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 = '1.81';
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) : $sql;
230 # Used by DBIx::Class::SQLMaker->insert
231 sub _insert_returning { shift->_returning(@_) }
234 my ($self, $options) = @_;
236 my $f = $options->{returning};
238 my $fieldlist = $self->_SWITCH_refkind($f, {
239 ARRAYREF => sub {join ', ', map { $self->_quote($_) } @$f;},
240 SCALAR => sub {$self->_quote($f)},
241 SCALARREF => sub {$$f},
243 return $self->_sqlcase(' returning ') . $fieldlist;
246 sub _insert_HASHREF { # explicit list of fields and then values
247 my ($self, $data) = @_;
249 my @fields = sort keys %$data;
251 my ($sql, @bind) = $self->_insert_values($data);
254 $_ = $self->_quote($_) foreach @fields;
255 $sql = "( ".join(", ", @fields).") ".$sql;
257 return ($sql, @bind);
260 sub _insert_ARRAYREF { # just generate values(?,?) part (no list of fields)
261 my ($self, $data) = @_;
263 # no names (arrayref) so can't generate bindtype
264 $self->{bindtype} ne 'columns'
265 or belch "can't do 'columns' bindtype when called with arrayref";
267 my (@values, @all_bind);
268 foreach my $value (@$data) {
269 my ($values, @bind) = $self->_insert_value(undef, $value);
270 push @values, $values;
271 push @all_bind, @bind;
273 my $sql = $self->_sqlcase('values')." ( ".join(", ", @values)." )";
274 return ($sql, @all_bind);
277 sub _insert_ARRAYREFREF { # literal SQL with bind
278 my ($self, $data) = @_;
280 my ($sql, @bind) = @${$data};
281 $self->_assert_bindval_matches_bindtype(@bind);
283 return ($sql, @bind);
287 sub _insert_SCALARREF { # literal SQL without bind
288 my ($self, $data) = @_;
294 my ($self, $data) = @_;
296 my (@values, @all_bind);
297 foreach my $column (sort keys %$data) {
298 my ($values, @bind) = $self->_insert_value($column, $data->{$column});
299 push @values, $values;
300 push @all_bind, @bind;
302 my $sql = $self->_sqlcase('values')." ( ".join(", ", @values)." )";
303 return ($sql, @all_bind);
307 my ($self, $column, $v) = @_;
309 my (@values, @all_bind);
310 $self->_SWITCH_refkind($v, {
313 if ($self->{array_datatypes}) { # if array datatype are activated
315 push @all_bind, $self->_bindtype($column, $v);
317 else { # else literal SQL with bind
318 my ($sql, @bind) = @$v;
319 $self->_assert_bindval_matches_bindtype(@bind);
321 push @all_bind, @bind;
325 ARRAYREFREF => sub { # literal SQL with bind
326 my ($sql, @bind) = @${$v};
327 $self->_assert_bindval_matches_bindtype(@bind);
329 push @all_bind, @bind;
332 # THINK : anything useful to do with a HASHREF ?
333 HASHREF => sub { # (nothing, but old SQLA passed it through)
334 #TODO in SQLA >= 2.0 it will die instead
335 belch "HASH ref as bind value in insert is not supported";
337 push @all_bind, $self->_bindtype($column, $v);
340 SCALARREF => sub { # literal SQL without bind
344 SCALAR_or_UNDEF => sub {
346 push @all_bind, $self->_bindtype($column, $v);
351 my $sql = join(", ", @values);
352 return ($sql, @all_bind);
357 #======================================================================
359 #======================================================================
364 my $table = $self->_table(shift);
365 my $data = shift || return;
369 # first build the 'SET' part of the sql statement
370 my (@set, @all_bind);
371 puke "Unsupported data type specified to \$sql->update"
372 unless ref $data eq 'HASH';
374 for my $k (sort keys %$data) {
377 my $label = $self->_quote($k);
379 $self->_SWITCH_refkind($v, {
381 if ($self->{array_datatypes}) { # array datatype
382 push @set, "$label = ?";
383 push @all_bind, $self->_bindtype($k, $v);
385 else { # literal SQL with bind
386 my ($sql, @bind) = @$v;
387 $self->_assert_bindval_matches_bindtype(@bind);
388 push @set, "$label = $sql";
389 push @all_bind, @bind;
392 ARRAYREFREF => sub { # literal SQL with bind
393 my ($sql, @bind) = @${$v};
394 $self->_assert_bindval_matches_bindtype(@bind);
395 push @set, "$label = $sql";
396 push @all_bind, @bind;
398 SCALARREF => sub { # literal SQL without bind
399 push @set, "$label = $$v";
402 my ($op, $arg, @rest) = %$v;
404 puke 'Operator calls in update must be in the form { -op => $arg }'
405 if (@rest or not $op =~ /^\-(.+)/);
407 local $self->{_nested_func_lhs} = $k;
408 my ($sql, @bind) = $self->_where_unary_op ($1, $arg);
410 push @set, "$label = $sql";
411 push @all_bind, @bind;
413 SCALAR_or_UNDEF => sub {
414 push @set, "$label = ?";
415 push @all_bind, $self->_bindtype($k, $v);
421 my $sql = $self->_sqlcase('update') . " $table " . $self->_sqlcase('set ')
425 my($where_sql, @where_bind) = $self->where($where);
427 push @all_bind, @where_bind;
430 if ($options->{returning}) {
431 my ($returning_sql, @returning_bind) = $self->_update_returning ($options);
432 $sql .= $returning_sql;
433 push @all_bind, @returning_bind;
436 return wantarray ? ($sql, @all_bind) : $sql;
439 sub _update_returning { shift->_returning(@_) }
443 #======================================================================
445 #======================================================================
450 my $table = $self->_table(shift);
451 my $fields = shift || '*';
455 my($where_sql, @bind) = $self->where($where, $order);
457 my $f = (ref $fields eq 'ARRAY') ? join ', ', map { $self->_quote($_) } @$fields
459 my $sql = join(' ', $self->_sqlcase('select'), $f,
460 $self->_sqlcase('from'), $table)
463 return wantarray ? ($sql, @bind) : $sql;
466 #======================================================================
468 #======================================================================
473 my $table = $self->_table(shift);
477 my($where_sql, @bind) = $self->where($where);
478 my $sql = $self->_sqlcase('delete from') . " $table" . $where_sql;
480 return wantarray ? ($sql, @bind) : $sql;
484 #======================================================================
486 #======================================================================
490 # Finally, a separate routine just to handle WHERE clauses
492 my ($self, $where, $order) = @_;
495 my ($sql, @bind) = $self->_recurse_where($where);
496 $sql = $sql ? $self->_sqlcase(' where ') . "( $sql )" : '';
500 my ($order_sql, @order_bind) = $self->_order_by($order);
502 push @bind, @order_bind;
505 return wantarray ? ($sql, @bind) : $sql;
510 my ($self, $where, $logic) = @_;
512 # dispatch on appropriate method according to refkind of $where
513 my $method = $self->_METHOD_FOR_refkind("_where", $where);
515 my ($sql, @bind) = $self->$method($where, $logic);
517 # DBIx::Class used to call _recurse_where in scalar context
518 # something else might too...
520 return ($sql, @bind);
523 belch "Calling _recurse_where in scalar context is deprecated and will go away before 2.0";
530 #======================================================================
531 # WHERE: top-level ARRAYREF
532 #======================================================================
535 sub _where_ARRAYREF {
536 my ($self, $where, $logic) = @_;
538 $logic = uc($logic || $self->{logic});
539 $logic eq 'AND' or $logic eq 'OR' or puke "unknown logic: $logic";
541 my @clauses = @$where;
543 my (@sql_clauses, @all_bind);
544 # need to use while() so can shift() for pairs
546 my $el = shift @clauses;
548 $el = undef if (defined $el and ! length $el);
550 # switch according to kind of $el and get corresponding ($sql, @bind)
551 my ($sql, @bind) = $self->_SWITCH_refkind($el, {
553 # skip empty elements, otherwise get invalid trailing AND stuff
554 ARRAYREF => sub {$self->_recurse_where($el) if @$el},
558 $self->_assert_bindval_matches_bindtype(@b);
562 HASHREF => sub {$self->_recurse_where($el, 'and') if %$el},
564 SCALARREF => sub { ($$el); },
567 # top-level arrayref with scalars, recurse in pairs
568 $self->_recurse_where({$el => shift(@clauses)})
571 UNDEF => sub {puke "Supplying an empty left hand side argument is not supported in array-pairs" },
575 push @sql_clauses, $sql;
576 push @all_bind, @bind;
580 return $self->_join_sql_clauses($logic, \@sql_clauses, \@all_bind);
583 #======================================================================
584 # WHERE: top-level ARRAYREFREF
585 #======================================================================
587 sub _where_ARRAYREFREF {
588 my ($self, $where) = @_;
589 my ($sql, @bind) = @$$where;
590 $self->_assert_bindval_matches_bindtype(@bind);
591 return ($sql, @bind);
594 #======================================================================
595 # WHERE: top-level HASHREF
596 #======================================================================
599 my ($self, $where) = @_;
600 my (@sql_clauses, @all_bind);
602 for my $k (sort keys %$where) {
603 my $v = $where->{$k};
605 # ($k => $v) is either a special unary op or a regular hashpair
606 my ($sql, @bind) = do {
608 # put the operator in canonical form
610 $op = substr $op, 1; # remove initial dash
611 $op =~ s/^\s+|\s+$//g;# remove leading/trailing space
612 $op =~ s/\s+/ /g; # compress whitespace
614 # so that -not_foo works correctly
615 $op =~ s/^not_/NOT /i;
617 $self->_debug("Unary OP(-$op) within hashref, recursing...");
618 my ($s, @b) = $self->_where_unary_op ($op, $v);
620 # top level vs nested
621 # we assume that handled unary ops will take care of their ()s
623 List::Util::first {$op =~ $_->{regex}} @{$self->{unary_ops}}
625 ( defined $self->{_nested_func_lhs} and $self->{_nested_func_lhs} eq $k )
631 if (is_literal_value ($v) ) {
632 belch 'Hash-pairs consisting of an empty string with a literal are deprecated, and will be removed in 2.0: use -and => [ $literal ] instead';
635 puke "Supplying an empty left hand side argument is not supported in hash-pairs";
639 my $method = $self->_METHOD_FOR_refkind("_where_hashpair", $v);
640 $self->$method($k, $v);
644 push @sql_clauses, $sql;
645 push @all_bind, @bind;
648 return $self->_join_sql_clauses('and', \@sql_clauses, \@all_bind);
651 sub _where_unary_op {
652 my ($self, $op, $rhs) = @_;
654 # top level special ops are illegal in general
655 # this includes the -ident/-value ops (dual purpose unary and special)
656 puke "Illegal use of top-level '-$op'"
657 if ! defined $self->{_nested_func_lhs} and List::Util::first {$op =~ $_->{regex}} @{$self->{special_ops}};
659 if (my $op_entry = List::Util::first {$op =~ $_->{regex}} @{$self->{unary_ops}}) {
660 my $handler = $op_entry->{handler};
662 if (not ref $handler) {
663 if ($op =~ s/ [_\s]? \d+ $//x ) {
664 belch 'Use of [and|or|nest]_N modifiers is deprecated and will be removed in SQLA v2.0. '
665 . "You probably wanted ...-and => [ -$op => COND1, -$op => COND2 ... ]";
667 return $self->$handler ($op, $rhs);
669 elsif (ref $handler eq 'CODE') {
670 return $handler->($self, $op, $rhs);
673 puke "Illegal handler for operator $op - expecting a method name or a coderef";
677 $self->_debug("Generic unary OP: $op - recursing as function");
679 $self->_assert_pass_injection_guard($op);
681 my ($sql, @bind) = $self->_SWITCH_refkind ($rhs, {
683 puke "Illegal use of top-level '-$op'"
684 unless defined $self->{_nested_func_lhs};
687 $self->_convert('?'),
688 $self->_bindtype($self->{_nested_func_lhs}, $rhs)
692 $self->_recurse_where ($rhs)
696 $sql = sprintf ('%s %s',
697 $self->_sqlcase($op),
701 return ($sql, @bind);
704 sub _where_op_ANDOR {
705 my ($self, $op, $v) = @_;
707 $self->_SWITCH_refkind($v, {
709 return $self->_where_ARRAYREF($v, $op);
713 return ( $op =~ /^or/i )
714 ? $self->_where_ARRAYREF( [ map { $_ => $v->{$_} } ( sort keys %$v ) ], $op )
715 : $self->_where_HASHREF($v);
719 puke "-$op => \\\$scalar makes little sense, use " .
