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 = '1.87';
32 # This would confuse some packagers
33 $VERSION = eval $VERSION if $VERSION =~ /_/; # numify for warning-free dev releases
37 # special operators (-in, -between). May be extended/overridden by user.
38 # See section WHERE: BUILTIN SPECIAL OPERATORS below for implementation
39 my @BUILTIN_SPECIAL_OPS = (
40 {regex => qr/^ (?: not \s )? between $/ix, handler => '_where_field_BETWEEN'},
41 {regex => qr/^ (?: not \s )? in $/ix, handler => '_where_field_IN'},
42 {regex => qr/^ ident $/ix, handler => '_where_op_IDENT'},
43 {regex => qr/^ value $/ix, handler => '_where_op_VALUE'},
44 {regex => qr/^ is (?: \s+ not )? $/ix, handler => '_where_field_IS'},
47 # unaryish operators - key maps to handler
48 my @BUILTIN_UNARY_OPS = (
49 # the digits are backcompat stuff
50 { regex => qr/^ and (?: [_\s]? \d+ )? $/xi, handler => '_where_op_ANDOR' },
51 { regex => qr/^ or (?: [_\s]? \d+ )? $/xi, handler => '_where_op_ANDOR' },
52 { regex => qr/^ nest (?: [_\s]? \d+ )? $/xi, handler => '_where_op_NEST' },
53 { regex => qr/^ (?: not \s )? bool $/xi, handler => '_where_op_BOOL' },
54 { regex => qr/^ ident $/xi, handler => '_where_op_IDENT' },
55 { regex => qr/^ value $/xi, handler => '_where_op_VALUE' },
58 #======================================================================
59 # DEBUGGING AND ERROR REPORTING
60 #======================================================================
63 return unless $_[0]->{debug}; shift; # a little faster
64 my $func = (caller(1))[3];
65 warn "[$func] ", @_, "\n";
69 my($func) = (caller(1))[3];
70 Carp::carp "[$func] Warning: ", @_;
74 my($func) = (caller(1))[3];
75 Carp::croak "[$func] Fatal: ", @_;
78 sub is_literal_value ($) {
79 ref $_[0] eq 'SCALAR' ? [ ${$_[0]} ]
80 : ( ref $_[0] eq 'REF' and ref ${$_[0]} eq 'ARRAY' ) ? [ @${ $_[0] } ]
84 # FIXME XSify - this can be done so much more efficiently
85 sub is_plain_value ($) {
87 ! length ref $_[0] ? \($_[0])
89 ref $_[0] eq 'HASH' and keys %{$_[0]} == 1
91 exists $_[0]->{-value}
92 ) ? \($_[0]->{-value})
94 # reuse @_ for even moar speedz
95 defined ( $_[1] = Scalar::Util::blessed $_[0] )
97 # deliberately not using Devel::OverloadInfo - the checks we are
98 # intersted in are much more limited than the fullblown thing, and
99 # this is a very hot piece of code
101 # simply using ->can('(""') can leave behind stub methods that
102 # break actually using the overload later (see L<perldiag/Stub
103 # found while resolving method "%s" overloading "%s" in package
104 # "%s"> and the source of overload::mycan())
106 # either has stringification which DBI SHOULD prefer out of the box
107 grep { *{ (qq[${_}::(""]) }{CODE} } @{ $_[2] = mro::get_linear_isa( $_[1] ) }
109 # has nummification or boolification, AND fallback is *not* disabled
111 SQL::Abstract::_ENV_::DETECT_AUTOGENERATED_STRINGIFICATION
114 grep { *{"${_}::(0+"}{CODE} } @{$_[2]}
116 grep { *{"${_}::(bool"}{CODE} } @{$_[2]}
120 # no fallback specified at all
121 ! ( ($_[3]) = grep { *{"${_}::()"}{CODE} } @{$_[2]} )
123 # fallback explicitly undef
124 ! defined ${"$_[3]::()"}
137 #======================================================================
139 #======================================================================
143 my $class = ref($self) || $self;
144 my %opt = (ref $_[0] eq 'HASH') ? %{$_[0]} : @_;
146 # choose our case by keeping an option around
147 delete $opt{case} if $opt{case} && $opt{case} ne 'lower';
149 # default logic for interpreting arrayrefs
150 $opt{logic} = $opt{logic} ? uc $opt{logic} : 'OR';
152 # how to return bind vars
153 $opt{bindtype} ||= 'normal';
155 # default comparison is "=", but can be overridden
158 # try to recognize which are the 'equality' and 'inequality' ops
159 # (temporary quickfix (in 2007), should go through a more seasoned API)
160 $opt{equality_op} = qr/^( \Q$opt{cmp}\E | \= )$/ix;
161 $opt{inequality_op} = qr/^( != | <> )$/ix;
163 $opt{like_op} = qr/^ (is\s+)? r?like $/xi;
164 $opt{not_like_op} = qr/^ (is\s+)? not \s+ r?like $/xi;
167 $opt{sqltrue} ||= '1=1';
168 $opt{sqlfalse} ||= '0=1';
171 $opt{special_ops} ||= [];
172 # regexes are applied in order, thus push after user-defines
173 push @{$opt{special_ops}}, @BUILTIN_SPECIAL_OPS;
176 $opt{unary_ops} ||= [];
177 push @{$opt{unary_ops}}, @BUILTIN_UNARY_OPS;
179 # rudimentary sanity-check for user supplied bits treated as functions/operators
180 # If a purported function matches this regular expression, an exception is thrown.
181 # Literal SQL is *NOT* subject to this check, only functions (and column names
182 # when quoting is not in effect)
185 # need to guard against ()'s in column names too, but this will break tons of
186 # hacks... ideas anyone?
187 $opt{injection_guard} ||= qr/
193 return bless \%opt, $class;
197 sub _assert_pass_injection_guard {
198 if ($_[1] =~ $_[0]->{injection_guard}) {
199 my $class = ref $_[0];
200 puke "Possible SQL injection attempt '$_[1]'. If this is indeed a part of the "
201 . "desired SQL use literal SQL ( \'...' or \[ '...' ] ) or supply your own "
202 . "{injection_guard} attribute to ${class}->new()"
207 #======================================================================
209 #======================================================================
213 my $table = $self->_table(shift);
214 my $data = shift || return;
217 my $method = $self->_METHOD_FOR_refkind("_insert", $data);
218 my ($sql, @bind) = $self->$method($data);
219 $sql = join " ", $self->_sqlcase('insert into'), $table, $sql;
221 if ($options->{returning}) {
222 my ($s, @b) = $self->_insert_returning($options);
227 return wantarray ? ($sql, @bind) : $sql;
230 # So that subclasses can override INSERT ... RETURNING separately from
231 # UPDATE and DELETE (e.g. DBIx::Class::SQLMaker::Oracle does this)
232 sub _insert_returning { shift->_returning(@_) }
235 my ($self, $options) = @_;
237 my $f = $options->{returning};
239 my $fieldlist = $self->_SWITCH_refkind($f, {
240 ARRAYREF => sub {join ', ', map { $self->_quote($_) } @$f;},
241 SCALAR => sub {$self->_quote($f)},
242 SCALARREF => sub {$$f},
244 return $self->_sqlcase(' returning ') . $fieldlist;
247 sub _insert_HASHREF { # explicit list of fields and then values
248 my ($self, $data) = @_;
250 my @fields = sort keys %$data;
252 my ($sql, @bind) = $self->_insert_values($data);
255 $_ = $self->_quote($_) foreach @fields;
256 $sql = "( ".join(", ", @fields).") ".$sql;
258 return ($sql, @bind);
261 sub _insert_ARRAYREF { # just generate values(?,?) part (no list of fields)
262 my ($self, $data) = @_;
264 # no names (arrayref) so can't generate bindtype
265 $self->{bindtype} ne 'columns'
266 or belch "can't do 'columns' bindtype when called with arrayref";
268 my (@values, @all_bind);
269 foreach my $value (@$data) {
270 my ($values, @bind) = $self->_insert_value(undef, $value);
271 push @values, $values;
272 push @all_bind, @bind;
274 my $sql = $self->_sqlcase('values')." ( ".join(", ", @values)." )";
275 return ($sql, @all_bind);
278 sub _insert_ARRAYREFREF { # literal SQL with bind
279 my ($self, $data) = @_;
281 my ($sql, @bind) = @${$data};
282 $self->_assert_bindval_matches_bindtype(@bind);
284 return ($sql, @bind);
288 sub _insert_SCALARREF { # literal SQL without bind
289 my ($self, $data) = @_;
295 my ($self, $data) = @_;
297 my (@values, @all_bind);
298 foreach my $column (sort keys %$data) {
299 my ($values, @bind) = $self->_insert_value($column, $data->{$column});
300 push @values, $values;
301 push @all_bind, @bind;
303 my $sql = $self->_sqlcase('values')." ( ".join(", ", @values)." )";
304 return ($sql, @all_bind);
308 my ($self, $column, $v) = @_;
310 my (@values, @all_bind);
311 $self->_SWITCH_refkind($v, {
314 if ($self->{array_datatypes}) { # if array datatype are activated
316 push @all_bind, $self->_bindtype($column, $v);
318 else { # else literal SQL with bind
319 my ($sql, @bind) = @$v;
320 $self->_assert_bindval_matches_bindtype(@bind);
322 push @all_bind, @bind;
326 ARRAYREFREF => sub { # literal SQL with bind
327 my ($sql, @bind) = @${$v};
328 $self->_assert_bindval_matches_bindtype(@bind);
330 push @all_bind, @bind;
333 # THINK: anything useful to do with a HASHREF ?
