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 = '1.84';
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 $fields = shift || '*';
467 my($where_sql, @bind) = $self->where($where, $order);
469 my $f = (ref $fields eq 'ARRAY') ? join ', ', map { $self->_quote($_) } @$fields
471 my $sql = join(' ', $self->_sqlcase('select'), $f,
472 $self->_sqlcase('from'), $table)
475 return wantarray ? ($sql, @bind) : $sql;
478 #======================================================================
480 #======================================================================
485 my $table = $self->_table(shift);
489 my($where_sql, @bind) = $self->where($where);
490 my $sql = $self->_sqlcase('delete from') . " $table" . $where_sql;
492 if ($options->{returning}) {
493 my ($returning_sql, @returning_bind) = $self->_delete_returning($options);
494 $sql .= $returning_sql;
495 push @bind, @returning_bind;
498 return wantarray ? ($sql, @bind) : $sql;
501 # So that subclasses can override DELETE ... RETURNING separately from
503 sub _delete_returning { shift->_returning(@_) }
507 #======================================================================
509 #======================================================================
513 # Finally, a separate routine just to handle WHERE clauses
515 my ($self, $where, $order) = @_;
518 my ($sql, @bind) = $self->_recurse_where($where);
519 $sql = $sql ? $self->_sqlcase(' where ') . "( $sql )" : '';
523 my ($order_sql, @order_bind) = $self->_order_by($order);
525 push @bind, @order_bind;
528 return wantarray ? ($sql, @bind) : $sql;
533 my ($self, $where, $logic) = @_;
535 # dispatch on appropriate method according to refkind of $where
536 my $method = $self->_METHOD_FOR_refkind("_where", $where);
538 my ($sql, @bind) = $self->$method($where, $logic);
540 # DBIx::Class used to call _recurse_where in scalar context
541 # something else might too...
543 return ($sql, @bind);
546 belch "Calling _recurse_where in scalar context is deprecated and will go away before 2.0";
553 #======================================================================
554 # WHERE: top-level ARRAYREF
555 #======================================================================
558 sub _where_ARRAYREF {
559 my ($self, $where, $logic) = @_;
561 $logic = uc($logic || $self->{logic});
562 $logic eq 'AND' or $logic eq 'OR' or puke "unknown logic: $logic";
564 my @clauses = @$where;
566 my (@sql_clauses, @all_bind);
567 # need to use while() so can shift() for pairs
569 my $el = shift @clauses;
571 $el = undef if (defined $el and ! length $el);
573 # switch according to kind of $el and get corresponding ($sql, @bind)
574 my ($sql, @bind) = $self->_SWITCH_refkind($el, {
576 # skip empty elements, otherwise get invalid trailing AND stuff
577 ARRAYREF => sub {$self->_recurse_where($el) if @$el},
581 $self->_assert_bindval_matches_bindtype(@b);
585 HASHREF => sub {$self->_recurse_where($el, 'and') if %$el},
587 SCALARREF => sub { ($$el); },
590 # top-level arrayref with scalars, recurse in pairs
591 $self->_recurse_where({$el => shift(@clauses)})
594 UNDEF => sub {puke "Supplying an empty left hand side argument is not supported in array-pairs" },
598 push @sql_clauses, $sql;
599 push @all_bind, @bind;
603 return $self->_join_sql_clauses($logic, \@sql_clauses, \@all_bind);
606 #======================================================================
607 # WHERE: top-level ARRAYREFREF
608 #======================================================================
610 sub _where_ARRAYREFREF {
611 my ($self, $where) = @_;
612 my ($sql, @bind) = @$$where;
613 $self->_assert_bindval_matches_bindtype(@bind);
614 return ($sql, @bind);
617 #======================================================================
618 # WHERE: top-level HASHREF
619 #======================================================================
622 my ($self, $where) = @_;
623 my (@sql_clauses, @all_bind);
625 for my $k (sort keys %$where) {
626 my $v = $where->{$k};
628 # ($k => $v) is either a special unary op or a regular hashpair
629 my ($sql, @bind) = do {
631 # put the operator in canonical form
633 $op = substr $op, 1; # remove initial dash
634 $op =~ s/^\s+|\s+$//g;# remove leading/trailing space
635 $op =~ s/\s+/ /g; # compress whitespace
637 # so that -not_foo works correctly
638 $op =~ s/^not_/NOT /i;
640 $self->_debug("Unary OP(-$op) within hashref, recursing...");
641 my ($s, @b) = $self->_where_unary_op($op, $v);
643 # top level vs nested
644 # we assume that handled unary ops will take care of their ()s
646 List::Util::first {$op =~ $_->{regex}} @{$self->{unary_ops}}
648 ( defined $self->{_nested_func_lhs} and $self->{_nested_func_lhs} eq $k )
654 if (is_literal_value ($v) ) {
655 belch 'Hash-pairs consisting of an empty string with a literal are deprecated, and will be removed in 2.0: use -and => [ $literal ] instead';
658 puke "Supplying an empty left hand side argument is not supported in hash-pairs";
662 my $method = $self->_METHOD_FOR_refkind("_where_hashpair", $v);
663 $self->$method($k, $v);
667 push @sql_clauses, $sql;
668 push @all_bind, @bind;
671 return $self->_join_sql_clauses('and', \@sql_clauses, \@all_bind);
674 sub _where_unary_op {
675 my ($self, $op, $rhs) = @_;
677 # top level special ops are illegal in general
678 # this includes the -ident/-value ops (dual purpose unary and special)
679 puke "Illegal use of top-level '-$op'"
680 if ! defined $self->{_nested_func_lhs} and List::Util::first { $op =~ $_->{regex} } @{$self->{special_ops}};
682 if (my $op_entry = List::Util::first { $op =~ $_->{regex} } @{$self->{unary_ops}}) {
683 my $handler = $op_entry->{handler};
685 if (not ref $handler) {
686 if ($op =~ s/ [_\s]? \d+ $//x ) {
687 belch 'Use of [and|or|nest]_N modifiers is deprecated and will be removed in SQLA v2.0. '
688 . "You probably wanted ...-and => [ -$op => COND1, -$op => COND2 ... ]";
690 return $self->$handler($op, $rhs);
692 elsif (ref $handler eq 'CODE') {
693 return $handler->($self, $op, $rhs);
696 puke "Illegal handler for operator $op - expecting a method name or a coderef";
700 $self->_debug("Generic unary OP: $op - recursing as function");
702 $self->_assert_pass_injection_guard($op);
704 my ($sql, @bind) = $self->_SWITCH_refkind($rhs, {
706 puke "Illegal use of top-level '-$op'"
707 unless defined $self->{_nested_func_lhs};
710 $self->_convert('?'),
711 $self->_bindtype($self->{_nested_func_lhs}, $rhs)
715 $self->_recurse_where($rhs)
719 $sql = sprintf('%s %s',
720 $self->_sqlcase($op),
724 return ($sql, @bind);
727 sub _where_op_ANDOR {
728 my ($self, $op, $v) = @_;
730 $self->_SWITCH_refkind($v, {
732 return $self->_where_ARRAYREF($v, $op);
736 return ($op =~ /^or/i)
737 ? $self->_where_ARRAYREF([ map { $_ => $v->{$_} } (sort keys %$v) ], $op)
738 : $self->_where_HASHREF($v);
742 puke "-$op => \\\$scalar makes little sense, use " .
744 ? '[ \$scalar, \%rest_of_conditions ] instead'
745 : '-and => [ \$scalar, \%rest_of_conditions ] instead'
750 puke "-$op => \\[...] makes little sense, use " .
