1 package SQL::Abstract; # see doc at end of file
10 our @EXPORT_OK = qw(is_plain_value is_literal_value);
20 *SQL::Abstract::_ENV_::DETECT_AUTOGENERATED_STRINGIFICATION = $ENV{SQLA_ISVALUE_IGNORE_AUTOGENERATED_STRINGIFICATION}
26 #======================================================================
28 #======================================================================
30 our $VERSION = '1.84';
32 # This would confuse some packagers
33 $VERSION = eval $VERSION if $VERSION =~ /_/; # numify for warning-free dev releases
37 # special operators (-in, -between). May be extended/overridden by user.
38 # See section WHERE: BUILTIN SPECIAL OPERATORS below for implementation
39 my @BUILTIN_SPECIAL_OPS = (
40 {regex => qr/^ (?: not \s )? between $/ix, handler => '_where_field_BETWEEN'},
41 {regex => qr/^ (?: not \s )? in $/ix, handler => '_where_field_IN'},
42 {regex => qr/^ ident $/ix, handler => '_where_op_IDENT'},
43 {regex => qr/^ value $/ix, handler => '_where_op_VALUE'},
44 {regex => qr/^ is (?: \s+ not )? $/ix, handler => '_where_field_IS'},
47 # unaryish operators - key maps to handler
48 my @BUILTIN_UNARY_OPS = (
49 # the digits are backcompat stuff
50 { regex => qr/^ and (?: [_\s]? \d+ )? $/xi, handler => '_where_op_ANDOR' },
51 { regex => qr/^ or (?: [_\s]? \d+ )? $/xi, handler => '_where_op_ANDOR' },
52 { regex => qr/^ nest (?: [_\s]? \d+ )? $/xi, handler => '_where_op_NEST' },
53 { regex => qr/^ (?: not \s )? bool $/xi, handler => '_where_op_BOOL' },
54 { regex => qr/^ ident $/xi, handler => '_where_op_IDENT' },
55 { regex => qr/^ value $/xi, handler => '_where_op_VALUE' },
58 #======================================================================
59 # DEBUGGING AND ERROR REPORTING
60 #======================================================================
63 return unless $_[0]->{debug}; shift; # a little faster
64 my $func = (caller(1))[3];
65 warn "[$func] ", @_, "\n";
69 my($func) = (caller(1))[3];
70 Carp::carp "[$func] Warning: ", @_;
74 my($func) = (caller(1))[3];
75 Carp::croak "[$func] Fatal: ", @_;
78 sub is_literal_value ($) {
79 ref $_[0] eq 'SCALAR' ? [ ${$_[0]} ]
80 : ( ref $_[0] eq 'REF' and ref ${$_[0]} eq 'ARRAY' ) ? [ @${ $_[0] } ]
84 # FIXME XSify - this can be done so much more efficiently
85 sub is_plain_value ($) {
87 ! length ref $_[0] ? \($_[0])
89 ref $_[0] eq 'HASH' and keys %{$_[0]} == 1
91 exists $_[0]->{-value}
92 ) ? \($_[0]->{-value})
94 # reuse @_ for even moar speedz
95 defined ( $_[1] = Scalar::Util::blessed $_[0] )
97 # deliberately not using Devel::OverloadInfo - the checks we are
98 # intersted in are much more limited than the fullblown thing, and
99 # this is a very hot piece of code
101 # simply using ->can('(""') can leave behind stub methods that
102 # break actually using the overload later (see L<perldiag/Stub
103 # found while resolving method "%s" overloading "%s" in package
104 # "%s"> and the source of overload::mycan())
106 # either has stringification which DBI SHOULD prefer out of the box
107 grep { *{ (qq[${_}::(""]) }{CODE} } @{ $_[2] = mro::get_linear_isa( $_[1] ) }
109 # has nummification or boolification, AND fallback is *not* disabled
111 SQL::Abstract::_ENV_::DETECT_AUTOGENERATED_STRINGIFICATION
114 grep { *{"${_}::(0+"}{CODE} } @{$_[2]}
116 grep { *{"${_}::(bool"}{CODE} } @{$_[2]}
120 # no fallback specified at all
121 ! ( ($_[3]) = grep { *{"${_}::()"}{CODE} } @{$_[2]} )
123 # fallback explicitly undef
124 ! defined ${"$_[3]::()"}
137 #======================================================================
139 #======================================================================
143 my $class = ref($self) || $self;
144 my %opt = (ref $_[0] eq 'HASH') ? %{$_[0]} : @_;
146 # choose our case by keeping an option around
147 delete $opt{case} if $opt{case} && $opt{case} ne 'lower';
149 # default logic for interpreting arrayrefs
150 $opt{logic} = $opt{logic} ? uc $opt{logic} : 'OR';
152 # how to return bind vars
153 $opt{bindtype} ||= 'normal';
155 # default comparison is "=", but can be overridden
158 # try to recognize which are the 'equality' and 'inequality' ops
159 # (temporary quickfix (in 2007), should go through a more seasoned API)
160 $opt{equality_op} = qr/^( \Q$opt{cmp}\E | \= )$/ix;
161 $opt{inequality_op} = qr/^( != | <> )$/ix;
163 $opt{like_op} = qr/^ (is\s+)? r?like $/xi;
164 $opt{not_like_op} = qr/^ (is\s+)? not \s+ r?like $/xi;
167 $opt{sqltrue} ||= '1=1';
168 $opt{sqlfalse} ||= '0=1';
171 $opt{special_ops} ||= [];
172 # regexes are applied in order, thus push after user-defines
173 push @{$opt{special_ops}}, @BUILTIN_SPECIAL_OPS;
176 $opt{unary_ops} ||= [];
177 push @{$opt{unary_ops}}, @BUILTIN_UNARY_OPS;
179 # rudimentary sanity-check for user supplied bits treated as functions/operators
180 # If a purported function matches this regular expression, an exception is thrown.
181 # Literal SQL is *NOT* subject to this check, only functions (and column names
182 # when quoting is not in effect)
185 # need to guard against ()'s in column names too, but this will break tons of
186 # hacks... ideas anyone?
187 $opt{injection_guard} ||= qr/
193 return bless \%opt, $class;
197 sub _assert_pass_injection_guard {
198 if ($_[1] =~ $_[0]->{injection_guard}) {
199 my $class = ref $_[0];
200 puke "Possible SQL injection attempt '$_[1]'. If this is indeed a part of the "
201 . "desired SQL use literal SQL ( \'...' or \[ '...' ] ) or supply your own "
202 . "{injection_guard} attribute to ${class}->new()"
207 #======================================================================
209 #======================================================================
213 my $table = $self->_table(shift);
214 my $data = shift || return;
217 my $method = $self->_METHOD_FOR_refkind("_insert", $data);
218 my ($sql, @bind) = $self->$method($data);
219 $sql = join " ", $self->_sqlcase('insert into'), $table, $sql;
221 if ($options->{returning}) {
222 my ($s, @b) = $self->_insert_returning($options);
227 return wantarray ? ($sql, @bind) : $sql;
230 # So that subclasses can override INSERT ... RETURNING separately from
231 # UPDATE and DELETE (e.g. DBIx::Class::SQLMaker::Oracle does this)
232 sub _insert_returning { shift->_returning(@_) }
235 my ($self, $options) = @_;
237 my $f = $options->{returning};
239 my $fieldlist = $self->_SWITCH_refkind($f, {
240 ARRAYREF => sub {join ', ', map { $self->_quote($_) } @$f;},
241 SCALAR => sub {$self->_quote($f)},
242 SCALARREF => sub {$$f},
244 return $self->_sqlcase(' returning ') . $fieldlist;
247 sub _insert_HASHREF { # explicit list of fields and then values
248 my ($self, $data) = @_;
250 my @fields = sort keys %$data;
252 my ($sql, @bind) = $self->_insert_values($data);
255 $_ = $self->_quote($_) foreach @fields;
256 $sql = "( ".join(", ", @fields).") ".$sql;
258 return ($sql, @bind);
261 sub _insert_ARRAYREF { # just generate values(?,?) part (no list of fields)
262 my ($self, $data) = @_;
264 # no names (arrayref) so can't generate bindtype
265 $self->{bindtype} ne 'columns'
266 or belch "can't do 'columns' bindtype when called with arrayref";
268 my (@values, @all_bind);
269 foreach my $value (@$data) {
270 my ($values, @bind) = $self->_insert_value(undef, $value);
271 push @values, $values;
272 push @all_bind, @bind;
274 my $sql = $self->_sqlcase('values')." ( ".join(", ", @values)." )";
275 return ($sql, @all_bind);
278 sub _insert_ARRAYREFREF { # literal SQL with bind
279 my ($self, $data) = @_;
281 my ($sql, @bind) = @${$data};
282 $self->_assert_bindval_matches_bindtype(@bind);
284 return ($sql, @bind);
288 sub _insert_SCALARREF { # literal SQL without bind
289 my ($self, $data) = @_;
295 my ($self, $data) = @_;
297 my (@values, @all_bind);
298 foreach my $column (sort keys %$data) {
299 my ($values, @bind) = $self->_insert_value($column, $data->{$column});
300 push @values, $values;
301 push @all_bind, @bind;
303 my $sql = $self->_sqlcase('values')." ( ".join(", ", @values)." )";
304 return ($sql, @all_bind);
308 my ($self, $column, $v) = @_;
310 my (@values, @all_bind);
311 $self->_SWITCH_refkind($v, {
314 if ($self->{array_datatypes}) { # if array datatype are activated
316 push @all_bind, $self->_bindtype($column, $v);
318 else { # else literal SQL with bind
319 my ($sql, @bind) = @$v;
320 $self->_assert_bindval_matches_bindtype(@bind);
322 push @all_bind, @bind;
326 ARRAYREFREF => sub { # literal SQL with bind
327 my ($sql, @bind) = @${$v};
328 $self->_assert_bindval_matches_bindtype(@bind);
330 push @all_bind, @bind;
333 # THINK : anything useful to do with a HASHREF ?