334 HASHREF => sub { # (nothing, but old SQLA passed it through)
335 #TODO in SQLA >= 2.0 it will die instead
336 belch "HASH ref as bind value in insert is not supported";
338 push @all_bind, $self->_bindtype($column, $v);
341 SCALARREF => sub { # literal SQL without bind
345 SCALAR_or_UNDEF => sub {
347 push @all_bind, $self->_bindtype($column, $v);
352 my $sql = join(", ", @values);
353 return ($sql, @all_bind);
358 #======================================================================
360 #======================================================================
365 my $table = $self->_table(shift);
366 my $data = shift || return;
370 # first build the 'SET' part of the sql statement
371 puke "Unsupported data type specified to \$sql->update"
372 unless ref $data eq 'HASH';
374 my ($sql, @all_bind) = $self->_update_set_values($data);
375 $sql = $self->_sqlcase('update ') . $table . $self->_sqlcase(' set ')
379 my($where_sql, @where_bind) = $self->where($where);
381 push @all_bind, @where_bind;
384 if ($options->{returning}) {
385 my ($returning_sql, @returning_bind) = $self->_update_returning($options);
386 $sql .= $returning_sql;
387 push @all_bind, @returning_bind;
390 return wantarray ? ($sql, @all_bind) : $sql;
393 sub _update_set_values {
394 my ($self, $data) = @_;
396 my (@set, @all_bind);
397 for my $k (sort keys %$data) {
400 my $label = $self->_quote($k);
402 $self->_SWITCH_refkind($v, {
404 if ($self->{array_datatypes}) { # array datatype
405 push @set, "$label = ?";
406 push @all_bind, $self->_bindtype($k, $v);
408 else { # literal SQL with bind
409 my ($sql, @bind) = @$v;
410 $self->_assert_bindval_matches_bindtype(@bind);
411 push @set, "$label = $sql";
412 push @all_bind, @bind;
415 ARRAYREFREF => sub { # literal SQL with bind
416 my ($sql, @bind) = @${$v};
417 $self->_assert_bindval_matches_bindtype(@bind);
418 push @set, "$label = $sql";
419 push @all_bind, @bind;
421 SCALARREF => sub { # literal SQL without bind
422 push @set, "$label = $$v";
425 my ($op, $arg, @rest) = %$v;
427 puke 'Operator calls in update must be in the form { -op => $arg }'
428 if (@rest or not $op =~ /^\-(.+)/);
430 local $self->{_nested_func_lhs} = $k;
431 my ($sql, @bind) = $self->_where_unary_op($1, $arg);
433 push @set, "$label = $sql";
434 push @all_bind, @bind;
436 SCALAR_or_UNDEF => sub {
437 push @set, "$label = ?";
438 push @all_bind, $self->_bindtype($k, $v);
444 my $sql = join ', ', @set;
446 return ($sql, @all_bind);
449 # So that subclasses can override UPDATE ... RETURNING separately from
451 sub _update_returning { shift->_returning(@_) }
455 #======================================================================
457 #======================================================================
462 my $table = $self->_table(shift);
463 my $fields = shift || '*';
467 my ($fields_sql, @bind) = $self->_select_fields($fields);
469 my ($where_sql, @where_bind) = $self->where($where, $order);
470 push @bind, @where_bind;
472 my $sql = join(' ', $self->_sqlcase('select'), $fields_sql,
473 $self->_sqlcase('from'), $table)
476 return wantarray ? ($sql, @bind) : $sql;
480 my ($self, $fields) = @_;
481 return ref $fields eq 'ARRAY' ? join ', ', map { $self->_quote($_) } @$fields
485 #======================================================================
487 #======================================================================
492 my $table = $self->_table(shift);
496 my($where_sql, @bind) = $self->where($where);
497 my $sql = $self->_sqlcase('delete from ') . $table . $where_sql;
499 if ($options->{returning}) {
500 my ($returning_sql, @returning_bind) = $self->_delete_returning($options);
501 $sql .= $returning_sql;
502 push @bind, @returning_bind;
505 return wantarray ? ($sql, @bind) : $sql;
508 # So that subclasses can override DELETE ... RETURNING separately from
510 sub _delete_returning { shift->_returning(@_) }
514 #======================================================================
516 #======================================================================
520 # Finally, a separate routine just to handle WHERE clauses
522 my ($self, $where, $order) = @_;
525 my ($sql, @bind) = $self->_recurse_where($where);
526 $sql = (defined $sql and length $sql) ? $self->_sqlcase(' where ') . "( $sql )" : '';
530 my ($order_sql, @order_bind) = $self->_order_by($order);
532 push @bind, @order_bind;
535 return wantarray ? ($sql, @bind) : $sql;
539 my ($self, $expr, $logic) = @_;
540 if (ref($expr) eq 'HASH') {
541 if (keys %$expr > 1) {
543 return +{ "-${logic}" => [
544 map $self->_expand_expr_hashpair($_ => $expr->{$_}, $logic),
548 return $self->_expand_expr_hashpair(%$expr, $logic);
553 sub _expand_expr_hashpair {
554 my ($self, $k, $v, $logic) = @_;
557 return $self->_expand_expr($v);
561 return +{ $k => { $self->{cmp} => $v } };
563 if (ref($v) eq 'ARRAY') {
564 return $self->{sqlfalse} unless @$v;
565 $self->_debug("ARRAY($k) means distribute over elements");
567 $v->[0] =~ /^-((?:and|or))$/i
568 ? ($v = [ @{$v}[1..$#$v] ], $1)
569 : ($self->{logic} || 'or')
571 return +{ "-${this_logic}" => [ map $self->_expand_expr({ $k => $_ }, $this_logic), @$v ] };
573 if (my $literal = is_literal_value($v)) {
575 belch 'Hash-pairs consisting of an empty string with a literal are deprecated, and will be removed in 2.0: use -and => [ $literal ] instead';
578 my ($sql, @bind) = @$literal;
579 return \[ $self->_quote($k).' '.$sql, @bind ];
586 my ($self, $where, $logic) = @_;
588 my $where_exp = $self->_expand_expr($where, $logic);
590 # dispatch on appropriate method according to refkind of $where
591 my $method = $self->_METHOD_FOR_refkind("_where", $where_exp);
593 my ($sql, @bind) = $self->$method($where_exp, $logic);
595 # DBIx::Class used to call _recurse_where in scalar context
596 # something else might too...
598 return ($sql, @bind);
601 belch "Calling _recurse_where in scalar context is deprecated and will go away before 2.0";
608 #======================================================================
609 # WHERE: top-level ARRAYREF
610 #======================================================================
613 sub _where_ARRAYREF {
614 my ($self, $where, $logic) = @_;
616 $logic = uc($logic || $self->{logic});
617 $logic eq 'AND' or $logic eq 'OR' or puke "unknown logic: $logic";
619 my @clauses = @$where;
621 my (@sql_clauses, @all_bind);
622 # need to use while() so can shift() for pairs
624 my $el = shift @clauses;
626 $el = undef if (defined $el and ! length $el);
628 # switch according to kind of $el and get corresponding ($sql, @bind)
629 my ($sql, @bind) = $self->_SWITCH_refkind($el, {
631 # skip empty elements, otherwise get invalid trailing AND stuff
632 ARRAYREF => sub {$self->_recurse_where($el) if @$el},
636 $self->_assert_bindval_matches_bindtype(@b);
640 HASHREF => sub {$self->_recurse_where($el, 'and') if %$el},
642 SCALARREF => sub { ($$el); },
645 # top-level arrayref with scalars, recurse in pairs
646 $self->_recurse_where({$el => shift(@clauses)})
649 UNDEF => sub {puke "Supplying an empty left hand side argument is not supported in array-pairs" },
653 push @sql_clauses, $sql;
654 push @all_bind, @bind;
658 return $self->_join_sql_clauses($logic, \@sql_clauses, \@all_bind);
661 #======================================================================
662 # WHERE: top-level ARRAYREFREF
663 #======================================================================
665 sub _where_ARRAYREFREF {
666 my ($self, $where) = @_;
667 my ($sql, @bind) = @$$where;
668 $self->_assert_bindval_matches_bindtype(@bind);
669 return ($sql, @bind);
672 #======================================================================
673 # WHERE: top-level HASHREF
674 #======================================================================
677 my ($self, $where) = @_;
678 my (@sql_clauses, @all_bind);
680 for my $k (sort keys %$where) {
681 my $v = $where->{$k};
683 # ($k => $v) is either a special unary op or a regular hashpair
684 my ($sql, @bind) = do {
686 # put the operator in canonical form
688 $op = substr $op, 1; # remove initial dash
689 $op =~ s/^\s+|\s+$//g;# remove leading/trailing space
690 $op =~ s/\s+/ /g; # compress whitespace
692 # so that -not_foo works correctly
693 $op =~ s/^not_/NOT /i;
695 $self->_debug("Unary OP(-$op) within hashref, recursing...");
696 my ($s, @b) = $self->_where_unary_op($op, $v);
698 # top level vs nested
699 # we assume that handled unary ops will take care of their ()s
701 List::Util::first {$op =~ $_->{regex}} @{$self->{unary_ops}}
703 ( defined $self->{_nested_func_lhs} and $self->{_nested_func_lhs} eq $k )
709 if (is_literal_value ($v) ) {
710 belch 'Hash-pairs consisting of an empty string with a literal are deprecated, and will be removed in 2.0: use -and => [ $literal ] instead';
713 puke "Supplying an empty left hand side argument is not supported in hash-pairs";
717 my $method = $self->_METHOD_FOR_refkind("_where_hashpair", $v);
718 $self->$method($k, $v);
722 push @sql_clauses, $sql;
723 push @all_bind, @bind;
726 return $self->_join_sql_clauses('and', \@sql_clauses, \@all_bind);
729 sub _where_unary_op {
730 my ($self, $op, $rhs) = @_;
732 # top level special ops are illegal in general
733 # this includes the -ident/-value ops (dual purpose unary and special)
734 puke "Illegal use of top-level '-$op'"
735 if ! defined $self->{_nested_func_lhs} and List::Util::first { $op =~ $_->{regex} } @{$self->{special_ops}};
737 if (my $op_entry = List::Util::first { $op =~ $_->{regex} } @{$self->{unary_ops}}) {
738 my $handler = $op_entry->{handler};
740 if (not ref $handler) {
741 if ($op =~ s/ [_\s]? \d+ $//x ) {
742 belch 'Use of [and|or|nest]_N modifiers is deprecated and will be removed in SQLA v2.0. '
743 . "You probably wanted ...-and => [ -$op => COND1, -$op => COND2 ... ]";
745 return $self->$handler($op, $rhs);
747 elsif (ref $handler eq 'CODE') {
748 return $handler->($self, $op, $rhs);
751 puke "Illegal handler for operator $op - expecting a method name or a coderef";
755 $self->_debug("Generic unary OP: $op - recursing as function");
757 $self->_assert_pass_injection_guard($op);
759 my ($sql, @bind) = $self->_SWITCH_refkind($rhs, {
761 puke "Illegal use of top-level '-$op'"
762 unless defined $self->{_nested_func_lhs};
765 $self->_convert('?'),
766 $self->_bindtype($self->{_nested_func_lhs}, $rhs)
770 $self->_recurse_where($rhs)
774 $sql = sprintf('%s %s',
775 $self->_sqlcase($op),
779 return ($sql, @bind);
782 sub _where_op_ANDOR {
783 my ($self, $op, $v) = @_;
785 $self->_SWITCH_refkind($v, {
787 return $self->_where_ARRAYREF($v, $op);
791 return ($op =~ /^or/i)
792 ? $self->_where_ARRAYREF([ map { $_ => $v->{$_} } (sort keys %$v) ], $op)
793 : $self->_where_HASHREF($v);
797 puke "-$op => \\\$scalar makes little sense, use " .
799 ? '[ \$scalar, \%rest_of_conditions ] instead'
800 : '-and => [ \$scalar, \%rest_of_conditions ] instead'
805 puke "-$op => \\[...] makes little sense, use " .