752 ? '[ \[...], \%rest_of_conditions ] instead'
753 : '-and => [ \[...], \%rest_of_conditions ] instead'
757 SCALAR => sub { # permissively interpreted as SQL
758 puke "-$op => \$value makes little sense, use -bool => \$value instead";
762 puke "-$op => undef not supported";
768 my ($self, $op, $v) = @_;
770 $self->_SWITCH_refkind($v, {
772 SCALAR => sub { # permissively interpreted as SQL
773 belch "literal SQL should be -nest => \\'scalar' "
774 . "instead of -nest => 'scalar' ";
779 puke "-$op => undef not supported";
783 $self->_recurse_where($v);
791 my ($self, $op, $v) = @_;
793 my ($s, @b) = $self->_SWITCH_refkind($v, {
794 SCALAR => sub { # interpreted as SQL column
795 $self->_convert($self->_quote($v));
799 puke "-$op => undef not supported";
803 $self->_recurse_where($v);
807 $s = "(NOT $s)" if $op =~ /^not/i;
812 sub _where_op_IDENT {
814 my ($op, $rhs) = splice @_, -2;
815 if (! defined $rhs or length ref $rhs) {
816 puke "-$op requires a single plain scalar argument (a quotable identifier)";
819 # in case we are called as a top level special op (no '=')
822 $_ = $self->_convert($self->_quote($_)) for ($lhs, $rhs);
830 sub _where_op_VALUE {
832 my ($op, $rhs) = splice @_, -2;
834 # in case we are called as a top level special op (no '=')
838 if (! defined $rhs) {
840 ? $self->_convert($self->_quote($lhs)) . ' IS NULL'
847 (defined $lhs ? $lhs : $self->{_nested_func_lhs}),
854 $self->_convert($self->_quote($lhs)) . ' = ' . $self->_convert('?'),
858 $self->_convert('?'),
864 sub _where_hashpair_ARRAYREF {
865 my ($self, $k, $v) = @_;
868 my @v = @$v; # need copy because of shift below
869 $self->_debug("ARRAY($k) means distribute over elements");
871 # put apart first element if it is an operator (-and, -or)
873 (defined $v[0] && $v[0] =~ /^ - (?: AND|OR ) $/ix)
877 my @distributed = map { {$k => $_} } @v;
880 $self->_debug("OP($op) reinjected into the distributed array");
881 unshift @distributed, $op;
884 my $logic = $op ? substr($op, 1) : '';
886 return $self->_recurse_where(\@distributed, $logic);
889 $self->_debug("empty ARRAY($k) means 0=1");
890 return ($self->{sqlfalse});
894 sub _where_hashpair_HASHREF {
895 my ($self, $k, $v, $logic) = @_;
898 local $self->{_nested_func_lhs} = defined $self->{_nested_func_lhs}
899 ? $self->{_nested_func_lhs}
903 my ($all_sql, @all_bind);
905 for my $orig_op (sort keys %$v) {
906 my $val = $v->{$orig_op};
908 # put the operator in canonical form
911 # FIXME - we need to phase out dash-less ops
912 $op =~ s/^-//; # remove possible initial dash
913 $op =~ s/^\s+|\s+$//g;# remove leading/trailing space
914 $op =~ s/\s+/ /g; # compress whitespace
916 $self->_assert_pass_injection_guard($op);
919 $op =~ s/^is_not/IS NOT/i;
921 # so that -not_foo works correctly
922 $op =~ s/^not_/NOT /i;
924 # another retarded special case: foo => { $op => { -value => undef } }
925 if (ref $val eq 'HASH' and keys %$val == 1 and exists $val->{-value} and ! defined $val->{-value} ) {
931 # CASE: col-value logic modifiers
932 if ($orig_op =~ /^ \- (and|or) $/xi) {
933 ($sql, @bind) = $self->_where_hashpair_HASHREF($k, $val, $1);
935 # CASE: special operators like -in or -between
936 elsif (my $special_op = List::Util::first { $op =~ $_->{regex} } @{$self->{special_ops}}) {
937 my $handler = $special_op->{handler};
939 puke "No handler supplied for special operator $orig_op";
941 elsif (not ref $handler) {
942 ($sql, @bind) = $self->$handler($k, $op, $val);
944 elsif (ref $handler eq 'CODE') {
945 ($sql, @bind) = $handler->($self, $k, $op, $val);
948 puke "Illegal handler for special operator $orig_op - expecting a method name or a coderef";
952 $self->_SWITCH_refkind($val, {
954 ARRAYREF => sub { # CASE: col => {op => \@vals}
955 ($sql, @bind) = $self->_where_field_op_ARRAYREF($k, $op, $val);
958 ARRAYREFREF => sub { # CASE: col => {op => \[$sql, @bind]} (literal SQL with bind)
959 my ($sub_sql, @sub_bind) = @$$val;
960 $self->_assert_bindval_matches_bindtype(@sub_bind);
961 $sql = join ' ', $self->_convert($self->_quote($k)),
962 $self->_sqlcase($op),
967 UNDEF => sub { # CASE: col => {op => undef} : sql "IS (NOT)? NULL"
969 $op =~ /^not$/i ? 'is not' # legacy
970 : $op =~ $self->{equality_op} ? 'is'
971 : $op =~ $self->{like_op} ? belch("Supplying an undefined argument to '@{[ uc $op]}' is deprecated") && 'is'
972 : $op =~ $self->{inequality_op} ? 'is not'
973 : $op =~ $self->{not_like_op} ? belch("Supplying an undefined argument to '@{[ uc $op]}' is deprecated") && 'is not'
974 : puke "unexpected operator '$orig_op' with undef operand";
976 $sql = $self->_quote($k) . $self->_sqlcase(" $is null");
979 FALLBACK => sub { # CASE: col => {op/func => $stuff}
980 ($sql, @bind) = $self->_where_unary_op($op, $val);
983 $self->_convert($self->_quote($k)),
984 $self->{_nested_func_lhs} eq $k ? $sql : "($sql)", # top level vs nested
990 ($all_sql) = (defined $all_sql and $all_sql) ? $self->_join_sql_clauses($logic, [$all_sql, $sql], []) : $sql;
991 push @all_bind, @bind;
993 return ($all_sql, @all_bind);
996 sub _where_field_IS {
997 my ($self, $k, $op, $v) = @_;
999 my ($s) = $self->_SWITCH_refkind($v, {
1002 $self->_convert($self->_quote($k)),
1003 map { $self->_sqlcase($_)} ($op, 'null')
1006 puke "$op can only take undef as argument";
1013 sub _where_field_op_ARRAYREF {
1014 my ($self, $k, $op, $vals) = @_;
1016 my @vals = @$vals; #always work on a copy
1019 $self->_debug(sprintf '%s means multiple elements: [ %s ]',
1021 join(', ', map { defined $_ ? "'$_'" : 'NULL' } @vals ),
1024 # see if the first element is an -and/-or op
1026 if (defined $vals[0] && $vals[0] =~ /^ - (AND|OR) $/ix) {
1031 # a long standing API wart - an attempt to change this behavior during
1032 # the 1.50 series failed *spectacularly*. Warn instead and leave the
1037 (!$logic or $logic eq 'OR')
1039 ($op =~ $self->{inequality_op} or $op =~ $self->{not_like_op})
1042 belch "A multi-element arrayref as an argument to the inequality op '$o' "
1043 . 'is technically equivalent to an always-true 1=1 (you probably wanted '
1044 . "to say ...{ \$inequality_op => [ -and => \@values ] }... instead)"
1048 # distribute $op over each remaining member of @vals, append logic if exists
1049 return $self->_recurse_where([map { {$k => {$op, $_}} } @vals], $logic);
1053 # try to DWIM on equality operators
1055 $op =~ $self->{equality_op} ? $self->{sqlfalse}
1056 : $op =~ $self->{like_op} ? belch("Supplying an empty arrayref to '@{[ uc $op]}' is deprecated") && $self->{sqlfalse}
1057 : $op =~ $self->{inequality_op} ? $self->{sqltrue}
1058 : $op =~ $self->{not_like_op} ? belch("Supplying an empty arrayref to '@{[ uc $op]}' is deprecated") && $self->{sqltrue}
1059 : puke "operator '$op' applied on an empty array (field '$k')";
1064 sub _where_hashpair_SCALARREF {
1065 my ($self, $k, $v) = @_;
1066 $self->_debug("SCALAR($k) means literal SQL: $$v");
1067 my $sql = $self->_quote($k) . " " . $$v;
1071 # literal SQL with bind
1072 sub _where_hashpair_ARRAYREFREF {
1073 my ($self, $k, $v) = @_;
1074 $self->_debug("REF($k) means literal SQL: @${$v}");
1075 my ($sql, @bind) = @$$v;
1076 $self->_assert_bindval_matches_bindtype(@bind);
1077 $sql = $self->_quote($k) . " " . $sql;
1078 return ($sql, @bind );
1081 # literal SQL without bind
1082 sub _where_hashpair_SCALAR {
1083 my ($self, $k, $v) = @_;
1084 $self->_debug("NOREF($k) means simple key=val: $k $self->{cmp} $v");
1085 my $sql = join ' ', $self->_convert($self->_quote($k)),
1086 $self->_sqlcase($self->{cmp}),
1087 $self->_convert('?');
1088 my @bind = $self->_bindtype($k, $v);
1089 return ($sql, @bind);
1093 sub _where_hashpair_UNDEF {
1094 my ($self, $k, $v) = @_;
1095 $self->_debug("UNDEF($k) means IS NULL");
1096 my $sql = $self->_quote($k) . $self->_sqlcase(' is null');
1100 #======================================================================
1101 # WHERE: TOP-LEVEL OTHERS (SCALARREF, SCALAR, UNDEF)
1102 #======================================================================
1105 sub _where_SCALARREF {
1106 my ($self, $where) = @_;
1109 $self->_debug("SCALAR(*top) means literal SQL: $$where");
1115 my ($self, $where) = @_;
1118 $self->_debug("NOREF(*top) means literal SQL: $where");
1129 #======================================================================
1130 # WHERE: BUILTIN SPECIAL OPERATORS (-in, -between)
1131 #======================================================================
1134 sub _where_field_BETWEEN {
1135 my ($self, $k, $op, $vals) = @_;
1137 my ($label, $and, $placeholder);
1138 $label = $self->_convert($self->_quote($k));
1139 $and = ' ' . $self->_sqlcase('and') . ' ';
1140 $placeholder = $self->_convert('?');
1141 $op = $self->_sqlcase($op);
1143 my $invalid_args = "Operator '$op' requires either an arrayref with two defined values or expressions, or a single literal scalarref/arrayref-ref";
1145 my ($clause, @bind) = $self->_SWITCH_refkind($vals, {
1146 ARRAYREFREF => sub {
1147 my ($s, @b) = @$$vals;
1148 $self->_assert_bindval_matches_bindtype(@b);
1155 puke $invalid_args if @$vals != 2;
1157 my (@all_sql, @all_bind);
