1 package SQL::Abstract; # see doc at end of file
10 our @EXPORT_OK = qw(is_plain_value is_literal_value);
20 *SQL::Abstract::_ENV_::DETECT_AUTOGENERATED_STRINGIFICATION = $ENV{SQLA_ISVALUE_IGNORE_AUTOGENERATED_STRINGIFICATION}
26 #======================================================================
28 #======================================================================
30 our $VERSION = '1.81';
32 # This would confuse some packagers
33 $VERSION = eval $VERSION if $VERSION =~ /_/; # numify for warning-free dev releases
37 # special operators (-in, -between). May be extended/overridden by user.
38 # See section WHERE: BUILTIN SPECIAL OPERATORS below for implementation
39 my @BUILTIN_SPECIAL_OPS = (
40 {regex => qr/^ (?: not \s )? between $/ix, handler => '_where_field_BETWEEN'},
41 {regex => qr/^ (?: not \s )? in $/ix, handler => '_where_field_IN'},
42 {regex => qr/^ ident $/ix, handler => '_where_op_IDENT'},
43 {regex => qr/^ value $/ix, handler => '_where_op_VALUE'},
44 {regex => qr/^ is (?: \s+ not )? $/ix, handler => '_where_field_IS'},
47 # unaryish operators - key maps to handler
48 my @BUILTIN_UNARY_OPS = (
49 # the digits are backcompat stuff
50 { regex => qr/^ and (?: [_\s]? \d+ )? $/xi, handler => '_where_op_ANDOR' },
51 { regex => qr/^ or (?: [_\s]? \d+ )? $/xi, handler => '_where_op_ANDOR' },
52 { regex => qr/^ nest (?: [_\s]? \d+ )? $/xi, handler => '_where_op_NEST' },
53 { regex => qr/^ (?: not \s )? bool $/xi, handler => '_where_op_BOOL' },
54 { regex => qr/^ ident $/xi, handler => '_where_op_IDENT' },
55 { regex => qr/^ value $/xi, handler => '_where_op_VALUE' },
58 #======================================================================
59 # DEBUGGING AND ERROR REPORTING
60 #======================================================================
63 return unless $_[0]->{debug}; shift; # a little faster
64 my $func = (caller(1))[3];
65 warn "[$func] ", @_, "\n";
69 my($func) = (caller(1))[3];
70 Carp::carp "[$func] Warning: ", @_;
74 my($func) = (caller(1))[3];
75 Carp::croak "[$func] Fatal: ", @_;
78 sub is_literal_value ($) {
79 ref $_[0] eq 'SCALAR' ? [ ${$_[0]} ]
80 : ( ref $_[0] eq 'REF' and ref ${$_[0]} eq 'ARRAY' ) ? [ @${ $_[0] } ]
84 # FIXME XSify - this can be done so much more efficiently
85 sub is_plain_value ($) {
87 ! length ref $_[0] ? \($_[0])
89 ref $_[0] eq 'HASH' and keys %{$_[0]} == 1
91 exists $_[0]->{-value}
92 ) ? \($_[0]->{-value})
94 # reuse @_ for even moar speedz
95 defined ( $_[1] = Scalar::Util::blessed $_[0] )
97 # deliberately not using Devel::OverloadInfo - the checks we are
98 # intersted in are much more limited than the fullblown thing, and
99 # this is a very hot piece of code
101 # simply using ->can('(""') can leave behind stub methods that
102 # break actually using the overload later (see L<perldiag/Stub
103 # found while resolving method "%s" overloading "%s" in package
104 # "%s"> and the source of overload::mycan())
106 # either has stringification which DBI SHOULD prefer out of the box
107 grep { *{ (qq[${_}::(""]) }{CODE} } @{ $_[2] = mro::get_linear_isa( $_[1] ) }
109 # has nummification or boolification, AND fallback is *not* disabled
111 SQL::Abstract::_ENV_::DETECT_AUTOGENERATED_STRINGIFICATION
114 grep { *{"${_}::(0+"}{CODE} } @{$_[2]}
116 grep { *{"${_}::(bool"}{CODE} } @{$_[2]}
120 # no fallback specified at all
121 ! ( ($_[3]) = grep { *{"${_}::()"}{CODE} } @{$_[2]} )
123 # fallback explicitly undef
124 ! defined ${"$_[3]::()"}
137 #======================================================================
139 #======================================================================
143 my $class = ref($self) || $self;
144 my %opt = (ref $_[0] eq 'HASH') ? %{$_[0]} : @_;
146 # choose our case by keeping an option around
147 delete $opt{case} if $opt{case} && $opt{case} ne 'lower';
149 # default logic for interpreting arrayrefs
150 $opt{logic} = $opt{logic} ? uc $opt{logic} : 'OR';
152 # how to return bind vars
153 $opt{bindtype} ||= 'normal';
155 # default comparison is "=", but can be overridden
158 # try to recognize which are the 'equality' and 'inequality' ops
159 # (temporary quickfix (in 2007), should go through a more seasoned API)
160 $opt{equality_op} = qr/^( \Q$opt{cmp}\E | \= )$/ix;
161 $opt{inequality_op} = qr/^( != | <> )$/ix;
163 $opt{like_op} = qr/^ (is\s+)? r?like $/xi;
164 $opt{not_like_op} = qr/^ (is\s+)? not \s+ r?like $/xi;
167 $opt{sqltrue} ||= '1=1';
168 $opt{sqlfalse} ||= '0=1';
171 $opt{special_ops} ||= [];
172 # regexes are applied in order, thus push after user-defines
173 push @{$opt{special_ops}}, @BUILTIN_SPECIAL_OPS;
176 $opt{unary_ops} ||= [];
177 push @{$opt{unary_ops}}, @BUILTIN_UNARY_OPS;
179 # rudimentary sanity-check for user supplied bits treated as functions/operators
180 # If a purported function matches this regular expression, an exception is thrown.
181 # Literal SQL is *NOT* subject to this check, only functions (and column names
182 # when quoting is not in effect)
185 # need to guard against ()'s in column names too, but this will break tons of
186 # hacks... ideas anyone?
187 $opt{injection_guard} ||= qr/
193 return bless \%opt, $class;
197 sub _assert_pass_injection_guard {
198 if ($_[1] =~ $_[0]->{injection_guard}) {
199 my $class = ref $_[0];
200 puke "Possible SQL injection attempt '$_[1]'. If this is indeed a part of the "
201 . "desired SQL use literal SQL ( \'...' or \[ '...' ] ) or supply your own "
202 . "{injection_guard} attribute to ${class}->new()"
207 #======================================================================
209 #======================================================================
213 my $table = $self->_table(shift);
214 my $data = shift || return;
217 my $method = $self->_METHOD_FOR_refkind("_insert", $data);
218 my ($sql, @bind) = $self->$method($data);
219 $sql = join " ", $self->_sqlcase('insert into'), $table, $sql;
221 if ($options->{returning}) {
222 my ($s, @b) = $self->_returning ($options);
227 return wantarray ? ($sql, @bind) : $sql;
230 # Used by DBIx::Class::SQLMaker->insert
231 sub _insert_returning { shift->_returning(@_) }
234 my ($self, $options) = @_;
236 my $f = $options->{returning};
238 my $fieldlist = $self->_SWITCH_refkind($f, {
239 ARRAYREF => sub {join ', ', map { $self->_quote($_) } @$f;},
240 SCALAR => sub {$self->_quote($f)},
241 SCALARREF => sub {$$f},
243 return $self->_sqlcase(' returning ') . $fieldlist;
246 sub _insert_HASHREF { # explicit list of fields and then values
247 my ($self, $data) = @_;
249 my @fields = sort keys %$data;
251 my ($sql, @bind) = $self->_insert_values($data);
254 $_ = $self->_quote($_) foreach @fields;
255 $sql = "( ".join(", ", @fields).") ".$sql;
257 return ($sql, @bind);
260 sub _insert_ARRAYREF { # just generate values(?,?) part (no list of fields)
261 my ($self, $data) = @_;
263 # no names (arrayref) so can't generate bindtype
264 $self->{bindtype} ne 'columns'
265 or belch "can't do 'columns' bindtype when called with arrayref";
267 # fold the list of values into a hash of column name - value pairs
268 # (where the column names are artificially generated, and their
269 # lexicographical ordering keep the ordering of the original list)
270 my $i = "a"; # incremented values will be in lexicographical order
271 my $data_in_hash = { map { ($i++ => $_) } @$data };
273 return $self->_insert_values($data_in_hash);
276 sub _insert_ARRAYREFREF { # literal SQL with bind
277 my ($self, $data) = @_;
279 my ($sql, @bind) = @${$data};
280 $self->_assert_bindval_matches_bindtype(@bind);
282 return ($sql, @bind);
286 sub _insert_SCALARREF { # literal SQL without bind
287 my ($self, $data) = @_;
293 my ($self, $data) = @_;
295 my (@values, @all_bind);
296 foreach my $column (sort keys %$data) {
297 my $v = $data->{$column};
299 $self->_SWITCH_refkind($v, {
302 if ($self->{array_datatypes}) { # if array datatype are activated
304 push @all_bind, $self->_bindtype($column, $v);
306 else { # else literal SQL with bind
307 my ($sql, @bind) = @$v;
308 $self->_assert_bindval_matches_bindtype(@bind);
310 push @all_bind, @bind;
314 ARRAYREFREF => sub { # literal SQL with bind
315 my ($sql, @bind) = @${$v};
316 $self->_assert_bindval_matches_bindtype(@bind);
318 push @all_bind, @bind;
321 # THINK : anything useful to do with a HASHREF ?
