'NULL' } @vals ),
1031 # see if the first element is an -and/-or op
1033 if (defined $vals[0] && $vals[0] =~ /^ - (AND|OR) $/ix) {
1038 # a long standing API wart - an attempt to change this behavior during
1039 # the 1.50 series failed *spectacularly*. Warn instead and leave the
1044 (!$logic or $logic eq 'OR')
1046 ($op =~ $self->{inequality_op} or $op =~ $self->{not_like_op})
1049 belch "A multi-element arrayref as an argument to the inequality op '$o' "
1050 . 'is technically equivalent to an always-true 1=1 (you probably wanted '
1051 . "to say ...{ \$inequality_op => [ -and => \@values ] }... instead)"
1055 # distribute $op over each remaining member of @vals, append logic if exists
1056 return $self->_recurse_where([map { {$k => {$op, $_}} } @vals], $logic);
1060 # try to DWIM on equality operators
1062 $op =~ $self->{equality_op} ? $self->{sqlfalse}
1063 : $op =~ $self->{like_op} ? belch("Supplying an empty arrayref to '@{[ uc $op]}' is deprecated") && $self->{sqlfalse}
1064 : $op =~ $self->{inequality_op} ? $self->{sqltrue}
1065 : $op =~ $self->{not_like_op} ? belch("Supplying an empty arrayref to '@{[ uc $op]}' is deprecated") && $self->{sqltrue}
1066 : puke "operator '$op' applied on an empty array (field '$k')";
1071 sub _where_hashpair_SCALARREF {
1072 my ($self, $k, $v) = @_;
1073 $self->_debug("SCALAR($k) means literal SQL: $$v");
1074 my $sql = $self->_quote($k) . " " . $$v;
1078 # literal SQL with bind
1079 sub _where_hashpair_ARRAYREFREF {
1080 my ($self, $k, $v) = @_;
1081 $self->_debug("REF($k) means literal SQL: @${$v}");
1082 my ($sql, @bind) = @$$v;
1083 $self->_assert_bindval_matches_bindtype(@bind);
1084 $sql = $self->_quote($k) . " " . $sql;
1085 return ($sql, @bind );
1088 # literal SQL without bind
1089 sub _where_hashpair_SCALAR {
1090 my ($self, $k, $v) = @_;
1091 $self->_debug("NOREF($k) means simple key=val: $k $self->{cmp} $v");
1092 return ($self->_where_hashpair_HASHREF($k, { $self->{cmp} => $v }));
1096 sub _where_hashpair_UNDEF {
1097 my ($self, $k, $v) = @_;
1098 $self->_debug("UNDEF($k) means IS NULL");
1099 return $self->_where_hashpair_HASHREF($k, { -is => undef });
1102 #======================================================================
1103 # WHERE: TOP-LEVEL OTHERS (SCALARREF, SCALAR, UNDEF)
1104 #======================================================================
1107 sub _where_SCALARREF {
1108 my ($self, $where) = @_;
1111 $self->_debug("SCALAR(*top) means literal SQL: $$where");
1117 my ($self, $where) = @_;
1120 $self->_debug("NOREF(*top) means literal SQL: $where");
1131 #======================================================================
1132 # WHERE: BUILTIN SPECIAL OPERATORS (-in, -between)
1133 #======================================================================
1136 sub _where_field_BETWEEN {
1137 my ($self, $k, $op, $vals) = @_;
1139 my ($label, $and, $placeholder);
1140 $label = $self->_convert($self->_quote($k));
1141 $and = ' ' . $self->_sqlcase('and') . ' ';
1142 $placeholder = $self->_convert('?');
1143 $op = $self->_sqlcase($op);
1145 my $invalid_args = "Operator '$op' requires either an arrayref with two defined values or expressions, or a single literal scalarref/arrayref-ref";
1147 my ($clause, @bind) = $self->_SWITCH_refkind($vals, {
1148 ARRAYREFREF => sub {
1149 my ($s, @b) = @$$vals;
1150 $self->_assert_bindval_matches_bindtype(@b);
1157 puke $invalid_args if @$vals != 2;
1159 my (@all_sql, @all_bind);
1160 foreach my $val (@$vals) {
1161 my ($sql, @bind) = $self->_SWITCH_refkind($val, {
1163 return ($placeholder, $self->_bindtype($k, $val) );
1168 ARRAYREFREF => sub {
1169 my ($sql, @bind) = @$$val;
1170 $self->_assert_bindval_matches_bindtype(@bind);
1171 return ($sql, @bind);
1174 my ($func, $arg, @rest) = %$val;
1175 puke "Only simple { -func => arg } functions accepted as sub-arguments to BETWEEN"
1176 if (@rest or $func !~ /^ \- (.+)/x);
1177 $self->_where_unary_op($1 => $arg);
1183 push @all_sql, $sql;
1184 push @all_bind, @bind;
1188 (join $and, @all_sql),
1197 my $sql = "( $label $op $clause )";
1198 return ($sql, @bind)
1202 sub _where_field_IN {
1203 my ($self, $k, $op, $vals) = @_;
1205 # backwards compatibility: if scalar, force into an arrayref
1206 $vals = [$vals] if defined $vals && ! ref $vals;
1208 my ($label) = $self->_convert($self->_quote($k));
1209 my ($placeholder) = $self->_convert('?');
1210 $op = $self->_sqlcase($op);
1212 my ($sql, @bind) = $self->_SWITCH_refkind($vals, {
1213 ARRAYREF => sub { # list of choices
1214 if (@$vals) { # nonempty list
1215 my (@all_sql, @all_bind);
1217 for my $val (@$vals) {
1218 my ($sql, @bind) = $self->_SWITCH_refkind($val, {
1220 return ($placeholder, $val);
1225 ARRAYREFREF => sub {
1226 my ($sql, @bind) = @$$val;
1227 $self->_assert_bindval_matches_bindtype(@bind);
1228 return ($sql, @bind);
1231 my ($func, $arg, @rest) = %$val;
1232 puke "Only simple { -func => arg } functions accepted as sub-arguments to IN"
1233 if (@rest or $func !~ /^ \- (.+)/x);
1234 $self->_where_unary_op($1 => $arg);
1238 'SQL::Abstract before v1.75 used to generate incorrect SQL when the '
1239 . "-$op operator was given an undef-containing list: !!!AUDIT YOUR CODE "
1240 . 'AND DATA!!! (the upcoming Data::Query-based version of SQL::Abstract '
1241 . 'will emit the logically correct SQL instead of raising this exception)'
1245 push @all_sql, $sql;
1246 push @all_bind, @bind;
1250 sprintf('%s %s ( %s )',
1253 join(', ', @all_sql)
1255 $self->_bindtype($k, @all_bind),
1258 else { # empty list: some databases won't understand "IN ()", so DWIM
1259 my $sql = ($op =~ /\bnot\b/i) ? $self->{sqltrue} : $self->{sqlfalse};
1264 SCALARREF => sub { # literal SQL
1265 my $sql = $self->_open_outer_paren($$vals);
1266 return ("$label $op ( $sql )");
1268 ARRAYREFREF => sub { # literal SQL with bind
1269 my ($sql, @bind) = @$$vals;
1270 $self->_assert_bindval_matches_bindtype(@bind);
1271 $sql = $self->_open_outer_paren($sql);
1272 return ("$label $op ( $sql )", @bind);
1276 puke "Argument passed to the '$op' operator can not be undefined";
1280 puke "special op $op requires an arrayref (or scalarref/arrayref-ref)";
1284 return ($sql, @bind);
1287 # Some databases (SQLite) treat col IN (1, 2) different from
