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 = '1.85';
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) = @_;
482 $self->_SWITCH_refkind($fields, {
483 ARRAYREF => sub { join ', ', map { $self->_quote($_) } @$fields },
486 my ($s, @b) = @$$fields;
487 $self->_assert_bindval_matches_bindtype(@b);
491 SCALARREF => sub { $$fields },
494 SCALAR => sub { $fields },
498 #======================================================================
500 #======================================================================
505 my $table = $self->_table(shift);
509 my($where_sql, @bind) = $self->where($where);
510 my $sql = $self->_sqlcase('delete from ') . $table . $where_sql;
512 if ($options->{returning}) {
513 my ($returning_sql, @returning_bind) = $self->_delete_returning($options);
514 $sql .= $returning_sql;
515 push @bind, @returning_bind;
518 return wantarray ? ($sql, @bind) : $sql;
521 # So that subclasses can override DELETE ... RETURNING separately from
523 sub _delete_returning { shift->_returning(@_) }
527 #======================================================================
529 #======================================================================
533 # Finally, a separate routine just to handle WHERE clauses
535 my ($self, $where, $order) = @_;
538 my ($sql, @bind) = $self->_recurse_where($where);
539 $sql = $sql ? $self->_sqlcase(' where ') . "( $sql )" : '';
543 my ($order_sql, @order_bind) = $self->_order_by($order);
545 push @bind, @order_bind;
548 return wantarray ? ($sql, @bind) : $sql;
553 my ($self, $where, $logic) = @_;
555 # dispatch on appropriate method according to refkind of $where
556 my $method = $self->_METHOD_FOR_refkind("_where", $where);
558 my ($sql, @bind) = $self->$method($where, $logic);
560 # DBIx::Class used to call _recurse_where in scalar context
561 # something else might too...
563 return ($sql, @bind);
566 belch "Calling _recurse_where in scalar context is deprecated and will go away before 2.0";
573 #======================================================================
574 # WHERE: top-level ARRAYREF
575 #======================================================================
578 sub _where_ARRAYREF {
579 my ($self, $where, $logic) = @_;
581 $logic = uc($logic || $self->{logic});
582 $logic eq 'AND' or $logic eq 'OR' or puke "unknown logic: $logic";
584 my @clauses = @$where;
586 my (@sql_clauses, @all_bind);
587 # need to use while() so can shift() for pairs
589 my $el = shift @clauses;
591 $el = undef if (defined $el and ! length $el);
593 # switch according to kind of $el and get corresponding ($sql, @bind)
594 my ($sql, @bind) = $self->_SWITCH_refkind($el, {
596 # skip empty elements, otherwise get invalid trailing AND stuff
597 ARRAYREF => sub {$self->_recurse_where($el) if @$el},
601 $self->_assert_bindval_matches_bindtype(@b);
605 HASHREF => sub {$self->_recurse_where($el, 'and') if %$el},
607 SCALARREF => sub { ($$el); },
610 # top-level arrayref with scalars, recurse in pairs
611 $self->_recurse_where({$el => shift(@clauses)})
614 UNDEF => sub {puke "Supplying an empty left hand side argument is not supported in array-pairs" },
618 push @sql_clauses, $sql;
619 push @all_bind, @bind;
623 return $self->_join_sql_clauses($logic, \@sql_clauses, \@all_bind);
626 #======================================================================
627 # WHERE: top-level ARRAYREFREF
628 #======================================================================
630 sub _where_ARRAYREFREF {
631 my ($self, $where) = @_;
632 my ($sql, @bind) = @$$where;
633 $self->_assert_bindval_matches_bindtype(@bind);
634 return ($sql, @bind);
637 #======================================================================
638 # WHERE: top-level HASHREF
639 #======================================================================
642 my ($self, $where) = @_;
643 my (@sql_clauses, @all_bind);
645 for my $k (sort keys %$where) {
646 my $v = $where->{$k};
648 # ($k => $v) is either a special unary op or a regular hashpair
649 my ($sql, @bind) = do {
651 # put the operator in canonical form
653 $op = substr $op, 1; # remove initial dash
654 $op =~ s/^\s+|\s+$//g;# remove leading/trailing space
655 $op =~ s/\s+/ /g; # compress whitespace
657 # so that -not_foo works correctly
658 $op =~ s/^not_/NOT /i;
660 $self->_debug("Unary OP(-$op) within hashref, recursing...");
661 my ($s, @b) = $self->_where_unary_op($op, $v);
663 # top level vs nested
664 # we assume that handled unary ops will take care of their ()s
666 List::Util::first {$op =~ $_->{regex}} @{$self->{unary_ops}}
668 ( defined $self->{_nested_func_lhs} and $self->{_nested_func_lhs} eq $k )
674 if (is_literal_value ($v) ) {
675 belch 'Hash-pairs consisting of an empty string with a literal are deprecated, and will be removed in 2.0: use -and => [ $literal ] instead';
678 puke "Supplying an empty left hand side argument is not supported in hash-pairs";
682 my $method = $self->_METHOD_FOR_refkind("_where_hashpair", $v);
683 $self->$method($k, $v);
687 push @sql_clauses, $sql;
688 push @all_bind, @bind;
691 return $self->_join_sql_clauses('and', \@sql_clauses, \@all_bind);
694 sub _where_unary_op {
695 my ($self, $op, $rhs) = @_;
697 # top level special ops are illegal in general
698 # this includes the -ident/-value ops (dual purpose unary and special)
699 puke "Illegal use of top-level '-$op'"
700 if ! defined $self->{_nested_func_lhs} and List::Util::first { $op =~ $_->{regex} } @{$self->{special_ops}};
702 if (my $op_entry = List::Util::first { $op =~ $_->{regex} } @{$self->{unary_ops}}) {
703 my $handler = $op_entry->{handler};
705 if (not ref $handler) {
706 if ($op =~ s/ [_\s]? \d+ $//x ) {
707 belch 'Use of [and|or|nest]_N modifiers is deprecated and will be removed in SQLA v2.0. '
708 . "You probably wanted ...-and => [ -$op => COND1, -$op => COND2 ... ]";
710 return $self->$handler($op, $rhs);
712 elsif (ref $handler eq 'CODE') {
713 return $handler->($self, $op, $rhs);
716 puke "Illegal handler for operator $op - expecting a method name or a coderef";
720 $self->_debug("Generic unary OP: $op - recursing as function");
722 $self->_assert_pass_injection_guard($op);
724 my ($sql, @bind) = $self->_SWITCH_refkind($rhs, {
726 puke "Illegal use of top-level '-$op'"
727 unless defined $self->{_nested_func_lhs};
730 $self->_convert('?'),
731 $self->_bindtype($self->{_nested_func_lhs}, $rhs)
735 $self->_recurse_where($rhs)
739 $sql = sprintf('%s %s',
740 $self->_sqlcase($op),
744 return ($sql, @bind);
747 sub _where_op_ANDOR {
748 my ($self, $op, $v) = @_;
750 $self->_SWITCH_refkind($v, {
752 return $self->_where_ARRAYREF($v, $op);
756 return ($op =~ /^or/i)
757 ? $self->_where_ARRAYREF([ map { $_ => $v->{$_} } (sort keys %$v) ], $op)
758 : $self->_where_HASHREF($v);
762 puke "-$op => \\\$scalar makes little sense, use " .
764 ? '[ \$scalar, \%rest_of_conditions ] instead'
765 : '-and => [ \$scalar, \%rest_of_conditions ] instead'
770 puke "-$op => \\[...] makes little sense, use " .