322 HASHREF => sub { # (nothing, but old SQLA passed it through)
323 #TODO in SQLA >= 2.0 it will die instead
324 belch "HASH ref as bind value in insert is not supported";
326 push @all_bind, $self->_bindtype($column, $v);
329 SCALARREF => sub { # literal SQL without bind
333 SCALAR_or_UNDEF => sub {
335 push @all_bind, $self->_bindtype($column, $v);
342 my $sql = $self->_sqlcase('values')." ( ".join(", ", @values)." )";
343 return ($sql, @all_bind);
348 #======================================================================
350 #======================================================================
355 my $table = $self->_table(shift);
356 my $data = shift || return;
360 # first build the 'SET' part of the sql statement
361 my (@set, @all_bind);
362 puke "Unsupported data type specified to \$sql->update"
363 unless ref $data eq 'HASH';
365 for my $k (sort keys %$data) {
368 my $label = $self->_quote($k);
370 $self->_SWITCH_refkind($v, {
372 if ($self->{array_datatypes}) { # array datatype
373 push @set, "$label = ?";
374 push @all_bind, $self->_bindtype($k, $v);
376 else { # literal SQL with bind
377 my ($sql, @bind) = @$v;
378 $self->_assert_bindval_matches_bindtype(@bind);
379 push @set, "$label = $sql";
380 push @all_bind, @bind;
383 ARRAYREFREF => sub { # literal SQL with bind
384 my ($sql, @bind) = @${$v};
385 $self->_assert_bindval_matches_bindtype(@bind);
386 push @set, "$label = $sql";
387 push @all_bind, @bind;
389 SCALARREF => sub { # literal SQL without bind
390 push @set, "$label = $$v";
393 my ($op, $arg, @rest) = %$v;
395 puke 'Operator calls in update must be in the form { -op => $arg }'
396 if (@rest or not $op =~ /^\-(.+)/);
398 local $self->{_nested_func_lhs} = $k;
399 my ($sql, @bind) = $self->_where_unary_op ($1, $arg);
401 push @set, "$label = $sql";
402 push @all_bind, @bind;
404 SCALAR_or_UNDEF => sub {
405 push @set, "$label = ?";
406 push @all_bind, $self->_bindtype($k, $v);
412 my $sql = $self->_sqlcase('update') . " $table " . $self->_sqlcase('set ')
416 my($where_sql, @where_bind) = $self->where($where);
418 push @all_bind, @where_bind;
421 if ($options->{returning}) {
422 my ($returning_sql, @returning_bind) = $self->_returning ($options);
423 $sql .= $returning_sql;
424 push @all_bind, @returning_bind;
427 return wantarray ? ($sql, @all_bind) : $sql;
433 #======================================================================
435 #======================================================================
440 my $table = $self->_table(shift);
441 my $fields = shift || '*';
445 my($where_sql, @bind) = $self->where($where, $order);
447 my $f = (ref $fields eq 'ARRAY') ? join ', ', map { $self->_quote($_) } @$fields
449 my $sql = join(' ', $self->_sqlcase('select'), $f,
450 $self->_sqlcase('from'), $table)
453 return wantarray ? ($sql, @bind) : $sql;
456 #======================================================================
458 #======================================================================
463 my $table = $self->_table(shift);
467 my($where_sql, @bind) = $self->where($where);
468 my $sql = $self->_sqlcase('delete from') . " $table" . $where_sql;
470 return wantarray ? ($sql, @bind) : $sql;
474 #======================================================================
476 #======================================================================
480 # Finally, a separate routine just to handle WHERE clauses
482 my ($self, $where, $order) = @_;
485 my ($sql, @bind) = $self->_recurse_where($where);
486 $sql = $sql ? $self->_sqlcase(' where ') . "( $sql )" : '';
490 $sql .= $self->_order_by($order);
493 return wantarray ? ($sql, @bind) : $sql;
498 my ($self, $where, $logic) = @_;
500 # dispatch on appropriate method according to refkind of $where
501 my $method = $self->_METHOD_FOR_refkind("_where", $where);
503 my ($sql, @bind) = $self->$method($where, $logic);
505 # DBIx::Class used to call _recurse_where in scalar context
506 # something else might too...
508 return ($sql, @bind);
511 belch "Calling _recurse_where in scalar context is deprecated and will go away before 2.0";
518 #======================================================================
519 # WHERE: top-level ARRAYREF
520 #======================================================================
523 sub _where_ARRAYREF {
524 my ($self, $where, $logic) = @_;
526 $logic = uc($logic || $self->{logic});
527 $logic eq 'AND' or $logic eq 'OR' or puke "unknown logic: $logic";
529 my @clauses = @$where;
531 my (@sql_clauses, @all_bind);
532 # need to use while() so can shift() for pairs
534 my $el = shift @clauses;
536 $el = undef if (defined $el and ! length $el);
538 # switch according to kind of $el and get corresponding ($sql, @bind)
539 my ($sql, @bind) = $self->_SWITCH_refkind($el, {
541 # skip empty elements, otherwise get invalid trailing AND stuff
542 ARRAYREF => sub {$self->_recurse_where($el) if @$el},
546 $self->_assert_bindval_matches_bindtype(@b);
550 HASHREF => sub {$self->_recurse_where($el, 'and') if %$el},
552 SCALARREF => sub { ($$el); },
555 # top-level arrayref with scalars, recurse in pairs
556 $self->_recurse_where({$el => shift(@clauses)})
559 UNDEF => sub {puke "Supplying an empty left hand side argument is not supported in array-pairs" },
563 push @sql_clauses, $sql;
564 push @all_bind, @bind;
568 return $self->_join_sql_clauses($logic, \@sql_clauses, \@all_bind);
571 #======================================================================
572 # WHERE: top-level ARRAYREFREF
573 #======================================================================
575 sub _where_ARRAYREFREF {
576 my ($self, $where) = @_;
577 my ($sql, @bind) = @$$where;
578 $self->_assert_bindval_matches_bindtype(@bind);
579 return ($sql, @bind);
582 #======================================================================
583 # WHERE: top-level HASHREF
584 #======================================================================
587 my ($self, $where) = @_;
588 my (@sql_clauses, @all_bind);
590 for my $k (sort keys %$where) {
591 my $v = $where->{$k};
593 # ($k => $v) is either a special unary op or a regular hashpair
594 my ($sql, @bind) = do {
596 # put the operator in canonical form
598 $op = substr $op, 1; # remove initial dash
599 $op =~ s/^\s+|\s+$//g;# remove leading/trailing space
600 $op =~ s/\s+/ /g; # compress whitespace
602 # so that -not_foo works correctly
603 $op =~ s/^not_/NOT /i;
605 $self->_debug("Unary OP(-$op) within hashref, recursing...");
606 my ($s, @b) = $self->_where_unary_op ($op, $v);
608 # top level vs nested
609 # we assume that handled unary ops will take care of their ()s
611 List::Util::first {$op =~ $_->{regex}} @{$self->{unary_ops}}
613 ( defined $self->{_nested_func_lhs} and $self->{_nested_func_lhs} eq $k )
619 if (is_literal_value ($v) ) {
620 belch 'Hash-pairs consisting of an empty string with a literal are deprecated, and will be removed in 2.0: use -and => [ $literal ] instead';
623 puke "Supplying an empty left hand side argument is not supported in hash-pairs";
627 my $method = $self->_METHOD_FOR_refkind("_where_hashpair", $v);
628 $self->$method($k, $v);
632 push @sql_clauses, $sql;
633 push @all_bind, @bind;
636 return $self->_join_sql_clauses('and', \@sql_clauses, \@all_bind);
639 sub _where_unary_op {
640 my ($self, $op, $rhs) = @_;
642 # top level special ops are illegal in general
643 # this includes the -ident/-value ops (dual purpose unary and special)
644 puke "Illegal use of top-level '-$op'"
645 if ! defined $self->{_nested_func_lhs} and List::Util::first {$op =~ $_->{regex}} @{$self->{special_ops}};
647 if (my $op_entry = List::Util::first {$op =~ $_->{regex}} @{$self->{unary_ops}}) {
648 my $handler = $op_entry->{handler};
650 if (not ref $handler) {
651 if ($op =~ s/ [_\s]? \d+ $//x ) {
652 belch 'Use of [and|or|nest]_N modifiers is deprecated and will be removed in SQLA v2.0. '
653 . "You probably wanted ...-and => [ -$op => COND1, -$op => COND2 ... ]";
655 return $self->$handler ($op, $rhs);
657 elsif (ref $handler eq 'CODE') {
658 return $handler->($self, $op, $rhs);
661 puke "Illegal handler for operator $op - expecting a method name or a coderef";
665 $self->_debug("Generic unary OP: $op - recursing as function");
667 $self->_assert_pass_injection_guard($op);
669 my ($sql, @bind) = $self->_SWITCH_refkind ($rhs, {
671 puke "Illegal use of top-level '-$op'"
672 unless defined $self->{_nested_func_lhs};
675 $self->_convert('?'),
676 $self->_bindtype($self->{_nested_func_lhs}, $rhs)
680 $self->_recurse_where ($rhs)
684 $sql = sprintf ('%s %s',
685 $self->_sqlcase($op),
689 return ($sql, @bind);
692 sub _where_op_ANDOR {
693 my ($self, $op, $v) = @_;
695 $self->_SWITCH_refkind($v, {
697 return $self->_where_ARRAYREF($v, $op);
701 return ( $op =~ /^or/i )
702 ? $self->_where_ARRAYREF( [ map { $_ => $v->{$_} } ( sort keys %$v ) ], $op )
703 : $self->_where_HASHREF($v);
707 puke "-$op => \\\$scalar makes little sense, use " .
709 ? '[ \$scalar, \%rest_of_conditions ] instead'
710 : '-and => [ \$scalar, \%rest_of_conditions ] instead'
715 puke "-$op => \\[...] makes little sense, use " .
717 ? '[ \[...], \%rest_of_conditions ] instead'
718 : '-and => [ \[...], \%rest_of_conditions ] instead'
722 SCALAR => sub { # permissively interpreted as SQL
723 puke "-$op => \$value makes little sense, use -bool => \$value instead";
727 puke "-$op => undef not supported";
733 my ($self, $op, $v) = @_;
735 $self->_SWITCH_refkind($v, {
737 SCALAR => sub { # permissively interpreted as SQL
738 belch "literal SQL should be -nest => \\'scalar' "
739 . "instead of -nest => 'scalar' ";
744 puke "-$op => undef not supported";
748 $self->_recurse_where ($v);
756 my ($self, $op, $v) = @_;
758 my ($s, @b) = $self->_SWITCH_refkind($v, {
759 SCALAR => sub { # interpreted as SQL column
760 $self->_convert($self->_quote($v));
764 puke "-$op => undef not supported";
768 $self->_recurse_where ($v);
772 $s = "(NOT $s)" if $op =~ /^not/i;
777 sub _where_op_IDENT {
779 my ($op, $rhs) = splice @_, -2;
780 if (! defined $rhs or length ref $rhs) {
781 puke "-$op requires a single plain scalar argument (a quotable identifier)";
784 # in case we are called as a top level special op (no '=')
787 $_ = $self->_convert($self->_quote($_)) for ($lhs, $rhs);
795 sub _where_op_VALUE {
797 my ($op, $rhs) = splice @_, -2;
799 # in case we are called as a top level special op (no '=')
