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 if (my $cols = $options->{columns}) {
218 $table_sql .= '('.join(', ', map $self->_quote($_), @$cols).')';
221 my $method = $self->_METHOD_FOR_refkind("_insert", $data);
222 my ($sql, @bind) = $self->$method($data);
223 $sql = join " ", $self->_sqlcase('insert into'), $table, $sql;
225 if ($options->{returning}) {
226 my ($s, @b) = $self->_insert_returning($options);
231 return wantarray ? ($sql, @bind) : $sql;
234 # So that subclasses can override INSERT ... RETURNING separately from
235 # UPDATE and DELETE (e.g. DBIx::Class::SQLMaker::Oracle does this)
236 sub _insert_returning { shift->_returning(@_) }
239 my ($self, $options) = @_;
241 my $f = $options->{returning};
243 my $fieldlist = $self->_SWITCH_refkind($f, {
244 ARRAYREF => sub {join ', ', map { $self->_quote($_) } @$f;},
245 SCALAR => sub {$self->_quote($f)},
246 SCALARREF => sub {$$f},
248 return $self->_sqlcase(' returning ') . $fieldlist;
251 sub _insert_HASHREF { # explicit list of fields and then values
252 my ($self, $data) = @_;
254 my @fields = sort keys %$data;
256 my ($sql, @bind) = $self->_insert_values($data);
259 $_ = $self->_quote($_) foreach @fields;
260 $sql = "( ".join(", ", @fields).") ".$sql;
262 return ($sql, @bind);
265 sub _insert_ARRAYREF { # just generate values(?,?) part (no list of fields)
266 my ($self, $data) = @_;
268 # no names (arrayref) so can't generate bindtype
269 $self->{bindtype} ne 'columns'
270 or belch "can't do 'columns' bindtype when called with arrayref";
272 my (@values, @all_bind);
273 foreach my $value (@$data) {
274 my ($values, @bind) = $self->_insert_value(undef, $value);
275 push @values, $values;
276 push @all_bind, @bind;
278 my $sql = $self->_sqlcase('values')." ( ".join(", ", @values)." )";
279 return ($sql, @all_bind);
282 sub _insert_ARRAYREFREF { # literal SQL with bind
283 my ($self, $data) = @_;
285 my ($sql, @bind) = @${$data};
286 $self->_assert_bindval_matches_bindtype(@bind);
288 return ($sql, @bind);
292 sub _insert_SCALARREF { # literal SQL without bind
293 my ($self, $data) = @_;
299 my ($self, $data) = @_;
301 my (@values, @all_bind);
302 foreach my $column (sort keys %$data) {
303 my ($values, @bind) = $self->_insert_value($column, $data->{$column});
304 push @values, $values;
305 push @all_bind, @bind;
307 my $sql = $self->_sqlcase('values')." ( ".join(", ", @values)." )";
308 return ($sql, @all_bind);
312 my ($self, $column, $v) = @_;
314 my (@values, @all_bind);
315 $self->_SWITCH_refkind($v, {
318 if ($self->{array_datatypes}) { # if array datatype are activated
320 push @all_bind, $self->_bindtype($column, $v);
322 else { # else literal SQL with bind
323 my ($sql, @bind) = @$v;
324 $self->_assert_bindval_matches_bindtype(@bind);
326 push @all_bind, @bind;
330 ARRAYREFREF => sub { # literal SQL with bind
331 my ($sql, @bind) = @${$v};
332 $self->_assert_bindval_matches_bindtype(@bind);
334 push @all_bind, @bind;
337 # THINK : anything useful to do with a HASHREF ?
338 HASHREF => sub { # (nothing, but old SQLA passed it through)
339 #TODO in SQLA >= 2.0 it will die instead
340 belch "HASH ref as bind value in insert is not supported";
342 push @all_bind, $self->_bindtype($column, $v);
345 SCALARREF => sub { # literal SQL without bind
349 SCALAR_or_UNDEF => sub {
351 push @all_bind, $self->_bindtype($column, $v);
356 my $sql = join(", ", @values);
357 return ($sql, @all_bind);
362 #======================================================================
364 #======================================================================
369 my $table = $self->_table(shift);
370 my $data = shift || return;
374 # first build the 'SET' part of the sql statement
375 my (@set, @all_bind);
376 puke "Unsupported data type specified to \$sql->update"
377 unless ref $data eq 'HASH';
379 for my $k (sort keys %$data) {
382 my $label = $self->_quote($k);
384 $self->_SWITCH_refkind($v, {
386 if ($self->{array_datatypes}) { # array datatype
387 push @set, "$label = ?";
388 push @all_bind, $self->_bindtype($k, $v);
390 else { # literal SQL with bind
391 my ($sql, @bind) = @$v;
392 $self->_assert_bindval_matches_bindtype(@bind);
393 push @set, "$label = $sql";
394 push @all_bind, @bind;
397 ARRAYREFREF => sub { # literal SQL with bind
398 my ($sql, @bind) = @${$v};
399 $self->_assert_bindval_matches_bindtype(@bind);
400 push @set, "$label = $sql";
401 push @all_bind, @bind;
403 SCALARREF => sub { # literal SQL without bind
404 push @set, "$label = $$v";
407 my ($op, $arg, @rest) = %$v;
409 puke 'Operator calls in update must be in the form { -op => $arg }'
410 if (@rest or not $op =~ /^\-(.+)/);
412 local $self->{_nested_func_lhs} = $k;
413 my ($sql, @bind) = $self->_where_unary_op($1, $arg);
415 push @set, "$label = $sql";
416 push @all_bind, @bind;
418 SCALAR_or_UNDEF => sub {
419 push @set, "$label = ?";
420 push @all_bind, $self->_bindtype($k, $v);
426 my $sql = $self->_sqlcase('update') . " $table " . $self->_sqlcase('set ')
430 my($where_sql, @where_bind) = $self->where($where);
432 push @all_bind, @where_bind;
435 if ($options->{returning}) {
436 my ($returning_sql, @returning_bind) = $self->_update_returning($options);
437 $sql .= $returning_sql;
438 push @all_bind, @returning_bind;
441 return wantarray ? ($sql, @all_bind) : $sql;
444 # So that subclasses can override UPDATE ... RETURNING separately from
446 sub _update_returning { shift->_returning(@_) }
450 #======================================================================
452 #======================================================================
457 my $table = $self->_table(shift);
458 my $fields = shift || '*';
462 my($where_sql, @bind) = $self->where($where, $order);
464 my $f = (ref $fields eq 'ARRAY') ? join ', ', map { $self->_quote($_) } @$fields
466 my $sql = join(' ', $self->_sqlcase('select'), $f,
467 $self->_sqlcase('from'), $table)
470 return wantarray ? ($sql, @bind) : $sql;
473 #======================================================================
475 #======================================================================
480 my $table = $self->_table(shift);
484 my($where_sql, @bind) = $self->where($where);
485 my $sql = $self->_sqlcase('delete from') . " $table" . $where_sql;
487 if ($options->{returning}) {
488 my ($returning_sql, @returning_bind) = $self->_delete_returning($options);
489 $sql .= $returning_sql;
490 push @bind, @returning_bind;
493 return wantarray ? ($sql, @bind) : $sql;
496 # So that subclasses can override DELETE ... RETURNING separately from
498 sub _delete_returning { shift->_returning(@_) }
502 #======================================================================
504 #======================================================================
508 # Finally, a separate routine just to handle WHERE clauses
510 my ($self, $where, $order) = @_;
513 my ($sql, @bind) = $self->_recurse_where($where);
514 $sql = $sql ? $self->_sqlcase(' where ') . "( $sql )" : '';
518 my ($order_sql, @order_bind) = $self->_order_by($order);
520 push @bind, @order_bind;
523 return wantarray ? ($sql, @bind) : $sql;
528 my ($self, $where, $logic) = @_;
530 # dispatch on appropriate method according to refkind of $where
531 my $method = $self->_METHOD_FOR_refkind("_where", $where);
533 my ($sql, @bind) = $self->$method($where, $logic);
535 # DBIx::Class used to call _recurse_where in scalar context
536 # something else might too...
538 return ($sql, @bind);
541 belch "Calling _recurse_where in scalar context is deprecated and will go away before 2.0";
548 #======================================================================
549 # WHERE: top-level ARRAYREF
550 #======================================================================
553 sub _where_ARRAYREF {
554 my ($self, $where, $logic) = @_;
556 $logic = uc($logic || $self->{logic});
557 $logic eq 'AND' or $logic eq 'OR' or puke "unknown logic: $logic";
559 my @clauses = @$where;
561 my (@sql_clauses, @all_bind);
562 # need to use while() so can shift() for pairs
564 my $el = shift @clauses;
566 $el = undef if (defined $el and ! length $el);
568 # switch according to kind of $el and get corresponding ($sql, @bind)
569 my ($sql, @bind) = $self->_SWITCH_refkind($el, {
571 # skip empty elements, otherwise get invalid trailing AND stuff
572 ARRAYREF => sub {$self->_recurse_where($el) if @$el},
576 $self->_assert_bindval_matches_bindtype(@b);
580 HASHREF => sub {$self->_recurse_where($el, 'and') if %$el},
582 SCALARREF => sub { ($$el); },
585 # top-level arrayref with scalars, recurse in pairs
586 $self->_recurse_where({$el => shift(@clauses)})
589 UNDEF => sub {puke "Supplying an empty left hand side argument is not supported in array-pairs" },
593 push @sql_clauses, $sql;
594 push @all_bind, @bind;
598 return $self->_join_sql_clauses($logic, \@sql_clauses, \@all_bind);
601 #======================================================================
602 # WHERE: top-level ARRAYREFREF
603 #======================================================================
605 sub _where_ARRAYREFREF {
606 my ($self, $where) = @_;
607 my ($sql, @bind) = @$$where;
608 $self->_assert_bindval_matches_bindtype(@bind);
609 return ($sql, @bind);
612 #======================================================================
613 # WHERE: top-level HASHREF
614 #======================================================================
617 my ($self, $where) = @_;
618 my (@sql_clauses, @all_bind);
620 for my $k (sort keys %$where) {
621 my $v = $where->{$k};
623 # ($k => $v) is either a special unary op or a regular hashpair
624 my ($sql, @bind) = do {
626 # put the operator in canonical form
628 $op = substr $op, 1; # remove initial dash
629 $op =~ s/^\s+|\s+$//g;# remove leading/trailing space
630 $op =~ s/\s+/ /g; # compress whitespace
632 # so that -not_foo works correctly
633 $op =~ s/^not_/NOT /i;
635 $self->_debug("Unary OP(-$op) within hashref, recursing...");
636 my ($s, @b) = $self->_where_unary_op($op, $v);
638 # top level vs nested
639 # we assume that handled unary ops will take care of their ()s
641 List::Util::first {$op =~ $_->{regex}} @{$self->{unary_ops}}
643 ( defined $self->{_nested_func_lhs} and $self->{_nested_func_lhs} eq $k )
649 if (is_literal_value ($v) ) {
650 belch 'Hash-pairs consisting of an empty string with a literal are deprecated, and will be removed in 2.0: use -and => [ $literal ] instead';
653 puke "Supplying an empty left hand side argument is not supported in hash-pairs";
657 my $method = $self->_METHOD_FOR_refkind("_where_hashpair", $v);
658 $self->$method($k, $v);
662 push @sql_clauses, $sql;
663 push @all_bind, @bind;
666 return $self->_join_sql_clauses('and', \@sql_clauses, \@all_bind);
669 sub _where_unary_op {
670 my ($self, $op, $rhs) = @_;
672 # top level special ops are illegal in general
673 # this includes the -ident/-value ops (dual purpose unary and special)
674 puke "Illegal use of top-level '-$op'"
675 if ! defined $self->{_nested_func_lhs} and List::Util::first { $op =~ $_->{regex} } @{$self->{special_ops}};
677 if (my $op_entry = List::Util::first { $op =~ $_->{regex} } @{$self->{unary_ops}}) {
678 my $handler = $op_entry->{handler};
680 if (not ref $handler) {
681 if ($op =~ s/ [_\s]? \d+ $//x ) {
682 belch 'Use of [and|or|nest]_N modifiers is deprecated and will be removed in SQLA v2.0. '
683 . "You probably wanted ...-and => [ -$op => COND1, -$op => COND2 ... ]";
685 return $self->$handler($op, $rhs);
687 elsif (ref $handler eq 'CODE') {
688 return $handler->($self, $op, $rhs);
691 puke "Illegal handler for operator $op - expecting a method name or a coderef";
695 $self->_debug("Generic unary OP: $op - recursing as function");
697 $self->_assert_pass_injection_guard($op);
699 my ($sql, @bind) = $self->_SWITCH_refkind($rhs, {
701 puke "Illegal use of top-level '-$op'"
702 unless defined $self->{_nested_func_lhs};
705 $self->_convert('?'),
706 $self->_bindtype($self->{_nested_func_lhs}, $rhs)
710 $self->_recurse_where($rhs)
714 $sql = sprintf('%s %s',
715 $self->_sqlcase($op),
719 return ($sql, @bind);
722 sub _where_op_ANDOR {
723 my ($self, $op, $v) = @_;
725 $self->_SWITCH_refkind($v, {
727 return $self->_where_ARRAYREF($v, $op);
731 return ($op =~ /^or/i)
732 ? $self->_where_ARRAYREF([ map { $_ => $v->{$_} } (sort keys %$v) ], $op)
733 : $self->_where_HASHREF($v);
737 puke "-$op => \\\$scalar makes little sense, use " .