1288 # col IN ( (1, 2) ). Use this to strip all outer parens while
1289 # adding them back in the corresponding method
1290 sub _open_outer_paren {
1291 my ($self, $sql) = @_;
1293 while (my ($inner) = $sql =~ /^ \s* \( (.*) \) \s* $/xs) {
1295 # there are closing parens inside, need the heavy duty machinery
1296 # to reevaluate the extraction starting from $sql (full reevaluation)
1297 if ($inner =~ /\)/) {
1298 require Text::Balanced;
1300 my (undef, $remainder) = do {
1301 # idiotic design - writes to $@ but *DOES NOT* throw exceptions
1303 Text::Balanced::extract_bracketed($sql, '()', qr/\s*/);
1306 # the entire expression needs to be a balanced bracketed thing
1307 # (after an extract no remainder sans trailing space)
1308 last if defined $remainder and $remainder =~ /\S/;
1318 #======================================================================
1320 #======================================================================
1323 my ($self, $arg) = @_;
1326 for my $c ($self->_order_by_chunks($arg) ) {
1327 $self->_SWITCH_refkind($c, {
1328 SCALAR => sub { push @sql, $c },
1329 ARRAYREF => sub { push @sql, shift @$c; push @bind, @$c },
1335 $self->_sqlcase(' order by'),
1341 return wantarray ? ($sql, @bind) : $sql;
1344 sub _order_by_chunks {
1345 my ($self, $arg) = @_;
1347 return $self->_SWITCH_refkind($arg, {
1350 map { $self->_order_by_chunks($_ ) } @$arg;
1353 ARRAYREFREF => sub {
1354 my ($s, @b) = @$$arg;
1355 $self->_assert_bindval_matches_bindtype(@b);
1359 SCALAR => sub {$self->_quote($arg)},
1361 UNDEF => sub {return () },
1363 SCALARREF => sub {$$arg}, # literal SQL, no quoting
1366 # get first pair in hash
1367 my ($key, $val, @rest) = %$arg;
1369 return () unless $key;
1371 if (@rest or not $key =~ /^-(desc|asc)/i) {
1372 puke "hash passed to _order_by must have exactly one key (-desc or -asc)";
1378 for my $c ($self->_order_by_chunks($val)) {
1381 $self->_SWITCH_refkind($c, {
1386 ($sql, @bind) = @$c;
1390 $sql = $sql . ' ' . $self->_sqlcase($direction);
1392 push @ret, [ $sql, @bind];
1401 #======================================================================
1402 # DATASOURCE (FOR NOW, JUST PLAIN TABLE OR LIST OF TABLES)
1403 #======================================================================
1408 $self->_SWITCH_refkind($from, {
1409 ARRAYREF => sub {join ', ', map { $self->_quote($_) } @$from;},
1410 SCALAR => sub {$self->_quote($from)},
1411 SCALARREF => sub {$$from},
1416 #======================================================================
1418 #======================================================================
1420 # highly optimized, as it's called way too often
1422 # my ($self, $label) = @_;
1424 return '' unless defined $_[1];
1425 return ${$_[1]} if ref($_[1]) eq 'SCALAR';
1427 $_[0]->{quote_char} or
1428 ($_[0]->_assert_pass_injection_guard($_[1]), return $_[1]);
1430 my $qref = ref $_[0]->{quote_char};
1432 !$qref ? ($_[0]->{quote_char}, $_[0]->{quote_char})
1433 : ($qref eq 'ARRAY') ? @{$_[0]->{quote_char}}
1434 : puke "Unsupported quote_char format: $_[0]->{quote_char}";
1436 my $esc = $_[0]->{escape_char} || $r;
1438 # parts containing * are naturally unquoted
1439 return join($_[0]->{name_sep}||'', map
1440 +( $_ eq '*' ? $_ : do { (my $n = $_) =~ s/(\Q$esc\E|\Q$r\E)/$esc$1/g; $l . $n . $r } ),
1441 ( $_[0]->{name_sep} ? split (/\Q$_[0]->{name_sep}\E/, $_[1] ) : $_[1] )
1446 # Conversion, if applicable
1448 #my ($self, $arg) = @_;
1449 if ($_[0]->{convert}) {
1450 return $_[0]->_sqlcase($_[0]->{convert}) .'(' . $_[1] . ')';
1457 #my ($self, $col, @vals) = @_;
1458 # called often - tighten code
1459 return $_[0]->{bindtype} eq 'columns'
1460 ? map {[$_[1], $_]} @_[2 .. $#_]
1465 # Dies if any element of @bind is not in [colname => value] format
1466 # if bindtype is 'columns'.
1467 sub _assert_bindval_matches_bindtype {
1468 # my ($self, @bind) = @_;
1470 if ($self->{bindtype} eq 'columns') {
1472 if (!defined $_ || ref($_) ne 'ARRAY' || @$_ != 2) {
1473 puke "bindtype 'columns' selected, you need to pass: [column_name => bind_value]"
1479 sub _join_sql_clauses {
1480 my ($self, $logic, $clauses_aref, $bind_aref) = @_;
1482 if (@$clauses_aref > 1) {
1483 my $join = " " . $self->_sqlcase($logic) . " ";
1484 my $sql = '( ' . join($join, @$clauses_aref) . ' )';
1485 return ($sql, @$bind_aref);
1487 elsif (@$clauses_aref) {
1488 return ($clauses_aref->[0], @$bind_aref); # no parentheses
1491 return (); # if no SQL, ignore @$bind_aref
1496 # Fix SQL case, if so requested
1498 # LDNOTE: if $self->{case} is true, then it contains 'lower', so we
1499 # don't touch the argument ... crooked logic, but let's not change it!
1500 return $_[0]->{case} ? $_[1] : uc($_[1]);
1504 #======================================================================
1505 # DISPATCHING FROM REFKIND
1506 #======================================================================
1509 my ($self, $data) = @_;
1511 return 'UNDEF' unless defined $data;
1513 # blessed objects are treated like scalars
1514 my $ref = (Scalar::Util::blessed $data) ? '' : ref $data;
1516 return 'SCALAR' unless $ref;
1519 while ($ref eq 'REF') {
1521 $ref = (Scalar::Util::blessed $data) ? '' : ref $data;
1525 return ($ref||'SCALAR') . ('REF' x $n_steps);
1529 my ($self, $data) = @_;
1530 my @try = ($self->_refkind($data));
1531 push @try, 'SCALAR_or_UNDEF' if $try[0] eq 'SCALAR' || $try[0] eq 'UNDEF';
1532 push @try, 'FALLBACK';
1536 sub _METHOD_FOR_refkind {
1537 my ($self, $meth_prefix, $data) = @_;
1540 for (@{$self->_try_refkind($data)}) {
1541 $method = $self->can($meth_prefix."_".$_)
1545 return $method || puke "cannot dispatch on '$meth_prefix' for ".$self->_refkind($data);
1549 sub _SWITCH_refkind {
1550 my ($self, $data, $dispatch_table) = @_;
1553 for (@{$self->_try_refkind($data)}) {
1554 $coderef = $dispatch_table->{$_}
1558 puke "no dispatch entry for ".$self->_refkind($data)
1567 #======================================================================
1568 # VALUES, GENERATE, AUTOLOAD
1569 #======================================================================
1571 # LDNOTE: original code from nwiger, didn't touch code in that section
1572 # I feel the AUTOLOAD stuff should not be the default, it should
1573 # only be activated on explicit demand by user.