803 if (! defined $rhs) {
805 ? $self->_convert($self->_quote($lhs)) . ' IS NULL'
812 ( defined $lhs ? $lhs : $self->{_nested_func_lhs} ),
819 $self->_convert($self->_quote($lhs)) . ' = ' . $self->_convert('?'),
823 $self->_convert('?'),
829 sub _where_hashpair_ARRAYREF {
830 my ($self, $k, $v) = @_;
833 my @v = @$v; # need copy because of shift below
834 $self->_debug("ARRAY($k) means distribute over elements");
836 # put apart first element if it is an operator (-and, -or)
838 (defined $v[0] && $v[0] =~ /^ - (?: AND|OR ) $/ix)
842 my @distributed = map { {$k => $_} } @v;
845 $self->_debug("OP($op) reinjected into the distributed array");
846 unshift @distributed, $op;
849 my $logic = $op ? substr($op, 1) : '';
851 return $self->_recurse_where(\@distributed, $logic);
854 $self->_debug("empty ARRAY($k) means 0=1");
855 return ($self->{sqlfalse});
859 sub _where_hashpair_HASHREF {
860 my ($self, $k, $v, $logic) = @_;
863 local $self->{_nested_func_lhs} = defined $self->{_nested_func_lhs}
864 ? $self->{_nested_func_lhs}
868 my ($all_sql, @all_bind);
870 for my $orig_op (sort keys %$v) {
871 my $val = $v->{$orig_op};
873 # put the operator in canonical form
876 # FIXME - we need to phase out dash-less ops
877 $op =~ s/^-//; # remove possible initial dash
878 $op =~ s/^\s+|\s+$//g;# remove leading/trailing space
879 $op =~ s/\s+/ /g; # compress whitespace
881 $self->_assert_pass_injection_guard($op);
884 $op =~ s/^is_not/IS NOT/i;
886 # so that -not_foo works correctly
887 $op =~ s/^not_/NOT /i;
889 # another retarded special case: foo => { $op => { -value => undef } }
890 if (ref $val eq 'HASH' and keys %$val == 1 and exists $val->{-value} and ! defined $val->{-value} ) {
896 # CASE: col-value logic modifiers
897 if ( $orig_op =~ /^ \- (and|or) $/xi ) {
898 ($sql, @bind) = $self->_where_hashpair_HASHREF($k, $val, $1);
900 # CASE: special operators like -in or -between
901 elsif ( my $special_op = List::Util::first {$op =~ $_->{regex}} @{$self->{special_ops}} ) {
902 my $handler = $special_op->{handler};
904 puke "No handler supplied for special operator $orig_op";
906 elsif (not ref $handler) {
907 ($sql, @bind) = $self->$handler ($k, $op, $val);
909 elsif (ref $handler eq 'CODE') {
910 ($sql, @bind) = $handler->($self, $k, $op, $val);
913 puke "Illegal handler for special operator $orig_op - expecting a method name or a coderef";
917 $self->_SWITCH_refkind($val, {
919 ARRAYREF => sub { # CASE: col => {op => \@vals}
920 ($sql, @bind) = $self->_where_field_op_ARRAYREF($k, $op, $val);
923 ARRAYREFREF => sub { # CASE: col => {op => \[$sql, @bind]} (literal SQL with bind)
924 my ($sub_sql, @sub_bind) = @$$val;
925 $self->_assert_bindval_matches_bindtype(@sub_bind);
926 $sql = join ' ', $self->_convert($self->_quote($k)),
927 $self->_sqlcase($op),
932 UNDEF => sub { # CASE: col => {op => undef} : sql "IS (NOT)? NULL"
934 $op =~ /^not$/i ? 'is not' # legacy
935 : $op =~ $self->{equality_op} ? 'is'
936 : $op =~ $self->{like_op} ? belch("Supplying an undefined argument to '@{[ uc $op]}' is deprecated") && 'is'
937 : $op =~ $self->{inequality_op} ? 'is not'
938 : $op =~ $self->{not_like_op} ? belch("Supplying an undefined argument to '@{[ uc $op]}' is deprecated") && 'is not'
939 : puke "unexpected operator '$orig_op' with undef operand";
941 $sql = $self->_quote($k) . $self->_sqlcase(" $is null");
944 FALLBACK => sub { # CASE: col => {op/func => $stuff}
945 ($sql, @bind) = $self->_where_unary_op ($op, $val);
948 $self->_convert($self->_quote($k)),
949 $self->{_nested_func_lhs} eq $k ? $sql : "($sql)", # top level vs nested
955 ($all_sql) = (defined $all_sql and $all_sql) ? $self->_join_sql_clauses($logic, [$all_sql, $sql], []) : $sql;
956 push @all_bind, @bind;
958 return ($all_sql, @all_bind);
961 sub _where_field_IS {
962 my ($self, $k, $op, $v) = @_;
964 my ($s) = $self->_SWITCH_refkind($v, {
967 $self->_convert($self->_quote($k)),
968 map { $self->_sqlcase($_)} ($op, 'null')
971 puke "$op can only take undef as argument";
978 sub _where_field_op_ARRAYREF {
979 my ($self, $k, $op, $vals) = @_;
981 my @vals = @$vals; #always work on a copy
984 $self->_debug(sprintf '%s means multiple elements: [ %s ]',
986 join (', ', map { defined $_ ? "'$_'" : 'NULL' } @vals ),
989 # see if the first element is an -and/-or op
991 if (defined $vals[0] && $vals[0] =~ /^ - ( AND|OR ) $/ix) {
996 # a long standing API wart - an attempt to change this behavior during
997 # the 1.50 series failed *spectacularly*. Warn instead and leave the
1002 (!$logic or $logic eq 'OR')
1004 ( $op =~ $self->{inequality_op} or $op =~ $self->{not_like_op} )
1007 belch "A multi-element arrayref as an argument to the inequality op '$o' "
1008 . 'is technically equivalent to an always-true 1=1 (you probably wanted '
1009 . "to say ...{ \$inequality_op => [ -and => \@values ] }... instead)"
1013 # distribute $op over each remaining member of @vals, append logic if exists
1014 return $self->_recurse_where([map { {$k => {$op, $_}} } @vals], $logic);
1018 # try to DWIM on equality operators
1020 $op =~ $self->{equality_op} ? $self->{sqlfalse}
1021 : $op =~ $self->{like_op} ? belch("Supplying an empty arrayref to '@{[ uc $op]}' is deprecated") && $self->{sqlfalse}
1022 : $op =~ $self->{inequality_op} ? $self->{sqltrue}
1023 : $op =~ $self->{not_like_op} ? belch("Supplying an empty arrayref to '@{[ uc $op]}' is deprecated") && $self->{sqltrue}
1024 : puke "operator '$op' applied on an empty array (field '$k')";
1029 sub _where_hashpair_SCALARREF {
1030 my ($self, $k, $v) = @_;
1031 $self->_debug("SCALAR($k) means literal SQL: $$v");
1032 my $sql = $self->_quote($k) . " " . $$v;
1036 # literal SQL with bind
1037 sub _where_hashpair_ARRAYREFREF {
1038 my ($self, $k, $v) = @_;
1039 $self->_debug("REF($k) means literal SQL: @${$v}");
1040 my ($sql, @bind) = @$$v;
1041 $self->_assert_bindval_matches_bindtype(@bind);
1042 $sql = $self->_quote($k) . " " . $sql;
1043 return ($sql, @bind );
1046 # literal SQL without bind
1047 sub _where_hashpair_SCALAR {
1048 my ($self, $k, $v) = @_;
1049 $self->_debug("NOREF($k) means simple key=val: $k $self->{cmp} $v");
1050 my $sql = join ' ', $self->_convert($self->_quote($k)),
1051 $self->_sqlcase($self->{cmp}),
1052 $self->_convert('?');
1053 my @bind = $self->_bindtype($k, $v);
1054 return ( $sql, @bind);
1058 sub _where_hashpair_UNDEF {
1059 my ($self, $k, $v) = @_;
1060 $self->_debug("UNDEF($k) means IS NULL");
1061 my $sql = $self->_quote($k) . $self->_sqlcase(' is null');
1065 #======================================================================
1066 # WHERE: TOP-LEVEL OTHERS (SCALARREF, SCALAR, UNDEF)
1067 #======================================================================
1070 sub _where_SCALARREF {
1071 my ($self, $where) = @_;
1074 $self->_debug("SCALAR(*top) means literal SQL: $$where");
1080 my ($self, $where) = @_;
1083 $self->_debug("NOREF(*top) means literal SQL: $where");
1094 #======================================================================
1095 # WHERE: BUILTIN SPECIAL OPERATORS (-in, -between)
1096 #======================================================================
1099 sub _where_field_BETWEEN {
1100 my ($self, $k, $op, $vals) = @_;
1102 my ($label, $and, $placeholder);
1103 $label = $self->_convert($self->_quote($k));
1104 $and = ' ' . $self->_sqlcase('and') . ' ';
1105 $placeholder = $self->_convert('?');
1106 $op = $self->_sqlcase($op);
1108 my $invalid_args = "Operator '$op' requires either an arrayref with two defined values or expressions, or a single literal scalarref/arrayref-ref";
1110 my ($clause, @bind) = $self->_SWITCH_refkind($vals, {
1111 ARRAYREFREF => sub {
1112 my ($s, @b) = @$$vals;
1113 $self->_assert_bindval_matches_bindtype(@b);
1120 puke $invalid_args if @$vals != 2;
1122 my (@all_sql, @all_bind);
1123 foreach my $val (@$vals) {
1124 my ($sql, @bind) = $self->_SWITCH_refkind($val, {
1126 return ($placeholder, $self->_bindtype($k, $val) );
1131 ARRAYREFREF => sub {
1132 my ($sql, @bind) = @$$val;
1133 $self->_assert_bindval_matches_bindtype(@bind);
1134 return ($sql, @bind);
1137 my ($func, $arg, @rest) = %$val;
1138 puke ("Only simple { -func => arg } functions accepted as sub-arguments to BETWEEN")
1139 if (@rest or $func !~ /^ \- (.+)/x);
1140 $self->_where_unary_op ($1 => $arg);
1146 push @all_sql, $sql;
1147 push @all_bind, @bind;
1151 (join $and, @all_sql),
1160 my $sql = "( $label $op $clause )";
1161 return ($sql, @bind)
1165 sub _where_field_IN {
1166 my ($self, $k, $op, $vals) = @_;
1168 # backwards compatibility : if scalar, force into an arrayref
1169 $vals = [$vals] if defined $vals && ! ref $vals;
1171 my ($label) = $self->_convert($self->_quote($k));
1172 my ($placeholder) = $self->_convert('?');
1173 $op = $self->_sqlcase($op);
1175 my ($sql, @bind) = $self->_SWITCH_refkind($vals, {
1176 ARRAYREF => sub { # list of choices
1177 if (@$vals) { # nonempty list
1178 my (@all_sql, @all_bind);
1180 for my $val (@$vals) {
1181 my ($sql, @bind) = $self->_SWITCH_refkind($val, {
1183 return ($placeholder, $val);
1188 ARRAYREFREF => sub {
1189 my ($sql, @bind) = @$$val;
1190 $self->_assert_bindval_matches_bindtype(@bind);
1191 return ($sql, @bind);
1194 my ($func, $arg, @rest) = %$val;
1195 puke ("Only simple { -func => arg } functions accepted as sub-arguments to IN")
1196 if (@rest or $func !~ /^ \- (.+)/x);
1197 $self->_where_unary_op ($1 => $arg);
1201 'SQL::Abstract before v1.75 used to generate incorrect SQL when the '
1202 . "-$op operator was given an undef-containing list: !!!AUDIT YOUR CODE "
1203 . 'AND DATA!!! (the upcoming Data::Query-based version of SQL::Abstract '
1204 . 'will emit the logically correct SQL instead of raising this exception)'
1208 push @all_sql, $sql;
1209 push @all_bind, @bind;
1213 sprintf ('%s %s ( %s )',
1216 join (', ', @all_sql)
1218 $self->_bindtype($k, @all_bind),
1221 else { # empty list : some databases won't understand "IN ()", so DWIM
1222 my $sql = ($op =~ /\bnot\b/i) ? $self->{sqltrue} : $self->{sqlfalse};
1227 SCALARREF => sub { # literal SQL
1228 my $sql = $self->_open_outer_paren ($$vals);
1229 return ("$label $op ( $sql )");
1231 ARRAYREFREF => sub { # literal SQL with bind
1232 my ($sql, @bind) = @$$vals;
1233 $self->_assert_bindval_matches_bindtype(@bind);
1234 $sql = $self->_open_outer_paren ($sql);