772 ? '[ \[...], \%rest_of_conditions ] instead'
773 : '-and => [ \[...], \%rest_of_conditions ] instead'
777 SCALAR => sub { # permissively interpreted as SQL
778 puke "-$op => \$value makes little sense, use -bool => \$value instead";
782 puke "-$op => undef not supported";
788 my ($self, $op, $v) = @_;
790 $self->_SWITCH_refkind($v, {
792 SCALAR => sub { # permissively interpreted as SQL
793 belch "literal SQL should be -nest => \\'scalar' "
794 . "instead of -nest => 'scalar' ";
799 puke "-$op => undef not supported";
803 $self->_recurse_where($v);
811 my ($self, $op, $v) = @_;
813 my ($s, @b) = $self->_SWITCH_refkind($v, {
814 SCALAR => sub { # interpreted as SQL column
815 $self->_convert($self->_quote($v));
819 puke "-$op => undef not supported";
823 $self->_recurse_where($v);
827 $s = "(NOT $s)" if $op =~ /^not/i;
832 sub _where_op_IDENT {
834 my ($op, $rhs) = splice @_, -2;
835 if (! defined $rhs or length ref $rhs) {
836 puke "-$op requires a single plain scalar argument (a quotable identifier)";
839 # in case we are called as a top level special op (no '=')
842 $_ = $self->_convert($self->_quote($_)) for ($lhs, $rhs);
850 sub _where_op_VALUE {
852 my ($op, $rhs) = splice @_, -2;
854 # in case we are called as a top level special op (no '=')
858 if (! defined $rhs) {
860 ? $self->_convert($self->_quote($lhs)) . ' IS NULL'
867 (defined $lhs ? $lhs : $self->{_nested_func_lhs}),
874 $self->_convert($self->_quote($lhs)) . ' = ' . $self->_convert('?'),
878 $self->_convert('?'),
884 sub _where_hashpair_ARRAYREF {
885 my ($self, $k, $v) = @_;
888 my @v = @$v; # need copy because of shift below
889 $self->_debug("ARRAY($k) means distribute over elements");
891 # put apart first element if it is an operator (-and, -or)
893 (defined $v[0] && $v[0] =~ /^ - (?: AND|OR ) $/ix)
897 my @distributed = map { {$k => $_} } @v;
900 $self->_debug("OP($op) reinjected into the distributed array");
901 unshift @distributed, $op;
904 my $logic = $op ? substr($op, 1) : '';
906 return $self->_recurse_where(\@distributed, $logic);
909 $self->_debug("empty ARRAY($k) means 0=1");
910 return ($self->{sqlfalse});
914 sub _where_hashpair_HASHREF {
915 my ($self, $k, $v, $logic) = @_;
918 local $self->{_nested_func_lhs} = defined $self->{_nested_func_lhs}
919 ? $self->{_nested_func_lhs}
923 my ($all_sql, @all_bind);
925 for my $orig_op (sort keys %$v) {
926 my $val = $v->{$orig_op};
928 # put the operator in canonical form
931 # FIXME - we need to phase out dash-less ops
932 $op =~ s/^-//; # remove possible initial dash
933 $op =~ s/^\s+|\s+$//g;# remove leading/trailing space
934 $op =~ s/\s+/ /g; # compress whitespace
936 $self->_assert_pass_injection_guard($op);
939 $op =~ s/^is_not/IS NOT/i;
941 # so that -not_foo works correctly
942 $op =~ s/^not_/NOT /i;
944 # another retarded special case: foo => { $op => { -value => undef } }
945 if (ref $val eq 'HASH' and keys %$val == 1 and exists $val->{-value} and ! defined $val->{-value} ) {
951 # CASE: col-value logic modifiers
952 if ($orig_op =~ /^ \- (and|or) $/xi) {
953 ($sql, @bind) = $self->_where_hashpair_HASHREF($k, $val, $1);
955 # CASE: special operators like -in or -between
956 elsif (my $special_op = List::Util::first { $op =~ $_->{regex} } @{$self->{special_ops}}) {
957 my $handler = $special_op->{handler};
959 puke "No handler supplied for special operator $orig_op";
961 elsif (not ref $handler) {
962 ($sql, @bind) = $self->$handler($k, $op, $val);
964 elsif (ref $handler eq 'CODE') {
965 ($sql, @bind) = $handler->($self, $k, $op, $val);
968 puke "Illegal handler for special operator $orig_op - expecting a method name or a coderef";
972 $self->_SWITCH_refkind($val, {
974 ARRAYREF => sub { # CASE: col => {op => \@vals}
975 ($sql, @bind) = $self->_where_field_op_ARRAYREF($k, $op, $val);
978 ARRAYREFREF => sub { # CASE: col => {op => \[$sql, @bind]} (literal SQL with bind)
979 my ($sub_sql, @sub_bind) = @$$val;
980 $self->_assert_bindval_matches_bindtype(@sub_bind);
981 $sql = join ' ', $self->_convert($self->_quote($k)),
982 $self->_sqlcase($op),
987 UNDEF => sub { # CASE: col => {op => undef} : sql "IS (NOT)? NULL"
989 $op =~ /^not$/i ? 'is not' # legacy
990 : $op =~ $self->{equality_op} ? 'is'
991 : $op =~ $self->{like_op} ? belch("Supplying an undefined argument to '@{[ uc $op]}' is deprecated") && 'is'
992 : $op =~ $self->{inequality_op} ? 'is not'
993 : $op =~ $self->{not_like_op} ? belch("Supplying an undefined argument to '@{[ uc $op]}' is deprecated") && 'is not'
994 : puke "unexpected operator '$orig_op' with undef operand";
996 $sql = $self->_quote($k) . $self->_sqlcase(" $is null");
999 FALLBACK => sub { # CASE: col => {op/func => $stuff}
1000 ($sql, @bind) = $self->_where_unary_op($op, $val);
1003 $self->_convert($self->_quote($k)),
1004 $self->{_nested_func_lhs} eq $k ? $sql : "($sql)", # top level vs nested
1010 ($all_sql) = (defined $all_sql and $all_sql) ? $self->_join_sql_clauses($logic, [$all_sql, $sql], []) : $sql;
1011 push @all_bind, @bind;
1013 return ($all_sql, @all_bind);
1016 sub _where_field_IS {
1017 my ($self, $k, $op, $v) = @_;
1019 my ($s) = $self->_SWITCH_refkind($v, {
1022 $self->_convert($self->_quote($k)),
1023 map { $self->_sqlcase($_)} ($op, 'null')
1026 puke "$op can only take undef as argument";
1033 sub _where_field_op_ARRAYREF {
1034 my ($self, $k, $op, $vals) = @_;
1036 my @vals = @$vals; #always work on a copy
1039 $self->_debug(sprintf '%s means multiple elements: [ %s ]',
1041 join(', ', map { defined $_ ? "'$_'" : 'NULL' } @vals ),
1044 # see if the first element is an -and/-or op
1046 if (defined $vals[0] && $vals[0] =~ /^ - (AND|OR) $/ix) {
1051 # a long standing API wart - an attempt to change this behavior during
1052 # the 1.50 series failed *spectacularly*. Warn instead and leave the
1057 (!$logic or $logic eq 'OR')
1059 ($op =~ $self->{inequality_op} or $op =~ $self->{not_like_op})
1062 belch "A multi-element arrayref as an argument to the inequality op '$o' "
1063 . 'is technically equivalent to an always-true 1=1 (you probably wanted '
1064 . "to say ...{ \$inequality_op => [ -and => \@values ] }... instead)"
1068 # distribute $op over each remaining member of @vals, append logic if exists
1069 return $self->_recurse_where([map { {$k => {$op, $_}} } @vals], $logic);
1073 # try to DWIM on equality operators
1075 $op =~ $self->{equality_op} ? $self->{sqlfalse}
1076 : $op =~ $self->{like_op} ? belch("Supplying an empty arrayref to '@{[ uc $op]}' is deprecated") && $self->{sqlfalse}
1077 : $op =~ $self->{inequality_op} ? $self->{sqltrue}
1078 : $op =~ $self->{not_like_op} ? belch("Supplying an empty arrayref to '@{[ uc $op]}' is deprecated") && $self->{sqltrue}
1079 : puke "operator '$op' applied on an empty array (field '$k')";
1084 sub _where_hashpair_SCALARREF {
1085 my ($self, $k, $v) = @_;
1086 $self->_debug("SCALAR($k) means literal SQL: $$v");
1087 my $sql = $self->_quote($k) . " " . $$v;
1091 # literal SQL with bind
1092 sub _where_hashpair_ARRAYREFREF {
1093 my ($self, $k, $v) = @_;
1094 $self->_debug("REF($k) means literal SQL: @${$v}");
1095 my ($sql, @bind) = @$$v;
1096 $self->_assert_bindval_matches_bindtype(@bind);
1097 $sql = $self->_quote($k) . " " . $sql;
1098 return ($sql, @bind );
1101 # literal SQL without bind
1102 sub _where_hashpair_SCALAR {
1103 my ($self, $k, $v) = @_;
1104 $self->_debug("NOREF($k) means simple key=val: $k $self->{cmp} $v");
1105 my $sql = join ' ', $self->_convert($self->_quote($k)),
1106 $self->_sqlcase($self->{cmp}),
1107 $self->_convert('?');
1108 my @bind = $self->_bindtype($k, $v);
1109 return ($sql, @bind);
1113 sub _where_hashpair_UNDEF {
1114 my ($self, $k, $v) = @_;
1115 $self->_debug("UNDEF($k) means IS NULL");
1116 my $sql = $self->_quote($k) . $self->_sqlcase(' is null');
1120 #======================================================================
1121 # WHERE: TOP-LEVEL OTHERS (SCALARREF, SCALAR, UNDEF)
1122 #======================================================================
1125 sub _where_SCALARREF {
1126 my ($self, $where) = @_;
1129 $self->_debug("SCALAR(*top) means literal SQL: $$where");
1135 my ($self, $where) = @_;
1138 $self->_debug("NOREF(*top) means literal SQL: $where");
1149 #======================================================================
1150 # WHERE: BUILTIN SPECIAL OPERATORS (-in, -between)
1151 #======================================================================
1154 sub _where_field_BETWEEN {
1155 my ($self, $k, $op, $vals) = @_;
1157 my ($label, $and, $placeholder);
1158 $label = $self->_convert($self->_quote($k));
1159 $and = ' ' . $self->_sqlcase('and') . ' ';
1160 $placeholder = $self->_convert('?');
1161 $op = $self->_sqlcase($op);
1163 my $invalid_args = "Operator '$op' requires either an arrayref with two defined values or expressions, or a single literal scalarref/arrayref-ref";
1165 my ($clause, @bind) = $self->_SWITCH_refkind($vals, {
1166 ARRAYREFREF => sub {
1167 my ($s, @b) = @$$vals;
1168 $self->_assert_bindval_matches_bindtype(@b);
1175 puke $invalid_args if @$vals != 2;
1177 my (@all_sql, @all_bind);