721 ? '[ \$scalar, \%rest_of_conditions ] instead'
722 : '-and => [ \$scalar, \%rest_of_conditions ] instead'
727 puke "-$op => \\[...] makes little sense, use " .
729 ? '[ \[...], \%rest_of_conditions ] instead'
730 : '-and => [ \[...], \%rest_of_conditions ] instead'
734 SCALAR => sub { # permissively interpreted as SQL
735 puke "-$op => \$value makes little sense, use -bool => \$value instead";
739 puke "-$op => undef not supported";
745 my ($self, $op, $v) = @_;
747 $self->_SWITCH_refkind($v, {
749 SCALAR => sub { # permissively interpreted as SQL
750 belch "literal SQL should be -nest => \\'scalar' "
751 . "instead of -nest => 'scalar' ";
756 puke "-$op => undef not supported";
760 $self->_recurse_where ($v);
768 my ($self, $op, $v) = @_;
770 my ($s, @b) = $self->_SWITCH_refkind($v, {
771 SCALAR => sub { # interpreted as SQL column
772 $self->_convert($self->_quote($v));
776 puke "-$op => undef not supported";
780 $self->_recurse_where ($v);
784 $s = "(NOT $s)" if $op =~ /^not/i;
789 sub _where_op_IDENT {
791 my ($op, $rhs) = splice @_, -2;
792 if (! defined $rhs or length ref $rhs) {
793 puke "-$op requires a single plain scalar argument (a quotable identifier)";
796 # in case we are called as a top level special op (no '=')
799 $_ = $self->_convert($self->_quote($_)) for ($lhs, $rhs);
807 sub _where_op_VALUE {
809 my ($op, $rhs) = splice @_, -2;
811 # in case we are called as a top level special op (no '=')
815 if (! defined $rhs) {
817 ? $self->_convert($self->_quote($lhs)) . ' IS NULL'
824 ( defined $lhs ? $lhs : $self->{_nested_func_lhs} ),
831 $self->_convert($self->_quote($lhs)) . ' = ' . $self->_convert('?'),
835 $self->_convert('?'),
841 sub _where_hashpair_ARRAYREF {
842 my ($self, $k, $v) = @_;
845 my @v = @$v; # need copy because of shift below
846 $self->_debug("ARRAY($k) means distribute over elements");
848 # put apart first element if it is an operator (-and, -or)
850 (defined $v[0] && $v[0] =~ /^ - (?: AND|OR ) $/ix)
854 my @distributed = map { {$k => $_} } @v;
857 $self->_debug("OP($op) reinjected into the distributed array");
858 unshift @distributed, $op;
861 my $logic = $op ? substr($op, 1) : '';
863 return $self->_recurse_where(\@distributed, $logic);
866 $self->_debug("empty ARRAY($k) means 0=1");
867 return ($self->{sqlfalse});
871 sub _where_hashpair_HASHREF {
872 my ($self, $k, $v, $logic) = @_;
875 local $self->{_nested_func_lhs} = defined $self->{_nested_func_lhs}
876 ? $self->{_nested_func_lhs}
880 my ($all_sql, @all_bind);
882 for my $orig_op (sort keys %$v) {
883 my $val = $v->{$orig_op};
885 # put the operator in canonical form
888 # FIXME - we need to phase out dash-less ops
889 $op =~ s/^-//; # remove possible initial dash
890 $op =~ s/^\s+|\s+$//g;# remove leading/trailing space
891 $op =~ s/\s+/ /g; # compress whitespace
893 $self->_assert_pass_injection_guard($op);
896 $op =~ s/^is_not/IS NOT/i;
898 # so that -not_foo works correctly
899 $op =~ s/^not_/NOT /i;
901 # another retarded special case: foo => { $op => { -value => undef } }
902 if (ref $val eq 'HASH' and keys %$val == 1 and exists $val->{-value} and ! defined $val->{-value} ) {
908 # CASE: col-value logic modifiers
909 if ( $orig_op =~ /^ \- (and|or) $/xi ) {
910 ($sql, @bind) = $self->_where_hashpair_HASHREF($k, $val, $1);
912 # CASE: special operators like -in or -between
913 elsif ( my $special_op = List::Util::first {$op =~ $_->{regex}} @{$self->{special_ops}} ) {
914 my $handler = $special_op->{handler};
916 puke "No handler supplied for special operator $orig_op";
918 elsif (not ref $handler) {
919 ($sql, @bind) = $self->$handler ($k, $op, $val);
921 elsif (ref $handler eq 'CODE') {
922 ($sql, @bind) = $handler->($self, $k, $op, $val);
925 puke "Illegal handler for special operator $orig_op - expecting a method name or a coderef";
929 $self->_SWITCH_refkind($val, {
931 ARRAYREF => sub { # CASE: col => {op => \@vals}
932 ($sql, @bind) = $self->_where_field_op_ARRAYREF($k, $op, $val);
935 ARRAYREFREF => sub { # CASE: col => {op => \[$sql, @bind]} (literal SQL with bind)
936 my ($sub_sql, @sub_bind) = @$$val;
937 $self->_assert_bindval_matches_bindtype(@sub_bind);
938 $sql = join ' ', $self->_convert($self->_quote($k)),
939 $self->_sqlcase($op),
944 UNDEF => sub { # CASE: col => {op => undef} : sql "IS (NOT)? NULL"
946 $op =~ /^not$/i ? 'is not' # legacy
947 : $op =~ $self->{equality_op} ? 'is'
948 : $op =~ $self->{like_op} ? belch("Supplying an undefined argument to '@{[ uc $op]}' is deprecated") && 'is'
949 : $op =~ $self->{inequality_op} ? 'is not'
950 : $op =~ $self->{not_like_op} ? belch("Supplying an undefined argument to '@{[ uc $op]}' is deprecated") && 'is not'
951 : puke "unexpected operator '$orig_op' with undef operand";
953 $sql = $self->_quote($k) . $self->_sqlcase(" $is null");
956 FALLBACK => sub { # CASE: col => {op/func => $stuff}
957 ($sql, @bind) = $self->_where_unary_op ($op, $val);
960 $self->_convert($self->_quote($k)),
961 $self->{_nested_func_lhs} eq $k ? $sql : "($sql)", # top level vs nested
967 ($all_sql) = (defined $all_sql and $all_sql) ? $self->_join_sql_clauses($logic, [$all_sql, $sql], []) : $sql;
968 push @all_bind, @bind;
970 return ($all_sql, @all_bind);
973 sub _where_field_IS {
974 my ($self, $k, $op, $v) = @_;
976 my ($s) = $self->_SWITCH_refkind($v, {
979 $self->_convert($self->_quote($k)),
980 map { $self->_sqlcase($_)} ($op, 'null')
983 puke "$op can only take undef as argument";
990 sub _where_field_op_ARRAYREF {
991 my ($self, $k, $op, $vals) = @_;
993 my @vals = @$vals; #always work on a copy
996 $self->_debug(sprintf '%s means multiple elements: [ %s ]',
998 join (', ', map { defined $_ ? "'$_'" : 'NULL' } @vals ),
1001 # see if the first element is an -and/-or op
1003 if (defined $vals[0] && $vals[0] =~ /^ - ( AND|OR ) $/ix) {
1008 # a long standing API wart - an attempt to change this behavior during
1009 # the 1.50 series failed *spectacularly*. Warn instead and leave the
1014 (!$logic or $logic eq 'OR')
1016 ( $op =~ $self->{inequality_op} or $op =~ $self->{not_like_op} )
1019 belch "A multi-element arrayref as an argument to the inequality op '$o' "
1020 . 'is technically equivalent to an always-true 1=1 (you probably wanted '
1021 . "to say ...{ \$inequality_op => [ -and => \@values ] }... instead)"
1025 # distribute $op over each remaining member of @vals, append logic if exists
1026 return $self->_recurse_where([map { {$k => {$op, $_}} } @vals], $logic);
1030 # try to DWIM on equality operators
1032 $op =~ $self->{equality_op} ? $self->{sqlfalse}
1033 : $op =~ $self->{like_op} ? belch("Supplying an empty arrayref to '@{[ uc $op]}' is deprecated") && $self->{sqlfalse}
1034 : $op =~ $self->{inequality_op} ? $self->{sqltrue}
1035 : $op =~ $self->{not_like_op} ? belch("Supplying an empty arrayref to '@{[ uc $op]}' is deprecated") && $self->{sqltrue}
1036 : puke "operator '$op' applied on an empty array (field '$k')";
1041 sub _where_hashpair_SCALARREF {
1042 my ($self, $k, $v) = @_;
1043 $self->_debug("SCALAR($k) means literal SQL: $$v");
1044 my $sql = $self->_quote($k) . " " . $$v;
1048 # literal SQL with bind
1049 sub _where_hashpair_ARRAYREFREF {
1050 my ($self, $k, $v) = @_;
1051 $self->_debug("REF($k) means literal SQL: @${$v}");
1052 my ($sql, @bind) = @$$v;
1053 $self->_assert_bindval_matches_bindtype(@bind);
1054 $sql = $self->_quote($k) . " " . $sql;
1055 return ($sql, @bind );
1058 # literal SQL without bind
1059 sub _where_hashpair_SCALAR {
1060 my ($self, $k, $v) = @_;
1061 $self->_debug("NOREF($k) means simple key=val: $k $self->{cmp} $v");
1062 my $sql = join ' ', $self->_convert($self->_quote($k)),
1063 $self->_sqlcase($self->{cmp}),
1064 $self->_convert('?');
1065 my @bind = $self->_bindtype($k, $v);
1066 return ( $sql, @bind);
1070 sub _where_hashpair_UNDEF {
1071 my ($self, $k, $v) = @_;
1072 $self->_debug("UNDEF($k) means IS NULL");
1073 my $sql = $self->_quote($k) . $self->_sqlcase(' is null');
1077 #======================================================================
1078 # WHERE: TOP-LEVEL OTHERS (SCALARREF, SCALAR, UNDEF)
1079 #======================================================================