1235 return ("$label $op ( $sql )", @bind);
1239 puke "Argument passed to the '$op' operator can not be undefined";
1243 puke "special op $op requires an arrayref (or scalarref/arrayref-ref)";
1247 return ($sql, @bind);
1250 # Some databases (SQLite) treat col IN (1, 2) different from
1251 # col IN ( (1, 2) ). Use this to strip all outer parens while
1252 # adding them back in the corresponding method
1253 sub _open_outer_paren {
1254 my ($self, $sql) = @_;
1256 while ( my ($inner) = $sql =~ /^ \s* \( (.*) \) \s* $/xs ) {
1258 # there are closing parens inside, need the heavy duty machinery
1259 # to reevaluate the extraction starting from $sql (full reevaluation)
1260 if ( $inner =~ /\)/ ) {
1261 require Text::Balanced;
1263 my (undef, $remainder) = do {
1264 # idiotic design - writes to $@ but *DOES NOT* throw exceptions
1266 Text::Balanced::extract_bracketed( $sql, '()', qr/\s*/ );
1269 # the entire expression needs to be a balanced bracketed thing
1270 # (after an extract no remainder sans trailing space)
1271 last if defined $remainder and $remainder =~ /\S/;
1281 #======================================================================
1283 #======================================================================
1286 my ($self, $arg) = @_;
1289 for my $c ($self->_order_by_chunks ($arg) ) {
1290 $self->_SWITCH_refkind ($c, {
1291 SCALAR => sub { push @sql, $c },
1292 ARRAYREF => sub { push @sql, shift @$c; push @bind, @$c },
1298 $self->_sqlcase(' order by'),
1304 return wantarray ? ($sql, @bind) : $sql;
1307 sub _order_by_chunks {
1308 my ($self, $arg) = @_;
1310 return $self->_SWITCH_refkind($arg, {
1313 map { $self->_order_by_chunks ($_ ) } @$arg;
1316 ARRAYREFREF => sub {
1317 my ($s, @b) = @$$arg;
1318 $self->_assert_bindval_matches_bindtype(@b);
1322 SCALAR => sub {$self->_quote($arg)},
1324 UNDEF => sub {return () },
1326 SCALARREF => sub {$$arg}, # literal SQL, no quoting
1329 # get first pair in hash
1330 my ($key, $val, @rest) = %$arg;
1332 return () unless $key;
1334 if ( @rest or not $key =~ /^-(desc|asc)/i ) {
1335 puke "hash passed to _order_by must have exactly one key (-desc or -asc)";
1341 for my $c ($self->_order_by_chunks ($val)) {
1344 $self->_SWITCH_refkind ($c, {
1349 ($sql, @bind) = @$c;
1353 $sql = $sql . ' ' . $self->_sqlcase($direction);
1355 push @ret, [ $sql, @bind];
1364 #======================================================================
1365 # DATASOURCE (FOR NOW, JUST PLAIN TABLE OR LIST OF TABLES)
1366 #======================================================================
1371 $self->_SWITCH_refkind($from, {
1372 ARRAYREF => sub {join ', ', map { $self->_quote($_) } @$from;},
1373 SCALAR => sub {$self->_quote($from)},
1374 SCALARREF => sub {$$from},
1379 #======================================================================
1381 #======================================================================
1383 # highly optimized, as it's called way too often
1385 # my ($self, $label) = @_;
1387 return '' unless defined $_[1];
1388 return ${$_[1]} if ref($_[1]) eq 'SCALAR';
1390 unless ($_[0]->{quote_char}) {
1391 $_[0]->_assert_pass_injection_guard($_[1]);
1395 my $qref = ref $_[0]->{quote_char};
1398 ($l, $r) = ( $_[0]->{quote_char}, $_[0]->{quote_char} );
1400 elsif ($qref eq 'ARRAY') {
1401 ($l, $r) = @{$_[0]->{quote_char}};
1404 puke "Unsupported quote_char format: $_[0]->{quote_char}";
1406 my $esc = $_[0]->{escape_char} || $r;
1408 # parts containing * are naturally unquoted
1409 return join( $_[0]->{name_sep}||'', map
1410 { $_ eq '*' ? $_ : do { (my $n = $_) =~ s/(\Q$esc\E|\Q$r\E)/$esc$1/g; $l . $n . $r } }
1411 ( $_[0]->{name_sep} ? split (/\Q$_[0]->{name_sep}\E/, $_[1] ) : $_[1] )
1416 # Conversion, if applicable
1418 #my ($self, $arg) = @_;
1419 if ($_[0]->{convert}) {
1420 return $_[0]->_sqlcase($_[0]->{convert}) .'(' . $_[1] . ')';
1427 #my ($self, $col, @vals) = @_;
1428 # called often - tighten code
1429 return $_[0]->{bindtype} eq 'columns'
1430 ? map {[$_[1], $_]} @_[2 .. $#_]
1435 # Dies if any element of @bind is not in [colname => value] format
1436 # if bindtype is 'columns'.
1437 sub _assert_bindval_matches_bindtype {
1438 # my ($self, @bind) = @_;
1440 if ($self->{bindtype} eq 'columns') {
1442 if (!defined $_ || ref($_) ne 'ARRAY' || @$_ != 2) {
1443 puke "bindtype 'columns' selected, you need to pass: [column_name => bind_value]"
1449 sub _join_sql_clauses {
1450 my ($self, $logic, $clauses_aref, $bind_aref) = @_;
1452 if (@$clauses_aref > 1) {
1453 my $join = " " . $self->_sqlcase($logic) . " ";
1454 my $sql = '( ' . join($join, @$clauses_aref) . ' )';
1455 return ($sql, @$bind_aref);
1457 elsif (@$clauses_aref) {
1458 return ($clauses_aref->[0], @$bind_aref); # no parentheses
1461 return (); # if no SQL, ignore @$bind_aref
1466 # Fix SQL case, if so requested
1468 # LDNOTE: if $self->{case} is true, then it contains 'lower', so we
1469 # don't touch the argument ... crooked logic, but let's not change it!
1470 return $_[0]->{case} ? $_[1] : uc($_[1]);
1474 #======================================================================
1475 # DISPATCHING FROM REFKIND
1476 #======================================================================
1479 my ($self, $data) = @_;
1481 return 'UNDEF' unless defined $data;
1483 # blessed objects are treated like scalars
1484 my $ref = (Scalar::Util::blessed $data) ? '' : ref $data;
1486 return 'SCALAR' unless $ref;
1489 while ($ref eq 'REF') {
1491 $ref = (Scalar::Util::blessed $data) ? '' : ref $data;
1495 return ($ref||'SCALAR') . ('REF' x $n_steps);
1499 my ($self, $data) = @_;
1500 my @try = ($self->_refkind($data));
1501 push @try, 'SCALAR_or_UNDEF' if $try[0] eq 'SCALAR' || $try[0] eq 'UNDEF';
1502 push @try, 'FALLBACK';
1506 sub _METHOD_FOR_refkind {
1507 my ($self, $meth_prefix, $data) = @_;
1510 for (@{$self->_try_refkind($data)}) {
1511 $method = $self->can($meth_prefix."_".$_)
1515 return $method || puke "cannot dispatch on '$meth_prefix' for ".$self->_refkind($data);
1519 sub _SWITCH_refkind {
1520 my ($self, $data, $dispatch_table) = @_;
1523 for (@{$self->_try_refkind($data)}) {
1524 $coderef = $dispatch_table->{$_}
1528 puke "no dispatch entry for ".$self->_refkind($data)
1537 #======================================================================
1538 # VALUES, GENERATE, AUTOLOAD
1539 #======================================================================
1541 # LDNOTE: original code from nwiger, didn't touch code in that section
1542 # I feel the AUTOLOAD stuff should not be the default, it should
1543 # only be activated on explicit demand by user.
1547 my $data = shift || return;
1548 puke "Argument to ", __PACKAGE__, "->values must be a \\%hash"
1549 unless ref $data eq 'HASH';
1552 foreach my $k ( sort keys %$data ) {
1553 my $v = $data->{$k};
1554 $self->_SWITCH_refkind($v, {
1556 if ($self->{array_datatypes}) { # array datatype
1557 push @all_bind, $self->_bindtype($k, $v);
1559 else { # literal SQL with bind
1560 my ($sql, @bind) = @$v;
1561 $self->_assert_bindval_matches_bindtype(@bind);
1562 push @all_bind, @bind;
1565 ARRAYREFREF => sub { # literal SQL with bind
1566 my ($sql, @bind) = @${$v};
1567 $self->_assert_bindval_matches_bindtype(@bind);
1568 push @all_bind, @bind;
1570 SCALARREF => sub { # literal SQL without bind
1572 SCALAR_or_UNDEF => sub {
1573 push @all_bind, $self->_bindtype($k, $v);
1584 my(@sql, @sqlq, @sqlv);
1588 if ($ref eq 'HASH') {
1589 for my $k (sort keys %$_) {
1592 my $label = $self->_quote($k);
1593 if ($r eq 'ARRAY') {
1594 # literal SQL with bind
1595 my ($sql, @bind) = @$v;
1596 $self->_assert_bindval_matches_bindtype(@bind);
1597 push @sqlq, "$label = $sql";
1599 } elsif ($r eq 'SCALAR') {
1600 # literal SQL without bind
1601 push @sqlq, "$label = $$v";
1603 push @sqlq, "$label = ?";
1604 push @sqlv, $self->_bindtype($k, $v);
1607 push @sql, $self->_sqlcase('set'), join ', ', @sqlq;
1608 } elsif ($ref eq 'ARRAY') {
1609 # unlike insert(), assume these are ONLY the column names, i.e. for SQL
1612 if ($r eq 'ARRAY') { # literal SQL with bind
1613 my ($sql, @bind) = @$v;
1614 $self->_assert_bindval_matches_bindtype(@bind);
1617 } elsif ($r eq 'SCALAR') { # literal SQL without bind
1618 # embedded literal SQL
1625 push @sql, '(' . join(', ', @sqlq) . ')';
1626 } elsif ($ref eq 'SCALAR') {
1630 # strings get case twiddled
1631 push @sql, $self->_sqlcase($_);
1635 my $sql = join ' ', @sql;
1637 # this is pretty tricky
1638 # if ask for an array, return ($stmt, @bind)
1639 # otherwise, s/?/shift @sqlv/ to put it inline
1641 return ($sql, @sqlv);
1643 1 while $sql =~ s/\?/my $d = shift(@sqlv);
1644 ref $d ? $d->[1] : $d/e;
1653 # This allows us to check for a local, then _form, attr
1655 my($name) = $AUTOLOAD =~ /.*::(.+)/;
1656 return $self->generate($name, @_);
1667 SQL::Abstract - Generate SQL from Perl data structures
1673 my $sql = SQL::Abstract->new;
1675 my($stmt, @bind) = $sql->select($source, \@fields, \%where, \@order);
1677 my($stmt, @bind) = $sql->insert($table, \%fieldvals || \@values);
1679 my($stmt, @bind) = $sql->update($table, \%fieldvals, \%where);
1681 my($stmt, @bind) = $sql->delete($table, \%where);
1683 # Then, use these in your DBI statements
1684 my $sth = $dbh->prepare($stmt);
1685 $sth->execute(@bind);
1687 # Just generate the WHERE clause
1688 my($stmt, @bind) = $sql->where(\%where, \@order);
1690 # Return values in the same order, for hashed queries
1691 # See PERFORMANCE section for more details
1692 my @bind = $sql->values(\%fieldvals);
1696 This module was inspired by the excellent L<DBIx::Abstract>.
1697 However, in using that module I found that what I really wanted
1698 to do was generate SQL, but still retain complete control over my
1699 statement handles and use the DBI interface. So, I set out to
1700 create an abstract SQL generation module.
1702 While based on the concepts used by L<DBIx::Abstract>, there are
1703 several important differences, especially when it comes to WHERE
1704 clauses. I have modified the concepts used to make the SQL easier
1705 to generate from Perl data structures and, IMO, more intuitive.
1706 The underlying idea is for this module to do what you mean, based
1707 on the data structures you provide it. The big advantage is that
1708 you don't have to modify your code every time your data changes,
1709 as this module figures it out.