334 HASHREF => sub { # (nothing, but old SQLA passed it through)
335 #TODO in SQLA >= 2.0 it will die instead
336 belch "HASH ref as bind value in insert is not supported";
338 push @all_bind, $self->_bindtype($column, $v);
341 SCALARREF => sub { # literal SQL without bind
345 SCALAR_or_UNDEF => sub {
347 push @all_bind, $self->_bindtype($column, $v);
352 my $sql = join(", ", @values);
353 return ($sql, @all_bind);
358 #======================================================================
360 #======================================================================
365 my $table = $self->_table(shift);
366 my $data = shift || return;
370 # first build the 'SET' part of the sql statement
371 my (@set, @all_bind);
372 puke "Unsupported data type specified to \$sql->update"
373 unless ref $data eq 'HASH';
375 for my $k (sort keys %$data) {
378 my $label = $self->_quote($k);
380 $self->_SWITCH_refkind($v, {
382 if ($self->{array_datatypes}) { # array datatype
383 push @set, "$label = ?";
384 push @all_bind, $self->_bindtype($k, $v);
386 else { # literal SQL with bind
387 my ($sql, @bind) = @$v;
388 $self->_assert_bindval_matches_bindtype(@bind);
389 push @set, "$label = $sql";
390 push @all_bind, @bind;
393 ARRAYREFREF => sub { # literal SQL with bind
394 my ($sql, @bind) = @${$v};
395 $self->_assert_bindval_matches_bindtype(@bind);
396 push @set, "$label = $sql";
397 push @all_bind, @bind;
399 SCALARREF => sub { # literal SQL without bind
400 push @set, "$label = $$v";
403 my ($op, $arg, @rest) = %$v;
405 puke 'Operator calls in update must be in the form { -op => $arg }'
406 if (@rest or not $op =~ /^\-(.+)/);
408 local $self->{_nested_func_lhs} = $k;
409 my ($sql, @bind) = $self->_where_unary_op($1, $arg);
411 push @set, "$label = $sql";
412 push @all_bind, @bind;
414 SCALAR_or_UNDEF => sub {
415 push @set, "$label = ?";
416 push @all_bind, $self->_bindtype($k, $v);
422 my $sql = $self->_sqlcase('update') . " $table " . $self->_sqlcase('set ')
426 my($where_sql, @where_bind) = $self->where($where);
428 push @all_bind, @where_bind;
431 if ($options->{returning}) {
432 my ($returning_sql, @returning_bind) = $self->_update_returning($options);
433 $sql .= $returning_sql;
434 push @all_bind, @returning_bind;
437 return wantarray ? ($sql, @all_bind) : $sql;
440 # So that subclasses can override UPDATE ... RETURNING separately from
442 sub _update_returning { shift->_returning(@_) }
446 #======================================================================
448 #======================================================================
453 my $table = $self->_table(shift);
454 my $fields = shift || '*';
458 my($where_sql, @bind) = $self->where($where, $order);
460 my $f = (ref $fields eq 'ARRAY') ? join ', ', map { $self->_quote($_) } @$fields
462 my $sql = join(' ', $self->_sqlcase('select'), $f,
463 $self->_sqlcase('from'), $table)
466 return wantarray ? ($sql, @bind) : $sql;
469 #======================================================================
471 #======================================================================
476 my $table = $self->_table(shift);
480 my($where_sql, @bind) = $self->where($where);
481 my $sql = $self->_sqlcase('delete from') . " $table" . $where_sql;
483 if ($options->{returning}) {
484 my ($returning_sql, @returning_bind) = $self->_delete_returning($options);
485 $sql .= $returning_sql;
486 push @bind, @returning_bind;
489 return wantarray ? ($sql, @bind) : $sql;
492 # So that subclasses can override DELETE ... RETURNING separately from
494 sub _delete_returning { shift->_returning(@_) }
498 #======================================================================
500 #======================================================================
504 # Finally, a separate routine just to handle WHERE clauses
506 my ($self, $where, $order) = @_;
509 my ($sql, @bind) = $self->_recurse_where($where);
510 $sql = $sql ? $self->_sqlcase(' where ') . "( $sql )" : '';
514 my ($order_sql, @order_bind) = $self->_order_by($order);
516 push @bind, @order_bind;
519 return wantarray ? ($sql, @bind) : $sql;
524 my ($self, $where, $logic) = @_;
526 # dispatch on appropriate method according to refkind of $where
527 my $method = $self->_METHOD_FOR_refkind("_where", $where);
529 my ($sql, @bind) = $self->$method($where, $logic);
531 # DBIx::Class used to call _recurse_where in scalar context
532 # something else might too...
534 return ($sql, @bind);
537 belch "Calling _recurse_where in scalar context is deprecated and will go away before 2.0";
544 #======================================================================
545 # WHERE: top-level ARRAYREF
546 #======================================================================
549 sub _where_ARRAYREF {
550 my ($self, $where, $logic) = @_;
552 $logic = uc($logic || $self->{logic});
553 $logic eq 'AND' or $logic eq 'OR' or puke "unknown logic: $logic";
555 my @clauses = @$where;
557 my (@sql_clauses, @all_bind);
558 # need to use while() so can shift() for pairs
560 my $el = shift @clauses;
562 $el = undef if (defined $el and ! length $el);
564 # switch according to kind of $el and get corresponding ($sql, @bind)
565 my ($sql, @bind) = $self->_SWITCH_refkind($el, {
567 # skip empty elements, otherwise get invalid trailing AND stuff
568 ARRAYREF => sub {$self->_recurse_where($el) if @$el},
572 $self->_assert_bindval_matches_bindtype(@b);
576 HASHREF => sub {$self->_recurse_where($el, 'and') if %$el},
578 SCALARREF => sub { ($$el); },
581 # top-level arrayref with scalars, recurse in pairs
582 $self->_recurse_where({$el => shift(@clauses)})
585 UNDEF => sub {puke "Supplying an empty left hand side argument is not supported in array-pairs" },
589 push @sql_clauses, $sql;
590 push @all_bind, @bind;
594 return $self->_join_sql_clauses($logic, \@sql_clauses, \@all_bind);
597 #======================================================================
598 # WHERE: top-level ARRAYREFREF
599 #======================================================================
601 sub _where_ARRAYREFREF {
602 my ($self, $where) = @_;
603 my ($sql, @bind) = @$$where;
604 $self->_assert_bindval_matches_bindtype(@bind);
605 return ($sql, @bind);
608 #======================================================================
609 # WHERE: top-level HASHREF
610 #======================================================================
613 my ($self, $where) = @_;
614 my (@sql_clauses, @all_bind);
616 for my $k (sort keys %$where) {
617 my $v = $where->{$k};
619 # ($k => $v) is either a special unary op or a regular hashpair
620 my ($sql, @bind) = do {
622 # put the operator in canonical form
624 $op = substr $op, 1; # remove initial dash
625 $op =~ s/^\s+|\s+$//g;# remove leading/trailing space
626 $op =~ s/\s+/ /g; # compress whitespace
628 # so that -not_foo works correctly
629 $op =~ s/^not_/NOT /i;
631 $self->_debug("Unary OP(-$op) within hashref, recursing...");
632 my ($s, @b) = $self->_where_unary_op($op, $v);
634 # top level vs nested
635 # we assume that handled unary ops will take care of their ()s
637 List::Util::first {$op =~ $_->{regex}} @{$self->{unary_ops}}
639 ( defined $self->{_nested_func_lhs} and $self->{_nested_func_lhs} eq $k )
645 if (is_literal_value ($v) ) {
646 belch 'Hash-pairs consisting of an empty string with a literal are deprecated, and will be removed in 2.0: use -and => [ $literal ] instead';
649 puke "Supplying an empty left hand side argument is not supported in hash-pairs";
653 my $method = $self->_METHOD_FOR_refkind("_where_hashpair", $v);
654 $self->$method($k, $v);
658 push @sql_clauses, $sql;
659 push @all_bind, @bind;
662 return $self->_join_sql_clauses('and', \@sql_clauses, \@all_bind);
665 sub _where_unary_op {
666 my ($self, $op, $rhs) = @_;
668 # top level special ops are illegal in general
669 # this includes the -ident/-value ops (dual purpose unary and special)
670 puke "Illegal use of top-level '-$op'"
671 if ! defined $self->{_nested_func_lhs} and List::Util::first { $op =~ $_->{regex} } @{$self->{special_ops}};
673 if (my $op_entry = List::Util::first { $op =~ $_->{regex} } @{$self->{unary_ops}}) {
674 my $handler = $op_entry->{handler};
676 if (not ref $handler) {
677 if ($op =~ s/ [_\s]? \d+ $//x ) {
678 belch 'Use of [and|or|nest]_N modifiers is deprecated and will be removed in SQLA v2.0. '
679 . "You probably wanted ...-and => [ -$op => COND1, -$op => COND2 ... ]";
681 return $self->$handler($op, $rhs);
683 elsif (ref $handler eq 'CODE') {
684 return $handler->($self, $op, $rhs);
687 puke "Illegal handler for operator $op - expecting a method name or a coderef";
691 $self->_debug("Generic unary OP: $op - recursing as function");
693 $self->_assert_pass_injection_guard($op);
695 my ($sql, @bind) = $self->_SWITCH_refkind($rhs, {
697 puke "Illegal use of top-level '-$op'"
698 unless defined $self->{_nested_func_lhs};
701 $self->_convert('?'),
702 $self->_bindtype($self->{_nested_func_lhs}, $rhs)
706 $self->_recurse_where($rhs)
710 $sql = sprintf('%s %s',
711 $self->_sqlcase($op),
715 return ($sql, @bind);
718 sub _where_op_ANDOR {
719 my ($self, $op, $v) = @_;
721 $self->_SWITCH_refkind($v, {
723 return $self->_where_ARRAYREF($v, $op);
727 return ($op =~ /^or/i)
728 ? $self->_where_ARRAYREF([ map { $_ => $v->{$_} } (sort keys %$v) ], $op)
729 : $self->_where_HASHREF($v);
733 puke "-$op => \\\$scalar makes little sense, use " .
735 ? '[ \$scalar, \%rest_of_conditions ] instead'
736 : '-and => [ \$scalar, \%rest_of_conditions ] instead'
741 puke "-$op => \\[...] makes little sense, use " .