1082 sub _where_SCALARREF {
1083 my ($self, $where) = @_;
1086 $self->_debug("SCALAR(*top) means literal SQL: $$where");
1092 my ($self, $where) = @_;
1095 $self->_debug("NOREF(*top) means literal SQL: $where");
1106 #======================================================================
1107 # WHERE: BUILTIN SPECIAL OPERATORS (-in, -between)
1108 #======================================================================
1111 sub _where_field_BETWEEN {
1112 my ($self, $k, $op, $vals) = @_;
1114 my ($label, $and, $placeholder);
1115 $label = $self->_convert($self->_quote($k));
1116 $and = ' ' . $self->_sqlcase('and') . ' ';
1117 $placeholder = $self->_convert('?');
1118 $op = $self->_sqlcase($op);
1120 my $invalid_args = "Operator '$op' requires either an arrayref with two defined values or expressions, or a single literal scalarref/arrayref-ref";
1122 my ($clause, @bind) = $self->_SWITCH_refkind($vals, {
1123 ARRAYREFREF => sub {
1124 my ($s, @b) = @$$vals;
1125 $self->_assert_bindval_matches_bindtype(@b);
1132 puke $invalid_args if @$vals != 2;
1134 my (@all_sql, @all_bind);
1135 foreach my $val (@$vals) {
1136 my ($sql, @bind) = $self->_SWITCH_refkind($val, {
1138 return ($placeholder, $self->_bindtype($k, $val) );
1143 ARRAYREFREF => sub {
1144 my ($sql, @bind) = @$$val;
1145 $self->_assert_bindval_matches_bindtype(@bind);
1146 return ($sql, @bind);
1149 my ($func, $arg, @rest) = %$val;
1150 puke ("Only simple { -func => arg } functions accepted as sub-arguments to BETWEEN")
1151 if (@rest or $func !~ /^ \- (.+)/x);
1152 $self->_where_unary_op ($1 => $arg);
1158 push @all_sql, $sql;
1159 push @all_bind, @bind;
1163 (join $and, @all_sql),
1172 my $sql = "( $label $op $clause )";
1173 return ($sql, @bind)
1177 sub _where_field_IN {
1178 my ($self, $k, $op, $vals) = @_;
1180 # backwards compatibility : if scalar, force into an arrayref
1181 $vals = [$vals] if defined $vals && ! ref $vals;
1183 my ($label) = $self->_convert($self->_quote($k));
1184 my ($placeholder) = $self->_convert('?');
1185 $op = $self->_sqlcase($op);
1187 my ($sql, @bind) = $self->_SWITCH_refkind($vals, {
1188 ARRAYREF => sub { # list of choices
1189 if (@$vals) { # nonempty list
1190 my (@all_sql, @all_bind);
1192 for my $val (@$vals) {
1193 my ($sql, @bind) = $self->_SWITCH_refkind($val, {
1195 return ($placeholder, $val);
1200 ARRAYREFREF => sub {
1201 my ($sql, @bind) = @$$val;
1202 $self->_assert_bindval_matches_bindtype(@bind);
1203 return ($sql, @bind);
1206 my ($func, $arg, @rest) = %$val;
1207 puke ("Only simple { -func => arg } functions accepted as sub-arguments to IN")
1208 if (@rest or $func !~ /^ \- (.+)/x);
1209 $self->_where_unary_op ($1 => $arg);
1213 'SQL::Abstract before v1.75 used to generate incorrect SQL when the '
1214 . "-$op operator was given an undef-containing list: !!!AUDIT YOUR CODE "
1215 . 'AND DATA!!! (the upcoming Data::Query-based version of SQL::Abstract '
1216 . 'will emit the logically correct SQL instead of raising this exception)'
1220 push @all_sql, $sql;
1221 push @all_bind, @bind;
1225 sprintf ('%s %s ( %s )',
1228 join (', ', @all_sql)
1230 $self->_bindtype($k, @all_bind),
1233 else { # empty list : some databases won't understand "IN ()", so DWIM
1234 my $sql = ($op =~ /\bnot\b/i) ? $self->{sqltrue} : $self->{sqlfalse};
1239 SCALARREF => sub { # literal SQL
1240 my $sql = $self->_open_outer_paren ($$vals);
1241 return ("$label $op ( $sql )");
1243 ARRAYREFREF => sub { # literal SQL with bind
1244 my ($sql, @bind) = @$$vals;
1245 $self->_assert_bindval_matches_bindtype(@bind);
1246 $sql = $self->_open_outer_paren ($sql);
1247 return ("$label $op ( $sql )", @bind);
1251 puke "Argument passed to the '$op' operator can not be undefined";
1255 puke "special op $op requires an arrayref (or scalarref/arrayref-ref)";
1259 return ($sql, @bind);
1262 # Some databases (SQLite) treat col IN (1, 2) different from
1263 # col IN ( (1, 2) ). Use this to strip all outer parens while
1264 # adding them back in the corresponding method
1265 sub _open_outer_paren {
1266 my ($self, $sql) = @_;
1268 while ( my ($inner) = $sql =~ /^ \s* \( (.*) \) \s* $/xs ) {
1270 # there are closing parens inside, need the heavy duty machinery
1271 # to reevaluate the extraction starting from $sql (full reevaluation)
1272 if ( $inner =~ /\)/ ) {
1273 require Text::Balanced;
1275 my (undef, $remainder) = do {
1276 # idiotic design - writes to $@ but *DOES NOT* throw exceptions
1278 Text::Balanced::extract_bracketed( $sql, '()', qr/\s*/ );
1281 # the entire expression needs to be a balanced bracketed thing
1282 # (after an extract no remainder sans trailing space)
1283 last if defined $remainder and $remainder =~ /\S/;
1293 #======================================================================
1295 #======================================================================
1298 my ($self, $arg) = @_;
1301 for my $c ($self->_order_by_chunks ($arg) ) {
1302 $self->_SWITCH_refkind ($c, {
1303 SCALAR => sub { push @sql, $c },
1304 ARRAYREF => sub { push @sql, shift @$c; push @bind, @$c },
1310 $self->_sqlcase(' order by'),
1316 return wantarray ? ($sql, @bind) : $sql;
1319 sub _order_by_chunks {
1320 my ($self, $arg) = @_;
1322 return $self->_SWITCH_refkind($arg, {
1325 map { $self->_order_by_chunks ($_ ) } @$arg;
1328 ARRAYREFREF => sub {
1329 my ($s, @b) = @$$arg;
1330 $self->_assert_bindval_matches_bindtype(@b);
1334 SCALAR => sub {$self->_quote($arg)},
1336 UNDEF => sub {return () },
1338 SCALARREF => sub {$$arg}, # literal SQL, no quoting
1341 return () unless %$arg;
1343 my ($direction, $nulls, $val);
1344 foreach my $key (keys %$arg) {
1345 if ( $key =~ /^-(desc|asc)/i ) {
1346 puke "hash passed to _order_by must have exactly one of -desc or -asc"
1347 if defined $direction;
1349 $val = $arg->{$key};
1350 } elsif ($key =~ /^-nulls$/i) {
1351 $nulls = $arg->{$key};
1352 puke "invalid value for -nulls" unless $nulls =~ /\A(?:first|last)\z/i;
1354 puke "invalid key in hash passed to _order_by";
1357 puke "hash passed to _order_by must have exactly one of -desc or -asc"
1358 unless defined $direction;
1361 for my $c ($self->_order_by_chunks ($val)) {
1364 $self->_SWITCH_refkind ($c, {
1369 ($sql, @bind) = @$c;
1373 $sql .= ' ' . $self->_sqlcase($direction);
1374 $sql .= ' ' . $self->_sqlcase("nulls $nulls")
1377 push @ret, [ $sql, @bind];
1386 #======================================================================
1387 # DATASOURCE (FOR NOW, JUST PLAIN TABLE OR LIST OF TABLES)
1388 #======================================================================
1393 $self->_SWITCH_refkind($from, {
1394 ARRAYREF => sub {join ', ', map { $self->_quote($_) } @$from;},
1395 SCALAR => sub {$self->_quote($from)},
1396 SCALARREF => sub {$$from},
1401 #======================================================================
1403 #======================================================================
1405 # highly optimized, as it's called way too often
1407 # my ($self, $label) = @_;
1409 return '' unless defined $_[1];
1410 return ${$_[1]} if ref($_[1]) eq 'SCALAR';
1412 $_[0]->{quote_char} or
1413 ($_[0]->_assert_pass_injection_guard($_[1]), return $_[1]);
1415 my $qref = ref $_[0]->{quote_char};
1417 !$qref ? ($_[0]->{quote_char}, $_[0]->{quote_char})
1418 : ($qref eq 'ARRAY') ? @{$_[0]->{quote_char}}
1419 : puke "Unsupported quote_char format: $_[0]->{quote_char}";
1421 my $esc = $_[0]->{escape_char} || $r;
1423 # parts containing * are naturally unquoted
1424 return join( $_[0]->{name_sep}||'', map
1425 +( $_ eq '*' ? $_ : do { (my $n = $_) =~ s/(\Q$esc\E|\Q$r\E)/$esc$1/g; $l . $n . $r } ),
1426 ( $_[0]->{name_sep} ? split (/\Q$_[0]->{name_sep}\E/, $_[1] ) : $_[1] )
1431 # Conversion, if applicable
1433 #my ($self, $arg) = @_;
1434 if ($_[0]->{convert}) {
1435 return $_[0]->_sqlcase($_[0]->{convert}) .'(' . $_[1] . ')';
1442 #my ($self, $col, @vals) = @_;
1443 # called often - tighten code
1444 return $_[0]->{bindtype} eq 'columns'
1445 ? map {[$_[1], $_]} @_[2 .. $#_]
1450 # Dies if any element of @bind is not in [colname => value] format
1451 # if bindtype is 'columns'.