1178 foreach my $val (@$vals) {
1179 my ($sql, @bind) = $self->_SWITCH_refkind($val, {
1181 return ($placeholder, $self->_bindtype($k, $val) );
1186 ARRAYREFREF => sub {
1187 my ($sql, @bind) = @$$val;
1188 $self->_assert_bindval_matches_bindtype(@bind);
1189 return ($sql, @bind);
1192 my ($func, $arg, @rest) = %$val;
1193 puke "Only simple { -func => arg } functions accepted as sub-arguments to BETWEEN"
1194 if (@rest or $func !~ /^ \- (.+)/x);
1195 $self->_where_unary_op($1 => $arg);
1201 push @all_sql, $sql;
1202 push @all_bind, @bind;
1206 (join $and, @all_sql),
1215 my $sql = "( $label $op $clause )";
1216 return ($sql, @bind)
1220 sub _where_field_IN {
1221 my ($self, $k, $op, $vals) = @_;
1223 # backwards compatibility: if scalar, force into an arrayref
1224 $vals = [$vals] if defined $vals && ! ref $vals;
1226 my ($label) = $self->_convert($self->_quote($k));
1227 my ($placeholder) = $self->_convert('?');
1228 $op = $self->_sqlcase($op);
1230 my ($sql, @bind) = $self->_SWITCH_refkind($vals, {
1231 ARRAYREF => sub { # list of choices
1232 if (@$vals) { # nonempty list
1233 my (@all_sql, @all_bind);
1235 for my $val (@$vals) {
1236 my ($sql, @bind) = $self->_SWITCH_refkind($val, {
1238 return ($placeholder, $val);
1243 ARRAYREFREF => sub {
1244 my ($sql, @bind) = @$$val;
1245 $self->_assert_bindval_matches_bindtype(@bind);
1246 return ($sql, @bind);
1249 my ($func, $arg, @rest) = %$val;
1250 puke "Only simple { -func => arg } functions accepted as sub-arguments to IN"
1251 if (@rest or $func !~ /^ \- (.+)/x);
1252 $self->_where_unary_op($1 => $arg);
1256 'SQL::Abstract before v1.75 used to generate incorrect SQL when the '
1257 . "-$op operator was given an undef-containing list: !!!AUDIT YOUR CODE "
1258 . 'AND DATA!!! (the upcoming Data::Query-based version of SQL::Abstract '
1259 . 'will emit the logically correct SQL instead of raising this exception)'
1263 push @all_sql, $sql;
1264 push @all_bind, @bind;
1268 sprintf('%s %s ( %s )',
1271 join(', ', @all_sql)
1273 $self->_bindtype($k, @all_bind),
1276 else { # empty list: some databases won't understand "IN ()", so DWIM
1277 my $sql = ($op =~ /\bnot\b/i) ? $self->{sqltrue} : $self->{sqlfalse};
1282 SCALARREF => sub { # literal SQL
1283 my $sql = $self->_open_outer_paren($$vals);
1284 return ("$label $op ( $sql )");
1286 ARRAYREFREF => sub { # literal SQL with bind
1287 my ($sql, @bind) = @$$vals;
1288 $self->_assert_bindval_matches_bindtype(@bind);
1289 $sql = $self->_open_outer_paren($sql);
1290 return ("$label $op ( $sql )", @bind);
1294 puke "Argument passed to the '$op' operator can not be undefined";
1298 puke "special op $op requires an arrayref (or scalarref/arrayref-ref)";
1302 return ($sql, @bind);
1305 # Some databases (SQLite) treat col IN (1, 2) different from
1306 # col IN ( (1, 2) ). Use this to strip all outer parens while
1307 # adding them back in the corresponding method
1308 sub _open_outer_paren {
1309 my ($self, $sql) = @_;
1311 while (my ($inner) = $sql =~ /^ \s* \( (.*) \) \s* $/xs) {
1313 # there are closing parens inside, need the heavy duty machinery
1314 # to reevaluate the extraction starting from $sql (full reevaluation)
1315 if ($inner =~ /\)/) {
1316 require Text::Balanced;
1318 my (undef, $remainder) = do {
1319 # idiotic design - writes to $@ but *DOES NOT* throw exceptions
1321 Text::Balanced::extract_bracketed($sql, '()', qr/\s*/);
1324 # the entire expression needs to be a balanced bracketed thing
1325 # (after an extract no remainder sans trailing space)
1326 last if defined $remainder and $remainder =~ /\S/;
1336 #======================================================================
1338 #======================================================================
1341 my ($self, $arg) = @_;
1344 for my $c ($self->_order_by_chunks($arg) ) {
1345 $self->_SWITCH_refkind($c, {
1346 SCALAR => sub { push @sql, $c },
1347 ARRAYREF => sub { push @sql, shift @$c; push @bind, @$c },
1353 $self->_sqlcase(' order by'),
1359 return wantarray ? ($sql, @bind) : $sql;
1362 sub _order_by_chunks {
1363 my ($self, $arg) = @_;
1365 return $self->_SWITCH_refkind($arg, {
1368 map { $self->_order_by_chunks($_ ) } @$arg;
1371 ARRAYREFREF => sub {
1372 my ($s, @b) = @$$arg;
1373 $self->_assert_bindval_matches_bindtype(@b);
1377 SCALAR => sub {$self->_quote($arg)},
1379 UNDEF => sub {return () },
1381 SCALARREF => sub {$$arg}, # literal SQL, no quoting
1384 # get first pair in hash
1385 my ($key, $val, @rest) = %$arg;
1387 return () unless $key;
1389 if (@rest or not $key =~ /^-(desc|asc)/i) {
1390 puke "hash passed to _order_by must have exactly one key (-desc or -asc)";
1396 for my $c ($self->_order_by_chunks($val)) {
1399 $self->_SWITCH_refkind($c, {
1404 ($sql, @bind) = @$c;
1408 $sql = $sql . ' ' . $self->_sqlcase($direction);
1410 push @ret, [ $sql, @bind];
1419 #======================================================================
1420 # DATASOURCE (FOR NOW, JUST PLAIN TABLE OR LIST OF TABLES)
1421 #======================================================================
1426 $self->_SWITCH_refkind($from, {
1427 ARRAYREF => sub {join ', ', map { $self->_quote($_) } @$from;},
1428 SCALAR => sub {$self->_quote($from)},
1429 SCALARREF => sub {$$from},
1434 #======================================================================
1436 #======================================================================
1438 # highly optimized, as it's called way too often
1440 # my ($self, $label) = @_;
1442 return '' unless defined $_[1];
1443 return ${$_[1]} if ref($_[1]) eq 'SCALAR';
1445 $_[0]->{quote_char} or
1446 ($_[0]->_assert_pass_injection_guard($_[1]), return $_[1]);
1448 my $qref = ref $_[0]->{quote_char};
1450 !$qref ? ($_[0]->{quote_char}, $_[0]->{quote_char})
1451 : ($qref eq 'ARRAY') ? @{$_[0]->{quote_char}}
1452 : puke "Unsupported quote_char format: $_[0]->{quote_char}";
1454 my $esc = $_[0]->{escape_char} || $r;
1456 # parts containing * are naturally unquoted
1457 return join($_[0]->{name_sep}||'', map
1458 +( $_ eq '*' ? $_ : do { (my $n = $_) =~ s/(\Q$esc\E|\Q$r\E)/$esc$1/g; $l . $n . $r } ),
1459 ( $_[0]->{name_sep} ? split (/\Q$_[0]->{name_sep}\E/, $_[1] ) : $_[1] )
1464 # Conversion, if applicable
1466 #my ($self, $arg) = @_;
1467 if ($_[0]->{convert}) {
1468 return $_[0]->_sqlcase($_[0]->{convert}) .'(' . $_[1] . ')';
1475 #my ($self, $col, @vals) = @_;
1476 # called often - tighten code
1477 return $_[0]->{bindtype} eq 'columns'
1478 ? map {[$_[1], $_]} @_[2 .. $#_]
1483 # Dies if any element of @bind is not in [colname => value] format
1484 # if bindtype is 'columns'.
1485 sub _assert_bindval_matches_bindtype {
1486 # my ($self, @bind) = @_;
1488 if ($self->{bindtype} eq 'columns') {
1490 if (!defined $_ || ref($_) ne 'ARRAY' || @$_ != 2) {
1491 puke "bindtype 'columns' selected, you need to pass: [column_name => bind_value]"
1497 sub _join_sql_clauses {
1498 my ($self, $logic, $clauses_aref, $bind_aref) = @_;
1500 if (@$clauses_aref > 1) {
1501 my $join = " " . $self->_sqlcase($logic) . " ";
1502 my $sql = '( ' . join($join, @$clauses_aref) . ' )';
1503 return ($sql, @$bind_aref);
1505 elsif (@$clauses_aref) {
1506 return ($clauses_aref->[0], @$bind_aref); # no parentheses
1509 return (); # if no SQL, ignore @$bind_aref
1514 # Fix SQL case, if so requested
1516 # LDNOTE: if $self->{case} is true, then it contains 'lower', so we
1517 # don't touch the argument ... crooked logic, but let's not change it!
1518 return $_[0]->{case} ? $_[1] : uc($_[1]);
1522 #======================================================================
1523 # DISPATCHING FROM REFKIND
1524 #======================================================================
1527 my ($self, $data) = @_;
1529 return 'UNDEF' unless defined $data;
1531 # blessed objects are treated like scalars
1532 my $ref = (Scalar::Util::blessed $data) ? '' : ref $data;
1534 return 'SCALAR' unless $ref;
1537 while ($ref eq 'REF') {
1539 $ref = (Scalar::Util::blessed $data) ? '' : ref $data;
1543 return ($ref||'SCALAR') . ('REF' x $n_steps);
1547 my ($self, $data) = @_;
1548 my @try = ($self->_refkind($data));
1549 push @try, 'SCALAR_or_UNDEF' if $try[0] eq 'SCALAR' || $try[0] eq 'UNDEF';
1550 push @try, 'FALLBACK';
1554 sub _METHOD_FOR_refkind {
1555 my ($self, $meth_prefix, $data) = @_;
1558 for (@{$self->_try_refkind($data)}) {
1559 $method = $self->can($meth_prefix."_".$_)
1563 return $method || puke "cannot dispatch on '$meth_prefix' for ".$self->_refkind($data);
1567 sub _SWITCH_refkind {
1568 my ($self, $data, $dispatch_table) = @_;
1571 for (@{$self->_try_refkind($data)}) {
1572 $coderef = $dispatch_table->{$_}
1576 puke "no dispatch entry for ".$self->_refkind($data)
1585 #======================================================================
1586 # VALUES, GENERATE, AUTOLOAD
1587 #======================================================================
1589 # LDNOTE: original code from nwiger, didn't touch code in that section
1590 # I feel the AUTOLOAD stuff should not be the default, it should
1591 # only be activated on explicit demand by user.