1158 foreach my $val (@$vals) {
1159 my ($sql, @bind) = $self->_SWITCH_refkind($val, {
1161 return ($placeholder, $self->_bindtype($k, $val) );
1166 ARRAYREFREF => sub {
1167 my ($sql, @bind) = @$$val;
1168 $self->_assert_bindval_matches_bindtype(@bind);
1169 return ($sql, @bind);
1172 my ($func, $arg, @rest) = %$val;
1173 puke "Only simple { -func => arg } functions accepted as sub-arguments to BETWEEN"
1174 if (@rest or $func !~ /^ \- (.+)/x);
1175 $self->_where_unary_op($1 => $arg);
1181 push @all_sql, $sql;
1182 push @all_bind, @bind;
1186 (join $and, @all_sql),
1195 my $sql = "( $label $op $clause )";
1196 return ($sql, @bind)
1200 sub _where_field_IN {
1201 my ($self, $k, $op, $vals) = @_;
1203 # backwards compatibility : if scalar, force into an arrayref
1204 $vals = [$vals] if defined $vals && ! ref $vals;
1206 my ($label) = $self->_convert($self->_quote($k));
1207 my ($placeholder) = $self->_convert('?');
1208 $op = $self->_sqlcase($op);
1210 my ($sql, @bind) = $self->_SWITCH_refkind($vals, {
1211 ARRAYREF => sub { # list of choices
1212 if (@$vals) { # nonempty list
1213 my (@all_sql, @all_bind);
1215 for my $val (@$vals) {
1216 my ($sql, @bind) = $self->_SWITCH_refkind($val, {
1218 return ($placeholder, $val);
1223 ARRAYREFREF => sub {
1224 my ($sql, @bind) = @$$val;
1225 $self->_assert_bindval_matches_bindtype(@bind);
1226 return ($sql, @bind);
1229 my ($func, $arg, @rest) = %$val;
1230 puke "Only simple { -func => arg } functions accepted as sub-arguments to IN"
1231 if (@rest or $func !~ /^ \- (.+)/x);
1232 $self->_where_unary_op($1 => $arg);
1236 'SQL::Abstract before v1.75 used to generate incorrect SQL when the '
1237 . "-$op operator was given an undef-containing list: !!!AUDIT YOUR CODE "
1238 . 'AND DATA!!! (the upcoming Data::Query-based version of SQL::Abstract '
1239 . 'will emit the logically correct SQL instead of raising this exception)'
1243 push @all_sql, $sql;
1244 push @all_bind, @bind;
1248 sprintf('%s %s ( %s )',
1251 join(', ', @all_sql)
1253 $self->_bindtype($k, @all_bind),
1256 else { # empty list : some databases won't understand "IN ()", so DWIM
1257 my $sql = ($op =~ /\bnot\b/i) ? $self->{sqltrue} : $self->{sqlfalse};
1262 SCALARREF => sub { # literal SQL
1263 my $sql = $self->_open_outer_paren($$vals);
1264 return ("$label $op ( $sql )");
1266 ARRAYREFREF => sub { # literal SQL with bind
1267 my ($sql, @bind) = @$$vals;
1268 $self->_assert_bindval_matches_bindtype(@bind);
1269 $sql = $self->_open_outer_paren($sql);
1270 return ("$label $op ( $sql )", @bind);
1274 puke "Argument passed to the '$op' operator can not be undefined";
1278 puke "special op $op requires an arrayref (or scalarref/arrayref-ref)";
1282 return ($sql, @bind);
1285 # Some databases (SQLite) treat col IN (1, 2) different from
1286 # col IN ( (1, 2) ). Use this to strip all outer parens while
1287 # adding them back in the corresponding method
1288 sub _open_outer_paren {
1289 my ($self, $sql) = @_;
1291 while (my ($inner) = $sql =~ /^ \s* \( (.*) \) \s* $/xs) {
1293 # there are closing parens inside, need the heavy duty machinery
1294 # to reevaluate the extraction starting from $sql (full reevaluation)
1295 if ($inner =~ /\)/) {
1296 require Text::Balanced;
1298 my (undef, $remainder) = do {
1299 # idiotic design - writes to $@ but *DOES NOT* throw exceptions
1301 Text::Balanced::extract_bracketed($sql, '()', qr/\s*/);
1304 # the entire expression needs to be a balanced bracketed thing
1305 # (after an extract no remainder sans trailing space)
1306 last if defined $remainder and $remainder =~ /\S/;
1316 #======================================================================
1318 #======================================================================
1321 my ($self, $arg) = @_;
1324 for my $c ($self->_order_by_chunks($arg) ) {
1325 $self->_SWITCH_refkind($c, {
1326 SCALAR => sub { push @sql, $c },
1327 ARRAYREF => sub { push @sql, shift @$c; push @bind, @$c },
1333 $self->_sqlcase(' order by'),
1339 return wantarray ? ($sql, @bind) : $sql;
1342 sub _order_by_chunks {
1343 my ($self, $arg) = @_;
1345 return $self->_SWITCH_refkind($arg, {
1348 map { $self->_order_by_chunks($_ ) } @$arg;
1351 ARRAYREFREF => sub {
1352 my ($s, @b) = @$$arg;
1353 $self->_assert_bindval_matches_bindtype(@b);
1357 SCALAR => sub {$self->_quote($arg)},
1359 UNDEF => sub {return () },
1361 SCALARREF => sub {$$arg}, # literal SQL, no quoting
1364 # get first pair in hash
1365 my ($key, $val, @rest) = %$arg;
1367 return () unless $key;
1369 if (@rest or not $key =~ /^-(desc|asc)/i) {
1370 puke "hash passed to _order_by must have exactly one key (-desc or -asc)";
1376 for my $c ($self->_order_by_chunks($val)) {
1379 $self->_SWITCH_refkind($c, {
1384 ($sql, @bind) = @$c;
1388 $sql = $sql . ' ' . $self->_sqlcase($direction);
1390 push @ret, [ $sql, @bind];
1399 #======================================================================
1400 # DATASOURCE (FOR NOW, JUST PLAIN TABLE OR LIST OF TABLES)
1401 #======================================================================
1406 $self->_SWITCH_refkind($from, {
1407 ARRAYREF => sub {join ', ', map { $self->_quote($_) } @$from;},
1408 SCALAR => sub {$self->_quote($from)},
1409 SCALARREF => sub {$$from},
1414 #======================================================================
1416 #======================================================================
1418 # highly optimized, as it's called way too often
1420 # my ($self, $label) = @_;
1422 return '' unless defined $_[1];
1423 return ${$_[1]} if ref($_[1]) eq 'SCALAR';
1425 $_[0]->{quote_char} or
1426 ($_[0]->_assert_pass_injection_guard($_[1]), return $_[1]);
1428 my $qref = ref $_[0]->{quote_char};
1430 !$qref ? ($_[0]->{quote_char}, $_[0]->{quote_char})
1431 : ($qref eq 'ARRAY') ? @{$_[0]->{quote_char}}
1432 : puke "Unsupported quote_char format: $_[0]->{quote_char}";
1434 my $esc = $_[0]->{escape_char} || $r;
1436 # parts containing * are naturally unquoted
1437 return join($_[0]->{name_sep}||'', map
1438 +( $_ eq '*' ? $_ : do { (my $n = $_) =~ s/(\Q$esc\E|\Q$r\E)/$esc$1/g; $l . $n . $r } ),
1439 ( $_[0]->{name_sep} ? split (/\Q$_[0]->{name_sep}\E/, $_[1] ) : $_[1] )
1444 # Conversion, if applicable
1446 #my ($self, $arg) = @_;
1447 if ($_[0]->{convert}) {
1448 return $_[0]->_sqlcase($_[0]->{convert}) .'(' . $_[1] . ')';
1455 #my ($self, $col, @vals) = @_;
1456 # called often - tighten code
1457 return $_[0]->{bindtype} eq 'columns'
1458 ? map {[$_[1], $_]} @_[2 .. $#_]
1463 # Dies if any element of @bind is not in [colname => value] format
1464 # if bindtype is 'columns'.
1465 sub _assert_bindval_matches_bindtype {
1466 # my ($self, @bind) = @_;
1468 if ($self->{bindtype} eq 'columns') {
1470 if (!defined $_ || ref($_) ne 'ARRAY' || @$_ != 2) {
1471 puke "bindtype 'columns' selected, you need to pass: [column_name => bind_value]"
1477 sub _join_sql_clauses {
1478 my ($self, $logic, $clauses_aref, $bind_aref) = @_;
1480 if (@$clauses_aref > 1) {
1481 my $join = " " . $self->_sqlcase($logic) . " ";
1482 my $sql = '( ' . join($join, @$clauses_aref) . ' )';
1483 return ($sql, @$bind_aref);
1485 elsif (@$clauses_aref) {
1486 return ($clauses_aref->[0], @$bind_aref); # no parentheses
1489 return (); # if no SQL, ignore @$bind_aref
1494 # Fix SQL case, if so requested
1496 # LDNOTE: if $self->{case} is true, then it contains 'lower', so we
1497 # don't touch the argument ... crooked logic, but let's not change it!
1498 return $_[0]->{case} ? $_[1] : uc($_[1]);
1502 #======================================================================
1503 # DISPATCHING FROM REFKIND
1504 #======================================================================
1507 my ($self, $data) = @_;
1509 return 'UNDEF' unless defined $data;
1511 # blessed objects are treated like scalars
1512 my $ref = (Scalar::Util::blessed $data) ? '' : ref $data;
1514 return 'SCALAR' unless $ref;
1517 while ($ref eq 'REF') {
1519 $ref = (Scalar::Util::blessed $data) ? '' : ref $data;
1523 return ($ref||'SCALAR') . ('REF' x $n_steps);
1527 my ($self, $data) = @_;
1528 my @try = ($self->_refkind($data));
1529 push @try, 'SCALAR_or_UNDEF' if $try[0] eq 'SCALAR' || $try[0] eq 'UNDEF';
1530 push @try, 'FALLBACK';
1534 sub _METHOD_FOR_refkind {
1535 my ($self, $meth_prefix, $data) = @_;
1538 for (@{$self->_try_refkind($data)}) {
1539 $method = $self->can($meth_prefix."_".$_)
1543 return $method || puke "cannot dispatch on '$meth_prefix' for ".$self->_refkind($data);
1547 sub _SWITCH_refkind {
1548 my ($self, $data, $dispatch_table) = @_;
1551 for (@{$self->_try_refkind($data)}) {
1552 $coderef = $dispatch_table->{$_}
1556 puke "no dispatch entry for ".$self->_refkind($data)
1565 #======================================================================
1566 # VALUES, GENERATE, AUTOLOAD
1567 #======================================================================
1569 # LDNOTE: original code from nwiger, didn't touch code in that section
1570 # I feel the AUTOLOAD stuff should not be the default, it should
1571 # only be activated on explicit demand by user.