1711 To begin with, an SQL INSERT is as easy as just specifying a hash
1712 of C<key=value> pairs:
1715 name => 'Jimbo Bobson',
1716 phone => '123-456-7890',
1717 address => '42 Sister Lane',
1718 city => 'St. Louis',
1719 state => 'Louisiana',
1722 The SQL can then be generated with this:
1724 my($stmt, @bind) = $sql->insert('people', \%data);
1726 Which would give you something like this:
1728 $stmt = "INSERT INTO people
1729 (address, city, name, phone, state)
1730 VALUES (?, ?, ?, ?, ?)";
1731 @bind = ('42 Sister Lane', 'St. Louis', 'Jimbo Bobson',
1732 '123-456-7890', 'Louisiana');
1734 These are then used directly in your DBI code:
1736 my $sth = $dbh->prepare($stmt);
1737 $sth->execute(@bind);
1739 =head2 Inserting and Updating Arrays
1741 If your database has array types (like for example Postgres),
1742 activate the special option C<< array_datatypes => 1 >>
1743 when creating the C<SQL::Abstract> object.
1744 Then you may use an arrayref to insert and update database array types:
1746 my $sql = SQL::Abstract->new(array_datatypes => 1);
1748 planets => [qw/Mercury Venus Earth Mars/]
1751 my($stmt, @bind) = $sql->insert('solar_system', \%data);
1755 $stmt = "INSERT INTO solar_system (planets) VALUES (?)"
1757 @bind = (['Mercury', 'Venus', 'Earth', 'Mars']);
1760 =head2 Inserting and Updating SQL
1762 In order to apply SQL functions to elements of your C<%data> you may
1763 specify a reference to an arrayref for the given hash value. For example,
1764 if you need to execute the Oracle C<to_date> function on a value, you can
1765 say something like this:
1769 date_entered => \[ "to_date(?,'MM/DD/YYYY')", "03/02/2003" ],
1772 The first value in the array is the actual SQL. Any other values are
1773 optional and would be included in the bind values array. This gives
1776 my($stmt, @bind) = $sql->insert('people', \%data);
1778 $stmt = "INSERT INTO people (name, date_entered)
1779 VALUES (?, to_date(?,'MM/DD/YYYY'))";
1780 @bind = ('Bill', '03/02/2003');
1782 An UPDATE is just as easy, all you change is the name of the function:
1784 my($stmt, @bind) = $sql->update('people', \%data);
1786 Notice that your C<%data> isn't touched; the module will generate
1787 the appropriately quirky SQL for you automatically. Usually you'll
1788 want to specify a WHERE clause for your UPDATE, though, which is
1789 where handling C<%where> hashes comes in handy...
1791 =head2 Complex where statements
1793 This module can generate pretty complicated WHERE statements
1794 easily. For example, simple C<key=value> pairs are taken to mean
1795 equality, and if you want to see if a field is within a set
1796 of values, you can use an arrayref. Let's say we wanted to
1797 SELECT some data based on this criteria:
1800 requestor => 'inna',
1801 worker => ['nwiger', 'rcwe', 'sfz'],
1802 status => { '!=', 'completed' }
1805 my($stmt, @bind) = $sql->select('tickets', '*', \%where);
1807 The above would give you something like this:
1809 $stmt = "SELECT * FROM tickets WHERE
1810 ( requestor = ? ) AND ( status != ? )
1811 AND ( worker = ? OR worker = ? OR worker = ? )";
1812 @bind = ('inna', 'completed', 'nwiger', 'rcwe', 'sfz');
1814 Which you could then use in DBI code like so:
1816 my $sth = $dbh->prepare($stmt);
1817 $sth->execute(@bind);
1823 The methods are simple. There's one for every major SQL operation,
1824 and a constructor you use first. The arguments are specified in a
1825 similar order for each method (table, then fields, then a where
1826 clause) to try and simplify things.
1828 =head2 new(option => 'value')
1830 The C<new()> function takes a list of options and values, and returns
1831 a new B<SQL::Abstract> object which can then be used to generate SQL
1832 through the methods below. The options accepted are:
1838 If set to 'lower', then SQL will be generated in all lowercase. By
1839 default SQL is generated in "textbook" case meaning something like:
1841 SELECT a_field FROM a_table WHERE some_field LIKE '%someval%'
1843 Any setting other than 'lower' is ignored.
1847 This determines what the default comparison operator is. By default
1848 it is C<=>, meaning that a hash like this:
1850 %where = (name => 'nwiger', email => 'nate@wiger.org');
1852 Will generate SQL like this:
1854 WHERE name = 'nwiger' AND email = 'nate@wiger.org'
1856 However, you may want loose comparisons by default, so if you set
1857 C<cmp> to C<like> you would get SQL such as:
1859 WHERE name like 'nwiger' AND email like 'nate@wiger.org'
1861 You can also override the comparison on an individual basis - see
1862 the huge section on L</"WHERE CLAUSES"> at the bottom.
1864 =item sqltrue, sqlfalse
1866 Expressions for inserting boolean values within SQL statements.
1867 By default these are C<1=1> and C<1=0>. They are used
1868 by the special operators C<-in> and C<-not_in> for generating
1869 correct SQL even when the argument is an empty array (see below).
1873 This determines the default logical operator for multiple WHERE
1874 statements in arrays or hashes. If absent, the default logic is "or"
1875 for arrays, and "and" for hashes. This means that a WHERE
1879 event_date => {'>=', '2/13/99'},
1880 event_date => {'<=', '4/24/03'},
1883 will generate SQL like this:
1885 WHERE event_date >= '2/13/99' OR event_date <= '4/24/03'
1887 This is probably not what you want given this query, though (look
1888 at the dates). To change the "OR" to an "AND", simply specify:
1890 my $sql = SQL::Abstract->new(logic => 'and');
1892 Which will change the above C<WHERE> to:
1894 WHERE event_date >= '2/13/99' AND event_date <= '4/24/03'
1896 The logic can also be changed locally by inserting
1897 a modifier in front of an arrayref :
1899 @where = (-and => [event_date => {'>=', '2/13/99'},
1900 event_date => {'<=', '4/24/03'} ]);
1902 See the L</"WHERE CLAUSES"> section for explanations.
1906 This will automatically convert comparisons using the specified SQL
1907 function for both column and value. This is mostly used with an argument
1908 of C<upper> or C<lower>, so that the SQL will have the effect of
1909 case-insensitive "searches". For example, this:
1911 $sql = SQL::Abstract->new(convert => 'upper');
1912 %where = (keywords => 'MaKe iT CAse inSeNSItive');
1914 Will turn out the following SQL:
1916 WHERE upper(keywords) like upper('MaKe iT CAse inSeNSItive')
1918 The conversion can be C<upper()>, C<lower()>, or any other SQL function
1919 that can be applied symmetrically to fields (actually B<SQL::Abstract> does
1920 not validate this option; it will just pass through what you specify verbatim).
1924 This is a kludge because many databases suck. For example, you can't
1925 just bind values using DBI's C<execute()> for Oracle C<CLOB> or C<BLOB> fields.
1926 Instead, you have to use C<bind_param()>:
1928 $sth->bind_param(1, 'reg data');
1929 $sth->bind_param(2, $lots, {ora_type => ORA_CLOB});
1931 The problem is, B<SQL::Abstract> will normally just return a C<@bind> array,
1932 which loses track of which field each slot refers to. Fear not.
1934 If you specify C<bindtype> in new, you can determine how C<@bind> is returned.
1935 Currently, you can specify either C<normal> (default) or C<columns>. If you
1936 specify C<columns>, you will get an array that looks like this:
1938 my $sql = SQL::Abstract->new(bindtype => 'columns');
1939 my($stmt, @bind) = $sql->insert(...);
1942 [ 'column1', 'value1' ],
1943 [ 'column2', 'value2' ],
1944 [ 'column3', 'value3' ],
1947 You can then iterate through this manually, using DBI's C<bind_param()>.
1949 $sth->prepare($stmt);
1952 my($col, $data) = @$_;
1953 if ($col eq 'details' || $col eq 'comments') {
1954 $sth->bind_param($i, $data, {ora_type => ORA_CLOB});
1955 } elsif ($col eq 'image') {
1956 $sth->bind_param($i, $data, {ora_type => ORA_BLOB});
1958 $sth->bind_param($i, $data);
1962 $sth->execute; # execute without @bind now
1964 Now, why would you still use B<SQL::Abstract> if you have to do this crap?
1965 Basically, the advantage is still that you don't have to care which fields
1966 are or are not included. You could wrap that above C<for> loop in a simple
1967 sub called C<bind_fields()> or something and reuse it repeatedly. You still
1968 get a layer of abstraction over manual SQL specification.
1970 Note that if you set L</bindtype> to C<columns>, the C<\[ $sql, @bind ]>
1971 construct (see L</Literal SQL with placeholders and bind values (subqueries)>)
1972 will expect the bind values in this format.
1976 This is the character that a table or column name will be quoted
1977 with. By default this is an empty string, but you could set it to
1978 the character C<`>, to generate SQL like this:
1980 SELECT `a_field` FROM `a_table` WHERE `some_field` LIKE '%someval%'
1982 Alternatively, you can supply an array ref of two items, the first being the left
1983 hand quote character, and the second the right hand quote character. For
1984 example, you could supply C<['[',']']> for SQL Server 2000 compliant quotes
1985 that generates SQL like this:
1987 SELECT [a_field] FROM [a_table] WHERE [some_field] LIKE '%someval%'
1989 Quoting is useful if you have tables or columns names that are reserved
1990 words in your database's SQL dialect.
1994 This is the character that will be used to escape L</quote_char>s appearing
1995 in an identifier before it has been quoted.
1997 The parameter default in case of a single L</quote_char> character is the quote
2000 When opening-closing-style quoting is used (L</quote_char> is an arrayref)
2001 this parameter defaults to the B<closing (right)> L</quote_char>. Occurences
2002 of the B<opening (left)> L</quote_char> within the identifier are currently left
2003 untouched. The default for opening-closing-style quotes may change in future
2004 versions, thus you are B<strongly encouraged> to specify the escape character
2009 This is the character that separates a table and column name. It is
2010 necessary to specify this when the C<quote_char> option is selected,
2011 so that tables and column names can be individually quoted like this:
2013 SELECT `table`.`one_field` FROM `table` WHERE `table`.`other_field` = 1
2015 =item injection_guard
2017 A regular expression C<qr/.../> that is applied to any C<-function> and unquoted
2018 column name specified in a query structure. This is a safety mechanism to avoid
2019 injection attacks when mishandling user input e.g.:
2021 my %condition_as_column_value_pairs = get_values_from_user();
2022 $sqla->select( ... , \%condition_as_column_value_pairs );
2024 If the expression matches an exception is thrown. Note that literal SQL
2025 supplied via C<\'...'> or C<\['...']> is B<not> checked in any way.
2027 Defaults to checking for C<;> and the C<GO> keyword (TransactSQL)
2029 =item array_datatypes
2031 When this option is true, arrayrefs in INSERT or UPDATE are
2032 interpreted as array datatypes and are passed directly
2034 When this option is false, arrayrefs are interpreted
2035 as literal SQL, just like refs to arrayrefs
2036 (but this behavior is for backwards compatibility; when writing
2037 new queries, use the "reference to arrayref" syntax
2043 Takes a reference to a list of "special operators"
2044 to extend the syntax understood by L<SQL::Abstract>.
2045 See section L</"SPECIAL OPERATORS"> for details.
2049 Takes a reference to a list of "unary operators"
2050 to extend the syntax understood by L<SQL::Abstract>.
2051 See section L</"UNARY OPERATORS"> for details.
2057 =head2 insert($table, \@values || \%fieldvals, \%options)
2059 This is the simplest function. You simply give it a table name
2060 and either an arrayref of values or hashref of field/value pairs.