1595 my $data = shift || return;
1596 puke "Argument to ", __PACKAGE__, "->values must be a \\%hash"
1597 unless ref $data eq 'HASH';
1600 foreach my $k (sort keys %$data) {
1601 my $v = $data->{$k};
1602 $self->_SWITCH_refkind($v, {
1604 if ($self->{array_datatypes}) { # array datatype
1605 push @all_bind, $self->_bindtype($k, $v);
1607 else { # literal SQL with bind
1608 my ($sql, @bind) = @$v;
1609 $self->_assert_bindval_matches_bindtype(@bind);
1610 push @all_bind, @bind;
1613 ARRAYREFREF => sub { # literal SQL with bind
1614 my ($sql, @bind) = @${$v};
1615 $self->_assert_bindval_matches_bindtype(@bind);
1616 push @all_bind, @bind;
1618 SCALARREF => sub { # literal SQL without bind
1620 SCALAR_or_UNDEF => sub {
1621 push @all_bind, $self->_bindtype($k, $v);
1632 my(@sql, @sqlq, @sqlv);
1636 if ($ref eq 'HASH') {
1637 for my $k (sort keys %$_) {
1640 my $label = $self->_quote($k);
1641 if ($r eq 'ARRAY') {
1642 # literal SQL with bind
1643 my ($sql, @bind) = @$v;
1644 $self->_assert_bindval_matches_bindtype(@bind);
1645 push @sqlq, "$label = $sql";
1647 } elsif ($r eq 'SCALAR') {
1648 # literal SQL without bind
1649 push @sqlq, "$label = $$v";
1651 push @sqlq, "$label = ?";
1652 push @sqlv, $self->_bindtype($k, $v);
1655 push @sql, $self->_sqlcase('set'), join ', ', @sqlq;
1656 } elsif ($ref eq 'ARRAY') {
1657 # unlike insert(), assume these are ONLY the column names, i.e. for SQL
1660 if ($r eq 'ARRAY') { # literal SQL with bind
1661 my ($sql, @bind) = @$v;
1662 $self->_assert_bindval_matches_bindtype(@bind);
1665 } elsif ($r eq 'SCALAR') { # literal SQL without bind
1666 # embedded literal SQL
1673 push @sql, '(' . join(', ', @sqlq) . ')';
1674 } elsif ($ref eq 'SCALAR') {
1678 # strings get case twiddled
1679 push @sql, $self->_sqlcase($_);
1683 my $sql = join ' ', @sql;
1685 # this is pretty tricky
1686 # if ask for an array, return ($stmt, @bind)
1687 # otherwise, s/?/shift @sqlv/ to put it inline
1689 return ($sql, @sqlv);
1691 1 while $sql =~ s/\?/my $d = shift(@sqlv);
1692 ref $d ? $d->[1] : $d/e;
1701 # This allows us to check for a local, then _form, attr
1703 my($name) = $AUTOLOAD =~ /.*::(.+)/;
1704 return $self->generate($name, @_);
1715 SQL::Abstract - Generate SQL from Perl data structures
1721 my $sql = SQL::Abstract->new;
1723 my($stmt, @bind) = $sql->select($source, \@fields, \%where, $order);
1725 my($stmt, @bind) = $sql->insert($table, \%fieldvals || \@values);
1727 my($stmt, @bind) = $sql->update($table, \%fieldvals, \%where);
1729 my($stmt, @bind) = $sql->delete($table, \%where);
1731 # Then, use these in your DBI statements
1732 my $sth = $dbh->prepare($stmt);
1733 $sth->execute(@bind);
1735 # Just generate the WHERE clause
1736 my($stmt, @bind) = $sql->where(\%where, $order);
1738 # Return values in the same order, for hashed queries
1739 # See PERFORMANCE section for more details
1740 my @bind = $sql->values(\%fieldvals);
1744 This module was inspired by the excellent L<DBIx::Abstract>.
1745 However, in using that module I found that what I really wanted
1746 to do was generate SQL, but still retain complete control over my
1747 statement handles and use the DBI interface. So, I set out to
1748 create an abstract SQL generation module.
1750 While based on the concepts used by L<DBIx::Abstract>, there are
1751 several important differences, especially when it comes to WHERE
1752 clauses. I have modified the concepts used to make the SQL easier
1753 to generate from Perl data structures and, IMO, more intuitive.
1754 The underlying idea is for this module to do what you mean, based
1755 on the data structures you provide it. The big advantage is that
1756 you don't have to modify your code every time your data changes,
1757 as this module figures it out.
1759 To begin with, an SQL INSERT is as easy as just specifying a hash
1760 of C<key=value> pairs:
1763 name => 'Jimbo Bobson',
1764 phone => '123-456-7890',
1765 address => '42 Sister Lane',
1766 city => 'St. Louis',
1767 state => 'Louisiana',
1770 The SQL can then be generated with this:
1772 my($stmt, @bind) = $sql->insert('people', \%data);
1774 Which would give you something like this:
1776 $stmt = "INSERT INTO people
1777 (address, city, name, phone, state)
1778 VALUES (?, ?, ?, ?, ?)";
1779 @bind = ('42 Sister Lane', 'St. Louis', 'Jimbo Bobson',
1780 '123-456-7890', 'Louisiana');
1782 These are then used directly in your DBI code:
1784 my $sth = $dbh->prepare($stmt);
1785 $sth->execute(@bind);
1787 =head2 Inserting and Updating Arrays
1789 If your database has array types (like for example Postgres),
1790 activate the special option C<< array_datatypes => 1 >>
1791 when creating the C<SQL::Abstract> object.
1792 Then you may use an arrayref to insert and update database array types:
1794 my $sql = SQL::Abstract->new(array_datatypes => 1);
1796 planets => [qw/Mercury Venus Earth Mars/]
1799 my($stmt, @bind) = $sql->insert('solar_system', \%data);
1803 $stmt = "INSERT INTO solar_system (planets) VALUES (?)"
1805 @bind = (['Mercury', 'Venus', 'Earth', 'Mars']);
1808 =head2 Inserting and Updating SQL
1810 In order to apply SQL functions to elements of your C<%data> you may
1811 specify a reference to an arrayref for the given hash value. For example,
1812 if you need to execute the Oracle C<to_date> function on a value, you can
1813 say something like this:
1817 date_entered => \[ "to_date(?,'MM/DD/YYYY')", "03/02/2003" ],
1820 The first value in the array is the actual SQL. Any other values are
1821 optional and would be included in the bind values array. This gives
1824 my($stmt, @bind) = $sql->insert('people', \%data);
1826 $stmt = "INSERT INTO people (name, date_entered)
1827 VALUES (?, to_date(?,'MM/DD/YYYY'))";
1828 @bind = ('Bill', '03/02/2003');
1830 An UPDATE is just as easy, all you change is the name of the function:
1832 my($stmt, @bind) = $sql->update('people', \%data);
1834 Notice that your C<%data> isn't touched; the module will generate
1835 the appropriately quirky SQL for you automatically. Usually you'll
1836 want to specify a WHERE clause for your UPDATE, though, which is
1837 where handling C<%where> hashes comes in handy...
1839 =head2 Complex where statements
1841 This module can generate pretty complicated WHERE statements
1842 easily. For example, simple C<key=value> pairs are taken to mean
1843 equality, and if you want to see if a field is within a set
1844 of values, you can use an arrayref. Let's say we wanted to
1845 SELECT some data based on this criteria:
1848 requestor => 'inna',
1849 worker => ['nwiger', 'rcwe', 'sfz'],
1850 status => { '!=', 'completed' }
1853 my($stmt, @bind) = $sql->select('tickets', '*', \%where);
1855 The above would give you something like this:
1857 $stmt = "SELECT * FROM tickets WHERE
1858 ( requestor = ? ) AND ( status != ? )
1859 AND ( worker = ? OR worker = ? OR worker = ? )";
1860 @bind = ('inna', 'completed', 'nwiger', 'rcwe', 'sfz');
1862 Which you could then use in DBI code like so:
1864 my $sth = $dbh->prepare($stmt);
1865 $sth->execute(@bind);
1871 The methods are simple. There's one for every major SQL operation,
1872 and a constructor you use first. The arguments are specified in a
1873 similar order for each method (table, then fields, then a where
1874 clause) to try and simplify things.
1876 =head2 new(option => 'value')
1878 The C<new()> function takes a list of options and values, and returns
1879 a new B<SQL::Abstract> object which can then be used to generate SQL
1880 through the methods below. The options accepted are:
1886 If set to 'lower', then SQL will be generated in all lowercase. By
1887 default SQL is generated in "textbook" case meaning something like:
1889 SELECT a_field FROM a_table WHERE some_field LIKE '%someval%'
1891 Any setting other than 'lower' is ignored.
1895 This determines what the default comparison operator is. By default
1896 it is C<=>, meaning that a hash like this:
1898 %where = (name => 'nwiger', email => 'nate@wiger.org');
1900 Will generate SQL like this:
1902 WHERE name = 'nwiger' AND email = 'nate@wiger.org'
1904 However, you may want loose comparisons by default, so if you set
1905 C<cmp> to C<like> you would get SQL such as:
1907 WHERE name like 'nwiger' AND email like 'nate@wiger.org'
1909 You can also override the comparison on an individual basis - see
1910 the huge section on L</"WHERE CLAUSES"> at the bottom.
1912 =item sqltrue, sqlfalse
1914 Expressions for inserting boolean values within SQL statements.
1915 By default these are C<1=1> and C<1=0>. They are used
1916 by the special operators C<-in> and C<-not_in> for generating
1917 correct SQL even when the argument is an empty array (see below).
1921 This determines the default logical operator for multiple WHERE
1922 statements in arrays or hashes. If absent, the default logic is "or"
1923 for arrays, and "and" for hashes. This means that a WHERE
1927 event_date => {'>=', '2/13/99'},
1928 event_date => {'<=', '4/24/03'},
1931 will generate SQL like this:
1933 WHERE event_date >= '2/13/99' OR event_date <= '4/24/03'
1935 This is probably not what you want given this query, though (look
1936 at the dates). To change the "OR" to an "AND", simply specify:
1938 my $sql = SQL::Abstract->new(logic => 'and');
1940 Which will change the above C<WHERE> to:
1942 WHERE event_date >= '2/13/99' AND event_date <= '4/24/03'
1944 The logic can also be changed locally by inserting
1945 a modifier in front of an arrayref:
1947 @where = (-and => [event_date => {'>=', '2/13/99'},
1948 event_date => {'<=', '4/24/03'} ]);
1950 See the L</"WHERE CLAUSES"> section for explanations.
1954 This will automatically convert comparisons using the specified SQL
1955 function for both column and value. This is mostly used with an argument
1956 of C<upper> or C<lower>, so that the SQL will have the effect of
1957 case-insensitive "searches". For example, this:
1959 $sql = SQL::Abstract->new(convert => 'upper');
1960 %where = (keywords => 'MaKe iT CAse inSeNSItive');
1962 Will turn out the following SQL:
1964 WHERE upper(keywords) like upper('MaKe iT CAse inSeNSItive')
1966 The conversion can be C<upper()>, C<lower()>, or any other SQL function
1967 that can be applied symmetrically to fields (actually B<SQL::Abstract> does
1968 not validate this option; it will just pass through what you specify verbatim).