739 ? '[ \$scalar, \%rest_of_conditions ] instead'
740 : '-and => [ \$scalar, \%rest_of_conditions ] instead'
745 puke "-$op => \\[...] makes little sense, use " .
747 ? '[ \[...], \%rest_of_conditions ] instead'
748 : '-and => [ \[...], \%rest_of_conditions ] instead'
752 SCALAR => sub { # permissively interpreted as SQL
753 puke "-$op => \$value makes little sense, use -bool => \$value instead";
757 puke "-$op => undef not supported";
763 my ($self, $op, $v) = @_;
765 $self->_SWITCH_refkind($v, {
767 SCALAR => sub { # permissively interpreted as SQL
768 belch "literal SQL should be -nest => \\'scalar' "
769 . "instead of -nest => 'scalar' ";
774 puke "-$op => undef not supported";
778 $self->_recurse_where($v);
786 my ($self, $op, $v) = @_;
788 my ($s, @b) = $self->_SWITCH_refkind($v, {
789 SCALAR => sub { # interpreted as SQL column
790 $self->_convert($self->_quote($v));
794 puke "-$op => undef not supported";
798 $self->_recurse_where($v);
802 $s = "(NOT $s)" if $op =~ /^not/i;
807 sub _where_op_IDENT {
809 my ($op, $rhs) = splice @_, -2;
810 if (! defined $rhs or length ref $rhs) {
811 puke "-$op requires a single plain scalar argument (a quotable identifier)";
814 # in case we are called as a top level special op (no '=')
817 $_ = $self->_convert($self->_quote($_)) for ($lhs, $rhs);
825 sub _where_op_VALUE {
827 my ($op, $rhs) = splice @_, -2;
829 # in case we are called as a top level special op (no '=')
833 if (! defined $rhs) {
835 ? $self->_convert($self->_quote($lhs)) . ' IS NULL'
842 (defined $lhs ? $lhs : $self->{_nested_func_lhs}),
849 $self->_convert($self->_quote($lhs)) . ' = ' . $self->_convert('?'),
853 $self->_convert('?'),
859 sub _where_hashpair_ARRAYREF {
860 my ($self, $k, $v) = @_;
863 my @v = @$v; # need copy because of shift below
864 $self->_debug("ARRAY($k) means distribute over elements");
866 # put apart first element if it is an operator (-and, -or)
868 (defined $v[0] && $v[0] =~ /^ - (?: AND|OR ) $/ix)
872 my @distributed = map { {$k => $_} } @v;
875 $self->_debug("OP($op) reinjected into the distributed array");
876 unshift @distributed, $op;
879 my $logic = $op ? substr($op, 1) : '';
881 return $self->_recurse_where(\@distributed, $logic);
884 $self->_debug("empty ARRAY($k) means 0=1");
885 return ($self->{sqlfalse});
889 sub _where_hashpair_HASHREF {
890 my ($self, $k, $v, $logic) = @_;
893 local $self->{_nested_func_lhs} = defined $self->{_nested_func_lhs}
894 ? $self->{_nested_func_lhs}
898 my ($all_sql, @all_bind);
900 for my $orig_op (sort keys %$v) {
901 my $val = $v->{$orig_op};
903 # put the operator in canonical form
906 # FIXME - we need to phase out dash-less ops
907 $op =~ s/^-//; # remove possible initial dash
908 $op =~ s/^\s+|\s+$//g;# remove leading/trailing space
909 $op =~ s/\s+/ /g; # compress whitespace
911 $self->_assert_pass_injection_guard($op);
914 $op =~ s/^is_not/IS NOT/i;
916 # so that -not_foo works correctly
917 $op =~ s/^not_/NOT /i;
919 # another retarded special case: foo => { $op => { -value => undef } }
920 if (ref $val eq 'HASH' and keys %$val == 1 and exists $val->{-value} and ! defined $val->{-value} ) {
926 # CASE: col-value logic modifiers
927 if ($orig_op =~ /^ \- (and|or) $/xi) {
928 ($sql, @bind) = $self->_where_hashpair_HASHREF($k, $val, $1);
930 # CASE: special operators like -in or -between
931 elsif (my $special_op = List::Util::first { $op =~ $_->{regex} } @{$self->{special_ops}}) {
932 my $handler = $special_op->{handler};
934 puke "No handler supplied for special operator $orig_op";
936 elsif (not ref $handler) {
937 ($sql, @bind) = $self->$handler($k, $op, $val);
939 elsif (ref $handler eq 'CODE') {
940 ($sql, @bind) = $handler->($self, $k, $op, $val);
943 puke "Illegal handler for special operator $orig_op - expecting a method name or a coderef";
947 $self->_SWITCH_refkind($val, {
949 ARRAYREF => sub { # CASE: col => {op => \@vals}
950 ($sql, @bind) = $self->_where_field_op_ARRAYREF($k, $op, $val);
953 ARRAYREFREF => sub { # CASE: col => {op => \[$sql, @bind]} (literal SQL with bind)
954 my ($sub_sql, @sub_bind) = @$$val;
955 $self->_assert_bindval_matches_bindtype(@sub_bind);
956 $sql = join ' ', $self->_convert($self->_quote($k)),
957 $self->_sqlcase($op),
962 UNDEF => sub { # CASE: col => {op => undef} : sql "IS (NOT)? NULL"
964 $op =~ /^not$/i ? 'is not' # legacy
965 : $op =~ $self->{equality_op} ? 'is'
966 : $op =~ $self->{like_op} ? belch("Supplying an undefined argument to '@{[ uc $op]}' is deprecated") && 'is'
967 : $op =~ $self->{inequality_op} ? 'is not'
968 : $op =~ $self->{not_like_op} ? belch("Supplying an undefined argument to '@{[ uc $op]}' is deprecated") && 'is not'
969 : puke "unexpected operator '$orig_op' with undef operand";
971 $sql = $self->_quote($k) . $self->_sqlcase(" $is null");
974 FALLBACK => sub { # CASE: col => {op/func => $stuff}
975 ($sql, @bind) = $self->_where_unary_op($op, $val);
978 $self->_convert($self->_quote($k)),
979 $self->{_nested_func_lhs} eq $k ? $sql : "($sql)", # top level vs nested
985 ($all_sql) = (defined $all_sql and $all_sql) ? $self->_join_sql_clauses($logic, [$all_sql, $sql], []) : $sql;
986 push @all_bind, @bind;
988 return ($all_sql, @all_bind);
991 sub _where_field_IS {
992 my ($self, $k, $op, $v) = @_;
994 my ($s) = $self->_SWITCH_refkind($v, {
997 $self->_convert($self->_quote($k)),
998 map { $self->_sqlcase($_)} ($op, 'null')
1001 puke "$op can only take undef as argument";
1008 sub _where_field_op_ARRAYREF {
1009 my ($self, $k, $op, $vals) = @_;
1011 my @vals = @$vals; #always work on a copy
1014 $self->_debug(sprintf '%s means multiple elements: [ %s ]',
1016 join(', ', map { defined $_ ? "'$_'" : 'NULL' } @vals ),
1019 # see if the first element is an -and/-or op
1021 if (defined $vals[0] && $vals[0] =~ /^ - (AND|OR) $/ix) {
1026 # a long standing API wart - an attempt to change this behavior during
1027 # the 1.50 series failed *spectacularly*. Warn instead and leave the
1032 (!$logic or $logic eq 'OR')
1034 ($op =~ $self->{inequality_op} or $op =~ $self->{not_like_op})
1037 belch "A multi-element arrayref as an argument to the inequality op '$o' "
1038 . 'is technically equivalent to an always-true 1=1 (you probably wanted '
1039 . "to say ...{ \$inequality_op => [ -and => \@values ] }... instead)"
1043 # distribute $op over each remaining member of @vals, append logic if exists
1044 return $self->_recurse_where([map { {$k => {$op, $_}} } @vals], $logic);
1048 # try to DWIM on equality operators
1050 $op =~ $self->{equality_op} ? $self->{sqlfalse}
1051 : $op =~ $self->{like_op} ? belch("Supplying an empty arrayref to '@{[ uc $op]}' is deprecated") && $self->{sqlfalse}
1052 : $op =~ $self->{inequality_op} ? $self->{sqltrue}
1053 : $op =~ $self->{not_like_op} ? belch("Supplying an empty arrayref to '@{[ uc $op]}' is deprecated") && $self->{sqltrue}
1054 : puke "operator '$op' applied on an empty array (field '$k')";
1059 sub _where_hashpair_SCALARREF {
1060 my ($self, $k, $v) = @_;
1061 $self->_debug("SCALAR($k) means literal SQL: $$v");
1062 my $sql = $self->_quote($k) . " " . $$v;
1066 # literal SQL with bind
1067 sub _where_hashpair_ARRAYREFREF {
1068 my ($self, $k, $v) = @_;
1069 $self->_debug("REF($k) means literal SQL: @${$v}");
1070 my ($sql, @bind) = @$$v;
1071 $self->_assert_bindval_matches_bindtype(@bind);
1072 $sql = $self->_quote($k) . " " . $sql;
1073 return ($sql, @bind );
1076 # literal SQL without bind
1077 sub _where_hashpair_SCALAR {
1078 my ($self, $k, $v) = @_;
1079 $self->_debug("NOREF($k) means simple key=val: $k $self->{cmp} $v");
1080 my $sql = join ' ', $self->_convert($self->_quote($k)),
1081 $self->_sqlcase($self->{cmp}),
1082 $self->_convert('?');
1083 my @bind = $self->_bindtype($k, $v);
1084 return ($sql, @bind);
1088 sub _where_hashpair_UNDEF {
1089 my ($self, $k, $v) = @_;
1090 $self->_debug("UNDEF($k) means IS NULL");
1091 my $sql = $self->_quote($k) . $self->_sqlcase(' is null');