743 ? '[ \[...], \%rest_of_conditions ] instead'
744 : '-and => [ \[...], \%rest_of_conditions ] instead'
748 SCALAR => sub { # permissively interpreted as SQL
749 puke "-$op => \$value makes little sense, use -bool => \$value instead";
753 puke "-$op => undef not supported";
759 my ($self, $op, $v) = @_;
761 $self->_SWITCH_refkind($v, {
763 SCALAR => sub { # permissively interpreted as SQL
764 belch "literal SQL should be -nest => \\'scalar' "
765 . "instead of -nest => 'scalar' ";
770 puke "-$op => undef not supported";
774 $self->_recurse_where($v);
782 my ($self, $op, $v) = @_;
784 my ($s, @b) = $self->_SWITCH_refkind($v, {
785 SCALAR => sub { # interpreted as SQL column
786 $self->_convert($self->_quote($v));
790 puke "-$op => undef not supported";
794 $self->_recurse_where($v);
798 $s = "(NOT $s)" if $op =~ /^not/i;
803 sub _where_op_IDENT {
805 my ($op, $rhs) = splice @_, -2;
806 if (! defined $rhs or length ref $rhs) {
807 puke "-$op requires a single plain scalar argument (a quotable identifier)";
810 # in case we are called as a top level special op (no '=')
813 $_ = $self->_convert($self->_quote($_)) for ($lhs, $rhs);
821 sub _where_op_VALUE {
823 my ($op, $rhs) = splice @_, -2;
825 # in case we are called as a top level special op (no '=')
829 if (! defined $rhs) {
831 ? $self->_convert($self->_quote($lhs)) . ' IS NULL'
838 (defined $lhs ? $lhs : $self->{_nested_func_lhs}),
845 $self->_convert($self->_quote($lhs)) . ' = ' . $self->_convert('?'),
849 $self->_convert('?'),
855 sub _where_hashpair_ARRAYREF {
856 my ($self, $k, $v) = @_;
859 my @v = @$v; # need copy because of shift below
860 $self->_debug("ARRAY($k) means distribute over elements");
862 # put apart first element if it is an operator (-and, -or)
864 (defined $v[0] && $v[0] =~ /^ - (?: AND|OR ) $/ix)
868 my @distributed = map { {$k => $_} } @v;
871 $self->_debug("OP($op) reinjected into the distributed array");
872 unshift @distributed, $op;
875 my $logic = $op ? substr($op, 1) : '';
877 return $self->_recurse_where(\@distributed, $logic);
880 $self->_debug("empty ARRAY($k) means 0=1");
881 return ($self->{sqlfalse});
885 sub _where_hashpair_HASHREF {
886 my ($self, $k, $v, $logic) = @_;
889 local $self->{_nested_func_lhs} = defined $self->{_nested_func_lhs}
890 ? $self->{_nested_func_lhs}
894 my ($all_sql, @all_bind);
896 for my $orig_op (sort keys %$v) {
897 my $val = $v->{$orig_op};
899 # put the operator in canonical form
902 # FIXME - we need to phase out dash-less ops
903 $op =~ s/^-//; # remove possible initial dash
904 $op =~ s/^\s+|\s+$//g;# remove leading/trailing space
905 $op =~ s/\s+/ /g; # compress whitespace
907 $self->_assert_pass_injection_guard($op);
910 $op =~ s/^is_not/IS NOT/i;
912 # so that -not_foo works correctly
913 $op =~ s/^not_/NOT /i;
915 # another retarded special case: foo => { $op => { -value => undef } }
916 if (ref $val eq 'HASH' and keys %$val == 1 and exists $val->{-value} and ! defined $val->{-value} ) {
922 # CASE: col-value logic modifiers
923 if ($orig_op =~ /^ \- (and|or) $/xi) {
924 ($sql, @bind) = $self->_where_hashpair_HASHREF($k, $val, $1);
926 # CASE: special operators like -in or -between
927 elsif (my $special_op = List::Util::first { $op =~ $_->{regex} } @{$self->{special_ops}}) {
928 my $handler = $special_op->{handler};
930 puke "No handler supplied for special operator $orig_op";
932 elsif (not ref $handler) {
933 ($sql, @bind) = $self->$handler($k, $op, $val);
935 elsif (ref $handler eq 'CODE') {
936 ($sql, @bind) = $handler->($self, $k, $op, $val);
939 puke "Illegal handler for special operator $orig_op - expecting a method name or a coderef";
943 $self->_SWITCH_refkind($val, {
945 ARRAYREF => sub { # CASE: col => {op => \@vals}
946 ($sql, @bind) = $self->_where_field_op_ARRAYREF($k, $op, $val);
949 ARRAYREFREF => sub { # CASE: col => {op => \[$sql, @bind]} (literal SQL with bind)
950 my ($sub_sql, @sub_bind) = @$$val;
951 $self->_assert_bindval_matches_bindtype(@sub_bind);
952 $sql = join ' ', $self->_convert($self->_quote($k)),
953 $self->_sqlcase($op),
958 UNDEF => sub { # CASE: col => {op => undef} : sql "IS (NOT)? NULL"
960 $op =~ /^not$/i ? 'is not' # legacy
961 : $op =~ $self->{equality_op} ? 'is'
962 : $op =~ $self->{like_op} ? belch("Supplying an undefined argument to '@{[ uc $op]}' is deprecated") && 'is'
963 : $op =~ $self->{inequality_op} ? 'is not'
964 : $op =~ $self->{not_like_op} ? belch("Supplying an undefined argument to '@{[ uc $op]}' is deprecated") && 'is not'
965 : puke "unexpected operator '$orig_op' with undef operand";
967 $sql = $self->_quote($k) . $self->_sqlcase(" $is null");
970 FALLBACK => sub { # CASE: col => {op/func => $stuff}
971 ($sql, @bind) = $self->_where_unary_op($op, $val);
974 $self->_convert($self->_quote($k)),
975 $self->{_nested_func_lhs} eq $k ? $sql : "($sql)", # top level vs nested
981 ($all_sql) = (defined $all_sql and $all_sql) ? $self->_join_sql_clauses($logic, [$all_sql, $sql], []) : $sql;
982 push @all_bind, @bind;
984 return ($all_sql, @all_bind);
987 sub _where_field_IS {
988 my ($self, $k, $op, $v) = @_;
990 my ($s) = $self->_SWITCH_refkind($v, {
993 $self->_convert($self->_quote($k)),
994 map { $self->_sqlcase($_)} ($op, 'null')
997 puke "$op can only take undef as argument";
1004 sub _where_field_op_ARRAYREF {
1005 my ($self, $k, $op, $vals) = @_;
1007 my @vals = @$vals; #always work on a copy
1010 $self->_debug(sprintf '%s means multiple elements: [ %s ]',
1012 join(', ', map { defined $_ ? "'$_'" : 'NULL' } @vals ),
1015 # see if the first element is an -and/-or op
1017 if (defined $vals[0] && $vals[0] =~ /^ - (AND|OR) $/ix) {
1022 # a long standing API wart - an attempt to change this behavior during
1023 # the 1.50 series failed *spectacularly*. Warn instead and leave the
1028 (!$logic or $logic eq 'OR')
1030 ($op =~ $self->{inequality_op} or $op =~ $self->{not_like_op})
1033 belch "A multi-element arrayref as an argument to the inequality op '$o' "
1034 . 'is technically equivalent to an always-true 1=1 (you probably wanted '
1035 . "to say ...{ \$inequality_op => [ -and => \@values ] }... instead)"
1039 # distribute $op over each remaining member of @vals, append logic if exists
1040 return $self->_recurse_where([map { {$k => {$op, $_}} } @vals], $logic);
1044 # try to DWIM on equality operators
1046 $op =~ $self->{equality_op} ? $self->{sqlfalse}
1047 : $op =~ $self->{like_op} ? belch("Supplying an empty arrayref to '@{[ uc $op]}' is deprecated") && $self->{sqlfalse}
1048 : $op =~ $self->{inequality_op} ? $self->{sqltrue}
1049 : $op =~ $self->{not_like_op} ? belch("Supplying an empty arrayref to '@{[ uc $op]}' is deprecated") && $self->{sqltrue}
1050 : puke "operator '$op' applied on an empty array (field '$k')";
1055 sub _where_hashpair_SCALARREF {
1056 my ($self, $k, $v) = @_;
1057 $self->_debug("SCALAR($k) means literal SQL: $$v");
1058 my $sql = $self->_quote($k) . " " . $$v;
1062 # literal SQL with bind
1063 sub _where_hashpair_ARRAYREFREF {
1064 my ($self, $k, $v) = @_;
1065 $self->_debug("REF($k) means literal SQL: @${$v}");
1066 my ($sql, @bind) = @$$v;
1067 $self->_assert_bindval_matches_bindtype(@bind);
1068 $sql = $self->_quote($k) . " " . $sql;
1069 return ($sql, @bind );
1072 # literal SQL without bind
1073 sub _where_hashpair_SCALAR {
1074 my ($self, $k, $v) = @_;
1075 $self->_debug("NOREF($k) means simple key=val: $k $self->{cmp} $v");
1076 my $sql = join ' ', $self->_convert($self->_quote($k)),
1077 $self->_sqlcase($self->{cmp}),
1078 $self->_convert('?');
1079 my @bind = $self->_bindtype($k, $v);
1080 return ($sql, @bind);
1084 sub _where_hashpair_UNDEF {
1085 my ($self, $k, $v) = @_;
1086 $self->_debug("UNDEF($k) means IS NULL");
1087 my $sql = $self->_quote($k) . $self->_sqlcase(' is null');
1091 #======================================================================
1092 # WHERE: TOP-LEVEL OTHERS (SCALARREF, SCALAR, UNDEF)
1093 #======================================================================
1096 sub _where_SCALARREF {
1097 my ($self, $where) = @_;
1100 $self->_debug("SCALAR(*top) means literal SQL: $$where");
1106 my ($self, $where) = @_;
1109 $self->_debug("NOREF(*top) means literal SQL: $where");
1120 #======================================================================
1121 # WHERE: BUILTIN SPECIAL OPERATORS (-in, -between)
1122 #======================================================================
1125 sub _where_field_BETWEEN {
1126 my ($self, $k, $op, $vals) = @_;
1128 my ($label, $and, $placeholder);
1129 $label = $self->_convert($self->_quote($k));
1130 $and = ' ' . $self->_sqlcase('and') . ' ';
1131 $placeholder = $self->_convert('?');
1132 $op = $self->_sqlcase($op);
1134 my $invalid_args = "Operator '$op' requires either an arrayref with two defined values or expressions, or a single literal scalarref/arrayref-ref";
1136 my ($clause, @bind) = $self->_SWITCH_refkind($vals, {
1137 ARRAYREFREF => sub {
1138 my ($s, @b) = @$$vals;
1139 $self->_assert_bindval_matches_bindtype(@b);
1146 puke $invalid_args if @$vals != 2;
1148 my (@all_sql, @all_bind);
1149 foreach my $val (@$vals) {
1150 my ($sql, @bind) = $self->_SWITCH_refkind($val, {
1152 return ($placeholder, $self->_bindtype($k, $val) );
1157 ARRAYREFREF => sub {
1158 my ($sql, @bind) = @$$val;
1159 $self->_assert_bindval_matches_bindtype(@bind);
1160 return ($sql, @bind);
1163 my ($func, $arg, @rest) = %$val;
1164 puke "Only simple { -func => arg } functions accepted as sub-arguments to BETWEEN"
1165 if (@rest or $func !~ /^ \- (.+)/x);
1166 $self->_where_unary_op($1 => $arg);
1172 push @all_sql, $sql;
1173 push @all_bind, @bind;
1177 (join $and, @all_sql),
1186 my $sql = "( $label $op $clause )";
1187 return ($sql, @bind)
1191 sub _where_field_IN {
1192 my ($self, $k, $op, $vals) = @_;
1194 # backwards compatibility : if scalar, force into an arrayref
1195 $vals = [$vals] if defined $vals && ! ref $vals;
1197 my ($label) = $self->_convert($self->_quote($k));
1198 my ($placeholder) = $self->_convert('?');
1199 $op = $self->_sqlcase($op);
1201 my ($sql, @bind) = $self->_SWITCH_refkind($vals, {
1202 ARRAYREF => sub { # list of choices
1203 if (@$vals) { # nonempty list
1204 my (@all_sql, @all_bind);
1206 for my $val (@$vals) {
1207 my ($sql, @bind) = $self->_SWITCH_refkind($val, {
1209 return ($placeholder, $val);
1214 ARRAYREFREF => sub {
1215 my ($sql, @bind) = @$$val;
1216 $self->_assert_bindval_matches_bindtype(@bind);
1217 return ($sql, @bind);
1220 my ($func, $arg, @rest) = %$val;
1221 puke "Only simple { -func => arg } functions accepted as sub-arguments to IN"
1222 if (@rest or $func !~ /^ \- (.+)/x);
1223 $self->_where_unary_op($1 => $arg);
1227 'SQL::Abstract before v1.75 used to generate incorrect SQL when the '
1228 . "-$op operator was given an undef-containing list: !!!AUDIT YOUR CODE "
1229 . 'AND DATA!!! (the upcoming Data::Query-based version of SQL::Abstract '
1230 . 'will emit the logically correct SQL instead of raising this exception)'