1972 This is a kludge because many databases suck. For example, you can't
1973 just bind values using DBI's C<execute()> for Oracle C<CLOB> or C<BLOB> fields.
1974 Instead, you have to use C<bind_param()>:
1976 $sth->bind_param(1, 'reg data');
1977 $sth->bind_param(2, $lots, {ora_type => ORA_CLOB});
1979 The problem is, B<SQL::Abstract> will normally just return a C<@bind> array,
1980 which loses track of which field each slot refers to. Fear not.
1982 If you specify C<bindtype> in new, you can determine how C<@bind> is returned.
1983 Currently, you can specify either C<normal> (default) or C<columns>. If you
1984 specify C<columns>, you will get an array that looks like this:
1986 my $sql = SQL::Abstract->new(bindtype => 'columns');
1987 my($stmt, @bind) = $sql->insert(...);
1990 [ 'column1', 'value1' ],
1991 [ 'column2', 'value2' ],
1992 [ 'column3', 'value3' ],
1995 You can then iterate through this manually, using DBI's C<bind_param()>.
1997 $sth->prepare($stmt);
2000 my($col, $data) = @$_;
2001 if ($col eq 'details' || $col eq 'comments') {
2002 $sth->bind_param($i, $data, {ora_type => ORA_CLOB});
2003 } elsif ($col eq 'image') {
2004 $sth->bind_param($i, $data, {ora_type => ORA_BLOB});
2006 $sth->bind_param($i, $data);
2010 $sth->execute; # execute without @bind now
2012 Now, why would you still use B<SQL::Abstract> if you have to do this crap?
2013 Basically, the advantage is still that you don't have to care which fields
2014 are or are not included. You could wrap that above C<for> loop in a simple
2015 sub called C<bind_fields()> or something and reuse it repeatedly. You still
2016 get a layer of abstraction over manual SQL specification.
2018 Note that if you set L</bindtype> to C<columns>, the C<\[ $sql, @bind ]>
2019 construct (see L</Literal SQL with placeholders and bind values (subqueries)>)
2020 will expect the bind values in this format.
2024 This is the character that a table or column name will be quoted
2025 with. By default this is an empty string, but you could set it to
2026 the character C<`>, to generate SQL like this:
2028 SELECT `a_field` FROM `a_table` WHERE `some_field` LIKE '%someval%'
2030 Alternatively, you can supply an array ref of two items, the first being the left
2031 hand quote character, and the second the right hand quote character. For
2032 example, you could supply C<['[',']']> for SQL Server 2000 compliant quotes
2033 that generates SQL like this:
2035 SELECT [a_field] FROM [a_table] WHERE [some_field] LIKE '%someval%'
2037 Quoting is useful if you have tables or columns names that are reserved
2038 words in your database's SQL dialect.
2042 This is the character that will be used to escape L</quote_char>s appearing
2043 in an identifier before it has been quoted.
2045 The parameter default in case of a single L</quote_char> character is the quote
2048 When opening-closing-style quoting is used (L</quote_char> is an arrayref)
2049 this parameter defaults to the B<closing (right)> L</quote_char>. Occurrences
2050 of the B<opening (left)> L</quote_char> within the identifier are currently left
2051 untouched. The default for opening-closing-style quotes may change in future
2052 versions, thus you are B<strongly encouraged> to specify the escape character
2057 This is the character that separates a table and column name. It is
2058 necessary to specify this when the C<quote_char> option is selected,
2059 so that tables and column names can be individually quoted like this:
2061 SELECT `table`.`one_field` FROM `table` WHERE `table`.`other_field` = 1
2063 =item injection_guard
2065 A regular expression C<qr/.../> that is applied to any C<-function> and unquoted
2066 column name specified in a query structure. This is a safety mechanism to avoid
2067 injection attacks when mishandling user input e.g.:
2069 my %condition_as_column_value_pairs = get_values_from_user();
2070 $sqla->select( ... , \%condition_as_column_value_pairs );
2072 If the expression matches an exception is thrown. Note that literal SQL
2073 supplied via C<\'...'> or C<\['...']> is B<not> checked in any way.
2075 Defaults to checking for C<;> and the C<GO> keyword (TransactSQL)
2077 =item array_datatypes
2079 When this option is true, arrayrefs in INSERT or UPDATE are
2080 interpreted as array datatypes and are passed directly
2082 When this option is false, arrayrefs are interpreted
2083 as literal SQL, just like refs to arrayrefs
2084 (but this behavior is for backwards compatibility; when writing
2085 new queries, use the "reference to arrayref" syntax
2091 Takes a reference to a list of "special operators"
2092 to extend the syntax understood by L<SQL::Abstract>.
2093 See section L</"SPECIAL OPERATORS"> for details.
2097 Takes a reference to a list of "unary operators"
2098 to extend the syntax understood by L<SQL::Abstract>.
2099 See section L</"UNARY OPERATORS"> for details.
2105 =head2 insert($table, \@values || \%fieldvals, \%options)
2107 This is the simplest function. You simply give it a table name
2108 and either an arrayref of values or hashref of field/value pairs.
2109 It returns an SQL INSERT statement and a list of bind values.
2110 See the sections on L</"Inserting and Updating Arrays"> and
2111 L</"Inserting and Updating SQL"> for information on how to insert
2112 with those data types.
2114 The optional C<\%options> hash reference may contain additional
2115 options to generate the insert SQL. Currently supported options
2122 Takes either a scalar of raw SQL fields, or an array reference of
2123 field names, and adds on an SQL C<RETURNING> statement at the end.
2124 This allows you to return data generated by the insert statement
2125 (such as row IDs) without performing another C<SELECT> statement.
2126 Note, however, this is not part of the SQL standard and may not
2127 be supported by all database engines.
2131 =head2 update($table, \%fieldvals, \%where, \%options)
2133 This takes a table, hashref of field/value pairs, and an optional
2134 hashref L<WHERE clause|/WHERE CLAUSES>. It returns an SQL UPDATE function and a list
2136 See the sections on L</"Inserting and Updating Arrays"> and
2137 L</"Inserting and Updating SQL"> for information on how to insert
2138 with those data types.
2140 The optional C<\%options> hash reference may contain additional
2141 options to generate the update SQL. Currently supported options
2148 See the C<returning> option to
2149 L<insert|/insert($table, \@values || \%fieldvals, \%options)>.
2153 =head2 select($source, $fields, $where, $order)
2155 This returns a SQL SELECT statement and associated list of bind values, as
2156 specified by the arguments:
2162 Specification of the 'FROM' part of the statement.
2163 The argument can be either a plain scalar (interpreted as a table
2164 name, will be quoted), or an arrayref (interpreted as a list
2165 of table names, joined by commas, quoted), or a scalarref
2166 (literal SQL, not quoted).
2170 Specification of the list of fields to retrieve from
2172 The argument can be either an arrayref (interpreted as a list
2173 of field names, will be joined by commas and quoted), or a
2174 plain scalar (literal SQL, not quoted).
2175 Please observe that this API is not as flexible as that of
2176 the first argument C<$source>, for backwards compatibility reasons.
2180 Optional argument to specify the WHERE part of the query.
2181 The argument is most often a hashref, but can also be
2182 an arrayref or plain scalar --
2183 see section L<WHERE clause|/"WHERE CLAUSES"> for details.
2187 Optional argument to specify the ORDER BY part of the query.
2188 The argument can be a scalar, a hashref or an arrayref
2189 -- see section L<ORDER BY clause|/"ORDER BY CLAUSES">
2195 =head2 delete($table, \%where, \%options)
2197 This takes a table name and optional hashref L<WHERE clause|/WHERE CLAUSES>.
2198 It returns an SQL DELETE statement and list of bind values.
2200 The optional C<\%options> hash reference may contain additional
2201 options to generate the delete SQL. Currently supported options
2208 See the C<returning> option to
2209 L<insert|/insert($table, \@values || \%fieldvals, \%options)>.
2213 =head2 where(\%where, $order)
2215 This is used to generate just the WHERE clause. For example,
2216 if you have an arbitrary data structure and know what the
2217 rest of your SQL is going to look like, but want an easy way
2218 to produce a WHERE clause, use this. It returns an SQL WHERE
2219 clause and list of bind values.
2222 =head2 values(\%data)
2224 This just returns the values from the hash C<%data>, in the same
2225 order that would be returned from any of the other above queries.
2226 Using this allows you to markedly speed up your queries if you
2227 are affecting lots of rows. See below under the L</"PERFORMANCE"> section.
2229 =head2 generate($any, 'number', $of, \@data, $struct, \%types)
2231 Warning: This is an experimental method and subject to change.
2233 This returns arbitrarily generated SQL. It's a really basic shortcut.
2234 It will return two different things, depending on return context:
2236 my($stmt, @bind) = $sql->generate('create table', \$table, \@fields);
2237 my $stmt_and_val = $sql->generate('create table', \$table, \@fields);
2239 These would return the following:
2241 # First calling form
2242 $stmt = "CREATE TABLE test (?, ?)";
2243 @bind = (field1, field2);
2245 # Second calling form
2246 $stmt_and_val = "CREATE TABLE test (field1, field2)";
2248 Depending on what you're trying to do, it's up to you to choose the correct
2249 format. In this example, the second form is what you would want.
2253 $sql->generate('alter session', { nls_date_format => 'MM/YY' });
2257 ALTER SESSION SET nls_date_format = 'MM/YY'
2259 You get the idea. Strings get their case twiddled, but everything
2260 else remains verbatim.
2262 =head1 EXPORTABLE FUNCTIONS
2264 =head2 is_plain_value
2266 Determines if the supplied argument is a plain value as understood by this
2271 =item * The value is C<undef>
2273 =item * The value is a non-reference
2275 =item * The value is an object with stringification overloading
2277 =item * The value is of the form C<< { -value => $anything } >>
2281 On failure returns C<undef>, on success returns a B<scalar> reference
2282 to the original supplied argument.