1095 #======================================================================
1096 # WHERE: TOP-LEVEL OTHERS (SCALARREF, SCALAR, UNDEF)
1097 #======================================================================
1100 sub _where_SCALARREF {
1101 my ($self, $where) = @_;
1104 $self->_debug("SCALAR(*top) means literal SQL: $$where");
1110 my ($self, $where) = @_;
1113 $self->_debug("NOREF(*top) means literal SQL: $where");
1124 #======================================================================
1125 # WHERE: BUILTIN SPECIAL OPERATORS (-in, -between)
1126 #======================================================================
1129 sub _where_field_BETWEEN {
1130 my ($self, $k, $op, $vals) = @_;
1132 my ($label, $and, $placeholder);
1133 $label = $self->_convert($self->_quote($k));
1134 $and = ' ' . $self->_sqlcase('and') . ' ';
1135 $placeholder = $self->_convert('?');
1136 $op = $self->_sqlcase($op);
1138 my $invalid_args = "Operator '$op' requires either an arrayref with two defined values or expressions, or a single literal scalarref/arrayref-ref";
1140 my ($clause, @bind) = $self->_SWITCH_refkind($vals, {
1141 ARRAYREFREF => sub {
1142 my ($s, @b) = @$$vals;
1143 $self->_assert_bindval_matches_bindtype(@b);
1150 puke $invalid_args if @$vals != 2;
1152 my (@all_sql, @all_bind);
1153 foreach my $val (@$vals) {
1154 my ($sql, @bind) = $self->_SWITCH_refkind($val, {
1156 return ($placeholder, $self->_bindtype($k, $val) );
1161 ARRAYREFREF => sub {
1162 my ($sql, @bind) = @$$val;
1163 $self->_assert_bindval_matches_bindtype(@bind);
1164 return ($sql, @bind);
1167 my ($func, $arg, @rest) = %$val;
1168 puke "Only simple { -func => arg } functions accepted as sub-arguments to BETWEEN"
1169 if (@rest or $func !~ /^ \- (.+)/x);
1170 $self->_where_unary_op($1 => $arg);
1176 push @all_sql, $sql;
1177 push @all_bind, @bind;
1181 (join $and, @all_sql),
1190 my $sql = "( $label $op $clause )";
1191 return ($sql, @bind)
1195 sub _where_field_IN {
1196 my ($self, $k, $op, $vals) = @_;
1198 # backwards compatibility : if scalar, force into an arrayref
1199 $vals = [$vals] if defined $vals && ! ref $vals;
1201 my ($label) = $self->_convert($self->_quote($k));
1202 my ($placeholder) = $self->_convert('?');
1203 $op = $self->_sqlcase($op);
1205 my ($sql, @bind) = $self->_SWITCH_refkind($vals, {
1206 ARRAYREF => sub { # list of choices
1207 if (@$vals) { # nonempty list
1208 my (@all_sql, @all_bind);
1210 for my $val (@$vals) {
1211 my ($sql, @bind) = $self->_SWITCH_refkind($val, {
1213 return ($placeholder, $val);
1218 ARRAYREFREF => sub {
1219 my ($sql, @bind) = @$$val;
1220 $self->_assert_bindval_matches_bindtype(@bind);
1221 return ($sql, @bind);
1224 my ($func, $arg, @rest) = %$val;
1225 puke "Only simple { -func => arg } functions accepted as sub-arguments to IN"
1226 if (@rest or $func !~ /^ \- (.+)/x);
1227 $self->_where_unary_op($1 => $arg);
1231 'SQL::Abstract before v1.75 used to generate incorrect SQL when the '
1232 . "-$op operator was given an undef-containing list: !!!AUDIT YOUR CODE "
1233 . 'AND DATA!!! (the upcoming Data::Query-based version of SQL::Abstract '
1234 . 'will emit the logically correct SQL instead of raising this exception)'
1238 push @all_sql, $sql;
1239 push @all_bind, @bind;
1243 sprintf('%s %s ( %s )',
1246 join(', ', @all_sql)
1248 $self->_bindtype($k, @all_bind),
1251 else { # empty list : some databases won't understand "IN ()", so DWIM
1252 my $sql = ($op =~ /\bnot\b/i) ? $self->{sqltrue} : $self->{sqlfalse};
1257 SCALARREF => sub { # literal SQL
1258 my $sql = $self->_open_outer_paren($$vals);
1259 return ("$label $op ( $sql )");
1261 ARRAYREFREF => sub { # literal SQL with bind
1262 my ($sql, @bind) = @$$vals;
1263 $self->_assert_bindval_matches_bindtype(@bind);
1264 $sql = $self->_open_outer_paren($sql);
1265 return ("$label $op ( $sql )", @bind);
1269 puke "Argument passed to the '$op' operator can not be undefined";
1273 puke "special op $op requires an arrayref (or scalarref/arrayref-ref)";
1277 return ($sql, @bind);
1280 # Some databases (SQLite) treat col IN (1, 2) different from
1281 # col IN ( (1, 2) ). Use this to strip all outer parens while
1282 # adding them back in the corresponding method
1283 sub _open_outer_paren {
1284 my ($self, $sql) = @_;
1286 while (my ($inner) = $sql =~ /^ \s* \( (.*) \) \s* $/xs) {
1288 # there are closing parens inside, need the heavy duty machinery
1289 # to reevaluate the extraction starting from $sql (full reevaluation)
1290 if ($inner =~ /\)/) {
1291 require Text::Balanced;
1293 my (undef, $remainder) = do {
1294 # idiotic design - writes to $@ but *DOES NOT* throw exceptions
1296 Text::Balanced::extract_bracketed($sql, '()', qr/\s*/);
1299 # the entire expression needs to be a balanced bracketed thing
1300 # (after an extract no remainder sans trailing space)
1301 last if defined $remainder and $remainder =~ /\S/;
1311 #======================================================================
1313 #======================================================================
1316 my ($self, $arg) = @_;
1319 for my $c ($self->_order_by_chunks($arg) ) {
1320 $self->_SWITCH_refkind($c, {
1321 SCALAR => sub { push @sql, $c },
1322 ARRAYREF => sub { push @sql, shift @$c; push @bind, @$c },
1328 $self->_sqlcase(' order by'),
1334 return wantarray ? ($sql, @bind) : $sql;
1337 sub _order_by_chunks {
1338 my ($self, $arg) = @_;
1340 return $self->_SWITCH_refkind($arg, {
1343 map { $self->_order_by_chunks($_ ) } @$arg;
1346 ARRAYREFREF => sub {
1347 my ($s, @b) = @$$arg;
1348 $self->_assert_bindval_matches_bindtype(@b);
1352 SCALAR => sub {$self->_quote($arg)},
1354 UNDEF => sub {return () },
1356 SCALARREF => sub {$$arg}, # literal SQL, no quoting
1359 # get first pair in hash
1360 my ($key, $val, @rest) = %$arg;
1362 return () unless $key;
1364 if (@rest or not $key =~ /^-(desc|asc)/i) {
1365 puke "hash passed to _order_by must have exactly one key (-desc or -asc)";
1371 for my $c ($self->_order_by_chunks($val)) {
1374 $self->_SWITCH_refkind($c, {
1379 ($sql, @bind) = @$c;
1383 $sql = $sql . ' ' . $self->_sqlcase($direction);
1385 push @ret, [ $sql, @bind];
1394 #======================================================================
1395 # DATASOURCE (FOR NOW, JUST PLAIN TABLE OR LIST OF TABLES)
1396 #======================================================================
1401 $self->_SWITCH_refkind($from, {
1402 ARRAYREF => sub {join ', ', map { $self->_quote($_) } @$from;},
1403 SCALAR => sub {$self->_quote($from)},
1404 SCALARREF => sub {$$from},
1409 #======================================================================
1411 #======================================================================
1413 # highly optimized, as it's called way too often
1415 # my ($self, $label) = @_;
1417 return '' unless defined $_[1];
1418 return ${$_[1]} if ref($_[1]) eq 'SCALAR';
1420 $_[0]->{quote_char} or
1421 ($_[0]->_assert_pass_injection_guard($_[1]), return $_[1]);
1423 my $qref = ref $_[0]->{quote_char};
1425 !$qref ? ($_[0]->{quote_char}, $_[0]->{quote_char})
1426 : ($qref eq 'ARRAY') ? @{$_[0]->{quote_char}}
1427 : puke "Unsupported quote_char format: $_[0]->{quote_char}";
1429 my $esc = $_[0]->{escape_char} || $r;
1431 # parts containing * are naturally unquoted
1432 return join($_[0]->{name_sep}||'', map
1433 +( $_ eq '*' ? $_ : do { (my $n = $_) =~ s/(\Q$esc\E|\Q$r\E)/$esc$1/g; $l . $n . $r } ),
1434 ( $_[0]->{name_sep} ? split (/\Q$_[0]->{name_sep}\E/, $_[1] ) : $_[1] )
1439 # Conversion, if applicable
1441 #my ($self, $arg) = @_;
1442 if ($_[0]->{convert}) {
1443 return $_[0]->_sqlcase($_[0]->{convert}) .'(' . $_[1] . ')';
1450 #my ($self, $col, @vals) = @_;
1451 # called often - tighten code
1452 return $_[0]->{bindtype} eq 'columns'
1453 ? map {[$_[1], $_]} @_[2 .. $#_]
1458 # Dies if any element of @bind is not in [colname => value] format
1459 # if bindtype is 'columns'.