1452 sub _assert_bindval_matches_bindtype {
1453 # my ($self, @bind) = @_;
1455 if ($self->{bindtype} eq 'columns') {
1457 if (!defined $_ || ref($_) ne 'ARRAY' || @$_ != 2) {
1458 puke "bindtype 'columns' selected, you need to pass: [column_name => bind_value]"
1464 sub _join_sql_clauses {
1465 my ($self, $logic, $clauses_aref, $bind_aref) = @_;
1467 if (@$clauses_aref > 1) {
1468 my $join = " " . $self->_sqlcase($logic) . " ";
1469 my $sql = '( ' . join($join, @$clauses_aref) . ' )';
1470 return ($sql, @$bind_aref);
1472 elsif (@$clauses_aref) {
1473 return ($clauses_aref->[0], @$bind_aref); # no parentheses
1476 return (); # if no SQL, ignore @$bind_aref
1481 # Fix SQL case, if so requested
1483 # LDNOTE: if $self->{case} is true, then it contains 'lower', so we
1484 # don't touch the argument ... crooked logic, but let's not change it!
1485 return $_[0]->{case} ? $_[1] : uc($_[1]);
1489 #======================================================================
1490 # DISPATCHING FROM REFKIND
1491 #======================================================================
1494 my ($self, $data) = @_;
1496 return 'UNDEF' unless defined $data;
1498 # blessed objects are treated like scalars
1499 my $ref = (Scalar::Util::blessed $data) ? '' : ref $data;
1501 return 'SCALAR' unless $ref;
1504 while ($ref eq 'REF') {
1506 $ref = (Scalar::Util::blessed $data) ? '' : ref $data;
1510 return ($ref||'SCALAR') . ('REF' x $n_steps);
1514 my ($self, $data) = @_;
1515 my @try = ($self->_refkind($data));
1516 push @try, 'SCALAR_or_UNDEF' if $try[0] eq 'SCALAR' || $try[0] eq 'UNDEF';
1517 push @try, 'FALLBACK';
1521 sub _METHOD_FOR_refkind {
1522 my ($self, $meth_prefix, $data) = @_;
1525 for (@{$self->_try_refkind($data)}) {
1526 $method = $self->can($meth_prefix."_".$_)
1530 return $method || puke "cannot dispatch on '$meth_prefix' for ".$self->_refkind($data);
1534 sub _SWITCH_refkind {
1535 my ($self, $data, $dispatch_table) = @_;
1538 for (@{$self->_try_refkind($data)}) {
1539 $coderef = $dispatch_table->{$_}
1543 puke "no dispatch entry for ".$self->_refkind($data)
1552 #======================================================================
1553 # VALUES, GENERATE, AUTOLOAD
1554 #======================================================================
1556 # LDNOTE: original code from nwiger, didn't touch code in that section
1557 # I feel the AUTOLOAD stuff should not be the default, it should
1558 # only be activated on explicit demand by user.
1562 my $data = shift || return;
1563 puke "Argument to ", __PACKAGE__, "->values must be a \\%hash"
1564 unless ref $data eq 'HASH';
1567 foreach my $k ( sort keys %$data ) {
1568 my $v = $data->{$k};
1569 $self->_SWITCH_refkind($v, {
1571 if ($self->{array_datatypes}) { # array datatype
1572 push @all_bind, $self->_bindtype($k, $v);
1574 else { # literal SQL with bind
1575 my ($sql, @bind) = @$v;
1576 $self->_assert_bindval_matches_bindtype(@bind);
1577 push @all_bind, @bind;
1580 ARRAYREFREF => sub { # literal SQL with bind
1581 my ($sql, @bind) = @${$v};
1582 $self->_assert_bindval_matches_bindtype(@bind);
1583 push @all_bind, @bind;
1585 SCALARREF => sub { # literal SQL without bind
1587 SCALAR_or_UNDEF => sub {
1588 push @all_bind, $self->_bindtype($k, $v);
1599 my(@sql, @sqlq, @sqlv);
1603 if ($ref eq 'HASH') {
1604 for my $k (sort keys %$_) {
1607 my $label = $self->_quote($k);
1608 if ($r eq 'ARRAY') {
1609 # literal SQL with bind
1610 my ($sql, @bind) = @$v;
1611 $self->_assert_bindval_matches_bindtype(@bind);
1612 push @sqlq, "$label = $sql";
1614 } elsif ($r eq 'SCALAR') {
1615 # literal SQL without bind
1616 push @sqlq, "$label = $$v";
1618 push @sqlq, "$label = ?";
1619 push @sqlv, $self->_bindtype($k, $v);
1622 push @sql, $self->_sqlcase('set'), join ', ', @sqlq;
1623 } elsif ($ref eq 'ARRAY') {
1624 # unlike insert(), assume these are ONLY the column names, i.e. for SQL
1627 if ($r eq 'ARRAY') { # literal SQL with bind
1628 my ($sql, @bind) = @$v;
1629 $self->_assert_bindval_matches_bindtype(@bind);
1632 } elsif ($r eq 'SCALAR') { # literal SQL without bind
1633 # embedded literal SQL
1640 push @sql, '(' . join(', ', @sqlq) . ')';
1641 } elsif ($ref eq 'SCALAR') {
1645 # strings get case twiddled
1646 push @sql, $self->_sqlcase($_);
1650 my $sql = join ' ', @sql;
1652 # this is pretty tricky
1653 # if ask for an array, return ($stmt, @bind)
1654 # otherwise, s/?/shift @sqlv/ to put it inline
1656 return ($sql, @sqlv);
1658 1 while $sql =~ s/\?/my $d = shift(@sqlv);
1659 ref $d ? $d->[1] : $d/e;
1668 # This allows us to check for a local, then _form, attr
1670 my($name) = $AUTOLOAD =~ /.*::(.+)/;
1671 return $self->generate($name, @_);
1682 SQL::Abstract - Generate SQL from Perl data structures
1688 my $sql = SQL::Abstract->new;
1690 my($stmt, @bind) = $sql->select($source, \@fields, \%where, $order);
1692 my($stmt, @bind) = $sql->insert($table, \%fieldvals || \@values);
1694 my($stmt, @bind) = $sql->update($table, \%fieldvals, \%where);
1696 my($stmt, @bind) = $sql->delete($table, \%where);
1698 # Then, use these in your DBI statements
1699 my $sth = $dbh->prepare($stmt);
1700 $sth->execute(@bind);
1702 # Just generate the WHERE clause
1703 my($stmt, @bind) = $sql->where(\%where, $order);
1705 # Return values in the same order, for hashed queries
1706 # See PERFORMANCE section for more details
1707 my @bind = $sql->values(\%fieldvals);
1711 This module was inspired by the excellent L<DBIx::Abstract>.
1712 However, in using that module I found that what I really wanted
1713 to do was generate SQL, but still retain complete control over my
1714 statement handles and use the DBI interface. So, I set out to
1715 create an abstract SQL generation module.
1717 While based on the concepts used by L<DBIx::Abstract>, there are
1718 several important differences, especially when it comes to WHERE
1719 clauses. I have modified the concepts used to make the SQL easier
1720 to generate from Perl data structures and, IMO, more intuitive.
1721 The underlying idea is for this module to do what you mean, based
1722 on the data structures you provide it. The big advantage is that
1723 you don't have to modify your code every time your data changes,
1724 as this module figures it out.
1726 To begin with, an SQL INSERT is as easy as just specifying a hash
1727 of C<key=value> pairs:
1730 name => 'Jimbo Bobson',
1731 phone => '123-456-7890',
1732 address => '42 Sister Lane',
1733 city => 'St. Louis',
1734 state => 'Louisiana',
1737 The SQL can then be generated with this:
1739 my($stmt, @bind) = $sql->insert('people', \%data);
1741 Which would give you something like this:
1743 $stmt = "INSERT INTO people
1744 (address, city, name, phone, state)
1745 VALUES (?, ?, ?, ?, ?)";
1746 @bind = ('42 Sister Lane', 'St. Louis', 'Jimbo Bobson',
1747 '123-456-7890', 'Louisiana');
1749 These are then used directly in your DBI code:
1751 my $sth = $dbh->prepare($stmt);
1752 $sth->execute(@bind);
1754 =head2 Inserting and Updating Arrays
1756 If your database has array types (like for example Postgres),
1757 activate the special option C<< array_datatypes => 1 >>
1758 when creating the C<SQL::Abstract> object.
1759 Then you may use an arrayref to insert and update database array types:
1761 my $sql = SQL::Abstract->new(array_datatypes => 1);
1763 planets => [qw/Mercury Venus Earth Mars/]
1766 my($stmt, @bind) = $sql->insert('solar_system', \%data);
1770 $stmt = "INSERT INTO solar_system (planets) VALUES (?)"
1772 @bind = (['Mercury', 'Venus', 'Earth', 'Mars']);
1775 =head2 Inserting and Updating SQL
1777 In order to apply SQL functions to elements of your C<%data> you may
1778 specify a reference to an arrayref for the given hash value. For example,
1779 if you need to execute the Oracle C<to_date> function on a value, you can
1780 say something like this:
1784 date_entered => \[ "to_date(?,'MM/DD/YYYY')", "03/02/2003" ],
1787 The first value in the array is the actual SQL. Any other values are
1788 optional and would be included in the bind values array. This gives
1791 my($stmt, @bind) = $sql->insert('people', \%data);
1793 $stmt = "INSERT INTO people (name, date_entered)
1794 VALUES (?, to_date(?,'MM/DD/YYYY'))";
1795 @bind = ('Bill', '03/02/2003');
1797 An UPDATE is just as easy, all you change is the name of the function:
1799 my($stmt, @bind) = $sql->update('people', \%data);
1801 Notice that your C<%data> isn't touched; the module will generate
1802 the appropriately quirky SQL for you automatically. Usually you'll
1803 want to specify a WHERE clause for your UPDATE, though, which is
1804 where handling C<%where> hashes comes in handy...
1806 =head2 Complex where statements
1808 This module can generate pretty complicated WHERE statements
1809 easily. For example, simple C<key=value> pairs are taken to mean
1810 equality, and if you want to see if a field is within a set
1811 of values, you can use an arrayref. Let's say we wanted to
1812 SELECT some data based on this criteria:
1815 requestor => 'inna',
1816 worker => ['nwiger', 'rcwe', 'sfz'],
1817 status => { '!=', 'completed' }
1820 my($stmt, @bind) = $sql->select('tickets', '*', \%where);
1822 The above would give you something like this:
1824 $stmt = "SELECT * FROM tickets WHERE
1825 ( requestor = ? ) AND ( status != ? )
1826 AND ( worker = ? OR worker = ? OR worker = ? )";
1827 @bind = ('inna', 'completed', 'nwiger', 'rcwe', 'sfz');
1829 Which you could then use in DBI code like so:
1831 my $sth = $dbh->prepare($stmt);
1832 $sth->execute(@bind);
1838 The methods are simple. There's one for every major SQL operation,
1839 and a constructor you use first. The arguments are specified in a
1840 similar order for each method (table, then fields, then a where
1841 clause) to try and simplify things.