2288 The stringification overloading detection is rather advanced: it takes
2289 into consideration not only the presence of a C<""> overload, but if that
2290 fails also checks for enabled
2291 L<autogenerated versions of C<"">|overload/Magic Autogeneration>, based
2292 on either C<0+> or C<bool>.
2294 Unfortunately testing in the field indicates that this
2295 detection B<< may tickle a latent bug in perl versions before 5.018 >>,
2296 but only when very large numbers of stringifying objects are involved.
2297 At the time of writing ( Sep 2014 ) there is no clear explanation of
2298 the direct cause, nor is there a manageably small test case that reliably
2299 reproduces the problem.
2301 If you encounter any of the following exceptions in B<random places within
2302 your application stack> - this module may be to blame:
2304 Operation "ne": no method found,
2305 left argument in overloaded package <something>,
2306 right argument in overloaded package <something>
2310 Stub found while resolving method "???" overloading """" in package <something>
2312 If you fall victim to the above - please attempt to reduce the problem
2313 to something that could be sent to the L<SQL::Abstract developers
2314 |DBIx::Class/GETTING HELP/SUPPORT>
2315 (either publicly or privately). As a workaround in the meantime you can
2316 set C<$ENV{SQLA_ISVALUE_IGNORE_AUTOGENERATED_STRINGIFICATION}> to a true
2317 value, which will most likely eliminate your problem (at the expense of
2318 not being able to properly detect exotic forms of stringification).
2320 This notice and environment variable will be removed in a future version,
2321 as soon as the underlying problem is found and a reliable workaround is
2326 =head2 is_literal_value
2328 Determines if the supplied argument is a literal value as understood by this
2333 =item * C<\$sql_string>
2335 =item * C<\[ $sql_string, @bind_values ]>
2339 On failure returns C<undef>, on success returns an B<array> reference
2340 containing the unpacked version of the supplied literal SQL and bind values.
2342 =head1 WHERE CLAUSES
2346 This module uses a variation on the idea from L<DBIx::Abstract>. It
2347 is B<NOT>, repeat I<not> 100% compatible. B<The main logic of this
2348 module is that things in arrays are OR'ed, and things in hashes
2351 The easiest way to explain is to show lots of examples. After
2352 each C<%where> hash shown, it is assumed you used:
2354 my($stmt, @bind) = $sql->where(\%where);
2356 However, note that the C<%where> hash can be used directly in any
2357 of the other functions as well, as described above.
2359 =head2 Key-value pairs
2361 So, let's get started. To begin, a simple hash:
2365 status => 'completed'
2368 Is converted to SQL C<key = val> statements:
2370 $stmt = "WHERE user = ? AND status = ?";
2371 @bind = ('nwiger', 'completed');
2373 One common thing I end up doing is having a list of values that
2374 a field can be in. To do this, simply specify a list inside of
2379 status => ['assigned', 'in-progress', 'pending'];
2382 This simple code will create the following:
2384 $stmt = "WHERE user = ? AND ( status = ? OR status = ? OR status = ? )";
2385 @bind = ('nwiger', 'assigned', 'in-progress', 'pending');
2387 A field associated to an empty arrayref will be considered a
2388 logical false and will generate 0=1.
2390 =head2 Tests for NULL values
2392 If the value part is C<undef> then this is converted to SQL <IS NULL>
2401 $stmt = "WHERE user = ? AND status IS NULL";
2404 To test if a column IS NOT NULL:
2408 status => { '!=', undef },
2411 =head2 Specific comparison operators
2413 If you want to specify a different type of operator for your comparison,
2414 you can use a hashref for a given column:
2418 status => { '!=', 'completed' }
2421 Which would generate:
2423 $stmt = "WHERE user = ? AND status != ?";
2424 @bind = ('nwiger', 'completed');
2426 To test against multiple values, just enclose the values in an arrayref:
2428 status => { '=', ['assigned', 'in-progress', 'pending'] };
2430 Which would give you:
2432 "WHERE status = ? OR status = ? OR status = ?"
2435 The hashref can also contain multiple pairs, in which case it is expanded
2436 into an C<AND> of its elements:
2440 status => { '!=', 'completed', -not_like => 'pending%' }
2443 # Or more dynamically, like from a form
2444 $where{user} = 'nwiger';
2445 $where{status}{'!='} = 'completed';
2446 $where{status}{'-not_like'} = 'pending%';
2448 # Both generate this
2449 $stmt = "WHERE user = ? AND status != ? AND status NOT LIKE ?";
2450 @bind = ('nwiger', 'completed', 'pending%');
2453 To get an OR instead, you can combine it with the arrayref idea:
2457 priority => [ { '=', 2 }, { '>', 5 } ]
2460 Which would generate:
2462 $stmt = "WHERE ( priority = ? OR priority > ? ) AND user = ?";
2463 @bind = ('2', '5', 'nwiger');
2465 If you want to include literal SQL (with or without bind values), just use a
2466 scalar reference or reference to an arrayref as the value:
2469 date_entered => { '>' => \["to_date(?, 'MM/DD/YYYY')", "11/26/2008"] },
2470 date_expires => { '<' => \"now()" }
2473 Which would generate:
2475 $stmt = "WHERE date_entered > to_date(?, 'MM/DD/YYYY') AND date_expires < now()";
2476 @bind = ('11/26/2008');
2479 =head2 Logic and nesting operators
2481 In the example above,
2482 there is a subtle trap if you want to say something like
2483 this (notice the C<AND>):
2485 WHERE priority != ? AND priority != ?
2487 Because, in Perl you I<can't> do this:
2489 priority => { '!=' => 2, '!=' => 1 }
2491 As the second C<!=> key will obliterate the first. The solution
2492 is to use the special C<-modifier> form inside an arrayref:
2494 priority => [ -and => {'!=', 2},
2498 Normally, these would be joined by C<OR>, but the modifier tells it
2499 to use C<AND> instead. (Hint: You can use this in conjunction with the
2500 C<logic> option to C<new()> in order to change the way your queries
2501 work by default.) B<Important:> Note that the C<-modifier> goes
2502 B<INSIDE> the arrayref, as an extra first element. This will
2503 B<NOT> do what you think it might:
2505 priority => -and => [{'!=', 2}, {'!=', 1}] # WRONG!
2507 Here is a quick list of equivalencies, since there is some overlap:
2510 status => {'!=', 'completed', 'not like', 'pending%' }
2511 status => [ -and => {'!=', 'completed'}, {'not like', 'pending%'}]
2514 status => {'=', ['assigned', 'in-progress']}
2515 status => [ -or => {'=', 'assigned'}, {'=', 'in-progress'}]
2516 status => [ {'=', 'assigned'}, {'=', 'in-progress'} ]
2520 =head2 Special operators: IN, BETWEEN, etc.
2522 You can also use the hashref format to compare a list of fields using the
2523 C<IN> comparison operator, by specifying the list as an arrayref:
2526 status => 'completed',
2527 reportid => { -in => [567, 2335, 2] }
2530 Which would generate:
2532 $stmt = "WHERE status = ? AND reportid IN (?,?,?)";
2533 @bind = ('completed', '567', '2335', '2');
2535 The reverse operator C<-not_in> generates SQL C<NOT IN> and is used in
2538 If the argument to C<-in> is an empty array, 'sqlfalse' is generated
2539 (by default: C<1=0>). Similarly, C<< -not_in => [] >> generates
2540 'sqltrue' (by default: C<1=1>).
2542 In addition to the array you can supply a chunk of literal sql or
2543 literal sql with bind:
2546 customer => { -in => \[
2547 'SELECT cust_id FROM cust WHERE balance > ?',
2550 status => { -in => \'SELECT status_codes FROM states' },
2556 customer IN ( SELECT cust_id FROM cust WHERE balance > ? )
2557 AND status IN ( SELECT status_codes FROM states )
2561 Finally, if the argument to C<-in> is not a reference, it will be
2562 treated as a single-element array.
2564 Another pair of operators is C<-between> and C<-not_between>,
2565 used with an arrayref of two values:
2569 completion_date => {
2570 -not_between => ['2002-10-01', '2003-02-06']
2576 WHERE user = ? AND completion_date NOT BETWEEN ( ? AND ? )
2578 Just like with C<-in> all plausible combinations of literal SQL
2582 start0 => { -between => [ 1, 2 ] },
2583 start1 => { -between => \["? AND ?", 1, 2] },
2584 start2 => { -between => \"lower(x) AND upper(y)" },
2585 start3 => { -between => [
2587 \["upper(?)", 'stuff' ],
2594 ( start0 BETWEEN ? AND ? )
2595 AND ( start1 BETWEEN ? AND ? )
2596 AND ( start2 BETWEEN lower(x) AND upper(y) )
2597 AND ( start3 BETWEEN lower(x) AND upper(?) )
2599 @bind = (1, 2, 1, 2, 'stuff');
2602 These are the two builtin "special operators"; but the
2603 list can be expanded: see section L</"SPECIAL OPERATORS"> below.
2605 =head2 Unary operators: bool
2607 If you wish to test against boolean columns or functions within your
2608 database you can use the C<-bool> and C<-not_bool> operators. For
2609 example to test the column C<is_user> being true and the column
2610 C<is_enabled> being false you would use:-
2614 -not_bool => 'is_enabled',
2619 WHERE is_user AND NOT is_enabled
2621 If a more complex combination is required, testing more conditions,
2622 then you should use the and/or operators:-
2627 -not_bool => { two=> { -rlike => 'bar' } },
2628 -not_bool => { three => [ { '=', 2 }, { '>', 5 } ] },
2639 (NOT ( three = ? OR three > ? ))
2642 =head2 Nested conditions, -and/-or prefixes
2644 So far, we've seen how multiple conditions are joined with a top-level
2645 C<AND>. We can change this by putting the different conditions we want in
2646 hashes and then putting those hashes in an array. For example:
2651 status => { -like => ['pending%', 'dispatched'] },
2655 status => 'unassigned',
2659 This data structure would create the following:
2661 $stmt = "WHERE ( user = ? AND ( status LIKE ? OR status LIKE ? ) )
2662 OR ( user = ? AND status = ? ) )";
2663 @bind = ('nwiger', 'pending', 'dispatched', 'robot', 'unassigned');
2666 Clauses in hashrefs or arrayrefs can be prefixed with an C<-and> or C<-or>
2667 to change the logic inside:
2673 -and => [ workhrs => {'>', 20}, geo => 'ASIA' ],
2674 -or => { workhrs => {'<', 50}, geo => 'EURO' },
2681 $stmt = "WHERE ( user = ?