807 ? '[ \[...], \%rest_of_conditions ] instead'
808 : '-and => [ \[...], \%rest_of_conditions ] instead'
812 SCALAR => sub { # permissively interpreted as SQL
813 puke "-$op => \$value makes little sense, use -bool => \$value instead";
817 puke "-$op => undef not supported";
823 my ($self, $op, $v) = @_;
825 $self->_SWITCH_refkind($v, {
827 SCALAR => sub { # permissively interpreted as SQL
828 belch "literal SQL should be -nest => \\'scalar' "
829 . "instead of -nest => 'scalar' ";
834 puke "-$op => undef not supported";
838 $self->_recurse_where($v);
846 my ($self, $op, $v) = @_;
848 my ($s, @b) = $self->_SWITCH_refkind($v, {
849 SCALAR => sub { # interpreted as SQL column
850 $self->_convert($self->_quote($v));
854 puke "-$op => undef not supported";
858 $self->_recurse_where($v);
862 $s = "(NOT $s)" if $op =~ /^not/i;
867 sub _where_op_IDENT {
869 my ($op, $rhs) = splice @_, -2;
870 if (! defined $rhs or length ref $rhs) {
871 puke "-$op requires a single plain scalar argument (a quotable identifier)";
874 # in case we are called as a top level special op (no '=')
877 $_ = $self->_convert($self->_quote($_)) for ($lhs, $rhs);
885 sub _where_op_VALUE {
887 my ($op, $rhs) = splice @_, -2;
889 # in case we are called as a top level special op (no '=')
893 if (! defined $rhs) {
895 ? $self->_where_hashpair_HASHREF($lhs, { -is => undef })
902 (defined $lhs ? $lhs : $self->{_nested_func_lhs}),
909 $self->_convert($self->_quote($lhs)) . ' = ' . $self->_convert('?'),
913 $self->_convert('?'),
919 sub _where_hashpair_ARRAYREF {
920 my ($self, $k, $v) = @_;
923 my @v = @$v; # need copy because of shift below
924 $self->_debug("ARRAY($k) means distribute over elements");
926 # put apart first element if it is an operator (-and, -or)
928 (defined $v[0] && $v[0] =~ /^ - (?: AND|OR ) $/ix)
932 my @distributed = map { {$k => $_} } @v;
935 $self->_debug("OP($op) reinjected into the distributed array");
936 unshift @distributed, $op;
939 my $logic = $op ? substr($op, 1) : '';
941 return $self->_recurse_where(\@distributed, $logic);
944 $self->_debug("empty ARRAY($k) means 0=1");
945 return ($self->{sqlfalse});
949 sub _where_hashpair_HASHREF {
950 my ($self, $k, $v, $logic) = @_;
953 local $self->{_nested_func_lhs} = defined $self->{_nested_func_lhs}
954 ? $self->{_nested_func_lhs}
958 my ($all_sql, @all_bind);
960 for my $orig_op (sort keys %$v) {
961 my $val = $v->{$orig_op};
963 # put the operator in canonical form
966 # FIXME - we need to phase out dash-less ops
967 $op =~ s/^-//; # remove possible initial dash
968 $op =~ s/^\s+|\s+$//g;# remove leading/trailing space
969 $op =~ s/\s+/ /g; # compress whitespace
971 $self->_assert_pass_injection_guard($op);
974 $op =~ s/^is_not/IS NOT/i;
976 # so that -not_foo works correctly
977 $op =~ s/^not_/NOT /i;
979 # another retarded special case: foo => { $op => { -value => undef } }
980 if (ref $val eq 'HASH' and keys %$val == 1 and exists $val->{-value} and ! defined $val->{-value} ) {
986 # CASE: col-value logic modifiers
987 if ($orig_op =~ /^ \- (and|or) $/xi) {
988 ($sql, @bind) = $self->_where_hashpair_HASHREF($k, $val, $1);
990 # CASE: special operators like -in or -between
991 elsif (my $special_op = List::Util::first { $op =~ $_->{regex} } @{$self->{special_ops}}) {
992 my $handler = $special_op->{handler};
994 puke "No handler supplied for special operator $orig_op";
996 elsif (not ref $handler) {
997 ($sql, @bind) = $self->$handler($k, $op, $val);
999 elsif (ref $handler eq 'CODE') {
1000 ($sql, @bind) = $handler->($self, $k, $op, $val);
1003 puke "Illegal handler for special operator $orig_op - expecting a method name or a coderef";
1007 $self->_SWITCH_refkind($val, {
1009 ARRAYREF => sub { # CASE: col => {op => \@vals}
1010 ($sql, @bind) = $self->_where_field_op_ARRAYREF($k, $op, $val);
1013 ARRAYREFREF => sub { # CASE: col => {op => \[$sql, @bind]} (literal SQL with bind)
1014 my ($sub_sql, @sub_bind) = @$$val;
1015 $self->_assert_bindval_matches_bindtype(@sub_bind);
1016 $sql = join ' ', $self->_convert($self->_quote($k)),
1017 $self->_sqlcase($op),
1022 UNDEF => sub { # CASE: col => {op => undef} : sql "IS (NOT)? NULL"
1024 $op =~ /^not$/i ? 'is not' # legacy
1025 : $op =~ $self->{equality_op} ? 'is'
1026 : $op =~ $self->{like_op} ? belch("Supplying an undefined argument to '@{[ uc $op]}' is deprecated") && 'is'
1027 : $op =~ $self->{inequality_op} ? 'is not'
1028 : $op =~ $self->{not_like_op} ? belch("Supplying an undefined argument to '@{[ uc $op]}' is deprecated") && 'is not'
1029 : puke "unexpected operator '$orig_op' with undef operand";
1031 $sql = $self->_quote($k) . $self->_sqlcase(" $is null");
1034 FALLBACK => sub { # CASE: col => {op/func => $stuff}
1035 ($sql, @bind) = $self->_where_unary_op($op, $val);
1038 $self->_convert($self->_quote($k)),
1039 $self->{_nested_func_lhs} eq $k ? $sql : "($sql)", # top level vs nested
1045 ($all_sql) = (defined $all_sql and $all_sql) ? $self->_join_sql_clauses($logic, [$all_sql, $sql], []) : $sql;
1046 push @all_bind, @bind;
1048 return ($all_sql, @all_bind);
1051 sub _where_field_IS {
1052 my ($self, $k, $op, $v) = @_;
1054 my ($s) = $self->_SWITCH_refkind($v, {
1057 $self->_convert($self->_quote($k)),
1058 map { $self->_sqlcase($_)} ($op, 'null')
1061 puke "$op can only take undef as argument";
1068 sub _where_field_op_ARRAYREF {
1069 my ($self, $k, $op, $vals) = @_;
1071 my @vals = @$vals; #always work on a copy
1074 $self->_debug(sprintf '%s means multiple elements: [ %s ]',
1076 join(', ', map { defined $_ ? "'$_'" : 'NULL' } @vals ),
1079 # see if the first element is an -and/-or op
1081 if (defined $vals[0] && $vals[0] =~ /^ - (AND|OR) $/ix) {
1086 # a long standing API wart - an attempt to change this behavior during
1087 # the 1.50 series failed *spectacularly*. Warn instead and leave the
1092 (!$logic or $logic eq 'OR')
1094 ($op =~ $self->{inequality_op} or $op =~ $self->{not_like_op})
1097 belch "A multi-element arrayref as an argument to the inequality op '$o' "
1098 . 'is technically equivalent to an always-true 1=1 (you probably wanted '
1099 . "to say ...{ \$inequality_op => [ -and => \@values ] }... instead)"
1103 # distribute $op over each remaining member of @vals, append logic if exists
1104 return $self->_recurse_where([map { {$k => {$op, $_}} } @vals], $logic);
1108 # try to DWIM on equality operators
1110 $op =~ $self->{equality_op} ? $self->{sqlfalse}
1111 : $op =~ $self->{like_op} ? belch("Supplying an empty arrayref to '@{[ uc $op]}' is deprecated") && $self->{sqlfalse}
1112 : $op =~ $self->{inequality_op} ? $self->{sqltrue}
1113 : $op =~ $self->{not_like_op} ? belch("Supplying an empty arrayref to '@{[ uc $op]}' is deprecated") && $self->{sqltrue}
1114 : puke "operator '$op' applied on an empty array (field '$k')";
1119 sub _where_hashpair_SCALARREF {
1120 my ($self, $k, $v) = @_;
1121 $self->_debug("SCALAR($k) means literal SQL: $$v");
1122 my $sql = $self->_quote($k) . " " . $$v;
1126 # literal SQL with bind
1127 sub _where_hashpair_ARRAYREFREF {
1128 my ($self, $k, $v) = @_;
1129 $self->_debug("REF($k) means literal SQL: @${$v}");
1130 my ($sql, @bind) = @$$v;
1131 $self->_assert_bindval_matches_bindtype(@bind);
1132 $sql = $self->_quote($k) . " " . $sql;
1133 return ($sql, @bind );
1136 # literal SQL without bind
1137 sub _where_hashpair_SCALAR {
1138 my ($self, $k, $v) = @_;
1139 $self->_debug("NOREF($k) means simple key=val: $k $self->{cmp} $v");
1140 return ($self->_where_hashpair_HASHREF($k, { $self->{cmp} => $v }));
1144 sub _where_hashpair_UNDEF {
1145 my ($self, $k, $v) = @_;
1146 $self->_debug("UNDEF($k) means IS NULL");
1147 return $self->_where_hashpair_HASHREF($k, { -is => undef });
1150 #======================================================================
1151 # WHERE: TOP-LEVEL OTHERS (SCALARREF, SCALAR, UNDEF)
1152 #======================================================================
1155 sub _where_SCALARREF {
1156 my ($self, $where) = @_;
1159 $self->_debug("SCALAR(*top) means literal SQL: $$where");
1165 my ($self, $where) = @_;
1168 $self->_debug("NOREF(*top) means literal SQL: $where");
1179 #======================================================================
1180 # WHERE: BUILTIN SPECIAL OPERATORS (-in, -between)
1181 #======================================================================
1184 sub _where_field_BETWEEN {
1185 my ($self, $k, $op, $vals) = @_;
1187 my ($label, $and, $placeholder);
1188 $label = $self->_convert($self->_quote($k));
1189 $and = ' ' . $self->_sqlcase('and') . ' ';
1190 $placeholder = $self->_convert('?');
1191 $op = $self->_sqlcase($op);
1193 my $invalid_args = "Operator '$op' requires either an arrayref with two defined values or expressions, or a single literal scalarref/arrayref-ref";
1195 my ($clause, @bind) = $self->_SWITCH_refkind($vals, {
1196 ARRAYREFREF => sub {
1197 my ($s, @b) = @$$vals;
1198 $self->_assert_bindval_matches_bindtype(@b);
1205 puke $invalid_args if @$vals != 2;
1207 my (@all_sql, @all_bind);
1208 foreach my $val (@$vals) {
1209 my ($sql, @bind) = $self->_SWITCH_refkind($val, {
1211 return ($placeholder, $self->_bindtype($k, $val) );
1216 ARRAYREFREF => sub {
1217 my ($sql, @bind) = @$$val;
1218 $self->_assert_bindval_matches_bindtype(@bind);
1219 return ($sql, @bind);
1222 my ($func, $arg, @rest) = %$val;
1223 puke "Only simple { -func => arg } functions accepted as sub-arguments to BETWEEN"
1224 if (@rest or $func !~ /^ \- (.+)/x);
1225 $self->_where_unary_op($1 => $arg);
1231 push @all_sql, $sql;
1232 push @all_bind, @bind;
1236 (join $and, @all_sql),
1245 my $sql = "( $label $op $clause )";
1246 return ($sql, @bind)
1250 sub _where_field_IN {
1251 my ($self, $k, $op, $vals) = @_;
1253 # backwards compatibility: if scalar, force into an arrayref
1254 $vals = [$vals] if defined $vals && ! ref $vals;
1256 my ($label) = $self->_convert($self->_quote($k));
1257 my ($placeholder) = $self->_convert('?');
1258 $op = $self->_sqlcase($op);
1260 my ($sql, @bind) = $self->_SWITCH_refkind($vals, {
1261 ARRAYREF => sub { # list of choices
1262 if (@$vals) { # nonempty list
1263 my (@all_sql, @all_bind);
1265 for my $val (@$vals) {
1266 my ($sql, @bind) = $self->_SWITCH_refkind($val, {
1268 return ($placeholder, $val);
1273 ARRAYREFREF => sub {
1274 my ($sql, @bind) = @$$val;
1275 $self->_assert_bindval_matches_bindtype(@bind);
1276 return ($sql, @bind);
1279 my ($func, $arg, @rest) = %$val;
1280 puke "Only simple { -func => arg } functions accepted as sub-arguments to IN"
1281 if (@rest or $func !~ /^ \- (.+)/x);
1282 $self->_where_unary_op($1 => $arg);
1286 'SQL::Abstract before v1.75 used to generate incorrect SQL when the '