1234 push @all_sql, $sql;
1235 push @all_bind, @bind;
1239 sprintf('%s %s ( %s )',
1242 join(', ', @all_sql)
1244 $self->_bindtype($k, @all_bind),
1247 else { # empty list : some databases won't understand "IN ()", so DWIM
1248 my $sql = ($op =~ /\bnot\b/i) ? $self->{sqltrue} : $self->{sqlfalse};
1253 SCALARREF => sub { # literal SQL
1254 my $sql = $self->_open_outer_paren($$vals);
1255 return ("$label $op ( $sql )");
1257 ARRAYREFREF => sub { # literal SQL with bind
1258 my ($sql, @bind) = @$$vals;
1259 $self->_assert_bindval_matches_bindtype(@bind);
1260 $sql = $self->_open_outer_paren($sql);
1261 return ("$label $op ( $sql )", @bind);
1265 puke "Argument passed to the '$op' operator can not be undefined";
1269 puke "special op $op requires an arrayref (or scalarref/arrayref-ref)";
1273 return ($sql, @bind);
1276 # Some databases (SQLite) treat col IN (1, 2) different from
1277 # col IN ( (1, 2) ). Use this to strip all outer parens while
1278 # adding them back in the corresponding method
1279 sub _open_outer_paren {
1280 my ($self, $sql) = @_;
1282 while (my ($inner) = $sql =~ /^ \s* \( (.*) \) \s* $/xs) {
1284 # there are closing parens inside, need the heavy duty machinery
1285 # to reevaluate the extraction starting from $sql (full reevaluation)
1286 if ($inner =~ /\)/) {
1287 require Text::Balanced;
1289 my (undef, $remainder) = do {
1290 # idiotic design - writes to $@ but *DOES NOT* throw exceptions
1292 Text::Balanced::extract_bracketed($sql, '()', qr/\s*/);
1295 # the entire expression needs to be a balanced bracketed thing
1296 # (after an extract no remainder sans trailing space)
1297 last if defined $remainder and $remainder =~ /\S/;
1307 #======================================================================
1309 #======================================================================
1312 my ($self, $arg) = @_;
1315 for my $c ($self->_order_by_chunks($arg) ) {
1316 $self->_SWITCH_refkind($c, {
1317 SCALAR => sub { push @sql, $c },
1318 ARRAYREF => sub { push @sql, shift @$c; push @bind, @$c },
1324 $self->_sqlcase(' order by'),
1330 return wantarray ? ($sql, @bind) : $sql;
1333 sub _order_by_chunks {
1334 my ($self, $arg) = @_;
1336 return $self->_SWITCH_refkind($arg, {
1339 map { $self->_order_by_chunks($_ ) } @$arg;
1342 ARRAYREFREF => sub {
1343 my ($s, @b) = @$$arg;
1344 $self->_assert_bindval_matches_bindtype(@b);
1348 SCALAR => sub {$self->_quote($arg)},
1350 UNDEF => sub {return () },
1352 SCALARREF => sub {$$arg}, # literal SQL, no quoting
1355 # get first pair in hash
1356 my ($key, $val, @rest) = %$arg;
1358 return () unless $key;
1360 if (@rest or not $key =~ /^-(desc|asc)/i) {
1361 puke "hash passed to _order_by must have exactly one key (-desc or -asc)";
1367 for my $c ($self->_order_by_chunks($val)) {
1370 $self->_SWITCH_refkind($c, {
1375 ($sql, @bind) = @$c;
1379 $sql = $sql . ' ' . $self->_sqlcase($direction);
1381 push @ret, [ $sql, @bind];
1390 #======================================================================
1391 # DATASOURCE (FOR NOW, JUST PLAIN TABLE OR LIST OF TABLES)
1392 #======================================================================
1397 $self->_SWITCH_refkind($from, {
1398 ARRAYREF => sub {join ', ', map { $self->_quote($_) } @$from;},
1399 SCALAR => sub {$self->_quote($from)},
1400 SCALARREF => sub {$$from},
1405 #======================================================================
1407 #======================================================================
1409 # highly optimized, as it's called way too often
1411 # my ($self, $label) = @_;
1413 return '' unless defined $_[1];
1414 return ${$_[1]} if ref($_[1]) eq 'SCALAR';
1416 $_[0]->{quote_char} or
1417 ($_[0]->_assert_pass_injection_guard($_[1]), return $_[1]);
1419 my $qref = ref $_[0]->{quote_char};
1421 !$qref ? ($_[0]->{quote_char}, $_[0]->{quote_char})
1422 : ($qref eq 'ARRAY') ? @{$_[0]->{quote_char}}
1423 : puke "Unsupported quote_char format: $_[0]->{quote_char}";
1425 my $esc = $_[0]->{escape_char} || $r;
1427 # parts containing * are naturally unquoted
1428 return join($_[0]->{name_sep}||'', map
1429 +( $_ eq '*' ? $_ : do { (my $n = $_) =~ s/(\Q$esc\E|\Q$r\E)/$esc$1/g; $l . $n . $r } ),
1430 ( $_[0]->{name_sep} ? split (/\Q$_[0]->{name_sep}\E/, $_[1] ) : $_[1] )
1435 # Conversion, if applicable
1437 #my ($self, $arg) = @_;
1438 if ($_[0]->{convert}) {
1439 return $_[0]->_sqlcase($_[0]->{convert}) .'(' . $_[1] . ')';
1446 #my ($self, $col, @vals) = @_;
1447 # called often - tighten code
1448 return $_[0]->{bindtype} eq 'columns'
1449 ? map {[$_[1], $_]} @_[2 .. $#_]
1454 # Dies if any element of @bind is not in [colname => value] format
1455 # if bindtype is 'columns'.
1456 sub _assert_bindval_matches_bindtype {
1457 # my ($self, @bind) = @_;
1459 if ($self->{bindtype} eq 'columns') {
1461 if (!defined $_ || ref($_) ne 'ARRAY' || @$_ != 2) {
1462 puke "bindtype 'columns' selected, you need to pass: [column_name => bind_value]"
1468 sub _join_sql_clauses {
1469 my ($self, $logic, $clauses_aref, $bind_aref) = @_;
1471 if (@$clauses_aref > 1) {
1472 my $join = " " . $self->_sqlcase($logic) . " ";
1473 my $sql = '( ' . join($join, @$clauses_aref) . ' )';
1474 return ($sql, @$bind_aref);
1476 elsif (@$clauses_aref) {
1477 return ($clauses_aref->[0], @$bind_aref); # no parentheses
1480 return (); # if no SQL, ignore @$bind_aref
1485 # Fix SQL case, if so requested
1487 # LDNOTE: if $self->{case} is true, then it contains 'lower', so we
1488 # don't touch the argument ... crooked logic, but let's not change it!
1489 return $_[0]->{case} ? $_[1] : uc($_[1]);
1493 #======================================================================
1494 # DISPATCHING FROM REFKIND
1495 #======================================================================
1498 my ($self, $data) = @_;
1500 return 'UNDEF' unless defined $data;
1502 # blessed objects are treated like scalars
1503 my $ref = (Scalar::Util::blessed $data) ? '' : ref $data;
1505 return 'SCALAR' unless $ref;
1508 while ($ref eq 'REF') {
1510 $ref = (Scalar::Util::blessed $data) ? '' : ref $data;
1514 return ($ref||'SCALAR') . ('REF' x $n_steps);
1518 my ($self, $data) = @_;
1519 my @try = ($self->_refkind($data));
1520 push @try, 'SCALAR_or_UNDEF' if $try[0] eq 'SCALAR' || $try[0] eq 'UNDEF';
1521 push @try, 'FALLBACK';
1525 sub _METHOD_FOR_refkind {
1526 my ($self, $meth_prefix, $data) = @_;
1529 for (@{$self->_try_refkind($data)}) {
1530 $method = $self->can($meth_prefix."_".$_)
1534 return $method || puke "cannot dispatch on '$meth_prefix' for ".$self->_refkind($data);
1538 sub _SWITCH_refkind {
1539 my ($self, $data, $dispatch_table) = @_;
1542 for (@{$self->_try_refkind($data)}) {
1543 $coderef = $dispatch_table->{$_}
1547 puke "no dispatch entry for ".$self->_refkind($data)
1556 #======================================================================
1557 # VALUES, GENERATE, AUTOLOAD
1558 #======================================================================
1560 # LDNOTE: original code from nwiger, didn't touch code in that section
1561 # I feel the AUTOLOAD stuff should not be the default, it should
1562 # only be activated on explicit demand by user.
1566 my $data = shift || return;
1567 puke "Argument to ", __PACKAGE__, "->values must be a \\%hash"
1568 unless ref $data eq 'HASH';
1571 foreach my $k (sort keys %$data) {
1572 my $v = $data->{$k};
1573 $self->_SWITCH_refkind($v, {
1575 if ($self->{array_datatypes}) { # array datatype
1576 push @all_bind, $self->_bindtype($k, $v);
1578 else { # literal SQL with bind
1579 my ($sql, @bind) = @$v;
1580 $self->_assert_bindval_matches_bindtype(@bind);
1581 push @all_bind, @bind;
1584 ARRAYREFREF => sub { # literal SQL with bind
1585 my ($sql, @bind) = @${$v};
1586 $self->_assert_bindval_matches_bindtype(@bind);
1587 push @all_bind, @bind;
1589 SCALARREF => sub { # literal SQL without bind
1591 SCALAR_or_UNDEF => sub {
1592 push @all_bind, $self->_bindtype($k, $v);
1603 my(@sql, @sqlq, @sqlv);
1607 if ($ref eq 'HASH') {
1608 for my $k (sort keys %$_) {
1611 my $label = $self->_quote($k);
1612 if ($r eq 'ARRAY') {
1613 # literal SQL with bind
1614 my ($sql, @bind) = @$v;
1615 $self->_assert_bindval_matches_bindtype(@bind);
1616 push @sqlq, "$label = $sql";
1618 } elsif ($r eq 'SCALAR') {
1619 # literal SQL without bind
1620 push @sqlq, "$label = $$v";
1622 push @sqlq, "$label = ?";
1623 push @sqlv, $self->_bindtype($k, $v);
1626 push @sql, $self->_sqlcase('set'), join ', ', @sqlq;
1627 } elsif ($ref eq 'ARRAY') {
1628 # unlike insert(), assume these are ONLY the column names, i.e. for SQL
1631 if ($r eq 'ARRAY') { # literal SQL with bind
1632 my ($sql, @bind) = @$v;
1633 $self->_assert_bindval_matches_bindtype(@bind);
1636 } elsif ($r eq 'SCALAR') { # literal SQL without bind
1637 # embedded literal SQL
1644 push @sql, '(' . join(', ', @sqlq) . ')';
1645 } elsif ($ref eq 'SCALAR') {
1649 # strings get case twiddled
1650 push @sql, $self->_sqlcase($_);
1654 my $sql = join ' ', @sql;
1656 # this is pretty tricky
1657 # if ask for an array, return ($stmt, @bind)
1658 # otherwise, s/?/shift @sqlv/ to put it inline
1660 return ($sql, @sqlv);
1662 1 while $sql =~ s/\?/my $d = shift(@sqlv);
1663 ref $d ? $d->[1] : $d/e;
1672 # This allows us to check for a local, then _form, attr
1674 my($name) = $AUTOLOAD =~ /.*::(.+)/;
1675 return $self->generate($name, @_);
1686 SQL::Abstract - Generate SQL from Perl data structures
1692 my $sql = SQL::Abstract->new;
1694 my($stmt, @bind) = $sql->select($source, \@fields, \%where, $order);
1696 my($stmt, @bind) = $sql->insert($table, \%fieldvals || \@values);
1698 my($stmt, @bind) = $sql->update($table, \%fieldvals, \%where);
1700 my($stmt, @bind) = $sql->delete($table, \%where);
1702 # Then, use these in your DBI statements
1703 my $sth = $dbh->prepare($stmt);
1704 $sth->execute(@bind);
1706 # Just generate the WHERE clause
1707 my($stmt, @bind) = $sql->where(\%where, $order);
1709 # Return values in the same order, for hashed queries
1710 # See PERFORMANCE section for more details
1711 my @bind = $sql->values(\%fieldvals);
1715 This module was inspired by the excellent L<DBIx::Abstract>.
1716 However, in using that module I found that what I really wanted
1717 to do was generate SQL, but still retain complete control over my
1718 statement handles and use the DBI interface. So, I set out to
1719 create an abstract SQL generation module.
1721 While based on the concepts used by L<DBIx::Abstract>, there are
1722 several important differences, especially when it comes to WHERE
1723 clauses. I have modified the concepts used to make the SQL easier
1724 to generate from Perl data structures and, IMO, more intuitive.
1725 The underlying idea is for this module to do what you mean, based
1726 on the data structures you provide it. The big advantage is that
1727 you don't have to modify your code every time your data changes,
1728 as this module figures it out.