1460 sub _assert_bindval_matches_bindtype {
1461 # my ($self, @bind) = @_;
1463 if ($self->{bindtype} eq 'columns') {
1465 if (!defined $_ || ref($_) ne 'ARRAY' || @$_ != 2) {
1466 puke "bindtype 'columns' selected, you need to pass: [column_name => bind_value]"
1472 sub _join_sql_clauses {
1473 my ($self, $logic, $clauses_aref, $bind_aref) = @_;
1475 if (@$clauses_aref > 1) {
1476 my $join = " " . $self->_sqlcase($logic) . " ";
1477 my $sql = '( ' . join($join, @$clauses_aref) . ' )';
1478 return ($sql, @$bind_aref);
1480 elsif (@$clauses_aref) {
1481 return ($clauses_aref->[0], @$bind_aref); # no parentheses
1484 return (); # if no SQL, ignore @$bind_aref
1489 # Fix SQL case, if so requested
1491 # LDNOTE: if $self->{case} is true, then it contains 'lower', so we
1492 # don't touch the argument ... crooked logic, but let's not change it!
1493 return $_[0]->{case} ? $_[1] : uc($_[1]);
1497 #======================================================================
1498 # DISPATCHING FROM REFKIND
1499 #======================================================================
1502 my ($self, $data) = @_;
1504 return 'UNDEF' unless defined $data;
1506 # blessed objects are treated like scalars
1507 my $ref = (Scalar::Util::blessed $data) ? '' : ref $data;
1509 return 'SCALAR' unless $ref;
1512 while ($ref eq 'REF') {
1514 $ref = (Scalar::Util::blessed $data) ? '' : ref $data;
1518 return ($ref||'SCALAR') . ('REF' x $n_steps);
1522 my ($self, $data) = @_;
1523 my @try = ($self->_refkind($data));
1524 push @try, 'SCALAR_or_UNDEF' if $try[0] eq 'SCALAR' || $try[0] eq 'UNDEF';
1525 push @try, 'FALLBACK';
1529 sub _METHOD_FOR_refkind {
1530 my ($self, $meth_prefix, $data) = @_;
1533 for (@{$self->_try_refkind($data)}) {
1534 $method = $self->can($meth_prefix."_".$_)
1538 return $method || puke "cannot dispatch on '$meth_prefix' for ".$self->_refkind($data);
1542 sub _SWITCH_refkind {
1543 my ($self, $data, $dispatch_table) = @_;
1546 for (@{$self->_try_refkind($data)}) {
1547 $coderef = $dispatch_table->{$_}
1551 puke "no dispatch entry for ".$self->_refkind($data)
1560 #======================================================================
1561 # VALUES, GENERATE, AUTOLOAD
1562 #======================================================================
1564 # LDNOTE: original code from nwiger, didn't touch code in that section
1565 # I feel the AUTOLOAD stuff should not be the default, it should
1566 # only be activated on explicit demand by user.
1570 my $data = shift || return;
1571 puke "Argument to ", __PACKAGE__, "->values must be a \\%hash"
1572 unless ref $data eq 'HASH';
1575 foreach my $k (sort keys %$data) {
1576 my $v = $data->{$k};
1577 $self->_SWITCH_refkind($v, {
1579 if ($self->{array_datatypes}) { # array datatype
1580 push @all_bind, $self->_bindtype($k, $v);
1582 else { # literal SQL with bind
1583 my ($sql, @bind) = @$v;
1584 $self->_assert_bindval_matches_bindtype(@bind);
1585 push @all_bind, @bind;
1588 ARRAYREFREF => sub { # literal SQL with bind
1589 my ($sql, @bind) = @${$v};
1590 $self->_assert_bindval_matches_bindtype(@bind);
1591 push @all_bind, @bind;
1593 SCALARREF => sub { # literal SQL without bind
1595 SCALAR_or_UNDEF => sub {
1596 push @all_bind, $self->_bindtype($k, $v);
1607 my(@sql, @sqlq, @sqlv);
1611 if ($ref eq 'HASH') {
1612 for my $k (sort keys %$_) {
1615 my $label = $self->_quote($k);
1616 if ($r eq 'ARRAY') {
1617 # literal SQL with bind
1618 my ($sql, @bind) = @$v;
1619 $self->_assert_bindval_matches_bindtype(@bind);
1620 push @sqlq, "$label = $sql";
1622 } elsif ($r eq 'SCALAR') {
1623 # literal SQL without bind
1624 push @sqlq, "$label = $$v";
1626 push @sqlq, "$label = ?";
1627 push @sqlv, $self->_bindtype($k, $v);
1630 push @sql, $self->_sqlcase('set'), join ', ', @sqlq;
1631 } elsif ($ref eq 'ARRAY') {
1632 # unlike insert(), assume these are ONLY the column names, i.e. for SQL
1635 if ($r eq 'ARRAY') { # literal SQL with bind
1636 my ($sql, @bind) = @$v;
1637 $self->_assert_bindval_matches_bindtype(@bind);
1640 } elsif ($r eq 'SCALAR') { # literal SQL without bind
1641 # embedded literal SQL
1648 push @sql, '(' . join(', ', @sqlq) . ')';
1649 } elsif ($ref eq 'SCALAR') {
1653 # strings get case twiddled
1654 push @sql, $self->_sqlcase($_);
1658 my $sql = join ' ', @sql;
1660 # this is pretty tricky
1661 # if ask for an array, return ($stmt, @bind)
1662 # otherwise, s/?/shift @sqlv/ to put it inline
1664 return ($sql, @sqlv);
1666 1 while $sql =~ s/\?/my $d = shift(@sqlv);
1667 ref $d ? $d->[1] : $d/e;
1676 # This allows us to check for a local, then _form, attr
1678 my($name) = $AUTOLOAD =~ /.*::(.+)/;
1679 return $self->generate($name, @_);
1690 SQL::Abstract - Generate SQL from Perl data structures
1696 my $sql = SQL::Abstract->new;
1698 my($stmt, @bind) = $sql->select($source, \@fields, \%where, $order);
1700 my($stmt, @bind) = $sql->insert($table, \%fieldvals || \@values);
1702 my($stmt, @bind) = $sql->update($table, \%fieldvals, \%where);
1704 my($stmt, @bind) = $sql->delete($table, \%where);
1706 # Then, use these in your DBI statements
1707 my $sth = $dbh->prepare($stmt);
1708 $sth->execute(@bind);
1710 # Just generate the WHERE clause
1711 my($stmt, @bind) = $sql->where(\%where, $order);
1713 # Return values in the same order, for hashed queries
1714 # See PERFORMANCE section for more details
1715 my @bind = $sql->values(\%fieldvals);
1719 This module was inspired by the excellent L<DBIx::Abstract>.
1720 However, in using that module I found that what I really wanted
1721 to do was generate SQL, but still retain complete control over my
1722 statement handles and use the DBI interface. So, I set out to
1723 create an abstract SQL generation module.
1725 While based on the concepts used by L<DBIx::Abstract>, there are
1726 several important differences, especially when it comes to WHERE
1727 clauses. I have modified the concepts used to make the SQL easier
1728 to generate from Perl data structures and, IMO, more intuitive.
1729 The underlying idea is for this module to do what you mean, based
1730 on the data structures you provide it. The big advantage is that
1731 you don't have to modify your code every time your data changes,
1732 as this module figures it out.
1734 To begin with, an SQL INSERT is as easy as just specifying a hash
1735 of C<key=value> pairs:
1738 name => 'Jimbo Bobson',
1739 phone => '123-456-7890',
1740 address => '42 Sister Lane',
1741 city => 'St. Louis',
1742 state => 'Louisiana',
1745 The SQL can then be generated with this:
1747 my($stmt, @bind) = $sql->insert('people', \%data);
1749 Which would give you something like this:
1751 $stmt = "INSERT INTO people
1752 (address, city, name, phone, state)
1753 VALUES (?, ?, ?, ?, ?)";
1754 @bind = ('42 Sister Lane', 'St. Louis', 'Jimbo Bobson',
1755 '123-456-7890', 'Louisiana');
1757 These are then used directly in your DBI code:
1759 my $sth = $dbh->prepare($stmt);
1760 $sth->execute(@bind);
1762 =head2 Inserting and Updating Arrays
1764 If your database has array types (like for example Postgres),
1765 activate the special option C<< array_datatypes => 1 >>
1766 when creating the C<SQL::Abstract> object.
1767 Then you may use an arrayref to insert and update database array types:
1769 my $sql = SQL::Abstract->new(array_datatypes => 1);
1771 planets => [qw/Mercury Venus Earth Mars/]
1774 my($stmt, @bind) = $sql->insert('solar_system', \%data);
1778 $stmt = "INSERT INTO solar_system (planets) VALUES (?)"
1780 @bind = (['Mercury', 'Venus', 'Earth', 'Mars']);
1783 =head2 Inserting and Updating SQL
1785 In order to apply SQL functions to elements of your C<%data> you may
1786 specify a reference to an arrayref for the given hash value. For example,
1787 if you need to execute the Oracle C<to_date> function on a value, you can
1788 say something like this:
1792 date_entered => \[ "to_date(?,'MM/DD/YYYY')", "03/02/2003" ],
1795 The first value in the array is the actual SQL. Any other values are
1796 optional and would be included in the bind values array. This gives
1799 my($stmt, @bind) = $sql->insert('people', \%data);
1801 $stmt = "INSERT INTO people (name, date_entered)
1802 VALUES (?, to_date(?,'MM/DD/YYYY'))";
1803 @bind = ('Bill', '03/02/2003');
1805 An UPDATE is just as easy, all you change is the name of the function:
1807 my($stmt, @bind) = $sql->update('people', \%data);
1809 Notice that your C<%data> isn't touched; the module will generate
1810 the appropriately quirky SQL for you automatically. Usually you'll
1811 want to specify a WHERE clause for your UPDATE, though, which is
1812 where handling C<%where> hashes comes in handy...
1814 =head2 Complex where statements
1816 This module can generate pretty complicated WHERE statements
1817 easily. For example, simple C<key=value> pairs are taken to mean
1818 equality, and if you want to see if a field is within a set
1819 of values, you can use an arrayref. Let's say we wanted to
1820 SELECT some data based on this criteria:
1823 requestor => 'inna',
1824 worker => ['nwiger', 'rcwe', 'sfz'],
1825 status => { '!=', 'completed' }
1828 my($stmt, @bind) = $sql->select('tickets', '*', \%where);
1830 The above would give you something like this:
1832 $stmt = "SELECT * FROM tickets WHERE
1833 ( requestor = ? ) AND ( status != ? )
1834 AND ( worker = ? OR worker = ? OR worker = ? )";
1835 @bind = ('inna', 'completed', 'nwiger', 'rcwe', 'sfz');
1837 Which you could then use in DBI code like so:
1839 my $sth = $dbh->prepare($stmt);
1840 $sth->execute(@bind);
1846 The methods are simple. There's one for every major SQL operation,
1847 and a constructor you use first. The arguments are specified in a
1848 similar order for each method (table, then fields, then a where
1849 clause) to try and simplify things.