1287 . "-$op operator was given an undef-containing list: !!!AUDIT YOUR CODE "
1288 . 'AND DATA!!! (the upcoming Data::Query-based version of SQL::Abstract '
1289 . 'will emit the logically correct SQL instead of raising this exception)'
1293 push @all_sql, $sql;
1294 push @all_bind, @bind;
1298 sprintf('%s %s ( %s )',
1301 join(', ', @all_sql)
1303 $self->_bindtype($k, @all_bind),
1306 else { # empty list: some databases won't understand "IN ()", so DWIM
1307 my $sql = ($op =~ /\bnot\b/i) ? $self->{sqltrue} : $self->{sqlfalse};
1312 SCALARREF => sub { # literal SQL
1313 my $sql = $self->_open_outer_paren($$vals);
1314 return ("$label $op ( $sql )");
1316 ARRAYREFREF => sub { # literal SQL with bind
1317 my ($sql, @bind) = @$$vals;
1318 $self->_assert_bindval_matches_bindtype(@bind);
1319 $sql = $self->_open_outer_paren($sql);
1320 return ("$label $op ( $sql )", @bind);
1324 puke "Argument passed to the '$op' operator can not be undefined";
1328 puke "special op $op requires an arrayref (or scalarref/arrayref-ref)";
1332 return ($sql, @bind);
1335 # Some databases (SQLite) treat col IN (1, 2) different from
1336 # col IN ( (1, 2) ). Use this to strip all outer parens while
1337 # adding them back in the corresponding method
1338 sub _open_outer_paren {
1339 my ($self, $sql) = @_;
1341 while (my ($inner) = $sql =~ /^ \s* \( (.*) \) \s* $/xs) {
1343 # there are closing parens inside, need the heavy duty machinery
1344 # to reevaluate the extraction starting from $sql (full reevaluation)
1345 if ($inner =~ /\)/) {
1346 require Text::Balanced;
1348 my (undef, $remainder) = do {
1349 # idiotic design - writes to $@ but *DOES NOT* throw exceptions
1351 Text::Balanced::extract_bracketed($sql, '()', qr/\s*/);
1354 # the entire expression needs to be a balanced bracketed thing
1355 # (after an extract no remainder sans trailing space)
1356 last if defined $remainder and $remainder =~ /\S/;
1366 #======================================================================
1368 #======================================================================
1371 my ($self, $arg) = @_;
1374 for my $c ($self->_order_by_chunks($arg) ) {
1375 $self->_SWITCH_refkind($c, {
1376 SCALAR => sub { push @sql, $c },
1377 ARRAYREF => sub { push @sql, shift @$c; push @bind, @$c },
1383 $self->_sqlcase(' order by'),
1389 return wantarray ? ($sql, @bind) : $sql;
1392 sub _order_by_chunks {
1393 my ($self, $arg) = @_;
1395 return $self->_SWITCH_refkind($arg, {
1398 map { $self->_order_by_chunks($_ ) } @$arg;
1401 ARRAYREFREF => sub {
1402 my ($s, @b) = @$$arg;
1403 $self->_assert_bindval_matches_bindtype(@b);
1407 SCALAR => sub {$self->_quote($arg)},
1409 UNDEF => sub {return () },
1411 SCALARREF => sub {$$arg}, # literal SQL, no quoting
1414 # get first pair in hash
1415 my ($key, $val, @rest) = %$arg;
1417 return () unless $key;
1419 if (@rest or not $key =~ /^-(desc|asc)/i) {
1420 puke "hash passed to _order_by must have exactly one key (-desc or -asc)";
1426 for my $c ($self->_order_by_chunks($val)) {
1429 $self->_SWITCH_refkind($c, {
1434 ($sql, @bind) = @$c;
1438 $sql = $sql . ' ' . $self->_sqlcase($direction);
1440 push @ret, [ $sql, @bind];
1449 #======================================================================
1450 # DATASOURCE (FOR NOW, JUST PLAIN TABLE OR LIST OF TABLES)
1451 #======================================================================
1456 $self->_SWITCH_refkind($from, {
1457 ARRAYREF => sub {join ', ', map { $self->_quote($_) } @$from;},
1458 SCALAR => sub {$self->_quote($from)},
1459 SCALARREF => sub {$$from},
1464 #======================================================================
1466 #======================================================================
1468 # highly optimized, as it's called way too often
1470 # my ($self, $label) = @_;
1472 return '' unless defined $_[1];
1473 return ${$_[1]} if ref($_[1]) eq 'SCALAR';
1475 $_[0]->{quote_char} or
1476 ($_[0]->_assert_pass_injection_guard($_[1]), return $_[1]);
1478 my $qref = ref $_[0]->{quote_char};
1480 !$qref ? ($_[0]->{quote_char}, $_[0]->{quote_char})
1481 : ($qref eq 'ARRAY') ? @{$_[0]->{quote_char}}
1482 : puke "Unsupported quote_char format: $_[0]->{quote_char}";
1484 my $esc = $_[0]->{escape_char} || $r;
1486 # parts containing * are naturally unquoted
1487 return join($_[0]->{name_sep}||'', map
1488 +( $_ eq '*' ? $_ : do { (my $n = $_) =~ s/(\Q$esc\E|\Q$r\E)/$esc$1/g; $l . $n . $r } ),
1489 ( $_[0]->{name_sep} ? split (/\Q$_[0]->{name_sep}\E/, $_[1] ) : $_[1] )
1494 # Conversion, if applicable
1496 #my ($self, $arg) = @_;
1497 if ($_[0]->{convert}) {
1498 return $_[0]->_sqlcase($_[0]->{convert}) .'(' . $_[1] . ')';
1505 #my ($self, $col, @vals) = @_;
1506 # called often - tighten code
1507 return $_[0]->{bindtype} eq 'columns'
1508 ? map {[$_[1], $_]} @_[2 .. $#_]
1513 # Dies if any element of @bind is not in [colname => value] format
1514 # if bindtype is 'columns'.
1515 sub _assert_bindval_matches_bindtype {
1516 # my ($self, @bind) = @_;
1518 if ($self->{bindtype} eq 'columns') {
1520 if (!defined $_ || ref($_) ne 'ARRAY' || @$_ != 2) {
1521 puke "bindtype 'columns' selected, you need to pass: [column_name => bind_value]"
1527 sub _join_sql_clauses {
1528 my ($self, $logic, $clauses_aref, $bind_aref) = @_;
1530 if (@$clauses_aref > 1) {
1531 my $join = " " . $self->_sqlcase($logic) . " ";
1532 my $sql = '( ' . join($join, @$clauses_aref) . ' )';
1533 return ($sql, @$bind_aref);
1535 elsif (@$clauses_aref) {
1536 return ($clauses_aref->[0], @$bind_aref); # no parentheses
1539 return (); # if no SQL, ignore @$bind_aref
1544 # Fix SQL case, if so requested
1546 # LDNOTE: if $self->{case} is true, then it contains 'lower', so we
1547 # don't touch the argument ... crooked logic, but let's not change it!
1548 return $_[0]->{case} ? $_[1] : uc($_[1]);
1552 #======================================================================
1553 # DISPATCHING FROM REFKIND
1554 #======================================================================
1557 my ($self, $data) = @_;
1559 return 'UNDEF' unless defined $data;
1561 # blessed objects are treated like scalars
1562 my $ref = (Scalar::Util::blessed $data) ? '' : ref $data;
1564 return 'SCALAR' unless $ref;
1567 while ($ref eq 'REF') {
1569 $ref = (Scalar::Util::blessed $data) ? '' : ref $data;
1573 return ($ref||'SCALAR') . ('REF' x $n_steps);
1577 my ($self, $data) = @_;
1578 my @try = ($self->_refkind($data));
1579 push @try, 'SCALAR_or_UNDEF' if $try[0] eq 'SCALAR' || $try[0] eq 'UNDEF';
1580 push @try, 'FALLBACK';
1584 sub _METHOD_FOR_refkind {
1585 my ($self, $meth_prefix, $data) = @_;
1588 for (@{$self->_try_refkind($data)}) {
1589 $method = $self->can($meth_prefix."_".$_)
1593 return $method || puke "cannot dispatch on '$meth_prefix' for ".$self->_refkind($data);
1597 sub _SWITCH_refkind {
1598 my ($self, $data, $dispatch_table) = @_;
1601 for (@{$self->_try_refkind($data)}) {
1602 $coderef = $dispatch_table->{$_}
1606 puke "no dispatch entry for ".$self->_refkind($data)
1615 #======================================================================
1616 # VALUES, GENERATE, AUTOLOAD
1617 #======================================================================
1619 # LDNOTE: original code from nwiger, didn't touch code in that section
1620 # I feel the AUTOLOAD stuff should not be the default, it should
1621 # only be activated on explicit demand by user.
1625 my $data = shift || return;
1626 puke "Argument to ", __PACKAGE__, "->values must be a \\%hash"
1627 unless ref $data eq 'HASH';
1630 foreach my $k (sort keys %$data) {
1631 my $v = $data->{$k};
1632 $self->_SWITCH_refkind($v, {
1634 if ($self->{array_datatypes}) { # array datatype
1635 push @all_bind, $self->_bindtype($k, $v);
1637 else { # literal SQL with bind
1638 my ($sql, @bind) = @$v;
1639 $self->_assert_bindval_matches_bindtype(@bind);
1640 push @all_bind, @bind;
1643 ARRAYREFREF => sub { # literal SQL with bind
1644 my ($sql, @bind) = @${$v};
1645 $self->_assert_bindval_matches_bindtype(@bind);
1646 push @all_bind, @bind;
1648 SCALARREF => sub { # literal SQL without bind
1650 SCALAR_or_UNDEF => sub {
1651 push @all_bind, $self->_bindtype($k, $v);
1662 my(@sql, @sqlq, @sqlv);
1666 if ($ref eq 'HASH') {
1667 for my $k (sort keys %$_) {
1670 my $label = $self->_quote($k);
1671 if ($r eq 'ARRAY') {
1672 # literal SQL with bind
1673 my ($sql, @bind) = @$v;
1674 $self->_assert_bindval_matches_bindtype(@bind);
1675 push @sqlq, "$label = $sql";
1677 } elsif ($r eq 'SCALAR') {
1678 # literal SQL without bind
1679 push @sqlq, "$label = $$v";
1681 push @sqlq, "$label = ?";
1682 push @sqlv, $self->_bindtype($k, $v);
1685 push @sql, $self->_sqlcase('set'), join ', ', @sqlq;
1686 } elsif ($ref eq 'ARRAY') {
1687 # unlike insert(), assume these are ONLY the column names, i.e. for SQL
1690 if ($r eq 'ARRAY') { # literal SQL with bind
1691 my ($sql, @bind) = @$v;
1692 $self->_assert_bindval_matches_bindtype(@bind);
1695 } elsif ($r eq 'SCALAR') { # literal SQL without bind
1696 # embedded literal SQL
1703 push @sql, '(' . join(', ', @sqlq) . ')';
1704 } elsif ($ref eq 'SCALAR') {
1708 # strings get case twiddled
1709 push @sql, $self->_sqlcase($_);
1713 my $sql = join ' ', @sql;
1715 # this is pretty tricky
1716 # if ask for an array, return ($stmt, @bind)
1717 # otherwise, s/?/shift @sqlv/ to put it inline
1719 return ($sql, @sqlv);
1721 1 while $sql =~ s/\?/my $d = shift(@sqlv);
1722 ref $d ? $d->[1] : $d/e;
1731 # This allows us to check for a local, then _form, attr
1733 my($name) = $AUTOLOAD =~ /.*::(.+)/;
1734 return $self->generate($name, @_);
1745 SQL::Abstract - Generate SQL from Perl data structures
1751 my $sql = SQL::Abstract->new;
1753 my($stmt, @bind) = $sql->select($source, \@fields, \%where, $order);
1755 my($stmt, @bind) = $sql->insert($table, \%fieldvals || \@values);
1757 my($stmt, @bind) = $sql->update($table, \%fieldvals, \%where);
1759 my($stmt, @bind) = $sql->delete($table, \%where);
1761 # Then, use these in your DBI statements
1762 my $sth = $dbh->prepare($stmt);
1763 $sth->execute(@bind);
1765 # Just generate the WHERE clause
1766 my($stmt, @bind) = $sql->where(\%where, $order);
1768 # Return values in the same order, for hashed queries
1769 # See PERFORMANCE section for more details
1770 my @bind = $sql->values(\%fieldvals);
1774 This module was inspired by the excellent L<DBIx::Abstract>.
1775 However, in using that module I found that what I really wanted
1776 to do was generate SQL, but still retain complete control over my
1777 statement handles and use the DBI interface. So, I set out to
1778 create an abstract SQL generation module.