1843 =head2 new(option => 'value')
1845 The C<new()> function takes a list of options and values, and returns
1846 a new B<SQL::Abstract> object which can then be used to generate SQL
1847 through the methods below. The options accepted are:
1853 If set to 'lower', then SQL will be generated in all lowercase. By
1854 default SQL is generated in "textbook" case meaning something like:
1856 SELECT a_field FROM a_table WHERE some_field LIKE '%someval%'
1858 Any setting other than 'lower' is ignored.
1862 This determines what the default comparison operator is. By default
1863 it is C<=>, meaning that a hash like this:
1865 %where = (name => 'nwiger', email => 'nate@wiger.org');
1867 Will generate SQL like this:
1869 WHERE name = 'nwiger' AND email = 'nate@wiger.org'
1871 However, you may want loose comparisons by default, so if you set
1872 C<cmp> to C<like> you would get SQL such as:
1874 WHERE name like 'nwiger' AND email like 'nate@wiger.org'
1876 You can also override the comparison on an individual basis - see
1877 the huge section on L</"WHERE CLAUSES"> at the bottom.
1879 =item sqltrue, sqlfalse
1881 Expressions for inserting boolean values within SQL statements.
1882 By default these are C<1=1> and C<1=0>. They are used
1883 by the special operators C<-in> and C<-not_in> for generating
1884 correct SQL even when the argument is an empty array (see below).
1888 This determines the default logical operator for multiple WHERE
1889 statements in arrays or hashes. If absent, the default logic is "or"
1890 for arrays, and "and" for hashes. This means that a WHERE
1894 event_date => {'>=', '2/13/99'},
1895 event_date => {'<=', '4/24/03'},
1898 will generate SQL like this:
1900 WHERE event_date >= '2/13/99' OR event_date <= '4/24/03'
1902 This is probably not what you want given this query, though (look
1903 at the dates). To change the "OR" to an "AND", simply specify:
1905 my $sql = SQL::Abstract->new(logic => 'and');
1907 Which will change the above C<WHERE> to:
1909 WHERE event_date >= '2/13/99' AND event_date <= '4/24/03'
1911 The logic can also be changed locally by inserting
1912 a modifier in front of an arrayref :
1914 @where = (-and => [event_date => {'>=', '2/13/99'},
1915 event_date => {'<=', '4/24/03'} ]);
1917 See the L</"WHERE CLAUSES"> section for explanations.
1921 This will automatically convert comparisons using the specified SQL
1922 function for both column and value. This is mostly used with an argument
1923 of C<upper> or C<lower>, so that the SQL will have the effect of
1924 case-insensitive "searches". For example, this:
1926 $sql = SQL::Abstract->new(convert => 'upper');
1927 %where = (keywords => 'MaKe iT CAse inSeNSItive');
1929 Will turn out the following SQL:
1931 WHERE upper(keywords) like upper('MaKe iT CAse inSeNSItive')
1933 The conversion can be C<upper()>, C<lower()>, or any other SQL function
1934 that can be applied symmetrically to fields (actually B<SQL::Abstract> does
1935 not validate this option; it will just pass through what you specify verbatim).
1939 This is a kludge because many databases suck. For example, you can't
1940 just bind values using DBI's C<execute()> for Oracle C<CLOB> or C<BLOB> fields.
1941 Instead, you have to use C<bind_param()>:
1943 $sth->bind_param(1, 'reg data');
1944 $sth->bind_param(2, $lots, {ora_type => ORA_CLOB});
1946 The problem is, B<SQL::Abstract> will normally just return a C<@bind> array,
1947 which loses track of which field each slot refers to. Fear not.
1949 If you specify C<bindtype> in new, you can determine how C<@bind> is returned.
1950 Currently, you can specify either C<normal> (default) or C<columns>. If you
1951 specify C<columns>, you will get an array that looks like this:
1953 my $sql = SQL::Abstract->new(bindtype => 'columns');
1954 my($stmt, @bind) = $sql->insert(...);
1957 [ 'column1', 'value1' ],
1958 [ 'column2', 'value2' ],
1959 [ 'column3', 'value3' ],
1962 You can then iterate through this manually, using DBI's C<bind_param()>.
1964 $sth->prepare($stmt);
1967 my($col, $data) = @$_;
1968 if ($col eq 'details' || $col eq 'comments') {
1969 $sth->bind_param($i, $data, {ora_type => ORA_CLOB});
1970 } elsif ($col eq 'image') {
1971 $sth->bind_param($i, $data, {ora_type => ORA_BLOB});
1973 $sth->bind_param($i, $data);
1977 $sth->execute; # execute without @bind now
1979 Now, why would you still use B<SQL::Abstract> if you have to do this crap?
1980 Basically, the advantage is still that you don't have to care which fields
1981 are or are not included. You could wrap that above C<for> loop in a simple
1982 sub called C<bind_fields()> or something and reuse it repeatedly. You still
1983 get a layer of abstraction over manual SQL specification.
1985 Note that if you set L</bindtype> to C<columns>, the C<\[ $sql, @bind ]>
1986 construct (see L</Literal SQL with placeholders and bind values (subqueries)>)
1987 will expect the bind values in this format.
1991 This is the character that a table or column name will be quoted
1992 with. By default this is an empty string, but you could set it to
1993 the character C<`>, to generate SQL like this:
1995 SELECT `a_field` FROM `a_table` WHERE `some_field` LIKE '%someval%'
1997 Alternatively, you can supply an array ref of two items, the first being the left
1998 hand quote character, and the second the right hand quote character. For
1999 example, you could supply C<['[',']']> for SQL Server 2000 compliant quotes
2000 that generates SQL like this:
2002 SELECT [a_field] FROM [a_table] WHERE [some_field] LIKE '%someval%'
2004 Quoting is useful if you have tables or columns names that are reserved
2005 words in your database's SQL dialect.
2009 This is the character that will be used to escape L</quote_char>s appearing
2010 in an identifier before it has been quoted.
2012 The parameter default in case of a single L</quote_char> character is the quote
2015 When opening-closing-style quoting is used (L</quote_char> is an arrayref)
2016 this parameter defaults to the B<closing (right)> L</quote_char>. Occurences
2017 of the B<opening (left)> L</quote_char> within the identifier are currently left
2018 untouched. The default for opening-closing-style quotes may change in future
2019 versions, thus you are B<strongly encouraged> to specify the escape character
2024 This is the character that separates a table and column name. It is
2025 necessary to specify this when the C<quote_char> option is selected,
2026 so that tables and column names can be individually quoted like this:
2028 SELECT `table`.`one_field` FROM `table` WHERE `table`.`other_field` = 1
2030 =item injection_guard
2032 A regular expression C<qr/.../> that is applied to any C<-function> and unquoted
2033 column name specified in a query structure. This is a safety mechanism to avoid
2034 injection attacks when mishandling user input e.g.:
2036 my %condition_as_column_value_pairs = get_values_from_user();
2037 $sqla->select( ... , \%condition_as_column_value_pairs );
2039 If the expression matches an exception is thrown. Note that literal SQL
2040 supplied via C<\'...'> or C<\['...']> is B<not> checked in any way.
2042 Defaults to checking for C<;> and the C<GO> keyword (TransactSQL)
2044 =item array_datatypes
2046 When this option is true, arrayrefs in INSERT or UPDATE are
2047 interpreted as array datatypes and are passed directly
2049 When this option is false, arrayrefs are interpreted
2050 as literal SQL, just like refs to arrayrefs
2051 (but this behavior is for backwards compatibility; when writing
2052 new queries, use the "reference to arrayref" syntax
2058 Takes a reference to a list of "special operators"
2059 to extend the syntax understood by L<SQL::Abstract>.
2060 See section L</"SPECIAL OPERATORS"> for details.
2064 Takes a reference to a list of "unary operators"
2065 to extend the syntax understood by L<SQL::Abstract>.
2066 See section L</"UNARY OPERATORS"> for details.
2072 =head2 insert($table, \@values || \%fieldvals, \%options)
2074 This is the simplest function. You simply give it a table name
2075 and either an arrayref of values or hashref of field/value pairs.
2076 It returns an SQL INSERT statement and a list of bind values.
2077 See the sections on L</"Inserting and Updating Arrays"> and
2078 L</"Inserting and Updating SQL"> for information on how to insert
2079 with those data types.
2081 The optional C<\%options> hash reference may contain additional
2082 options to generate the insert SQL. Currently supported options
2089 Takes either a scalar of raw SQL fields, or an array reference of
2090 field names, and adds on an SQL C<RETURNING> statement at the end.
2091 This allows you to return data generated by the insert statement
2092 (such as row IDs) without performing another C<SELECT> statement.
2093 Note, however, this is not part of the SQL standard and may not
2094 be supported by all database engines.
2098 =head2 update($table, \%fieldvals, \%where, \%options)
2100 This takes a table, hashref of field/value pairs, and an optional
2101 hashref L<WHERE clause|/WHERE CLAUSES>. It returns an SQL UPDATE function and a list
2103 See the sections on L</"Inserting and Updating Arrays"> and
2104 L</"Inserting and Updating SQL"> for information on how to insert
2105 with those data types.
2107 The optional C<\%options> hash reference may contain additional
2108 options to generate the update SQL. Currently supported options
2115 See the C<returning> option to
2116 L<insert|/insert($table, \@values || \%fieldvals, \%options)>.
2120 =head2 select($source, $fields, $where, $order)
2122 This returns a SQL SELECT statement and associated list of bind values, as
2123 specified by the arguments :
2129 Specification of the 'FROM' part of the statement.
2130 The argument can be either a plain scalar (interpreted as a table
2131 name, will be quoted), or an arrayref (interpreted as a list
2132 of table names, joined by commas, quoted), or a scalarref
2133 (literal table name, not quoted), or a ref to an arrayref
2134 (list of literal table names, joined by commas, not quoted).
2138 Specification of the list of fields to retrieve from
2140 The argument can be either an arrayref (interpreted as a list
2141 of field names, will be joined by commas and quoted), or a
2142 plain scalar (literal SQL, not quoted).
2143 Please observe that this API is not as flexible as that of
2144 the first argument C<$source>, for backwards compatibility reasons.
2148 Optional argument to specify the WHERE part of the query.
2149 The argument is most often a hashref, but can also be
2150 an arrayref or plain scalar --
2151 see section L<WHERE clause|/"WHERE CLAUSES"> for details.