1730 To begin with, an SQL INSERT is as easy as just specifying a hash
1731 of C<key=value> pairs:
1734 name => 'Jimbo Bobson',
1735 phone => '123-456-7890',
1736 address => '42 Sister Lane',
1737 city => 'St. Louis',
1738 state => 'Louisiana',
1741 The SQL can then be generated with this:
1743 my($stmt, @bind) = $sql->insert('people', \%data);
1745 Which would give you something like this:
1747 $stmt = "INSERT INTO people
1748 (address, city, name, phone, state)
1749 VALUES (?, ?, ?, ?, ?)";
1750 @bind = ('42 Sister Lane', 'St. Louis', 'Jimbo Bobson',
1751 '123-456-7890', 'Louisiana');
1753 These are then used directly in your DBI code:
1755 my $sth = $dbh->prepare($stmt);
1756 $sth->execute(@bind);
1758 =head2 Inserting and Updating Arrays
1760 If your database has array types (like for example Postgres),
1761 activate the special option C<< array_datatypes => 1 >>
1762 when creating the C<SQL::Abstract> object.
1763 Then you may use an arrayref to insert and update database array types:
1765 my $sql = SQL::Abstract->new(array_datatypes => 1);
1767 planets => [qw/Mercury Venus Earth Mars/]
1770 my($stmt, @bind) = $sql->insert('solar_system', \%data);
1774 $stmt = "INSERT INTO solar_system (planets) VALUES (?)"
1776 @bind = (['Mercury', 'Venus', 'Earth', 'Mars']);
1779 =head2 Inserting and Updating SQL
1781 In order to apply SQL functions to elements of your C<%data> you may
1782 specify a reference to an arrayref for the given hash value. For example,
1783 if you need to execute the Oracle C<to_date> function on a value, you can
1784 say something like this:
1788 date_entered => \[ "to_date(?,'MM/DD/YYYY')", "03/02/2003" ],
1791 The first value in the array is the actual SQL. Any other values are
1792 optional and would be included in the bind values array. This gives
1795 my($stmt, @bind) = $sql->insert('people', \%data);
1797 $stmt = "INSERT INTO people (name, date_entered)
1798 VALUES (?, to_date(?,'MM/DD/YYYY'))";
1799 @bind = ('Bill', '03/02/2003');
1801 An UPDATE is just as easy, all you change is the name of the function:
1803 my($stmt, @bind) = $sql->update('people', \%data);
1805 Notice that your C<%data> isn't touched; the module will generate
1806 the appropriately quirky SQL for you automatically. Usually you'll
1807 want to specify a WHERE clause for your UPDATE, though, which is
1808 where handling C<%where> hashes comes in handy...
1810 =head2 Complex where statements
1812 This module can generate pretty complicated WHERE statements
1813 easily. For example, simple C<key=value> pairs are taken to mean
1814 equality, and if you want to see if a field is within a set
1815 of values, you can use an arrayref. Let's say we wanted to
1816 SELECT some data based on this criteria:
1819 requestor => 'inna',
1820 worker => ['nwiger', 'rcwe', 'sfz'],
1821 status => { '!=', 'completed' }
1824 my($stmt, @bind) = $sql->select('tickets', '*', \%where);
1826 The above would give you something like this:
1828 $stmt = "SELECT * FROM tickets WHERE
1829 ( requestor = ? ) AND ( status != ? )
1830 AND ( worker = ? OR worker = ? OR worker = ? )";
1831 @bind = ('inna', 'completed', 'nwiger', 'rcwe', 'sfz');
1833 Which you could then use in DBI code like so:
1835 my $sth = $dbh->prepare($stmt);
1836 $sth->execute(@bind);
1842 The methods are simple. There's one for every major SQL operation,
1843 and a constructor you use first. The arguments are specified in a
1844 similar order for each method (table, then fields, then a where
1845 clause) to try and simplify things.
1847 =head2 new(option => 'value')
1849 The C<new()> function takes a list of options and values, and returns
1850 a new B<SQL::Abstract> object which can then be used to generate SQL
1851 through the methods below. The options accepted are:
1857 If set to 'lower', then SQL will be generated in all lowercase. By
1858 default SQL is generated in "textbook" case meaning something like:
1860 SELECT a_field FROM a_table WHERE some_field LIKE '%someval%'
1862 Any setting other than 'lower' is ignored.
1866 This determines what the default comparison operator is. By default
1867 it is C<=>, meaning that a hash like this:
1869 %where = (name => 'nwiger', email => 'nate@wiger.org');
1871 Will generate SQL like this:
1873 WHERE name = 'nwiger' AND email = 'nate@wiger.org'
1875 However, you may want loose comparisons by default, so if you set
1876 C<cmp> to C<like> you would get SQL such as:
1878 WHERE name like 'nwiger' AND email like 'nate@wiger.org'
1880 You can also override the comparison on an individual basis - see
1881 the huge section on L</"WHERE CLAUSES"> at the bottom.
1883 =item sqltrue, sqlfalse
1885 Expressions for inserting boolean values within SQL statements.
1886 By default these are C<1=1> and C<1=0>. They are used
1887 by the special operators C<-in> and C<-not_in> for generating
1888 correct SQL even when the argument is an empty array (see below).
1892 This determines the default logical operator for multiple WHERE
1893 statements in arrays or hashes. If absent, the default logic is "or"
1894 for arrays, and "and" for hashes. This means that a WHERE
1898 event_date => {'>=', '2/13/99'},
1899 event_date => {'<=', '4/24/03'},
1902 will generate SQL like this:
1904 WHERE event_date >= '2/13/99' OR event_date <= '4/24/03'
1906 This is probably not what you want given this query, though (look
1907 at the dates). To change the "OR" to an "AND", simply specify:
1909 my $sql = SQL::Abstract->new(logic => 'and');
1911 Which will change the above C<WHERE> to:
1913 WHERE event_date >= '2/13/99' AND event_date <= '4/24/03'
1915 The logic can also be changed locally by inserting
1916 a modifier in front of an arrayref :
1918 @where = (-and => [event_date => {'>=', '2/13/99'},
1919 event_date => {'<=', '4/24/03'} ]);
1921 See the L</"WHERE CLAUSES"> section for explanations.
1925 This will automatically convert comparisons using the specified SQL
1926 function for both column and value. This is mostly used with an argument
1927 of C<upper> or C<lower>, so that the SQL will have the effect of
1928 case-insensitive "searches". For example, this:
1930 $sql = SQL::Abstract->new(convert => 'upper');
1931 %where = (keywords => 'MaKe iT CAse inSeNSItive');
1933 Will turn out the following SQL:
1935 WHERE upper(keywords) like upper('MaKe iT CAse inSeNSItive')
1937 The conversion can be C<upper()>, C<lower()>, or any other SQL function
1938 that can be applied symmetrically to fields (actually B<SQL::Abstract> does
1939 not validate this option; it will just pass through what you specify verbatim).
1943 This is a kludge because many databases suck. For example, you can't
1944 just bind values using DBI's C<execute()> for Oracle C<CLOB> or C<BLOB> fields.
1945 Instead, you have to use C<bind_param()>:
1947 $sth->bind_param(1, 'reg data');
1948 $sth->bind_param(2, $lots, {ora_type => ORA_CLOB});
1950 The problem is, B<SQL::Abstract> will normally just return a C<@bind> array,
1951 which loses track of which field each slot refers to. Fear not.
1953 If you specify C<bindtype> in new, you can determine how C<@bind> is returned.
1954 Currently, you can specify either C<normal> (default) or C<columns>. If you
1955 specify C<columns>, you will get an array that looks like this:
1957 my $sql = SQL::Abstract->new(bindtype => 'columns');
1958 my($stmt, @bind) = $sql->insert(...);
1961 [ 'column1', 'value1' ],
1962 [ 'column2', 'value2' ],
1963 [ 'column3', 'value3' ],
1966 You can then iterate through this manually, using DBI's C<bind_param()>.
1968 $sth->prepare($stmt);
1971 my($col, $data) = @$_;
1972 if ($col eq 'details' || $col eq 'comments') {
1973 $sth->bind_param($i, $data, {ora_type => ORA_CLOB});
1974 } elsif ($col eq 'image') {
1975 $sth->bind_param($i, $data, {ora_type => ORA_BLOB});
1977 $sth->bind_param($i, $data);
1981 $sth->execute; # execute without @bind now
1983 Now, why would you still use B<SQL::Abstract> if you have to do this crap?
1984 Basically, the advantage is still that you don't have to care which fields
1985 are or are not included. You could wrap that above C<for> loop in a simple
1986 sub called C<bind_fields()> or something and reuse it repeatedly. You still
1987 get a layer of abstraction over manual SQL specification.
1989 Note that if you set L</bindtype> to C<columns>, the C<\[ $sql, @bind ]>
1990 construct (see L</Literal SQL with placeholders and bind values (subqueries)>)
1991 will expect the bind values in this format.
1995 This is the character that a table or column name will be quoted
1996 with. By default this is an empty string, but you could set it to
1997 the character C<`>, to generate SQL like this:
1999 SELECT `a_field` FROM `a_table` WHERE `some_field` LIKE '%someval%'
2001 Alternatively, you can supply an array ref of two items, the first being the left
2002 hand quote character, and the second the right hand quote character. For
2003 example, you could supply C<['[',']']> for SQL Server 2000 compliant quotes
2004 that generates SQL like this:
2006 SELECT [a_field] FROM [a_table] WHERE [some_field] LIKE '%someval%'
2008 Quoting is useful if you have tables or columns names that are reserved
2009 words in your database's SQL dialect.
2013 This is the character that will be used to escape L</quote_char>s appearing
2014 in an identifier before it has been quoted.
2016 The parameter default in case of a single L</quote_char> character is the quote
2019 When opening-closing-style quoting is used (L</quote_char> is an arrayref)
2020 this parameter defaults to the B<closing (right)> L</quote_char>. Occurences
2021 of the B<opening (left)> L</quote_char> within the identifier are currently left
2022 untouched. The default for opening-closing-style quotes may change in future
2023 versions, thus you are B<strongly encouraged> to specify the escape character
2028 This is the character that separates a table and column name. It is
2029 necessary to specify this when the C<quote_char> option is selected,
2030 so that tables and column names can be individually quoted like this:
2032 SELECT `table`.`one_field` FROM `table` WHERE `table`.`other_field` = 1
2034 =item injection_guard
2036 A regular expression C<qr/.../> that is applied to any C<-function> and unquoted
2037 column name specified in a query structure. This is a safety mechanism to avoid
2038 injection attacks when mishandling user input e.g.:
2040 my %condition_as_column_value_pairs = get_values_from_user();
2041 $sqla->select( ... , \%condition_as_column_value_pairs );
2043 If the expression matches an exception is thrown. Note that literal SQL
2044 supplied via C<\'...'> or C<\['...']> is B<not> checked in any way.
2046 Defaults to checking for C<;> and the C<GO> keyword (TransactSQL)
2048 =item array_datatypes
2050 When this option is true, arrayrefs in INSERT or UPDATE are
2051 interpreted as array datatypes and are passed directly
2053 When this option is false, arrayrefs are interpreted
2054 as literal SQL, just like refs to arrayrefs
2055 (but this behavior is for backwards compatibility; when writing
2056 new queries, use the "reference to arrayref" syntax
2062 Takes a reference to a list of "special operators"
2063 to extend the syntax understood by L<SQL::Abstract>.
2064 See section L</"SPECIAL OPERATORS"> for details.
2068 Takes a reference to a list of "unary operators"
2069 to extend the syntax understood by L<SQL::Abstract>.
2070 See section L</"UNARY OPERATORS"> for details.
2076 =head2 insert($table, \@values || \%fieldvals, \%options)
2078 This is the simplest function. You simply give it a table name
2079 and either an arrayref of values or hashref of field/value pairs.
2080 It returns an SQL INSERT statement and a list of bind values.
2081 See the sections on L</"Inserting and Updating Arrays"> and
2082 L</"Inserting and Updating SQL"> for information on how to insert
2083 with those data types.