1577 my $data = shift || return;
1578 puke "Argument to ", __PACKAGE__, "->values must be a \\%hash"
1579 unless ref $data eq 'HASH';
1582 foreach my $k (sort keys %$data) {
1583 my $v = $data->{$k};
1584 $self->_SWITCH_refkind($v, {
1586 if ($self->{array_datatypes}) { # array datatype
1587 push @all_bind, $self->_bindtype($k, $v);
1589 else { # literal SQL with bind
1590 my ($sql, @bind) = @$v;
1591 $self->_assert_bindval_matches_bindtype(@bind);
1592 push @all_bind, @bind;
1595 ARRAYREFREF => sub { # literal SQL with bind
1596 my ($sql, @bind) = @${$v};
1597 $self->_assert_bindval_matches_bindtype(@bind);
1598 push @all_bind, @bind;
1600 SCALARREF => sub { # literal SQL without bind
1602 SCALAR_or_UNDEF => sub {
1603 push @all_bind, $self->_bindtype($k, $v);
1614 my(@sql, @sqlq, @sqlv);
1618 if ($ref eq 'HASH') {
1619 for my $k (sort keys %$_) {
1622 my $label = $self->_quote($k);
1623 if ($r eq 'ARRAY') {
1624 # literal SQL with bind
1625 my ($sql, @bind) = @$v;
1626 $self->_assert_bindval_matches_bindtype(@bind);
1627 push @sqlq, "$label = $sql";
1629 } elsif ($r eq 'SCALAR') {
1630 # literal SQL without bind
1631 push @sqlq, "$label = $$v";
1633 push @sqlq, "$label = ?";
1634 push @sqlv, $self->_bindtype($k, $v);
1637 push @sql, $self->_sqlcase('set'), join ', ', @sqlq;
1638 } elsif ($ref eq 'ARRAY') {
1639 # unlike insert(), assume these are ONLY the column names, i.e. for SQL
1642 if ($r eq 'ARRAY') { # literal SQL with bind
1643 my ($sql, @bind) = @$v;
1644 $self->_assert_bindval_matches_bindtype(@bind);
1647 } elsif ($r eq 'SCALAR') { # literal SQL without bind
1648 # embedded literal SQL
1655 push @sql, '(' . join(', ', @sqlq) . ')';
1656 } elsif ($ref eq 'SCALAR') {
1660 # strings get case twiddled
1661 push @sql, $self->_sqlcase($_);
1665 my $sql = join ' ', @sql;
1667 # this is pretty tricky
1668 # if ask for an array, return ($stmt, @bind)
1669 # otherwise, s/?/shift @sqlv/ to put it inline
1671 return ($sql, @sqlv);
1673 1 while $sql =~ s/\?/my $d = shift(@sqlv);
1674 ref $d ? $d->[1] : $d/e;
1683 # This allows us to check for a local, then _form, attr
1685 my($name) = $AUTOLOAD =~ /.*::(.+)/;
1686 return $self->generate($name, @_);
1697 SQL::Abstract - Generate SQL from Perl data structures
1703 my $sql = SQL::Abstract->new;
1705 my($stmt, @bind) = $sql->select($source, \@fields, \%where, $order);
1707 my($stmt, @bind) = $sql->insert($table, \%fieldvals || \@values);
1709 my($stmt, @bind) = $sql->update($table, \%fieldvals, \%where);
1711 my($stmt, @bind) = $sql->delete($table, \%where);
1713 # Then, use these in your DBI statements
1714 my $sth = $dbh->prepare($stmt);
1715 $sth->execute(@bind);
1717 # Just generate the WHERE clause
1718 my($stmt, @bind) = $sql->where(\%where, $order);
1720 # Return values in the same order, for hashed queries
1721 # See PERFORMANCE section for more details
1722 my @bind = $sql->values(\%fieldvals);
1726 This module was inspired by the excellent L<DBIx::Abstract>.
1727 However, in using that module I found that what I really wanted
1728 to do was generate SQL, but still retain complete control over my
1729 statement handles and use the DBI interface. So, I set out to
1730 create an abstract SQL generation module.
1732 While based on the concepts used by L<DBIx::Abstract>, there are
1733 several important differences, especially when it comes to WHERE
1734 clauses. I have modified the concepts used to make the SQL easier
1735 to generate from Perl data structures and, IMO, more intuitive.
1736 The underlying idea is for this module to do what you mean, based
1737 on the data structures you provide it. The big advantage is that
1738 you don't have to modify your code every time your data changes,
1739 as this module figures it out.
1741 To begin with, an SQL INSERT is as easy as just specifying a hash
1742 of C<key=value> pairs:
1745 name => 'Jimbo Bobson',
1746 phone => '123-456-7890',
1747 address => '42 Sister Lane',
1748 city => 'St. Louis',
1749 state => 'Louisiana',
1752 The SQL can then be generated with this:
1754 my($stmt, @bind) = $sql->insert('people', \%data);
1756 Which would give you something like this:
1758 $stmt = "INSERT INTO people
1759 (address, city, name, phone, state)
1760 VALUES (?, ?, ?, ?, ?)";
1761 @bind = ('42 Sister Lane', 'St. Louis', 'Jimbo Bobson',
1762 '123-456-7890', 'Louisiana');
1764 These are then used directly in your DBI code:
1766 my $sth = $dbh->prepare($stmt);
1767 $sth->execute(@bind);
1769 =head2 Inserting and Updating Arrays
1771 If your database has array types (like for example Postgres),
1772 activate the special option C<< array_datatypes => 1 >>
1773 when creating the C<SQL::Abstract> object.
1774 Then you may use an arrayref to insert and update database array types:
1776 my $sql = SQL::Abstract->new(array_datatypes => 1);
1778 planets => [qw/Mercury Venus Earth Mars/]
1781 my($stmt, @bind) = $sql->insert('solar_system', \%data);
1785 $stmt = "INSERT INTO solar_system (planets) VALUES (?)"
1787 @bind = (['Mercury', 'Venus', 'Earth', 'Mars']);
1790 =head2 Inserting and Updating SQL
1792 In order to apply SQL functions to elements of your C<%data> you may
1793 specify a reference to an arrayref for the given hash value. For example,
1794 if you need to execute the Oracle C<to_date> function on a value, you can
1795 say something like this:
1799 date_entered => \[ "to_date(?,'MM/DD/YYYY')", "03/02/2003" ],
1802 The first value in the array is the actual SQL. Any other values are
1803 optional and would be included in the bind values array. This gives
1806 my($stmt, @bind) = $sql->insert('people', \%data);
1808 $stmt = "INSERT INTO people (name, date_entered)
1809 VALUES (?, to_date(?,'MM/DD/YYYY'))";
1810 @bind = ('Bill', '03/02/2003');
1812 An UPDATE is just as easy, all you change is the name of the function:
1814 my($stmt, @bind) = $sql->update('people', \%data);
1816 Notice that your C<%data> isn't touched; the module will generate
1817 the appropriately quirky SQL for you automatically. Usually you'll
1818 want to specify a WHERE clause for your UPDATE, though, which is
1819 where handling C<%where> hashes comes in handy...
1821 =head2 Complex where statements
1823 This module can generate pretty complicated WHERE statements
1824 easily. For example, simple C<key=value> pairs are taken to mean
1825 equality, and if you want to see if a field is within a set
1826 of values, you can use an arrayref. Let's say we wanted to
1827 SELECT some data based on this criteria:
1830 requestor => 'inna',
1831 worker => ['nwiger', 'rcwe', 'sfz'],
1832 status => { '!=', 'completed' }
1835 my($stmt, @bind) = $sql->select('tickets', '*', \%where);
1837 The above would give you something like this:
1839 $stmt = "SELECT * FROM tickets WHERE
1840 ( requestor = ? ) AND ( status != ? )
1841 AND ( worker = ? OR worker = ? OR worker = ? )";
1842 @bind = ('inna', 'completed', 'nwiger', 'rcwe', 'sfz');
1844 Which you could then use in DBI code like so:
1846 my $sth = $dbh->prepare($stmt);
1847 $sth->execute(@bind);
1853 The methods are simple. There's one for every major SQL operation,
1854 and a constructor you use first. The arguments are specified in a
1855 similar order for each method (table, then fields, then a where
1856 clause) to try and simplify things.
1858 =head2 new(option => 'value')
1860 The C<new()> function takes a list of options and values, and returns
1861 a new B<SQL::Abstract> object which can then be used to generate SQL
1862 through the methods below. The options accepted are:
1868 If set to 'lower', then SQL will be generated in all lowercase. By
1869 default SQL is generated in "textbook" case meaning something like:
1871 SELECT a_field FROM a_table WHERE some_field LIKE '%someval%'
1873 Any setting other than 'lower' is ignored.
1877 This determines what the default comparison operator is. By default
1878 it is C<=>, meaning that a hash like this:
1880 %where = (name => 'nwiger', email => 'nate@wiger.org');
1882 Will generate SQL like this:
1884 WHERE name = 'nwiger' AND email = 'nate@wiger.org'
1886 However, you may want loose comparisons by default, so if you set
1887 C<cmp> to C<like> you would get SQL such as:
1889 WHERE name like 'nwiger' AND email like 'nate@wiger.org'
1891 You can also override the comparison on an individual basis - see
1892 the huge section on L</"WHERE CLAUSES"> at the bottom.
1894 =item sqltrue, sqlfalse
1896 Expressions for inserting boolean values within SQL statements.
1897 By default these are C<1=1> and C<1=0>. They are used
1898 by the special operators C<-in> and C<-not_in> for generating
1899 correct SQL even when the argument is an empty array (see below).