2155 Optional argument to specify the ORDER BY part of the query.
2156 The argument can be a scalar, a hashref or an arrayref
2157 -- see section L<ORDER BY clause|/"ORDER BY CLAUSES">
2163 =head2 delete($table, \%where)
2165 This takes a table name and optional hashref L<WHERE clause|/WHERE CLAUSES>.
2166 It returns an SQL DELETE statement and list of bind values.
2168 =head2 where(\%where, $order)
2170 This is used to generate just the WHERE clause. For example,
2171 if you have an arbitrary data structure and know what the
2172 rest of your SQL is going to look like, but want an easy way
2173 to produce a WHERE clause, use this. It returns an SQL WHERE
2174 clause and list of bind values.
2177 =head2 values(\%data)
2179 This just returns the values from the hash C<%data>, in the same
2180 order that would be returned from any of the other above queries.
2181 Using this allows you to markedly speed up your queries if you
2182 are affecting lots of rows. See below under the L</"PERFORMANCE"> section.
2184 =head2 generate($any, 'number', $of, \@data, $struct, \%types)
2186 Warning: This is an experimental method and subject to change.
2188 This returns arbitrarily generated SQL. It's a really basic shortcut.
2189 It will return two different things, depending on return context:
2191 my($stmt, @bind) = $sql->generate('create table', \$table, \@fields);
2192 my $stmt_and_val = $sql->generate('create table', \$table, \@fields);
2194 These would return the following:
2196 # First calling form
2197 $stmt = "CREATE TABLE test (?, ?)";
2198 @bind = (field1, field2);
2200 # Second calling form
2201 $stmt_and_val = "CREATE TABLE test (field1, field2)";
2203 Depending on what you're trying to do, it's up to you to choose the correct
2204 format. In this example, the second form is what you would want.
2208 $sql->generate('alter session', { nls_date_format => 'MM/YY' });
2212 ALTER SESSION SET nls_date_format = 'MM/YY'
2214 You get the idea. Strings get their case twiddled, but everything
2215 else remains verbatim.
2217 =head1 EXPORTABLE FUNCTIONS
2219 =head2 is_plain_value
2221 Determines if the supplied argument is a plain value as understood by this
2226 =item * The value is C<undef>
2228 =item * The value is a non-reference
2230 =item * The value is an object with stringification overloading
2232 =item * The value is of the form C<< { -value => $anything } >>
2236 On failure returns C<undef>, on sucess returns a B<scalar> reference
2237 to the original supplied argument.
2243 The stringification overloading detection is rather advanced: it takes
2244 into consideration not only the presence of a C<""> overload, but if that
2245 fails also checks for enabled
2246 L<autogenerated versions of C<"">|overload/Magic Autogeneration>, based
2247 on either C<0+> or C<bool>.
2249 Unfortunately testing in the field indicates that this
2250 detection B<< may tickle a latent bug in perl versions before 5.018 >>,
2251 but only when very large numbers of stringifying objects are involved.
2252 At the time of writing ( Sep 2014 ) there is no clear explanation of
2253 the direct cause, nor is there a manageably small test case that reliably
2254 reproduces the problem.
2256 If you encounter any of the following exceptions in B<random places within
2257 your application stack> - this module may be to blame:
2259 Operation "ne": no method found,
2260 left argument in overloaded package <something>,
2261 right argument in overloaded package <something>
2265 Stub found while resolving method "???" overloading """" in package <something>
2267 If you fall victim to the above - please attempt to reduce the problem
2268 to something that could be sent to the L<SQL::Abstract developers
2269 |DBIx::Class/GETTING HELP/SUPPORT>
2270 (either publicly or privately). As a workaround in the meantime you can
2271 set C<$ENV{SQLA_ISVALUE_IGNORE_AUTOGENERATED_STRINGIFICATION}> to a true
2272 value, which will most likely eliminate your problem (at the expense of
2273 not being able to properly detect exotic forms of stringification).
2275 This notice and environment variable will be removed in a future version,
2276 as soon as the underlying problem is found and a reliable workaround is
2281 =head2 is_literal_value
2283 Determines if the supplied argument is a literal value as understood by this
2288 =item * C<\$sql_string>
2290 =item * C<\[ $sql_string, @bind_values ]>
2294 On failure returns C<undef>, on sucess returns an B<array> reference
2295 containing the unpacked version of the supplied literal SQL and bind values.
2297 =head1 WHERE CLAUSES
2301 This module uses a variation on the idea from L<DBIx::Abstract>. It
2302 is B<NOT>, repeat I<not> 100% compatible. B<The main logic of this
2303 module is that things in arrays are OR'ed, and things in hashes
2306 The easiest way to explain is to show lots of examples. After
2307 each C<%where> hash shown, it is assumed you used:
2309 my($stmt, @bind) = $sql->where(\%where);
2311 However, note that the C<%where> hash can be used directly in any
2312 of the other functions as well, as described above.
2314 =head2 Key-value pairs
2316 So, let's get started. To begin, a simple hash:
2320 status => 'completed'
2323 Is converted to SQL C<key = val> statements:
2325 $stmt = "WHERE user = ? AND status = ?";
2326 @bind = ('nwiger', 'completed');
2328 One common thing I end up doing is having a list of values that
2329 a field can be in. To do this, simply specify a list inside of
2334 status => ['assigned', 'in-progress', 'pending'];
2337 This simple code will create the following:
2339 $stmt = "WHERE user = ? AND ( status = ? OR status = ? OR status = ? )";
2340 @bind = ('nwiger', 'assigned', 'in-progress', 'pending');
2342 A field associated to an empty arrayref will be considered a
2343 logical false and will generate 0=1.
2345 =head2 Tests for NULL values
2347 If the value part is C<undef> then this is converted to SQL <IS NULL>
2356 $stmt = "WHERE user = ? AND status IS NULL";
2359 To test if a column IS NOT NULL:
2363 status => { '!=', undef },
2366 =head2 Specific comparison operators
2368 If you want to specify a different type of operator for your comparison,
2369 you can use a hashref for a given column:
2373 status => { '!=', 'completed' }
2376 Which would generate:
2378 $stmt = "WHERE user = ? AND status != ?";
2379 @bind = ('nwiger', 'completed');
2381 To test against multiple values, just enclose the values in an arrayref:
2383 status => { '=', ['assigned', 'in-progress', 'pending'] };
2385 Which would give you:
2387 "WHERE status = ? OR status = ? OR status = ?"
2390 The hashref can also contain multiple pairs, in which case it is expanded
2391 into an C<AND> of its elements:
2395 status => { '!=', 'completed', -not_like => 'pending%' }
2398 # Or more dynamically, like from a form
2399 $where{user} = 'nwiger';
2400 $where{status}{'!='} = 'completed';
2401 $where{status}{'-not_like'} = 'pending%';
2403 # Both generate this
2404 $stmt = "WHERE user = ? AND status != ? AND status NOT LIKE ?";
2405 @bind = ('nwiger', 'completed', 'pending%');
2408 To get an OR instead, you can combine it with the arrayref idea:
2412 priority => [ { '=', 2 }, { '>', 5 } ]
2415 Which would generate:
2417 $stmt = "WHERE ( priority = ? OR priority > ? ) AND user = ?";
2418 @bind = ('2', '5', 'nwiger');
2420 If you want to include literal SQL (with or without bind values), just use a
2421 scalar reference or reference to an arrayref as the value:
2424 date_entered => { '>' => \["to_date(?, 'MM/DD/YYYY')", "11/26/2008"] },
2425 date_expires => { '<' => \"now()" }
2428 Which would generate:
2430 $stmt = "WHERE date_entered > to_date(?, 'MM/DD/YYYY') AND date_expires < now()";
2431 @bind = ('11/26/2008');
2434 =head2 Logic and nesting operators
2436 In the example above,
2437 there is a subtle trap if you want to say something like
2438 this (notice the C<AND>):
2440 WHERE priority != ? AND priority != ?
2442 Because, in Perl you I<can't> do this:
2444 priority => { '!=' => 2, '!=' => 1 }
2446 As the second C<!=> key will obliterate the first. The solution
2447 is to use the special C<-modifier> form inside an arrayref:
2449 priority => [ -and => {'!=', 2},
2453 Normally, these would be joined by C<OR>, but the modifier tells it
2454 to use C<AND> instead. (Hint: You can use this in conjunction with the
2455 C<logic> option to C<new()> in order to change the way your queries
2456 work by default.) B<Important:> Note that the C<-modifier> goes
2457 B<INSIDE> the arrayref, as an extra first element. This will
2458 B<NOT> do what you think it might:
2460 priority => -and => [{'!=', 2}, {'!=', 1}] # WRONG!
2462 Here is a quick list of equivalencies, since there is some overlap:
2465 status => {'!=', 'completed', 'not like', 'pending%' }
2466 status => [ -and => {'!=', 'completed'}, {'not like', 'pending%'}]
2469 status => {'=', ['assigned', 'in-progress']}
2470 status => [ -or => {'=', 'assigned'}, {'=', 'in-progress'}]
2471 status => [ {'=', 'assigned'}, {'=', 'in-progress'} ]
2475 =head2 Special operators : IN, BETWEEN, etc.
2477 You can also use the hashref format to compare a list of fields using the
2478 C<IN> comparison operator, by specifying the list as an arrayref:
2481 status => 'completed',
2482 reportid => { -in => [567, 2335, 2] }
2485 Which would generate:
2487 $stmt = "WHERE status = ? AND reportid IN (?,?,?)";
2488 @bind = ('completed', '567', '2335', '2');
2490 The reverse operator C<-not_in> generates SQL C<NOT IN> and is used in
2493 If the argument to C<-in> is an empty array, 'sqlfalse' is generated
2494 (by default : C<1=0>). Similarly, C<< -not_in => [] >> generates
2495 'sqltrue' (by default : C<1=1>).
2497 In addition to the array you can supply a chunk of literal sql or
2498 literal sql with bind:
2501 customer => { -in => \[
2502 'SELECT cust_id FROM cust WHERE balance > ?',
2505 status => { -in => \'SELECT status_codes FROM states' },
2511 customer IN ( SELECT cust_id FROM cust WHERE balance > ? )
2512 AND status IN ( SELECT status_codes FROM states )
2516 Finally, if the argument to C<-in> is not a reference, it will be
2517 treated as a single-element array.
2519 Another pair of operators is C<-between> and C<-not_between>,
2520 used with an arrayref of two values:
2524 completion_date => {
2525 -not_between => ['2002-10-01', '2003-02-06']
2531 WHERE user = ? AND completion_date NOT BETWEEN ( ? AND ? )
2533 Just like with C<-in> all plausible combinations of literal SQL
2537 start0 => { -between => [ 1, 2 ] },
2538 start1 => { -between => \["? AND ?", 1, 2] },
2539 start2 => { -between => \"lower(x) AND upper(y)" },
2540 start3 => { -between => [
2542 \["upper(?)", 'stuff' ],
2549 ( start0 BETWEEN ? AND ? )
2550 AND ( start1 BETWEEN ? AND ? )
2551 AND ( start2 BETWEEN lower(x) AND upper(y) )
2552 AND ( start3 BETWEEN lower(x) AND upper(?) )
2554 @bind = (1, 2, 1, 2, 'stuff');
2557 These are the two builtin "special operators"; but the
2558 list can be expanded : see section L</"SPECIAL OPERATORS"> below.