1780 While based on the concepts used by L<DBIx::Abstract>, there are
1781 several important differences, especially when it comes to WHERE
1782 clauses. I have modified the concepts used to make the SQL easier
1783 to generate from Perl data structures and, IMO, more intuitive.
1784 The underlying idea is for this module to do what you mean, based
1785 on the data structures you provide it. The big advantage is that
1786 you don't have to modify your code every time your data changes,
1787 as this module figures it out.
1789 To begin with, an SQL INSERT is as easy as just specifying a hash
1790 of C<key=value> pairs:
1793 name => 'Jimbo Bobson',
1794 phone => '123-456-7890',
1795 address => '42 Sister Lane',
1796 city => 'St. Louis',
1797 state => 'Louisiana',
1800 The SQL can then be generated with this:
1802 my($stmt, @bind) = $sql->insert('people', \%data);
1804 Which would give you something like this:
1806 $stmt = "INSERT INTO people
1807 (address, city, name, phone, state)
1808 VALUES (?, ?, ?, ?, ?)";
1809 @bind = ('42 Sister Lane', 'St. Louis', 'Jimbo Bobson',
1810 '123-456-7890', 'Louisiana');
1812 These are then used directly in your DBI code:
1814 my $sth = $dbh->prepare($stmt);
1815 $sth->execute(@bind);
1817 =head2 Inserting and Updating Arrays
1819 If your database has array types (like for example Postgres),
1820 activate the special option C<< array_datatypes => 1 >>
1821 when creating the C<SQL::Abstract> object.
1822 Then you may use an arrayref to insert and update database array types:
1824 my $sql = SQL::Abstract->new(array_datatypes => 1);
1826 planets => [qw/Mercury Venus Earth Mars/]
1829 my($stmt, @bind) = $sql->insert('solar_system', \%data);
1833 $stmt = "INSERT INTO solar_system (planets) VALUES (?)"
1835 @bind = (['Mercury', 'Venus', 'Earth', 'Mars']);
1838 =head2 Inserting and Updating SQL
1840 In order to apply SQL functions to elements of your C<%data> you may
1841 specify a reference to an arrayref for the given hash value. For example,
1842 if you need to execute the Oracle C<to_date> function on a value, you can
1843 say something like this:
1847 date_entered => \[ "to_date(?,'MM/DD/YYYY')", "03/02/2003" ],
1850 The first value in the array is the actual SQL. Any other values are
1851 optional and would be included in the bind values array. This gives
1854 my($stmt, @bind) = $sql->insert('people', \%data);
1856 $stmt = "INSERT INTO people (name, date_entered)
1857 VALUES (?, to_date(?,'MM/DD/YYYY'))";
1858 @bind = ('Bill', '03/02/2003');
1860 An UPDATE is just as easy, all you change is the name of the function:
1862 my($stmt, @bind) = $sql->update('people', \%data);
1864 Notice that your C<%data> isn't touched; the module will generate
1865 the appropriately quirky SQL for you automatically. Usually you'll
1866 want to specify a WHERE clause for your UPDATE, though, which is
1867 where handling C<%where> hashes comes in handy...
1869 =head2 Complex where statements
1871 This module can generate pretty complicated WHERE statements
1872 easily. For example, simple C<key=value> pairs are taken to mean
1873 equality, and if you want to see if a field is within a set
1874 of values, you can use an arrayref. Let's say we wanted to
1875 SELECT some data based on this criteria:
1878 requestor => 'inna',
1879 worker => ['nwiger', 'rcwe', 'sfz'],
1880 status => { '!=', 'completed' }
1883 my($stmt, @bind) = $sql->select('tickets', '*', \%where);
1885 The above would give you something like this:
1887 $stmt = "SELECT * FROM tickets WHERE
1888 ( requestor = ? ) AND ( status != ? )
1889 AND ( worker = ? OR worker = ? OR worker = ? )";
1890 @bind = ('inna', 'completed', 'nwiger', 'rcwe', 'sfz');
1892 Which you could then use in DBI code like so:
1894 my $sth = $dbh->prepare($stmt);
1895 $sth->execute(@bind);
1901 The methods are simple. There's one for every major SQL operation,
1902 and a constructor you use first. The arguments are specified in a
1903 similar order for each method (table, then fields, then a where
1904 clause) to try and simplify things.
1906 =head2 new(option => 'value')
1908 The C<new()> function takes a list of options and values, and returns
1909 a new B<SQL::Abstract> object which can then be used to generate SQL
1910 through the methods below. The options accepted are:
1916 If set to 'lower', then SQL will be generated in all lowercase. By
1917 default SQL is generated in "textbook" case meaning something like:
1919 SELECT a_field FROM a_table WHERE some_field LIKE '%someval%'
1921 Any setting other than 'lower' is ignored.
1925 This determines what the default comparison operator is. By default
1926 it is C<=>, meaning that a hash like this:
1928 %where = (name => 'nwiger', email => 'nate@wiger.org');
1930 Will generate SQL like this:
1932 WHERE name = 'nwiger' AND email = 'nate@wiger.org'
1934 However, you may want loose comparisons by default, so if you set
1935 C<cmp> to C<like> you would get SQL such as:
1937 WHERE name like 'nwiger' AND email like 'nate@wiger.org'
1939 You can also override the comparison on an individual basis - see
1940 the huge section on L</"WHERE CLAUSES"> at the bottom.
1942 =item sqltrue, sqlfalse
1944 Expressions for inserting boolean values within SQL statements.
1945 By default these are C<1=1> and C<1=0>. They are used
1946 by the special operators C<-in> and C<-not_in> for generating
1947 correct SQL even when the argument is an empty array (see below).
1951 This determines the default logical operator for multiple WHERE
1952 statements in arrays or hashes. If absent, the default logic is "or"
1953 for arrays, and "and" for hashes. This means that a WHERE
1957 event_date => {'>=', '2/13/99'},
1958 event_date => {'<=', '4/24/03'},
1961 will generate SQL like this:
1963 WHERE event_date >= '2/13/99' OR event_date <= '4/24/03'
1965 This is probably not what you want given this query, though (look
1966 at the dates). To change the "OR" to an "AND", simply specify:
1968 my $sql = SQL::Abstract->new(logic => 'and');
1970 Which will change the above C<WHERE> to:
1972 WHERE event_date >= '2/13/99' AND event_date <= '4/24/03'
1974 The logic can also be changed locally by inserting
1975 a modifier in front of an arrayref:
1977 @where = (-and => [event_date => {'>=', '2/13/99'},
1978 event_date => {'<=', '4/24/03'} ]);
1980 See the L</"WHERE CLAUSES"> section for explanations.
1984 This will automatically convert comparisons using the specified SQL
1985 function for both column and value. This is mostly used with an argument
1986 of C<upper> or C<lower>, so that the SQL will have the effect of
1987 case-insensitive "searches". For example, this:
1989 $sql = SQL::Abstract->new(convert => 'upper');
1990 %where = (keywords => 'MaKe iT CAse inSeNSItive');
1992 Will turn out the following SQL:
1994 WHERE upper(keywords) like upper('MaKe iT CAse inSeNSItive')
1996 The conversion can be C<upper()>, C<lower()>, or any other SQL function
1997 that can be applied symmetrically to fields (actually B<SQL::Abstract> does
1998 not validate this option; it will just pass through what you specify verbatim).
2002 This is a kludge because many databases suck. For example, you can't
2003 just bind values using DBI's C<execute()> for Oracle C<CLOB> or C<BLOB> fields.
2004 Instead, you have to use C<bind_param()>:
2006 $sth->bind_param(1, 'reg data');
2007 $sth->bind_param(2, $lots, {ora_type => ORA_CLOB});
2009 The problem is, B<SQL::Abstract> will normally just return a C<@bind> array,
2010 which loses track of which field each slot refers to. Fear not.
2012 If you specify C<bindtype> in new, you can determine how C<@bind> is returned.
2013 Currently, you can specify either C<normal> (default) or C<columns>. If you
2014 specify C<columns>, you will get an array that looks like this:
2016 my $sql = SQL::Abstract->new(bindtype => 'columns');
2017 my($stmt, @bind) = $sql->insert(...);
2020 [ 'column1', 'value1' ],
2021 [ 'column2', 'value2' ],
2022 [ 'column3', 'value3' ],
2025 You can then iterate through this manually, using DBI's C<bind_param()>.
2027 $sth->prepare($stmt);
2030 my($col, $data) = @$_;
2031 if ($col eq 'details' || $col eq 'comments') {
2032 $sth->bind_param($i, $data, {ora_type => ORA_CLOB});
2033 } elsif ($col eq 'image') {
2034 $sth->bind_param($i, $data, {ora_type => ORA_BLOB});
2036 $sth->bind_param($i, $data);
2040 $sth->execute; # execute without @bind now
2042 Now, why would you still use B<SQL::Abstract> if you have to do this crap?
2043 Basically, the advantage is still that you don't have to care which fields
2044 are or are not included. You could wrap that above C<for> loop in a simple
2045 sub called C<bind_fields()> or something and reuse it repeatedly. You still
2046 get a layer of abstraction over manual SQL specification.
2048 Note that if you set L</bindtype> to C<columns>, the C<\[ $sql, @bind ]>
2049 construct (see L</Literal SQL with placeholders and bind values (subqueries)>)
2050 will expect the bind values in this format.
2054 This is the character that a table or column name will be quoted
2055 with. By default this is an empty string, but you could set it to
2056 the character C<`>, to generate SQL like this:
2058 SELECT `a_field` FROM `a_table` WHERE `some_field` LIKE '%someval%'
2060 Alternatively, you can supply an array ref of two items, the first being the left
2061 hand quote character, and the second the right hand quote character. For
2062 example, you could supply C<['[',']']> for SQL Server 2000 compliant quotes
2063 that generates SQL like this:
2065 SELECT [a_field] FROM [a_table] WHERE [some_field] LIKE '%someval%'
2067 Quoting is useful if you have tables or columns names that are reserved
2068 words in your database's SQL dialect.
2072 This is the character that will be used to escape L</quote_char>s appearing
2073 in an identifier before it has been quoted.
2075 The parameter default in case of a single L</quote_char> character is the quote
2078 When opening-closing-style quoting is used (L</quote_char> is an arrayref)
2079 this parameter defaults to the B<closing (right)> L</quote_char>. Occurrences
2080 of the B<opening (left)> L</quote_char> within the identifier are currently left
2081 untouched. The default for opening-closing-style quotes may change in future
2082 versions, thus you are B<strongly encouraged> to specify the escape character
2087 This is the character that separates a table and column name. It is
2088 necessary to specify this when the C<quote_char> option is selected,
2089 so that tables and column names can be individually quoted like this:
2091 SELECT `table`.`one_field` FROM `table` WHERE `table`.`other_field` = 1
2093 =item injection_guard
2095 A regular expression C<qr/.../> that is applied to any C<-function> and unquoted
2096 column name specified in a query structure. This is a safety mechanism to avoid
2097 injection attacks when mishandling user input e.g.:
2099 my %condition_as_column_value_pairs = get_values_from_user();
2100 $sqla->select( ... , \%condition_as_column_value_pairs );
2102 If the expression matches an exception is thrown. Note that literal SQL
2103 supplied via C<\'...'> or C<\['...']> is B<not> checked in any way.
2105 Defaults to checking for C<;> and the C<GO> keyword (TransactSQL)
2107 =item array_datatypes
2109 When this option is true, arrayrefs in INSERT or UPDATE are
2110 interpreted as array datatypes and are passed directly
2112 When this option is false, arrayrefs are interpreted
2113 as literal SQL, just like refs to arrayrefs
2114 (but this behavior is for backwards compatibility; when writing
2115 new queries, use the "reference to arrayref" syntax
2121 Takes a reference to a list of "special operators"
2122 to extend the syntax understood by L<SQL::Abstract>.