2061 It returns an SQL INSERT statement and a list of bind values.
2062 See the sections on L</"Inserting and Updating Arrays"> and
2063 L</"Inserting and Updating SQL"> for information on how to insert
2064 with those data types.
2066 The optional C<\%options> hash reference may contain additional
2067 options to generate the insert SQL. Currently supported options
2074 Takes either a scalar of raw SQL fields, or an array reference of
2075 field names, and adds on an SQL C<RETURNING> statement at the end.
2076 This allows you to return data generated by the insert statement
2077 (such as row IDs) without performing another C<SELECT> statement.
2078 Note, however, this is not part of the SQL standard and may not
2079 be supported by all database engines.
2083 =head2 update($table, \%fieldvals, \%where, \%options)
2085 This takes a table, hashref of field/value pairs, and an optional
2086 hashref L<WHERE clause|/WHERE CLAUSES>. It returns an SQL UPDATE function and a list
2088 See the sections on L</"Inserting and Updating Arrays"> and
2089 L</"Inserting and Updating SQL"> for information on how to insert
2090 with those data types.
2092 The optional C<\%options> hash reference may contain additional
2093 options to generate the update SQL. Currently supported options
2100 See the C<returning> option to
2101 L<insert|/insert($table, \@values || \%fieldvals, \%options)>.
2105 =head2 select($source, $fields, $where, $order)
2107 This returns a SQL SELECT statement and associated list of bind values, as
2108 specified by the arguments :
2114 Specification of the 'FROM' part of the statement.
2115 The argument can be either a plain scalar (interpreted as a table
2116 name, will be quoted), or an arrayref (interpreted as a list
2117 of table names, joined by commas, quoted), or a scalarref
2118 (literal table name, not quoted), or a ref to an arrayref
2119 (list of literal table names, joined by commas, not quoted).
2123 Specification of the list of fields to retrieve from
2125 The argument can be either an arrayref (interpreted as a list
2126 of field names, will be joined by commas and quoted), or a
2127 plain scalar (literal SQL, not quoted).
2128 Please observe that this API is not as flexible as that of
2129 the first argument C<$source>, for backwards compatibility reasons.
2133 Optional argument to specify the WHERE part of the query.
2134 The argument is most often a hashref, but can also be
2135 an arrayref or plain scalar --
2136 see section L<WHERE clause|/"WHERE CLAUSES"> for details.
2140 Optional argument to specify the ORDER BY part of the query.
2141 The argument can be a scalar, a hashref or an arrayref
2142 -- see section L<ORDER BY clause|/"ORDER BY CLAUSES">
2148 =head2 delete($table, \%where)
2150 This takes a table name and optional hashref L<WHERE clause|/WHERE CLAUSES>.
2151 It returns an SQL DELETE statement and list of bind values.
2153 =head2 where(\%where, \@order)
2155 This is used to generate just the WHERE clause. For example,
2156 if you have an arbitrary data structure and know what the
2157 rest of your SQL is going to look like, but want an easy way
2158 to produce a WHERE clause, use this. It returns an SQL WHERE
2159 clause and list of bind values.
2162 =head2 values(\%data)
2164 This just returns the values from the hash C<%data>, in the same
2165 order that would be returned from any of the other above queries.
2166 Using this allows you to markedly speed up your queries if you
2167 are affecting lots of rows. See below under the L</"PERFORMANCE"> section.
2169 =head2 generate($any, 'number', $of, \@data, $struct, \%types)
2171 Warning: This is an experimental method and subject to change.
2173 This returns arbitrarily generated SQL. It's a really basic shortcut.
2174 It will return two different things, depending on return context:
2176 my($stmt, @bind) = $sql->generate('create table', \$table, \@fields);
2177 my $stmt_and_val = $sql->generate('create table', \$table, \@fields);
2179 These would return the following:
2181 # First calling form
2182 $stmt = "CREATE TABLE test (?, ?)";
2183 @bind = (field1, field2);
2185 # Second calling form
2186 $stmt_and_val = "CREATE TABLE test (field1, field2)";
2188 Depending on what you're trying to do, it's up to you to choose the correct
2189 format. In this example, the second form is what you would want.
2193 $sql->generate('alter session', { nls_date_format => 'MM/YY' });
2197 ALTER SESSION SET nls_date_format = 'MM/YY'
2199 You get the idea. Strings get their case twiddled, but everything
2200 else remains verbatim.
2202 =head1 EXPORTABLE FUNCTIONS
2204 =head2 is_plain_value
2206 Determines if the supplied argument is a plain value as understood by this
2211 =item * The value is C<undef>
2213 =item * The value is a non-reference
2215 =item * The value is an object with stringification overloading
2217 =item * The value is of the form C<< { -value => $anything } >>
2221 On failure returns C<undef>, on sucess returns a B<scalar> reference
2222 to the original supplied argument.
2228 The stringification overloading detection is rather advanced: it takes
2229 into consideration not only the presence of a C<""> overload, but if that
2230 fails also checks for enabled
2231 L<autogenerated versions of C<"">|overload/Magic Autogeneration>, based
2232 on either C<0+> or C<bool>.
2234 Unfortunately testing in the field indicates that this
2235 detection B<< may tickle a latent bug in perl versions before 5.018 >>,
2236 but only when very large numbers of stringifying objects are involved.
2237 At the time of writing ( Sep 2014 ) there is no clear explanation of
2238 the direct cause, nor is there a manageably small test case that reliably
2239 reproduces the problem.
2241 If you encounter any of the following exceptions in B<random places within
2242 your application stack> - this module may be to blame:
2244 Operation "ne": no method found,
2245 left argument in overloaded package <something>,
2246 right argument in overloaded package <something>
2250 Stub found while resolving method "???" overloading """" in package <something>
2252 If you fall victim to the above - please attempt to reduce the problem
2253 to something that could be sent to the L<SQL::Abstract developers
2254 |DBIx::Class/GETTING HELP/SUPPORT>
2255 (either publicly or privately). As a workaround in the meantime you can
2256 set C<$ENV{SQLA_ISVALUE_IGNORE_AUTOGENERATED_STRINGIFICATION}> to a true
2257 value, which will most likely eliminate your problem (at the expense of
2258 not being able to properly detect exotic forms of stringification).
2260 This notice and environment variable will be removed in a future version,
2261 as soon as the underlying problem is found and a reliable workaround is
2266 =head2 is_literal_value
2268 Determines if the supplied argument is a literal value as understood by this
2273 =item * C<\$sql_string>
2275 =item * C<\[ $sql_string, @bind_values ]>
2279 On failure returns C<undef>, on sucess returns an B<array> reference
2280 containing the unpacked version of the supplied literal SQL and bind values.
2282 =head1 WHERE CLAUSES
2286 This module uses a variation on the idea from L<DBIx::Abstract>. It
2287 is B<NOT>, repeat I<not> 100% compatible. B<The main logic of this
2288 module is that things in arrays are OR'ed, and things in hashes
2291 The easiest way to explain is to show lots of examples. After
2292 each C<%where> hash shown, it is assumed you used:
2294 my($stmt, @bind) = $sql->where(\%where);
2296 However, note that the C<%where> hash can be used directly in any
2297 of the other functions as well, as described above.
2299 =head2 Key-value pairs
2301 So, let's get started. To begin, a simple hash:
2305 status => 'completed'
2308 Is converted to SQL C<key = val> statements:
2310 $stmt = "WHERE user = ? AND status = ?";
2311 @bind = ('nwiger', 'completed');
2313 One common thing I end up doing is having a list of values that
2314 a field can be in. To do this, simply specify a list inside of
2319 status => ['assigned', 'in-progress', 'pending'];
2322 This simple code will create the following:
2324 $stmt = "WHERE user = ? AND ( status = ? OR status = ? OR status = ? )";
2325 @bind = ('nwiger', 'assigned', 'in-progress', 'pending');
2327 A field associated to an empty arrayref will be considered a
2328 logical false and will generate 0=1.
2330 =head2 Tests for NULL values
2332 If the value part is C<undef> then this is converted to SQL <IS NULL>
2341 $stmt = "WHERE user = ? AND status IS NULL";
2344 To test if a column IS NOT NULL:
2348 status => { '!=', undef },
2351 =head2 Specific comparison operators
2353 If you want to specify a different type of operator for your comparison,
2354 you can use a hashref for a given column:
2358 status => { '!=', 'completed' }
2361 Which would generate:
2363 $stmt = "WHERE user = ? AND status != ?";
2364 @bind = ('nwiger', 'completed');
2366 To test against multiple values, just enclose the values in an arrayref:
2368 status => { '=', ['assigned', 'in-progress', 'pending'] };
2370 Which would give you:
2372 "WHERE status = ? OR status = ? OR status = ?"
2375 The hashref can also contain multiple pairs, in which case it is expanded
2376 into an C<AND> of its elements:
2380 status => { '!=', 'completed', -not_like => 'pending%' }
2383 # Or more dynamically, like from a form
2384 $where{user} = 'nwiger';
2385 $where{status}{'!='} = 'completed';
2386 $where{status}{'-not_like'} = 'pending%';
2388 # Both generate this
2389 $stmt = "WHERE user = ? AND status != ? AND status NOT LIKE ?";
2390 @bind = ('nwiger', 'completed', 'pending%');
2393 To get an OR instead, you can combine it with the arrayref idea:
2397 priority => [ { '=', 2 }, { '>', 5 } ]
2400 Which would generate:
2402 $stmt = "WHERE ( priority = ? OR priority > ? ) AND user = ?";
2403 @bind = ('2', '5', 'nwiger');
2405 If you want to include literal SQL (with or without bind values), just use a
2406 scalar reference or reference to an arrayref as the value:
2409 date_entered => { '>' => \["to_date(?, 'MM/DD/YYYY')", "11/26/2008"] },
2410 date_expires => { '<' => \"now()" }
2413 Which would generate:
2415 $stmt = "WHERE date_entered > to_date(?, 'MM/DD/YYYY') AND date_expires < now()";
2416 @bind = ('11/26/2008');
2419 =head2 Logic and nesting operators
2421 In the example above,
2422 there is a subtle trap if you want to say something like
2423 this (notice the C<AND>):
2425 WHERE priority != ? AND priority != ?
2427 Because, in Perl you I<can't> do this:
2429 priority => { '!=' => 2, '!=' => 1 }
2431 As the second C<!=> key will obliterate the first. The solution
2432 is to use the special C<-modifier> form inside an arrayref:
2434 priority => [ -and => {'!=', 2},
2438 Normally, these would be joined by C<OR>, but the modifier tells it
2439 to use C<AND> instead. (Hint: You can use this in conjunction with the
2440 C<logic> option to C<new()> in order to change the way your queries
2441 work by default.) B<Important:> Note that the C<-modifier> goes
2442 B<INSIDE> the arrayref, as an extra first element. This will
2443 B<NOT> do what you think it might:
2445 priority => -and => [{'!=', 2}, {'!=', 1}] # WRONG!
2447 Here is a quick list of equivalencies, since there is some overlap:
2450 status => {'!=', 'completed', 'not like', 'pending%' }
2451 status => [ -and => {'!=', 'completed'}, {'not like', 'pending%'}]
2454 status => {'=', ['assigned', 'in-progress']}
2455 status => [ -or => {'=', 'assigned'}, {'=', 'in-progress'}]
2456 status => [ {'=', 'assigned'}, {'=', 'in-progress'} ]
2460 =head2 Special operators : IN, BETWEEN, etc.