2560 =head2 Unary operators: bool
2562 If you wish to test against boolean columns or functions within your
2563 database you can use the C<-bool> and C<-not_bool> operators. For
2564 example to test the column C<is_user> being true and the column
2565 C<is_enabled> being false you would use:-
2569 -not_bool => 'is_enabled',
2574 WHERE is_user AND NOT is_enabled
2576 If a more complex combination is required, testing more conditions,
2577 then you should use the and/or operators:-
2582 -not_bool => { two=> { -rlike => 'bar' } },
2583 -not_bool => { three => [ { '=', 2 }, { '>', 5 } ] },
2594 (NOT ( three = ? OR three > ? ))
2597 =head2 Nested conditions, -and/-or prefixes
2599 So far, we've seen how multiple conditions are joined with a top-level
2600 C<AND>. We can change this by putting the different conditions we want in
2601 hashes and then putting those hashes in an array. For example:
2606 status => { -like => ['pending%', 'dispatched'] },
2610 status => 'unassigned',
2614 This data structure would create the following:
2616 $stmt = "WHERE ( user = ? AND ( status LIKE ? OR status LIKE ? ) )
2617 OR ( user = ? AND status = ? ) )";
2618 @bind = ('nwiger', 'pending', 'dispatched', 'robot', 'unassigned');
2621 Clauses in hashrefs or arrayrefs can be prefixed with an C<-and> or C<-or>
2622 to change the logic inside :
2628 -and => [ workhrs => {'>', 20}, geo => 'ASIA' ],
2629 -or => { workhrs => {'<', 50}, geo => 'EURO' },
2636 $stmt = "WHERE ( user = ?
2637 AND ( ( workhrs > ? AND geo = ? )
2638 OR ( workhrs < ? OR geo = ? ) ) )";
2639 @bind = ('nwiger', '20', 'ASIA', '50', 'EURO');
2641 =head3 Algebraic inconsistency, for historical reasons
2643 C<Important note>: when connecting several conditions, the C<-and->|C<-or>
2644 operator goes C<outside> of the nested structure; whereas when connecting
2645 several constraints on one column, the C<-and> operator goes
2646 C<inside> the arrayref. Here is an example combining both features :
2649 -and => [a => 1, b => 2],
2650 -or => [c => 3, d => 4],
2651 e => [-and => {-like => 'foo%'}, {-like => '%bar'} ]
2656 WHERE ( ( ( a = ? AND b = ? )
2657 OR ( c = ? OR d = ? )
2658 OR ( e LIKE ? AND e LIKE ? ) ) )
2660 This difference in syntax is unfortunate but must be preserved for
2661 historical reasons. So be careful : the two examples below would
2662 seem algebraically equivalent, but they are not
2665 { -like => 'foo%' },
2666 { -like => '%bar' },
2668 # yields : WHERE ( ( col LIKE ? AND col LIKE ? ) )
2671 { col => { -like => 'foo%' } },
2672 { col => { -like => '%bar' } },
2674 # yields : WHERE ( ( col LIKE ? OR col LIKE ? ) )
2677 =head2 Literal SQL and value type operators
2679 The basic premise of SQL::Abstract is that in WHERE specifications the "left
2680 side" is a column name and the "right side" is a value (normally rendered as
2681 a placeholder). This holds true for both hashrefs and arrayref pairs as you
2682 see in the L</WHERE CLAUSES> examples above. Sometimes it is necessary to
2683 alter this behavior. There are several ways of doing so.
2687 This is a virtual operator that signals the string to its right side is an
2688 identifier (a column name) and not a value. For example to compare two
2689 columns you would write:
2692 priority => { '<', 2 },
2693 requestor => { -ident => 'submitter' },
2698 $stmt = "WHERE priority < ? AND requestor = submitter";
2701 If you are maintaining legacy code you may see a different construct as
2702 described in L</Deprecated usage of Literal SQL>, please use C<-ident> in new
2707 This is a virtual operator that signals that the construct to its right side
2708 is a value to be passed to DBI. This is for example necessary when you want
2709 to write a where clause against an array (for RDBMS that support such
2710 datatypes). For example:
2713 array => { -value => [1, 2, 3] }
2718 $stmt = 'WHERE array = ?';
2719 @bind = ([1, 2, 3]);
2721 Note that if you were to simply say:
2727 the result would probably not be what you wanted:
2729 $stmt = 'WHERE array = ? OR array = ? OR array = ?';
2734 Finally, sometimes only literal SQL will do. To include a random snippet
2735 of SQL verbatim, you specify it as a scalar reference. Consider this only
2736 as a last resort. Usually there is a better way. For example:
2739 priority => { '<', 2 },
2740 requestor => { -in => \'(SELECT name FROM hitmen)' },
2745 $stmt = "WHERE priority < ? AND requestor IN (SELECT name FROM hitmen)"
2748 Note that in this example, you only get one bind parameter back, since
2749 the verbatim SQL is passed as part of the statement.
2753 Never use untrusted input as a literal SQL argument - this is a massive
2754 security risk (there is no way to check literal snippets for SQL
2755 injections and other nastyness). If you need to deal with untrusted input
2756 use literal SQL with placeholders as described next.
2758 =head3 Literal SQL with placeholders and bind values (subqueries)
2760 If the literal SQL to be inserted has placeholders and bind values,
2761 use a reference to an arrayref (yes this is a double reference --
2762 not so common, but perfectly legal Perl). For example, to find a date
2763 in Postgres you can use something like this:
2766 date_column => \[ "= date '2008-09-30' - ?::integer", 10 ]
2771 $stmt = "WHERE ( date_column = date '2008-09-30' - ?::integer )"
2774 Note that you must pass the bind values in the same format as they are returned
2775 by L<where|/where(\%where, $order)>. This means that if you set L</bindtype>
2776 to C<columns>, you must provide the bind values in the
2777 C<< [ column_meta => value ] >> format, where C<column_meta> is an opaque
2778 scalar value; most commonly the column name, but you can use any scalar value
2779 (including references and blessed references), L<SQL::Abstract> will simply
2780 pass it through intact. So if C<bindtype> is set to C<columns> the above
2781 example will look like:
2784 date_column => \[ "= date '2008-09-30' - ?::integer", [ {} => 10 ] ]
2787 Literal SQL is especially useful for nesting parenthesized clauses in the
2788 main SQL query. Here is a first example :
2790 my ($sub_stmt, @sub_bind) = ("SELECT c1 FROM t1 WHERE c2 < ? AND c3 LIKE ?",
2794 bar => \["IN ($sub_stmt)" => @sub_bind],
2799 $stmt = "WHERE (foo = ? AND bar IN (SELECT c1 FROM t1
2800 WHERE c2 < ? AND c3 LIKE ?))";
2801 @bind = (1234, 100, "foo%");
2803 Other subquery operators, like for example C<"E<gt> ALL"> or C<"NOT IN">,
2804 are expressed in the same way. Of course the C<$sub_stmt> and
2805 its associated bind values can be generated through a former call
2808 my ($sub_stmt, @sub_bind)
2809 = $sql->select("t1", "c1", {c2 => {"<" => 100},
2810 c3 => {-like => "foo%"}});
2813 bar => \["> ALL ($sub_stmt)" => @sub_bind],
2816 In the examples above, the subquery was used as an operator on a column;
2817 but the same principle also applies for a clause within the main C<%where>
2818 hash, like an EXISTS subquery :
2820 my ($sub_stmt, @sub_bind)
2821 = $sql->select("t1", "*", {c1 => 1, c2 => \"> t0.c0"});
2822 my %where = ( -and => [
2824 \["EXISTS ($sub_stmt)" => @sub_bind],
2829 $stmt = "WHERE (foo = ? AND EXISTS (SELECT * FROM t1
2830 WHERE c1 = ? AND c2 > t0.c0))";
2834 Observe that the condition on C<c2> in the subquery refers to
2835 column C<t0.c0> of the main query : this is I<not> a bind
2836 value, so we have to express it through a scalar ref.
2837 Writing C<< c2 => {">" => "t0.c0"} >> would have generated
2838 C<< c2 > ? >> with bind value C<"t0.c0"> ... not exactly
2839 what we wanted here.
2841 Finally, here is an example where a subquery is used
2842 for expressing unary negation:
2844 my ($sub_stmt, @sub_bind)
2845 = $sql->where({age => [{"<" => 10}, {">" => 20}]});
2846 $sub_stmt =~ s/^ where //i; # don't want "WHERE" in the subclause
2848 lname => {like => '%son%'},
2849 \["NOT ($sub_stmt)" => @sub_bind],
2854 $stmt = "lname LIKE ? AND NOT ( age < ? OR age > ? )"
2855 @bind = ('%son%', 10, 20)
2857 =head3 Deprecated usage of Literal SQL
2859 Below are some examples of archaic use of literal SQL. It is shown only as
2860 reference for those who deal with legacy code. Each example has a much
2861 better, cleaner and safer alternative that users should opt for in new code.
2867 my %where = ( requestor => \'IS NOT NULL' )
2869 $stmt = "WHERE requestor IS NOT NULL"
2871 This used to be the way of generating NULL comparisons, before the handling
2872 of C<undef> got formalized. For new code please use the superior syntax as
2873 described in L</Tests for NULL values>.
2877 my %where = ( requestor => \'= submitter' )
2879 $stmt = "WHERE requestor = submitter"
2881 This used to be the only way to compare columns. Use the superior L</-ident>
2882 method for all new code. For example an identifier declared in such a way
2883 will be properly quoted if L</quote_char> is properly set, while the legacy
2884 form will remain as supplied.
2888 my %where = ( is_ready => \"", completed => { '>', '2012-12-21' } )
2890 $stmt = "WHERE completed > ? AND is_ready"
2891 @bind = ('2012-12-21')
2893 Using an empty string literal used to be the only way to express a boolean.
2894 For all new code please use the much more readable
2895 L<-bool|/Unary operators: bool> operator.
2901 These pages could go on for a while, since the nesting of the data
2902 structures this module can handle are pretty much unlimited (the
2903 module implements the C<WHERE> expansion as a recursive function
2904 internally). Your best bet is to "play around" with the module a
2905 little to see how the data structures behave, and choose the best
2906 format for your data based on that.