1903 This determines the default logical operator for multiple WHERE
1904 statements in arrays or hashes. If absent, the default logic is "or"
1905 for arrays, and "and" for hashes. This means that a WHERE
1909 event_date => {'>=', '2/13/99'},
1910 event_date => {'<=', '4/24/03'},
1913 will generate SQL like this:
1915 WHERE event_date >= '2/13/99' OR event_date <= '4/24/03'
1917 This is probably not what you want given this query, though (look
1918 at the dates). To change the "OR" to an "AND", simply specify:
1920 my $sql = SQL::Abstract->new(logic => 'and');
1922 Which will change the above C<WHERE> to:
1924 WHERE event_date >= '2/13/99' AND event_date <= '4/24/03'
1926 The logic can also be changed locally by inserting
1927 a modifier in front of an arrayref:
1929 @where = (-and => [event_date => {'>=', '2/13/99'},
1930 event_date => {'<=', '4/24/03'} ]);
1932 See the L</"WHERE CLAUSES"> section for explanations.
1936 This will automatically convert comparisons using the specified SQL
1937 function for both column and value. This is mostly used with an argument
1938 of C<upper> or C<lower>, so that the SQL will have the effect of
1939 case-insensitive "searches". For example, this:
1941 $sql = SQL::Abstract->new(convert => 'upper');
1942 %where = (keywords => 'MaKe iT CAse inSeNSItive');
1944 Will turn out the following SQL:
1946 WHERE upper(keywords) like upper('MaKe iT CAse inSeNSItive')
1948 The conversion can be C<upper()>, C<lower()>, or any other SQL function
1949 that can be applied symmetrically to fields (actually B<SQL::Abstract> does
1950 not validate this option; it will just pass through what you specify verbatim).
1954 This is a kludge because many databases suck. For example, you can't
1955 just bind values using DBI's C<execute()> for Oracle C<CLOB> or C<BLOB> fields.
1956 Instead, you have to use C<bind_param()>:
1958 $sth->bind_param(1, 'reg data');
1959 $sth->bind_param(2, $lots, {ora_type => ORA_CLOB});
1961 The problem is, B<SQL::Abstract> will normally just return a C<@bind> array,
1962 which loses track of which field each slot refers to. Fear not.
1964 If you specify C<bindtype> in new, you can determine how C<@bind> is returned.
1965 Currently, you can specify either C<normal> (default) or C<columns>. If you
1966 specify C<columns>, you will get an array that looks like this:
1968 my $sql = SQL::Abstract->new(bindtype => 'columns');
1969 my($stmt, @bind) = $sql->insert(...);
1972 [ 'column1', 'value1' ],
1973 [ 'column2', 'value2' ],
1974 [ 'column3', 'value3' ],
1977 You can then iterate through this manually, using DBI's C<bind_param()>.
1979 $sth->prepare($stmt);
1982 my($col, $data) = @$_;
1983 if ($col eq 'details' || $col eq 'comments') {
1984 $sth->bind_param($i, $data, {ora_type => ORA_CLOB});
1985 } elsif ($col eq 'image') {
1986 $sth->bind_param($i, $data, {ora_type => ORA_BLOB});
1988 $sth->bind_param($i, $data);
1992 $sth->execute; # execute without @bind now
1994 Now, why would you still use B<SQL::Abstract> if you have to do this crap?
1995 Basically, the advantage is still that you don't have to care which fields
1996 are or are not included. You could wrap that above C<for> loop in a simple
1997 sub called C<bind_fields()> or something and reuse it repeatedly. You still
1998 get a layer of abstraction over manual SQL specification.
2000 Note that if you set L</bindtype> to C<columns>, the C<\[ $sql, @bind ]>
2001 construct (see L</Literal SQL with placeholders and bind values (subqueries)>)
2002 will expect the bind values in this format.
2006 This is the character that a table or column name will be quoted
2007 with. By default this is an empty string, but you could set it to
2008 the character C<`>, to generate SQL like this:
2010 SELECT `a_field` FROM `a_table` WHERE `some_field` LIKE '%someval%'
2012 Alternatively, you can supply an array ref of two items, the first being the left
2013 hand quote character, and the second the right hand quote character. For
2014 example, you could supply C<['[',']']> for SQL Server 2000 compliant quotes
2015 that generates SQL like this:
2017 SELECT [a_field] FROM [a_table] WHERE [some_field] LIKE '%someval%'
2019 Quoting is useful if you have tables or columns names that are reserved
2020 words in your database's SQL dialect.
2024 This is the character that will be used to escape L</quote_char>s appearing
2025 in an identifier before it has been quoted.
2027 The parameter default in case of a single L</quote_char> character is the quote
2030 When opening-closing-style quoting is used (L</quote_char> is an arrayref)
2031 this parameter defaults to the B<closing (right)> L</quote_char>. Occurrences
2032 of the B<opening (left)> L</quote_char> within the identifier are currently left
2033 untouched. The default for opening-closing-style quotes may change in future
2034 versions, thus you are B<strongly encouraged> to specify the escape character
2039 This is the character that separates a table and column name. It is
2040 necessary to specify this when the C<quote_char> option is selected,
2041 so that tables and column names can be individually quoted like this:
2043 SELECT `table`.`one_field` FROM `table` WHERE `table`.`other_field` = 1
2045 =item injection_guard
2047 A regular expression C<qr/.../> that is applied to any C<-function> and unquoted
2048 column name specified in a query structure. This is a safety mechanism to avoid
2049 injection attacks when mishandling user input e.g.:
2051 my %condition_as_column_value_pairs = get_values_from_user();
2052 $sqla->select( ... , \%condition_as_column_value_pairs );
2054 If the expression matches an exception is thrown. Note that literal SQL
2055 supplied via C<\'...'> or C<\['...']> is B<not> checked in any way.
2057 Defaults to checking for C<;> and the C<GO> keyword (TransactSQL)
2059 =item array_datatypes
2061 When this option is true, arrayrefs in INSERT or UPDATE are
2062 interpreted as array datatypes and are passed directly
2064 When this option is false, arrayrefs are interpreted
2065 as literal SQL, just like refs to arrayrefs
2066 (but this behavior is for backwards compatibility; when writing
2067 new queries, use the "reference to arrayref" syntax
2073 Takes a reference to a list of "special operators"
2074 to extend the syntax understood by L<SQL::Abstract>.
2075 See section L</"SPECIAL OPERATORS"> for details.
2079 Takes a reference to a list of "unary operators"
2080 to extend the syntax understood by L<SQL::Abstract>.
2081 See section L</"UNARY OPERATORS"> for details.
2087 =head2 insert($table, \@values || \%fieldvals, \%options)
2089 This is the simplest function. You simply give it a table name
2090 and either an arrayref of values or hashref of field/value pairs.
2091 It returns an SQL INSERT statement and a list of bind values.
2092 See the sections on L</"Inserting and Updating Arrays"> and
2093 L</"Inserting and Updating SQL"> for information on how to insert
2094 with those data types.
2096 The optional C<\%options> hash reference may contain additional
2097 options to generate the insert SQL. Currently supported options
2104 Takes either a scalar of raw SQL fields, or an array reference of
2105 field names, and adds on an SQL C<RETURNING> statement at the end.
2106 This allows you to return data generated by the insert statement
2107 (such as row IDs) without performing another C<SELECT> statement.
2108 Note, however, this is not part of the SQL standard and may not
2109 be supported by all database engines.
2113 =head2 update($table, \%fieldvals, \%where, \%options)
2115 This takes a table, hashref of field/value pairs, and an optional
2116 hashref L<WHERE clause|/WHERE CLAUSES>. It returns an SQL UPDATE function and a list
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 update SQL. Currently supported options
2130 See the C<returning> option to
2131 L<insert|/insert($table, \@values || \%fieldvals, \%options)>.
2135 =head2 select($source, $fields, $where, $order)
2137 This returns a SQL SELECT statement and associated list of bind values, as
2138 specified by the arguments:
2144 Specification of the 'FROM' part of the statement.
2145 The argument can be either a plain scalar (interpreted as a table
2146 name, will be quoted), or an arrayref (interpreted as a list
2147 of table names, joined by commas, quoted), or a scalarref
2148 (literal SQL, not quoted).
2152 Specification of the list of fields to retrieve from
2154 The argument can be either an arrayref (interpreted as a list
2155 of field names, will be joined by commas and quoted), or a
2156 plain scalar (literal SQL, not quoted).
2157 Please observe that this API is not as flexible as that of
2158 the first argument C<$source>, for backwards compatibility reasons.