2462 You can also use the hashref format to compare a list of fields using the
2463 C<IN> comparison operator, by specifying the list as an arrayref:
2466 status => 'completed',
2467 reportid => { -in => [567, 2335, 2] }
2470 Which would generate:
2472 $stmt = "WHERE status = ? AND reportid IN (?,?,?)";
2473 @bind = ('completed', '567', '2335', '2');
2475 The reverse operator C<-not_in> generates SQL C<NOT IN> and is used in
2478 If the argument to C<-in> is an empty array, 'sqlfalse' is generated
2479 (by default : C<1=0>). Similarly, C<< -not_in => [] >> generates
2480 'sqltrue' (by default : C<1=1>).
2482 In addition to the array you can supply a chunk of literal sql or
2483 literal sql with bind:
2486 customer => { -in => \[
2487 'SELECT cust_id FROM cust WHERE balance > ?',
2490 status => { -in => \'SELECT status_codes FROM states' },
2496 customer IN ( SELECT cust_id FROM cust WHERE balance > ? )
2497 AND status IN ( SELECT status_codes FROM states )
2501 Finally, if the argument to C<-in> is not a reference, it will be
2502 treated as a single-element array.
2504 Another pair of operators is C<-between> and C<-not_between>,
2505 used with an arrayref of two values:
2509 completion_date => {
2510 -not_between => ['2002-10-01', '2003-02-06']
2516 WHERE user = ? AND completion_date NOT BETWEEN ( ? AND ? )
2518 Just like with C<-in> all plausible combinations of literal SQL
2522 start0 => { -between => [ 1, 2 ] },
2523 start1 => { -between => \["? AND ?", 1, 2] },
2524 start2 => { -between => \"lower(x) AND upper(y)" },
2525 start3 => { -between => [
2527 \["upper(?)", 'stuff' ],
2534 ( start0 BETWEEN ? AND ? )
2535 AND ( start1 BETWEEN ? AND ? )
2536 AND ( start2 BETWEEN lower(x) AND upper(y) )
2537 AND ( start3 BETWEEN lower(x) AND upper(?) )
2539 @bind = (1, 2, 1, 2, 'stuff');
2542 These are the two builtin "special operators"; but the
2543 list can be expanded : see section L</"SPECIAL OPERATORS"> below.
2545 =head2 Unary operators: bool
2547 If you wish to test against boolean columns or functions within your
2548 database you can use the C<-bool> and C<-not_bool> operators. For
2549 example to test the column C<is_user> being true and the column
2550 C<is_enabled> being false you would use:-
2554 -not_bool => 'is_enabled',
2559 WHERE is_user AND NOT is_enabled
2561 If a more complex combination is required, testing more conditions,
2562 then you should use the and/or operators:-
2567 -not_bool => { two=> { -rlike => 'bar' } },
2568 -not_bool => { three => [ { '=', 2 }, { '>', 5 } ] },
2579 (NOT ( three = ? OR three > ? ))
2582 =head2 Nested conditions, -and/-or prefixes
2584 So far, we've seen how multiple conditions are joined with a top-level
2585 C<AND>. We can change this by putting the different conditions we want in
2586 hashes and then putting those hashes in an array. For example:
2591 status => { -like => ['pending%', 'dispatched'] },
2595 status => 'unassigned',
2599 This data structure would create the following:
2601 $stmt = "WHERE ( user = ? AND ( status LIKE ? OR status LIKE ? ) )
2602 OR ( user = ? AND status = ? ) )";
2603 @bind = ('nwiger', 'pending', 'dispatched', 'robot', 'unassigned');
2606 Clauses in hashrefs or arrayrefs can be prefixed with an C<-and> or C<-or>
2607 to change the logic inside :
2613 -and => [ workhrs => {'>', 20}, geo => 'ASIA' ],
2614 -or => { workhrs => {'<', 50}, geo => 'EURO' },
2621 $stmt = "WHERE ( user = ?
2622 AND ( ( workhrs > ? AND geo = ? )
2623 OR ( workhrs < ? OR geo = ? ) ) )";
2624 @bind = ('nwiger', '20', 'ASIA', '50', 'EURO');
2626 =head3 Algebraic inconsistency, for historical reasons
2628 C<Important note>: when connecting several conditions, the C<-and->|C<-or>
2629 operator goes C<outside> of the nested structure; whereas when connecting
2630 several constraints on one column, the C<-and> operator goes
2631 C<inside> the arrayref. Here is an example combining both features :
2634 -and => [a => 1, b => 2],
2635 -or => [c => 3, d => 4],
2636 e => [-and => {-like => 'foo%'}, {-like => '%bar'} ]
2641 WHERE ( ( ( a = ? AND b = ? )
2642 OR ( c = ? OR d = ? )
2643 OR ( e LIKE ? AND e LIKE ? ) ) )
2645 This difference in syntax is unfortunate but must be preserved for
2646 historical reasons. So be careful : the two examples below would
2647 seem algebraically equivalent, but they are not
2650 { -like => 'foo%' },
2651 { -like => '%bar' },
2653 # yields : WHERE ( ( col LIKE ? AND col LIKE ? ) )
2656 { col => { -like => 'foo%' } },
2657 { col => { -like => '%bar' } },
2659 # yields : WHERE ( ( col LIKE ? OR col LIKE ? ) )
2662 =head2 Literal SQL and value type operators
2664 The basic premise of SQL::Abstract is that in WHERE specifications the "left
2665 side" is a column name and the "right side" is a value (normally rendered as
2666 a placeholder). This holds true for both hashrefs and arrayref pairs as you
2667 see in the L</WHERE CLAUSES> examples above. Sometimes it is necessary to
2668 alter this behavior. There are several ways of doing so.
2672 This is a virtual operator that signals the string to its right side is an
2673 identifier (a column name) and not a value. For example to compare two
2674 columns you would write:
2677 priority => { '<', 2 },
2678 requestor => { -ident => 'submitter' },
2683 $stmt = "WHERE priority < ? AND requestor = submitter";
2686 If you are maintaining legacy code you may see a different construct as
2687 described in L</Deprecated usage of Literal SQL>, please use C<-ident> in new
2692 This is a virtual operator that signals that the construct to its right side
2693 is a value to be passed to DBI. This is for example necessary when you want
2694 to write a where clause against an array (for RDBMS that support such
2695 datatypes). For example:
2698 array => { -value => [1, 2, 3] }
2703 $stmt = 'WHERE array = ?';
2704 @bind = ([1, 2, 3]);
2706 Note that if you were to simply say:
2712 the result would probably not be what you wanted:
2714 $stmt = 'WHERE array = ? OR array = ? OR array = ?';
2719 Finally, sometimes only literal SQL will do. To include a random snippet
2720 of SQL verbatim, you specify it as a scalar reference. Consider this only
2721 as a last resort. Usually there is a better way. For example:
2724 priority => { '<', 2 },
2725 requestor => { -in => \'(SELECT name FROM hitmen)' },
2730 $stmt = "WHERE priority < ? AND requestor IN (SELECT name FROM hitmen)"
2733 Note that in this example, you only get one bind parameter back, since
2734 the verbatim SQL is passed as part of the statement.
2738 Never use untrusted input as a literal SQL argument - this is a massive
2739 security risk (there is no way to check literal snippets for SQL
2740 injections and other nastyness). If you need to deal with untrusted input
2741 use literal SQL with placeholders as described next.
2743 =head3 Literal SQL with placeholders and bind values (subqueries)
2745 If the literal SQL to be inserted has placeholders and bind values,
2746 use a reference to an arrayref (yes this is a double reference --
2747 not so common, but perfectly legal Perl). For example, to find a date
2748 in Postgres you can use something like this:
2751 date_column => \[ "= date '2008-09-30' - ?::integer", 10 ]
2756 $stmt = "WHERE ( date_column = date '2008-09-30' - ?::integer )"
2759 Note that you must pass the bind values in the same format as they are returned
2760 by L<where|/where(\%where, \@order)>. This means that if you set L</bindtype>
2761 to C<columns>, you must provide the bind values in the
2762 C<< [ column_meta => value ] >> format, where C<column_meta> is an opaque
2763 scalar value; most commonly the column name, but you can use any scalar value
2764 (including references and blessed references), L<SQL::Abstract> will simply
2765 pass it through intact. So if C<bindtype> is set to C<columns> the above
2766 example will look like:
2769 date_column => \[ "= date '2008-09-30' - ?::integer", [ {} => 10 ] ]
2772 Literal SQL is especially useful for nesting parenthesized clauses in the
2773 main SQL query. Here is a first example :
2775 my ($sub_stmt, @sub_bind) = ("SELECT c1 FROM t1 WHERE c2 < ? AND c3 LIKE ?",
2779 bar => \["IN ($sub_stmt)" => @sub_bind],
2784 $stmt = "WHERE (foo = ? AND bar IN (SELECT c1 FROM t1
2785 WHERE c2 < ? AND c3 LIKE ?))";
2786 @bind = (1234, 100, "foo%");
2788 Other subquery operators, like for example C<"E<gt> ALL"> or C<"NOT IN">,
2789 are expressed in the same way. Of course the C<$sub_stmt> and
2790 its associated bind values can be generated through a former call
2793 my ($sub_stmt, @sub_bind)
2794 = $sql->select("t1", "c1", {c2 => {"<" => 100},
2795 c3 => {-like => "foo%"}});
2798 bar => \["> ALL ($sub_stmt)" => @sub_bind],
2801 In the examples above, the subquery was used as an operator on a column;
2802 but the same principle also applies for a clause within the main C<%where>
2803 hash, like an EXISTS subquery :
2805 my ($sub_stmt, @sub_bind)
2806 = $sql->select("t1", "*", {c1 => 1, c2 => \"> t0.c0"});
2807 my %where = ( -and => [
2809 \["EXISTS ($sub_stmt)" => @sub_bind],
2814 $stmt = "WHERE (foo = ? AND EXISTS (SELECT * FROM t1
2815 WHERE c1 = ? AND c2 > t0.c0))";
2819 Observe that the condition on C<c2> in the subquery refers to
2820 column C<t0.c0> of the main query : this is I<not> a bind
2821 value, so we have to express it through a scalar ref.
2822 Writing C<< c2 => {">" => "t0.c0"} >> would have generated
2823 C<< c2 > ? >> with bind value C<"t0.c0"> ... not exactly
2824 what we wanted here.
2826 Finally, here is an example where a subquery is used
2827 for expressing unary negation:
2829 my ($sub_stmt, @sub_bind)
2830 = $sql->where({age => [{"<" => 10}, {">" => 20}]});
2831 $sub_stmt =~ s/^ where //i; # don't want "WHERE" in the subclause
2833 lname => {like => '%son%'},
2834 \["NOT ($sub_stmt)" => @sub_bind],
2839 $stmt = "lname LIKE ? AND NOT ( age < ? OR age > ? )"
2840 @bind = ('%son%', 10, 20)
2842 =head3 Deprecated usage of Literal SQL
2844 Below are some examples of archaic use of literal SQL. It is shown only as
2845 reference for those who deal with legacy code. Each example has a much
2846 better, cleaner and safer alternative that users should opt for in new code.
2852 my %where = ( requestor => \'IS NOT NULL' )
2854 $stmt = "WHERE requestor IS NOT NULL"
2856 This used to be the way of generating NULL comparisons, before the handling
2857 of C<undef> got formalized. For new code please use the superior syntax as
2858 described in L</Tests for NULL values>.
2862 my %where = ( requestor => \'= submitter' )
2864 $stmt = "WHERE requestor = submitter"
2866 This used to be the only way to compare columns. Use the superior L</-ident>
2867 method for all new code. For example an identifier declared in such a way
2868 will be properly quoted if L</quote_char> is properly set, while the legacy
2869 form will remain as supplied.