2682 AND ( ( workhrs > ? AND geo = ? )
2683 OR ( workhrs < ? OR geo = ? ) ) )";
2684 @bind = ('nwiger', '20', 'ASIA', '50', 'EURO');
2686 =head3 Algebraic inconsistency, for historical reasons
2688 C<Important note>: when connecting several conditions, the C<-and->|C<-or>
2689 operator goes C<outside> of the nested structure; whereas when connecting
2690 several constraints on one column, the C<-and> operator goes
2691 C<inside> the arrayref. Here is an example combining both features:
2694 -and => [a => 1, b => 2],
2695 -or => [c => 3, d => 4],
2696 e => [-and => {-like => 'foo%'}, {-like => '%bar'} ]
2701 WHERE ( ( ( a = ? AND b = ? )
2702 OR ( c = ? OR d = ? )
2703 OR ( e LIKE ? AND e LIKE ? ) ) )
2705 This difference in syntax is unfortunate but must be preserved for
2706 historical reasons. So be careful: the two examples below would
2707 seem algebraically equivalent, but they are not
2710 { -like => 'foo%' },
2711 { -like => '%bar' },
2713 # yields: WHERE ( ( col LIKE ? AND col LIKE ? ) )
2716 { col => { -like => 'foo%' } },
2717 { col => { -like => '%bar' } },
2719 # yields: WHERE ( ( col LIKE ? OR col LIKE ? ) )
2722 =head2 Literal SQL and value type operators
2724 The basic premise of SQL::Abstract is that in WHERE specifications the "left
2725 side" is a column name and the "right side" is a value (normally rendered as
2726 a placeholder). This holds true for both hashrefs and arrayref pairs as you
2727 see in the L</WHERE CLAUSES> examples above. Sometimes it is necessary to
2728 alter this behavior. There are several ways of doing so.
2732 This is a virtual operator that signals the string to its right side is an
2733 identifier (a column name) and not a value. For example to compare two
2734 columns you would write:
2737 priority => { '<', 2 },
2738 requestor => { -ident => 'submitter' },
2743 $stmt = "WHERE priority < ? AND requestor = submitter";
2746 If you are maintaining legacy code you may see a different construct as
2747 described in L</Deprecated usage of Literal SQL>, please use C<-ident> in new
2752 This is a virtual operator that signals that the construct to its right side
2753 is a value to be passed to DBI. This is for example necessary when you want
2754 to write a where clause against an array (for RDBMS that support such
2755 datatypes). For example:
2758 array => { -value => [1, 2, 3] }
2763 $stmt = 'WHERE array = ?';
2764 @bind = ([1, 2, 3]);
2766 Note that if you were to simply say:
2772 the result would probably not be what you wanted:
2774 $stmt = 'WHERE array = ? OR array = ? OR array = ?';
2779 Finally, sometimes only literal SQL will do. To include a random snippet
2780 of SQL verbatim, you specify it as a scalar reference. Consider this only
2781 as a last resort. Usually there is a better way. For example:
2784 priority => { '<', 2 },
2785 requestor => { -in => \'(SELECT name FROM hitmen)' },
2790 $stmt = "WHERE priority < ? AND requestor IN (SELECT name FROM hitmen)"
2793 Note that in this example, you only get one bind parameter back, since
2794 the verbatim SQL is passed as part of the statement.
2798 Never use untrusted input as a literal SQL argument - this is a massive
2799 security risk (there is no way to check literal snippets for SQL
2800 injections and other nastyness). If you need to deal with untrusted input
2801 use literal SQL with placeholders as described next.
2803 =head3 Literal SQL with placeholders and bind values (subqueries)
2805 If the literal SQL to be inserted has placeholders and bind values,
2806 use a reference to an arrayref (yes this is a double reference --
2807 not so common, but perfectly legal Perl). For example, to find a date
2808 in Postgres you can use something like this:
2811 date_column => \[ "= date '2008-09-30' - ?::integer", 10 ]
2816 $stmt = "WHERE ( date_column = date '2008-09-30' - ?::integer )"
2819 Note that you must pass the bind values in the same format as they are returned
2820 by L<where|/where(\%where, $order)>. This means that if you set L</bindtype>
2821 to C<columns>, you must provide the bind values in the
2822 C<< [ column_meta => value ] >> format, where C<column_meta> is an opaque
2823 scalar value; most commonly the column name, but you can use any scalar value
2824 (including references and blessed references), L<SQL::Abstract> will simply
2825 pass it through intact. So if C<bindtype> is set to C<columns> the above
2826 example will look like:
2829 date_column => \[ "= date '2008-09-30' - ?::integer", [ {} => 10 ] ]
2832 Literal SQL is especially useful for nesting parenthesized clauses in the
2833 main SQL query. Here is a first example:
2835 my ($sub_stmt, @sub_bind) = ("SELECT c1 FROM t1 WHERE c2 < ? AND c3 LIKE ?",
2839 bar => \["IN ($sub_stmt)" => @sub_bind],
2844 $stmt = "WHERE (foo = ? AND bar IN (SELECT c1 FROM t1
2845 WHERE c2 < ? AND c3 LIKE ?))";
2846 @bind = (1234, 100, "foo%");
2848 Other subquery operators, like for example C<"E<gt> ALL"> or C<"NOT IN">,
2849 are expressed in the same way. Of course the C<$sub_stmt> and
2850 its associated bind values can be generated through a former call
2853 my ($sub_stmt, @sub_bind)
2854 = $sql->select("t1", "c1", {c2 => {"<" => 100},
2855 c3 => {-like => "foo%"}});
2858 bar => \["> ALL ($sub_stmt)" => @sub_bind],
2861 In the examples above, the subquery was used as an operator on a column;
2862 but the same principle also applies for a clause within the main C<%where>
2863 hash, like an EXISTS subquery:
2865 my ($sub_stmt, @sub_bind)
2866 = $sql->select("t1", "*", {c1 => 1, c2 => \"> t0.c0"});
2867 my %where = ( -and => [
2869 \["EXISTS ($sub_stmt)" => @sub_bind],
2874 $stmt = "WHERE (foo = ? AND EXISTS (SELECT * FROM t1
2875 WHERE c1 = ? AND c2 > t0.c0))";
2879 Observe that the condition on C<c2> in the subquery refers to
2880 column C<t0.c0> of the main query: this is I<not> a bind
2881 value, so we have to express it through a scalar ref.
2882 Writing C<< c2 => {">" => "t0.c0"} >> would have generated
2883 C<< c2 > ? >> with bind value C<"t0.c0"> ... not exactly
2884 what we wanted here.
2886 Finally, here is an example where a subquery is used
2887 for expressing unary negation:
2889 my ($sub_stmt, @sub_bind)
2890 = $sql->where({age => [{"<" => 10}, {">" => 20}]});
2891 $sub_stmt =~ s/^ where //i; # don't want "WHERE" in the subclause
2893 lname => {like => '%son%'},
2894 \["NOT ($sub_stmt)" => @sub_bind],
2899 $stmt = "lname LIKE ? AND NOT ( age < ? OR age > ? )"
2900 @bind = ('%son%', 10, 20)
2902 =head3 Deprecated usage of Literal SQL
2904 Below are some examples of archaic use of literal SQL. It is shown only as
2905 reference for those who deal with legacy code. Each example has a much
2906 better, cleaner and safer alternative that users should opt for in new code.
2912 my %where = ( requestor => \'IS NOT NULL' )
2914 $stmt = "WHERE requestor IS NOT NULL"
2916 This used to be the way of generating NULL comparisons, before the handling
2917 of C<undef> got formalized. For new code please use the superior syntax as
2918 described in L</Tests for NULL values>.
2922 my %where = ( requestor => \'= submitter' )
2924 $stmt = "WHERE requestor = submitter"
2926 This used to be the only way to compare columns. Use the superior L</-ident>
2927 method for all new code. For example an identifier declared in such a way
2928 will be properly quoted if L</quote_char> is properly set, while the legacy
2929 form will remain as supplied.
2933 my %where = ( is_ready => \"", completed => { '>', '2012-12-21' } )
2935 $stmt = "WHERE completed > ? AND is_ready"
2936 @bind = ('2012-12-21')
2938 Using an empty string literal used to be the only way to express a boolean.
2939 For all new code please use the much more readable
2940 L<-bool|/Unary operators: bool> operator.
2946 These pages could go on for a while, since the nesting of the data
2947 structures this module can handle are pretty much unlimited (the
2948 module implements the C<WHERE> expansion as a recursive function
2949 internally). Your best bet is to "play around" with the module a
2950 little to see how the data structures behave, and choose the best
2951 format for your data based on that.
2953 And of course, all the values above will probably be replaced with
2954 variables gotten from forms or the command line. After all, if you
2955 knew everything ahead of time, you wouldn't have to worry about
2956 dynamically-generating SQL and could just hardwire it into your
2959 =head1 ORDER BY CLAUSES
2961 Some functions take an order by clause. This can either be a scalar (just a
2962 column name), a hashref of C<< { -desc => 'col' } >> or C<< { -asc => 'col' }
2963 >>, a scalarref, an arrayref-ref, or an arrayref of any of the previous
2966 Given | Will Generate
2967 ---------------------------------------------------------------
2969 'colA' | ORDER BY colA
2971 [qw/colA colB/] | ORDER BY colA, colB
2973 {-asc => 'colA'} | ORDER BY colA ASC
2975 {-desc => 'colB'} | ORDER BY colB DESC
2977 ['colA', {-asc => 'colB'}] | ORDER BY colA, colB ASC
2979 { -asc => [qw/colA colB/] } | ORDER BY colA ASC, colB ASC
2981 \'colA DESC' | ORDER BY colA DESC
2983 \[ 'FUNC(colA, ?)', $x ] | ORDER BY FUNC(colA, ?)