2162 Optional argument to specify the WHERE part of the query.
2163 The argument is most often a hashref, but can also be
2164 an arrayref or plain scalar --
2165 see section L<WHERE clause|/"WHERE CLAUSES"> for details.
2169 Optional argument to specify the ORDER BY part of the query.
2170 The argument can be a scalar, a hashref or an arrayref
2171 -- see section L<ORDER BY clause|/"ORDER BY CLAUSES">
2177 =head2 delete($table, \%where, \%options)
2179 This takes a table name and optional hashref L<WHERE clause|/WHERE CLAUSES>.
2180 It returns an SQL DELETE statement and list of bind values.
2182 The optional C<\%options> hash reference may contain additional
2183 options to generate the delete SQL. Currently supported options
2190 See the C<returning> option to
2191 L<insert|/insert($table, \@values || \%fieldvals, \%options)>.
2195 =head2 where(\%where, $order)
2197 This is used to generate just the WHERE clause. For example,
2198 if you have an arbitrary data structure and know what the
2199 rest of your SQL is going to look like, but want an easy way
2200 to produce a WHERE clause, use this. It returns an SQL WHERE
2201 clause and list of bind values.
2204 =head2 values(\%data)
2206 This just returns the values from the hash C<%data>, in the same
2207 order that would be returned from any of the other above queries.
2208 Using this allows you to markedly speed up your queries if you
2209 are affecting lots of rows. See below under the L</"PERFORMANCE"> section.
2211 =head2 generate($any, 'number', $of, \@data, $struct, \%types)
2213 Warning: This is an experimental method and subject to change.
2215 This returns arbitrarily generated SQL. It's a really basic shortcut.
2216 It will return two different things, depending on return context:
2218 my($stmt, @bind) = $sql->generate('create table', \$table, \@fields);
2219 my $stmt_and_val = $sql->generate('create table', \$table, \@fields);
2221 These would return the following:
2223 # First calling form
2224 $stmt = "CREATE TABLE test (?, ?)";
2225 @bind = (field1, field2);
2227 # Second calling form
2228 $stmt_and_val = "CREATE TABLE test (field1, field2)";
2230 Depending on what you're trying to do, it's up to you to choose the correct
2231 format. In this example, the second form is what you would want.
2235 $sql->generate('alter session', { nls_date_format => 'MM/YY' });
2239 ALTER SESSION SET nls_date_format = 'MM/YY'
2241 You get the idea. Strings get their case twiddled, but everything
2242 else remains verbatim.
2244 =head1 EXPORTABLE FUNCTIONS
2246 =head2 is_plain_value
2248 Determines if the supplied argument is a plain value as understood by this
2253 =item * The value is C<undef>
2255 =item * The value is a non-reference
2257 =item * The value is an object with stringification overloading
2259 =item * The value is of the form C<< { -value => $anything } >>
2263 On failure returns C<undef>, on success returns a B<scalar> reference
2264 to the original supplied argument.
2270 The stringification overloading detection is rather advanced: it takes
2271 into consideration not only the presence of a C<""> overload, but if that
2272 fails also checks for enabled
2273 L<autogenerated versions of C<"">|overload/Magic Autogeneration>, based
2274 on either C<0+> or C<bool>.
2276 Unfortunately testing in the field indicates that this
2277 detection B<< may tickle a latent bug in perl versions before 5.018 >>,
2278 but only when very large numbers of stringifying objects are involved.
2279 At the time of writing ( Sep 2014 ) there is no clear explanation of
2280 the direct cause, nor is there a manageably small test case that reliably
2281 reproduces the problem.
2283 If you encounter any of the following exceptions in B<random places within
2284 your application stack> - this module may be to blame:
2286 Operation "ne": no method found,
2287 left argument in overloaded package <something>,
2288 right argument in overloaded package <something>
2292 Stub found while resolving method "???" overloading """" in package <something>
2294 If you fall victim to the above - please attempt to reduce the problem
2295 to something that could be sent to the L<SQL::Abstract developers
2296 |DBIx::Class/GETTING HELP/SUPPORT>
2297 (either publicly or privately). As a workaround in the meantime you can
2298 set C<$ENV{SQLA_ISVALUE_IGNORE_AUTOGENERATED_STRINGIFICATION}> to a true
2299 value, which will most likely eliminate your problem (at the expense of
2300 not being able to properly detect exotic forms of stringification).
2302 This notice and environment variable will be removed in a future version,
2303 as soon as the underlying problem is found and a reliable workaround is
2308 =head2 is_literal_value
2310 Determines if the supplied argument is a literal value as understood by this
2315 =item * C<\$sql_string>
2317 =item * C<\[ $sql_string, @bind_values ]>
2321 On failure returns C<undef>, on success returns an B<array> reference
2322 containing the unpacked version of the supplied literal SQL and bind values.
2324 =head1 WHERE CLAUSES
2328 This module uses a variation on the idea from L<DBIx::Abstract>. It
2329 is B<NOT>, repeat I<not> 100% compatible. B<The main logic of this
2330 module is that things in arrays are OR'ed, and things in hashes
2333 The easiest way to explain is to show lots of examples. After
2334 each C<%where> hash shown, it is assumed you used:
2336 my($stmt, @bind) = $sql->where(\%where);
2338 However, note that the C<%where> hash can be used directly in any
2339 of the other functions as well, as described above.
2341 =head2 Key-value pairs
2343 So, let's get started. To begin, a simple hash:
2347 status => 'completed'
2350 Is converted to SQL C<key = val> statements:
2352 $stmt = "WHERE user = ? AND status = ?";
2353 @bind = ('nwiger', 'completed');
2355 One common thing I end up doing is having a list of values that
2356 a field can be in. To do this, simply specify a list inside of
2361 status => ['assigned', 'in-progress', 'pending'];
2364 This simple code will create the following:
2366 $stmt = "WHERE user = ? AND ( status = ? OR status = ? OR status = ? )";
2367 @bind = ('nwiger', 'assigned', 'in-progress', 'pending');
2369 A field associated to an empty arrayref will be considered a
2370 logical false and will generate 0=1.
2372 =head2 Tests for NULL values
2374 If the value part is C<undef> then this is converted to SQL <IS NULL>
2383 $stmt = "WHERE user = ? AND status IS NULL";
2386 To test if a column IS NOT NULL:
2390 status => { '!=', undef },
2393 =head2 Specific comparison operators
2395 If you want to specify a different type of operator for your comparison,
2396 you can use a hashref for a given column:
2400 status => { '!=', 'completed' }
2403 Which would generate:
2405 $stmt = "WHERE user = ? AND status != ?";
2406 @bind = ('nwiger', 'completed');
2408 To test against multiple values, just enclose the values in an arrayref:
2410 status => { '=', ['assigned', 'in-progress', 'pending'] };
2412 Which would give you:
2414 "WHERE status = ? OR status = ? OR status = ?"
2417 The hashref can also contain multiple pairs, in which case it is expanded
2418 into an C<AND> of its elements:
2422 status => { '!=', 'completed', -not_like => 'pending%' }
2425 # Or more dynamically, like from a form
2426 $where{user} = 'nwiger';
2427 $where{status}{'!='} = 'completed';
2428 $where{status}{'-not_like'} = 'pending%';
2430 # Both generate this
2431 $stmt = "WHERE user = ? AND status != ? AND status NOT LIKE ?";
2432 @bind = ('nwiger', 'completed', 'pending%');
2435 To get an OR instead, you can combine it with the arrayref idea:
2439 priority => [ { '=', 2 }, { '>', 5 } ]
2442 Which would generate:
2444 $stmt = "WHERE ( priority = ? OR priority > ? ) AND user = ?";
2445 @bind = ('2', '5', 'nwiger');
2447 If you want to include literal SQL (with or without bind values), just use a
2448 scalar reference or reference to an arrayref as the value:
2451 date_entered => { '>' => \["to_date(?, 'MM/DD/YYYY')", "11/26/2008"] },
2452 date_expires => { '<' => \"now()" }
2455 Which would generate:
2457 $stmt = "WHERE date_entered > to_date(?, 'MM/DD/YYYY') AND date_expires < now()";
2458 @bind = ('11/26/2008');
2461 =head2 Logic and nesting operators
2463 In the example above,
2464 there is a subtle trap if you want to say something like
2465 this (notice the C<AND>):
2467 WHERE priority != ? AND priority != ?
2469 Because, in Perl you I<can't> do this:
2471 priority => { '!=' => 2, '!=' => 1 }
2473 As the second C<!=> key will obliterate the first. The solution
2474 is to use the special C<-modifier> form inside an arrayref:
2476 priority => [ -and => {'!=', 2},
2480 Normally, these would be joined by C<OR>, but the modifier tells it
2481 to use C<AND> instead. (Hint: You can use this in conjunction with the
2482 C<logic> option to C<new()> in order to change the way your queries
2483 work by default.) B<Important:> Note that the C<-modifier> goes
2484 B<INSIDE> the arrayref, as an extra first element. This will
2485 B<NOT> do what you think it might:
2487 priority => -and => [{'!=', 2}, {'!=', 1}] # WRONG!