2085 The optional C<\%options> hash reference may contain additional
2086 options to generate the insert SQL. Currently supported options
2093 Takes either a scalar of raw SQL fields, or an array reference of
2094 field names, and adds on an SQL C<RETURNING> statement at the end.
2095 This allows you to return data generated by the insert statement
2096 (such as row IDs) without performing another C<SELECT> statement.
2097 Note, however, this is not part of the SQL standard and may not
2098 be supported by all database engines.
2102 =head2 update($table, \%fieldvals, \%where, \%options)
2104 This takes a table, hashref of field/value pairs, and an optional
2105 hashref L<WHERE clause|/WHERE CLAUSES>. It returns an SQL UPDATE function and a list
2107 See the sections on L</"Inserting and Updating Arrays"> and
2108 L</"Inserting and Updating SQL"> for information on how to insert
2109 with those data types.
2111 The optional C<\%options> hash reference may contain additional
2112 options to generate the update SQL. Currently supported options
2119 See the C<returning> option to
2120 L<insert|/insert($table, \@values || \%fieldvals, \%options)>.
2124 =head2 select($source, $fields, $where, $order)
2126 This returns a SQL SELECT statement and associated list of bind values, as
2127 specified by the arguments :
2133 Specification of the 'FROM' part of the statement.
2134 The argument can be either a plain scalar (interpreted as a table
2135 name, will be quoted), or an arrayref (interpreted as a list
2136 of table names, joined by commas, quoted), or a scalarref
2137 (literal table name, not quoted), or a ref to an arrayref
2138 (list of literal table names, joined by commas, not quoted).
2142 Specification of the list of fields to retrieve from
2144 The argument can be either an arrayref (interpreted as a list
2145 of field names, will be joined by commas and quoted), or a
2146 plain scalar (literal SQL, not quoted).
2147 Please observe that this API is not as flexible as that of
2148 the first argument C<$source>, for backwards compatibility reasons.
2152 Optional argument to specify the WHERE part of the query.
2153 The argument is most often a hashref, but can also be
2154 an arrayref or plain scalar --
2155 see section L<WHERE clause|/"WHERE CLAUSES"> for details.
2159 Optional argument to specify the ORDER BY part of the query.
2160 The argument can be a scalar, a hashref or an arrayref
2161 -- see section L<ORDER BY clause|/"ORDER BY CLAUSES">
2167 =head2 delete($table, \%where, \%options)
2169 This takes a table name and optional hashref L<WHERE clause|/WHERE CLAUSES>.
2170 It returns an SQL DELETE statement and list of bind values.
2172 The optional C<\%options> hash reference may contain additional
2173 options to generate the delete SQL. Currently supported options
2180 See the C<returning> option to
2181 L<insert|/insert($table, \@values || \%fieldvals, \%options)>.
2185 =head2 where(\%where, $order)
2187 This is used to generate just the WHERE clause. For example,
2188 if you have an arbitrary data structure and know what the
2189 rest of your SQL is going to look like, but want an easy way
2190 to produce a WHERE clause, use this. It returns an SQL WHERE
2191 clause and list of bind values.
2194 =head2 values(\%data)
2196 This just returns the values from the hash C<%data>, in the same
2197 order that would be returned from any of the other above queries.
2198 Using this allows you to markedly speed up your queries if you
2199 are affecting lots of rows. See below under the L</"PERFORMANCE"> section.
2201 =head2 generate($any, 'number', $of, \@data, $struct, \%types)
2203 Warning: This is an experimental method and subject to change.
2205 This returns arbitrarily generated SQL. It's a really basic shortcut.
2206 It will return two different things, depending on return context:
2208 my($stmt, @bind) = $sql->generate('create table', \$table, \@fields);
2209 my $stmt_and_val = $sql->generate('create table', \$table, \@fields);
2211 These would return the following:
2213 # First calling form
2214 $stmt = "CREATE TABLE test (?, ?)";
2215 @bind = (field1, field2);
2217 # Second calling form
2218 $stmt_and_val = "CREATE TABLE test (field1, field2)";
2220 Depending on what you're trying to do, it's up to you to choose the correct
2221 format. In this example, the second form is what you would want.
2225 $sql->generate('alter session', { nls_date_format => 'MM/YY' });
2229 ALTER SESSION SET nls_date_format = 'MM/YY'
2231 You get the idea. Strings get their case twiddled, but everything
2232 else remains verbatim.
2234 =head1 EXPORTABLE FUNCTIONS
2236 =head2 is_plain_value
2238 Determines if the supplied argument is a plain value as understood by this
2243 =item * The value is C<undef>
2245 =item * The value is a non-reference
2247 =item * The value is an object with stringification overloading
2249 =item * The value is of the form C<< { -value => $anything } >>
2253 On failure returns C<undef>, on sucess returns a B<scalar> reference
2254 to the original supplied argument.
2260 The stringification overloading detection is rather advanced: it takes
2261 into consideration not only the presence of a C<""> overload, but if that
2262 fails also checks for enabled
2263 L<autogenerated versions of C<"">|overload/Magic Autogeneration>, based
2264 on either C<0+> or C<bool>.
2266 Unfortunately testing in the field indicates that this
2267 detection B<< may tickle a latent bug in perl versions before 5.018 >>,
2268 but only when very large numbers of stringifying objects are involved.
2269 At the time of writing ( Sep 2014 ) there is no clear explanation of
2270 the direct cause, nor is there a manageably small test case that reliably
2271 reproduces the problem.
2273 If you encounter any of the following exceptions in B<random places within
2274 your application stack> - this module may be to blame:
2276 Operation "ne": no method found,
2277 left argument in overloaded package <something>,
2278 right argument in overloaded package <something>
2282 Stub found while resolving method "???" overloading """" in package <something>
2284 If you fall victim to the above - please attempt to reduce the problem
2285 to something that could be sent to the L<SQL::Abstract developers
2286 |DBIx::Class/GETTING HELP/SUPPORT>
2287 (either publicly or privately). As a workaround in the meantime you can
2288 set C<$ENV{SQLA_ISVALUE_IGNORE_AUTOGENERATED_STRINGIFICATION}> to a true
2289 value, which will most likely eliminate your problem (at the expense of
2290 not being able to properly detect exotic forms of stringification).
2292 This notice and environment variable will be removed in a future version,
2293 as soon as the underlying problem is found and a reliable workaround is
2298 =head2 is_literal_value
2300 Determines if the supplied argument is a literal value as understood by this
2305 =item * C<\$sql_string>
2307 =item * C<\[ $sql_string, @bind_values ]>
2311 On failure returns C<undef>, on sucess returns an B<array> reference
2312 containing the unpacked version of the supplied literal SQL and bind values.
2314 =head1 WHERE CLAUSES
2318 This module uses a variation on the idea from L<DBIx::Abstract>. It
2319 is B<NOT>, repeat I<not> 100% compatible. B<The main logic of this
2320 module is that things in arrays are OR'ed, and things in hashes
2323 The easiest way to explain is to show lots of examples. After
2324 each C<%where> hash shown, it is assumed you used:
2326 my($stmt, @bind) = $sql->where(\%where);
2328 However, note that the C<%where> hash can be used directly in any
2329 of the other functions as well, as described above.
2331 =head2 Key-value pairs
2333 So, let's get started. To begin, a simple hash:
2337 status => 'completed'
2340 Is converted to SQL C<key = val> statements:
2342 $stmt = "WHERE user = ? AND status = ?";
2343 @bind = ('nwiger', 'completed');
2345 One common thing I end up doing is having a list of values that
2346 a field can be in. To do this, simply specify a list inside of
2351 status => ['assigned', 'in-progress', 'pending'];
2354 This simple code will create the following:
2356 $stmt = "WHERE user = ? AND ( status = ? OR status = ? OR status = ? )";
2357 @bind = ('nwiger', 'assigned', 'in-progress', 'pending');
2359 A field associated to an empty arrayref will be considered a
2360 logical false and will generate 0=1.
2362 =head2 Tests for NULL values
2364 If the value part is C<undef> then this is converted to SQL <IS NULL>
2373 $stmt = "WHERE user = ? AND status IS NULL";
2376 To test if a column IS NOT NULL:
2380 status => { '!=', undef },
2383 =head2 Specific comparison operators
2385 If you want to specify a different type of operator for your comparison,
2386 you can use a hashref for a given column:
2390 status => { '!=', 'completed' }
2393 Which would generate:
2395 $stmt = "WHERE user = ? AND status != ?";
2396 @bind = ('nwiger', 'completed');
2398 To test against multiple values, just enclose the values in an arrayref:
2400 status => { '=', ['assigned', 'in-progress', 'pending'] };
2402 Which would give you:
2404 "WHERE status = ? OR status = ? OR status = ?"
2407 The hashref can also contain multiple pairs, in which case it is expanded
2408 into an C<AND> of its elements:
2412 status => { '!=', 'completed', -not_like => 'pending%' }
2415 # Or more dynamically, like from a form
2416 $where{user} = 'nwiger';
2417 $where{status}{'!='} = 'completed';
2418 $where{status}{'-not_like'} = 'pending%';
2420 # Both generate this
2421 $stmt = "WHERE user = ? AND status != ? AND status NOT LIKE ?";
2422 @bind = ('nwiger', 'completed', 'pending%');
2425 To get an OR instead, you can combine it with the arrayref idea:
2429 priority => [ { '=', 2 }, { '>', 5 } ]
2432 Which would generate:
2434 $stmt = "WHERE ( priority = ? OR priority > ? ) AND user = ?";
2435 @bind = ('2', '5', 'nwiger');
2437 If you want to include literal SQL (with or without bind values), just use a
2438 scalar reference or reference to an arrayref as the value:
2441 date_entered => { '>' => \["to_date(?, 'MM/DD/YYYY')", "11/26/2008"] },
2442 date_expires => { '<' => \"now()" }
2445 Which would generate:
2447 $stmt = "WHERE date_entered > to_date(?, 'MM/DD/YYYY') AND date_expires < now()";
2448 @bind = ('11/26/2008');
2451 =head2 Logic and nesting operators
2453 In the example above,
2454 there is a subtle trap if you want to say something like
2455 this (notice the C<AND>):
2457 WHERE priority != ? AND priority != ?
2459 Because, in Perl you I<can't> do this:
2461 priority => { '!=' => 2, '!=' => 1 }
2463 As the second C<!=> key will obliterate the first. The solution
2464 is to use the special C<-modifier> form inside an arrayref:
2466 priority => [ -and => {'!=', 2},
2470 Normally, these would be joined by C<OR>, but the modifier tells it
2471 to use C<AND> instead. (Hint: You can use this in conjunction with the
2472 C<logic> option to C<new()> in order to change the way your queries
2473 work by default.) B<Important:> Note that the C<-modifier> goes
2474 B<INSIDE> the arrayref, as an extra first element. This will
2475 B<NOT> do what you think it might:
2477 priority => -and => [{'!=', 2}, {'!=', 1}] # WRONG!
2479 Here is a quick list of equivalencies, since there is some overlap:
2482 status => {'!=', 'completed', 'not like', 'pending%' }
2483 status => [ -and => {'!=', 'completed'}, {'not like', 'pending%'}]
2486 status => {'=', ['assigned', 'in-progress']}
2487 status => [ -or => {'=', 'assigned'}, {'=', 'in-progress'}]
2488 status => [ {'=', 'assigned'}, {'=', 'in-progress'} ]
2492 =head2 Special operators : IN, BETWEEN, etc.