1851 =head2 new(option => 'value')
1853 The C<new()> function takes a list of options and values, and returns
1854 a new B<SQL::Abstract> object which can then be used to generate SQL
1855 through the methods below. The options accepted are:
1861 If set to 'lower', then SQL will be generated in all lowercase. By
1862 default SQL is generated in "textbook" case meaning something like:
1864 SELECT a_field FROM a_table WHERE some_field LIKE '%someval%'
1866 Any setting other than 'lower' is ignored.
1870 This determines what the default comparison operator is. By default
1871 it is C<=>, meaning that a hash like this:
1873 %where = (name => 'nwiger', email => 'nate@wiger.org');
1875 Will generate SQL like this:
1877 WHERE name = 'nwiger' AND email = 'nate@wiger.org'
1879 However, you may want loose comparisons by default, so if you set
1880 C<cmp> to C<like> you would get SQL such as:
1882 WHERE name like 'nwiger' AND email like 'nate@wiger.org'
1884 You can also override the comparison on an individual basis - see
1885 the huge section on L</"WHERE CLAUSES"> at the bottom.
1887 =item sqltrue, sqlfalse
1889 Expressions for inserting boolean values within SQL statements.
1890 By default these are C<1=1> and C<1=0>. They are used
1891 by the special operators C<-in> and C<-not_in> for generating
1892 correct SQL even when the argument is an empty array (see below).
1896 This determines the default logical operator for multiple WHERE
1897 statements in arrays or hashes. If absent, the default logic is "or"
1898 for arrays, and "and" for hashes. This means that a WHERE
1902 event_date => {'>=', '2/13/99'},
1903 event_date => {'<=', '4/24/03'},
1906 will generate SQL like this:
1908 WHERE event_date >= '2/13/99' OR event_date <= '4/24/03'
1910 This is probably not what you want given this query, though (look
1911 at the dates). To change the "OR" to an "AND", simply specify:
1913 my $sql = SQL::Abstract->new(logic => 'and');
1915 Which will change the above C<WHERE> to:
1917 WHERE event_date >= '2/13/99' AND event_date <= '4/24/03'
1919 The logic can also be changed locally by inserting
1920 a modifier in front of an arrayref :
1922 @where = (-and => [event_date => {'>=', '2/13/99'},
1923 event_date => {'<=', '4/24/03'} ]);
1925 See the L</"WHERE CLAUSES"> section for explanations.
1929 This will automatically convert comparisons using the specified SQL
1930 function for both column and value. This is mostly used with an argument
1931 of C<upper> or C<lower>, so that the SQL will have the effect of
1932 case-insensitive "searches". For example, this:
1934 $sql = SQL::Abstract->new(convert => 'upper');
1935 %where = (keywords => 'MaKe iT CAse inSeNSItive');
1937 Will turn out the following SQL:
1939 WHERE upper(keywords) like upper('MaKe iT CAse inSeNSItive')
1941 The conversion can be C<upper()>, C<lower()>, or any other SQL function
1942 that can be applied symmetrically to fields (actually B<SQL::Abstract> does
1943 not validate this option; it will just pass through what you specify verbatim).
1947 This is a kludge because many databases suck. For example, you can't
1948 just bind values using DBI's C<execute()> for Oracle C<CLOB> or C<BLOB> fields.
1949 Instead, you have to use C<bind_param()>:
1951 $sth->bind_param(1, 'reg data');
1952 $sth->bind_param(2, $lots, {ora_type => ORA_CLOB});
1954 The problem is, B<SQL::Abstract> will normally just return a C<@bind> array,
1955 which loses track of which field each slot refers to. Fear not.
1957 If you specify C<bindtype> in new, you can determine how C<@bind> is returned.
1958 Currently, you can specify either C<normal> (default) or C<columns>. If you
1959 specify C<columns>, you will get an array that looks like this:
1961 my $sql = SQL::Abstract->new(bindtype => 'columns');
1962 my($stmt, @bind) = $sql->insert(...);
1965 [ 'column1', 'value1' ],
1966 [ 'column2', 'value2' ],
1967 [ 'column3', 'value3' ],
1970 You can then iterate through this manually, using DBI's C<bind_param()>.
1972 $sth->prepare($stmt);
1975 my($col, $data) = @$_;
1976 if ($col eq 'details' || $col eq 'comments') {
1977 $sth->bind_param($i, $data, {ora_type => ORA_CLOB});
1978 } elsif ($col eq 'image') {
1979 $sth->bind_param($i, $data, {ora_type => ORA_BLOB});
1981 $sth->bind_param($i, $data);
1985 $sth->execute; # execute without @bind now
1987 Now, why would you still use B<SQL::Abstract> if you have to do this crap?
1988 Basically, the advantage is still that you don't have to care which fields
1989 are or are not included. You could wrap that above C<for> loop in a simple
1990 sub called C<bind_fields()> or something and reuse it repeatedly. You still
1991 get a layer of abstraction over manual SQL specification.
1993 Note that if you set L</bindtype> to C<columns>, the C<\[ $sql, @bind ]>
1994 construct (see L</Literal SQL with placeholders and bind values (subqueries)>)
1995 will expect the bind values in this format.
1999 This is the character that a table or column name will be quoted
2000 with. By default this is an empty string, but you could set it to
2001 the character C<`>, to generate SQL like this:
2003 SELECT `a_field` FROM `a_table` WHERE `some_field` LIKE '%someval%'
2005 Alternatively, you can supply an array ref of two items, the first being the left
2006 hand quote character, and the second the right hand quote character. For
2007 example, you could supply C<['[',']']> for SQL Server 2000 compliant quotes
2008 that generates SQL like this:
2010 SELECT [a_field] FROM [a_table] WHERE [some_field] LIKE '%someval%'
2012 Quoting is useful if you have tables or columns names that are reserved
2013 words in your database's SQL dialect.
2017 This is the character that will be used to escape L</quote_char>s appearing
2018 in an identifier before it has been quoted.
2020 The parameter default in case of a single L</quote_char> character is the quote
2023 When opening-closing-style quoting is used (L</quote_char> is an arrayref)
2024 this parameter defaults to the B<closing (right)> L</quote_char>. Occurences
2025 of the B<opening (left)> L</quote_char> within the identifier are currently left
2026 untouched. The default for opening-closing-style quotes may change in future
2027 versions, thus you are B<strongly encouraged> to specify the escape character
2032 This is the character that separates a table and column name. It is
2033 necessary to specify this when the C<quote_char> option is selected,
2034 so that tables and column names can be individually quoted like this:
2036 SELECT `table`.`one_field` FROM `table` WHERE `table`.`other_field` = 1
2038 =item injection_guard
2040 A regular expression C<qr/.../> that is applied to any C<-function> and unquoted
2041 column name specified in a query structure. This is a safety mechanism to avoid
2042 injection attacks when mishandling user input e.g.:
2044 my %condition_as_column_value_pairs = get_values_from_user();
2045 $sqla->select( ... , \%condition_as_column_value_pairs );
2047 If the expression matches an exception is thrown. Note that literal SQL
2048 supplied via C<\'...'> or C<\['...']> is B<not> checked in any way.
2050 Defaults to checking for C<;> and the C<GO> keyword (TransactSQL)
2052 =item array_datatypes
2054 When this option is true, arrayrefs in INSERT or UPDATE are
2055 interpreted as array datatypes and are passed directly
2057 When this option is false, arrayrefs are interpreted
2058 as literal SQL, just like refs to arrayrefs
2059 (but this behavior is for backwards compatibility; when writing
2060 new queries, use the "reference to arrayref" syntax
2066 Takes a reference to a list of "special operators"
2067 to extend the syntax understood by L<SQL::Abstract>.
2068 See section L</"SPECIAL OPERATORS"> for details.
2072 Takes a reference to a list of "unary operators"
2073 to extend the syntax understood by L<SQL::Abstract>.
2074 See section L</"UNARY OPERATORS"> for details.
2080 =head2 insert($table, \@values || \%fieldvals, \%options)
2082 This is the simplest function. You simply give it a table name
2083 and either an arrayref of values or hashref of field/value pairs.
2084 It returns an SQL INSERT statement and a list of bind values.
2085 See the sections on L</"Inserting and Updating Arrays"> and
2086 L</"Inserting and Updating SQL"> for information on how to insert
2087 with those data types.
2089 The optional C<\%options> hash reference may contain additional
2090 options to generate the insert SQL. Currently supported options
2097 Takes either a scalar of raw SQL fields, or an array reference of
2098 field names, and adds on an SQL C<RETURNING> statement at the end.
2099 This allows you to return data generated by the insert statement
2100 (such as row IDs) without performing another C<SELECT> statement.
2101 Note, however, this is not part of the SQL standard and may not
2102 be supported by all database engines.
2106 =head2 update($table, \%fieldvals, \%where, \%options)
2108 This takes a table, hashref of field/value pairs, and an optional
2109 hashref L<WHERE clause|/WHERE CLAUSES>. It returns an SQL UPDATE function and a list
2111 See the sections on L</"Inserting and Updating Arrays"> and
2112 L</"Inserting and Updating SQL"> for information on how to insert
2113 with those data types.
2115 The optional C<\%options> hash reference may contain additional
2116 options to generate the update SQL. Currently supported options
2123 See the C<returning> option to
2124 L<insert|/insert($table, \@values || \%fieldvals, \%options)>.
2128 =head2 select($source, $fields, $where, $order)
2130 This returns a SQL SELECT statement and associated list of bind values, as
2131 specified by the arguments :
2137 Specification of the 'FROM' part of the statement.
2138 The argument can be either a plain scalar (interpreted as a table
2139 name, will be quoted), or an arrayref (interpreted as a list
2140 of table names, joined by commas, quoted), or a scalarref
2141 (literal table name, not quoted), or a ref to an arrayref
2142 (list of literal table names, joined by commas, not quoted).
2146 Specification of the list of fields to retrieve from
2148 The argument can be either an arrayref (interpreted as a list
2149 of field names, will be joined by commas and quoted), or a
2150 plain scalar (literal SQL, not quoted).
2151 Please observe that this API is not as flexible as that of
2152 the first argument C<$source>, for backwards compatibility reasons.
2156 Optional argument to specify the WHERE part of the query.
2157 The argument is most often a hashref, but can also be
2158 an arrayref or plain scalar --
2159 see section L<WHERE clause|/"WHERE CLAUSES"> for details.