2489 Here is a quick list of equivalencies, since there is some overlap:
2492 status => {'!=', 'completed', 'not like', 'pending%' }
2493 status => [ -and => {'!=', 'completed'}, {'not like', 'pending%'}]
2496 status => {'=', ['assigned', 'in-progress']}
2497 status => [ -or => {'=', 'assigned'}, {'=', 'in-progress'}]
2498 status => [ {'=', 'assigned'}, {'=', 'in-progress'} ]
2502 =head2 Special operators: IN, BETWEEN, etc.
2504 You can also use the hashref format to compare a list of fields using the
2505 C<IN> comparison operator, by specifying the list as an arrayref:
2508 status => 'completed',
2509 reportid => { -in => [567, 2335, 2] }
2512 Which would generate:
2514 $stmt = "WHERE status = ? AND reportid IN (?,?,?)";
2515 @bind = ('completed', '567', '2335', '2');
2517 The reverse operator C<-not_in> generates SQL C<NOT IN> and is used in
2520 If the argument to C<-in> is an empty array, 'sqlfalse' is generated
2521 (by default: C<1=0>). Similarly, C<< -not_in => [] >> generates
2522 'sqltrue' (by default: C<1=1>).
2524 In addition to the array you can supply a chunk of literal sql or
2525 literal sql with bind:
2528 customer => { -in => \[
2529 'SELECT cust_id FROM cust WHERE balance > ?',
2532 status => { -in => \'SELECT status_codes FROM states' },
2538 customer IN ( SELECT cust_id FROM cust WHERE balance > ? )
2539 AND status IN ( SELECT status_codes FROM states )
2543 Finally, if the argument to C<-in> is not a reference, it will be
2544 treated as a single-element array.
2546 Another pair of operators is C<-between> and C<-not_between>,
2547 used with an arrayref of two values:
2551 completion_date => {
2552 -not_between => ['2002-10-01', '2003-02-06']
2558 WHERE user = ? AND completion_date NOT BETWEEN ( ? AND ? )
2560 Just like with C<-in> all plausible combinations of literal SQL
2564 start0 => { -between => [ 1, 2 ] },
2565 start1 => { -between => \["? AND ?", 1, 2] },
2566 start2 => { -between => \"lower(x) AND upper(y)" },
2567 start3 => { -between => [
2569 \["upper(?)", 'stuff' ],
2576 ( start0 BETWEEN ? AND ? )
2577 AND ( start1 BETWEEN ? AND ? )
2578 AND ( start2 BETWEEN lower(x) AND upper(y) )
2579 AND ( start3 BETWEEN lower(x) AND upper(?) )
2581 @bind = (1, 2, 1, 2, 'stuff');
2584 These are the two builtin "special operators"; but the
2585 list can be expanded: see section L</"SPECIAL OPERATORS"> below.
2587 =head2 Unary operators: bool
2589 If you wish to test against boolean columns or functions within your
2590 database you can use the C<-bool> and C<-not_bool> operators. For
2591 example to test the column C<is_user> being true and the column
2592 C<is_enabled> being false you would use:-
2596 -not_bool => 'is_enabled',
2601 WHERE is_user AND NOT is_enabled
2603 If a more complex combination is required, testing more conditions,
2604 then you should use the and/or operators:-
2609 -not_bool => { two=> { -rlike => 'bar' } },
2610 -not_bool => { three => [ { '=', 2 }, { '>', 5 } ] },
2621 (NOT ( three = ? OR three > ? ))
2624 =head2 Nested conditions, -and/-or prefixes
2626 So far, we've seen how multiple conditions are joined with a top-level
2627 C<AND>. We can change this by putting the different conditions we want in
2628 hashes and then putting those hashes in an array. For example:
2633 status => { -like => ['pending%', 'dispatched'] },
2637 status => 'unassigned',
2641 This data structure would create the following:
2643 $stmt = "WHERE ( user = ? AND ( status LIKE ? OR status LIKE ? ) )
2644 OR ( user = ? AND status = ? ) )";
2645 @bind = ('nwiger', 'pending', 'dispatched', 'robot', 'unassigned');
2648 Clauses in hashrefs or arrayrefs can be prefixed with an C<-and> or C<-or>
2649 to change the logic inside:
2655 -and => [ workhrs => {'>', 20}, geo => 'ASIA' ],
2656 -or => { workhrs => {'<', 50}, geo => 'EURO' },
2663 $stmt = "WHERE ( user = ?
2664 AND ( ( workhrs > ? AND geo = ? )
2665 OR ( workhrs < ? OR geo = ? ) ) )";
2666 @bind = ('nwiger', '20', 'ASIA', '50', 'EURO');
2668 =head3 Algebraic inconsistency, for historical reasons
2670 C<Important note>: when connecting several conditions, the C<-and->|C<-or>
2671 operator goes C<outside> of the nested structure; whereas when connecting
2672 several constraints on one column, the C<-and> operator goes
2673 C<inside> the arrayref. Here is an example combining both features:
2676 -and => [a => 1, b => 2],
2677 -or => [c => 3, d => 4],
2678 e => [-and => {-like => 'foo%'}, {-like => '%bar'} ]
2683 WHERE ( ( ( a = ? AND b = ? )
2684 OR ( c = ? OR d = ? )
2685 OR ( e LIKE ? AND e LIKE ? ) ) )
2687 This difference in syntax is unfortunate but must be preserved for
2688 historical reasons. So be careful: the two examples below would
2689 seem algebraically equivalent, but they are not
2692 { -like => 'foo%' },
2693 { -like => '%bar' },
2695 # yields: WHERE ( ( col LIKE ? AND col LIKE ? ) )
2698 { col => { -like => 'foo%' } },
2699 { col => { -like => '%bar' } },
2701 # yields: WHERE ( ( col LIKE ? OR col LIKE ? ) )
2704 =head2 Literal SQL and value type operators
2706 The basic premise of SQL::Abstract is that in WHERE specifications the "left
2707 side" is a column name and the "right side" is a value (normally rendered as
2708 a placeholder). This holds true for both hashrefs and arrayref pairs as you
2709 see in the L</WHERE CLAUSES> examples above. Sometimes it is necessary to
2710 alter this behavior. There are several ways of doing so.
2714 This is a virtual operator that signals the string to its right side is an
2715 identifier (a column name) and not a value. For example to compare two
2716 columns you would write:
2719 priority => { '<', 2 },
2720 requestor => { -ident => 'submitter' },
2725 $stmt = "WHERE priority < ? AND requestor = submitter";
2728 If you are maintaining legacy code you may see a different construct as
2729 described in L</Deprecated usage of Literal SQL>, please use C<-ident> in new
2734 This is a virtual operator that signals that the construct to its right side
2735 is a value to be passed to DBI. This is for example necessary when you want
2736 to write a where clause against an array (for RDBMS that support such
2737 datatypes). For example:
2740 array => { -value => [1, 2, 3] }
2745 $stmt = 'WHERE array = ?';
2746 @bind = ([1, 2, 3]);
2748 Note that if you were to simply say:
2754 the result would probably not be what you wanted:
2756 $stmt = 'WHERE array = ? OR array = ? OR array = ?';
2761 Finally, sometimes only literal SQL will do. To include a random snippet
2762 of SQL verbatim, you specify it as a scalar reference. Consider this only
2763 as a last resort. Usually there is a better way. For example:
2766 priority => { '<', 2 },
2767 requestor => { -in => \'(SELECT name FROM hitmen)' },
2772 $stmt = "WHERE priority < ? AND requestor IN (SELECT name FROM hitmen)"
2775 Note that in this example, you only get one bind parameter back, since
2776 the verbatim SQL is passed as part of the statement.