2494 You can also use the hashref format to compare a list of fields using the
2495 C<IN> comparison operator, by specifying the list as an arrayref:
2498 status => 'completed',
2499 reportid => { -in => [567, 2335, 2] }
2502 Which would generate:
2504 $stmt = "WHERE status = ? AND reportid IN (?,?,?)";
2505 @bind = ('completed', '567', '2335', '2');
2507 The reverse operator C<-not_in> generates SQL C<NOT IN> and is used in
2510 If the argument to C<-in> is an empty array, 'sqlfalse' is generated
2511 (by default : C<1=0>). Similarly, C<< -not_in => [] >> generates
2512 'sqltrue' (by default : C<1=1>).
2514 In addition to the array you can supply a chunk of literal sql or
2515 literal sql with bind:
2518 customer => { -in => \[
2519 'SELECT cust_id FROM cust WHERE balance > ?',
2522 status => { -in => \'SELECT status_codes FROM states' },
2528 customer IN ( SELECT cust_id FROM cust WHERE balance > ? )
2529 AND status IN ( SELECT status_codes FROM states )
2533 Finally, if the argument to C<-in> is not a reference, it will be
2534 treated as a single-element array.
2536 Another pair of operators is C<-between> and C<-not_between>,
2537 used with an arrayref of two values:
2541 completion_date => {
2542 -not_between => ['2002-10-01', '2003-02-06']
2548 WHERE user = ? AND completion_date NOT BETWEEN ( ? AND ? )
2550 Just like with C<-in> all plausible combinations of literal SQL
2554 start0 => { -between => [ 1, 2 ] },
2555 start1 => { -between => \["? AND ?", 1, 2] },
2556 start2 => { -between => \"lower(x) AND upper(y)" },
2557 start3 => { -between => [
2559 \["upper(?)", 'stuff' ],
2566 ( start0 BETWEEN ? AND ? )
2567 AND ( start1 BETWEEN ? AND ? )
2568 AND ( start2 BETWEEN lower(x) AND upper(y) )
2569 AND ( start3 BETWEEN lower(x) AND upper(?) )
2571 @bind = (1, 2, 1, 2, 'stuff');
2574 These are the two builtin "special operators"; but the
2575 list can be expanded : see section L</"SPECIAL OPERATORS"> below.
2577 =head2 Unary operators: bool
2579 If you wish to test against boolean columns or functions within your
2580 database you can use the C<-bool> and C<-not_bool> operators. For
2581 example to test the column C<is_user> being true and the column
2582 C<is_enabled> being false you would use:-
2586 -not_bool => 'is_enabled',
2591 WHERE is_user AND NOT is_enabled
2593 If a more complex combination is required, testing more conditions,
2594 then you should use the and/or operators:-
2599 -not_bool => { two=> { -rlike => 'bar' } },
2600 -not_bool => { three => [ { '=', 2 }, { '>', 5 } ] },
2611 (NOT ( three = ? OR three > ? ))
2614 =head2 Nested conditions, -and/-or prefixes
2616 So far, we've seen how multiple conditions are joined with a top-level
2617 C<AND>. We can change this by putting the different conditions we want in
2618 hashes and then putting those hashes in an array. For example:
2623 status => { -like => ['pending%', 'dispatched'] },
2627 status => 'unassigned',
2631 This data structure would create the following:
2633 $stmt = "WHERE ( user = ? AND ( status LIKE ? OR status LIKE ? ) )
2634 OR ( user = ? AND status = ? ) )";
2635 @bind = ('nwiger', 'pending', 'dispatched', 'robot', 'unassigned');
2638 Clauses in hashrefs or arrayrefs can be prefixed with an C<-and> or C<-or>
2639 to change the logic inside :
2645 -and => [ workhrs => {'>', 20}, geo => 'ASIA' ],
2646 -or => { workhrs => {'<', 50}, geo => 'EURO' },
2653 $stmt = "WHERE ( user = ?
2654 AND ( ( workhrs > ? AND geo = ? )
2655 OR ( workhrs < ? OR geo = ? ) ) )";
2656 @bind = ('nwiger', '20', 'ASIA', '50', 'EURO');
2658 =head3 Algebraic inconsistency, for historical reasons
2660 C<Important note>: when connecting several conditions, the C<-and->|C<-or>
2661 operator goes C<outside> of the nested structure; whereas when connecting
2662 several constraints on one column, the C<-and> operator goes
2663 C<inside> the arrayref. Here is an example combining both features :
2666 -and => [a => 1, b => 2],
2667 -or => [c => 3, d => 4],
2668 e => [-and => {-like => 'foo%'}, {-like => '%bar'} ]
2673 WHERE ( ( ( a = ? AND b = ? )
2674 OR ( c = ? OR d = ? )
2675 OR ( e LIKE ? AND e LIKE ? ) ) )
2677 This difference in syntax is unfortunate but must be preserved for
2678 historical reasons. So be careful : the two examples below would
2679 seem algebraically equivalent, but they are not
2682 { -like => 'foo%' },
2683 { -like => '%bar' },
2685 # yields : WHERE ( ( col LIKE ? AND col LIKE ? ) )
2688 { col => { -like => 'foo%' } },
2689 { col => { -like => '%bar' } },
2691 # yields : WHERE ( ( col LIKE ? OR col LIKE ? ) )
2694 =head2 Literal SQL and value type operators
2696 The basic premise of SQL::Abstract is that in WHERE specifications the "left
2697 side" is a column name and the "right side" is a value (normally rendered as
2698 a placeholder). This holds true for both hashrefs and arrayref pairs as you
2699 see in the L</WHERE CLAUSES> examples above. Sometimes it is necessary to
2700 alter this behavior. There are several ways of doing so.
2704 This is a virtual operator that signals the string to its right side is an
2705 identifier (a column name) and not a value. For example to compare two
2706 columns you would write:
2709 priority => { '<', 2 },
2710 requestor => { -ident => 'submitter' },
2715 $stmt = "WHERE priority < ? AND requestor = submitter";
2718 If you are maintaining legacy code you may see a different construct as
2719 described in L</Deprecated usage of Literal SQL>, please use C<-ident> in new
2724 This is a virtual operator that signals that the construct to its right side
2725 is a value to be passed to DBI. This is for example necessary when you want
2726 to write a where clause against an array (for RDBMS that support such
2727 datatypes). For example:
2730 array => { -value => [1, 2, 3] }
2735 $stmt = 'WHERE array = ?';
2736 @bind = ([1, 2, 3]);
2738 Note that if you were to simply say:
2744 the result would probably not be what you wanted:
2746 $stmt = 'WHERE array = ? OR array = ? OR array = ?';
2751 Finally, sometimes only literal SQL will do. To include a random snippet
2752 of SQL verbatim, you specify it as a scalar reference. Consider this only
2753 as a last resort. Usually there is a better way. For example:
2756 priority => { '<', 2 },
2757 requestor => { -in => \'(SELECT name FROM hitmen)' },
2762 $stmt = "WHERE priority < ? AND requestor IN (SELECT name FROM hitmen)"
2765 Note that in this example, you only get one bind parameter back, since
2766 the verbatim SQL is passed as part of the statement.
2770 Never use untrusted input as a literal SQL argument - this is a massive
2771 security risk (there is no way to check literal snippets for SQL
2772 injections and other nastyness). If you need to deal with untrusted input
2773 use literal SQL with placeholders as described next.
2775 =head3 Literal SQL with placeholders and bind values (subqueries)
2777 If the literal SQL to be inserted has placeholders and bind values,
2778 use a reference to an arrayref (yes this is a double reference --
2779 not so common, but perfectly legal Perl). For example, to find a date
2780 in Postgres you can use something like this:
2783 date_column => \[ "= date '2008-09-30' - ?::integer", 10 ]
2788 $stmt = "WHERE ( date_column = date '2008-09-30' - ?::integer )"
2791 Note that you must pass the bind values in the same format as they are returned
2792 by L<where|/where(\%where, $order)>. This means that if you set L</bindtype>
2793 to C<columns>, you must provide the bind values in the
2794 C<< [ column_meta => value ] >> format, where C<column_meta> is an opaque
2795 scalar value; most commonly the column name, but you can use any scalar value
2796 (including references and blessed references), L<SQL::Abstract> will simply
2797 pass it through intact. So if C<bindtype> is set to C<columns> the above
2798 example will look like:
2801 date_column => \[ "= date '2008-09-30' - ?::integer", [ {} => 10 ] ]
2804 Literal SQL is especially useful for nesting parenthesized clauses in the
2805 main SQL query. Here is a first example :
2807 my ($sub_stmt, @sub_bind) = ("SELECT c1 FROM t1 WHERE c2 < ? AND c3 LIKE ?",
2811 bar => \["IN ($sub_stmt)" => @sub_bind],
2816 $stmt = "WHERE (foo = ? AND bar IN (SELECT c1 FROM t1
2817 WHERE c2 < ? AND c3 LIKE ?))";
2818 @bind = (1234, 100, "foo%");
2820 Other subquery operators, like for example C<"E<gt> ALL"> or C<"NOT IN">,
2821 are expressed in the same way. Of course the C<$sub_stmt> and
2822 its associated bind values can be generated through a former call
2825 my ($sub_stmt, @sub_bind)
2826 = $sql->select("t1", "c1", {c2 => {"<" => 100},
2827 c3 => {-like => "foo%"}});
2830 bar => \["> ALL ($sub_stmt)" => @sub_bind],
2833 In the examples above, the subquery was used as an operator on a column;
2834 but the same principle also applies for a clause within the main C<%where>
2835 hash, like an EXISTS subquery :
2837 my ($sub_stmt, @sub_bind)
2838 = $sql->select("t1", "*", {c1 => 1, c2 => \"> t0.c0"});
2839 my %where = ( -and => [
2841 \["EXISTS ($sub_stmt)" => @sub_bind],
2846 $stmt = "WHERE (foo = ? AND EXISTS (SELECT * FROM t1
2847 WHERE c1 = ? AND c2 > t0.c0))";
2851 Observe that the condition on C<c2> in the subquery refers to
2852 column C<t0.c0> of the main query : this is I<not> a bind
2853 value, so we have to express it through a scalar ref.
2854 Writing C<< c2 => {">" => "t0.c0"} >> would have generated
2855 C<< c2 > ? >> with bind value C<"t0.c0"> ... not exactly
2856 what we wanted here.
2858 Finally, here is an example where a subquery is used
2859 for expressing unary negation:
2861 my ($sub_stmt, @sub_bind)
2862 = $sql->where({age => [{"<" => 10}, {">" => 20}]});
2863 $sub_stmt =~ s/^ where //i; # don't want "WHERE" in the subclause
2865 lname => {like => '%son%'},
2866 \["NOT ($sub_stmt)" => @sub_bind],
2871 $stmt = "lname LIKE ? AND NOT ( age < ? OR age > ? )"
2872 @bind = ('%son%', 10, 20)
2874 =head3 Deprecated usage of Literal SQL
2876 Below are some examples of archaic use of literal SQL. It is shown only as
2877 reference for those who deal with legacy code. Each example has a much
2878 better, cleaner and safer alternative that users should opt for in new code.
2884 my %where = ( requestor => \'IS NOT NULL' )
2886 $stmt = "WHERE requestor IS NOT NULL"
2888 This used to be the way of generating NULL comparisons, before the handling
2889 of C<undef> got formalized. For new code please use the superior syntax as
2890 described in L</Tests for NULL values>.
2894 my %where = ( requestor => \'= submitter' )
2896 $stmt = "WHERE requestor = submitter"
2898 This used to be the only way to compare columns. Use the superior L</-ident>
2899 method for all new code. For example an identifier declared in such a way
2900 will be properly quoted if L</quote_char> is properly set, while the legacy
2901 form will remain as supplied.