2163 Optional argument to specify the ORDER BY part of the query.
2164 The argument can be a scalar, a hashref or an arrayref
2165 -- see section L<ORDER BY clause|/"ORDER BY CLAUSES">
2171 =head2 delete($table, \%where, \%options)
2173 This takes a table name and optional hashref L<WHERE clause|/WHERE CLAUSES>.
2174 It returns an SQL DELETE statement and list of bind values.
2176 The optional C<\%options> hash reference may contain additional
2177 options to generate the delete SQL. Currently supported options
2184 See the C<returning> option to
2185 L<insert|/insert($table, \@values || \%fieldvals, \%options)>.
2189 =head2 where(\%where, $order)
2191 This is used to generate just the WHERE clause. For example,
2192 if you have an arbitrary data structure and know what the
2193 rest of your SQL is going to look like, but want an easy way
2194 to produce a WHERE clause, use this. It returns an SQL WHERE
2195 clause and list of bind values.
2198 =head2 values(\%data)
2200 This just returns the values from the hash C<%data>, in the same
2201 order that would be returned from any of the other above queries.
2202 Using this allows you to markedly speed up your queries if you
2203 are affecting lots of rows. See below under the L</"PERFORMANCE"> section.
2205 =head2 generate($any, 'number', $of, \@data, $struct, \%types)
2207 Warning: This is an experimental method and subject to change.
2209 This returns arbitrarily generated SQL. It's a really basic shortcut.
2210 It will return two different things, depending on return context:
2212 my($stmt, @bind) = $sql->generate('create table', \$table, \@fields);
2213 my $stmt_and_val = $sql->generate('create table', \$table, \@fields);
2215 These would return the following:
2217 # First calling form
2218 $stmt = "CREATE TABLE test (?, ?)";
2219 @bind = (field1, field2);
2221 # Second calling form
2222 $stmt_and_val = "CREATE TABLE test (field1, field2)";
2224 Depending on what you're trying to do, it's up to you to choose the correct
2225 format. In this example, the second form is what you would want.
2229 $sql->generate('alter session', { nls_date_format => 'MM/YY' });
2233 ALTER SESSION SET nls_date_format = 'MM/YY'
2235 You get the idea. Strings get their case twiddled, but everything
2236 else remains verbatim.
2238 =head1 EXPORTABLE FUNCTIONS
2240 =head2 is_plain_value
2242 Determines if the supplied argument is a plain value as understood by this
2247 =item * The value is C<undef>
2249 =item * The value is a non-reference
2251 =item * The value is an object with stringification overloading
2253 =item * The value is of the form C<< { -value => $anything } >>
2257 On failure returns C<undef>, on sucess returns a B<scalar> reference
2258 to the original supplied argument.
2264 The stringification overloading detection is rather advanced: it takes
2265 into consideration not only the presence of a C<""> overload, but if that
2266 fails also checks for enabled
2267 L<autogenerated versions of C<"">|overload/Magic Autogeneration>, based
2268 on either C<0+> or C<bool>.
2270 Unfortunately testing in the field indicates that this
2271 detection B<< may tickle a latent bug in perl versions before 5.018 >>,
2272 but only when very large numbers of stringifying objects are involved.
2273 At the time of writing ( Sep 2014 ) there is no clear explanation of
2274 the direct cause, nor is there a manageably small test case that reliably
2275 reproduces the problem.
2277 If you encounter any of the following exceptions in B<random places within
2278 your application stack> - this module may be to blame:
2280 Operation "ne": no method found,
2281 left argument in overloaded package <something>,
2282 right argument in overloaded package <something>
2286 Stub found while resolving method "???" overloading """" in package <something>
2288 If you fall victim to the above - please attempt to reduce the problem
2289 to something that could be sent to the L<SQL::Abstract developers
2290 |DBIx::Class/GETTING HELP/SUPPORT>
2291 (either publicly or privately). As a workaround in the meantime you can
2292 set C<$ENV{SQLA_ISVALUE_IGNORE_AUTOGENERATED_STRINGIFICATION}> to a true
2293 value, which will most likely eliminate your problem (at the expense of
2294 not being able to properly detect exotic forms of stringification).
2296 This notice and environment variable will be removed in a future version,
2297 as soon as the underlying problem is found and a reliable workaround is
2302 =head2 is_literal_value
2304 Determines if the supplied argument is a literal value as understood by this
2309 =item * C<\$sql_string>
2311 =item * C<\[ $sql_string, @bind_values ]>
2315 On failure returns C<undef>, on sucess returns an B<array> reference
2316 containing the unpacked version of the supplied literal SQL and bind values.
2318 =head1 WHERE CLAUSES
2322 This module uses a variation on the idea from L<DBIx::Abstract>. It
2323 is B<NOT>, repeat I<not> 100% compatible. B<The main logic of this
2324 module is that things in arrays are OR'ed, and things in hashes
2327 The easiest way to explain is to show lots of examples. After
2328 each C<%where> hash shown, it is assumed you used:
2330 my($stmt, @bind) = $sql->where(\%where);
2332 However, note that the C<%where> hash can be used directly in any
2333 of the other functions as well, as described above.
2335 =head2 Key-value pairs
2337 So, let's get started. To begin, a simple hash:
2341 status => 'completed'
2344 Is converted to SQL C<key = val> statements:
2346 $stmt = "WHERE user = ? AND status = ?";
2347 @bind = ('nwiger', 'completed');
2349 One common thing I end up doing is having a list of values that
2350 a field can be in. To do this, simply specify a list inside of
2355 status => ['assigned', 'in-progress', 'pending'];
2358 This simple code will create the following:
2360 $stmt = "WHERE user = ? AND ( status = ? OR status = ? OR status = ? )";
2361 @bind = ('nwiger', 'assigned', 'in-progress', 'pending');
2363 A field associated to an empty arrayref will be considered a
2364 logical false and will generate 0=1.
2366 =head2 Tests for NULL values
2368 If the value part is C<undef> then this is converted to SQL <IS NULL>
2377 $stmt = "WHERE user = ? AND status IS NULL";
2380 To test if a column IS NOT NULL:
2384 status => { '!=', undef },
2387 =head2 Specific comparison operators
2389 If you want to specify a different type of operator for your comparison,
2390 you can use a hashref for a given column:
2394 status => { '!=', 'completed' }
2397 Which would generate:
2399 $stmt = "WHERE user = ? AND status != ?";
2400 @bind = ('nwiger', 'completed');
2402 To test against multiple values, just enclose the values in an arrayref:
2404 status => { '=', ['assigned', 'in-progress', 'pending'] };
2406 Which would give you:
2408 "WHERE status = ? OR status = ? OR status = ?"
2411 The hashref can also contain multiple pairs, in which case it is expanded
2412 into an C<AND> of its elements:
2416 status => { '!=', 'completed', -not_like => 'pending%' }
2419 # Or more dynamically, like from a form
2420 $where{user} = 'nwiger';
2421 $where{status}{'!='} = 'completed';
2422 $where{status}{'-not_like'} = 'pending%';
2424 # Both generate this
2425 $stmt = "WHERE user = ? AND status != ? AND status NOT LIKE ?";
2426 @bind = ('nwiger', 'completed', 'pending%');
2429 To get an OR instead, you can combine it with the arrayref idea:
2433 priority => [ { '=', 2 }, { '>', 5 } ]
2436 Which would generate:
2438 $stmt = "WHERE ( priority = ? OR priority > ? ) AND user = ?";
2439 @bind = ('2', '5', 'nwiger');
2441 If you want to include literal SQL (with or without bind values), just use a
2442 scalar reference or reference to an arrayref as the value:
2445 date_entered => { '>' => \["to_date(?, 'MM/DD/YYYY')", "11/26/2008"] },
2446 date_expires => { '<' => \"now()" }
2449 Which would generate:
2451 $stmt = "WHERE date_entered > to_date(?, 'MM/DD/YYYY') AND date_expires < now()";
2452 @bind = ('11/26/2008');
2455 =head2 Logic and nesting operators
2457 In the example above,
2458 there is a subtle trap if you want to say something like
2459 this (notice the C<AND>):
2461 WHERE priority != ? AND priority != ?
2463 Because, in Perl you I<can't> do this:
2465 priority => { '!=' => 2, '!=' => 1 }
2467 As the second C<!=> key will obliterate the first. The solution
2468 is to use the special C<-modifier> form inside an arrayref:
2470 priority => [ -and => {'!=', 2},
2474 Normally, these would be joined by C<OR>, but the modifier tells it
2475 to use C<AND> instead. (Hint: You can use this in conjunction with the
2476 C<logic> option to C<new()> in order to change the way your queries
2477 work by default.) B<Important:> Note that the C<-modifier> goes
2478 B<INSIDE> the arrayref, as an extra first element. This will
2479 B<NOT> do what you think it might:
2481 priority => -and => [{'!=', 2}, {'!=', 1}] # WRONG!
2483 Here is a quick list of equivalencies, since there is some overlap:
2486 status => {'!=', 'completed', 'not like', 'pending%' }
2487 status => [ -and => {'!=', 'completed'}, {'not like', 'pending%'}]
2490 status => {'=', ['assigned', 'in-progress']}
2491 status => [ -or => {'=', 'assigned'}, {'=', 'in-progress'}]
2492 status => [ {'=', 'assigned'}, {'=', 'in-progress'} ]
2496 =head2 Special operators : IN, BETWEEN, etc.
2498 You can also use the hashref format to compare a list of fields using the
2499 C<IN> comparison operator, by specifying the list as an arrayref:
2502 status => 'completed',
2503 reportid => { -in => [567, 2335, 2] }
2506 Which would generate:
2508 $stmt = "WHERE status = ? AND reportid IN (?,?,?)";
2509 @bind = ('completed', '567', '2335', '2');
2511 The reverse operator C<-not_in> generates SQL C<NOT IN> and is used in
2514 If the argument to C<-in> is an empty array, 'sqlfalse' is generated
2515 (by default : C<1=0>). Similarly, C<< -not_in => [] >> generates
2516 'sqltrue' (by default : C<1=1>).
2518 In addition to the array you can supply a chunk of literal sql or
2519 literal sql with bind:
2522 customer => { -in => \[
2523 'SELECT cust_id FROM cust WHERE balance > ?',
2526 status => { -in => \'SELECT status_codes FROM states' },
2532 customer IN ( SELECT cust_id FROM cust WHERE balance > ? )
2533 AND status IN ( SELECT status_codes FROM states )
2537 Finally, if the argument to C<-in> is not a reference, it will be
2538 treated as a single-element array.
2540 Another pair of operators is C<-between> and C<-not_between>,
2541 used with an arrayref of two values:
2545 completion_date => {
2546 -not_between => ['2002-10-01', '2003-02-06']
2552 WHERE user = ? AND completion_date NOT BETWEEN ( ? AND ? )
2554 Just like with C<-in> all plausible combinations of literal SQL
2558 start0 => { -between => [ 1, 2 ] },
2559 start1 => { -between => \["? AND ?", 1, 2] },
2560 start2 => { -between => \"lower(x) AND upper(y)" },
2561 start3 => { -between => [
2563 \["upper(?)", 'stuff' ],
2570 ( start0 BETWEEN ? AND ? )
2571 AND ( start1 BETWEEN ? AND ? )
2572 AND ( start2 BETWEEN lower(x) AND upper(y) )
2573 AND ( start3 BETWEEN lower(x) AND upper(?) )
2575 @bind = (1, 2, 1, 2, 'stuff');
2578 These are the two builtin "special operators"; but the
2579 list can be expanded : see section L</"SPECIAL OPERATORS"> below.