2780 Never use untrusted input as a literal SQL argument - this is a massive
2781 security risk (there is no way to check literal snippets for SQL
2782 injections and other nastyness). If you need to deal with untrusted input
2783 use literal SQL with placeholders as described next.
2785 =head3 Literal SQL with placeholders and bind values (subqueries)
2787 If the literal SQL to be inserted has placeholders and bind values,
2788 use a reference to an arrayref (yes this is a double reference --
2789 not so common, but perfectly legal Perl). For example, to find a date
2790 in Postgres you can use something like this:
2793 date_column => \[ "= date '2008-09-30' - ?::integer", 10 ]
2798 $stmt = "WHERE ( date_column = date '2008-09-30' - ?::integer )"
2801 Note that you must pass the bind values in the same format as they are returned
2802 by L<where|/where(\%where, $order)>. This means that if you set L</bindtype>
2803 to C<columns>, you must provide the bind values in the
2804 C<< [ column_meta => value ] >> format, where C<column_meta> is an opaque
2805 scalar value; most commonly the column name, but you can use any scalar value
2806 (including references and blessed references), L<SQL::Abstract> will simply
2807 pass it through intact. So if C<bindtype> is set to C<columns> the above
2808 example will look like:
2811 date_column => \[ "= date '2008-09-30' - ?::integer", [ {} => 10 ] ]
2814 Literal SQL is especially useful for nesting parenthesized clauses in the
2815 main SQL query. Here is a first example:
2817 my ($sub_stmt, @sub_bind) = ("SELECT c1 FROM t1 WHERE c2 < ? AND c3 LIKE ?",
2821 bar => \["IN ($sub_stmt)" => @sub_bind],
2826 $stmt = "WHERE (foo = ? AND bar IN (SELECT c1 FROM t1
2827 WHERE c2 < ? AND c3 LIKE ?))";
2828 @bind = (1234, 100, "foo%");
2830 Other subquery operators, like for example C<"E<gt> ALL"> or C<"NOT IN">,
2831 are expressed in the same way. Of course the C<$sub_stmt> and
2832 its associated bind values can be generated through a former call
2835 my ($sub_stmt, @sub_bind)
2836 = $sql->select("t1", "c1", {c2 => {"<" => 100},
2837 c3 => {-like => "foo%"}});
2840 bar => \["> ALL ($sub_stmt)" => @sub_bind],
2843 In the examples above, the subquery was used as an operator on a column;
2844 but the same principle also applies for a clause within the main C<%where>
2845 hash, like an EXISTS subquery:
2847 my ($sub_stmt, @sub_bind)
2848 = $sql->select("t1", "*", {c1 => 1, c2 => \"> t0.c0"});
2849 my %where = ( -and => [
2851 \["EXISTS ($sub_stmt)" => @sub_bind],
2856 $stmt = "WHERE (foo = ? AND EXISTS (SELECT * FROM t1
2857 WHERE c1 = ? AND c2 > t0.c0))";
2861 Observe that the condition on C<c2> in the subquery refers to
2862 column C<t0.c0> of the main query: this is I<not> a bind
2863 value, so we have to express it through a scalar ref.
2864 Writing C<< c2 => {">" => "t0.c0"} >> would have generated
2865 C<< c2 > ? >> with bind value C<"t0.c0"> ... not exactly
2866 what we wanted here.
2868 Finally, here is an example where a subquery is used
2869 for expressing unary negation:
2871 my ($sub_stmt, @sub_bind)
2872 = $sql->where({age => [{"<" => 10}, {">" => 20}]});
2873 $sub_stmt =~ s/^ where //i; # don't want "WHERE" in the subclause
2875 lname => {like => '%son%'},
2876 \["NOT ($sub_stmt)" => @sub_bind],
2881 $stmt = "lname LIKE ? AND NOT ( age < ? OR age > ? )"
2882 @bind = ('%son%', 10, 20)
2884 =head3 Deprecated usage of Literal SQL
2886 Below are some examples of archaic use of literal SQL. It is shown only as
2887 reference for those who deal with legacy code. Each example has a much
2888 better, cleaner and safer alternative that users should opt for in new code.
2894 my %where = ( requestor => \'IS NOT NULL' )
2896 $stmt = "WHERE requestor IS NOT NULL"
2898 This used to be the way of generating NULL comparisons, before the handling
2899 of C<undef> got formalized. For new code please use the superior syntax as
2900 described in L</Tests for NULL values>.
2904 my %where = ( requestor => \'= submitter' )
2906 $stmt = "WHERE requestor = submitter"
2908 This used to be the only way to compare columns. Use the superior L</-ident>
2909 method for all new code. For example an identifier declared in such a way
2910 will be properly quoted if L</quote_char> is properly set, while the legacy
2911 form will remain as supplied.
2915 my %where = ( is_ready => \"", completed => { '>', '2012-12-21' } )
2917 $stmt = "WHERE completed > ? AND is_ready"
2918 @bind = ('2012-12-21')
2920 Using an empty string literal used to be the only way to express a boolean.
2921 For all new code please use the much more readable
2922 L<-bool|/Unary operators: bool> operator.
2928 These pages could go on for a while, since the nesting of the data
2929 structures this module can handle are pretty much unlimited (the
2930 module implements the C<WHERE> expansion as a recursive function
2931 internally). Your best bet is to "play around" with the module a
2932 little to see how the data structures behave, and choose the best
2933 format for your data based on that.
2935 And of course, all the values above will probably be replaced with
2936 variables gotten from forms or the command line. After all, if you
2937 knew everything ahead of time, you wouldn't have to worry about
2938 dynamically-generating SQL and could just hardwire it into your
2941 =head1 ORDER BY CLAUSES
2943 Some functions take an order by clause. This can either be a scalar (just a
2944 column name), a hashref of C<< { -desc => 'col' } >> or C<< { -asc => 'col' }
2945 >>, a scalarref, an arrayref-ref, or an arrayref of any of the previous
2948 Given | Will Generate
2949 ---------------------------------------------------------------
2951 'colA' | ORDER BY colA
2953 [qw/colA colB/] | ORDER BY colA, colB
2955 {-asc => 'colA'} | ORDER BY colA ASC
2957 {-desc => 'colB'} | ORDER BY colB DESC
2959 ['colA', {-asc => 'colB'}] | ORDER BY colA, colB ASC
2961 { -asc => [qw/colA colB/] } | ORDER BY colA ASC, colB ASC
2963 \'colA DESC' | ORDER BY colA DESC
2965 \[ 'FUNC(colA, ?)', $x ] | ORDER BY FUNC(colA, ?)
2966 | /* ...with $x bound to ? */
2969 { -asc => 'colA' }, | colA ASC,
2970 { -desc => [qw/colB/] }, | colB DESC,
2971 { -asc => [qw/colC colD/] },| colC ASC, colD ASC,
2972 \'colE DESC', | colE DESC,
2973 \[ 'FUNC(colF, ?)', $x ], | FUNC(colF, ?)
2974 ] | /* ...with $x bound to ? */
2975 ===============================================================
2979 =head1 SPECIAL OPERATORS
2981 my $sqlmaker = SQL::Abstract->new(special_ops => [
2985 my ($self, $field, $op, $arg) = @_;
2991 handler => 'method_name',
2995 A "special operator" is a SQL syntactic clause that can be
2996 applied to a field, instead of a usual binary operator.
2999 WHERE field IN (?, ?, ?)
3000 WHERE field BETWEEN ? AND ?
3001 WHERE MATCH(field) AGAINST (?, ?)
3003 Special operators IN and BETWEEN are fairly standard and therefore
3004 are builtin within C<SQL::Abstract> (as the overridable methods
3005 C<_where_field_IN> and C<_where_field_BETWEEN>). For other operators,
3006 like the MATCH .. AGAINST example above which is specific to MySQL,
3007 you can write your own operator handlers - supply a C<special_ops>
3008 argument to the C<new> method. That argument takes an arrayref of
3009 operator definitions; each operator definition is a hashref with two
3016 the regular expression to match the operator
3020 Either a coderef or a plain scalar method name. In both cases
3021 the expected return is C<< ($sql, @bind) >>.