1575 my $data = shift || return;
1576 puke "Argument to ", __PACKAGE__, "->values must be a \\%hash"
1577 unless ref $data eq 'HASH';
1580 foreach my $k (sort keys %$data) {
1581 my $v = $data->{$k};
1582 $self->_SWITCH_refkind($v, {
1584 if ($self->{array_datatypes}) { # array datatype
1585 push @all_bind, $self->_bindtype($k, $v);
1587 else { # literal SQL with bind
1588 my ($sql, @bind) = @$v;
1589 $self->_assert_bindval_matches_bindtype(@bind);
1590 push @all_bind, @bind;
1593 ARRAYREFREF => sub { # literal SQL with bind
1594 my ($sql, @bind) = @${$v};
1595 $self->_assert_bindval_matches_bindtype(@bind);
1596 push @all_bind, @bind;
1598 SCALARREF => sub { # literal SQL without bind
1600 SCALAR_or_UNDEF => sub {
1601 push @all_bind, $self->_bindtype($k, $v);
1612 my(@sql, @sqlq, @sqlv);
1616 if ($ref eq 'HASH') {
1617 for my $k (sort keys %$_) {
1620 my $label = $self->_quote($k);
1621 if ($r eq 'ARRAY') {
1622 # literal SQL with bind
1623 my ($sql, @bind) = @$v;
1624 $self->_assert_bindval_matches_bindtype(@bind);
1625 push @sqlq, "$label = $sql";
1627 } elsif ($r eq 'SCALAR') {
1628 # literal SQL without bind
1629 push @sqlq, "$label = $$v";
1631 push @sqlq, "$label = ?";
1632 push @sqlv, $self->_bindtype($k, $v);
1635 push @sql, $self->_sqlcase('set'), join ', ', @sqlq;
1636 } elsif ($ref eq 'ARRAY') {
1637 # unlike insert(), assume these are ONLY the column names, i.e. for SQL
1640 if ($r eq 'ARRAY') { # literal SQL with bind
1641 my ($sql, @bind) = @$v;
1642 $self->_assert_bindval_matches_bindtype(@bind);
1645 } elsif ($r eq 'SCALAR') { # literal SQL without bind
1646 # embedded literal SQL
1653 push @sql, '(' . join(', ', @sqlq) . ')';
1654 } elsif ($ref eq 'SCALAR') {
1658 # strings get case twiddled
1659 push @sql, $self->_sqlcase($_);
1663 my $sql = join ' ', @sql;
1665 # this is pretty tricky
1666 # if ask for an array, return ($stmt, @bind)
1667 # otherwise, s/?/shift @sqlv/ to put it inline
1669 return ($sql, @sqlv);
1671 1 while $sql =~ s/\?/my $d = shift(@sqlv);
1672 ref $d ? $d->[1] : $d/e;
1681 # This allows us to check for a local, then _form, attr
1683 my($name) = $AUTOLOAD =~ /.*::(.+)/;
1684 return $self->generate($name, @_);
1695 SQL::Abstract - Generate SQL from Perl data structures
1701 my $sql = SQL::Abstract->new;
1703 my($stmt, @bind) = $sql->select($source, \@fields, \%where, $order);
1705 my($stmt, @bind) = $sql->insert($table, \%fieldvals || \@values);
1707 my($stmt, @bind) = $sql->update($table, \%fieldvals, \%where);
1709 my($stmt, @bind) = $sql->delete($table, \%where);
1711 # Then, use these in your DBI statements
1712 my $sth = $dbh->prepare($stmt);
1713 $sth->execute(@bind);
1715 # Just generate the WHERE clause
1716 my($stmt, @bind) = $sql->where(\%where, $order);
1718 # Return values in the same order, for hashed queries
1719 # See PERFORMANCE section for more details
1720 my @bind = $sql->values(\%fieldvals);
1724 This module was inspired by the excellent L<DBIx::Abstract>.
1725 However, in using that module I found that what I really wanted
1726 to do was generate SQL, but still retain complete control over my
1727 statement handles and use the DBI interface. So, I set out to
1728 create an abstract SQL generation module.
1730 While based on the concepts used by L<DBIx::Abstract>, there are
1731 several important differences, especially when it comes to WHERE
1732 clauses. I have modified the concepts used to make the SQL easier
1733 to generate from Perl data structures and, IMO, more intuitive.
1734 The underlying idea is for this module to do what you mean, based
1735 on the data structures you provide it. The big advantage is that
1736 you don't have to modify your code every time your data changes,
1737 as this module figures it out.
1739 To begin with, an SQL INSERT is as easy as just specifying a hash
1740 of C<key=value> pairs:
1743 name => 'Jimbo Bobson',
1744 phone => '123-456-7890',
1745 address => '42 Sister Lane',
1746 city => 'St. Louis',
1747 state => 'Louisiana',
1750 The SQL can then be generated with this:
1752 my($stmt, @bind) = $sql->insert('people', \%data);
1754 Which would give you something like this:
1756 $stmt = "INSERT INTO people
1757 (address, city, name, phone, state)
1758 VALUES (?, ?, ?, ?, ?)";
1759 @bind = ('42 Sister Lane', 'St. Louis', 'Jimbo Bobson',
1760 '123-456-7890', 'Louisiana');
1762 These are then used directly in your DBI code:
1764 my $sth = $dbh->prepare($stmt);
1765 $sth->execute(@bind);
1767 =head2 Inserting and Updating Arrays
1769 If your database has array types (like for example Postgres),
1770 activate the special option C<< array_datatypes => 1 >>
1771 when creating the C<SQL::Abstract> object.
1772 Then you may use an arrayref to insert and update database array types:
1774 my $sql = SQL::Abstract->new(array_datatypes => 1);
1776 planets => [qw/Mercury Venus Earth Mars/]
1779 my($stmt, @bind) = $sql->insert('solar_system', \%data);
1783 $stmt = "INSERT INTO solar_system (planets) VALUES (?)"
1785 @bind = (['Mercury', 'Venus', 'Earth', 'Mars']);
1788 =head2 Inserting and Updating SQL
1790 In order to apply SQL functions to elements of your C<%data> you may
1791 specify a reference to an arrayref for the given hash value. For example,
1792 if you need to execute the Oracle C<to_date> function on a value, you can
1793 say something like this:
1797 date_entered => \[ "to_date(?,'MM/DD/YYYY')", "03/02/2003" ],
1800 The first value in the array is the actual SQL. Any other values are
1801 optional and would be included in the bind values array. This gives
1804 my($stmt, @bind) = $sql->insert('people', \%data);
1806 $stmt = "INSERT INTO people (name, date_entered)
1807 VALUES (?, to_date(?,'MM/DD/YYYY'))";
1808 @bind = ('Bill', '03/02/2003');
1810 An UPDATE is just as easy, all you change is the name of the function:
1812 my($stmt, @bind) = $sql->update('people', \%data);
1814 Notice that your C<%data> isn't touched; the module will generate
1815 the appropriately quirky SQL for you automatically. Usually you'll
1816 want to specify a WHERE clause for your UPDATE, though, which is
1817 where handling C<%where> hashes comes in handy...
1819 =head2 Complex where statements
1821 This module can generate pretty complicated WHERE statements
1822 easily. For example, simple C<key=value> pairs are taken to mean
1823 equality, and if you want to see if a field is within a set
1824 of values, you can use an arrayref. Let's say we wanted to
1825 SELECT some data based on this criteria:
1828 requestor => 'inna',
1829 worker => ['nwiger', 'rcwe', 'sfz'],
1830 status => { '!=', 'completed' }
1833 my($stmt, @bind) = $sql->select('tickets', '*', \%where);
1835 The above would give you something like this:
1837 $stmt = "SELECT * FROM tickets WHERE
1838 ( requestor = ? ) AND ( status != ? )
1839 AND ( worker = ? OR worker = ? OR worker = ? )";
1840 @bind = ('inna', 'completed', 'nwiger', 'rcwe', 'sfz');
1842 Which you could then use in DBI code like so:
1844 my $sth = $dbh->prepare($stmt);
1845 $sth->execute(@bind);
1851 The methods are simple. There's one for every major SQL operation,
1852 and a constructor you use first. The arguments are specified in a
1853 similar order for each method (table, then fields, then a where
1854 clause) to try and simplify things.
1856 =head2 new(option => 'value')
1858 The C<new()> function takes a list of options and values, and returns
1859 a new B<SQL::Abstract> object which can then be used to generate SQL
1860 through the methods below. The options accepted are:
1866 If set to 'lower', then SQL will be generated in all lowercase. By
1867 default SQL is generated in "textbook" case meaning something like:
1869 SELECT a_field FROM a_table WHERE some_field LIKE '%someval%'
1871 Any setting other than 'lower' is ignored.
1875 This determines what the default comparison operator is. By default
1876 it is C<=>, meaning that a hash like this:
1878 %where = (name => 'nwiger', email => 'nate@wiger.org');
1880 Will generate SQL like this:
1882 WHERE name = 'nwiger' AND email = 'nate@wiger.org'
1884 However, you may want loose comparisons by default, so if you set
1885 C<cmp> to C<like> you would get SQL such as:
1887 WHERE name like 'nwiger' AND email like 'nate@wiger.org'
1889 You can also override the comparison on an individual basis - see
1890 the huge section on L</"WHERE CLAUSES"> at the bottom.
1892 =item sqltrue, sqlfalse
1894 Expressions for inserting boolean values within SQL statements.
1895 By default these are C<1=1> and C<1=0>. They are used
1896 by the special operators C<-in> and C<-not_in> for generating
1897 correct SQL even when the argument is an empty array (see below).
1901 This determines the default logical operator for multiple WHERE
1902 statements in arrays or hashes. If absent, the default logic is "or"
1903 for arrays, and "and" for hashes. This means that a WHERE
1907 event_date => {'>=', '2/13/99'},
1908 event_date => {'<=', '4/24/03'},
1911 will generate SQL like this:
1913 WHERE event_date >= '2/13/99' OR event_date <= '4/24/03'
1915 This is probably not what you want given this query, though (look
1916 at the dates). To change the "OR" to an "AND", simply specify:
1918 my $sql = SQL::Abstract->new(logic => 'and');
1920 Which will change the above C<WHERE> to:
1922 WHERE event_date >= '2/13/99' AND event_date <= '4/24/03'
1924 The logic can also be changed locally by inserting
1925 a modifier in front of an arrayref :
1927 @where = (-and => [event_date => {'>=', '2/13/99'},
1928 event_date => {'<=', '4/24/03'} ]);
1930 See the L</"WHERE CLAUSES"> section for explanations.
1934 This will automatically convert comparisons using the specified SQL
1935 function for both column and value. This is mostly used with an argument
1936 of C<upper> or C<lower>, so that the SQL will have the effect of
1937 case-insensitive "searches". For example, this:
1939 $sql = SQL::Abstract->new(convert => 'upper');
1940 %where = (keywords => 'MaKe iT CAse inSeNSItive');
1942 Will turn out the following SQL:
1944 WHERE upper(keywords) like upper('MaKe iT CAse inSeNSItive')
1946 The conversion can be C<upper()>, C<lower()>, or any other SQL function
1947 that can be applied symmetrically to fields (actually B<SQL::Abstract> does
1948 not validate this option; it will just pass through what you specify verbatim).