2581 =head2 Unary operators: bool
2583 If you wish to test against boolean columns or functions within your
2584 database you can use the C<-bool> and C<-not_bool> operators. For
2585 example to test the column C<is_user> being true and the column
2586 C<is_enabled> being false you would use:-
2590 -not_bool => 'is_enabled',
2595 WHERE is_user AND NOT is_enabled
2597 If a more complex combination is required, testing more conditions,
2598 then you should use the and/or operators:-
2603 -not_bool => { two=> { -rlike => 'bar' } },
2604 -not_bool => { three => [ { '=', 2 }, { '>', 5 } ] },
2615 (NOT ( three = ? OR three > ? ))
2618 =head2 Nested conditions, -and/-or prefixes
2620 So far, we've seen how multiple conditions are joined with a top-level
2621 C<AND>. We can change this by putting the different conditions we want in
2622 hashes and then putting those hashes in an array. For example:
2627 status => { -like => ['pending%', 'dispatched'] },
2631 status => 'unassigned',
2635 This data structure would create the following:
2637 $stmt = "WHERE ( user = ? AND ( status LIKE ? OR status LIKE ? ) )
2638 OR ( user = ? AND status = ? ) )";
2639 @bind = ('nwiger', 'pending', 'dispatched', 'robot', 'unassigned');
2642 Clauses in hashrefs or arrayrefs can be prefixed with an C<-and> or C<-or>
2643 to change the logic inside :
2649 -and => [ workhrs => {'>', 20}, geo => 'ASIA' ],
2650 -or => { workhrs => {'<', 50}, geo => 'EURO' },
2657 $stmt = "WHERE ( user = ?
2658 AND ( ( workhrs > ? AND geo = ? )
2659 OR ( workhrs < ? OR geo = ? ) ) )";
2660 @bind = ('nwiger', '20', 'ASIA', '50', 'EURO');
2662 =head3 Algebraic inconsistency, for historical reasons
2664 C<Important note>: when connecting several conditions, the C<-and->|C<-or>
2665 operator goes C<outside> of the nested structure; whereas when connecting
2666 several constraints on one column, the C<-and> operator goes
2667 C<inside> the arrayref. Here is an example combining both features :
2670 -and => [a => 1, b => 2],
2671 -or => [c => 3, d => 4],
2672 e => [-and => {-like => 'foo%'}, {-like => '%bar'} ]
2677 WHERE ( ( ( a = ? AND b = ? )
2678 OR ( c = ? OR d = ? )
2679 OR ( e LIKE ? AND e LIKE ? ) ) )
2681 This difference in syntax is unfortunate but must be preserved for
2682 historical reasons. So be careful : the two examples below would
2683 seem algebraically equivalent, but they are not
2686 { -like => 'foo%' },
2687 { -like => '%bar' },
2689 # yields : WHERE ( ( col LIKE ? AND col LIKE ? ) )
2692 { col => { -like => 'foo%' } },
2693 { col => { -like => '%bar' } },
2695 # yields : WHERE ( ( col LIKE ? OR col LIKE ? ) )
2698 =head2 Literal SQL and value type operators
2700 The basic premise of SQL::Abstract is that in WHERE specifications the "left
2701 side" is a column name and the "right side" is a value (normally rendered as
2702 a placeholder). This holds true for both hashrefs and arrayref pairs as you
2703 see in the L</WHERE CLAUSES> examples above. Sometimes it is necessary to
2704 alter this behavior. There are several ways of doing so.
2708 This is a virtual operator that signals the string to its right side is an
2709 identifier (a column name) and not a value. For example to compare two
2710 columns you would write:
2713 priority => { '<', 2 },
2714 requestor => { -ident => 'submitter' },
2719 $stmt = "WHERE priority < ? AND requestor = submitter";
2722 If you are maintaining legacy code you may see a different construct as
2723 described in L</Deprecated usage of Literal SQL>, please use C<-ident> in new
2728 This is a virtual operator that signals that the construct to its right side
2729 is a value to be passed to DBI. This is for example necessary when you want
2730 to write a where clause against an array (for RDBMS that support such
2731 datatypes). For example:
2734 array => { -value => [1, 2, 3] }
2739 $stmt = 'WHERE array = ?';
2740 @bind = ([1, 2, 3]);
2742 Note that if you were to simply say:
2748 the result would probably not be what you wanted:
2750 $stmt = 'WHERE array = ? OR array = ? OR array = ?';
2755 Finally, sometimes only literal SQL will do. To include a random snippet
2756 of SQL verbatim, you specify it as a scalar reference. Consider this only
2757 as a last resort. Usually there is a better way. For example:
2760 priority => { '<', 2 },
2761 requestor => { -in => \'(SELECT name FROM hitmen)' },
2766 $stmt = "WHERE priority < ? AND requestor IN (SELECT name FROM hitmen)"
2769 Note that in this example, you only get one bind parameter back, since
2770 the verbatim SQL is passed as part of the statement.
2774 Never use untrusted input as a literal SQL argument - this is a massive
2775 security risk (there is no way to check literal snippets for SQL
2776 injections and other nastyness). If you need to deal with untrusted input
2777 use literal SQL with placeholders as described next.
2779 =head3 Literal SQL with placeholders and bind values (subqueries)
2781 If the literal SQL to be inserted has placeholders and bind values,
2782 use a reference to an arrayref (yes this is a double reference --
2783 not so common, but perfectly legal Perl). For example, to find a date
2784 in Postgres you can use something like this:
2787 date_column => \[ "= date '2008-09-30' - ?::integer", 10 ]
2792 $stmt = "WHERE ( date_column = date '2008-09-30' - ?::integer )"
2795 Note that you must pass the bind values in the same format as they are returned
2796 by L<where|/where(\%where, $order)>. This means that if you set L</bindtype>
2797 to C<columns>, you must provide the bind values in the
2798 C<< [ column_meta => value ] >> format, where C<column_meta> is an opaque
2799 scalar value; most commonly the column name, but you can use any scalar value
2800 (including references and blessed references), L<SQL::Abstract> will simply
2801 pass it through intact. So if C<bindtype> is set to C<columns> the above
2802 example will look like:
2805 date_column => \[ "= date '2008-09-30' - ?::integer", [ {} => 10 ] ]
2808 Literal SQL is especially useful for nesting parenthesized clauses in the
2809 main SQL query. Here is a first example :
2811 my ($sub_stmt, @sub_bind) = ("SELECT c1 FROM t1 WHERE c2 < ? AND c3 LIKE ?",
2815 bar => \["IN ($sub_stmt)" => @sub_bind],
2820 $stmt = "WHERE (foo = ? AND bar IN (SELECT c1 FROM t1
2821 WHERE c2 < ? AND c3 LIKE ?))";
2822 @bind = (1234, 100, "foo%");
2824 Other subquery operators, like for example C<"E<gt> ALL"> or C<"NOT IN">,
2825 are expressed in the same way. Of course the C<$sub_stmt> and
2826 its associated bind values can be generated through a former call
2829 my ($sub_stmt, @sub_bind)
2830 = $sql->select("t1", "c1", {c2 => {"<" => 100},
2831 c3 => {-like => "foo%"}});
2834 bar => \["> ALL ($sub_stmt)" => @sub_bind],
2837 In the examples above, the subquery was used as an operator on a column;
2838 but the same principle also applies for a clause within the main C<%where>
2839 hash, like an EXISTS subquery :
2841 my ($sub_stmt, @sub_bind)
2842 = $sql->select("t1", "*", {c1 => 1, c2 => \"> t0.c0"});
2843 my %where = ( -and => [
2845 \["EXISTS ($sub_stmt)" => @sub_bind],
2850 $stmt = "WHERE (foo = ? AND EXISTS (SELECT * FROM t1
2851 WHERE c1 = ? AND c2 > t0.c0))";
2855 Observe that the condition on C<c2> in the subquery refers to
2856 column C<t0.c0> of the main query : this is I<not> a bind
2857 value, so we have to express it through a scalar ref.
2858 Writing C<< c2 => {">" => "t0.c0"} >> would have generated
2859 C<< c2 > ? >> with bind value C<"t0.c0"> ... not exactly
2860 what we wanted here.
2862 Finally, here is an example where a subquery is used
2863 for expressing unary negation:
2865 my ($sub_stmt, @sub_bind)
2866 = $sql->where({age => [{"<" => 10}, {">" => 20}]});
2867 $sub_stmt =~ s/^ where //i; # don't want "WHERE" in the subclause
2869 lname => {like => '%son%'},
2870 \["NOT ($sub_stmt)" => @sub_bind],
2875 $stmt = "lname LIKE ? AND NOT ( age < ? OR age > ? )"
2876 @bind = ('%son%', 10, 20)
2878 =head3 Deprecated usage of Literal SQL
2880 Below are some examples of archaic use of literal SQL. It is shown only as
2881 reference for those who deal with legacy code. Each example has a much
2882 better, cleaner and safer alternative that users should opt for in new code.
2888 my %where = ( requestor => \'IS NOT NULL' )
2890 $stmt = "WHERE requestor IS NOT NULL"
2892 This used to be the way of generating NULL comparisons, before the handling
2893 of C<undef> got formalized. For new code please use the superior syntax as
2894 described in L</Tests for NULL values>.
2898 my %where = ( requestor => \'= submitter' )
2900 $stmt = "WHERE requestor = submitter"
2902 This used to be the only way to compare columns. Use the superior L</-ident>
2903 method for all new code. For example an identifier declared in such a way
2904 will be properly quoted if L</quote_char> is properly set, while the legacy
2905 form will remain as supplied.
2909 my %where = ( is_ready => \"", completed => { '>', '2012-12-21' } )
2911 $stmt = "WHERE completed > ? AND is_ready"
2912 @bind = ('2012-12-21')
2914 Using an empty string literal used to be the only way to express a boolean.
2915 For all new code please use the much more readable
2916 L<-bool|/Unary operators: bool> operator.
2922 These pages could go on for a while, since the nesting of the data
2923 structures this module can handle are pretty much unlimited (the
2924 module implements the C<WHERE> expansion as a recursive function
2925 internally). Your best bet is to "play around" with the module a
2926 little to see how the data structures behave, and choose the best
2927 format for your data based on that.