2905 my %where = ( is_ready => \"", completed => { '>', '2012-12-21' } )
2907 $stmt = "WHERE completed > ? AND is_ready"
2908 @bind = ('2012-12-21')
2910 Using an empty string literal used to be the only way to express a boolean.
2911 For all new code please use the much more readable
2912 L<-bool|/Unary operators: bool> operator.
2918 These pages could go on for a while, since the nesting of the data
2919 structures this module can handle are pretty much unlimited (the
2920 module implements the C<WHERE> expansion as a recursive function
2921 internally). Your best bet is to "play around" with the module a
2922 little to see how the data structures behave, and choose the best
2923 format for your data based on that.
2925 And of course, all the values above will probably be replaced with
2926 variables gotten from forms or the command line. After all, if you
2927 knew everything ahead of time, you wouldn't have to worry about
2928 dynamically-generating SQL and could just hardwire it into your
2931 =head1 ORDER BY CLAUSES
2933 Some functions take an order by clause. This can either be a scalar (just a
2934 column name), a hashref of C<< { -desc => 'col' } >> or C<< { -asc => 'col' }
2935 >>, a scalarref, an arrayref-ref, or an arrayref of any of the previous
2938 Given | Will Generate
2939 ---------------------------------------------------------------
2941 'colA' | ORDER BY colA
2943 [qw/colA colB/] | ORDER BY colA, colB
2945 {-asc => 'colA'} | ORDER BY colA ASC
2947 {-desc => 'colB'} | ORDER BY colB DESC
2949 ['colA', {-asc => 'colB'}] | ORDER BY colA, colB ASC
2951 { -asc => [qw/colA colB/] } | ORDER BY colA ASC, colB ASC
2953 \'colA DESC' | ORDER BY colA DESC
2955 \[ 'FUNC(colA, ?)', $x ] | ORDER BY FUNC(colA, ?)
2956 | /* ...with $x bound to ? */
2959 { -asc => 'colA' }, | colA ASC,
2960 { -desc => [qw/colB/] }, | colB DESC,
2961 { -asc => [qw/colC colD/] },| colC ASC, colD ASC,
2962 \'colE DESC', | colE DESC,
2963 \[ 'FUNC(colF, ?)', $x ], | FUNC(colF, ?)
2964 ] | /* ...with $x bound to ? */
2965 ===============================================================
2969 =head1 SPECIAL OPERATORS
2971 my $sqlmaker = SQL::Abstract->new(special_ops => [
2975 my ($self, $field, $op, $arg) = @_;
2981 handler => 'method_name',
2985 A "special operator" is a SQL syntactic clause that can be
2986 applied to a field, instead of a usual binary operator.
2989 WHERE field IN (?, ?, ?)
2990 WHERE field BETWEEN ? AND ?
2991 WHERE MATCH(field) AGAINST (?, ?)
2993 Special operators IN and BETWEEN are fairly standard and therefore
2994 are builtin within C<SQL::Abstract> (as the overridable methods
2995 C<_where_field_IN> and C<_where_field_BETWEEN>). For other operators,
2996 like the MATCH .. AGAINST example above which is specific to MySQL,
2997 you can write your own operator handlers - supply a C<special_ops>
2998 argument to the C<new> method. That argument takes an arrayref of
2999 operator definitions; each operator definition is a hashref with two
3006 the regular expression to match the operator
3010 Either a coderef or a plain scalar method name. In both cases
3011 the expected return is C<< ($sql, @bind) >>.
3013 When supplied with a method name, it is simply called on the
3014 L<SQL::Abstract> object as:
3016 $self->$method_name($field, $op, $arg)
3020 $field is the LHS of the operator
3021 $op is the part that matched the handler regex
3024 When supplied with a coderef, it is called as:
3026 $coderef->($self, $field, $op, $arg)
3031 For example, here is an implementation
3032 of the MATCH .. AGAINST syntax for MySQL
3034 my $sqlmaker = SQL::Abstract->new(special_ops => [
3036 # special op for MySql MATCH (field) AGAINST(word1, word2, ...)
3037 {regex => qr/^match$/i,
3039 my ($self, $field, $op, $arg) = @_;
3040 $arg = [$arg] if not ref $arg;
3041 my $label = $self->_quote($field);
3042 my ($placeholder) = $self->_convert('?');
3043 my $placeholders = join ", ", (($placeholder) x @$arg);
3044 my $sql = $self->_sqlcase('match') . " ($label) "
3045 . $self->_sqlcase('against') . " ($placeholders) ";
3046 my @bind = $self->_bindtype($field, @$arg);
3047 return ($sql, @bind);
3054 =head1 UNARY OPERATORS
3056 my $sqlmaker = SQL::Abstract->new(unary_ops => [
3060 my ($self, $op, $arg) = @_;
3066 handler => 'method_name',
3070 A "unary operator" is a SQL syntactic clause that can be
3071 applied to a field - the operator goes before the field
3073 You can write your own operator handlers - supply a C<unary_ops>
3074 argument to the C<new> method. That argument takes an arrayref of
3075 operator definitions; each operator definition is a hashref with two
3082 the regular expression to match the operator
3086 Either a coderef or a plain scalar method name. In both cases
3087 the expected return is C<< $sql >>.
3089 When supplied with a method name, it is simply called on the
3090 L<SQL::Abstract> object as:
3092 $self->$method_name($op, $arg)
3096 $op is the part that matched the handler regex
3097 $arg is the RHS or argument of the operator
3099 When supplied with a coderef, it is called as:
3101 $coderef->($self, $op, $arg)
3109 Thanks to some benchmarking by Mark Stosberg, it turns out that
3110 this module is many orders of magnitude faster than using C<DBIx::Abstract>.
3111 I must admit this wasn't an intentional design issue, but it's a
3112 byproduct of the fact that you get to control your C<DBI> handles
3115 To maximize performance, use a code snippet like the following:
3117 # prepare a statement handle using the first row
3118 # and then reuse it for the rest of the rows
3120 for my $href (@array_of_hashrefs) {
3121 $stmt ||= $sql->insert('table', $href);
3122 $sth ||= $dbh->prepare($stmt);
3123 $sth->execute($sql->values($href));
3126 The reason this works is because the keys in your C<$href> are sorted
3127 internally by B<SQL::Abstract>. Thus, as long as your data retains
3128 the same structure, you only have to generate the SQL the first time
3129 around. On subsequent queries, simply use the C<values> function provided
3130 by this module to return your values in the correct order.
3132 However this depends on the values having the same type - if, for
3133 example, the values of a where clause may either have values
3134 (resulting in sql of the form C<column = ?> with a single bind
3135 value), or alternatively the values might be C<undef> (resulting in
3136 sql of the form C<column IS NULL> with no bind value) then the
3137 caching technique suggested will not work.
3141 If you use my C<CGI::FormBuilder> module at all, you'll hopefully
3142 really like this part (I do, at least). Building up a complex query
3143 can be as simple as the following:
3150 use CGI::FormBuilder;
3153 my $form = CGI::FormBuilder->new(...);
3154 my $sql = SQL::Abstract->new;
3156 if ($form->submitted) {
3157 my $field = $form->field;
3158 my $id = delete $field->{id};
3159 my($stmt, @bind) = $sql->update('table', $field, {id => $id});
3162 Of course, you would still have to connect using C<DBI> to run the
3163 query, but the point is that if you make your form look like your
3164 table, the actual query script can be extremely simplistic.
3166 If you're B<REALLY> lazy (I am), check out C<HTML::QuickTable> for
3167 a fast interface to returning and formatting data. I frequently
3168 use these three modules together to write complex database query
3169 apps in under 50 lines.
3171 =head1 HOW TO CONTRIBUTE
3173 Contributions are always welcome, in all usable forms (we especially
3174 welcome documentation improvements). The delivery methods include git-
3175 or unified-diff formatted patches, GitHub pull requests, or plain bug
3176 reports either via RT or the Mailing list. Contributors are generally
3177 granted full access to the official repository after their first several
3178 patches pass successful review.
3180 This project is maintained in a git repository. The code and related tools are
3181 accessible at the following locations:
3185 =item * Official repo: L<git://git.shadowcat.co.uk/dbsrgits/SQL-Abstract.git>
3187 =item * Official gitweb: L<http://git.shadowcat.co.uk/gitweb/gitweb.cgi?p=dbsrgits/SQL-Abstract.git>
3189 =item * GitHub mirror: L<https://github.com/dbsrgits/sql-abstract>
3191 =item * Authorized committers: L<ssh://dbsrgits@git.shadowcat.co.uk/SQL-Abstract.git>
3197 Version 1.50 was a major internal refactoring of C<SQL::Abstract>.
3198 Great care has been taken to preserve the I<published> behavior
3199 documented in previous versions in the 1.* family; however,
3200 some features that were previously undocumented, or behaved
3201 differently from the documentation, had to be changed in order
3202 to clarify the semantics. Hence, client code that was relying
3203 on some dark areas of C<SQL::Abstract> v1.*
3204 B<might behave differently> in v1.50.
3206 The main changes are :
3212 support for literal SQL through the C<< \ [ $sql, @bind ] >> syntax.
3216 support for the { operator => \"..." } construct (to embed literal SQL)
3220 support for the { operator => \["...", @bind] } construct (to embed literal SQL with bind values)
3224 optional support for L<array datatypes|/"Inserting and Updating Arrays">
3228 defensive programming : check arguments
3232 fixed bug with global logic, which was previously implemented
3233 through global variables yielding side-effects. Prior versions would
3234 interpret C<< [ {cond1, cond2}, [cond3, cond4] ] >>
3235 as C<< "(cond1 AND cond2) OR (cond3 AND cond4)" >>.
3236 Now this is interpreted
3237 as C<< "(cond1 AND cond2) OR (cond3 OR cond4)" >>.
3242 fixed semantics of _bindtype on array args
3246 dropped the C<_anoncopy> of the %where tree. No longer necessary,
3247 we just avoid shifting arrays within that tree.
3251 dropped the C<_modlogic> function
3255 =head1 ACKNOWLEDGEMENTS
3257 There are a number of individuals that have really helped out with
3258 this module. Unfortunately, most of them submitted bugs via CPAN
3259 so I have no idea who they are! But the people I do know are:
3261 Ash Berlin (order_by hash term support)
3262 Matt Trout (DBIx::Class support)
3263 Mark Stosberg (benchmarking)
3264 Chas Owens (initial "IN" operator support)
3265 Philip Collins (per-field SQL functions)
3266 Eric Kolve (hashref "AND" support)
3267 Mike Fragassi (enhancements to "BETWEEN" and "LIKE")
3268 Dan Kubb (support for "quote_char" and "name_sep")
3269 Guillermo Roditi (patch to cleanup "IN" and "BETWEEN", fix and tests for _order_by)
3270 Laurent Dami (internal refactoring, extensible list of special operators, literal SQL)
3271 Norbert Buchmuller (support for literal SQL in hashpair, misc. fixes & tests)
3272 Peter Rabbitson (rewrite of SQLA::Test, misc. fixes & tests)
3273 Oliver Charles (support for "RETURNING" after "INSERT")
3279 L<DBIx::Class>, L<DBIx::Abstract>, L<CGI::FormBuilder>, L<HTML::QuickTable>.
3283 Copyright (c) 2001-2007 Nathan Wiger <nwiger@cpan.org>. All Rights Reserved.
3285 This module is actively maintained by Matt Trout <mst@shadowcatsystems.co.uk>
3287 For support, your best bet is to try the C<DBIx::Class> users mailing list.
3288 While not an official support venue, C<DBIx::Class> makes heavy use of
3289 C<SQL::Abstract>, and as such list members there are very familiar with
3290 how to create queries.
3294 This module is free software; you may copy this under the same
3295 terms as perl itself (either the GNU General Public License or
3296 the Artistic License)