1952 This is a kludge because many databases suck. For example, you can't
1953 just bind values using DBI's C<execute()> for Oracle C<CLOB> or C<BLOB> fields.
1954 Instead, you have to use C<bind_param()>:
1956 $sth->bind_param(1, 'reg data');
1957 $sth->bind_param(2, $lots, {ora_type => ORA_CLOB});
1959 The problem is, B<SQL::Abstract> will normally just return a C<@bind> array,
1960 which loses track of which field each slot refers to. Fear not.
1962 If you specify C<bindtype> in new, you can determine how C<@bind> is returned.
1963 Currently, you can specify either C<normal> (default) or C<columns>. If you
1964 specify C<columns>, you will get an array that looks like this:
1966 my $sql = SQL::Abstract->new(bindtype => 'columns');
1967 my($stmt, @bind) = $sql->insert(...);
1970 [ 'column1', 'value1' ],
1971 [ 'column2', 'value2' ],
1972 [ 'column3', 'value3' ],
1975 You can then iterate through this manually, using DBI's C<bind_param()>.
1977 $sth->prepare($stmt);
1980 my($col, $data) = @$_;
1981 if ($col eq 'details' || $col eq 'comments') {
1982 $sth->bind_param($i, $data, {ora_type => ORA_CLOB});
1983 } elsif ($col eq 'image') {
1984 $sth->bind_param($i, $data, {ora_type => ORA_BLOB});
1986 $sth->bind_param($i, $data);
1990 $sth->execute; # execute without @bind now
1992 Now, why would you still use B<SQL::Abstract> if you have to do this crap?
1993 Basically, the advantage is still that you don't have to care which fields
1994 are or are not included. You could wrap that above C<for> loop in a simple
1995 sub called C<bind_fields()> or something and reuse it repeatedly. You still
1996 get a layer of abstraction over manual SQL specification.
1998 Note that if you set L</bindtype> to C<columns>, the C<\[ $sql, @bind ]>
1999 construct (see L</Literal SQL with placeholders and bind values (subqueries)>)
2000 will expect the bind values in this format.
2004 This is the character that a table or column name will be quoted
2005 with. By default this is an empty string, but you could set it to
2006 the character C<`>, to generate SQL like this:
2008 SELECT `a_field` FROM `a_table` WHERE `some_field` LIKE '%someval%'
2010 Alternatively, you can supply an array ref of two items, the first being the left
2011 hand quote character, and the second the right hand quote character. For
2012 example, you could supply C<['[',']']> for SQL Server 2000 compliant quotes
2013 that generates SQL like this:
2015 SELECT [a_field] FROM [a_table] WHERE [some_field] LIKE '%someval%'
2017 Quoting is useful if you have tables or columns names that are reserved
2018 words in your database's SQL dialect.
2022 This is the character that will be used to escape L</quote_char>s appearing
2023 in an identifier before it has been quoted.
2025 The parameter default in case of a single L</quote_char> character is the quote
2028 When opening-closing-style quoting is used (L</quote_char> is an arrayref)
2029 this parameter defaults to the B<closing (right)> L</quote_char>. Occurrences
2030 of the B<opening (left)> L</quote_char> within the identifier are currently left
2031 untouched. The default for opening-closing-style quotes may change in future
2032 versions, thus you are B<strongly encouraged> to specify the escape character
2037 This is the character that separates a table and column name. It is
2038 necessary to specify this when the C<quote_char> option is selected,
2039 so that tables and column names can be individually quoted like this:
2041 SELECT `table`.`one_field` FROM `table` WHERE `table`.`other_field` = 1
2043 =item injection_guard
2045 A regular expression C<qr/.../> that is applied to any C<-function> and unquoted
2046 column name specified in a query structure. This is a safety mechanism to avoid
2047 injection attacks when mishandling user input e.g.:
2049 my %condition_as_column_value_pairs = get_values_from_user();
2050 $sqla->select( ... , \%condition_as_column_value_pairs );
2052 If the expression matches an exception is thrown. Note that literal SQL
2053 supplied via C<\'...'> or C<\['...']> is B<not> checked in any way.
2055 Defaults to checking for C<;> and the C<GO> keyword (TransactSQL)
2057 =item array_datatypes
2059 When this option is true, arrayrefs in INSERT or UPDATE are
2060 interpreted as array datatypes and are passed directly
2062 When this option is false, arrayrefs are interpreted
2063 as literal SQL, just like refs to arrayrefs
2064 (but this behavior is for backwards compatibility; when writing
2065 new queries, use the "reference to arrayref" syntax
2071 Takes a reference to a list of "special operators"
2072 to extend the syntax understood by L<SQL::Abstract>.
2073 See section L</"SPECIAL OPERATORS"> for details.
2077 Takes a reference to a list of "unary operators"
2078 to extend the syntax understood by L<SQL::Abstract>.
2079 See section L</"UNARY OPERATORS"> for details.
2085 =head2 insert($table, \@values || \%fieldvals, \%options)
2087 This is the simplest function. You simply give it a table name
2088 and either an arrayref of values or hashref of field/value pairs.
2089 It returns an SQL INSERT statement and a list of bind values.
2090 See the sections on L</"Inserting and Updating Arrays"> and
2091 L</"Inserting and Updating SQL"> for information on how to insert
2092 with those data types.
2094 The optional C<\%options> hash reference may contain additional
2095 options to generate the insert SQL. Currently supported options
2102 Takes either a scalar of raw SQL fields, or an array reference of
2103 field names, and adds on an SQL C<RETURNING> statement at the end.
2104 This allows you to return data generated by the insert statement
2105 (such as row IDs) without performing another C<SELECT> statement.
2106 Note, however, this is not part of the SQL standard and may not
2107 be supported by all database engines.
2111 =head2 update($table, \%fieldvals, \%where, \%options)
2113 This takes a table, hashref of field/value pairs, and an optional
2114 hashref L<WHERE clause|/WHERE CLAUSES>. It returns an SQL UPDATE function and a list
2116 See the sections on L</"Inserting and Updating Arrays"> and
2117 L</"Inserting and Updating SQL"> for information on how to insert
2118 with those data types.
2120 The optional C<\%options> hash reference may contain additional
2121 options to generate the update SQL. Currently supported options
2128 See the C<returning> option to
2129 L<insert|/insert($table, \@values || \%fieldvals, \%options)>.
2133 =head2 select($source, $fields, $where, $order)
2135 This returns a SQL SELECT statement and associated list of bind values, as
2136 specified by the arguments :
2142 Specification of the 'FROM' part of the statement.
2143 The argument can be either a plain scalar (interpreted as a table
2144 name, will be quoted), or an arrayref (interpreted as a list
2145 of table names, joined by commas, quoted), or a scalarref
2146 (literal table name, not quoted), or a ref to an arrayref
2147 (list of literal table names, joined by commas, not quoted).
2151 Specification of the list of fields to retrieve from
2153 The argument can be either an arrayref (interpreted as a list
2154 of field names, will be joined by commas and quoted), or a
2155 plain scalar (literal SQL, not quoted).
2156 Please observe that this API is not as flexible as that of
2157 the first argument C<$source>, for backwards compatibility reasons.
2161 Optional argument to specify the WHERE part of the query.
2162 The argument is most often a hashref, but can also be
2163 an arrayref or plain scalar --
2164 see section L<WHERE clause|/"WHERE CLAUSES"> for details.
2168 Optional argument to specify the ORDER BY part of the query.
2169 The argument can be a scalar, a hashref or an arrayref
2170 -- see section L<ORDER BY clause|/"ORDER BY CLAUSES">
2176 =head2 delete($table, \%where, \%options)
2178 This takes a table name and optional hashref L<WHERE clause|/WHERE CLAUSES>.
2179 It returns an SQL DELETE statement and list of bind values.
2181 The optional C<\%options> hash reference may contain additional
2182 options to generate the delete SQL. Currently supported options
2189 See the C<returning> option to
2190 L<insert|/insert($table, \@values || \%fieldvals, \%options)>.
2194 =head2 where(\%where, $order)
2196 This is used to generate just the WHERE clause. For example,
2197 if you have an arbitrary data structure and know what the
2198 rest of your SQL is going to look like, but want an easy way
2199 to produce a WHERE clause, use this. It returns an SQL WHERE
2200 clause and list of bind values.
2203 =head2 values(\%data)
2205 This just returns the values from the hash C<%data>, in the same
2206 order that would be returned from any of the other above queries.
2207 Using this allows you to markedly speed up your queries if you
2208 are affecting lots of rows. See below under the L</"PERFORMANCE"> section.
2210 =head2 generate($any, 'number', $of, \@data, $struct, \%types)
2212 Warning: This is an experimental method and subject to change.
2214 This returns arbitrarily generated SQL. It's a really basic shortcut.
2215 It will return two different things, depending on return context:
2217 my($stmt, @bind) = $sql->generate('create table', \$table, \@fields);
2218 my $stmt_and_val = $sql->generate('create table', \$table, \@fields);
2220 These would return the following:
2222 # First calling form
2223 $stmt = "CREATE TABLE test (?, ?)";
2224 @bind = (field1, field2);
2226 # Second calling form
2227 $stmt_and_val = "CREATE TABLE test (field1, field2)";
2229 Depending on what you're trying to do, it's up to you to choose the correct
2230 format. In this example, the second form is what you would want.
2234 $sql->generate('alter session', { nls_date_format => 'MM/YY' });
2238 ALTER SESSION SET nls_date_format = 'MM/YY'
2240 You get the idea. Strings get their case twiddled, but everything
2241 else remains verbatim.
2243 =head1 EXPORTABLE FUNCTIONS
2245 =head2 is_plain_value
2247 Determines if the supplied argument is a plain value as understood by this
2252 =item * The value is C<undef>
2254 =item * The value is a non-reference
2256 =item * The value is an object with stringification overloading
2258 =item * The value is of the form C<< { -value => $anything } >>
2262 On failure returns C<undef>, on success returns a B<scalar> reference
2263 to the original supplied argument.