2123 See section L</"SPECIAL OPERATORS"> for details.
2127 Takes a reference to a list of "unary operators"
2128 to extend the syntax understood by L<SQL::Abstract>.
2129 See section L</"UNARY OPERATORS"> for details.
2135 =head2 insert($table, \@values || \%fieldvals, \%options)
2137 This is the simplest function. You simply give it a table name
2138 and either an arrayref of values or hashref of field/value pairs.
2139 It returns an SQL INSERT statement and a list of bind values.
2140 See the sections on L</"Inserting and Updating Arrays"> and
2141 L</"Inserting and Updating SQL"> for information on how to insert
2142 with those data types.
2144 The optional C<\%options> hash reference may contain additional
2145 options to generate the insert SQL. Currently supported options
2152 Takes either a scalar of raw SQL fields, or an array reference of
2153 field names, and adds on an SQL C<RETURNING> statement at the end.
2154 This allows you to return data generated by the insert statement
2155 (such as row IDs) without performing another C<SELECT> statement.
2156 Note, however, this is not part of the SQL standard and may not
2157 be supported by all database engines.
2161 =head2 update($table, \%fieldvals, \%where, \%options)
2163 This takes a table, hashref of field/value pairs, and an optional
2164 hashref L<WHERE clause|/WHERE CLAUSES>. It returns an SQL UPDATE function and a list
2166 See the sections on L</"Inserting and Updating Arrays"> and
2167 L</"Inserting and Updating SQL"> for information on how to insert
2168 with those data types.
2170 The optional C<\%options> hash reference may contain additional
2171 options to generate the update SQL. Currently supported options
2178 See the C<returning> option to
2179 L<insert|/insert($table, \@values || \%fieldvals, \%options)>.
2183 =head2 select($source, $fields, $where, $order)
2185 This returns a SQL SELECT statement and associated list of bind values, as
2186 specified by the arguments:
2192 Specification of the 'FROM' part of the statement.
2193 The argument can be either a plain scalar (interpreted as a table
2194 name, will be quoted), or an arrayref (interpreted as a list
2195 of table names, joined by commas, quoted), or a scalarref
2196 (literal SQL, not quoted).
2200 Specification of the list of fields to retrieve from
2202 The argument can be either an arrayref (interpreted as a list
2203 of field names, will be joined by commas and quoted), or a
2204 plain scalar (literal SQL, not quoted).
2205 Please observe that this API is not as flexible as that of
2206 the first argument C<$source>, for backwards compatibility reasons.
2210 Optional argument to specify the WHERE part of the query.
2211 The argument is most often a hashref, but can also be
2212 an arrayref or plain scalar --
2213 see section L<WHERE clause|/"WHERE CLAUSES"> for details.
2217 Optional argument to specify the ORDER BY part of the query.
2218 The argument can be a scalar, a hashref or an arrayref
2219 -- see section L<ORDER BY clause|/"ORDER BY CLAUSES">
2225 =head2 delete($table, \%where, \%options)
2227 This takes a table name and optional hashref L<WHERE clause|/WHERE CLAUSES>.
2228 It returns an SQL DELETE statement and list of bind values.
2230 The optional C<\%options> hash reference may contain additional
2231 options to generate the delete SQL. Currently supported options
2238 See the C<returning> option to
2239 L<insert|/insert($table, \@values || \%fieldvals, \%options)>.
2243 =head2 where(\%where, $order)
2245 This is used to generate just the WHERE clause. For example,
2246 if you have an arbitrary data structure and know what the
2247 rest of your SQL is going to look like, but want an easy way
2248 to produce a WHERE clause, use this. It returns an SQL WHERE
2249 clause and list of bind values.
2252 =head2 values(\%data)
2254 This just returns the values from the hash C<%data>, in the same
2255 order that would be returned from any of the other above queries.
2256 Using this allows you to markedly speed up your queries if you
2257 are affecting lots of rows. See below under the L</"PERFORMANCE"> section.
2259 =head2 generate($any, 'number', $of, \@data, $struct, \%types)
2261 Warning: This is an experimental method and subject to change.
2263 This returns arbitrarily generated SQL. It's a really basic shortcut.
2264 It will return two different things, depending on return context:
2266 my($stmt, @bind) = $sql->generate('create table', \$table, \@fields);
2267 my $stmt_and_val = $sql->generate('create table', \$table, \@fields);
2269 These would return the following:
2271 # First calling form
2272 $stmt = "CREATE TABLE test (?, ?)";
2273 @bind = (field1, field2);
2275 # Second calling form
2276 $stmt_and_val = "CREATE TABLE test (field1, field2)";
2278 Depending on what you're trying to do, it's up to you to choose the correct
2279 format. In this example, the second form is what you would want.
2283 $sql->generate('alter session', { nls_date_format => 'MM/YY' });
2287 ALTER SESSION SET nls_date_format = 'MM/YY'
2289 You get the idea. Strings get their case twiddled, but everything
2290 else remains verbatim.
2292 =head1 EXPORTABLE FUNCTIONS
2294 =head2 is_plain_value
2296 Determines if the supplied argument is a plain value as understood by this
2301 =item * The value is C<undef>
2303 =item * The value is a non-reference
2305 =item * The value is an object with stringification overloading
2307 =item * The value is of the form C<< { -value => $anything } >>
2311 On failure returns C<undef>, on success returns a B<scalar> reference
2312 to the original supplied argument.
2318 The stringification overloading detection is rather advanced: it takes
2319 into consideration not only the presence of a C<""> overload, but if that
2320 fails also checks for enabled
2321 L<autogenerated versions of C<"">|overload/Magic Autogeneration>, based
2322 on either C<0+> or C<bool>.
2324 Unfortunately testing in the field indicates that this
2325 detection B<< may tickle a latent bug in perl versions before 5.018 >>,
2326 but only when very large numbers of stringifying objects are involved.
2327 At the time of writing ( Sep 2014 ) there is no clear explanation of
2328 the direct cause, nor is there a manageably small test case that reliably
2329 reproduces the problem.
2331 If you encounter any of the following exceptions in B<random places within
2332 your application stack> - this module may be to blame:
2334 Operation "ne": no method found,
2335 left argument in overloaded package <something>,
2336 right argument in overloaded package <something>
2340 Stub found while resolving method "???" overloading """" in package <something>
2342 If you fall victim to the above - please attempt to reduce the problem
2343 to something that could be sent to the L<SQL::Abstract developers
2344 |DBIx::Class/GETTING HELP/SUPPORT>
2345 (either publicly or privately). As a workaround in the meantime you can
2346 set C<$ENV{SQLA_ISVALUE_IGNORE_AUTOGENERATED_STRINGIFICATION}> to a true
2347 value, which will most likely eliminate your problem (at the expense of
2348 not being able to properly detect exotic forms of stringification).
2350 This notice and environment variable will be removed in a future version,
2351 as soon as the underlying problem is found and a reliable workaround is
2356 =head2 is_literal_value
2358 Determines if the supplied argument is a literal value as understood by this
2363 =item * C<\$sql_string>
2365 =item * C<\[ $sql_string, @bind_values ]>
2369 On failure returns C<undef>, on success returns an B<array> reference
2370 containing the unpacked version of the supplied literal SQL and bind values.
2372 =head1 WHERE CLAUSES
2376 This module uses a variation on the idea from L<DBIx::Abstract>. It
2377 is B<NOT>, repeat I<not> 100% compatible. B<The main logic of this
2378 module is that things in arrays are OR'ed, and things in hashes
2381 The easiest way to explain is to show lots of examples. After
2382 each C<%where> hash shown, it is assumed you used:
2384 my($stmt, @bind) = $sql->where(\%where);
2386 However, note that the C<%where> hash can be used directly in any
2387 of the other functions as well, as described above.
2389 =head2 Key-value pairs
2391 So, let's get started. To begin, a simple hash:
2395 status => 'completed'
2398 Is converted to SQL C<key = val> statements:
2400 $stmt = "WHERE user = ? AND status = ?";
2401 @bind = ('nwiger', 'completed');
2403 One common thing I end up doing is having a list of values that
2404 a field can be in. To do this, simply specify a list inside of
2409 status => ['assigned', 'in-progress', 'pending'];
2412 This simple code will create the following:
2414 $stmt = "WHERE user = ? AND ( status = ? OR status = ? OR status = ? )";
2415 @bind = ('nwiger', 'assigned', 'in-progress', 'pending');
2417 A field associated to an empty arrayref will be considered a
2418 logical false and will generate 0=1.
2420 =head2 Tests for NULL values
2422 If the value part is C<undef> then this is converted to SQL <IS NULL>
2431 $stmt = "WHERE user = ? AND status IS NULL";
2434 To test if a column IS NOT NULL:
2438 status => { '!=', undef },
2441 =head2 Specific comparison operators
2443 If you want to specify a different type of operator for your comparison,
2444 you can use a hashref for a given column:
2448 status => { '!=', 'completed' }
2451 Which would generate:
2453 $stmt = "WHERE user = ? AND status != ?";
2454 @bind = ('nwiger', 'completed');
2456 To test against multiple values, just enclose the values in an arrayref:
2458 status => { '=', ['assigned', 'in-progress', 'pending'] };
2460 Which would give you:
2462 "WHERE status = ? OR status = ? OR status = ?"
2465 The hashref can also contain multiple pairs, in which case it is expanded
2466 into an C<AND> of its elements:
2470 status => { '!=', 'completed', -not_like => 'pending%' }
2473 # Or more dynamically, like from a form
2474 $where{user} = 'nwiger';
2475 $where{status}{'!='} = 'completed';
2476 $where{status}{'-not_like'} = 'pending%';
2478 # Both generate this
2479 $stmt = "WHERE user = ? AND status != ? AND status NOT LIKE ?";
2480 @bind = ('nwiger', 'completed', 'pending%');
2483 To get an OR instead, you can combine it with the arrayref idea:
2487 priority => [ { '=', 2 }, { '>', 5 } ]
2490 Which would generate:
2492 $stmt = "WHERE ( priority = ? OR priority > ? ) AND user = ?";
2493 @bind = ('2', '5', 'nwiger');
2495 If you want to include literal SQL (with or without bind values), just use a
2496 scalar reference or reference to an arrayref as the value:
2499 date_entered => { '>' => \["to_date(?, 'MM/DD/YYYY')", "11/26/2008"] },
2500 date_expires => { '<' => \"now()" }
2503 Which would generate:
2505 $stmt = "WHERE date_entered > to_date(?, 'MM/DD/YYYY') AND date_expires < now()";
2506 @bind = ('11/26/2008');
2509 =head2 Logic and nesting operators
2511 In the example above,
2512 there is a subtle trap if you want to say something like
2513 this (notice the C<AND>):
2515 WHERE priority != ? AND priority != ?
2517 Because, in Perl you I<can't> do this:
2519 priority => { '!=' => 2, '!=' => 1 }
2521 As the second C<!=> key will obliterate the first. The solution
2522 is to use the special C<-modifier> form inside an arrayref:
2524 priority => [ -and => {'!=', 2},
2528 Normally, these would be joined by C<OR>, but the modifier tells it
2529 to use C<AND> instead. (Hint: You can use this in conjunction with the
2530 C<logic> option to C<new()> in order to change the way your queries
2531 work by default.) B<Important:> Note that the C<-modifier> goes
2532 B<INSIDE> the arrayref, as an extra first element. This will
2533 B<NOT> do what you think it might:
2535 priority => -and => [{'!=', 2}, {'!=', 1}] # WRONG!
2537 Here is a quick list of equivalencies, since there is some overlap:
2540 status => {'!=', 'completed', 'not like', 'pending%' }
2541 status => [ -and => {'!=', 'completed'}, {'not like', 'pending%'}]
2544 status => {'=', ['assigned', 'in-progress']}
2545 status => [ -or => {'=', 'assigned'}, {'=', 'in-progress'}]
2546 status => [ {'=', 'assigned'}, {'=', 'in-progress'} ]
2550 =head2 Special operators: IN, BETWEEN, etc.