3023 When supplied with a method name, it is simply called on the
3024 L<SQL::Abstract> object as:
3026 $self->$method_name($field, $op, $arg)
3030 $field is the LHS of the operator
3031 $op is the part that matched the handler regex
3034 When supplied with a coderef, it is called as:
3036 $coderef->($self, $field, $op, $arg)
3041 For example, here is an implementation
3042 of the MATCH .. AGAINST syntax for MySQL
3044 my $sqlmaker = SQL::Abstract->new(special_ops => [
3046 # special op for MySql MATCH (field) AGAINST(word1, word2, ...)
3047 {regex => qr/^match$/i,
3049 my ($self, $field, $op, $arg) = @_;
3050 $arg = [$arg] if not ref $arg;
3051 my $label = $self->_quote($field);
3052 my ($placeholder) = $self->_convert('?');
3053 my $placeholders = join ", ", (($placeholder) x @$arg);
3054 my $sql = $self->_sqlcase('match') . " ($label) "
3055 . $self->_sqlcase('against') . " ($placeholders) ";
3056 my @bind = $self->_bindtype($field, @$arg);
3057 return ($sql, @bind);
3064 =head1 UNARY OPERATORS
3066 my $sqlmaker = SQL::Abstract->new(unary_ops => [
3070 my ($self, $op, $arg) = @_;
3076 handler => 'method_name',
3080 A "unary operator" is a SQL syntactic clause that can be
3081 applied to a field - the operator goes before the field
3083 You can write your own operator handlers - supply a C<unary_ops>
3084 argument to the C<new> method. That argument takes an arrayref of
3085 operator definitions; each operator definition is a hashref with two
3092 the regular expression to match the operator
3096 Either a coderef or a plain scalar method name. In both cases
3097 the expected return is C<< $sql >>.
3099 When supplied with a method name, it is simply called on the
3100 L<SQL::Abstract> object as:
3102 $self->$method_name($op, $arg)
3106 $op is the part that matched the handler regex
3107 $arg is the RHS or argument of the operator
3109 When supplied with a coderef, it is called as:
3111 $coderef->($self, $op, $arg)
3119 Thanks to some benchmarking by Mark Stosberg, it turns out that
3120 this module is many orders of magnitude faster than using C<DBIx::Abstract>.
3121 I must admit this wasn't an intentional design issue, but it's a
3122 byproduct of the fact that you get to control your C<DBI> handles
3125 To maximize performance, use a code snippet like the following:
3127 # prepare a statement handle using the first row
3128 # and then reuse it for the rest of the rows
3130 for my $href (@array_of_hashrefs) {
3131 $stmt ||= $sql->insert('table', $href);
3132 $sth ||= $dbh->prepare($stmt);
3133 $sth->execute($sql->values($href));
3136 The reason this works is because the keys in your C<$href> are sorted
3137 internally by B<SQL::Abstract>. Thus, as long as your data retains
3138 the same structure, you only have to generate the SQL the first time
3139 around. On subsequent queries, simply use the C<values> function provided
3140 by this module to return your values in the correct order.
3142 However this depends on the values having the same type - if, for
3143 example, the values of a where clause may either have values
3144 (resulting in sql of the form C<column = ?> with a single bind
3145 value), or alternatively the values might be C<undef> (resulting in
3146 sql of the form C<column IS NULL> with no bind value) then the
3147 caching technique suggested will not work.
3151 If you use my C<CGI::FormBuilder> module at all, you'll hopefully
3152 really like this part (I do, at least). Building up a complex query
3153 can be as simple as the following:
3160 use CGI::FormBuilder;
3163 my $form = CGI::FormBuilder->new(...);
3164 my $sql = SQL::Abstract->new;
3166 if ($form->submitted) {
3167 my $field = $form->field;
3168 my $id = delete $field->{id};
3169 my($stmt, @bind) = $sql->update('table', $field, {id => $id});
3172 Of course, you would still have to connect using C<DBI> to run the
3173 query, but the point is that if you make your form look like your
3174 table, the actual query script can be extremely simplistic.
3176 If you're B<REALLY> lazy (I am), check out C<HTML::QuickTable> for
3177 a fast interface to returning and formatting data. I frequently
3178 use these three modules together to write complex database query
3179 apps in under 50 lines.
3181 =head1 HOW TO CONTRIBUTE
3183 Contributions are always welcome, in all usable forms (we especially
3184 welcome documentation improvements). The delivery methods include git-
3185 or unified-diff formatted patches, GitHub pull requests, or plain bug
3186 reports either via RT or the Mailing list. Contributors are generally
3187 granted full access to the official repository after their first several
3188 patches pass successful review.
3190 This project is maintained in a git repository. The code and related tools are
3191 accessible at the following locations:
3195 =item * Official repo: L<git://git.shadowcat.co.uk/dbsrgits/SQL-Abstract.git>
3197 =item * Official gitweb: L<http://git.shadowcat.co.uk/gitweb/gitweb.cgi?p=dbsrgits/SQL-Abstract.git>
3199 =item * GitHub mirror: L<https://github.com/dbsrgits/sql-abstract>
3201 =item * Authorized committers: L<ssh://dbsrgits@git.shadowcat.co.uk/SQL-Abstract.git>
3207 Version 1.50 was a major internal refactoring of C<SQL::Abstract>.
3208 Great care has been taken to preserve the I<published> behavior
3209 documented in previous versions in the 1.* family; however,
3210 some features that were previously undocumented, or behaved
3211 differently from the documentation, had to be changed in order
3212 to clarify the semantics. Hence, client code that was relying
3213 on some dark areas of C<SQL::Abstract> v1.*
3214 B<might behave differently> in v1.50.
3216 The main changes are:
3222 support for literal SQL through the C<< \ [ $sql, @bind ] >> syntax.
3226 support for the { operator => \"..." } construct (to embed literal SQL)
3230 support for the { operator => \["...", @bind] } construct (to embed literal SQL with bind values)
3234 optional support for L<array datatypes|/"Inserting and Updating Arrays">
3238 defensive programming: check arguments
3242 fixed bug with global logic, which was previously implemented
3243 through global variables yielding side-effects. Prior versions would
3244 interpret C<< [ {cond1, cond2}, [cond3, cond4] ] >>
3245 as C<< "(cond1 AND cond2) OR (cond3 AND cond4)" >>.
3246 Now this is interpreted
3247 as C<< "(cond1 AND cond2) OR (cond3 OR cond4)" >>.
3252 fixed semantics of _bindtype on array args
3256 dropped the C<_anoncopy> of the %where tree. No longer necessary,
3257 we just avoid shifting arrays within that tree.
3261 dropped the C<_modlogic> function
3265 =head1 ACKNOWLEDGEMENTS
3267 There are a number of individuals that have really helped out with
3268 this module. Unfortunately, most of them submitted bugs via CPAN
3269 so I have no idea who they are! But the people I do know are:
3271 Ash Berlin (order_by hash term support)
3272 Matt Trout (DBIx::Class support)
3273 Mark Stosberg (benchmarking)
3274 Chas Owens (initial "IN" operator support)
3275 Philip Collins (per-field SQL functions)
3276 Eric Kolve (hashref "AND" support)
3277 Mike Fragassi (enhancements to "BETWEEN" and "LIKE")
3278 Dan Kubb (support for "quote_char" and "name_sep")
3279 Guillermo Roditi (patch to cleanup "IN" and "BETWEEN", fix and tests for _order_by)
3280 Laurent Dami (internal refactoring, extensible list of special operators, literal SQL)
3281 Norbert Buchmuller (support for literal SQL in hashpair, misc. fixes & tests)
3282 Peter Rabbitson (rewrite of SQLA::Test, misc. fixes & tests)
3283 Oliver Charles (support for "RETURNING" after "INSERT")
3289 L<DBIx::Class>, L<DBIx::Abstract>, L<CGI::FormBuilder>, L<HTML::QuickTable>.
3293 Copyright (c) 2001-2007 Nathan Wiger <nwiger@cpan.org>. All Rights Reserved.
3295 This module is actively maintained by Matt Trout <mst@shadowcatsystems.co.uk>
3297 For support, your best bet is to try the C<DBIx::Class> users mailing list.
3298 While not an official support venue, C<DBIx::Class> makes heavy use of
3299 C<SQL::Abstract>, and as such list members there are very familiar with
3300 how to create queries.
3304 This module is free software; you may copy this under the same
3305 terms as perl itself (either the GNU General Public License or
3306 the Artistic License)