2269 The stringification overloading detection is rather advanced: it takes
2270 into consideration not only the presence of a C<""> overload, but if that
2271 fails also checks for enabled
2272 L<autogenerated versions of C<"">|overload/Magic Autogeneration>, based
2273 on either C<0+> or C<bool>.
2275 Unfortunately testing in the field indicates that this
2276 detection B<< may tickle a latent bug in perl versions before 5.018 >>,
2277 but only when very large numbers of stringifying objects are involved.
2278 At the time of writing ( Sep 2014 ) there is no clear explanation of
2279 the direct cause, nor is there a manageably small test case that reliably
2280 reproduces the problem.
2282 If you encounter any of the following exceptions in B<random places within
2283 your application stack> - this module may be to blame:
2285 Operation "ne": no method found,
2286 left argument in overloaded package <something>,
2287 right argument in overloaded package <something>
2291 Stub found while resolving method "???" overloading """" in package <something>
2293 If you fall victim to the above - please attempt to reduce the problem
2294 to something that could be sent to the L<SQL::Abstract developers
2295 |DBIx::Class/GETTING HELP/SUPPORT>
2296 (either publicly or privately). As a workaround in the meantime you can
2297 set C<$ENV{SQLA_ISVALUE_IGNORE_AUTOGENERATED_STRINGIFICATION}> to a true
2298 value, which will most likely eliminate your problem (at the expense of
2299 not being able to properly detect exotic forms of stringification).
2301 This notice and environment variable will be removed in a future version,
2302 as soon as the underlying problem is found and a reliable workaround is
2307 =head2 is_literal_value
2309 Determines if the supplied argument is a literal value as understood by this
2314 =item * C<\$sql_string>
2316 =item * C<\[ $sql_string, @bind_values ]>
2320 On failure returns C<undef>, on success returns an B<array> reference
2321 containing the unpacked version of the supplied literal SQL and bind values.
2323 =head1 WHERE CLAUSES
2327 This module uses a variation on the idea from L<DBIx::Abstract>. It
2328 is B<NOT>, repeat I<not> 100% compatible. B<The main logic of this
2329 module is that things in arrays are OR'ed, and things in hashes
2332 The easiest way to explain is to show lots of examples. After
2333 each C<%where> hash shown, it is assumed you used:
2335 my($stmt, @bind) = $sql->where(\%where);
2337 However, note that the C<%where> hash can be used directly in any
2338 of the other functions as well, as described above.
2340 =head2 Key-value pairs
2342 So, let's get started. To begin, a simple hash:
2346 status => 'completed'
2349 Is converted to SQL C<key = val> statements:
2351 $stmt = "WHERE user = ? AND status = ?";
2352 @bind = ('nwiger', 'completed');
2354 One common thing I end up doing is having a list of values that
2355 a field can be in. To do this, simply specify a list inside of
2360 status => ['assigned', 'in-progress', 'pending'];
2363 This simple code will create the following:
2365 $stmt = "WHERE user = ? AND ( status = ? OR status = ? OR status = ? )";
2366 @bind = ('nwiger', 'assigned', 'in-progress', 'pending');
2368 A field associated to an empty arrayref will be considered a
2369 logical false and will generate 0=1.
2371 =head2 Tests for NULL values
2373 If the value part is C<undef> then this is converted to SQL <IS NULL>
2382 $stmt = "WHERE user = ? AND status IS NULL";
2385 To test if a column IS NOT NULL:
2389 status => { '!=', undef },
2392 =head2 Specific comparison operators
2394 If you want to specify a different type of operator for your comparison,
2395 you can use a hashref for a given column:
2399 status => { '!=', 'completed' }
2402 Which would generate:
2404 $stmt = "WHERE user = ? AND status != ?";
2405 @bind = ('nwiger', 'completed');
2407 To test against multiple values, just enclose the values in an arrayref:
2409 status => { '=', ['assigned', 'in-progress', 'pending'] };
2411 Which would give you:
2413 "WHERE status = ? OR status = ? OR status = ?"
2416 The hashref can also contain multiple pairs, in which case it is expanded
2417 into an C<AND> of its elements:
2421 status => { '!=', 'completed', -not_like => 'pending%' }
2424 # Or more dynamically, like from a form
2425 $where{user} = 'nwiger';
2426 $where{status}{'!='} = 'completed';
2427 $where{status}{'-not_like'} = 'pending%';
2429 # Both generate this
2430 $stmt = "WHERE user = ? AND status != ? AND status NOT LIKE ?";
2431 @bind = ('nwiger', 'completed', 'pending%');
2434 To get an OR instead, you can combine it with the arrayref idea:
2438 priority => [ { '=', 2 }, { '>', 5 } ]
2441 Which would generate:
2443 $stmt = "WHERE ( priority = ? OR priority > ? ) AND user = ?";
2444 @bind = ('2', '5', 'nwiger');
2446 If you want to include literal SQL (with or without bind values), just use a
2447 scalar reference or reference to an arrayref as the value:
2450 date_entered => { '>' => \["to_date(?, 'MM/DD/YYYY')", "11/26/2008"] },
2451 date_expires => { '<' => \"now()" }
2454 Which would generate:
2456 $stmt = "WHERE date_entered > to_date(?, 'MM/DD/YYYY') AND date_expires < now()";
2457 @bind = ('11/26/2008');
2460 =head2 Logic and nesting operators
2462 In the example above,
2463 there is a subtle trap if you want to say something like
2464 this (notice the C<AND>):
2466 WHERE priority != ? AND priority != ?
2468 Because, in Perl you I<can't> do this:
2470 priority => { '!=' => 2, '!=' => 1 }
2472 As the second C<!=> key will obliterate the first. The solution
2473 is to use the special C<-modifier> form inside an arrayref:
2475 priority => [ -and => {'!=', 2},
2479 Normally, these would be joined by C<OR>, but the modifier tells it
2480 to use C<AND> instead. (Hint: You can use this in conjunction with the
2481 C<logic> option to C<new()> in order to change the way your queries
2482 work by default.) B<Important:> Note that the C<-modifier> goes
2483 B<INSIDE> the arrayref, as an extra first element. This will
2484 B<NOT> do what you think it might:
2486 priority => -and => [{'!=', 2}, {'!=', 1}] # WRONG!
2488 Here is a quick list of equivalencies, since there is some overlap:
2491 status => {'!=', 'completed', 'not like', 'pending%' }
2492 status => [ -and => {'!=', 'completed'}, {'not like', 'pending%'}]
2495 status => {'=', ['assigned', 'in-progress']}
2496 status => [ -or => {'=', 'assigned'}, {'=', 'in-progress'}]
2497 status => [ {'=', 'assigned'}, {'=', 'in-progress'} ]
2501 =head2 Special operators : IN, BETWEEN, etc.
2503 You can also use the hashref format to compare a list of fields using the
2504 C<IN> comparison operator, by specifying the list as an arrayref:
2507 status => 'completed',
2508 reportid => { -in => [567, 2335, 2] }
2511 Which would generate:
2513 $stmt = "WHERE status = ? AND reportid IN (?,?,?)";
2514 @bind = ('completed', '567', '2335', '2');
2516 The reverse operator C<-not_in> generates SQL C<NOT IN> and is used in
2519 If the argument to C<-in> is an empty array, 'sqlfalse' is generated
2520 (by default : C<1=0>). Similarly, C<< -not_in => [] >> generates
2521 'sqltrue' (by default : C<1=1>).
2523 In addition to the array you can supply a chunk of literal sql or
2524 literal sql with bind:
2527 customer => { -in => \[
2528 'SELECT cust_id FROM cust WHERE balance > ?',
2531 status => { -in => \'SELECT status_codes FROM states' },
2537 customer IN ( SELECT cust_id FROM cust WHERE balance > ? )
2538 AND status IN ( SELECT status_codes FROM states )
2542 Finally, if the argument to C<-in> is not a reference, it will be
2543 treated as a single-element array.
2545 Another pair of operators is C<-between> and C<-not_between>,
2546 used with an arrayref of two values:
2550 completion_date => {
2551 -not_between => ['2002-10-01', '2003-02-06']
2557 WHERE user = ? AND completion_date NOT BETWEEN ( ? AND ? )
2559 Just like with C<-in> all plausible combinations of literal SQL
2563 start0 => { -between => [ 1, 2 ] },
2564 start1 => { -between => \["? AND ?", 1, 2] },
2565 start2 => { -between => \"lower(x) AND upper(y)" },
2566 start3 => { -between => [
2568 \["upper(?)", 'stuff' ],
2575 ( start0 BETWEEN ? AND ? )
2576 AND ( start1 BETWEEN ? AND ? )
2577 AND ( start2 BETWEEN lower(x) AND upper(y) )
2578 AND ( start3 BETWEEN lower(x) AND upper(?) )
2580 @bind = (1, 2, 1, 2, 'stuff');
2583 These are the two builtin "special operators"; but the
2584 list can be expanded : see section L</"SPECIAL OPERATORS"> below.
2586 =head2 Unary operators: bool
2588 If you wish to test against boolean columns or functions within your
2589 database you can use the C<-bool> and C<-not_bool> operators. For
2590 example to test the column C<is_user> being true and the column
2591 C<is_enabled> being false you would use:-
2595 -not_bool => 'is_enabled',
2600 WHERE is_user AND NOT is_enabled
2602 If a more complex combination is required, testing more conditions,
2603 then you should use the and/or operators:-
2608 -not_bool => { two=> { -rlike => 'bar' } },
2609 -not_bool => { three => [ { '=', 2 }, { '>', 5 } ] },
2620 (NOT ( three = ? OR three > ? ))
2623 =head2 Nested conditions, -and/-or prefixes
2625 So far, we've seen how multiple conditions are joined with a top-level
2626 C<AND>. We can change this by putting the different conditions we want in
2627 hashes and then putting those hashes in an array. For example:
2632 status => { -like => ['pending%', 'dispatched'] },
2636 status => 'unassigned',
2640 This data structure would create the following:
2642 $stmt = "WHERE ( user = ? AND ( status LIKE ? OR status LIKE ? ) )
2643 OR ( user = ? AND status = ? ) )";
2644 @bind = ('nwiger', 'pending', 'dispatched', 'robot', 'unassigned');
2647 Clauses in hashrefs or arrayrefs can be prefixed with an C<-and> or C<-or>
2648 to change the logic inside :
2654 -and => [ workhrs => {'>', 20}, geo => 'ASIA' ],
2655 -or => { workhrs => {'<', 50}, geo => 'EURO' },
2662 $stmt = "WHERE ( user = ?