2908 And of course, all the values above will probably be replaced with
2909 variables gotten from forms or the command line. After all, if you
2910 knew everything ahead of time, you wouldn't have to worry about
2911 dynamically-generating SQL and could just hardwire it into your
2914 =head1 ORDER BY CLAUSES
2916 Some functions take an order by clause. This can either be a scalar (just a
2917 column name,) a hash of C<< { -desc => 'col' } >> or C<< { -asc => 'col' } >>,
2918 optionally with C<< -nulls => 'first' >> or C<< -nulls => 'last' >>,
2919 or an array of either of the two previous forms. Examples:
2921 Given | Will Generate
2922 ----------------------------------------------------------
2924 \'colA DESC' | ORDER BY colA DESC
2926 'colA' | ORDER BY colA
2928 [qw/colA colB/] | ORDER BY colA, colB
2930 {-asc => 'colA'} | ORDER BY colA ASC
2932 {-desc => 'colB'} | ORDER BY colB DESC
2935 -asc => 'colA', | ORDER BY colA ASC NULLS LAST
2939 ['colA', {-asc => 'colB'}] | ORDER BY colA, colB ASC
2941 { -asc => [qw/colA colB/] } | ORDER BY colA ASC, colB ASC
2944 -asc => [qw/colA colB/] | ORDER BY colA ASC NULLS FIRST,
2945 -nulls => 'first' | colB ASC NULLS FIRST
2949 { -asc => 'colA' }, | ORDER BY colA ASC, colB DESC,
2950 { -desc => [qw/colB/], | colC ASC, colD ASC
2951 { -asc => [qw/colC colD/],|
2953 ===========================================================
2957 =head1 SPECIAL OPERATORS
2959 my $sqlmaker = SQL::Abstract->new(special_ops => [
2963 my ($self, $field, $op, $arg) = @_;
2969 handler => 'method_name',
2973 A "special operator" is a SQL syntactic clause that can be
2974 applied to a field, instead of a usual binary operator.
2977 WHERE field IN (?, ?, ?)
2978 WHERE field BETWEEN ? AND ?
2979 WHERE MATCH(field) AGAINST (?, ?)
2981 Special operators IN and BETWEEN are fairly standard and therefore
2982 are builtin within C<SQL::Abstract> (as the overridable methods
2983 C<_where_field_IN> and C<_where_field_BETWEEN>). For other operators,
2984 like the MATCH .. AGAINST example above which is specific to MySQL,
2985 you can write your own operator handlers - supply a C<special_ops>
2986 argument to the C<new> method. That argument takes an arrayref of
2987 operator definitions; each operator definition is a hashref with two
2994 the regular expression to match the operator
2998 Either a coderef or a plain scalar method name. In both cases
2999 the expected return is C<< ($sql, @bind) >>.
3001 When supplied with a method name, it is simply called on the
3002 L<SQL::Abstract> object as:
3004 $self->$method_name ($field, $op, $arg)
3008 $field is the LHS of the operator
3009 $op is the part that matched the handler regex
3012 When supplied with a coderef, it is called as:
3014 $coderef->($self, $field, $op, $arg)
3019 For example, here is an implementation
3020 of the MATCH .. AGAINST syntax for MySQL
3022 my $sqlmaker = SQL::Abstract->new(special_ops => [
3024 # special op for MySql MATCH (field) AGAINST(word1, word2, ...)
3025 {regex => qr/^match$/i,
3027 my ($self, $field, $op, $arg) = @_;
3028 $arg = [$arg] if not ref $arg;
3029 my $label = $self->_quote($field);
3030 my ($placeholder) = $self->_convert('?');
3031 my $placeholders = join ", ", (($placeholder) x @$arg);
3032 my $sql = $self->_sqlcase('match') . " ($label) "
3033 . $self->_sqlcase('against') . " ($placeholders) ";
3034 my @bind = $self->_bindtype($field, @$arg);
3035 return ($sql, @bind);
3042 =head1 UNARY OPERATORS
3044 my $sqlmaker = SQL::Abstract->new(unary_ops => [
3048 my ($self, $op, $arg) = @_;
3054 handler => 'method_name',
3058 A "unary operator" is a SQL syntactic clause that can be
3059 applied to a field - the operator goes before the field
3061 You can write your own operator handlers - supply a C<unary_ops>
3062 argument to the C<new> method. That argument takes an arrayref of
3063 operator definitions; each operator definition is a hashref with two
3070 the regular expression to match the operator
3074 Either a coderef or a plain scalar method name. In both cases
3075 the expected return is C<< $sql >>.
3077 When supplied with a method name, it is simply called on the
3078 L<SQL::Abstract> object as:
3080 $self->$method_name ($op, $arg)
3084 $op is the part that matched the handler regex
3085 $arg is the RHS or argument of the operator
3087 When supplied with a coderef, it is called as:
3089 $coderef->($self, $op, $arg)
3097 Thanks to some benchmarking by Mark Stosberg, it turns out that
3098 this module is many orders of magnitude faster than using C<DBIx::Abstract>.
3099 I must admit this wasn't an intentional design issue, but it's a
3100 byproduct of the fact that you get to control your C<DBI> handles
3103 To maximize performance, use a code snippet like the following:
3105 # prepare a statement handle using the first row
3106 # and then reuse it for the rest of the rows
3108 for my $href (@array_of_hashrefs) {
3109 $stmt ||= $sql->insert('table', $href);
3110 $sth ||= $dbh->prepare($stmt);
3111 $sth->execute($sql->values($href));
3114 The reason this works is because the keys in your C<$href> are sorted
3115 internally by B<SQL::Abstract>. Thus, as long as your data retains
3116 the same structure, you only have to generate the SQL the first time
3117 around. On subsequent queries, simply use the C<values> function provided
3118 by this module to return your values in the correct order.
3120 However this depends on the values having the same type - if, for
3121 example, the values of a where clause may either have values
3122 (resulting in sql of the form C<column = ?> with a single bind
3123 value), or alternatively the values might be C<undef> (resulting in
3124 sql of the form C<column IS NULL> with no bind value) then the
3125 caching technique suggested will not work.
3129 If you use my C<CGI::FormBuilder> module at all, you'll hopefully
3130 really like this part (I do, at least). Building up a complex query
3131 can be as simple as the following:
3138 use CGI::FormBuilder;
3141 my $form = CGI::FormBuilder->new(...);
3142 my $sql = SQL::Abstract->new;
3144 if ($form->submitted) {
3145 my $field = $form->field;
3146 my $id = delete $field->{id};
3147 my($stmt, @bind) = $sql->update('table', $field, {id => $id});
3150 Of course, you would still have to connect using C<DBI> to run the
3151 query, but the point is that if you make your form look like your
3152 table, the actual query script can be extremely simplistic.
3154 If you're B<REALLY> lazy (I am), check out C<HTML::QuickTable> for
3155 a fast interface to returning and formatting data. I frequently
3156 use these three modules together to write complex database query
3157 apps in under 50 lines.
3159 =head1 HOW TO CONTRIBUTE
3161 Contributions are always welcome, in all usable forms (we especially
3162 welcome documentation improvements). The delivery methods include git-
3163 or unified-diff formatted patches, GitHub pull requests, or plain bug
3164 reports either via RT or the Mailing list. Contributors are generally
3165 granted full access to the official repository after their first several
3166 patches pass successful review.
3168 This project is maintained in a git repository. The code and related tools are
3169 accessible at the following locations:
3173 =item * Official repo: L<git://git.shadowcat.co.uk/dbsrgits/SQL-Abstract.git>
3175 =item * Official gitweb: L<http://git.shadowcat.co.uk/gitweb/gitweb.cgi?p=dbsrgits/SQL-Abstract.git>
3177 =item * GitHub mirror: L<https://github.com/dbsrgits/sql-abstract>
3179 =item * Authorized committers: L<ssh://dbsrgits@git.shadowcat.co.uk/SQL-Abstract.git>
3185 Version 1.50 was a major internal refactoring of C<SQL::Abstract>.
3186 Great care has been taken to preserve the I<published> behavior
3187 documented in previous versions in the 1.* family; however,
3188 some features that were previously undocumented, or behaved
3189 differently from the documentation, had to be changed in order
3190 to clarify the semantics. Hence, client code that was relying
3191 on some dark areas of C<SQL::Abstract> v1.*
3192 B<might behave differently> in v1.50.
3194 The main changes are :
3200 support for literal SQL through the C<< \ [ $sql, @bind ] >> syntax.
3204 support for the { operator => \"..." } construct (to embed literal SQL)
3208 support for the { operator => \["...", @bind] } construct (to embed literal SQL with bind values)
3212 optional support for L<array datatypes|/"Inserting and Updating Arrays">
3216 defensive programming : check arguments
3220 fixed bug with global logic, which was previously implemented
3221 through global variables yielding side-effects. Prior versions would
3222 interpret C<< [ {cond1, cond2}, [cond3, cond4] ] >>
3223 as C<< "(cond1 AND cond2) OR (cond3 AND cond4)" >>.
3224 Now this is interpreted
3225 as C<< "(cond1 AND cond2) OR (cond3 OR cond4)" >>.
3230 fixed semantics of _bindtype on array args
3234 dropped the C<_anoncopy> of the %where tree. No longer necessary,
3235 we just avoid shifting arrays within that tree.
3239 dropped the C<_modlogic> function
3243 =head1 ACKNOWLEDGEMENTS
3245 There are a number of individuals that have really helped out with
3246 this module. Unfortunately, most of them submitted bugs via CPAN
3247 so I have no idea who they are! But the people I do know are:
3249 Ash Berlin (order_by hash term support)
3250 Matt Trout (DBIx::Class support)
3251 Mark Stosberg (benchmarking)
3252 Chas Owens (initial "IN" operator support)
3253 Philip Collins (per-field SQL functions)
3254 Eric Kolve (hashref "AND" support)
3255 Mike Fragassi (enhancements to "BETWEEN" and "LIKE")
3256 Dan Kubb (support for "quote_char" and "name_sep")
3257 Guillermo Roditi (patch to cleanup "IN" and "BETWEEN", fix and tests for _order_by)
3258 Laurent Dami (internal refactoring, extensible list of special operators, literal SQL)
3259 Norbert Buchmuller (support for literal SQL in hashpair, misc. fixes & tests)
3260 Peter Rabbitson (rewrite of SQLA::Test, misc. fixes & tests)
3261 Oliver Charles (support for "RETURNING" after "INSERT")
3267 L<DBIx::Class>, L<DBIx::Abstract>, L<CGI::FormBuilder>, L<HTML::QuickTable>.
3271 Copyright (c) 2001-2007 Nathan Wiger <nwiger@cpan.org>. All Rights Reserved.
3273 This module is actively maintained by Matt Trout <mst@shadowcatsystems.co.uk>
3275 For support, your best bet is to try the C<DBIx::Class> users mailing list.
3276 While not an official support venue, C<DBIx::Class> makes heavy use of
3277 C<SQL::Abstract>, and as such list members there are very familiar with
3278 how to create queries.
3282 This module is free software; you may copy this under the same
3283 terms as perl itself (either the GNU General Public License or
3284 the Artistic License)