2552 You can also use the hashref format to compare a list of fields using the
2553 C<IN> comparison operator, by specifying the list as an arrayref:
2556 status => 'completed',
2557 reportid => { -in => [567, 2335, 2] }
2560 Which would generate:
2562 $stmt = "WHERE status = ? AND reportid IN (?,?,?)";
2563 @bind = ('completed', '567', '2335', '2');
2565 The reverse operator C<-not_in> generates SQL C<NOT IN> and is used in
2568 If the argument to C<-in> is an empty array, 'sqlfalse' is generated
2569 (by default: C<1=0>). Similarly, C<< -not_in => [] >> generates
2570 'sqltrue' (by default: C<1=1>).
2572 In addition to the array you can supply a chunk of literal sql or
2573 literal sql with bind:
2576 customer => { -in => \[
2577 'SELECT cust_id FROM cust WHERE balance > ?',
2580 status => { -in => \'SELECT status_codes FROM states' },
2586 customer IN ( SELECT cust_id FROM cust WHERE balance > ? )
2587 AND status IN ( SELECT status_codes FROM states )
2591 Finally, if the argument to C<-in> is not a reference, it will be
2592 treated as a single-element array.
2594 Another pair of operators is C<-between> and C<-not_between>,
2595 used with an arrayref of two values:
2599 completion_date => {
2600 -not_between => ['2002-10-01', '2003-02-06']
2606 WHERE user = ? AND completion_date NOT BETWEEN ( ? AND ? )
2608 Just like with C<-in> all plausible combinations of literal SQL
2612 start0 => { -between => [ 1, 2 ] },
2613 start1 => { -between => \["? AND ?", 1, 2] },
2614 start2 => { -between => \"lower(x) AND upper(y)" },
2615 start3 => { -between => [
2617 \["upper(?)", 'stuff' ],
2624 ( start0 BETWEEN ? AND ? )
2625 AND ( start1 BETWEEN ? AND ? )
2626 AND ( start2 BETWEEN lower(x) AND upper(y) )
2627 AND ( start3 BETWEEN lower(x) AND upper(?) )
2629 @bind = (1, 2, 1, 2, 'stuff');
2632 These are the two builtin "special operators"; but the
2633 list can be expanded: see section L</"SPECIAL OPERATORS"> below.
2635 =head2 Unary operators: bool
2637 If you wish to test against boolean columns or functions within your
2638 database you can use the C<-bool> and C<-not_bool> operators. For
2639 example to test the column C<is_user> being true and the column
2640 C<is_enabled> being false you would use:-
2644 -not_bool => 'is_enabled',
2649 WHERE is_user AND NOT is_enabled
2651 If a more complex combination is required, testing more conditions,
2652 then you should use the and/or operators:-
2657 -not_bool => { two=> { -rlike => 'bar' } },
2658 -not_bool => { three => [ { '=', 2 }, { '>', 5 } ] },
2669 (NOT ( three = ? OR three > ? ))
2672 =head2 Nested conditions, -and/-or prefixes
2674 So far, we've seen how multiple conditions are joined with a top-level
2675 C<AND>. We can change this by putting the different conditions we want in
2676 hashes and then putting those hashes in an array. For example:
2681 status => { -like => ['pending%', 'dispatched'] },
2685 status => 'unassigned',
2689 This data structure would create the following:
2691 $stmt = "WHERE ( user = ? AND ( status LIKE ? OR status LIKE ? ) )
2692 OR ( user = ? AND status = ? ) )";
2693 @bind = ('nwiger', 'pending', 'dispatched', 'robot', 'unassigned');
2696 Clauses in hashrefs or arrayrefs can be prefixed with an C<-and> or C<-or>
2697 to change the logic inside:
2703 -and => [ workhrs => {'>', 20}, geo => 'ASIA' ],
2704 -or => { workhrs => {'<', 50}, geo => 'EURO' },
2711 $stmt = "WHERE ( user = ?
2712 AND ( ( workhrs > ? AND geo = ? )
2713 OR ( workhrs < ? OR geo = ? ) ) )";
2714 @bind = ('nwiger', '20', 'ASIA', '50', 'EURO');
2716 =head3 Algebraic inconsistency, for historical reasons
2718 C<Important note>: when connecting several conditions, the C<-and->|C<-or>
2719 operator goes C<outside> of the nested structure; whereas when connecting
2720 several constraints on one column, the C<-and> operator goes
2721 C<inside> the arrayref. Here is an example combining both features:
2724 -and => [a => 1, b => 2],
2725 -or => [c => 3, d => 4],
2726 e => [-and => {-like => 'foo%'}, {-like => '%bar'} ]
2731 WHERE ( ( ( a = ? AND b = ? )
2732 OR ( c = ? OR d = ? )
2733 OR ( e LIKE ? AND e LIKE ? ) ) )
2735 This difference in syntax is unfortunate but must be preserved for
2736 historical reasons. So be careful: the two examples below would
2737 seem algebraically equivalent, but they are not
2740 { -like => 'foo%' },
2741 { -like => '%bar' },
2743 # yields: WHERE ( ( col LIKE ? AND col LIKE ? ) )
2746 { col => { -like => 'foo%' } },
2747 { col => { -like => '%bar' } },
2749 # yields: WHERE ( ( col LIKE ? OR col LIKE ? ) )
2752 =head2 Literal SQL and value type operators
2754 The basic premise of SQL::Abstract is that in WHERE specifications the "left
2755 side" is a column name and the "right side" is a value (normally rendered as
2756 a placeholder). This holds true for both hashrefs and arrayref pairs as you
2757 see in the L</WHERE CLAUSES> examples above. Sometimes it is necessary to
2758 alter this behavior. There are several ways of doing so.
2762 This is a virtual operator that signals the string to its right side is an
2763 identifier (a column name) and not a value. For example to compare two
2764 columns you would write:
2767 priority => { '<', 2 },
2768 requestor => { -ident => 'submitter' },
2773 $stmt = "WHERE priority < ? AND requestor = submitter";
2776 If you are maintaining legacy code you may see a different construct as
2777 described in L</Deprecated usage of Literal SQL>, please use C<-ident> in new
2782 This is a virtual operator that signals that the construct to its right side
2783 is a value to be passed to DBI. This is for example necessary when you want
2784 to write a where clause against an array (for RDBMS that support such
2785 datatypes). For example:
2788 array => { -value => [1, 2, 3] }
2793 $stmt = 'WHERE array = ?';
2794 @bind = ([1, 2, 3]);
2796 Note that if you were to simply say:
2802 the result would probably not be what you wanted:
2804 $stmt = 'WHERE array = ? OR array = ? OR array = ?';
2809 Finally, sometimes only literal SQL will do. To include a random snippet
2810 of SQL verbatim, you specify it as a scalar reference. Consider this only
2811 as a last resort. Usually there is a better way. For example:
2814 priority => { '<', 2 },
2815 requestor => { -in => \'(SELECT name FROM hitmen)' },
2820 $stmt = "WHERE priority < ? AND requestor IN (SELECT name FROM hitmen)"
2823 Note that in this example, you only get one bind parameter back, since
2824 the verbatim SQL is passed as part of the statement.
2828 Never use untrusted input as a literal SQL argument - this is a massive
2829 security risk (there is no way to check literal snippets for SQL
2830 injections and other nastyness). If you need to deal with untrusted input
2831 use literal SQL with placeholders as described next.
2833 =head3 Literal SQL with placeholders and bind values (subqueries)
2835 If the literal SQL to be inserted has placeholders and bind values,
2836 use a reference to an arrayref (yes this is a double reference --
2837 not so common, but perfectly legal Perl). For example, to find a date
2838 in Postgres you can use something like this:
2841 date_column => \[ "= date '2008-09-30' - ?::integer", 10 ]
2846 $stmt = "WHERE ( date_column = date '2008-09-30' - ?::integer )"
2849 Note that you must pass the bind values in the same format as they are returned
2850 by L<where|/where(\%where, $order)>. This means that if you set L</bindtype>
2851 to C<columns>, you must provide the bind values in the
2852 C<< [ column_meta => value ] >> format, where C<column_meta> is an opaque
2853 scalar value; most commonly the column name, but you can use any scalar value
2854 (including references and blessed references), L<SQL::Abstract> will simply
2855 pass it through intact. So if C<bindtype> is set to C<columns> the above
2856 example will look like:
2859 date_column => \[ "= date '2008-09-30' - ?::integer", [ {} => 10 ] ]
2862 Literal SQL is especially useful for nesting parenthesized clauses in the
2863 main SQL query. Here is a first example:
2865 my ($sub_stmt, @sub_bind) = ("SELECT c1 FROM t1 WHERE c2 < ? AND c3 LIKE ?",
2869 bar => \["IN ($sub_stmt)" => @sub_bind],
2874 $stmt = "WHERE (foo = ? AND bar IN (SELECT c1 FROM t1
2875 WHERE c2 < ? AND c3 LIKE ?))";
2876 @bind = (1234, 100, "foo%");
2878 Other subquery operators, like for example C<"E<gt> ALL"> or C<"NOT IN">,
2879 are expressed in the same way. Of course the C<$sub_stmt> and
2880 its associated bind values can be generated through a former call
2883 my ($sub_stmt, @sub_bind)
2884 = $sql->select("t1", "c1", {c2 => {"<" => 100},
2885 c3 => {-like => "foo%"}});
2888 bar => \["> ALL ($sub_stmt)" => @sub_bind],
2891 In the examples above, the subquery was used as an operator on a column;
2892 but the same principle also applies for a clause within the main C<%where>
2893 hash, like an EXISTS subquery:
2895 my ($sub_stmt, @sub_bind)
2896 = $sql->select("t1", "*", {c1 => 1, c2 => \"> t0.c0"});
2897 my %where = ( -and => [
2899 \["EXISTS ($sub_stmt)" => @sub_bind],
2904 $stmt = "WHERE (foo = ? AND EXISTS (SELECT * FROM t1
2905 WHERE c1 = ? AND c2 > t0.c0))";
2909 Observe that the condition on C<c2> in the subquery refers to
2910 column C<t0.c0> of the main query: this is I<not> a bind
2911 value, so we have to express it through a scalar ref.
2912 Writing C<< c2 => {">" => "t0.c0"} >> would have generated
2913 C<< c2 > ? >> with bind value C<"t0.c0"> ... not exactly
2914 what we wanted here.
2916 Finally, here is an example where a subquery is used
2917 for expressing unary negation:
2919 my ($sub_stmt, @sub_bind)
2920 = $sql->where({age => [{"<" => 10}, {">" => 20}]});
2921 $sub_stmt =~ s/^ where //i; # don't want "WHERE" in the subclause
2923 lname => {like => '%son%'},
2924 \["NOT ($sub_stmt)" => @sub_bind],
2929 $stmt = "lname LIKE ? AND NOT ( age < ? OR age > ? )"
2930 @bind = ('%son%', 10, 20)
2932 =head3 Deprecated usage of Literal SQL
2934 Below are some examples of archaic use of literal SQL. It is shown only as
2935 reference for those who deal with legacy code. Each example has a much
2936 better, cleaner and safer alternative that users should opt for in new code.
2942 my %where = ( requestor => \'IS NOT NULL' )
2944 $stmt = "WHERE requestor IS NOT NULL"
2946 This used to be the way of generating NULL comparisons, before the handling
2947 of C<undef> got formalized. For new code please use the superior syntax as
2948 described in L</Tests for NULL values>.
2952 my %where = ( requestor => \'= submitter' )
2954 $stmt = "WHERE requestor = submitter"
2956 This used to be the only way to compare columns. Use the superior L</-ident>
2957 method for all new code. For example an identifier declared in such a way
2958 will be properly quoted if L</quote_char> is properly set, while the legacy
2959 form will remain as supplied.