2663 AND ( ( workhrs > ? AND geo = ? )
2664 OR ( workhrs < ? OR geo = ? ) ) )";
2665 @bind = ('nwiger', '20', 'ASIA', '50', 'EURO');
2667 =head3 Algebraic inconsistency, for historical reasons
2669 C<Important note>: when connecting several conditions, the C<-and->|C<-or>
2670 operator goes C<outside> of the nested structure; whereas when connecting
2671 several constraints on one column, the C<-and> operator goes
2672 C<inside> the arrayref. Here is an example combining both features :
2675 -and => [a => 1, b => 2],
2676 -or => [c => 3, d => 4],
2677 e => [-and => {-like => 'foo%'}, {-like => '%bar'} ]
2682 WHERE ( ( ( a = ? AND b = ? )
2683 OR ( c = ? OR d = ? )
2684 OR ( e LIKE ? AND e LIKE ? ) ) )
2686 This difference in syntax is unfortunate but must be preserved for
2687 historical reasons. So be careful : the two examples below would
2688 seem algebraically equivalent, but they are not
2691 { -like => 'foo%' },
2692 { -like => '%bar' },
2694 # yields : WHERE ( ( col LIKE ? AND col LIKE ? ) )
2697 { col => { -like => 'foo%' } },
2698 { col => { -like => '%bar' } },
2700 # yields : WHERE ( ( col LIKE ? OR col LIKE ? ) )
2703 =head2 Literal SQL and value type operators
2705 The basic premise of SQL::Abstract is that in WHERE specifications the "left
2706 side" is a column name and the "right side" is a value (normally rendered as
2707 a placeholder). This holds true for both hashrefs and arrayref pairs as you
2708 see in the L</WHERE CLAUSES> examples above. Sometimes it is necessary to
2709 alter this behavior. There are several ways of doing so.
2713 This is a virtual operator that signals the string to its right side is an
2714 identifier (a column name) and not a value. For example to compare two
2715 columns you would write:
2718 priority => { '<', 2 },
2719 requestor => { -ident => 'submitter' },
2724 $stmt = "WHERE priority < ? AND requestor = submitter";
2727 If you are maintaining legacy code you may see a different construct as
2728 described in L</Deprecated usage of Literal SQL>, please use C<-ident> in new
2733 This is a virtual operator that signals that the construct to its right side
2734 is a value to be passed to DBI. This is for example necessary when you want
2735 to write a where clause against an array (for RDBMS that support such
2736 datatypes). For example:
2739 array => { -value => [1, 2, 3] }
2744 $stmt = 'WHERE array = ?';
2745 @bind = ([1, 2, 3]);
2747 Note that if you were to simply say:
2753 the result would probably not be what you wanted:
2755 $stmt = 'WHERE array = ? OR array = ? OR array = ?';
2760 Finally, sometimes only literal SQL will do. To include a random snippet
2761 of SQL verbatim, you specify it as a scalar reference. Consider this only
2762 as a last resort. Usually there is a better way. For example:
2765 priority => { '<', 2 },
2766 requestor => { -in => \'(SELECT name FROM hitmen)' },
2771 $stmt = "WHERE priority < ? AND requestor IN (SELECT name FROM hitmen)"
2774 Note that in this example, you only get one bind parameter back, since
2775 the verbatim SQL is passed as part of the statement.
2779 Never use untrusted input as a literal SQL argument - this is a massive
2780 security risk (there is no way to check literal snippets for SQL
2781 injections and other nastyness). If you need to deal with untrusted input
2782 use literal SQL with placeholders as described next.
2784 =head3 Literal SQL with placeholders and bind values (subqueries)
2786 If the literal SQL to be inserted has placeholders and bind values,
2787 use a reference to an arrayref (yes this is a double reference --
2788 not so common, but perfectly legal Perl). For example, to find a date
2789 in Postgres you can use something like this:
2792 date_column => \[ "= date '2008-09-30' - ?::integer", 10 ]
2797 $stmt = "WHERE ( date_column = date '2008-09-30' - ?::integer )"
2800 Note that you must pass the bind values in the same format as they are returned
2801 by L<where|/where(\%where, $order)>. This means that if you set L</bindtype>
2802 to C<columns>, you must provide the bind values in the
2803 C<< [ column_meta => value ] >> format, where C<column_meta> is an opaque
2804 scalar value; most commonly the column name, but you can use any scalar value
2805 (including references and blessed references), L<SQL::Abstract> will simply
2806 pass it through intact. So if C<bindtype> is set to C<columns> the above
2807 example will look like:
2810 date_column => \[ "= date '2008-09-30' - ?::integer", [ {} => 10 ] ]
2813 Literal SQL is especially useful for nesting parenthesized clauses in the
2814 main SQL query. Here is a first example :
2816 my ($sub_stmt, @sub_bind) = ("SELECT c1 FROM t1 WHERE c2 < ? AND c3 LIKE ?",
2820 bar => \["IN ($sub_stmt)" => @sub_bind],
2825 $stmt = "WHERE (foo = ? AND bar IN (SELECT c1 FROM t1
2826 WHERE c2 < ? AND c3 LIKE ?))";
2827 @bind = (1234, 100, "foo%");
2829 Other subquery operators, like for example C<"E<gt> ALL"> or C<"NOT IN">,
2830 are expressed in the same way. Of course the C<$sub_stmt> and
2831 its associated bind values can be generated through a former call
2834 my ($sub_stmt, @sub_bind)
2835 = $sql->select("t1", "c1", {c2 => {"<" => 100},
2836 c3 => {-like => "foo%"}});
2839 bar => \["> ALL ($sub_stmt)" => @sub_bind],
2842 In the examples above, the subquery was used as an operator on a column;
2843 but the same principle also applies for a clause within the main C<%where>
2844 hash, like an EXISTS subquery :
2846 my ($sub_stmt, @sub_bind)
2847 = $sql->select("t1", "*", {c1 => 1, c2 => \"> t0.c0"});
2848 my %where = ( -and => [
2850 \["EXISTS ($sub_stmt)" => @sub_bind],
2855 $stmt = "WHERE (foo = ? AND EXISTS (SELECT * FROM t1
2856 WHERE c1 = ? AND c2 > t0.c0))";
2860 Observe that the condition on C<c2> in the subquery refers to
2861 column C<t0.c0> of the main query : this is I<not> a bind
2862 value, so we have to express it through a scalar ref.
2863 Writing C<< c2 => {">" => "t0.c0"} >> would have generated
2864 C<< c2 > ? >> with bind value C<"t0.c0"> ... not exactly
2865 what we wanted here.
2867 Finally, here is an example where a subquery is used
2868 for expressing unary negation:
2870 my ($sub_stmt, @sub_bind)
2871 = $sql->where({age => [{"<" => 10}, {">" => 20}]});
2872 $sub_stmt =~ s/^ where //i; # don't want "WHERE" in the subclause
2874 lname => {like => '%son%'},
2875 \["NOT ($sub_stmt)" => @sub_bind],
2880 $stmt = "lname LIKE ? AND NOT ( age < ? OR age > ? )"
2881 @bind = ('%son%', 10, 20)
2883 =head3 Deprecated usage of Literal SQL
2885 Below are some examples of archaic use of literal SQL. It is shown only as
2886 reference for those who deal with legacy code. Each example has a much
2887 better, cleaner and safer alternative that users should opt for in new code.
2893 my %where = ( requestor => \'IS NOT NULL' )
2895 $stmt = "WHERE requestor IS NOT NULL"
2897 This used to be the way of generating NULL comparisons, before the handling
2898 of C<undef> got formalized. For new code please use the superior syntax as
2899 described in L</Tests for NULL values>.
2903 my %where = ( requestor => \'= submitter' )
2905 $stmt = "WHERE requestor = submitter"
2907 This used to be the only way to compare columns. Use the superior L</-ident>
2908 method for all new code. For example an identifier declared in such a way
2909 will be properly quoted if L</quote_char> is properly set, while the legacy
2910 form will remain as supplied.
2914 my %where = ( is_ready => \"", completed => { '>', '2012-12-21' } )
2916 $stmt = "WHERE completed > ? AND is_ready"
2917 @bind = ('2012-12-21')
2919 Using an empty string literal used to be the only way to express a boolean.
2920 For all new code please use the much more readable
2921 L<-bool|/Unary operators: bool> operator.
2927 These pages could go on for a while, since the nesting of the data
2928 structures this module can handle are pretty much unlimited (the
2929 module implements the C<WHERE> expansion as a recursive function
2930 internally). Your best bet is to "play around" with the module a
2931 little to see how the data structures behave, and choose the best
2932 format for your data based on that.
2934 And of course, all the values above will probably be replaced with
2935 variables gotten from forms or the command line. After all, if you
2936 knew everything ahead of time, you wouldn't have to worry about
2937 dynamically-generating SQL and could just hardwire it into your
2940 =head1 ORDER BY CLAUSES
2942 Some functions take an order by clause. This can either be a scalar (just a
2943 column name), a hashref of C<< { -desc => 'col' } >> or C<< { -asc => 'col' }
2944 >>, a scalarref, an arrayref-ref, or an arrayref of any of the previous
2947 Given | Will Generate
2948 ---------------------------------------------------------------
2950 'colA' | ORDER BY colA
2952 [qw/colA colB/] | ORDER BY colA, colB
2954 {-asc => 'colA'} | ORDER BY colA ASC
2956 {-desc => 'colB'} | ORDER BY colB DESC
2958 ['colA', {-asc => 'colB'}] | ORDER BY colA, colB ASC
2960 { -asc => [qw/colA colB/] } | ORDER BY colA ASC, colB ASC
2962 \'colA DESC' | ORDER BY colA DESC
2964 \[ 'FUNC(colA, ?)', $x ] | ORDER BY FUNC(colA, ?)