2929 And of course, all the values above will probably be replaced with
2930 variables gotten from forms or the command line. After all, if you
2931 knew everything ahead of time, you wouldn't have to worry about
2932 dynamically-generating SQL and could just hardwire it into your
2935 =head1 ORDER BY CLAUSES
2937 Some functions take an order by clause. This can either be a scalar (just a
2938 column name), a hashref of C<< { -desc => 'col' } >> or C<< { -asc => 'col' }
2939 >>, a scalarref, an arrayref-ref, or an arrayref of any of the previous
2942 Given | Will Generate
2943 ---------------------------------------------------------------
2945 'colA' | ORDER BY colA
2947 [qw/colA colB/] | ORDER BY colA, colB
2949 {-asc => 'colA'} | ORDER BY colA ASC
2951 {-desc => 'colB'} | ORDER BY colB DESC
2953 ['colA', {-asc => 'colB'}] | ORDER BY colA, colB ASC
2955 { -asc => [qw/colA colB/] } | ORDER BY colA ASC, colB ASC
2957 \'colA DESC' | ORDER BY colA DESC
2959 \[ 'FUNC(colA, ?)', $x ] | ORDER BY FUNC(colA, ?)
2960 | /* ...with $x bound to ? */
2963 { -asc => 'colA' }, | colA ASC,
2964 { -desc => [qw/colB/] }, | colB DESC,
2965 { -asc => [qw/colC colD/] },| colC ASC, colD ASC,
2966 \'colE DESC', | colE DESC,
2967 \[ 'FUNC(colF, ?)', $x ], | FUNC(colF, ?)
2968 ] | /* ...with $x bound to ? */
2969 ===============================================================
2973 =head1 SPECIAL OPERATORS
2975 my $sqlmaker = SQL::Abstract->new(special_ops => [
2979 my ($self, $field, $op, $arg) = @_;
2985 handler => 'method_name',
2989 A "special operator" is a SQL syntactic clause that can be
2990 applied to a field, instead of a usual binary operator.
2993 WHERE field IN (?, ?, ?)
2994 WHERE field BETWEEN ? AND ?
2995 WHERE MATCH(field) AGAINST (?, ?)
2997 Special operators IN and BETWEEN are fairly standard and therefore
2998 are builtin within C<SQL::Abstract> (as the overridable methods
2999 C<_where_field_IN> and C<_where_field_BETWEEN>). For other operators,
3000 like the MATCH .. AGAINST example above which is specific to MySQL,
3001 you can write your own operator handlers - supply a C<special_ops>
3002 argument to the C<new> method. That argument takes an arrayref of
3003 operator definitions; each operator definition is a hashref with two
3010 the regular expression to match the operator
3014 Either a coderef or a plain scalar method name. In both cases
3015 the expected return is C<< ($sql, @bind) >>.
3017 When supplied with a method name, it is simply called on the
3018 L<SQL::Abstract> object as:
3020 $self->$method_name($field, $op, $arg)
3024 $field is the LHS of the operator
3025 $op is the part that matched the handler regex
3028 When supplied with a coderef, it is called as:
3030 $coderef->($self, $field, $op, $arg)
3035 For example, here is an implementation
3036 of the MATCH .. AGAINST syntax for MySQL
3038 my $sqlmaker = SQL::Abstract->new(special_ops => [
3040 # special op for MySql MATCH (field) AGAINST(word1, word2, ...)
3041 {regex => qr/^match$/i,
3043 my ($self, $field, $op, $arg) = @_;
3044 $arg = [$arg] if not ref $arg;
3045 my $label = $self->_quote($field);
3046 my ($placeholder) = $self->_convert('?');
3047 my $placeholders = join ", ", (($placeholder) x @$arg);
3048 my $sql = $self->_sqlcase('match') . " ($label) "
3049 . $self->_sqlcase('against') . " ($placeholders) ";
3050 my @bind = $self->_bindtype($field, @$arg);
3051 return ($sql, @bind);
3058 =head1 UNARY OPERATORS
3060 my $sqlmaker = SQL::Abstract->new(unary_ops => [
3064 my ($self, $op, $arg) = @_;
3070 handler => 'method_name',
3074 A "unary operator" is a SQL syntactic clause that can be
3075 applied to a field - the operator goes before the field
3077 You can write your own operator handlers - supply a C<unary_ops>
3078 argument to the C<new> method. That argument takes an arrayref of
3079 operator definitions; each operator definition is a hashref with two
3086 the regular expression to match the operator
3090 Either a coderef or a plain scalar method name. In both cases
3091 the expected return is C<< $sql >>.
3093 When supplied with a method name, it is simply called on the
3094 L<SQL::Abstract> object as:
3096 $self->$method_name($op, $arg)
3100 $op is the part that matched the handler regex
3101 $arg is the RHS or argument of the operator
3103 When supplied with a coderef, it is called as:
3105 $coderef->($self, $op, $arg)
3113 Thanks to some benchmarking by Mark Stosberg, it turns out that
3114 this module is many orders of magnitude faster than using C<DBIx::Abstract>.
3115 I must admit this wasn't an intentional design issue, but it's a
3116 byproduct of the fact that you get to control your C<DBI> handles
3119 To maximize performance, use a code snippet like the following:
3121 # prepare a statement handle using the first row
3122 # and then reuse it for the rest of the rows
3124 for my $href (@array_of_hashrefs) {
3125 $stmt ||= $sql->insert('table', $href);
3126 $sth ||= $dbh->prepare($stmt);
3127 $sth->execute($sql->values($href));
3130 The reason this works is because the keys in your C<$href> are sorted
3131 internally by B<SQL::Abstract>. Thus, as long as your data retains
3132 the same structure, you only have to generate the SQL the first time
3133 around. On subsequent queries, simply use the C<values> function provided
3134 by this module to return your values in the correct order.
3136 However this depends on the values having the same type - if, for
3137 example, the values of a where clause may either have values
3138 (resulting in sql of the form C<column = ?> with a single bind
3139 value), or alternatively the values might be C<undef> (resulting in
3140 sql of the form C<column IS NULL> with no bind value) then the
3141 caching technique suggested will not work.
3145 If you use my C<CGI::FormBuilder> module at all, you'll hopefully
3146 really like this part (I do, at least). Building up a complex query
3147 can be as simple as the following:
3154 use CGI::FormBuilder;
3157 my $form = CGI::FormBuilder->new(...);
3158 my $sql = SQL::Abstract->new;
3160 if ($form->submitted) {
3161 my $field = $form->field;
3162 my $id = delete $field->{id};
3163 my($stmt, @bind) = $sql->update('table', $field, {id => $id});
3166 Of course, you would still have to connect using C<DBI> to run the
3167 query, but the point is that if you make your form look like your
3168 table, the actual query script can be extremely simplistic.
3170 If you're B<REALLY> lazy (I am), check out C<HTML::QuickTable> for
3171 a fast interface to returning and formatting data. I frequently
3172 use these three modules together to write complex database query
3173 apps in under 50 lines.
3175 =head1 HOW TO CONTRIBUTE
3177 Contributions are always welcome, in all usable forms (we especially
3178 welcome documentation improvements). The delivery methods include git-
3179 or unified-diff formatted patches, GitHub pull requests, or plain bug
3180 reports either via RT or the Mailing list. Contributors are generally
3181 granted full access to the official repository after their first several
3182 patches pass successful review.
3184 This project is maintained in a git repository. The code and related tools are
3185 accessible at the following locations:
3189 =item * Official repo: L<git://git.shadowcat.co.uk/dbsrgits/SQL-Abstract.git>
3191 =item * Official gitweb: L<http://git.shadowcat.co.uk/gitweb/gitweb.cgi?p=dbsrgits/SQL-Abstract.git>
3193 =item * GitHub mirror: L<https://github.com/dbsrgits/sql-abstract>
3195 =item * Authorized committers: L<ssh://dbsrgits@git.shadowcat.co.uk/SQL-Abstract.git>
3201 Version 1.50 was a major internal refactoring of C<SQL::Abstract>.
3202 Great care has been taken to preserve the I<published> behavior
3203 documented in previous versions in the 1.* family; however,
3204 some features that were previously undocumented, or behaved
3205 differently from the documentation, had to be changed in order
3206 to clarify the semantics. Hence, client code that was relying
3207 on some dark areas of C<SQL::Abstract> v1.*
3208 B<might behave differently> in v1.50.
3210 The main changes are :
3216 support for literal SQL through the C<< \ [ $sql, @bind ] >> syntax.
3220 support for the { operator => \"..." } construct (to embed literal SQL)
3224 support for the { operator => \["...", @bind] } construct (to embed literal SQL with bind values)
3228 optional support for L<array datatypes|/"Inserting and Updating Arrays">
3232 defensive programming : check arguments
3236 fixed bug with global logic, which was previously implemented
3237 through global variables yielding side-effects. Prior versions would
3238 interpret C<< [ {cond1, cond2}, [cond3, cond4] ] >>
3239 as C<< "(cond1 AND cond2) OR (cond3 AND cond4)" >>.
3240 Now this is interpreted
3241 as C<< "(cond1 AND cond2) OR (cond3 OR cond4)" >>.
3246 fixed semantics of _bindtype on array args
3250 dropped the C<_anoncopy> of the %where tree. No longer necessary,
3251 we just avoid shifting arrays within that tree.
3255 dropped the C<_modlogic> function
3259 =head1 ACKNOWLEDGEMENTS
3261 There are a number of individuals that have really helped out with
3262 this module. Unfortunately, most of them submitted bugs via CPAN
3263 so I have no idea who they are! But the people I do know are:
3265 Ash Berlin (order_by hash term support)
3266 Matt Trout (DBIx::Class support)
3267 Mark Stosberg (benchmarking)
3268 Chas Owens (initial "IN" operator support)
3269 Philip Collins (per-field SQL functions)
3270 Eric Kolve (hashref "AND" support)
3271 Mike Fragassi (enhancements to "BETWEEN" and "LIKE")
3272 Dan Kubb (support for "quote_char" and "name_sep")
3273 Guillermo Roditi (patch to cleanup "IN" and "BETWEEN", fix and tests for _order_by)
3274 Laurent Dami (internal refactoring, extensible list of special operators, literal SQL)
3275 Norbert Buchmuller (support for literal SQL in hashpair, misc. fixes & tests)
3276 Peter Rabbitson (rewrite of SQLA::Test, misc. fixes & tests)
3277 Oliver Charles (support for "RETURNING" after "INSERT")
3283 L<DBIx::Class>, L<DBIx::Abstract>, L<CGI::FormBuilder>, L<HTML::QuickTable>.
3287 Copyright (c) 2001-2007 Nathan Wiger <nwiger@cpan.org>. All Rights Reserved.
3289 This module is actively maintained by Matt Trout <mst@shadowcatsystems.co.uk>
3291 For support, your best bet is to try the C<DBIx::Class> users mailing list.
3292 While not an official support venue, C<DBIx::Class> makes heavy use of
3293 C<SQL::Abstract>, and as such list members there are very familiar with
3294 how to create queries.
3298 This module is free software; you may copy this under the same
3299 terms as perl itself (either the GNU General Public License or
3300 the Artistic License)