2873 my %where = ( is_ready => \"", completed => { '>', '2012-12-21' } )
2875 $stmt = "WHERE completed > ? AND is_ready"
2876 @bind = ('2012-12-21')
2878 Using an empty string literal used to be the only way to express a boolean.
2879 For all new code please use the much more readable
2880 L<-bool|/Unary operators: bool> operator.
2886 These pages could go on for a while, since the nesting of the data
2887 structures this module can handle are pretty much unlimited (the
2888 module implements the C<WHERE> expansion as a recursive function
2889 internally). Your best bet is to "play around" with the module a
2890 little to see how the data structures behave, and choose the best
2891 format for your data based on that.
2893 And of course, all the values above will probably be replaced with
2894 variables gotten from forms or the command line. After all, if you
2895 knew everything ahead of time, you wouldn't have to worry about
2896 dynamically-generating SQL and could just hardwire it into your
2899 =head1 ORDER BY CLAUSES
2901 Some functions take an order by clause. This can either be a scalar (just a
2902 column name,) a hash of C<< { -desc => 'col' } >> or C<< { -asc => 'col' } >>,
2903 or an array of either of the two previous forms. Examples:
2905 Given | Will Generate
2906 ----------------------------------------------------------
2908 \'colA DESC' | ORDER BY colA DESC
2910 'colA' | ORDER BY colA
2912 [qw/colA colB/] | ORDER BY colA, colB
2914 {-asc => 'colA'} | ORDER BY colA ASC
2916 {-desc => 'colB'} | ORDER BY colB DESC
2918 ['colA', {-asc => 'colB'}] | ORDER BY colA, colB ASC
2920 { -asc => [qw/colA colB/] } | ORDER BY colA ASC, colB ASC
2923 { -asc => 'colA' }, | ORDER BY colA ASC, colB DESC,
2924 { -desc => [qw/colB/], | colC ASC, colD ASC
2925 { -asc => [qw/colC colD/],|
2927 ===========================================================
2931 =head1 SPECIAL OPERATORS
2933 my $sqlmaker = SQL::Abstract->new(special_ops => [
2937 my ($self, $field, $op, $arg) = @_;
2943 handler => 'method_name',
2947 A "special operator" is a SQL syntactic clause that can be
2948 applied to a field, instead of a usual binary operator.
2951 WHERE field IN (?, ?, ?)
2952 WHERE field BETWEEN ? AND ?
2953 WHERE MATCH(field) AGAINST (?, ?)
2955 Special operators IN and BETWEEN are fairly standard and therefore
2956 are builtin within C<SQL::Abstract> (as the overridable methods
2957 C<_where_field_IN> and C<_where_field_BETWEEN>). For other operators,
2958 like the MATCH .. AGAINST example above which is specific to MySQL,
2959 you can write your own operator handlers - supply a C<special_ops>
2960 argument to the C<new> method. That argument takes an arrayref of
2961 operator definitions; each operator definition is a hashref with two
2968 the regular expression to match the operator
2972 Either a coderef or a plain scalar method name. In both cases
2973 the expected return is C<< ($sql, @bind) >>.
2975 When supplied with a method name, it is simply called on the
2976 L<SQL::Abstract> object as:
2978 $self->$method_name ($field, $op, $arg)
2982 $field is the LHS of the operator
2983 $op is the part that matched the handler regex
2986 When supplied with a coderef, it is called as:
2988 $coderef->($self, $field, $op, $arg)
2993 For example, here is an implementation
2994 of the MATCH .. AGAINST syntax for MySQL
2996 my $sqlmaker = SQL::Abstract->new(special_ops => [
2998 # special op for MySql MATCH (field) AGAINST(word1, word2, ...)
2999 {regex => qr/^match$/i,
3001 my ($self, $field, $op, $arg) = @_;
3002 $arg = [$arg] if not ref $arg;
3003 my $label = $self->_quote($field);
3004 my ($placeholder) = $self->_convert('?');
3005 my $placeholders = join ", ", (($placeholder) x @$arg);
3006 my $sql = $self->_sqlcase('match') . " ($label) "
3007 . $self->_sqlcase('against') . " ($placeholders) ";
3008 my @bind = $self->_bindtype($field, @$arg);
3009 return ($sql, @bind);
3016 =head1 UNARY OPERATORS
3018 my $sqlmaker = SQL::Abstract->new(unary_ops => [
3022 my ($self, $op, $arg) = @_;
3028 handler => 'method_name',
3032 A "unary operator" is a SQL syntactic clause that can be
3033 applied to a field - the operator goes before the field
3035 You can write your own operator handlers - supply a C<unary_ops>
3036 argument to the C<new> method. That argument takes an arrayref of
3037 operator definitions; each operator definition is a hashref with two
3044 the regular expression to match the operator
3048 Either a coderef or a plain scalar method name. In both cases
3049 the expected return is C<< $sql >>.
3051 When supplied with a method name, it is simply called on the
3052 L<SQL::Abstract> object as:
3054 $self->$method_name ($op, $arg)
3058 $op is the part that matched the handler regex
3059 $arg is the RHS or argument of the operator
3061 When supplied with a coderef, it is called as:
3063 $coderef->($self, $op, $arg)
3071 Thanks to some benchmarking by Mark Stosberg, it turns out that
3072 this module is many orders of magnitude faster than using C<DBIx::Abstract>.
3073 I must admit this wasn't an intentional design issue, but it's a
3074 byproduct of the fact that you get to control your C<DBI> handles
3077 To maximize performance, use a code snippet like the following:
3079 # prepare a statement handle using the first row
3080 # and then reuse it for the rest of the rows
3082 for my $href (@array_of_hashrefs) {
3083 $stmt ||= $sql->insert('table', $href);
3084 $sth ||= $dbh->prepare($stmt);
3085 $sth->execute($sql->values($href));
3088 The reason this works is because the keys in your C<$href> are sorted
3089 internally by B<SQL::Abstract>. Thus, as long as your data retains
3090 the same structure, you only have to generate the SQL the first time
3091 around. On subsequent queries, simply use the C<values> function provided
3092 by this module to return your values in the correct order.
3094 However this depends on the values having the same type - if, for
3095 example, the values of a where clause may either have values
3096 (resulting in sql of the form C<column = ?> with a single bind
3097 value), or alternatively the values might be C<undef> (resulting in
3098 sql of the form C<column IS NULL> with no bind value) then the
3099 caching technique suggested will not work.
3103 If you use my C<CGI::FormBuilder> module at all, you'll hopefully
3104 really like this part (I do, at least). Building up a complex query
3105 can be as simple as the following:
3112 use CGI::FormBuilder;
3115 my $form = CGI::FormBuilder->new(...);
3116 my $sql = SQL::Abstract->new;
3118 if ($form->submitted) {
3119 my $field = $form->field;
3120 my $id = delete $field->{id};
3121 my($stmt, @bind) = $sql->update('table', $field, {id => $id});
3124 Of course, you would still have to connect using C<DBI> to run the
3125 query, but the point is that if you make your form look like your
3126 table, the actual query script can be extremely simplistic.
3128 If you're B<REALLY> lazy (I am), check out C<HTML::QuickTable> for
3129 a fast interface to returning and formatting data. I frequently
3130 use these three modules together to write complex database query
3131 apps in under 50 lines.
3133 =head1 HOW TO CONTRIBUTE
3135 Contributions are always welcome, in all usable forms (we especially
3136 welcome documentation improvements). The delivery methods include git-
3137 or unified-diff formatted patches, GitHub pull requests, or plain bug
3138 reports either via RT or the Mailing list. Contributors are generally
3139 granted full access to the official repository after their first several
3140 patches pass successful review.
3142 This project is maintained in a git repository. The code and related tools are
3143 accessible at the following locations:
3147 =item * Official repo: L<git://git.shadowcat.co.uk/dbsrgits/SQL-Abstract.git>
3149 =item * Official gitweb: L<http://git.shadowcat.co.uk/gitweb/gitweb.cgi?p=dbsrgits/SQL-Abstract.git>
3151 =item * GitHub mirror: L<https://github.com/dbsrgits/sql-abstract>
3153 =item * Authorized committers: L<ssh://dbsrgits@git.shadowcat.co.uk/SQL-Abstract.git>
3159 Version 1.50 was a major internal refactoring of C<SQL::Abstract>.
3160 Great care has been taken to preserve the I<published> behavior
3161 documented in previous versions in the 1.* family; however,
3162 some features that were previously undocumented, or behaved
3163 differently from the documentation, had to be changed in order
3164 to clarify the semantics. Hence, client code that was relying
3165 on some dark areas of C<SQL::Abstract> v1.*
3166 B<might behave differently> in v1.50.
3168 The main changes are :
3174 support for literal SQL through the C<< \ [ $sql, @bind ] >> syntax.
3178 support for the { operator => \"..." } construct (to embed literal SQL)
3182 support for the { operator => \["...", @bind] } construct (to embed literal SQL with bind values)
3186 optional support for L<array datatypes|/"Inserting and Updating Arrays">
3190 defensive programming : check arguments
3194 fixed bug with global logic, which was previously implemented
3195 through global variables yielding side-effects. Prior versions would
3196 interpret C<< [ {cond1, cond2}, [cond3, cond4] ] >>
3197 as C<< "(cond1 AND cond2) OR (cond3 AND cond4)" >>.
3198 Now this is interpreted
3199 as C<< "(cond1 AND cond2) OR (cond3 OR cond4)" >>.
3204 fixed semantics of _bindtype on array args
3208 dropped the C<_anoncopy> of the %where tree. No longer necessary,
3209 we just avoid shifting arrays within that tree.
3213 dropped the C<_modlogic> function
3217 =head1 ACKNOWLEDGEMENTS
3219 There are a number of individuals that have really helped out with
3220 this module. Unfortunately, most of them submitted bugs via CPAN
3221 so I have no idea who they are! But the people I do know are:
3223 Ash Berlin (order_by hash term support)
3224 Matt Trout (DBIx::Class support)
3225 Mark Stosberg (benchmarking)
3226 Chas Owens (initial "IN" operator support)
3227 Philip Collins (per-field SQL functions)
3228 Eric Kolve (hashref "AND" support)
3229 Mike Fragassi (enhancements to "BETWEEN" and "LIKE")
3230 Dan Kubb (support for "quote_char" and "name_sep")
3231 Guillermo Roditi (patch to cleanup "IN" and "BETWEEN", fix and tests for _order_by)
3232 Laurent Dami (internal refactoring, extensible list of special operators, literal SQL)
3233 Norbert Buchmuller (support for literal SQL in hashpair, misc. fixes & tests)
3234 Peter Rabbitson (rewrite of SQLA::Test, misc. fixes & tests)
3235 Oliver Charles (support for "RETURNING" after "INSERT")
3241 L<DBIx::Class>, L<DBIx::Abstract>, L<CGI::FormBuilder>, L<HTML::QuickTable>.
3245 Copyright (c) 2001-2007 Nathan Wiger <nwiger@cpan.org>. All Rights Reserved.
3247 This module is actively maintained by Matt Trout <mst@shadowcatsystems.co.uk>
3249 For support, your best bet is to try the C<DBIx::Class> users mailing list.
3250 While not an official support venue, C<DBIx::Class> makes heavy use of
3251 C<SQL::Abstract>, and as such list members there are very familiar with
3252 how to create queries.
3256 This module is free software; you may copy this under the same
3257 terms as perl itself (either the GNU General Public License or
3258 the Artistic License)