2965 | /* ...with $x bound to ? */
2968 { -asc => 'colA' }, | colA ASC,
2969 { -desc => [qw/colB/] }, | colB DESC,
2970 { -asc => [qw/colC colD/] },| colC ASC, colD ASC,
2971 \'colE DESC', | colE DESC,
2972 \[ 'FUNC(colF, ?)', $x ], | FUNC(colF, ?)
2973 ] | /* ...with $x bound to ? */
2974 ===============================================================
2978 =head1 SPECIAL OPERATORS
2980 my $sqlmaker = SQL::Abstract->new(special_ops => [
2984 my ($self, $field, $op, $arg) = @_;
2990 handler => 'method_name',
2994 A "special operator" is a SQL syntactic clause that can be
2995 applied to a field, instead of a usual binary operator.
2998 WHERE field IN (?, ?, ?)
2999 WHERE field BETWEEN ? AND ?
3000 WHERE MATCH(field) AGAINST (?, ?)
3002 Special operators IN and BETWEEN are fairly standard and therefore
3003 are builtin within C<SQL::Abstract> (as the overridable methods
3004 C<_where_field_IN> and C<_where_field_BETWEEN>). For other operators,
3005 like the MATCH .. AGAINST example above which is specific to MySQL,
3006 you can write your own operator handlers - supply a C<special_ops>
3007 argument to the C<new> method. That argument takes an arrayref of
3008 operator definitions; each operator definition is a hashref with two
3015 the regular expression to match the operator
3019 Either a coderef or a plain scalar method name. In both cases
3020 the expected return is C<< ($sql, @bind) >>.
3022 When supplied with a method name, it is simply called on the
3023 L<SQL::Abstract> object as:
3025 $self->$method_name($field, $op, $arg)
3029 $field is the LHS of the operator
3030 $op is the part that matched the handler regex
3033 When supplied with a coderef, it is called as:
3035 $coderef->($self, $field, $op, $arg)
3040 For example, here is an implementation
3041 of the MATCH .. AGAINST syntax for MySQL
3043 my $sqlmaker = SQL::Abstract->new(special_ops => [
3045 # special op for MySql MATCH (field) AGAINST(word1, word2, ...)
3046 {regex => qr/^match$/i,
3048 my ($self, $field, $op, $arg) = @_;
3049 $arg = [$arg] if not ref $arg;
3050 my $label = $self->_quote($field);
3051 my ($placeholder) = $self->_convert('?');
3052 my $placeholders = join ", ", (($placeholder) x @$arg);
3053 my $sql = $self->_sqlcase('match') . " ($label) "
3054 . $self->_sqlcase('against') . " ($placeholders) ";
3055 my @bind = $self->_bindtype($field, @$arg);
3056 return ($sql, @bind);
3063 =head1 UNARY OPERATORS
3065 my $sqlmaker = SQL::Abstract->new(unary_ops => [
3069 my ($self, $op, $arg) = @_;
3075 handler => 'method_name',
3079 A "unary operator" is a SQL syntactic clause that can be
3080 applied to a field - the operator goes before the field
3082 You can write your own operator handlers - supply a C<unary_ops>
3083 argument to the C<new> method. That argument takes an arrayref of
3084 operator definitions; each operator definition is a hashref with two
3091 the regular expression to match the operator
3095 Either a coderef or a plain scalar method name. In both cases
3096 the expected return is C<< $sql >>.
3098 When supplied with a method name, it is simply called on the
3099 L<SQL::Abstract> object as:
3101 $self->$method_name($op, $arg)
3105 $op is the part that matched the handler regex
3106 $arg is the RHS or argument of the operator
3108 When supplied with a coderef, it is called as:
3110 $coderef->($self, $op, $arg)
3118 Thanks to some benchmarking by Mark Stosberg, it turns out that
3119 this module is many orders of magnitude faster than using C<DBIx::Abstract>.
3120 I must admit this wasn't an intentional design issue, but it's a
3121 byproduct of the fact that you get to control your C<DBI> handles
3124 To maximize performance, use a code snippet like the following:
3126 # prepare a statement handle using the first row
3127 # and then reuse it for the rest of the rows
3129 for my $href (@array_of_hashrefs) {
3130 $stmt ||= $sql->insert('table', $href);
3131 $sth ||= $dbh->prepare($stmt);
3132 $sth->execute($sql->values($href));
3135 The reason this works is because the keys in your C<$href> are sorted
3136 internally by B<SQL::Abstract>. Thus, as long as your data retains
3137 the same structure, you only have to generate the SQL the first time
3138 around. On subsequent queries, simply use the C<values> function provided
3139 by this module to return your values in the correct order.
3141 However this depends on the values having the same type - if, for
3142 example, the values of a where clause may either have values
3143 (resulting in sql of the form C<column = ?> with a single bind
3144 value), or alternatively the values might be C<undef> (resulting in
3145 sql of the form C<column IS NULL> with no bind value) then the
3146 caching technique suggested will not work.
3150 If you use my C<CGI::FormBuilder> module at all, you'll hopefully
3151 really like this part (I do, at least). Building up a complex query
3152 can be as simple as the following:
3159 use CGI::FormBuilder;
3162 my $form = CGI::FormBuilder->new(...);
3163 my $sql = SQL::Abstract->new;
3165 if ($form->submitted) {
3166 my $field = $form->field;
3167 my $id = delete $field->{id};
3168 my($stmt, @bind) = $sql->update('table', $field, {id => $id});
3171 Of course, you would still have to connect using C<DBI> to run the
3172 query, but the point is that if you make your form look like your
3173 table, the actual query script can be extremely simplistic.
3175 If you're B<REALLY> lazy (I am), check out C<HTML::QuickTable> for
3176 a fast interface to returning and formatting data. I frequently
3177 use these three modules together to write complex database query
3178 apps in under 50 lines.
3180 =head1 HOW TO CONTRIBUTE
3182 Contributions are always welcome, in all usable forms (we especially
3183 welcome documentation improvements). The delivery methods include git-
3184 or unified-diff formatted patches, GitHub pull requests, or plain bug
3185 reports either via RT or the Mailing list. Contributors are generally
3186 granted full access to the official repository after their first several
3187 patches pass successful review.
3189 This project is maintained in a git repository. The code and related tools are
3190 accessible at the following locations:
3194 =item * Official repo: L<git://git.shadowcat.co.uk/dbsrgits/SQL-Abstract.git>
3196 =item * Official gitweb: L<http://git.shadowcat.co.uk/gitweb/gitweb.cgi?p=dbsrgits/SQL-Abstract.git>
3198 =item * GitHub mirror: L<https://github.com/dbsrgits/sql-abstract>
3200 =item * Authorized committers: L<ssh://dbsrgits@git.shadowcat.co.uk/SQL-Abstract.git>
3206 Version 1.50 was a major internal refactoring of C<SQL::Abstract>.
3207 Great care has been taken to preserve the I<published> behavior
3208 documented in previous versions in the 1.* family; however,
3209 some features that were previously undocumented, or behaved
3210 differently from the documentation, had to be changed in order
3211 to clarify the semantics. Hence, client code that was relying
3212 on some dark areas of C<SQL::Abstract> v1.*
3213 B<might behave differently> in v1.50.
3215 The main changes are :
3221 support for literal SQL through the C<< \ [ $sql, @bind ] >> syntax.
3225 support for the { operator => \"..." } construct (to embed literal SQL)
3229 support for the { operator => \["...", @bind] } construct (to embed literal SQL with bind values)
3233 optional support for L<array datatypes|/"Inserting and Updating Arrays">
3237 defensive programming : check arguments
3241 fixed bug with global logic, which was previously implemented
3242 through global variables yielding side-effects. Prior versions would
3243 interpret C<< [ {cond1, cond2}, [cond3, cond4] ] >>
3244 as C<< "(cond1 AND cond2) OR (cond3 AND cond4)" >>.
3245 Now this is interpreted
3246 as C<< "(cond1 AND cond2) OR (cond3 OR cond4)" >>.
3251 fixed semantics of _bindtype on array args
3255 dropped the C<_anoncopy> of the %where tree. No longer necessary,
3256 we just avoid shifting arrays within that tree.
3260 dropped the C<_modlogic> function
3264 =head1 ACKNOWLEDGEMENTS
3266 There are a number of individuals that have really helped out with
3267 this module. Unfortunately, most of them submitted bugs via CPAN
3268 so I have no idea who they are! But the people I do know are:
3270 Ash Berlin (order_by hash term support)
3271 Matt Trout (DBIx::Class support)
3272 Mark Stosberg (benchmarking)
3273 Chas Owens (initial "IN" operator support)
3274 Philip Collins (per-field SQL functions)
3275 Eric Kolve (hashref "AND" support)
3276 Mike Fragassi (enhancements to "BETWEEN" and "LIKE")
3277 Dan Kubb (support for "quote_char" and "name_sep")
3278 Guillermo Roditi (patch to cleanup "IN" and "BETWEEN", fix and tests for _order_by)
3279 Laurent Dami (internal refactoring, extensible list of special operators, literal SQL)
3280 Norbert Buchmuller (support for literal SQL in hashpair, misc. fixes & tests)
3281 Peter Rabbitson (rewrite of SQLA::Test, misc. fixes & tests)
3282 Oliver Charles (support for "RETURNING" after "INSERT")
3288 L<DBIx::Class>, L<DBIx::Abstract>, L<CGI::FormBuilder>, L<HTML::QuickTable>.
3292 Copyright (c) 2001-2007 Nathan Wiger <nwiger@cpan.org>. All Rights Reserved.
3294 This module is actively maintained by Matt Trout <mst@shadowcatsystems.co.uk>
3296 For support, your best bet is to try the C<DBIx::Class> users mailing list.
3297 While not an official support venue, C<DBIx::Class> makes heavy use of
3298 C<SQL::Abstract>, and as such list members there are very familiar with
3299 how to create queries.
3303 This module is free software; you may copy this under the same
3304 terms as perl itself (either the GNU General Public License or
3305 the Artistic License)