2984 | /* ...with $x bound to ? */
2987 { -asc => 'colA' }, | colA ASC,
2988 { -desc => [qw/colB/] }, | colB DESC,
2989 { -asc => [qw/colC colD/] },| colC ASC, colD ASC,
2990 \'colE DESC', | colE DESC,
2991 \[ 'FUNC(colF, ?)', $x ], | FUNC(colF, ?)
2992 ] | /* ...with $x bound to ? */
2993 ===============================================================
2997 =head1 SPECIAL OPERATORS
2999 my $sqlmaker = SQL::Abstract->new(special_ops => [
3003 my ($self, $field, $op, $arg) = @_;
3009 handler => 'method_name',
3013 A "special operator" is a SQL syntactic clause that can be
3014 applied to a field, instead of a usual binary operator.
3017 WHERE field IN (?, ?, ?)
3018 WHERE field BETWEEN ? AND ?
3019 WHERE MATCH(field) AGAINST (?, ?)
3021 Special operators IN and BETWEEN are fairly standard and therefore
3022 are builtin within C<SQL::Abstract> (as the overridable methods
3023 C<_where_field_IN> and C<_where_field_BETWEEN>). For other operators,
3024 like the MATCH .. AGAINST example above which is specific to MySQL,
3025 you can write your own operator handlers - supply a C<special_ops>
3026 argument to the C<new> method. That argument takes an arrayref of
3027 operator definitions; each operator definition is a hashref with two
3034 the regular expression to match the operator
3038 Either a coderef or a plain scalar method name. In both cases
3039 the expected return is C<< ($sql, @bind) >>.
3041 When supplied with a method name, it is simply called on the
3042 L<SQL::Abstract> object as:
3044 $self->$method_name($field, $op, $arg)
3048 $field is the LHS of the operator
3049 $op is the part that matched the handler regex
3052 When supplied with a coderef, it is called as:
3054 $coderef->($self, $field, $op, $arg)
3059 For example, here is an implementation
3060 of the MATCH .. AGAINST syntax for MySQL
3062 my $sqlmaker = SQL::Abstract->new(special_ops => [
3064 # special op for MySql MATCH (field) AGAINST(word1, word2, ...)
3065 {regex => qr/^match$/i,
3067 my ($self, $field, $op, $arg) = @_;
3068 $arg = [$arg] if not ref $arg;
3069 my $label = $self->_quote($field);
3070 my ($placeholder) = $self->_convert('?');
3071 my $placeholders = join ", ", (($placeholder) x @$arg);
3072 my $sql = $self->_sqlcase('match') . " ($label) "
3073 . $self->_sqlcase('against') . " ($placeholders) ";
3074 my @bind = $self->_bindtype($field, @$arg);
3075 return ($sql, @bind);
3082 =head1 UNARY OPERATORS
3084 my $sqlmaker = SQL::Abstract->new(unary_ops => [
3088 my ($self, $op, $arg) = @_;
3094 handler => 'method_name',
3098 A "unary operator" is a SQL syntactic clause that can be
3099 applied to a field - the operator goes before the field
3101 You can write your own operator handlers - supply a C<unary_ops>
3102 argument to the C<new> method. That argument takes an arrayref of
3103 operator definitions; each operator definition is a hashref with two
3110 the regular expression to match the operator
3114 Either a coderef or a plain scalar method name. In both cases
3115 the expected return is C<< $sql >>.
3117 When supplied with a method name, it is simply called on the
3118 L<SQL::Abstract> object as:
3120 $self->$method_name($op, $arg)
3124 $op is the part that matched the handler regex
3125 $arg is the RHS or argument of the operator
3127 When supplied with a coderef, it is called as:
3129 $coderef->($self, $op, $arg)
3137 Thanks to some benchmarking by Mark Stosberg, it turns out that
3138 this module is many orders of magnitude faster than using C<DBIx::Abstract>.
3139 I must admit this wasn't an intentional design issue, but it's a
3140 byproduct of the fact that you get to control your C<DBI> handles
3143 To maximize performance, use a code snippet like the following:
3145 # prepare a statement handle using the first row
3146 # and then reuse it for the rest of the rows
3148 for my $href (@array_of_hashrefs) {
3149 $stmt ||= $sql->insert('table', $href);
3150 $sth ||= $dbh->prepare($stmt);
3151 $sth->execute($sql->values($href));
3154 The reason this works is because the keys in your C<$href> are sorted
3155 internally by B<SQL::Abstract>. Thus, as long as your data retains
3156 the same structure, you only have to generate the SQL the first time
3157 around. On subsequent queries, simply use the C<values> function provided
3158 by this module to return your values in the correct order.
3160 However this depends on the values having the same type - if, for
3161 example, the values of a where clause may either have values
3162 (resulting in sql of the form C<column = ?> with a single bind
3163 value), or alternatively the values might be C<undef> (resulting in
3164 sql of the form C<column IS NULL> with no bind value) then the
3165 caching technique suggested will not work.
3169 If you use my C<CGI::FormBuilder> module at all, you'll hopefully
3170 really like this part (I do, at least). Building up a complex query
3171 can be as simple as the following:
3178 use CGI::FormBuilder;
3181 my $form = CGI::FormBuilder->new(...);
3182 my $sql = SQL::Abstract->new;
3184 if ($form->submitted) {
3185 my $field = $form->field;
3186 my $id = delete $field->{id};
3187 my($stmt, @bind) = $sql->update('table', $field, {id => $id});
3190 Of course, you would still have to connect using C<DBI> to run the
3191 query, but the point is that if you make your form look like your
3192 table, the actual query script can be extremely simplistic.
3194 If you're B<REALLY> lazy (I am), check out C<HTML::QuickTable> for
3195 a fast interface to returning and formatting data. I frequently
3196 use these three modules together to write complex database query
3197 apps in under 50 lines.
3199 =head1 HOW TO CONTRIBUTE
3201 Contributions are always welcome, in all usable forms (we especially
3202 welcome documentation improvements). The delivery methods include git-
3203 or unified-diff formatted patches, GitHub pull requests, or plain bug
3204 reports either via RT or the Mailing list. Contributors are generally
3205 granted full access to the official repository after their first several
3206 patches pass successful review.
3208 This project is maintained in a git repository. The code and related tools are
3209 accessible at the following locations:
3213 =item * Official repo: L<git://git.shadowcat.co.uk/dbsrgits/SQL-Abstract.git>
3215 =item * Official gitweb: L<http://git.shadowcat.co.uk/gitweb/gitweb.cgi?p=dbsrgits/SQL-Abstract.git>
3217 =item * GitHub mirror: L<https://github.com/dbsrgits/sql-abstract>
3219 =item * Authorized committers: L<ssh://dbsrgits@git.shadowcat.co.uk/SQL-Abstract.git>
3225 Version 1.50 was a major internal refactoring of C<SQL::Abstract>.
3226 Great care has been taken to preserve the I<published> behavior
3227 documented in previous versions in the 1.* family; however,
3228 some features that were previously undocumented, or behaved
3229 differently from the documentation, had to be changed in order
3230 to clarify the semantics. Hence, client code that was relying
3231 on some dark areas of C<SQL::Abstract> v1.*
3232 B<might behave differently> in v1.50.
3234 The main changes are:
3240 support for literal SQL through the C<< \ [ $sql, @bind ] >> syntax.
3244 support for the { operator => \"..." } construct (to embed literal SQL)
3248 support for the { operator => \["...", @bind] } construct (to embed literal SQL with bind values)
3252 optional support for L<array datatypes|/"Inserting and Updating Arrays">
3256 defensive programming: check arguments
3260 fixed bug with global logic, which was previously implemented
3261 through global variables yielding side-effects. Prior versions would
3262 interpret C<< [ {cond1, cond2}, [cond3, cond4] ] >>
3263 as C<< "(cond1 AND cond2) OR (cond3 AND cond4)" >>.
3264 Now this is interpreted
3265 as C<< "(cond1 AND cond2) OR (cond3 OR cond4)" >>.
3270 fixed semantics of _bindtype on array args
3274 dropped the C<_anoncopy> of the %where tree. No longer necessary,
3275 we just avoid shifting arrays within that tree.
3279 dropped the C<_modlogic> function
3283 =head1 ACKNOWLEDGEMENTS
3285 There are a number of individuals that have really helped out with
3286 this module. Unfortunately, most of them submitted bugs via CPAN
3287 so I have no idea who they are! But the people I do know are:
3289 Ash Berlin (order_by hash term support)
3290 Matt Trout (DBIx::Class support)
3291 Mark Stosberg (benchmarking)
3292 Chas Owens (initial "IN" operator support)
3293 Philip Collins (per-field SQL functions)
3294 Eric Kolve (hashref "AND" support)
3295 Mike Fragassi (enhancements to "BETWEEN" and "LIKE")
3296 Dan Kubb (support for "quote_char" and "name_sep")
3297 Guillermo Roditi (patch to cleanup "IN" and "BETWEEN", fix and tests for _order_by)
3298 Laurent Dami (internal refactoring, extensible list of special operators, literal SQL)
3299 Norbert Buchmuller (support for literal SQL in hashpair, misc. fixes & tests)
3300 Peter Rabbitson (rewrite of SQLA::Test, misc. fixes & tests)
3301 Oliver Charles (support for "RETURNING" after "INSERT")
3307 L<DBIx::Class>, L<DBIx::Abstract>, L<CGI::FormBuilder>, L<HTML::QuickTable>.
3311 Copyright (c) 2001-2007 Nathan Wiger <nwiger@cpan.org>. All Rights Reserved.
3313 This module is actively maintained by Matt Trout <mst@shadowcatsystems.co.uk>
3315 For support, your best bet is to try the C<DBIx::Class> users mailing list.
3316 While not an official support venue, C<DBIx::Class> makes heavy use of
3317 C<SQL::Abstract>, and as such list members there are very familiar with
3318 how to create queries.
3322 This module is free software; you may copy this under the same
3323 terms as perl itself (either the GNU General Public License or
3324 the Artistic License)