2963 my %where = ( is_ready => \"", completed => { '>', '2012-12-21' } )
2965 $stmt = "WHERE completed > ? AND is_ready"
2966 @bind = ('2012-12-21')
2968 Using an empty string literal used to be the only way to express a boolean.
2969 For all new code please use the much more readable
2970 L<-bool|/Unary operators: bool> operator.
2976 These pages could go on for a while, since the nesting of the data
2977 structures this module can handle are pretty much unlimited (the
2978 module implements the C<WHERE> expansion as a recursive function
2979 internally). Your best bet is to "play around" with the module a
2980 little to see how the data structures behave, and choose the best
2981 format for your data based on that.
2983 And of course, all the values above will probably be replaced with
2984 variables gotten from forms or the command line. After all, if you
2985 knew everything ahead of time, you wouldn't have to worry about
2986 dynamically-generating SQL and could just hardwire it into your
2989 =head1 ORDER BY CLAUSES
2991 Some functions take an order by clause. This can either be a scalar (just a
2992 column name), a hashref of C<< { -desc => 'col' } >> or C<< { -asc => 'col' }
2993 >>, a scalarref, an arrayref-ref, or an arrayref of any of the previous
2996 Given | Will Generate
2997 ---------------------------------------------------------------
2999 'colA' | ORDER BY colA
3001 [qw/colA colB/] | ORDER BY colA, colB
3003 {-asc => 'colA'} | ORDER BY colA ASC
3005 {-desc => 'colB'} | ORDER BY colB DESC
3007 ['colA', {-asc => 'colB'}] | ORDER BY colA, colB ASC
3009 { -asc => [qw/colA colB/] } | ORDER BY colA ASC, colB ASC
3011 \'colA DESC' | ORDER BY colA DESC
3013 \[ 'FUNC(colA, ?)', $x ] | ORDER BY FUNC(colA, ?)
3014 | /* ...with $x bound to ? */
3017 { -asc => 'colA' }, | colA ASC,
3018 { -desc => [qw/colB/] }, | colB DESC,
3019 { -asc => [qw/colC colD/] },| colC ASC, colD ASC,
3020 \'colE DESC', | colE DESC,
3021 \[ 'FUNC(colF, ?)', $x ], | FUNC(colF, ?)
3022 ] | /* ...with $x bound to ? */
3023 ===============================================================
3027 =head1 SPECIAL OPERATORS
3029 my $sqlmaker = SQL::Abstract->new(special_ops => [
3033 my ($self, $field, $op, $arg) = @_;
3039 handler => 'method_name',
3043 A "special operator" is a SQL syntactic clause that can be
3044 applied to a field, instead of a usual binary operator.
3047 WHERE field IN (?, ?, ?)
3048 WHERE field BETWEEN ? AND ?
3049 WHERE MATCH(field) AGAINST (?, ?)
3051 Special operators IN and BETWEEN are fairly standard and therefore
3052 are builtin within C<SQL::Abstract> (as the overridable methods
3053 C<_where_field_IN> and C<_where_field_BETWEEN>). For other operators,
3054 like the MATCH .. AGAINST example above which is specific to MySQL,
3055 you can write your own operator handlers - supply a C<special_ops>
3056 argument to the C<new> method. That argument takes an arrayref of
3057 operator definitions; each operator definition is a hashref with two
3064 the regular expression to match the operator
3068 Either a coderef or a plain scalar method name. In both cases
3069 the expected return is C<< ($sql, @bind) >>.
3071 When supplied with a method name, it is simply called on the
3072 L<SQL::Abstract> object as:
3074 $self->$method_name($field, $op, $arg)
3078 $field is the LHS of the operator
3079 $op is the part that matched the handler regex
3082 When supplied with a coderef, it is called as:
3084 $coderef->($self, $field, $op, $arg)
3089 For example, here is an implementation
3090 of the MATCH .. AGAINST syntax for MySQL
3092 my $sqlmaker = SQL::Abstract->new(special_ops => [
3094 # special op for MySql MATCH (field) AGAINST(word1, word2, ...)
3095 {regex => qr/^match$/i,
3097 my ($self, $field, $op, $arg) = @_;
3098 $arg = [$arg] if not ref $arg;
3099 my $label = $self->_quote($field);
3100 my ($placeholder) = $self->_convert('?');
3101 my $placeholders = join ", ", (($placeholder) x @$arg);
3102 my $sql = $self->_sqlcase('match') . " ($label) "
3103 . $self->_sqlcase('against') . " ($placeholders) ";
3104 my @bind = $self->_bindtype($field, @$arg);
3105 return ($sql, @bind);
3112 =head1 UNARY OPERATORS
3114 my $sqlmaker = SQL::Abstract->new(unary_ops => [
3118 my ($self, $op, $arg) = @_;
3124 handler => 'method_name',
3128 A "unary operator" is a SQL syntactic clause that can be
3129 applied to a field - the operator goes before the field
3131 You can write your own operator handlers - supply a C<unary_ops>
3132 argument to the C<new> method. That argument takes an arrayref of
3133 operator definitions; each operator definition is a hashref with two
3140 the regular expression to match the operator
3144 Either a coderef or a plain scalar method name. In both cases
3145 the expected return is C<< $sql >>.
3147 When supplied with a method name, it is simply called on the
3148 L<SQL::Abstract> object as:
3150 $self->$method_name($op, $arg)
3154 $op is the part that matched the handler regex
3155 $arg is the RHS or argument of the operator
3157 When supplied with a coderef, it is called as:
3159 $coderef->($self, $op, $arg)
3167 Thanks to some benchmarking by Mark Stosberg, it turns out that
3168 this module is many orders of magnitude faster than using C<DBIx::Abstract>.
3169 I must admit this wasn't an intentional design issue, but it's a
3170 byproduct of the fact that you get to control your C<DBI> handles
3173 To maximize performance, use a code snippet like the following:
3175 # prepare a statement handle using the first row
3176 # and then reuse it for the rest of the rows
3178 for my $href (@array_of_hashrefs) {
3179 $stmt ||= $sql->insert('table', $href);
3180 $sth ||= $dbh->prepare($stmt);
3181 $sth->execute($sql->values($href));
3184 The reason this works is because the keys in your C<$href> are sorted
3185 internally by B<SQL::Abstract>. Thus, as long as your data retains
3186 the same structure, you only have to generate the SQL the first time
3187 around. On subsequent queries, simply use the C<values> function provided
3188 by this module to return your values in the correct order.
3190 However this depends on the values having the same type - if, for
3191 example, the values of a where clause may either have values
3192 (resulting in sql of the form C<column = ?> with a single bind
3193 value), or alternatively the values might be C<undef> (resulting in
3194 sql of the form C<column IS NULL> with no bind value) then the
3195 caching technique suggested will not work.
3199 If you use my C<CGI::FormBuilder> module at all, you'll hopefully
3200 really like this part (I do, at least). Building up a complex query
3201 can be as simple as the following:
3208 use CGI::FormBuilder;
3211 my $form = CGI::FormBuilder->new(...);
3212 my $sql = SQL::Abstract->new;
3214 if ($form->submitted) {
3215 my $field = $form->field;
3216 my $id = delete $field->{id};
3217 my($stmt, @bind) = $sql->update('table', $field, {id => $id});
3220 Of course, you would still have to connect using C<DBI> to run the
3221 query, but the point is that if you make your form look like your
3222 table, the actual query script can be extremely simplistic.
3224 If you're B<REALLY> lazy (I am), check out C<HTML::QuickTable> for
3225 a fast interface to returning and formatting data. I frequently
3226 use these three modules together to write complex database query
3227 apps in under 50 lines.
3229 =head1 HOW TO CONTRIBUTE
3231 Contributions are always welcome, in all usable forms (we especially
3232 welcome documentation improvements). The delivery methods include git-
3233 or unified-diff formatted patches, GitHub pull requests, or plain bug
3234 reports either via RT or the Mailing list. Contributors are generally
3235 granted full access to the official repository after their first several
3236 patches pass successful review.
3238 This project is maintained in a git repository. The code and related tools are
3239 accessible at the following locations:
3243 =item * Official repo: L<git://git.shadowcat.co.uk/dbsrgits/SQL-Abstract.git>
3245 =item * Official gitweb: L<http://git.shadowcat.co.uk/gitweb/gitweb.cgi?p=dbsrgits/SQL-Abstract.git>
3247 =item * GitHub mirror: L<https://github.com/dbsrgits/sql-abstract>
3249 =item * Authorized committers: L<ssh://dbsrgits@git.shadowcat.co.uk/SQL-Abstract.git>
3255 Version 1.50 was a major internal refactoring of C<SQL::Abstract>.
3256 Great care has been taken to preserve the I<published> behavior
3257 documented in previous versions in the 1.* family; however,
3258 some features that were previously undocumented, or behaved
3259 differently from the documentation, had to be changed in order
3260 to clarify the semantics. Hence, client code that was relying
3261 on some dark areas of C<SQL::Abstract> v1.*
3262 B<might behave differently> in v1.50.
3264 The main changes are:
3270 support for literal SQL through the C<< \ [ $sql, @bind ] >> syntax.
3274 support for the { operator => \"..." } construct (to embed literal SQL)
3278 support for the { operator => \["...", @bind] } construct (to embed literal SQL with bind values)
3282 optional support for L<array datatypes|/"Inserting and Updating Arrays">
3286 defensive programming: check arguments
3290 fixed bug with global logic, which was previously implemented
3291 through global variables yielding side-effects. Prior versions would
3292 interpret C<< [ {cond1, cond2}, [cond3, cond4] ] >>
3293 as C<< "(cond1 AND cond2) OR (cond3 AND cond4)" >>.
3294 Now this is interpreted
3295 as C<< "(cond1 AND cond2) OR (cond3 OR cond4)" >>.
3300 fixed semantics of _bindtype on array args
3304 dropped the C<_anoncopy> of the %where tree. No longer necessary,
3305 we just avoid shifting arrays within that tree.
3309 dropped the C<_modlogic> function
3313 =head1 ACKNOWLEDGEMENTS
3315 There are a number of individuals that have really helped out with
3316 this module. Unfortunately, most of them submitted bugs via CPAN
3317 so I have no idea who they are! But the people I do know are:
3319 Ash Berlin (order_by hash term support)
3320 Matt Trout (DBIx::Class support)
3321 Mark Stosberg (benchmarking)
3322 Chas Owens (initial "IN" operator support)
3323 Philip Collins (per-field SQL functions)
3324 Eric Kolve (hashref "AND" support)
3325 Mike Fragassi (enhancements to "BETWEEN" and "LIKE")
3326 Dan Kubb (support for "quote_char" and "name_sep")
3327 Guillermo Roditi (patch to cleanup "IN" and "BETWEEN", fix and tests for _order_by)
3328 Laurent Dami (internal refactoring, extensible list of special operators, literal SQL)
3329 Norbert Buchmuller (support for literal SQL in hashpair, misc. fixes & tests)
3330 Peter Rabbitson (rewrite of SQLA::Test, misc. fixes & tests)
3331 Oliver Charles (support for "RETURNING" after "INSERT")
3337 L<DBIx::Class>, L<DBIx::Abstract>, L<CGI::FormBuilder>, L<HTML::QuickTable>.
3341 Copyright (c) 2001-2007 Nathan Wiger <nwiger@cpan.org>. All Rights Reserved.
3343 This module is actively maintained by Matt Trout <mst@shadowcatsystems.co.uk>
3345 For support, your best bet is to try the C<DBIx::Class> users mailing list.
3346 While not an official support venue, C<DBIx::Class> makes heavy use of
3347 C<SQL::Abstract>, and as such list members there are very familiar with
3348 how to create queries.
3352 This module is free software; you may copy this under the same
3353 terms as perl itself (either the GNU General Public License or
3354 the Artistic License)